Overview
x265.changes
Changed
x
1
2
-------------------------------------------------------------------3
+Thu Jun 13 05:58:19 UTC 2024 - Luigi Baldoni <aloisio@gmx.com>4
+5
+- Update to version 3.66
+ New features:7
+ * Segment based Ratecontrol (SBRC) feature8
+ * Motion-Compensated Spatio-Temporal Filtering9
+ * Scene-cut aware qp - BBAQ (Bidirectional Boundary Aware10
+ Quantization)11
+ * Histogram-Based Scene Change Detection12
+ * Film-Grain characteristics as a SEI message to support Film13
+ Grain Synthesis(FGS)14
+ * Add temporal layer implementation(Hierarchical B-frame15
+ implementation)16
+ Enhancements to existing features:17
+ * Added Dolby Vision 8.4 Profile Support18
+ API changes:19
+ * Add Segment based Ratecontrol(SBRC) feature: "--no-sbrc".20
+ * Add command line parameter for mcstf feature: "--no-mctf".21
+ * Add command line parameters for the scene cut aware qp22
+ feature: "--scenecut-aware-qp" and "--masking-strength".23
+ * Add command line parameters for Histogram-Based Scene Change24
+ Detection: "--hist-scenecut".25
+ * Add film grain characteristics as a SEI message to the26
+ bitstream: "--film-grain <filename>"27
+ * cli: add new option --cra-nal (Force nal type to CRA to all28
+ frames expect for the first frame, works only with keyint 1)29
+ Optimizations:30
+ * ARM64 NEON optimizations:- Several time-consuming C31
+ functions have been optimized for the targeted platform -32
+ aarch64. The overall performance increased by around 20%.33
+ * SVE/SVE2 optimizations34
+ Bug fixes:35
+ * Linux bug to utilize all the cores36
+ * Crash with hist-scenecut build when source resolution is not37
+ multiple of minCuSize38
+ * 32bit and 64bit builds generation for ARM39
+ * bugs in zonefile feature (Reflect Zonefile Parameters inside40
+ Lookahead, extra IDR issue, Avg I Slice QP value issue etc..)41
+ * Add x86 ASM implementation for subsampling luma42
+ * Fix for abrladder segfault with load reuse level 143
+ * Reorder miniGOP based on temporal layer hierarchy and add44
+ support for more B frame45
+ * Add MacOS aarch64 build support46
+ * Fix boundary condition issue for Gaussian filter47
+- Drop arm.patch and replace it with 0001-Fix-arm-flags.patch48
+ and 0004-Do-not-build-with-assembly-support-on-arm.patch49
+ (courtesy of Debian)50
+51
+-------------------------------------------------------------------52
Wed May 19 13:21:09 UTC 2021 - Luigi Baldoni <aloisio@gmx.com>53
54
- Build libx265_main10 and libx265_main12 unconditionally and55
x265.spec
Changed
46
1
2
#3
# spec file for package x2654
#5
-# Copyright (c) 2021 Packman Team <packman@links2linux.de>6
+# Copyright (c) 2024 Packman Team <packman@links2linux.de>7
# Copyright (c) 2014 Torsten Gruner <t.gruner@katodev.de>8
#9
# All modifications and additions to the file contributed by third parties10
11
#12
13
14
-%define sover 19915
+%define sover 20916
%define libname lib%{name}17
%define libsoname %{libname}-%{sover}18
-%define uver 3_519
+%define uver 3_620
Name: x26521
-Version: 3.522
+Version: 3.623
Release: 024
Summary: A free h265/HEVC encoder - encoder binary25
License: GPL-2.0-or-later26
Group: Productivity/Multimedia/Video/Editors and Convertors27
URL: https://bitbucket.org/multicoreware/x265_git28
Source0: https://bitbucket.org/multicoreware/x265_git/downloads/%{name}_%{version}.tar.gz29
-Patch0: arm.patch30
Patch1: x265.pkgconfig.patch31
Patch2: x265-fix_enable512.patch32
+Patch3: 0001-Fix-arm-flags.patch33
+Patch4: 0004-Do-not-build-with-assembly-support-on-arm.patch34
BuildRequires: cmake >= 2.8.835
BuildRequires: gcc-c++36
BuildRequires: nasm >= 2.1337
38
%cmake_install39
find %{buildroot} -type f -name "*.a" -delete -print040
41
+%check42
+43
%post -n %{libsoname} -p /sbin/ldconfig44
%postun -n %{libsoname} -p /sbin/ldconfig45
46
0001-Fix-arm-flags.patch
Added
41
1
2
+From: Sebastian Ramacher <sramacher@debian.org>3
+Date: Sun, 21 Jun 2020 17:54:56 +02004
+Subject: Fix arm* flags5
+6
+---7
+ source/CMakeLists.txt | 7 ++-----8
+ 1 file changed, 2 insertions(+), 5 deletions(-)9
+10
+diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt11
+index ab5ddfe..eb9b19b 10075512
+--- a/source/CMakeLists.txt13
++++ b/source/CMakeLists.txt14
+@@ -253,10 +253,7 @@ if(GCC)15
+ elseif(ARM)16
+ find_package(Neon)17
+ if(CPU_HAS_NEON)18
+- set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=neon -marm -fPIC)19
+ add_definitions(-DHAVE_NEON)20
+- else()21
+- set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=vfp -marm)22
+ endif()23
+ endif()24
+ if(ARM64 OR CROSS_COMPILE_ARM64)25
+@@ -265,13 +262,13 @@ if(GCC)26
+ find_package(SVE2)27
+ if(CPU_HAS_SVE2 OR CROSS_COMPILE_SVE2)28
+ message(STATUS "Found SVE2")29
+- set(ARM_ARGS -O3 -march=armv8-a+sve2 -fPIC -flax-vector-conversions)30
++ set(ARM_ARGS -fPIC -flax-vector-conversions)31
+ add_definitions(-DHAVE_SVE2)32
+ add_definitions(-DHAVE_SVE)33
+ add_definitions(-DHAVE_NEON) # for NEON c/c++ primitives, as currently there is no implementation that use SVE234
+ elseif(CPU_HAS_SVE OR CROSS_COMPILE_SVE)35
+ message(STATUS "Found SVE")36
+- set(ARM_ARGS -O3 -march=armv8-a+sve -fPIC -flax-vector-conversions)37
++ set(ARM_ARGS -fPIC -flax-vector-conversions)38
+ add_definitions(-DHAVE_SVE)39
+ add_definitions(-DHAVE_NEON) # for NEON c/c++ primitives, as currently there is no implementation that use SVE40
+ elseif(CPU_HAS_NEON)41
0004-Do-not-build-with-assembly-support-on-arm.patch
Added
30
1
2
+From: Sebastian Ramacher <sramacher@debian.org>3
+Date: Fri, 31 May 2024 23:38:23 +02004
+Subject: Do not build with assembly support on arm*5
+6
+---7
+ source/CMakeLists.txt | 9 ---------8
+ 1 file changed, 9 deletions(-)9
+10
+diff --git a/source/CMakeLists.txt b/source/CMakeLists.txt11
+index 672cc2d..f112330 10075512
+--- a/source/CMakeLists.txt13
++++ b/source/CMakeLists.txt14
+@@ -73,15 +73,6 @@ elseif(POWERMATCH GREATER "-1")15
+ add_definitions(-DPPC64=1)16
+ message(STATUS "Detected POWER PPC64 target processor")17
+ endif()18
+-elseif(ARMMATCH GREATER "-1")19
+- if(CROSS_COMPILE_ARM)20
+- message(STATUS "Cross compiling for ARM arch")21
+- else()22
+- set(CROSS_COMPILE_ARM 0)23
+- endif()24
+- message(STATUS "Detected ARM target processor")25
+- set(ARM 1)26
+- add_definitions(-DX265_ARCH_ARM=1 -DHAVE_ARMV6=1)27
+ elseif(ARM64MATCH GREATER "-1")28
+ #if(CROSS_COMPILE_ARM64)29
+ #message(STATUS "Cross compiling for ARM64 arch")30
arm.patch
Deleted
110
1
2
-Index: x265_3.4/source/CMakeLists.txt3
-===================================================================4
---- x265_3.4.orig/source/CMakeLists.txt5
-+++ x265_3.4/source/CMakeLists.txt6
-@@ -64,26 +64,26 @@ elseif(POWERMATCH GREATER "-1")7
- add_definitions(-DPPC64=1)8
- message(STATUS "Detected POWER PPC64 target processor")9
- endif()10
--elseif(ARMMATCH GREATER "-1")11
-- if(CROSS_COMPILE_ARM)12
-- message(STATUS "Cross compiling for ARM arch")13
-- else()14
-- set(CROSS_COMPILE_ARM 0)15
-- endif()16
-- set(ARM 1)17
-- if("${CMAKE_SIZEOF_VOID_P}" MATCHES 8)18
-- message(STATUS "Detected ARM64 target processor")19
-- set(ARM64 1)20
-- add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=1 -DHAVE_ARMV6=0)21
-- else()22
-- message(STATUS "Detected ARM target processor")23
-- add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=0 -DHAVE_ARMV6=1)24
-- endif()25
-+elseif(${SYSPROC} MATCHES "armv5.*")26
-+ message(STATUS "Detected ARMV5 system processor")27
-+ set(ARMV5 1)28
-+ add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=0 -DHAVE_ARMV6=0 -DHAVE_NEON=0)29
-+elseif(${SYSPROC} STREQUAL "armv6l")30
-+ message(STATUS "Detected ARMV6 system processor")31
-+ set(ARMV6 1)32
-+ add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=0 -DHAVE_ARMV6=1 -DHAVE_NEON=0)33
-+elseif(${SYSPROC} STREQUAL "armv7l")34
-+ message(STATUS "Detected ARMV7 system processor")35
-+ set(ARMV7 1)36
-+ add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=0 -DHAVE_ARMV6=1 -DHAVE_NEON=0)37
-+elseif(${SYSPROC} STREQUAL "aarch64")38
-+ message(STATUS "Detected AArch64 system processor")39
-+ set(ARMV7 1)40
-+ add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=1 -DHAVE_ARMV6=0 -DHAVE_NEON=0)41
- else()42
- message(STATUS "CMAKE_SYSTEM_PROCESSOR value `${CMAKE_SYSTEM_PROCESSOR}` is unknown")43
- message(STATUS "Please add this value near ${CMAKE_CURRENT_LIST_FILE}:${CMAKE_CURRENT_LIST_LINE}")44
- endif()45
--46
- if(UNIX)47
- list(APPEND PLATFORM_LIBS pthread)48
- find_library(LIBRT rt)49
-@@ -238,28 +238,9 @@ if(GCC)50
- endif()51
- endif()52
- endif()53
-- if(ARM AND CROSS_COMPILE_ARM)54
-- if(ARM64)55
-- set(ARM_ARGS -fPIC)56
-- else()57
-- set(ARM_ARGS -march=armv6 -mfloat-abi=soft -mfpu=vfp -marm -fPIC)58
-- endif()59
-- message(STATUS "cross compile arm")60
-- elseif(ARM)61
-- if(ARM64)62
-- set(ARM_ARGS -fPIC)63
-- add_definitions(-DHAVE_NEON)64
-- else()65
-- find_package(Neon)66
-- if(CPU_HAS_NEON)67
-- set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=neon -marm -fPIC)68
-- add_definitions(-DHAVE_NEON)69
-- else()70
-- set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=vfp -marm)71
-- endif()72
-- endif()73
-+ if(ARMV7)74
-+ add_definitions(-fPIC)75
- endif()76
-- add_definitions(${ARM_ARGS})77
- if(FPROFILE_GENERATE)78
- if(INTEL_CXX)79
- add_definitions(-prof-gen -prof-dir="${CMAKE_CURRENT_BINARY_DIR}")80
-Index: x265_3.4/source/common/cpu.cpp81
-===================================================================82
---- x265_3.4.orig/source/common/cpu.cpp83
-+++ x265_3.4/source/common/cpu.cpp84
-@@ -39,7 +39,7 @@85
- #include <machine/cpu.h>86
- #endif87
- 88
--#if X265_ARCH_ARM && !defined(HAVE_NEON)89
-+#if X265_ARCH_ARM && (!defined(HAVE_NEON) || HAVE_NEON==0)90
- #include <signal.h>91
- #include <setjmp.h>92
- static sigjmp_buf jmpbuf;93
-@@ -350,7 +350,6 @@ uint32_t cpu_detect(bool benableavx512)94
- }95
- 96
- canjump = 1;97
-- PFX(cpu_neon_test)();98
- canjump = 0;99
- signal(SIGILL, oldsig);100
- #endif // if !HAVE_NEON101
-@@ -366,7 +365,7 @@ uint32_t cpu_detect(bool benableavx512)102
- // which may result in incorrect detection and the counters stuck enabled.103
- // right now Apple does not seem to support performance counters for this test104
- #ifndef __MACH__105
-- flags |= PFX(cpu_fast_neon_mrc_test)() ? X265_CPU_FAST_NEON_MRC : 0;106
-+ //flags |= PFX(cpu_fast_neon_mrc_test)() ? X265_CPU_FAST_NEON_MRC : 0;107
- #endif108
- // TODO: write dual issue test? currently it's A8 (dual issue) vs. A9 (fast mrc)109
- #elif X265_ARCH_ARM64110
baselibs.conf
Changed
4
1
2
-libx265-1993
+libx265-2094
x265_3.5.tar.gz/source/common/aarch64/ipfilter8.S
Deleted
416
1
2
-/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
- *5
- * Authors: Yimeng Su <yimeng.su@huawei.com>6
- *7
- * This program is free software; you can redistribute it and/or modify8
- * it under the terms of the GNU General Public License as published by9
- * the Free Software Foundation; either version 2 of the License, or10
- * (at your option) any later version.11
- *12
- * This program is distributed in the hope that it will be useful,13
- * but WITHOUT ANY WARRANTY; without even the implied warranty of14
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
- * GNU General Public License for more details.16
- *17
- * You should have received a copy of the GNU General Public License18
- * along with this program; if not, write to the Free Software19
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
- *21
- * This program is also available under a commercial proprietary license.22
- * For more information, contact us at license @ x265.com.23
- *****************************************************************************/24
-25
-#include "asm.S"26
-27
-.section .rodata28
-29
-.align 430
-31
-.text32
-33
-34
-35
-.macro qpel_filter_0_32b36
- movi v24.8h, #6437
- uxtl v19.8h, v5.8b38
- smull v17.4s, v19.4h, v24.4h39
- smull2 v18.4s, v19.8h, v24.8h40
-.endm41
-42
-.macro qpel_filter_1_32b43
- movi v16.8h, #5844
- uxtl v19.8h, v5.8b45
- smull v17.4s, v19.4h, v16.4h46
- smull2 v18.4s, v19.8h, v16.8h47
-48
- movi v24.8h, #1049
- uxtl v21.8h, v1.8b50
- smull v19.4s, v21.4h, v24.4h51
- smull2 v20.4s, v21.8h, v24.8h52
-53
- movi v16.8h, #1754
- uxtl v23.8h, v2.8b55
- smull v21.4s, v23.4h, v16.4h56
- smull2 v22.4s, v23.8h, v16.8h57
-58
- movi v24.8h, #559
- uxtl v1.8h, v6.8b60
- smull v23.4s, v1.4h, v24.4h61
- smull2 v16.4s, v1.8h, v24.8h62
-63
- sub v17.4s, v17.4s, v19.4s64
- sub v18.4s, v18.4s, v20.4s65
-66
- uxtl v1.8h, v4.8b67
- sshll v19.4s, v1.4h, #268
- sshll2 v20.4s, v1.8h, #269
-70
- add v17.4s, v17.4s, v21.4s71
- add v18.4s, v18.4s, v22.4s72
-73
- uxtl v1.8h, v0.8b74
- uxtl v2.8h, v3.8b75
- ssubl v21.4s, v2.4h, v1.4h76
- ssubl2 v22.4s, v2.8h, v1.8h77
-78
- add v17.4s, v17.4s, v19.4s79
- add v18.4s, v18.4s, v20.4s80
- sub v21.4s, v21.4s, v23.4s81
- sub v22.4s, v22.4s, v16.4s82
- add v17.4s, v17.4s, v21.4s83
- add v18.4s, v18.4s, v22.4s84
-.endm85
-86
-.macro qpel_filter_2_32b87
- movi v16.4s, #1188
- uxtl v19.8h, v5.8b89
- uxtl v20.8h, v2.8b90
- saddl v17.4s, v19.4h, v20.4h91
- saddl2 v18.4s, v19.8h, v20.8h92
-93
- uxtl v21.8h, v1.8b94
- uxtl v22.8h, v6.8b95
- saddl v19.4s, v21.4h, v22.4h96
- saddl2 v20.4s, v21.8h, v22.8h97
-98
- mul v19.4s, v19.4s, v16.4s99
- mul v20.4s, v20.4s, v16.4s100
-101
- movi v16.4s, #40102
- mul v17.4s, v17.4s, v16.4s103
- mul v18.4s, v18.4s, v16.4s104
-105
- uxtl v21.8h, v4.8b106
- uxtl v22.8h, v3.8b107
- saddl v23.4s, v21.4h, v22.4h108
- saddl2 v16.4s, v21.8h, v22.8h109
-110
- uxtl v1.8h, v0.8b111
- uxtl v2.8h, v7.8b112
- saddl v21.4s, v1.4h, v2.4h113
- saddl2 v22.4s, v1.8h, v2.8h114
-115
- shl v23.4s, v23.4s, #2116
- shl v16.4s, v16.4s, #2117
-118
- add v19.4s, v19.4s, v21.4s119
- add v20.4s, v20.4s, v22.4s120
- add v17.4s, v17.4s, v23.4s121
- add v18.4s, v18.4s, v16.4s122
- sub v17.4s, v17.4s, v19.4s123
- sub v18.4s, v18.4s, v20.4s124
-.endm125
-126
-.macro qpel_filter_3_32b127
- movi v16.8h, #17128
- movi v24.8h, #5129
-130
- uxtl v19.8h, v5.8b131
- smull v17.4s, v19.4h, v16.4h132
- smull2 v18.4s, v19.8h, v16.8h133
-134
- uxtl v21.8h, v1.8b135
- smull v19.4s, v21.4h, v24.4h136
- smull2 v20.4s, v21.8h, v24.8h137
-138
- movi v16.8h, #58139
- uxtl v23.8h, v2.8b140
- smull v21.4s, v23.4h, v16.4h141
- smull2 v22.4s, v23.8h, v16.8h142
-143
- movi v24.8h, #10144
- uxtl v1.8h, v6.8b145
- smull v23.4s, v1.4h, v24.4h146
- smull2 v16.4s, v1.8h, v24.8h147
-148
- sub v17.4s, v17.4s, v19.4s149
- sub v18.4s, v18.4s, v20.4s150
-151
- uxtl v1.8h, v3.8b152
- sshll v19.4s, v1.4h, #2153
- sshll2 v20.4s, v1.8h, #2154
-155
- add v17.4s, v17.4s, v21.4s156
- add v18.4s, v18.4s, v22.4s157
-158
- uxtl v1.8h, v4.8b159
- uxtl v2.8h, v7.8b160
- ssubl v21.4s, v1.4h, v2.4h161
- ssubl2 v22.4s, v1.8h, v2.8h162
-163
- add v17.4s, v17.4s, v19.4s164
- add v18.4s, v18.4s, v20.4s165
- sub v21.4s, v21.4s, v23.4s166
- sub v22.4s, v22.4s, v16.4s167
- add v17.4s, v17.4s, v21.4s168
- add v18.4s, v18.4s, v22.4s169
-.endm170
-171
-172
-173
-174
-.macro vextin8175
- ld1 {v3.16b}, x11, #16176
- mov v7.d0, v3.d1177
- ext v0.8b, v3.8b, v7.8b, #1178
- ext v4.8b, v3.8b, v7.8b, #2179
- ext v1.8b, v3.8b, v7.8b, #3180
- ext v5.8b, v3.8b, v7.8b, #4181
- ext v2.8b, v3.8b, v7.8b, #5182
- ext v6.8b, v3.8b, v7.8b, #6183
- ext v3.8b, v3.8b, v7.8b, #7184
-.endm185
-186
-187
-188
-// void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)189
-.macro HPS_FILTER a b filterhps190
- mov w12, #8192191
- mov w6, w10192
- sub x3, x3, #\a193
- lsl x3, x3, #1194
- mov w9, #\a195
- cmp w9, #4196
- b.eq 14f197
- cmp w9, #12198
- b.eq 15f199
- b 7f200
-14:201
- HPS_FILTER_4 \a \b \filterhps202
- b 10f203
-15:204
- HPS_FILTER_12 \a \b \filterhps205
- b 10f206
-7:207
- cmp w5, #0208
- b.eq 8f209
- cmp w5, #1210
- b.eq 9f211
-8:212
-loop1_hps_\filterhps\()_\a\()x\b\()_rowext0:213
- mov w7, #\a214
- lsr w7, w7, #3215
- mov x11, x0216
- sub x11, x11, #4217
-loop2_hps_\filterhps\()_\a\()x\b\()_rowext0:218
- vextin8219
- \filterhps220
- dup v16.4s, w12221
- sub v17.4s, v17.4s, v16.4s222
- sub v18.4s, v18.4s, v16.4s223
- xtn v0.4h, v17.4s224
- xtn2 v0.8h, v18.4s225
- st1 {v0.8h}, x2, #16226
- subs w7, w7, #1227
- sub x11, x11, #8228
- b.ne loop2_hps_\filterhps\()_\a\()x\b\()_rowext0229
- subs w6, w6, #1230
- add x0, x0, x1231
- add x2, x2, x3232
- b.ne loop1_hps_\filterhps\()_\a\()x\b\()_rowext0233
- b 10f234
-9:235
-loop3_hps_\filterhps\()_\a\()x\b\()_rowext1:236
- mov w7, #\a237
- lsr w7, w7, #3238
- mov x11, x0239
- sub x11, x11, #4240
-loop4_hps_\filterhps\()_\a\()x\b\()_rowext1:241
- vextin8242
- \filterhps243
- dup v16.4s, w12244
- sub v17.4s, v17.4s, v16.4s245
- sub v18.4s, v18.4s, v16.4s246
- xtn v0.4h, v17.4s247
- xtn2 v0.8h, v18.4s248
- st1 {v0.8h}, x2, #16249
- subs w7, w7, #1250
- sub x11, x11, #8251
- b.ne loop4_hps_\filterhps\()_\a\()x\b\()_rowext1252
- subs w6, w6, #1253
- add x0, x0, x1254
- add x2, x2, x3255
- b.ne loop3_hps_\filterhps\()_\a\()x\b\()_rowext1256
-10:257
-.endm258
-259
-.macro HPS_FILTER_4 w h filterhps260
- cmp w5, #0261
- b.eq 11f262
- cmp w5, #1263
- b.eq 12f264
-11:265
-loop4_hps_\filterhps\()_\w\()x\h\()_rowext0:266
- mov x11, x0267
- sub x11, x11, #4268
- vextin8269
- \filterhps270
- dup v16.4s, w12271
- sub v17.4s, v17.4s, v16.4s272
- xtn v0.4h, v17.4s273
- st1 {v0.4h}, x2, #8274
- sub x11, x11, #8275
- subs w6, w6, #1276
- add x0, x0, x1277
- add x2, x2, x3278
- b.ne loop4_hps_\filterhps\()_\w\()x\h\()_rowext0279
- b 13f280
-12:281
-loop5_hps_\filterhps\()_\w\()x\h\()_rowext1:282
- mov x11, x0283
- sub x11, x11, #4284
- vextin8285
- \filterhps286
- dup v16.4s, w12287
- sub v17.4s, v17.4s, v16.4s288
- xtn v0.4h, v17.4s289
- st1 {v0.4h}, x2, #8290
- sub x11, x11, #8291
- subs w6, w6, #1292
- add x0, x0, x1293
- add x2, x2, x3294
- b.ne loop5_hps_\filterhps\()_\w\()x\h\()_rowext1295
-13:296
-.endm297
-298
-.macro HPS_FILTER_12 w h filterhps299
- cmp w5, #0300
- b.eq 14f301
- cmp w5, #1302
- b.eq 15f303
-14:304
-loop12_hps_\filterhps\()_\w\()x\h\()_rowext0:305
- mov x11, x0306
- sub x11, x11, #4307
- vextin8308
- \filterhps309
- dup v16.4s, w12310
- sub v17.4s, v17.4s, v16.4s311
- sub v18.4s, v18.4s, v16.4s312
- xtn v0.4h, v17.4s313
- xtn2 v0.8h, v18.4s314
- st1 {v0.8h}, x2, #16315
- sub x11, x11, #8316
-317
- vextin8318
- \filterhps319
- dup v16.4s, w12320
- sub v17.4s, v17.4s, v16.4s321
- xtn v0.4h, v17.4s322
- st1 {v0.4h}, x2, #8323
- add x2, x2, x3324
- subs w6, w6, #1325
- add x0, x0, x1326
- b.ne loop12_hps_\filterhps\()_\w\()x\h\()_rowext0327
- b 16f328
-15:329
-loop12_hps_\filterhps\()_\w\()x\h\()_rowext1:330
- mov x11, x0331
- sub x11, x11, #4332
- vextin8333
- \filterhps334
- dup v16.4s, w12335
- sub v17.4s, v17.4s, v16.4s336
- sub v18.4s, v18.4s, v16.4s337
- xtn v0.4h, v17.4s338
- xtn2 v0.8h, v18.4s339
- st1 {v0.8h}, x2, #16340
- sub x11, x11, #8341
-342
- vextin8343
- \filterhps344
- dup v16.4s, w12345
- sub v17.4s, v17.4s, v16.4s346
- xtn v0.4h, v17.4s347
- st1 {v0.4h}, x2, #8348
- add x2, x2, x3349
- subs w6, w6, #1350
- add x0, x0, x1351
- b.ne loop12_hps_\filterhps\()_\w\()x\h\()_rowext1352
-16:353
-.endm354
-355
-// void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)356
-.macro LUMA_HPS w h357
-function x265_interp_8tap_horiz_ps_\w\()x\h\()_neon358
- mov w10, #\h359
- cmp w5, #0360
- b.eq 6f361
- sub x0, x0, x1, lsl #2362
-363
- add x0, x0, x1364
- add w10, w10, #7365
-6:366
- cmp w4, #0367
- b.eq 0f368
- cmp w4, #1369
- b.eq 1f370
- cmp w4, #2371
- b.eq 2f372
- cmp w4, #3373
- b.eq 3f374
-0:375
- HPS_FILTER \w \h qpel_filter_0_32b376
- b 5f377
-1:378
- HPS_FILTER \w \h qpel_filter_1_32b379
- b 5f380
-2:381
- HPS_FILTER \w \h qpel_filter_2_32b382
- b 5f383
-3:384
- HPS_FILTER \w \h qpel_filter_3_32b385
- b 5f386
-5:387
- ret388
-endfunc389
-.endm390
-391
-LUMA_HPS 4 4392
-LUMA_HPS 4 8393
-LUMA_HPS 4 16394
-LUMA_HPS 8 4395
-LUMA_HPS 8 8396
-LUMA_HPS 8 16397
-LUMA_HPS 8 32398
-LUMA_HPS 12 16399
-LUMA_HPS 16 4400
-LUMA_HPS 16 8401
-LUMA_HPS 16 12402
-LUMA_HPS 16 16403
-LUMA_HPS 16 32404
-LUMA_HPS 16 64405
-LUMA_HPS 24 32406
-LUMA_HPS 32 8407
-LUMA_HPS 32 16408
-LUMA_HPS 32 24409
-LUMA_HPS 32 32410
-LUMA_HPS 32 64411
-LUMA_HPS 48 64412
-LUMA_HPS 64 16413
-LUMA_HPS 64 32414
-LUMA_HPS 64 48415
-LUMA_HPS 64 64416
x265_3.5.tar.gz/source/common/aarch64/ipfilter8.h
Deleted
57
1
2
-/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
- *5
- * Authors: Yimeng Su <yimeng.su@huawei.com>6
- *7
- * This program is free software; you can redistribute it and/or modify8
- * it under the terms of the GNU General Public License as published by9
- * the Free Software Foundation; either version 2 of the License, or10
- * (at your option) any later version.11
- *12
- * This program is distributed in the hope that it will be useful,13
- * but WITHOUT ANY WARRANTY; without even the implied warranty of14
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
- * GNU General Public License for more details.16
- *17
- * You should have received a copy of the GNU General Public License18
- * along with this program; if not, write to the Free Software19
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
- *21
- * This program is also available under a commercial proprietary license.22
- * For more information, contact us at license @ x265.com.23
- *****************************************************************************/24
-25
-#ifndef X265_IPFILTER8_AARCH64_H26
-#define X265_IPFILTER8_AARCH64_H27
-28
-29
-void x265_interp_8tap_horiz_ps_4x4_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);30
-void x265_interp_8tap_horiz_ps_4x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);31
-void x265_interp_8tap_horiz_ps_4x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);32
-void x265_interp_8tap_horiz_ps_8x4_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);33
-void x265_interp_8tap_horiz_ps_8x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);34
-void x265_interp_8tap_horiz_ps_8x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);35
-void x265_interp_8tap_horiz_ps_8x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);36
-void x265_interp_8tap_horiz_ps_12x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);37
-void x265_interp_8tap_horiz_ps_16x4_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);38
-void x265_interp_8tap_horiz_ps_16x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);39
-void x265_interp_8tap_horiz_ps_16x12_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);40
-void x265_interp_8tap_horiz_ps_16x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);41
-void x265_interp_8tap_horiz_ps_16x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);42
-void x265_interp_8tap_horiz_ps_16x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);43
-void x265_interp_8tap_horiz_ps_24x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);44
-void x265_interp_8tap_horiz_ps_32x8_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);45
-void x265_interp_8tap_horiz_ps_32x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);46
-void x265_interp_8tap_horiz_ps_32x24_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);47
-void x265_interp_8tap_horiz_ps_32x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);48
-void x265_interp_8tap_horiz_ps_32x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);49
-void x265_interp_8tap_horiz_ps_48x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);50
-void x265_interp_8tap_horiz_ps_64x16_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);51
-void x265_interp_8tap_horiz_ps_64x32_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);52
-void x265_interp_8tap_horiz_ps_64x48_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);53
-void x265_interp_8tap_horiz_ps_64x64_neon(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt);54
-55
-56
-#endif // ifndef X265_IPFILTER8_AARCH64_H57
x265_3.5.tar.gz/source/common/aarch64/pixel-util.h
Deleted
42
1
2
-/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
- *5
- * Authors: Yimeng Su <yimeng.su@huawei.com>6
- * Hongbin Liu <liuhongbin1@huawei.com>7
- *8
- * This program is free software; you can redistribute it and/or modify9
- * it under the terms of the GNU General Public License as published by10
- * the Free Software Foundation; either version 2 of the License, or11
- * (at your option) any later version.12
- *13
- * This program is distributed in the hope that it will be useful,14
- * but WITHOUT ANY WARRANTY; without even the implied warranty of15
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the16
- * GNU General Public License for more details.17
- *18
- * You should have received a copy of the GNU General Public License19
- * along with this program; if not, write to the Free Software20
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.21
- *22
- * This program is also available under a commercial proprietary license.23
- * For more information, contact us at license @ x265.com.24
- *****************************************************************************/25
-26
-#ifndef X265_PIXEL_UTIL_AARCH64_H27
-#define X265_PIXEL_UTIL_AARCH64_H28
-29
-int x265_pixel_satd_4x4_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);30
-int x265_pixel_satd_4x8_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);31
-int x265_pixel_satd_4x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);32
-int x265_pixel_satd_4x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);33
-int x265_pixel_satd_8x4_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);34
-int x265_pixel_satd_8x8_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);35
-int x265_pixel_satd_12x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);36
-int x265_pixel_satd_12x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);37
-38
-uint32_t x265_quant_neon(const int16_t* coef, const int32_t* quantCoeff, int32_t* deltaU, int16_t* qCoef, int qBits, int add, int numCoeff);39
-int PFX(psyCost_4x4_neon)(const pixel* source, intptr_t sstride, const pixel* recon, intptr_t rstride);40
-41
-#endif // ifndef X265_PIXEL_UTIL_AARCH64_H42
x265_3.5.tar.gz/source/common/aarch64/pixel.h
Deleted
107
1
2
-/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
- *5
- * Authors: Hongbin Liu <liuhongbin1@huawei.com>6
- *7
- * This program is free software; you can redistribute it and/or modify8
- * it under the terms of the GNU General Public License as published by9
- * the Free Software Foundation; either version 2 of the License, or10
- * (at your option) any later version.11
- *12
- * This program is distributed in the hope that it will be useful,13
- * but WITHOUT ANY WARRANTY; without even the implied warranty of14
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
- * GNU General Public License for more details.16
- *17
- * You should have received a copy of the GNU General Public License18
- * along with this program; if not, write to the Free Software19
- * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
- *21
- * This program is also available under a commercial proprietary license.22
- * For more information, contact us at license @ x265.com.23
- *****************************************************************************/24
-25
-#ifndef X265_I386_PIXEL_AARCH64_H26
-#define X265_I386_PIXEL_AARCH64_H27
-28
-void x265_pixel_avg_pp_4x4_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);29
-void x265_pixel_avg_pp_4x8_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);30
-void x265_pixel_avg_pp_4x16_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);31
-void x265_pixel_avg_pp_8x4_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);32
-void x265_pixel_avg_pp_8x8_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);33
-void x265_pixel_avg_pp_8x16_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);34
-void x265_pixel_avg_pp_8x32_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);35
-void x265_pixel_avg_pp_12x16_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);36
-void x265_pixel_avg_pp_16x4_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);37
-void x265_pixel_avg_pp_16x8_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);38
-void x265_pixel_avg_pp_16x12_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);39
-void x265_pixel_avg_pp_16x16_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);40
-void x265_pixel_avg_pp_16x32_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);41
-void x265_pixel_avg_pp_16x64_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);42
-void x265_pixel_avg_pp_24x32_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);43
-void x265_pixel_avg_pp_32x8_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);44
-void x265_pixel_avg_pp_32x16_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);45
-void x265_pixel_avg_pp_32x24_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);46
-void x265_pixel_avg_pp_32x32_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);47
-void x265_pixel_avg_pp_32x64_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);48
-void x265_pixel_avg_pp_48x64_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);49
-void x265_pixel_avg_pp_64x16_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);50
-void x265_pixel_avg_pp_64x32_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);51
-void x265_pixel_avg_pp_64x48_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);52
-void x265_pixel_avg_pp_64x64_neon (pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int);53
-54
-void x265_sad_x3_4x4_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);55
-void x265_sad_x3_4x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);56
-void x265_sad_x3_4x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);57
-void x265_sad_x3_8x4_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);58
-void x265_sad_x3_8x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);59
-void x265_sad_x3_8x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);60
-void x265_sad_x3_8x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);61
-void x265_sad_x3_12x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);62
-void x265_sad_x3_16x4_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);63
-void x265_sad_x3_16x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);64
-void x265_sad_x3_16x12_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);65
-void x265_sad_x3_16x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);66
-void x265_sad_x3_16x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);67
-void x265_sad_x3_16x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);68
-void x265_sad_x3_24x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);69
-void x265_sad_x3_32x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);70
-void x265_sad_x3_32x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);71
-void x265_sad_x3_32x24_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);72
-void x265_sad_x3_32x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);73
-void x265_sad_x3_32x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);74
-void x265_sad_x3_48x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);75
-void x265_sad_x3_64x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);76
-void x265_sad_x3_64x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);77
-void x265_sad_x3_64x48_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);78
-void x265_sad_x3_64x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, intptr_t frefstride, int32_t* res);79
-80
-void x265_sad_x4_4x4_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);81
-void x265_sad_x4_4x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);82
-void x265_sad_x4_4x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);83
-void x265_sad_x4_8x4_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);84
-void x265_sad_x4_8x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);85
-void x265_sad_x4_8x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);86
-void x265_sad_x4_8x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);87
-void x265_sad_x4_12x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);88
-void x265_sad_x4_16x4_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);89
-void x265_sad_x4_16x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);90
-void x265_sad_x4_16x12_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);91
-void x265_sad_x4_16x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);92
-void x265_sad_x4_16x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);93
-void x265_sad_x4_16x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);94
-void x265_sad_x4_24x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);95
-void x265_sad_x4_32x8_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);96
-void x265_sad_x4_32x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);97
-void x265_sad_x4_32x24_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);98
-void x265_sad_x4_32x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);99
-void x265_sad_x4_32x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);100
-void x265_sad_x4_48x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);101
-void x265_sad_x4_64x16_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);102
-void x265_sad_x4_64x32_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);103
-void x265_sad_x4_64x48_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);104
-void x265_sad_x4_64x64_neon(const pixel* fenc, const pixel* fref0, const pixel* fref1, const pixel* fref2, const pixel* fref3, intptr_t frefstride, int32_t* res);105
-106
-#endif // ifndef X265_I386_PIXEL_AARCH64_H107
x265_3.6.tar.gz/.gitignore
Added
38
1
2
+# Prerequisites3
+*.d4
+5
+# Compiled Object files6
+*.slo7
+*.lo8
+*.o9
+*.obj10
+11
+# Precompiled Headers12
+*.gch13
+*.pch14
+15
+# Compiled Dynamic libraries16
+*.so17
+*.dylib18
+*.dll19
+20
+# Fortran module files21
+*.mod22
+*.smod23
+24
+# Compiled Static libraries25
+*.lai26
+*.la27
+*.a28
+*.lib29
+30
+# Executables31
+*.exe32
+*.out33
+*.app34
+35
+# Build directory36
+build/37
+38
x265_3.5.tar.gz/build/README.txt -> x265_3.6.tar.gz/build/README.txt
Changed
37
1
2
3
Note: MSVC12 requires cmake 2.8.11 or later4
5
+Note: When the SVE/SVE2 instruction set of Arm AArch64 architecture is to be used, the GCC10.x and onwards must6
+ be installed in order to compile x265.7
+8
9
= Optional Prerequisites =10
11
12
building out of a Mercurial source repository. If you are building out of13
a release source package, the version will not change. If Mercurial is not14
found, the version will be "unknown".15
+16
+= Build Instructions for cross-compilation for Arm AArch64 Targets=17
+18
+When the target platform is based on Arm AArch64 architecture, the x265 can be19
+built in x86 platforms. However, the CMAKE_C_COMPILER and CMAKE_CXX_COMPILER20
+enviroment variables should be set to point to the cross compilers of the21
+appropriate gcc. For example:22
+23
+1. export CMAKE_C_COMPILER=aarch64-unknown-linux-gnu-gcc24
+2. export CMAKE_CXX_COMPILER=aarch64-unknown-linux-gnu-g++25
+26
+The default ones are aarch64-linux-gnu-gcc and aarch64-linux-gnu-g++.27
+Then, the normal building process can be followed.28
+29
+Moreover, if the target platform supports SVE or SVE2 instruction set, the30
+CROSS_COMPILE_SVE or CROSS_COMPILE_SVE2 environment variables should be set31
+to true, respectively. For example:32
+33
+1. export CROSS_COMPILE_SVE2=true34
+2. export CROSS_COMPILE_SVE=true35
+36
+Then, the normal building process can be followed.37
x265_3.6.tar.gz/build/aarch64-darwin
Added
2
1
+(directory)2
x265_3.6.tar.gz/build/aarch64-darwin/crosscompile.cmake
Added
25
1
2
+# CMake toolchain file for cross compiling x265 for aarch643
+# This feature is only supported as experimental. Use with caution.4
+# Please report bugs on bitbucket5
+# Run cmake with: cmake -DCMAKE_TOOLCHAIN_FILE=crosscompile.cmake -G "Unix Makefiles" ../../source && ccmake ../../source6
+7
+set(CROSS_COMPILE_ARM64 1)8
+set(CMAKE_SYSTEM_NAME Darwin)9
+set(CMAKE_SYSTEM_PROCESSOR aarch64)10
+11
+# specify the cross compiler12
+set(CMAKE_C_COMPILER gcc-12)13
+set(CMAKE_CXX_COMPILER g++-12)14
+15
+# specify the target environment16
+SET(CMAKE_FIND_ROOT_PATH /opt/homebrew/bin/)17
+18
+# specify whether SVE/SVE2 is supported by the target platform19
+if(DEFINED ENV{CROSS_COMPILE_SVE2})20
+ set(CROSS_COMPILE_SVE2 1)21
+elseif(DEFINED ENV{CROSS_COMPILE_SVE})22
+ set(CROSS_COMPILE_SVE 1)23
+endif()24
+25
x265_3.6.tar.gz/build/aarch64-darwin/make-Makefiles.bash
Added
6
1
2
+#!/bin/bash3
+# Run this from within a bash shell4
+5
+cmake -DCMAKE_TOOLCHAIN_FILE="crosscompile.cmake" -G "Unix Makefiles" ../../source && ccmake ../../source6
x265_3.5.tar.gz/build/aarch64-linux/crosscompile.cmake -> x265_3.6.tar.gz/build/aarch64-linux/crosscompile.cmake
Changed
34
1
2
# Please report bugs on bitbucket3
# Run cmake with: cmake -DCMAKE_TOOLCHAIN_FILE=crosscompile.cmake -G "Unix Makefiles" ../../source && ccmake ../../source4
5
-set(CROSS_COMPILE_ARM 1)6
+set(CROSS_COMPILE_ARM64 1)7
set(CMAKE_SYSTEM_NAME Linux)8
set(CMAKE_SYSTEM_PROCESSOR aarch64)9
10
# specify the cross compiler11
-set(CMAKE_C_COMPILER aarch64-linux-gnu-gcc)12
-set(CMAKE_CXX_COMPILER aarch64-linux-gnu-g++)13
+if(DEFINED ENV{CMAKE_C_COMPILER})14
+ set(CMAKE_C_COMPILER $ENV{CMAKE_C_COMPILER})15
+else()16
+ set(CMAKE_C_COMPILER aarch64-linux-gnu-gcc)17
+endif()18
+if(DEFINED ENV{CMAKE_CXX_COMPILER})19
+ set(CMAKE_CXX_COMPILER $ENV{CMAKE_CXX_COMPILER})20
+else()21
+ set(CMAKE_CXX_COMPILER aarch64-linux-gnu-g++)22
+endif()23
24
# specify the target environment25
SET(CMAKE_FIND_ROOT_PATH /usr/aarch64-linux-gnu)26
+27
+# specify whether SVE/SVE2 is supported by the target platform28
+if(DEFINED ENV{CROSS_COMPILE_SVE2})29
+ set(CROSS_COMPILE_SVE2 1)30
+elseif(DEFINED ENV{CROSS_COMPILE_SVE})31
+ set(CROSS_COMPILE_SVE 1)32
+endif()33
+34
x265_3.5.tar.gz/build/arm-linux/make-Makefiles.bash -> x265_3.6.tar.gz/build/arm-linux/make-Makefiles.bash
Changed
7
1
2
#!/bin/bash3
# Run this from within a bash shell4
5
-cmake -G "Unix Makefiles" ../../source && ccmake ../../source6
+cmake -DCMAKE_TOOLCHAIN_FILE="crosscompile.cmake" -G "Unix Makefiles" ../../source && ccmake ../../source7
x265_3.5.tar.gz/doc/reST/cli.rst -> x265_3.6.tar.gz/doc/reST/cli.rst
Changed
405
1
2
auto-detection by the encoder. If specified, the encoder will3
attempt to bring the encode specifications within that specified4
level. If the encoder is unable to reach the level it issues a5
- warning and aborts the encode. If the requested requirement level is6
- higher than the actual level, the actual requirement level is7
- signaled.8
+ warning and aborts the encode. The requested level will be signaled 9
+ in the bitstream even if it is higher than the actual level.10
11
Beware, specifying a decoder level will force the encoder to enable12
VBV for constant rate factor encodes, which may introduce13
14
(main, main10, etc). Second, an encoder is created from this15
x265_param instance and the :option:`--level-idc` and16
:option:`--high-tier` parameters are used to reduce bitrate or other17
- features in order to enforce the target level. Finally, the encoder18
- re-examines the final set of parameters and detects the actual19
- minimum decoder requirement level and this is what is signaled in20
- the bitstream headers. The detected decoder level will only use High21
- tier if the user specified a High tier level.22
+ features in order to enforce the target level. The detected decoder level23
+ will only use High tier if the user specified a High tier level.24
25
The signaled profile will be determined by the encoder's internal26
bitdepth and input color space. If :option:`--keyint` is 0 or 1,27
28
Note that :option:`--analysis-save-reuse-level` and :option:`--analysis-load-reuse-level` must be paired29
with :option:`--analysis-save` and :option:`--analysis-load` respectively.30
31
- +--------------+------------------------------------------+32
- | Level | Description |33
- +==============+==========================================+34
- | 1 | Lookahead information |35
- +--------------+------------------------------------------+36
- | 2 to 4 | Level 1 + intra/inter modes, ref's |37
- +--------------+------------------------------------------+38
- | 5 and 6 | Level 2 + rect-amp |39
- +--------------+------------------------------------------+40
- | 7 | Level 5 + AVC size CU refinement |41
- +--------------+------------------------------------------+42
- | 8 and 9 | Level 5 + AVC size Full CU analysis-info |43
- +--------------+------------------------------------------+44
- | 10 | Level 5 + Full CU analysis-info |45
- +--------------+------------------------------------------+46
+ +--------------+---------------------------------------------------+47
+ | Level | Description |48
+ +==============+===================================================+49
+ | 1 | Lookahead information |50
+ +--------------+---------------------------------------------------+51
+ | 2 to 4 | Level 1 + intra/inter modes, depth, ref's, cutree |52
+ +--------------+---------------------------------------------------+53
+ | 5 and 6 | Level 2 + rect-amp |54
+ +--------------+---------------------------------------------------+55
+ | 7 | Level 5 + AVC size CU refinement |56
+ +--------------+---------------------------------------------------+57
+ | 8 and 9 | Level 5 + AVC size Full CU analysis-info |58
+ +--------------+---------------------------------------------------+59
+ | 10 | Level 5 + Full CU analysis-info |60
+ +--------------+---------------------------------------------------+61
62
.. option:: --refine-mv-type <string>63
64
65
Search range for HME level 0, 1 and 2.66
The Search Range for each HME level must be between 0 and 32768(excluding).67
Default search range is 16,32,48 for level 0,1,2 respectively.68
+ 69
+.. option:: --mcstf, --no-mcstf70
+71
+ Enable Motion Compensated Temporal filtering.72
+ Default: disabled73
74
Spatial/intra options75
=====================76
77
78
.. option:: --hist-scenecut, --no-hist-scenecut79
80
- Indicates that scenecuts need to be detected using luma edge and chroma histograms.81
- :option:`--hist-scenecut` enables scenecut detection using the histograms and disables the default scene cut algorithm.82
- :option:`--no-hist-scenecut` disables histogram based scenecut algorithm.83
- 84
-.. option:: --hist-threshold <0.0..1.0>85
-86
- This value represents the threshold for normalized SAD of edge histograms used in scenecut detection.87
- This requires :option:`--hist-scenecut` to be enabled. For example, a value of 0.2 indicates that a frame with normalized SAD value 88
- greater than 0.2 against the previous frame as scenecut. 89
- Increasing the threshold reduces the number of scenecuts detected.90
- Default 0.03.91
+ Scenecuts detected based on histogram, intensity and variance of the picture.92
+ :option:`--hist-scenecut` enables or :option:`--no-hist-scenecut` disables scenecut detection based on93
+ histogram.94
95
.. option:: --radl <integer>96
97
98
Default 1.0.99
**Range of values:** 0.0 to 3.0100
101
+.. option:: --sbrc --no-sbrc102
+103
+ To enable and disable segment based rate control.Segment duration depends on the104
+ keyframe interval specified.If unspecified,default keyframe interval will be used.105
+ Default: disabled.106
+107
.. option:: --hevc-aq108
109
Enable adaptive quantization110
111
112
**CLI ONLY**113
114
+.. option:: --scenecut-qp-config <filename>115
+116
+ Specify a text file which contains the scenecut aware QP options.117
+ The options include :option:`--scenecut-aware-qp` and :option:`--masking-strength`118
+119
+ **CLI ONLY**120
+121
.. option:: --scenecut-aware-qp <integer>122
123
It reduces the bits spent on the inter-frames within the scenecut window124
before and after a scenecut by increasing their QP in ratecontrol pass2 algorithm125
- without any deterioration in visual quality. If a scenecut falls within the window,126
- the QP of the inter-frames after this scenecut will not be modified.127
+ without any deterioration in visual quality.128
:option:`--scenecut-aware-qp` works only with --pass 2. Default 0.129
130
+-------+---------------------------------------------------------------+131
132
for the QP increment for inter-frames when :option:`--scenecut-aware-qp`133
is enabled.134
135
- When :option:`--scenecut-aware-qp` is::136
+ When :option:`--scenecut-aware-qp` is:137
+138
* 1 (Forward masking):139
- --masking-strength <fwdWindow,fwdRefQPDelta,fwdNonRefQPDelta>140
+ --masking-strength <fwdMaxWindow,fwdRefQPDelta,fwdNonRefQPDelta>141
+ or 142
+ --masking-strength <fwdWindow1,fwdRefQPDelta1,fwdNonRefQPDelta1,fwdWindow2,fwdRefQPDelta2,fwdNonRefQPDelta2,143
+ fwdWindow3,fwdRefQPDelta3,fwdNonRefQPDelta3,fwdWindow4,fwdRefQPDelta4,fwdNonRefQPDelta4,144
+ fwdWindow5,fwdRefQPDelta5,fwdNonRefQPDelta5,fwdWindow6,fwdRefQPDelta6,fwdNonRefQPDelta6>145
* 2 (Backward masking):146
- --masking-strength <bwdWindow,bwdRefQPDelta,bwdNonRefQPDelta>147
+ --masking-strength <bwdMaxWindow,bwdRefQPDelta,bwdNonRefQPDelta>148
+ or 149
+ --masking-strength <bwdWindow1,bwdRefQPDelta1,bwdNonRefQPDelta1,bwdWindow2,bwdRefQPDelta2,bwdNonRefQPDelta2,150
+ bwdWindow3,bwdRefQPDelta3,bwdNonRefQPDelta3,bwdWindow4,bwdRefQPDelta4,bwdNonRefQPDelta4,151
+ bwdWindow5,bwdRefQPDelta5,bwdNonRefQPDelta5,bwdWindow6,bwdRefQPDelta6,bwdNonRefQPDelta6>152
* 3 (Bi-directional masking):153
- --masking-strength <fwdWindow,fwdRefQPDelta,fwdNonRefQPDelta,bwdWindow,bwdRefQPDelta,bwdNonRefQPDelta>154
+ --masking-strength <fwdMaxWindow,fwdRefQPDelta,fwdNonRefQPDelta,bwdMaxWindow,bwdRefQPDelta,bwdNonRefQPDelta>155
+ or 156
+ --masking-strength <fwdWindow1,fwdRefQPDelta1,fwdNonRefQPDelta1,fwdWindow2,fwdRefQPDelta2,fwdNonRefQPDelta2,157
+ fwdWindow3,fwdRefQPDelta3,fwdNonRefQPDelta3,fwdWindow4,fwdRefQPDelta4,fwdNonRefQPDelta4,158
+ fwdWindow5,fwdRefQPDelta5,fwdNonRefQPDelta5,fwdWindow6,fwdRefQPDelta6,fwdNonRefQPDelta6,159
+ bwdWindow1,bwdRefQPDelta1,bwdNonRefQPDelta1,bwdWindow2,bwdRefQPDelta2,bwdNonRefQPDelta2,160
+ bwdWindow3,bwdRefQPDelta3,bwdNonRefQPDelta3,bwdWindow4,bwdRefQPDelta4,bwdNonRefQPDelta4,161
+ bwdWindow5,bwdRefQPDelta5,bwdNonRefQPDelta5,bwdWindow6,bwdRefQPDelta6,bwdNonRefQPDelta6>162
163
+-----------------+---------------------------------------------------------------+164
| Parameter | Description |165
+=================+===============================================================+166
- | fwdWindow | The duration(in milliseconds) for which there is a reduction |167
- | | in the bits spent on the inter-frames after a scenecut by |168
- | | increasing their QP. Default 500ms. |169
- | | **Range of values:** 0 to 1000 |170
+ | fwdMaxWindow | The maximum duration(in milliseconds) for which there is a |171
+ | | reduction in the bits spent on the inter-frames after a |172
+ | | scenecut by increasing their QP. Default 500ms. |173
+ | | **Range of values:** 0 to 2000 |174
+ +-----------------+---------------------------------------------------------------+175
+ | fwdWindow | The duration of a sub-window(in milliseconds) for which there |176
+ | | is a reduction in the bits spent on the inter-frames after a |177
+ | | scenecut by increasing their QP. Default 500ms. |178
+ | | **Range of values:** 0 to 2000 |179
+-----------------+---------------------------------------------------------------+180
| fwdRefQPDelta | The offset by which QP is incremented for inter-frames |181
| | after a scenecut. Default 5. |182
- | | **Range of values:** 0 to 10 |183
+ | | **Range of values:** 0 to 20 |184
+-----------------+---------------------------------------------------------------+185
| fwdNonRefQPDelta| The offset by which QP is incremented for non-referenced |186
| | inter-frames after a scenecut. The offset is computed from |187
| | fwdRefQPDelta when it is not explicitly specified. |188
- | | **Range of values:** 0 to 10 |189
+ | | **Range of values:** 0 to 20 |190
+ +-----------------+---------------------------------------------------------------+191
+ | bwdMaxWindow | The maximum duration(in milliseconds) for which there is a |192
+ | | reduction in the bits spent on the inter-frames before a |193
+ | | scenecut by increasing their QP. Default 100ms. |194
+ | | **Range of values:** 0 to 2000 |195
+-----------------+---------------------------------------------------------------+196
- | bwdWindow | The duration(in milliseconds) for which there is a reduction |197
- | | in the bits spent on the inter-frames before a scenecut by |198
- | | increasing their QP. Default 100ms. |199
- | | **Range of values:** 0 to 1000 |200
+ | bwdWindow | The duration of a sub-window(in milliseconds) for which there |201
+ | | is a reduction in the bits spent on the inter-frames before a |202
+ | | scenecut by increasing their QP. Default 100ms. |203
+ | | **Range of values:** 0 to 2000 |204
+-----------------+---------------------------------------------------------------+205
| bwdRefQPDelta | The offset by which QP is incremented for inter-frames |206
| | before a scenecut. The offset is computed from |207
| | fwdRefQPDelta when it is not explicitly specified. |208
- | | **Range of values:** 0 to 10 |209
+ | | **Range of values:** 0 to 20 |210
+-----------------+---------------------------------------------------------------+211
| bwdNonRefQPDelta| The offset by which QP is incremented for non-referenced |212
| | inter-frames before a scenecut. The offset is computed from |213
| | bwdRefQPDelta when it is not explicitly specified. |214
- | | **Range of values:** 0 to 10 |215
+ | | **Range of values:** 0 to 20 |216
+-----------------+---------------------------------------------------------------+217
218
- **CLI ONLY**219
+ We can specify the value for the Use :option:`--masking-strength` parameter in different formats.220
+ 1. If we don't specify --masking-strength and specify only --scenecut-aware-qp, then default offset and window size values are considered.221
+ 2. If we specify --masking-strength with the format 1 mentioned above, the values of window, refQpDelta and nonRefQpDelta given by the user are taken for window 1 and the offsets for the remaining windows are derived with 15% difference between windows.222
+ 3. If we specify the --masking-strength with the format 2 mentioned above, the values of window, refQpDelta and nonRefQpDelta given by the user for each window from 1 to 6 are directly used.NOTE: We can use this format to specify zero offsets for any particular window223
+224
+ Sample config file:: (Format 2 Forward masking explained here)225
+226
+ --scenecut-aware-qp 1 --masking-strength 1000,8,12227
+ 228
+ The above sample config file is available in `the downloads page <https://bitbucket.org/multicoreware/x265_git/downloads/scenecut_qp_config.txt>`_229
230
.. option:: --vbv-live-multi-pass, --no-vbv-live-multi-pass231
232
233
rate control mode.234
235
Default disabled. **Experimental feature**236
+ 237
+238
+.. option:: bEncFocusedFramesOnly239
+240
+ Used to trigger encoding of selective GOPs; Disabled by default.241
+ 242
+ **API ONLY**243
+ 244
245
Quantization Options246
====================247
248
Values in the range 0..12. See D.3.3 of the HEVC spec. for a detailed explanation.249
Required for HLG (Hybrid Log Gamma) signaling. Not signaled by default.250
251
+.. option:: --video-signal-type-preset <string>252
+253
+ Specify combinations of color primaries, transfer characteristics, color matrix,254
+ range of luma and chroma signals, and chroma sample location.255
+ String format: <system-id>:<color-volume>256
+ 257
+ This has higher precedence than individual VUI parameters. If any individual VUI option258
+ is specified together with this, which changes the values set corresponding to the system-id259
+ or color-volume, it will be discarded.260
+261
+ system-id options and their corresponding values:262
+ +----------------+---------------------------------------------------------------+263
+ | system-id | Value |264
+ +================+===============================================================+265
+ | BT601_525 | --colorprim smpte170m --transfer smpte170m |266
+ | | --colormatrix smpte170m --range limited --chromaloc 0 |267
+ +----------------+---------------------------------------------------------------+268
+ | BT601_626 | --colorprim bt470bg --transfer smpte170m --colormatrix bt470bg|269
+ | | --range limited --chromaloc 0 |270
+ +----------------+---------------------------------------------------------------+271
+ | BT709_YCC | --colorprim bt709 --transfer bt709 --colormatrix bt709 |272
+ | | --range limited --chromaloc 0 |273
+ +----------------+---------------------------------------------------------------+274
+ | BT709_RGB | --colorprim bt709 --transfer bt709 --colormatrix gbr |275
+ | | --range limited |276
+ +----------------+---------------------------------------------------------------+277
+ | BT2020_YCC_NCL | --colorprim bt2020 --transfer bt2020-10 --colormatrix bt709 |278
+ | | --range limited --chromaloc 2 |279
+ +----------------+---------------------------------------------------------------+280
+ | BT2020_RGB | --colorprim bt2020 --transfer smpte2084 --colormatrix bt2020nc|281
+ | | --range limited |282
+ +----------------+---------------------------------------------------------------+283
+ | BT2100_PQ_YCC | --colorprim bt2020 --transfer smpte2084 --colormatrix bt2020nc|284
+ | | --range limited --chromaloc 2 |285
+ +----------------+---------------------------------------------------------------+286
+ | BT2100_PQ_ICTCP| --colorprim bt2020 --transfer smpte2084 --colormatrix ictcp |287
+ | | --range limited --chromaloc 2 |288
+ +----------------+---------------------------------------------------------------+289
+ | BT2100_PQ_RGB | --colorprim bt2020 --transfer smpte2084 --colormatrix gbr |290
+ | | --range limited |291
+ +----------------+---------------------------------------------------------------+292
+ | BT2100_HLG_YCC | --colorprim bt2020 --transfer arib-std-b67 |293
+ | | --colormatrix bt2020nc --range limited --chromaloc 2 |294
+ +----------------+---------------------------------------------------------------+295
+ | BT2100_HLG_RGB | --colorprim bt2020 --transfer arib-std-b67 --colormatrix gbr |296
+ | | --range limited |297
+ +----------------+---------------------------------------------------------------+298
+ | FR709_RGB | --colorprim bt709 --transfer bt709 --colormatrix gbr |299
+ | | --range full |300
+ +----------------+---------------------------------------------------------------+301
+ | FR2020_RGB | --colorprim bt2020 --transfer bt2020-10 --colormatrix gbr |302
+ | | --range full |303
+ +----------------+---------------------------------------------------------------+304
+ | FRP3D65_YCC | --colorprim smpte432 --transfer bt709 --colormatrix smpte170m |305
+ | | --range full --chromaloc 1 |306
+ +----------------+---------------------------------------------------------------+307
+308
+ color-volume options and their corresponding values:309
+ +----------------+---------------------------------------------------------------+310
+ | color-volume | Value |311
+ +================+===============================================================+312
+ | P3D65x1000n0005| --master-display G(13250,34500)B(7500,3000)R(34000,16000) |313
+ | | WP(15635,16450)L(10000000,5) |314
+ +----------------+---------------------------------------------------------------+315
+ | P3D65x4000n005 | --master-display G(13250,34500)B(7500,3000)R(34000,16000) |316
+ | | WP(15635,16450)L(40000000,50) |317
+ +----------------+---------------------------------------------------------------+318
+ | BT2100x108n0005| --master-display G(8500,39850)B(6550,2300)R(34000,146000) |319
+ | | WP(15635,16450)L(10000000,1) |320
+ +----------------+---------------------------------------------------------------+321
+322
+ Note: The color-volume options can be used only with the system-id options BT2100_PQ_YCC,323
+ BT2100_PQ_ICTCP, and BT2100_PQ_RGB. It is incompatible with other options.324
+325
+326
Bitstream options327
=================328
329
330
the very first AUD will be skipped since it cannot be placed at the331
start of the access unit, where it belongs. Default disabled332
333
+.. option:: --eob, --no-eob334
+335
+ Emit an end of bitstream NAL unit at the end of the bitstream.336
+ Default disabled337
+338
+.. option:: --eos, --no-eos339
+340
+ Emit an end of sequence NAL unit at the end of every coded341
+ video sequence. Default disabled342
+343
.. option:: --hrd, --no-hrd344
345
Enable the signaling of HRD parameters to the decoder. The HRD346
347
The value is specified as a float or as an integer with the profile times 10,348
for example profile 5 is specified as "5" or "5.0" or "50".349
350
- Currently only profile 5, profile 8.1 and profile 8.2 enabled, Default 0 (disabled)351
+ Currently only profile 5, profile 8.1, profile 8.2 and profile 8.4 enabled, Default 0 (disabled)352
353
.. option:: --dolby-vision-rpu <filename>354
355
356
2. CRC357
3. Checksum358
359
-.. option:: --temporal-layers,--no-temporal-layers360
+.. option:: --temporal-layers <integer>361
362
- Enable a temporal sub layer. All referenced I/P/B frames are in the363
- base layer and all unreferenced B frames are placed in a temporal364
- enhancement layer. A decoder may choose to drop the enhancement layer 365
- and only decode and display the base layer slices.366
- 367
- If used with a fixed GOP (:option:`--b-adapt` 0) and :option:`--bframes`368
- 3 then the two layers evenly split the frame rate, with a cadence of369
- PbBbP. You probably also want :option:`--no-scenecut` and a keyframe370
- interval that is a multiple of 4.371
+ Enable specified number of temporal sub layers. For any frame in layer N,372
+ all referenced frames are in the layer N or N-1.A decoder may choose to drop the enhancement layer373
+ and only decode and display the base layer slices.Allowed number of temporal sub-layers374
+ are 2 to 5.(2 and 5 inclusive)375
+376
+ When enabled,temporal layers 3 through 5 configures a fixed miniGOP with the number of bframes as shown below377
+ unless miniGOP size is modified due to lookahead decisions.Temporal layer 2 is a special case that has378
+ all reference frames in base layer and non-reference frames in enhancement layer without any constraint on the379
+ number of bframes.Default disabled.380
+ +----------------+--------+381
+ | temporal layer | bframes|382
+ +================+========+383
+ | 3 | 3 |384
+ +----------------+--------+385
+ | 4 | 7 |386
+ +----------------+--------+387
+ | 5 | 15 |388
+ +----------------+--------+389
390
.. option:: --log2-max-poc-lsb <integer>391
392
393
Emit SEI messages in a single NAL unit instead of multiple NALs. Default disabled.394
When HRD SEI is enabled the HM decoder will throw a warning.395
396
+.. option:: --film-grain <filename>397
+398
+ Refers to the film grain model characteristics for signal enhancement information transmission399
+400
+ **CLI_ONLY**401
+402
DCT Approximations403
=================404
405
x265_3.5.tar.gz/doc/reST/introduction.rst -> x265_3.6.tar.gz/doc/reST/introduction.rst
Changed
9
1
2
to start is with the `Motion Picture Experts Group - Licensing Authority3
- HEVC Licensing Program <http://www.mpegla.com/main/PID/HEVC/default.aspx>`_.4
5
-x265 is a registered trademark of MulticoreWare, Inc. The x265 logo is6
+x265 is a registered trademark of MulticoreWare, Inc. The X265 logo is7
a trademark of MulticoreWare, and may only be used with explicit written8
permission. All rights reserved.9
x265_3.5.tar.gz/doc/reST/releasenotes.rst -> x265_3.6.tar.gz/doc/reST/releasenotes.rst
Changed
55
1
2
Release Notes3
*************4
5
+Version 3.66
+===========7
+8
+Release date - 4th April, 2024.9
+10
+New feature11
+-----------12
+1. Segment based Ratecontrol (SBRC) feature13
+2. Motion-Compensated Spatio-Temporal Filtering14
+3. Scene-cut aware qp - BBAQ (Bidirectional Boundary Aware Quantization)15
+4. Histogram-Based Scene Change Detection16
+5. Film-Grain characteristics as a SEI message to support Film Grain Synthesis(FGS)17
+6. Add temporal layer implementation(Hierarchical B-frame implementation)18
+ 19
+Enhancements to existing features20
+---------------------------------21
+1. Added Dolby Vision 8.4 Profile Support22
+23
+24
+API changes25
+-----------26
+1. Add Segment based Ratecontrol(SBRC) feature: "--no-sbrc".27
+2. Add command line parameter for mcstf feature: "--no-mctf".28
+3. Add command line parameters for the scene cut aware qp feature: "--scenecut-aware-qp" and "--masking-strength".29
+4. Add command line parameters for Histogram-Based Scene Change Detection: "--hist-scenecut".30
+5. Add film grain characteristics as a SEI message to the bitstream: "--film-grain <filename>"31
+6. cli: add new option --cra-nal (Force nal type to CRA to all frames expect for the first frame, works only with keyint 1)32
+33
+Optimizations34
+---------------------35
+ARM64 NEON optimizations:- Several time-consuming C functions have been optimized for the targeted platform - aarch64. The overall performance increased by around 20%.36
+SVE/SVE2 optimizations37
+38
+39
+Bug fixes40
+---------41
+1. Linux bug to utilize all the cores42
+2. Crash with hist-scenecut build when source resolution is not multiple of minCuSize43
+3. 32bit and 64bit builds generation for ARM44
+4. bugs in zonefile feature (Reflect Zonefile Parameters inside Lookahead, extra IDR issue, Avg I Slice QP value issue etc..)45
+5. Add x86 ASM implementation for subsampling luma 46
+6. Fix for abrladder segfault with load reuse level 1 47
+7. Reorder miniGOP based on temporal layer hierarchy and add support for more B frame 48
+8. Add MacOS aarch64 build support 49
+9. Fix boundary condition issue for Gaussian filter50
+51
+52
Version 3.553
===========54
55
x265_3.5.tar.gz/readme.rst -> x265_3.6.tar.gz/readme.rst
Changed
10
1
2
x265 HEVC Encoder3
=================4
5
-| **Read:** | Online `documentation <http://x265.readthedocs.org/en/default/>`_ | Developer `wiki <http://bitbucket.org/multicoreware/x265/wiki/>`_6
+| **Read:** | Online `documentation <http://x265.readthedocs.org/en/master/>`_ | Developer `wiki <http://bitbucket.org/multicoreware/x265_git/wiki/>`_7
| **Download:** | `releases <http://ftp.videolan.org/pub/videolan/x265/>`_ 8
| **Interact:** | #x265 on freenode.irc.net | `x265-devel@videolan.org <http://mailman.videolan.org/listinfo/x265-devel>`_ | `Report an issue <https://bitbucket.org/multicoreware/x265/issues?status=new&status=open>`_9
10
x265_3.5.tar.gz/source/CMakeLists.txt -> x265_3.6.tar.gz/source/CMakeLists.txt
Changed
232
1
2
option(STATIC_LINK_CRT "Statically link C runtime for release builds" OFF)3
mark_as_advanced(FPROFILE_USE FPROFILE_GENERATE NATIVE_BUILD)4
# X265_BUILD must be incremented each time the public API is changed5
-set(X265_BUILD 199)6
+set(X265_BUILD 209)7
configure_file("${PROJECT_SOURCE_DIR}/x265.def.in"8
"${PROJECT_BINARY_DIR}/x265.def")9
configure_file("${PROJECT_SOURCE_DIR}/x265_config.h.in"10
11
SET(CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake" "${CMAKE_MODULE_PATH}")12
13
# System architecture detection14
-string(TOLOWER "${CMAKE_SYSTEM_PROCESSOR}" SYSPROC)15
+if (APPLE AND CMAKE_OSX_ARCHITECTURES)16
+ string(TOLOWER "${CMAKE_OSX_ARCHITECTURES}" SYSPROC)17
+else()18
+ string(TOLOWER "${CMAKE_SYSTEM_PROCESSOR}" SYSPROC)19
+endif()20
set(X86_ALIASES x86 i386 i686 x86_64 amd64)21
-set(ARM_ALIASES armv6l armv7l aarch64)22
+set(ARM_ALIASES armv6l armv7l)23
+set(ARM64_ALIASES arm64 arm64e aarch64)24
list(FIND X86_ALIASES "${SYSPROC}" X86MATCH)25
list(FIND ARM_ALIASES "${SYSPROC}" ARMMATCH)26
-set(POWER_ALIASES ppc64 ppc64le)27
+list(FIND ARM64_ALIASES "${SYSPROC}" ARM64MATCH)28
+set(POWER_ALIASES powerpc64 powerpc64le ppc64 ppc64le)29
list(FIND POWER_ALIASES "${SYSPROC}" POWERMATCH)30
-if("${SYSPROC}" STREQUAL "" OR X86MATCH GREATER "-1")31
+if(X86MATCH GREATER "-1")32
set(X86 1)33
add_definitions(-DX265_ARCH_X86=1)34
if(CMAKE_CXX_FLAGS STREQUAL "-m32")35
36
else()37
set(CROSS_COMPILE_ARM 0)38
endif()39
+ message(STATUS "Detected ARM target processor")40
set(ARM 1)41
- if("${CMAKE_SIZEOF_VOID_P}" MATCHES 8)42
- message(STATUS "Detected ARM64 target processor")43
- set(ARM64 1)44
- add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=1 -DHAVE_ARMV6=0)45
- else()46
- message(STATUS "Detected ARM target processor")47
- add_definitions(-DX265_ARCH_ARM=1 -DX265_ARCH_ARM64=0 -DHAVE_ARMV6=1)48
- endif()49
+ add_definitions(-DX265_ARCH_ARM=1 -DHAVE_ARMV6=1)50
+elseif(ARM64MATCH GREATER "-1")51
+ #if(CROSS_COMPILE_ARM64)52
+ #message(STATUS "Cross compiling for ARM64 arch")53
+ #else()54
+ #set(CROSS_COMPILE_ARM64 0)55
+ #endif()56
+ message(STATUS "Detected ARM64 target processor")57
+ set(ARM64 1)58
+ add_definitions(-DX265_ARCH_ARM64=1 -DHAVE_NEON)59
else()60
message(STATUS "CMAKE_SYSTEM_PROCESSOR value `${CMAKE_SYSTEM_PROCESSOR}` is unknown")61
message(STATUS "Please add this value near ${CMAKE_CURRENT_LIST_FILE}:${CMAKE_CURRENT_LIST_LINE}")62
63
endif()64
endif()65
if(ARM AND CROSS_COMPILE_ARM)66
- if(ARM64)67
- set(ARM_ARGS -fPIC)68
- else()69
- set(ARM_ARGS -march=armv6 -mfloat-abi=soft -mfpu=vfp -marm -fPIC)70
- endif()71
message(STATUS "cross compile arm")72
+ set(ARM_ARGS -march=armv6 -mfloat-abi=soft -mfpu=vfp -marm -fPIC)73
elseif(ARM)74
- if(ARM64)75
- set(ARM_ARGS -fPIC)76
+ find_package(Neon)77
+ if(CPU_HAS_NEON)78
+ set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=neon -marm -fPIC)79
add_definitions(-DHAVE_NEON)80
else()81
- find_package(Neon)82
- if(CPU_HAS_NEON)83
- set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=neon -marm -fPIC)84
- add_definitions(-DHAVE_NEON)85
- else()86
- set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=vfp -marm)87
- endif()88
+ set(ARM_ARGS -mcpu=native -mfloat-abi=hard -mfpu=vfp -marm)89
endif()90
endif()91
+ if(ARM64 OR CROSS_COMPILE_ARM64)92
+ find_package(Neon)93
+ find_package(SVE)94
+ find_package(SVE2)95
+ if(CPU_HAS_SVE2 OR CROSS_COMPILE_SVE2)96
+ message(STATUS "Found SVE2")97
+ set(ARM_ARGS -O3 -march=armv8-a+sve2 -fPIC -flax-vector-conversions)98
+ add_definitions(-DHAVE_SVE2)99
+ add_definitions(-DHAVE_SVE)100
+ add_definitions(-DHAVE_NEON) # for NEON c/c++ primitives, as currently there is no implementation that use SVE2101
+ elseif(CPU_HAS_SVE OR CROSS_COMPILE_SVE)102
+ message(STATUS "Found SVE")103
+ set(ARM_ARGS -O3 -march=armv8-a+sve -fPIC -flax-vector-conversions)104
+ add_definitions(-DHAVE_SVE)105
+ add_definitions(-DHAVE_NEON) # for NEON c/c++ primitives, as currently there is no implementation that use SVE106
+ elseif(CPU_HAS_NEON)107
+ message(STATUS "Found NEON")108
+ set(ARM_ARGS -fPIC -flax-vector-conversions)109
+ add_definitions(-DHAVE_NEON)110
+ else()111
+ set(ARM_ARGS -fPIC -flax-vector-conversions)112
+ endif() 113
+ endif()114
+ if(ENABLE_PIC)115
+ list(APPEND ARM_ARGS -DPIC)116
+ endif()117
add_definitions(${ARM_ARGS})118
if(FPROFILE_GENERATE)119
if(INTEL_CXX)120
121
endif(GCC)122
123
find_package(Nasm)124
-if(ARM OR CROSS_COMPILE_ARM)125
+if(ARM OR CROSS_COMPILE_ARM OR ARM64 OR CROSS_COMPILE_ARM64)126
option(ENABLE_ASSEMBLY "Enable use of assembly coded primitives" ON)127
elseif(NASM_FOUND AND X86)128
if (NASM_VERSION_STRING VERSION_LESS "2.13.0")129
130
endif(EXTRA_LIB)131
mark_as_advanced(EXTRA_LIB EXTRA_LINK_FLAGS)132
133
-if(X64)134
+if(X64 OR ARM64 OR PPC64)135
# NOTE: We only officially support high-bit-depth compiles of x265136
# on 64bit architectures. Main10 plus large resolution plus slow137
# preset plus 32bit address space usually means malloc failure. You138
139
# license" so to speak. If it breaks you get to keep both halves.140
# You will need to disable assembly manually.141
option(HIGH_BIT_DEPTH "Store pixel samples as 16bit values (Main10/Main12)" OFF)142
-endif(X64)143
+endif(X64 OR ARM64 OR PPC64)144
if(HIGH_BIT_DEPTH)145
option(MAIN12 "Support Main12 instead of Main10" OFF)146
if(MAIN12)147
148
endif()149
add_definitions(-DX265_NS=${X265_NS})150
151
+if(ARM64)152
+ if(HIGH_BIT_DEPTH)153
+ if(MAIN12)154
+ list(APPEND ASM_FLAGS -DHIGH_BIT_DEPTH=1 -DBIT_DEPTH=12 -DX265_NS=${X265_NS})155
+ else()156
+ list(APPEND ASM_FLAGS -DHIGH_BIT_DEPTH=1 -DBIT_DEPTH=10 -DX265_NS=${X265_NS})157
+ endif()158
+ else()159
+ list(APPEND ASM_FLAGS -DHIGH_BIT_DEPTH=0 -DBIT_DEPTH=8 -DX265_NS=${X265_NS})160
+ endif()161
+endif(ARM64)162
+163
option(WARNINGS_AS_ERRORS "Stop compiles on first warning" OFF)164
if(WARNINGS_AS_ERRORS)165
if(GCC)166
167
# compile ARM arch asm files here168
enable_language(ASM)169
foreach(ASM ${ARM_ASMS})170
- if(ARM64)171
- set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/aarch64/${ASM})172
- else()173
- set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/arm/${ASM})174
- endif()175
+ set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/arm/${ASM})176
list(APPEND ASM_SRCS ${ASM_SRC})177
list(APPEND ASM_OBJS ${ASM}.${SUFFIX})178
add_custom_command(179
180
ARGS ${ARM_ARGS} -c ${ASM_SRC} -o ${ASM}.${SUFFIX}181
DEPENDS ${ASM_SRC})182
endforeach()183
+ elseif(ARM64 OR CROSS_COMPILE_ARM64)184
+ # compile ARM64 arch asm files here185
+ enable_language(ASM)186
+ foreach(ASM ${ARM_ASMS})187
+ set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/aarch64/${ASM})188
+ list(APPEND ASM_SRCS ${ASM_SRC})189
+ list(APPEND ASM_OBJS ${ASM}.${SUFFIX})190
+ add_custom_command(191
+ OUTPUT ${ASM}.${SUFFIX}192
+ COMMAND ${CMAKE_CXX_COMPILER}193
+ ARGS ${ARM_ARGS} ${ASM_FLAGS} -c ${ASM_SRC} -o ${ASM}.${SUFFIX}194
+ DEPENDS ${ASM_SRC})195
+ endforeach()196
+ if(CPU_HAS_SVE2 OR CROSS_COMPILE_SVE2)197
+ foreach(ASM ${ARM_ASMS_SVE})198
+ set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/aarch64/${ASM})199
+ list(APPEND ASM_SRCS ${ASM_SRC})200
+ list(APPEND ASM_OBJS ${ASM}.${SUFFIX})201
+ add_custom_command(202
+ OUTPUT ${ASM}.${SUFFIX}203
+ COMMAND ${CMAKE_CXX_COMPILER}204
+ ARGS ${ARM_ARGS} ${ASM_FLAGS} -c ${ASM_SRC} -o ${ASM}.${SUFFIX}205
+ DEPENDS ${ASM_SRC})206
+ endforeach()207
+ foreach(ASM ${ARM_ASMS_SVE2})208
+ set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/aarch64/${ASM})209
+ list(APPEND ASM_SRCS ${ASM_SRC})210
+ list(APPEND ASM_OBJS ${ASM}.${SUFFIX})211
+ add_custom_command(212
+ OUTPUT ${ASM}.${SUFFIX}213
+ COMMAND ${CMAKE_CXX_COMPILER}214
+ ARGS ${ARM_ARGS} ${ASM_FLAGS} -c ${ASM_SRC} -o ${ASM}.${SUFFIX}215
+ DEPENDS ${ASM_SRC})216
+ endforeach()217
+ elseif(CPU_HAS_SVE OR CROSS_COMPILE_SVE)218
+ foreach(ASM ${ARM_ASMS_SVE})219
+ set(ASM_SRC ${CMAKE_CURRENT_SOURCE_DIR}/common/aarch64/${ASM})220
+ list(APPEND ASM_SRCS ${ASM_SRC})221
+ list(APPEND ASM_OBJS ${ASM}.${SUFFIX})222
+ add_custom_command(223
+ OUTPUT ${ASM}.${SUFFIX}224
+ COMMAND ${CMAKE_CXX_COMPILER}225
+ ARGS ${ARM_ARGS} ${ASM_FLAGS} -c ${ASM_SRC} -o ${ASM}.${SUFFIX}226
+ DEPENDS ${ASM_SRC})227
+ endforeach()228
+ endif()229
elseif(X86)230
# compile X86 arch asm files here231
foreach(ASM ${MSVC_ASMS})232
x265_3.5.tar.gz/source/abrEncApp.cpp -> x265_3.6.tar.gz/source/abrEncApp.cpp
Changed
2220
1
2
-/*****************************************************************************3
-* Copyright (C) 2013-2020 MulticoreWare, Inc4
-*5
-* Authors: Pooja Venkatesan <pooja@multicorewareinc.com>6
-* Aruna Matheswaran <aruna@multicorewareinc.com>7
-*8
-* This program is free software; you can redistribute it and/or modify9
-* it under the terms of the GNU General Public License as published by10
-* the Free Software Foundation; either version 2 of the License, or11
-* (at your option) any later version.12
-*13
-* This program is distributed in the hope that it will be useful,14
-* but WITHOUT ANY WARRANTY; without even the implied warranty of15
-* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the16
-* GNU General Public License for more details.17
-*18
-* You should have received a copy of the GNU General Public License19
-* along with this program; if not, write to the Free Software20
-* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.21
-*22
-* This program is also available under a commercial proprietary license.23
-* For more information, contact us at license @ x265.com.24
-*****************************************************************************/25
-26
-#include "abrEncApp.h"27
-#include "mv.h"28
-#include "slice.h"29
-#include "param.h"30
-31
-#include <signal.h>32
-#include <errno.h>33
-34
-#include <queue>35
-36
-using namespace X265_NS;37
-38
-/* Ctrl-C handler */39
-static volatile sig_atomic_t b_ctrl_c /* = 0 */;40
-static void sigint_handler(int)41
-{42
- b_ctrl_c = 1;43
-}44
-45
-namespace X265_NS {46
- // private namespace47
-#define X265_INPUT_QUEUE_SIZE 25048
-49
- AbrEncoder::AbrEncoder(CLIOptions cliopt, uint8_t numEncodes, int &ret)50
- {51
- m_numEncodes = numEncodes;52
- m_numActiveEncodes.set(numEncodes);53
- m_queueSize = (numEncodes > 1) ? X265_INPUT_QUEUE_SIZE : 1;54
- m_passEnc = X265_MALLOC(PassEncoder*, m_numEncodes);55
-56
- for (uint8_t i = 0; i < m_numEncodes; i++)57
- {58
- m_passEnci = new PassEncoder(i, cliopti, this);59
- if (!m_passEnci)60
- {61
- x265_log(NULL, X265_LOG_ERROR, "Unable to allocate memory for passEncoder\n");62
- ret = 4;63
- }64
- m_passEnci->init(ret);65
- }66
-67
- if (!allocBuffers())68
- {69
- x265_log(NULL, X265_LOG_ERROR, "Unable to allocate memory for buffers\n");70
- ret = 4;71
- }72
-73
- /* start passEncoder worker threads */74
- for (uint8_t pass = 0; pass < m_numEncodes; pass++)75
- m_passEncpass->startThreads();76
- }77
-78
- bool AbrEncoder::allocBuffers()79
- {80
- m_inputPicBuffer = X265_MALLOC(x265_picture**, m_numEncodes);81
- m_analysisBuffer = X265_MALLOC(x265_analysis_data*, m_numEncodes);82
-83
- m_picWriteCnt = new ThreadSafeIntegerm_numEncodes;84
- m_picReadCnt = new ThreadSafeIntegerm_numEncodes;85
- m_analysisWriteCnt = new ThreadSafeIntegerm_numEncodes;86
- m_analysisReadCnt = new ThreadSafeIntegerm_numEncodes;87
-88
- m_picIdxReadCnt = X265_MALLOC(ThreadSafeInteger*, m_numEncodes);89
- m_analysisWrite = X265_MALLOC(ThreadSafeInteger*, m_numEncodes);90
- m_analysisRead = X265_MALLOC(ThreadSafeInteger*, m_numEncodes);91
- m_readFlag = X265_MALLOC(int*, m_numEncodes);92
-93
- for (uint8_t pass = 0; pass < m_numEncodes; pass++)94
- {95
- m_inputPicBufferpass = X265_MALLOC(x265_picture*, m_queueSize);96
- for (uint32_t idx = 0; idx < m_queueSize; idx++)97
- {98
- m_inputPicBufferpassidx = x265_picture_alloc();99
- x265_picture_init(m_passEncpass->m_param, m_inputPicBufferpassidx);100
- }101
-102
- CHECKED_MALLOC_ZERO(m_analysisBufferpass, x265_analysis_data, m_queueSize);103
- m_picIdxReadCntpass = new ThreadSafeIntegerm_queueSize;104
- m_analysisWritepass = new ThreadSafeIntegerm_queueSize;105
- m_analysisReadpass = new ThreadSafeIntegerm_queueSize;106
- m_readFlagpass = X265_MALLOC(int, m_queueSize);107
- }108
- return true;109
- fail:110
- return false;111
- }112
-113
- void AbrEncoder::destroy()114
- {115
- x265_cleanup(); /* Free library singletons */116
- for (uint8_t pass = 0; pass < m_numEncodes; pass++)117
- {118
- for (uint32_t index = 0; index < m_queueSize; index++)119
- {120
- X265_FREE(m_inputPicBufferpassindex->planes0);121
- x265_picture_free(m_inputPicBufferpassindex);122
- }123
-124
- X265_FREE(m_inputPicBufferpass);125
- X265_FREE(m_analysisBufferpass);126
- X265_FREE(m_readFlagpass);127
- delete m_picIdxReadCntpass;128
- delete m_analysisWritepass;129
- delete m_analysisReadpass;130
- m_passEncpass->destroy();131
- delete m_passEncpass;132
- }133
- X265_FREE(m_inputPicBuffer);134
- X265_FREE(m_analysisBuffer);135
- X265_FREE(m_readFlag);136
-137
- delete m_picWriteCnt;138
- delete m_picReadCnt;139
- delete m_analysisWriteCnt;140
- delete m_analysisReadCnt;141
-142
- X265_FREE(m_picIdxReadCnt);143
- X265_FREE(m_analysisWrite);144
- X265_FREE(m_analysisRead);145
-146
- X265_FREE(m_passEnc);147
- }148
-149
- PassEncoder::PassEncoder(uint32_t id, CLIOptions cliopt, AbrEncoder *parent)150
- {151
- m_id = id;152
- m_cliopt = cliopt;153
- m_parent = parent;154
- if(!(m_cliopt.enableScaler && m_id))155
- m_input = m_cliopt.input;156
- m_param = cliopt.param;157
- m_inputOver = false;158
- m_lastIdx = -1;159
- m_encoder = NULL;160
- m_scaler = NULL;161
- m_reader = NULL;162
- m_ret = 0;163
- }164
-165
- int PassEncoder::init(int &result)166
- {167
- if (m_parent->m_numEncodes > 1)168
- setReuseLevel();169
- 170
- if (!(m_cliopt.enableScaler && m_id))171
- m_reader = new Reader(m_id, this);172
- else173
- {174
- VideoDesc *src = NULL, *dst = NULL;175
- dst = new VideoDesc(m_param->sourceWidth, m_param->sourceHeight, m_param->internalCsp, m_param->internalBitDepth);176
- int dstW = m_parent->m_passEncm_id - 1->m_param->sourceWidth;177
- int dstH = m_parent->m_passEncm_id - 1->m_param->sourceHeight;178
- src = new VideoDesc(dstW, dstH, m_param->internalCsp, m_param->internalBitDepth);179
- if (src != NULL && dst != NULL)180
- {181
- m_scaler = new Scaler(0, 1, m_id, src, dst, this);182
- if (!m_scaler)183
- {184
- x265_log(m_param, X265_LOG_ERROR, "\n MALLOC failure in Scaler");185
- result = 4;186
- }187
- }188
- }189
-190
- /* note: we could try to acquire a different libx265 API here based on191
- * the profile found during option parsing, but it must be done before192
- * opening an encoder */193
-194
- if (m_param)195
- m_encoder = m_cliopt.api->encoder_open(m_param);196
- if (!m_encoder)197
- {198
- x265_log(NULL, X265_LOG_ERROR, "x265_encoder_open() failed for Enc, \n");199
- m_ret = 2;200
- return -1;201
- }202
-203
- /* get the encoder parameters post-initialization */204
- m_cliopt.api->encoder_parameters(m_encoder, m_param);205
-206
- return 1;207
- }208
-209
- void PassEncoder::setReuseLevel()210
- {211
- uint32_t r, padh = 0, padw = 0;212
-213
- m_param->confWinBottomOffset = m_param->confWinRightOffset = 0;214
-215
- m_param->analysisLoadReuseLevel = m_cliopt.loadLevel;216
- m_param->analysisSaveReuseLevel = m_cliopt.saveLevel;217
- m_param->analysisSave = m_cliopt.saveLevel ? "save.dat" : NULL;218
- m_param->analysisLoad = m_cliopt.loadLevel ? "load.dat" : NULL;219
- m_param->bUseAnalysisFile = 0;220
-221
- if (m_cliopt.loadLevel)222
- {223
- x265_param *refParam = m_parent->m_passEncm_cliopt.refId->m_param;224
-225
- if (m_param->sourceHeight == (refParam->sourceHeight - refParam->confWinBottomOffset) &&226
- m_param->sourceWidth == (refParam->sourceWidth - refParam->confWinRightOffset))227
- {228
- m_parent->m_passEncm_id->m_param->confWinBottomOffset = refParam->confWinBottomOffset;229
- m_parent->m_passEncm_id->m_param->confWinRightOffset = refParam->confWinRightOffset;230
- }231
- else232
- {233
- int srcH = refParam->sourceHeight - refParam->confWinBottomOffset;234
- int srcW = refParam->sourceWidth - refParam->confWinRightOffset;235
-236
- double scaleFactorH = double(m_param->sourceHeight / srcH);237
- double scaleFactorW = double(m_param->sourceWidth / srcW);238
-239
- int absScaleFactorH = (int)(10 * scaleFactorH + 0.5);240
- int absScaleFactorW = (int)(10 * scaleFactorW + 0.5);241
-242
- if (absScaleFactorH == 20 && absScaleFactorW == 20)243
- {244
- m_param->scaleFactor = 2;245
-246
- m_parent->m_passEncm_id->m_param->confWinBottomOffset = refParam->confWinBottomOffset * 2;247
- m_parent->m_passEncm_id->m_param->confWinRightOffset = refParam->confWinRightOffset * 2;248
-249
- }250
- }251
- }252
-253
- int h = m_param->sourceHeight + m_param->confWinBottomOffset;254
- int w = m_param->sourceWidth + m_param->confWinRightOffset;255
- if (h & (m_param->minCUSize - 1))256
- {257
- r = h & (m_param->minCUSize - 1);258
- padh = m_param->minCUSize - r;259
- m_param->confWinBottomOffset += padh;260
-261
- }262
-263
- if (w & (m_param->minCUSize - 1))264
- {265
- r = w & (m_param->minCUSize - 1);266
- padw = m_param->minCUSize - r;267
- m_param->confWinRightOffset += padw;268
- }269
- }270
-271
- void PassEncoder::startThreads()272
- {273
- /* Start slave worker threads */274
- m_threadActive = true;275
- start();276
- /* Start reader threads*/277
- if (m_reader != NULL)278
- {279
- m_reader->m_threadActive = true;280
- m_reader->start();281
- }282
- /* Start scaling worker threads */283
- if (m_scaler != NULL)284
- {285
- m_scaler->m_threadActive = true;286
- m_scaler->start();287
- }288
- }289
-290
- void PassEncoder::copyInfo(x265_analysis_data * src)291
- {292
-293
- uint32_t written = m_parent->m_analysisWriteCntm_id.get();294
-295
- int index = written % m_parent->m_queueSize;296
- //If all streams have read analysis data, reuse that position in Queue297
-298
- int read = m_parent->m_analysisReadm_idindex.get();299
- int write = m_parent->m_analysisWritem_idindex.get();300
-301
- int overwrite = written / m_parent->m_queueSize;302
- bool emptyIdxFound = 0;303
- while (!emptyIdxFound && overwrite)304
- {305
- for (uint32_t i = 0; i < m_parent->m_queueSize; i++)306
- {307
- read = m_parent->m_analysisReadm_idi.get();308
- write = m_parent->m_analysisWritem_idi.get();309
- write *= m_cliopt.numRefs;310
-311
- if (read == write)312
- {313
- index = i;314
- emptyIdxFound = 1;315
- }316
- }317
- }318
-319
- x265_analysis_data *m_analysisInfo = &m_parent->m_analysisBufferm_idindex;320
-321
- x265_free_analysis_data(m_param, m_analysisInfo);322
- memcpy(m_analysisInfo, src, sizeof(x265_analysis_data));323
- x265_alloc_analysis_data(m_param, m_analysisInfo);324
-325
- bool isVbv = m_param->rc.vbvBufferSize && m_param->rc.vbvMaxBitrate;326
- if (m_param->bDisableLookahead && isVbv)327
- {328
- memcpy(m_analysisInfo->lookahead.intraSatdForVbv, src->lookahead.intraSatdForVbv, src->numCuInHeight * sizeof(uint32_t));329
- memcpy(m_analysisInfo->lookahead.satdForVbv, src->lookahead.satdForVbv, src->numCuInHeight * sizeof(uint32_t));330
- memcpy(m_analysisInfo->lookahead.intraVbvCost, src->lookahead.intraVbvCost, src->numCUsInFrame * sizeof(uint32_t));331
- memcpy(m_analysisInfo->lookahead.vbvCost, src->lookahead.vbvCost, src->numCUsInFrame * sizeof(uint32_t));332
- }333
-334
- if (src->sliceType == X265_TYPE_IDR || src->sliceType == X265_TYPE_I)335
- {336
- if (m_param->analysisSaveReuseLevel < 2)337
- goto ret;338
- x265_analysis_intra_data *intraDst, *intraSrc;339
- intraDst = (x265_analysis_intra_data*)m_analysisInfo->intraData;340
- intraSrc = (x265_analysis_intra_data*)src->intraData;341
- memcpy(intraDst->depth, intraSrc->depth, sizeof(uint8_t) * src->depthBytes);342
- memcpy(intraDst->modes, intraSrc->modes, sizeof(uint8_t) * src->numCUsInFrame * src->numPartitions);343
- memcpy(intraDst->partSizes, intraSrc->partSizes, sizeof(char) * src->depthBytes);344
- memcpy(intraDst->chromaModes, intraSrc->chromaModes, sizeof(uint8_t) * src->depthBytes);345
- if (m_param->rc.cuTree)346
- memcpy(intraDst->cuQPOff, intraSrc->cuQPOff, sizeof(int8_t) * src->depthBytes);347
- }348
- else349
- {350
- bool bIntraInInter = (src->sliceType == X265_TYPE_P || m_param->bIntraInBFrames);351
- int numDir = src->sliceType == X265_TYPE_P ? 1 : 2;352
- memcpy(m_analysisInfo->wt, src->wt, sizeof(WeightParam) * 3 * numDir);353
- if (m_param->analysisSaveReuseLevel < 2)354
- goto ret;355
- x265_analysis_inter_data *interDst, *interSrc;356
- interDst = (x265_analysis_inter_data*)m_analysisInfo->interData;357
- interSrc = (x265_analysis_inter_data*)src->interData;358
- memcpy(interDst->depth, interSrc->depth, sizeof(uint8_t) * src->depthBytes);359
- memcpy(interDst->modes, interSrc->modes, sizeof(uint8_t) * src->depthBytes);360
- if (m_param->rc.cuTree)361
- memcpy(interDst->cuQPOff, interSrc->cuQPOff, sizeof(int8_t) * src->depthBytes);362
- if (m_param->analysisSaveReuseLevel > 4)363
- {364
- memcpy(interDst->partSize, interSrc->partSize, sizeof(uint8_t) * src->depthBytes);365
- memcpy(interDst->mergeFlag, interSrc->mergeFlag, sizeof(uint8_t) * src->depthBytes);366
- if (m_param->analysisSaveReuseLevel == 10)367
- {368
- memcpy(interDst->interDir, interSrc->interDir, sizeof(uint8_t) * src->depthBytes);369
- for (int dir = 0; dir < numDir; dir++)370
- {371
- memcpy(interDst->mvpIdxdir, interSrc->mvpIdxdir, sizeof(uint8_t) * src->depthBytes);372
- memcpy(interDst->refIdxdir, interSrc->refIdxdir, sizeof(int8_t) * src->depthBytes);373
- memcpy(interDst->mvdir, interSrc->mvdir, sizeof(MV) * src->depthBytes);374
- }375
- if (bIntraInInter)376
- {377
- x265_analysis_intra_data *intraDst = (x265_analysis_intra_data*)m_analysisInfo->intraData;378
- x265_analysis_intra_data *intraSrc = (x265_analysis_intra_data*)src->intraData;379
- memcpy(intraDst->modes, intraSrc->modes, sizeof(uint8_t) * src->numPartitions * src->numCUsInFrame);380
- memcpy(intraDst->chromaModes, intraSrc->chromaModes, sizeof(uint8_t) * src->depthBytes);381
- }382
- }383
- }384
- if (m_param->analysisSaveReuseLevel != 10)385
- memcpy(interDst->ref, interSrc->ref, sizeof(int32_t) * src->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir);386
- }387
-388
-ret:389
- //increment analysis Write counter 390
- m_parent->m_analysisWriteCntm_id.incr();391
- m_parent->m_analysisWritem_idindex.incr();392
- return;393
- }394
-395
-396
- bool PassEncoder::readPicture(x265_picture *dstPic)397
- {398
- /*Check and wait if there any input frames to read*/399
- int ipread = m_parent->m_picReadCntm_id.get();400
- int ipwrite = m_parent->m_picWriteCntm_id.get();401
-402
- bool isAbrLoad = m_cliopt.loadLevel && (m_parent->m_numEncodes > 1);403
- while (!m_inputOver && (ipread == ipwrite))404
- {405
- ipwrite = m_parent->m_picWriteCntm_id.waitForChange(ipwrite);406
- }407
-408
- if (m_threadActive && ipread < ipwrite)409
- {410
- /*Get input index to read from inputQueue. If doesn't need analysis info, it need not wait to fetch poc from analysisQueue*/411
- int readPos = ipread % m_parent->m_queueSize;412
- x265_analysis_data* analysisData = 0;413
-414
- if (isAbrLoad)415
- {416
- /*If stream is master of each slave pass, then fetch analysis data from prev pass*/417
- int analysisQId = m_cliopt.refId;418
- /*Check and wait if there any analysis Data to read*/419
- int analysisWrite = m_parent->m_analysisWriteCntanalysisQId.get();420
- int written = analysisWrite * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;421
- int analysisRead = m_parent->m_analysisReadCntanalysisQId.get();422
- 423
- while (m_threadActive && written == analysisRead)424
- {425
- analysisWrite = m_parent->m_analysisWriteCntanalysisQId.waitForChange(analysisWrite);426
- written = analysisWrite * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;427
- }428
-429
- if (analysisRead < written)430
- {431
- int analysisIdx = 0;432
- if (!m_param->bDisableLookahead)433
- {434
- bool analysisdRead = false;435
- while ((analysisRead < written) && !analysisdRead)436
- {437
- while (analysisWrite < ipread)438
- {439
- analysisWrite = m_parent->m_analysisWriteCntanalysisQId.waitForChange(analysisWrite);440
- written = analysisWrite * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;441
- }442
- for (uint32_t i = 0; i < m_parent->m_queueSize; i++)443
- {444
- analysisData = &m_parent->m_analysisBufferanalysisQIdi;445
- int read = m_parent->m_analysisReadanalysisQIdi.get();446
- int write = m_parent->m_analysisWriteanalysisQIdi.get() * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;447
- if ((analysisData->poc == (uint32_t)(ipread)) && (read < write))448
- {449
- analysisIdx = i;450
- analysisdRead = true;451
- break;452
- }453
- }454
- }455
- }456
- else457
- {458
- analysisIdx = analysisRead % m_parent->m_queueSize;459
- analysisData = &m_parent->m_analysisBufferanalysisQIdanalysisIdx;460
- readPos = analysisData->poc % m_parent->m_queueSize;461
- while ((ipwrite < readPos) || ((ipwrite - 1) < (int)analysisData->poc))462
- {463
- ipwrite = m_parent->m_picWriteCntm_id.waitForChange(ipwrite);464
- }465
- }466
-467
- m_lastIdx = analysisIdx;468
- }469
- else470
- return false;471
- }472
-473
-474
- x265_picture *srcPic = (x265_picture*)(m_parent->m_inputPicBufferm_idreadPos);475
-476
- x265_picture *pic = (x265_picture*)(dstPic);477
- pic->colorSpace = srcPic->colorSpace;478
- pic->bitDepth = srcPic->bitDepth;479
- pic->framesize = srcPic->framesize;480
- pic->height = srcPic->height;481
- pic->pts = srcPic->pts;482
- pic->dts = srcPic->dts;483
- pic->reorderedPts = srcPic->reorderedPts;484
- pic->width = srcPic->width;485
- pic->analysisData = srcPic->analysisData;486
- pic->userSEI = srcPic->userSEI;487
- pic->stride0 = srcPic->stride0;488
- pic->stride1 = srcPic->stride1;489
- pic->stride2 = srcPic->stride2;490
- pic->planes0 = srcPic->planes0;491
- pic->planes1 = srcPic->planes1;492
- pic->planes2 = srcPic->planes2;493
- if (isAbrLoad)494
- pic->analysisData = *analysisData;495
- return true;496
- }497
- else498
- return false;499
- }500
-501
- void PassEncoder::threadMain()502
- {503
+/*****************************************************************************504
+* Copyright (C) 2013-2020 MulticoreWare, Inc505
+*506
+* Authors: Pooja Venkatesan <pooja@multicorewareinc.com>507
+* Aruna Matheswaran <aruna@multicorewareinc.com>508
+*509
+* This program is free software; you can redistribute it and/or modify510
+* it under the terms of the GNU General Public License as published by511
+* the Free Software Foundation; either version 2 of the License, or512
+* (at your option) any later version.513
+*514
+* This program is distributed in the hope that it will be useful,515
+* but WITHOUT ANY WARRANTY; without even the implied warranty of516
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the517
+* GNU General Public License for more details.518
+*519
+* You should have received a copy of the GNU General Public License520
+* along with this program; if not, write to the Free Software521
+* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.522
+*523
+* This program is also available under a commercial proprietary license.524
+* For more information, contact us at license @ x265.com.525
+*****************************************************************************/526
+527
+#include "abrEncApp.h"528
+#include "mv.h"529
+#include "slice.h"530
+#include "param.h"531
+532
+#include <signal.h>533
+#include <errno.h>534
+535
+#include <queue>536
+537
+using namespace X265_NS;538
+539
+/* Ctrl-C handler */540
+static volatile sig_atomic_t b_ctrl_c /* = 0 */;541
+static void sigint_handler(int)542
+{543
+ b_ctrl_c = 1;544
+}545
+546
+namespace X265_NS {547
+ // private namespace548
+#define X265_INPUT_QUEUE_SIZE 250549
+550
+ AbrEncoder::AbrEncoder(CLIOptions cliopt, uint8_t numEncodes, int &ret)551
+ {552
+ m_numEncodes = numEncodes;553
+ m_numActiveEncodes.set(numEncodes);554
+ m_queueSize = (numEncodes > 1) ? X265_INPUT_QUEUE_SIZE : 1;555
+ m_passEnc = X265_MALLOC(PassEncoder*, m_numEncodes);556
+557
+ for (uint8_t i = 0; i < m_numEncodes; i++)558
+ {559
+ m_passEnci = new PassEncoder(i, cliopti, this);560
+ if (!m_passEnci)561
+ {562
+ x265_log(NULL, X265_LOG_ERROR, "Unable to allocate memory for passEncoder\n");563
+ ret = 4;564
+ }565
+ m_passEnci->init(ret);566
+ }567
+568
+ if (!allocBuffers())569
+ {570
+ x265_log(NULL, X265_LOG_ERROR, "Unable to allocate memory for buffers\n");571
+ ret = 4;572
+ }573
+574
+ /* start passEncoder worker threads */575
+ for (uint8_t pass = 0; pass < m_numEncodes; pass++)576
+ m_passEncpass->startThreads();577
+ }578
+579
+ bool AbrEncoder::allocBuffers()580
+ {581
+ m_inputPicBuffer = X265_MALLOC(x265_picture**, m_numEncodes);582
+ m_analysisBuffer = X265_MALLOC(x265_analysis_data*, m_numEncodes);583
+584
+ m_picWriteCnt = new ThreadSafeIntegerm_numEncodes;585
+ m_picReadCnt = new ThreadSafeIntegerm_numEncodes;586
+ m_analysisWriteCnt = new ThreadSafeIntegerm_numEncodes;587
+ m_analysisReadCnt = new ThreadSafeIntegerm_numEncodes;588
+589
+ m_picIdxReadCnt = X265_MALLOC(ThreadSafeInteger*, m_numEncodes);590
+ m_analysisWrite = X265_MALLOC(ThreadSafeInteger*, m_numEncodes);591
+ m_analysisRead = X265_MALLOC(ThreadSafeInteger*, m_numEncodes);592
+ m_readFlag = X265_MALLOC(int*, m_numEncodes);593
+594
+ for (uint8_t pass = 0; pass < m_numEncodes; pass++)595
+ {596
+ m_inputPicBufferpass = X265_MALLOC(x265_picture*, m_queueSize);597
+ for (uint32_t idx = 0; idx < m_queueSize; idx++)598
+ {599
+ m_inputPicBufferpassidx = x265_picture_alloc();600
+ x265_picture_init(m_passEncpass->m_param, m_inputPicBufferpassidx);601
+ }602
+603
+ CHECKED_MALLOC_ZERO(m_analysisBufferpass, x265_analysis_data, m_queueSize);604
+ m_picIdxReadCntpass = new ThreadSafeIntegerm_queueSize;605
+ m_analysisWritepass = new ThreadSafeIntegerm_queueSize;606
+ m_analysisReadpass = new ThreadSafeIntegerm_queueSize;607
+ m_readFlagpass = X265_MALLOC(int, m_queueSize);608
+ }609
+ return true;610
+ fail:611
+ return false;612
+ }613
+614
+ void AbrEncoder::destroy()615
+ {616
+ x265_cleanup(); /* Free library singletons */617
+ for (uint8_t pass = 0; pass < m_numEncodes; pass++)618
+ {619
+ for (uint32_t index = 0; index < m_queueSize; index++)620
+ {621
+ X265_FREE(m_inputPicBufferpassindex->planes0);622
+ x265_picture_free(m_inputPicBufferpassindex);623
+ }624
+625
+ X265_FREE(m_inputPicBufferpass);626
+ X265_FREE(m_analysisBufferpass);627
+ X265_FREE(m_readFlagpass);628
+ delete m_picIdxReadCntpass;629
+ delete m_analysisWritepass;630
+ delete m_analysisReadpass;631
+ m_passEncpass->destroy();632
+ delete m_passEncpass;633
+ }634
+ X265_FREE(m_inputPicBuffer);635
+ X265_FREE(m_analysisBuffer);636
+ X265_FREE(m_readFlag);637
+638
+ delete m_picWriteCnt;639
+ delete m_picReadCnt;640
+ delete m_analysisWriteCnt;641
+ delete m_analysisReadCnt;642
+643
+ X265_FREE(m_picIdxReadCnt);644
+ X265_FREE(m_analysisWrite);645
+ X265_FREE(m_analysisRead);646
+647
+ X265_FREE(m_passEnc);648
+ }649
+650
+ PassEncoder::PassEncoder(uint32_t id, CLIOptions cliopt, AbrEncoder *parent)651
+ {652
+ m_id = id;653
+ m_cliopt = cliopt;654
+ m_parent = parent;655
+ if(!(m_cliopt.enableScaler && m_id))656
+ m_input = m_cliopt.input;657
+ m_param = cliopt.param;658
+ m_inputOver = false;659
+ m_lastIdx = -1;660
+ m_encoder = NULL;661
+ m_scaler = NULL;662
+ m_reader = NULL;663
+ m_ret = 0;664
+ }665
+666
+ int PassEncoder::init(int &result)667
+ {668
+ if (m_parent->m_numEncodes > 1)669
+ setReuseLevel();670
+ 671
+ if (!(m_cliopt.enableScaler && m_id))672
+ m_reader = new Reader(m_id, this);673
+ else674
+ {675
+ VideoDesc *src = NULL, *dst = NULL;676
+ dst = new VideoDesc(m_param->sourceWidth, m_param->sourceHeight, m_param->internalCsp, m_param->internalBitDepth);677
+ int dstW = m_parent->m_passEncm_id - 1->m_param->sourceWidth;678
+ int dstH = m_parent->m_passEncm_id - 1->m_param->sourceHeight;679
+ src = new VideoDesc(dstW, dstH, m_param->internalCsp, m_param->internalBitDepth);680
+ if (src != NULL && dst != NULL)681
+ {682
+ m_scaler = new Scaler(0, 1, m_id, src, dst, this);683
+ if (!m_scaler)684
+ {685
+ x265_log(m_param, X265_LOG_ERROR, "\n MALLOC failure in Scaler");686
+ result = 4;687
+ }688
+ }689
+ }690
+691
+ if (m_cliopt.zoneFile)692
+ {693
+ if (!m_cliopt.parseZoneFile())694
+ {695
+ x265_log(NULL, X265_LOG_ERROR, "Unable to parse zonefile in %s\n");696
+ fclose(m_cliopt.zoneFile);697
+ m_cliopt.zoneFile = NULL;698
+ }699
+ }700
+701
+ /* note: we could try to acquire a different libx265 API here based on702
+ * the profile found during option parsing, but it must be done before703
+ * opening an encoder */704
+705
+ if (m_param)706
+ m_encoder = m_cliopt.api->encoder_open(m_param);707
+ if (!m_encoder)708
+ {709
+ x265_log(NULL, X265_LOG_ERROR, "x265_encoder_open() failed for Enc, \n");710
+ m_ret = 2;711
+ return -1;712
+ }713
+714
+ /* get the encoder parameters post-initialization */715
+ m_cliopt.api->encoder_parameters(m_encoder, m_param);716
+717
+ return 1;718
+ }719
+720
+ void PassEncoder::setReuseLevel()721
+ {722
+ uint32_t r, padh = 0, padw = 0;723
+724
+ m_param->confWinBottomOffset = m_param->confWinRightOffset = 0;725
+726
+ m_param->analysisLoadReuseLevel = m_cliopt.loadLevel;727
+ m_param->analysisSaveReuseLevel = m_cliopt.saveLevel;728
+ m_param->analysisSave = m_cliopt.saveLevel ? "save.dat" : NULL;729
+ m_param->analysisLoad = m_cliopt.loadLevel ? "load.dat" : NULL;730
+ m_param->bUseAnalysisFile = 0;731
+732
+ if (m_cliopt.loadLevel)733
+ {734
+ x265_param *refParam = m_parent->m_passEncm_cliopt.refId->m_param;735
+736
+ if (m_param->sourceHeight == (refParam->sourceHeight - refParam->confWinBottomOffset) &&737
+ m_param->sourceWidth == (refParam->sourceWidth - refParam->confWinRightOffset))738
+ {739
+ m_parent->m_passEncm_id->m_param->confWinBottomOffset = refParam->confWinBottomOffset;740
+ m_parent->m_passEncm_id->m_param->confWinRightOffset = refParam->confWinRightOffset;741
+ }742
+ else743
+ {744
+ int srcH = refParam->sourceHeight - refParam->confWinBottomOffset;745
+ int srcW = refParam->sourceWidth - refParam->confWinRightOffset;746
+747
+ double scaleFactorH = double(m_param->sourceHeight / srcH);748
+ double scaleFactorW = double(m_param->sourceWidth / srcW);749
+750
+ int absScaleFactorH = (int)(10 * scaleFactorH + 0.5);751
+ int absScaleFactorW = (int)(10 * scaleFactorW + 0.5);752
+753
+ if (absScaleFactorH == 20 && absScaleFactorW == 20)754
+ {755
+ m_param->scaleFactor = 2;756
+757
+ m_parent->m_passEncm_id->m_param->confWinBottomOffset = refParam->confWinBottomOffset * 2;758
+ m_parent->m_passEncm_id->m_param->confWinRightOffset = refParam->confWinRightOffset * 2;759
+760
+ }761
+ }762
+ }763
+764
+ int h = m_param->sourceHeight + m_param->confWinBottomOffset;765
+ int w = m_param->sourceWidth + m_param->confWinRightOffset;766
+ if (h & (m_param->minCUSize - 1))767
+ {768
+ r = h & (m_param->minCUSize - 1);769
+ padh = m_param->minCUSize - r;770
+ m_param->confWinBottomOffset += padh;771
+772
+ }773
+774
+ if (w & (m_param->minCUSize - 1))775
+ {776
+ r = w & (m_param->minCUSize - 1);777
+ padw = m_param->minCUSize - r;778
+ m_param->confWinRightOffset += padw;779
+ }780
+ }781
+782
+ void PassEncoder::startThreads()783
+ {784
+ /* Start slave worker threads */785
+ m_threadActive = true;786
+ start();787
+ /* Start reader threads*/788
+ if (m_reader != NULL)789
+ {790
+ m_reader->m_threadActive = true;791
+ m_reader->start();792
+ }793
+ /* Start scaling worker threads */794
+ if (m_scaler != NULL)795
+ {796
+ m_scaler->m_threadActive = true;797
+ m_scaler->start();798
+ }799
+ }800
+801
+ void PassEncoder::copyInfo(x265_analysis_data * src)802
+ {803
+804
+ uint32_t written = m_parent->m_analysisWriteCntm_id.get();805
+806
+ int index = written % m_parent->m_queueSize;807
+ //If all streams have read analysis data, reuse that position in Queue808
+809
+ int read = m_parent->m_analysisReadm_idindex.get();810
+ int write = m_parent->m_analysisWritem_idindex.get();811
+812
+ int overwrite = written / m_parent->m_queueSize;813
+ bool emptyIdxFound = 0;814
+ while (!emptyIdxFound && overwrite)815
+ {816
+ for (uint32_t i = 0; i < m_parent->m_queueSize; i++)817
+ {818
+ read = m_parent->m_analysisReadm_idi.get();819
+ write = m_parent->m_analysisWritem_idi.get();820
+ write *= m_cliopt.numRefs;821
+822
+ if (read == write)823
+ {824
+ index = i;825
+ emptyIdxFound = 1;826
+ }827
+ }828
+ }829
+830
+ x265_analysis_data *m_analysisInfo = &m_parent->m_analysisBufferm_idindex;831
+832
+ x265_free_analysis_data(m_param, m_analysisInfo);833
+ memcpy(m_analysisInfo, src, sizeof(x265_analysis_data));834
+ x265_alloc_analysis_data(m_param, m_analysisInfo);835
+836
+ bool isVbv = m_param->rc.vbvBufferSize && m_param->rc.vbvMaxBitrate;837
+ if (m_param->bDisableLookahead && isVbv)838
+ {839
+ memcpy(m_analysisInfo->lookahead.intraSatdForVbv, src->lookahead.intraSatdForVbv, src->numCuInHeight * sizeof(uint32_t));840
+ memcpy(m_analysisInfo->lookahead.satdForVbv, src->lookahead.satdForVbv, src->numCuInHeight * sizeof(uint32_t));841
+ memcpy(m_analysisInfo->lookahead.intraVbvCost, src->lookahead.intraVbvCost, src->numCUsInFrame * sizeof(uint32_t));842
+ memcpy(m_analysisInfo->lookahead.vbvCost, src->lookahead.vbvCost, src->numCUsInFrame * sizeof(uint32_t));843
+ }844
+845
+ if (src->sliceType == X265_TYPE_IDR || src->sliceType == X265_TYPE_I)846
+ {847
+ if (m_param->analysisSaveReuseLevel < 2)848
+ goto ret;849
+ x265_analysis_intra_data *intraDst, *intraSrc;850
+ intraDst = (x265_analysis_intra_data*)m_analysisInfo->intraData;851
+ intraSrc = (x265_analysis_intra_data*)src->intraData;852
+ memcpy(intraDst->depth, intraSrc->depth, sizeof(uint8_t) * src->depthBytes);853
+ memcpy(intraDst->modes, intraSrc->modes, sizeof(uint8_t) * src->numCUsInFrame * src->numPartitions);854
+ memcpy(intraDst->partSizes, intraSrc->partSizes, sizeof(char) * src->depthBytes);855
+ memcpy(intraDst->chromaModes, intraSrc->chromaModes, sizeof(uint8_t) * src->depthBytes);856
+ if (m_param->rc.cuTree)857
+ memcpy(intraDst->cuQPOff, intraSrc->cuQPOff, sizeof(int8_t) * src->depthBytes);858
+ }859
+ else860
+ {861
+ bool bIntraInInter = (src->sliceType == X265_TYPE_P || m_param->bIntraInBFrames);862
+ int numDir = src->sliceType == X265_TYPE_P ? 1 : 2;863
+ memcpy(m_analysisInfo->wt, src->wt, sizeof(WeightParam) * 3 * numDir);864
+ if (m_param->analysisSaveReuseLevel < 2)865
+ goto ret;866
+ x265_analysis_inter_data *interDst, *interSrc;867
+ interDst = (x265_analysis_inter_data*)m_analysisInfo->interData;868
+ interSrc = (x265_analysis_inter_data*)src->interData;869
+ memcpy(interDst->depth, interSrc->depth, sizeof(uint8_t) * src->depthBytes);870
+ memcpy(interDst->modes, interSrc->modes, sizeof(uint8_t) * src->depthBytes);871
+ if (m_param->rc.cuTree)872
+ memcpy(interDst->cuQPOff, interSrc->cuQPOff, sizeof(int8_t) * src->depthBytes);873
+ if (m_param->analysisSaveReuseLevel > 4)874
+ {875
+ memcpy(interDst->partSize, interSrc->partSize, sizeof(uint8_t) * src->depthBytes);876
+ memcpy(interDst->mergeFlag, interSrc->mergeFlag, sizeof(uint8_t) * src->depthBytes);877
+ if (m_param->analysisSaveReuseLevel == 10)878
+ {879
+ memcpy(interDst->interDir, interSrc->interDir, sizeof(uint8_t) * src->depthBytes);880
+ for (int dir = 0; dir < numDir; dir++)881
+ {882
+ memcpy(interDst->mvpIdxdir, interSrc->mvpIdxdir, sizeof(uint8_t) * src->depthBytes);883
+ memcpy(interDst->refIdxdir, interSrc->refIdxdir, sizeof(int8_t) * src->depthBytes);884
+ memcpy(interDst->mvdir, interSrc->mvdir, sizeof(MV) * src->depthBytes);885
+ }886
+ if (bIntraInInter)887
+ {888
+ x265_analysis_intra_data *intraDst = (x265_analysis_intra_data*)m_analysisInfo->intraData;889
+ x265_analysis_intra_data *intraSrc = (x265_analysis_intra_data*)src->intraData;890
+ memcpy(intraDst->modes, intraSrc->modes, sizeof(uint8_t) * src->numPartitions * src->numCUsInFrame);891
+ memcpy(intraDst->chromaModes, intraSrc->chromaModes, sizeof(uint8_t) * src->depthBytes);892
+ }893
+ }894
+ }895
+ if (m_param->analysisSaveReuseLevel != 10)896
+ memcpy(interDst->ref, interSrc->ref, sizeof(int32_t) * src->numCUsInFrame * X265_MAX_PRED_MODE_PER_CTU * numDir);897
+ }898
+899
+ret:900
+ //increment analysis Write counter 901
+ m_parent->m_analysisWriteCntm_id.incr();902
+ m_parent->m_analysisWritem_idindex.incr();903
+ return;904
+ }905
+906
+907
+ bool PassEncoder::readPicture(x265_picture *dstPic)908
+ {909
+ /*Check and wait if there any input frames to read*/910
+ int ipread = m_parent->m_picReadCntm_id.get();911
+ int ipwrite = m_parent->m_picWriteCntm_id.get();912
+913
+ bool isAbrLoad = m_cliopt.loadLevel && (m_parent->m_numEncodes > 1);914
+ while (!m_inputOver && (ipread == ipwrite))915
+ {916
+ ipwrite = m_parent->m_picWriteCntm_id.waitForChange(ipwrite);917
+ }918
+919
+ if (m_threadActive && ipread < ipwrite)920
+ {921
+ /*Get input index to read from inputQueue. If doesn't need analysis info, it need not wait to fetch poc from analysisQueue*/922
+ int readPos = ipread % m_parent->m_queueSize;923
+ x265_analysis_data* analysisData = 0;924
+925
+ if (isAbrLoad)926
+ {927
+ /*If stream is master of each slave pass, then fetch analysis data from prev pass*/928
+ int analysisQId = m_cliopt.refId;929
+ /*Check and wait if there any analysis Data to read*/930
+ int analysisWrite = m_parent->m_analysisWriteCntanalysisQId.get();931
+ int written = analysisWrite * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;932
+ int analysisRead = m_parent->m_analysisReadCntanalysisQId.get();933
+ 934
+ while (m_threadActive && written == analysisRead)935
+ {936
+ analysisWrite = m_parent->m_analysisWriteCntanalysisQId.waitForChange(analysisWrite);937
+ written = analysisWrite * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;938
+ }939
+940
+ if (analysisRead < written)941
+ {942
+ int analysisIdx = 0;943
+ if (!m_param->bDisableLookahead)944
+ {945
+ bool analysisdRead = false;946
+ while ((analysisRead < written) && !analysisdRead)947
+ {948
+ while (analysisWrite < ipread)949
+ {950
+ analysisWrite = m_parent->m_analysisWriteCntanalysisQId.waitForChange(analysisWrite);951
+ written = analysisWrite * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;952
+ }953
+ for (uint32_t i = 0; i < m_parent->m_queueSize; i++)954
+ {955
+ analysisData = &m_parent->m_analysisBufferanalysisQIdi;956
+ int read = m_parent->m_analysisReadanalysisQIdi.get();957
+ int write = m_parent->m_analysisWriteanalysisQIdi.get() * m_parent->m_passEncanalysisQId->m_cliopt.numRefs;958
+ if ((analysisData->poc == (uint32_t)(ipread)) && (read < write))959
+ {960
+ analysisIdx = i;961
+ analysisdRead = true;962
+ break;963
+ }964
+ }965
+ }966
+ }967
+ else968
+ {969
+ analysisIdx = analysisRead % m_parent->m_queueSize;970
+ analysisData = &m_parent->m_analysisBufferanalysisQIdanalysisIdx;971
+ readPos = analysisData->poc % m_parent->m_queueSize;972
+ while ((ipwrite < readPos) || ((ipwrite - 1) < (int)analysisData->poc))973
+ {974
+ ipwrite = m_parent->m_picWriteCntm_id.waitForChange(ipwrite);975
+ }976
+ }977
+978
+ m_lastIdx = analysisIdx;979
+ }980
+ else981
+ return false;982
+ }983
+984
+985
+ x265_picture *srcPic = (x265_picture*)(m_parent->m_inputPicBufferm_idreadPos);986
+987
+ x265_picture *pic = (x265_picture*)(dstPic);988
+ pic->colorSpace = srcPic->colorSpace;989
+ pic->bitDepth = srcPic->bitDepth;990
+ pic->framesize = srcPic->framesize;991
+ pic->height = srcPic->height;992
+ pic->pts = srcPic->pts;993
+ pic->dts = srcPic->dts;994
+ pic->reorderedPts = srcPic->reorderedPts;995
+ pic->width = srcPic->width;996
+ pic->analysisData = srcPic->analysisData;997
+ pic->userSEI = srcPic->userSEI;998
+ pic->stride0 = srcPic->stride0;999
+ pic->stride1 = srcPic->stride1;1000
+ pic->stride2 = srcPic->stride2;1001
+ pic->planes0 = srcPic->planes0;1002
+ pic->planes1 = srcPic->planes1;1003
+ pic->planes2 = srcPic->planes2;1004
+ if (isAbrLoad)1005
+ pic->analysisData = *analysisData;1006
+ return true;1007
+ }1008
+ else1009
+ return false;1010
+ }1011
+1012
+ void PassEncoder::threadMain()1013
+ {1014
THREAD_NAME("PassEncoder", m_id);1015
1016
while (m_threadActive)1017
{1018
-1019
-#if ENABLE_LIBVMAF1020
- x265_vmaf_data* vmafdata = m_cliopt.vmafData;1021
-#endif1022
- /* This allows muxers to modify bitstream format */1023
- m_cliopt.output->setParam(m_param);1024
- const x265_api* api = m_cliopt.api;1025
- ReconPlay* reconPlay = NULL;1026
- if (m_cliopt.reconPlayCmd)1027
- reconPlay = new ReconPlay(m_cliopt.reconPlayCmd, *m_param);1028
- char* profileName = m_cliopt.encName ? m_cliopt.encName : (char *)"x265";1029
-1030
- if (m_cliopt.zoneFile)1031
- {1032
- if (!m_cliopt.parseZoneFile())1033
- {1034
- x265_log(NULL, X265_LOG_ERROR, "Unable to parse zonefile in %s\n", profileName);1035
- fclose(m_cliopt.zoneFile);1036
- m_cliopt.zoneFile = NULL;1037
- }1038
- }1039
-1040
- if (signal(SIGINT, sigint_handler) == SIG_ERR)1041
- x265_log(m_param, X265_LOG_ERROR, "Unable to register CTRL+C handler: %s in %s\n",1042
- strerror(errno), profileName);1043
-1044
- x265_picture pic_orig, pic_out;1045
- x265_picture *pic_in = &pic_orig;1046
- /* Allocate recon picture if analysis save/load is enabled */1047
- std::priority_queue<int64_t>* pts_queue = m_cliopt.output->needPTS() ? new std::priority_queue<int64_t>() : NULL;1048
- x265_picture *pic_recon = (m_cliopt.recon || m_param->analysisSave || m_param->analysisLoad || pts_queue || reconPlay || m_param->csvLogLevel) ? &pic_out : NULL;1049
- uint32_t inFrameCount = 0;1050
- uint32_t outFrameCount = 0;1051
- x265_nal *p_nal;1052
- x265_stats stats;1053
- uint32_t nal;1054
- int16_t *errorBuf = NULL;1055
- bool bDolbyVisionRPU = false;1056
- uint8_t *rpuPayload = NULL;1057
- int inputPicNum = 1;1058
- x265_picture picField1, picField2;1059
- x265_analysis_data* analysisInfo = (x265_analysis_data*)(&pic_out.analysisData);1060
- bool isAbrSave = m_cliopt.saveLevel && (m_parent->m_numEncodes > 1);1061
-1062
- if (!m_param->bRepeatHeaders && !m_param->bEnableSvtHevc)1063
- {1064
- if (api->encoder_headers(m_encoder, &p_nal, &nal) < 0)1065
- {1066
- x265_log(m_param, X265_LOG_ERROR, "Failure generating stream headers in %s\n", profileName);1067
- m_ret = 3;1068
- goto fail;1069
- }1070
- else1071
- m_cliopt.totalbytes += m_cliopt.output->writeHeaders(p_nal, nal);1072
- }1073
-1074
- if (m_param->bField && m_param->interlaceMode)1075
- {1076
- api->picture_init(m_param, &picField1);1077
- api->picture_init(m_param, &picField2);1078
- // return back the original height of input1079
- m_param->sourceHeight *= 2;1080
- api->picture_init(m_param, &pic_orig);1081
- }1082
- else1083
- api->picture_init(m_param, &pic_orig);1084
-1085
- if (m_param->dolbyProfile && m_cliopt.dolbyVisionRpu)1086
- {1087
- rpuPayload = X265_MALLOC(uint8_t, 1024);1088
- pic_in->rpu.payload = rpuPayload;1089
- if (pic_in->rpu.payload)1090
- bDolbyVisionRPU = true;1091
- }1092
-1093
- if (m_cliopt.bDither)1094
- {1095
- errorBuf = X265_MALLOC(int16_t, m_param->sourceWidth + 1);1096
- if (errorBuf)1097
- memset(errorBuf, 0, (m_param->sourceWidth + 1) * sizeof(int16_t));1098
- else1099
- m_cliopt.bDither = false;1100
- }1101
-1102
- // main encoder loop1103
- while (pic_in && !b_ctrl_c)1104
- {1105
- pic_orig.poc = (m_param->bField && m_param->interlaceMode) ? inFrameCount * 2 : inFrameCount;1106
- if (m_cliopt.qpfile)1107
- {1108
- if (!m_cliopt.parseQPFile(pic_orig))1109
- {1110
- x265_log(NULL, X265_LOG_ERROR, "can't parse qpfile for frame %d in %s\n",1111
- pic_in->poc, profileName);1112
- fclose(m_cliopt.qpfile);1113
- m_cliopt.qpfile = NULL;1114
- }1115
- }1116
-1117
- if (m_cliopt.framesToBeEncoded && inFrameCount >= m_cliopt.framesToBeEncoded)1118
- pic_in = NULL;1119
- else if (readPicture(pic_in))1120
- inFrameCount++;1121
- else1122
- pic_in = NULL;1123
-1124
- if (pic_in)1125
- {1126
- if (pic_in->bitDepth > m_param->internalBitDepth && m_cliopt.bDither)1127
- {1128
- x265_dither_image(pic_in, m_cliopt.input->getWidth(), m_cliopt.input->getHeight(), errorBuf, m_param->internalBitDepth);1129
- pic_in->bitDepth = m_param->internalBitDepth;1130
- }1131
- /* Overwrite PTS */1132
- pic_in->pts = pic_in->poc;1133
-1134
- // convert to field1135
- if (m_param->bField && m_param->interlaceMode)1136
- {1137
- int height = pic_in->height >> 1;1138
-1139
- int static bCreated = 0;1140
- if (bCreated == 0)1141
- {1142
- bCreated = 1;1143
- inputPicNum = 2;1144
- picField1.fieldNum = 1;1145
- picField2.fieldNum = 2;1146
-1147
- picField1.bitDepth = picField2.bitDepth = pic_in->bitDepth;1148
- picField1.colorSpace = picField2.colorSpace = pic_in->colorSpace;1149
- picField1.height = picField2.height = pic_in->height >> 1;1150
- picField1.framesize = picField2.framesize = pic_in->framesize >> 1;1151
-1152
- size_t fieldFrameSize = (size_t)pic_in->framesize >> 1;1153
- char* field1Buf = X265_MALLOC(char, fieldFrameSize);1154
- char* field2Buf = X265_MALLOC(char, fieldFrameSize);1155
-1156
- int stride = picField1.stride0 = picField2.stride0 = pic_in->stride0;1157
- uint64_t framesize = stride * (height >> x265_cli_cspspic_in->colorSpace.height0);1158
- picField1.planes0 = field1Buf;1159
- picField2.planes0 = field2Buf;1160
- for (int i = 1; i < x265_cli_cspspic_in->colorSpace.planes; i++)1161
- {1162
- picField1.planesi = field1Buf + framesize;1163
- picField2.planesi = field2Buf + framesize;1164
-1165
- stride = picField1.stridei = picField2.stridei = pic_in->stridei;1166
- framesize += (stride * (height >> x265_cli_cspspic_in->colorSpace.heighti));1167
- }1168
- assert(framesize == picField1.framesize);1169
- }1170
-1171
- picField1.pts = picField1.poc = pic_in->poc;1172
- picField2.pts = picField2.poc = pic_in->poc + 1;1173
-1174
- picField1.userSEI = picField2.userSEI = pic_in->userSEI;1175
-1176
- //if (pic_in->userData)1177
- //{1178
- // // Have to handle userData here1179
- //}1180
-1181
- if (pic_in->framesize)1182
- {1183
- for (int i = 0; i < x265_cli_cspspic_in->colorSpace.planes; i++)1184
- {1185
- char* srcP1 = (char*)pic_in->planesi;1186
- char* srcP2 = (char*)pic_in->planesi + pic_in->stridei;1187
- char* p1 = (char*)picField1.planesi;1188
- char* p2 = (char*)picField2.planesi;1189
-1190
- int stride = picField1.stridei;1191
-1192
- for (int y = 0; y < (height >> x265_cli_cspspic_in->colorSpace.heighti); y++)1193
- {1194
- memcpy(p1, srcP1, stride);1195
- memcpy(p2, srcP2, stride);1196
- srcP1 += 2 * stride;1197
- srcP2 += 2 * stride;1198
- p1 += stride;1199
- p2 += stride;1200
- }1201
- }1202
- }1203
- }1204
-1205
- if (bDolbyVisionRPU)1206
- {1207
- if (m_param->bField && m_param->interlaceMode)1208
- {1209
- if (m_cliopt.rpuParser(&picField1) > 0)1210
- goto fail;1211
- if (m_cliopt.rpuParser(&picField2) > 0)1212
- goto fail;1213
- }1214
- else1215
- {1216
- if (m_cliopt.rpuParser(pic_in) > 0)1217
- goto fail;1218
- }1219
- }1220
- }1221
-1222
- for (int inputNum = 0; inputNum < inputPicNum; inputNum++)1223
- {1224
- x265_picture *picInput = NULL;1225
- if (inputPicNum == 2)1226
- picInput = pic_in ? (inputNum ? &picField2 : &picField1) : NULL;1227
- else1228
- picInput = pic_in;1229
-1230
- int numEncoded = api->encoder_encode(m_encoder, &p_nal, &nal, picInput, pic_recon);1231
-1232
- int idx = (inFrameCount - 1) % m_parent->m_queueSize;1233
- m_parent->m_picIdxReadCntm_ididx.incr();1234
- m_parent->m_picReadCntm_id.incr();1235
- if (m_cliopt.loadLevel && picInput)1236
- {1237
- m_parent->m_analysisReadCntm_cliopt.refId.incr();1238
- m_parent->m_analysisReadm_cliopt.refIdm_lastIdx.incr();1239
- }1240
-1241
- if (numEncoded < 0)1242
- {1243
- b_ctrl_c = 1;1244
- m_ret = 4;1245
- break;1246
- }1247
-1248
- if (reconPlay && numEncoded)1249
- reconPlay->writePicture(*pic_recon);1250
-1251
- outFrameCount += numEncoded;1252
-1253
- if (isAbrSave && numEncoded)1254
- {1255
- copyInfo(analysisInfo);1256
- }1257
-1258
- if (numEncoded && pic_recon && m_cliopt.recon)1259
- m_cliopt.recon->writePicture(pic_out);1260
- if (nal)1261
- {1262
- m_cliopt.totalbytes += m_cliopt.output->writeFrame(p_nal, nal, pic_out);1263
- if (pts_queue)1264
- {1265
- pts_queue->push(-pic_out.pts);1266
- if (pts_queue->size() > 2)1267
- pts_queue->pop();1268
- }1269
- }1270
- m_cliopt.printStatus(outFrameCount);1271
- }1272
- }1273
-1274
- /* Flush the encoder */1275
- while (!b_ctrl_c)1276
- {1277
- int numEncoded = api->encoder_encode(m_encoder, &p_nal, &nal, NULL, pic_recon);1278
- if (numEncoded < 0)1279
- {1280
- m_ret = 4;1281
- break;1282
- }1283
-1284
- if (reconPlay && numEncoded)1285
- reconPlay->writePicture(*pic_recon);1286
-1287
- outFrameCount += numEncoded;1288
- if (isAbrSave && numEncoded)1289
- {1290
- copyInfo(analysisInfo);1291
- }1292
-1293
- if (numEncoded && pic_recon && m_cliopt.recon)1294
- m_cliopt.recon->writePicture(pic_out);1295
- if (nal)1296
- {1297
- m_cliopt.totalbytes += m_cliopt.output->writeFrame(p_nal, nal, pic_out);1298
- if (pts_queue)1299
- {1300
- pts_queue->push(-pic_out.pts);1301
- if (pts_queue->size() > 2)1302
- pts_queue->pop();1303
- }1304
- }1305
-1306
- m_cliopt.printStatus(outFrameCount);1307
-1308
- if (!numEncoded)1309
- break;1310
- }1311
-1312
- if (bDolbyVisionRPU)1313
- {1314
- if (fgetc(m_cliopt.dolbyVisionRpu) != EOF)1315
- x265_log(NULL, X265_LOG_WARNING, "Dolby Vision RPU count is greater than frame count in %s\n",1316
- profileName);1317
- x265_log(NULL, X265_LOG_INFO, "VES muxing with Dolby Vision RPU file successful in %s\n",1318
- profileName);1319
- }1320
-1321
- /* clear progress report */1322
- if (m_cliopt.bProgress)1323
- fprintf(stderr, "%*s\r", 80, " ");1324
-1325
- fail:1326
-1327
- delete reconPlay;1328
-1329
- api->encoder_get_stats(m_encoder, &stats, sizeof(stats));1330
- if (m_param->csvfn && !b_ctrl_c)1331
-#if ENABLE_LIBVMAF1332
- api->vmaf_encoder_log(m_encoder, m_cliopt.argCnt, m_cliopt.argString, m_cliopt.param, vmafdata);1333
-#else1334
- api->encoder_log(m_encoder, m_cliopt.argCnt, m_cliopt.argString);1335
-#endif1336
- api->encoder_close(m_encoder);1337
-1338
- int64_t second_largest_pts = 0;1339
- int64_t largest_pts = 0;1340
- if (pts_queue && pts_queue->size() >= 2)1341
- {1342
- second_largest_pts = -pts_queue->top();1343
- pts_queue->pop();1344
- largest_pts = -pts_queue->top();1345
- pts_queue->pop();1346
- delete pts_queue;1347
- pts_queue = NULL;1348
- }1349
- m_cliopt.output->closeFile(largest_pts, second_largest_pts);1350
-1351
- if (b_ctrl_c)1352
- general_log(m_param, NULL, X265_LOG_INFO, "aborted at input frame %d, output frame %d in %s\n",1353
- m_cliopt.seek + inFrameCount, stats.encodedPictureCount, profileName);1354
-1355
- api->param_free(m_param);1356
-1357
- X265_FREE(errorBuf);1358
- X265_FREE(rpuPayload);1359
-1360
- m_threadActive = false;1361
- m_parent->m_numActiveEncodes.decr();1362
- }1363
- }1364
-1365
- void PassEncoder::destroy()1366
- {1367
- stop();1368
- if (m_reader)1369
- {1370
- m_reader->stop();1371
- delete m_reader;1372
- }1373
- else1374
- {1375
- m_scaler->stop();1376
- m_scaler->destroy();1377
- delete m_scaler;1378
- }1379
- }1380
-1381
- Scaler::Scaler(int threadId, int threadNum, int id, VideoDesc *src, VideoDesc *dst, PassEncoder *parentEnc)1382
- {1383
- m_parentEnc = parentEnc;1384
- m_id = id;1385
- m_srcFormat = src;1386
- m_dstFormat = dst;1387
- m_threadActive = false;1388
- m_scaleFrameSize = 0;1389
- m_filterManager = NULL;1390
- m_threadId = threadId;1391
- m_threadTotal = threadNum;1392
-1393
- int csp = dst->m_csp;1394
- uint32_t pixelbytes = dst->m_inputDepth > 8 ? 2 : 1;1395
- for (int i = 0; i < x265_cli_cspscsp.planes; i++)1396
- {1397
- int w = dst->m_width >> x265_cli_cspscsp.widthi;1398
- int h = dst->m_height >> x265_cli_cspscsp.heighti;1399
- m_scalePlanesi = w * h * pixelbytes;1400
- m_scaleFrameSize += m_scalePlanesi;1401
- }1402
-1403
- if (src->m_height != dst->m_height || src->m_width != dst->m_width)1404
- {1405
- m_filterManager = new ScalerFilterManager;1406
- m_filterManager->init(4, m_srcFormat, m_dstFormat);1407
- }1408
- }1409
-1410
- bool Scaler::scalePic(x265_picture * destination, x265_picture * source)1411
- {1412
- if (!destination || !source)1413
- return false;1414
- x265_param* param = m_parentEnc->m_param;1415
- int pixelBytes = m_dstFormat->m_inputDepth > 8 ? 2 : 1;1416
- if (m_srcFormat->m_height != m_dstFormat->m_height || m_srcFormat->m_width != m_dstFormat->m_width)1417
- {1418
- void **srcPlane = NULL, **dstPlane = NULL;1419
- int srcStride3, dstStride3;1420
- destination->bitDepth = source->bitDepth;1421
- destination->colorSpace = source->colorSpace;1422
- destination->pts = source->pts;1423
- destination->dts = source->dts;1424
- destination->reorderedPts = source->reorderedPts;1425
- destination->poc = source->poc;1426
- destination->userSEI = source->userSEI;1427
- srcPlane = source->planes;1428
- dstPlane = destination->planes;1429
- srcStride0 = source->stride0;1430
- destination->stride0 = m_dstFormat->m_width * pixelBytes;1431
- dstStride0 = destination->stride0;1432
- if (param->internalCsp != X265_CSP_I400)1433
- {1434
- srcStride1 = source->stride1;1435
- srcStride2 = source->stride2;1436
- destination->stride1 = destination->stride0 >> x265_cli_cspsparam->internalCsp.width1;1437
- destination->stride2 = destination->stride0 >> x265_cli_cspsparam->internalCsp.width2;1438
- dstStride1 = destination->stride1;1439
- dstStride2 = destination->stride2;1440
- }1441
- if (m_scaleFrameSize)1442
- {1443
- m_filterManager->scale_pic(srcPlane, dstPlane, srcStride, dstStride);1444
- return true;1445
- }1446
- else1447
- x265_log(param, X265_LOG_INFO, "Empty frame received\n");1448
- }1449
- return false;1450
- }1451
-1452
- void Scaler::threadMain()1453
- {1454
- THREAD_NAME("Scaler", m_id);1455
-1456
- /* unscaled picture is stored in the last index */1457
- uint32_t srcId = m_id - 1;1458
- int QDepth = m_parentEnc->m_parent->m_queueSize;1459
- while (!m_parentEnc->m_inputOver)1460
- {1461
-1462
- uint32_t scaledWritten = m_parentEnc->m_parent->m_picWriteCntm_id.get();1463
-1464
- if (m_parentEnc->m_cliopt.framesToBeEncoded && scaledWritten >= m_parentEnc->m_cliopt.framesToBeEncoded)1465
- break;1466
-1467
- if (m_threadTotal > 1 && (m_threadId != scaledWritten % m_threadTotal))1468
- {1469
- continue;1470
- }1471
- uint32_t written = m_parentEnc->m_parent->m_picWriteCntsrcId.get();1472
-1473
- /*If all the input pictures are scaled by the current scale worker thread wait for input pictures*/1474
- while (m_threadActive && (scaledWritten == written)) {1475
- written = m_parentEnc->m_parent->m_picWriteCntsrcId.waitForChange(written);1476
- }1477
-1478
- if (m_threadActive && scaledWritten < written)1479
- {1480
-1481
- int scaledWriteIdx = scaledWritten % QDepth;1482
- int overWritePicBuffer = scaledWritten / QDepth;1483
- int read = m_parentEnc->m_parent->m_picIdxReadCntm_idscaledWriteIdx.get();1484
-1485
- while (overWritePicBuffer && read < overWritePicBuffer)1486
- {1487
- read = m_parentEnc->m_parent->m_picIdxReadCntm_idscaledWriteIdx.waitForChange(read);1488
- }1489
-1490
- if (!m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx)1491
- {1492
- int framesize = 0;1493
- int planesize3;1494
- int csp = m_dstFormat->m_csp;1495
- int stride3;1496
- stride0 = m_dstFormat->m_width;1497
- stride1 = stride0 >> x265_cli_cspscsp.width1;1498
- stride2 = stride0 >> x265_cli_cspscsp.width2;1499
- for (int i = 0; i < x265_cli_cspscsp.planes; i++)1500
- {1501
- uint32_t h = m_dstFormat->m_height >> x265_cli_cspscsp.heighti;1502
- planesizei = h * stridei;1503
- framesize += planesizei;1504
- }1505
-1506
- m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx = x265_picture_alloc();1507
- x265_picture_init(m_parentEnc->m_param, m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx);1508
-1509
- ((x265_picture*)m_parentEnc->m_parent->m_inputPicBufferm_idscaledWritten % QDepth)->framesize = framesize;1510
- for (int32_t j = 0; j < x265_cli_cspscsp.planes; j++)1511
- {1512
- m_parentEnc->m_parent->m_inputPicBufferm_idscaledWritten % QDepth->planesj = X265_MALLOC(char, planesizej);1513
- }1514
- }1515
-1516
- x265_picture *srcPic = m_parentEnc->m_parent->m_inputPicBuffersrcIdscaledWritten % QDepth;1517
- x265_picture* destPic = m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx;1518
-1519
- // Enqueue this picture up with the current encoder so that it will asynchronously encode1520
- if (!scalePic(destPic, srcPic))1521
- x265_log(NULL, X265_LOG_ERROR, "Unable to copy scaled input picture to input queue \n");1522
- else1523
- m_parentEnc->m_parent->m_picWriteCntm_id.incr();1524
- m_scaledWriteCnt.incr();1525
- m_parentEnc->m_parent->m_picIdxReadCntsrcIdscaledWriteIdx.incr();1526
- }1527
- if (m_threadTotal > 1)1528
- {1529
- written = m_parentEnc->m_parent->m_picWriteCntsrcId.get();1530
- int totalWrite = written / m_threadTotal;1531
- if (written % m_threadTotal > m_threadId)1532
- totalWrite++;1533
- if (totalWrite == m_scaledWriteCnt.get())1534
- {1535
- m_parentEnc->m_parent->m_picWriteCntsrcId.poke();1536
- m_parentEnc->m_parent->m_picWriteCntm_id.poke();1537
- break;1538
- }1539
- }1540
- else1541
- {1542
- /* Once end of video is reached and all frames are scaled, release wait on picwritecount */1543
- scaledWritten = m_parentEnc->m_parent->m_picWriteCntm_id.get();1544
- written = m_parentEnc->m_parent->m_picWriteCntsrcId.get();1545
- if (written == scaledWritten)1546
- {1547
- m_parentEnc->m_parent->m_picWriteCntsrcId.poke();1548
- m_parentEnc->m_parent->m_picWriteCntm_id.poke();1549
- break;1550
- }1551
- }1552
-1553
- }1554
- m_threadActive = false;1555
- destroy();1556
- }1557
-1558
- Reader::Reader(int id, PassEncoder *parentEnc)1559
- {1560
- m_parentEnc = parentEnc;1561
- m_id = id;1562
- m_input = parentEnc->m_input;1563
- }1564
-1565
- void Reader::threadMain()1566
- {1567
- THREAD_NAME("Reader", m_id);1568
-1569
- int QDepth = m_parentEnc->m_parent->m_queueSize;1570
- x265_picture* src = x265_picture_alloc();1571
- x265_picture_init(m_parentEnc->m_param, src);1572
-1573
- while (m_threadActive)1574
- {1575
- uint32_t written = m_parentEnc->m_parent->m_picWriteCntm_id.get();1576
- uint32_t writeIdx = written % QDepth;1577
- uint32_t read = m_parentEnc->m_parent->m_picIdxReadCntm_idwriteIdx.get();1578
- uint32_t overWritePicBuffer = written / QDepth;1579
-1580
- if (m_parentEnc->m_cliopt.framesToBeEncoded && written >= m_parentEnc->m_cliopt.framesToBeEncoded)1581
- break;1582
-1583
- while (overWritePicBuffer && read < overWritePicBuffer)1584
- {1585
- read = m_parentEnc->m_parent->m_picIdxReadCntm_idwriteIdx.waitForChange(read);1586
- }1587
-1588
- x265_picture* dest = m_parentEnc->m_parent->m_inputPicBufferm_idwriteIdx;1589
- if (m_input->readPicture(*src))1590
- {1591
- dest->poc = src->poc;1592
- dest->pts = src->pts;1593
- dest->userSEI = src->userSEI;1594
- dest->bitDepth = src->bitDepth;1595
- dest->framesize = src->framesize;1596
- dest->height = src->height;1597
- dest->width = src->width;1598
- dest->colorSpace = src->colorSpace;1599
- dest->userSEI = src->userSEI;1600
- dest->rpu.payload = src->rpu.payload;1601
- dest->picStruct = src->picStruct;1602
- dest->stride0 = src->stride0;1603
- dest->stride1 = src->stride1;1604
- dest->stride2 = src->stride2;1605
-1606
- if (!dest->planes0)1607
- dest->planes0 = X265_MALLOC(char, dest->framesize);1608
-1609
- memcpy(dest->planes0, src->planes0, src->framesize * sizeof(char));1610
- dest->planes1 = (char*)dest->planes0 + src->stride0 * src->height;1611
- dest->planes2 = (char*)dest->planes1 + src->stride1 * (src->height >> x265_cli_cspssrc->colorSpace.height1);1612
- m_parentEnc->m_parent->m_picWriteCntm_id.incr();1613
- }1614
- else1615
- {1616
- m_threadActive = false;1617
- m_parentEnc->m_inputOver = true;1618
- m_parentEnc->m_parent->m_picWriteCntm_id.poke();1619
- }1620
- }1621
- x265_picture_free(src);1622
- }1623
-}1624
+1625
+#if ENABLE_LIBVMAF1626
+ x265_vmaf_data* vmafdata = m_cliopt.vmafData;1627
+#endif1628
+ /* This allows muxers to modify bitstream format */1629
+ m_cliopt.output->setParam(m_param);1630
+ const x265_api* api = m_cliopt.api;1631
+ ReconPlay* reconPlay = NULL;1632
+ if (m_cliopt.reconPlayCmd)1633
+ reconPlay = new ReconPlay(m_cliopt.reconPlayCmd, *m_param);1634
+ char* profileName = m_cliopt.encName ? m_cliopt.encName : (char *)"x265";1635
+1636
+ if (signal(SIGINT, sigint_handler) == SIG_ERR)1637
+ x265_log(m_param, X265_LOG_ERROR, "Unable to register CTRL+C handler: %s in %s\n",1638
+ strerror(errno), profileName);1639
+1640
+ x265_picture pic_orig, pic_out;1641
+ x265_picture *pic_in = &pic_orig;1642
+ /* Allocate recon picture if analysis save/load is enabled */1643
+ std::priority_queue<int64_t>* pts_queue = m_cliopt.output->needPTS() ? new std::priority_queue<int64_t>() : NULL;1644
+ x265_picture *pic_recon = (m_cliopt.recon || m_param->analysisSave || m_param->analysisLoad || pts_queue || reconPlay || m_param->csvLogLevel) ? &pic_out : NULL;1645
+ uint32_t inFrameCount = 0;1646
+ uint32_t outFrameCount = 0;1647
+ x265_nal *p_nal;1648
+ x265_stats stats;1649
+ uint32_t nal;1650
+ int16_t *errorBuf = NULL;1651
+ bool bDolbyVisionRPU = false;1652
+ uint8_t *rpuPayload = NULL;1653
+ int inputPicNum = 1;1654
+ x265_picture picField1, picField2;1655
+ x265_analysis_data* analysisInfo = (x265_analysis_data*)(&pic_out.analysisData);1656
+ bool isAbrSave = m_cliopt.saveLevel && (m_parent->m_numEncodes > 1);1657
+1658
+ if (!m_param->bRepeatHeaders && !m_param->bEnableSvtHevc)1659
+ {1660
+ if (api->encoder_headers(m_encoder, &p_nal, &nal) < 0)1661
+ {1662
+ x265_log(m_param, X265_LOG_ERROR, "Failure generating stream headers in %s\n", profileName);1663
+ m_ret = 3;1664
+ goto fail;1665
+ }1666
+ else1667
+ m_cliopt.totalbytes += m_cliopt.output->writeHeaders(p_nal, nal);1668
+ }1669
+1670
+ if (m_param->bField && m_param->interlaceMode)1671
+ {1672
+ api->picture_init(m_param, &picField1);1673
+ api->picture_init(m_param, &picField2);1674
+ // return back the original height of input1675
+ m_param->sourceHeight *= 2;1676
+ api->picture_init(m_param, &pic_orig);1677
+ }1678
+ else1679
+ api->picture_init(m_param, &pic_orig);1680
+1681
+ if (m_param->dolbyProfile && m_cliopt.dolbyVisionRpu)1682
+ {1683
+ rpuPayload = X265_MALLOC(uint8_t, 1024);1684
+ pic_in->rpu.payload = rpuPayload;1685
+ if (pic_in->rpu.payload)1686
+ bDolbyVisionRPU = true;1687
+ }1688
+1689
+ if (m_cliopt.bDither)1690
+ {1691
+ errorBuf = X265_MALLOC(int16_t, m_param->sourceWidth + 1);1692
+ if (errorBuf)1693
+ memset(errorBuf, 0, (m_param->sourceWidth + 1) * sizeof(int16_t));1694
+ else1695
+ m_cliopt.bDither = false;1696
+ }1697
+1698
+ // main encoder loop1699
+ while (pic_in && !b_ctrl_c)1700
+ {1701
+ pic_orig.poc = (m_param->bField && m_param->interlaceMode) ? inFrameCount * 2 : inFrameCount;1702
+ if (m_cliopt.qpfile)1703
+ {1704
+ if (!m_cliopt.parseQPFile(pic_orig))1705
+ {1706
+ x265_log(NULL, X265_LOG_ERROR, "can't parse qpfile for frame %d in %s\n",1707
+ pic_in->poc, profileName);1708
+ fclose(m_cliopt.qpfile);1709
+ m_cliopt.qpfile = NULL;1710
+ }1711
+ }1712
+1713
+ if (m_cliopt.framesToBeEncoded && inFrameCount >= m_cliopt.framesToBeEncoded)1714
+ pic_in = NULL;1715
+ else if (readPicture(pic_in))1716
+ inFrameCount++;1717
+ else1718
+ pic_in = NULL;1719
+1720
+ if (pic_in)1721
+ {1722
+ if (pic_in->bitDepth > m_param->internalBitDepth && m_cliopt.bDither)1723
+ {1724
+ x265_dither_image(pic_in, m_cliopt.input->getWidth(), m_cliopt.input->getHeight(), errorBuf, m_param->internalBitDepth);1725
+ pic_in->bitDepth = m_param->internalBitDepth;1726
+ }1727
+ /* Overwrite PTS */1728
+ pic_in->pts = pic_in->poc;1729
+1730
+ // convert to field1731
+ if (m_param->bField && m_param->interlaceMode)1732
+ {1733
+ int height = pic_in->height >> 1;1734
+1735
+ int static bCreated = 0;1736
+ if (bCreated == 0)1737
+ {1738
+ bCreated = 1;1739
+ inputPicNum = 2;1740
+ picField1.fieldNum = 1;1741
+ picField2.fieldNum = 2;1742
+1743
+ picField1.bitDepth = picField2.bitDepth = pic_in->bitDepth;1744
+ picField1.colorSpace = picField2.colorSpace = pic_in->colorSpace;1745
+ picField1.height = picField2.height = pic_in->height >> 1;1746
+ picField1.framesize = picField2.framesize = pic_in->framesize >> 1;1747
+1748
+ size_t fieldFrameSize = (size_t)pic_in->framesize >> 1;1749
+ char* field1Buf = X265_MALLOC(char, fieldFrameSize);1750
+ char* field2Buf = X265_MALLOC(char, fieldFrameSize);1751
+1752
+ int stride = picField1.stride0 = picField2.stride0 = pic_in->stride0;1753
+ uint64_t framesize = stride * (height >> x265_cli_cspspic_in->colorSpace.height0);1754
+ picField1.planes0 = field1Buf;1755
+ picField2.planes0 = field2Buf;1756
+ for (int i = 1; i < x265_cli_cspspic_in->colorSpace.planes; i++)1757
+ {1758
+ picField1.planesi = field1Buf + framesize;1759
+ picField2.planesi = field2Buf + framesize;1760
+1761
+ stride = picField1.stridei = picField2.stridei = pic_in->stridei;1762
+ framesize += (stride * (height >> x265_cli_cspspic_in->colorSpace.heighti));1763
+ }1764
+ assert(framesize == picField1.framesize);1765
+ }1766
+1767
+ picField1.pts = picField1.poc = pic_in->poc;1768
+ picField2.pts = picField2.poc = pic_in->poc + 1;1769
+1770
+ picField1.userSEI = picField2.userSEI = pic_in->userSEI;1771
+1772
+ //if (pic_in->userData)1773
+ //{1774
+ // // Have to handle userData here1775
+ //}1776
+1777
+ if (pic_in->framesize)1778
+ {1779
+ for (int i = 0; i < x265_cli_cspspic_in->colorSpace.planes; i++)1780
+ {1781
+ char* srcP1 = (char*)pic_in->planesi;1782
+ char* srcP2 = (char*)pic_in->planesi + pic_in->stridei;1783
+ char* p1 = (char*)picField1.planesi;1784
+ char* p2 = (char*)picField2.planesi;1785
+1786
+ int stride = picField1.stridei;1787
+1788
+ for (int y = 0; y < (height >> x265_cli_cspspic_in->colorSpace.heighti); y++)1789
+ {1790
+ memcpy(p1, srcP1, stride);1791
+ memcpy(p2, srcP2, stride);1792
+ srcP1 += 2 * stride;1793
+ srcP2 += 2 * stride;1794
+ p1 += stride;1795
+ p2 += stride;1796
+ }1797
+ }1798
+ }1799
+ }1800
+1801
+ if (bDolbyVisionRPU)1802
+ {1803
+ if (m_param->bField && m_param->interlaceMode)1804
+ {1805
+ if (m_cliopt.rpuParser(&picField1) > 0)1806
+ goto fail;1807
+ if (m_cliopt.rpuParser(&picField2) > 0)1808
+ goto fail;1809
+ }1810
+ else1811
+ {1812
+ if (m_cliopt.rpuParser(pic_in) > 0)1813
+ goto fail;1814
+ }1815
+ }1816
+ }1817
+1818
+ for (int inputNum = 0; inputNum < inputPicNum; inputNum++)1819
+ {1820
+ x265_picture *picInput = NULL;1821
+ if (inputPicNum == 2)1822
+ picInput = pic_in ? (inputNum ? &picField2 : &picField1) : NULL;1823
+ else1824
+ picInput = pic_in;1825
+1826
+ int numEncoded = api->encoder_encode(m_encoder, &p_nal, &nal, picInput, pic_recon);1827
+1828
+ int idx = (inFrameCount - 1) % m_parent->m_queueSize;1829
+ m_parent->m_picIdxReadCntm_ididx.incr();1830
+ m_parent->m_picReadCntm_id.incr();1831
+ if (m_cliopt.loadLevel && picInput)1832
+ {1833
+ m_parent->m_analysisReadCntm_cliopt.refId.incr();1834
+ m_parent->m_analysisReadm_cliopt.refIdm_lastIdx.incr();1835
+ }1836
+1837
+ if (numEncoded < 0)1838
+ {1839
+ b_ctrl_c = 1;1840
+ m_ret = 4;1841
+ break;1842
+ }1843
+1844
+ if (reconPlay && numEncoded)1845
+ reconPlay->writePicture(*pic_recon);1846
+1847
+ outFrameCount += numEncoded;1848
+1849
+ if (isAbrSave && numEncoded)1850
+ {1851
+ copyInfo(analysisInfo);1852
+ }1853
+1854
+ if (numEncoded && pic_recon && m_cliopt.recon)1855
+ m_cliopt.recon->writePicture(pic_out);1856
+ if (nal)1857
+ {1858
+ m_cliopt.totalbytes += m_cliopt.output->writeFrame(p_nal, nal, pic_out);1859
+ if (pts_queue)1860
+ {1861
+ pts_queue->push(-pic_out.pts);1862
+ if (pts_queue->size() > 2)1863
+ pts_queue->pop();1864
+ }1865
+ }1866
+ m_cliopt.printStatus(outFrameCount);1867
+ }1868
+ }1869
+1870
+ /* Flush the encoder */1871
+ while (!b_ctrl_c)1872
+ {1873
+ int numEncoded = api->encoder_encode(m_encoder, &p_nal, &nal, NULL, pic_recon);1874
+ if (numEncoded < 0)1875
+ {1876
+ m_ret = 4;1877
+ break;1878
+ }1879
+1880
+ if (reconPlay && numEncoded)1881
+ reconPlay->writePicture(*pic_recon);1882
+1883
+ outFrameCount += numEncoded;1884
+ if (isAbrSave && numEncoded)1885
+ {1886
+ copyInfo(analysisInfo);1887
+ }1888
+1889
+ if (numEncoded && pic_recon && m_cliopt.recon)1890
+ m_cliopt.recon->writePicture(pic_out);1891
+ if (nal)1892
+ {1893
+ m_cliopt.totalbytes += m_cliopt.output->writeFrame(p_nal, nal, pic_out);1894
+ if (pts_queue)1895
+ {1896
+ pts_queue->push(-pic_out.pts);1897
+ if (pts_queue->size() > 2)1898
+ pts_queue->pop();1899
+ }1900
+ }1901
+1902
+ m_cliopt.printStatus(outFrameCount);1903
+1904
+ if (!numEncoded)1905
+ break;1906
+ }1907
+1908
+ if (bDolbyVisionRPU)1909
+ {1910
+ if (fgetc(m_cliopt.dolbyVisionRpu) != EOF)1911
+ x265_log(NULL, X265_LOG_WARNING, "Dolby Vision RPU count is greater than frame count in %s\n",1912
+ profileName);1913
+ x265_log(NULL, X265_LOG_INFO, "VES muxing with Dolby Vision RPU file successful in %s\n",1914
+ profileName);1915
+ }1916
+1917
+ /* clear progress report */1918
+ if (m_cliopt.bProgress)1919
+ fprintf(stderr, "%*s\r", 80, " ");1920
+1921
+ fail:1922
+1923
+ delete reconPlay;1924
+1925
+ api->encoder_get_stats(m_encoder, &stats, sizeof(stats));1926
+ if (m_param->csvfn && !b_ctrl_c)1927
+#if ENABLE_LIBVMAF1928
+ api->vmaf_encoder_log(m_encoder, m_cliopt.argCnt, m_cliopt.argString, m_cliopt.param, vmafdata);1929
+#else1930
+ api->encoder_log(m_encoder, m_cliopt.argCnt, m_cliopt.argString);1931
+#endif1932
+ api->encoder_close(m_encoder);1933
+1934
+ int64_t second_largest_pts = 0;1935
+ int64_t largest_pts = 0;1936
+ if (pts_queue && pts_queue->size() >= 2)1937
+ {1938
+ second_largest_pts = -pts_queue->top();1939
+ pts_queue->pop();1940
+ largest_pts = -pts_queue->top();1941
+ pts_queue->pop();1942
+ delete pts_queue;1943
+ pts_queue = NULL;1944
+ }1945
+ m_cliopt.output->closeFile(largest_pts, second_largest_pts);1946
+1947
+ if (b_ctrl_c)1948
+ general_log(m_param, NULL, X265_LOG_INFO, "aborted at input frame %d, output frame %d in %s\n",1949
+ m_cliopt.seek + inFrameCount, stats.encodedPictureCount, profileName);1950
+1951
+ api->param_free(m_param);1952
+1953
+ X265_FREE(errorBuf);1954
+ X265_FREE(rpuPayload);1955
+1956
+ m_threadActive = false;1957
+ m_parent->m_numActiveEncodes.decr();1958
+ }1959
+ }1960
+1961
+ void PassEncoder::destroy()1962
+ {1963
+ stop();1964
+ if (m_reader)1965
+ {1966
+ m_reader->stop();1967
+ delete m_reader;1968
+ }1969
+ else1970
+ {1971
+ m_scaler->stop();1972
+ m_scaler->destroy();1973
+ delete m_scaler;1974
+ }1975
+ }1976
+1977
+ Scaler::Scaler(int threadId, int threadNum, int id, VideoDesc *src, VideoDesc *dst, PassEncoder *parentEnc)1978
+ {1979
+ m_parentEnc = parentEnc;1980
+ m_id = id;1981
+ m_srcFormat = src;1982
+ m_dstFormat = dst;1983
+ m_threadActive = false;1984
+ m_scaleFrameSize = 0;1985
+ m_filterManager = NULL;1986
+ m_threadId = threadId;1987
+ m_threadTotal = threadNum;1988
+1989
+ int csp = dst->m_csp;1990
+ uint32_t pixelbytes = dst->m_inputDepth > 8 ? 2 : 1;1991
+ for (int i = 0; i < x265_cli_cspscsp.planes; i++)1992
+ {1993
+ int w = dst->m_width >> x265_cli_cspscsp.widthi;1994
+ int h = dst->m_height >> x265_cli_cspscsp.heighti;1995
+ m_scalePlanesi = w * h * pixelbytes;1996
+ m_scaleFrameSize += m_scalePlanesi;1997
+ }1998
+1999
+ if (src->m_height != dst->m_height || src->m_width != dst->m_width)2000
+ {2001
+ m_filterManager = new ScalerFilterManager;2002
+ m_filterManager->init(4, m_srcFormat, m_dstFormat);2003
+ }2004
+ }2005
+2006
+ bool Scaler::scalePic(x265_picture * destination, x265_picture * source)2007
+ {2008
+ if (!destination || !source)2009
+ return false;2010
+ x265_param* param = m_parentEnc->m_param;2011
+ int pixelBytes = m_dstFormat->m_inputDepth > 8 ? 2 : 1;2012
+ if (m_srcFormat->m_height != m_dstFormat->m_height || m_srcFormat->m_width != m_dstFormat->m_width)2013
+ {2014
+ void **srcPlane = NULL, **dstPlane = NULL;2015
+ int srcStride3, dstStride3;2016
+ destination->bitDepth = source->bitDepth;2017
+ destination->colorSpace = source->colorSpace;2018
+ destination->pts = source->pts;2019
+ destination->dts = source->dts;2020
+ destination->reorderedPts = source->reorderedPts;2021
+ destination->poc = source->poc;2022
+ destination->userSEI = source->userSEI;2023
+ srcPlane = source->planes;2024
+ dstPlane = destination->planes;2025
+ srcStride0 = source->stride0;2026
+ destination->stride0 = m_dstFormat->m_width * pixelBytes;2027
+ dstStride0 = destination->stride0;2028
+ if (param->internalCsp != X265_CSP_I400)2029
+ {2030
+ srcStride1 = source->stride1;2031
+ srcStride2 = source->stride2;2032
+ destination->stride1 = destination->stride0 >> x265_cli_cspsparam->internalCsp.width1;2033
+ destination->stride2 = destination->stride0 >> x265_cli_cspsparam->internalCsp.width2;2034
+ dstStride1 = destination->stride1;2035
+ dstStride2 = destination->stride2;2036
+ }2037
+ if (m_scaleFrameSize)2038
+ {2039
+ m_filterManager->scale_pic(srcPlane, dstPlane, srcStride, dstStride);2040
+ return true;2041
+ }2042
+ else2043
+ x265_log(param, X265_LOG_INFO, "Empty frame received\n");2044
+ }2045
+ return false;2046
+ }2047
+2048
+ void Scaler::threadMain()2049
+ {2050
+ THREAD_NAME("Scaler", m_id);2051
+2052
+ /* unscaled picture is stored in the last index */2053
+ uint32_t srcId = m_id - 1;2054
+ int QDepth = m_parentEnc->m_parent->m_queueSize;2055
+ while (!m_parentEnc->m_inputOver)2056
+ {2057
+2058
+ uint32_t scaledWritten = m_parentEnc->m_parent->m_picWriteCntm_id.get();2059
+2060
+ if (m_parentEnc->m_cliopt.framesToBeEncoded && scaledWritten >= m_parentEnc->m_cliopt.framesToBeEncoded)2061
+ break;2062
+2063
+ if (m_threadTotal > 1 && (m_threadId != scaledWritten % m_threadTotal))2064
+ {2065
+ continue;2066
+ }2067
+ uint32_t written = m_parentEnc->m_parent->m_picWriteCntsrcId.get();2068
+2069
+ /*If all the input pictures are scaled by the current scale worker thread wait for input pictures*/2070
+ while (m_threadActive && (scaledWritten == written)) {2071
+ written = m_parentEnc->m_parent->m_picWriteCntsrcId.waitForChange(written);2072
+ }2073
+2074
+ if (m_threadActive && scaledWritten < written)2075
+ {2076
+2077
+ int scaledWriteIdx = scaledWritten % QDepth;2078
+ int overWritePicBuffer = scaledWritten / QDepth;2079
+ int read = m_parentEnc->m_parent->m_picIdxReadCntm_idscaledWriteIdx.get();2080
+2081
+ while (overWritePicBuffer && read < overWritePicBuffer)2082
+ {2083
+ read = m_parentEnc->m_parent->m_picIdxReadCntm_idscaledWriteIdx.waitForChange(read);2084
+ }2085
+2086
+ if (!m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx)2087
+ {2088
+ int framesize = 0;2089
+ int planesize3;2090
+ int csp = m_dstFormat->m_csp;2091
+ int stride3;2092
+ stride0 = m_dstFormat->m_width;2093
+ stride1 = stride0 >> x265_cli_cspscsp.width1;2094
+ stride2 = stride0 >> x265_cli_cspscsp.width2;2095
+ for (int i = 0; i < x265_cli_cspscsp.planes; i++)2096
+ {2097
+ uint32_t h = m_dstFormat->m_height >> x265_cli_cspscsp.heighti;2098
+ planesizei = h * stridei;2099
+ framesize += planesizei;2100
+ }2101
+2102
+ m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx = x265_picture_alloc();2103
+ x265_picture_init(m_parentEnc->m_param, m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx);2104
+2105
+ ((x265_picture*)m_parentEnc->m_parent->m_inputPicBufferm_idscaledWritten % QDepth)->framesize = framesize;2106
+ for (int32_t j = 0; j < x265_cli_cspscsp.planes; j++)2107
+ {2108
+ m_parentEnc->m_parent->m_inputPicBufferm_idscaledWritten % QDepth->planesj = X265_MALLOC(char, planesizej);2109
+ }2110
+ }2111
+2112
+ x265_picture *srcPic = m_parentEnc->m_parent->m_inputPicBuffersrcIdscaledWritten % QDepth;2113
+ x265_picture* destPic = m_parentEnc->m_parent->m_inputPicBufferm_idscaledWriteIdx;2114
+2115
+ // Enqueue this picture up with the current encoder so that it will asynchronously encode2116
+ if (!scalePic(destPic, srcPic))2117
+ x265_log(NULL, X265_LOG_ERROR, "Unable to copy scaled input picture to input queue \n");2118
+ else2119
+ m_parentEnc->m_parent->m_picWriteCntm_id.incr();2120
+ m_scaledWriteCnt.incr();2121
+ m_parentEnc->m_parent->m_picIdxReadCntsrcIdscaledWriteIdx.incr();2122
+ }2123
+ if (m_threadTotal > 1)2124
+ {2125
+ written = m_parentEnc->m_parent->m_picWriteCntsrcId.get();2126
+ int totalWrite = written / m_threadTotal;2127
+ if (written % m_threadTotal > m_threadId)2128
+ totalWrite++;2129
+ if (totalWrite == m_scaledWriteCnt.get())2130
+ {2131
+ m_parentEnc->m_parent->m_picWriteCntsrcId.poke();2132
+ m_parentEnc->m_parent->m_picWriteCntm_id.poke();2133
+ break;2134
+ }2135
+ }2136
+ else2137
+ {2138
+ /* Once end of video is reached and all frames are scaled, release wait on picwritecount */2139
+ scaledWritten = m_parentEnc->m_parent->m_picWriteCntm_id.get();2140
+ written = m_parentEnc->m_parent->m_picWriteCntsrcId.get();2141
+ if (written == scaledWritten)2142
+ {2143
+ m_parentEnc->m_parent->m_picWriteCntsrcId.poke();2144
+ m_parentEnc->m_parent->m_picWriteCntm_id.poke();2145
+ break;2146
+ }2147
+ }2148
+2149
+ }2150
+ m_threadActive = false;2151
+ destroy();2152
+ }2153
+2154
+ Reader::Reader(int id, PassEncoder *parentEnc)2155
+ {2156
+ m_parentEnc = parentEnc;2157
+ m_id = id;2158
+ m_input = parentEnc->m_input;2159
+ }2160
+2161
+ void Reader::threadMain()2162
+ {2163
+ THREAD_NAME("Reader", m_id);2164
+2165
+ int QDepth = m_parentEnc->m_parent->m_queueSize;2166
+ x265_picture* src = x265_picture_alloc();2167
+ x265_picture_init(m_parentEnc->m_param, src);2168
+2169
+ while (m_threadActive)2170
+ {2171
+ uint32_t written = m_parentEnc->m_parent->m_picWriteCntm_id.get();2172
+ uint32_t writeIdx = written % QDepth;2173
+ uint32_t read = m_parentEnc->m_parent->m_picIdxReadCntm_idwriteIdx.get();2174
+ uint32_t overWritePicBuffer = written / QDepth;2175
+2176
+ if (m_parentEnc->m_cliopt.framesToBeEncoded && written >= m_parentEnc->m_cliopt.framesToBeEncoded)2177
+ break;2178
+2179
+ while (overWritePicBuffer && read < overWritePicBuffer)2180
+ {2181
+ read = m_parentEnc->m_parent->m_picIdxReadCntm_idwriteIdx.waitForChange(read);2182
+ }2183
+2184
+ x265_picture* dest = m_parentEnc->m_parent->m_inputPicBufferm_idwriteIdx;2185
+ if (m_input->readPicture(*src))2186
+ {2187
+ dest->poc = src->poc;2188
+ dest->pts = src->pts;2189
+ dest->userSEI = src->userSEI;2190
+ dest->bitDepth = src->bitDepth;2191
+ dest->framesize = src->framesize;2192
+ dest->height = src->height;2193
+ dest->width = src->width;2194
+ dest->colorSpace = src->colorSpace;2195
+ dest->userSEI = src->userSEI;2196
+ dest->rpu.payload = src->rpu.payload;2197
+ dest->picStruct = src->picStruct;2198
+ dest->stride0 = src->stride0;2199
+ dest->stride1 = src->stride1;2200
+ dest->stride2 = src->stride2;2201
+2202
+ if (!dest->planes0)2203
+ dest->planes0 = X265_MALLOC(char, dest->framesize);2204
+2205
+ memcpy(dest->planes0, src->planes0, src->framesize * sizeof(char));2206
+ dest->planes1 = (char*)dest->planes0 + src->stride0 * src->height;2207
+ dest->planes2 = (char*)dest->planes1 + src->stride1 * (src->height >> x265_cli_cspssrc->colorSpace.height1);2208
+ m_parentEnc->m_parent->m_picWriteCntm_id.incr();2209
+ }2210
+ else2211
+ {2212
+ m_threadActive = false;2213
+ m_parentEnc->m_inputOver = true;2214
+ m_parentEnc->m_parent->m_picWriteCntm_id.poke();2215
+ }2216
+ }2217
+ x265_picture_free(src);2218
+ }2219
+}2220
x265_3.5.tar.gz/source/abrEncApp.h -> x265_3.6.tar.gz/source/abrEncApp.h
Changed
9
1
2
FILE* m_qpfile;3
FILE* m_zoneFile;4
FILE* m_dolbyVisionRpu;/* File containing Dolby Vision BL RPU metadata */5
+ FILE* m_scenecutAwareQpConfig;6
7
int m_ret;8
9
x265_3.5.tar.gz/source/cmake/FindNeon.cmake -> x265_3.6.tar.gz/source/cmake/FindNeon.cmake
Changed
27
1
2
include(FindPackageHandleStandardArgs)3
4
# Check the version of neon supported by the ARM CPU5
-execute_process(COMMAND cat /proc/cpuinfo | grep Features | grep neon6
- OUTPUT_VARIABLE neon_version7
- ERROR_QUIET8
- OUTPUT_STRIP_TRAILING_WHITESPACE)9
+if(APPLE)10
+ execute_process(COMMAND sysctl -a11
+ COMMAND grep "hw.optional.neon: 1"12
+ OUTPUT_VARIABLE neon_version13
+ ERROR_QUIET14
+ OUTPUT_STRIP_TRAILING_WHITESPACE)15
+else()16
+ execute_process(COMMAND cat /proc/cpuinfo17
+ COMMAND grep Features18
+ COMMAND grep neon19
+ OUTPUT_VARIABLE neon_version20
+ ERROR_QUIET21
+ OUTPUT_STRIP_TRAILING_WHITESPACE)22
+endif()23
+24
if(neon_version)25
set(CPU_HAS_NEON 1)26
endif()27
x265_3.6.tar.gz/source/cmake/FindSVE.cmake
Added
23
1
2
+include(FindPackageHandleStandardArgs)3
+4
+# Check the version of SVE supported by the ARM CPU5
+if(APPLE)6
+ execute_process(COMMAND sysctl -a7
+ COMMAND grep "hw.optional.sve: 1"8
+ OUTPUT_VARIABLE sve_version9
+ ERROR_QUIET10
+ OUTPUT_STRIP_TRAILING_WHITESPACE)11
+else()12
+ execute_process(COMMAND cat /proc/cpuinfo13
+ COMMAND grep Features14
+ COMMAND grep -e "sve$" -e "sve:space:"15
+ OUTPUT_VARIABLE sve_version16
+ ERROR_QUIET17
+ OUTPUT_STRIP_TRAILING_WHITESPACE)18
+endif()19
+20
+if(sve_version)21
+ set(CPU_HAS_SVE 1)22
+endif()23
x265_3.6.tar.gz/source/cmake/FindSVE2.cmake
Added
24
1
2
+include(FindPackageHandleStandardArgs)3
+4
+# Check the version of SVE2 supported by the ARM CPU5
+if(APPLE)6
+ execute_process(COMMAND sysctl -a7
+ COMMAND grep "hw.optional.sve2: 1"8
+ OUTPUT_VARIABLE sve2_version9
+ ERROR_QUIET10
+ OUTPUT_STRIP_TRAILING_WHITESPACE)11
+else()12
+ execute_process(COMMAND cat /proc/cpuinfo13
+ COMMAND grep Features14
+ COMMAND grep sve215
+ OUTPUT_VARIABLE sve2_version16
+ ERROR_QUIET17
+ OUTPUT_STRIP_TRAILING_WHITESPACE)18
+endif()19
+20
+if(sve2_version)21
+ set(CPU_HAS_SVE 1)22
+ set(CPU_HAS_SVE2 1)23
+endif()24
x265_3.5.tar.gz/source/common/CMakeLists.txt -> x265_3.6.tar.gz/source/common/CMakeLists.txt
Changed
76
1
2
endif(ENABLE_ASSEMBLY AND X86)3
4
if(ENABLE_ASSEMBLY AND (ARM OR CROSS_COMPILE_ARM))5
- if(ARM64)6
- if(GCC AND (CMAKE_CXX_FLAGS_RELEASE MATCHES "-O3"))7
- message(STATUS "Detected CXX compiler using -O3 optimization level")8
- add_definitions(-DAUTO_VECTORIZE=1)9
- endif()10
- set(C_SRCS asm-primitives.cpp pixel.h ipfilter8.h)11
-12
- # add ARM assembly/intrinsic files here13
- set(A_SRCS asm.S mc-a.S sad-a.S pixel-util.S ipfilter8.S)14
- set(VEC_PRIMITIVES)15
+ set(C_SRCS asm-primitives.cpp pixel.h mc.h ipfilter8.h blockcopy8.h dct8.h loopfilter.h)16
17
- set(ARM_ASMS "${A_SRCS}" CACHE INTERNAL "ARM Assembly Sources")18
- foreach(SRC ${C_SRCS})19
- set(ASM_PRIMITIVES ${ASM_PRIMITIVES} aarch64/${SRC})20
- endforeach()21
- else()22
- set(C_SRCS asm-primitives.cpp pixel.h mc.h ipfilter8.h blockcopy8.h dct8.h loopfilter.h)23
+ # add ARM assembly/intrinsic files here24
+ set(A_SRCS asm.S cpu-a.S mc-a.S sad-a.S pixel-util.S ssd-a.S blockcopy8.S ipfilter8.S dct-a.S)25
+ set(VEC_PRIMITIVES)26
27
- # add ARM assembly/intrinsic files here28
- set(A_SRCS asm.S cpu-a.S mc-a.S sad-a.S pixel-util.S ssd-a.S blockcopy8.S ipfilter8.S dct-a.S)29
- set(VEC_PRIMITIVES)30
+ set(ARM_ASMS "${A_SRCS}" CACHE INTERNAL "ARM Assembly Sources")31
+ foreach(SRC ${C_SRCS})32
+ set(ASM_PRIMITIVES ${ASM_PRIMITIVES} arm/${SRC})33
+ endforeach()34
+ source_group(Assembly FILES ${ASM_PRIMITIVES})35
+endif(ENABLE_ASSEMBLY AND (ARM OR CROSS_COMPILE_ARM))36
37
- set(ARM_ASMS "${A_SRCS}" CACHE INTERNAL "ARM Assembly Sources")38
- foreach(SRC ${C_SRCS})39
- set(ASM_PRIMITIVES ${ASM_PRIMITIVES} arm/${SRC})40
- endforeach()41
+if(ENABLE_ASSEMBLY AND (ARM64 OR CROSS_COMPILE_ARM64))42
+ if(GCC AND (CMAKE_CXX_FLAGS_RELEASE MATCHES "-O3"))43
+ message(STATUS "Detected CXX compiler using -O3 optimization level")44
+ add_definitions(-DAUTO_VECTORIZE=1)45
endif()46
+47
+ set(C_SRCS asm-primitives.cpp pixel-prim.h pixel-prim.cpp filter-prim.h filter-prim.cpp dct-prim.h dct-prim.cpp loopfilter-prim.cpp loopfilter-prim.h intrapred-prim.cpp arm64-utils.cpp arm64-utils.h fun-decls.h)48
+ enable_language(ASM)49
+50
+ # add ARM assembly/intrinsic files here51
+ set(A_SRCS asm.S mc-a.S mc-a-common.S sad-a.S sad-a-common.S pixel-util.S pixel-util-common.S p2s.S p2s-common.S ipfilter.S ipfilter-common.S blockcopy8.S blockcopy8-common.S ssd-a.S ssd-a-common.S)52
+ set(A_SRCS_SVE asm-sve.S blockcopy8-sve.S p2s-sve.S pixel-util-sve.S ssd-a-sve.S)53
+ set(A_SRCS_SVE2 mc-a-sve2.S sad-a-sve2.S pixel-util-sve2.S ipfilter-sve2.S ssd-a-sve2.S)54
+ set(VEC_PRIMITIVES)55
+56
+ set(ARM_ASMS "${A_SRCS}" CACHE INTERNAL "ARM Assembly Sources")57
+ set(ARM_ASMS_SVE "${A_SRCS_SVE}" CACHE INTERNAL "ARM Assembly Sources that use SVE instruction set")58
+ set(ARM_ASMS_SVE2 "${A_SRCS_SVE2}" CACHE INTERNAL "ARM Assembly Sources that use SVE2 instruction set")59
+ foreach(SRC ${C_SRCS})60
+ set(ASM_PRIMITIVES ${ASM_PRIMITIVES} aarch64/${SRC})61
+ endforeach()62
source_group(Assembly FILES ${ASM_PRIMITIVES})63
-endif(ENABLE_ASSEMBLY AND (ARM OR CROSS_COMPILE_ARM))64
+endif(ENABLE_ASSEMBLY AND (ARM64 OR CROSS_COMPILE_ARM64))65
66
if(POWER)67
set_source_files_properties(version.cpp PROPERTIES COMPILE_FLAGS -DX265_VERSION=${X265_VERSION})68
69
scalinglist.cpp scalinglist.h70
quant.cpp quant.h contexts.h71
deblock.cpp deblock.h72
- scaler.cpp scaler.h)73
+ scaler.cpp scaler.h74
+ ringmem.cpp ringmem.h75
+ temporalfilter.cpp temporalfilter.h)76
x265_3.6.tar.gz/source/common/aarch64/arm64-utils.cpp
Added
302
1
2
+#include "common.h"3
+#include "x265.h"4
+#include "arm64-utils.h"5
+#include <arm_neon.h>6
+7
+#define COPY_16(d,s) *(uint8x16_t *)(d) = *(uint8x16_t *)(s)8
+namespace X265_NS9
+{10
+11
+12
+13
+void transpose8x8(uint8_t *dst, const uint8_t *src, intptr_t dstride, intptr_t sstride)14
+{15
+ uint8x8_t a0, a1, a2, a3, a4, a5, a6, a7;16
+ uint8x8_t b0, b1, b2, b3, b4, b5, b6, b7;17
+18
+ a0 = *(uint8x8_t *)(src + 0 * sstride);19
+ a1 = *(uint8x8_t *)(src + 1 * sstride);20
+ a2 = *(uint8x8_t *)(src + 2 * sstride);21
+ a3 = *(uint8x8_t *)(src + 3 * sstride);22
+ a4 = *(uint8x8_t *)(src + 4 * sstride);23
+ a5 = *(uint8x8_t *)(src + 5 * sstride);24
+ a6 = *(uint8x8_t *)(src + 6 * sstride);25
+ a7 = *(uint8x8_t *)(src + 7 * sstride);26
+27
+ b0 = vtrn1_u32(a0, a4);28
+ b1 = vtrn1_u32(a1, a5);29
+ b2 = vtrn1_u32(a2, a6);30
+ b3 = vtrn1_u32(a3, a7);31
+ b4 = vtrn2_u32(a0, a4);32
+ b5 = vtrn2_u32(a1, a5);33
+ b6 = vtrn2_u32(a2, a6);34
+ b7 = vtrn2_u32(a3, a7);35
+36
+ a0 = vtrn1_u16(b0, b2);37
+ a1 = vtrn1_u16(b1, b3);38
+ a2 = vtrn2_u16(b0, b2);39
+ a3 = vtrn2_u16(b1, b3);40
+ a4 = vtrn1_u16(b4, b6);41
+ a5 = vtrn1_u16(b5, b7);42
+ a6 = vtrn2_u16(b4, b6);43
+ a7 = vtrn2_u16(b5, b7);44
+45
+ b0 = vtrn1_u8(a0, a1);46
+ b1 = vtrn2_u8(a0, a1);47
+ b2 = vtrn1_u8(a2, a3);48
+ b3 = vtrn2_u8(a2, a3);49
+ b4 = vtrn1_u8(a4, a5);50
+ b5 = vtrn2_u8(a4, a5);51
+ b6 = vtrn1_u8(a6, a7);52
+ b7 = vtrn2_u8(a6, a7);53
+54
+ *(uint8x8_t *)(dst + 0 * dstride) = b0;55
+ *(uint8x8_t *)(dst + 1 * dstride) = b1;56
+ *(uint8x8_t *)(dst + 2 * dstride) = b2;57
+ *(uint8x8_t *)(dst + 3 * dstride) = b3;58
+ *(uint8x8_t *)(dst + 4 * dstride) = b4;59
+ *(uint8x8_t *)(dst + 5 * dstride) = b5;60
+ *(uint8x8_t *)(dst + 6 * dstride) = b6;61
+ *(uint8x8_t *)(dst + 7 * dstride) = b7;62
+}63
+64
+65
+66
+67
+68
+69
+void transpose16x16(uint8_t *dst, const uint8_t *src, intptr_t dstride, intptr_t sstride)70
+{71
+ uint16x8_t a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, aA, aB, aC, aD, aE, aF;72
+ uint16x8_t b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, bA, bB, bC, bD, bE, bF;73
+ uint16x8_t c0, c1, c2, c3, c4, c5, c6, c7, c8, c9, cA, cB, cC, cD, cE, cF;74
+ uint16x8_t d0, d1, d2, d3, d4, d5, d6, d7, d8, d9, dA, dB, dC, dD, dE, dF;75
+76
+ a0 = *(uint16x8_t *)(src + 0 * sstride);77
+ a1 = *(uint16x8_t *)(src + 1 * sstride);78
+ a2 = *(uint16x8_t *)(src + 2 * sstride);79
+ a3 = *(uint16x8_t *)(src + 3 * sstride);80
+ a4 = *(uint16x8_t *)(src + 4 * sstride);81
+ a5 = *(uint16x8_t *)(src + 5 * sstride);82
+ a6 = *(uint16x8_t *)(src + 6 * sstride);83
+ a7 = *(uint16x8_t *)(src + 7 * sstride);84
+ a8 = *(uint16x8_t *)(src + 8 * sstride);85
+ a9 = *(uint16x8_t *)(src + 9 * sstride);86
+ aA = *(uint16x8_t *)(src + 10 * sstride);87
+ aB = *(uint16x8_t *)(src + 11 * sstride);88
+ aC = *(uint16x8_t *)(src + 12 * sstride);89
+ aD = *(uint16x8_t *)(src + 13 * sstride);90
+ aE = *(uint16x8_t *)(src + 14 * sstride);91
+ aF = *(uint16x8_t *)(src + 15 * sstride);92
+93
+ b0 = vtrn1q_u64(a0, a8);94
+ b1 = vtrn1q_u64(a1, a9);95
+ b2 = vtrn1q_u64(a2, aA);96
+ b3 = vtrn1q_u64(a3, aB);97
+ b4 = vtrn1q_u64(a4, aC);98
+ b5 = vtrn1q_u64(a5, aD);99
+ b6 = vtrn1q_u64(a6, aE);100
+ b7 = vtrn1q_u64(a7, aF);101
+ b8 = vtrn2q_u64(a0, a8);102
+ b9 = vtrn2q_u64(a1, a9);103
+ bA = vtrn2q_u64(a2, aA);104
+ bB = vtrn2q_u64(a3, aB);105
+ bC = vtrn2q_u64(a4, aC);106
+ bD = vtrn2q_u64(a5, aD);107
+ bE = vtrn2q_u64(a6, aE);108
+ bF = vtrn2q_u64(a7, aF);109
+110
+ c0 = vtrn1q_u32(b0, b4);111
+ c1 = vtrn1q_u32(b1, b5);112
+ c2 = vtrn1q_u32(b2, b6);113
+ c3 = vtrn1q_u32(b3, b7);114
+ c4 = vtrn2q_u32(b0, b4);115
+ c5 = vtrn2q_u32(b1, b5);116
+ c6 = vtrn2q_u32(b2, b6);117
+ c7 = vtrn2q_u32(b3, b7);118
+ c8 = vtrn1q_u32(b8, bC);119
+ c9 = vtrn1q_u32(b9, bD);120
+ cA = vtrn1q_u32(bA, bE);121
+ cB = vtrn1q_u32(bB, bF);122
+ cC = vtrn2q_u32(b8, bC);123
+ cD = vtrn2q_u32(b9, bD);124
+ cE = vtrn2q_u32(bA, bE);125
+ cF = vtrn2q_u32(bB, bF);126
+127
+ d0 = vtrn1q_u16(c0, c2);128
+ d1 = vtrn1q_u16(c1, c3);129
+ d2 = vtrn2q_u16(c0, c2);130
+ d3 = vtrn2q_u16(c1, c3);131
+ d4 = vtrn1q_u16(c4, c6);132
+ d5 = vtrn1q_u16(c5, c7);133
+ d6 = vtrn2q_u16(c4, c6);134
+ d7 = vtrn2q_u16(c5, c7);135
+ d8 = vtrn1q_u16(c8, cA);136
+ d9 = vtrn1q_u16(c9, cB);137
+ dA = vtrn2q_u16(c8, cA);138
+ dB = vtrn2q_u16(c9, cB);139
+ dC = vtrn1q_u16(cC, cE);140
+ dD = vtrn1q_u16(cD, cF);141
+ dE = vtrn2q_u16(cC, cE);142
+ dF = vtrn2q_u16(cD, cF);143
+144
+ *(uint16x8_t *)(dst + 0 * dstride) = vtrn1q_u8(d0, d1);145
+ *(uint16x8_t *)(dst + 1 * dstride) = vtrn2q_u8(d0, d1);146
+ *(uint16x8_t *)(dst + 2 * dstride) = vtrn1q_u8(d2, d3);147
+ *(uint16x8_t *)(dst + 3 * dstride) = vtrn2q_u8(d2, d3);148
+ *(uint16x8_t *)(dst + 4 * dstride) = vtrn1q_u8(d4, d5);149
+ *(uint16x8_t *)(dst + 5 * dstride) = vtrn2q_u8(d4, d5);150
+ *(uint16x8_t *)(dst + 6 * dstride) = vtrn1q_u8(d6, d7);151
+ *(uint16x8_t *)(dst + 7 * dstride) = vtrn2q_u8(d6, d7);152
+ *(uint16x8_t *)(dst + 8 * dstride) = vtrn1q_u8(d8, d9);153
+ *(uint16x8_t *)(dst + 9 * dstride) = vtrn2q_u8(d8, d9);154
+ *(uint16x8_t *)(dst + 10 * dstride) = vtrn1q_u8(dA, dB);155
+ *(uint16x8_t *)(dst + 11 * dstride) = vtrn2q_u8(dA, dB);156
+ *(uint16x8_t *)(dst + 12 * dstride) = vtrn1q_u8(dC, dD);157
+ *(uint16x8_t *)(dst + 13 * dstride) = vtrn2q_u8(dC, dD);158
+ *(uint16x8_t *)(dst + 14 * dstride) = vtrn1q_u8(dE, dF);159
+ *(uint16x8_t *)(dst + 15 * dstride) = vtrn2q_u8(dE, dF);160
+161
+162
+}163
+164
+165
+void transpose32x32(uint8_t *dst, const uint8_t *src, intptr_t dstride, intptr_t sstride)166
+{167
+ //assumption: there is no partial overlap168
+ transpose16x16(dst, src, dstride, sstride);169
+ transpose16x16(dst + 16 * dstride + 16, src + 16 * sstride + 16, dstride, sstride);170
+ if (dst == src)171
+ {172
+ uint8_t tmp16 * 16 __attribute__((aligned(64)));173
+ transpose16x16(tmp, src + 16, 16, sstride);174
+ transpose16x16(dst + 16, src + 16 * sstride, dstride, sstride);175
+ for (int i = 0; i < 16; i++)176
+ {177
+ COPY_16(dst + (16 + i)*dstride, tmp + 16 * i);178
+ }179
+ }180
+ else181
+ {182
+ transpose16x16(dst + 16 * dstride, src + 16, dstride, sstride);183
+ transpose16x16(dst + 16, src + 16 * sstride, dstride, sstride);184
+ }185
+186
+}187
+188
+189
+190
+void transpose8x8(uint16_t *dst, const uint16_t *src, intptr_t dstride, intptr_t sstride)191
+{192
+ uint16x8_t a0, a1, a2, a3, a4, a5, a6, a7;193
+ uint16x8_t b0, b1, b2, b3, b4, b5, b6, b7;194
+195
+ a0 = *(uint16x8_t *)(src + 0 * sstride);196
+ a1 = *(uint16x8_t *)(src + 1 * sstride);197
+ a2 = *(uint16x8_t *)(src + 2 * sstride);198
+ a3 = *(uint16x8_t *)(src + 3 * sstride);199
+ a4 = *(uint16x8_t *)(src + 4 * sstride);200
+ a5 = *(uint16x8_t *)(src + 5 * sstride);201
+ a6 = *(uint16x8_t *)(src + 6 * sstride);202
+ a7 = *(uint16x8_t *)(src + 7 * sstride);203
+204
+ b0 = vtrn1q_u64(a0, a4);205
+ b1 = vtrn1q_u64(a1, a5);206
+ b2 = vtrn1q_u64(a2, a6);207
+ b3 = vtrn1q_u64(a3, a7);208
+ b4 = vtrn2q_u64(a0, a4);209
+ b5 = vtrn2q_u64(a1, a5);210
+ b6 = vtrn2q_u64(a2, a6);211
+ b7 = vtrn2q_u64(a3, a7);212
+213
+ a0 = vtrn1q_u32(b0, b2);214
+ a1 = vtrn1q_u32(b1, b3);215
+ a2 = vtrn2q_u32(b0, b2);216
+ a3 = vtrn2q_u32(b1, b3);217
+ a4 = vtrn1q_u32(b4, b6);218
+ a5 = vtrn1q_u32(b5, b7);219
+ a6 = vtrn2q_u32(b4, b6);220
+ a7 = vtrn2q_u32(b5, b7);221
+222
+ b0 = vtrn1q_u16(a0, a1);223
+ b1 = vtrn2q_u16(a0, a1);224
+ b2 = vtrn1q_u16(a2, a3);225
+ b3 = vtrn2q_u16(a2, a3);226
+ b4 = vtrn1q_u16(a4, a5);227
+ b5 = vtrn2q_u16(a4, a5);228
+ b6 = vtrn1q_u16(a6, a7);229
+ b7 = vtrn2q_u16(a6, a7);230
+231
+ *(uint16x8_t *)(dst + 0 * dstride) = b0;232
+ *(uint16x8_t *)(dst + 1 * dstride) = b1;233
+ *(uint16x8_t *)(dst + 2 * dstride) = b2;234
+ *(uint16x8_t *)(dst + 3 * dstride) = b3;235
+ *(uint16x8_t *)(dst + 4 * dstride) = b4;236
+ *(uint16x8_t *)(dst + 5 * dstride) = b5;237
+ *(uint16x8_t *)(dst + 6 * dstride) = b6;238
+ *(uint16x8_t *)(dst + 7 * dstride) = b7;239
+}240
+241
+void transpose16x16(uint16_t *dst, const uint16_t *src, intptr_t dstride, intptr_t sstride)242
+{243
+ //assumption: there is no partial overlap244
+ transpose8x8(dst, src, dstride, sstride);245
+ transpose8x8(dst + 8 * dstride + 8, src + 8 * sstride + 8, dstride, sstride);246
+247
+ if (dst == src)248
+ {249
+ uint16_t tmp8 * 8;250
+ transpose8x8(tmp, src + 8, 8, sstride);251
+ transpose8x8(dst + 8, src + 8 * sstride, dstride, sstride);252
+ for (int i = 0; i < 8; i++)253
+ {254
+ COPY_16(dst + (8 + i)*dstride, tmp + 8 * i);255
+ }256
+ }257
+ else258
+ {259
+ transpose8x8(dst + 8 * dstride, src + 8, dstride, sstride);260
+ transpose8x8(dst + 8, src + 8 * sstride, dstride, sstride);261
+ }262
+263
+}264
+265
+266
+267
+void transpose32x32(uint16_t *dst, const uint16_t *src, intptr_t dstride, intptr_t sstride)268
+{269
+ //assumption: there is no partial overlap270
+ for (int i = 0; i < 4; i++)271
+ {272
+ transpose8x8(dst + i * 8 * (1 + dstride), src + i * 8 * (1 + sstride), dstride, sstride);273
+ for (int j = i + 1; j < 4; j++)274
+ {275
+ if (dst == src)276
+ {277
+ uint16_t tmp8 * 8 __attribute__((aligned(64)));278
+ transpose8x8(tmp, src + 8 * i + 8 * j * sstride, 8, sstride);279
+ transpose8x8(dst + 8 * i + 8 * j * dstride, src + 8 * j + 8 * i * sstride, dstride, sstride);280
+ for (int k = 0; k < 8; k++)281
+ {282
+ COPY_16(dst + 8 * j + (8 * i + k)*dstride, tmp + 8 * k);283
+ }284
+ }285
+ else286
+ {287
+ transpose8x8(dst + 8 * (j + i * dstride), src + 8 * (i + j * sstride), dstride, sstride);288
+ transpose8x8(dst + 8 * (i + j * dstride), src + 8 * (j + i * sstride), dstride, sstride);289
+ }290
+291
+ }292
+ }293
+}294
+295
+296
+297
+298
+}299
+300
+301
+302
x265_3.6.tar.gz/source/common/aarch64/arm64-utils.h
Added
17
1
2
+#ifndef __ARM64_UTILS_H__3
+#define __ARM64_UTILS_H__4
+5
+6
+namespace X265_NS7
+{8
+void transpose8x8(uint8_t *dst, const uint8_t *src, intptr_t dstride, intptr_t sstride);9
+void transpose16x16(uint8_t *dst, const uint8_t *src, intptr_t dstride, intptr_t sstride);10
+void transpose32x32(uint8_t *dst, const uint8_t *src, intptr_t dstride, intptr_t sstride);11
+void transpose8x8(uint16_t *dst, const uint16_t *src, intptr_t dstride, intptr_t sstride);12
+void transpose16x16(uint16_t *dst, const uint16_t *src, intptr_t dstride, intptr_t sstride);13
+void transpose32x32(uint16_t *dst, const uint16_t *src, intptr_t dstride, intptr_t sstride);14
+}15
+16
+#endif17
x265_3.5.tar.gz/source/common/aarch64/asm-primitives.cpp -> x265_3.6.tar.gz/source/common/aarch64/asm-primitives.cpp
Changed
2102
1
2
*3
* Authors: Hongbin Liu <liuhongbin1@huawei.com>4
* Yimeng Su <yimeng.su@huawei.com>5
+ * Sebastian Pop <spop@amazon.com>6
*7
* This program is free software; you can redistribute it and/or modify8
* it under the terms of the GNU General Public License as published by9
10
* For more information, contact us at license @ x265.com.11
*****************************************************************************/12
13
+14
#include "common.h"15
#include "primitives.h"16
#include "x265.h"17
#include "cpu.h"18
19
+extern "C" {20
+#include "fun-decls.h"21
+}22
+23
+#define ALL_LUMA_TU_TYPED(prim, fncdef, fname, cpu) \24
+ p.cuBLOCK_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \25
+ p.cuBLOCK_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \26
+ p.cuBLOCK_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \27
+ p.cuBLOCK_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \28
+ p.cuBLOCK_64x64.prim = fncdef PFX(fname ## _64x64_ ## cpu)29
+#define LUMA_TU_TYPED_NEON(prim, fncdef, fname) \30
+ p.cuBLOCK_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \31
+ p.cuBLOCK_8x8.prim = fncdef PFX(fname ## _8x8_ ## neon); \32
+ p.cuBLOCK_16x16.prim = fncdef PFX(fname ## _16x16_ ## neon); \33
+ p.cuBLOCK_64x64.prim = fncdef PFX(fname ## _64x64_ ## neon)34
+#define LUMA_TU_TYPED_CAN_USE_SVE(prim, fncdef, fname) \35
+ p.cuBLOCK_32x32.prim = fncdef PFX(fname ## _32x32_ ## sve)36
+#define ALL_LUMA_TU(prim, fname, cpu) ALL_LUMA_TU_TYPED(prim, , fname, cpu)37
+#define LUMA_TU_NEON(prim, fname) LUMA_TU_TYPED_NEON(prim, , fname)38
+#define LUMA_TU_CAN_USE_SVE(prim, fname) LUMA_TU_TYPED_CAN_USE_SVE(prim, , fname)39
+40
+#define ALL_LUMA_PU_TYPED(prim, fncdef, fname, cpu) \41
+ p.puLUMA_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \42
+ p.puLUMA_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \43
+ p.puLUMA_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \44
+ p.puLUMA_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \45
+ p.puLUMA_64x64.prim = fncdef PFX(fname ## _64x64_ ## cpu); \46
+ p.puLUMA_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \47
+ p.puLUMA_4x8.prim = fncdef PFX(fname ## _4x8_ ## cpu); \48
+ p.puLUMA_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \49
+ p.puLUMA_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \50
+ p.puLUMA_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \51
+ p.puLUMA_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \52
+ p.puLUMA_64x32.prim = fncdef PFX(fname ## _64x32_ ## cpu); \53
+ p.puLUMA_32x64.prim = fncdef PFX(fname ## _32x64_ ## cpu); \54
+ p.puLUMA_16x12.prim = fncdef PFX(fname ## _16x12_ ## cpu); \55
+ p.puLUMA_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \56
+ p.puLUMA_16x4.prim = fncdef PFX(fname ## _16x4_ ## cpu); \57
+ p.puLUMA_4x16.prim = fncdef PFX(fname ## _4x16_ ## cpu); \58
+ p.puLUMA_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \59
+ p.puLUMA_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \60
+ p.puLUMA_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \61
+ p.puLUMA_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu); \62
+ p.puLUMA_64x48.prim = fncdef PFX(fname ## _64x48_ ## cpu); \63
+ p.puLUMA_48x64.prim = fncdef PFX(fname ## _48x64_ ## cpu); \64
+ p.puLUMA_64x16.prim = fncdef PFX(fname ## _64x16_ ## cpu); \65
+ p.puLUMA_16x64.prim = fncdef PFX(fname ## _16x64_ ## cpu)66
+#define LUMA_PU_TYPED_MULTIPLE_ARCHS_1(prim, fncdef, fname, cpu) \67
+ p.puLUMA_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \68
+ p.puLUMA_4x8.prim = fncdef PFX(fname ## _4x8_ ## cpu); \69
+ p.puLUMA_4x16.prim = fncdef PFX(fname ## _4x16_ ## cpu)70
+#define LUMA_PU_TYPED_MULTIPLE_ARCHS_2(prim, fncdef, fname, cpu) \71
+ p.puLUMA_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \72
+ p.puLUMA_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \73
+ p.puLUMA_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \74
+ p.puLUMA_64x64.prim = fncdef PFX(fname ## _64x64_ ## cpu); \75
+ p.puLUMA_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \76
+ p.puLUMA_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \77
+ p.puLUMA_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \78
+ p.puLUMA_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \79
+ p.puLUMA_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \80
+ p.puLUMA_64x32.prim = fncdef PFX(fname ## _64x32_ ## cpu); \81
+ p.puLUMA_32x64.prim = fncdef PFX(fname ## _32x64_ ## cpu); \82
+ p.puLUMA_16x12.prim = fncdef PFX(fname ## _16x12_ ## cpu); \83
+ p.puLUMA_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \84
+ p.puLUMA_16x4.prim = fncdef PFX(fname ## _16x4_ ## cpu); \85
+ p.puLUMA_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \86
+ p.puLUMA_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \87
+ p.puLUMA_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \88
+ p.puLUMA_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu); \89
+ p.puLUMA_64x48.prim = fncdef PFX(fname ## _64x48_ ## cpu); \90
+ p.puLUMA_48x64.prim = fncdef PFX(fname ## _48x64_ ## cpu); \91
+ p.puLUMA_64x16.prim = fncdef PFX(fname ## _64x16_ ## cpu); \92
+ p.puLUMA_16x64.prim = fncdef PFX(fname ## _16x64_ ## cpu)93
+#define LUMA_PU_TYPED_NEON_1(prim, fncdef, fname) \94
+ p.puLUMA_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \95
+ p.puLUMA_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \96
+ p.puLUMA_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon); \97
+ p.puLUMA_12x16.prim = fncdef PFX(fname ## _12x16_ ## neon); \98
+ p.puLUMA_8x8.prim = fncdef PFX(fname ## _8x8_ ## neon); \99
+ p.puLUMA_16x16.prim = fncdef PFX(fname ## _16x16_ ## neon); \100
+ p.puLUMA_8x4.prim = fncdef PFX(fname ## _8x4_ ## neon); \101
+ p.puLUMA_16x8.prim = fncdef PFX(fname ## _16x8_ ## neon); \102
+ p.puLUMA_8x16.prim = fncdef PFX(fname ## _8x16_ ## neon); \103
+ p.puLUMA_16x12.prim = fncdef PFX(fname ## _16x12_ ## neon); \104
+ p.puLUMA_16x32.prim = fncdef PFX(fname ## _16x32_ ## neon); \105
+ p.puLUMA_16x4.prim = fncdef PFX(fname ## _16x4_ ## neon); \106
+ p.puLUMA_24x32.prim = fncdef PFX(fname ## _24x32_ ## neon); \107
+ p.puLUMA_8x32.prim = fncdef PFX(fname ## _8x32_ ## neon); \108
+ p.puLUMA_48x64.prim = fncdef PFX(fname ## _48x64_ ## neon); \109
+ p.puLUMA_16x64.prim = fncdef PFX(fname ## _16x64_ ## neon)110
+#define LUMA_PU_TYPED_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, fncdef, fname) \111
+ p.puLUMA_32x32.prim = fncdef PFX(fname ## _32x32_ ## sve); \112
+ p.puLUMA_64x64.prim = fncdef PFX(fname ## _64x64_ ## sve); \113
+ p.puLUMA_32x16.prim = fncdef PFX(fname ## _32x16_ ## sve); \114
+ p.puLUMA_64x32.prim = fncdef PFX(fname ## _64x32_ ## sve); \115
+ p.puLUMA_32x64.prim = fncdef PFX(fname ## _32x64_ ## sve); \116
+ p.puLUMA_32x24.prim = fncdef PFX(fname ## _32x24_ ## sve); \117
+ p.puLUMA_32x8.prim = fncdef PFX(fname ## _32x8_ ## sve); \118
+ p.puLUMA_64x48.prim = fncdef PFX(fname ## _64x48_ ## sve); \119
+ p.puLUMA_64x16.prim = fncdef PFX(fname ## _64x16_ ## sve)120
+#define LUMA_PU_TYPED_NEON_2(prim, fncdef, fname) \121
+ p.puLUMA_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \122
+ p.puLUMA_8x4.prim = fncdef PFX(fname ## _8x4_ ## neon); \123
+ p.puLUMA_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \124
+ p.puLUMA_8x8.prim = fncdef PFX(fname ## _8x8_ ## neon); \125
+ p.puLUMA_16x8.prim = fncdef PFX(fname ## _16x8_ ## neon); \126
+ p.puLUMA_8x16.prim = fncdef PFX(fname ## _8x16_ ## neon); \127
+ p.puLUMA_16x16.prim = fncdef PFX(fname ## _16x16_ ## neon); \128
+ p.puLUMA_16x32.prim = fncdef PFX(fname ## _16x32_ ## neon); \129
+ p.puLUMA_16x12.prim = fncdef PFX(fname ## _16x12_ ## neon); \130
+ p.puLUMA_16x4.prim = fncdef PFX(fname ## _16x4_ ## neon); \131
+ p.puLUMA_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon); \132
+ p.puLUMA_8x32.prim = fncdef PFX(fname ## _8x32_ ## neon); \133
+ p.puLUMA_16x64.prim = fncdef PFX(fname ## _16x64_ ## neon)134
+#define LUMA_PU_TYPED_MULTIPLE_ARCHS_3(prim, fncdef, fname, cpu) \135
+ p.puLUMA_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \136
+ p.puLUMA_64x64.prim = fncdef PFX(fname ## _64x64_ ## cpu); \137
+ p.puLUMA_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \138
+ p.puLUMA_64x32.prim = fncdef PFX(fname ## _64x32_ ## cpu); \139
+ p.puLUMA_32x64.prim = fncdef PFX(fname ## _32x64_ ## cpu); \140
+ p.puLUMA_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \141
+ p.puLUMA_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \142
+ p.puLUMA_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \143
+ p.puLUMA_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \144
+ p.puLUMA_64x48.prim = fncdef PFX(fname ## _64x48_ ## cpu); \145
+ p.puLUMA_48x64.prim = fncdef PFX(fname ## _48x64_ ## cpu); \146
+ p.puLUMA_64x16.prim = fncdef PFX(fname ## _64x16_ ## cpu)147
+#define LUMA_PU_TYPED_NEON_3(prim, fncdef, fname) \148
+ p.puLUMA_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \149
+ p.puLUMA_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \150
+ p.puLUMA_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon)151
+#define LUMA_PU_TYPED_CAN_USE_SVE2(prim, fncdef, fname) \152
+ p.puLUMA_8x8.prim = fncdef PFX(fname ## _8x8_ ## sve2); \153
+ p.puLUMA_16x16.prim = fncdef PFX(fname ## _16x16_ ## sve2); \154
+ p.puLUMA_32x32.prim = fncdef PFX(fname ## _32x32_ ## sve2); \155
+ p.puLUMA_64x64.prim = fncdef PFX(fname ## _64x64_ ## sve2); \156
+ p.puLUMA_8x4.prim = fncdef PFX(fname ## _8x4_ ## sve2); \157
+ p.puLUMA_16x8.prim = fncdef PFX(fname ## _16x8_ ## sve2); \158
+ p.puLUMA_8x16.prim = fncdef PFX(fname ## _8x16_ ## sve2); \159
+ p.puLUMA_16x32.prim = fncdef PFX(fname ## _16x32_ ## sve2); \160
+ p.puLUMA_32x16.prim = fncdef PFX(fname ## _32x16_ ## sve2); \161
+ p.puLUMA_64x32.prim = fncdef PFX(fname ## _64x32_ ## sve2); \162
+ p.puLUMA_32x64.prim = fncdef PFX(fname ## _32x64_ ## sve2); \163
+ p.puLUMA_16x12.prim = fncdef PFX(fname ## _16x12_ ## sve2); \164
+ p.puLUMA_12x16.prim = fncdef PFX(fname ## _12x16_ ## sve2); \165
+ p.puLUMA_16x4.prim = fncdef PFX(fname ## _16x4_ ## sve2); \166
+ p.puLUMA_32x24.prim = fncdef PFX(fname ## _32x24_ ## sve2); \167
+ p.puLUMA_24x32.prim = fncdef PFX(fname ## _24x32_ ## sve2); \168
+ p.puLUMA_32x8.prim = fncdef PFX(fname ## _32x8_ ## sve2); \169
+ p.puLUMA_8x32.prim = fncdef PFX(fname ## _8x32_ ## sve2); \170
+ p.puLUMA_64x48.prim = fncdef PFX(fname ## _64x48_ ## sve2); \171
+ p.puLUMA_48x64.prim = fncdef PFX(fname ## _48x64_ ## sve2); \172
+ p.puLUMA_64x16.prim = fncdef PFX(fname ## _64x16_ ## sve2); \173
+ p.puLUMA_16x64.prim = fncdef PFX(fname ## _16x64_ ## sve2)174
+#define LUMA_PU_TYPED_NEON_FILTER_PIXEL_TO_SHORT(prim, fncdef) \175
+ p.puLUMA_4x4.prim = fncdef PFX(filterPixelToShort ## _4x4_ ## neon); \176
+ p.puLUMA_8x8.prim = fncdef PFX(filterPixelToShort ## _8x8_ ## neon); \177
+ p.puLUMA_16x16.prim = fncdef PFX(filterPixelToShort ## _16x16_ ## neon); \178
+ p.puLUMA_8x4.prim = fncdef PFX(filterPixelToShort ## _8x4_ ## neon); \179
+ p.puLUMA_4x8.prim = fncdef PFX(filterPixelToShort ## _4x8_ ## neon); \180
+ p.puLUMA_16x8.prim = fncdef PFX(filterPixelToShort ## _16x8_ ## neon); \181
+ p.puLUMA_8x16.prim = fncdef PFX(filterPixelToShort ## _8x16_ ## neon); \182
+ p.puLUMA_16x32.prim = fncdef PFX(filterPixelToShort ## _16x32_ ## neon); \183
+ p.puLUMA_16x12.prim = fncdef PFX(filterPixelToShort ## _16x12_ ## neon); \184
+ p.puLUMA_12x16.prim = fncdef PFX(filterPixelToShort ## _12x16_ ## neon); \185
+ p.puLUMA_16x4.prim = fncdef PFX(filterPixelToShort ## _16x4_ ## neon); \186
+ p.puLUMA_4x16.prim = fncdef PFX(filterPixelToShort ## _4x16_ ## neon); \187
+ p.puLUMA_24x32.prim = fncdef PFX(filterPixelToShort ## _24x32_ ## neon); \188
+ p.puLUMA_8x32.prim = fncdef PFX(filterPixelToShort ## _8x32_ ## neon); \189
+ p.puLUMA_16x64.prim = fncdef PFX(filterPixelToShort ## _16x64_ ## neon)190
+#define LUMA_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, fncdef) \191
+ p.puLUMA_32x32.prim = fncdef PFX(filterPixelToShort ## _32x32_ ## sve); \192
+ p.puLUMA_32x16.prim = fncdef PFX(filterPixelToShort ## _32x16_ ## sve); \193
+ p.puLUMA_32x64.prim = fncdef PFX(filterPixelToShort ## _32x64_ ## sve); \194
+ p.puLUMA_32x24.prim = fncdef PFX(filterPixelToShort ## _32x24_ ## sve); \195
+ p.puLUMA_32x8.prim = fncdef PFX(filterPixelToShort ## _32x8_ ## sve); \196
+ p.puLUMA_64x64.prim = fncdef PFX(filterPixelToShort ## _64x64_ ## sve); \197
+ p.puLUMA_64x32.prim = fncdef PFX(filterPixelToShort ## _64x32_ ## sve); \198
+ p.puLUMA_64x48.prim = fncdef PFX(filterPixelToShort ## _64x48_ ## sve); \199
+ p.puLUMA_64x16.prim = fncdef PFX(filterPixelToShort ## _64x16_ ## sve); \200
+ p.puLUMA_48x64.prim = fncdef PFX(filterPixelToShort ## _48x64_ ## sve)201
+#define ALL_LUMA_PU(prim, fname, cpu) ALL_LUMA_PU_TYPED(prim, , fname, cpu)202
+#define LUMA_PU_MULTIPLE_ARCHS_1(prim, fname, cpu) LUMA_PU_TYPED_MULTIPLE_ARCHS_1(prim, , fname, cpu)203
+#define LUMA_PU_MULTIPLE_ARCHS_2(prim, fname, cpu) LUMA_PU_TYPED_MULTIPLE_ARCHS_2(prim, , fname, cpu)204
+#define LUMA_PU_NEON_1(prim, fname) LUMA_PU_TYPED_NEON_1(prim, , fname)205
+#define LUMA_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, fname) LUMA_PU_TYPED_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, , fname)206
+#define LUMA_PU_NEON_2(prim, fname) LUMA_PU_TYPED_NEON_2(prim, , fname)207
+#define LUMA_PU_MULTIPLE_ARCHS_3(prim, fname, cpu) LUMA_PU_TYPED_MULTIPLE_ARCHS_3(prim, , fname, cpu)208
+#define LUMA_PU_NEON_3(prim, fname) LUMA_PU_TYPED_NEON_3(prim, , fname)209
+#define LUMA_PU_CAN_USE_SVE2(prim, fname) LUMA_PU_TYPED_CAN_USE_SVE2(prim, , fname)210
+#define LUMA_PU_NEON_FILTER_PIXEL_TO_SHORT(prim) LUMA_PU_TYPED_NEON_FILTER_PIXEL_TO_SHORT(prim, )211
+#define LUMA_PU_SVE_FILTER_PIXEL_TO_SHORT(prim) LUMA_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, )212
+213
+214
+#define ALL_LUMA_PU_T(prim, fname) \215
+ p.puLUMA_4x4.prim = fname<LUMA_4x4>; \216
+ p.puLUMA_8x8.prim = fname<LUMA_8x8>; \217
+ p.puLUMA_16x16.prim = fname<LUMA_16x16>; \218
+ p.puLUMA_32x32.prim = fname<LUMA_32x32>; \219
+ p.puLUMA_64x64.prim = fname<LUMA_64x64>; \220
+ p.puLUMA_8x4.prim = fname<LUMA_8x4>; \221
+ p.puLUMA_4x8.prim = fname<LUMA_4x8>; \222
+ p.puLUMA_16x8.prim = fname<LUMA_16x8>; \223
+ p.puLUMA_8x16.prim = fname<LUMA_8x16>; \224
+ p.puLUMA_16x32.prim = fname<LUMA_16x32>; \225
+ p.puLUMA_32x16.prim = fname<LUMA_32x16>; \226
+ p.puLUMA_64x32.prim = fname<LUMA_64x32>; \227
+ p.puLUMA_32x64.prim = fname<LUMA_32x64>; \228
+ p.puLUMA_16x12.prim = fname<LUMA_16x12>; \229
+ p.puLUMA_12x16.prim = fname<LUMA_12x16>; \230
+ p.puLUMA_16x4.prim = fname<LUMA_16x4>; \231
+ p.puLUMA_4x16.prim = fname<LUMA_4x16>; \232
+ p.puLUMA_32x24.prim = fname<LUMA_32x24>; \233
+ p.puLUMA_24x32.prim = fname<LUMA_24x32>; \234
+ p.puLUMA_32x8.prim = fname<LUMA_32x8>; \235
+ p.puLUMA_8x32.prim = fname<LUMA_8x32>; \236
+ p.puLUMA_64x48.prim = fname<LUMA_64x48>; \237
+ p.puLUMA_48x64.prim = fname<LUMA_48x64>; \238
+ p.puLUMA_64x16.prim = fname<LUMA_64x16>; \239
+ p.puLUMA_16x64.prim = fname<LUMA_16x64>240
+241
+#define ALL_CHROMA_420_PU_TYPED(prim, fncdef, fname, cpu) \242
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \243
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \244
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \245
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \246
+ p.chromaX265_CSP_I420.puCHROMA_420_4x2.prim = fncdef PFX(fname ## _4x2_ ## cpu); \247
+ p.chromaX265_CSP_I420.puCHROMA_420_2x4.prim = fncdef PFX(fname ## _2x4_ ## cpu); \248
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \249
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.prim = fncdef PFX(fname ## _4x8_ ## cpu); \250
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \251
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \252
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \253
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \254
+ p.chromaX265_CSP_I420.puCHROMA_420_8x6.prim = fncdef PFX(fname ## _8x6_ ## cpu); \255
+ p.chromaX265_CSP_I420.puCHROMA_420_6x8.prim = fncdef PFX(fname ## _6x8_ ## cpu); \256
+ p.chromaX265_CSP_I420.puCHROMA_420_8x2.prim = fncdef PFX(fname ## _8x2_ ## cpu); \257
+ p.chromaX265_CSP_I420.puCHROMA_420_2x8.prim = fncdef PFX(fname ## _2x8_ ## cpu); \258
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.prim = fncdef PFX(fname ## _16x12_ ## cpu); \259
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \260
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.prim = fncdef PFX(fname ## _16x4_ ## cpu); \261
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.prim = fncdef PFX(fname ## _4x16_ ## cpu); \262
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \263
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \264
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \265
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu)266
+#define CHROMA_420_PU_TYPED_NEON_1(prim, fncdef, fname) \267
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \268
+ p.chromaX265_CSP_I420.puCHROMA_420_4x2.prim = fncdef PFX(fname ## _4x2_ ## neon); \269
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \270
+ p.chromaX265_CSP_I420.puCHROMA_420_6x8.prim = fncdef PFX(fname ## _6x8_ ## neon); \271
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.prim = fncdef PFX(fname ## _12x16_ ## neon); \272
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon); \273
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.prim = fncdef PFX(fname ## _32x24_ ## neon); \274
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.prim = fncdef PFX(fname ## _24x32_ ## neon); \275
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.prim = fncdef PFX(fname ## _32x8_ ## neon); \276
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.prim = fncdef PFX(fname ## _8x32_ ## neon); \277
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.prim = fncdef PFX(fname ## _8x8_ ## neon); \278
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.prim = fncdef PFX(fname ## _16x16_ ## neon); \279
+ p.chromaX265_CSP_I420.puCHROMA_420_2x4.prim = fncdef PFX(fname ## _2x4_ ## neon); \280
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.prim = fncdef PFX(fname ## _8x4_ ## neon); \281
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.prim = fncdef PFX(fname ## _16x8_ ## neon); \282
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.prim = fncdef PFX(fname ## _8x16_ ## neon); \283
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.prim = fncdef PFX(fname ## _16x32_ ## neon); \284
+ p.chromaX265_CSP_I420.puCHROMA_420_8x6.prim = fncdef PFX(fname ## _8x6_ ## neon); \285
+ p.chromaX265_CSP_I420.puCHROMA_420_8x2.prim = fncdef PFX(fname ## _8x2_ ## neon); \286
+ p.chromaX265_CSP_I420.puCHROMA_420_2x8.prim = fncdef PFX(fname ## _2x8_ ## neon); \287
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.prim = fncdef PFX(fname ## _16x12_ ## neon); \288
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.prim = fncdef PFX(fname ## _16x4_ ## neon)289
+#define CHROMA_420_PU_TYPED_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, fncdef, fname) \290
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.prim = fncdef PFX(fname ## _32x32_ ## sve); \291
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.prim = fncdef PFX(fname ## _32x16_ ## sve)292
+#define CHROMA_420_PU_TYPED_NEON_2(prim, fncdef, fname) \293
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \294
+ p.chromaX265_CSP_I420.puCHROMA_420_4x2.prim = fncdef PFX(fname ## _4x2_ ## neon); \295
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \296
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon)297
+#define CHROMA_420_PU_TYPED_MULTIPLE_ARCHS(prim, fncdef, fname, cpu) \298
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \299
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \300
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \301
+ p.chromaX265_CSP_I420.puCHROMA_420_2x4.prim = fncdef PFX(fname ## _2x4_ ## cpu); \302
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \303
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \304
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \305
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \306
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \307
+ p.chromaX265_CSP_I420.puCHROMA_420_8x6.prim = fncdef PFX(fname ## _8x6_ ## cpu); \308
+ p.chromaX265_CSP_I420.puCHROMA_420_6x8.prim = fncdef PFX(fname ## _6x8_ ## cpu); \309
+ p.chromaX265_CSP_I420.puCHROMA_420_8x2.prim = fncdef PFX(fname ## _8x2_ ## cpu); \310
+ p.chromaX265_CSP_I420.puCHROMA_420_2x8.prim = fncdef PFX(fname ## _2x8_ ## cpu); \311
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.prim = fncdef PFX(fname ## _16x12_ ## cpu); \312
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \313
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.prim = fncdef PFX(fname ## _16x4_ ## cpu); \314
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \315
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \316
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \317
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu)318
+#define CHROMA_420_PU_TYPED_FILTER_PIXEL_TO_SHORT_NEON(prim, fncdef) \319
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.prim = fncdef PFX(filterPixelToShort ## _4x4_ ## neon); \320
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.prim = fncdef PFX(filterPixelToShort ## _8x8_ ## neon); \321
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.prim = fncdef PFX(filterPixelToShort ## _16x16_ ## neon); \322
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.prim = fncdef PFX(filterPixelToShort ## _8x4_ ## neon); \323
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.prim = fncdef PFX(filterPixelToShort ## _4x8_ ## neon); \324
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.prim = fncdef PFX(filterPixelToShort ## _16x8_ ## neon); \325
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.prim = fncdef PFX(filterPixelToShort ## _8x16_ ## neon); \326
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.prim = fncdef PFX(filterPixelToShort ## _16x32_ ## neon); \327
+ p.chromaX265_CSP_I420.puCHROMA_420_8x6.prim = fncdef PFX(filterPixelToShort ## _8x6_ ## neon); \328
+ p.chromaX265_CSP_I420.puCHROMA_420_8x2.prim = fncdef PFX(filterPixelToShort ## _8x2_ ## neon); \329
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.prim = fncdef PFX(filterPixelToShort ## _16x12_ ## neon); \330
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.prim = fncdef PFX(filterPixelToShort ## _12x16_ ## neon); \331
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.prim = fncdef PFX(filterPixelToShort ## _16x4_ ## neon); \332
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.prim = fncdef PFX(filterPixelToShort ## _4x16_ ## neon); \333
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.prim = fncdef PFX(filterPixelToShort ## _24x32_ ## neon); \334
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.prim = fncdef PFX(filterPixelToShort ## _8x32_ ## neon)335
+#define CHROMA_420_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, fncdef) \336
+ p.chromaX265_CSP_I420.puCHROMA_420_2x4.prim = fncdef PFX(filterPixelToShort ## _2x4_ ## sve); \337
+ p.chromaX265_CSP_I420.puCHROMA_420_2x8.prim = fncdef PFX(filterPixelToShort ## _2x8_ ## sve); \338
+ p.chromaX265_CSP_I420.puCHROMA_420_6x8.prim = fncdef PFX(filterPixelToShort ## _6x8_ ## sve); \339
+ p.chromaX265_CSP_I420.puCHROMA_420_4x2.prim = fncdef PFX(filterPixelToShort ## _4x2_ ## sve); \340
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.prim = fncdef PFX(filterPixelToShort ## _32x32_ ## sve); \341
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.prim = fncdef PFX(filterPixelToShort ## _32x16_ ## sve); \342
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.prim = fncdef PFX(filterPixelToShort ## _32x24_ ## sve); \343
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.prim = fncdef PFX(filterPixelToShort ## _32x8_ ## sve)344
+#define ALL_CHROMA_420_PU(prim, fname, cpu) ALL_CHROMA_420_PU_TYPED(prim, , fname, cpu)345
+#define CHROMA_420_PU_NEON_1(prim, fname) CHROMA_420_PU_TYPED_NEON_1(prim, , fname)346
+#define CHROMA_420_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, fname) CHROMA_420_PU_TYPED_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, , fname)347
+#define CHROMA_420_PU_NEON_2(prim, fname) CHROMA_420_PU_TYPED_NEON_2(prim, , fname)348
+#define CHROMA_420_PU_MULTIPLE_ARCHS(prim, fname, cpu) CHROMA_420_PU_TYPED_MULTIPLE_ARCHS(prim, , fname, cpu)349
+#define CHROMA_420_PU_FILTER_PIXEL_TO_SHORT_NEON(prim) CHROMA_420_PU_TYPED_FILTER_PIXEL_TO_SHORT_NEON(prim, )350
+#define CHROMA_420_PU_SVE_FILTER_PIXEL_TO_SHORT(prim) CHROMA_420_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, )351
+352
+353
+#define ALL_CHROMA_420_4x4_PU_TYPED(prim, fncdef, fname, cpu) \354
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \355
+ p.chromaX265_CSP_I420.puCHROMA_420_8x2.prim = fncdef PFX(fname ## _8x2_ ## cpu); \356
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \357
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \358
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \359
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \360
+ p.chromaX265_CSP_I420.puCHROMA_420_8x6.prim = fncdef PFX(fname ## _8x6_ ## cpu); \361
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.prim = fncdef PFX(fname ## _4x8_ ## cpu); \362
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \363
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \364
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \365
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \366
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.prim = fncdef PFX(fname ## _16x12_ ## cpu); \367
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \368
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.prim = fncdef PFX(fname ## _16x4_ ## cpu); \369
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.prim = fncdef PFX(fname ## _4x16_ ## cpu); \370
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \371
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \372
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \373
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu)374
+#define ALL_CHROMA_420_4x4_PU(prim, fname, cpu) ALL_CHROMA_420_4x4_PU_TYPED(prim, , fname, cpu)375
+376
+#define ALL_CHROMA_422_PU_TYPED(prim, fncdef, fname, cpu) \377
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.prim = fncdef PFX(fname ## _4x8_ ## cpu); \378
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \379
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \380
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.prim = fncdef PFX(fname ## _32x64_ ## cpu); \381
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \382
+ p.chromaX265_CSP_I422.puCHROMA_422_2x8.prim = fncdef PFX(fname ## _2x8_ ## cpu); \383
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \384
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.prim = fncdef PFX(fname ## _4x16_ ## cpu); \385
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \386
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu); \387
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \388
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.prim = fncdef PFX(fname ## _16x64_ ## cpu); \389
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.prim = fncdef PFX(fname ## _8x12_ ## cpu); \390
+ p.chromaX265_CSP_I422.puCHROMA_422_6x16.prim = fncdef PFX(fname ## _6x16_ ## cpu); \391
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \392
+ p.chromaX265_CSP_I422.puCHROMA_422_2x16.prim = fncdef PFX(fname ## _2x16_ ## cpu); \393
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.prim = fncdef PFX(fname ## _16x24_ ## cpu); \394
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.prim = fncdef PFX(fname ## _12x32_ ## cpu); \395
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \396
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.prim = fncdef PFX(fname ## _4x32_ ## cpu); \397
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.prim = fncdef PFX(fname ## _32x48_ ## cpu); \398
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.prim = fncdef PFX(fname ## _24x64_ ## cpu); \399
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \400
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.prim = fncdef PFX(fname ## _8x64_ ## cpu)401
+#define CHROMA_422_PU_TYPED_NEON_1(prim, fncdef, fname) \402
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \403
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \404
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon); \405
+ p.chromaX265_CSP_I422.puCHROMA_422_6x16.prim = fncdef PFX(fname ## _6x16_ ## neon); \406
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.prim = fncdef PFX(fname ## _12x32_ ## neon); \407
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.prim = fncdef PFX(fname ## _4x32_ ## neon); \408
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.prim = fncdef PFX(fname ## _8x16_ ## neon); \409
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.prim = fncdef PFX(fname ## _16x32_ ## neon); \410
+ p.chromaX265_CSP_I422.puCHROMA_422_2x8.prim = fncdef PFX(fname ## _2x8_ ## neon); \411
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.prim = fncdef PFX(fname ## _8x8_ ## neon); \412
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.prim = fncdef PFX(fname ## _16x16_ ## neon); \413
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.prim = fncdef PFX(fname ## _8x32_ ## neon); \414
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.prim = fncdef PFX(fname ## _16x64_ ## neon); \415
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.prim = fncdef PFX(fname ## _8x12_ ## neon); \416
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.prim = fncdef PFX(fname ## _8x4_ ## neon); \417
+ p.chromaX265_CSP_I422.puCHROMA_422_2x16.prim = fncdef PFX(fname ## _2x16_ ## neon); \418
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.prim = fncdef PFX(fname ## _16x24_ ## neon); \419
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.prim = fncdef PFX(fname ## _16x8_ ## neon); \420
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.prim = fncdef PFX(fname ## _24x64_ ## neon); \421
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.prim = fncdef PFX(fname ## _8x64_ ## neon)422
+#define CHROMA_422_PU_TYPED_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, fncdef, fname) \423
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.prim = fncdef PFX(fname ## _32x64_ ## sve); \424
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.prim = fncdef PFX(fname ## _32x32_ ## sve); \425
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.prim = fncdef PFX(fname ## _32x48_ ## sve); \426
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.prim = fncdef PFX(fname ## _32x16_ ## sve)427
+#define CHROMA_422_PU_TYPED_NEON_2(prim, fncdef, fname) \428
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.prim = fncdef PFX(fname ## _4x8_ ## neon); \429
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.prim = fncdef PFX(fname ## _4x4_ ## neon); \430
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.prim = fncdef PFX(fname ## _4x16_ ## neon); \431
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.prim = fncdef PFX(fname ## _4x32_ ## neon)432
+#define CHROMA_422_PU_TYPED_CAN_USE_SVE2(prim, fncdef, fname) \433
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.prim = fncdef PFX(fname ## _8x16_ ## sve2); \434
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.prim = fncdef PFX(fname ## _16x32_ ## sve2); \435
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.prim = fncdef PFX(fname ## _32x64_ ## sve2); \436
+ p.chromaX265_CSP_I422.puCHROMA_422_2x8.prim = fncdef PFX(fname ## _2x8_ ## sve2); \437
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.prim = fncdef PFX(fname ## _8x8_ ## sve2); \438
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.prim = fncdef PFX(fname ## _16x16_ ## sve2); \439
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.prim = fncdef PFX(fname ## _8x32_ ## sve2); \440
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.prim = fncdef PFX(fname ## _32x32_ ## sve2); \441
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.prim = fncdef PFX(fname ## _16x64_ ## sve2); \442
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.prim = fncdef PFX(fname ## _8x12_ ## sve2); \443
+ p.chromaX265_CSP_I422.puCHROMA_422_6x16.prim = fncdef PFX(fname ## _6x16_ ## sve2); \444
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.prim = fncdef PFX(fname ## _8x4_ ## sve2); \445
+ p.chromaX265_CSP_I422.puCHROMA_422_2x16.prim = fncdef PFX(fname ## _2x16_ ## sve2); \446
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.prim = fncdef PFX(fname ## _16x24_ ## sve2); \447
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.prim = fncdef PFX(fname ## _12x32_ ## sve2); \448
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.prim = fncdef PFX(fname ## _16x8_ ## sve2); \449
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.prim = fncdef PFX(fname ## _32x48_ ## sve2); \450
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.prim = fncdef PFX(fname ## _24x64_ ## sve2); \451
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.prim = fncdef PFX(fname ## _32x16_ ## sve2); \452
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.prim = fncdef PFX(fname ## _8x64_ ## sve2)453
+#define CHROMA_422_PU_TYPED_NEON_FILTER_PIXEL_TO_SHORT(prim, fncdef) \454
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.prim = fncdef PFX(filterPixelToShort ## _4x8_ ## neon); \455
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.prim = fncdef PFX(filterPixelToShort ## _8x16_ ## neon); \456
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.prim = fncdef PFX(filterPixelToShort ## _16x32_ ## neon); \457
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.prim = fncdef PFX(filterPixelToShort ## _4x4_ ## neon); \458
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.prim = fncdef PFX(filterPixelToShort ## _8x8_ ## neon); \459
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.prim = fncdef PFX(filterPixelToShort ## _4x16_ ## neon); \460
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.prim = fncdef PFX(filterPixelToShort ## _16x16_ ## neon); \461
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.prim = fncdef PFX(filterPixelToShort ## _8x32_ ## neon); \462
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.prim = fncdef PFX(filterPixelToShort ## _16x64_ ## neon); \463
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.prim = fncdef PFX(filterPixelToShort ## _8x12_ ## neon); \464
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.prim = fncdef PFX(filterPixelToShort ## _8x4_ ## neon); \465
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.prim = fncdef PFX(filterPixelToShort ## _16x24_ ## neon); \466
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.prim = fncdef PFX(filterPixelToShort ## _12x32_ ## neon); \467
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.prim = fncdef PFX(filterPixelToShort ## _16x8_ ## neon); \468
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.prim = fncdef PFX(filterPixelToShort ## _4x32_ ## neon); \469
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.prim = fncdef PFX(filterPixelToShort ## _24x64_ ## neon); \470
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.prim = fncdef PFX(filterPixelToShort ## _8x64_ ## neon)471
+#define CHROMA_422_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, fncdef) \472
+ p.chromaX265_CSP_I422.puCHROMA_422_2x8.prim = fncdef PFX(filterPixelToShort ## _2x8_ ## sve); \473
+ p.chromaX265_CSP_I422.puCHROMA_422_2x16.prim = fncdef PFX(filterPixelToShort ## _2x16_ ## sve); \474
+ p.chromaX265_CSP_I422.puCHROMA_422_6x16.prim = fncdef PFX(filterPixelToShort ## _6x16_ ## sve); \475
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.prim = fncdef PFX(filterPixelToShort ## _32x64_ ## sve); \476
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.prim = fncdef PFX(filterPixelToShort ## _32x32_ ## sve); \477
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.prim = fncdef PFX(filterPixelToShort ## _32x48_ ## sve); \478
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.prim = fncdef PFX(filterPixelToShort ## _32x16_ ## sve)479
+#define ALL_CHROMA_422_PU(prim, fname, cpu) ALL_CHROMA_422_PU_TYPED(prim, , fname, cpu)480
+#define CHROMA_422_PU_NEON_1(prim, fname) CHROMA_422_PU_TYPED_NEON_1(prim, , fname)481
+#define CHROMA_422_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, fname) CHROMA_422_PU_TYPED_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(prim, , fname)482
+#define CHROMA_422_PU_NEON_2(prim, fname) CHROMA_422_PU_TYPED_NEON_2(prim, , fname)483
+#define CHROMA_422_PU_CAN_USE_SVE2(prim, fname) CHROMA_422_PU_TYPED_CAN_USE_SVE2(prim, , fname)484
+#define CHROMA_422_PU_NEON_FILTER_PIXEL_TO_SHORT(prim) CHROMA_422_PU_TYPED_NEON_FILTER_PIXEL_TO_SHORT(prim, )485
+#define CHROMA_422_PU_SVE_FILTER_PIXEL_TO_SHORT(prim) CHROMA_422_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, )486
+487
+#define ALL_CHROMA_444_PU_TYPED(prim, fncdef, fname, cpu) \488
+ p.chromaX265_CSP_I444.puLUMA_4x4.prim = fncdef PFX(fname ## _4x4_ ## cpu); \489
+ p.chromaX265_CSP_I444.puLUMA_8x8.prim = fncdef PFX(fname ## _8x8_ ## cpu); \490
+ p.chromaX265_CSP_I444.puLUMA_16x16.prim = fncdef PFX(fname ## _16x16_ ## cpu); \491
+ p.chromaX265_CSP_I444.puLUMA_32x32.prim = fncdef PFX(fname ## _32x32_ ## cpu); \492
+ p.chromaX265_CSP_I444.puLUMA_64x64.prim = fncdef PFX(fname ## _64x64_ ## cpu); \493
+ p.chromaX265_CSP_I444.puLUMA_8x4.prim = fncdef PFX(fname ## _8x4_ ## cpu); \494
+ p.chromaX265_CSP_I444.puLUMA_4x8.prim = fncdef PFX(fname ## _4x8_ ## cpu); \495
+ p.chromaX265_CSP_I444.puLUMA_16x8.prim = fncdef PFX(fname ## _16x8_ ## cpu); \496
+ p.chromaX265_CSP_I444.puLUMA_8x16.prim = fncdef PFX(fname ## _8x16_ ## cpu); \497
+ p.chromaX265_CSP_I444.puLUMA_16x32.prim = fncdef PFX(fname ## _16x32_ ## cpu); \498
+ p.chromaX265_CSP_I444.puLUMA_32x16.prim = fncdef PFX(fname ## _32x16_ ## cpu); \499
+ p.chromaX265_CSP_I444.puLUMA_64x32.prim = fncdef PFX(fname ## _64x32_ ## cpu); \500
+ p.chromaX265_CSP_I444.puLUMA_32x64.prim = fncdef PFX(fname ## _32x64_ ## cpu); \501
+ p.chromaX265_CSP_I444.puLUMA_16x12.prim = fncdef PFX(fname ## _16x12_ ## cpu); \502
+ p.chromaX265_CSP_I444.puLUMA_12x16.prim = fncdef PFX(fname ## _12x16_ ## cpu); \503
+ p.chromaX265_CSP_I444.puLUMA_16x4.prim = fncdef PFX(fname ## _16x4_ ## cpu); \504
+ p.chromaX265_CSP_I444.puLUMA_4x16.prim = fncdef PFX(fname ## _4x16_ ## cpu); \505
+ p.chromaX265_CSP_I444.puLUMA_32x24.prim = fncdef PFX(fname ## _32x24_ ## cpu); \506
+ p.chromaX265_CSP_I444.puLUMA_24x32.prim = fncdef PFX(fname ## _24x32_ ## cpu); \507
+ p.chromaX265_CSP_I444.puLUMA_32x8.prim = fncdef PFX(fname ## _32x8_ ## cpu); \508
+ p.chromaX265_CSP_I444.puLUMA_8x32.prim = fncdef PFX(fname ## _8x32_ ## cpu); \509
+ p.chromaX265_CSP_I444.puLUMA_64x48.prim = fncdef PFX(fname ## _64x48_ ## cpu); \510
+ p.chromaX265_CSP_I444.puLUMA_48x64.prim = fncdef PFX(fname ## _48x64_ ## cpu); \511
+ p.chromaX265_CSP_I444.puLUMA_64x16.prim = fncdef PFX(fname ## _64x16_ ## cpu); \512
+ p.chromaX265_CSP_I444.puLUMA_16x64.prim = fncdef PFX(fname ## _16x64_ ## cpu)513
+#define CHROMA_444_PU_TYPED_NEON_FILTER_PIXEL_TO_SHORT(prim, fncdef) \514
+ p.chromaX265_CSP_I444.puLUMA_4x4.prim = fncdef PFX(filterPixelToShort ## _4x4_ ## neon); \515
+ p.chromaX265_CSP_I444.puLUMA_8x8.prim = fncdef PFX(filterPixelToShort ## _8x8_ ## neon); \516
+ p.chromaX265_CSP_I444.puLUMA_16x16.prim = fncdef PFX(filterPixelToShort ## _16x16_ ## neon); \517
+ p.chromaX265_CSP_I444.puLUMA_8x4.prim = fncdef PFX(filterPixelToShort ## _8x4_ ## neon); \518
+ p.chromaX265_CSP_I444.puLUMA_4x8.prim = fncdef PFX(filterPixelToShort ## _4x8_ ## neon); \519
+ p.chromaX265_CSP_I444.puLUMA_16x8.prim = fncdef PFX(filterPixelToShort ## _16x8_ ## neon); \520
+ p.chromaX265_CSP_I444.puLUMA_8x16.prim = fncdef PFX(filterPixelToShort ## _8x16_ ## neon); \521
+ p.chromaX265_CSP_I444.puLUMA_16x32.prim = fncdef PFX(filterPixelToShort ## _16x32_ ## neon); \522
+ p.chromaX265_CSP_I444.puLUMA_16x12.prim = fncdef PFX(filterPixelToShort ## _16x12_ ## neon); \523
+ p.chromaX265_CSP_I444.puLUMA_12x16.prim = fncdef PFX(filterPixelToShort ## _12x16_ ## neon); \524
+ p.chromaX265_CSP_I444.puLUMA_16x4.prim = fncdef PFX(filterPixelToShort ## _16x4_ ## neon); \525
+ p.chromaX265_CSP_I444.puLUMA_4x16.prim = fncdef PFX(filterPixelToShort ## _4x16_ ## neon); \526
+ p.chromaX265_CSP_I444.puLUMA_24x32.prim = fncdef PFX(filterPixelToShort ## _24x32_ ## neon); \527
+ p.chromaX265_CSP_I444.puLUMA_8x32.prim = fncdef PFX(filterPixelToShort ## _8x32_ ## neon); \528
+ p.chromaX265_CSP_I444.puLUMA_16x64.prim = fncdef PFX(filterPixelToShort ## _16x64_ ## neon)529
+#define CHROMA_444_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, fncdef) \530
+ p.chromaX265_CSP_I444.puLUMA_32x32.prim = fncdef PFX(filterPixelToShort ## _32x32_ ## sve); \531
+ p.chromaX265_CSP_I444.puLUMA_32x16.prim = fncdef PFX(filterPixelToShort ## _32x16_ ## sve); \532
+ p.chromaX265_CSP_I444.puLUMA_32x64.prim = fncdef PFX(filterPixelToShort ## _32x64_ ## sve); \533
+ p.chromaX265_CSP_I444.puLUMA_32x24.prim = fncdef PFX(filterPixelToShort ## _32x24_ ## sve); \534
+ p.chromaX265_CSP_I444.puLUMA_32x8.prim = fncdef PFX(filterPixelToShort ## _32x8_ ## sve); \535
+ p.chromaX265_CSP_I444.puLUMA_64x64.prim = fncdef PFX(filterPixelToShort ## _64x64_ ## sve); \536
+ p.chromaX265_CSP_I444.puLUMA_64x32.prim = fncdef PFX(filterPixelToShort ## _64x32_ ## sve); \537
+ p.chromaX265_CSP_I444.puLUMA_64x48.prim = fncdef PFX(filterPixelToShort ## _64x48_ ## sve); \538
+ p.chromaX265_CSP_I444.puLUMA_64x16.prim = fncdef PFX(filterPixelToShort ## _64x16_ ## sve); \539
+ p.chromaX265_CSP_I444.puLUMA_48x64.prim = fncdef PFX(filterPixelToShort ## _48x64_ ## sve)540
+#define ALL_CHROMA_444_PU(prim, fname, cpu) ALL_CHROMA_444_PU_TYPED(prim, , fname, cpu)541
+#define CHROMA_444_PU_NEON_FILTER_PIXEL_TO_SHORT(prim) CHROMA_444_PU_TYPED_NEON_FILTER_PIXEL_TO_SHORT(prim, )542
+#define CHROMA_444_PU_SVE_FILTER_PIXEL_TO_SHORT(prim) CHROMA_444_PU_TYPED_SVE_FILTER_PIXEL_TO_SHORT(prim, )543
+544
+#define ALL_CHROMA_420_VERT_FILTERS(cpu) \545
+ ALL_CHROMA_420_4x4_PU(filter_vpp, interp_4tap_vert_pp, cpu); \546
+ ALL_CHROMA_420_4x4_PU(filter_vps, interp_4tap_vert_ps, cpu); \547
+ ALL_CHROMA_420_4x4_PU(filter_vsp, interp_4tap_vert_sp, cpu); \548
+ ALL_CHROMA_420_4x4_PU(filter_vss, interp_4tap_vert_ss, cpu)549
+550
+#define CHROMA_420_VERT_FILTERS_NEON() \551
+ ALL_CHROMA_420_4x4_PU(filter_vsp, interp_4tap_vert_sp, neon)552
+553
+#define CHROMA_420_VERT_FILTERS_CAN_USE_SVE2() \554
+ ALL_CHROMA_420_4x4_PU(filter_vpp, interp_4tap_vert_pp, sve2); \555
+ ALL_CHROMA_420_4x4_PU(filter_vps, interp_4tap_vert_ps, sve2); \556
+ ALL_CHROMA_420_4x4_PU(filter_vss, interp_4tap_vert_ss, sve2)557
+558
+#define SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(W, H) \559
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vsp = PFX(interp_4tap_vert_sp_ ## W ## x ## H ## _ ## neon)560
+561
+#define SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(W, H, cpu) \562
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vpp = PFX(interp_4tap_vert_pp_ ## W ## x ## H ## _ ## cpu); \563
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vps = PFX(interp_4tap_vert_ps_ ## W ## x ## H ## _ ## cpu); \564
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vss = PFX(interp_4tap_vert_ss_ ## W ## x ## H ## _ ## cpu)565
+566
+#define CHROMA_422_VERT_FILTERS_NEON() \567
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(4, 8); \568
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(8, 16); \569
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(8, 8); \570
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(4, 16); \571
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(8, 12); \572
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(8, 4); \573
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(16, 32); \574
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(16, 16); \575
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(8, 32); \576
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(16, 24); \577
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(12, 32); \578
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(16, 8); \579
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(4, 32); \580
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(32, 64); \581
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(32, 32); \582
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(16, 64); \583
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(32, 48); \584
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(24, 64); \585
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(32, 16); \586
+ SETUP_CHROMA_422_VERT_FUNC_DEF_NEON(8, 64)587
+588
+#define CHROMA_422_VERT_FILTERS_CAN_USE_SVE2(cpu) \589
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(4, 8, cpu); \590
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(8, 16, cpu); \591
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(8, 8, cpu); \592
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(4, 16, cpu); \593
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(8, 12, cpu); \594
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(8, 4, cpu); \595
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(16, 32, cpu); \596
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(16, 16, cpu); \597
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(8, 32, cpu); \598
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(16, 24, cpu); \599
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(12, 32, cpu); \600
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(16, 8, cpu); \601
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(4, 32, cpu); \602
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(32, 64, cpu); \603
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(32, 32, cpu); \604
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(16, 64, cpu); \605
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(32, 48, cpu); \606
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(24, 64, cpu); \607
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(32, 16, cpu); \608
+ SETUP_CHROMA_422_VERT_FUNC_DEF_CAN_USE_SVE2(8, 64, cpu)609
+610
+#define ALL_CHROMA_444_VERT_FILTERS(cpu) \611
+ ALL_CHROMA_444_PU(filter_vpp, interp_4tap_vert_pp, cpu); \612
+ ALL_CHROMA_444_PU(filter_vps, interp_4tap_vert_ps, cpu); \613
+ ALL_CHROMA_444_PU(filter_vsp, interp_4tap_vert_sp, cpu); \614
+ ALL_CHROMA_444_PU(filter_vss, interp_4tap_vert_ss, cpu)615
+616
+#define CHROMA_444_VERT_FILTERS_NEON() \617
+ ALL_CHROMA_444_PU(filter_vsp, interp_4tap_vert_sp, neon)618
+619
+#define CHROMA_444_VERT_FILTERS_CAN_USE_SVE2() \620
+ ALL_CHROMA_444_PU(filter_vpp, interp_4tap_vert_pp, sve2); \621
+ ALL_CHROMA_444_PU(filter_vps, interp_4tap_vert_ps, sve2); \622
+ ALL_CHROMA_444_PU(filter_vss, interp_4tap_vert_ss, sve2)623
+624
+#define ALL_CHROMA_420_FILTERS(cpu) \625
+ ALL_CHROMA_420_PU(filter_hpp, interp_4tap_horiz_pp, cpu); \626
+ ALL_CHROMA_420_PU(filter_hps, interp_4tap_horiz_ps, cpu); \627
+ ALL_CHROMA_420_PU(filter_vpp, interp_4tap_vert_pp, cpu); \628
+ ALL_CHROMA_420_PU(filter_vps, interp_4tap_vert_ps, cpu)629
+630
+#define CHROMA_420_FILTERS_NEON() \631
+ ALL_CHROMA_420_PU(filter_hpp, interp_4tap_horiz_pp, neon); \632
+ ALL_CHROMA_420_PU(filter_hps, interp_4tap_horiz_ps, neon)633
+634
+#define CHROMA_420_FILTERS_CAN_USE_SVE2() \635
+ ALL_CHROMA_420_PU(filter_vpp, interp_4tap_vert_pp, sve2); \636
+ ALL_CHROMA_420_PU(filter_vps, interp_4tap_vert_ps, sve2)637
+638
+#define ALL_CHROMA_422_FILTERS(cpu) \639
+ ALL_CHROMA_422_PU(filter_hpp, interp_4tap_horiz_pp, cpu); \640
+ ALL_CHROMA_422_PU(filter_hps, interp_4tap_horiz_ps, cpu); \641
+ ALL_CHROMA_422_PU(filter_vpp, interp_4tap_vert_pp, cpu); \642
+ ALL_CHROMA_422_PU(filter_vps, interp_4tap_vert_ps, cpu)643
+644
+#define CHROMA_422_FILTERS_NEON() \645
+ ALL_CHROMA_422_PU(filter_hpp, interp_4tap_horiz_pp, neon); \646
+ ALL_CHROMA_422_PU(filter_hps, interp_4tap_horiz_ps, neon)647
+648
+#define CHROMA_422_FILTERS_CAN_USE_SVE2() \649
+ ALL_CHROMA_422_PU(filter_vpp, interp_4tap_vert_pp, sve2); \650
+ ALL_CHROMA_422_PU(filter_vps, interp_4tap_vert_ps, sve2)651
+652
+#define ALL_CHROMA_444_FILTERS(cpu) \653
+ ALL_CHROMA_444_PU(filter_hpp, interp_4tap_horiz_pp, cpu); \654
+ ALL_CHROMA_444_PU(filter_hps, interp_4tap_horiz_ps, cpu); \655
+ ALL_CHROMA_444_PU(filter_vpp, interp_4tap_vert_pp, cpu); \656
+ ALL_CHROMA_444_PU(filter_vps, interp_4tap_vert_ps, cpu)657
+658
+#define CHROMA_444_FILTERS_NEON() \659
+ ALL_CHROMA_444_PU(filter_hpp, interp_4tap_horiz_pp, neon); \660
+ ALL_CHROMA_444_PU(filter_hps, interp_4tap_horiz_ps, neon)661
+662
+#define CHROMA_444_FILTERS_CAN_USE_SVE2() \663
+ ALL_CHROMA_444_PU(filter_vpp, interp_4tap_vert_pp, sve2); \664
+ ALL_CHROMA_444_PU(filter_vps, interp_4tap_vert_ps, sve2)665
+666
667
#if defined(__GNUC__)668
#define GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__)669
670
#define GCC_4_9_0 40900671
#define GCC_5_1_0 50100672
673
-extern "C" {674
-#include "pixel.h"675
-#include "pixel-util.h"676
-#include "ipfilter8.h"677
-}678
+#include "pixel-prim.h"679
+#include "filter-prim.h"680
+#include "dct-prim.h"681
+#include "loopfilter-prim.h"682
+#include "intrapred-prim.h"683
684
-namespace X265_NS {685
+namespace X265_NS686
+{687
// private x265 namespace688
689
690
template<int size>691
-void interp_8tap_hv_pp_cpu(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int idxX, int idxY)692
+void interp_8tap_hv_pp_cpu(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)693
{694
ALIGN_VAR_32(int16_t, immedMAX_CU_SIZE * (MAX_CU_SIZE + NTAPS_LUMA - 1));695
const int halfFilterSize = NTAPS_LUMA >> 1;696
697
primitives.pusize.luma_vsp(immed + (halfFilterSize - 1) * immedStride, immedStride, dst, dstStride, idxY);698
}699
700
-701
-/* Temporary workaround because luma_vsp assembly primitive has not been completed702
- * but interp_8tap_hv_pp_cpu uses mixed C primitive and assembly primitive.703
- * Otherwise, segment fault occurs. */704
-void setupAliasCPrimitives(EncoderPrimitives &cp, EncoderPrimitives &asmp, int cpuMask)705
+void setupNeonPrimitives(EncoderPrimitives &p)706
{707
- if (cpuMask & X265_CPU_NEON)708
- {709
- asmp.puLUMA_8x4.luma_vsp = cp.puLUMA_8x4.luma_vsp;710
- asmp.puLUMA_8x8.luma_vsp = cp.puLUMA_8x8.luma_vsp;711
- asmp.puLUMA_8x16.luma_vsp = cp.puLUMA_8x16.luma_vsp;712
- asmp.puLUMA_8x32.luma_vsp = cp.puLUMA_8x32.luma_vsp;713
- asmp.puLUMA_12x16.luma_vsp = cp.puLUMA_12x16.luma_vsp;714
-#if !AUTO_VECTORIZE || GCC_VERSION < GCC_5_1_0 /* gcc_version < gcc-5.1.0 */715
- asmp.puLUMA_16x4.luma_vsp = cp.puLUMA_16x4.luma_vsp;716
- asmp.puLUMA_16x8.luma_vsp = cp.puLUMA_16x8.luma_vsp;717
- asmp.puLUMA_16x12.luma_vsp = cp.puLUMA_16x12.luma_vsp;718
- asmp.puLUMA_16x16.luma_vsp = cp.puLUMA_16x16.luma_vsp;719
- asmp.puLUMA_16x32.luma_vsp = cp.puLUMA_16x32.luma_vsp;720
- asmp.puLUMA_16x64.luma_vsp = cp.puLUMA_16x64.luma_vsp;721
- asmp.puLUMA_32x16.luma_vsp = cp.puLUMA_32x16.luma_vsp;722
- asmp.puLUMA_32x24.luma_vsp = cp.puLUMA_32x24.luma_vsp;723
- asmp.puLUMA_32x32.luma_vsp = cp.puLUMA_32x32.luma_vsp;724
- asmp.puLUMA_32x64.luma_vsp = cp.puLUMA_32x64.luma_vsp;725
- asmp.puLUMA_48x64.luma_vsp = cp.puLUMA_48x64.luma_vsp;726
- asmp.puLUMA_64x16.luma_vsp = cp.puLUMA_64x16.luma_vsp;727
- asmp.puLUMA_64x32.luma_vsp = cp.puLUMA_64x32.luma_vsp;728
- asmp.puLUMA_64x48.luma_vsp = cp.puLUMA_64x48.luma_vsp;729
- asmp.puLUMA_64x64.luma_vsp = cp.puLUMA_64x64.luma_vsp; 730
-#if !AUTO_VECTORIZE || GCC_VERSION < GCC_4_9_0 /* gcc_version < gcc-4.9.0 */731
- asmp.puLUMA_4x4.luma_vsp = cp.puLUMA_4x4.luma_vsp;732
- asmp.puLUMA_4x8.luma_vsp = cp.puLUMA_4x8.luma_vsp;733
- asmp.puLUMA_4x16.luma_vsp = cp.puLUMA_4x16.luma_vsp;734
- asmp.puLUMA_24x32.luma_vsp = cp.puLUMA_24x32.luma_vsp;735
- asmp.puLUMA_32x8.luma_vsp = cp.puLUMA_32x8.luma_vsp;736
+ setupPixelPrimitives_neon(p);737
+ setupFilterPrimitives_neon(p);738
+ setupDCTPrimitives_neon(p);739
+ setupLoopFilterPrimitives_neon(p);740
+ setupIntraPrimitives_neon(p);741
+742
+ ALL_CHROMA_420_PU(p2sNONALIGNED, filterPixelToShort, neon);743
+ ALL_CHROMA_422_PU(p2sALIGNED, filterPixelToShort, neon);744
+ ALL_CHROMA_444_PU(p2sALIGNED, filterPixelToShort, neon);745
+ ALL_LUMA_PU(convert_p2sALIGNED, filterPixelToShort, neon);746
+ ALL_CHROMA_420_PU(p2sALIGNED, filterPixelToShort, neon);747
+ ALL_CHROMA_422_PU(p2sNONALIGNED, filterPixelToShort, neon);748
+ ALL_CHROMA_444_PU(p2sNONALIGNED, filterPixelToShort, neon);749
+ ALL_LUMA_PU(convert_p2sNONALIGNED, filterPixelToShort, neon);750
+751
+#if !HIGH_BIT_DEPTH752
+ ALL_LUMA_PU(luma_vpp, interp_8tap_vert_pp, neon);753
+ ALL_LUMA_PU(luma_vsp, interp_8tap_vert_sp, neon);754
+ ALL_LUMA_PU(luma_vps, interp_8tap_vert_ps, neon);755
+ ALL_LUMA_PU(luma_hpp, interp_horiz_pp, neon);756
+ ALL_LUMA_PU(luma_hps, interp_horiz_ps, neon);757
+ ALL_LUMA_PU(luma_vss, interp_8tap_vert_ss, neon);758
+ ALL_LUMA_PU_T(luma_hvpp, interp_8tap_hv_pp_cpu);759
+ ALL_CHROMA_420_VERT_FILTERS(neon);760
+ CHROMA_422_VERT_FILTERS_NEON();761
+ CHROMA_422_VERT_FILTERS_CAN_USE_SVE2(neon);762
+ ALL_CHROMA_444_VERT_FILTERS(neon);763
+ ALL_CHROMA_420_FILTERS(neon);764
+ ALL_CHROMA_422_FILTERS(neon);765
+ ALL_CHROMA_444_FILTERS(neon);766
+767
+ // Blockcopy_pp768
+ ALL_LUMA_PU(copy_pp, blockcopy_pp, neon);769
+ ALL_CHROMA_420_PU(copy_pp, blockcopy_pp, neon);770
+ ALL_CHROMA_422_PU(copy_pp, blockcopy_pp, neon);771
+ p.cuBLOCK_4x4.copy_pp = PFX(blockcopy_pp_4x4_neon);772
+ p.cuBLOCK_8x8.copy_pp = PFX(blockcopy_pp_8x8_neon);773
+ p.cuBLOCK_16x16.copy_pp = PFX(blockcopy_pp_16x16_neon);774
+ p.cuBLOCK_32x32.copy_pp = PFX(blockcopy_pp_32x32_neon);775
+ p.cuBLOCK_64x64.copy_pp = PFX(blockcopy_pp_64x64_neon);776
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_pp = PFX(blockcopy_pp_4x4_neon);777
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_pp = PFX(blockcopy_pp_8x8_neon);778
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_pp = PFX(blockcopy_pp_16x16_neon);779
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_pp = PFX(blockcopy_pp_32x32_neon);780
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_pp = PFX(blockcopy_pp_4x8_neon);781
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_pp = PFX(blockcopy_pp_8x16_neon);782
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_pp = PFX(blockcopy_pp_16x32_neon);783
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_pp = PFX(blockcopy_pp_32x64_neon);784
+785
+#endif // !HIGH_BIT_DEPTH786
+787
+ // Blockcopy_ss788
+ p.cuBLOCK_4x4.copy_ss = PFX(blockcopy_ss_4x4_neon);789
+ p.cuBLOCK_8x8.copy_ss = PFX(blockcopy_ss_8x8_neon);790
+ p.cuBLOCK_16x16.copy_ss = PFX(blockcopy_ss_16x16_neon);791
+ p.cuBLOCK_32x32.copy_ss = PFX(blockcopy_ss_32x32_neon);792
+ p.cuBLOCK_64x64.copy_ss = PFX(blockcopy_ss_64x64_neon);793
+794
+ // Blockcopy_ps795
+ p.cuBLOCK_4x4.copy_ps = PFX(blockcopy_ps_4x4_neon);796
+ p.cuBLOCK_8x8.copy_ps = PFX(blockcopy_ps_8x8_neon);797
+ p.cuBLOCK_16x16.copy_ps = PFX(blockcopy_ps_16x16_neon);798
+ p.cuBLOCK_32x32.copy_ps = PFX(blockcopy_ps_32x32_neon);799
+ p.cuBLOCK_64x64.copy_ps = PFX(blockcopy_ps_64x64_neon);800
+801
+ // Blockcopy_sp802
+ p.cuBLOCK_4x4.copy_sp = PFX(blockcopy_sp_4x4_neon);803
+ p.cuBLOCK_8x8.copy_sp = PFX(blockcopy_sp_8x8_neon);804
+ p.cuBLOCK_16x16.copy_sp = PFX(blockcopy_sp_16x16_neon);805
+ p.cuBLOCK_32x32.copy_sp = PFX(blockcopy_sp_32x32_neon);806
+ p.cuBLOCK_64x64.copy_sp = PFX(blockcopy_sp_64x64_neon);807
+808
+ // chroma blockcopy_ss809
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_ss = PFX(blockcopy_ss_4x4_neon);810
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_ss = PFX(blockcopy_ss_8x8_neon);811
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_ss = PFX(blockcopy_ss_16x16_neon);812
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_ss = PFX(blockcopy_ss_32x32_neon);813
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_ss = PFX(blockcopy_ss_4x8_neon);814
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_ss = PFX(blockcopy_ss_8x16_neon);815
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_ss = PFX(blockcopy_ss_16x32_neon);816
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_ss = PFX(blockcopy_ss_32x64_neon);817
+818
+ // chroma blockcopy_ps819
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_ps = PFX(blockcopy_ps_4x4_neon);820
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_ps = PFX(blockcopy_ps_8x8_neon);821
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_ps = PFX(blockcopy_ps_16x16_neon);822
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_ps = PFX(blockcopy_ps_32x32_neon);823
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_ps = PFX(blockcopy_ps_4x8_neon);824
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_ps = PFX(blockcopy_ps_8x16_neon);825
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_ps = PFX(blockcopy_ps_16x32_neon);826
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_ps = PFX(blockcopy_ps_32x64_neon);827
+828
+ // chroma blockcopy_sp829
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_sp = PFX(blockcopy_sp_4x4_neon);830
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_sp = PFX(blockcopy_sp_8x8_neon);831
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_sp = PFX(blockcopy_sp_16x16_neon);832
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_sp = PFX(blockcopy_sp_32x32_neon);833
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_sp = PFX(blockcopy_sp_4x8_neon);834
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_sp = PFX(blockcopy_sp_8x16_neon);835
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_sp = PFX(blockcopy_sp_16x32_neon);836
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_sp = PFX(blockcopy_sp_32x64_neon);837
+838
+ // Block_fill839
+ ALL_LUMA_TU(blockfill_sALIGNED, blockfill_s, neon);840
+ ALL_LUMA_TU(blockfill_sNONALIGNED, blockfill_s, neon);841
+842
+ // copy_count843
+ p.cuBLOCK_4x4.copy_cnt = PFX(copy_cnt_4_neon);844
+ p.cuBLOCK_8x8.copy_cnt = PFX(copy_cnt_8_neon);845
+ p.cuBLOCK_16x16.copy_cnt = PFX(copy_cnt_16_neon);846
+ p.cuBLOCK_32x32.copy_cnt = PFX(copy_cnt_32_neon);847
+848
+ // count nonzero849
+ p.cuBLOCK_4x4.count_nonzero = PFX(count_nonzero_4_neon);850
+ p.cuBLOCK_8x8.count_nonzero = PFX(count_nonzero_8_neon);851
+ p.cuBLOCK_16x16.count_nonzero = PFX(count_nonzero_16_neon);852
+ p.cuBLOCK_32x32.count_nonzero = PFX(count_nonzero_32_neon);853
+854
+ // cpy2Dto1D_shl855
+ p.cuBLOCK_4x4.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_4x4_neon);856
+ p.cuBLOCK_8x8.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_8x8_neon);857
+ p.cuBLOCK_16x16.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_16x16_neon);858
+ p.cuBLOCK_32x32.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_32x32_neon);859
+ p.cuBLOCK_64x64.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_64x64_neon);860
+861
+ // cpy2Dto1D_shr862
+ p.cuBLOCK_4x4.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_4x4_neon);863
+ p.cuBLOCK_8x8.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_8x8_neon);864
+ p.cuBLOCK_16x16.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_16x16_neon);865
+ p.cuBLOCK_32x32.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_32x32_neon);866
+867
+ // cpy1Dto2D_shl868
+ p.cuBLOCK_4x4.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_4x4_neon);869
+ p.cuBLOCK_8x8.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_8x8_neon);870
+ p.cuBLOCK_16x16.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_16x16_neon);871
+ p.cuBLOCK_32x32.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_32x32_neon);872
+ p.cuBLOCK_64x64.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_64x64_neon);873
+874
+ p.cuBLOCK_4x4.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_4x4_neon);875
+ p.cuBLOCK_8x8.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_8x8_neon);876
+ p.cuBLOCK_16x16.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_16x16_neon);877
+ p.cuBLOCK_32x32.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_32x32_neon);878
+ p.cuBLOCK_64x64.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_64x64_neon);879
+880
+ // cpy1Dto2D_shr881
+ p.cuBLOCK_4x4.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_4x4_neon);882
+ p.cuBLOCK_8x8.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_8x8_neon);883
+ p.cuBLOCK_16x16.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_16x16_neon);884
+ p.cuBLOCK_32x32.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_32x32_neon);885
+ p.cuBLOCK_64x64.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_64x64_neon);886
+887
+#if !HIGH_BIT_DEPTH888
+ // pixel_avg_pp889
+ ALL_LUMA_PU(pixelavg_ppNONALIGNED, pixel_avg_pp, neon);890
+ ALL_LUMA_PU(pixelavg_ppALIGNED, pixel_avg_pp, neon);891
+892
+ // addAvg893
+ ALL_LUMA_PU(addAvgNONALIGNED, addAvg, neon);894
+ ALL_LUMA_PU(addAvgALIGNED, addAvg, neon);895
+ ALL_CHROMA_420_PU(addAvgNONALIGNED, addAvg, neon);896
+ ALL_CHROMA_422_PU(addAvgNONALIGNED, addAvg, neon);897
+ ALL_CHROMA_420_PU(addAvgALIGNED, addAvg, neon);898
+ ALL_CHROMA_422_PU(addAvgALIGNED, addAvg, neon);899
+900
+ // sad901
+ ALL_LUMA_PU(sad, pixel_sad, neon);902
+ ALL_LUMA_PU(sad_x3, sad_x3, neon);903
+ ALL_LUMA_PU(sad_x4, sad_x4, neon);904
+905
+ // sse_pp906
+ p.cuBLOCK_4x4.sse_pp = PFX(pixel_sse_pp_4x4_neon);907
+ p.cuBLOCK_8x8.sse_pp = PFX(pixel_sse_pp_8x8_neon);908
+ p.cuBLOCK_16x16.sse_pp = PFX(pixel_sse_pp_16x16_neon);909
+ p.cuBLOCK_32x32.sse_pp = PFX(pixel_sse_pp_32x32_neon);910
+ p.cuBLOCK_64x64.sse_pp = PFX(pixel_sse_pp_64x64_neon);911
+912
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.sse_pp = PFX(pixel_sse_pp_4x4_neon);913
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.sse_pp = PFX(pixel_sse_pp_8x8_neon);914
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.sse_pp = PFX(pixel_sse_pp_16x16_neon);915
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.sse_pp = PFX(pixel_sse_pp_32x32_neon);916
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.sse_pp = PFX(pixel_sse_pp_4x8_neon);917
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sse_pp = PFX(pixel_sse_pp_8x16_neon);918
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sse_pp = PFX(pixel_sse_pp_16x32_neon);919
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sse_pp = PFX(pixel_sse_pp_32x64_neon);920
+921
+ // sse_ss922
+ p.cuBLOCK_4x4.sse_ss = PFX(pixel_sse_ss_4x4_neon);923
+ p.cuBLOCK_8x8.sse_ss = PFX(pixel_sse_ss_8x8_neon);924
+ p.cuBLOCK_16x16.sse_ss = PFX(pixel_sse_ss_16x16_neon);925
+ p.cuBLOCK_32x32.sse_ss = PFX(pixel_sse_ss_32x32_neon);926
+ p.cuBLOCK_64x64.sse_ss = PFX(pixel_sse_ss_64x64_neon);927
+928
+ // ssd_s929
+ p.cuBLOCK_4x4.ssd_sNONALIGNED = PFX(pixel_ssd_s_4x4_neon);930
+ p.cuBLOCK_8x8.ssd_sNONALIGNED = PFX(pixel_ssd_s_8x8_neon);931
+ p.cuBLOCK_16x16.ssd_sNONALIGNED = PFX(pixel_ssd_s_16x16_neon);932
+ p.cuBLOCK_32x32.ssd_sNONALIGNED = PFX(pixel_ssd_s_32x32_neon);933
+934
+ p.cuBLOCK_4x4.ssd_sALIGNED = PFX(pixel_ssd_s_4x4_neon);935
+ p.cuBLOCK_8x8.ssd_sALIGNED = PFX(pixel_ssd_s_8x8_neon);936
+ p.cuBLOCK_16x16.ssd_sALIGNED = PFX(pixel_ssd_s_16x16_neon);937
+ p.cuBLOCK_32x32.ssd_sALIGNED = PFX(pixel_ssd_s_32x32_neon);938
+939
+ // pixel_var940
+ p.cuBLOCK_8x8.var = PFX(pixel_var_8x8_neon);941
+ p.cuBLOCK_16x16.var = PFX(pixel_var_16x16_neon);942
+ p.cuBLOCK_32x32.var = PFX(pixel_var_32x32_neon);943
+ p.cuBLOCK_64x64.var = PFX(pixel_var_64x64_neon);944
+945
+ // calc_Residual946
+ p.cuBLOCK_4x4.calcresidualNONALIGNED = PFX(getResidual4_neon);947
+ p.cuBLOCK_8x8.calcresidualNONALIGNED = PFX(getResidual8_neon);948
+ p.cuBLOCK_16x16.calcresidualNONALIGNED = PFX(getResidual16_neon);949
+ p.cuBLOCK_32x32.calcresidualNONALIGNED = PFX(getResidual32_neon);950
+951
+ p.cuBLOCK_4x4.calcresidualALIGNED = PFX(getResidual4_neon);952
+ p.cuBLOCK_8x8.calcresidualALIGNED = PFX(getResidual8_neon);953
+ p.cuBLOCK_16x16.calcresidualALIGNED = PFX(getResidual16_neon);954
+ p.cuBLOCK_32x32.calcresidualALIGNED = PFX(getResidual32_neon);955
+956
+ // pixel_sub_ps957
+ p.cuBLOCK_4x4.sub_ps = PFX(pixel_sub_ps_4x4_neon);958
+ p.cuBLOCK_8x8.sub_ps = PFX(pixel_sub_ps_8x8_neon);959
+ p.cuBLOCK_16x16.sub_ps = PFX(pixel_sub_ps_16x16_neon);960
+ p.cuBLOCK_32x32.sub_ps = PFX(pixel_sub_ps_32x32_neon);961
+ p.cuBLOCK_64x64.sub_ps = PFX(pixel_sub_ps_64x64_neon);962
+963
+ // chroma sub_ps964
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.sub_ps = PFX(pixel_sub_ps_4x4_neon);965
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.sub_ps = PFX(pixel_sub_ps_8x8_neon);966
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.sub_ps = PFX(pixel_sub_ps_16x16_neon);967
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.sub_ps = PFX(pixel_sub_ps_32x32_neon);968
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.sub_ps = PFX(pixel_sub_ps_4x8_neon);969
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sub_ps = PFX(pixel_sub_ps_8x16_neon);970
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sub_ps = PFX(pixel_sub_ps_16x32_neon);971
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sub_ps = PFX(pixel_sub_ps_32x64_neon);972
+973
+ // pixel_add_ps974
+ p.cuBLOCK_4x4.add_psNONALIGNED = PFX(pixel_add_ps_4x4_neon);975
+ p.cuBLOCK_8x8.add_psNONALIGNED = PFX(pixel_add_ps_8x8_neon);976
+ p.cuBLOCK_16x16.add_psNONALIGNED = PFX(pixel_add_ps_16x16_neon);977
+ p.cuBLOCK_32x32.add_psNONALIGNED = PFX(pixel_add_ps_32x32_neon);978
+ p.cuBLOCK_64x64.add_psNONALIGNED = PFX(pixel_add_ps_64x64_neon);979
+980
+ p.cuBLOCK_4x4.add_psALIGNED = PFX(pixel_add_ps_4x4_neon);981
+ p.cuBLOCK_8x8.add_psALIGNED = PFX(pixel_add_ps_8x8_neon);982
+ p.cuBLOCK_16x16.add_psALIGNED = PFX(pixel_add_ps_16x16_neon);983
+ p.cuBLOCK_32x32.add_psALIGNED = PFX(pixel_add_ps_32x32_neon);984
+ p.cuBLOCK_64x64.add_psALIGNED = PFX(pixel_add_ps_64x64_neon);985
+986
+ // chroma add_ps987
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.add_psNONALIGNED = PFX(pixel_add_ps_4x4_neon);988
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.add_psNONALIGNED = PFX(pixel_add_ps_8x8_neon);989
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.add_psNONALIGNED = PFX(pixel_add_ps_16x16_neon);990
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.add_psNONALIGNED = PFX(pixel_add_ps_32x32_neon);991
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.add_psNONALIGNED = PFX(pixel_add_ps_4x8_neon);992
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.add_psNONALIGNED = PFX(pixel_add_ps_8x16_neon);993
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.add_psNONALIGNED = PFX(pixel_add_ps_16x32_neon);994
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.add_psNONALIGNED = PFX(pixel_add_ps_32x64_neon);995
+996
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.add_psALIGNED = PFX(pixel_add_ps_4x4_neon);997
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.add_psALIGNED = PFX(pixel_add_ps_8x8_neon);998
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.add_psALIGNED = PFX(pixel_add_ps_16x16_neon);999
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.add_psALIGNED = PFX(pixel_add_ps_32x32_neon);1000
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.add_psALIGNED = PFX(pixel_add_ps_4x8_neon);1001
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.add_psALIGNED = PFX(pixel_add_ps_8x16_neon);1002
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.add_psALIGNED = PFX(pixel_add_ps_16x32_neon);1003
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.add_psALIGNED = PFX(pixel_add_ps_32x64_neon);1004
+1005
+ //scale2D_64to321006
+ p.scale2D_64to32 = PFX(scale2D_64to32_neon);1007
+1008
+ // scale1D_128to641009
+ p.scale1D_128to64NONALIGNED = PFX(scale1D_128to64_neon);1010
+ p.scale1D_128to64ALIGNED = PFX(scale1D_128to64_neon);1011
+1012
+ // planecopy1013
+ p.planecopy_cp = PFX(pixel_planecopy_cp_neon);1014
+1015
+ // satd1016
+ ALL_LUMA_PU(satd, pixel_satd, neon);1017
+1018
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.satd = PFX(pixel_satd_4x4_neon);1019
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.satd = PFX(pixel_satd_8x8_neon);1020
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.satd = PFX(pixel_satd_16x16_neon);1021
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.satd = PFX(pixel_satd_32x32_neon);1022
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.satd = PFX(pixel_satd_8x4_neon);1023
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.satd = PFX(pixel_satd_4x8_neon);1024
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.satd = PFX(pixel_satd_16x8_neon);1025
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.satd = PFX(pixel_satd_8x16_neon);1026
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.satd = PFX(pixel_satd_32x16_neon);1027
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.satd = PFX(pixel_satd_16x32_neon);1028
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.satd = PFX(pixel_satd_16x12_neon);1029
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.satd = PFX(pixel_satd_12x16_neon);1030
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.satd = PFX(pixel_satd_16x4_neon);1031
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.satd = PFX(pixel_satd_4x16_neon);1032
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.satd = PFX(pixel_satd_32x24_neon);1033
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.satd = PFX(pixel_satd_24x32_neon);1034
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.satd = PFX(pixel_satd_32x8_neon);1035
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.satd = PFX(pixel_satd_8x32_neon);1036
+1037
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.satd = PFX(pixel_satd_4x8_neon);1038
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.satd = PFX(pixel_satd_8x16_neon);1039
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.satd = PFX(pixel_satd_16x32_neon);1040
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.satd = PFX(pixel_satd_32x64_neon);1041
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.satd = PFX(pixel_satd_4x4_neon);1042
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.satd = PFX(pixel_satd_8x8_neon);1043
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.satd = PFX(pixel_satd_4x16_neon);1044
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.satd = PFX(pixel_satd_16x16_neon);1045
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.satd = PFX(pixel_satd_8x32_neon);1046
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.satd = PFX(pixel_satd_32x32_neon);1047
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.satd = PFX(pixel_satd_16x64_neon);1048
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.satd = PFX(pixel_satd_8x12_neon);1049
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.satd = PFX(pixel_satd_8x4_neon);1050
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.satd = PFX(pixel_satd_16x24_neon);1051
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.satd = PFX(pixel_satd_12x32_neon);1052
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.satd = PFX(pixel_satd_16x8_neon);1053
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.satd = PFX(pixel_satd_4x32_neon);1054
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.satd = PFX(pixel_satd_32x48_neon);1055
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.satd = PFX(pixel_satd_24x64_neon);1056
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.satd = PFX(pixel_satd_32x16_neon);1057
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.satd = PFX(pixel_satd_8x64_neon);1058
+1059
+ // sa8d1060
+ p.cuBLOCK_4x4.sa8d = PFX(pixel_satd_4x4_neon);1061
+ p.cuBLOCK_8x8.sa8d = PFX(pixel_sa8d_8x8_neon);1062
+ p.cuBLOCK_16x16.sa8d = PFX(pixel_sa8d_16x16_neon);1063
+ p.cuBLOCK_32x32.sa8d = PFX(pixel_sa8d_32x32_neon);1064
+ p.cuBLOCK_64x64.sa8d = PFX(pixel_sa8d_64x64_neon);1065
+ p.chromaX265_CSP_I420.cuBLOCK_8x8.sa8d = PFX(pixel_satd_4x4_neon);1066
+ p.chromaX265_CSP_I420.cuBLOCK_16x16.sa8d = PFX(pixel_sa8d_16x16_neon);1067
+ p.chromaX265_CSP_I420.cuBLOCK_32x32.sa8d = PFX(pixel_sa8d_32x32_neon);1068
+ p.chromaX265_CSP_I420.cuBLOCK_64x64.sa8d = PFX(pixel_sa8d_64x64_neon);1069
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sa8d = PFX(pixel_sa8d_8x16_neon);1070
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sa8d = PFX(pixel_sa8d_16x32_neon);1071
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sa8d = PFX(pixel_sa8d_32x64_neon);1072
+1073
+ // dequant_scaling1074
+ p.dequant_scaling = PFX(dequant_scaling_neon);1075
+ p.dequant_normal = PFX(dequant_normal_neon);1076
+1077
+ // ssim_4x4x2_core1078
+ p.ssim_4x4x2_core = PFX(ssim_4x4x2_core_neon);1079
+1080
+ // ssimDist1081
+ p.cuBLOCK_4x4.ssimDist = PFX(ssimDist4_neon);1082
+ p.cuBLOCK_8x8.ssimDist = PFX(ssimDist8_neon);1083
+ p.cuBLOCK_16x16.ssimDist = PFX(ssimDist16_neon);1084
+ p.cuBLOCK_32x32.ssimDist = PFX(ssimDist32_neon);1085
+ p.cuBLOCK_64x64.ssimDist = PFX(ssimDist64_neon);1086
+1087
+ // normFact1088
+ p.cuBLOCK_8x8.normFact = PFX(normFact8_neon);1089
+ p.cuBLOCK_16x16.normFact = PFX(normFact16_neon);1090
+ p.cuBLOCK_32x32.normFact = PFX(normFact32_neon);1091
+ p.cuBLOCK_64x64.normFact = PFX(normFact64_neon);1092
+1093
+ // psy_cost_pp1094
+ p.cuBLOCK_4x4.psy_cost_pp = PFX(psyCost_4x4_neon);1095
+1096
+ p.weight_pp = PFX(weight_pp_neon);1097
+#if !defined(__APPLE__)1098
+ p.scanPosLast = PFX(scanPosLast_neon);1099
#endif1100
+ p.costCoeffNxN = PFX(costCoeffNxN_neon);1101
#endif1102
- }1103
-}1104
1105
+ // quant1106
+ p.quant = PFX(quant_neon);1107
+ p.nquant = PFX(nquant_neon);1108
+}1109
1110
-void setupAssemblyPrimitives(EncoderPrimitives &p, int cpuMask) 1111
+#if defined(HAVE_SVE2) || defined(HAVE_SVE)1112
+void setupSvePrimitives(EncoderPrimitives &p)1113
{1114
- if (cpuMask & X265_CPU_NEON)1115
- {1116
- p.puLUMA_4x4.satd = PFX(pixel_satd_4x4_neon);1117
- p.puLUMA_4x8.satd = PFX(pixel_satd_4x8_neon);1118
- p.puLUMA_4x16.satd = PFX(pixel_satd_4x16_neon);1119
- p.puLUMA_8x4.satd = PFX(pixel_satd_8x4_neon);1120
- p.puLUMA_8x8.satd = PFX(pixel_satd_8x8_neon);1121
- p.puLUMA_12x16.satd = PFX(pixel_satd_12x16_neon);1122
- 1123
- p.chromaX265_CSP_I420.puCHROMA_420_4x4.satd = PFX(pixel_satd_4x4_neon);1124
- p.chromaX265_CSP_I420.puCHROMA_420_4x8.satd = PFX(pixel_satd_4x8_neon);1125
- p.chromaX265_CSP_I420.puCHROMA_420_4x16.satd = PFX(pixel_satd_4x16_neon);1126
- p.chromaX265_CSP_I420.puCHROMA_420_8x4.satd = PFX(pixel_satd_8x4_neon);1127
- p.chromaX265_CSP_I420.puCHROMA_420_8x8.satd = PFX(pixel_satd_8x8_neon);1128
- p.chromaX265_CSP_I420.puCHROMA_420_12x16.satd = PFX(pixel_satd_12x16_neon);1129
- 1130
- p.chromaX265_CSP_I422.puCHROMA_422_4x4.satd = PFX(pixel_satd_4x4_neon);1131
- p.chromaX265_CSP_I422.puCHROMA_422_4x8.satd = PFX(pixel_satd_4x8_neon);1132
- p.chromaX265_CSP_I422.puCHROMA_422_4x16.satd = PFX(pixel_satd_4x16_neon);1133
- p.chromaX265_CSP_I422.puCHROMA_422_4x32.satd = PFX(pixel_satd_4x32_neon);1134
- p.chromaX265_CSP_I422.puCHROMA_422_8x4.satd = PFX(pixel_satd_8x4_neon);1135
- p.chromaX265_CSP_I422.puCHROMA_422_8x8.satd = PFX(pixel_satd_8x8_neon);1136
- p.chromaX265_CSP_I422.puCHROMA_422_12x32.satd = PFX(pixel_satd_12x32_neon);1137
-1138
- p.puLUMA_4x4.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_4x4_neon);1139
- p.puLUMA_4x8.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_4x8_neon);1140
- p.puLUMA_4x16.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_4x16_neon);1141
- p.puLUMA_8x4.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_8x4_neon);1142
- p.puLUMA_8x8.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_8x8_neon);1143
- p.puLUMA_8x16.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_8x16_neon);1144
- p.puLUMA_8x32.pixelavg_ppNONALIGNED = PFX(pixel_avg_pp_8x32_neon);1145
-1146
- p.puLUMA_4x4.pixelavg_ppALIGNED = PFX(pixel_avg_pp_4x4_neon);1147
- p.puLUMA_4x8.pixelavg_ppALIGNED = PFX(pixel_avg_pp_4x8_neon);1148
- p.puLUMA_4x16.pixelavg_ppALIGNED = PFX(pixel_avg_pp_4x16_neon);1149
- p.puLUMA_8x4.pixelavg_ppALIGNED = PFX(pixel_avg_pp_8x4_neon);1150
- p.puLUMA_8x8.pixelavg_ppALIGNED = PFX(pixel_avg_pp_8x8_neon);1151
- p.puLUMA_8x16.pixelavg_ppALIGNED = PFX(pixel_avg_pp_8x16_neon);1152
- p.puLUMA_8x32.pixelavg_ppALIGNED = PFX(pixel_avg_pp_8x32_neon);1153
-1154
- p.puLUMA_8x4.sad_x3 = PFX(sad_x3_8x4_neon);1155
- p.puLUMA_8x8.sad_x3 = PFX(sad_x3_8x8_neon);1156
- p.puLUMA_8x16.sad_x3 = PFX(sad_x3_8x16_neon);1157
- p.puLUMA_8x32.sad_x3 = PFX(sad_x3_8x32_neon);1158
-1159
- p.puLUMA_8x4.sad_x4 = PFX(sad_x4_8x4_neon);1160
- p.puLUMA_8x8.sad_x4 = PFX(sad_x4_8x8_neon);1161
- p.puLUMA_8x16.sad_x4 = PFX(sad_x4_8x16_neon);1162
- p.puLUMA_8x32.sad_x4 = PFX(sad_x4_8x32_neon);1163
-1164
- // quant1165
- p.quant = PFX(quant_neon);1166
- // luma_hps1167
- p.puLUMA_4x4.luma_hps = PFX(interp_8tap_horiz_ps_4x4_neon);1168
- p.puLUMA_4x8.luma_hps = PFX(interp_8tap_horiz_ps_4x8_neon);1169
- p.puLUMA_4x16.luma_hps = PFX(interp_8tap_horiz_ps_4x16_neon);1170
- p.puLUMA_8x4.luma_hps = PFX(interp_8tap_horiz_ps_8x4_neon);1171
- p.puLUMA_8x8.luma_hps = PFX(interp_8tap_horiz_ps_8x8_neon);1172
- p.puLUMA_8x16.luma_hps = PFX(interp_8tap_horiz_ps_8x16_neon);1173
- p.puLUMA_8x32.luma_hps = PFX(interp_8tap_horiz_ps_8x32_neon);1174
- p.puLUMA_12x16.luma_hps = PFX(interp_8tap_horiz_ps_12x16_neon);1175
- p.puLUMA_24x32.luma_hps = PFX(interp_8tap_horiz_ps_24x32_neon);1176
-#if !AUTO_VECTORIZE || GCC_VERSION < GCC_5_1_0 /* gcc_version < gcc-5.1.0 */1177
- p.puLUMA_16x4.luma_hps = PFX(interp_8tap_horiz_ps_16x4_neon);1178
- p.puLUMA_16x8.luma_hps = PFX(interp_8tap_horiz_ps_16x8_neon);1179
- p.puLUMA_16x12.luma_hps = PFX(interp_8tap_horiz_ps_16x12_neon);1180
- p.puLUMA_16x16.luma_hps = PFX(interp_8tap_horiz_ps_16x16_neon);1181
- p.puLUMA_16x32.luma_hps = PFX(interp_8tap_horiz_ps_16x32_neon);1182
- p.puLUMA_16x64.luma_hps = PFX(interp_8tap_horiz_ps_16x64_neon);1183
- p.puLUMA_32x8.luma_hps = PFX(interp_8tap_horiz_ps_32x8_neon);1184
- p.puLUMA_32x16.luma_hps = PFX(interp_8tap_horiz_ps_32x16_neon);1185
- p.puLUMA_32x24.luma_hps = PFX(interp_8tap_horiz_ps_32x24_neon);1186
- p.puLUMA_32x32.luma_hps = PFX(interp_8tap_horiz_ps_32x32_neon);1187
- p.puLUMA_32x64.luma_hps = PFX(interp_8tap_horiz_ps_32x64_neon);1188
- p.puLUMA_48x64.luma_hps = PFX(interp_8tap_horiz_ps_48x64_neon);1189
- p.puLUMA_64x16.luma_hps = PFX(interp_8tap_horiz_ps_64x16_neon);1190
- p.puLUMA_64x32.luma_hps = PFX(interp_8tap_horiz_ps_64x32_neon);1191
- p.puLUMA_64x48.luma_hps = PFX(interp_8tap_horiz_ps_64x48_neon);1192
- p.puLUMA_64x64.luma_hps = PFX(interp_8tap_horiz_ps_64x64_neon);1193
-#endif1194
-1195
- p.puLUMA_8x4.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_8x4>;1196
- p.puLUMA_8x8.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_8x8>;1197
- p.puLUMA_8x16.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_8x16>;1198
- p.puLUMA_8x32.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_8x32>;1199
- p.puLUMA_12x16.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_12x16>;1200
-#if !AUTO_VECTORIZE || GCC_VERSION < GCC_5_1_0 /* gcc_version < gcc-5.1.0 */1201
- p.puLUMA_16x4.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_16x4>;1202
- p.puLUMA_16x8.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_16x8>;1203
- p.puLUMA_16x12.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_16x12>;1204
- p.puLUMA_16x16.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_16x16>;1205
- p.puLUMA_16x32.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_16x32>;1206
- p.puLUMA_16x64.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_16x64>;1207
- p.puLUMA_32x16.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_32x16>;1208
- p.puLUMA_32x24.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_32x24>;1209
- p.puLUMA_32x32.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_32x32>;1210
- p.puLUMA_32x64.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_32x64>;1211
- p.puLUMA_48x64.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_48x64>;1212
- p.puLUMA_64x16.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_64x16>;1213
- p.puLUMA_64x32.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_64x32>;1214
- p.puLUMA_64x48.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_64x48>;1215
- p.puLUMA_64x64.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_64x64>;1216
-#if !AUTO_VECTORIZE || GCC_VERSION < GCC_4_9_0 /* gcc_version < gcc-4.9.0 */1217
- p.puLUMA_4x4.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_4x4>;1218
- p.puLUMA_4x8.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_4x8>;1219
- p.puLUMA_4x16.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_4x16>;1220
- p.puLUMA_24x32.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_24x32>;1221
- p.puLUMA_32x8.luma_hvpp = interp_8tap_hv_pp_cpu<LUMA_32x8>;1222
+ // When these primitives will use SVE/SVE2 instructions set,1223
+ // change the following definitions to point to the SVE/SVE2 implementation1224
+ setupPixelPrimitives_neon(p);1225
+ setupFilterPrimitives_neon(p);1226
+ setupDCTPrimitives_neon(p);1227
+ setupLoopFilterPrimitives_neon(p);1228
+ setupIntraPrimitives_neon(p);1229
+1230
+ CHROMA_420_PU_FILTER_PIXEL_TO_SHORT_NEON(p2sNONALIGNED);1231
+ CHROMA_420_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1232
+ CHROMA_422_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1233
+ CHROMA_422_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1234
+ CHROMA_444_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1235
+ CHROMA_444_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1236
+ LUMA_PU_NEON_FILTER_PIXEL_TO_SHORT(convert_p2sALIGNED);1237
+ LUMA_PU_SVE_FILTER_PIXEL_TO_SHORT(convert_p2sALIGNED);1238
+ CHROMA_420_PU_FILTER_PIXEL_TO_SHORT_NEON(p2sALIGNED);1239
+ CHROMA_420_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1240
+ CHROMA_422_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1241
+ CHROMA_422_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1242
+ CHROMA_444_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1243
+ CHROMA_444_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1244
+ LUMA_PU_NEON_FILTER_PIXEL_TO_SHORT(convert_p2sNONALIGNED);1245
+ LUMA_PU_SVE_FILTER_PIXEL_TO_SHORT(convert_p2sNONALIGNED);1246
+1247
+#if !HIGH_BIT_DEPTH1248
+ ALL_LUMA_PU(luma_vpp, interp_8tap_vert_pp, neon);1249
+ ALL_LUMA_PU(luma_vsp, interp_8tap_vert_sp, neon);1250
+ ALL_LUMA_PU(luma_vps, interp_8tap_vert_ps, neon);1251
+ ALL_LUMA_PU(luma_hpp, interp_horiz_pp, neon);1252
+ ALL_LUMA_PU(luma_hps, interp_horiz_ps, neon);1253
+ ALL_LUMA_PU(luma_vss, interp_8tap_vert_ss, neon);1254
+ ALL_LUMA_PU_T(luma_hvpp, interp_8tap_hv_pp_cpu);1255
+ ALL_CHROMA_420_VERT_FILTERS(neon);1256
+ CHROMA_422_VERT_FILTERS_NEON();1257
+ CHROMA_422_VERT_FILTERS_CAN_USE_SVE2(neon);1258
+ ALL_CHROMA_444_VERT_FILTERS(neon);1259
+ ALL_CHROMA_420_FILTERS(neon);1260
+ ALL_CHROMA_422_FILTERS(neon);1261
+ ALL_CHROMA_444_FILTERS(neon);1262
+1263
+1264
+ // Blockcopy_pp1265
+ LUMA_PU_NEON_1(copy_pp, blockcopy_pp);1266
+ LUMA_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(copy_pp, blockcopy_pp);1267
+ CHROMA_420_PU_NEON_1(copy_pp, blockcopy_pp);1268
+ CHROMA_420_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(copy_pp, blockcopy_pp);1269
+ CHROMA_422_PU_NEON_1(copy_pp, blockcopy_pp);1270
+ CHROMA_422_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(copy_pp, blockcopy_pp);1271
+ p.cuBLOCK_4x4.copy_pp = PFX(blockcopy_pp_4x4_neon);1272
+ p.cuBLOCK_8x8.copy_pp = PFX(blockcopy_pp_8x8_neon);1273
+ p.cuBLOCK_16x16.copy_pp = PFX(blockcopy_pp_16x16_neon);1274
+ p.cuBLOCK_32x32.copy_pp = PFX(blockcopy_pp_32x32_sve);1275
+ p.cuBLOCK_64x64.copy_pp = PFX(blockcopy_pp_64x64_sve);1276
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_pp = PFX(blockcopy_pp_4x4_neon);1277
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_pp = PFX(blockcopy_pp_8x8_neon);1278
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_pp = PFX(blockcopy_pp_16x16_neon);1279
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_pp = PFX(blockcopy_pp_32x32_sve);1280
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_pp = PFX(blockcopy_pp_4x8_neon);1281
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_pp = PFX(blockcopy_pp_8x16_neon);1282
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_pp = PFX(blockcopy_pp_16x32_neon);1283
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_pp = PFX(blockcopy_pp_32x64_sve);1284
+1285
+#endif // !HIGH_BIT_DEPTH1286
+1287
+ // Blockcopy_ss1288
+ p.cuBLOCK_4x4.copy_ss = PFX(blockcopy_ss_4x4_neon);1289
+ p.cuBLOCK_8x8.copy_ss = PFX(blockcopy_ss_8x8_neon);1290
+ p.cuBLOCK_16x16.copy_ss = PFX(blockcopy_ss_16x16_sve);1291
+ p.cuBLOCK_32x32.copy_ss = PFX(blockcopy_ss_32x32_sve);1292
+ p.cuBLOCK_64x64.copy_ss = PFX(blockcopy_ss_64x64_sve);1293
+1294
+ // Blockcopy_ps1295
+ p.cuBLOCK_4x4.copy_ps = PFX(blockcopy_ps_4x4_neon);1296
+ p.cuBLOCK_8x8.copy_ps = PFX(blockcopy_ps_8x8_neon);1297
+ p.cuBLOCK_16x16.copy_ps = PFX(blockcopy_ps_16x16_sve);1298
+ p.cuBLOCK_32x32.copy_ps = PFX(blockcopy_ps_32x32_sve);1299
+ p.cuBLOCK_64x64.copy_ps = PFX(blockcopy_ps_64x64_sve);1300
+1301
+ // Blockcopy_sp1302
+ p.cuBLOCK_4x4.copy_sp = PFX(blockcopy_sp_4x4_sve);1303
+ p.cuBLOCK_8x8.copy_sp = PFX(blockcopy_sp_8x8_sve);1304
+ p.cuBLOCK_16x16.copy_sp = PFX(blockcopy_sp_16x16_sve);1305
+ p.cuBLOCK_32x32.copy_sp = PFX(blockcopy_sp_32x32_sve);1306
+ p.cuBLOCK_64x64.copy_sp = PFX(blockcopy_sp_64x64_neon);1307
+1308
+ // chroma blockcopy_ss1309
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_ss = PFX(blockcopy_ss_4x4_neon);1310
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_ss = PFX(blockcopy_ss_8x8_neon);1311
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_ss = PFX(blockcopy_ss_16x16_sve);1312
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_ss = PFX(blockcopy_ss_32x32_sve);1313
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_ss = PFX(blockcopy_ss_4x8_neon);1314
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_ss = PFX(blockcopy_ss_8x16_neon);1315
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_ss = PFX(blockcopy_ss_16x32_sve);1316
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_ss = PFX(blockcopy_ss_32x64_sve);1317
+1318
+ // chroma blockcopy_ps1319
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_ps = PFX(blockcopy_ps_4x4_neon);1320
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_ps = PFX(blockcopy_ps_8x8_neon);1321
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_ps = PFX(blockcopy_ps_16x16_sve);1322
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_ps = PFX(blockcopy_ps_32x32_sve);1323
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_ps = PFX(blockcopy_ps_4x8_sve);1324
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_ps = PFX(blockcopy_ps_8x16_sve);1325
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_ps = PFX(blockcopy_ps_16x32_sve);1326
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_ps = PFX(blockcopy_ps_32x64_sve);1327
+1328
+ // chroma blockcopy_sp1329
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_sp = PFX(blockcopy_sp_4x4_sve);1330
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_sp = PFX(blockcopy_sp_8x8_sve);1331
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_sp = PFX(blockcopy_sp_16x16_sve);1332
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_sp = PFX(blockcopy_sp_32x32_sve);1333
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_sp = PFX(blockcopy_sp_4x8_sve);1334
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_sp = PFX(blockcopy_sp_8x16_sve);1335
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_sp = PFX(blockcopy_sp_16x32_sve);1336
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_sp = PFX(blockcopy_sp_32x64_sve);1337
+1338
+ // Block_fill1339
+ LUMA_TU_NEON(blockfill_sALIGNED, blockfill_s);1340
+ LUMA_TU_CAN_USE_SVE(blockfill_sALIGNED, blockfill_s);1341
+ LUMA_TU_NEON(blockfill_sNONALIGNED, blockfill_s);1342
+ LUMA_TU_CAN_USE_SVE(blockfill_sNONALIGNED, blockfill_s);1343
+1344
+ // copy_count1345
+ p.cuBLOCK_4x4.copy_cnt = PFX(copy_cnt_4_neon);1346
+ p.cuBLOCK_8x8.copy_cnt = PFX(copy_cnt_8_neon);1347
+ p.cuBLOCK_16x16.copy_cnt = PFX(copy_cnt_16_neon);1348
+ p.cuBLOCK_32x32.copy_cnt = PFX(copy_cnt_32_neon);1349
+1350
+ // count nonzero1351
+ p.cuBLOCK_4x4.count_nonzero = PFX(count_nonzero_4_neon);1352
+ p.cuBLOCK_8x8.count_nonzero = PFX(count_nonzero_8_neon);1353
+ p.cuBLOCK_16x16.count_nonzero = PFX(count_nonzero_16_neon);1354
+ p.cuBLOCK_32x32.count_nonzero = PFX(count_nonzero_32_neon);1355
+1356
+ // cpy2Dto1D_shl1357
+ p.cuBLOCK_4x4.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_4x4_neon);1358
+ p.cuBLOCK_8x8.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_8x8_neon);1359
+ p.cuBLOCK_16x16.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_16x16_sve);1360
+ p.cuBLOCK_32x32.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_32x32_sve);1361
+ p.cuBLOCK_64x64.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_64x64_sve);1362
+1363
+ // cpy2Dto1D_shr1364
+ p.cuBLOCK_4x4.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_4x4_neon);1365
+ p.cuBLOCK_8x8.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_8x8_neon);1366
+ p.cuBLOCK_16x16.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_16x16_sve);1367
+ p.cuBLOCK_32x32.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_32x32_sve);1368
+1369
+ // cpy1Dto2D_shl1370
+ p.cuBLOCK_4x4.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_4x4_neon);1371
+ p.cuBLOCK_8x8.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_8x8_neon);1372
+ p.cuBLOCK_16x16.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_16x16_sve);1373
+ p.cuBLOCK_32x32.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_32x32_sve);1374
+ p.cuBLOCK_64x64.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_64x64_sve);1375
+1376
+ p.cuBLOCK_4x4.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_4x4_neon);1377
+ p.cuBLOCK_8x8.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_8x8_neon);1378
+ p.cuBLOCK_16x16.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_16x16_sve);1379
+ p.cuBLOCK_32x32.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_32x32_sve);1380
+ p.cuBLOCK_64x64.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_64x64_sve);1381
+1382
+ // cpy1Dto2D_shr1383
+ p.cuBLOCK_4x4.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_4x4_neon);1384
+ p.cuBLOCK_8x8.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_8x8_neon);1385
+ p.cuBLOCK_16x16.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_16x16_sve);1386
+ p.cuBLOCK_32x32.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_32x32_sve);1387
+ p.cuBLOCK_64x64.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_64x64_sve);1388
+1389
+#if !HIGH_BIT_DEPTH1390
+ // pixel_avg_pp1391
+ ALL_LUMA_PU(pixelavg_ppNONALIGNED, pixel_avg_pp, neon);1392
+ ALL_LUMA_PU(pixelavg_ppALIGNED, pixel_avg_pp, neon);1393
+1394
+ // addAvg1395
+ ALL_LUMA_PU(addAvgNONALIGNED, addAvg, neon);1396
+ ALL_LUMA_PU(addAvgALIGNED, addAvg, neon);1397
+ ALL_CHROMA_420_PU(addAvgNONALIGNED, addAvg, neon);1398
+ ALL_CHROMA_422_PU(addAvgNONALIGNED, addAvg, neon);1399
+ ALL_CHROMA_420_PU(addAvgALIGNED, addAvg, neon);1400
+ ALL_CHROMA_422_PU(addAvgALIGNED, addAvg, neon);1401
+1402
+ // sad1403
+ ALL_LUMA_PU(sad, pixel_sad, neon);1404
+ ALL_LUMA_PU(sad_x3, sad_x3, neon);1405
+ ALL_LUMA_PU(sad_x4, sad_x4, neon);1406
+1407
+ // sse_pp1408
+ p.cuBLOCK_4x4.sse_pp = PFX(pixel_sse_pp_4x4_sve);1409
+ p.cuBLOCK_8x8.sse_pp = PFX(pixel_sse_pp_8x8_neon);1410
+ p.cuBLOCK_16x16.sse_pp = PFX(pixel_sse_pp_16x16_neon);1411
+ p.cuBLOCK_32x32.sse_pp = PFX(pixel_sse_pp_32x32_neon);1412
+ p.cuBLOCK_64x64.sse_pp = PFX(pixel_sse_pp_64x64_neon);1413
+1414
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.sse_pp = PFX(pixel_sse_pp_4x4_sve);1415
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.sse_pp = PFX(pixel_sse_pp_8x8_neon);1416
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.sse_pp = PFX(pixel_sse_pp_16x16_neon);1417
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.sse_pp = PFX(pixel_sse_pp_32x32_neon);1418
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.sse_pp = PFX(pixel_sse_pp_4x8_sve);1419
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sse_pp = PFX(pixel_sse_pp_8x16_neon);1420
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sse_pp = PFX(pixel_sse_pp_16x32_neon);1421
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sse_pp = PFX(pixel_sse_pp_32x64_neon);1422
+1423
+ // sse_ss1424
+ p.cuBLOCK_4x4.sse_ss = PFX(pixel_sse_ss_4x4_neon);1425
+ p.cuBLOCK_8x8.sse_ss = PFX(pixel_sse_ss_8x8_neon);1426
+ p.cuBLOCK_16x16.sse_ss = PFX(pixel_sse_ss_16x16_neon);1427
+ p.cuBLOCK_32x32.sse_ss = PFX(pixel_sse_ss_32x32_neon);1428
+ p.cuBLOCK_64x64.sse_ss = PFX(pixel_sse_ss_64x64_neon);1429
+1430
+ // ssd_s1431
+ p.cuBLOCK_4x4.ssd_sNONALIGNED = PFX(pixel_ssd_s_4x4_neon);1432
+ p.cuBLOCK_8x8.ssd_sNONALIGNED = PFX(pixel_ssd_s_8x8_neon);1433
+ p.cuBLOCK_16x16.ssd_sNONALIGNED = PFX(pixel_ssd_s_16x16_neon);1434
+ p.cuBLOCK_32x32.ssd_sNONALIGNED = PFX(pixel_ssd_s_32x32_neon);1435
+1436
+ p.cuBLOCK_4x4.ssd_sALIGNED = PFX(pixel_ssd_s_4x4_neon);1437
+ p.cuBLOCK_8x8.ssd_sALIGNED = PFX(pixel_ssd_s_8x8_neon);1438
+ p.cuBLOCK_16x16.ssd_sALIGNED = PFX(pixel_ssd_s_16x16_neon);1439
+ p.cuBLOCK_32x32.ssd_sALIGNED = PFX(pixel_ssd_s_32x32_neon);1440
+1441
+ // pixel_var1442
+ p.cuBLOCK_8x8.var = PFX(pixel_var_8x8_neon);1443
+ p.cuBLOCK_16x16.var = PFX(pixel_var_16x16_neon);1444
+ p.cuBLOCK_32x32.var = PFX(pixel_var_32x32_neon);1445
+ p.cuBLOCK_64x64.var = PFX(pixel_var_64x64_neon);1446
+1447
+ // calc_Residual1448
+ p.cuBLOCK_4x4.calcresidualNONALIGNED = PFX(getResidual4_neon);1449
+ p.cuBLOCK_8x8.calcresidualNONALIGNED = PFX(getResidual8_neon);1450
+ p.cuBLOCK_16x16.calcresidualNONALIGNED = PFX(getResidual16_neon);1451
+ p.cuBLOCK_32x32.calcresidualNONALIGNED = PFX(getResidual32_neon);1452
+1453
+ p.cuBLOCK_4x4.calcresidualALIGNED = PFX(getResidual4_neon);1454
+ p.cuBLOCK_8x8.calcresidualALIGNED = PFX(getResidual8_neon);1455
+ p.cuBLOCK_16x16.calcresidualALIGNED = PFX(getResidual16_neon);1456
+ p.cuBLOCK_32x32.calcresidualALIGNED = PFX(getResidual32_neon);1457
+1458
+ // pixel_sub_ps1459
+ p.cuBLOCK_4x4.sub_ps = PFX(pixel_sub_ps_4x4_neon);1460
+ p.cuBLOCK_8x8.sub_ps = PFX(pixel_sub_ps_8x8_neon);1461
+ p.cuBLOCK_16x16.sub_ps = PFX(pixel_sub_ps_16x16_neon);1462
+ p.cuBLOCK_32x32.sub_ps = PFX(pixel_sub_ps_32x32_neon);1463
+ p.cuBLOCK_64x64.sub_ps = PFX(pixel_sub_ps_64x64_neon);1464
+1465
+ // chroma sub_ps1466
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.sub_ps = PFX(pixel_sub_ps_4x4_neon);1467
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.sub_ps = PFX(pixel_sub_ps_8x8_neon);1468
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.sub_ps = PFX(pixel_sub_ps_16x16_neon);1469
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.sub_ps = PFX(pixel_sub_ps_32x32_neon);1470
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.sub_ps = PFX(pixel_sub_ps_4x8_neon);1471
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sub_ps = PFX(pixel_sub_ps_8x16_sve);1472
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sub_ps = PFX(pixel_sub_ps_16x32_neon);1473
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sub_ps = PFX(pixel_sub_ps_32x64_neon);1474
+1475
+ // pixel_add_ps1476
+ p.cuBLOCK_4x4.add_psNONALIGNED = PFX(pixel_add_ps_4x4_neon);1477
+ p.cuBLOCK_8x8.add_psNONALIGNED = PFX(pixel_add_ps_8x8_neon);1478
+ p.cuBLOCK_16x16.add_psNONALIGNED = PFX(pixel_add_ps_16x16_neon);1479
+ p.cuBLOCK_32x32.add_psNONALIGNED = PFX(pixel_add_ps_32x32_neon);1480
+ p.cuBLOCK_64x64.add_psNONALIGNED = PFX(pixel_add_ps_64x64_neon);1481
+1482
+ p.cuBLOCK_4x4.add_psALIGNED = PFX(pixel_add_ps_4x4_neon);1483
+ p.cuBLOCK_8x8.add_psALIGNED = PFX(pixel_add_ps_8x8_neon);1484
+ p.cuBLOCK_16x16.add_psALIGNED = PFX(pixel_add_ps_16x16_neon);1485
+ p.cuBLOCK_32x32.add_psALIGNED = PFX(pixel_add_ps_32x32_neon);1486
+ p.cuBLOCK_64x64.add_psALIGNED = PFX(pixel_add_ps_64x64_neon);1487
+1488
+ // chroma add_ps1489
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.add_psNONALIGNED = PFX(pixel_add_ps_4x4_neon);1490
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.add_psNONALIGNED = PFX(pixel_add_ps_8x8_neon);1491
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.add_psNONALIGNED = PFX(pixel_add_ps_16x16_neon);1492
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.add_psNONALIGNED = PFX(pixel_add_ps_32x32_neon);1493
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.add_psNONALIGNED = PFX(pixel_add_ps_4x8_neon);1494
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.add_psNONALIGNED = PFX(pixel_add_ps_8x16_neon);1495
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.add_psNONALIGNED = PFX(pixel_add_ps_16x32_neon);1496
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.add_psNONALIGNED = PFX(pixel_add_ps_32x64_neon);1497
+1498
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.add_psALIGNED = PFX(pixel_add_ps_4x4_neon);1499
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.add_psALIGNED = PFX(pixel_add_ps_8x8_neon);1500
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.add_psALIGNED = PFX(pixel_add_ps_16x16_neon);1501
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.add_psALIGNED = PFX(pixel_add_ps_32x32_neon);1502
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.add_psALIGNED = PFX(pixel_add_ps_4x8_neon);1503
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.add_psALIGNED = PFX(pixel_add_ps_8x16_neon);1504
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.add_psALIGNED = PFX(pixel_add_ps_16x32_neon);1505
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.add_psALIGNED = PFX(pixel_add_ps_32x64_neon);1506
+1507
+ //scale2D_64to321508
+ p.scale2D_64to32 = PFX(scale2D_64to32_neon);1509
+1510
+ // scale1D_128to641511
+ p.scale1D_128to64NONALIGNED = PFX(scale1D_128to64_neon);1512
+ p.scale1D_128to64ALIGNED = PFX(scale1D_128to64_neon);1513
+1514
+ // planecopy1515
+ p.planecopy_cp = PFX(pixel_planecopy_cp_neon);1516
+1517
+ // satd1518
+ p.puLUMA_4x4.satd = PFX(pixel_satd_4x4_sve);1519
+ p.puLUMA_8x8.satd = PFX(pixel_satd_8x8_neon);1520
+ p.puLUMA_16x16.satd = PFX(pixel_satd_16x16_neon);1521
+ p.puLUMA_32x32.satd = PFX(pixel_satd_32x32_sve);1522
+ p.puLUMA_64x64.satd = PFX(pixel_satd_64x64_neon);1523
+ p.puLUMA_8x4.satd = PFX(pixel_satd_8x4_sve);1524
+ p.puLUMA_4x8.satd = PFX(pixel_satd_4x8_neon);1525
+ p.puLUMA_16x8.satd = PFX(pixel_satd_16x8_neon);1526
+ p.puLUMA_8x16.satd = PFX(pixel_satd_8x16_neon);1527
+ p.puLUMA_16x32.satd = PFX(pixel_satd_16x32_neon);1528
+ p.puLUMA_32x16.satd = PFX(pixel_satd_32x16_sve);1529
+ p.puLUMA_64x32.satd = PFX(pixel_satd_64x32_neon);1530
+ p.puLUMA_32x64.satd = PFX(pixel_satd_32x64_neon);1531
+ p.puLUMA_16x12.satd = PFX(pixel_satd_16x12_neon);1532
+ p.puLUMA_12x16.satd = PFX(pixel_satd_12x16_neon);1533
+ p.puLUMA_16x4.satd = PFX(pixel_satd_16x4_neon);1534
+ p.puLUMA_4x16.satd = PFX(pixel_satd_4x16_neon);1535
+ p.puLUMA_32x24.satd = PFX(pixel_satd_32x24_neon);1536
+ p.puLUMA_24x32.satd = PFX(pixel_satd_24x32_neon);1537
+ p.puLUMA_32x8.satd = PFX(pixel_satd_32x8_neon);1538
+ p.puLUMA_8x32.satd = PFX(pixel_satd_8x32_neon);1539
+ p.puLUMA_64x48.satd = PFX(pixel_satd_64x48_sve);1540
+ p.puLUMA_48x64.satd = PFX(pixel_satd_48x64_neon);1541
+ p.puLUMA_64x16.satd = PFX(pixel_satd_64x16_neon);1542
+ p.puLUMA_16x64.satd = PFX(pixel_satd_16x64_neon);1543
+1544
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.satd = PFX(pixel_satd_4x4_sve);1545
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.satd = PFX(pixel_satd_8x8_neon);1546
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.satd = PFX(pixel_satd_16x16_neon);1547
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.satd = PFX(pixel_satd_32x32_neon);1548
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.satd = PFX(pixel_satd_8x4_sve);1549
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.satd = PFX(pixel_satd_4x8_neon);1550
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.satd = PFX(pixel_satd_16x8_neon);1551
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.satd = PFX(pixel_satd_8x16_neon);1552
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.satd = PFX(pixel_satd_32x16_neon);1553
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.satd = PFX(pixel_satd_16x32_neon);1554
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.satd = PFX(pixel_satd_16x12_neon);1555
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.satd = PFX(pixel_satd_12x16_neon);1556
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.satd = PFX(pixel_satd_16x4_neon);1557
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.satd = PFX(pixel_satd_4x16_neon);1558
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.satd = PFX(pixel_satd_32x24_neon);1559
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.satd = PFX(pixel_satd_24x32_neon);1560
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.satd = PFX(pixel_satd_32x8_neon);1561
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.satd = PFX(pixel_satd_8x32_neon);1562
+1563
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.satd = PFX(pixel_satd_4x8_neon);1564
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.satd = PFX(pixel_satd_8x16_neon);1565
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.satd = PFX(pixel_satd_16x32_neon);1566
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.satd = PFX(pixel_satd_32x64_neon);1567
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.satd = PFX(pixel_satd_4x4_sve);1568
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.satd = PFX(pixel_satd_8x8_neon);1569
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.satd = PFX(pixel_satd_4x16_neon);1570
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.satd = PFX(pixel_satd_16x16_neon);1571
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.satd = PFX(pixel_satd_8x32_neon);1572
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.satd = PFX(pixel_satd_32x32_neon);1573
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.satd = PFX(pixel_satd_16x64_neon);1574
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.satd = PFX(pixel_satd_8x12_sve);1575
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.satd = PFX(pixel_satd_8x4_sve);1576
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.satd = PFX(pixel_satd_16x24_neon);1577
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.satd = PFX(pixel_satd_12x32_neon);1578
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.satd = PFX(pixel_satd_16x8_neon);1579
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.satd = PFX(pixel_satd_4x32_neon);1580
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.satd = PFX(pixel_satd_32x48_neon);1581
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.satd = PFX(pixel_satd_24x64_neon);1582
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.satd = PFX(pixel_satd_32x16_neon);1583
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.satd = PFX(pixel_satd_8x64_neon);1584
+1585
+ // sa8d1586
+ p.cuBLOCK_4x4.sa8d = PFX(pixel_satd_4x4_sve);1587
+ p.cuBLOCK_8x8.sa8d = PFX(pixel_sa8d_8x8_neon);1588
+ p.cuBLOCK_16x16.sa8d = PFX(pixel_sa8d_16x16_neon);1589
+ p.cuBLOCK_32x32.sa8d = PFX(pixel_sa8d_32x32_neon);1590
+ p.cuBLOCK_64x64.sa8d = PFX(pixel_sa8d_64x64_neon);1591
+ p.chromaX265_CSP_I420.cuBLOCK_8x8.sa8d = PFX(pixel_satd_4x4_sve);1592
+ p.chromaX265_CSP_I420.cuBLOCK_16x16.sa8d = PFX(pixel_sa8d_16x16_neon);1593
+ p.chromaX265_CSP_I420.cuBLOCK_32x32.sa8d = PFX(pixel_sa8d_32x32_neon);1594
+ p.chromaX265_CSP_I420.cuBLOCK_64x64.sa8d = PFX(pixel_sa8d_64x64_neon);1595
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sa8d = PFX(pixel_sa8d_8x16_neon);1596
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sa8d = PFX(pixel_sa8d_16x32_neon);1597
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sa8d = PFX(pixel_sa8d_32x64_neon);1598
+1599
+ // dequant_scaling1600
+ p.dequant_scaling = PFX(dequant_scaling_neon);1601
+ p.dequant_normal = PFX(dequant_normal_neon);1602
+1603
+ // ssim_4x4x2_core1604
+ p.ssim_4x4x2_core = PFX(ssim_4x4x2_core_neon);1605
+1606
+ // ssimDist1607
+ p.cuBLOCK_4x4.ssimDist = PFX(ssimDist4_neon);1608
+ p.cuBLOCK_8x8.ssimDist = PFX(ssimDist8_neon);1609
+ p.cuBLOCK_16x16.ssimDist = PFX(ssimDist16_neon);1610
+ p.cuBLOCK_32x32.ssimDist = PFX(ssimDist32_neon);1611
+ p.cuBLOCK_64x64.ssimDist = PFX(ssimDist64_neon);1612
+1613
+ // normFact1614
+ p.cuBLOCK_8x8.normFact = PFX(normFact8_neon);1615
+ p.cuBLOCK_16x16.normFact = PFX(normFact16_neon);1616
+ p.cuBLOCK_32x32.normFact = PFX(normFact32_neon);1617
+ p.cuBLOCK_64x64.normFact = PFX(normFact64_neon);1618
+1619
+ // psy_cost_pp1620
+ p.cuBLOCK_4x4.psy_cost_pp = PFX(psyCost_4x4_neon);1621
+1622
+ p.weight_pp = PFX(weight_pp_neon);1623
+#if !defined(__APPLE__)1624
+ p.scanPosLast = PFX(scanPosLast_neon);1625
+#endif1626
+ p.costCoeffNxN = PFX(costCoeffNxN_neon);1627
#endif1628
+1629
+ // quant1630
+ p.quant = PFX(quant_sve);1631
+ p.nquant = PFX(nquant_neon);1632
+}1633
#endif1634
1635
+#if defined(HAVE_SVE2)1636
+void setupSve2Primitives(EncoderPrimitives &p)1637
+{1638
+ // When these primitives will use SVE/SVE2 instructions set,1639
+ // change the following definitions to point to the SVE/SVE2 implementation1640
+ setupPixelPrimitives_neon(p);1641
+ setupFilterPrimitives_neon(p);1642
+ setupDCTPrimitives_neon(p);1643
+ setupLoopFilterPrimitives_neon(p);1644
+ setupIntraPrimitives_neon(p);1645
+1646
+ CHROMA_420_PU_FILTER_PIXEL_TO_SHORT_NEON(p2sNONALIGNED);1647
+ CHROMA_420_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1648
+ CHROMA_422_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1649
+ CHROMA_422_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1650
+ CHROMA_444_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1651
+ CHROMA_444_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1652
+ LUMA_PU_NEON_FILTER_PIXEL_TO_SHORT(convert_p2sALIGNED);1653
+ LUMA_PU_SVE_FILTER_PIXEL_TO_SHORT(convert_p2sALIGNED);1654
+ CHROMA_420_PU_FILTER_PIXEL_TO_SHORT_NEON(p2sALIGNED);1655
+ CHROMA_420_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sALIGNED);1656
+ CHROMA_422_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1657
+ CHROMA_422_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1658
+ CHROMA_444_PU_NEON_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1659
+ CHROMA_444_PU_SVE_FILTER_PIXEL_TO_SHORT(p2sNONALIGNED);1660
+ LUMA_PU_NEON_FILTER_PIXEL_TO_SHORT(convert_p2sNONALIGNED);1661
+ LUMA_PU_SVE_FILTER_PIXEL_TO_SHORT(convert_p2sNONALIGNED);1662
+1663
#if !HIGH_BIT_DEPTH1664
- p.cuBLOCK_4x4.psy_cost_pp = PFX(psyCost_4x4_neon);1665
+ LUMA_PU_MULTIPLE_ARCHS_1(luma_vpp, interp_8tap_vert_pp, neon);1666
+ LUMA_PU_MULTIPLE_ARCHS_2(luma_vpp, interp_8tap_vert_pp, sve2);1667
+ LUMA_PU_MULTIPLE_ARCHS_1(luma_vsp, interp_8tap_vert_sp, sve2);1668
+ LUMA_PU_MULTIPLE_ARCHS_2(luma_vsp, interp_8tap_vert_sp, neon);1669
+ ALL_LUMA_PU(luma_vps, interp_8tap_vert_ps, sve2);1670
+ ALL_LUMA_PU(luma_hpp, interp_horiz_pp, neon);1671
+ ALL_LUMA_PU(luma_hps, interp_horiz_ps, neon);1672
+ ALL_LUMA_PU(luma_vss, interp_8tap_vert_ss, sve2);1673
+ ALL_LUMA_PU_T(luma_hvpp, interp_8tap_hv_pp_cpu);1674
+ CHROMA_420_VERT_FILTERS_NEON();1675
+ CHROMA_420_VERT_FILTERS_CAN_USE_SVE2();1676
+ CHROMA_422_VERT_FILTERS_NEON();1677
+ CHROMA_422_VERT_FILTERS_CAN_USE_SVE2(sve2);1678
+ CHROMA_444_VERT_FILTERS_NEON();1679
+ CHROMA_444_VERT_FILTERS_CAN_USE_SVE2();1680
+ CHROMA_420_FILTERS_NEON();1681
+ CHROMA_420_FILTERS_CAN_USE_SVE2();1682
+ CHROMA_422_FILTERS_NEON();1683
+ CHROMA_422_FILTERS_CAN_USE_SVE2();1684
+ CHROMA_444_FILTERS_NEON();1685
+ CHROMA_444_FILTERS_CAN_USE_SVE2();1686
+1687
+ // Blockcopy_pp1688
+ LUMA_PU_NEON_1(copy_pp, blockcopy_pp);1689
+ LUMA_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(copy_pp, blockcopy_pp);1690
+ CHROMA_420_PU_NEON_1(copy_pp, blockcopy_pp);1691
+ CHROMA_420_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(copy_pp, blockcopy_pp);1692
+ CHROMA_422_PU_NEON_1(copy_pp, blockcopy_pp);1693
+ CHROMA_422_PU_CAN_USE_SVE_EXCEPT_FILTER_PIXEL_TO_SHORT(copy_pp, blockcopy_pp);1694
+ p.cuBLOCK_4x4.copy_pp = PFX(blockcopy_pp_4x4_neon);1695
+ p.cuBLOCK_8x8.copy_pp = PFX(blockcopy_pp_8x8_neon);1696
+ p.cuBLOCK_16x16.copy_pp = PFX(blockcopy_pp_16x16_neon);1697
+ p.cuBLOCK_32x32.copy_pp = PFX(blockcopy_pp_32x32_sve);1698
+ p.cuBLOCK_64x64.copy_pp = PFX(blockcopy_pp_64x64_sve);1699
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_pp = PFX(blockcopy_pp_4x4_neon);1700
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_pp = PFX(blockcopy_pp_8x8_neon);1701
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_pp = PFX(blockcopy_pp_16x16_neon);1702
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_pp = PFX(blockcopy_pp_32x32_sve);1703
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_pp = PFX(blockcopy_pp_4x8_neon);1704
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_pp = PFX(blockcopy_pp_8x16_neon);1705
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_pp = PFX(blockcopy_pp_16x32_neon);1706
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_pp = PFX(blockcopy_pp_32x64_sve);1707
+1708
#endif // !HIGH_BIT_DEPTH1709
1710
+ // Blockcopy_ss1711
+ p.cuBLOCK_4x4.copy_ss = PFX(blockcopy_ss_4x4_neon);1712
+ p.cuBLOCK_8x8.copy_ss = PFX(blockcopy_ss_8x8_neon);1713
+ p.cuBLOCK_16x16.copy_ss = PFX(blockcopy_ss_16x16_sve);1714
+ p.cuBLOCK_32x32.copy_ss = PFX(blockcopy_ss_32x32_sve);1715
+ p.cuBLOCK_64x64.copy_ss = PFX(blockcopy_ss_64x64_sve);1716
+1717
+ // Blockcopy_ps1718
+ p.cuBLOCK_4x4.copy_ps = PFX(blockcopy_ps_4x4_neon);1719
+ p.cuBLOCK_8x8.copy_ps = PFX(blockcopy_ps_8x8_neon);1720
+ p.cuBLOCK_16x16.copy_ps = PFX(blockcopy_ps_16x16_sve);1721
+ p.cuBLOCK_32x32.copy_ps = PFX(blockcopy_ps_32x32_sve);1722
+ p.cuBLOCK_64x64.copy_ps = PFX(blockcopy_ps_64x64_sve);1723
+1724
+ // Blockcopy_sp1725
+ p.cuBLOCK_4x4.copy_sp = PFX(blockcopy_sp_4x4_sve);1726
+ p.cuBLOCK_8x8.copy_sp = PFX(blockcopy_sp_8x8_sve);1727
+ p.cuBLOCK_16x16.copy_sp = PFX(blockcopy_sp_16x16_sve);1728
+ p.cuBLOCK_32x32.copy_sp = PFX(blockcopy_sp_32x32_sve);1729
+ p.cuBLOCK_64x64.copy_sp = PFX(blockcopy_sp_64x64_neon);1730
+1731
+ // chroma blockcopy_ss1732
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_ss = PFX(blockcopy_ss_4x4_neon);1733
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_ss = PFX(blockcopy_ss_8x8_neon);1734
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_ss = PFX(blockcopy_ss_16x16_sve);1735
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_ss = PFX(blockcopy_ss_32x32_sve);1736
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_ss = PFX(blockcopy_ss_4x8_neon);1737
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_ss = PFX(blockcopy_ss_8x16_neon);1738
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_ss = PFX(blockcopy_ss_16x32_sve);1739
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_ss = PFX(blockcopy_ss_32x64_sve);1740
+1741
+ // chroma blockcopy_ps1742
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_ps = PFX(blockcopy_ps_4x4_neon);1743
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_ps = PFX(blockcopy_ps_8x8_neon);1744
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_ps = PFX(blockcopy_ps_16x16_sve);1745
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_ps = PFX(blockcopy_ps_32x32_sve);1746
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_ps = PFX(blockcopy_ps_4x8_sve);1747
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_ps = PFX(blockcopy_ps_8x16_sve);1748
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_ps = PFX(blockcopy_ps_16x32_sve);1749
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_ps = PFX(blockcopy_ps_32x64_sve);1750
+1751
+ // chroma blockcopy_sp1752
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.copy_sp = PFX(blockcopy_sp_4x4_sve);1753
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.copy_sp = PFX(blockcopy_sp_8x8_sve);1754
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.copy_sp = PFX(blockcopy_sp_16x16_sve);1755
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.copy_sp = PFX(blockcopy_sp_32x32_sve);1756
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.copy_sp = PFX(blockcopy_sp_4x8_sve);1757
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.copy_sp = PFX(blockcopy_sp_8x16_sve);1758
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.copy_sp = PFX(blockcopy_sp_16x32_sve);1759
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.copy_sp = PFX(blockcopy_sp_32x64_sve);1760
+1761
+ // Block_fill1762
+ LUMA_TU_NEON(blockfill_sALIGNED, blockfill_s);1763
+ LUMA_TU_CAN_USE_SVE(blockfill_sALIGNED, blockfill_s);1764
+ LUMA_TU_NEON(blockfill_sNONALIGNED, blockfill_s);1765
+ LUMA_TU_CAN_USE_SVE(blockfill_sNONALIGNED, blockfill_s);1766
+1767
+ // copy_count1768
+ p.cuBLOCK_4x4.copy_cnt = PFX(copy_cnt_4_neon);1769
+ p.cuBLOCK_8x8.copy_cnt = PFX(copy_cnt_8_neon);1770
+ p.cuBLOCK_16x16.copy_cnt = PFX(copy_cnt_16_neon);1771
+ p.cuBLOCK_32x32.copy_cnt = PFX(copy_cnt_32_neon);1772
+1773
+ // count nonzero1774
+ p.cuBLOCK_4x4.count_nonzero = PFX(count_nonzero_4_neon);1775
+ p.cuBLOCK_8x8.count_nonzero = PFX(count_nonzero_8_neon);1776
+ p.cuBLOCK_16x16.count_nonzero = PFX(count_nonzero_16_neon);1777
+ p.cuBLOCK_32x32.count_nonzero = PFX(count_nonzero_32_neon);1778
+1779
+ // cpy2Dto1D_shl1780
+ p.cuBLOCK_4x4.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_4x4_neon);1781
+ p.cuBLOCK_8x8.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_8x8_neon);1782
+ p.cuBLOCK_16x16.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_16x16_sve);1783
+ p.cuBLOCK_32x32.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_32x32_sve);1784
+ p.cuBLOCK_64x64.cpy2Dto1D_shl = PFX(cpy2Dto1D_shl_64x64_sve);1785
+1786
+ // cpy2Dto1D_shr1787
+ p.cuBLOCK_4x4.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_4x4_neon);1788
+ p.cuBLOCK_8x8.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_8x8_neon);1789
+ p.cuBLOCK_16x16.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_16x16_sve);1790
+ p.cuBLOCK_32x32.cpy2Dto1D_shr = PFX(cpy2Dto1D_shr_32x32_sve);1791
+1792
+ // cpy1Dto2D_shl1793
+ p.cuBLOCK_4x4.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_4x4_neon);1794
+ p.cuBLOCK_8x8.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_8x8_neon);1795
+ p.cuBLOCK_16x16.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_16x16_sve);1796
+ p.cuBLOCK_32x32.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_32x32_sve);1797
+ p.cuBLOCK_64x64.cpy1Dto2D_shlALIGNED = PFX(cpy1Dto2D_shl_64x64_sve);1798
+1799
+ p.cuBLOCK_4x4.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_4x4_neon);1800
+ p.cuBLOCK_8x8.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_8x8_neon);1801
+ p.cuBLOCK_16x16.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_16x16_sve);1802
+ p.cuBLOCK_32x32.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_32x32_sve);1803
+ p.cuBLOCK_64x64.cpy1Dto2D_shlNONALIGNED = PFX(cpy1Dto2D_shl_64x64_sve);1804
+1805
+ // cpy1Dto2D_shr1806
+ p.cuBLOCK_4x4.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_4x4_neon);1807
+ p.cuBLOCK_8x8.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_8x8_neon);1808
+ p.cuBLOCK_16x16.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_16x16_sve);1809
+ p.cuBLOCK_32x32.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_32x32_sve);1810
+ p.cuBLOCK_64x64.cpy1Dto2D_shr = PFX(cpy1Dto2D_shr_64x64_sve);1811
+1812
+#if !HIGH_BIT_DEPTH1813
+ // pixel_avg_pp1814
+ LUMA_PU_NEON_2(pixelavg_ppNONALIGNED, pixel_avg_pp);1815
+ LUMA_PU_MULTIPLE_ARCHS_3(pixelavg_ppNONALIGNED, pixel_avg_pp, sve2);1816
+ LUMA_PU_NEON_2(pixelavg_ppALIGNED, pixel_avg_pp);1817
+ LUMA_PU_MULTIPLE_ARCHS_3(pixelavg_ppALIGNED, pixel_avg_pp, sve2);1818
+1819
+ // addAvg1820
+ LUMA_PU_NEON_3(addAvgNONALIGNED, addAvg);1821
+ LUMA_PU_CAN_USE_SVE2(addAvgNONALIGNED, addAvg);1822
+ LUMA_PU_NEON_3(addAvgALIGNED, addAvg);1823
+ LUMA_PU_CAN_USE_SVE2(addAvgALIGNED, addAvg);1824
+ CHROMA_420_PU_NEON_2(addAvgNONALIGNED, addAvg);1825
+ CHROMA_420_PU_MULTIPLE_ARCHS(addAvgNONALIGNED, addAvg, sve2);1826
+ CHROMA_420_PU_NEON_2(addAvgALIGNED, addAvg);1827
+ CHROMA_420_PU_MULTIPLE_ARCHS(addAvgALIGNED, addAvg, sve2);1828
+ CHROMA_422_PU_NEON_2(addAvgNONALIGNED, addAvg);1829
+ CHROMA_422_PU_CAN_USE_SVE2(addAvgNONALIGNED, addAvg);1830
+ CHROMA_422_PU_NEON_2(addAvgALIGNED, addAvg);1831
+ CHROMA_422_PU_CAN_USE_SVE2(addAvgALIGNED, addAvg);1832
+1833
+ // sad1834
+ ALL_LUMA_PU(sad, pixel_sad, sve2);1835
+ ALL_LUMA_PU(sad_x3, sad_x3, sve2);1836
+ ALL_LUMA_PU(sad_x4, sad_x4, sve2);1837
+1838
+ // sse_pp1839
+ p.cuBLOCK_4x4.sse_pp = PFX(pixel_sse_pp_4x4_sve);1840
+ p.cuBLOCK_8x8.sse_pp = PFX(pixel_sse_pp_8x8_neon);1841
+ p.cuBLOCK_16x16.sse_pp = PFX(pixel_sse_pp_16x16_neon);1842
+ p.cuBLOCK_32x32.sse_pp = PFX(pixel_sse_pp_32x32_sve2);1843
+ p.cuBLOCK_64x64.sse_pp = PFX(pixel_sse_pp_64x64_sve2);1844
+1845
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.sse_pp = PFX(pixel_sse_pp_4x4_sve);1846
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.sse_pp = PFX(pixel_sse_pp_8x8_neon);1847
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.sse_pp = PFX(pixel_sse_pp_16x16_neon);1848
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.sse_pp = PFX(pixel_sse_pp_32x32_sve2);1849
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.sse_pp = PFX(pixel_sse_pp_4x8_sve);1850
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sse_pp = PFX(pixel_sse_pp_8x16_neon);1851
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sse_pp = PFX(pixel_sse_pp_16x32_neon);1852
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sse_pp = PFX(pixel_sse_pp_32x64_sve2);1853
+1854
+ // sse_ss1855
+ p.cuBLOCK_4x4.sse_ss = PFX(pixel_sse_ss_4x4_sve2);1856
+ p.cuBLOCK_8x8.sse_ss = PFX(pixel_sse_ss_8x8_sve2);1857
+ p.cuBLOCK_16x16.sse_ss = PFX(pixel_sse_ss_16x16_sve2);1858
+ p.cuBLOCK_32x32.sse_ss = PFX(pixel_sse_ss_32x32_sve2);1859
+ p.cuBLOCK_64x64.sse_ss = PFX(pixel_sse_ss_64x64_sve2);1860
+1861
+ // ssd_s1862
+ p.cuBLOCK_4x4.ssd_sNONALIGNED = PFX(pixel_ssd_s_4x4_sve2);1863
+ p.cuBLOCK_8x8.ssd_sNONALIGNED = PFX(pixel_ssd_s_8x8_sve2);1864
+ p.cuBLOCK_16x16.ssd_sNONALIGNED = PFX(pixel_ssd_s_16x16_sve2);1865
+ p.cuBLOCK_32x32.ssd_sNONALIGNED = PFX(pixel_ssd_s_32x32_sve2);1866
+1867
+ p.cuBLOCK_4x4.ssd_sALIGNED = PFX(pixel_ssd_s_4x4_sve2);1868
+ p.cuBLOCK_8x8.ssd_sALIGNED = PFX(pixel_ssd_s_8x8_sve2);1869
+ p.cuBLOCK_16x16.ssd_sALIGNED = PFX(pixel_ssd_s_16x16_sve2);1870
+ p.cuBLOCK_32x32.ssd_sALIGNED = PFX(pixel_ssd_s_32x32_sve2);1871
+1872
+ // pixel_var1873
+ p.cuBLOCK_8x8.var = PFX(pixel_var_8x8_sve2);1874
+ p.cuBLOCK_16x16.var = PFX(pixel_var_16x16_sve2);1875
+ p.cuBLOCK_32x32.var = PFX(pixel_var_32x32_sve2);1876
+ p.cuBLOCK_64x64.var = PFX(pixel_var_64x64_sve2);1877
+1878
+ // calc_Residual1879
+ p.cuBLOCK_4x4.calcresidualNONALIGNED = PFX(getResidual4_neon);1880
+ p.cuBLOCK_8x8.calcresidualNONALIGNED = PFX(getResidual8_neon);1881
+ p.cuBLOCK_16x16.calcresidualNONALIGNED = PFX(getResidual16_sve2);1882
+ p.cuBLOCK_32x32.calcresidualNONALIGNED = PFX(getResidual32_sve2);1883
+1884
+ p.cuBLOCK_4x4.calcresidualALIGNED = PFX(getResidual4_neon);1885
+ p.cuBLOCK_8x8.calcresidualALIGNED = PFX(getResidual8_neon);1886
+ p.cuBLOCK_16x16.calcresidualALIGNED = PFX(getResidual16_sve2);1887
+ p.cuBLOCK_32x32.calcresidualALIGNED = PFX(getResidual32_sve2);1888
+1889
+ // pixel_sub_ps1890
+ p.cuBLOCK_4x4.sub_ps = PFX(pixel_sub_ps_4x4_neon);1891
+ p.cuBLOCK_8x8.sub_ps = PFX(pixel_sub_ps_8x8_neon);1892
+ p.cuBLOCK_16x16.sub_ps = PFX(pixel_sub_ps_16x16_neon);1893
+ p.cuBLOCK_32x32.sub_ps = PFX(pixel_sub_ps_32x32_sve2);1894
+ p.cuBLOCK_64x64.sub_ps = PFX(pixel_sub_ps_64x64_sve2);1895
+1896
+ // chroma sub_ps1897
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.sub_ps = PFX(pixel_sub_ps_4x4_neon);1898
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.sub_ps = PFX(pixel_sub_ps_8x8_neon);1899
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.sub_ps = PFX(pixel_sub_ps_16x16_neon);1900
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.sub_ps = PFX(pixel_sub_ps_32x32_sve2);1901
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.sub_ps = PFX(pixel_sub_ps_4x8_neon);1902
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sub_ps = PFX(pixel_sub_ps_8x16_sve);1903
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sub_ps = PFX(pixel_sub_ps_16x32_neon);1904
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sub_ps = PFX(pixel_sub_ps_32x64_sve2);1905
+1906
+ // pixel_add_ps1907
+ p.cuBLOCK_4x4.add_psNONALIGNED = PFX(pixel_add_ps_4x4_sve2);1908
+ p.cuBLOCK_8x8.add_psNONALIGNED = PFX(pixel_add_ps_8x8_sve2);1909
+ p.cuBLOCK_16x16.add_psNONALIGNED = PFX(pixel_add_ps_16x16_sve2);1910
+ p.cuBLOCK_32x32.add_psNONALIGNED = PFX(pixel_add_ps_32x32_sve2);1911
+ p.cuBLOCK_64x64.add_psNONALIGNED = PFX(pixel_add_ps_64x64_sve2);1912
+1913
+ p.cuBLOCK_4x4.add_psALIGNED = PFX(pixel_add_ps_4x4_sve2);1914
+ p.cuBLOCK_8x8.add_psALIGNED = PFX(pixel_add_ps_8x8_sve2);1915
+ p.cuBLOCK_16x16.add_psALIGNED = PFX(pixel_add_ps_16x16_sve2);1916
+ p.cuBLOCK_32x32.add_psALIGNED = PFX(pixel_add_ps_32x32_sve2);1917
+ p.cuBLOCK_64x64.add_psALIGNED = PFX(pixel_add_ps_64x64_sve2);1918
+1919
+ // chroma add_ps1920
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.add_psNONALIGNED = PFX(pixel_add_ps_4x4_sve2);1921
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.add_psNONALIGNED = PFX(pixel_add_ps_8x8_sve2);1922
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.add_psNONALIGNED = PFX(pixel_add_ps_16x16_sve2);1923
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.add_psNONALIGNED = PFX(pixel_add_ps_32x32_sve2);1924
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.add_psNONALIGNED = PFX(pixel_add_ps_4x8_sve2);1925
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.add_psNONALIGNED = PFX(pixel_add_ps_8x16_sve2);1926
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.add_psNONALIGNED = PFX(pixel_add_ps_16x32_sve2);1927
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.add_psNONALIGNED = PFX(pixel_add_ps_32x64_sve2);1928
+1929
+ p.chromaX265_CSP_I420.cuBLOCK_420_4x4.add_psALIGNED = PFX(pixel_add_ps_4x4_sve2);1930
+ p.chromaX265_CSP_I420.cuBLOCK_420_8x8.add_psALIGNED = PFX(pixel_add_ps_8x8_sve2);1931
+ p.chromaX265_CSP_I420.cuBLOCK_420_16x16.add_psALIGNED = PFX(pixel_add_ps_16x16_sve2);1932
+ p.chromaX265_CSP_I420.cuBLOCK_420_32x32.add_psALIGNED = PFX(pixel_add_ps_32x32_sve2);1933
+ p.chromaX265_CSP_I422.cuBLOCK_422_4x8.add_psALIGNED = PFX(pixel_add_ps_4x8_sve2);1934
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.add_psALIGNED = PFX(pixel_add_ps_8x16_sve2);1935
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.add_psALIGNED = PFX(pixel_add_ps_16x32_sve2);1936
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.add_psALIGNED = PFX(pixel_add_ps_32x64_sve2);1937
+1938
+ //scale2D_64to321939
+ p.scale2D_64to32 = PFX(scale2D_64to32_neon);1940
+1941
+ // scale1D_128to641942
+ p.scale1D_128to64NONALIGNED = PFX(scale1D_128to64_sve2);1943
+ p.scale1D_128to64ALIGNED = PFX(scale1D_128to64_sve2);1944
+1945
+ // planecopy1946
+ p.planecopy_cp = PFX(pixel_planecopy_cp_neon);1947
+1948
+ // satd1949
+ p.puLUMA_4x4.satd = PFX(pixel_satd_4x4_sve);1950
+ p.puLUMA_8x8.satd = PFX(pixel_satd_8x8_neon);1951
+ p.puLUMA_16x16.satd = PFX(pixel_satd_16x16_neon);1952
+ p.puLUMA_32x32.satd = PFX(pixel_satd_32x32_sve);1953
+ p.puLUMA_64x64.satd = PFX(pixel_satd_64x64_neon);1954
+ p.puLUMA_8x4.satd = PFX(pixel_satd_8x4_sve);1955
+ p.puLUMA_4x8.satd = PFX(pixel_satd_4x8_neon);1956
+ p.puLUMA_16x8.satd = PFX(pixel_satd_16x8_neon);1957
+ p.puLUMA_8x16.satd = PFX(pixel_satd_8x16_neon);1958
+ p.puLUMA_16x32.satd = PFX(pixel_satd_16x32_neon);1959
+ p.puLUMA_32x16.satd = PFX(pixel_satd_32x16_sve);1960
+ p.puLUMA_64x32.satd = PFX(pixel_satd_64x32_neon);1961
+ p.puLUMA_32x64.satd = PFX(pixel_satd_32x64_neon);1962
+ p.puLUMA_16x12.satd = PFX(pixel_satd_16x12_neon);1963
+ p.puLUMA_12x16.satd = PFX(pixel_satd_12x16_neon);1964
+ p.puLUMA_16x4.satd = PFX(pixel_satd_16x4_neon);1965
+ p.puLUMA_4x16.satd = PFX(pixel_satd_4x16_neon);1966
+ p.puLUMA_32x24.satd = PFX(pixel_satd_32x24_neon);1967
+ p.puLUMA_24x32.satd = PFX(pixel_satd_24x32_neon);1968
+ p.puLUMA_32x8.satd = PFX(pixel_satd_32x8_neon);1969
+ p.puLUMA_8x32.satd = PFX(pixel_satd_8x32_neon);1970
+ p.puLUMA_64x48.satd = PFX(pixel_satd_64x48_sve);1971
+ p.puLUMA_48x64.satd = PFX(pixel_satd_48x64_neon);1972
+ p.puLUMA_64x16.satd = PFX(pixel_satd_64x16_neon);1973
+ p.puLUMA_16x64.satd = PFX(pixel_satd_16x64_neon);1974
+1975
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.satd = PFX(pixel_satd_4x4_sve);1976
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.satd = PFX(pixel_satd_8x8_neon);1977
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.satd = PFX(pixel_satd_16x16_neon);1978
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.satd = PFX(pixel_satd_32x32_neon);1979
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.satd = PFX(pixel_satd_8x4_sve);1980
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.satd = PFX(pixel_satd_4x8_neon);1981
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.satd = PFX(pixel_satd_16x8_neon);1982
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.satd = PFX(pixel_satd_8x16_neon);1983
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.satd = PFX(pixel_satd_32x16_neon);1984
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.satd = PFX(pixel_satd_16x32_neon);1985
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.satd = PFX(pixel_satd_16x12_neon);1986
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.satd = PFX(pixel_satd_12x16_neon);1987
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.satd = PFX(pixel_satd_16x4_neon);1988
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.satd = PFX(pixel_satd_4x16_neon);1989
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.satd = PFX(pixel_satd_32x24_neon);1990
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.satd = PFX(pixel_satd_24x32_neon);1991
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.satd = PFX(pixel_satd_32x8_neon);1992
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.satd = PFX(pixel_satd_8x32_neon);1993
+1994
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.satd = PFX(pixel_satd_4x8_neon);1995
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.satd = PFX(pixel_satd_8x16_neon);1996
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.satd = PFX(pixel_satd_16x32_neon);1997
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.satd = PFX(pixel_satd_32x64_neon);1998
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.satd = PFX(pixel_satd_4x4_sve);1999
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.satd = PFX(pixel_satd_8x8_neon);2000
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.satd = PFX(pixel_satd_4x16_neon);2001
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.satd = PFX(pixel_satd_16x16_neon);2002
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.satd = PFX(pixel_satd_8x32_neon);2003
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.satd = PFX(pixel_satd_32x32_neon);2004
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.satd = PFX(pixel_satd_16x64_neon);2005
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.satd = PFX(pixel_satd_8x12_sve);2006
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.satd = PFX(pixel_satd_8x4_sve);2007
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.satd = PFX(pixel_satd_16x24_neon);2008
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.satd = PFX(pixel_satd_12x32_neon);2009
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.satd = PFX(pixel_satd_16x8_neon);2010
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.satd = PFX(pixel_satd_4x32_neon);2011
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.satd = PFX(pixel_satd_32x48_neon);2012
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.satd = PFX(pixel_satd_24x64_neon);2013
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.satd = PFX(pixel_satd_32x16_neon);2014
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.satd = PFX(pixel_satd_8x64_neon);2015
+2016
+ // sa8d2017
+ p.cuBLOCK_4x4.sa8d = PFX(pixel_satd_4x4_sve);2018
+ p.cuBLOCK_8x8.sa8d = PFX(pixel_sa8d_8x8_neon);2019
+ p.cuBLOCK_16x16.sa8d = PFX(pixel_sa8d_16x16_neon);2020
+ p.cuBLOCK_32x32.sa8d = PFX(pixel_sa8d_32x32_neon);2021
+ p.cuBLOCK_64x64.sa8d = PFX(pixel_sa8d_64x64_neon);2022
+ p.chromaX265_CSP_I420.cuBLOCK_8x8.sa8d = PFX(pixel_satd_4x4_sve);2023
+ p.chromaX265_CSP_I420.cuBLOCK_16x16.sa8d = PFX(pixel_sa8d_16x16_neon);2024
+ p.chromaX265_CSP_I420.cuBLOCK_32x32.sa8d = PFX(pixel_sa8d_32x32_neon);2025
+ p.chromaX265_CSP_I420.cuBLOCK_64x64.sa8d = PFX(pixel_sa8d_64x64_neon);2026
+ p.chromaX265_CSP_I422.cuBLOCK_422_8x16.sa8d = PFX(pixel_sa8d_8x16_neon);2027
+ p.chromaX265_CSP_I422.cuBLOCK_422_16x32.sa8d = PFX(pixel_sa8d_16x32_neon);2028
+ p.chromaX265_CSP_I422.cuBLOCK_422_32x64.sa8d = PFX(pixel_sa8d_32x64_neon);2029
+2030
+ // dequant_scaling2031
+ p.dequant_scaling = PFX(dequant_scaling_sve2);2032
+ p.dequant_normal = PFX(dequant_normal_sve2);2033
+2034
+ // ssim_4x4x2_core2035
+ p.ssim_4x4x2_core = PFX(ssim_4x4x2_core_sve2);2036
+2037
+ // ssimDist2038
+ p.cuBLOCK_4x4.ssimDist = PFX(ssimDist4_sve2);2039
+ p.cuBLOCK_8x8.ssimDist = PFX(ssimDist8_sve2);2040
+ p.cuBLOCK_16x16.ssimDist = PFX(ssimDist16_sve2);2041
+ p.cuBLOCK_32x32.ssimDist = PFX(ssimDist32_sve2);2042
+ p.cuBLOCK_64x64.ssimDist = PFX(ssimDist64_sve2);2043
+2044
+ // normFact2045
+ p.cuBLOCK_8x8.normFact = PFX(normFact8_sve2);2046
+ p.cuBLOCK_16x16.normFact = PFX(normFact16_sve2);2047
+ p.cuBLOCK_32x32.normFact = PFX(normFact32_sve2);2048
+ p.cuBLOCK_64x64.normFact = PFX(normFact64_sve2);2049
+2050
+ // psy_cost_pp2051
+ p.cuBLOCK_4x4.psy_cost_pp = PFX(psyCost_4x4_neon);2052
+2053
+ p.weight_pp = PFX(weight_pp_neon);2054
+#if !defined(__APPLE__)2055
+ p.scanPosLast = PFX(scanPosLast_neon);2056
+#endif2057
+ p.costCoeffNxN = PFX(costCoeffNxN_neon);2058
+#endif2059
+2060
+ // quant2061
+ p.quant = PFX(quant_sve);2062
+ p.nquant = PFX(nquant_neon);2063
+}2064
+#endif2065
+2066
+void setupAssemblyPrimitives(EncoderPrimitives &p, int cpuMask)2067
+{2068
+2069
+#ifdef HAVE_SVE22070
+ if (cpuMask & X265_CPU_SVE2)2071
+ {2072
+ setupSve2Primitives(p);2073
}2074
+ else if (cpuMask & X265_CPU_SVE)2075
+ {2076
+ setupSvePrimitives(p);2077
+ }2078
+ else if (cpuMask & X265_CPU_NEON)2079
+ {2080
+ setupNeonPrimitives(p);2081
+ }2082
+2083
+#elif defined(HAVE_SVE)2084
+ if (cpuMask & X265_CPU_SVE)2085
+ {2086
+ setupSvePrimitives(p);2087
+ }2088
+ else if (cpuMask & X265_CPU_NEON)2089
+ {2090
+ setupNeonPrimitives(p);2091
+ }2092
+2093
+#else2094
+ if (cpuMask & X265_CPU_NEON)2095
+ {2096
+ setupNeonPrimitives(p);2097
+ }2098
+#endif2099
+2100
}2101
} // namespace X265_NS2102
x265_3.6.tar.gz/source/common/aarch64/asm-sve.S
Added
41
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm.S"26
+27
+.arch armv8-a+sve28
+29
+.macro ABS2_SVE a b c30
+ abs \a, \c\()/m, \a31
+ abs \b, \c\()/m, \b32
+.endm33
+34
+.macro ABS8_SVE z0, z1, z2, z3, z4, z5, z6, z7, p035
+ ABS2_SVE \z0, \z1, p036
+ ABS2_SVE \z2, \z3, p037
+ ABS2_SVE \z4, \z5, p038
+ ABS2_SVE \z6, \z7, p039
+.endm40
+41
x265_3.5.tar.gz/source/common/aarch64/asm.S -> x265_3.6.tar.gz/source/common/aarch64/asm.S
Changed
173
1
2
/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
+ * Copyright (C) 2020-2021 MulticoreWare, Inc5
*6
* Authors: Hongbin Liu <liuhongbin1@huawei.com>7
+ * Sebastian Pop <spop@amazon.com>8
*9
* This program is free software; you can redistribute it and/or modify10
* it under the terms of the GNU General Public License as published by11
12
* For more information, contact us at license @ x265.com.13
*****************************************************************************/14
15
+#ifndef ASM_S_ // #include guards16
+#define ASM_S_17
+18
.arch armv8-a19
20
+#define PFX3(prefix, name) prefix ## _ ## name21
+#define PFX2(prefix, name) PFX3(prefix, name)22
+#define PFX(name) PFX2(X265_NS, name)23
+24
+#ifdef __APPLE__25
+#define PREFIX 126
+#endif27
+28
#ifdef PREFIX29
#define EXTERN_ASM _30
+#define HAVE_AS_FUNC 031
+#elif defined __clang__32
+#define EXTERN_ASM33
+#define HAVE_AS_FUNC 034
+#define PREFIX 135
#else36
#define EXTERN_ASM37
+#define HAVE_AS_FUNC 138
#endif39
40
#ifdef __ELF__41
#define ELF42
#else43
+#ifdef PREFIX44
+#define ELF #45
+#else46
#define ELF @47
#endif48
-49
-#define HAVE_AS_FUNC 150
+#endif51
52
#if HAVE_AS_FUNC53
#define FUNC54
#else55
+#ifdef PREFIX56
+#define FUNC #57
+#else58
#define FUNC @59
#endif60
+#endif61
+62
+#define GLUE(a, b) a ## b63
+#define JOIN(a, b) GLUE(a, b)64
+65
+#define PFX_C(name) JOIN(JOIN(JOIN(EXTERN_ASM, X265_NS), _), name)66
+67
+#ifdef __APPLE__68
+.macro endfunc69
+ELF .size \name, . - \name70
+FUNC .endfunc71
+.endm72
+#endif73
74
.macro function name, export=175
+#ifdef __APPLE__76
+ .global \name77
+ endfunc78
+#else79
.macro endfunc80
ELF .size \name, . - \name81
FUNC .endfunc82
.purgem endfunc83
.endm84
+#endif85
.align 286
.if \export == 187
.global EXTERN_ASM\name88
89
.endif90
.endm91
92
+.macro const name, align=293
+ .macro endconst94
+ELF .size \name, . - \name95
+ .purgem endconst96
+ .endm97
+#ifdef __MACH__98
+ .const_data99
+#else100
+ .section .rodata101
+#endif102
+ .align \align103
+\name:104
+.endm105
+106
+.macro movrel rd, val, offset=0107
+#if defined(__APPLE__)108
+ .if \offset < 0109
+ adrp \rd, \val@PAGE110
+ add \rd, \rd, \val@PAGEOFF111
+ sub \rd, \rd, -(\offset)112
+ .else113
+ adrp \rd, \val+(\offset)@PAGE114
+ add \rd, \rd, \val+(\offset)@PAGEOFF115
+ .endif116
+#elif defined(PIC) && defined(_WIN32)117
+ .if \offset < 0118
+ adrp \rd, \val119
+ add \rd, \rd, :lo12:\val120
+ sub \rd, \rd, -(\offset)121
+ .else122
+ adrp \rd, \val+(\offset)123
+ add \rd, \rd, :lo12:\val+(\offset)124
+ .endif125
+#else126
+ adrp \rd, \val+(\offset)127
+ add \rd, \rd, :lo12:\val+(\offset)128
+#endif129
+.endm130
131
#define FENC_STRIDE 64132
#define FDEC_STRIDE 32133
+134
+.macro SUMSUB_AB sum, diff, a, b135
+ add \sum, \a, \b136
+ sub \diff, \a, \b137
+.endm138
+139
+.macro SUMSUB_ABCD s1, d1, s2, d2, a, b, c, d140
+ SUMSUB_AB \s1, \d1, \a, \b141
+ SUMSUB_AB \s2, \d2, \c, \d142
+.endm143
+144
+.macro HADAMARD4_V r1, r2, r3, r4, t1, t2, t3, t4145
+ SUMSUB_ABCD \t1, \t2, \t3, \t4, \r1, \r2, \r3, \r4146
+ SUMSUB_ABCD \r1, \r3, \r2, \r4, \t1, \t3, \t2, \t4147
+.endm148
+149
+.macro ABS2 a b150
+ abs \a, \a151
+ abs \b, \b152
+.endm153
+154
+.macro ABS8 v0, v1, v2, v3, v4, v5, v6, v7155
+ ABS2 \v0, \v1156
+ ABS2 \v2, \v3157
+ ABS2 \v4, \v5158
+ ABS2 \v6, \v7159
+.endm160
+161
+.macro vtrn t1, t2, s1, s2162
+ trn1 \t1, \s1, \s2163
+ trn2 \t2, \s1, \s2164
+.endm165
+166
+.macro trn4 t1, t2, t3, t4, s1, s2, s3, s4167
+ vtrn \t1, \t2, \s1, \s2168
+ vtrn \t3, \t4, \s3, \s4169
+.endm170
+171
+#endif172
\ No newline at end of file173
x265_3.6.tar.gz/source/common/aarch64/blockcopy8-common.S
Added
56
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+#include "asm.S"29
+30
+.arch armv8-a31
+32
+// void cpy1Dto2D_shr(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift)33
+.macro cpy1Dto2D_shr_start34
+ add x2, x2, x235
+ dup v0.8h, w336
+ cmeq v1.8h, v1.8h, v1.8h37
+ sshl v1.8h, v1.8h, v0.8h38
+ sri v1.8h, v1.8h, #139
+ neg v0.8h, v0.8h40
+.endm41
+42
+.macro cpy2Dto1D_shr_start43
+ add x2, x2, x244
+ dup v0.8h, w345
+ cmeq v1.8h, v1.8h, v1.8h46
+ sshl v1.8h, v1.8h, v0.8h47
+ sri v1.8h, v1.8h, #148
+ neg v0.8h, v0.8h49
+.endm50
+51
+const xtn_xtn2_table, align=452
+.byte 0, 2, 4, 6, 8, 10, 12, 1453
+.byte 16, 18, 20, 22, 24, 26, 28, 3054
+endconst55
+56
x265_3.6.tar.gz/source/common/aarch64/blockcopy8-sve.S
Added
1418
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ 7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "blockcopy8-common.S"27
+28
+.arch armv8-a+sve29
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+/* void blockcopy_sp(pixel* a, intptr_t stridea, const int16_t* b, intptr_t strideb)41
+ *42
+ * r0 - a43
+ * r1 - stridea44
+ * r2 - b45
+ * r3 - strideb */46
+47
+function PFX(blockcopy_sp_4x4_sve)48
+ ptrue p0.h, vl449
+.rept 250
+ ld1h {z0.h}, p0/z, x251
+ add x2, x2, x3, lsl #152
+ st1b {z0.h}, p0, x053
+ add x0, x0, x154
+ ld1h {z1.h}, p0/z, x255
+ add x2, x2, x3, lsl #156
+ st1b {z1.h}, p0, x057
+ add x0, x0, x158
+.endr59
+ ret60
+endfunc61
+62
+function PFX(blockcopy_sp_8x8_sve)63
+ ptrue p0.h, vl864
+.rept 465
+ ld1h {z0.h}, p0/z, x266
+ add x2, x2, x3, lsl #167
+ st1b {z0.h}, p0, x068
+ add x0, x0, x169
+ ld1h {z1.h}, p0/z, x270
+ add x2, x2, x3, lsl #171
+ st1b {z1.h}, p0, x072
+ add x0, x0, x173
+.endr74
+ ret75
+endfunc76
+77
+function PFX(blockcopy_sp_16x16_sve)78
+ rdvl x9, #179
+ cmp x9, #1680
+ bgt .vl_gt_16_blockcopy_sp_16_1681
+ lsl x3, x3, #182
+ movrel x11, xtn_xtn2_table83
+ ld1 {v31.16b}, x1184
+.rept 885
+ ld1 {v0.8h-v1.8h}, x2, x386
+ ld1 {v2.8h-v3.8h}, x2, x387
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b88
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b89
+ st1 {v0.16b}, x0, x190
+ st1 {v1.16b}, x0, x191
+.endr92
+ ret93
+.vl_gt_16_blockcopy_sp_16_16:94
+ ptrue p0.h, vl1695
+.rept 896
+ ld1h {z0.h}, p0/z, x297
+ st1b {z0.h}, p0, x098
+ add x2, x2, x3, lsl #199
+ add x0, x0, x1100
+ ld1h {z1.h}, p0/z, x2101
+ st1b {z1.h}, p0, x0102
+ add x2, x2, x3, lsl #1103
+ add x0, x0, x1104
+.endr105
+ ret106
+endfunc107
+108
+function PFX(blockcopy_sp_32x32_sve)109
+ mov w12, #4110
+ rdvl x9, #1111
+ cmp x9, #16112
+ bgt .vl_gt_16_blockcopy_sp_32_32113
+ lsl x3, x3, #1114
+ movrel x11, xtn_xtn2_table115
+ ld1 {v31.16b}, x11116
+.loop_csp32_sve:117
+ sub w12, w12, #1118
+.rept 4119
+ ld1 {v0.8h-v3.8h}, x2, x3120
+ ld1 {v4.8h-v7.8h}, x2, x3121
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b122
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b123
+ tbl v2.16b, {v4.16b,v5.16b}, v31.16b124
+ tbl v3.16b, {v6.16b,v7.16b}, v31.16b125
+ st1 {v0.16b-v1.16b}, x0, x1126
+ st1 {v2.16b-v3.16b}, x0, x1127
+.endr128
+ cbnz w12, .loop_csp32_sve129
+ ret130
+.vl_gt_16_blockcopy_sp_32_32:131
+ cmp x9, #48132
+ bgt .vl_gt_48_blockcopy_sp_32_32133
+ ptrue p0.h, vl16134
+.vl_gt_16_loop_csp32_sve:135
+ sub w12, w12, #1136
+.rept 4137
+ ld1h {z0.h}, p0/z, x2138
+ ld1h {z1.h}, p0/z, x2, #1, mul vl139
+ st1b {z0.h}, p0, x0140
+ st1b {z1.h}, p0, x0, #1, mul vl141
+ add x2, x2, x3, lsl #1142
+ add x0, x0, x1143
+ ld1h {z2.h}, p0/z, x2144
+ ld1h {z3.h}, p0/z, x2, #1, mul vl145
+ st1b {z2.h}, p0, x0146
+ st1b {z3.h}, p0, x0, #1, mul vl147
+ add x2, x2, x3, lsl #1148
+ add x0, x0, x1149
+.endr150
+ cbnz w12, .vl_gt_16_loop_csp32_sve151
+ ret152
+.vl_gt_48_blockcopy_sp_32_32:153
+ ptrue p0.h, vl32154
+.vl_gt_48_loop_csp32_sve:155
+ sub w12, w12, #1156
+.rept 4157
+ ld1h {z0.h}, p0/z, x2158
+ st1b {z0.h}, p0, x0159
+ add x2, x2, x3, lsl #1160
+ add x0, x0, x1161
+ ld1h {z1.h}, p0/z, x2162
+ st1b {z1.h}, p0, x0163
+ add x2, x2, x3, lsl #1164
+ add x0, x0, x1165
+.endr166
+ cbnz w12, .vl_gt_48_loop_csp32_sve167
+ ret168
+endfunc169
+170
+function PFX(blockcopy_ps_16x16_sve)171
+ rdvl x9, #1172
+ cmp x9, #16173
+ bgt .vl_gt_16_blockcopy_ps_16_16174
+ lsl x1, x1, #1175
+.rept 8176
+ ld1 {v4.16b}, x2, x3177
+ ld1 {v5.16b}, x2, x3178
+ uxtl v0.8h, v4.8b179
+ uxtl2 v1.8h, v4.16b180
+ uxtl v2.8h, v5.8b181
+ uxtl2 v3.8h, v5.16b182
+ st1 {v0.8h-v1.8h}, x0, x1183
+ st1 {v2.8h-v3.8h}, x0, x1184
+.endr185
+ ret186
+.vl_gt_16_blockcopy_ps_16_16:187
+ ptrue p0.b, vl32188
+.rept 16189
+ ld1b {z1.h}, p0/z, x2190
+ st1h {z1.h}, p0, x0191
+ add x0, x0, x1, lsl #1192
+ add x2, x2, x3193
+.endr194
+ ret195
+endfunc196
+197
+function PFX(blockcopy_ps_32x32_sve)198
+ rdvl x9, #1199
+ cmp x9, #16200
+ bgt .vl_gt_16_blockcopy_ps_32_32201
+ lsl x1, x1, #1202
+ mov w12, #4203
+.loop_cps32_sve:204
+ sub w12, w12, #1205
+.rept 4206
+ ld1 {v16.16b-v17.16b}, x2, x3207
+ ld1 {v18.16b-v19.16b}, x2, x3208
+ uxtl v0.8h, v16.8b209
+ uxtl2 v1.8h, v16.16b210
+ uxtl v2.8h, v17.8b211
+ uxtl2 v3.8h, v17.16b212
+ uxtl v4.8h, v18.8b213
+ uxtl2 v5.8h, v18.16b214
+ uxtl v6.8h, v19.8b215
+ uxtl2 v7.8h, v19.16b216
+ st1 {v0.8h-v3.8h}, x0, x1217
+ st1 {v4.8h-v7.8h}, x0, x1218
+.endr219
+ cbnz w12, .loop_cps32_sve220
+ ret221
+.vl_gt_16_blockcopy_ps_32_32:222
+ cmp x9, #48223
+ bgt .vl_gt_48_blockcopy_ps_32_32224
+ ptrue p0.b, vl32225
+.rept 32226
+ ld1b {z2.h}, p0/z, x2227
+ ld1b {z3.h}, p0/z, x2, #1, mul vl228
+ st1h {z2.h}, p0, x0229
+ st1h {z3.h}, p0, x0, #1, mul vl230
+ add x0, x0, x1, lsl #1231
+ add x2, x2, x3232
+.endr233
+ ret234
+.vl_gt_48_blockcopy_ps_32_32:235
+ ptrue p0.b, vl64236
+.rept 32237
+ ld1b {z2.h}, p0/z, x2238
+ st1h {z2.h}, p0, x0239
+ add x0, x0, x1, lsl #1240
+ add x2, x2, x3241
+.endr242
+ ret243
+endfunc244
+245
+function PFX(blockcopy_ps_64x64_sve)246
+ rdvl x9, #1247
+ cmp x9, #16248
+ bgt .vl_gt_16_blockcopy_ps_64_64249
+ lsl x1, x1, #1250
+ sub x1, x1, #64251
+ mov w12, #16252
+.loop_cps64_sve:253
+ sub w12, w12, #1254
+.rept 4255
+ ld1 {v16.16b-v19.16b}, x2, x3256
+ uxtl v0.8h, v16.8b257
+ uxtl2 v1.8h, v16.16b258
+ uxtl v2.8h, v17.8b259
+ uxtl2 v3.8h, v17.16b260
+ uxtl v4.8h, v18.8b261
+ uxtl2 v5.8h, v18.16b262
+ uxtl v6.8h, v19.8b263
+ uxtl2 v7.8h, v19.16b264
+ st1 {v0.8h-v3.8h}, x0, #64265
+ st1 {v4.8h-v7.8h}, x0, x1266
+.endr267
+ cbnz w12, .loop_cps64_sve268
+ ret269
+.vl_gt_16_blockcopy_ps_64_64:270
+ cmp x9, #48271
+ bgt .vl_gt_48_blockcopy_ps_64_64272
+ ptrue p0.b, vl32273
+.rept 64274
+ ld1b {z4.h}, p0/z, x2275
+ ld1b {z5.h}, p0/z, x2, #1, mul vl276
+ ld1b {z6.h}, p0/z, x2, #2, mul vl277
+ ld1b {z7.h}, p0/z, x2, #3, mul vl278
+ st1h {z4.h}, p0, x0279
+ st1h {z5.h}, p0, x0, #1, mul vl280
+ st1h {z6.h}, p0, x0, #2, mul vl281
+ st1h {z7.h}, p0, x0, #3, mul vl282
+ add x0, x0, x1, lsl #1283
+ add x2, x2, x3284
+.endr285
+ ret286
+.vl_gt_48_blockcopy_ps_64_64:287
+ cmp x9, #112288
+ bgt .vl_gt_112_blockcopy_ps_64_64289
+ ptrue p0.b, vl64290
+.rept 64291
+ ld1b {z4.h}, p0/z, x2292
+ ld1b {z5.h}, p0/z, x2, #1, mul vl293
+ st1h {z4.h}, p0, x0294
+ st1h {z5.h}, p0, x0, #1, mul vl295
+ add x0, x0, x1, lsl #1296
+ add x2, x2, x3297
+.endr298
+ ret299
+.vl_gt_112_blockcopy_ps_64_64:300
+ ptrue p0.b, vl128301
+.rept 64302
+ ld1b {z4.h}, p0/z, x2303
+ st1h {z4.h}, p0, x0304
+ add x0, x0, x1, lsl #1305
+ add x2, x2, x3306
+.endr307
+ ret308
+309
+endfunc310
+311
+function PFX(blockcopy_ss_16x16_sve)312
+ rdvl x9, #1313
+ cmp x9, #16314
+ bgt .vl_gt_16_blockcopy_ss_16_16315
+ lsl x1, x1, #1316
+ lsl x3, x3, #1317
+.rept 8318
+ ld1 {v0.8h-v1.8h}, x2, x3319
+ ld1 {v2.8h-v3.8h}, x2, x3320
+ st1 {v0.8h-v1.8h}, x0, x1321
+ st1 {v2.8h-v3.8h}, x0, x1322
+.endr323
+ ret324
+.vl_gt_16_blockcopy_ss_16_16:325
+ ptrue p0.h, vl16326
+.rept 16327
+ ld1h {z0.h}, p0/z, x2328
+ st1h {z0.h}, p0, x0329
+ add x2, x2, x3, lsl #1330
+ add x0, x0, x1, lsl #1331
+.endr332
+ ret333
+endfunc334
+335
+function PFX(blockcopy_ss_32x32_sve)336
+ rdvl x9, #1337
+ cmp x9, #16338
+ bgt .vl_gt_16_blockcopy_ss_32_32339
+ lsl x1, x1, #1340
+ lsl x3, x3, #1341
+ mov w12, #4342
+.loop_css32_sve:343
+ sub w12, w12, #1344
+.rept 8345
+ ld1 {v0.8h-v3.8h}, x2, x3346
+ st1 {v0.8h-v3.8h}, x0, x1347
+.endr348
+ cbnz w12, .loop_css32_sve349
+ ret350
+.vl_gt_16_blockcopy_ss_32_32:351
+ cmp x9, #48352
+ bgt .vl_gt_48_blockcopy_ss_32_32353
+ ptrue p0.h, vl16354
+.rept 32355
+ ld1h {z0.h}, p0/z, x2356
+ ld1h {z1.h}, p0/z, x2, #1, mul vl357
+ st1h {z0.h}, p0, x0358
+ st1h {z1.h}, p0, x0, #1, mul vl359
+ add x2, x2, x3, lsl #1360
+ add x0, x0, x1, lsl #1361
+.endr362
+ ret363
+.vl_gt_48_blockcopy_ss_32_32:364
+ ptrue p0.h, vl32365
+.rept 32366
+ ld1h {z0.h}, p0/z, x2367
+ st1h {z0.h}, p0, x0368
+ add x2, x2, x3, lsl #1369
+ add x0, x0, x1, lsl #1370
+.endr371
+ ret372
+endfunc373
+374
+function PFX(blockcopy_ss_64x64_sve)375
+ rdvl x9, #1376
+ cmp x9, #16377
+ bgt .vl_gt_16_blockcopy_ss_64_64378
+ lsl x1, x1, #1379
+ sub x1, x1, #64380
+ lsl x3, x3, #1381
+ sub x3, x3, #64382
+ mov w12, #8383
+.loop_css64_sve:384
+ sub w12, w12, #1385
+.rept 8386
+ ld1 {v0.8h-v3.8h}, x2, #64387
+ ld1 {v4.8h-v7.8h}, x2, x3388
+ st1 {v0.8h-v3.8h}, x0, #64389
+ st1 {v4.8h-v7.8h}, x0, x1390
+.endr391
+ cbnz w12, .loop_css64_sve392
+ ret393
+.vl_gt_16_blockcopy_ss_64_64:394
+ cmp x9, #48395
+ bgt .vl_gt_48_blockcopy_ss_64_64396
+ mov w12, #8397
+ ptrue p0.b, vl32398
+.vl_gt_16_loop_css64_sve:399
+ sub w12, w12, #1400
+.rept 8401
+ ld1b {z0.b}, p0/z, x2402
+ ld1b {z1.b}, p0/z, x2, #1, mul vl403
+ ld1b {z2.b}, p0/z, x2, #2, mul vl404
+ ld1b {z3.b}, p0/z, x2, #3, mul vl405
+ st1b {z0.b}, p0, x0406
+ st1b {z1.b}, p0, x0, #1, mul vl407
+ st1b {z2.b}, p0, x0, #2, mul vl408
+ st1b {z3.b}, p0, x0, #3, mul vl409
+ add x2, x2, x3, lsl #1410
+ add x0, x0, x1, lsl #1411
+.endr412
+ cbnz w12, .vl_gt_16_loop_css64_sve413
+ ret414
+.vl_gt_48_blockcopy_ss_64_64:415
+ cmp x9, #112416
+ bgt .vl_gt_112_blockcopy_ss_64_64417
+ mov w12, #8418
+ ptrue p0.b, vl64419
+.vl_gt_48_loop_css64_sve:420
+ sub w12, w12, #1421
+.rept 8422
+ ld1b {z0.b}, p0/z, x2423
+ ld1b {z1.b}, p0/z, x2, #1, mul vl424
+ st1b {z0.b}, p0, x0425
+ st1b {z1.b}, p0, x0, #1, mul vl426
+ add x2, x2, x3, lsl #1427
+ add x0, x0, x1, lsl #1428
+.endr429
+ cbnz w12, .vl_gt_48_loop_css64_sve430
+ ret431
+.vl_gt_112_blockcopy_ss_64_64:432
+ mov w12, #8433
+ ptrue p0.b, vl128434
+.vl_gt_112_loop_css64_sve:435
+ sub w12, w12, #1436
+.rept 8437
+ ld1b {z0.b}, p0/z, x2438
+ st1b {z0.b}, p0, x0439
+ add x2, x2, x3, lsl #1440
+ add x0, x0, x1, lsl #1441
+.endr442
+ cbnz w12, .vl_gt_112_loop_css64_sve443
+ ret444
+endfunc445
+446
+/******** Chroma blockcopy********/447
+function PFX(blockcopy_ss_16x32_sve)448
+ rdvl x9, #1449
+ cmp x9, #16450
+ bgt .vl_gt_16_blockcopy_ss_16_32451
+ lsl x1, x1, #1452
+ lsl x3, x3, #1453
+.rept 16454
+ ld1 {v0.8h-v1.8h}, x2, x3455
+ ld1 {v2.8h-v3.8h}, x2, x3456
+ st1 {v0.8h-v1.8h}, x0, x1457
+ st1 {v2.8h-v3.8h}, x0, x1458
+.endr459
+ ret460
+.vl_gt_16_blockcopy_ss_16_32:461
+ ptrue p0.h, vl16462
+.rept 32463
+ ld1h {z0.h}, p0/z, x2464
+ st1h {z0.h}, p0, x0465
+ add x2, x2, x3, lsl #1466
+ add x0, x0, x1, lsl #1467
+.endr468
+ ret469
+endfunc470
+471
+function PFX(blockcopy_ss_32x64_sve)472
+ rdvl x9, #1473
+ cmp x9, #16474
+ bgt .vl_gt_16_blockcopy_ss_32_64475
+ lsl x1, x1, #1476
+ lsl x3, x3, #1477
+ mov w12, #8478
+.loop_css32x64_sve:479
+ sub w12, w12, #1480
+.rept 8481
+ ld1 {v0.8h-v3.8h}, x2, x3482
+ st1 {v0.8h-v3.8h}, x0, x1483
+.endr484
+ cbnz w12, .loop_css32x64_sve485
+ ret486
+.vl_gt_16_blockcopy_ss_32_64:487
+ cmp x9, #48488
+ bgt .vl_gt_48_blockcopy_ss_32_64489
+ mov w12, #8490
+ ptrue p0.b, vl32491
+.vl_gt_32_loop_css32x64_sve:492
+ sub w12, w12, #1493
+.rept 8494
+ ld1b {z0.b}, p0/z, x2495
+ ld1b {z1.b}, p0/z, x2, #1, mul vl496
+ st1b {z0.b}, p0, x0497
+ st1b {z1.b}, p0, x0, #1, mul vl498
+ add x2, x2, x3, lsl #1499
+ add x0, x0, x1, lsl #1500
+.endr501
+ cbnz w12, .vl_gt_32_loop_css32x64_sve502
+ ret503
+.vl_gt_48_blockcopy_ss_32_64:504
+ mov w12, #8505
+ ptrue p0.b, vl64506
+.vl_gt_48_loop_css32x64_sve:507
+ sub w12, w12, #1508
+.rept 8509
+ ld1b {z0.b}, p0/z, x2510
+ st1b {z0.b}, p0, x0511
+ add x2, x2, x3, lsl #1512
+ add x0, x0, x1, lsl #1513
+.endr514
+ cbnz w12, .vl_gt_48_loop_css32x64_sve515
+ ret516
+endfunc517
+518
+// chroma blockcopy_ps519
+function PFX(blockcopy_ps_4x8_sve)520
+ ptrue p0.h, vl4521
+.rept 8522
+ ld1b {z0.h}, p0/z, x2523
+ st1h {z0.h}, p0, x0524
+ add x0, x0, x1, lsl #1525
+ add x2, x2, x3526
+.endr527
+ ret528
+endfunc529
+530
+function PFX(blockcopy_ps_8x16_sve)531
+ ptrue p0.h, vl8532
+.rept 16533
+ ld1b {z0.h}, p0/z, x2534
+ st1h {z0.h}, p0, x0535
+ add x0, x0, x1, lsl #1536
+ add x2, x2, x3537
+.endr538
+ ret539
+endfunc540
+541
+function PFX(blockcopy_ps_16x32_sve)542
+ rdvl x9, #1543
+ cmp x9, #16544
+ bgt .vl_gt_16_blockcopy_ps_16_32545
+ lsl x1, x1, #1546
+.rept 16547
+ ld1 {v4.16b}, x2, x3548
+ ld1 {v5.16b}, x2, x3549
+ uxtl v0.8h, v4.8b550
+ uxtl2 v1.8h, v4.16b551
+ uxtl v2.8h, v5.8b552
+ uxtl2 v3.8h, v5.16b553
+ st1 {v0.8h-v1.8h}, x0, x1554
+ st1 {v2.8h-v3.8h}, x0, x1555
+.endr556
+ ret557
+.vl_gt_16_blockcopy_ps_16_32:558
+ ptrue p0.b, vl32559
+.rept 32560
+ ld1b {z1.h}, p0/z, x2561
+ st1h {z1.h}, p0, x0562
+ add x0, x0, x1, lsl #1563
+ add x2, x2, x3564
+.endr565
+ ret566
+endfunc567
+568
+function PFX(blockcopy_ps_32x64_sve)569
+ rdvl x9, #1570
+ cmp x9, #16571
+ bgt .vl_gt_16_blockcopy_ps_32_64572
+ lsl x1, x1, #1573
+ mov w12, #8574
+.loop_cps32x64_sve:575
+ sub w12, w12, #1576
+.rept 4577
+ ld1 {v16.16b-v17.16b}, x2, x3578
+ ld1 {v18.16b-v19.16b}, x2, x3579
+ uxtl v0.8h, v16.8b580
+ uxtl2 v1.8h, v16.16b581
+ uxtl v2.8h, v17.8b582
+ uxtl2 v3.8h, v17.16b583
+ uxtl v4.8h, v18.8b584
+ uxtl2 v5.8h, v18.16b585
+ uxtl v6.8h, v19.8b586
+ uxtl2 v7.8h, v19.16b587
+ st1 {v0.8h-v3.8h}, x0, x1588
+ st1 {v4.8h-v7.8h}, x0, x1589
+.endr590
+ cbnz w12, .loop_cps32x64_sve591
+ ret592
+.vl_gt_16_blockcopy_ps_32_64:593
+ cmp x9, #48594
+ bgt .vl_gt_48_blockcopy_ps_32_64595
+ ptrue p0.b, vl32596
+.rept 64597
+ ld1b {z2.h}, p0/z, x2598
+ ld1b {z3.h}, p0/z, x2, #1, mul vl599
+ st1h {z2.h}, p0, x0600
+ st1h {z3.h}, p0, x0, #1, mul vl601
+ add x0, x0, x1, lsl #1602
+ add x2, x2, x3603
+.endr604
+ ret605
+.vl_gt_48_blockcopy_ps_32_64:606
+ ptrue p0.b, vl64607
+.rept 64608
+ ld1b {z2.h}, p0/z, x2609
+ st1h {z2.h}, p0, x0610
+ add x0, x0, x1, lsl #1611
+ add x2, x2, x3612
+.endr613
+ ret614
+endfunc615
+616
+// chroma blockcopy_sp617
+function PFX(blockcopy_sp_4x8_sve)618
+ ptrue p0.h, vl4619
+.rept 8620
+ ld1h {z0.h}, p0/z, x2621
+ st1b {z0.h}, p0, x0622
+ add x2, x2, x3, lsl #1623
+ add x0, x0, x1624
+.endr625
+ ret626
+endfunc627
+628
+function PFX(blockcopy_sp_8x16_sve)629
+ ptrue p0.h, vl8630
+.rept 16631
+ ld1h {z0.h}, p0/z, x2632
+ st1b {z0.h}, p0, x0633
+ add x2, x2, x3, lsl #1634
+ add x0, x0, x1635
+.endr636
+ ret637
+endfunc638
+639
+function PFX(blockcopy_sp_16x32_sve)640
+ rdvl x9, #1641
+ cmp x9, #16642
+ bgt .vl_gt_16_blockcopy_sp_16_32643
+ ptrue p0.h, vl8644
+.rept 32645
+ ld1h {z0.h}, p0/z, x2646
+ ld1h {z1.h}, p0/z, x2, #1, mul vl647
+ st1b {z0.h}, p0, x0648
+ st1b {z1.h}, p0, x0, #1, mul vl649
+ add x2, x2, x3, lsl #1650
+ add x0, x0, x1651
+.endr652
+ ret653
+.vl_gt_16_blockcopy_sp_16_32:654
+ ptrue p0.h, vl16655
+.rept 32656
+ ld1h {z0.h}, p0/z, x2657
+ st1b {z0.h}, p0, x0658
+ add x2, x2, x3, lsl #1659
+ add x0, x0, x1660
+.endr661
+ ret662
+endfunc663
+664
+function PFX(blockcopy_sp_32x64_sve)665
+ rdvl x9, #1666
+ cmp x9, #16667
+ bgt .vl_gt_16_blockcopy_sp_32_64668
+ ptrue p0.h, vl8669
+.rept 64670
+ ld1h {z0.h}, p0/z, x2671
+ ld1h {z1.h}, p0/z, x2, #1, mul vl672
+ ld1h {z2.h}, p0/z, x2, #2, mul vl673
+ ld1h {z3.h}, p0/z, x2, #3, mul vl674
+ st1b {z0.h}, p0, x0675
+ st1b {z1.h}, p0, x0, #1, mul vl676
+ st1b {z2.h}, p0, x0, #2, mul vl677
+ st1b {z3.h}, p0, x0, #3, mul vl678
+ add x2, x2, x3, lsl #1679
+ add x0, x0, x1680
+.endr681
+ ret682
+.vl_gt_16_blockcopy_sp_32_64:683
+ cmp x9, #48684
+ bgt .vl_gt_48_blockcopy_sp_32_64685
+ ptrue p0.h, vl16686
+.rept 64687
+ ld1h {z0.h}, p0/z, x2688
+ ld1h {z1.h}, p0/z, x2, #1, mul vl689
+ st1b {z0.h}, p0, x0690
+ st1b {z1.h}, p0, x0, #1, mul vl691
+ add x2, x2, x3, lsl #1692
+ add x0, x0, x1693
+.endr694
+ ret695
+.vl_gt_48_blockcopy_sp_32_64:696
+ ptrue p0.h, vl32697
+.rept 64698
+ ld1h {z0.h}, p0/z, x2699
+ st1b {z0.h}, p0, x0700
+ add x2, x2, x3, lsl #1701
+ add x0, x0, x1702
+.endr703
+ ret704
+endfunc705
+706
+/* blockcopy_pp(pixel* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride) */707
+708
+function PFX(blockcopy_pp_32x8_sve)709
+ rdvl x9, #1710
+ cmp x9, #16711
+ bgt .vl_gt_16_blockcopy_pp_32_8712
+.rept 8713
+ ld1 {v0.16b-v1.16b}, x2, x3714
+ st1 {v0.16b-v1.16b}, x0, x1715
+.endr716
+ ret717
+.vl_gt_16_blockcopy_pp_32_8:718
+ ptrue p0.b, vl32719
+.rept 8720
+ ld1b {z0.b}, p0/z, x2721
+ st1b {z0.b}, p0, x0722
+ add x2, x2, x3723
+ add x0, x0, x1724
+.endr725
+ ret726
+endfunc727
+728
+.macro blockcopy_pp_32xN_sve h729
+function PFX(blockcopy_pp_32x\h\()_sve)730
+ mov w12, #\h / 8731
+ rdvl x9, #1732
+ cmp x9, #16733
+ bgt .vl_gt_16_blockcopy_pp_32xN_\h734
+.loop_sve_32x\h\():735
+ sub w12, w12, #1736
+.rept 8737
+ ld1 {v0.16b-v1.16b}, x2, x3738
+ st1 {v0.16b-v1.16b}, x0, x1739
+.endr740
+ cbnz w12, .loop_sve_32x\h741
+ ret742
+.vl_gt_16_blockcopy_pp_32xN_\h:743
+ ptrue p0.b, vl32744
+.L_gt_16_blockcopy_pp_32xN_\h:745
+ sub w12, w12, #1746
+.rept 8747
+ ld1b {z0.b}, p0/z, x2748
+ st1b {z0.b}, p0, x0749
+ add x2, x2, x3750
+ add x0, x0, x1751
+.endr752
+ cbnz w12, .L_gt_16_blockcopy_pp_32xN_\h753
+ ret754
+endfunc755
+.endm756
+757
+blockcopy_pp_32xN_sve 16758
+blockcopy_pp_32xN_sve 24759
+blockcopy_pp_32xN_sve 32760
+blockcopy_pp_32xN_sve 64761
+blockcopy_pp_32xN_sve 48762
+763
+.macro blockcopy_pp_64xN_sve h764
+function PFX(blockcopy_pp_64x\h\()_sve)765
+ mov w12, #\h / 4766
+ rdvl x9, #1767
+ cmp x9, #16768
+ bgt .vl_gt_16_blockcopy_pp_64xN_\h769
+.loop_sve_64x\h\():770
+ sub w12, w12, #1771
+.rept 4772
+ ld1 {v0.16b-v3.16b}, x2, x3773
+ st1 {v0.16b-v3.16b}, x0, x1774
+.endr775
+ cbnz w12, .loop_sve_64x\h776
+ ret777
+.vl_gt_16_blockcopy_pp_64xN_\h:778
+ cmp x9, #48779
+ bgt .vl_gt_48_blockcopy_pp_64xN_\h780
+ ptrue p0.b, vl32781
+.L_le_32_blockcopy_pp_64xN_\h:782
+ sub w12, w12, #1783
+.rept 4784
+ ld1b {z0.b}, p0/z, x2785
+ ld1b {z1.b}, p0/z, x2, #1, mul vl786
+ st1b {z0.b}, p0, x0787
+ st1b {z1.b}, p0, x0, #1, mul vl788
+ add x2, x2, x3789
+ add x0, x0, x1790
+.endr791
+ cbnz w12, .L_le_32_blockcopy_pp_64xN_\h792
+ ret793
+.vl_gt_48_blockcopy_pp_64xN_\h:794
+ ptrue p0.b, vl64795
+.L_blockcopy_pp_64xN_\h:796
+ sub w12, w12, #1797
+.rept 4798
+ ld1b {z0.b}, p0/z, x2799
+ st1b {z0.b}, p0, x0800
+ add x2, x2, x3801
+ add x0, x0, x1802
+.endr803
+ cbnz w12, .L_blockcopy_pp_64xN_\h804
+ ret805
+endfunc806
+.endm807
+808
+blockcopy_pp_64xN_sve 16809
+blockcopy_pp_64xN_sve 32810
+blockcopy_pp_64xN_sve 48811
+blockcopy_pp_64xN_sve 64812
+813
+function PFX(blockfill_s_32x32_sve)814
+ rdvl x9, #1815
+ cmp x9, #16816
+ bgt .vl_gt_16_blockfill_s_32_32817
+ dup v0.8h, w2818
+ mov v1.16b, v0.16b819
+ mov v2.16b, v0.16b820
+ mov v3.16b, v0.16b821
+ lsl x1, x1, #1822
+.rept 32823
+ st1 {v0.8h-v3.8h}, x0, x1824
+.endr825
+ ret826
+.vl_gt_16_blockfill_s_32_32:827
+ cmp x9, #48828
+ bgt .vl_gt_48_blockfill_s_32_32829
+ dup z0.h, w2830
+ ptrue p0.h, vl16831
+.rept 32832
+ st1h {z0.h}, p0, x0833
+ st1h {z0.h}, p0, x0, #1, mul vl834
+ add x0, x0, x1, lsl #1835
+.endr836
+ ret837
+.vl_gt_48_blockfill_s_32_32:838
+ dup z0.h, w2839
+ ptrue p0.h, vl32840
+.rept 32841
+ st1h {z0.h}, p0, x0842
+ add x0, x0, x1, lsl #1843
+.endr844
+ ret845
+endfunc846
+847
+// void cpy2Dto1D_shl(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift)848
+.macro cpy2Dto1D_shl_start_sve849
+ add x2, x2, x2850
+ mov z0.h, w3851
+.endm852
+853
+function PFX(cpy2Dto1D_shl_16x16_sve)854
+ dup z0.h, w3855
+ rdvl x9, #1856
+ cmp x9, #16857
+ bgt .vl_gt_16_cpy2Dto1D_shl_16x16858
+ cpy2Dto1D_shl_start_sve859
+ mov w12, #4860
+.loop_cpy2Dto1D_shl_16_sve:861
+ sub w12, w12, #1862
+.rept 4863
+ ld1 {v2.16b-v3.16b}, x1, x2864
+ sshl v2.8h, v2.8h, v0.8h865
+ sshl v3.8h, v3.8h, v0.8h866
+ st1 {v2.16b-v3.16b}, x0, #32867
+.endr868
+ cbnz w12, .loop_cpy2Dto1D_shl_16_sve869
+ ret870
+.vl_gt_16_cpy2Dto1D_shl_16x16:871
+ ptrue p0.h, vl16872
+.rept 16873
+ ld1h {z1.h}, p0/z, x1874
+ lsl z1.h, p0/m, z1.h, z0.h875
+ st1h {z1.h}, p0, x0876
+ add x1, x1, x2, lsl #1877
+ add x0, x0, #32878
+.endr879
+ ret880
+endfunc881
+882
+function PFX(cpy2Dto1D_shl_32x32_sve)883
+ dup z0.h, w3884
+ rdvl x9, #1885
+ cmp x9, #16886
+ bgt .vl_gt_16_cpy2Dto1D_shl_32x32887
+ cpy2Dto1D_shl_start_sve888
+ mov w12, #16889
+.loop_cpy2Dto1D_shl_32_sve:890
+ sub w12, w12, #1891
+.rept 2892
+ ld1 {v2.16b-v5.16b}, x1, x2893
+ sshl v2.8h, v2.8h, v0.8h894
+ sshl v3.8h, v3.8h, v0.8h895
+ sshl v4.8h, v4.8h, v0.8h896
+ sshl v5.8h, v5.8h, v0.8h897
+ st1 {v2.16b-v5.16b}, x0, #64898
+.endr899
+ cbnz w12, .loop_cpy2Dto1D_shl_32_sve900
+ ret901
+.vl_gt_16_cpy2Dto1D_shl_32x32:902
+ cmp x9, #48903
+ bgt .vl_gt_48_cpy2Dto1D_shl_32x32904
+ ptrue p0.h, vl16905
+.rept 32906
+ ld1h {z1.h}, p0/z, x1907
+ ld1h {z2.h}, p0/z, x1, #1, mul vl908
+ lsl z1.h, p0/m, z1.h, z0.h909
+ lsl z2.h, p0/m, z2.h, z0.h910
+ st1h {z1.h}, p0, x0911
+ st1h {z2.h}, p0, x0, #1, mul vl912
+ add x1, x1, x2, lsl #1913
+ add x0, x0, #64914
+.endr915
+ ret916
+.vl_gt_48_cpy2Dto1D_shl_32x32:917
+ ptrue p0.h, vl32918
+.rept 32919
+ ld1h {z1.h}, p0/z, x1920
+ lsl z1.h, p0/m, z1.h, z0.h921
+ st1h {z1.h}, p0, x0922
+ add x1, x1, x2, lsl #1923
+ add x0, x0, #64924
+.endr925
+ ret926
+endfunc927
+928
+function PFX(cpy2Dto1D_shl_64x64_sve)929
+ rdvl x9, #1930
+ cmp x9, #16931
+ bgt .vl_gt_16_cpy2Dto1D_shl_64x64932
+ cpy2Dto1D_shl_start_sve933
+ mov w12, #32934
+ sub x2, x2, #64935
+.loop_cpy2Dto1D_shl_64_sve:936
+ sub w12, w12, #1937
+.rept 2938
+ ld1 {v2.16b-v5.16b}, x1, #64939
+ ld1 {v16.16b-v19.16b}, x1, x2940
+ sshl v2.8h, v2.8h, v0.8h941
+ sshl v3.8h, v3.8h, v0.8h942
+ sshl v4.8h, v4.8h, v0.8h943
+ sshl v5.8h, v5.8h, v0.8h944
+ sshl v16.8h, v16.8h, v0.8h945
+ sshl v17.8h, v17.8h, v0.8h946
+ sshl v18.8h, v18.8h, v0.8h947
+ sshl v19.8h, v19.8h, v0.8h948
+ st1 {v2.16b-v5.16b}, x0, #64949
+ st1 {v16.16b-v19.16b}, x0, #64950
+.endr951
+ cbnz w12, .loop_cpy2Dto1D_shl_64_sve952
+ ret953
+.vl_gt_16_cpy2Dto1D_shl_64x64:954
+ dup z0.h, w3955
+ mov x8, #64956
+ mov w12, #64957
+.L_init_cpy2Dto1D_shl_64x64:958
+ sub w12, w12, 1959
+ mov x9, #0960
+ whilelt p0.h, x9, x8961
+.L_cpy2Dto1D_shl_64x64:962
+ ld1h {z1.h}, p0/z, x1, x9, lsl #1963
+ lsl z1.h, p0/m, z1.h, z0.h964
+ st1h {z1.h}, p0, x0, x9, lsl #1965
+ inch x9966
+ whilelt p0.h, x9, x8967
+ b.first .L_cpy2Dto1D_shl_64x64968
+ add x1, x1, x2, lsl #1969
+ addvl x0, x0, #1970
+ cbnz w12, .L_init_cpy2Dto1D_shl_64x64971
+ ret972
+endfunc973
+974
+// void cpy2Dto1D_shr(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift)975
+976
+function PFX(cpy2Dto1D_shr_4x4_sve)977
+ dup z0.h, w3978
+ sub w4, w3, #1979
+ dup z1.h, w4980
+ ptrue p0.h, vl8981
+ mov z2.h, #1982
+ lsl z2.h, p0/m, z2.h, z1.h983
+ lsl x2, x2, #1984
+ index z3.d, #0, x2985
+ index z4.d, #0, #8986
+.rept 2987
+ ld1d {z5.d}, p0/z, x1, z3.d988
+ add x1, x1, x2, lsl #1989
+ add z5.h, p0/m, z5.h, z2.h990
+ asr z5.h, p0/m, z5.h, z0.h991
+ st1d {z5.d}, p0, x0, z4.d992
+ add x0, x0, #16993
+.endr994
+ ret995
+endfunc996
+997
+function PFX(cpy2Dto1D_shr_8x8_sve)998
+ dup z0.h, w3999
+ sub w4, w3, #11000
+ dup z1.h, w41001
+ ptrue p0.h, vl81002
+ mov z2.h, #11003
+ lsl z2.h, p0/m, z2.h, z1.h1004
+.rept 81005
+ ld1d {z5.d}, p0/z, x11006
+ add x1, x1, x2, lsl #11007
+ add z5.h, p0/m, z5.h, z2.h1008
+ asr z5.h, p0/m, z5.h, z0.h1009
+ st1d {z5.d}, p0, x01010
+ add x0, x0, #161011
+.endr1012
+ ret1013
+endfunc1014
+1015
+function PFX(cpy2Dto1D_shr_16x16_sve)1016
+ dup z0.h, w31017
+ sub w4, w3, #11018
+ dup z1.h, w41019
+ rdvl x9, #11020
+ cmp x9, #161021
+ bgt .vl_gt_16_cpy2Dto1D_shr_16x161022
+ ptrue p0.h, vl81023
+ mov z2.h, #11024
+ lsl z2.h, p0/m, z2.h, z1.h1025
+.rept 161026
+ ld1d {z5.d}, p0/z, x11027
+ ld1d {z6.d}, p0/z, x1, #1, mul vl1028
+ add x1, x1, x2, lsl #11029
+ add z5.h, p0/m, z5.h, z2.h1030
+ add z6.h, p0/m, z6.h, z2.h1031
+ asr z5.h, p0/m, z5.h, z0.h1032
+ asr z6.h, p0/m, z6.h, z0.h1033
+ st1d {z5.d}, p0, x01034
+ st1d {z6.d}, p0, x0, #1, mul vl1035
+ add x0, x0, #321036
+.endr1037
+ ret1038
+.vl_gt_16_cpy2Dto1D_shr_16x16:1039
+ ptrue p0.h, vl161040
+ mov z2.h, #11041
+ lsl z2.h, p0/m, z2.h, z1.h1042
+.rept 161043
+ ld1d {z5.d}, p0/z, x11044
+ add x1, x1, x2, lsl #11045
+ add z5.h, p0/m, z5.h, z2.h1046
+ asr z5.h, p0/m, z5.h, z0.h1047
+ st1d {z5.d}, p0, x01048
+ add x0, x0, #321049
+.endr1050
+ ret1051
+endfunc1052
+1053
+function PFX(cpy2Dto1D_shr_32x32_sve)1054
+ rdvl x9, #11055
+ cmp x9, #161056
+ bgt .vl_gt_16_cpy2Dto1D_shr_32x321057
+ cpy2Dto1D_shr_start1058
+ mov w12, #161059
+.loop_cpy2Dto1D_shr_32_sve:1060
+ sub w12, w12, #11061
+.rept 21062
+ ld1 {v2.8h-v5.8h}, x1, x21063
+ sub v2.8h, v2.8h, v1.8h1064
+ sub v3.8h, v3.8h, v1.8h1065
+ sub v4.8h, v4.8h, v1.8h1066
+ sub v5.8h, v5.8h, v1.8h1067
+ sshl v2.8h, v2.8h, v0.8h1068
+ sshl v3.8h, v3.8h, v0.8h1069
+ sshl v4.8h, v4.8h, v0.8h1070
+ sshl v5.8h, v5.8h, v0.8h1071
+ st1 {v2.8h-v5.8h}, x0, #641072
+.endr1073
+ cbnz w12, .loop_cpy2Dto1D_shr_32_sve1074
+ ret1075
+.vl_gt_16_cpy2Dto1D_shr_32x32:1076
+ dup z0.h, w31077
+ sub w4, w3, #11078
+ dup z1.h, w41079
+ cmp x9, #481080
+ bgt .vl_gt_48_cpy2Dto1D_shr_32x321081
+ ptrue p0.h, vl161082
+ mov z2.h, #11083
+ lsl z2.h, p0/m, z2.h, z1.h1084
+.rept 321085
+ ld1d {z5.d}, p0/z, x11086
+ ld1d {z6.d}, p0/z, x1, #1, mul vl1087
+ add x1, x1, x2, lsl #11088
+ add z5.h, p0/m, z5.h, z2.h1089
+ add z6.h, p0/m, z6.h, z2.h1090
+ asr z5.h, p0/m, z5.h, z0.h1091
+ asr z6.h, p0/m, z6.h, z0.h1092
+ st1d {z5.d}, p0, x01093
+ st1d {z6.d}, p0, x0, #1, mul vl1094
+ add x0, x0, #641095
+.endr1096
+ ret1097
+.vl_gt_48_cpy2Dto1D_shr_32x32:1098
+ ptrue p0.h, vl321099
+ mov z2.h, #11100
+ lsl z2.h, p0/m, z2.h, z1.h1101
+.rept 321102
+ ld1d {z5.d}, p0/z, x11103
+ add x1, x1, x2, lsl #11104
+ add z5.h, p0/m, z5.h, z2.h1105
+ asr z5.h, p0/m, z5.h, z0.h1106
+ st1d {z5.d}, p0, x01107
+ add x0, x0, #641108
+.endr1109
+ ret1110
+endfunc1111
+1112
+// void cpy1Dto2D_shl(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift)1113
+1114
+function PFX(cpy1Dto2D_shl_16x16_sve)1115
+ dup z0.h, w31116
+ rdvl x9, #11117
+ cmp x9, #161118
+ bgt .vl_gt_16_cpy1Dto2D_shl_16x161119
+ ptrue p0.h, vl81120
+.rept 161121
+ ld1h {z1.h}, p0/z, x11122
+ ld1h {z2.h}, p0/z, x1, #1, mul vl1123
+ lsl z1.h, p0/m, z1.h, z0.h1124
+ lsl z2.h, p0/m, z2.h, z0.h1125
+ st1h {z1.h}, p0, x01126
+ st1h {z2.h}, p0, x0, #1, mul vl1127
+ add x1, x1, #321128
+ add x0, x0, x2, lsl #11129
+.endr1130
+ ret1131
+.vl_gt_16_cpy1Dto2D_shl_16x16:1132
+ ptrue p0.h, vl161133
+.rept 161134
+ ld1h {z1.h}, p0/z, x11135
+ lsl z1.h, p0/m, z1.h, z0.h1136
+ st1h {z1.h}, p0, x01137
+ add x1, x1, #321138
+ add x0, x0, x2, lsl #11139
+.endr1140
+ ret1141
+endfunc1142
+1143
+function PFX(cpy1Dto2D_shl_32x32_sve)1144
+ dup z0.h, w31145
+ rdvl x9, #11146
+ cmp x9, #161147
+ bgt .vl_gt_16_cpy1Dto2D_shl_32x321148
+ ptrue p0.h, vl81149
+.rept 321150
+ ld1h {z1.h}, p0/z, x11151
+ ld1h {z2.h}, p0/z, x1, #1, mul vl1152
+ ld1h {z3.h}, p0/z, x1, #2, mul vl1153
+ ld1h {z4.h}, p0/z, x1, #3, mul vl1154
+ lsl z1.h, p0/m, z1.h, z0.h1155
+ lsl z2.h, p0/m, z2.h, z0.h1156
+ lsl z3.h, p0/m, z3.h, z0.h1157
+ lsl z4.h, p0/m, z4.h, z0.h1158
+ st1h {z1.h}, p0, x01159
+ st1h {z2.h}, p0, x0, #1, mul vl1160
+ st1h {z3.h}, p0, x0, #2, mul vl1161
+ st1h {z4.h}, p0, x0, #3, mul vl1162
+ add x1, x1, #641163
+ add x0, x0, x2, lsl #11164
+.endr1165
+ ret1166
+.vl_gt_16_cpy1Dto2D_shl_32x32:1167
+ cmp x9, #481168
+ bgt .vl_gt_48_cpy1Dto2D_shl_32x321169
+ ptrue p0.h, vl161170
+.rept 321171
+ ld1h {z1.h}, p0/z, x11172
+ ld1h {z2.h}, p0/z, x1, #1, mul vl1173
+ lsl z1.h, p0/m, z1.h, z0.h1174
+ lsl z2.h, p0/m, z2.h, z0.h1175
+ st1h {z1.h}, p0, x01176
+ st1h {z2.h}, p0, x0, #1, mul vl1177
+ add x1, x1, #641178
+ add x0, x0, x2, lsl #11179
+.endr1180
+ ret1181
+.vl_gt_48_cpy1Dto2D_shl_32x32:1182
+ ptrue p0.h, vl321183
+.rept 321184
+ ld1h {z1.h}, p0/z, x11185
+ lsl z1.h, p0/m, z1.h, z0.h1186
+ st1h {z1.h}, p0, x01187
+ add x1, x1, #641188
+ add x0, x0, x2, lsl #11189
+.endr1190
+ ret1191
+endfunc1192
+1193
+function PFX(cpy1Dto2D_shl_64x64_sve)1194
+ dup z0.h, w31195
+ mov x8, #641196
+ mov w12, #641197
+.L_init_cpy1Dto2D_shl_64x64:1198
+ sub w12, w12, 11199
+ mov x9, #01200
+ whilelt p0.h, x9, x81201
+.L_cpy1Dto2D_shl_64x64:1202
+ ld1h {z1.h}, p0/z, x1, x9, lsl #11203
+ lsl z1.h, p0/m, z1.h, z0.h1204
+ st1h {z1.h}, p0, x0, x9, lsl #11205
+ inch x91206
+ whilelt p0.h, x9, x81207
+ b.first .L_cpy1Dto2D_shl_64x641208
+ addvl x1, x1, #11209
+ add x0, x0, x2, lsl #11210
+ cbnz w12, .L_init_cpy1Dto2D_shl_64x641211
+ ret1212
+endfunc1213
+1214
+// void cpy1Dto2D_shr(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift)1215
+1216
+function PFX(cpy1Dto2D_shr_16x16_sve)1217
+ rdvl x9, #11218
+ cmp x9, #161219
+ bgt .vl_gt_16_cpy1Dto2D_shr_16x161220
+ cpy1Dto2D_shr_start1221
+ mov w12, #41222
+.loop_cpy1Dto2D_shr_16:1223
+ sub w12, w12, #11224
+.rept 41225
+ ld1 {v2.8h-v3.8h}, x1, #321226
+ sub v2.8h, v2.8h, v1.8h1227
+ sub v3.8h, v3.8h, v1.8h1228
+ sshl v2.8h, v2.8h, v0.8h1229
+ sshl v3.8h, v3.8h, v0.8h1230
+ st1 {v2.8h-v3.8h}, x0, x21231
+.endr1232
+ cbnz w12, .loop_cpy1Dto2D_shr_161233
+ ret1234
+.vl_gt_16_cpy1Dto2D_shr_16x16:1235
+ dup z0.h, w31236
+ sub w4, w3, #11237
+ dup z1.h, w41238
+ ptrue p0.h, vl161239
+ mov z2.h, #11240
+ lsl z2.h, p0/m, z2.h, z1.h1241
+.rept 161242
+ ld1d {z5.d}, p0/z, x11243
+ add x1, x1, #321244
+ add z5.h, p0/m, z5.h, z2.h1245
+ asr z5.h, p0/m, z5.h, z0.h1246
+ st1d {z5.d}, p0, x01247
+ add x0, x0, x2, lsl #11248
+.endr1249
+ ret1250
+endfunc1251
+1252
+function PFX(cpy1Dto2D_shr_32x32_sve)1253
+ rdvl x9, #11254
+ cmp x9, #161255
+ bgt .vl_gt_16_cpy1Dto2D_shr_32x321256
+ cpy1Dto2D_shr_start1257
+ mov w12, #161258
+.loop_cpy1Dto2D_shr_32_sve:1259
+ sub w12, w12, #11260
+.rept 21261
+ ld1 {v2.16b-v5.16b}, x1, #641262
+ sub v2.8h, v2.8h, v1.8h1263
+ sub v3.8h, v3.8h, v1.8h1264
+ sub v4.8h, v4.8h, v1.8h1265
+ sub v5.8h, v5.8h, v1.8h1266
+ sshl v2.8h, v2.8h, v0.8h1267
+ sshl v3.8h, v3.8h, v0.8h1268
+ sshl v4.8h, v4.8h, v0.8h1269
+ sshl v5.8h, v5.8h, v0.8h1270
+ st1 {v2.16b-v5.16b}, x0, x21271
+.endr1272
+ cbnz w12, .loop_cpy1Dto2D_shr_32_sve1273
+ ret1274
+.vl_gt_16_cpy1Dto2D_shr_32x32:1275
+ dup z0.h, w31276
+ sub w4, w3, #11277
+ dup z1.h, w41278
+ cmp x9, #481279
+ bgt .vl_gt_48_cpy2Dto1D_shr_32x321280
+ ptrue p0.h, vl161281
+ mov z2.h, #11282
+ lsl z2.h, p0/m, z2.h, z1.h1283
+.rept 321284
+ ld1d {z5.d}, p0/z, x11285
+ ld1d {z6.d}, p0/z, x1, #1, mul vl1286
+ add x1, x1, #641287
+ add z5.h, p0/m, z5.h, z2.h1288
+ add z6.h, p0/m, z6.h, z2.h1289
+ asr z5.h, p0/m, z5.h, z0.h1290
+ asr z6.h, p0/m, z6.h, z0.h1291
+ st1d {z5.d}, p0, x01292
+ st1d {z6.d}, p0, x0, #1, mul vl1293
+ add x0, x0, x2, lsl #11294
+.endr1295
+ ret1296
+.vl_gt_48_cpy1Dto2D_shr_32x32:1297
+ ptrue p0.h, vl321298
+ mov z2.h, #11299
+ lsl z2.h, p0/m, z2.h, z1.h1300
+.rept 321301
+ ld1d {z5.d}, p0/z, x11302
+ add x1, x1, #641303
+ add z5.h, p0/m, z5.h, z2.h1304
+ asr z5.h, p0/m, z5.h, z0.h1305
+ st1d {z5.d}, p0, x01306
+ add x0, x0, x2, lsl #11307
+.endr1308
+ ret1309
+endfunc1310
+1311
+function PFX(cpy1Dto2D_shr_64x64_sve)1312
+ dup z0.h, w31313
+ sub w4, w3, #11314
+ dup z1.h, w41315
+ rdvl x9, #11316
+ cmp x9, #161317
+ bgt .vl_gt_16_cpy1Dto2D_shr_64x641318
+ ptrue p0.h, vl81319
+ mov z2.h, #11320
+ lsl z2.h, p0/m, z2.h, z1.h1321
+.rept 1281322
+ ld1d {z5.d}, p0/z, x11323
+ ld1d {z6.d}, p0/z, x1, #1, mul vl1324
+ ld1d {z7.d}, p0/z, x1, #2, mul vl1325
+ ld1d {z8.d}, p0/z, x1, #3, mul vl1326
+ ld1d {z9.d}, p0/z, x1, #4, mul vl1327
+ ld1d {z10.d}, p0/z, x1, #5, mul vl1328
+ ld1d {z11.d}, p0/z, x1, #6, mul vl1329
+ ld1d {z12.d}, p0/z, x1, #7, mul vl1330
+ add x1, x1, #1281331
+ add z5.h, p0/m, z5.h, z2.h1332
+ add z6.h, p0/m, z6.h, z2.h1333
+ add z7.h, p0/m, z7.h, z2.h1334
+ add z8.h, p0/m, z8.h, z2.h1335
+ add z9.h, p0/m, z9.h, z2.h1336
+ add z10.h, p0/m, z10.h, z2.h1337
+ add z11.h, p0/m, z11.h, z2.h1338
+ add z12.h, p0/m, z12.h, z2.h1339
+ asr z5.h, p0/m, z5.h, z0.h1340
+ asr z6.h, p0/m, z6.h, z0.h1341
+ asr z7.h, p0/m, z7.h, z0.h1342
+ asr z8.h, p0/m, z8.h, z0.h1343
+ asr z9.h, p0/m, z9.h, z0.h1344
+ asr z10.h, p0/m, z10.h, z0.h1345
+ asr z11.h, p0/m, z11.h, z0.h1346
+ asr z12.h, p0/m, z12.h, z0.h1347
+ st1d {z5.d}, p0, x01348
+ st1d {z6.d}, p0, x0, #1, mul vl1349
+ st1d {z7.d}, p0, x0, #2, mul vl1350
+ st1d {z8.d}, p0, x0, #3, mul vl1351
+ st1d {z9.d}, p0, x0, #4, mul vl1352
+ st1d {z10.d}, p0, x0, #5, mul vl1353
+ st1d {z11.d}, p0, x0, #6, mul vl1354
+ st1d {z12.d}, p0, x0, #7, mul vl1355
+ add x0, x0, x2, lsl #11356
+.endr1357
+ ret1358
+.vl_gt_16_cpy1Dto2D_shr_64x64:1359
+ cmp x9, #481360
+ bgt .vl_gt_48_cpy1Dto2D_shr_64x641361
+ ptrue p0.h, vl161362
+ mov z2.h, #11363
+ lsl z2.h, p0/m, z2.h, z1.h1364
+.rept 1281365
+ ld1d {z5.d}, p0/z, x11366
+ ld1d {z6.d}, p0/z, x1, #1, mul vl1367
+ ld1d {z7.d}, p0/z, x1, #2, mul vl1368
+ ld1d {z8.d}, p0/z, x1, #3, mul vl1369
+ add x1, x1, #1281370
+ add z5.h, p0/m, z5.h, z2.h1371
+ add z6.h, p0/m, z6.h, z2.h1372
+ add z7.h, p0/m, z7.h, z2.h1373
+ add z8.h, p0/m, z8.h, z2.h1374
+ asr z5.h, p0/m, z5.h, z0.h1375
+ asr z6.h, p0/m, z6.h, z0.h1376
+ asr z7.h, p0/m, z7.h, z0.h1377
+ asr z8.h, p0/m, z8.h, z0.h1378
+ st1d {z5.d}, p0, x01379
+ st1d {z6.d}, p0, x0, #1, mul vl1380
+ st1d {z7.d}, p0, x0, #2, mul vl1381
+ st1d {z8.d}, p0, x0, #3, mul vl1382
+ add x0, x0, x2, lsl #11383
+.endr1384
+ ret1385
+.vl_gt_48_cpy1Dto2D_shr_64x64:1386
+ cmp x9, #1121387
+ bgt .vl_gt_112_cpy1Dto2D_shr_64x641388
+ ptrue p0.h, vl321389
+ mov z2.h, #11390
+ lsl z2.h, p0/m, z2.h, z1.h1391
+.rept 1281392
+ ld1d {z5.d}, p0/z, x11393
+ ld1d {z6.d}, p0/z, x1, #1, mul vl1394
+ add x1, x1, #1281395
+ add z5.h, p0/m, z5.h, z2.h1396
+ add z6.h, p0/m, z6.h, z2.h1397
+ asr z5.h, p0/m, z5.h, z0.h1398
+ asr z6.h, p0/m, z6.h, z0.h1399
+ st1d {z5.d}, p0, x01400
+ st1d {z6.d}, p0, x0, #1, mul vl1401
+ add x0, x0, x2, lsl #11402
+.endr1403
+ ret1404
+.vl_gt_112_cpy1Dto2D_shr_64x64:1405
+ ptrue p0.h, vl641406
+ mov z2.h, #11407
+ lsl z2.h, p0/m, z2.h, z1.h1408
+.rept 1281409
+ ld1d {z5.d}, p0/z, x11410
+ add x1, x1, #1281411
+ add z5.h, p0/m, z5.h, z2.h1412
+ asr z5.h, p0/m, z5.h, z0.h1413
+ st1d {z5.d}, p0, x01414
+ add x0, x0, x2, lsl #11415
+.endr1416
+ ret1417
+endfunc1418
x265_3.6.tar.gz/source/common/aarch64/blockcopy8.S
Added
1301
1
2
+/*****************************************************************************3
+ * Copyright (C) 2021 MulticoreWare, Inc4
+ *5
+ * Authors: Sebastian Pop <spop@amazon.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm.S"26
+#include "blockcopy8-common.S"27
+28
+#ifdef __APPLE__29
+.section __RODATA,__rodata30
+#else31
+.section .rodata32
+#endif33
+34
+.align 435
+36
+.text37
+38
+/* void blockcopy_sp(pixel* a, intptr_t stridea, const int16_t* b, intptr_t strideb)39
+ *40
+ * r0 - a41
+ * r1 - stridea42
+ * r2 - b43
+ * r3 - strideb */44
+function PFX(blockcopy_sp_4x4_neon)45
+ lsl x3, x3, #146
+.rept 247
+ ld1 {v0.8h}, x2, x348
+ ld1 {v1.8h}, x2, x349
+ xtn v0.8b, v0.8h50
+ xtn v1.8b, v1.8h51
+ st1 {v0.s}0, x0, x152
+ st1 {v1.s}0, x0, x153
+.endr54
+ ret55
+endfunc56
+57
+function PFX(blockcopy_sp_8x8_neon)58
+ lsl x3, x3, #159
+.rept 460
+ ld1 {v0.8h}, x2, x361
+ ld1 {v1.8h}, x2, x362
+ xtn v0.8b, v0.8h63
+ xtn v1.8b, v1.8h64
+ st1 {v0.d}0, x0, x165
+ st1 {v1.d}0, x0, x166
+.endr67
+ ret68
+endfunc69
+70
+function PFX(blockcopy_sp_16x16_neon)71
+ lsl x3, x3, #172
+ movrel x11, xtn_xtn2_table73
+ ld1 {v31.16b}, x1174
+.rept 875
+ ld1 {v0.8h-v1.8h}, x2, x376
+ ld1 {v2.8h-v3.8h}, x2, x377
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b78
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b79
+ st1 {v0.16b}, x0, x180
+ st1 {v1.16b}, x0, x181
+.endr82
+ ret83
+endfunc84
+85
+function PFX(blockcopy_sp_32x32_neon)86
+ mov w12, #487
+ lsl x3, x3, #188
+ movrel x11, xtn_xtn2_table89
+ ld1 {v31.16b}, x1190
+.loop_csp32:91
+ sub w12, w12, #192
+.rept 493
+ ld1 {v0.8h-v3.8h}, x2, x394
+ ld1 {v4.8h-v7.8h}, x2, x395
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b96
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b97
+ tbl v2.16b, {v4.16b,v5.16b}, v31.16b98
+ tbl v3.16b, {v6.16b,v7.16b}, v31.16b99
+ st1 {v0.16b-v1.16b}, x0, x1100
+ st1 {v2.16b-v3.16b}, x0, x1101
+.endr102
+ cbnz w12, .loop_csp32103
+ ret104
+endfunc105
+106
+function PFX(blockcopy_sp_64x64_neon)107
+ mov w12, #16108
+ lsl x3, x3, #1109
+ sub x3, x3, #64110
+ movrel x11, xtn_xtn2_table111
+ ld1 {v31.16b}, x11112
+.loop_csp64:113
+ sub w12, w12, #1114
+.rept 4115
+ ld1 {v0.8h-v3.8h}, x2, #64116
+ ld1 {v4.8h-v7.8h}, x2, x3117
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b118
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b119
+ tbl v2.16b, {v4.16b,v5.16b}, v31.16b120
+ tbl v3.16b, {v6.16b,v7.16b}, v31.16b121
+ st1 {v0.16b-v3.16b}, x0, x1122
+.endr123
+ cbnz w12, .loop_csp64124
+ ret125
+endfunc126
+127
+// void blockcopy_ps(int16_t* a, intptr_t stridea, const pixel* b, intptr_t strideb)128
+function PFX(blockcopy_ps_4x4_neon)129
+ lsl x1, x1, #1130
+.rept 2131
+ ld1 {v0.8b}, x2, x3132
+ ld1 {v1.8b}, x2, x3133
+ uxtl v0.8h, v0.8b134
+ uxtl v1.8h, v1.8b135
+ st1 {v0.4h}, x0, x1136
+ st1 {v1.4h}, x0, x1137
+.endr138
+ ret139
+endfunc140
+141
+function PFX(blockcopy_ps_8x8_neon)142
+ lsl x1, x1, #1143
+.rept 4144
+ ld1 {v0.8b}, x2, x3145
+ ld1 {v1.8b}, x2, x3146
+ uxtl v0.8h, v0.8b147
+ uxtl v1.8h, v1.8b148
+ st1 {v0.8h}, x0, x1149
+ st1 {v1.8h}, x0, x1150
+.endr151
+ ret152
+endfunc153
+154
+function PFX(blockcopy_ps_16x16_neon)155
+ lsl x1, x1, #1156
+.rept 8157
+ ld1 {v4.16b}, x2, x3158
+ ld1 {v5.16b}, x2, x3159
+ uxtl v0.8h, v4.8b160
+ uxtl2 v1.8h, v4.16b161
+ uxtl v2.8h, v5.8b162
+ uxtl2 v3.8h, v5.16b163
+ st1 {v0.8h-v1.8h}, x0, x1164
+ st1 {v2.8h-v3.8h}, x0, x1165
+.endr166
+ ret167
+endfunc168
+169
+function PFX(blockcopy_ps_32x32_neon)170
+ lsl x1, x1, #1171
+ mov w12, #4172
+.loop_cps32:173
+ sub w12, w12, #1174
+.rept 4175
+ ld1 {v16.16b-v17.16b}, x2, x3176
+ ld1 {v18.16b-v19.16b}, x2, x3177
+ uxtl v0.8h, v16.8b178
+ uxtl2 v1.8h, v16.16b179
+ uxtl v2.8h, v17.8b180
+ uxtl2 v3.8h, v17.16b181
+ uxtl v4.8h, v18.8b182
+ uxtl2 v5.8h, v18.16b183
+ uxtl v6.8h, v19.8b184
+ uxtl2 v7.8h, v19.16b185
+ st1 {v0.8h-v3.8h}, x0, x1186
+ st1 {v4.8h-v7.8h}, x0, x1187
+.endr188
+ cbnz w12, .loop_cps32189
+ ret190
+endfunc191
+192
+function PFX(blockcopy_ps_64x64_neon)193
+ lsl x1, x1, #1194
+ sub x1, x1, #64195
+ mov w12, #16196
+.loop_cps64:197
+ sub w12, w12, #1198
+.rept 4199
+ ld1 {v16.16b-v19.16b}, x2, x3200
+ uxtl v0.8h, v16.8b201
+ uxtl2 v1.8h, v16.16b202
+ uxtl v2.8h, v17.8b203
+ uxtl2 v3.8h, v17.16b204
+ uxtl v4.8h, v18.8b205
+ uxtl2 v5.8h, v18.16b206
+ uxtl v6.8h, v19.8b207
+ uxtl2 v7.8h, v19.16b208
+ st1 {v0.8h-v3.8h}, x0, #64209
+ st1 {v4.8h-v7.8h}, x0, x1210
+.endr211
+ cbnz w12, .loop_cps64212
+ ret213
+endfunc214
+215
+// void x265_blockcopy_ss(int16_t* a, intptr_t stridea, const int16_t* b, intptr_t strideb)216
+function PFX(blockcopy_ss_4x4_neon)217
+ lsl x1, x1, #1218
+ lsl x3, x3, #1219
+.rept 2220
+ ld1 {v0.8b}, x2, x3221
+ ld1 {v1.8b}, x2, x3222
+ st1 {v0.8b}, x0, x1223
+ st1 {v1.8b}, x0, x1224
+.endr225
+ ret226
+endfunc227
+228
+function PFX(blockcopy_ss_8x8_neon)229
+ lsl x1, x1, #1230
+ lsl x3, x3, #1231
+.rept 4232
+ ld1 {v0.8h}, x2, x3233
+ ld1 {v1.8h}, x2, x3234
+ st1 {v0.8h}, x0, x1235
+ st1 {v1.8h}, x0, x1236
+.endr237
+ ret238
+endfunc239
+240
+function PFX(blockcopy_ss_16x16_neon)241
+ lsl x1, x1, #1242
+ lsl x3, x3, #1243
+.rept 8244
+ ld1 {v0.8h-v1.8h}, x2, x3245
+ ld1 {v2.8h-v3.8h}, x2, x3246
+ st1 {v0.8h-v1.8h}, x0, x1247
+ st1 {v2.8h-v3.8h}, x0, x1248
+.endr249
+ ret250
+endfunc251
+252
+function PFX(blockcopy_ss_32x32_neon)253
+ lsl x1, x1, #1254
+ lsl x3, x3, #1255
+ mov w12, #4256
+.loop_css32:257
+ sub w12, w12, #1258
+.rept 8259
+ ld1 {v0.8h-v3.8h}, x2, x3260
+ st1 {v0.8h-v3.8h}, x0, x1261
+.endr262
+ cbnz w12, .loop_css32263
+ ret264
+endfunc265
+266
+function PFX(blockcopy_ss_64x64_neon)267
+ lsl x1, x1, #1268
+ sub x1, x1, #64269
+ lsl x3, x3, #1270
+ sub x3, x3, #64271
+ mov w12, #8272
+.loop_css64:273
+ sub w12, w12, #1274
+.rept 8275
+ ld1 {v0.8h-v3.8h}, x2, #64276
+ ld1 {v4.8h-v7.8h}, x2, x3277
+ st1 {v0.8h-v3.8h}, x0, #64278
+ st1 {v4.8h-v7.8h}, x0, x1279
+.endr280
+ cbnz w12, .loop_css64281
+ ret282
+endfunc283
+284
+/******** Chroma blockcopy********/285
+function PFX(blockcopy_ss_4x8_neon)286
+ lsl x1, x1, #1287
+ lsl x3, x3, #1288
+.rept 4289
+ ld1 {v0.8b}, x2, x3290
+ ld1 {v1.8b}, x2, x3291
+ st1 {v0.8b}, x0, x1292
+ st1 {v1.8b}, x0, x1293
+.endr294
+ ret295
+endfunc296
+297
+function PFX(blockcopy_ss_8x16_neon)298
+ lsl x1, x1, #1299
+ lsl x3, x3, #1300
+.rept 8301
+ ld1 {v0.8h}, x2, x3302
+ ld1 {v1.8h}, x2, x3303
+ st1 {v0.8h}, x0, x1304
+ st1 {v1.8h}, x0, x1305
+.endr306
+ ret307
+endfunc308
+309
+function PFX(blockcopy_ss_16x32_neon)310
+ lsl x1, x1, #1311
+ lsl x3, x3, #1312
+.rept 16313
+ ld1 {v0.8h-v1.8h}, x2, x3314
+ ld1 {v2.8h-v3.8h}, x2, x3315
+ st1 {v0.8h-v1.8h}, x0, x1316
+ st1 {v2.8h-v3.8h}, x0, x1317
+.endr318
+ ret319
+endfunc320
+321
+function PFX(blockcopy_ss_32x64_neon)322
+ lsl x1, x1, #1323
+ lsl x3, x3, #1324
+ mov w12, #8325
+.loop_css32x64:326
+ sub w12, w12, #1327
+.rept 8328
+ ld1 {v0.8h-v3.8h}, x2, x3329
+ st1 {v0.8h-v3.8h}, x0, x1330
+.endr331
+ cbnz w12, .loop_css32x64332
+ ret333
+endfunc334
+335
+// chroma blockcopy_ps336
+function PFX(blockcopy_ps_4x8_neon)337
+ lsl x1, x1, #1338
+.rept 4339
+ ld1 {v0.8b}, x2, x3340
+ ld1 {v1.8b}, x2, x3341
+ uxtl v0.8h, v0.8b342
+ uxtl v1.8h, v1.8b343
+ st1 {v0.4h}, x0, x1344
+ st1 {v1.4h}, x0, x1345
+.endr346
+ ret347
+endfunc348
+349
+function PFX(blockcopy_ps_8x16_neon)350
+ lsl x1, x1, #1351
+.rept 8352
+ ld1 {v0.8b}, x2, x3353
+ ld1 {v1.8b}, x2, x3354
+ uxtl v0.8h, v0.8b355
+ uxtl v1.8h, v1.8b356
+ st1 {v0.8h}, x0, x1357
+ st1 {v1.8h}, x0, x1358
+.endr359
+ ret360
+endfunc361
+362
+function PFX(blockcopy_ps_16x32_neon)363
+ lsl x1, x1, #1364
+.rept 16365
+ ld1 {v4.16b}, x2, x3366
+ ld1 {v5.16b}, x2, x3367
+ uxtl v0.8h, v4.8b368
+ uxtl2 v1.8h, v4.16b369
+ uxtl v2.8h, v5.8b370
+ uxtl2 v3.8h, v5.16b371
+ st1 {v0.8h-v1.8h}, x0, x1372
+ st1 {v2.8h-v3.8h}, x0, x1373
+.endr374
+ ret375
+endfunc376
+377
+function PFX(blockcopy_ps_32x64_neon)378
+ lsl x1, x1, #1379
+ mov w12, #8380
+.loop_cps32x64:381
+ sub w12, w12, #1382
+.rept 4383
+ ld1 {v16.16b-v17.16b}, x2, x3384
+ ld1 {v18.16b-v19.16b}, x2, x3385
+ uxtl v0.8h, v16.8b386
+ uxtl2 v1.8h, v16.16b387
+ uxtl v2.8h, v17.8b388
+ uxtl2 v3.8h, v17.16b389
+ uxtl v4.8h, v18.8b390
+ uxtl2 v5.8h, v18.16b391
+ uxtl v6.8h, v19.8b392
+ uxtl2 v7.8h, v19.16b393
+ st1 {v0.8h-v3.8h}, x0, x1394
+ st1 {v4.8h-v7.8h}, x0, x1395
+.endr396
+ cbnz w12, .loop_cps32x64397
+ ret398
+endfunc399
+400
+// chroma blockcopy_sp401
+function PFX(blockcopy_sp_4x8_neon)402
+ lsl x3, x3, #1403
+.rept 4404
+ ld1 {v0.8h}, x2, x3405
+ ld1 {v1.8h}, x2, x3406
+ xtn v0.8b, v0.8h407
+ xtn v1.8b, v1.8h408
+ st1 {v0.s}0, x0, x1409
+ st1 {v1.s}0, x0, x1410
+.endr411
+ ret412
+endfunc413
+414
+function PFX(blockcopy_sp_8x16_neon)415
+ lsl x3, x3, #1416
+.rept 8417
+ ld1 {v0.8h}, x2, x3418
+ ld1 {v1.8h}, x2, x3419
+ xtn v0.8b, v0.8h420
+ xtn v1.8b, v1.8h421
+ st1 {v0.d}0, x0, x1422
+ st1 {v1.d}0, x0, x1423
+.endr424
+ ret425
+endfunc426
+427
+function PFX(blockcopy_sp_16x32_neon)428
+ lsl x3, x3, #1429
+ movrel x11, xtn_xtn2_table430
+ ld1 {v31.16b}, x11431
+.rept 16432
+ ld1 {v0.8h-v1.8h}, x2, x3433
+ ld1 {v2.8h-v3.8h}, x2, x3434
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b435
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b436
+ st1 {v0.16b}, x0, x1437
+ st1 {v1.16b}, x0, x1438
+.endr439
+ ret440
+endfunc441
+442
+function PFX(blockcopy_sp_32x64_neon)443
+ mov w12, #8444
+ lsl x3, x3, #1445
+ movrel x11, xtn_xtn2_table446
+ ld1 {v31.16b}, x11447
+.loop_csp32x64:448
+ sub w12, w12, #1449
+.rept 4450
+ ld1 {v0.8h-v3.8h}, x2, x3451
+ ld1 {v4.8h-v7.8h}, x2, x3452
+ tbl v0.16b, {v0.16b,v1.16b}, v31.16b453
+ tbl v1.16b, {v2.16b,v3.16b}, v31.16b454
+ tbl v2.16b, {v4.16b,v5.16b}, v31.16b455
+ tbl v3.16b, {v6.16b,v7.16b}, v31.16b456
+ st1 {v0.16b-v1.16b}, x0, x1457
+ st1 {v2.16b-v3.16b}, x0, x1458
+.endr459
+ cbnz w12, .loop_csp32x64460
+ ret461
+endfunc462
+463
+/* blockcopy_pp(pixel* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride) */464
+465
+function PFX(blockcopy_pp_2x4_neon)466
+ ldrh w9, x2467
+ add x4, x1, x1468
+ add x14, x3, x3469
+ strh w9, x0470
+ ldrh w10, x2, x3471
+ add x5, x4, x1472
+ add x15, x14, x3473
+ strh w10, x0, x1474
+ ldrh w11, x2, x14475
+ strh w11, x0, x4476
+ ldrh w12, x2, x15477
+ strh w12, x0, x5478
+ ret479
+endfunc480
+481
+.macro blockcopy_pp_2xN_neon h482
+function PFX(blockcopy_pp_2x\h\()_neon)483
+ add x4, x1, x1484
+ add x5, x4, x1485
+ add x6, x5, x1486
+487
+ add x14, x3, x3488
+ add x15, x14, x3489
+ add x16, x15, x3490
+491
+.rept \h / 4492
+ ldrh w9, x2493
+ strh w9, x0494
+ ldrh w10, x2, x3495
+ strh w10, x0, x1496
+ ldrh w11, x2, x14497
+ strh w11, x0, x4498
+ ldrh w12, x2, x15499
+ strh w12, x0, x5500
+ add x2, x2, x16501
+ add x0, x0, x6502
+.endr503
+ ret504
+endfunc505
+.endm506
+507
+blockcopy_pp_2xN_neon 8508
+blockcopy_pp_2xN_neon 16509
+510
+function PFX(blockcopy_pp_4x2_neon)511
+ ldr w9, x2512
+ str w9, x0513
+ ldr w10, x2, x3514
+ str w10, x0, x1515
+ ret516
+endfunc517
+518
+function PFX(blockcopy_pp_4x4_neon)519
+ ldr w9, x2520
+ add x4, x1, x1521
+ add x14, x3, x3522
+ str w9, x0523
+ ldr w10, x2, x3524
+ add x5, x4, x1525
+ add x15, x14, x3526
+ str w10, x0, x1527
+ ldr w11, x2, x14528
+ str w11, x0, x4529
+ ldr w12, x2, x15530
+ str w12, x0, x5531
+ ret532
+endfunc533
+534
+.macro blockcopy_pp_4xN_neon h535
+function PFX(blockcopy_pp_4x\h\()_neon)536
+ add x4, x1, x1537
+ add x5, x4, x1538
+ add x6, x5, x1539
+540
+ add x14, x3, x3541
+ add x15, x14, x3542
+ add x16, x15, x3543
+544
+.rept \h / 4545
+ ldr w9, x2546
+ str w9, x0547
+ ldr w10, x2, x3548
+ str w10, x0, x1549
+ ldr w11, x2, x14550
+ str w11, x0, x4551
+ ldr w12, x2, x15552
+ str w12, x0, x5553
+ add x2, x2, x16554
+ add x0, x0, x6555
+.endr556
+ ret557
+endfunc558
+.endm559
+560
+blockcopy_pp_4xN_neon 8561
+blockcopy_pp_4xN_neon 16562
+blockcopy_pp_4xN_neon 32563
+564
+.macro blockcopy_pp_6xN_neon h565
+function PFX(blockcopy_pp_6x\h\()_neon)566
+ sub x1, x1, #4567
+.rept \h568
+ ld1 {v0.8b}, x2, x3569
+ st1 {v0.s}0, x0, #4570
+ st1 {v0.h}2, x0, x1571
+.endr572
+ ret573
+endfunc574
+.endm575
+576
+blockcopy_pp_6xN_neon 8577
+blockcopy_pp_6xN_neon 16578
+579
+.macro blockcopy_pp_8xN_neon h580
+function PFX(blockcopy_pp_8x\h\()_neon)581
+.rept \h582
+ ld1 {v0.4h}, x2, x3583
+ st1 {v0.4h}, x0, x1584
+.endr585
+ ret586
+endfunc587
+.endm588
+589
+blockcopy_pp_8xN_neon 2590
+blockcopy_pp_8xN_neon 4591
+blockcopy_pp_8xN_neon 6592
+blockcopy_pp_8xN_neon 8593
+blockcopy_pp_8xN_neon 12594
+blockcopy_pp_8xN_neon 16595
+blockcopy_pp_8xN_neon 32596
+597
+function PFX(blockcopy_pp_8x64_neon)598
+ mov w12, #4599
+.loop_pp_8x64:600
+ sub w12, w12, #1601
+.rept 16602
+ ld1 {v0.4h}, x2, x3603
+ st1 {v0.4h}, x0, x1604
+.endr605
+ cbnz w12, .loop_pp_8x64606
+ ret607
+endfunc608
+609
+.macro blockcopy_pp_16xN_neon h610
+function PFX(blockcopy_pp_16x\h\()_neon)611
+.rept \h612
+ ld1 {v0.8h}, x2, x3613
+ st1 {v0.8h}, x0, x1614
+.endr615
+ ret616
+endfunc617
+.endm618
+619
+blockcopy_pp_16xN_neon 4620
+blockcopy_pp_16xN_neon 8621
+blockcopy_pp_16xN_neon 12622
+blockcopy_pp_16xN_neon 16623
+624
+.macro blockcopy_pp_16xN1_neon h625
+function PFX(blockcopy_pp_16x\h\()_neon)626
+ mov w12, #\h / 8627
+.loop_16x\h\():628
+.rept 8629
+ ld1 {v0.8h}, x2, x3630
+ st1 {v0.8h}, x0, x1631
+.endr632
+ sub w12, w12, #1633
+ cbnz w12, .loop_16x\h634
+ ret635
+endfunc636
+.endm637
+638
+blockcopy_pp_16xN1_neon 24639
+blockcopy_pp_16xN1_neon 32640
+blockcopy_pp_16xN1_neon 64641
+642
+function PFX(blockcopy_pp_12x16_neon)643
+ sub x1, x1, #8644
+.rept 16645
+ ld1 {v0.16b}, x2, x3646
+ str d0, x0, #8647
+ st1 {v0.s}2, x0, x1648
+.endr649
+ ret650
+endfunc651
+652
+function PFX(blockcopy_pp_12x32_neon)653
+ sub x1, x1, #8654
+ mov w12, #4655
+.loop_pp_12x32:656
+ sub w12, w12, #1657
+.rept 8658
+ ld1 {v0.16b}, x2, x3659
+ str d0, x0, #8660
+ st1 {v0.s}2, x0, x1661
+.endr662
+ cbnz w12, .loop_pp_12x32663
+ ret664
+endfunc665
+666
+function PFX(blockcopy_pp_24x32_neon)667
+ mov w12, #4668
+.loop_24x32:669
+ sub w12, w12, #1670
+.rept 8671
+ ld1 {v0.8b-v2.8b}, x2, x3672
+ st1 {v0.8b-v2.8b}, x0, x1673
+.endr674
+ cbnz w12, .loop_24x32675
+ ret676
+endfunc677
+678
+function PFX(blockcopy_pp_24x64_neon)679
+ mov w12, #4680
+.loop_24x64:681
+ sub w12, w12, #1682
+.rept 16683
+ ld1 {v0.8b-v2.8b}, x2, x3684
+ st1 {v0.8b-v2.8b}, x0, x1685
+.endr686
+ cbnz w12, .loop_24x64687
+ ret688
+endfunc689
+690
+function PFX(blockcopy_pp_32x8_neon)691
+.rept 8692
+ ld1 {v0.16b-v1.16b}, x2, x3693
+ st1 {v0.16b-v1.16b}, x0, x1694
+.endr695
+ ret696
+endfunc697
+698
+.macro blockcopy_pp_32xN_neon h699
+function PFX(blockcopy_pp_32x\h\()_neon)700
+ mov w12, #\h / 8701
+.loop_32x\h\():702
+ sub w12, w12, #1703
+.rept 8704
+ ld1 {v0.16b-v1.16b}, x2, x3705
+ st1 {v0.16b-v1.16b}, x0, x1706
+.endr707
+ cbnz w12, .loop_32x\h708
+ ret709
+endfunc710
+.endm711
+712
+blockcopy_pp_32xN_neon 16713
+blockcopy_pp_32xN_neon 24714
+blockcopy_pp_32xN_neon 32715
+blockcopy_pp_32xN_neon 64716
+blockcopy_pp_32xN_neon 48717
+718
+function PFX(blockcopy_pp_48x64_neon)719
+ mov w12, #8720
+.loop_48x64:721
+ sub w12, w12, #1722
+.rept 8723
+ ld1 {v0.16b-v2.16b}, x2, x3724
+ st1 {v0.16b-v2.16b}, x0, x1725
+.endr726
+ cbnz w12, .loop_48x64727
+ ret728
+endfunc729
+730
+.macro blockcopy_pp_64xN_neon h731
+function PFX(blockcopy_pp_64x\h\()_neon)732
+ mov w12, #\h / 4733
+.loop_64x\h\():734
+ sub w12, w12, #1735
+.rept 4736
+ ld1 {v0.16b-v3.16b}, x2, x3737
+ st1 {v0.16b-v3.16b}, x0, x1738
+.endr739
+ cbnz w12, .loop_64x\h740
+ ret741
+endfunc742
+.endm743
+744
+blockcopy_pp_64xN_neon 16745
+blockcopy_pp_64xN_neon 32746
+blockcopy_pp_64xN_neon 48747
+blockcopy_pp_64xN_neon 64748
+749
+// void x265_blockfill_s_neon(int16_t* dst, intptr_t dstride, int16_t val)750
+function PFX(blockfill_s_4x4_neon)751
+ dup v0.4h, w2752
+ lsl x1, x1, #1753
+.rept 4754
+ st1 {v0.4h}, x0, x1755
+.endr756
+ ret757
+endfunc758
+759
+function PFX(blockfill_s_8x8_neon)760
+ dup v0.8h, w2761
+ lsl x1, x1, #1762
+.rept 8763
+ st1 {v0.8h}, x0, x1764
+.endr765
+ ret766
+endfunc767
+768
+function PFX(blockfill_s_16x16_neon)769
+ dup v0.8h, w2770
+ mov v1.16b, v0.16b771
+ lsl x1, x1, #1772
+.rept 16773
+ stp q0, q1, x0774
+ add x0, x0, x1775
+.endr776
+ ret777
+endfunc778
+779
+function PFX(blockfill_s_32x32_neon)780
+ dup v0.8h, w2781
+ mov v1.16b, v0.16b782
+ mov v2.16b, v0.16b783
+ mov v3.16b, v0.16b784
+ lsl x1, x1, #1785
+.rept 32786
+ st1 {v0.8h-v3.8h}, x0, x1787
+.endr788
+ ret789
+endfunc790
+791
+function PFX(blockfill_s_64x64_neon)792
+ dup v0.8h, w2793
+ mov v1.16b, v0.16b794
+ mov v2.16b, v0.16b795
+ mov v3.16b, v0.16b796
+ lsl x1, x1, #1797
+ sub x1, x1, #64798
+.rept 64799
+ st1 {v0.8h-v3.8h}, x0, #64800
+ st1 {v0.8h-v3.8h}, x0, x1801
+.endr802
+ ret803
+endfunc804
+805
+// uint32_t copy_count(int16_t* coeff, const int16_t* residual, intptr_t resiStride)806
+function PFX(copy_cnt_4_neon)807
+ lsl x2, x2, #1808
+ movi v4.8b, #0809
+.rept 2810
+ ld1 {v0.8b}, x1, x2811
+ ld1 {v1.8b}, x1, x2812
+ stp d0, d1, x0, #16813
+ cmeq v0.4h, v0.4h, #0814
+ cmeq v1.4h, v1.4h, #0815
+ add v4.4h, v4.4h, v0.4h816
+ add v4.4h, v4.4h, v1.4h817
+.endr818
+ saddlv s4, v4.4h819
+ fmov w12, s4820
+ add w0, w12, #16821
+ ret822
+endfunc823
+824
+function PFX(copy_cnt_8_neon)825
+ lsl x2, x2, #1826
+ movi v4.8b, #0827
+.rept 4828
+ ld1 {v0.16b}, x1, x2829
+ ld1 {v1.16b}, x1, x2830
+ stp q0, q1, x0, #32831
+ cmeq v0.8h, v0.8h, #0832
+ cmeq v1.8h, v1.8h, #0833
+ add v4.8h, v4.8h, v0.8h834
+ add v4.8h, v4.8h, v1.8h835
+.endr836
+ saddlv s4, v4.8h837
+ fmov w12, s4838
+ add w0, w12, #64839
+ ret840
+endfunc841
+842
+function PFX(copy_cnt_16_neon)843
+ lsl x2, x2, #1844
+ movi v4.8b, #0845
+.rept 16846
+ ld1 {v0.16b-v1.16b}, x1, x2847
+ st1 {v0.16b-v1.16b}, x0, #32848
+ cmeq v0.8h, v0.8h, #0849
+ cmeq v1.8h, v1.8h, #0850
+ add v4.8h, v4.8h, v0.8h851
+ add v4.8h, v4.8h, v1.8h852
+.endr853
+ saddlv s4, v4.8h854
+ fmov w12, s4855
+ add w0, w12, #256856
+ ret857
+endfunc858
+859
+function PFX(copy_cnt_32_neon)860
+ lsl x2, x2, #1861
+ movi v4.8b, #0862
+.rept 32863
+ ld1 {v0.16b-v3.16b}, x1, x2864
+ st1 {v0.16b-v3.16b}, x0, #64865
+ cmeq v0.8h, v0.8h, #0866
+ cmeq v1.8h, v1.8h, #0867
+ cmeq v2.8h, v2.8h, #0868
+ cmeq v3.8h, v3.8h, #0869
+ add v0.8h, v0.8h, v1.8h870
+ add v2.8h, v2.8h, v3.8h871
+ add v4.8h, v4.8h, v0.8h872
+ add v4.8h, v4.8h, v2.8h873
+.endr874
+ saddlv s4, v4.8h875
+ fmov w12, s4876
+ add w0, w12, #1024877
+ ret878
+endfunc879
+880
+// int count_nonzero_c(const int16_t* quantCoeff)881
+function PFX(count_nonzero_4_neon)882
+ movi v16.16b, #1883
+ movi v17.16b, #0884
+ trn1 v16.16b, v16.16b, v17.16b885
+ ldp q0, q1, x0886
+ cmhi v0.8h, v0.8h, v17.8h887
+ cmhi v1.8h, v1.8h, v17.8h888
+ and v0.16b, v0.16b, v16.16b889
+ and v1.16b, v1.16b, v16.16b890
+ add v0.8h, v0.8h, v1.8h891
+ uaddlv s0, v0.8h892
+ fmov w0, s0893
+ ret894
+endfunc895
+896
+.macro COUNT_NONZERO_8897
+ ld1 {v0.16b-v3.16b}, x0, #64898
+ ld1 {v4.16b-v7.16b}, x0, #64899
+ cmhi v0.8h, v0.8h, v17.8h900
+ cmhi v1.8h, v1.8h, v17.8h901
+ cmhi v2.8h, v2.8h, v17.8h902
+ cmhi v3.8h, v3.8h, v17.8h903
+ cmhi v4.8h, v4.8h, v17.8h904
+ cmhi v5.8h, v5.8h, v17.8h905
+ cmhi v6.8h, v6.8h, v17.8h906
+ cmhi v7.8h, v7.8h, v17.8h907
+ and v0.16b, v0.16b, v16.16b908
+ and v1.16b, v1.16b, v16.16b909
+ and v2.16b, v2.16b, v16.16b910
+ and v3.16b, v3.16b, v16.16b911
+ and v4.16b, v4.16b, v16.16b912
+ and v5.16b, v5.16b, v16.16b913
+ and v6.16b, v6.16b, v16.16b914
+ and v7.16b, v7.16b, v16.16b915
+ add v0.8h, v0.8h, v1.8h916
+ add v2.8h, v2.8h, v3.8h917
+ add v4.8h, v4.8h, v5.8h918
+ add v6.8h, v6.8h, v7.8h919
+ add v0.8h, v0.8h, v2.8h920
+ add v4.8h, v4.8h, v6.8h921
+ add v0.8h, v0.8h, v4.8h922
+.endm923
+924
+function PFX(count_nonzero_8_neon)925
+ movi v16.16b, #1926
+ movi v17.16b, #0927
+ trn1 v16.16b, v16.16b, v17.16b928
+ COUNT_NONZERO_8929
+ uaddlv s0, v0.8h930
+ fmov w0, s0931
+ ret932
+endfunc933
+934
+function PFX(count_nonzero_16_neon)935
+ movi v16.16b, #1936
+ movi v17.16b, #0937
+ trn1 v16.16b, v16.16b, v17.16b938
+ movi v18.16b, #0939
+.rept 4940
+ COUNT_NONZERO_8941
+ add v18.16b, v18.16b, v0.16b942
+.endr943
+ uaddlv s0, v18.8h944
+ fmov w0, s0945
+ ret946
+endfunc947
+948
+function PFX(count_nonzero_32_neon)949
+ movi v16.16b, #1950
+ movi v17.16b, #0951
+ trn1 v16.16b, v16.16b, v17.16b952
+ movi v18.16b, #0953
+ mov w12, #16954
+.loop_count_nonzero_32:955
+ sub w12, w12, #1956
+ COUNT_NONZERO_8957
+ add v18.16b, v18.16b, v0.16b958
+ cbnz w12, .loop_count_nonzero_32959
+960
+ uaddlv s0, v18.8h961
+ fmov w0, s0962
+ ret963
+endfunc964
+965
+// void cpy2Dto1D_shl(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift)966
+.macro cpy2Dto1D_shl_start967
+ add x2, x2, x2968
+ dup v0.8h, w3969
+.endm970
+971
+function PFX(cpy2Dto1D_shl_4x4_neon)972
+ cpy2Dto1D_shl_start973
+ ld1 {v2.d}0, x1, x2974
+ ld1 {v2.d}1, x1, x2975
+ ld1 {v3.d}0, x1, x2976
+ ld1 {v3.d}1, x1, x2977
+ sshl v2.8h, v2.8h, v0.8h978
+ sshl v3.8h, v3.8h, v0.8h979
+ st1 {v2.16b-v3.16b}, x0980
+ ret981
+endfunc982
+983
+function PFX(cpy2Dto1D_shl_8x8_neon)984
+ cpy2Dto1D_shl_start985
+.rept 4986
+ ld1 {v2.16b}, x1, x2987
+ ld1 {v3.16b}, x1, x2988
+ sshl v2.8h, v2.8h, v0.8h989
+ sshl v3.8h, v3.8h, v0.8h990
+ st1 {v2.16b-v3.16b}, x0, #32991
+.endr992
+ ret993
+endfunc994
+995
+function PFX(cpy2Dto1D_shl_16x16_neon)996
+ cpy2Dto1D_shl_start997
+ mov w12, #4998
+.loop_cpy2Dto1D_shl_16:999
+ sub w12, w12, #11000
+.rept 41001
+ ld1 {v2.16b-v3.16b}, x1, x21002
+ sshl v2.8h, v2.8h, v0.8h1003
+ sshl v3.8h, v3.8h, v0.8h1004
+ st1 {v2.16b-v3.16b}, x0, #321005
+.endr1006
+ cbnz w12, .loop_cpy2Dto1D_shl_161007
+ ret1008
+endfunc1009
+1010
+function PFX(cpy2Dto1D_shl_32x32_neon)1011
+ cpy2Dto1D_shl_start1012
+ mov w12, #161013
+.loop_cpy2Dto1D_shl_32:1014
+ sub w12, w12, #11015
+.rept 21016
+ ld1 {v2.16b-v5.16b}, x1, x21017
+ sshl v2.8h, v2.8h, v0.8h1018
+ sshl v3.8h, v3.8h, v0.8h1019
+ sshl v4.8h, v4.8h, v0.8h1020
+ sshl v5.8h, v5.8h, v0.8h1021
+ st1 {v2.16b-v5.16b}, x0, #641022
+.endr1023
+ cbnz w12, .loop_cpy2Dto1D_shl_321024
+ ret1025
+endfunc1026
+1027
+function PFX(cpy2Dto1D_shl_64x64_neon)1028
+ cpy2Dto1D_shl_start1029
+ mov w12, #321030
+ sub x2, x2, #641031
+.loop_cpy2Dto1D_shl_64:1032
+ sub w12, w12, #11033
+.rept 21034
+ ld1 {v2.16b-v5.16b}, x1, #641035
+ ld1 {v16.16b-v19.16b}, x1, x21036
+ sshl v2.8h, v2.8h, v0.8h1037
+ sshl v3.8h, v3.8h, v0.8h1038
+ sshl v4.8h, v4.8h, v0.8h1039
+ sshl v5.8h, v5.8h, v0.8h1040
+ sshl v16.8h, v16.8h, v0.8h1041
+ sshl v17.8h, v17.8h, v0.8h1042
+ sshl v18.8h, v18.8h, v0.8h1043
+ sshl v19.8h, v19.8h, v0.8h1044
+ st1 {v2.16b-v5.16b}, x0, #641045
+ st1 {v16.16b-v19.16b}, x0, #641046
+.endr1047
+ cbnz w12, .loop_cpy2Dto1D_shl_641048
+ ret1049
+endfunc1050
+1051
+// void cpy2Dto1D_shr(int16_t* dst, const int16_t* src, intptr_t srcStride, int shift)1052
+function PFX(cpy2Dto1D_shr_4x4_neon)1053
+ cpy2Dto1D_shr_start1054
+ ld1 {v2.d}0, x1, x21055
+ ld1 {v2.d}1, x1, x21056
+ ld1 {v3.d}0, x1, x21057
+ ld1 {v3.d}1, x1, x21058
+ sub v2.8h, v2.8h, v1.8h1059
+ sub v3.8h, v3.8h, v1.8h1060
+ sshl v2.8h, v2.8h, v0.8h1061
+ sshl v3.8h, v3.8h, v0.8h1062
+ stp q2, q3, x01063
+ ret1064
+endfunc1065
+1066
+function PFX(cpy2Dto1D_shr_8x8_neon)1067
+ cpy2Dto1D_shr_start1068
+.rept 41069
+ ld1 {v2.16b}, x1, x21070
+ ld1 {v3.16b}, x1, x21071
+ sub v2.8h, v2.8h, v1.8h1072
+ sub v3.8h, v3.8h, v1.8h1073
+ sshl v2.8h, v2.8h, v0.8h1074
+ sshl v3.8h, v3.8h, v0.8h1075
+ stp q2, q3, x0, #321076
+.endr1077
+ ret1078
+endfunc1079
+1080
+function PFX(cpy2Dto1D_shr_16x16_neon)1081
+ cpy2Dto1D_shr_start1082
+ mov w12, #41083
+.loop_cpy2Dto1D_shr_16:1084
+ sub w12, w12, #11085
+.rept 41086
+ ld1 {v2.8h-v3.8h}, x1, x21087
+ sub v2.8h, v2.8h, v1.8h1088
+ sub v3.8h, v3.8h, v1.8h1089
+ sshl v2.8h, v2.8h, v0.8h1090
+ sshl v3.8h, v3.8h, v0.8h1091
+ st1 {v2.8h-v3.8h}, x0, #321092
+.endr1093
+ cbnz w12, .loop_cpy2Dto1D_shr_161094
+ ret1095
+endfunc1096
+1097
+function PFX(cpy2Dto1D_shr_32x32_neon)1098
+ cpy2Dto1D_shr_start1099
+ mov w12, #161100
+.loop_cpy2Dto1D_shr_32:1101
+ sub w12, w12, #11102
+.rept 21103
+ ld1 {v2.8h-v5.8h}, x1, x21104
+ sub v2.8h, v2.8h, v1.8h1105
+ sub v3.8h, v3.8h, v1.8h1106
+ sub v4.8h, v4.8h, v1.8h1107
+ sub v5.8h, v5.8h, v1.8h1108
+ sshl v2.8h, v2.8h, v0.8h1109
+ sshl v3.8h, v3.8h, v0.8h1110
+ sshl v4.8h, v4.8h, v0.8h1111
+ sshl v5.8h, v5.8h, v0.8h1112
+ st1 {v2.8h-v5.8h}, x0, #641113
+.endr1114
+ cbnz w12, .loop_cpy2Dto1D_shr_321115
+ ret1116
+endfunc1117
+1118
+// void cpy1Dto2D_shl(int16_t* dst, const int16_t* src, intptr_t dstStride, int shift)1119
+.macro cpy1Dto2D_shl_start1120
+ add x2, x2, x21121
+ dup v0.8h, w31122
+.endm1123
+1124
+function PFX(cpy1Dto2D_shl_4x4_neon)1125
+ cpy1Dto2D_shl_start1126
+ ld1 {v2.16b-v3.16b}, x11127
+ sshl v2.8h, v2.8h, v0.8h1128
+ sshl v3.8h, v3.8h, v0.8h1129
+ st1 {v2.d}0, x0, x21130
+ st1 {v2.d}1, x0, x21131
+ st1 {v3.d}0, x0, x21132
+ st1 {v3.d}1, x0, x21133
+ ret1134
+endfunc1135
+1136
+function PFX(cpy1Dto2D_shl_8x8_neon)1137
+ cpy1Dto2D_shl_start1138
+.rept 41139
+ ld1 {v2.16b-v3.16b}, x1, #321140
+ sshl v2.8h, v2.8h, v0.8h1141
+ sshl v3.8h, v3.8h, v0.8h1142
+ st1 {v2.16b}, x0, x21143
+ st1 {v3.16b}, x0, x21144
+.endr1145
+ ret1146
+endfunc1147
+1148
+function PFX(cpy1Dto2D_shl_16x16_neon)1149
+ cpy1Dto2D_shl_start1150
+ mov w12, #41151
+.loop_cpy1Dto2D_shl_16:1152
+ sub w12, w12, #11153
+.rept 41154
+ ld1 {v2.16b-v3.16b}, x1, #321155
+ sshl v2.8h, v2.8h, v0.8h1156
+ sshl v3.8h, v3.8h, v0.8h1157
+ st1 {v2.16b-v3.16b}, x0, x21158
+.endr1159
+ cbnz w12, .loop_cpy1Dto2D_shl_161160
+ ret1161
+endfunc1162
+1163
+function PFX(cpy1Dto2D_shl_32x32_neon)1164
+ cpy1Dto2D_shl_start1165
+ mov w12, #161166
+.loop_cpy1Dto2D_shl_32:1167
+ sub w12, w12, #11168
+.rept 21169
+ ld1 {v2.16b-v5.16b}, x1, #641170
+ sshl v2.8h, v2.8h, v0.8h1171
+ sshl v3.8h, v3.8h, v0.8h1172
+ sshl v4.8h, v4.8h, v0.8h1173
+ sshl v5.8h, v5.8h, v0.8h1174
+ st1 {v2.16b-v5.16b}, x0, x21175
+.endr1176
+ cbnz w12, .loop_cpy1Dto2D_shl_321177
+ ret1178
+endfunc1179
+1180
+function PFX(cpy1Dto2D_shl_64x64_neon)1181
+ cpy1Dto2D_shl_start1182
+ mov w12, #321183
+ sub x2, x2, #641184
+.loop_cpy1Dto2D_shl_64:1185
+ sub w12, w12, #11186
+.rept 21187
+ ld1 {v2.16b-v5.16b}, x1, #641188
+ ld1 {v16.16b-v19.16b}, x1, #641189
+ sshl v2.8h, v2.8h, v0.8h1190
+ sshl v3.8h, v3.8h, v0.8h1191
+ sshl v4.8h, v4.8h, v0.8h1192
+ sshl v5.8h, v5.8h, v0.8h1193
+ sshl v16.8h, v16.8h, v0.8h1194
+ sshl v17.8h, v17.8h, v0.8h1195
+ sshl v18.8h, v18.8h, v0.8h1196
+ sshl v19.8h, v19.8h, v0.8h1197
+ st1 {v2.16b-v5.16b}, x0, #641198
+ st1 {v16.16b-v19.16b}, x0, x21199
+.endr1200
+ cbnz w12, .loop_cpy1Dto2D_shl_641201
+ ret1202
+endfunc1203
+1204
+function PFX(cpy1Dto2D_shr_4x4_neon)1205
+ cpy1Dto2D_shr_start1206
+ ld1 {v2.16b-v3.16b}, x11207
+ sub v2.8h, v2.8h, v1.8h1208
+ sub v3.8h, v3.8h, v1.8h1209
+ sshl v2.8h, v2.8h, v0.8h1210
+ sshl v3.8h, v3.8h, v0.8h1211
+ st1 {v2.d}0, x0, x21212
+ st1 {v2.d}1, x0, x21213
+ st1 {v3.d}0, x0, x21214
+ st1 {v3.d}1, x0, x21215
+ ret1216
+endfunc1217
+1218
+function PFX(cpy1Dto2D_shr_8x8_neon)1219
+ cpy1Dto2D_shr_start1220
+.rept 41221
+ ld1 {v2.16b-v3.16b}, x1, #321222
+ sub v2.8h, v2.8h, v1.8h1223
+ sub v3.8h, v3.8h, v1.8h1224
+ sshl v2.8h, v2.8h, v0.8h1225
+ sshl v3.8h, v3.8h, v0.8h1226
+ st1 {v2.16b}, x0, x21227
+ st1 {v3.16b}, x0, x21228
+.endr1229
+ ret1230
+endfunc1231
+1232
+function PFX(cpy1Dto2D_shr_16x16_neon)1233
+ cpy1Dto2D_shr_start1234
+ mov w12, #41235
+.loop_cpy1Dto2D_shr_16:1236
+ sub w12, w12, #11237
+.rept 41238
+ ld1 {v2.8h-v3.8h}, x1, #321239
+ sub v2.8h, v2.8h, v1.8h1240
+ sub v3.8h, v3.8h, v1.8h1241
+ sshl v2.8h, v2.8h, v0.8h1242
+ sshl v3.8h, v3.8h, v0.8h1243
+ st1 {v2.8h-v3.8h}, x0, x21244
+.endr1245
+ cbnz w12, .loop_cpy1Dto2D_shr_161246
+ ret1247
+endfunc1248
+1249
+function PFX(cpy1Dto2D_shr_32x32_neon)1250
+ cpy1Dto2D_shr_start1251
+ mov w12, #161252
+.loop_cpy1Dto2D_shr_32:1253
+ sub w12, w12, #11254
+.rept 21255
+ ld1 {v2.16b-v5.16b}, x1, #641256
+ sub v2.8h, v2.8h, v1.8h1257
+ sub v3.8h, v3.8h, v1.8h1258
+ sub v4.8h, v4.8h, v1.8h1259
+ sub v5.8h, v5.8h, v1.8h1260
+ sshl v2.8h, v2.8h, v0.8h1261
+ sshl v3.8h, v3.8h, v0.8h1262
+ sshl v4.8h, v4.8h, v0.8h1263
+ sshl v5.8h, v5.8h, v0.8h1264
+ st1 {v2.16b-v5.16b}, x0, x21265
+.endr1266
+ cbnz w12, .loop_cpy1Dto2D_shr_321267
+ ret1268
+endfunc1269
+1270
+function PFX(cpy1Dto2D_shr_64x64_neon)1271
+ cpy1Dto2D_shr_start1272
+ mov w12, #321273
+ sub x2, x2, #641274
+.loop_cpy1Dto2D_shr_64:1275
+ sub w12, w12, #11276
+.rept 21277
+ ld1 {v2.16b-v5.16b}, x1, #641278
+ ld1 {v16.16b-v19.16b}, x1, #641279
+ sub v2.8h, v2.8h, v1.8h1280
+ sub v3.8h, v3.8h, v1.8h1281
+ sub v4.8h, v4.8h, v1.8h1282
+ sub v5.8h, v5.8h, v1.8h1283
+ sub v16.8h, v16.8h, v1.8h1284
+ sub v17.8h, v17.8h, v1.8h1285
+ sub v18.8h, v18.8h, v1.8h1286
+ sub v19.8h, v19.8h, v1.8h1287
+ sshl v2.8h, v2.8h, v0.8h1288
+ sshl v3.8h, v3.8h, v0.8h1289
+ sshl v4.8h, v4.8h, v0.8h1290
+ sshl v5.8h, v5.8h, v0.8h1291
+ sshl v16.8h, v16.8h, v0.8h1292
+ sshl v17.8h, v17.8h, v0.8h1293
+ sshl v18.8h, v18.8h, v0.8h1294
+ sshl v19.8h, v19.8h, v0.8h1295
+ st1 {v2.16b-v5.16b}, x0, #641296
+ st1 {v16.16b-v19.16b}, x0, x21297
+.endr1298
+ cbnz w12, .loop_cpy1Dto2D_shr_641299
+ ret1300
+endfunc1301
x265_3.6.tar.gz/source/common/aarch64/dct-prim.cpp
Added
950
1
2
+#include "dct-prim.h"3
+4
+5
+#if HAVE_NEON6
+7
+#include <arm_neon.h>8
+9
+10
+namespace11
+{12
+using namespace X265_NS;13
+14
+15
+static int16x8_t rev16(const int16x8_t a)16
+{17
+ static const int8x16_t tbl = {14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1};18
+ return vqtbx1q_u8(a, a, tbl);19
+}20
+21
+static int32x4_t rev32(const int32x4_t a)22
+{23
+ static const int8x16_t tbl = {12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3};24
+ return vqtbx1q_u8(a, a, tbl);25
+}26
+27
+static void transpose_4x4x16(int16x4_t &x0, int16x4_t &x1, int16x4_t &x2, int16x4_t &x3)28
+{29
+ int16x4_t s0, s1, s2, s3;30
+ s0 = vtrn1_s32(x0, x2);31
+ s1 = vtrn1_s32(x1, x3);32
+ s2 = vtrn2_s32(x0, x2);33
+ s3 = vtrn2_s32(x1, x3);34
+35
+ x0 = vtrn1_s16(s0, s1);36
+ x1 = vtrn2_s16(s0, s1);37
+ x2 = vtrn1_s16(s2, s3);38
+ x3 = vtrn2_s16(s2, s3);39
+}40
+41
+42
+43
+static int scanPosLast_opt(const uint16_t *scan, const coeff_t *coeff, uint16_t *coeffSign, uint16_t *coeffFlag,44
+ uint8_t *coeffNum, int numSig, const uint16_t * /*scanCG4x4*/, const int /*trSize*/)45
+{46
+47
+ // This is an optimized function for scanPosLast, which removes the rmw dependency, once integrated into mainline x265, should replace reference implementation48
+ // For clarity, left the original reference code in comments49
+ int scanPosLast = 0;50
+51
+ uint16_t cSign = 0;52
+ uint16_t cFlag = 0;53
+ uint8_t cNum = 0;54
+55
+ uint32_t prevcgIdx = 0;56
+ do57
+ {58
+ const uint32_t cgIdx = (uint32_t)scanPosLast >> MLS_CG_SIZE;59
+60
+ const uint32_t posLast = scanscanPosLast;61
+62
+ const int curCoeff = coeffposLast;63
+ const uint32_t isNZCoeff = (curCoeff != 0);64
+ /*65
+ NOTE: the new algorithm is complicated, so I keep reference code here66
+ uint32_t posy = posLast >> log2TrSize;67
+ uint32_t posx = posLast - (posy << log2TrSize);68
+ uint32_t blkIdx0 = ((posy >> MLS_CG_LOG2_SIZE) << codingParameters.log2TrSizeCG) + (posx >> MLS_CG_LOG2_SIZE);69
+ const uint32_t blkIdx = ((posLast >> (2 * MLS_CG_LOG2_SIZE)) & ~maskPosXY) + ((posLast >> MLS_CG_LOG2_SIZE) & maskPosXY);70
+ sigCoeffGroupFlag64 |= ((uint64_t)isNZCoeff << blkIdx);71
+ */72
+73
+ // get L1 sig map74
+ numSig -= isNZCoeff;75
+76
+ if (scanPosLast % (1 << MLS_CG_SIZE) == 0)77
+ {78
+ coeffSignprevcgIdx = cSign;79
+ coeffFlagprevcgIdx = cFlag;80
+ coeffNumprevcgIdx = cNum;81
+ cSign = 0;82
+ cFlag = 0;83
+ cNum = 0;84
+ }85
+ // TODO: optimize by instruction BTS86
+ cSign += (uint16_t)(((curCoeff < 0) ? 1 : 0) << cNum);87
+ cFlag = (cFlag << 1) + (uint16_t)isNZCoeff;88
+ cNum += (uint8_t)isNZCoeff;89
+ prevcgIdx = cgIdx;90
+ scanPosLast++;91
+ }92
+ while (numSig > 0);93
+94
+ coeffSignprevcgIdx = cSign;95
+ coeffFlagprevcgIdx = cFlag;96
+ coeffNumprevcgIdx = cNum;97
+ return scanPosLast - 1;98
+}99
+100
+101
+#if (MLS_CG_SIZE == 4)102
+template<int log2TrSize>103
+static void nonPsyRdoQuant_neon(int16_t *m_resiDctCoeff, int64_t *costUncoded, int64_t *totalUncodedCost,104
+ int64_t *totalRdCost, uint32_t blkPos)105
+{106
+ const int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH -107
+ log2TrSize; /* Represents scaling through forward transform */108
+ const int scaleBits = SCALE_BITS - 2 * transformShift;109
+ const uint32_t trSize = 1 << log2TrSize;110
+111
+ int64x2_t vcost_sum_0 = vdupq_n_s64(0);112
+ int64x2_t vcost_sum_1 = vdupq_n_s64(0);113
+ for (int y = 0; y < MLS_CG_SIZE; y++)114
+ {115
+ int16x4_t in = *(int16x4_t *)&m_resiDctCoeffblkPos;116
+ int32x4_t mul = vmull_s16(in, in);117
+ int64x2_t cost0, cost1;118
+ cost0 = vshll_n_s32(vget_low_s32(mul), scaleBits);119
+ cost1 = vshll_high_n_s32(mul, scaleBits);120
+ *(int64x2_t *)&costUncodedblkPos + 0 = cost0;121
+ *(int64x2_t *)&costUncodedblkPos + 2 = cost1;122
+ vcost_sum_0 = vaddq_s64(vcost_sum_0, cost0);123
+ vcost_sum_1 = vaddq_s64(vcost_sum_1, cost1);124
+ blkPos += trSize;125
+ }126
+ int64_t sum = vaddvq_s64(vaddq_s64(vcost_sum_0, vcost_sum_1));127
+ *totalUncodedCost += sum;128
+ *totalRdCost += sum;129
+}130
+131
+template<int log2TrSize>132
+static void psyRdoQuant_neon(int16_t *m_resiDctCoeff, int16_t *m_fencDctCoeff, int64_t *costUncoded,133
+ int64_t *totalUncodedCost, int64_t *totalRdCost, int64_t *psyScale, uint32_t blkPos)134
+{135
+ const int transformShift = MAX_TR_DYNAMIC_RANGE - X265_DEPTH -136
+ log2TrSize; /* Represents scaling through forward transform */137
+ const int scaleBits = SCALE_BITS - 2 * transformShift;138
+ const uint32_t trSize = 1 << log2TrSize;139
+ //using preprocessor to bypass clang bug140
+ const int max = X265_MAX(0, (2 * transformShift + 1));141
+142
+ int64x2_t vcost_sum_0 = vdupq_n_s64(0);143
+ int64x2_t vcost_sum_1 = vdupq_n_s64(0);144
+ int32x4_t vpsy = vdupq_n_s32(*psyScale);145
+ for (int y = 0; y < MLS_CG_SIZE; y++)146
+ {147
+ int32x4_t signCoef = vmovl_s16(*(int16x4_t *)&m_resiDctCoeffblkPos);148
+ int32x4_t predictedCoef = vsubq_s32(vmovl_s16(*(int16x4_t *)&m_fencDctCoeffblkPos), signCoef);149
+ int64x2_t cost0, cost1;150
+ cost0 = vmull_s32(vget_low_s32(signCoef), vget_low_s32(signCoef));151
+ cost1 = vmull_high_s32(signCoef, signCoef);152
+ cost0 = vshlq_n_s64(cost0, scaleBits);153
+ cost1 = vshlq_n_s64(cost1, scaleBits);154
+ int64x2_t neg0 = vmull_s32(vget_low_s32(predictedCoef), vget_low_s32(vpsy));155
+ int64x2_t neg1 = vmull_high_s32(predictedCoef, vpsy);156
+ if (max > 0)157
+ {158
+ int64x2_t shift = vdupq_n_s64(-max);159
+ neg0 = vshlq_s64(neg0, shift);160
+ neg1 = vshlq_s64(neg1, shift);161
+ }162
+ cost0 = vsubq_s64(cost0, neg0);163
+ cost1 = vsubq_s64(cost1, neg1);164
+ *(int64x2_t *)&costUncodedblkPos + 0 = cost0;165
+ *(int64x2_t *)&costUncodedblkPos + 2 = cost1;166
+ vcost_sum_0 = vaddq_s64(vcost_sum_0, cost0);167
+ vcost_sum_1 = vaddq_s64(vcost_sum_1, cost1);168
+169
+ blkPos += trSize;170
+ }171
+ int64_t sum = vaddvq_s64(vaddq_s64(vcost_sum_0, vcost_sum_1));172
+ *totalUncodedCost += sum;173
+ *totalRdCost += sum;174
+}175
+176
+#else177
+#error "MLS_CG_SIZE must be 4 for neon version"178
+#endif179
+180
+181
+182
+template<int trSize>183
+int count_nonzero_neon(const int16_t *quantCoeff)184
+{185
+ X265_CHECK(((intptr_t)quantCoeff & 15) == 0, "quant buffer not aligned\n");186
+ int count = 0;187
+ int16x8_t vcount = vdupq_n_s16(0);188
+ const int numCoeff = trSize * trSize;189
+ int i = 0;190
+ for (; (i + 8) <= numCoeff; i += 8)191
+ {192
+ int16x8_t in = *(int16x8_t *)&quantCoeffi;193
+ vcount = vaddq_s16(vcount, vtstq_s16(in, in));194
+ }195
+ for (; i < numCoeff; i++)196
+ {197
+ count += quantCoeffi != 0;198
+ }199
+200
+ return count - vaddvq_s16(vcount);201
+}202
+203
+template<int trSize>204
+uint32_t copy_count_neon(int16_t *coeff, const int16_t *residual, intptr_t resiStride)205
+{206
+ uint32_t numSig = 0;207
+ int16x8_t vcount = vdupq_n_s16(0);208
+ for (int k = 0; k < trSize; k++)209
+ {210
+ int j = 0;211
+ for (; (j + 8) <= trSize; j += 8)212
+ {213
+ int16x8_t in = *(int16x8_t *)&residualj;214
+ *(int16x8_t *)&coeffj = in;215
+ vcount = vaddq_s16(vcount, vtstq_s16(in, in));216
+ }217
+ for (; j < trSize; j++)218
+ {219
+ coeffj = residualj;220
+ numSig += (residualj != 0);221
+ }222
+ residual += resiStride;223
+ coeff += trSize;224
+ }225
+226
+ return numSig - vaddvq_s16(vcount);227
+}228
+229
+230
+static void partialButterfly16(const int16_t *src, int16_t *dst, int shift, int line)231
+{232
+ int j, k;233
+ int32x4_t E2, O2;234
+ int32x4_t EE, EO;235
+ int32x2_t EEE, EEO;236
+ const int add = 1 << (shift - 1);237
+ const int32x4_t _vadd = {add, 0};238
+239
+ for (j = 0; j < line; j++)240
+ {241
+ int16x8_t in0 = *(int16x8_t *)src;242
+ int16x8_t in1 = rev16(*(int16x8_t *)&src8);243
+244
+ E0 = vaddl_s16(vget_low_s16(in0), vget_low_s16(in1));245
+ O0 = vsubl_s16(vget_low_s16(in0), vget_low_s16(in1));246
+ E1 = vaddl_high_s16(in0, in1);247
+ O1 = vsubl_high_s16(in0, in1);248
+249
+ for (k = 1; k < 16; k += 2)250
+ {251
+ int32x4_t c0 = vmovl_s16(*(int16x4_t *)&g_t16k0);252
+ int32x4_t c1 = vmovl_s16(*(int16x4_t *)&g_t16k4);253
+254
+ int32x4_t res = _vadd;255
+ res = vmlaq_s32(res, c0, O0);256
+ res = vmlaq_s32(res, c1, O1);257
+ dstk * line = (int16_t)(vaddvq_s32(res) >> shift);258
+ }259
+260
+ /* EE and EO */261
+ EE = vaddq_s32(E0, rev32(E1));262
+ EO = vsubq_s32(E0, rev32(E1));263
+264
+ for (k = 2; k < 16; k += 4)265
+ {266
+ int32x4_t c0 = vmovl_s16(*(int16x4_t *)&g_t16k0);267
+ int32x4_t res = _vadd;268
+ res = vmlaq_s32(res, c0, EO);269
+ dstk * line = (int16_t)(vaddvq_s32(res) >> shift);270
+ }271
+272
+ /* EEE and EEO */273
+ EEE0 = EE0 + EE3;274
+ EEO0 = EE0 - EE3;275
+ EEE1 = EE1 + EE2;276
+ EEO1 = EE1 - EE2;277
+278
+ dst0 = (int16_t)((g_t1600 * EEE0 + g_t1601 * EEE1 + add) >> shift);279
+ dst8 * line = (int16_t)((g_t1680 * EEE0 + g_t1681 * EEE1 + add) >> shift);280
+ dst4 * line = (int16_t)((g_t1640 * EEO0 + g_t1641 * EEO1 + add) >> shift);281
+ dst12 * line = (int16_t)((g_t16120 * EEO0 + g_t16121 * EEO1 + add) >> shift);282
+283
+284
+ src += 16;285
+ dst++;286
+ }287
+}288
+289
+290
+static void partialButterfly32(const int16_t *src, int16_t *dst, int shift, int line)291
+{292
+ int j, k;293
+ const int add = 1 << (shift - 1);294
+295
+296
+ for (j = 0; j < line; j++)297
+ {298
+ int32x4_t VE4, VO0, VO1, VO2, VO3;299
+ int32x4_t VEE2, VEO2;300
+ int32x4_t VEEE, VEEO;301
+ int EEEE2, EEEO2;302
+303
+ int16x8x4_t inputs;304
+ inputs = *(int16x8x4_t *)&src0;305
+ int16x8x4_t in_rev;306
+307
+ in_rev.val1 = rev16(inputs.val2);308
+ in_rev.val0 = rev16(inputs.val3);309
+310
+ VE0 = vaddl_s16(vget_low_s16(inputs.val0), vget_low_s16(in_rev.val0));311
+ VE1 = vaddl_high_s16(inputs.val0, in_rev.val0);312
+ VO0 = vsubl_s16(vget_low_s16(inputs.val0), vget_low_s16(in_rev.val0));313
+ VO1 = vsubl_high_s16(inputs.val0, in_rev.val0);314
+ VE2 = vaddl_s16(vget_low_s16(inputs.val1), vget_low_s16(in_rev.val1));315
+ VE3 = vaddl_high_s16(inputs.val1, in_rev.val1);316
+ VO2 = vsubl_s16(vget_low_s16(inputs.val1), vget_low_s16(in_rev.val1));317
+ VO3 = vsubl_high_s16(inputs.val1, in_rev.val1);318
+319
+ for (k = 1; k < 32; k += 2)320
+ {321
+ int32x4_t c0 = vmovl_s16(*(int16x4_t *)&g_t32k0);322
+ int32x4_t c1 = vmovl_s16(*(int16x4_t *)&g_t32k4);323
+ int32x4_t c2 = vmovl_s16(*(int16x4_t *)&g_t32k8);324
+ int32x4_t c3 = vmovl_s16(*(int16x4_t *)&g_t32k12);325
+ int32x4_t s = vmulq_s32(c0, VO0);326
+ s = vmlaq_s32(s, c1, VO1);327
+ s = vmlaq_s32(s, c2, VO2);328
+ s = vmlaq_s32(s, c3, VO3);329
+330
+ dstk * line = (int16_t)((vaddvq_s32(s) + add) >> shift);331
+332
+ }333
+334
+ int32x4_t rev_VE2;335
+336
+337
+ rev_VE0 = rev32(VE3);338
+ rev_VE1 = rev32(VE2);339
+340
+ /* EE and EO */341
+ for (k = 0; k < 2; k++)342
+ {343
+ VEEk = vaddq_s32(VEk, rev_VEk);344
+ VEOk = vsubq_s32(VEk, rev_VEk);345
+ }346
+ for (k = 2; k < 32; k += 4)347
+ {348
+ int32x4_t c0 = vmovl_s16(*(int16x4_t *)&g_t32k0);349
+ int32x4_t c1 = vmovl_s16(*(int16x4_t *)&g_t32k4);350
+ int32x4_t s = vmulq_s32(c0, VEO0);351
+ s = vmlaq_s32(s, c1, VEO1);352
+353
+ dstk * line = (int16_t)((vaddvq_s32(s) + add) >> shift);354
+355
+ }356
+357
+ int32x4_t tmp = rev32(VEE1);358
+ VEEE = vaddq_s32(VEE0, tmp);359
+ VEEO = vsubq_s32(VEE0, tmp);360
+ for (k = 4; k < 32; k += 8)361
+ {362
+ int32x4_t c = vmovl_s16(*(int16x4_t *)&g_t32k0);363
+ int32x4_t s = vmulq_s32(c, VEEO);364
+365
+ dstk * line = (int16_t)((vaddvq_s32(s) + add) >> shift);366
+ }367
+368
+ /* EEEE and EEEO */369
+ EEEE0 = VEEE0 + VEEE3;370
+ EEEO0 = VEEE0 - VEEE3;371
+ EEEE1 = VEEE1 + VEEE2;372
+ EEEO1 = VEEE1 - VEEE2;373
+374
+ dst0 = (int16_t)((g_t3200 * EEEE0 + g_t3201 * EEEE1 + add) >> shift);375
+ dst16 * line = (int16_t)((g_t32160 * EEEE0 + g_t32161 * EEEE1 + add) >> shift);376
+ dst8 * line = (int16_t)((g_t3280 * EEEO0 + g_t3281 * EEEO1 + add) >> shift);377
+ dst24 * line = (int16_t)((g_t32240 * EEEO0 + g_t32241 * EEEO1 + add) >> shift);378
+379
+380
+381
+ src += 32;382
+ dst++;383
+ }384
+}385
+386
+static void partialButterfly8(const int16_t *src, int16_t *dst, int shift, int line)387
+{388
+ int j, k;389
+ int E4, O4;390
+ int EE2, EO2;391
+ int add = 1 << (shift - 1);392
+393
+ for (j = 0; j < line; j++)394
+ {395
+ /* E and O*/396
+ for (k = 0; k < 4; k++)397
+ {398
+ Ek = srck + src7 - k;399
+ Ok = srck - src7 - k;400
+ }401
+402
+ /* EE and EO */403
+ EE0 = E0 + E3;404
+ EO0 = E0 - E3;405
+ EE1 = E1 + E2;406
+ EO1 = E1 - E2;407
+408
+ dst0 = (int16_t)((g_t800 * EE0 + g_t801 * EE1 + add) >> shift);409
+ dst4 * line = (int16_t)((g_t840 * EE0 + g_t841 * EE1 + add) >> shift);410
+ dst2 * line = (int16_t)((g_t820 * EO0 + g_t821 * EO1 + add) >> shift);411
+ dst6 * line = (int16_t)((g_t860 * EO0 + g_t861 * EO1 + add) >> shift);412
+413
+ dstline = (int16_t)((g_t810 * O0 + g_t811 * O1 + g_t812 * O2 + g_t813 * O3 + add) >> shift);414
+ dst3 * line = (int16_t)((g_t830 * O0 + g_t831 * O1 + g_t832 * O2 + g_t833 * O3 + add) >>415
+ shift);416
+ dst5 * line = (int16_t)((g_t850 * O0 + g_t851 * O1 + g_t852 * O2 + g_t853 * O3 + add) >>417
+ shift);418
+ dst7 * line = (int16_t)((g_t870 * O0 + g_t871 * O1 + g_t872 * O2 + g_t873 * O3 + add) >>419
+ shift);420
+421
+ src += 8;422
+ dst++;423
+ }424
+}425
+426
+static void partialButterflyInverse4(const int16_t *src, int16_t *dst, int shift, int line)427
+{428
+ int j;429
+ int E2, O2;430
+ int add = 1 << (shift - 1);431
+432
+ for (j = 0; j < line; j++)433
+ {434
+ /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */435
+ O0 = g_t410 * srcline + g_t430 * src3 * line;436
+ O1 = g_t411 * srcline + g_t431 * src3 * line;437
+ E0 = g_t400 * src0 + g_t420 * src2 * line;438
+ E1 = g_t401 * src0 + g_t421 * src2 * line;439
+440
+ /* Combining even and odd terms at each hierarchy levels to calculate the final spatial domain vector */441
+ dst0 = (int16_t)(x265_clip3(-32768, 32767, (E0 + O0 + add) >> shift));442
+ dst1 = (int16_t)(x265_clip3(-32768, 32767, (E1 + O1 + add) >> shift));443
+ dst2 = (int16_t)(x265_clip3(-32768, 32767, (E1 - O1 + add) >> shift));444
+ dst3 = (int16_t)(x265_clip3(-32768, 32767, (E0 - O0 + add) >> shift));445
+446
+ src++;447
+ dst += 4;448
+ }449
+}450
+451
+452
+453
+static void partialButterflyInverse16_neon(const int16_t *src, int16_t *orig_dst, int shift, int line)454
+{455
+#define FMAK(x,l) sl = vmlal_lane_s16(sl,*(int16x4_t*)&src(x)*line,*(int16x4_t *)&g_t16xk,l)456
+#define MULK(x,l) vmull_lane_s16(*(int16x4_t*)&srcx*line,*(int16x4_t *)&g_t16xk,l);457
+#define ODD3_15(k) FMAK(3,k);FMAK(5,k);FMAK(7,k);FMAK(9,k);FMAK(11,k);FMAK(13,k);FMAK(15,k);458
+#define EVEN6_14_STEP4(k) FMAK(6,k);FMAK(10,k);FMAK(14,k);459
+460
+461
+ int j, k;462
+ int32x4_t E8, O8;463
+ int32x4_t EE4, EO4;464
+ int32x4_t EEE2, EEO2;465
+ const int add = 1 << (shift - 1);466
+467
+468
+#pragma unroll(4)469
+ for (j = 0; j < line; j += 4)470
+ {471
+ /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */472
+473
+#pragma unroll(2)474
+ for (k = 0; k < 2; k++)475
+ {476
+ int32x4_t s;477
+ s = vmull_s16(vdup_n_s16(g_t164k), *(int16x4_t *)&src4 * line);;478
+ EEOk = vmlal_s16(s, vdup_n_s16(g_t1612k), *(int16x4_t *)&src(12) * line);479
+ s = vmull_s16(vdup_n_s16(g_t160k), *(int16x4_t *)&src0 * line);;480
+ EEEk = vmlal_s16(s, vdup_n_s16(g_t168k), *(int16x4_t *)&src(8) * line);481
+ }482
+483
+ /* Combining even and odd terms at each hierarchy levels to calculate the final spatial domain vector */484
+ EE0 = vaddq_s32(EEE0 , EEO0);485
+ EE2 = vsubq_s32(EEE1 , EEO1);486
+ EE1 = vaddq_s32(EEE1 , EEO1);487
+ EE3 = vsubq_s32(EEE0 , EEO0);488
+489
+490
+#pragma unroll(1)491
+ for (k = 0; k < 4; k += 4)492
+ {493
+ int32x4_t s4;494
+ s0 = MULK(2, 0);495
+ s1 = MULK(2, 1);496
+ s2 = MULK(2, 2);497
+ s3 = MULK(2, 3);498
+499
+ EVEN6_14_STEP4(0);500
+ EVEN6_14_STEP4(1);501
+ EVEN6_14_STEP4(2);502
+ EVEN6_14_STEP4(3);503
+504
+ EOk = s0;505
+ EOk + 1 = s1;506
+ EOk + 2 = s2;507
+ EOk + 3 = s3;508
+ }509
+510
+511
+512
+ static const int32x4_t min = vdupq_n_s32(-32768);513
+ static const int32x4_t max = vdupq_n_s32(32767);514
+ const int32x4_t minus_shift = vdupq_n_s32(-shift);515
+516
+#pragma unroll(4)517
+ for (k = 0; k < 4; k++)518
+ {519
+ Ek = vaddq_s32(EEk , EOk);520
+ Ek + 4 = vsubq_s32(EE3 - k , EO3 - k);521
+ }522
+523
+#pragma unroll(2)524
+ for (k = 0; k < 8; k += 4)525
+ {526
+ int32x4_t s4;527
+ s0 = MULK(1, 0);528
+ s1 = MULK(1, 1);529
+ s2 = MULK(1, 2);530
+ s3 = MULK(1, 3);531
+ ODD3_15(0);532
+ ODD3_15(1);533
+ ODD3_15(2);534
+ ODD3_15(3);535
+ Ok = s0;536
+ Ok + 1 = s1;537
+ Ok + 2 = s2;538
+ Ok + 3 = s3;539
+ int32x4_t t;540
+ int16x4_t x0, x1, x2, x3;541
+542
+ Ek = vaddq_s32(vdupq_n_s32(add), Ek);543
+ t = vaddq_s32(Ek, Ok);544
+ t = vshlq_s32(t, minus_shift);545
+ t = vmaxq_s32(t, min);546
+ t = vminq_s32(t, max);547
+ x0 = vmovn_s32(t);548
+549
+ Ek + 1 = vaddq_s32(vdupq_n_s32(add), Ek + 1);550
+ t = vaddq_s32(Ek + 1, Ok + 1);551
+ t = vshlq_s32(t, minus_shift);552
+ t = vmaxq_s32(t, min);553
+ t = vminq_s32(t, max);554
+ x1 = vmovn_s32(t);555
+556
+ Ek + 2 = vaddq_s32(vdupq_n_s32(add), Ek + 2);557
+ t = vaddq_s32(Ek + 2, Ok + 2);558
+ t = vshlq_s32(t, minus_shift);559
+ t = vmaxq_s32(t, min);560
+ t = vminq_s32(t, max);561
+ x2 = vmovn_s32(t);562
+563
+ Ek + 3 = vaddq_s32(vdupq_n_s32(add), Ek + 3);564
+ t = vaddq_s32(Ek + 3, Ok + 3);565
+ t = vshlq_s32(t, minus_shift);566
+ t = vmaxq_s32(t, min);567
+ t = vminq_s32(t, max);568
+ x3 = vmovn_s32(t);569
+570
+ transpose_4x4x16(x0, x1, x2, x3);571
+ *(int16x4_t *)&orig_dst0 * 16 + k = x0;572
+ *(int16x4_t *)&orig_dst1 * 16 + k = x1;573
+ *(int16x4_t *)&orig_dst2 * 16 + k = x2;574
+ *(int16x4_t *)&orig_dst3 * 16 + k = x3;575
+ }576
+577
+578
+#pragma unroll(2)579
+ for (k = 0; k < 8; k += 4)580
+ {581
+ int32x4_t t;582
+ int16x4_t x0, x1, x2, x3;583
+584
+ t = vsubq_s32(E7 - k, O7 - k);585
+ t = vshlq_s32(t, minus_shift);586
+ t = vmaxq_s32(t, min);587
+ t = vminq_s32(t, max);588
+ x0 = vmovn_s32(t);589
+590
+ t = vsubq_s32(E6 - k, O6 - k);591
+ t = vshlq_s32(t, minus_shift);592
+ t = vmaxq_s32(t, min);593
+ t = vminq_s32(t, max);594
+ x1 = vmovn_s32(t);595
+596
+ t = vsubq_s32(E5 - k, O5 - k);597
+598
+ t = vshlq_s32(t, minus_shift);599
+ t = vmaxq_s32(t, min);600
+ t = vminq_s32(t, max);601
+ x2 = vmovn_s32(t);602
+603
+ t = vsubq_s32(E4 - k, O4 - k);604
+ t = vshlq_s32(t, minus_shift);605
+ t = vmaxq_s32(t, min);606
+ t = vminq_s32(t, max);607
+ x3 = vmovn_s32(t);608
+609
+ transpose_4x4x16(x0, x1, x2, x3);610
+ *(int16x4_t *)&orig_dst0 * 16 + k + 8 = x0;611
+ *(int16x4_t *)&orig_dst1 * 16 + k + 8 = x1;612
+ *(int16x4_t *)&orig_dst2 * 16 + k + 8 = x2;613
+ *(int16x4_t *)&orig_dst3 * 16 + k + 8 = x3;614
+ }615
+ orig_dst += 4 * 16;616
+ src += 4;617
+ }618
+619
+#undef MUL620
+#undef FMA621
+#undef FMAK622
+#undef MULK623
+#undef ODD3_15624
+#undef EVEN6_14_STEP4625
+626
+627
+}628
+629
+630
+631
+static void partialButterflyInverse32_neon(const int16_t *src, int16_t *orig_dst, int shift, int line)632
+{633
+#define MUL(x) vmull_s16(vdup_n_s16(g_t32xk),*(int16x4_t*)&srcx*line);634
+#define FMA(x) s = vmlal_s16(s,vdup_n_s16(g_t32xk),*(int16x4_t*)&src(x)*line)635
+#define FMAK(x,l) sl = vmlal_lane_s16(sl,*(int16x4_t*)&src(x)*line,*(int16x4_t *)&g_t32xk,l)636
+#define MULK(x,l) vmull_lane_s16(*(int16x4_t*)&srcx*line,*(int16x4_t *)&g_t32xk,l);637
+#define ODD31(k) FMAK(3,k);FMAK(5,k);FMAK(7,k);FMAK(9,k);FMAK(11,k);FMAK(13,k);FMAK(15,k);FMAK(17,k);FMAK(19,k);FMAK(21,k);FMAK(23,k);FMAK(25,k);FMAK(27,k);FMAK(29,k);FMAK(31,k);638
+639
+#define ODD15(k) FMAK(6,k);FMAK(10,k);FMAK(14,k);FMAK(18,k);FMAK(22,k);FMAK(26,k);FMAK(30,k);640
+#define ODD7(k) FMAK(12,k);FMAK(20,k);FMAK(28,k);641
+642
+643
+ int j, k;644
+ int32x4_t E16, O16;645
+ int32x4_t EE8, EO8;646
+ int32x4_t EEE4, EEO4;647
+ int32x4_t EEEE2, EEEO2;648
+ int16x4_t dst32;649
+ int add = 1 << (shift - 1);650
+651
+#pragma unroll (8)652
+ for (j = 0; j < line; j += 4)653
+ {654
+#pragma unroll (4)655
+ for (k = 0; k < 16; k += 4)656
+ {657
+ int32x4_t s4;658
+ s0 = MULK(1, 0);659
+ s1 = MULK(1, 1);660
+ s2 = MULK(1, 2);661
+ s3 = MULK(1, 3);662
+ ODD31(0);663
+ ODD31(1);664
+ ODD31(2);665
+ ODD31(3);666
+ Ok = s0;667
+ Ok + 1 = s1;668
+ Ok + 2 = s2;669
+ Ok + 3 = s3;670
+671
+672
+ }673
+674
+675
+#pragma unroll (2)676
+ for (k = 0; k < 8; k += 4)677
+ {678
+ int32x4_t s4;679
+ s0 = MULK(2, 0);680
+ s1 = MULK(2, 1);681
+ s2 = MULK(2, 2);682
+ s3 = MULK(2, 3);683
+684
+ ODD15(0);685
+ ODD15(1);686
+ ODD15(2);687
+ ODD15(3);688
+689
+ EOk = s0;690
+ EOk + 1 = s1;691
+ EOk + 2 = s2;692
+ EOk + 3 = s3;693
+ }694
+695
+696
+ for (k = 0; k < 4; k += 4)697
+ {698
+ int32x4_t s4;699
+ s0 = MULK(4, 0);700
+ s1 = MULK(4, 1);701
+ s2 = MULK(4, 2);702
+ s3 = MULK(4, 3);703
+704
+ ODD7(0);705
+ ODD7(1);706
+ ODD7(2);707
+ ODD7(3);708
+709
+ EEOk = s0;710
+ EEOk + 1 = s1;711
+ EEOk + 2 = s2;712
+ EEOk + 3 = s3;713
+ }714
+715
+#pragma unroll (2)716
+ for (k = 0; k < 2; k++)717
+ {718
+ int32x4_t s;719
+ s = MUL(8);720
+ EEEOk = FMA(24);721
+ s = MUL(0);722
+ EEEEk = FMA(16);723
+ }724
+ /* Combining even and odd terms at each hierarchy levels to calculate the final spatial domain vector */725
+ EEE0 = vaddq_s32(EEEE0, EEEO0);726
+ EEE3 = vsubq_s32(EEEE0, EEEO0);727
+ EEE1 = vaddq_s32(EEEE1, EEEO1);728
+ EEE2 = vsubq_s32(EEEE1, EEEO1);729
+730
+#pragma unroll (4)731
+ for (k = 0; k < 4; k++)732
+ {733
+ EEk = vaddq_s32(EEEk, EEOk);734
+ EEk + 4 = vsubq_s32((EEE3 - k), (EEO3 - k));735
+ }736
+737
+#pragma unroll (8)738
+ for (k = 0; k < 8; k++)739
+ {740
+ Ek = vaddq_s32(EEk, EOk);741
+ Ek + 8 = vsubq_s32((EE7 - k), (EO7 - k));742
+ }743
+744
+ static const int32x4_t min = vdupq_n_s32(-32768);745
+ static const int32x4_t max = vdupq_n_s32(32767);746
+747
+748
+749
+#pragma unroll (16)750
+ for (k = 0; k < 16; k++)751
+ {752
+ int32x4_t adde = vaddq_s32(vdupq_n_s32(add), Ek);753
+ int32x4_t s = vaddq_s32(adde, Ok);754
+ s = vshlq_s32(s, vdupq_n_s32(-shift));755
+ s = vmaxq_s32(s, min);756
+ s = vminq_s32(s, max);757
+758
+759
+760
+ dstk = vmovn_s32(s);761
+ adde = vaddq_s32(vdupq_n_s32(add), (E15 - k));762
+ s = vsubq_s32(adde, (O15 - k));763
+ s = vshlq_s32(s, vdupq_n_s32(-shift));764
+ s = vmaxq_s32(s, min);765
+ s = vminq_s32(s, max);766
+767
+ dstk + 16 = vmovn_s32(s);768
+ }769
+770
+771
+#pragma unroll (8)772
+ for (k = 0; k < 32; k += 4)773
+ {774
+ int16x4_t x0 = dstk + 0;775
+ int16x4_t x1 = dstk + 1;776
+ int16x4_t x2 = dstk + 2;777
+ int16x4_t x3 = dstk + 3;778
+ transpose_4x4x16(x0, x1, x2, x3);779
+ *(int16x4_t *)&orig_dst0 * 32 + k = x0;780
+ *(int16x4_t *)&orig_dst1 * 32 + k = x1;781
+ *(int16x4_t *)&orig_dst2 * 32 + k = x2;782
+ *(int16x4_t *)&orig_dst3 * 32 + k = x3;783
+ }784
+ orig_dst += 4 * 32;785
+ src += 4;786
+ }787
+#undef MUL788
+#undef FMA789
+#undef FMAK790
+#undef MULK791
+#undef ODD31792
+#undef ODD15793
+#undef ODD7794
+795
+}796
+797
+798
+static void dct8_neon(const int16_t *src, int16_t *dst, intptr_t srcStride)799
+{800
+ const int shift_1st = 2 + X265_DEPTH - 8;801
+ const int shift_2nd = 9;802
+803
+ ALIGN_VAR_32(int16_t, coef8 * 8);804
+ ALIGN_VAR_32(int16_t, block8 * 8);805
+806
+ for (int i = 0; i < 8; i++)807
+ {808
+ memcpy(&blocki * 8, &srci * srcStride, 8 * sizeof(int16_t));809
+ }810
+811
+ partialButterfly8(block, coef, shift_1st, 8);812
+ partialButterfly8(coef, dst, shift_2nd, 8);813
+}814
+815
+static void dct16_neon(const int16_t *src, int16_t *dst, intptr_t srcStride)816
+{817
+ const int shift_1st = 3 + X265_DEPTH - 8;818
+ const int shift_2nd = 10;819
+820
+ ALIGN_VAR_32(int16_t, coef16 * 16);821
+ ALIGN_VAR_32(int16_t, block16 * 16);822
+823
+ for (int i = 0; i < 16; i++)824
+ {825
+ memcpy(&blocki * 16, &srci * srcStride, 16 * sizeof(int16_t));826
+ }827
+828
+ partialButterfly16(block, coef, shift_1st, 16);829
+ partialButterfly16(coef, dst, shift_2nd, 16);830
+}831
+832
+static void dct32_neon(const int16_t *src, int16_t *dst, intptr_t srcStride)833
+{834
+ const int shift_1st = 4 + X265_DEPTH - 8;835
+ const int shift_2nd = 11;836
+837
+ ALIGN_VAR_32(int16_t, coef32 * 32);838
+ ALIGN_VAR_32(int16_t, block32 * 32);839
+840
+ for (int i = 0; i < 32; i++)841
+ {842
+ memcpy(&blocki * 32, &srci * srcStride, 32 * sizeof(int16_t));843
+ }844
+845
+ partialButterfly32(block, coef, shift_1st, 32);846
+ partialButterfly32(coef, dst, shift_2nd, 32);847
+}848
+849
+static void idct4_neon(const int16_t *src, int16_t *dst, intptr_t dstStride)850
+{851
+ const int shift_1st = 7;852
+ const int shift_2nd = 12 - (X265_DEPTH - 8);853
+854
+ ALIGN_VAR_32(int16_t, coef4 * 4);855
+ ALIGN_VAR_32(int16_t, block4 * 4);856
+857
+ partialButterflyInverse4(src, coef, shift_1st, 4); // Forward DST BY FAST ALGORITHM, block input, coef output858
+ partialButterflyInverse4(coef, block, shift_2nd, 4); // Forward DST BY FAST ALGORITHM, coef input, coeff output859
+860
+ for (int i = 0; i < 4; i++)861
+ {862
+ memcpy(&dsti * dstStride, &blocki * 4, 4 * sizeof(int16_t));863
+ }864
+}865
+866
+static void idct16_neon(const int16_t *src, int16_t *dst, intptr_t dstStride)867
+{868
+ const int shift_1st = 7;869
+ const int shift_2nd = 12 - (X265_DEPTH - 8);870
+871
+ ALIGN_VAR_32(int16_t, coef16 * 16);872
+ ALIGN_VAR_32(int16_t, block16 * 16);873
+874
+ partialButterflyInverse16_neon(src, coef, shift_1st, 16);875
+ partialButterflyInverse16_neon(coef, block, shift_2nd, 16);876
+877
+ for (int i = 0; i < 16; i++)878
+ {879
+ memcpy(&dsti * dstStride, &blocki * 16, 16 * sizeof(int16_t));880
+ }881
+}882
+883
+static void idct32_neon(const int16_t *src, int16_t *dst, intptr_t dstStride)884
+{885
+ const int shift_1st = 7;886
+ const int shift_2nd = 12 - (X265_DEPTH - 8);887
+888
+ ALIGN_VAR_32(int16_t, coef32 * 32);889
+ ALIGN_VAR_32(int16_t, block32 * 32);890
+891
+ partialButterflyInverse32_neon(src, coef, shift_1st, 32);892
+ partialButterflyInverse32_neon(coef, block, shift_2nd, 32);893
+894
+ for (int i = 0; i < 32; i++)895
+ {896
+ memcpy(&dsti * dstStride, &blocki * 32, 32 * sizeof(int16_t));897
+ }898
+}899
+900
+901
+902
+}903
+904
+namespace X265_NS905
+{906
+// x265 private namespace907
+void setupDCTPrimitives_neon(EncoderPrimitives &p)908
+{909
+ p.cuBLOCK_4x4.nonPsyRdoQuant = nonPsyRdoQuant_neon<2>;910
+ p.cuBLOCK_8x8.nonPsyRdoQuant = nonPsyRdoQuant_neon<3>;911
+ p.cuBLOCK_16x16.nonPsyRdoQuant = nonPsyRdoQuant_neon<4>;912
+ p.cuBLOCK_32x32.nonPsyRdoQuant = nonPsyRdoQuant_neon<5>;913
+ p.cuBLOCK_4x4.psyRdoQuant = psyRdoQuant_neon<2>;914
+ p.cuBLOCK_8x8.psyRdoQuant = psyRdoQuant_neon<3>;915
+ p.cuBLOCK_16x16.psyRdoQuant = psyRdoQuant_neon<4>;916
+ p.cuBLOCK_32x32.psyRdoQuant = psyRdoQuant_neon<5>;917
+ p.cuBLOCK_8x8.dct = dct8_neon;918
+ p.cuBLOCK_16x16.dct = dct16_neon;919
+ p.cuBLOCK_32x32.dct = dct32_neon;920
+ p.cuBLOCK_4x4.idct = idct4_neon;921
+ p.cuBLOCK_16x16.idct = idct16_neon;922
+ p.cuBLOCK_32x32.idct = idct32_neon;923
+ p.cuBLOCK_4x4.count_nonzero = count_nonzero_neon<4>;924
+ p.cuBLOCK_8x8.count_nonzero = count_nonzero_neon<8>;925
+ p.cuBLOCK_16x16.count_nonzero = count_nonzero_neon<16>;926
+ p.cuBLOCK_32x32.count_nonzero = count_nonzero_neon<32>;927
+928
+ p.cuBLOCK_4x4.copy_cnt = copy_count_neon<4>;929
+ p.cuBLOCK_8x8.copy_cnt = copy_count_neon<8>;930
+ p.cuBLOCK_16x16.copy_cnt = copy_count_neon<16>;931
+ p.cuBLOCK_32x32.copy_cnt = copy_count_neon<32>;932
+ p.cuBLOCK_4x4.psyRdoQuant_1p = nonPsyRdoQuant_neon<2>;933
+ p.cuBLOCK_4x4.psyRdoQuant_2p = psyRdoQuant_neon<2>;934
+ p.cuBLOCK_8x8.psyRdoQuant_1p = nonPsyRdoQuant_neon<3>;935
+ p.cuBLOCK_8x8.psyRdoQuant_2p = psyRdoQuant_neon<3>;936
+ p.cuBLOCK_16x16.psyRdoQuant_1p = nonPsyRdoQuant_neon<4>;937
+ p.cuBLOCK_16x16.psyRdoQuant_2p = psyRdoQuant_neon<4>;938
+ p.cuBLOCK_32x32.psyRdoQuant_1p = nonPsyRdoQuant_neon<5>;939
+ p.cuBLOCK_32x32.psyRdoQuant_2p = psyRdoQuant_neon<5>;940
+941
+ p.scanPosLast = scanPosLast_opt;942
+943
+}944
+945
+};946
+947
+948
+949
+#endif950
x265_3.6.tar.gz/source/common/aarch64/dct-prim.h
Added
21
1
2
+#ifndef __DCT_PRIM_NEON_H__3
+#define __DCT_PRIM_NEON_H__4
+5
+6
+#include "common.h"7
+#include "primitives.h"8
+#include "contexts.h" // costCoeffNxN_c9
+#include "threading.h" // CLZ10
+11
+namespace X265_NS12
+{13
+// x265 private namespace14
+void setupDCTPrimitives_neon(EncoderPrimitives &p);15
+};16
+17
+18
+19
+#endif20
+21
x265_3.6.tar.gz/source/common/aarch64/filter-prim.cpp
Added
997
1
2
+#if HAVE_NEON3
+4
+#include "filter-prim.h"5
+#include <arm_neon.h>6
+7
+namespace8
+{9
+10
+using namespace X265_NS;11
+12
+13
+template<int width, int height>14
+void filterPixelToShort_neon(const pixel *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride)15
+{16
+ const int shift = IF_INTERNAL_PREC - X265_DEPTH;17
+ int row, col;18
+ const int16x8_t off = vdupq_n_s16(IF_INTERNAL_OFFS);19
+ for (row = 0; row < height; row++)20
+ {21
+22
+ for (col = 0; col < width; col += 8)23
+ {24
+ int16x8_t in;25
+26
+#if HIGH_BIT_DEPTH27
+ in = *(int16x8_t *)&srccol;28
+#else29
+ in = vmovl_u8(*(uint8x8_t *)&srccol);30
+#endif31
+32
+ int16x8_t tmp = vshlq_n_s16(in, shift);33
+ tmp = vsubq_s16(tmp, off);34
+ *(int16x8_t *)&dstcol = tmp;35
+36
+ }37
+38
+ src += srcStride;39
+ dst += dstStride;40
+ }41
+}42
+43
+44
+template<int N, int width, int height>45
+void interp_horiz_pp_neon(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)46
+{47
+ const int16_t *coeff = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;48
+ int headRoom = IF_FILTER_PREC;49
+ int offset = (1 << (headRoom - 1));50
+ uint16_t maxVal = (1 << X265_DEPTH) - 1;51
+ int cStride = 1;52
+53
+ src -= (N / 2 - 1) * cStride;54
+ int16x8_t vc;55
+ vc = *(int16x8_t *)coeff;56
+ int16x4_t low_vc = vget_low_s16(vc);57
+ int16x4_t high_vc = vget_high_s16(vc);58
+59
+ const int32x4_t voffset = vdupq_n_s32(offset);60
+ const int32x4_t vhr = vdupq_n_s32(-headRoom);61
+62
+ int row, col;63
+ for (row = 0; row < height; row++)64
+ {65
+ for (col = 0; col < width; col += 8)66
+ {67
+ int32x4_t vsum1, vsum2;68
+69
+ int16x8_t inputN;70
+71
+ for (int i = 0; i < N; i++)72
+ {73
+#if HIGH_BIT_DEPTH74
+ inputi = *(int16x8_t *)&srccol + i;75
+#else76
+ inputi = vmovl_u8(*(uint8x8_t *)&srccol + i);77
+#endif78
+ }79
+ vsum1 = voffset;80
+ vsum2 = voffset;81
+82
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input0), low_vc, 0);83
+ vsum2 = vmlal_high_lane_s16(vsum2, input0, low_vc, 0);84
+85
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input1), low_vc, 1);86
+ vsum2 = vmlal_high_lane_s16(vsum2, input1, low_vc, 1);87
+88
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input2), low_vc, 2);89
+ vsum2 = vmlal_high_lane_s16(vsum2, input2, low_vc, 2);90
+91
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input3), low_vc, 3);92
+ vsum2 = vmlal_high_lane_s16(vsum2, input3, low_vc, 3);93
+94
+ if (N == 8)95
+ {96
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input4), high_vc, 0);97
+ vsum2 = vmlal_high_lane_s16(vsum2, input4, high_vc, 0);98
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input5), high_vc, 1);99
+ vsum2 = vmlal_high_lane_s16(vsum2, input5, high_vc, 1);100
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input6), high_vc, 2);101
+ vsum2 = vmlal_high_lane_s16(vsum2, input6, high_vc, 2);102
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input7), high_vc, 3);103
+ vsum2 = vmlal_high_lane_s16(vsum2, input7, high_vc, 3);104
+105
+ }106
+107
+ vsum1 = vshlq_s32(vsum1, vhr);108
+ vsum2 = vshlq_s32(vsum2, vhr);109
+110
+ int16x8_t vsum = vuzp1q_s16(vsum1, vsum2);111
+ vsum = vminq_s16(vsum, vdupq_n_s16(maxVal));112
+ vsum = vmaxq_s16(vsum, vdupq_n_s16(0));113
+#if HIGH_BIT_DEPTH114
+ *(int16x8_t *)&dstcol = vsum;115
+#else116
+ uint8x16_t usum = vuzp1q_u8(vsum, vsum);117
+ *(uint8x8_t *)&dstcol = vget_low_u8(usum);118
+#endif119
+120
+ }121
+122
+ src += srcStride;123
+ dst += dstStride;124
+ }125
+}126
+127
+#if HIGH_BIT_DEPTH128
+129
+template<int N, int width, int height>130
+void interp_horiz_ps_neon(const uint16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx,131
+ int isRowExt)132
+{133
+ const int16_t *coeff = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;134
+ const int headRoom = IF_INTERNAL_PREC - X265_DEPTH;135
+ const int shift = IF_FILTER_PREC - headRoom;136
+ const int offset = (unsigned) - IF_INTERNAL_OFFS << shift;137
+138
+ int blkheight = height;139
+ src -= N / 2 - 1;140
+141
+ if (isRowExt)142
+ {143
+ src -= (N / 2 - 1) * srcStride;144
+ blkheight += N - 1;145
+ }146
+ int16x8_t vc3 = vld1q_s16(coeff);147
+ const int32x4_t voffset = vdupq_n_s32(offset);148
+ const int32x4_t vhr = vdupq_n_s32(-shift);149
+150
+ int row, col;151
+ for (row = 0; row < blkheight; row++)152
+ {153
+ for (col = 0; col < width; col += 8)154
+ {155
+ int32x4_t vsum, vsum2;156
+157
+ int16x8_t inputN;158
+ for (int i = 0; i < N; i++)159
+ {160
+ inputi = vld1q_s16((int16_t *)&srccol + i);161
+ }162
+163
+ vsum = voffset;164
+ vsum2 = voffset;165
+166
+ vsum = vmlal_lane_s16(vsum, vget_low_u16(input0), vget_low_s16(vc3), 0);167
+ vsum2 = vmlal_high_lane_s16(vsum2, input0, vget_low_s16(vc3), 0);168
+169
+ vsum = vmlal_lane_s16(vsum, vget_low_u16(input1), vget_low_s16(vc3), 1);170
+ vsum2 = vmlal_high_lane_s16(vsum2, input1, vget_low_s16(vc3), 1);171
+172
+ vsum = vmlal_lane_s16(vsum, vget_low_u16(input2), vget_low_s16(vc3), 2);173
+ vsum2 = vmlal_high_lane_s16(vsum2, input2, vget_low_s16(vc3), 2);174
+175
+ vsum = vmlal_lane_s16(vsum, vget_low_u16(input3), vget_low_s16(vc3), 3);176
+ vsum2 = vmlal_high_lane_s16(vsum2, input3, vget_low_s16(vc3), 3);177
+178
+ if (N == 8)179
+ {180
+ vsum = vmlal_lane_s16(vsum, vget_low_s16(input4), vget_high_s16(vc3), 0);181
+ vsum2 = vmlal_high_lane_s16(vsum2, input4, vget_high_s16(vc3), 0);182
+183
+ vsum = vmlal_lane_s16(vsum, vget_low_s16(input5), vget_high_s16(vc3), 1);184
+ vsum2 = vmlal_high_lane_s16(vsum2, input5, vget_high_s16(vc3), 1);185
+186
+ vsum = vmlal_lane_s16(vsum, vget_low_s16(input6), vget_high_s16(vc3), 2);187
+ vsum2 = vmlal_high_lane_s16(vsum2, input6, vget_high_s16(vc3), 2);188
+189
+ vsum = vmlal_lane_s16(vsum, vget_low_s16(input7), vget_high_s16(vc3), 3);190
+ vsum2 = vmlal_high_lane_s16(vsum2, input7, vget_high_s16(vc3), 3);191
+ }192
+193
+ vsum = vshlq_s32(vsum, vhr);194
+ vsum2 = vshlq_s32(vsum2, vhr);195
+ *(int16x4_t *)&dstcol = vmovn_u32(vsum);196
+ *(int16x4_t *)&dstcol+4 = vmovn_u32(vsum2);197
+ }198
+199
+ src += srcStride;200
+ dst += dstStride;201
+ }202
+}203
+204
+205
+#else206
+207
+template<int N, int width, int height>208
+void interp_horiz_ps_neon(const uint8_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx,209
+ int isRowExt)210
+{211
+ const int16_t *coeff = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;212
+ const int headRoom = IF_INTERNAL_PREC - X265_DEPTH;213
+ const int shift = IF_FILTER_PREC - headRoom;214
+ const int offset = (unsigned) - IF_INTERNAL_OFFS << shift;215
+216
+ int blkheight = height;217
+ src -= N / 2 - 1;218
+219
+ if (isRowExt)220
+ {221
+ src -= (N / 2 - 1) * srcStride;222
+ blkheight += N - 1;223
+ }224
+ int16x8_t vc;225
+ vc = *(int16x8_t *)coeff;226
+227
+ const int16x8_t voffset = vdupq_n_s16(offset);228
+ const int16x8_t vhr = vdupq_n_s16(-shift);229
+230
+ int row, col;231
+ for (row = 0; row < blkheight; row++)232
+ {233
+ for (col = 0; col < width; col += 8)234
+ {235
+ int16x8_t vsum;236
+237
+ int16x8_t inputN;238
+239
+ for (int i = 0; i < N; i++)240
+ {241
+ inputi = vmovl_u8(*(uint8x8_t *)&srccol + i);242
+ }243
+ vsum = voffset;244
+ vsum = vmlaq_laneq_s16(vsum, (input0), vc, 0);245
+ vsum = vmlaq_laneq_s16(vsum, (input1), vc, 1);246
+ vsum = vmlaq_laneq_s16(vsum, (input2), vc, 2);247
+ vsum = vmlaq_laneq_s16(vsum, (input3), vc, 3);248
+249
+250
+ if (N == 8)251
+ {252
+ vsum = vmlaq_laneq_s16(vsum, (input4), vc, 4);253
+ vsum = vmlaq_laneq_s16(vsum, (input5), vc, 5);254
+ vsum = vmlaq_laneq_s16(vsum, (input6), vc, 6);255
+ vsum = vmlaq_laneq_s16(vsum, (input7), vc, 7);256
+257
+ }258
+259
+ vsum = vshlq_s16(vsum, vhr);260
+ *(int16x8_t *)&dstcol = vsum;261
+ }262
+263
+ src += srcStride;264
+ dst += dstStride;265
+ }266
+}267
+268
+#endif269
+270
+271
+template<int N, int width, int height>272
+void interp_vert_ss_neon(const int16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)273
+{274
+ const int16_t *c = (N == 8 ? g_lumaFiltercoeffIdx : g_chromaFiltercoeffIdx);275
+ int shift = IF_FILTER_PREC;276
+ src -= (N / 2 - 1) * srcStride;277
+ int16x8_t vc;278
+ vc = *(int16x8_t *)c;279
+ int16x4_t low_vc = vget_low_s16(vc);280
+ int16x4_t high_vc = vget_high_s16(vc);281
+282
+ const int32x4_t vhr = vdupq_n_s32(-shift);283
+284
+ int row, col;285
+ for (row = 0; row < height; row++)286
+ {287
+ for (col = 0; col < width; col += 8)288
+ {289
+ int32x4_t vsum1, vsum2;290
+291
+ int16x8_t inputN;292
+293
+ for (int i = 0; i < N; i++)294
+ {295
+ inputi = *(int16x8_t *)&srccol + i * srcStride;296
+ }297
+298
+ vsum1 = vmull_lane_s16(vget_low_s16(input0), low_vc, 0);299
+ vsum2 = vmull_high_lane_s16(input0, low_vc, 0);300
+301
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input1), low_vc, 1);302
+ vsum2 = vmlal_high_lane_s16(vsum2, input1, low_vc, 1);303
+304
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input2), low_vc, 2);305
+ vsum2 = vmlal_high_lane_s16(vsum2, input2, low_vc, 2);306
+307
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input3), low_vc, 3);308
+ vsum2 = vmlal_high_lane_s16(vsum2, input3, low_vc, 3);309
+310
+ if (N == 8)311
+ {312
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input4), high_vc, 0);313
+ vsum2 = vmlal_high_lane_s16(vsum2, input4, high_vc, 0);314
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input5), high_vc, 1);315
+ vsum2 = vmlal_high_lane_s16(vsum2, input5, high_vc, 1);316
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input6), high_vc, 2);317
+ vsum2 = vmlal_high_lane_s16(vsum2, input6, high_vc, 2);318
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input7), high_vc, 3);319
+ vsum2 = vmlal_high_lane_s16(vsum2, input7, high_vc, 3);320
+321
+ }322
+323
+ vsum1 = vshlq_s32(vsum1, vhr);324
+ vsum2 = vshlq_s32(vsum2, vhr);325
+326
+ int16x8_t vsum = vuzp1q_s16(vsum1, vsum2);327
+ *(int16x8_t *)&dstcol = vsum;328
+ }329
+330
+ src += srcStride;331
+ dst += dstStride;332
+ }333
+334
+}335
+336
+337
+#if HIGH_BIT_DEPTH338
+339
+template<int N, int width, int height>340
+void interp_vert_pp_neon(const uint16_t *src, intptr_t srcStride, uint16_t *dst, intptr_t dstStride, int coeffIdx)341
+{342
+343
+ const int16_t *c = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;344
+ int shift = IF_FILTER_PREC;345
+ int offset = 1 << (shift - 1);346
+ const uint16_t maxVal = (1 << X265_DEPTH) - 1;347
+348
+ src -= (N / 2 - 1) * srcStride;349
+ int16x8_t vc;350
+ vc = *(int16x8_t *)c;351
+ int32x4_t low_vc = vmovl_s16(vget_low_s16(vc));352
+ int32x4_t high_vc = vmovl_s16(vget_high_s16(vc));353
+354
+ const int32x4_t voffset = vdupq_n_s32(offset);355
+ const int32x4_t vhr = vdupq_n_s32(-shift);356
+357
+ int row, col;358
+ for (row = 0; row < height; row++)359
+ {360
+ for (col = 0; col < width; col += 4)361
+ {362
+ int32x4_t vsum;363
+364
+ int32x4_t inputN;365
+366
+ for (int i = 0; i < N; i++)367
+ {368
+ inputi = vmovl_u16(*(uint16x4_t *)&srccol + i * srcStride);369
+ }370
+ vsum = voffset;371
+372
+ vsum = vmlaq_laneq_s32(vsum, (input0), low_vc, 0);373
+ vsum = vmlaq_laneq_s32(vsum, (input1), low_vc, 1);374
+ vsum = vmlaq_laneq_s32(vsum, (input2), low_vc, 2);375
+ vsum = vmlaq_laneq_s32(vsum, (input3), low_vc, 3);376
+377
+ if (N == 8)378
+ {379
+ vsum = vmlaq_laneq_s32(vsum, (input4), high_vc, 0);380
+ vsum = vmlaq_laneq_s32(vsum, (input5), high_vc, 1);381
+ vsum = vmlaq_laneq_s32(vsum, (input6), high_vc, 2);382
+ vsum = vmlaq_laneq_s32(vsum, (input7), high_vc, 3);383
+ }384
+385
+ vsum = vshlq_s32(vsum, vhr);386
+ vsum = vminq_s32(vsum, vdupq_n_s32(maxVal));387
+ vsum = vmaxq_s32(vsum, vdupq_n_s32(0));388
+ *(uint16x4_t *)&dstcol = vmovn_u32(vsum);389
+ }390
+ src += srcStride;391
+ dst += dstStride;392
+ }393
+}394
+395
+396
+397
+398
+#else399
+400
+template<int N, int width, int height>401
+void interp_vert_pp_neon(const uint8_t *src, intptr_t srcStride, uint8_t *dst, intptr_t dstStride, int coeffIdx)402
+{403
+404
+ const int16_t *c = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;405
+ int shift = IF_FILTER_PREC;406
+ int offset = 1 << (shift - 1);407
+ const uint16_t maxVal = (1 << X265_DEPTH) - 1;408
+409
+ src -= (N / 2 - 1) * srcStride;410
+ int16x8_t vc;411
+ vc = *(int16x8_t *)c;412
+413
+ const int16x8_t voffset = vdupq_n_s16(offset);414
+ const int16x8_t vhr = vdupq_n_s16(-shift);415
+416
+ int row, col;417
+ for (row = 0; row < height; row++)418
+ {419
+ for (col = 0; col < width; col += 8)420
+ {421
+ int16x8_t vsum;422
+423
+ int16x8_t inputN;424
+425
+ for (int i = 0; i < N; i++)426
+ {427
+ inputi = vmovl_u8(*(uint8x8_t *)&srccol + i * srcStride);428
+ }429
+ vsum = voffset;430
+431
+ vsum = vmlaq_laneq_s16(vsum, (input0), vc, 0);432
+ vsum = vmlaq_laneq_s16(vsum, (input1), vc, 1);433
+ vsum = vmlaq_laneq_s16(vsum, (input2), vc, 2);434
+ vsum = vmlaq_laneq_s16(vsum, (input3), vc, 3);435
+436
+ if (N == 8)437
+ {438
+ vsum = vmlaq_laneq_s16(vsum, (input4), vc, 4);439
+ vsum = vmlaq_laneq_s16(vsum, (input5), vc, 5);440
+ vsum = vmlaq_laneq_s16(vsum, (input6), vc, 6);441
+ vsum = vmlaq_laneq_s16(vsum, (input7), vc, 7);442
+443
+ }444
+445
+ vsum = vshlq_s16(vsum, vhr);446
+447
+ vsum = vminq_s16(vsum, vdupq_n_s16(maxVal));448
+ vsum = vmaxq_s16(vsum, vdupq_n_s16(0));449
+ uint8x16_t usum = vuzp1q_u8(vsum, vsum);450
+ *(uint8x8_t *)&dstcol = vget_low_u8(usum);451
+452
+ }453
+454
+ src += srcStride;455
+ dst += dstStride;456
+ }457
+}458
+459
+460
+#endif461
+462
+463
+#if HIGH_BIT_DEPTH464
+465
+template<int N, int width, int height>466
+void interp_vert_ps_neon(const uint16_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)467
+{468
+ const int16_t *c = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;469
+ int headRoom = IF_INTERNAL_PREC - X265_DEPTH;470
+ int shift = IF_FILTER_PREC - headRoom;471
+ int offset = (unsigned) - IF_INTERNAL_OFFS << shift;472
+ src -= (N / 2 - 1) * srcStride;473
+474
+ int16x8_t vc;475
+ vc = *(int16x8_t *)c;476
+ int32x4_t low_vc = vmovl_s16(vget_low_s16(vc));477
+ int32x4_t high_vc = vmovl_s16(vget_high_s16(vc));478
+479
+ const int32x4_t voffset = vdupq_n_s32(offset);480
+ const int32x4_t vhr = vdupq_n_s32(-shift);481
+482
+ int row, col;483
+ for (row = 0; row < height; row++)484
+ {485
+ for (col = 0; col < width; col += 4)486
+ {487
+ int16x8_t vsum;488
+489
+ int16x8_t inputN;490
+491
+ for (int i = 0; i < N; i++)492
+ {493
+ inputi = vmovl_u16(*(uint16x4_t *)&srccol + i * srcStride);494
+ }495
+ vsum = voffset;496
+497
+ vsum = vmlaq_laneq_s32(vsum, (input0), low_vc, 0);498
+ vsum = vmlaq_laneq_s32(vsum, (input1), low_vc, 1);499
+ vsum = vmlaq_laneq_s32(vsum, (input2), low_vc, 2);500
+ vsum = vmlaq_laneq_s32(vsum, (input3), low_vc, 3);501
+502
+ if (N == 8)503
+ {504
+ int16x8_t vsum1 = vmulq_laneq_s32((input4), high_vc, 0);505
+ vsum1 = vmlaq_laneq_s32(vsum1, (input5), high_vc, 1);506
+ vsum1 = vmlaq_laneq_s32(vsum1, (input6), high_vc, 2);507
+ vsum1 = vmlaq_laneq_s32(vsum1, (input7), high_vc, 3);508
+ vsum = vaddq_s32(vsum, vsum1);509
+ }510
+511
+ vsum = vshlq_s32(vsum, vhr);512
+513
+ *(uint16x4_t *)&dstcol = vmovn_s32(vsum);514
+ }515
+516
+ src += srcStride;517
+ dst += dstStride;518
+ }519
+}520
+521
+#else522
+523
+template<int N, int width, int height>524
+void interp_vert_ps_neon(const uint8_t *src, intptr_t srcStride, int16_t *dst, intptr_t dstStride, int coeffIdx)525
+{526
+ const int16_t *c = (N == 4) ? g_chromaFiltercoeffIdx : g_lumaFiltercoeffIdx;527
+ int headRoom = IF_INTERNAL_PREC - X265_DEPTH;528
+ int shift = IF_FILTER_PREC - headRoom;529
+ int offset = (unsigned) - IF_INTERNAL_OFFS << shift;530
+ src -= (N / 2 - 1) * srcStride;531
+532
+ int16x8_t vc;533
+ vc = *(int16x8_t *)c;534
+535
+ const int16x8_t voffset = vdupq_n_s16(offset);536
+ const int16x8_t vhr = vdupq_n_s16(-shift);537
+538
+ int row, col;539
+ for (row = 0; row < height; row++)540
+ {541
+ for (col = 0; col < width; col += 8)542
+ {543
+ int16x8_t vsum;544
+545
+ int16x8_t inputN;546
+547
+ for (int i = 0; i < N; i++)548
+ {549
+ inputi = vmovl_u8(*(uint8x8_t *)&srccol + i * srcStride);550
+ }551
+ vsum = voffset;552
+553
+ vsum = vmlaq_laneq_s16(vsum, (input0), vc, 0);554
+ vsum = vmlaq_laneq_s16(vsum, (input1), vc, 1);555
+ vsum = vmlaq_laneq_s16(vsum, (input2), vc, 2);556
+ vsum = vmlaq_laneq_s16(vsum, (input3), vc, 3);557
+558
+ if (N == 8)559
+ {560
+ int16x8_t vsum1 = vmulq_laneq_s16((input4), vc, 4);561
+ vsum1 = vmlaq_laneq_s16(vsum1, (input5), vc, 5);562
+ vsum1 = vmlaq_laneq_s16(vsum1, (input6), vc, 6);563
+ vsum1 = vmlaq_laneq_s16(vsum1, (input7), vc, 7);564
+ vsum = vaddq_s16(vsum, vsum1);565
+ }566
+567
+ vsum = vshlq_s32(vsum, vhr);568
+ *(int16x8_t *)&dstcol = vsum;569
+ }570
+571
+ src += srcStride;572
+ dst += dstStride;573
+ }574
+}575
+576
+#endif577
+578
+579
+580
+template<int N, int width, int height>581
+void interp_vert_sp_neon(const int16_t *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)582
+{583
+ int headRoom = IF_INTERNAL_PREC - X265_DEPTH;584
+ int shift = IF_FILTER_PREC + headRoom;585
+ int offset = (1 << (shift - 1)) + (IF_INTERNAL_OFFS << IF_FILTER_PREC);586
+ uint16_t maxVal = (1 << X265_DEPTH) - 1;587
+ const int16_t *coeff = (N == 8 ? g_lumaFiltercoeffIdx : g_chromaFiltercoeffIdx);588
+589
+ src -= (N / 2 - 1) * srcStride;590
+591
+ int16x8_t vc;592
+ vc = *(int16x8_t *)coeff;593
+ int16x4_t low_vc = vget_low_s16(vc);594
+ int16x4_t high_vc = vget_high_s16(vc);595
+596
+ const int32x4_t voffset = vdupq_n_s32(offset);597
+ const int32x4_t vhr = vdupq_n_s32(-shift);598
+599
+ int row, col;600
+ for (row = 0; row < height; row++)601
+ {602
+ for (col = 0; col < width; col += 8)603
+ {604
+ int32x4_t vsum1, vsum2;605
+606
+ int16x8_t inputN;607
+608
+ for (int i = 0; i < N; i++)609
+ {610
+ inputi = *(int16x8_t *)&srccol + i * srcStride;611
+ }612
+ vsum1 = voffset;613
+ vsum2 = voffset;614
+615
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input0), low_vc, 0);616
+ vsum2 = vmlal_high_lane_s16(vsum2, input0, low_vc, 0);617
+618
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input1), low_vc, 1);619
+ vsum2 = vmlal_high_lane_s16(vsum2, input1, low_vc, 1);620
+621
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input2), low_vc, 2);622
+ vsum2 = vmlal_high_lane_s16(vsum2, input2, low_vc, 2);623
+624
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input3), low_vc, 3);625
+ vsum2 = vmlal_high_lane_s16(vsum2, input3, low_vc, 3);626
+627
+ if (N == 8)628
+ {629
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input4), high_vc, 0);630
+ vsum2 = vmlal_high_lane_s16(vsum2, input4, high_vc, 0);631
+632
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input5), high_vc, 1);633
+ vsum2 = vmlal_high_lane_s16(vsum2, input5, high_vc, 1);634
+635
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input6), high_vc, 2);636
+ vsum2 = vmlal_high_lane_s16(vsum2, input6, high_vc, 2);637
+638
+ vsum1 = vmlal_lane_s16(vsum1, vget_low_s16(input7), high_vc, 3);639
+ vsum2 = vmlal_high_lane_s16(vsum2, input7, high_vc, 3);640
+ }641
+642
+ vsum1 = vshlq_s32(vsum1, vhr);643
+ vsum2 = vshlq_s32(vsum2, vhr);644
+645
+ int16x8_t vsum = vuzp1q_s16(vsum1, vsum2);646
+ vsum = vminq_s16(vsum, vdupq_n_s16(maxVal));647
+ vsum = vmaxq_s16(vsum, vdupq_n_s16(0));648
+#if HIGH_BIT_DEPTH649
+ *(int16x8_t *)&dstcol = vsum;650
+#else651
+ uint8x16_t usum = vuzp1q_u8(vsum, vsum);652
+ *(uint8x8_t *)&dstcol = vget_low_u8(usum);653
+#endif654
+655
+ }656
+657
+ src += srcStride;658
+ dst += dstStride;659
+ }660
+}661
+662
+663
+664
+665
+666
+667
+template<int N, int width, int height>668
+void interp_hv_pp_neon(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int idxX, int idxY)669
+{670
+ ALIGN_VAR_32(int16_t, immedwidth * (height + N - 1));671
+672
+ interp_horiz_ps_neon<N, width, height>(src, srcStride, immed, width, idxX, 1);673
+ interp_vert_sp_neon<N, width, height>(immed + (N / 2 - 1) * width, width, dst, dstStride, idxY);674
+}675
+676
+677
+678
+}679
+680
+681
+682
+683
+namespace X265_NS684
+{685
+#if defined(__APPLE__)686
+#define CHROMA_420(W, H) \687
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_hpp = interp_horiz_pp_neon<4, W, H>; \688
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vpp = interp_vert_pp_neon<4, W, H>; \689
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vps = interp_vert_ps_neon<4, W, H>; \690
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vsp = interp_vert_sp_neon<4, W, H>; \691
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vss = interp_vert_ss_neon<4, W, H>; \692
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.p2sNONALIGNED = filterPixelToShort_neon<W, H>;\693
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.p2sALIGNED = filterPixelToShort_neon<W, H>;694
+ 695
+#define CHROMA_FILTER_420(W, H) \696
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_hps = interp_horiz_ps_neon<4, W, H>;697
+ 698
+#else // defined(__APPLE__)699
+#define CHROMA_420(W, H) \700
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vss = interp_vert_ss_neon<4, W, H>; \701
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.p2sNONALIGNED = filterPixelToShort_neon<W, H>;\702
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.p2sALIGNED = filterPixelToShort_neon<W, H>;703
+ 704
+#define CHROMA_FILTER_420(W, H) \705
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_hpp = interp_horiz_pp_neon<4, W, H>; \706
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_hps = interp_horiz_ps_neon<4, W, H>; \707
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vpp = interp_vert_pp_neon<4, W, H>; \708
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vps = interp_vert_ps_neon<4, W, H>; \709
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.filter_vsp = interp_vert_sp_neon<4, W, H>;710
+#endif // defined(__APPLE__)711
+712
+#if defined(__APPLE__)713
+#define CHROMA_422(W, H) \714
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_hpp = interp_horiz_pp_neon<4, W, H>; \715
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vpp = interp_vert_pp_neon<4, W, H>; \716
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vps = interp_vert_ps_neon<4, W, H>; \717
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vsp = interp_vert_sp_neon<4, W, H>; \718
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vss = interp_vert_ss_neon<4, W, H>; \719
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.p2sNONALIGNED = filterPixelToShort_neon<W, H>;\720
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.p2sALIGNED = filterPixelToShort_neon<W, H>;721
+ 722
+#define CHROMA_FILTER_422(W, H) \723
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_hps = interp_horiz_ps_neon<4, W, H>;724
+ 725
+#else // defined(__APPLE__)726
+#define CHROMA_422(W, H) \727
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vss = interp_vert_ss_neon<4, W, H>; \728
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.p2sNONALIGNED = filterPixelToShort_neon<W, H>;\729
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.p2sALIGNED = filterPixelToShort_neon<W, H>;730
+ 731
+#define CHROMA_FILTER_422(W, H) \732
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_hpp = interp_horiz_pp_neon<4, W, H>; \733
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_hps = interp_horiz_ps_neon<4, W, H>; \734
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vpp = interp_vert_pp_neon<4, W, H>; \735
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vps = interp_vert_ps_neon<4, W, H>; \736
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.filter_vsp = interp_vert_sp_neon<4, W, H>;737
+#endif // defined(__APPLE__)738
+739
+#if defined(__APPLE__)740
+#define CHROMA_444(W, H) \741
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_hpp = interp_horiz_pp_neon<4, W, H>; \742
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vpp = interp_vert_pp_neon<4, W, H>; \743
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vps = interp_vert_ps_neon<4, W, H>; \744
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vsp = interp_vert_sp_neon<4, W, H>; \745
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vss = interp_vert_ss_neon<4, W, H>; \746
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.p2sNONALIGNED = filterPixelToShort_neon<W, H>;\747
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.p2sALIGNED = filterPixelToShort_neon<W, H>;748
+749
+#define CHROMA_FILTER_444(W, H) \750
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_hps = interp_horiz_ps_neon<4, W, H>;751
+ 752
+#else // defined(__APPLE__)753
+#define CHROMA_444(W, H) \754
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.p2sNONALIGNED = filterPixelToShort_neon<W, H>;\755
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.p2sALIGNED = filterPixelToShort_neon<W, H>;756
+ 757
+#define CHROMA_FILTER_444(W, H) \758
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_hpp = interp_horiz_pp_neon<4, W, H>; \759
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_hps = interp_horiz_ps_neon<4, W, H>; \760
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vpp = interp_vert_pp_neon<4, W, H>; \761
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vps = interp_vert_ps_neon<4, W, H>; \762
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vsp = interp_vert_sp_neon<4, W, H>; \763
+ p.chromaX265_CSP_I444.puLUMA_ ## W ## x ## H.filter_vss = interp_vert_ss_neon<4, W, H>;764
+#endif // defined(__APPLE__)765
+766
+#if defined(__APPLE__)767
+#define LUMA(W, H) \768
+ p.puLUMA_ ## W ## x ## H.luma_hpp = interp_horiz_pp_neon<8, W, H>; \769
+ p.puLUMA_ ## W ## x ## H.luma_vpp = interp_vert_pp_neon<8, W, H>; \770
+ p.puLUMA_ ## W ## x ## H.luma_vps = interp_vert_ps_neon<8, W, H>; \771
+ p.puLUMA_ ## W ## x ## H.luma_vsp = interp_vert_sp_neon<8, W, H>; \772
+ p.puLUMA_ ## W ## x ## H.luma_vss = interp_vert_ss_neon<8, W, H>; \773
+ p.puLUMA_ ## W ## x ## H.luma_hvpp = interp_hv_pp_neon<8, W, H>; \774
+ p.puLUMA_ ## W ## x ## H.convert_p2sNONALIGNED = filterPixelToShort_neon<W, H>;\775
+ p.puLUMA_ ## W ## x ## H.convert_p2sALIGNED = filterPixelToShort_neon<W, H>;776
+ 777
+#else // defined(__APPLE__)778
+#define LUMA(W, H) \779
+ p.puLUMA_ ## W ## x ## H.luma_vss = interp_vert_ss_neon<8, W, H>; \780
+ p.puLUMA_ ## W ## x ## H.convert_p2sNONALIGNED = filterPixelToShort_neon<W, H>;\781
+ p.puLUMA_ ## W ## x ## H.convert_p2sALIGNED = filterPixelToShort_neon<W, H>;782
+ 783
+#define LUMA_FILTER(W, H) \784
+ p.puLUMA_ ## W ## x ## H.luma_hpp = interp_horiz_pp_neon<8, W, H>; \785
+ p.puLUMA_ ## W ## x ## H.luma_vpp = interp_vert_pp_neon<8, W, H>; \786
+ p.puLUMA_ ## W ## x ## H.luma_vps = interp_vert_ps_neon<8, W, H>; \787
+ p.puLUMA_ ## W ## x ## H.luma_vsp = interp_vert_sp_neon<8, W, H>; \788
+ p.puLUMA_ ## W ## x ## H.luma_hvpp = interp_hv_pp_neon<8, W, H>;789
+#endif // defined(__APPLE__)790
+791
+void setupFilterPrimitives_neon(EncoderPrimitives &p)792
+{793
+794
+ // All neon functions assume width of multiple of 8, (2,4,12 variants are not optimized)795
+796
+ LUMA(8, 8);797
+ LUMA(8, 4);798
+ LUMA(16, 16);799
+ CHROMA_420(8, 8);800
+ LUMA(16, 8);801
+ CHROMA_420(8, 4);802
+ LUMA(8, 16);803
+ LUMA(16, 12);804
+ CHROMA_420(8, 6);805
+ LUMA(16, 4);806
+ CHROMA_420(8, 2);807
+ LUMA(32, 32);808
+ CHROMA_420(16, 16);809
+ LUMA(32, 16);810
+ CHROMA_420(16, 8);811
+ LUMA(16, 32);812
+ CHROMA_420(8, 16);813
+ LUMA(32, 24);814
+ CHROMA_420(16, 12);815
+ LUMA(24, 32);816
+ LUMA(32, 8);817
+ CHROMA_420(16, 4);818
+ LUMA(8, 32);819
+ LUMA(64, 64);820
+ CHROMA_420(32, 32);821
+ LUMA(64, 32);822
+ CHROMA_420(32, 16);823
+ LUMA(32, 64);824
+ CHROMA_420(16, 32);825
+ LUMA(64, 48);826
+ CHROMA_420(32, 24);827
+ LUMA(48, 64);828
+ CHROMA_420(24, 32);829
+ LUMA(64, 16);830
+ CHROMA_420(32, 8);831
+ LUMA(16, 64);832
+ CHROMA_420(8, 32);833
+ CHROMA_422(8, 16);834
+ CHROMA_422(8, 8);835
+ CHROMA_422(8, 12);836
+ CHROMA_422(8, 4);837
+ CHROMA_422(16, 32);838
+ CHROMA_422(16, 16);839
+ CHROMA_422(8, 32);840
+ CHROMA_422(16, 24);841
+ CHROMA_422(16, 8);842
+ CHROMA_422(32, 64);843
+ CHROMA_422(32, 32);844
+ CHROMA_422(16, 64);845
+ CHROMA_422(32, 48);846
+ CHROMA_422(24, 64);847
+ CHROMA_422(32, 16);848
+ CHROMA_422(8, 64);849
+ CHROMA_444(8, 8);850
+ CHROMA_444(8, 4);851
+ CHROMA_444(16, 16);852
+ CHROMA_444(16, 8);853
+ CHROMA_444(8, 16);854
+ CHROMA_444(16, 12);855
+ CHROMA_444(16, 4);856
+ CHROMA_444(32, 32);857
+ CHROMA_444(32, 16);858
+ CHROMA_444(16, 32);859
+ CHROMA_444(32, 24);860
+ CHROMA_444(24, 32);861
+ CHROMA_444(32, 8);862
+ CHROMA_444(8, 32);863
+ CHROMA_444(64, 64);864
+ CHROMA_444(64, 32);865
+ CHROMA_444(32, 64);866
+ CHROMA_444(64, 48);867
+ CHROMA_444(48, 64);868
+ CHROMA_444(64, 16);869
+ CHROMA_444(16, 64);870
+871
+#if defined(__APPLE__) || HIGH_BIT_DEPTH872
+ p.puLUMA_8x4.luma_hps = interp_horiz_ps_neon<8, 8, 4>;873
+ p.puLUMA_8x8.luma_hps = interp_horiz_ps_neon<8, 8, 8>;874
+ p.puLUMA_8x16.luma_hps = interp_horiz_ps_neon<8, 8, 16>;875
+ p.puLUMA_8x32.luma_hps = interp_horiz_ps_neon<8, 8, 32>;876
+#endif // HIGH_BIT_DEPTH877
+878
+#if !defined(__APPLE__) && HIGH_BIT_DEPTH879
+ p.puLUMA_24x32.luma_hps = interp_horiz_ps_neon<8, 24, 32>;880
+#endif // !defined(__APPLE__)881
+882
+#if !defined(__APPLE__)883
+ p.puLUMA_32x8.luma_hpp = interp_horiz_pp_neon<8, 32, 8>;884
+ p.puLUMA_32x16.luma_hpp = interp_horiz_pp_neon<8, 32, 16>;885
+ p.puLUMA_32x24.luma_hpp = interp_horiz_pp_neon<8, 32, 24>;886
+ p.puLUMA_32x32.luma_hpp = interp_horiz_pp_neon<8, 32, 32>;887
+ p.puLUMA_32x64.luma_hpp = interp_horiz_pp_neon<8, 32, 64>;888
+ p.puLUMA_48x64.luma_hpp = interp_horiz_pp_neon<8, 48, 64>;889
+ p.puLUMA_64x16.luma_hpp = interp_horiz_pp_neon<8, 64, 16>;890
+ p.puLUMA_64x32.luma_hpp = interp_horiz_pp_neon<8, 64, 32>;891
+ p.puLUMA_64x48.luma_hpp = interp_horiz_pp_neon<8, 64, 48>;892
+ p.puLUMA_64x64.luma_hpp = interp_horiz_pp_neon<8, 64, 64>;893
+894
+ LUMA_FILTER(8, 4);895
+ LUMA_FILTER(8, 8);896
+ LUMA_FILTER(8, 16);897
+ LUMA_FILTER(8, 32);898
+ LUMA_FILTER(24, 32);899
+900
+ LUMA_FILTER(16, 32);901
+ LUMA_FILTER(32, 16);902
+ LUMA_FILTER(32, 24);903
+ LUMA_FILTER(32, 32);904
+ LUMA_FILTER(32, 64);905
+ LUMA_FILTER(48, 64);906
+ LUMA_FILTER(64, 32);907
+ LUMA_FILTER(64, 48);908
+ LUMA_FILTER(64, 64);909
+ 910
+ CHROMA_FILTER_420(24, 32);911
+ 912
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.filter_hpp = interp_horiz_pp_neon<4, 32, 8>;913
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.filter_hpp = interp_horiz_pp_neon<4, 32, 16>;914
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.filter_hpp = interp_horiz_pp_neon<4, 32, 24>;915
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.filter_hpp = interp_horiz_pp_neon<4, 32, 32>;916
+ 917
+ CHROMA_FILTER_422(24, 64);918
+ 919
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.filter_hpp = interp_horiz_pp_neon<4, 32, 16>;920
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.filter_hpp = interp_horiz_pp_neon<4, 32, 32>;921
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.filter_hpp = interp_horiz_pp_neon<4, 32, 48>;922
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.filter_hpp = interp_horiz_pp_neon<4, 32, 64>;923
+ 924
+ CHROMA_FILTER_444(24, 32);925
+ 926
+ p.chromaX265_CSP_I444.puLUMA_32x8.filter_hpp = interp_horiz_pp_neon<4, 32, 8>;927
+ p.chromaX265_CSP_I444.puLUMA_32x16.filter_hpp = interp_horiz_pp_neon<4, 32, 16>;928
+ p.chromaX265_CSP_I444.puLUMA_32x24.filter_hpp = interp_horiz_pp_neon<4, 32, 24>;929
+ p.chromaX265_CSP_I444.puLUMA_32x32.filter_hpp = interp_horiz_pp_neon<4, 32, 32>;930
+ p.chromaX265_CSP_I444.puLUMA_32x64.filter_hpp = interp_horiz_pp_neon<4, 32, 64>;931
+ p.chromaX265_CSP_I444.puLUMA_48x64.filter_hpp = interp_horiz_pp_neon<4, 48, 64>;932
+ p.chromaX265_CSP_I444.puLUMA_64x16.filter_hpp = interp_horiz_pp_neon<4, 64, 16>;933
+ p.chromaX265_CSP_I444.puLUMA_64x32.filter_hpp = interp_horiz_pp_neon<4, 64, 32>;934
+ p.chromaX265_CSP_I444.puLUMA_64x48.filter_hpp = interp_horiz_pp_neon<4, 64, 48>;935
+ p.chromaX265_CSP_I444.puLUMA_64x64.filter_hpp = interp_horiz_pp_neon<4, 64, 64>;936
+ 937
+ p.chromaX265_CSP_I444.puLUMA_16x4.filter_vss = interp_vert_ss_neon<4, 16, 4>;938
+ p.chromaX265_CSP_I444.puLUMA_16x8.filter_vss = interp_vert_ss_neon<4, 16, 8>;939
+ p.chromaX265_CSP_I444.puLUMA_16x12.filter_vss = interp_vert_ss_neon<4, 16, 12>;940
+ p.chromaX265_CSP_I444.puLUMA_16x16.filter_vss = interp_vert_ss_neon<4, 16, 16>;941
+ p.chromaX265_CSP_I444.puLUMA_16x32.filter_vss = interp_vert_ss_neon<4, 16, 32>;942
+ p.chromaX265_CSP_I444.puLUMA_16x64.filter_vss = interp_vert_ss_neon<4, 16, 64>;943
+ p.chromaX265_CSP_I444.puLUMA_32x8.filter_vss = interp_vert_ss_neon<4, 32, 8>;944
+ p.chromaX265_CSP_I444.puLUMA_32x16.filter_vss = interp_vert_ss_neon<4, 32, 16>;945
+ p.chromaX265_CSP_I444.puLUMA_32x24.filter_vss = interp_vert_ss_neon<4, 32, 24>;946
+ p.chromaX265_CSP_I444.puLUMA_32x32.filter_vss = interp_vert_ss_neon<4, 32, 32>;947
+ p.chromaX265_CSP_I444.puLUMA_32x64.filter_vss = interp_vert_ss_neon<4, 32, 64>;948
+#endif // !defined(__APPLE__)949
+950
+ CHROMA_FILTER_420(8, 2);951
+ CHROMA_FILTER_420(8, 4);952
+ CHROMA_FILTER_420(8, 6);953
+ CHROMA_FILTER_420(8, 8);954
+ CHROMA_FILTER_420(8, 16);955
+ CHROMA_FILTER_420(8, 32);956
+ 957
+ CHROMA_FILTER_422(8, 4);958
+ CHROMA_FILTER_422(8, 8);959
+ CHROMA_FILTER_422(8, 12);960
+ CHROMA_FILTER_422(8, 16);961
+ CHROMA_FILTER_422(8, 32);962
+ CHROMA_FILTER_422(8, 64);963
+ 964
+ CHROMA_FILTER_444(8, 4);965
+ CHROMA_FILTER_444(8, 8);966
+ CHROMA_FILTER_444(8, 16);967
+ CHROMA_FILTER_444(8, 32);968
+ 969
+#if defined(__APPLE__)970
+ CHROMA_FILTER_420(16, 4);971
+ CHROMA_FILTER_420(16, 8);972
+ CHROMA_FILTER_420(16, 12);973
+ CHROMA_FILTER_420(16, 16);974
+ CHROMA_FILTER_420(16, 32);975
+976
+ CHROMA_FILTER_422(16, 8);977
+ CHROMA_FILTER_422(16, 16);978
+ CHROMA_FILTER_422(16, 24);979
+ CHROMA_FILTER_422(16, 32);980
+ CHROMA_FILTER_422(16, 64);981
+ 982
+ CHROMA_FILTER_444(16, 4);983
+ CHROMA_FILTER_444(16, 8);984
+ CHROMA_FILTER_444(16, 12);985
+ CHROMA_FILTER_444(16, 16);986
+ CHROMA_FILTER_444(16, 32);987
+ CHROMA_FILTER_444(16, 64);988
+#endif // defined(__APPLE__)989
+}990
+991
+};992
+993
+994
+#endif995
+996
+997
x265_3.6.tar.gz/source/common/aarch64/filter-prim.h
Added
23
1
2
+#ifndef _FILTER_PRIM_ARM64_H__3
+#define _FILTER_PRIM_ARM64_H__4
+5
+6
+#include "common.h"7
+#include "slicetype.h" // LOWRES_COST_MASK8
+#include "primitives.h"9
+#include "x265.h"10
+11
+12
+namespace X265_NS13
+{14
+15
+16
+void setupFilterPrimitives_neon(EncoderPrimitives &p);17
+18
+};19
+20
+21
+#endif22
+23
x265_3.6.tar.gz/source/common/aarch64/fun-decls.h
Added
258
1
2
+/*****************************************************************************3
+ * Copyright (C) 2021 MulticoreWare, Inc4
+ *5
+ * Authors: Sebastian Pop <spop@amazon.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#define FUNCDEF_TU(ret, name, cpu, ...) \26
+ ret PFX(name ## _4x4_ ## cpu(__VA_ARGS__)); \27
+ ret PFX(name ## _8x8_ ## cpu(__VA_ARGS__)); \28
+ ret PFX(name ## _16x16_ ## cpu(__VA_ARGS__)); \29
+ ret PFX(name ## _32x32_ ## cpu(__VA_ARGS__)); \30
+ ret PFX(name ## _64x64_ ## cpu(__VA_ARGS__))31
+32
+#define FUNCDEF_TU_S(ret, name, cpu, ...) \33
+ ret PFX(name ## _4_ ## cpu(__VA_ARGS__)); \34
+ ret PFX(name ## _8_ ## cpu(__VA_ARGS__)); \35
+ ret PFX(name ## _16_ ## cpu(__VA_ARGS__)); \36
+ ret PFX(name ## _32_ ## cpu(__VA_ARGS__)); \37
+ ret PFX(name ## _64_ ## cpu(__VA_ARGS__))38
+39
+#define FUNCDEF_TU_S2(ret, name, cpu, ...) \40
+ ret PFX(name ## 4_ ## cpu(__VA_ARGS__)); \41
+ ret PFX(name ## 8_ ## cpu(__VA_ARGS__)); \42
+ ret PFX(name ## 16_ ## cpu(__VA_ARGS__)); \43
+ ret PFX(name ## 32_ ## cpu(__VA_ARGS__)); \44
+ ret PFX(name ## 64_ ## cpu(__VA_ARGS__))45
+46
+#define FUNCDEF_PU(ret, name, cpu, ...) \47
+ ret PFX(name ## _4x4_ ## cpu)(__VA_ARGS__); \48
+ ret PFX(name ## _8x8_ ## cpu)(__VA_ARGS__); \49
+ ret PFX(name ## _16x16_ ## cpu)(__VA_ARGS__); \50
+ ret PFX(name ## _32x32_ ## cpu)(__VA_ARGS__); \51
+ ret PFX(name ## _64x64_ ## cpu)(__VA_ARGS__); \52
+ ret PFX(name ## _8x4_ ## cpu)(__VA_ARGS__); \53
+ ret PFX(name ## _4x8_ ## cpu)(__VA_ARGS__); \54
+ ret PFX(name ## _16x8_ ## cpu)(__VA_ARGS__); \55
+ ret PFX(name ## _8x16_ ## cpu)(__VA_ARGS__); \56
+ ret PFX(name ## _16x32_ ## cpu)(__VA_ARGS__); \57
+ ret PFX(name ## _32x16_ ## cpu)(__VA_ARGS__); \58
+ ret PFX(name ## _64x32_ ## cpu)(__VA_ARGS__); \59
+ ret PFX(name ## _32x64_ ## cpu)(__VA_ARGS__); \60
+ ret PFX(name ## _16x12_ ## cpu)(__VA_ARGS__); \61
+ ret PFX(name ## _12x16_ ## cpu)(__VA_ARGS__); \62
+ ret PFX(name ## _16x4_ ## cpu)(__VA_ARGS__); \63
+ ret PFX(name ## _4x16_ ## cpu)(__VA_ARGS__); \64
+ ret PFX(name ## _32x24_ ## cpu)(__VA_ARGS__); \65
+ ret PFX(name ## _24x32_ ## cpu)(__VA_ARGS__); \66
+ ret PFX(name ## _32x8_ ## cpu)(__VA_ARGS__); \67
+ ret PFX(name ## _8x32_ ## cpu)(__VA_ARGS__); \68
+ ret PFX(name ## _64x48_ ## cpu)(__VA_ARGS__); \69
+ ret PFX(name ## _48x64_ ## cpu)(__VA_ARGS__); \70
+ ret PFX(name ## _64x16_ ## cpu)(__VA_ARGS__); \71
+ ret PFX(name ## _16x64_ ## cpu)(__VA_ARGS__)72
+73
+#define FUNCDEF_CHROMA_PU(ret, name, cpu, ...) \74
+ FUNCDEF_PU(ret, name, cpu, __VA_ARGS__); \75
+ ret PFX(name ## _4x2_ ## cpu)(__VA_ARGS__); \76
+ ret PFX(name ## _4x4_ ## cpu)(__VA_ARGS__); \77
+ ret PFX(name ## _2x4_ ## cpu)(__VA_ARGS__); \78
+ ret PFX(name ## _8x2_ ## cpu)(__VA_ARGS__); \79
+ ret PFX(name ## _2x8_ ## cpu)(__VA_ARGS__); \80
+ ret PFX(name ## _8x6_ ## cpu)(__VA_ARGS__); \81
+ ret PFX(name ## _6x8_ ## cpu)(__VA_ARGS__); \82
+ ret PFX(name ## _8x12_ ## cpu)(__VA_ARGS__); \83
+ ret PFX(name ## _12x8_ ## cpu)(__VA_ARGS__); \84
+ ret PFX(name ## _6x16_ ## cpu)(__VA_ARGS__); \85
+ ret PFX(name ## _16x6_ ## cpu)(__VA_ARGS__); \86
+ ret PFX(name ## _2x16_ ## cpu)(__VA_ARGS__); \87
+ ret PFX(name ## _16x2_ ## cpu)(__VA_ARGS__); \88
+ ret PFX(name ## _4x12_ ## cpu)(__VA_ARGS__); \89
+ ret PFX(name ## _12x4_ ## cpu)(__VA_ARGS__); \90
+ ret PFX(name ## _32x12_ ## cpu)(__VA_ARGS__); \91
+ ret PFX(name ## _12x32_ ## cpu)(__VA_ARGS__); \92
+ ret PFX(name ## _32x4_ ## cpu)(__VA_ARGS__); \93
+ ret PFX(name ## _4x32_ ## cpu)(__VA_ARGS__); \94
+ ret PFX(name ## _32x48_ ## cpu)(__VA_ARGS__); \95
+ ret PFX(name ## _48x32_ ## cpu)(__VA_ARGS__); \96
+ ret PFX(name ## _16x24_ ## cpu)(__VA_ARGS__); \97
+ ret PFX(name ## _24x16_ ## cpu)(__VA_ARGS__); \98
+ ret PFX(name ## _8x64_ ## cpu)(__VA_ARGS__); \99
+ ret PFX(name ## _64x8_ ## cpu)(__VA_ARGS__); \100
+ ret PFX(name ## _64x24_ ## cpu)(__VA_ARGS__); \101
+ ret PFX(name ## _24x64_ ## cpu)(__VA_ARGS__);102
+103
+#define DECLS(cpu) \104
+ FUNCDEF_TU(void, cpy2Dto1D_shl, cpu, int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); \105
+ FUNCDEF_TU(void, cpy2Dto1D_shr, cpu, int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); \106
+ FUNCDEF_TU(void, cpy1Dto2D_shl, cpu, int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); \107
+ FUNCDEF_TU(void, cpy1Dto2D_shl_aligned, cpu, int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); \108
+ FUNCDEF_TU(void, cpy1Dto2D_shr, cpu, int16_t* dst, const int16_t* src, intptr_t srcStride, int shift); \109
+ FUNCDEF_TU_S(uint32_t, copy_cnt, cpu, int16_t* dst, const int16_t* src, intptr_t srcStride); \110
+ FUNCDEF_TU_S(int, count_nonzero, cpu, const int16_t* quantCoeff); \111
+ FUNCDEF_TU(void, blockfill_s, cpu, int16_t* dst, intptr_t dstride, int16_t val); \112
+ FUNCDEF_TU(void, blockfill_s_aligned, cpu, int16_t* dst, intptr_t dstride, int16_t val); \113
+ FUNCDEF_CHROMA_PU(void, blockcopy_ss, cpu, int16_t* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); \114
+ FUNCDEF_CHROMA_PU(void, blockcopy_pp, cpu, pixel* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride); \115
+ FUNCDEF_PU(void, blockcopy_sp, cpu, pixel* dst, intptr_t dstStride, const int16_t* src, intptr_t srcStride); \116
+ FUNCDEF_PU(void, blockcopy_ps, cpu, int16_t* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride); \117
+ FUNCDEF_PU(void, interp_8tap_horiz_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \118
+ FUNCDEF_PU(void, interp_8tap_horiz_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt); \119
+ FUNCDEF_PU(void, interp_8tap_vert_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \120
+ FUNCDEF_PU(void, interp_8tap_vert_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx); \121
+ FUNCDEF_PU(void, interp_8tap_vert_sp, cpu, const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \122
+ FUNCDEF_PU(void, interp_8tap_vert_ss, cpu, const int16_t* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx); \123
+ FUNCDEF_PU(void, interp_8tap_hv_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int idxX, int idxY); \124
+ FUNCDEF_CHROMA_PU(void, filterPixelToShort, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride); \125
+ FUNCDEF_CHROMA_PU(void, filterPixelToShort_aligned, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride); \126
+ FUNCDEF_CHROMA_PU(void, interp_horiz_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \127
+ FUNCDEF_CHROMA_PU(void, interp_4tap_horiz_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \128
+ FUNCDEF_CHROMA_PU(void, interp_horiz_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt); \129
+ FUNCDEF_CHROMA_PU(void, interp_4tap_horiz_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt); \130
+ FUNCDEF_CHROMA_PU(void, interp_4tap_vert_pp, cpu, const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \131
+ FUNCDEF_CHROMA_PU(void, interp_4tap_vert_ps, cpu, const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx); \132
+ FUNCDEF_CHROMA_PU(void, interp_4tap_vert_sp, cpu, const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx); \133
+ FUNCDEF_CHROMA_PU(void, interp_4tap_vert_ss, cpu, const int16_t* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx); \134
+ FUNCDEF_CHROMA_PU(void, addAvg, cpu, const int16_t*, const int16_t*, pixel*, intptr_t, intptr_t, intptr_t); \135
+ FUNCDEF_CHROMA_PU(void, addAvg_aligned, cpu, const int16_t*, const int16_t*, pixel*, intptr_t, intptr_t, intptr_t); \136
+ FUNCDEF_PU(void, pixel_avg_pp, cpu, pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int); \137
+ FUNCDEF_PU(void, pixel_avg_pp_aligned, cpu, pixel* dst, intptr_t dstride, const pixel* src0, intptr_t sstride0, const pixel* src1, intptr_t sstride1, int); \138
+ FUNCDEF_PU(void, sad_x3, cpu, const pixel*, const pixel*, const pixel*, const pixel*, intptr_t, int32_t*); \139
+ FUNCDEF_PU(void, sad_x4, cpu, const pixel*, const pixel*, const pixel*, const pixel*, const pixel*, intptr_t, int32_t*); \140
+ FUNCDEF_CHROMA_PU(int, pixel_sad, cpu, const pixel*, intptr_t, const pixel*, intptr_t); \141
+ FUNCDEF_CHROMA_PU(sse_t, pixel_ssd_s, cpu, const int16_t*, intptr_t); \142
+ FUNCDEF_CHROMA_PU(sse_t, pixel_ssd_s_aligned, cpu, const int16_t*, intptr_t); \143
+ FUNCDEF_TU_S(sse_t, pixel_ssd_s, cpu, const int16_t*, intptr_t); \144
+ FUNCDEF_TU_S(sse_t, pixel_ssd_s_aligned, cpu, const int16_t*, intptr_t); \145
+ FUNCDEF_PU(sse_t, pixel_sse_pp, cpu, const pixel*, intptr_t, const pixel*, intptr_t); \146
+ FUNCDEF_CHROMA_PU(sse_t, pixel_sse_ss, cpu, const int16_t*, intptr_t, const int16_t*, intptr_t); \147
+ FUNCDEF_PU(void, pixel_sub_ps, cpu, int16_t* a, intptr_t dstride, const pixel* b0, const pixel* b1, intptr_t sstride0, intptr_t sstride1); \148
+ FUNCDEF_PU(void, pixel_add_ps, cpu, pixel* a, intptr_t dstride, const pixel* b0, const int16_t* b1, intptr_t sstride0, intptr_t sstride1); \149
+ FUNCDEF_PU(void, pixel_add_ps_aligned, cpu, pixel* a, intptr_t dstride, const pixel* b0, const int16_t* b1, intptr_t sstride0, intptr_t sstride1); \150
+ FUNCDEF_CHROMA_PU(int, pixel_satd, cpu, const pixel*, intptr_t, const pixel*, intptr_t); \151
+ FUNCDEF_TU_S2(void, ssimDist, cpu, const pixel *fenc, uint32_t fStride, const pixel *recon, intptr_t rstride, uint64_t *ssBlock, int shift, uint64_t *ac_k); \152
+ FUNCDEF_TU_S2(void, normFact, cpu, const pixel *src, uint32_t blockSize, int shift, uint64_t *z_k)153
+154
+DECLS(neon);155
+DECLS(sve);156
+DECLS(sve2);157
+158
+159
+void x265_pixel_planecopy_cp_neon(const uint8_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int width, int height, int shift);160
+161
+uint64_t x265_pixel_var_8x8_neon(const pixel* pix, intptr_t stride);162
+uint64_t x265_pixel_var_16x16_neon(const pixel* pix, intptr_t stride);163
+uint64_t x265_pixel_var_32x32_neon(const pixel* pix, intptr_t stride);164
+uint64_t x265_pixel_var_64x64_neon(const pixel* pix, intptr_t stride);165
+166
+void x265_getResidual4_neon(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride);167
+void x265_getResidual8_neon(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride);168
+void x265_getResidual16_neon(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride);169
+void x265_getResidual32_neon(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride);170
+171
+void x265_scale1D_128to64_neon(pixel *dst, const pixel *src);172
+void x265_scale2D_64to32_neon(pixel* dst, const pixel* src, intptr_t stride);173
+174
+int x265_pixel_satd_4x4_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);175
+int x265_pixel_satd_4x8_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);176
+int x265_pixel_satd_4x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);177
+int x265_pixel_satd_4x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);178
+int x265_pixel_satd_8x4_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);179
+int x265_pixel_satd_8x8_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);180
+int x265_pixel_satd_8x12_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);181
+int x265_pixel_satd_8x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);182
+int x265_pixel_satd_8x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);183
+int x265_pixel_satd_8x64_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);184
+int x265_pixel_satd_12x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);185
+int x265_pixel_satd_12x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);186
+int x265_pixel_satd_16x4_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);187
+int x265_pixel_satd_16x8_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);188
+int x265_pixel_satd_16x12_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);189
+int x265_pixel_satd_16x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);190
+int x265_pixel_satd_16x24_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);191
+int x265_pixel_satd_16x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);192
+int x265_pixel_satd_16x64_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);193
+int x265_pixel_satd_24x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);194
+int x265_pixel_satd_24x64_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);195
+int x265_pixel_satd_32x8_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);196
+int x265_pixel_satd_32x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);197
+int x265_pixel_satd_32x24_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);198
+int x265_pixel_satd_32x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);199
+int x265_pixel_satd_32x48_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);200
+int x265_pixel_satd_32x64_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);201
+int x265_pixel_satd_48x64_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);202
+int x265_pixel_satd_64x16_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);203
+int x265_pixel_satd_64x32_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);204
+int x265_pixel_satd_64x48_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);205
+int x265_pixel_satd_64x64_neon(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);206
+207
+int x265_pixel_sa8d_8x8_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);208
+int x265_pixel_sa8d_8x16_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);209
+int x265_pixel_sa8d_16x16_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);210
+int x265_pixel_sa8d_16x32_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);211
+int x265_pixel_sa8d_32x32_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);212
+int x265_pixel_sa8d_32x64_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);213
+int x265_pixel_sa8d_64x64_neon(const pixel* pix1, intptr_t i_pix1, const pixel* pix2, intptr_t i_pix2);214
+215
+uint32_t PFX(quant_neon)(const int16_t* coef, const int32_t* quantCoeff, int32_t* deltaU, int16_t* qCoef, int qBits, int add, int numCoeff);216
+uint32_t PFX(nquant_neon)(const int16_t* coef, const int32_t* quantCoeff, int16_t* qCoef, int qBits, int add, int numCoeff);217
+218
+void x265_dequant_scaling_neon(const int16_t* quantCoef, const int32_t* deQuantCoef, int16_t* coef, int num, int per, int shift);219
+void x265_dequant_normal_neon(const int16_t* quantCoef, int16_t* coef, int num, int scale, int shift);220
+221
+void x265_ssim_4x4x2_core_neon(const pixel* pix1, intptr_t stride1, const pixel* pix2, intptr_t stride2, int sums24);222
+223
+int PFX(psyCost_4x4_neon)(const pixel* source, intptr_t sstride, const pixel* recon, intptr_t rstride);224
+int PFX(psyCost_8x8_neon)(const pixel* source, intptr_t sstride, const pixel* recon, intptr_t rstride);225
+void PFX(weight_pp_neon)(const pixel* src, pixel* dst, intptr_t stride, int width, int height, int w0, int round, int shift, int offset);226
+void PFX(weight_sp_neon)(const int16_t* src, pixel* dst, intptr_t srcStride, intptr_t dstStride, int width, int height, int w0, int round, int shift, int offset);227
+int PFX(scanPosLast_neon)(const uint16_t *scan, const coeff_t *coeff, uint16_t *coeffSign, uint16_t *coeffFlag, uint8_t *coeffNum, int numSig, const uint16_t* scanCG4x4, const int trSize);228
+uint32_t PFX(costCoeffNxN_neon)(const uint16_t *scan, const coeff_t *coeff, intptr_t trSize, uint16_t *absCoeff, const uint8_t *tabSigCtx, uint32_t scanFlagMask, uint8_t *baseCtx, int offset, int scanPosSigOff, int subPosBase);229
+230
+uint64_t x265_pixel_var_8x8_sve2(const pixel* pix, intptr_t stride);231
+uint64_t x265_pixel_var_16x16_sve2(const pixel* pix, intptr_t stride);232
+uint64_t x265_pixel_var_32x32_sve2(const pixel* pix, intptr_t stride);233
+uint64_t x265_pixel_var_64x64_sve2(const pixel* pix, intptr_t stride);234
+235
+void x265_getResidual16_sve2(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride);236
+void x265_getResidual32_sve2(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride);237
+238
+void x265_scale1D_128to64_sve2(pixel *dst, const pixel *src);239
+void x265_scale2D_64to32_sve2(pixel* dst, const pixel* src, intptr_t stride);240
+241
+int x265_pixel_satd_4x4_sve(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);242
+int x265_pixel_satd_8x4_sve(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);243
+int x265_pixel_satd_8x12_sve(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);244
+int x265_pixel_satd_32x16_sve(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);245
+int x265_pixel_satd_32x32_sve(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);246
+int x265_pixel_satd_64x48_sve(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2);247
+248
+uint32_t PFX(quant_sve)(const int16_t* coef, const int32_t* quantCoeff, int32_t* deltaU, int16_t* qCoef, int qBits, int add, int numCoeff);249
+250
+void x265_dequant_scaling_sve2(const int16_t* quantCoef, const int32_t* deQuantCoef, int16_t* coef, int num, int per, int shift);251
+void x265_dequant_normal_sve2(const int16_t* quantCoef, int16_t* coef, int num, int scale, int shift);252
+253
+void x265_ssim_4x4x2_core_sve2(const pixel* pix1, intptr_t stride1, const pixel* pix2, intptr_t stride2, int sums24);254
+255
+int PFX(psyCost_8x8_sve2)(const pixel* source, intptr_t sstride, const pixel* recon, intptr_t rstride);256
+void PFX(weight_sp_sve2)(const int16_t* src, pixel* dst, intptr_t srcStride, intptr_t dstStride, int width, int height, int w0, int round, int shift, int offset);257
+int PFX(scanPosLast_sve2)(const uint16_t *scan, const coeff_t *coeff, uint16_t *coeffSign, uint16_t *coeffFlag, uint8_t *coeffNum, int numSig, const uint16_t* scanCG4x4, const int trSize);258
x265_3.6.tar.gz/source/common/aarch64/intrapred-prim.cpp
Added
267
1
2
+#include "common.h"3
+#include "primitives.h"4
+5
+6
+#if 17
+#include "arm64-utils.h"8
+#include <arm_neon.h>9
+10
+using namespace X265_NS;11
+12
+namespace13
+{14
+15
+16
+17
+template<int width>18
+void intra_pred_ang_neon(pixel *dst, intptr_t dstStride, const pixel *srcPix0, int dirMode, int bFilter)19
+{20
+ int width2 = width << 1;21
+ // Flip the neighbours in the horizontal case.22
+ int horMode = dirMode < 18;23
+ pixel neighbourBuf129;24
+ const pixel *srcPix = srcPix0;25
+26
+ if (horMode)27
+ {28
+ neighbourBuf0 = srcPix0;29
+ //for (int i = 0; i < width << 1; i++)30
+ //{31
+ // neighbourBuf1 + i = srcPixwidth2 + 1 + i;32
+ // neighbourBufwidth2 + 1 + i = srcPix1 + i;33
+ //}34
+ memcpy(&neighbourBuf1, &srcPixwidth2 + 1, sizeof(pixel) * (width << 1));35
+ memcpy(&neighbourBufwidth2 + 1, &srcPix1, sizeof(pixel) * (width << 1));36
+ srcPix = neighbourBuf;37
+ }38
+39
+ // Intra prediction angle and inverse angle tables.40
+ const int8_t angleTable17 = { -32, -26, -21, -17, -13, -9, -5, -2, 0, 2, 5, 9, 13, 17, 21, 26, 32 };41
+ const int16_t invAngleTable8 = { 4096, 1638, 910, 630, 482, 390, 315, 256 };42
+43
+ // Get the prediction angle.44
+ int angleOffset = horMode ? 10 - dirMode : dirMode - 26;45
+ int angle = angleTable8 + angleOffset;46
+47
+ // Vertical Prediction.48
+ if (!angle)49
+ {50
+ for (int y = 0; y < width; y++)51
+ {52
+ memcpy(&dsty * dstStride, srcPix + 1, sizeof(pixel)*width);53
+ }54
+ if (bFilter)55
+ {56
+ int topLeft = srcPix0, top = srcPix1;57
+ for (int y = 0; y < width; y++)58
+ {59
+ dsty * dstStride = x265_clip((int16_t)(top + ((srcPixwidth2 + 1 + y - topLeft) >> 1)));60
+ }61
+ }62
+ }63
+ else // Angular prediction.64
+ {65
+ // Get the reference pixels. The reference base is the first pixel to the top (neighbourBuf1).66
+ pixel refBuf64;67
+ const pixel *ref;68
+69
+ // Use the projected left neighbours and the top neighbours.70
+ if (angle < 0)71
+ {72
+ // Number of neighbours projected.73
+ int nbProjected = -((width * angle) >> 5) - 1;74
+ pixel *ref_pix = refBuf + nbProjected + 1;75
+76
+ // Project the neighbours.77
+ int invAngle = invAngleTable- angleOffset - 1;78
+ int invAngleSum = 128;79
+ for (int i = 0; i < nbProjected; i++)80
+ {81
+ invAngleSum += invAngle;82
+ ref_pix- 2 - i = srcPixwidth2 + (invAngleSum >> 8);83
+ }84
+85
+ // Copy the top-left and top pixels.86
+ //for (int i = 0; i < width + 1; i++)87
+ //ref_pix-1 + i = srcPixi;88
+89
+ memcpy(&ref_pix-1, srcPix, (width + 1)*sizeof(pixel));90
+ ref = ref_pix;91
+ }92
+ else // Use the top and top-right neighbours.93
+ {94
+ ref = srcPix + 1;95
+ }96
+97
+ // Pass every row.98
+ int angleSum = 0;99
+ for (int y = 0; y < width; y++)100
+ {101
+ angleSum += angle;102
+ int offset = angleSum >> 5;103
+ int fraction = angleSum & 31;104
+105
+ if (fraction) // Interpolate106
+ {107
+ if (width >= 8 && sizeof(pixel) == 1)108
+ {109
+ const int16x8_t f0 = vdupq_n_s16(32 - fraction);110
+ const int16x8_t f1 = vdupq_n_s16(fraction);111
+ for (int x = 0; x < width; x += 8)112
+ {113
+ uint8x8_t in0 = *(uint8x8_t *)&refoffset + x;114
+ uint8x8_t in1 = *(uint8x8_t *)&refoffset + x + 1;115
+ int16x8_t lo = vmlaq_s16(vdupq_n_s16(16), vmovl_u8(in0), f0);116
+ lo = vmlaq_s16(lo, vmovl_u8(in1), f1);117
+ lo = vshrq_n_s16(lo, 5);118
+ *(uint8x8_t *)&dsty * dstStride + x = vmovn_u16(lo);119
+ }120
+ }121
+ else if (width >= 4 && sizeof(pixel) == 2)122
+ {123
+ const int32x4_t f0 = vdupq_n_s32(32 - fraction);124
+ const int32x4_t f1 = vdupq_n_s32(fraction);125
+ for (int x = 0; x < width; x += 4)126
+ {127
+ uint16x4_t in0 = *(uint16x4_t *)&refoffset + x;128
+ uint16x4_t in1 = *(uint16x4_t *)&refoffset + x + 1;129
+ int32x4_t lo = vmlaq_s32(vdupq_n_s32(16), vmovl_u16(in0), f0);130
+ lo = vmlaq_s32(lo, vmovl_u16(in1), f1);131
+ lo = vshrq_n_s32(lo, 5);132
+ *(uint16x4_t *)&dsty * dstStride + x = vmovn_u32(lo);133
+ }134
+ }135
+ else136
+ {137
+ for (int x = 0; x < width; x++)138
+ {139
+ dsty * dstStride + x = (pixel)(((32 - fraction) * refoffset + x + fraction * refoffset + x + 1 + 16) >> 5);140
+ }141
+ }142
+ }143
+ else // Copy.144
+ {145
+ memcpy(&dsty * dstStride, &refoffset, sizeof(pixel)*width);146
+ }147
+ }148
+ }149
+150
+ // Flip for horizontal.151
+ if (horMode)152
+ {153
+ if (width == 8)154
+ {155
+ transpose8x8(dst, dst, dstStride, dstStride);156
+ }157
+ else if (width == 16)158
+ {159
+ transpose16x16(dst, dst, dstStride, dstStride);160
+ }161
+ else if (width == 32)162
+ {163
+ transpose32x32(dst, dst, dstStride, dstStride);164
+ }165
+ else166
+ {167
+ for (int y = 0; y < width - 1; y++)168
+ {169
+ for (int x = y + 1; x < width; x++)170
+ {171
+ pixel tmp = dsty * dstStride + x;172
+ dsty * dstStride + x = dstx * dstStride + y;173
+ dstx * dstStride + y = tmp;174
+ }175
+ }176
+ }177
+ }178
+}179
+180
+template<int log2Size>181
+void all_angs_pred_neon(pixel *dest, pixel *refPix, pixel *filtPix, int bLuma)182
+{183
+ const int size = 1 << log2Size;184
+ for (int mode = 2; mode <= 34; mode++)185
+ {186
+ pixel *srcPix = (g_intraFilterFlagsmode & size ? filtPix : refPix);187
+ pixel *out = dest + ((mode - 2) << (log2Size * 2));188
+189
+ intra_pred_ang_neon<size>(out, size, srcPix, mode, bLuma);190
+191
+ // Optimize code don't flip buffer192
+ bool modeHor = (mode < 18);193
+194
+ // transpose the block if this is a horizontal mode195
+ if (modeHor)196
+ {197
+ if (size == 8)198
+ {199
+ transpose8x8(out, out, size, size);200
+ }201
+ else if (size == 16)202
+ {203
+ transpose16x16(out, out, size, size);204
+ }205
+ else if (size == 32)206
+ {207
+ transpose32x32(out, out, size, size);208
+ }209
+ else210
+ {211
+ for (int k = 0; k < size - 1; k++)212
+ {213
+ for (int l = k + 1; l < size; l++)214
+ {215
+ pixel tmp = outk * size + l;216
+ outk * size + l = outl * size + k;217
+ outl * size + k = tmp;218
+ }219
+ }220
+ }221
+ }222
+ }223
+}224
+}225
+226
+namespace X265_NS227
+{228
+// x265 private namespace229
+230
+void setupIntraPrimitives_neon(EncoderPrimitives &p)231
+{232
+ for (int i = 2; i < NUM_INTRA_MODE; i++)233
+ {234
+ p.cuBLOCK_8x8.intra_predi = intra_pred_ang_neon<8>;235
+ p.cuBLOCK_16x16.intra_predi = intra_pred_ang_neon<16>;236
+ p.cuBLOCK_32x32.intra_predi = intra_pred_ang_neon<32>;237
+ }238
+ p.cuBLOCK_4x4.intra_pred2 = intra_pred_ang_neon<4>;239
+ p.cuBLOCK_4x4.intra_pred10 = intra_pred_ang_neon<4>;240
+ p.cuBLOCK_4x4.intra_pred18 = intra_pred_ang_neon<4>;241
+ p.cuBLOCK_4x4.intra_pred26 = intra_pred_ang_neon<4>;242
+ p.cuBLOCK_4x4.intra_pred34 = intra_pred_ang_neon<4>;243
+244
+ p.cuBLOCK_4x4.intra_pred_allangs = all_angs_pred_neon<2>;245
+ p.cuBLOCK_8x8.intra_pred_allangs = all_angs_pred_neon<3>;246
+ p.cuBLOCK_16x16.intra_pred_allangs = all_angs_pred_neon<4>;247
+ p.cuBLOCK_32x32.intra_pred_allangs = all_angs_pred_neon<5>;248
+}249
+250
+}251
+252
+253
+254
+#else255
+256
+namespace X265_NS257
+{258
+// x265 private namespace259
+void setupIntraPrimitives_neon(EncoderPrimitives &p)260
+{}261
+}262
+263
+#endif264
+265
+266
+267
x265_3.6.tar.gz/source/common/aarch64/intrapred-prim.h
Added
17
1
2
+#ifndef INTRAPRED_PRIM_H__3
+4
+#if defined(__aarch64__)5
+6
+namespace X265_NS7
+{8
+// x265 private namespace9
+10
+void setupIntraPrimitives_neon(EncoderPrimitives &p);11
+}12
+13
+#endif14
+15
+#endif16
+17
x265_3.6.tar.gz/source/common/aarch64/ipfilter-common.S
Added
1438
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+// Macros below follow these conventions:29
+// - input data in registers: v0, v1, v2, v3, v4, v5, v6, v730
+// - constants in registers: v24, v25, v26, v27, v3131
+// - temporary registers: v16, v17, v18, v19, v20, v21, v22, v23, v28, v29, v30.32
+// - _32b macros output a result in v17.4s33
+// - _64b and _32b_1 macros output results in v17.4s, v18.4s34
+35
+#include "asm.S"36
+37
+.arch armv8-a38
+39
+#ifdef __APPLE__40
+.section __RODATA,__rodata41
+#else42
+.section .rodata43
+#endif44
+45
+.align 446
+47
+.macro vextin8 v48
+ ldp d6, d7, x11, #1649
+.if \v == 050
+ // qpel_filter_0 only uses values in v351
+ ext v3.8b, v6.8b, v7.8b, #452
+.else53
+.if \v != 354
+ ext v0.8b, v6.8b, v7.8b, #155
+.endif56
+ ext v1.8b, v6.8b, v7.8b, #257
+ ext v2.8b, v6.8b, v7.8b, #358
+ ext v3.8b, v6.8b, v7.8b, #459
+ ext v4.8b, v6.8b, v7.8b, #560
+ ext v5.8b, v6.8b, v7.8b, #661
+ ext v6.8b, v6.8b, v7.8b, #762
+.endif63
+.endm64
+65
+.macro vextin8_64 v66
+ ldp q6, q7, x11, #3267
+.if \v == 068
+ // qpel_filter_0 only uses values in v369
+ ext v3.16b, v6.16b, v7.16b, #470
+.else71
+.if \v != 372
+ // qpel_filter_3 does not use values in v073
+ ext v0.16b, v6.16b, v7.16b, #174
+.endif75
+ ext v1.16b, v6.16b, v7.16b, #276
+ ext v2.16b, v6.16b, v7.16b, #377
+ ext v3.16b, v6.16b, v7.16b, #478
+ ext v4.16b, v6.16b, v7.16b, #579
+ ext v5.16b, v6.16b, v7.16b, #680
+.if \v == 181
+ ext v6.16b, v6.16b, v7.16b, #782
+ // qpel_filter_1 does not use v783
+.else84
+ ext v16.16b, v6.16b, v7.16b, #785
+ ext v7.16b, v6.16b, v7.16b, #886
+ mov v6.16b, v16.16b87
+.endif88
+.endif89
+.endm90
+91
+.macro vextin8_chroma v92
+ ldp d6, d7, x11, #1693
+.if \v == 094
+ // qpel_filter_chroma_0 only uses values in v195
+ ext v1.8b, v6.8b, v7.8b, #296
+.else97
+ ext v0.8b, v6.8b, v7.8b, #198
+ ext v1.8b, v6.8b, v7.8b, #299
+ ext v2.8b, v6.8b, v7.8b, #3100
+ ext v3.8b, v6.8b, v7.8b, #4101
+.endif102
+.endm103
+104
+.macro vextin8_chroma_64 v105
+ ldp q16, q17, x11, #32106
+.if \v == 0107
+ // qpel_filter_chroma_0 only uses values in v1108
+ ext v1.16b, v16.16b, v17.16b, #2109
+.else110
+ ext v0.16b, v16.16b, v17.16b, #1111
+ ext v1.16b, v16.16b, v17.16b, #2112
+ ext v2.16b, v16.16b, v17.16b, #3113
+ ext v3.16b, v16.16b, v17.16b, #4114
+.endif115
+.endm116
+117
+.macro qpel_load_32b v118
+.if \v == 0119
+ add x6, x6, x11 // do not load 3 values that are not used in qpel_filter_0120
+ ld1 {v3.8b}, x6, x1121
+.elseif \v == 1 || \v == 2 || \v == 3122
+.if \v != 3 // not used in qpel_filter_3123
+ ld1 {v0.8b}, x6, x1124
+.else125
+ add x6, x6, x1126
+.endif127
+ ld1 {v1.8b}, x6, x1128
+ ld1 {v2.8b}, x6, x1129
+ ld1 {v3.8b}, x6, x1130
+ ld1 {v4.8b}, x6, x1131
+ ld1 {v5.8b}, x6, x1132
+.if \v != 1 // not used in qpel_filter_1133
+ ld1 {v6.8b}, x6, x1134
+ ld1 {v7.8b}, x6135
+.else136
+ ld1 {v6.8b}, x6137
+.endif138
+.endif139
+.endm140
+141
+.macro qpel_load_64b v142
+.if \v == 0143
+ add x6, x6, x11 // do not load 3 values that are not used in qpel_filter_0144
+ ld1 {v3.16b}, x6, x1145
+.elseif \v == 1 || \v == 2 || \v == 3146
+.if \v != 3 // not used in qpel_filter_3147
+ ld1 {v0.16b}, x6, x1148
+.else149
+ add x6, x6, x1150
+.endif151
+ ld1 {v1.16b}, x6, x1152
+ ld1 {v2.16b}, x6, x1153
+ ld1 {v3.16b}, x6, x1154
+ ld1 {v4.16b}, x6, x1155
+ ld1 {v5.16b}, x6, x1156
+.if \v != 1 // not used in qpel_filter_1157
+ ld1 {v6.16b}, x6, x1158
+ ld1 {v7.16b}, x6159
+.else160
+ ld1 {v6.16b}, x6161
+.endif162
+.endif163
+.endm164
+165
+.macro qpel_chroma_load_32b v166
+.if \v == 0167
+ // qpel_filter_chroma_0 only uses values in v1168
+ add x6, x6, x1169
+ ldr d1, x6170
+.else171
+ ld1 {v0.8b}, x6, x1172
+ ld1 {v1.8b}, x6, x1173
+ ld1 {v2.8b}, x6, x1174
+ ld1 {v3.8b}, x6175
+.endif176
+.endm177
+178
+.macro qpel_chroma_load_64b v179
+.if \v == 0180
+ // qpel_filter_chroma_0 only uses values in v1181
+ add x6, x6, x1182
+ ldr q1, x6183
+.else184
+ ld1 {v0.16b}, x6, x1185
+ ld1 {v1.16b}, x6, x1186
+ ld1 {v2.16b}, x6, x1187
+ ld1 {v3.16b}, x6188
+.endif189
+.endm190
+191
+// a, b, c, d, e, f, g, h192
+// .hword 0, 0, 0, 64, 0, 0, 0, 0193
+.macro qpel_start_0194
+ movi v24.16b, #64195
+.endm196
+197
+.macro qpel_filter_0_32b198
+ umull v17.8h, v3.8b, v24.8b // 64*d199
+.endm200
+201
+.macro qpel_filter_0_64b202
+ qpel_filter_0_32b203
+ umull2 v18.8h, v3.16b, v24.16b // 64*d204
+.endm205
+206
+.macro qpel_start_0_1207
+ movi v24.8h, #64208
+.endm209
+210
+.macro qpel_filter_0_32b_1211
+ smull v17.4s, v3.4h, v24.4h // 64*d0212
+ smull2 v18.4s, v3.8h, v24.8h // 64*d1213
+.endm214
+215
+// a, b, c, d, e, f, g, h216
+// .hword -1, 4, -10, 58, 17, -5, 1, 0217
+.macro qpel_start_1218
+ movi v24.16b, #58219
+ movi v25.16b, #10220
+ movi v26.16b, #17221
+ movi v27.16b, #5222
+.endm223
+224
+.macro qpel_filter_1_32b225
+ umull v19.8h, v2.8b, v25.8b // c*10226
+ umull v17.8h, v3.8b, v24.8b // d*58227
+ umull v21.8h, v4.8b, v26.8b // e*17228
+ umull v23.8h, v5.8b, v27.8b // f*5229
+ sub v17.8h, v17.8h, v19.8h // d*58 - c*10230
+ ushll v18.8h, v1.8b, #2 // b*4231
+ add v17.8h, v17.8h, v21.8h // d*58 - c*10 + e*17232
+ usubl v21.8h, v6.8b, v0.8b // g - a233
+ add v17.8h, v17.8h, v18.8h // d*58 - c*10 + e*17 + b*4234
+ sub v21.8h, v21.8h, v23.8h // g - a - f*5235
+ add v17.8h, v17.8h, v21.8h // d*58 - c*10 + e*17 + b*4 + g - a - f*5236
+.endm237
+238
+.macro qpel_filter_1_64b239
+ qpel_filter_1_32b240
+ umull2 v20.8h, v2.16b, v25.16b // c*10241
+ umull2 v18.8h, v3.16b, v24.16b // d*58242
+ umull2 v21.8h, v4.16b, v26.16b // e*17243
+ umull2 v23.8h, v5.16b, v27.16b // f*5244
+ sub v18.8h, v18.8h, v20.8h // d*58 - c*10245
+ ushll2 v28.8h, v1.16b, #2 // b*4246
+ add v18.8h, v18.8h, v21.8h // d*58 - c*10 + e*17247
+ usubl2 v21.8h, v6.16b, v0.16b // g - a248
+ add v18.8h, v18.8h, v28.8h // d*58 - c*10 + e*17 + b*4249
+ sub v21.8h, v21.8h, v23.8h // g - a - f*5250
+ add v18.8h, v18.8h, v21.8h // d*58 - c*10 + e*17 + b*4 + g - a - f*5251
+.endm252
+253
+.macro qpel_start_1_1254
+ movi v24.8h, #58255
+ movi v25.8h, #10256
+ movi v26.8h, #17257
+ movi v27.8h, #5258
+.endm259
+260
+.macro qpel_filter_1_32b_1261
+ smull v17.4s, v3.4h, v24.4h // 58 * d0262
+ smull2 v18.4s, v3.8h, v24.8h // 58 * d1263
+ smull v19.4s, v2.4h, v25.4h // 10 * c0264
+ smull2 v20.4s, v2.8h, v25.8h // 10 * c1265
+ smull v21.4s, v4.4h, v26.4h // 17 * e0266
+ smull2 v22.4s, v4.8h, v26.8h // 17 * e1267
+ smull v23.4s, v5.4h, v27.4h // 5 * f0268
+ smull2 v16.4s, v5.8h, v27.8h // 5 * f1269
+ sub v17.4s, v17.4s, v19.4s // 58 * d0 - 10 * c0270
+ sub v18.4s, v18.4s, v20.4s // 58 * d1 - 10 * c1271
+ sshll v19.4s, v1.4h, #2 // 4 * b0272
+ sshll2 v20.4s, v1.8h, #2 // 4 * b1273
+ add v17.4s, v17.4s, v21.4s // 58 * d0 - 10 * c0 + 17 * e0274
+ add v18.4s, v18.4s, v22.4s // 58 * d1 - 10 * c1 + 17 * e1275
+ ssubl v21.4s, v6.4h, v0.4h // g0 - a0276
+ ssubl2 v22.4s, v6.8h, v0.8h // g1 - a1277
+ add v17.4s, v17.4s, v19.4s // 58 * d0 - 10 * c0 + 17 * e0 + 4 * b0278
+ add v18.4s, v18.4s, v20.4s // 58 * d1 - 10 * c1 + 17 * e1 + 4 * b1279
+ sub v21.4s, v21.4s, v23.4s // g0 - a0 - 5 * f0280
+ sub v22.4s, v22.4s, v16.4s // g1 - a1 - 5 * f1281
+ add v17.4s, v17.4s, v21.4s // 58 * d0 - 10 * c0 + 17 * e0 + 4 * b0 + g0 - a0 - 5 * f0282
+ add v18.4s, v18.4s, v22.4s // 58 * d1 - 10 * c1 + 17 * e1 + 4 * b1 + g1 - a1 - 5 * f1283
+.endm284
+285
+// a, b, c, d, e, f, g, h286
+// .hword -1, 4, -11, 40, 40, -11, 4, -1287
+.macro qpel_start_2288
+ movi v24.8h, #11289
+ movi v25.8h, #40290
+.endm291
+292
+.macro qpel_filter_2_32b293
+ uaddl v17.8h, v3.8b, v4.8b // d + e294
+ uaddl v19.8h, v2.8b, v5.8b // c + f295
+ uaddl v23.8h, v1.8b, v6.8b // b + g296
+ uaddl v21.8h, v0.8b, v7.8b // a + h297
+ mul v17.8h, v17.8h, v25.8h // 40 * (d + e)298
+ mul v19.8h, v19.8h, v24.8h // 11 * (c + f)299
+ shl v23.8h, v23.8h, #2 // (b + g) * 4300
+ add v19.8h, v19.8h, v21.8h // 11 * (c + f) + a + h301
+ add v17.8h, v17.8h, v23.8h // 40 * (d + e) + (b + g) * 4302
+ sub v17.8h, v17.8h, v19.8h // 40 * (d + e) + (b + g) * 4 - 11 * (c + f) - a - h303
+.endm304
+305
+.macro qpel_filter_2_64b306
+ qpel_filter_2_32b307
+ uaddl2 v27.8h, v3.16b, v4.16b // d + e308
+ uaddl2 v16.8h, v2.16b, v5.16b // c + f309
+ uaddl2 v23.8h, v1.16b, v6.16b // b + g310
+ uaddl2 v21.8h, v0.16b, v7.16b // a + h311
+ mul v27.8h, v27.8h, v25.8h // 40 * (d + e)312
+ mul v16.8h, v16.8h, v24.8h // 11 * (c + f)313
+ shl v23.8h, v23.8h, #2 // (b + g) * 4314
+ add v16.8h, v16.8h, v21.8h // 11 * (c + f) + a + h315
+ add v27.8h, v27.8h, v23.8h // 40 * (d + e) + (b + g) * 4316
+ sub v18.8h, v27.8h, v16.8h // 40 * (d + e) + (b + g) * 4 - 11 * (c + f) - a - h317
+.endm318
+319
+.macro qpel_start_2_1320
+ movi v24.4s, #11321
+ movi v25.4s, #40322
+.endm323
+324
+.macro qpel_filter_2_32b_1325
+ saddl v17.4s, v3.4h, v4.4h // d0 + e0326
+ saddl2 v18.4s, v3.8h, v4.8h // d1 + e1327
+ saddl v19.4s, v2.4h, v5.4h // c0 + f0328
+ saddl2 v20.4s, v2.8h, v5.8h // c1 + f1329
+ mul v19.4s, v19.4s, v24.4s // 11 * (c0 + f0)330
+ mul v20.4s, v20.4s, v24.4s // 11 * (c1 + f1)331
+ saddl v23.4s, v1.4h, v6.4h // b0 + g0332
+ mul v17.4s, v17.4s, v25.4s // 40 * (d0 + e0)333
+ mul v18.4s, v18.4s, v25.4s // 40 * (d1 + e1)334
+ saddl2 v16.4s, v1.8h, v6.8h // b1 + g1335
+ saddl v21.4s, v0.4h, v7.4h // a0 + h0336
+ saddl2 v22.4s, v0.8h, v7.8h // a1 + h1337
+ shl v23.4s, v23.4s, #2 // 4*(b0+g0)338
+ shl v16.4s, v16.4s, #2 // 4*(b1+g1)339
+ add v19.4s, v19.4s, v21.4s // 11 * (c0 + f0) + a0 + h0340
+ add v20.4s, v20.4s, v22.4s // 11 * (c1 + f1) + a1 + h1341
+ add v17.4s, v17.4s, v23.4s // 40 * (d0 + e0) + 4*(b0+g0)342
+ add v18.4s, v18.4s, v16.4s // 40 * (d1 + e1) + 4*(b1+g1)343
+ sub v17.4s, v17.4s, v19.4s // 40 * (d0 + e0) + 4*(b0+g0) - (11 * (c0 + f0) + a0 + h0)344
+ sub v18.4s, v18.4s, v20.4s // 40 * (d1 + e1) + 4*(b1+g1) - (11 * (c1 + f1) + a1 + h1)345
+.endm346
+347
+// a, b, c, d, e, f, g, h348
+// .hword 0, 1, -5, 17, 58, -10, 4, -1349
+.macro qpel_start_3350
+ movi v24.16b, #17351
+ movi v25.16b, #5352
+ movi v26.16b, #58353
+ movi v27.16b, #10354
+.endm355
+356
+.macro qpel_filter_3_32b357
+ umull v19.8h, v2.8b, v25.8b // c * 5358
+ umull v17.8h, v3.8b, v24.8b // d * 17359
+ umull v21.8h, v4.8b, v26.8b // e * 58360
+ umull v23.8h, v5.8b, v27.8b // f * 10361
+ sub v17.8h, v17.8h, v19.8h // d * 17 - c * 5362
+ ushll v19.8h, v6.8b, #2 // g * 4363
+ add v17.8h, v17.8h, v21.8h // d * 17 - c * 5 + e * 58364
+ usubl v21.8h, v1.8b, v7.8b // b - h365
+ add v17.8h, v17.8h, v19.8h // d * 17 - c * 5 + e * 58 + g * 4366
+ sub v21.8h, v21.8h, v23.8h // b - h - f * 10367
+ add v17.8h, v17.8h, v21.8h // d * 17 - c * 5 + e * 58 + g * 4 + b - h - f * 10368
+.endm369
+370
+.macro qpel_filter_3_64b371
+ qpel_filter_3_32b372
+ umull2 v16.8h, v2.16b, v25.16b // c * 5373
+ umull2 v18.8h, v3.16b, v24.16b // d * 17374
+ umull2 v21.8h, v4.16b, v26.16b // e * 58375
+ umull2 v23.8h, v5.16b, v27.16b // f * 10376
+ sub v18.8h, v18.8h, v16.8h // d * 17 - c * 5377
+ ushll2 v16.8h, v6.16b, #2 // g * 4378
+ add v18.8h, v18.8h, v21.8h // d * 17 - c * 5 + e * 58379
+ usubl2 v21.8h, v1.16b, v7.16b // b - h380
+ add v18.8h, v18.8h, v16.8h // d * 17 - c * 5 + e * 58 + g * 4381
+ sub v21.8h, v21.8h, v23.8h // b - h - f * 10382
+ add v18.8h, v18.8h, v21.8h // d * 17 - c * 5 + e * 58 + g * 4 + b - h - f * 10383
+.endm384
+385
+.macro qpel_start_3_1386
+ movi v24.8h, #17387
+ movi v25.8h, #5388
+ movi v26.8h, #58389
+ movi v27.8h, #10390
+.endm391
+392
+.macro qpel_filter_3_32b_1393
+ smull v17.4s, v3.4h, v24.4h // 17 * d0394
+ smull2 v18.4s, v3.8h, v24.8h // 17 * d1395
+ smull v19.4s, v2.4h, v25.4h // 5 * c0396
+ smull2 v20.4s, v2.8h, v25.8h // 5 * c1397
+ smull v21.4s, v4.4h, v26.4h // 58 * e0398
+ smull2 v22.4s, v4.8h, v26.8h // 58 * e1399
+ smull v23.4s, v5.4h, v27.4h // 10 * f0400
+ smull2 v16.4s, v5.8h, v27.8h // 10 * f1401
+ sub v17.4s, v17.4s, v19.4s // 17 * d0 - 5 * c0402
+ sub v18.4s, v18.4s, v20.4s // 17 * d1 - 5 * c1403
+ sshll v19.4s, v6.4h, #2 // 4 * g0404
+ sshll2 v20.4s, v6.8h, #2 // 4 * g1405
+ add v17.4s, v17.4s, v21.4s // 17 * d0 - 5 * c0 + 58 * e0406
+ add v18.4s, v18.4s, v22.4s // 17 * d1 - 5 * c1 + 58 * e1407
+ ssubl v21.4s, v1.4h, v7.4h // b0 - h0408
+ ssubl2 v22.4s, v1.8h, v7.8h // b1 - h1409
+ add v17.4s, v17.4s, v19.4s // 17 * d0 - 5 * c0 + 58 * e0 + 4 * g0410
+ add v18.4s, v18.4s, v20.4s // 17 * d1 - 5 * c1 + 58 * e1 + 4 * g1411
+ sub v21.4s, v21.4s, v23.4s // b0 - h0 - 10 * f0412
+ sub v22.4s, v22.4s, v16.4s // b1 - h1 - 10 * f1413
+ add v17.4s, v17.4s, v21.4s // 17 * d0 - 5 * c0 + 58 * e0 + 4 * g0 + b0 - h0 - 10 * f0414
+ add v18.4s, v18.4s, v22.4s // 17 * d1 - 5 * c1 + 58 * e1 + 4 * g1 + b1 - h1 - 10 * f1415
+.endm416
+417
+.macro qpel_start_chroma_0418
+ movi v24.16b, #64419
+.endm420
+421
+.macro qpel_filter_chroma_0_32b422
+ umull v17.8h, v1.8b, v24.8b // 64*b423
+.endm424
+425
+.macro qpel_filter_chroma_0_64b426
+ umull v17.8h, v1.8b, v24.8b // 64*b427
+ umull2 v18.8h, v1.16b, v24.16b // 64*b428
+.endm429
+430
+.macro qpel_start_chroma_0_1431
+ movi v24.8h, #64432
+.endm433
+434
+.macro qpel_filter_chroma_0_32b_1435
+ smull v17.4s, v1.4h, v24.4h // 64*b0436
+ smull2 v18.4s, v1.8h, v24.8h // 64*b1437
+.endm438
+439
+.macro qpel_start_chroma_1440
+ movi v24.16b, #58441
+ movi v25.16b, #10442
+.endm443
+444
+.macro qpel_filter_chroma_1_32b445
+ umull v17.8h, v1.8b, v24.8b // 58 * b446
+ umull v19.8h, v2.8b, v25.8b // 10 * c447
+ uaddl v22.8h, v0.8b, v3.8b // a + d448
+ shl v22.8h, v22.8h, #1 // 2 * (a+d)449
+ sub v17.8h, v17.8h, v22.8h // 58*b - 2*(a+d)450
+ add v17.8h, v17.8h, v19.8h // 58*b-2*(a+d) + 10*c451
+.endm452
+453
+.macro qpel_filter_chroma_1_64b454
+ umull v17.8h, v1.8b, v24.8b // 58 * b455
+ umull2 v18.8h, v1.16b, v24.16b // 58 * b456
+ umull v19.8h, v2.8b, v25.8b // 10 * c457
+ umull2 v20.8h, v2.16b, v25.16b // 10 * c458
+ uaddl v22.8h, v0.8b, v3.8b // a + d459
+ uaddl2 v23.8h, v0.16b, v3.16b // a + d460
+ shl v22.8h, v22.8h, #1 // 2 * (a+d)461
+ shl v23.8h, v23.8h, #1 // 2 * (a+d)462
+ sub v17.8h, v17.8h, v22.8h // 58*b - 2*(a+d)463
+ sub v18.8h, v18.8h, v23.8h // 58*b - 2*(a+d)464
+ add v17.8h, v17.8h, v19.8h // 58*b-2*(a+d) + 10*c465
+ add v18.8h, v18.8h, v20.8h // 58*b-2*(a+d) + 10*c466
+.endm467
+468
+.macro qpel_start_chroma_1_1469
+ movi v24.8h, #58470
+ movi v25.8h, #10471
+.endm472
+473
+.macro qpel_filter_chroma_1_32b_1474
+ smull v17.4s, v1.4h, v24.4h // 58 * b0475
+ smull2 v18.4s, v1.8h, v24.8h // 58 * b1476
+ smull v19.4s, v2.4h, v25.4h // 10 * c0477
+ smull2 v20.4s, v2.8h, v25.8h // 10 * c1478
+ add v22.8h, v0.8h, v3.8h // a + d479
+ sshll v21.4s, v22.4h, #1 // 2 * (a0+d0)480
+ sshll2 v22.4s, v22.8h, #1 // 2 * (a1+d1)481
+ sub v17.4s, v17.4s, v21.4s // 58*b0 - 2*(a0+d0)482
+ sub v18.4s, v18.4s, v22.4s // 58*b1 - 2*(a1+d1)483
+ add v17.4s, v17.4s, v19.4s // 58*b0-2*(a0+d0) + 10*c0484
+ add v18.4s, v18.4s, v20.4s // 58*b1-2*(a1+d1) + 10*c1485
+.endm486
+487
+.macro qpel_start_chroma_2488
+ movi v25.16b, #54489
+.endm490
+491
+.macro qpel_filter_chroma_2_32b492
+ umull v17.8h, v1.8b, v25.8b // 54 * b493
+ ushll v19.8h, v0.8b, #2 // 4 * a494
+ ushll v21.8h, v2.8b, #4 // 16 * c495
+ ushll v23.8h, v3.8b, #1 // 2 * d496
+ add v17.8h, v17.8h, v21.8h // 54*b + 16*c497
+ add v19.8h, v19.8h, v23.8h // 4*a + 2*d498
+ sub v17.8h, v17.8h, v19.8h // 54*b+16*c - (4*a+2*d)499
+.endm500
+501
+.macro qpel_filter_chroma_2_64b502
+ umull v17.8h, v1.8b, v25.8b // 54 * b503
+ umull2 v18.8h, v1.16b, v25.16b // 54 * b504
+ ushll v19.8h, v0.8b, #2 // 4 * a505
+ ushll2 v20.8h, v0.16b, #2 // 4 * a506
+ ushll v21.8h, v2.8b, #4 // 16 * c507
+ ushll2 v22.8h, v2.16b, #4 // 16 * c508
+ ushll v23.8h, v3.8b, #1 // 2 * d509
+ ushll2 v24.8h, v3.16b, #1 // 2 * d510
+ add v17.8h, v17.8h, v21.8h // 54*b + 16*c511
+ add v18.8h, v18.8h, v22.8h // 54*b + 16*c512
+ add v19.8h, v19.8h, v23.8h // 4*a + 2*d513
+ add v20.8h, v20.8h, v24.8h // 4*a + 2*d514
+ sub v17.8h, v17.8h, v19.8h // 54*b+16*c - (4*a+2*d)515
+ sub v18.8h, v18.8h, v20.8h // 54*b+16*c - (4*a+2*d)516
+.endm517
+518
+.macro qpel_start_chroma_2_1519
+ movi v25.8h, #54520
+.endm521
+522
+.macro qpel_filter_chroma_2_32b_1523
+ smull v17.4s, v1.4h, v25.4h // 54 * b0524
+ smull2 v18.4s, v1.8h, v25.8h // 54 * b1525
+ sshll v19.4s, v0.4h, #2 // 4 * a0526
+ sshll2 v20.4s, v0.8h, #2 // 4 * a1527
+ sshll v21.4s, v2.4h, #4 // 16 * c0528
+ sshll2 v22.4s, v2.8h, #4 // 16 * c1529
+ sshll v23.4s, v3.4h, #1 // 2 * d0530
+ sshll2 v24.4s, v3.8h, #1 // 2 * d1531
+ add v17.4s, v17.4s, v21.4s // 54*b0 + 16*c0532
+ add v18.4s, v18.4s, v22.4s // 54*b1 + 16*c1533
+ add v19.4s, v19.4s, v23.4s // 4*a0 + 2*d0534
+ add v20.4s, v20.4s, v24.4s // 4*a1 + 2*d1535
+ sub v17.4s, v17.4s, v19.4s // 54*b0+16*c0 - (4*a0+2*d0)536
+ sub v18.4s, v18.4s, v20.4s // 54*b1+16*c1 - (4*a1+2*d1)537
+.endm538
+539
+.macro qpel_start_chroma_3540
+ movi v25.16b, #46541
+ movi v26.16b, #28542
+ movi v27.16b, #6543
+.endm544
+545
+.macro qpel_filter_chroma_3_32b546
+ umull v17.8h, v1.8b, v25.8b // 46 * b547
+ umull v19.8h, v2.8b, v26.8b // 28 * c548
+ ushll v21.8h, v3.8b, #2 // 4 * d549
+ umull v23.8h, v0.8b, v27.8b // 6 * a550
+ add v17.8h, v17.8h, v19.8h // 46*b + 28*c551
+ add v21.8h, v21.8h, v23.8h // 4*d + 6*a552
+ sub v17.8h, v17.8h, v21.8h // 46*b+28*c - (4*d+6*a)553
+.endm554
+555
+.macro qpel_filter_chroma_3_64b556
+ umull v17.8h, v1.8b, v25.8b // 46 * b557
+ umull2 v18.8h, v1.16b, v25.16b // 46 * b558
+ umull v19.8h, v2.8b, v26.8b // 28 * c559
+ umull2 v20.8h, v2.16b, v26.16b // 28 * c560
+ ushll v21.8h, v3.8b, #2 // 4 * d561
+ ushll2 v22.8h, v3.16b, #2 // 4 * d562
+ umull v23.8h, v0.8b, v27.8b // 6 * a563
+ umull2 v24.8h, v0.16b, v27.16b // 6 * a564
+ add v17.8h, v17.8h, v19.8h // 46*b + 28*c565
+ add v18.8h, v18.8h, v20.8h // 46*b + 28*c566
+ add v21.8h, v21.8h, v23.8h // 4*d + 6*a567
+ add v22.8h, v22.8h, v24.8h // 4*d + 6*a568
+ sub v17.8h, v17.8h, v21.8h // 46*b+28*c - (4*d+6*a)569
+ sub v18.8h, v18.8h, v22.8h // 46*b+28*c - (4*d+6*a)570
+.endm571
+572
+.macro qpel_start_chroma_3_1573
+ movi v25.8h, #46574
+ movi v26.8h, #28575
+ movi v27.8h, #6576
+.endm577
+578
+.macro qpel_filter_chroma_3_32b_1579
+ smull v17.4s, v1.4h, v25.4h // 46 * b0580
+ smull2 v18.4s, v1.8h, v25.8h // 46 * b1581
+ smull v19.4s, v2.4h, v26.4h // 28 * c0582
+ smull2 v20.4s, v2.8h, v26.8h // 28 * c1583
+ sshll v21.4s, v3.4h, #2 // 4 * d0584
+ sshll2 v22.4s, v3.8h, #2 // 4 * d1585
+ smull v23.4s, v0.4h, v27.4h // 6 * a0586
+ smull2 v24.4s, v0.8h, v27.8h // 6 * a1587
+ add v17.4s, v17.4s, v19.4s // 46*b0 + 28*c0588
+ add v18.4s, v18.4s, v20.4s // 46*b1 + 28*c1589
+ add v21.4s, v21.4s, v23.4s // 4*d0 + 6*a0590
+ add v22.4s, v22.4s, v24.4s // 4*d1 + 6*a1591
+ sub v17.4s, v17.4s, v21.4s // 46*b0+28*c0 - (4*d0+6*a0)592
+ sub v18.4s, v18.4s, v22.4s // 46*b1+28*c1 - (4*d1+6*a1)593
+.endm594
+595
+.macro qpel_start_chroma_4596
+ movi v24.8h, #36597
+.endm598
+599
+.macro qpel_filter_chroma_4_32b600
+ uaddl v20.8h, v0.8b, v3.8b // a + d601
+ uaddl v17.8h, v1.8b, v2.8b // b + c602
+ shl v20.8h, v20.8h, #2 // 4 * (a+d)603
+ mul v17.8h, v17.8h, v24.8h // 36 * (b+c)604
+ sub v17.8h, v17.8h, v20.8h // 36*(b+c) - 4*(a+d)605
+.endm606
+607
+.macro qpel_filter_chroma_4_64b608
+ uaddl v20.8h, v0.8b, v3.8b // a + d609
+ uaddl2 v21.8h, v0.16b, v3.16b // a + d610
+ uaddl v17.8h, v1.8b, v2.8b // b + c611
+ uaddl2 v18.8h, v1.16b, v2.16b // b + c612
+ shl v20.8h, v20.8h, #2 // 4 * (a+d)613
+ shl v21.8h, v21.8h, #2 // 4 * (a+d)614
+ mul v17.8h, v17.8h, v24.8h // 36 * (b+c)615
+ mul v18.8h, v18.8h, v24.8h // 36 * (b+c)616
+ sub v17.8h, v17.8h, v20.8h // 36*(b+c) - 4*(a+d)617
+ sub v18.8h, v18.8h, v21.8h // 36*(b+c) - 4*(a+d)618
+.endm619
+620
+.macro qpel_start_chroma_4_1621
+ movi v24.8h, #36622
+.endm623
+624
+.macro qpel_filter_chroma_4_32b_1625
+ add v20.8h, v0.8h, v3.8h // a + d626
+ add v21.8h, v1.8h, v2.8h // b + c627
+ smull v17.4s, v21.4h, v24.4h // 36 * (b0+c0)628
+ smull2 v18.4s, v21.8h, v24.8h // 36 * (b1+c1)629
+ sshll v21.4s, v20.4h, #2 // 4 * (a0+d0)630
+ sshll2 v22.4s, v20.8h, #2 // 4 * (a1+d1)631
+ sub v17.4s, v17.4s, v21.4s // 36*(b0+c0) - 4*(a0+d0)632
+ sub v18.4s, v18.4s, v22.4s // 36*(b1+c1) - 4*(a1+d1)633
+.endm634
+635
+.macro qpel_start_chroma_5636
+ movi v25.16b, #28637
+ movi v26.16b, #46638
+ movi v27.16b, #6639
+.endm640
+641
+.macro qpel_filter_chroma_5_32b642
+ umull v17.8h, v1.8b, v25.8b // 28 * b643
+ umull v19.8h, v2.8b, v26.8b // 46 * c644
+ ushll v21.8h, v0.8b, #2 // 4 * a645
+ umull v23.8h, v3.8b, v27.8b // 6 * d646
+ add v17.8h, v17.8h, v19.8h // 28*b + 46*c647
+ add v21.8h, v21.8h, v23.8h // 4*a + 6*d648
+ sub v17.8h, v17.8h, v21.8h // 28*b+46*c - (4*a+6*d)649
+.endm650
+651
+.macro qpel_filter_chroma_5_64b652
+ umull v17.8h, v1.8b, v25.8b // 28 * b653
+ umull2 v18.8h, v1.16b, v25.16b // 28 * b654
+ umull v19.8h, v2.8b, v26.8b // 46 * c655
+ umull2 v20.8h, v2.16b, v26.16b // 46 * c656
+ ushll v21.8h, v0.8b, #2 // 4 * a657
+ ushll2 v22.8h, v0.16b, #2 // 4 * a658
+ umull v23.8h, v3.8b, v27.8b // 6 * d659
+ umull2 v24.8h, v3.16b, v27.16b // 6 * d660
+ add v17.8h, v17.8h, v19.8h // 28*b + 46*c661
+ add v18.8h, v18.8h, v20.8h // 28*b + 46*c662
+ add v21.8h, v21.8h, v23.8h // 4*a + 6*d663
+ add v22.8h, v22.8h, v24.8h // 4*a + 6*d664
+ sub v17.8h, v17.8h, v21.8h // 28*b+46*c - (4*a+6*d)665
+ sub v18.8h, v18.8h, v22.8h // 28*b+46*c - (4*a+6*d)666
+.endm667
+668
+.macro qpel_start_chroma_5_1669
+ movi v25.8h, #28670
+ movi v26.8h, #46671
+ movi v27.8h, #6672
+.endm673
+674
+.macro qpel_filter_chroma_5_32b_1675
+ smull v17.4s, v1.4h, v25.4h // 28 * b0676
+ smull2 v18.4s, v1.8h, v25.8h // 28 * b1677
+ smull v19.4s, v2.4h, v26.4h // 46 * c0678
+ smull2 v20.4s, v2.8h, v26.8h // 46 * c1679
+ sshll v21.4s, v0.4h, #2 // 4 * a0680
+ sshll2 v22.4s, v0.8h, #2 // 4 * a1681
+ smull v23.4s, v3.4h, v27.4h // 6 * d0682
+ smull2 v24.4s, v3.8h, v27.8h // 6 * d1683
+ add v17.4s, v17.4s, v19.4s // 28*b0 + 46*c0684
+ add v18.4s, v18.4s, v20.4s // 28*b1 + 46*c1685
+ add v21.4s, v21.4s, v23.4s // 4*a0 + 6*d0686
+ add v22.4s, v22.4s, v24.4s // 4*a1 + 6*d1687
+ sub v17.4s, v17.4s, v21.4s // 28*b0+46*c0 - (4*a0+6*d0)688
+ sub v18.4s, v18.4s, v22.4s // 28*b1+46*c1 - (4*a1+6*d1)689
+.endm690
+691
+.macro qpel_start_chroma_6692
+ movi v25.16b, #54693
+.endm694
+695
+.macro qpel_filter_chroma_6_32b696
+ umull v17.8h, v2.8b, v25.8b // 54 * c697
+ ushll v19.8h, v0.8b, #1 // 2 * a698
+ ushll v21.8h, v1.8b, #4 // 16 * b699
+ ushll v23.8h, v3.8b, #2 // 4 * d700
+ add v17.8h, v17.8h, v21.8h // 54*c + 16*b701
+ add v19.8h, v19.8h, v23.8h // 2*a + 4*d702
+ sub v17.8h, v17.8h, v19.8h // 54*c+16*b - (2*a+4*d)703
+.endm704
+705
+.macro qpel_filter_chroma_6_64b706
+ umull v17.8h, v2.8b, v25.8b // 54 * c707
+ umull2 v18.8h, v2.16b, v25.16b // 54 * c708
+ ushll v19.8h, v0.8b, #1 // 2 * a709
+ ushll2 v20.8h, v0.16b, #1 // 2 * a710
+ ushll v21.8h, v1.8b, #4 // 16 * b711
+ ushll2 v22.8h, v1.16b, #4 // 16 * b712
+ ushll v23.8h, v3.8b, #2 // 4 * d713
+ ushll2 v24.8h, v3.16b, #2 // 4 * d714
+ add v17.8h, v17.8h, v21.8h // 54*c + 16*b715
+ add v18.8h, v18.8h, v22.8h // 54*c + 16*b716
+ add v19.8h, v19.8h, v23.8h // 2*a + 4*d717
+ add v20.8h, v20.8h, v24.8h // 2*a + 4*d718
+ sub v17.8h, v17.8h, v19.8h // 54*c+16*b - (2*a+4*d)719
+ sub v18.8h, v18.8h, v20.8h // 54*c+16*b - (2*a+4*d)720
+.endm721
+722
+.macro qpel_start_chroma_6_1723
+ movi v25.8h, #54724
+.endm725
+726
+.macro qpel_filter_chroma_6_32b_1727
+ smull v17.4s, v2.4h, v25.4h // 54 * c0728
+ smull2 v18.4s, v2.8h, v25.8h // 54 * c1729
+ sshll v19.4s, v0.4h, #1 // 2 * a0730
+ sshll2 v20.4s, v0.8h, #1 // 2 * a1731
+ sshll v21.4s, v1.4h, #4 // 16 * b0732
+ sshll2 v22.4s, v1.8h, #4 // 16 * b1733
+ sshll v23.4s, v3.4h, #2 // 4 * d0734
+ sshll2 v24.4s, v3.8h, #2 // 4 * d1735
+ add v17.4s, v17.4s, v21.4s // 54*c0 + 16*b0736
+ add v18.4s, v18.4s, v22.4s // 54*c1 + 16*b1737
+ add v19.4s, v19.4s, v23.4s // 2*a0 + 4*d0738
+ add v20.4s, v20.4s, v24.4s // 2*a1 + 4*d1739
+ sub v17.4s, v17.4s, v19.4s // 54*c0+16*b0 - (2*a0+4*d0)740
+ sub v18.4s, v18.4s, v20.4s // 54*c1+16*b1 - (2*a1+4*d1)741
+.endm742
+743
+.macro qpel_start_chroma_7744
+ movi v24.16b, #58745
+ movi v25.16b, #10746
+.endm747
+748
+.macro qpel_filter_chroma_7_32b749
+ uaddl v20.8h, v0.8b, v3.8b // a + d750
+ umull v17.8h, v2.8b, v24.8b // 58 * c751
+ shl v20.8h, v20.8h, #1 // 2 * (a+d)752
+ umull v19.8h, v1.8b, v25.8b // 10 * b753
+ sub v17.8h, v17.8h, v20.8h // 58*c - 2*(a+d)754
+ add v17.8h, v17.8h, v19.8h // 58*c-2*(a+d) + 10*b755
+.endm756
+757
+.macro qpel_filter_chroma_7_64b758
+ uaddl v20.8h, v0.8b, v3.8b // a + d759
+ uaddl2 v21.8h, v0.16b, v3.16b // a + d760
+ umull v17.8h, v2.8b, v24.8b // 58 * c761
+ umull2 v18.8h, v2.16b, v24.16b // 58 * c762
+ shl v20.8h, v20.8h, #1 // 2 * (a+d)763
+ shl v21.8h, v21.8h, #1 // 2 * (a+d)764
+ umull v22.8h, v1.8b, v25.8b // 10 * b765
+ umull2 v23.8h, v1.16b, v25.16b // 10 * b766
+ sub v17.8h, v17.8h, v20.8h // 58*c - 2*(a+d)767
+ sub v18.8h, v18.8h, v21.8h // 58*c - 2*(a+d)768
+ add v17.8h, v17.8h, v22.8h // 58*c-2*(a+d) + 10*b769
+ add v18.8h, v18.8h, v23.8h // 58*c-2*(a+d) + 10*b770
+.endm771
+772
+.macro qpel_start_chroma_7_1773
+ movi v24.8h, #58774
+ movi v25.8h, #10775
+.endm776
+777
+.macro qpel_filter_chroma_7_32b_1778
+ add v20.8h, v0.8h, v3.8h // a + d779
+ smull v17.4s, v2.4h, v24.4h // 58 * c0780
+ smull2 v18.4s, v2.8h, v24.8h // 58 * c1781
+ sshll v21.4s, v20.4h, #1 // 2 * (a0+d0)782
+ sshll2 v22.4s, v20.8h, #1 // 2 * (a1+d1)783
+ smull v19.4s, v1.4h, v25.4h // 10 * b0784
+ smull2 v20.4s, v1.8h, v25.8h // 10 * b1785
+ sub v17.4s, v17.4s, v21.4s // 58*c0 - 2*(a0+d0)786
+ sub v18.4s, v18.4s, v22.4s // 58*c1 - 2*(a1+d1)787
+ add v17.4s, v17.4s, v19.4s // 58*c0-2*(a0+d0) + 10*b0788
+ add v18.4s, v18.4s, v20.4s // 58*c1-2*(a1+d1) + 10*b1789
+.endm790
+791
+.macro vpp_end792
+ add v17.8h, v17.8h, v31.8h793
+ sqshrun v17.8b, v17.8h, #6794
+.endm795
+796
+.macro FILTER_LUMA_VPP w, h, v797
+ lsl x10, x1, #2 // x10 = 4 * x1798
+ sub x11, x10, x1 // x11 = 3 * x1799
+ sub x0, x0, x11 // src -= (8 / 2 - 1) * srcStride800
+ mov x5, #\h801
+ mov w12, #32802
+ dup v31.8h, w12803
+ qpel_start_\v804
+.loop_luma_vpp_\v\()_\w\()x\h:805
+ mov x7, x2806
+ mov x9, #0807
+.loop_luma_vpp_w8_\v\()_\w\()x\h:808
+ add x6, x0, x9809
+.if \w == 8 || \w == 24810
+ qpel_load_32b \v811
+ qpel_filter_\v\()_32b812
+ vpp_end813
+ str d17, x7, #8814
+ add x9, x9, #8815
+.elseif \w == 12816
+ qpel_load_32b \v817
+ qpel_filter_\v\()_32b818
+ vpp_end819
+ str d17, x7, #8820
+ add x6, x0, #8821
+ qpel_load_32b \v822
+ qpel_filter_\v\()_32b823
+ vpp_end824
+ fmov w6, s17825
+ str w6, x7, #4826
+ add x9, x9, #12827
+.else828
+ qpel_load_64b \v829
+ qpel_filter_\v\()_64b830
+ vpp_end831
+ add v18.8h, v18.8h, v31.8h832
+ sqshrun2 v17.16b, v18.8h, #6833
+ str q17, x7, #16834
+ add x9, x9, #16835
+.endif836
+ cmp x9, #\w837
+ blt .loop_luma_vpp_w8_\v\()_\w\()x\h838
+ add x0, x0, x1839
+ add x2, x2, x3840
+ sub x5, x5, #1841
+ cbnz x5, .loop_luma_vpp_\v\()_\w\()x\h842
+ ret843
+.endm844
+845
+.macro vps_end846
+ sub v17.8h, v17.8h, v31.8h847
+.endm848
+849
+.macro FILTER_VPS w, h, v850
+ lsl x3, x3, #1851
+ lsl x10, x1, #2 // x10 = 4 * x1852
+ sub x11, x10, x1 // x11 = 3 * x1853
+ sub x0, x0, x11 // src -= (8 / 2 - 1) * srcStride854
+ mov x5, #\h855
+ mov w12, #8192856
+ dup v31.8h, w12857
+ qpel_start_\v858
+.loop_ps_\v\()_\w\()x\h:859
+ mov x7, x2860
+ mov x9, #0861
+.loop_ps_w8_\v\()_\w\()x\h:862
+ add x6, x0, x9863
+.if \w == 8 || \w == 24864
+ qpel_load_32b \v865
+ qpel_filter_\v\()_32b866
+ vps_end867
+ str q17, x7, #16868
+ add x9, x9, #8869
+.elseif \w == 12870
+ qpel_load_32b \v871
+ qpel_filter_\v\()_32b872
+ vps_end873
+ str q17, x7, #16874
+ add x6, x0, #8875
+ qpel_load_32b \v876
+ qpel_filter_\v\()_32b877
+ vps_end878
+ str d17, x7, #8879
+ add x9, x9, #12880
+.else881
+ qpel_load_64b \v882
+ qpel_filter_\v\()_64b883
+ vps_end884
+ sub v18.8h, v18.8h, v31.8h885
+ stp q17, q18, x7, #32886
+ add x9, x9, #16887
+.endif888
+ cmp x9, #\w889
+ blt .loop_ps_w8_\v\()_\w\()x\h890
+ add x0, x0, x1891
+ add x2, x2, x3892
+ sub x5, x5, #1893
+ cbnz x5, .loop_ps_\v\()_\w\()x\h894
+ ret895
+.endm896
+897
+.macro vsp_end898
+ add v17.4s, v17.4s, v31.4s899
+ add v18.4s, v18.4s, v31.4s900
+ sqshrun v17.4h, v17.4s, #12901
+ sqshrun2 v17.8h, v18.4s, #12902
+ sqxtun v17.8b, v17.8h903
+.endm904
+905
+.macro FILTER_VSP w, h, v906
+ lsl x1, x1, #1907
+ lsl x10, x1, #2 // x10 = 4 * x1908
+ sub x11, x10, x1 // x11 = 3 * x1909
+ sub x0, x0, x11910
+ mov x5, #\h911
+ mov w12, #1912
+ lsl w12, w12, #19913
+ add w12, w12, #2048914
+ dup v31.4s, w12915
+ mov x12, #\w916
+ lsl x12, x12, #1917
+ qpel_start_\v\()_1918
+.loop_luma_vsp_\v\()_\w\()x\h:919
+ mov x7, x2920
+ mov x9, #0921
+.loop_luma_vsp_w8_\v\()_\w\()x\h:922
+ add x6, x0, x9923
+ qpel_load_64b \v924
+ qpel_filter_\v\()_32b_1925
+ vsp_end926
+ str d17, x7, #8927
+ add x9, x9, #16928
+.if \w == 12929
+ add x6, x0, #16930
+ qpel_load_64b \v931
+ qpel_filter_\v\()_32b_1932
+ vsp_end933
+ str s17, x7, #4934
+ add x9, x9, #8935
+.endif936
+ cmp x9, x12937
+ blt .loop_luma_vsp_w8_\v\()_\w\()x\h938
+ add x0, x0, x1939
+ add x2, x2, x3940
+ sub x5, x5, #1941
+ cbnz x5, .loop_luma_vsp_\v\()_\w\()x\h942
+ ret943
+.endm944
+945
+.macro vss_end946
+ sshr v17.4s, v17.4s, #6947
+ sshr v18.4s, v18.4s, #6948
+ uzp1 v17.8h, v17.8h, v18.8h949
+.endm950
+951
+.macro FILTER_VSS w, h, v952
+ lsl x1, x1, #1953
+ lsl x10, x1, #2 // x10 = 4 * x1954
+ sub x11, x10, x1 // x11 = 3 * x1955
+ sub x0, x0, x11956
+ lsl x3, x3, #1957
+ mov x5, #\h958
+ mov x12, #\w959
+ lsl x12, x12, #1960
+ qpel_start_\v\()_1961
+.loop_luma_vss_\v\()_\w\()x\h:962
+ mov x7, x2963
+ mov x9, #0964
+.loop_luma_vss_w8_\v\()_\w\()x\h:965
+ add x6, x0, x9966
+ qpel_load_64b \v967
+ qpel_filter_\v\()_32b_1968
+ vss_end969
+.if \w == 4970
+ str s17, x7, #4971
+ add x9, x9, #4972
+.else973
+ str q17, x7, #16974
+ add x9, x9, #16975
+.if \w == 12976
+ add x6, x0, x9977
+ qpel_load_64b \v978
+ qpel_filter_\v\()_32b_1979
+ vss_end980
+ str d17, x7, #8981
+ add x9, x9, #8982
+.endif983
+.endif984
+ cmp x9, x12985
+ blt .loop_luma_vss_w8_\v\()_\w\()x\h986
+ add x0, x0, x1987
+ add x2, x2, x3988
+ sub x5, x5, #1989
+ cbnz x5, .loop_luma_vss_\v\()_\w\()x\h990
+ ret991
+.endm992
+993
+.macro hpp_end994
+ add v17.8h, v17.8h, v31.8h995
+ sqshrun v17.8b, v17.8h, #6996
+.endm997
+998
+.macro FILTER_HPP w, h, v999
+ mov w6, #\h1000
+ sub x3, x3, #\w1001
+ mov w12, #321002
+ dup v31.8h, w121003
+ qpel_start_\v1004
+.if \w == 41005
+.rept \h1006
+ mov x11, x01007
+ sub x11, x11, #41008
+ vextin8 \v1009
+ qpel_filter_\v\()_32b1010
+ hpp_end1011
+ str s17, x2, #41012
+ add x0, x0, x11013
+ add x2, x2, x31014
+.endr1015
+ ret1016
+.else1017
+.loop1_hpp_\v\()_\w\()x\h:1018
+ mov x7, #\w1019
+ mov x11, x01020
+ sub x11, x11, #41021
+.loop2_hpp_\v\()_\w\()x\h:1022
+ vextin8 \v1023
+ qpel_filter_\v\()_32b1024
+ hpp_end1025
+ str d17, x2, #81026
+ sub x11, x11, #81027
+ sub x7, x7, #81028
+.if \w == 121029
+ vextin8 \v1030
+ qpel_filter_\v\()_32b1031
+ hpp_end1032
+ str s17, x2, #41033
+ sub x7, x7, #41034
+.endif1035
+ cbnz x7, .loop2_hpp_\v\()_\w\()x\h1036
+ sub x6, x6, #11037
+ add x0, x0, x11038
+ add x2, x2, x31039
+ cbnz x6, .loop1_hpp_\v\()_\w\()x\h1040
+ ret1041
+.endif1042
+.endm1043
+1044
+.macro hps_end1045
+ sub v17.8h, v17.8h, v31.8h1046
+.endm1047
+1048
+.macro FILTER_HPS w, h, v1049
+ sub x3, x3, #\w1050
+ lsl x3, x3, #11051
+ mov w12, #81921052
+ dup v31.8h, w121053
+ qpel_start_\v1054
+.if \w == 41055
+.loop_hps_\v\()_\w\()x\h\():1056
+ mov x11, x01057
+ sub x11, x11, #41058
+ vextin8 \v1059
+ qpel_filter_\v\()_32b1060
+ hps_end1061
+ str d17, x2, #81062
+ sub w6, w6, #11063
+ add x0, x0, x11064
+ add x2, x2, x31065
+ cbnz w6, .loop_hps_\v\()_\w\()x\h1066
+ ret1067
+.else1068
+.loop1_hps_\v\()_\w\()x\h\():1069
+ mov w7, #\w1070
+ mov x11, x01071
+ sub x11, x11, #41072
+.loop2_hps_\v\()_\w\()x\h\():1073
+.if \w == 8 || \w == 12 || \w == 241074
+ vextin8 \v1075
+ qpel_filter_\v\()_32b1076
+ hps_end1077
+ str q17, x2, #161078
+ sub w7, w7, #81079
+ sub x11, x11, #81080
+.if \w == 121081
+ vextin8 \v1082
+ qpel_filter_\v\()_32b1083
+ hps_end1084
+ str d17, x2, #81085
+ sub w7, w7, #41086
+.endif1087
+.elseif \w == 16 || \w == 32 || \w == 48 || \w == 641088
+ vextin8_64 \v1089
+ qpel_filter_\v\()_64b1090
+ hps_end1091
+ sub v18.8h, v18.8h, v31.8h1092
+ stp q17, q18, x2, #321093
+ sub w7, w7, #161094
+ sub x11, x11, #161095
+.endif1096
+ cbnz w7, .loop2_hps_\v\()_\w\()x\h1097
+ sub w6, w6, #11098
+ add x0, x0, x11099
+ add x2, x2, x31100
+ cbnz w6, .loop1_hps_\v\()_\w\()x\h1101
+ ret1102
+.endif1103
+.endm1104
+1105
+.macro FILTER_CHROMA_VPP w, h, v1106
+ qpel_start_chroma_\v1107
+ mov w12, #321108
+ dup v31.8h, w121109
+ sub x0, x0, x11110
+ mov x5, #\h1111
+.loop_chroma_vpp_\v\()_\w\()x\h:1112
+ mov x7, x21113
+ mov x9, #01114
+.loop_chroma_vpp_w8_\v\()_\w\()x\h:1115
+ add x6, x0, x91116
+ qpel_chroma_load_32b \v1117
+ qpel_filter_chroma_\v\()_32b1118
+ vpp_end1119
+ add x9, x9, #81120
+.if \w == 21121
+ fmov w12, s171122
+ strh w12, x7, #21123
+.elseif \w == 41124
+ str s17, x7, #41125
+.elseif \w == 61126
+ str s17, x7, #41127
+ umov w12, v17.h21128
+ strh w12, x7, #21129
+.elseif \w == 121130
+ str d17, x7, #81131
+ add x6, x0, x91132
+ qpel_chroma_load_32b \v1133
+ qpel_filter_chroma_\v\()_32b1134
+ vpp_end1135
+ str s17, x7, #41136
+ add x9, x9, #81137
+.else1138
+ str d17, x7, #81139
+.endif1140
+ cmp x9, #\w1141
+ blt .loop_chroma_vpp_w8_\v\()_\w\()x\h1142
+ add x0, x0, x11143
+ add x2, x2, x31144
+ sub x5, x5, #11145
+ cbnz x5, .loop_chroma_vpp_\v\()_\w\()x\h1146
+ ret1147
+.endm1148
+1149
+.macro FILTER_CHROMA_VPS w, h, v1150
+ qpel_start_chroma_\v1151
+ mov w12, #81921152
+ dup v31.8h, w121153
+ lsl x3, x3, #11154
+ sub x0, x0, x11155
+ mov x5, #\h1156
+.loop_vps_\v\()_\w\()x\h:1157
+ mov x7, x21158
+ mov x9, #01159
+.loop_vps_w8_\v\()_\w\()x\h:1160
+ add x6, x0, x91161
+ qpel_chroma_load_32b \v1162
+ qpel_filter_chroma_\v\()_32b1163
+ vps_end1164
+ add x9, x9, #81165
+.if \w == 21166
+ str s17, x7, #41167
+.elseif \w == 41168
+ str d17, x7, #81169
+.elseif \w == 61170
+ str d17, x7, #81171
+ st1 {v17.s}2, x7, #41172
+.elseif \w == 121173
+ str q17, x7, #161174
+ add x6, x0, x91175
+ qpel_chroma_load_32b \v1176
+ qpel_filter_chroma_\v\()_32b1177
+ vps_end1178
+ str d17, x7, #81179
+ add x9, x9, #81180
+.else1181
+ str q17, x7, #161182
+.endif1183
+ cmp x9, #\w1184
+ blt .loop_vps_w8_\v\()_\w\()x\h1185
+1186
+ add x0, x0, x11187
+ add x2, x2, x31188
+ sub x5, x5, #11189
+ cbnz x5, .loop_vps_\v\()_\w\()x\h1190
+ ret1191
+.endm1192
+1193
+.macro FILTER_CHROMA_VSP w, h, v1194
+ lsl x1, x1, #11195
+ sub x0, x0, x11196
+ mov x5, #\h1197
+ mov w12, #11198
+ lsl w12, w12, #191199
+ add w12, w12, #20481200
+ dup v31.4s, w121201
+ mov x12, #\w1202
+ lsl x12, x12, #11203
+ qpel_start_chroma_\v\()_11204
+.loop_vsp_\v\()_\w\()x\h:1205
+ mov x7, x21206
+ mov x9, #01207
+.loop_vsp_w8_\v\()_\w\()x\h:1208
+ add x6, x0, x91209
+ qpel_chroma_load_64b \v1210
+ qpel_filter_chroma_\v\()_32b_11211
+ vsp_end1212
+ add x9, x9, #161213
+.if \w == 41214
+ str s17, x7, #41215
+.elseif \w == 121216
+ str d17, x7, #81217
+ add x6, x0, x91218
+ qpel_chroma_load_64b \v1219
+ qpel_filter_chroma_\v\()_32b_11220
+ vsp_end1221
+ str s17, x7, #41222
+ add x9, x9, #81223
+.else1224
+ str d17, x7, #81225
+.endif1226
+ cmp x9, x121227
+ blt .loop_vsp_w8_\v\()_\w\()x\h1228
+ add x0, x0, x11229
+ add x2, x2, x31230
+ sub x5, x5, #11231
+ cbnz x5, .loop_vsp_\v\()_\w\()x\h1232
+ ret1233
+.endm1234
+1235
+.macro FILTER_CHROMA_VSS w, h, v1236
+ lsl x1, x1, #11237
+ sub x0, x0, x11238
+ lsl x3, x3, #11239
+ mov x5, #\h1240
+ mov x12, #\w1241
+ lsl x12, x12, #11242
+ qpel_start_chroma_\v\()_11243
+.loop_vss_\v\()_\w\()x\h:1244
+ mov x7, x21245
+ mov x9, #01246
+.if \w == 41247
+.rept 21248
+ add x6, x0, x91249
+ qpel_chroma_load_64b \v1250
+ qpel_filter_chroma_\v\()_32b_11251
+ vss_end1252
+ str s17, x7, #41253
+ add x9, x9, #41254
+.endr1255
+.else1256
+.loop_vss_w8_\v\()_\w\()x\h:1257
+ add x6, x0, x91258
+ qpel_chroma_load_64b \v1259
+ qpel_filter_chroma_\v\()_32b_11260
+ vss_end1261
+ str q17, x7, #161262
+ add x9, x9, #161263
+.if \w == 121264
+ add x6, x0, x91265
+ qpel_chroma_load_64b \v1266
+ qpel_filter_chroma_\v\()_32b_11267
+ vss_end1268
+ str d17, x7, #81269
+ add x9, x9, #81270
+.endif1271
+ cmp x9, x121272
+ blt .loop_vss_w8_\v\()_\w\()x\h1273
+.endif1274
+ add x0, x0, x11275
+ add x2, x2, x31276
+ sub x5, x5, #11277
+ cbnz x5, .loop_vss_\v\()_\w\()x\h1278
+ ret1279
+.endm1280
+1281
+.macro FILTER_CHROMA_HPP w, h, v1282
+ qpel_start_chroma_\v1283
+ mov w12, #321284
+ dup v31.8h, w121285
+ mov w6, #\h1286
+ sub x3, x3, #\w1287
+.if \w == 2 || \w == 4 || \w == 6 || \w == 121288
+.loop4_chroma_hpp_\v\()_\w\()x\h:1289
+ mov x11, x01290
+ sub x11, x11, #21291
+ vextin8_chroma \v1292
+ qpel_filter_chroma_\v\()_32b1293
+ hpp_end1294
+.if \w == 21295
+ fmov w12, s171296
+ strh w12, x2, #21297
+.elseif \w == 41298
+ str s17, x2, #41299
+.elseif \w == 61300
+ str s17, x2, #41301
+ umov w12, v17.h21302
+ strh w12, x2, #21303
+.elseif \w == 121304
+ str d17, x2, #81305
+ sub x11, x11, #81306
+ vextin8_chroma \v1307
+ qpel_filter_chroma_\v\()_32b1308
+ hpp_end1309
+ str s17, x2, #41310
+.endif1311
+ sub w6, w6, #11312
+ add x0, x0, x11313
+ add x2, x2, x31314
+ cbnz w6, .loop4_chroma_hpp_\v\()_\w\()x\h1315
+ ret1316
+.else1317
+.loop2_chroma_hpp_\v\()_\w\()x\h:1318
+ mov x7, #\w1319
+ lsr x7, x7, #31320
+ mov x11, x01321
+ sub x11, x11, #21322
+.loop3_chroma_hpp_\v\()_\w\()x\h:1323
+.if \w == 8 || \w == 241324
+ vextin8_chroma \v1325
+ qpel_filter_chroma_\v\()_32b1326
+ hpp_end1327
+ str d17, x2, #81328
+ sub x7, x7, #11329
+ sub x11, x11, #81330
+.elseif \w == 16 || \w == 32 || \w == 48 || \w == 641331
+ vextin8_chroma_64 \v1332
+ qpel_filter_chroma_\v\()_64b1333
+ hpp_end1334
+ add v18.8h, v18.8h, v31.8h1335
+ sqshrun2 v17.16b, v18.8h, #61336
+ str q17, x2, #161337
+ sub x7, x7, #21338
+ sub x11, x11, #161339
+.endif1340
+ cbnz x7, .loop3_chroma_hpp_\v\()_\w\()x\h1341
+ sub w6, w6, #11342
+ add x0, x0, x11343
+ add x2, x2, x31344
+ cbnz w6, .loop2_chroma_hpp_\v\()_\w\()x\h1345
+ ret1346
+.endif1347
+.endm1348
+1349
+.macro CHROMA_HPS_2_4_6_12 w, v1350
+ mov x11, x01351
+ sub x11, x11, #21352
+ vextin8_chroma \v1353
+ qpel_filter_chroma_\v\()_32b1354
+ hps_end1355
+ sub x11, x11, #81356
+.if \w == 21357
+ str s17, x2, #41358
+.elseif \w == 41359
+ str d17, x2, #81360
+.elseif \w == 61361
+ str d17, x2, #81362
+ st1 {v17.s}2, x2, #41363
+.elseif \w == 121364
+ str q17, x2, #161365
+ vextin8_chroma \v1366
+ qpel_filter_chroma_\v\()_32b1367
+ sub v17.8h, v17.8h, v31.8h1368
+ str d17, x2, #81369
+.endif1370
+ add x0, x0, x11371
+ add x2, x2, x31372
+.endm1373
+1374
+.macro FILTER_CHROMA_HPS w, h, v1375
+ qpel_start_chroma_\v1376
+ mov w12, #81921377
+ dup v31.8h, w121378
+ sub x3, x3, #\w1379
+ lsl x3, x3, #11380
+1381
+.if \w == 2 || \w == 4 || \w == 6 || \w == 121382
+ cmp x5, #01383
+ beq 0f1384
+ sub x0, x0, x11385
+.rept 31386
+ CHROMA_HPS_2_4_6_12 \w, \v1387
+.endr1388
+0:1389
+.rept \h1390
+ CHROMA_HPS_2_4_6_12 \w, \v1391
+.endr1392
+ ret1393
+.else1394
+ mov w10, #\h1395
+ cmp x5, #01396
+ beq 9f1397
+ sub x0, x0, x11398
+ add w10, w10, #31399
+9:1400
+ mov w6, w101401
+.loop1_chroma_hps_\v\()_\w\()x\h\():1402
+ mov x7, #\w1403
+ lsr x7, x7, #31404
+ mov x11, x01405
+ sub x11, x11, #21406
+.loop2_chroma_hps_\v\()_\w\()x\h\():1407
+.if \w == 8 || \w == 241408
+ vextin8_chroma \v1409
+ qpel_filter_chroma_\v\()_32b1410
+ hps_end1411
+ str q17, x2, #161412
+ sub x7, x7, #11413
+ sub x11, x11, #81414
+.elseif \w == 16 || \w == 32 || \w == 48 || \w == 641415
+ vextin8_chroma_64 \v1416
+ qpel_filter_chroma_\v\()_64b1417
+ hps_end1418
+ sub v18.8h, v18.8h, v31.8h1419
+ stp q17, q18, x2, #321420
+ sub x7, x7, #21421
+ sub x11, x11, #161422
+.endif1423
+ cbnz x7, .loop2_chroma_hps_\v\()_\w\()x\h\()1424
+ sub w6, w6, #11425
+ add x0, x0, x11426
+ add x2, x2, x31427
+ cbnz w6, .loop1_chroma_hps_\v\()_\w\()x\h\()1428
+ ret1429
+.endif1430
+.endm1431
+1432
+const g_lumaFilter, align=81433
+.word 0,0,0,0,0,0,64,64,0,0,0,0,0,0,0,01434
+.word -1,-1,4,4,-10,-10,58,58,17,17,-5,-5,1,1,0,01435
+.word -1,-1,4,4,-11,-11,40,40,40,40,-11,-11,4,4,-1,-11436
+.word 0,0,1,1,-5,-5,17,17,58,58,-10,-10,4,4,-1,-11437
+endconst1438
x265_3.6.tar.gz/source/common/aarch64/ipfilter-sve2.S
Added
1284
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// Functions in this file:26
+// ***** luma_vpp *****27
+// ***** luma_vps *****28
+// ***** luma_vsp *****29
+// ***** luma_vss *****30
+// ***** luma_hpp *****31
+// ***** luma_hps *****32
+// ***** chroma_vpp *****33
+// ***** chroma_vps *****34
+// ***** chroma_vsp *****35
+// ***** chroma_vss *****36
+// ***** chroma_hpp *****37
+// ***** chroma_hps *****38
+39
+#include "asm-sve.S"40
+#include "ipfilter-common.S"41
+42
+.arch armv8-a+sve243
+44
+#ifdef __APPLE__45
+.section __RODATA,__rodata46
+#else47
+.section .rodata48
+#endif49
+50
+.align 451
+52
+.text53
+54
+.macro qpel_load_32b_sve2 v55
+.if \v == 056
+ add x6, x6, x11 // do not load 3 values that are not used in qpel_filter_057
+ ld1b {z3.h}, p0/z, x658
+ add x6, x6, x159
+.elseif \v == 1 || \v == 2 || \v == 360
+.if \v != 3 // not used in qpel_filter_361
+ ld1b {z0.h}, p0/z, x662
+ add x6, x6, x163
+.else64
+ add x6, x6, x165
+.endif66
+ ld1b {z1.h}, p0/z, x667
+ add x6, x6, x168
+ ld1b {z2.h}, p0/z, x669
+ add x6, x6, x170
+ ld1b {z3.h}, p0/z, x671
+ add x6, x6, x172
+ ld1b {z4.h}, p0/z, x673
+ add x6, x6, x174
+ ld1b {z5.h}, p0/z, x675
+ add x6, x6, x176
+.if \v != 1 // not used in qpel_filter_177
+ ld1b {z6.h}, p0/z, x678
+ add x6, x6, x179
+ ld1b {z7.h}, p0/z, x680
+.else81
+ ld1b {z6.h}, p0/z, x682
+.endif83
+.endif84
+.endm85
+86
+.macro qpel_load_64b_sve2_gt_16 v87
+.if \v == 088
+ add x6, x6, x11 // do not load 3 values that are not used in qpel_filter_089
+ ld1b {z3.h}, p2/z, x690
+ add x6, x6, x191
+.elseif \v == 1 || \v == 2 || \v == 392
+.if \v != 3 // not used in qpel_filter_393
+ ld1b {z0.h}, p2/z, x694
+ add x6, x6, x195
+.else96
+ add x6, x6, x197
+.endif98
+ ld1b {z1.h}, p2/z, x699
+ add x6, x6, x1100
+ ld1b {z2.h}, p2/z, x6101
+ add x6, x6, x1102
+ ld1b {z3.h}, p2/z, x6103
+ add x6, x6, x1104
+ ld1b {z4.h}, p2/z, x6105
+ add x6, x6, x1106
+ ld1b {z5.h}, p2/z, x6107
+ add x6, x6, x1108
+.if \v != 1 // not used in qpel_filter_1109
+ ld1b {z6.h}, p2/z, x6110
+ add x6, x6, x1111
+ ld1b {z7.h}, p2/z, x6112
+.else113
+ ld1b {z6.h}, p2/z, x6114
+.endif115
+.endif116
+.endm117
+118
+.macro qpel_chroma_load_32b_sve2 v119
+.if \v == 0120
+ // qpel_filter_chroma_0 only uses values in v1121
+ add x6, x6, x1122
+ ld1b {z1.h}, p0/z, x6123
+.else124
+ ld1b {z0.h}, p0/z, x6125
+ add x6, x6, x1126
+ ld1b {z1.h}, p0/z, x6127
+ add x6, x6, x1128
+ ld1b {z2.h}, p0/z, x6129
+ add x6, x6, x1130
+ ld1b {z3.h}, p0/z, x6131
+.endif132
+.endm133
+134
+.macro qpel_start_sve2_0135
+ mov z24.h, #64136
+.endm137
+138
+.macro qpel_filter_sve2_0_32b139
+ mul z17.h, z3.h, z24.h // 64*d140
+.endm141
+142
+.macro qpel_filter_sve2_0_64b143
+ qpel_filter_sve2_0_32b144
+ mul z18.h, z11.h, z24.h145
+.endm146
+147
+.macro qpel_start_sve2_1148
+ mov z24.h, #58149
+ mov z25.h, #10150
+ mov z26.h, #17151
+ mov z27.h, #5152
+.endm153
+154
+.macro qpel_filter_sve2_1_32b155
+ mul z19.h, z2.h, z25.h // c*10156
+ mul z17.h, z3.h, z24.h // d*58157
+ mul z21.h, z4.h, z26.h // e*17158
+ mul z23.h, z5.h, z27.h // f*5159
+ sub z17.h, z17.h, z19.h // d*58 - c*10160
+ lsl z18.h, z1.h, #2 // b*4161
+ add z17.h, z17.h, z21.h // d*58 - c*10 + e*17162
+ sub z21.h, z6.h, z0.h // g - a163
+ add z17.h, z17.h, z18.h // d*58 - c*10 + e*17 + b*4164
+ sub z21.h, z21.h, z23.h // g - a - f*5165
+ add z17.h, z17.h, z21.h // d*58 - c*10 + e*17 + b*4 + g - a - f*5166
+.endm167
+168
+.macro qpel_filter_sve2_1_64b169
+ qpel_filter_sve2_1_32b170
+ mul z20.h, z10.h, z25.h // c*10171
+ mul z18.h, z11.h, z24.h // d*58172
+ mul z21.h, z12.h, z26.h // e*17173
+ mul z23.h, z13.h, z27.h // f*5174
+ sub z18.h, z18.h, z20.h // d*58 - c*10175
+ lsl z28.h, z30.h, #2 // b*4176
+ add z18.h, z18.h, z21.h // d*58 - c*10 + e*17177
+ sub z21.h, z14.h, z29.h // g - a178
+ add z18.h, z18.h, z28.h // d*58 - c*10 + e*17 + b*4179
+ sub z21.h, z21.h, z23.h // g - a - f*5180
+ add z18.h, z18.h, z21.h // d*58 - c*10 + e*17 + b*4 + g - a - f*5181
+.endm182
+183
+.macro qpel_start_sve2_2184
+ mov z24.h, #11185
+ mov z25.h, #40186
+.endm187
+188
+.macro qpel_filter_sve2_2_32b189
+ add z17.h, z3.h, z4.h // d + e190
+ add z19.h, z2.h, z5.h // c + f191
+ add z23.h, z1.h, z6.h // b + g192
+ add z21.h, z0.h, z7.h // a + h193
+ mul z17.h, z17.h, z25.h // 40 * (d + e)194
+ mul z19.h, z19.h, z24.h // 11 * (c + f)195
+ lsl z23.h, z23.h, #2 // (b + g) * 4196
+ add z19.h, z19.h, z21.h // 11 * (c + f) + a + h197
+ add z17.h, z17.h, z23.h // 40 * (d + e) + (b + g) * 4198
+ sub z17.h, z17.h, z19.h // 40 * (d + e) + (b + g) * 4 - 11 * (c + f) - a - h199
+.endm200
+201
+.macro qpel_filter_sve2_2_64b202
+ qpel_filter_sve2_2_32b203
+ add z27.h, z11.h, z12.h // d + e204
+ add z16.h, z10.h, z13.h // c + f205
+ add z23.h, z30.h, z14.h // b + g206
+ add z21.h, z29.h, z15.h // a + h207
+ mul z27.h, z27.h, z25.h // 40 * (d + e)208
+ mul z16.h, z16.h, z24.h // 11 * (c + f)209
+ lsl z23.h, z23.h, #2 // (b + g) * 4210
+ add z16.h, z16.h, z21.h // 11 * (c + f) + a + h211
+ add z27.h, z27.h, z23.h // 40 * (d + e) + (b + g) * 4212
+ sub z18.h, z27.h, z16.h // 40 * (d + e) + (b + g) * 4 - 11 * (c + f) - a - h213
+.endm214
+215
+.macro qpel_start_sve2_3216
+ mov z24.h, #17217
+ mov z25.h, #5218
+ mov z26.h, #58219
+ mov z27.h, #10220
+.endm221
+222
+.macro qpel_filter_sve2_3_32b223
+ mul z19.h, z2.h, z25.h // c * 5224
+ mul z17.h, z3.h, z24.h // d * 17225
+ mul z21.h, z4.h, z26.h // e * 58226
+ mul z23.h, z5.h, z27.h // f * 10227
+ sub z17.h, z17.h, z19.h // d * 17 - c * 5228
+ lsl z19.h, z6.h, #2 // g * 4229
+ add z17.h, z17.h, z21.h // d * 17 - c * 5 + e * 58230
+ sub z21.h, z1.h, z7.h // b - h231
+ add z17.h, z17.h, z19.h // d * 17 - c * 5 + e * 58 + g * 4232
+ sub z21.h, z21.h, z23.h // b - h - f * 10233
+ add z17.h, z17.h, z21.h // d * 17 - c * 5 + e * 58 + g * 4 + b - h - f * 10234
+.endm235
+236
+.macro qpel_filter_sve2_3_64b237
+ qpel_filter_sve2_3_32b238
+ mul z16.h, z10.h, z25.h // c * 5239
+ mul z18.h, z11.h, z24.h // d * 17240
+ mul z21.h, z12.h, z26.h // e * 58241
+ mul z23.h, z13.h, z27.h // f * 10242
+ sub z18.h, z18.h, z16.h // d * 17 - c * 5243
+ lsl z16.h, z14.h, #2 // g * 4244
+ add z18.h, z18.h, z21.h // d * 17 - c * 5 + e * 58245
+ sub z21.h, z30.h, z15.h // b - h246
+ add z18.h, z18.h, z16.h // d * 17 - c * 5 + e * 58 + g * 4247
+ sub z21.h, z21.h, z23.h // b - h - f * 10248
+ add z18.h, z18.h, z21.h // d * 17 - c * 5 + e * 58 + g * 4 + b - h - f * 10249
+.endm250
+251
+.macro qpel_start_chroma_sve2_0252
+ mov z29.h, #64253
+.endm254
+255
+.macro qpel_filter_chroma_sve2_0_32b256
+ mul z17.h, z1.h, z29.h // 64*b257
+.endm258
+259
+.macro qpel_start_chroma_sve2_1260
+ mov z29.h, #58261
+ mov z30.h, #10262
+.endm263
+264
+.macro qpel_filter_chroma_sve2_1_32b265
+ mul z17.h, z1.h, z29.h // 58 * b266
+ mul z19.h, z2.h, z30.h // 10 * c267
+ add z22.h, z0.h, z3.h // a + d268
+ lsl z22.h, z22.h, #1 // 2 * (a+d)269
+ sub z17.h, z17.h, z22.h // 58*b - 2*(a+d)270
+ add z17.h, z17.h, z19.h // 58*b-2*(a+d) + 10*c271
+.endm272
+273
+.macro qpel_start_chroma_sve2_2274
+ mov z30.h, #54275
+.endm276
+277
+.macro qpel_filter_chroma_sve2_2_32b278
+ mul z17.h, z1.h, z30.h // 54 * b279
+ lsl z19.h, z0.h, #2 // 4 * a280
+ lsl z21.h, z2.h, #4 // 16 * c281
+ lsl z23.h, z3.h, #1 // 2 * d282
+ add z17.h, z17.h, z21.h // 54*b + 16*c283
+ add z19.h, z19.h, z23.h // 4*a + 2*d284
+ sub z17.h, z17.h, z19.h // 54*b+16*c - (4*a+2*d)285
+.endm286
+287
+.macro qpel_start_chroma_sve2_3288
+ mov z28.h, #46289
+ mov z29.h, #28290
+ mov z30.h, #6291
+.endm292
+293
+.macro qpel_filter_chroma_sve2_3_32b294
+ mul z17.h, z1.h, z28.h // 46 * b295
+ mul z19.h, z2.h, z29.h // 28 * c296
+ lsl z21.h, z3.h, #2 // 4 * d297
+ mul z23.h, z0.h, z30.h // 6 * a298
+ add z17.h, z17.h, z19.h // 46*b + 28*c299
+ add z21.h, z21.h, z23.h // 4*d + 6*a300
+ sub z17.h, z17.h, z21.h // 46*b+28*c - (4*d+6*a)301
+.endm302
+303
+.macro qpel_start_chroma_sve2_4304
+ mov z29.h, #36305
+.endm306
+307
+.macro qpel_filter_chroma_sve2_4_32b308
+ add z20.h, z0.h, z3.h // a + d309
+ add z17.h, z1.h, z2.h // b + c310
+ lsl z20.h, z20.h, #2 // 4 * (a+d)311
+ mul z17.h, z17.h, z29.h // 36 * (b+c)312
+ sub z17.h, z17.h, z20.h // 36*(b+c) - 4*(a+d)313
+.endm314
+315
+.macro qpel_start_chroma_sve2_5316
+ mov z28.h, #28317
+ mov z29.h, #46318
+ mov z30.h, #6319
+.endm320
+321
+.macro qpel_filter_chroma_sve2_5_32b322
+ mul z17.h, z1.h, z28.h // 28 * b323
+ mul z19.h, z2.h, z29.h // 46 * c324
+ lsl z21.h, z0.h, #2 // 4 * a325
+ mul z23.h, z3.h, z30.h // 6 * d326
+ add z17.h, z17.h, z19.h // 28*b + 46*c327
+ add z21.h, z21.h, z23.h // 4*a + 6*d328
+ sub z17.h, z17.h, z21.h // 28*b+46*c - (4*a+6*d)329
+.endm330
+331
+.macro qpel_start_chroma_sve2_6332
+ mov z30.h, #54333
+.endm334
+335
+.macro qpel_filter_chroma_sve2_6_32b336
+ mul z17.h, z2.h, z30.h // 54 * c337
+ lsl z19.h, z0.h, #1 // 2 * a338
+ lsl z21.h, z1.h, #4 // 16 * b339
+ lsl z23.h, z3.h, #2 // 4 * d340
+ add z17.h, z17.h, z21.h // 54*c + 16*b341
+ add z19.h, z19.h, z23.h // 2*a + 4*d342
+ sub z17.h, z17.h, z19.h // 54*c+16*b - (2*a+4*d)343
+.endm344
+345
+.macro qpel_start_chroma_sve2_7346
+ mov z29.h, #58347
+ mov z30.h, #10348
+.endm349
+350
+.macro qpel_filter_chroma_sve2_7_32b351
+ add z20.h, z0.h, z3.h // a + d352
+ mul z17.h, z2.h, z29.h // 58 * c353
+ lsl z20.h, z20.h, #1 // 2 * (a+d)354
+ mul z19.h, z1.h, z30.h // 10 * b355
+ sub z17.h, z17.h, z20.h // 58*c - 2*(a+d)356
+ add z17.h, z17.h, z19.h // 58*c-2*(a+d) + 10*b357
+.endm358
+359
+.macro vpp_end_sve2360
+ add z17.h, z17.h, z31.h361
+ sqshrun v17.8b, v17.8h, #6362
+.endm363
+364
+.macro FILTER_LUMA_VPP_SVE2 w, h, v365
+ lsl x10, x1, #2 // x10 = 4 * x1366
+ sub x11, x10, x1 // x11 = 3 * x1367
+ sub x0, x0, x11 // src -= (8 / 2 - 1) * srcStride368
+ mov x5, #\h369
+ mov z31.h, #32370
+ rdvl x9, #1371
+ cmp x9, #16372
+ bgt .vl_gt_16_FILTER_LUMA_VPP_\v\()_\w\()x\h373
+ qpel_start_\v374
+.loop_luma_vpp_sve2_\v\()_\w\()x\h:375
+ mov x7, x2376
+ mov x9, #0377
+.loop_luma_vpp_w8_sve2_\v\()_\w\()x\h:378
+ add x6, x0, x9379
+.if \w == 8 || \w == 24380
+ qpel_load_32b \v381
+ qpel_filter_\v\()_32b382
+ vpp_end383
+ str d17, x7, #8384
+ add x9, x9, #8385
+.elseif \w == 12386
+ qpel_load_32b \v387
+ qpel_filter_\v\()_32b388
+ vpp_end389
+ str d17, x7, #8390
+ add x6, x0, #8391
+ qpel_load_32b \v392
+ qpel_filter_\v\()_32b393
+ vpp_end394
+ fmov w6, s17395
+ str w6, x7, #4396
+ add x9, x9, #12397
+.else398
+ qpel_load_64b \v399
+ qpel_filter_\v\()_64b400
+ vpp_end401
+ add v18.8h, v18.8h, v31.8h402
+ sqshrun2 v17.16b, v18.8h, #6403
+ str q17, x7, #16404
+ add x9, x9, #16405
+.endif406
+ cmp x9, #\w407
+ blt .loop_luma_vpp_w8_sve2_\v\()_\w\()x\h408
+ add x0, x0, x1409
+ add x2, x2, x3410
+ sub x5, x5, #1411
+ cbnz x5, .loop_luma_vpp_sve2_\v\()_\w\()x\h412
+ ret413
+.vl_gt_16_FILTER_LUMA_VPP_\v\()_\w\()x\h:414
+ ptrue p0.h, vl8415
+ ptrue p2.h, vl16416
+ qpel_start_sve2_\v417
+.gt_16_loop_luma_vpp_sve2_\v\()_\w\()x\h:418
+ mov x7, x2419
+ mov x9, #0420
+.gt_16_loop_luma_vpp_w8_sve2_\v\()_\w\()x\h:421
+ add x6, x0, x9422
+.if \w == 8 || \w == 24423
+ qpel_load_32b_sve2 \v424
+ qpel_filter_sve2_\v\()_32b425
+ vpp_end_sve2426
+ str d17, x7, #8427
+ add x9, x9, #8428
+.elseif \w == 12429
+ qpel_load_32b_sve2 \v430
+ qpel_filter_sve2_\v\()_32b431
+ vpp_end_sve2432
+ str d17, x7, #8433
+ add x6, x0, #8434
+ qpel_load_32b_sve2 \v435
+ qpel_filter_sve2_\v\()_32b436
+ vpp_end_sve2437
+ fmov w6, s17438
+ str w6, x7, #4439
+ add x9, x9, #12440
+.else441
+ qpel_load_64b_sve2_gt_16 \v442
+ qpel_filter_sve2_\v\()_32b443
+ vpp_end_sve2444
+ add z18.h, z18.h, z31.h445
+ sqshrun2 v17.16b, v18.8h, #6446
+ str q17, x7, #16447
+ add x9, x9, #16448
+.endif449
+ cmp x9, #\w450
+ blt .gt_16_loop_luma_vpp_w8_sve2_\v\()_\w\()x\h451
+ add x0, x0, x1452
+ add x2, x2, x3453
+ sub x5, x5, #1454
+ cbnz x5, .gt_16_loop_luma_vpp_sve2_\v\()_\w\()x\h455
+ ret456
+.endm457
+458
+// void interp_vert_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)459
+.macro LUMA_VPP_SVE2 w, h460
+function x265_interp_8tap_vert_pp_\w\()x\h\()_sve2461
+ cmp x4, #0462
+ b.eq 0f463
+ cmp x4, #1464
+ b.eq 1f465
+ cmp x4, #2466
+ b.eq 2f467
+ cmp x4, #3468
+ b.eq 3f469
+0:470
+ FILTER_LUMA_VPP_SVE2 \w, \h, 0471
+1:472
+ FILTER_LUMA_VPP_SVE2 \w, \h, 1473
+2:474
+ FILTER_LUMA_VPP_SVE2 \w, \h, 2475
+3:476
+ FILTER_LUMA_VPP_SVE2 \w, \h, 3477
+endfunc478
+.endm479
+480
+LUMA_VPP_SVE2 8, 4481
+LUMA_VPP_SVE2 8, 8482
+LUMA_VPP_SVE2 8, 16483
+LUMA_VPP_SVE2 8, 32484
+LUMA_VPP_SVE2 12, 16485
+LUMA_VPP_SVE2 16, 4486
+LUMA_VPP_SVE2 16, 8487
+LUMA_VPP_SVE2 16, 16488
+LUMA_VPP_SVE2 16, 32489
+LUMA_VPP_SVE2 16, 64490
+LUMA_VPP_SVE2 16, 12491
+LUMA_VPP_SVE2 24, 32492
+LUMA_VPP_SVE2 32, 8493
+LUMA_VPP_SVE2 32, 16494
+LUMA_VPP_SVE2 32, 32495
+LUMA_VPP_SVE2 32, 64496
+LUMA_VPP_SVE2 32, 24497
+LUMA_VPP_SVE2 48, 64498
+LUMA_VPP_SVE2 64, 16499
+LUMA_VPP_SVE2 64, 32500
+LUMA_VPP_SVE2 64, 64501
+LUMA_VPP_SVE2 64, 48502
+503
+// void interp_vert_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)504
+.macro LUMA_VPS_4xN_SVE2 h505
+function x265_interp_8tap_vert_ps_4x\h\()_sve2506
+ lsl x3, x3, #1507
+ lsl x5, x4, #6508
+ lsl x4, x1, #2509
+ sub x4, x4, x1510
+ sub x0, x0, x4511
+512
+ mov z28.s, #8192513
+ mov x4, #\h514
+ movrel x12, g_lumaFilter515
+ add x12, x12, x5516
+ ptrue p0.s, vl4517
+ ld1rd {z16.d}, p0/z, x12518
+ ld1rd {z17.d}, p0/z, x12, #8519
+ ld1rd {z18.d}, p0/z, x12, #16520
+ ld1rd {z19.d}, p0/z, x12, #24521
+ ld1rd {z20.d}, p0/z, x12, #32522
+ ld1rd {z21.d}, p0/z, x12, #40523
+ ld1rd {z22.d}, p0/z, x12, #48524
+ ld1rd {z23.d}, p0/z, x12, #56525
+526
+.loop_vps_sve2_4x\h:527
+ mov x6, x0528
+529
+ ld1b {z0.s}, p0/z, x6530
+ add x6, x6, x1531
+ ld1b {z1.s}, p0/z, x6532
+ add x6, x6, x1533
+ ld1b {z2.s}, p0/z, x6534
+ add x6, x6, x1535
+ ld1b {z3.s}, p0/z, x6536
+ add x6, x6, x1537
+ ld1b {z4.s}, p0/z, x6538
+ add x6, x6, x1539
+ ld1b {z5.s}, p0/z, x6540
+ add x6, x6, x1541
+ ld1b {z6.s}, p0/z, x6542
+ add x6, x6, x1543
+ ld1b {z7.s}, p0/z, x6544
+ add x6, x6, x1545
+546
+ mul z0.s, z0.s, z16.s547
+ mla z0.s, p0/m, z1.s, z17.s548
+ mla z0.s, p0/m, z2.s, z18.s549
+ mla z0.s, p0/m, z3.s, z19.s550
+ mla z0.s, p0/m, z4.s, z20.s551
+ mla z0.s, p0/m, z5.s, z21.s552
+ mla z0.s, p0/m, z6.s, z22.s553
+ mla z0.s, p0/m, z7.s, z23.s554
+555
+ sub z0.s, z0.s, z28.s556
+ sqxtn v0.4h, v0.4s557
+ st1 {v0.8b}, x2, x3558
+559
+ add x0, x0, x1560
+ sub x4, x4, #1561
+ cbnz x4, .loop_vps_sve2_4x\h562
+ ret563
+endfunc564
+.endm565
+566
+LUMA_VPS_4xN_SVE2 4567
+LUMA_VPS_4xN_SVE2 8568
+LUMA_VPS_4xN_SVE2 16569
+570
+// void interp_vert_sp_c(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)571
+.macro LUMA_VSP_4xN_SVE2 h572
+function x265_interp_8tap_vert_sp_4x\h\()_sve2573
+ lsl x5, x4, #6574
+ lsl x1, x1, #1575
+ lsl x4, x1, #2576
+ sub x4, x4, x1577
+ sub x0, x0, x4578
+579
+ mov w12, #1580
+ lsl w12, w12, #19581
+ add w12, w12, #2048582
+ dup v24.4s, w12583
+ mov x4, #\h584
+ movrel x12, g_lumaFilter585
+ add x12, x12, x5586
+587
+ ptrue p0.s, vl4588
+ ld1rd {z16.d}, p0/z, x12589
+ ld1rd {z17.d}, p0/z, x12, #8590
+ ld1rd {z18.d}, p0/z, x12, #16591
+ ld1rd {z19.d}, p0/z, x12, #24592
+ ld1rd {z20.d}, p0/z, x12, #32593
+ ld1rd {z21.d}, p0/z, x12, #40594
+ ld1rd {z22.d}, p0/z, x12, #48595
+ ld1rd {z23.d}, p0/z, x12, #56596
+597
+.loop_vsp_sve2_4x\h:598
+ mov x6, x0599
+600
+ ld1 {v0.8b}, x6, x1601
+ ld1 {v1.8b}, x6, x1602
+ ld1 {v2.8b}, x6, x1603
+ ld1 {v3.8b}, x6, x1604
+ ld1 {v4.8b}, x6, x1605
+ ld1 {v5.8b}, x6, x1606
+ ld1 {v6.8b}, x6, x1607
+ ld1 {v7.8b}, x6, x1608
+609
+ sunpklo z0.s, z0.h610
+ sunpklo z1.s, z1.h611
+ mul z0.s, z0.s, z16.s612
+ sunpklo z2.s, z2.h613
+ mla z0.s, p0/m, z1.s, z17.s614
+ sunpklo z3.s, z3.h615
+ mla z0.s, p0/m, z2.s, z18.s616
+ sunpklo z4.s, z4.h617
+ mla z0.s, p0/m, z3.s, z19.s618
+ sunpklo z5.s, z5.h619
+ mla z0.s, p0/m, z4.s, z20.s620
+ sunpklo z6.s, z6.h621
+ mla z0.s, p0/m, z5.s, z21.s622
+ sunpklo z7.s, z7.h623
+ mla z0.s, p0/m, z6.s, z22.s624
+625
+ mla z0.s, p0/m, z7.s, z23.s626
+627
+ add z0.s, z0.s, z24.s628
+ sqshrun v0.4h, v0.4s, #12629
+ sqxtun v0.8b, v0.8h630
+ st1 {v0.s}0, x2, x3631
+632
+ add x0, x0, x1633
+ sub x4, x4, #1634
+ cbnz x4, .loop_vsp_sve2_4x\h635
+ ret636
+endfunc637
+.endm638
+639
+LUMA_VSP_4xN_SVE2 4640
+LUMA_VSP_4xN_SVE2 8641
+LUMA_VSP_4xN_SVE2 16642
+643
+.macro vps_end_sve2644
+ sub z17.h, z17.h, z31.h645
+.endm646
+647
+.macro FILTER_VPS_SVE2 w, h, v648
+ lsl x3, x3, #1649
+ lsl x10, x1, #2 // x10 = 4 * x1650
+ sub x11, x10, x1 // x11 = 3 * x1651
+ sub x0, x0, x11 // src -= (8 / 2 - 1) * srcStride652
+ mov x5, #\h653
+ mov z31.h, #8192654
+ rdvl x14, #1655
+ cmp x14, #16656
+ bgt .vl_gt_16_FILTER_VPS_\v\()_\w\()x\h657
+ qpel_start_\v658
+.loop_ps_sve2_\v\()_\w\()x\h:659
+ mov x7, x2660
+ mov x9, #0661
+.loop_ps_w8_sve2_\v\()_\w\()x\h:662
+ add x6, x0, x9663
+.if \w == 8 || \w == 24664
+ qpel_load_32b \v665
+ qpel_filter_\v\()_32b666
+ vps_end667
+ str q17, x7, #16668
+ add x9, x9, #8669
+.elseif \w == 12670
+ qpel_load_32b \v671
+ qpel_filter_\v\()_32b672
+ vps_end673
+ str q17, x7, #16674
+ add x6, x0, #8675
+ qpel_load_32b \v676
+ qpel_filter_\v\()_32b677
+ vps_end678
+ str d17, x7, #8679
+ add x9, x9, #12680
+.else681
+ qpel_load_64b \v682
+ qpel_filter_\v\()_64b683
+ vps_end684
+ sub v18.8h, v18.8h, v31.8h685
+ stp q17, q18, x7, #32686
+ add x9, x9, #16687
+.endif688
+ cmp x9, #\w689
+ blt .loop_ps_w8_sve2_\v\()_\w\()x\h690
+ add x0, x0, x1691
+ add x2, x2, x3692
+ sub x5, x5, #1693
+ cbnz x5, .loop_ps_sve2_\v\()_\w\()x\h694
+ ret695
+.vl_gt_16_FILTER_VPS_\v\()_\w\()x\h:696
+ ptrue p0.h, vl8697
+ ptrue p2.h, vl16698
+ qpel_start_sve2_\v699
+.gt_16_loop_ps_sve2_\v\()_\w\()x\h:700
+ mov x7, x2701
+ mov x9, #0702
+.gt_16_loop_ps_w8_sve2_\v\()_\w\()x\h:703
+ add x6, x0, x9704
+.if \w == 8 || \w == 24705
+ qpel_load_32b_sve2 \v706
+ qpel_filter_sve2_\v\()_32b707
+ vps_end_sve2708
+ str q17, x7, #16709
+ add x9, x9, #8710
+.elseif \w == 12711
+ qpel_load_32b_sve2 \v712
+ qpel_filter_sve2_\v\()_32b713
+ vps_end_sve2714
+ str q17, x7, #16715
+ add x6, x0, #8716
+ qpel_load_32b_sve2 \v717
+ qpel_filter_sve2_\v\()_32b718
+ vps_end_sve2719
+ str d17, x7, #8720
+ add x9, x9, #12721
+.else722
+ qpel_load_64b_sve2_gt_16 \v723
+ qpel_filter_sve2_\v\()_32b724
+ vps_end_sve2725
+ sub z18.h, z18.h, z31.h726
+ stp q17, q18, x7, #32727
+ add x9, x9, #16728
+.endif729
+ cmp x9, #\w730
+ blt .gt_16_loop_ps_w8_sve2_\v\()_\w\()x\h731
+ add x0, x0, x1732
+ add x2, x2, x3733
+ sub x5, x5, #1734
+ cbnz x5, .gt_16_loop_ps_sve2_\v\()_\w\()x\h735
+ ret736
+.endm737
+738
+// void interp_vert_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)739
+.macro LUMA_VPS_SVE2 w, h740
+function x265_interp_8tap_vert_ps_\w\()x\h\()_sve2741
+ cmp x4, #0742
+ beq 0f743
+ cmp x4, #1744
+ beq 1f745
+ cmp x4, #2746
+ beq 2f747
+ cmp x4, #3748
+ beq 3f749
+0:750
+ FILTER_VPS_SVE2 \w, \h, 0751
+1:752
+ FILTER_VPS_SVE2 \w, \h, 1753
+2:754
+ FILTER_VPS_SVE2 \w, \h, 2755
+3:756
+ FILTER_VPS_SVE2 \w, \h, 3757
+endfunc758
+.endm759
+760
+LUMA_VPS_SVE2 8, 4761
+LUMA_VPS_SVE2 8, 8762
+LUMA_VPS_SVE2 8, 16763
+LUMA_VPS_SVE2 8, 32764
+LUMA_VPS_SVE2 12, 16765
+LUMA_VPS_SVE2 16, 4766
+LUMA_VPS_SVE2 16, 8767
+LUMA_VPS_SVE2 16, 16768
+LUMA_VPS_SVE2 16, 32769
+LUMA_VPS_SVE2 16, 64770
+LUMA_VPS_SVE2 16, 12771
+LUMA_VPS_SVE2 24, 32772
+LUMA_VPS_SVE2 32, 8773
+LUMA_VPS_SVE2 32, 16774
+LUMA_VPS_SVE2 32, 32775
+LUMA_VPS_SVE2 32, 64776
+LUMA_VPS_SVE2 32, 24777
+LUMA_VPS_SVE2 48, 64778
+LUMA_VPS_SVE2 64, 16779
+LUMA_VPS_SVE2 64, 32780
+LUMA_VPS_SVE2 64, 64781
+LUMA_VPS_SVE2 64, 48782
+783
+// ***** luma_vss *****784
+.macro vss_end_sve2785
+ asr z17.s, z17.s, #6786
+ asr z18.s, z18.s, #6787
+ uzp1 v17.8h, v17.8h, v18.8h788
+.endm789
+790
+.macro FILTER_VSS_SVE2 w, h, v791
+ lsl x1, x1, #1792
+ lsl x10, x1, #2 // x10 = 4 * x1793
+ sub x11, x10, x1 // x11 = 3 * x1794
+ sub x0, x0, x11795
+ lsl x3, x3, #1796
+ mov x5, #\h797
+ mov x12, #\w798
+ lsl x12, x12, #1799
+ qpel_start_\v\()_1800
+.loop_luma_vss_sve2_\v\()_\w\()x\h:801
+ mov x7, x2802
+ mov x9, #0803
+.loop_luma_vss_w8_sve2_\v\()_\w\()x\h:804
+ add x6, x0, x9805
+ qpel_load_64b \v806
+ qpel_filter_\v\()_32b_1807
+ vss_end_sve2808
+.if \w == 4809
+ str s17, x7, #4810
+ add x9, x9, #4811
+.else812
+ str q17, x7, #16813
+ add x9, x9, #16814
+.if \w == 12815
+ add x6, x0, x9816
+ qpel_load_64b \v817
+ qpel_filter_\v\()_32b_1818
+ vss_end_sve2819
+ str d17, x7, #8820
+ add x9, x9, #8821
+.endif822
+.endif823
+ cmp x9, x12824
+ blt .loop_luma_vss_w8_sve2_\v\()_\w\()x\h825
+ add x0, x0, x1826
+ add x2, x2, x3827
+ sub x5, x5, #1828
+ cbnz x5, .loop_luma_vss_sve2_\v\()_\w\()x\h829
+ ret830
+.endm831
+832
+// void interp_vert_ss_c(const int16_t* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)833
+.macro LUMA_VSS_SVE2 w, h834
+function x265_interp_8tap_vert_ss_\w\()x\h\()_sve2835
+ cmp x4, #0836
+ beq 0f837
+ cmp x4, #1838
+ beq 1f839
+ cmp x4, #2840
+ beq 2f841
+ cmp x4, #3842
+ beq 3f843
+0:844
+ FILTER_VSS_SVE2 \w, \h, 0845
+1:846
+ FILTER_VSS_SVE2 \w, \h, 1847
+2:848
+ FILTER_VSS_SVE2 \w, \h, 2849
+3:850
+ FILTER_VSS_SVE2 \w, \h, 3851
+endfunc852
+.endm853
+854
+LUMA_VSS_SVE2 4, 4855
+LUMA_VSS_SVE2 4, 8856
+LUMA_VSS_SVE2 4, 16857
+LUMA_VSS_SVE2 8, 4858
+LUMA_VSS_SVE2 8, 8859
+LUMA_VSS_SVE2 8, 16860
+LUMA_VSS_SVE2 8, 32861
+LUMA_VSS_SVE2 12, 16862
+LUMA_VSS_SVE2 16, 4863
+LUMA_VSS_SVE2 16, 8864
+LUMA_VSS_SVE2 16, 16865
+LUMA_VSS_SVE2 16, 32866
+LUMA_VSS_SVE2 16, 64867
+LUMA_VSS_SVE2 16, 12868
+LUMA_VSS_SVE2 32, 8869
+LUMA_VSS_SVE2 32, 16870
+LUMA_VSS_SVE2 32, 32871
+LUMA_VSS_SVE2 32, 64872
+LUMA_VSS_SVE2 32, 24873
+LUMA_VSS_SVE2 64, 16874
+LUMA_VSS_SVE2 64, 32875
+LUMA_VSS_SVE2 64, 64876
+LUMA_VSS_SVE2 64, 48877
+LUMA_VSS_SVE2 24, 32878
+LUMA_VSS_SVE2 48, 64879
+880
+// ***** luma_hps *****881
+882
+.macro FILTER_CHROMA_VPP_SVE2 w, h, v883
+ ptrue p0.h, vl8884
+ qpel_start_chroma_sve2_\v885
+ mov z31.h, #32886
+ sub x0, x0, x1887
+ mov x5, #\h888
+.loop_chroma_vpp_sve2_\v\()_\w\()x\h:889
+ mov x7, x2890
+ mov x9, #0891
+.loop_chroma_vpp_w8_sve2_\v\()_\w\()x\h:892
+ add x6, x0, x9893
+ qpel_chroma_load_32b_sve2 \v894
+ qpel_filter_chroma_sve2_\v\()_32b895
+ vpp_end_sve2896
+ add x9, x9, #8897
+.if \w == 2898
+ fmov w12, s17899
+ strh w12, x7, #2900
+.elseif \w == 4901
+ str s17, x7, #4902
+.elseif \w == 6903
+ str s17, x7, #4904
+ umov w12, v17.h2905
+ strh w12, x7, #2906
+.elseif \w == 12907
+ str d17, x7, #8908
+ add x6, x0, x9909
+ qpel_chroma_load_32b_sve2 \v910
+ qpel_filter_chroma_sve2_\v\()_32b911
+ vpp_end_sve2912
+ str s17, x7, #4913
+ add x9, x9, #8914
+.else915
+ str d17, x7, #8916
+.endif917
+ cmp x9, #\w918
+ blt .loop_chroma_vpp_w8_sve2_\v\()_\w\()x\h919
+ add x0, x0, x1920
+ add x2, x2, x3921
+ sub x5, x5, #1922
+ cbnz x5, .loop_chroma_vpp_sve2_\v\()_\w\()x\h923
+ ret924
+.endm925
+926
+// void interp_vert_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)927
+.macro CHROMA_VPP_SVE2 w, h928
+function x265_interp_4tap_vert_pp_\w\()x\h\()_sve2929
+ cmp x4, #0930
+ beq 0f931
+ cmp x4, #1932
+ beq 1f933
+ cmp x4, #2934
+ beq 2f935
+ cmp x4, #3936
+ beq 3f937
+ cmp x4, #4938
+ beq 4f939
+ cmp x4, #5940
+ beq 5f941
+ cmp x4, #6942
+ beq 6f943
+ cmp x4, #7944
+ beq 7f945
+0:946
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 0947
+1:948
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 1949
+2:950
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 2951
+3:952
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 3953
+4:954
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 4955
+5:956
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 5957
+6:958
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 6959
+7:960
+ FILTER_CHROMA_VPP_SVE2 \w, \h, 7961
+endfunc962
+.endm963
+964
+CHROMA_VPP_SVE2 2, 4965
+CHROMA_VPP_SVE2 2, 8966
+CHROMA_VPP_SVE2 2, 16967
+CHROMA_VPP_SVE2 4, 2968
+CHROMA_VPP_SVE2 4, 4969
+CHROMA_VPP_SVE2 4, 8970
+CHROMA_VPP_SVE2 4, 16971
+CHROMA_VPP_SVE2 4, 32972
+CHROMA_VPP_SVE2 6, 8973
+CHROMA_VPP_SVE2 6, 16974
+CHROMA_VPP_SVE2 8, 2975
+CHROMA_VPP_SVE2 8, 4976
+CHROMA_VPP_SVE2 8, 6977
+CHROMA_VPP_SVE2 8, 8978
+CHROMA_VPP_SVE2 8, 16979
+CHROMA_VPP_SVE2 8, 32980
+CHROMA_VPP_SVE2 8, 12981
+CHROMA_VPP_SVE2 8, 64982
+CHROMA_VPP_SVE2 12, 16983
+CHROMA_VPP_SVE2 12, 32984
+CHROMA_VPP_SVE2 16, 4985
+CHROMA_VPP_SVE2 16, 8986
+CHROMA_VPP_SVE2 16, 12987
+CHROMA_VPP_SVE2 16, 16988
+CHROMA_VPP_SVE2 16, 32989
+CHROMA_VPP_SVE2 16, 64990
+CHROMA_VPP_SVE2 16, 24991
+CHROMA_VPP_SVE2 32, 8992
+CHROMA_VPP_SVE2 32, 16993
+CHROMA_VPP_SVE2 32, 24994
+CHROMA_VPP_SVE2 32, 32995
+CHROMA_VPP_SVE2 32, 64996
+CHROMA_VPP_SVE2 32, 48997
+CHROMA_VPP_SVE2 24, 32998
+CHROMA_VPP_SVE2 24, 64999
+CHROMA_VPP_SVE2 64, 161000
+CHROMA_VPP_SVE2 64, 321001
+CHROMA_VPP_SVE2 64, 481002
+CHROMA_VPP_SVE2 64, 641003
+CHROMA_VPP_SVE2 48, 641004
+1005
+.macro FILTER_CHROMA_VPS_SVE2 w, h, v1006
+ ptrue p0.h, vl81007
+ qpel_start_chroma_sve2_\v1008
+ mov z31.h, #81921009
+ lsl x3, x3, #11010
+ sub x0, x0, x11011
+ mov x5, #\h1012
+.loop_vps_sve2_\v\()_\w\()x\h:1013
+ mov x7, x21014
+ mov x9, #01015
+.loop_vps_w8_sve2_\v\()_\w\()x\h:1016
+ add x6, x0, x91017
+ qpel_chroma_load_32b_sve2 \v1018
+ qpel_filter_chroma_sve2_\v\()_32b1019
+ vps_end_sve21020
+ add x9, x9, #81021
+.if \w == 21022
+ str s17, x7, #41023
+.elseif \w == 41024
+ str d17, x7, #81025
+.elseif \w == 61026
+ str d17, x7, #81027
+ st1 {v17.s}2, x7, #41028
+.elseif \w == 121029
+ str q17, x7, #161030
+ add x6, x0, x91031
+ qpel_chroma_load_32b_sve2 \v1032
+ qpel_filter_chroma_sve2_\v\()_32b1033
+ vps_end_sve21034
+ str d17, x7, #81035
+ add x9, x9, #81036
+.else1037
+ str q17, x7, #161038
+.endif1039
+ cmp x9, #\w1040
+ blt .loop_vps_w8_sve2_\v\()_\w\()x\h1041
+1042
+ add x0, x0, x11043
+ add x2, x2, x31044
+ sub x5, x5, #11045
+ cbnz x5, .loop_vps_sve2_\v\()_\w\()x\h1046
+ ret1047
+.endm1048
+1049
+// void interp_vert_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)1050
+.macro CHROMA_VPS_SVE2 w, h1051
+function x265_interp_4tap_vert_ps_\w\()x\h\()_sve21052
+ cmp x4, #01053
+ beq 0f1054
+ cmp x4, #11055
+ beq 1f1056
+ cmp x4, #21057
+ beq 2f1058
+ cmp x4, #31059
+ beq 3f1060
+ cmp x4, #41061
+ beq 4f1062
+ cmp x4, #51063
+ beq 5f1064
+ cmp x4, #61065
+ beq 6f1066
+ cmp x4, #71067
+ beq 7f1068
+0:1069
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 01070
+1:1071
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 11072
+2:1073
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 21074
+3:1075
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 31076
+4:1077
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 41078
+5:1079
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 51080
+6:1081
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 61082
+7:1083
+ FILTER_CHROMA_VPS_SVE2 \w, \h, 71084
+endfunc1085
+.endm1086
+1087
+CHROMA_VPS_SVE2 2, 41088
+CHROMA_VPS_SVE2 2, 81089
+CHROMA_VPS_SVE2 2, 161090
+CHROMA_VPS_SVE2 4, 21091
+CHROMA_VPS_SVE2 4, 41092
+CHROMA_VPS_SVE2 4, 81093
+CHROMA_VPS_SVE2 4, 161094
+CHROMA_VPS_SVE2 4, 321095
+CHROMA_VPS_SVE2 6, 81096
+CHROMA_VPS_SVE2 6, 161097
+CHROMA_VPS_SVE2 8, 21098
+CHROMA_VPS_SVE2 8, 41099
+CHROMA_VPS_SVE2 8, 61100
+CHROMA_VPS_SVE2 8, 81101
+CHROMA_VPS_SVE2 8, 161102
+CHROMA_VPS_SVE2 8, 321103
+CHROMA_VPS_SVE2 8, 121104
+CHROMA_VPS_SVE2 8, 641105
+CHROMA_VPS_SVE2 12, 161106
+CHROMA_VPS_SVE2 12, 321107
+CHROMA_VPS_SVE2 16, 41108
+CHROMA_VPS_SVE2 16, 81109
+CHROMA_VPS_SVE2 16, 121110
+CHROMA_VPS_SVE2 16, 161111
+CHROMA_VPS_SVE2 16, 321112
+CHROMA_VPS_SVE2 16, 641113
+CHROMA_VPS_SVE2 16, 241114
+CHROMA_VPS_SVE2 32, 81115
+CHROMA_VPS_SVE2 32, 161116
+CHROMA_VPS_SVE2 32, 241117
+CHROMA_VPS_SVE2 32, 321118
+CHROMA_VPS_SVE2 32, 641119
+CHROMA_VPS_SVE2 32, 481120
+CHROMA_VPS_SVE2 24, 321121
+CHROMA_VPS_SVE2 24, 641122
+CHROMA_VPS_SVE2 64, 161123
+CHROMA_VPS_SVE2 64, 321124
+CHROMA_VPS_SVE2 64, 481125
+CHROMA_VPS_SVE2 64, 641126
+CHROMA_VPS_SVE2 48, 641127
+1128
+.macro qpel_start_chroma_sve2_0_11129
+ mov z24.h, #641130
+.endm1131
+1132
+.macro qpel_start_chroma_sve2_1_11133
+ mov z24.h, #581134
+ mov z25.h, #101135
+.endm1136
+1137
+.macro qpel_start_chroma_sve2_2_11138
+ mov z25.h, #541139
+.endm1140
+1141
+.macro qpel_start_chroma_sve2_3_11142
+ mov z25.h, #461143
+ mov z26.h, #281144
+ mov z27.h, #61145
+.endm1146
+1147
+.macro qpel_start_chroma_sve2_4_11148
+ mov z24.h, #361149
+.endm1150
+1151
+.macro qpel_start_chroma_sve2_5_11152
+ mov z25.h, #281153
+ mov z26.h, #461154
+ mov z27.h, #61155
+.endm1156
+1157
+.macro qpel_start_chroma_sve2_6_11158
+ mov z25.h, #541159
+.endm1160
+1161
+.macro qpel_start_chroma_sve2_7_11162
+ mov z24.h, #581163
+ mov z25.h, #101164
+.endm1165
+1166
+.macro FILTER_CHROMA_VSS_SVE2 w, h, v1167
+ lsl x1, x1, #11168
+ sub x0, x0, x11169
+ lsl x3, x3, #11170
+ mov x5, #\h1171
+ mov x12, #\w1172
+ lsl x12, x12, #11173
+ qpel_start_chroma_sve2_\v\()_11174
+.loop_vss_sve2_\v\()_\w\()x\h:1175
+ mov x7, x21176
+ mov x9, #01177
+.if \w == 41178
+.rept 21179
+ add x6, x0, x91180
+ qpel_chroma_load_64b \v1181
+ qpel_filter_chroma_\v\()_32b_11182
+ vss_end_sve21183
+ str s17, x7, #41184
+ add x9, x9, #41185
+.endr1186
+.else1187
+.loop_vss_w8_sve2_\v\()_\w\()x\h:1188
+ add x6, x0, x91189
+ qpel_chroma_load_64b \v1190
+ qpel_filter_chroma_\v\()_32b_11191
+ vss_end_sve21192
+ str q17, x7, #161193
+ add x9, x9, #161194
+.if \w == 121195
+ add x6, x0, x91196
+ qpel_chroma_load_64b \v1197
+ qpel_filter_chroma_\v\()_32b_11198
+ vss_end_sve21199
+ str d17, x7, #81200
+ add x9, x9, #81201
+.endif1202
+ cmp x9, x121203
+ blt .loop_vss_w8_sve2_\v\()_\w\()x\h1204
+.endif1205
+ add x0, x0, x11206
+ add x2, x2, x31207
+ sub x5, x5, #11208
+ cbnz x5, .loop_vss_sve2_\v\()_\w\()x\h1209
+ ret1210
+.endm1211
+1212
+// void interp_vert_ss_c(const int16_t* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)1213
+.macro CHROMA_VSS_SVE2 w, h1214
+function x265_interp_4tap_vert_ss_\w\()x\h\()_sve21215
+ cmp x4, #01216
+ beq 0f1217
+ cmp x4, #11218
+ beq 1f1219
+ cmp x4, #21220
+ beq 2f1221
+ cmp x4, #31222
+ beq 3f1223
+ cmp x4, #41224
+ beq 4f1225
+ cmp x4, #51226
+ beq 5f1227
+ cmp x4, #61228
+ beq 6f1229
+ cmp x4, #71230
+ beq 7f1231
+0:1232
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 01233
+1:1234
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 11235
+2:1236
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 21237
+3:1238
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 31239
+4:1240
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 41241
+5:1242
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 51243
+6:1244
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 61245
+7:1246
+ FILTER_CHROMA_VSS_SVE2 \w, \h, 71247
+endfunc1248
+.endm1249
+1250
+CHROMA_VSS_SVE2 4, 41251
+CHROMA_VSS_SVE2 4, 81252
+CHROMA_VSS_SVE2 4, 161253
+CHROMA_VSS_SVE2 4, 321254
+CHROMA_VSS_SVE2 8, 21255
+CHROMA_VSS_SVE2 8, 41256
+CHROMA_VSS_SVE2 8, 61257
+CHROMA_VSS_SVE2 8, 81258
+CHROMA_VSS_SVE2 8, 161259
+CHROMA_VSS_SVE2 8, 321260
+CHROMA_VSS_SVE2 8, 121261
+CHROMA_VSS_SVE2 8, 641262
+CHROMA_VSS_SVE2 12, 161263
+CHROMA_VSS_SVE2 12, 321264
+CHROMA_VSS_SVE2 16, 41265
+CHROMA_VSS_SVE2 16, 81266
+CHROMA_VSS_SVE2 16, 121267
+CHROMA_VSS_SVE2 16, 161268
+CHROMA_VSS_SVE2 16, 321269
+CHROMA_VSS_SVE2 16, 641270
+CHROMA_VSS_SVE2 16, 241271
+CHROMA_VSS_SVE2 32, 81272
+CHROMA_VSS_SVE2 32, 161273
+CHROMA_VSS_SVE2 32, 241274
+CHROMA_VSS_SVE2 32, 321275
+CHROMA_VSS_SVE2 32, 641276
+CHROMA_VSS_SVE2 32, 481277
+CHROMA_VSS_SVE2 24, 321278
+CHROMA_VSS_SVE2 24, 641279
+CHROMA_VSS_SVE2 64, 161280
+CHROMA_VSS_SVE2 64, 321281
+CHROMA_VSS_SVE2 64, 481282
+CHROMA_VSS_SVE2 64, 641283
+CHROMA_VSS_SVE2 48, 641284
x265_3.6.tar.gz/source/common/aarch64/ipfilter.S
Added
1056
1
2
+/*****************************************************************************3
+ * Copyright (C) 2021 MulticoreWare, Inc4
+ *5
+ * Authors: Sebastian Pop <spop@amazon.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// Functions in this file:26
+// ***** luma_vpp *****27
+// ***** luma_vps *****28
+// ***** luma_vsp *****29
+// ***** luma_vss *****30
+// ***** luma_hpp *****31
+// ***** luma_hps *****32
+// ***** chroma_vpp *****33
+// ***** chroma_vps *****34
+// ***** chroma_vsp *****35
+// ***** chroma_vss *****36
+// ***** chroma_hpp *****37
+// ***** chroma_hps *****38
+39
+#include "asm.S"40
+#include "ipfilter-common.S"41
+42
+#ifdef __APPLE__43
+.section __RODATA,__rodata44
+#else45
+.section .rodata46
+#endif47
+48
+.align 449
+50
+.text51
+52
+// ***** luma_vpp *****53
+// void interp_vert_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)54
+.macro LUMA_VPP_4xN h55
+function x265_interp_8tap_vert_pp_4x\h\()_neon56
+ movrel x10, g_luma_s1657
+ sub x0, x0, x158
+ sub x0, x0, x1, lsl #1 // src -= 3 * srcStride59
+ lsl x4, x4, #460
+ ldr q0, x10, x4 // q0 = luma interpolate coeff61
+ dup v24.8h, v0.h062
+ dup v25.8h, v0.h163
+ trn1 v24.2d, v24.2d, v25.2d64
+ dup v26.8h, v0.h265
+ dup v27.8h, v0.h366
+ trn1 v26.2d, v26.2d, v27.2d67
+ dup v28.8h, v0.h468
+ dup v29.8h, v0.h569
+ trn1 v28.2d, v28.2d, v29.2d70
+ dup v30.8h, v0.h671
+ dup v31.8h, v0.h772
+ trn1 v30.2d, v30.2d, v31.2d73
+74
+ // prepare to load 8 lines75
+ ld1 {v0.s}0, x0, x176
+ ld1 {v0.s}1, x0, x177
+ ushll v0.8h, v0.8b, #078
+ ld1 {v1.s}0, x0, x179
+ ld1 {v1.s}1, x0, x180
+ ushll v1.8h, v1.8b, #081
+ ld1 {v2.s}0, x0, x182
+ ld1 {v2.s}1, x0, x183
+ ushll v2.8h, v2.8b, #084
+ ld1 {v3.s}0, x0, x185
+ ld1 {v3.s}1, x0, x186
+ ushll v3.8h, v3.8b, #087
+88
+ mov x9, #\h89
+.loop_4x\h:90
+ ld1 {v4.s}0, x0, x191
+ ld1 {v4.s}1, x0, x192
+ ushll v4.8h, v4.8b, #093
+94
+ // row0-195
+ mul v16.8h, v0.8h, v24.8h96
+ ext v21.16b, v0.16b, v1.16b, #897
+ mul v17.8h, v21.8h, v24.8h98
+ mov v0.16b, v1.16b99
+100
+ // row2-3101
+ mla v16.8h, v1.8h, v26.8h102
+ ext v21.16b, v1.16b, v2.16b, #8103
+ mla v17.8h, v21.8h, v26.8h104
+ mov v1.16b, v2.16b105
+106
+ // row4-5107
+ mla v16.8h, v2.8h, v28.8h108
+ ext v21.16b, v2.16b, v3.16b, #8109
+ mla v17.8h, v21.8h, v28.8h110
+ mov v2.16b, v3.16b111
+112
+ // row6-7113
+ mla v16.8h, v3.8h, v30.8h114
+ ext v21.16b, v3.16b, v4.16b, #8115
+ mla v17.8h, v21.8h, v30.8h116
+ mov v3.16b, v4.16b117
+118
+ // sum row0-7119
+ trn1 v20.2d, v16.2d, v17.2d120
+ trn2 v21.2d, v16.2d, v17.2d121
+ add v16.8h, v20.8h, v21.8h122
+123
+ sqrshrun v16.8b, v16.8h, #6124
+ st1 {v16.s}0, x2, x3125
+ st1 {v16.s}1, x2, x3126
+127
+ sub x9, x9, #2128
+ cbnz x9, .loop_4x\h129
+ ret130
+endfunc131
+.endm132
+133
+LUMA_VPP_4xN 4134
+LUMA_VPP_4xN 8135
+LUMA_VPP_4xN 16136
+137
+// void interp_vert_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)138
+.macro LUMA_VPP w, h139
+function x265_interp_8tap_vert_pp_\w\()x\h\()_neon140
+ cmp x4, #0141
+ b.eq 0f142
+ cmp x4, #1143
+ b.eq 1f144
+ cmp x4, #2145
+ b.eq 2f146
+ cmp x4, #3147
+ b.eq 3f148
+0:149
+ FILTER_LUMA_VPP \w, \h, 0150
+1:151
+ FILTER_LUMA_VPP \w, \h, 1152
+2:153
+ FILTER_LUMA_VPP \w, \h, 2154
+3:155
+ FILTER_LUMA_VPP \w, \h, 3156
+endfunc157
+.endm158
+159
+LUMA_VPP 8, 4160
+LUMA_VPP 8, 8161
+LUMA_VPP 8, 16162
+LUMA_VPP 8, 32163
+LUMA_VPP 12, 16164
+LUMA_VPP 16, 4165
+LUMA_VPP 16, 8166
+LUMA_VPP 16, 16167
+LUMA_VPP 16, 32168
+LUMA_VPP 16, 64169
+LUMA_VPP 16, 12170
+LUMA_VPP 24, 32171
+LUMA_VPP 32, 8172
+LUMA_VPP 32, 16173
+LUMA_VPP 32, 32174
+LUMA_VPP 32, 64175
+LUMA_VPP 32, 24176
+LUMA_VPP 48, 64177
+LUMA_VPP 64, 16178
+LUMA_VPP 64, 32179
+LUMA_VPP 64, 64180
+LUMA_VPP 64, 48181
+182
+// ***** luma_vps *****183
+// void interp_vert_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)184
+.macro LUMA_VPS_4xN h185
+function x265_interp_8tap_vert_ps_4x\h\()_neon186
+ lsl x3, x3, #1187
+ lsl x5, x4, #6188
+ lsl x4, x1, #2189
+ sub x4, x4, x1190
+ sub x0, x0, x4191
+192
+ mov w6, #8192193
+ dup v28.4s, w6194
+ mov x4, #\h195
+ movrel x12, g_lumaFilter196
+ add x12, x12, x5197
+ ld1r {v16.2d}, x12, #8198
+ ld1r {v17.2d}, x12, #8199
+ ld1r {v18.2d}, x12, #8200
+ ld1r {v19.2d}, x12, #8201
+ ld1r {v20.2d}, x12, #8202
+ ld1r {v21.2d}, x12, #8203
+ ld1r {v22.2d}, x12, #8204
+ ld1r {v23.2d}, x12, #8205
+206
+.loop_vps_4x\h:207
+ mov x6, x0208
+209
+ ld1 {v0.s}0, x6, x1210
+ ld1 {v1.s}0, x6, x1211
+ ld1 {v2.s}0, x6, x1212
+ ld1 {v3.s}0, x6, x1213
+ ld1 {v4.s}0, x6, x1214
+ ld1 {v5.s}0, x6, x1215
+ ld1 {v6.s}0, x6, x1216
+ ld1 {v7.s}0, x6, x1217
+ uxtl v0.8h, v0.8b218
+ uxtl v0.4s, v0.4h219
+220
+ uxtl v1.8h, v1.8b221
+ uxtl v1.4s, v1.4h222
+ mul v0.4s, v0.4s, v16.4s223
+224
+ uxtl v2.8h, v2.8b225
+ uxtl v2.4s, v2.4h226
+ mla v0.4s, v1.4s, v17.4s227
+228
+ uxtl v3.8h, v3.8b229
+ uxtl v3.4s, v3.4h230
+ mla v0.4s, v2.4s, v18.4s231
+232
+ uxtl v4.8h, v4.8b233
+ uxtl v4.4s, v4.4h234
+ mla v0.4s, v3.4s, v19.4s235
+236
+ uxtl v5.8h, v5.8b237
+ uxtl v5.4s, v5.4h238
+ mla v0.4s, v4.4s, v20.4s239
+240
+ uxtl v6.8h, v6.8b241
+ uxtl v6.4s, v6.4h242
+ mla v0.4s, v5.4s, v21.4s243
+244
+ uxtl v7.8h, v7.8b245
+ uxtl v7.4s, v7.4h246
+ mla v0.4s, v6.4s, v22.4s247
+248
+ mla v0.4s, v7.4s, v23.4s249
+250
+ sub v0.4s, v0.4s, v28.4s251
+ sqxtn v0.4h, v0.4s252
+ st1 {v0.8b}, x2, x3253
+254
+ add x0, x0, x1255
+ sub x4, x4, #1256
+ cbnz x4, .loop_vps_4x\h257
+ ret258
+endfunc259
+.endm260
+261
+LUMA_VPS_4xN 4262
+LUMA_VPS_4xN 8263
+LUMA_VPS_4xN 16264
+265
+// void interp_vert_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)266
+.macro LUMA_VPS w, h267
+function x265_interp_8tap_vert_ps_\w\()x\h\()_neon268
+ cmp x4, #0269
+ beq 0f270
+ cmp x4, #1271
+ beq 1f272
+ cmp x4, #2273
+ beq 2f274
+ cmp x4, #3275
+ beq 3f276
+0:277
+ FILTER_VPS \w, \h, 0278
+1:279
+ FILTER_VPS \w, \h, 1280
+2:281
+ FILTER_VPS \w, \h, 2282
+3:283
+ FILTER_VPS \w, \h, 3284
+endfunc285
+.endm286
+287
+LUMA_VPS 8, 4288
+LUMA_VPS 8, 8289
+LUMA_VPS 8, 16290
+LUMA_VPS 8, 32291
+LUMA_VPS 12, 16292
+LUMA_VPS 16, 4293
+LUMA_VPS 16, 8294
+LUMA_VPS 16, 16295
+LUMA_VPS 16, 32296
+LUMA_VPS 16, 64297
+LUMA_VPS 16, 12298
+LUMA_VPS 24, 32299
+LUMA_VPS 32, 8300
+LUMA_VPS 32, 16301
+LUMA_VPS 32, 32302
+LUMA_VPS 32, 64303
+LUMA_VPS 32, 24304
+LUMA_VPS 48, 64305
+LUMA_VPS 64, 16306
+LUMA_VPS 64, 32307
+LUMA_VPS 64, 64308
+LUMA_VPS 64, 48309
+310
+// ***** luma_vsp *****311
+// void interp_vert_sp_c(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)312
+.macro LUMA_VSP_4xN h313
+function x265_interp_8tap_vert_sp_4x\h\()_neon314
+ lsl x5, x4, #6315
+ lsl x1, x1, #1316
+ lsl x4, x1, #2317
+ sub x4, x4, x1318
+ sub x0, x0, x4319
+320
+ mov w12, #1321
+ lsl w12, w12, #19322
+ add w12, w12, #2048323
+ dup v24.4s, w12324
+ mov x4, #\h325
+ movrel x12, g_lumaFilter326
+ add x12, x12, x5327
+ ld1r {v16.2d}, x12, #8328
+ ld1r {v17.2d}, x12, #8329
+ ld1r {v18.2d}, x12, #8330
+ ld1r {v19.2d}, x12, #8331
+ ld1r {v20.2d}, x12, #8332
+ ld1r {v21.2d}, x12, #8333
+ ld1r {v22.2d}, x12, #8334
+ ld1r {v23.2d}, x12, #8335
+.loop_vsp_4x\h:336
+ mov x6, x0337
+338
+ ld1 {v0.8b}, x6, x1339
+ ld1 {v1.8b}, x6, x1340
+ ld1 {v2.8b}, x6, x1341
+ ld1 {v3.8b}, x6, x1342
+ ld1 {v4.8b}, x6, x1343
+ ld1 {v5.8b}, x6, x1344
+ ld1 {v6.8b}, x6, x1345
+ ld1 {v7.8b}, x6, x1346
+347
+ sshll v0.4s, v0.4h, #0348
+ sshll v1.4s, v1.4h, #0349
+ mul v0.4s, v0.4s, v16.4s350
+ sshll v2.4s, v2.4h, #0351
+ mla v0.4s, v1.4s, v17.4s352
+ sshll v3.4s, v3.4h, #0353
+ mla v0.4s, v2.4s, v18.4s354
+ sshll v4.4s, v4.4h, #0355
+ mla v0.4s, v3.4s, v19.4s356
+ sshll v5.4s, v5.4h, #0357
+ mla v0.4s, v4.4s, v20.4s358
+ sshll v6.4s, v6.4h, #0359
+ mla v0.4s, v5.4s, v21.4s360
+ sshll v7.4s, v7.4h, #0361
+ mla v0.4s, v6.4s, v22.4s362
+363
+ mla v0.4s, v7.4s, v23.4s364
+365
+ add v0.4s, v0.4s, v24.4s366
+ sqshrun v0.4h, v0.4s, #12367
+ sqxtun v0.8b, v0.8h368
+ st1 {v0.s}0, x2, x3369
+370
+ add x0, x0, x1371
+ sub x4, x4, #1372
+ cbnz x4, .loop_vsp_4x\h373
+ ret374
+endfunc375
+.endm376
+377
+LUMA_VSP_4xN 4378
+LUMA_VSP_4xN 8379
+LUMA_VSP_4xN 16380
+381
+// void interp_vert_sp_c(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)382
+.macro LUMA_VSP w, h383
+function x265_interp_8tap_vert_sp_\w\()x\h\()_neon384
+ cmp x4, #0385
+ beq 0f386
+ cmp x4, #1387
+ beq 1f388
+ cmp x4, #2389
+ beq 2f390
+ cmp x4, #3391
+ beq 3f392
+0:393
+ FILTER_VSP \w, \h, 0394
+1:395
+ FILTER_VSP \w, \h, 1396
+2:397
+ FILTER_VSP \w, \h, 2398
+3:399
+ FILTER_VSP \w, \h, 3400
+endfunc401
+.endm402
+403
+LUMA_VSP 8, 4404
+LUMA_VSP 8, 8405
+LUMA_VSP 8, 16406
+LUMA_VSP 8, 32407
+LUMA_VSP 12, 16408
+LUMA_VSP 16, 4409
+LUMA_VSP 16, 8410
+LUMA_VSP 16, 16411
+LUMA_VSP 16, 32412
+LUMA_VSP 16, 64413
+LUMA_VSP 16, 12414
+LUMA_VSP 32, 8415
+LUMA_VSP 32, 16416
+LUMA_VSP 32, 32417
+LUMA_VSP 32, 64418
+LUMA_VSP 32, 24419
+LUMA_VSP 64, 16420
+LUMA_VSP 64, 32421
+LUMA_VSP 64, 64422
+LUMA_VSP 64, 48423
+LUMA_VSP 24, 32424
+LUMA_VSP 48, 64425
+426
+// ***** luma_vss *****427
+// void interp_vert_ss_c(const int16_t* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)428
+.macro LUMA_VSS w, h429
+function x265_interp_8tap_vert_ss_\w\()x\h\()_neon430
+ cmp x4, #0431
+ beq 0f432
+ cmp x4, #1433
+ beq 1f434
+ cmp x4, #2435
+ beq 2f436
+ cmp x4, #3437
+ beq 3f438
+0:439
+ FILTER_VSS \w, \h, 0440
+1:441
+ FILTER_VSS \w, \h, 1442
+2:443
+ FILTER_VSS \w, \h, 2444
+3:445
+ FILTER_VSS \w, \h, 3446
+endfunc447
+.endm448
+449
+LUMA_VSS 4, 4450
+LUMA_VSS 4, 8451
+LUMA_VSS 4, 16452
+LUMA_VSS 8, 4453
+LUMA_VSS 8, 8454
+LUMA_VSS 8, 16455
+LUMA_VSS 8, 32456
+LUMA_VSS 12, 16457
+LUMA_VSS 16, 4458
+LUMA_VSS 16, 8459
+LUMA_VSS 16, 16460
+LUMA_VSS 16, 32461
+LUMA_VSS 16, 64462
+LUMA_VSS 16, 12463
+LUMA_VSS 32, 8464
+LUMA_VSS 32, 16465
+LUMA_VSS 32, 32466
+LUMA_VSS 32, 64467
+LUMA_VSS 32, 24468
+LUMA_VSS 64, 16469
+LUMA_VSS 64, 32470
+LUMA_VSS 64, 64471
+LUMA_VSS 64, 48472
+LUMA_VSS 24, 32473
+LUMA_VSS 48, 64474
+475
+// ***** luma_hpp *****476
+// void interp_horiz_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)477
+.macro LUMA_HPP w, h478
+function x265_interp_horiz_pp_\w\()x\h\()_neon479
+ cmp x4, #0480
+ beq 0f481
+ cmp x4, #1482
+ beq 1f483
+ cmp x4, #2484
+ beq 2f485
+ cmp x4, #3486
+ beq 3f487
+0:488
+ FILTER_HPP \w, \h, 0489
+1:490
+ FILTER_HPP \w, \h, 1491
+2:492
+ FILTER_HPP \w, \h, 2493
+3:494
+ FILTER_HPP \w, \h, 3495
+endfunc496
+.endm497
+498
+LUMA_HPP 4, 4499
+LUMA_HPP 4, 8500
+LUMA_HPP 4, 16501
+LUMA_HPP 8, 4502
+LUMA_HPP 8, 8503
+LUMA_HPP 8, 16504
+LUMA_HPP 8, 32505
+LUMA_HPP 12, 16506
+LUMA_HPP 16, 4507
+LUMA_HPP 16, 8508
+LUMA_HPP 16, 12509
+LUMA_HPP 16, 16510
+LUMA_HPP 16, 32511
+LUMA_HPP 16, 64512
+LUMA_HPP 24, 32513
+LUMA_HPP 32, 8514
+LUMA_HPP 32, 16515
+LUMA_HPP 32, 24516
+LUMA_HPP 32, 32517
+LUMA_HPP 32, 64518
+LUMA_HPP 48, 64519
+LUMA_HPP 64, 16520
+LUMA_HPP 64, 32521
+LUMA_HPP 64, 48522
+LUMA_HPP 64, 64523
+524
+// ***** luma_hps *****525
+// void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)526
+.macro LUMA_HPS w, h527
+function x265_interp_horiz_ps_\w\()x\h\()_neon528
+ mov w10, #\h529
+ cmp w5, #0530
+ b.eq 6f531
+ sub x0, x0, x1, lsl #2532
+ add x0, x0, x1533
+ add w10, w10, #7534
+6:535
+ mov w6, w10536
+ cmp w4, #0537
+ b.eq 0f538
+ cmp w4, #1539
+ b.eq 1f540
+ cmp w4, #2541
+ b.eq 2f542
+ cmp w4, #3543
+ b.eq 3f544
+0:545
+ FILTER_HPS \w, \h, 0546
+1:547
+ FILTER_HPS \w, \h, 1548
+2:549
+ FILTER_HPS \w, \h, 2550
+3:551
+ FILTER_HPS \w, \h, 3552
+endfunc553
+.endm554
+555
+LUMA_HPS 4, 4556
+LUMA_HPS 4, 8557
+LUMA_HPS 4, 16558
+LUMA_HPS 8, 4559
+LUMA_HPS 8, 8560
+LUMA_HPS 8, 16561
+LUMA_HPS 8, 32562
+LUMA_HPS 12, 16563
+LUMA_HPS 16, 4564
+LUMA_HPS 16, 8565
+LUMA_HPS 16, 12566
+LUMA_HPS 16, 16567
+LUMA_HPS 16, 32568
+LUMA_HPS 16, 64569
+LUMA_HPS 24, 32570
+LUMA_HPS 32, 8571
+LUMA_HPS 32, 16572
+LUMA_HPS 32, 24573
+LUMA_HPS 32, 32574
+LUMA_HPS 32, 64575
+LUMA_HPS 48, 64576
+LUMA_HPS 64, 16577
+LUMA_HPS 64, 32578
+LUMA_HPS 64, 48579
+LUMA_HPS 64, 64580
+581
+// ***** chroma_vpp *****582
+// void interp_vert_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)583
+.macro CHROMA_VPP w, h584
+function x265_interp_4tap_vert_pp_\w\()x\h\()_neon585
+ cmp x4, #0586
+ beq 0f587
+ cmp x4, #1588
+ beq 1f589
+ cmp x4, #2590
+ beq 2f591
+ cmp x4, #3592
+ beq 3f593
+ cmp x4, #4594
+ beq 4f595
+ cmp x4, #5596
+ beq 5f597
+ cmp x4, #6598
+ beq 6f599
+ cmp x4, #7600
+ beq 7f601
+0:602
+ FILTER_CHROMA_VPP \w, \h, 0603
+1:604
+ FILTER_CHROMA_VPP \w, \h, 1605
+2:606
+ FILTER_CHROMA_VPP \w, \h, 2607
+3:608
+ FILTER_CHROMA_VPP \w, \h, 3609
+4:610
+ FILTER_CHROMA_VPP \w, \h, 4611
+5:612
+ FILTER_CHROMA_VPP \w, \h, 5613
+6:614
+ FILTER_CHROMA_VPP \w, \h, 6615
+7:616
+ FILTER_CHROMA_VPP \w, \h, 7617
+endfunc618
+.endm619
+620
+CHROMA_VPP 2, 4621
+CHROMA_VPP 2, 8622
+CHROMA_VPP 2, 16623
+CHROMA_VPP 4, 2624
+CHROMA_VPP 4, 4625
+CHROMA_VPP 4, 8626
+CHROMA_VPP 4, 16627
+CHROMA_VPP 4, 32628
+CHROMA_VPP 6, 8629
+CHROMA_VPP 6, 16630
+CHROMA_VPP 8, 2631
+CHROMA_VPP 8, 4632
+CHROMA_VPP 8, 6633
+CHROMA_VPP 8, 8634
+CHROMA_VPP 8, 16635
+CHROMA_VPP 8, 32636
+CHROMA_VPP 8, 12637
+CHROMA_VPP 8, 64638
+CHROMA_VPP 12, 16639
+CHROMA_VPP 12, 32640
+CHROMA_VPP 16, 4641
+CHROMA_VPP 16, 8642
+CHROMA_VPP 16, 12643
+CHROMA_VPP 16, 16644
+CHROMA_VPP 16, 32645
+CHROMA_VPP 16, 64646
+CHROMA_VPP 16, 24647
+CHROMA_VPP 32, 8648
+CHROMA_VPP 32, 16649
+CHROMA_VPP 32, 24650
+CHROMA_VPP 32, 32651
+CHROMA_VPP 32, 64652
+CHROMA_VPP 32, 48653
+CHROMA_VPP 24, 32654
+CHROMA_VPP 24, 64655
+CHROMA_VPP 64, 16656
+CHROMA_VPP 64, 32657
+CHROMA_VPP 64, 48658
+CHROMA_VPP 64, 64659
+CHROMA_VPP 48, 64660
+661
+// ***** chroma_vps *****662
+// void interp_vert_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)663
+.macro CHROMA_VPS w, h664
+function x265_interp_4tap_vert_ps_\w\()x\h\()_neon665
+ cmp x4, #0666
+ beq 0f667
+ cmp x4, #1668
+ beq 1f669
+ cmp x4, #2670
+ beq 2f671
+ cmp x4, #3672
+ beq 3f673
+ cmp x4, #4674
+ beq 4f675
+ cmp x4, #5676
+ beq 5f677
+ cmp x4, #6678
+ beq 6f679
+ cmp x4, #7680
+ beq 7f681
+0:682
+ FILTER_CHROMA_VPS \w, \h, 0683
+1:684
+ FILTER_CHROMA_VPS \w, \h, 1685
+2:686
+ FILTER_CHROMA_VPS \w, \h, 2687
+3:688
+ FILTER_CHROMA_VPS \w, \h, 3689
+4:690
+ FILTER_CHROMA_VPS \w, \h, 4691
+5:692
+ FILTER_CHROMA_VPS \w, \h, 5693
+6:694
+ FILTER_CHROMA_VPS \w, \h, 6695
+7:696
+ FILTER_CHROMA_VPS \w, \h, 7697
+endfunc698
+.endm699
+700
+CHROMA_VPS 2, 4701
+CHROMA_VPS 2, 8702
+CHROMA_VPS 2, 16703
+CHROMA_VPS 4, 2704
+CHROMA_VPS 4, 4705
+CHROMA_VPS 4, 8706
+CHROMA_VPS 4, 16707
+CHROMA_VPS 4, 32708
+CHROMA_VPS 6, 8709
+CHROMA_VPS 6, 16710
+CHROMA_VPS 8, 2711
+CHROMA_VPS 8, 4712
+CHROMA_VPS 8, 6713
+CHROMA_VPS 8, 8714
+CHROMA_VPS 8, 16715
+CHROMA_VPS 8, 32716
+CHROMA_VPS 8, 12717
+CHROMA_VPS 8, 64718
+CHROMA_VPS 12, 16719
+CHROMA_VPS 12, 32720
+CHROMA_VPS 16, 4721
+CHROMA_VPS 16, 8722
+CHROMA_VPS 16, 12723
+CHROMA_VPS 16, 16724
+CHROMA_VPS 16, 32725
+CHROMA_VPS 16, 64726
+CHROMA_VPS 16, 24727
+CHROMA_VPS 32, 8728
+CHROMA_VPS 32, 16729
+CHROMA_VPS 32, 24730
+CHROMA_VPS 32, 32731
+CHROMA_VPS 32, 64732
+CHROMA_VPS 32, 48733
+CHROMA_VPS 24, 32734
+CHROMA_VPS 24, 64735
+CHROMA_VPS 64, 16736
+CHROMA_VPS 64, 32737
+CHROMA_VPS 64, 48738
+CHROMA_VPS 64, 64739
+CHROMA_VPS 48, 64740
+741
+// ***** chroma_vsp *****742
+// void interp_vert_sp_c(const int16_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)743
+.macro CHROMA_VSP w, h744
+function x265_interp_4tap_vert_sp_\w\()x\h\()_neon745
+ cmp x4, #0746
+ beq 0f747
+ cmp x4, #1748
+ beq 1f749
+ cmp x4, #2750
+ beq 2f751
+ cmp x4, #3752
+ beq 3f753
+ cmp x4, #4754
+ beq 4f755
+ cmp x4, #5756
+ beq 5f757
+ cmp x4, #6758
+ beq 6f759
+ cmp x4, #7760
+ beq 7f761
+0:762
+ FILTER_CHROMA_VSP \w, \h, 0763
+1:764
+ FILTER_CHROMA_VSP \w, \h, 1765
+2:766
+ FILTER_CHROMA_VSP \w, \h, 2767
+3:768
+ FILTER_CHROMA_VSP \w, \h, 3769
+4:770
+ FILTER_CHROMA_VSP \w, \h, 4771
+5:772
+ FILTER_CHROMA_VSP \w, \h, 5773
+6:774
+ FILTER_CHROMA_VSP \w, \h, 6775
+7:776
+ FILTER_CHROMA_VSP \w, \h, 7777
+endfunc778
+.endm779
+780
+CHROMA_VSP 4, 4781
+CHROMA_VSP 4, 8782
+CHROMA_VSP 4, 16783
+CHROMA_VSP 4, 32784
+CHROMA_VSP 8, 2785
+CHROMA_VSP 8, 4786
+CHROMA_VSP 8, 6787
+CHROMA_VSP 8, 8788
+CHROMA_VSP 8, 16789
+CHROMA_VSP 8, 32790
+CHROMA_VSP 8, 12791
+CHROMA_VSP 8, 64792
+CHROMA_VSP 12, 16793
+CHROMA_VSP 12, 32794
+CHROMA_VSP 16, 4795
+CHROMA_VSP 16, 8796
+CHROMA_VSP 16, 12797
+CHROMA_VSP 16, 16798
+CHROMA_VSP 16, 32799
+CHROMA_VSP 16, 64800
+CHROMA_VSP 16, 24801
+CHROMA_VSP 32, 8802
+CHROMA_VSP 32, 16803
+CHROMA_VSP 32, 24804
+CHROMA_VSP 32, 32805
+CHROMA_VSP 32, 64806
+CHROMA_VSP 32, 48807
+CHROMA_VSP 24, 32808
+CHROMA_VSP 24, 64809
+CHROMA_VSP 64, 16810
+CHROMA_VSP 64, 32811
+CHROMA_VSP 64, 48812
+CHROMA_VSP 64, 64813
+CHROMA_VSP 48, 64814
+815
+// ***** chroma_vss *****816
+// void interp_vert_ss_c(const int16_t* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx)817
+.macro CHROMA_VSS w, h818
+function x265_interp_4tap_vert_ss_\w\()x\h\()_neon819
+ cmp x4, #0820
+ beq 0f821
+ cmp x4, #1822
+ beq 1f823
+ cmp x4, #2824
+ beq 2f825
+ cmp x4, #3826
+ beq 3f827
+ cmp x4, #4828
+ beq 4f829
+ cmp x4, #5830
+ beq 5f831
+ cmp x4, #6832
+ beq 6f833
+ cmp x4, #7834
+ beq 7f835
+0:836
+ FILTER_CHROMA_VSS \w, \h, 0837
+1:838
+ FILTER_CHROMA_VSS \w, \h, 1839
+2:840
+ FILTER_CHROMA_VSS \w, \h, 2841
+3:842
+ FILTER_CHROMA_VSS \w, \h, 3843
+4:844
+ FILTER_CHROMA_VSS \w, \h, 4845
+5:846
+ FILTER_CHROMA_VSS \w, \h, 5847
+6:848
+ FILTER_CHROMA_VSS \w, \h, 6849
+7:850
+ FILTER_CHROMA_VSS \w, \h, 7851
+endfunc852
+.endm853
+854
+CHROMA_VSS 4, 4855
+CHROMA_VSS 4, 8856
+CHROMA_VSS 4, 16857
+CHROMA_VSS 4, 32858
+CHROMA_VSS 8, 2859
+CHROMA_VSS 8, 4860
+CHROMA_VSS 8, 6861
+CHROMA_VSS 8, 8862
+CHROMA_VSS 8, 16863
+CHROMA_VSS 8, 32864
+CHROMA_VSS 8, 12865
+CHROMA_VSS 8, 64866
+CHROMA_VSS 12, 16867
+CHROMA_VSS 12, 32868
+CHROMA_VSS 16, 4869
+CHROMA_VSS 16, 8870
+CHROMA_VSS 16, 12871
+CHROMA_VSS 16, 16872
+CHROMA_VSS 16, 32873
+CHROMA_VSS 16, 64874
+CHROMA_VSS 16, 24875
+CHROMA_VSS 32, 8876
+CHROMA_VSS 32, 16877
+CHROMA_VSS 32, 24878
+CHROMA_VSS 32, 32879
+CHROMA_VSS 32, 64880
+CHROMA_VSS 32, 48881
+CHROMA_VSS 24, 32882
+CHROMA_VSS 24, 64883
+CHROMA_VSS 64, 16884
+CHROMA_VSS 64, 32885
+CHROMA_VSS 64, 48886
+CHROMA_VSS 64, 64887
+CHROMA_VSS 48, 64888
+889
+// ***** chroma_hpp *****890
+// void interp_horiz_pp_c(const pixel* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int coeffIdx)891
+.macro CHROMA_HPP w, h892
+function x265_interp_4tap_horiz_pp_\w\()x\h\()_neon893
+ cmp x4, #0894
+ beq 0f895
+ cmp x4, #1896
+ beq 1f897
+ cmp x4, #2898
+ beq 2f899
+ cmp x4, #3900
+ beq 3f901
+ cmp x4, #4902
+ beq 4f903
+ cmp x4, #5904
+ beq 5f905
+ cmp x4, #6906
+ beq 6f907
+ cmp x4, #7908
+ beq 7f909
+0:910
+ FILTER_CHROMA_HPP \w, \h, 0911
+1:912
+ FILTER_CHROMA_HPP \w, \h, 1913
+2:914
+ FILTER_CHROMA_HPP \w, \h, 2915
+3:916
+ FILTER_CHROMA_HPP \w, \h, 3917
+4:918
+ FILTER_CHROMA_HPP \w, \h, 4919
+5:920
+ FILTER_CHROMA_HPP \w, \h, 5921
+6:922
+ FILTER_CHROMA_HPP \w, \h, 6923
+7:924
+ FILTER_CHROMA_HPP \w, \h, 7925
+endfunc926
+.endm927
+928
+CHROMA_HPP 2, 4929
+CHROMA_HPP 2, 8930
+CHROMA_HPP 2, 16931
+CHROMA_HPP 4, 2932
+CHROMA_HPP 4, 4933
+CHROMA_HPP 4, 8934
+CHROMA_HPP 4, 16935
+CHROMA_HPP 4, 32936
+CHROMA_HPP 6, 8937
+CHROMA_HPP 6, 16938
+CHROMA_HPP 8, 2939
+CHROMA_HPP 8, 4940
+CHROMA_HPP 8, 6941
+CHROMA_HPP 8, 8942
+CHROMA_HPP 8, 12943
+CHROMA_HPP 8, 16944
+CHROMA_HPP 8, 32945
+CHROMA_HPP 8, 64946
+CHROMA_HPP 12, 16947
+CHROMA_HPP 12, 32948
+CHROMA_HPP 16, 4949
+CHROMA_HPP 16, 8950
+CHROMA_HPP 16, 12951
+CHROMA_HPP 16, 16952
+CHROMA_HPP 16, 24953
+CHROMA_HPP 16, 32954
+CHROMA_HPP 16, 64955
+CHROMA_HPP 24, 32956
+CHROMA_HPP 24, 64957
+CHROMA_HPP 32, 8958
+CHROMA_HPP 32, 16959
+CHROMA_HPP 32, 24960
+CHROMA_HPP 32, 32961
+CHROMA_HPP 32, 48962
+CHROMA_HPP 32, 64963
+CHROMA_HPP 48, 64964
+CHROMA_HPP 64, 16965
+CHROMA_HPP 64, 32966
+CHROMA_HPP 64, 48967
+CHROMA_HPP 64, 64968
+969
+// ***** chroma_hps *****970
+// void interp_horiz_ps_c(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride, int coeffIdx, int isRowExt)971
+.macro CHROMA_HPS w, h972
+function x265_interp_4tap_horiz_ps_\w\()x\h\()_neon973
+ cmp x4, #0974
+ beq 0f975
+ cmp x4, #1976
+ beq 1f977
+ cmp x4, #2978
+ beq 2f979
+ cmp x4, #3980
+ beq 3f981
+ cmp x4, #4982
+ beq 4f983
+ cmp x4, #5984
+ beq 5f985
+ cmp x4, #6986
+ beq 6f987
+ cmp x4, #7988
+ beq 7f989
+0:990
+ FILTER_CHROMA_HPS \w, \h, 0991
+1:992
+ FILTER_CHROMA_HPS \w, \h, 1993
+2:994
+ FILTER_CHROMA_HPS \w, \h, 2995
+3:996
+ FILTER_CHROMA_HPS \w, \h, 3997
+4:998
+ FILTER_CHROMA_HPS \w, \h, 4999
+5:1000
+ FILTER_CHROMA_HPS \w, \h, 51001
+6:1002
+ FILTER_CHROMA_HPS \w, \h, 61003
+7:1004
+ FILTER_CHROMA_HPS \w, \h, 71005
+endfunc1006
+.endm1007
+1008
+CHROMA_HPS 2, 41009
+CHROMA_HPS 2, 81010
+CHROMA_HPS 2, 161011
+CHROMA_HPS 4, 21012
+CHROMA_HPS 4, 41013
+CHROMA_HPS 4, 81014
+CHROMA_HPS 4, 161015
+CHROMA_HPS 4, 321016
+CHROMA_HPS 6, 81017
+CHROMA_HPS 6, 161018
+CHROMA_HPS 8, 21019
+CHROMA_HPS 8, 41020
+CHROMA_HPS 8, 61021
+CHROMA_HPS 8, 81022
+CHROMA_HPS 8, 121023
+CHROMA_HPS 8, 161024
+CHROMA_HPS 8, 321025
+CHROMA_HPS 8, 641026
+CHROMA_HPS 12, 161027
+CHROMA_HPS 12, 321028
+CHROMA_HPS 16, 41029
+CHROMA_HPS 16, 81030
+CHROMA_HPS 16, 121031
+CHROMA_HPS 16, 161032
+CHROMA_HPS 16, 241033
+CHROMA_HPS 16, 321034
+CHROMA_HPS 16, 641035
+CHROMA_HPS 24, 321036
+CHROMA_HPS 24, 641037
+CHROMA_HPS 32, 81038
+CHROMA_HPS 32, 161039
+CHROMA_HPS 32, 241040
+CHROMA_HPS 32, 321041
+CHROMA_HPS 32, 481042
+CHROMA_HPS 32, 641043
+CHROMA_HPS 48, 641044
+CHROMA_HPS 64, 161045
+CHROMA_HPS 64, 321046
+CHROMA_HPS 64, 481047
+CHROMA_HPS 64, 641048
+1049
+const g_luma_s16, align=81050
+// a, b, c, d, e, f, g, h1051
+.hword 0, 0, 0, 64, 0, 0, 0, 01052
+.hword -1, 4, -10, 58, 17, -5, 1, 01053
+.hword -1, 4, -11, 40, 40, -11, 4, -11054
+.hword 0, 1, -5, 17, 58, -10, 4, -11055
+endconst1056
x265_3.6.tar.gz/source/common/aarch64/loopfilter-prim.cpp
Added
293
1
2
+#include "loopfilter-prim.h"3
+4
+#define PIXEL_MIN 05
+6
+7
+8
+#if !(HIGH_BIT_DEPTH) && defined(HAVE_NEON)9
+#include<arm_neon.h>10
+11
+namespace12
+{13
+14
+15
+/* get the sign of input variable (TODO: this is a dup, make common) */16
+static inline int8_t signOf(int x)17
+{18
+ return (x >> 31) | ((int)((((uint32_t) - x)) >> 31));19
+}20
+21
+static inline int8x8_t sign_diff_neon(const uint8x8_t in0, const uint8x8_t in1)22
+{23
+ int16x8_t in = vsubl_u8(in0, in1);24
+ return vmovn_s16(vmaxq_s16(vminq_s16(in, vdupq_n_s16(1)), vdupq_n_s16(-1)));25
+}26
+27
+static void calSign_neon(int8_t *dst, const pixel *src1, const pixel *src2, const int endX)28
+{29
+ int x = 0;30
+ for (; (x + 8) <= endX; x += 8)31
+ {32
+ *(int8x8_t *)&dstx = sign_diff_neon(*(uint8x8_t *)&src1x, *(uint8x8_t *)&src2x);33
+ }34
+35
+ for (; x < endX; x++)36
+ {37
+ dstx = signOf(src1x - src2x);38
+ }39
+}40
+41
+static void processSaoCUE0_neon(pixel *rec, int8_t *offsetEo, int width, int8_t *signLeft, intptr_t stride)42
+{43
+44
+45
+ int y;46
+ int8_t signRight, signLeft0;47
+ int8_t edgeType;48
+49
+ for (y = 0; y < 2; y++)50
+ {51
+ signLeft0 = signLefty;52
+ int x = 0;53
+54
+ if (width >= 8)55
+ {56
+ int8x8_t vsignRight;57
+ int8x8x2_t shifter;58
+ shifter.val10 = signLeft0;59
+ static const int8x8_t index = {8, 0, 1, 2, 3, 4, 5, 6};60
+ int8x8_t tbl = *(int8x8_t *)offsetEo;61
+ for (; (x + 8) <= width; x += 8)62
+ {63
+ uint8x8_t in = *(uint8x8_t *)&recx;64
+ vsignRight = sign_diff_neon(in, *(uint8x8_t *)&recx + 1);65
+ shifter.val0 = vneg_s8(vsignRight);66
+ int8x8_t tmp = shifter.val0;67
+ int8x8_t edge = vtbl2_s8(shifter, index);68
+ int8x8_t vedgeType = vadd_s8(vadd_s8(vsignRight, edge), vdup_n_s8(2));69
+ shifter.val10 = tmp7;70
+ int16x8_t t1 = vmovl_s8(vtbl1_s8(tbl, vedgeType));71
+ t1 = vaddw_u8(t1, in);72
+ t1 = vmaxq_s16(t1, vdupq_n_s16(0));73
+ t1 = vminq_s16(t1, vdupq_n_s16(255));74
+ *(uint8x8_t *)&recx = vmovn_u16(t1);75
+ }76
+ signLeft0 = shifter.val10;77
+ }78
+ for (; x < width; x++)79
+ {80
+ signRight = ((recx - recx + 1) < 0) ? -1 : ((recx - recx + 1) > 0) ? 1 : 0;81
+ edgeType = signRight + signLeft0 + 2;82
+ signLeft0 = -signRight;83
+ recx = x265_clip(recx + offsetEoedgeType);84
+ }85
+ rec += stride;86
+ }87
+}88
+89
+static void processSaoCUE1_neon(pixel *rec, int8_t *upBuff1, int8_t *offsetEo, intptr_t stride, int width)90
+{91
+ int x = 0;92
+ int8_t signDown;93
+ int edgeType;94
+95
+ if (width >= 8)96
+ {97
+ int8x8_t tbl = *(int8x8_t *)offsetEo;98
+ for (; (x + 8) <= width; x += 8)99
+ {100
+ uint8x8_t in0 = *(uint8x8_t *)&recx;101
+ uint8x8_t in1 = *(uint8x8_t *)&recx + stride;102
+ int8x8_t vsignDown = sign_diff_neon(in0, in1);103
+ int8x8_t vedgeType = vadd_s8(vadd_s8(vsignDown, *(int8x8_t *)&upBuff1x), vdup_n_s8(2));104
+ *(int8x8_t *)&upBuff1x = vneg_s8(vsignDown);105
+ int16x8_t t1 = vmovl_s8(vtbl1_s8(tbl, vedgeType));106
+ t1 = vaddw_u8(t1, in0);107
+ *(uint8x8_t *)&recx = vqmovun_s16(t1);108
+ }109
+ }110
+ for (; x < width; x++)111
+ {112
+ signDown = signOf(recx - recx + stride);113
+ edgeType = signDown + upBuff1x + 2;114
+ upBuff1x = -signDown;115
+ recx = x265_clip(recx + offsetEoedgeType);116
+ }117
+}118
+119
+static void processSaoCUE1_2Rows_neon(pixel *rec, int8_t *upBuff1, int8_t *offsetEo, intptr_t stride, int width)120
+{121
+ int y;122
+ int8_t signDown;123
+ int edgeType;124
+125
+ for (y = 0; y < 2; y++)126
+ {127
+ int x = 0;128
+ if (width >= 8)129
+ {130
+ int8x8_t tbl = *(int8x8_t *)offsetEo;131
+ for (; (x + 8) <= width; x += 8)132
+ {133
+ uint8x8_t in0 = *(uint8x8_t *)&recx;134
+ uint8x8_t in1 = *(uint8x8_t *)&recx + stride;135
+ int8x8_t vsignDown = sign_diff_neon(in0, in1);136
+ int8x8_t vedgeType = vadd_s8(vadd_s8(vsignDown, *(int8x8_t *)&upBuff1x), vdup_n_s8(2));137
+ *(int8x8_t *)&upBuff1x = vneg_s8(vsignDown);138
+ int16x8_t t1 = vmovl_s8(vtbl1_s8(tbl, vedgeType));139
+ t1 = vaddw_u8(t1, in0);140
+ t1 = vmaxq_s16(t1, vdupq_n_s16(0));141
+ t1 = vminq_s16(t1, vdupq_n_s16(255));142
+ *(uint8x8_t *)&recx = vmovn_u16(t1);143
+144
+ }145
+ }146
+ for (; x < width; x++)147
+ {148
+ signDown = signOf(recx - recx + stride);149
+ edgeType = signDown + upBuff1x + 2;150
+ upBuff1x = -signDown;151
+ recx = x265_clip(recx + offsetEoedgeType);152
+ }153
+ rec += stride;154
+ }155
+}156
+157
+static void processSaoCUE2_neon(pixel *rec, int8_t *bufft, int8_t *buff1, int8_t *offsetEo, int width, intptr_t stride)158
+{159
+ int x;160
+161
+ if (abs(buff1 - bufft) < 16)162
+ {163
+ for (x = 0; x < width; x++)164
+ {165
+ int8_t signDown = signOf(recx - recx + stride + 1);166
+ int edgeType = signDown + buff1x + 2;167
+ bufftx + 1 = -signDown;168
+ recx = x265_clip(recx + offsetEoedgeType);;169
+ }170
+ }171
+ else172
+ {173
+ int8x8_t tbl = *(int8x8_t *)offsetEo;174
+ x = 0;175
+ for (; (x + 8) <= width; x += 8)176
+ {177
+ uint8x8_t in0 = *(uint8x8_t *)&recx;178
+ uint8x8_t in1 = *(uint8x8_t *)&recx + stride + 1;179
+ int8x8_t vsignDown = sign_diff_neon(in0, in1);180
+ int8x8_t vedgeType = vadd_s8(vadd_s8(vsignDown, *(int8x8_t *)&buff1x), vdup_n_s8(2));181
+ *(int8x8_t *)&bufftx + 1 = vneg_s8(vsignDown);182
+ int16x8_t t1 = vmovl_s8(vtbl1_s8(tbl, vedgeType));183
+ t1 = vaddw_u8(t1, in0);184
+ t1 = vmaxq_s16(t1, vdupq_n_s16(0));185
+ t1 = vminq_s16(t1, vdupq_n_s16(255));186
+ *(uint8x8_t *)&recx = vmovn_u16(t1);187
+ }188
+ for (; x < width; x++)189
+ {190
+ int8_t signDown = signOf(recx - recx + stride + 1);191
+ int edgeType = signDown + buff1x + 2;192
+ bufftx + 1 = -signDown;193
+ recx = x265_clip(recx + offsetEoedgeType);;194
+ }195
+196
+ }197
+}198
+199
+200
+static void processSaoCUE3_neon(pixel *rec, int8_t *upBuff1, int8_t *offsetEo, intptr_t stride, int startX, int endX)201
+{202
+ int8_t signDown;203
+ int8_t edgeType;204
+ int8x8_t tbl = *(int8x8_t *)offsetEo;205
+206
+ int x = startX + 1;207
+ for (; (x + 8) <= endX; x += 8)208
+ {209
+ uint8x8_t in0 = *(uint8x8_t *)&recx;210
+ uint8x8_t in1 = *(uint8x8_t *)&recx + stride;211
+ int8x8_t vsignDown = sign_diff_neon(in0, in1);212
+ int8x8_t vedgeType = vadd_s8(vadd_s8(vsignDown, *(int8x8_t *)&upBuff1x), vdup_n_s8(2));213
+ *(int8x8_t *)&upBuff1x - 1 = vneg_s8(vsignDown);214
+ int16x8_t t1 = vmovl_s8(vtbl1_s8(tbl, vedgeType));215
+ t1 = vaddw_u8(t1, in0);216
+ t1 = vmaxq_s16(t1, vdupq_n_s16(0));217
+ t1 = vminq_s16(t1, vdupq_n_s16(255));218
+ *(uint8x8_t *)&recx = vmovn_u16(t1);219
+220
+ }221
+ for (; x < endX; x++)222
+ {223
+ signDown = signOf(recx - recx + stride);224
+ edgeType = signDown + upBuff1x + 2;225
+ upBuff1x - 1 = -signDown;226
+ recx = x265_clip(recx + offsetEoedgeType);227
+ }228
+}229
+230
+static void processSaoCUB0_neon(pixel *rec, const int8_t *offset, int ctuWidth, int ctuHeight, intptr_t stride)231
+{232
+#define SAO_BO_BITS 5233
+ const int boShift = X265_DEPTH - SAO_BO_BITS;234
+ int x, y;235
+ int8x8x4_t table;236
+ table = *(int8x8x4_t *)offset;237
+238
+ for (y = 0; y < ctuHeight; y++)239
+ {240
+241
+ for (x = 0; (x + 8) <= ctuWidth; x += 8)242
+ {243
+ int8x8_t in = *(int8x8_t *)&recx;244
+ int8x8_t offsets = vtbl4_s8(table, vshr_n_u8(in, boShift));245
+ int16x8_t tmp = vmovl_s8(offsets);246
+ tmp = vaddw_u8(tmp, in);247
+ tmp = vmaxq_s16(tmp, vdupq_n_s16(0));248
+ tmp = vminq_s16(tmp, vdupq_n_s16(255));249
+ *(uint8x8_t *)&recx = vmovn_u16(tmp);250
+ }251
+ for (; x < ctuWidth; x++)252
+ {253
+ recx = x265_clip(recx + offsetrecx >> boShift);254
+ }255
+ rec += stride;256
+ }257
+}258
+259
+}260
+261
+262
+263
+namespace X265_NS264
+{265
+void setupLoopFilterPrimitives_neon(EncoderPrimitives &p)266
+{267
+ p.saoCuOrgE0 = processSaoCUE0_neon;268
+ p.saoCuOrgE1 = processSaoCUE1_neon;269
+ p.saoCuOrgE1_2Rows = processSaoCUE1_2Rows_neon;270
+ p.saoCuOrgE20 = processSaoCUE2_neon;271
+ p.saoCuOrgE21 = processSaoCUE2_neon;272
+ p.saoCuOrgE30 = processSaoCUE3_neon;273
+ p.saoCuOrgE31 = processSaoCUE3_neon;274
+ p.saoCuOrgB0 = processSaoCUB0_neon;275
+ p.sign = calSign_neon;276
+277
+}278
+279
+280
+#else //HIGH_BIT_DEPTH281
+282
+283
+namespace X265_NS284
+{285
+void setupLoopFilterPrimitives_neon(EncoderPrimitives &)286
+{287
+}288
+289
+#endif290
+291
+292
+}293
x265_3.6.tar.gz/source/common/aarch64/loopfilter-prim.h
Added
18
1
2
+#ifndef _LOOPFILTER_NEON_H__3
+#define _LOOPFILTER_NEON_H__4
+5
+#include "common.h"6
+#include "primitives.h"7
+8
+#define PIXEL_MIN 09
+10
+namespace X265_NS11
+{12
+void setupLoopFilterPrimitives_neon(EncoderPrimitives &p);13
+14
+};15
+16
+17
+#endif18
x265_3.6.tar.gz/source/common/aarch64/mc-a-common.S
Added
50
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+.arch armv8-a29
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.macro addAvg_start37
+ lsl x3, x3, #138
+ lsl x4, x4, #139
+ mov w11, #0x4040
+ dup v30.16b, w1141
+.endm42
+43
+.macro addavg_1 v0, v144
+ add \v0\().8h, \v0\().8h, \v1\().8h45
+ saddl v16.4s, \v0\().4h, v30.4h46
+ saddl2 v17.4s, \v0\().8h, v30.8h47
+ shrn \v0\().4h, v16.4s, #748
+ shrn2 \v0\().8h, v17.4s, #749
+.endm50
x265_3.6.tar.gz/source/common/aarch64/mc-a-sve2.S
Added
926
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "mc-a-common.S"27
+28
+.arch armv8-a+sve229
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+function PFX(pixel_avg_pp_12x16_sve2)41
+ sub x1, x1, #442
+ sub x3, x3, #443
+ sub x5, x5, #444
+ ptrue p0.s, vl145
+ ptrue p1.b, vl846
+ mov x11, #447
+.rept 1648
+ ld1w {z0.s}, p0/z, x249
+ ld1b {z1.b}, p1/z, x2, x1150
+ ld1w {z2.s}, p0/z, x451
+ ld1b {z3.b}, p1/z, x4, x1152
+ add x2, x2, #453
+ add x2, x2, x354
+ add x4, x4, #455
+ add x4, x4, x556
+ urhadd z0.b, p1/m, z0.b, z2.b57
+ urhadd z1.b, p1/m, z1.b, z3.b58
+ st1b {z0.b}, p1, x059
+ st1b {z1.b}, p1, x0, x1160
+ add x0, x0, #461
+ add x0, x0, x162
+.endr63
+ ret64
+endfunc65
+66
+function PFX(pixel_avg_pp_24x32_sve2)67
+ mov w12, #468
+ rdvl x9, #169
+ cmp x9, #1670
+ bgt .vl_gt_16_pixel_avg_pp_24x3271
+ sub x1, x1, #1672
+ sub x3, x3, #1673
+ sub x5, x5, #1674
+.lpavg_24x32_sve2:75
+ sub w12, w12, #176
+.rept 877
+ ld1 {v0.16b}, x2, #1678
+ ld1 {v1.8b}, x2, x379
+ ld1 {v2.16b}, x4, #1680
+ ld1 {v3.8b}, x4, x581
+ urhadd v0.16b, v0.16b, v2.16b82
+ urhadd v1.8b, v1.8b, v3.8b83
+ st1 {v0.16b}, x0, #1684
+ st1 {v1.8b}, x0, x185
+.endr86
+ cbnz w12, .lpavg_24x32_sve287
+ ret88
+.vl_gt_16_pixel_avg_pp_24x32:89
+ mov x10, #2490
+ mov x11, #091
+ whilelt p0.b, x11, x1092
+.vl_gt_16_loop_pixel_avg_pp_24x32:93
+ sub w12, w12, #194
+.rept 895
+ ld1b {z0.b}, p0/z, x296
+ ld1b {z2.b}, p0/z, x497
+ add x2, x2, x398
+ add x4, x4, x599
+ urhadd z0.b, p0/m, z0.b, z2.b100
+ st1b {z0.b}, p0, x0101
+ add x0, x0, x1102
+.endr103
+ cbnz w12, .vl_gt_16_loop_pixel_avg_pp_24x32104
+ ret105
+endfunc106
+107
+.macro pixel_avg_pp_32xN_sve2 h108
+function PFX(pixel_avg_pp_32x\h\()_sve2)109
+ rdvl x9, #1110
+ cmp x9, #16111
+ bgt .vl_gt_16_pixel_avg_pp_32_\h112
+.rept \h113
+ ld1 {v0.16b-v1.16b}, x2, x3114
+ ld1 {v2.16b-v3.16b}, x4, x5115
+ urhadd v0.16b, v0.16b, v2.16b116
+ urhadd v1.16b, v1.16b, v3.16b117
+ st1 {v0.16b-v1.16b}, x0, x1118
+.endr119
+ ret120
+.vl_gt_16_pixel_avg_pp_32_\h:121
+ ptrue p0.b, vl32122
+.rept \h123
+ ld1b {z0.b}, p0/z, x2124
+ ld1b {z2.b}, p0/z, x4125
+ add x2, x2, x3126
+ add x4, x4, x5127
+ urhadd z0.b, p0/m, z0.b, z2.b128
+ st1b {z0.b}, p0, x0129
+ add x0, x0, x1130
+.endr131
+ ret132
+endfunc133
+.endm134
+135
+pixel_avg_pp_32xN_sve2 8136
+pixel_avg_pp_32xN_sve2 16137
+pixel_avg_pp_32xN_sve2 24138
+139
+.macro pixel_avg_pp_32xN1_sve2 h140
+function PFX(pixel_avg_pp_32x\h\()_sve2)141
+ rdvl x9, #1142
+ cmp x9, #16143
+ bgt .vl_gt_16_pixel_avg_pp_32xN1_\h144
+ mov w12, #\h / 8145
+.lpavg_sve2_32x\h\():146
+ sub w12, w12, #1147
+.rept 8148
+ ld1 {v0.16b-v1.16b}, x2, x3149
+ ld1 {v2.16b-v3.16b}, x4, x5150
+ urhadd v0.16b, v0.16b, v2.16b151
+ urhadd v1.16b, v1.16b, v3.16b152
+ st1 {v0.16b-v1.16b}, x0, x1153
+.endr154
+ cbnz w12, .lpavg_sve2_32x\h155
+ ret156
+.vl_gt_16_pixel_avg_pp_32xN1_\h:157
+ ptrue p0.b, vl32158
+ mov w12, #\h / 8159
+.eq_32_loop_pixel_avg_pp_32xN1_\h\():160
+ sub w12, w12, #1161
+.rept 8162
+ ld1b {z0.b}, p0/z, x2163
+ ld1b {z2.b}, p0/z, x4164
+ add x2, x2, x3165
+ add x4, x4, x5166
+ urhadd z0.b, p0/m, z0.b, z2.b167
+ st1b {z0.b}, p0, x0168
+ add x0, x0, x1169
+.endr170
+ cbnz w12, .eq_32_loop_pixel_avg_pp_32xN1_\h171
+ ret172
+endfunc173
+.endm174
+175
+pixel_avg_pp_32xN1_sve2 32176
+pixel_avg_pp_32xN1_sve2 64177
+178
+function PFX(pixel_avg_pp_48x64_sve2)179
+ rdvl x9, #1180
+ cmp x9, #16181
+ bgt .vl_gt_16_pixel_avg_pp_48x64182
+ mov w12, #8183
+.lpavg_48x64_sve2:184
+ sub w12, w12, #1185
+.rept 8186
+ ld1 {v0.16b-v2.16b}, x2, x3187
+ ld1 {v3.16b-v5.16b}, x4, x5188
+ urhadd v0.16b, v0.16b, v3.16b189
+ urhadd v1.16b, v1.16b, v4.16b190
+ urhadd v2.16b, v2.16b, v5.16b191
+ st1 {v0.16b-v2.16b}, x0, x1192
+.endr193
+ cbnz w12, .lpavg_48x64_sve2194
+ ret195
+.vl_gt_16_pixel_avg_pp_48x64:196
+ cmp x9, #32197
+ bgt .vl_gt_32_pixel_avg_pp_48x64198
+ ptrue p0.b, vl32199
+ ptrue p1.b, vl16200
+ mov w12, #8201
+.vl_eq_32_pixel_avg_pp_48x64:202
+ sub w12, w12, #1203
+.rept 8204
+ ld1b {z0.b}, p0/z, x2205
+ ld1b {z1.b}, p1/z, x2, #1, mul vl206
+ ld1b {z2.b}, p0/z, x4207
+ ld1b {z3.b}, p1/z, x4, #1, mul vl208
+ add x2, x2, x3209
+ add x4, x4, x5210
+ urhadd z0.b, p0/m, z0.b, z2.b211
+ urhadd z1.b, p1/m, z1.b, z3.b212
+ st1b {z0.b}, p0, x0213
+ st1b {z1.b}, p1, x0, #1, mul vl214
+ add x0, x0, x1215
+.endr216
+ cbnz w12, .vl_eq_32_pixel_avg_pp_48x64217
+ ret218
+.vl_gt_32_pixel_avg_pp_48x64:219
+ mov x10, #48220
+ mov x11, #0221
+ whilelt p0.b, x11, x10222
+ mov w12, #8223
+.loop_gt_32_pixel_avg_pp_48x64:224
+ sub w12, w12, #1225
+.rept 8226
+ ld1b {z0.b}, p0/z, x2227
+ ld1b {z2.b}, p0/z, x4228
+ add x2, x2, x3229
+ add x4, x4, x5230
+ urhadd z0.b, p0/m, z0.b, z2.b231
+ st1b {z0.b}, p0, x0232
+ add x0, x0, x1233
+.endr234
+ cbnz w12, .loop_gt_32_pixel_avg_pp_48x64235
+ ret236
+endfunc237
+238
+.macro pixel_avg_pp_64xN_sve2 h239
+function PFX(pixel_avg_pp_64x\h\()_sve2)240
+ rdvl x9, #1241
+ cmp x9, #16242
+ bgt .vl_gt_16_pixel_avg_pp_64x\h243
+ mov w12, #\h / 4244
+.lpavg_sve2_64x\h\():245
+ sub w12, w12, #1246
+.rept 4247
+ ld1 {v0.16b-v3.16b}, x2, x3248
+ ld1 {v4.16b-v7.16b}, x4, x5249
+ urhadd v0.16b, v0.16b, v4.16b250
+ urhadd v1.16b, v1.16b, v5.16b251
+ urhadd v2.16b, v2.16b, v6.16b252
+ urhadd v3.16b, v3.16b, v7.16b253
+ st1 {v0.16b-v3.16b}, x0, x1254
+.endr255
+ cbnz w12, .lpavg_sve2_64x\h256
+ ret257
+.vl_gt_16_pixel_avg_pp_64x\h\():258
+ cmp x9, #48259
+ bgt .vl_gt_48_pixel_avg_pp_64x\h260
+ ptrue p0.b, vl32261
+ mov w12, #\h / 4262
+.vl_eq_32_pixel_avg_pp_64x\h\():263
+ sub w12, w12, #1264
+.rept 4265
+ ld1b {z0.b}, p0/z, x2266
+ ld1b {z1.b}, p0/z, x2, #1, mul vl267
+ ld1b {z2.b}, p0/z, x4268
+ ld1b {z3.b}, p0/z, x4, #1, mul vl269
+ add x2, x2, x3270
+ add x4, x4, x5271
+ urhadd z0.b, p0/m, z0.b, z2.b272
+ urhadd z1.b, p0/m, z1.b, z3.b273
+ st1b {z0.b}, p0, x0274
+ st1b {z1.b}, p0, x0, #1, mul vl275
+ add x0, x0, x1276
+.endr277
+ cbnz w12, .vl_eq_32_pixel_avg_pp_64x\h278
+ ret279
+.vl_gt_48_pixel_avg_pp_64x\h\():280
+ ptrue p0.b, vl64281
+ mov w12, #\h / 4282
+.vl_eq_64_pixel_avg_pp_64x\h\():283
+ sub w12, w12, #1284
+.rept 4285
+ ld1b {z0.b}, p0/z, x2286
+ ld1b {z2.b}, p0/z, x4287
+ add x2, x2, x3288
+ add x4, x4, x5289
+ urhadd z0.b, p0/m, z0.b, z2.b290
+ st1b {z0.b}, p0, x0291
+ add x0, x0, x1292
+.endr293
+ cbnz w12, .vl_eq_64_pixel_avg_pp_64x\h294
+ ret295
+endfunc296
+.endm297
+298
+pixel_avg_pp_64xN_sve2 16299
+pixel_avg_pp_64xN_sve2 32300
+pixel_avg_pp_64xN_sve2 48301
+pixel_avg_pp_64xN_sve2 64302
+303
+// void addAvg(const int16_t* src0, const int16_t* src1, pixel* dst, intptr_t src0Stride, intptr_t src1Stride, intptr_t dstStride)304
+305
+.macro addAvg_2xN_sve2 h306
+function PFX(addAvg_2x\h\()_sve2)307
+ ptrue p0.s, vl2308
+ ptrue p1.h, vl4309
+ ptrue p2.h, vl2310
+.rept \h / 2311
+ ld1rw {z0.s}, p0/z, x0312
+ ld1rw {z1.s}, p0/z, x1313
+ add x0, x0, x3, lsl #1314
+ add x1, x1, x4, lsl #1315
+ ld1rw {z2.s}, p0/z, x0316
+ ld1rw {z3.s}, p0/z, x1317
+ add x0, x0, x3, lsl #1318
+ add x1, x1, x4, lsl #1319
+ add z0.h, p1/m, z0.h, z1.h320
+ add z2.h, p1/m, z2.h, z3.h321
+ sqrshrnb z0.b, z0.h, #7322
+ add z0.b, z0.b, #0x80323
+ sqrshrnb z2.b, z2.h, #7324
+ add z2.b, z2.b, #0x80325
+ st1b {z0.h}, p2, x2326
+ add x2, x2, x5327
+ st1b {z2.h}, p2, x2328
+ add x2, x2, x5329
+.endr330
+ ret331
+endfunc332
+.endm333
+334
+addAvg_2xN_sve2 4335
+addAvg_2xN_sve2 8336
+addAvg_2xN_sve2 16337
+338
+.macro addAvg_6xN_sve2 h339
+function PFX(addAvg_6x\h\()_sve2)340
+ mov w12, #\h / 2341
+ ptrue p0.b, vl16342
+ ptrue p2.h, vl6343
+.loop_sve2_addavg_6x\h\():344
+ sub w12, w12, #1345
+ ld1b {z0.b}, p0/z, x0346
+ ld1b {z1.b}, p0/z, x1347
+ add x0, x0, x3, lsl #1348
+ add x1, x1, x4, lsl #1349
+ ld1b {z2.b}, p0/z, x0350
+ ld1b {z3.b}, p0/z, x1351
+ add x0, x0, x3, lsl #1352
+ add x1, x1, x4, lsl #1353
+ add z0.h, p0/m, z0.h, z1.h354
+ add z2.h, p0/m, z2.h, z3.h355
+ sqrshrnb z0.b, z0.h, #7356
+ sqrshrnb z2.b, z2.h, #7357
+ add z0.b, z0.b, #0x80358
+ add z2.b, z2.b, #0x80359
+ st1b {z0.h}, p2, x2360
+ add x2, x2, x5361
+ st1b {z2.h}, p2, x2362
+ add x2, x2, x5363
+ cbnz w12, .loop_sve2_addavg_6x\h364
+ ret365
+endfunc366
+.endm367
+368
+addAvg_6xN_sve2 8369
+addAvg_6xN_sve2 16370
+371
+.macro addAvg_8xN_sve2 h372
+function PFX(addAvg_8x\h\()_sve2)373
+ ptrue p0.b, vl16374
+.rept \h / 2375
+ ld1b {z0.b}, p0/z, x0376
+ ld1b {z1.b}, p0/z, x1377
+ add x0, x0, x3, lsl #1378
+ add x1, x1, x4, lsl #1379
+ ld1b {z2.b}, p0/z, x0380
+ ld1b {z3.b}, p0/z, x1381
+ add x0, x0, x3, lsl #1382
+ add x1, x1, x4, lsl #1383
+ add z0.h, p0/m, z0.h, z1.h384
+ add z2.h, p0/m, z2.h, z3.h385
+ sqrshrnb z0.b, z0.h, #7386
+ add z0.b, z0.b, #0x80387
+ sqrshrnb z2.b, z2.h, #7388
+ add z2.b, z2.b, #0x80389
+ st1b {z0.h}, p0, x2390
+ add x2, x2, x5391
+ st1b {z2.h}, p0, x2392
+ add x2, x2, x5393
+.endr394
+ ret395
+endfunc396
+.endm397
+398
+.macro addAvg_8xN1_sve2 h399
+function PFX(addAvg_8x\h\()_sve2)400
+ mov w12, #\h / 2401
+ ptrue p0.b, vl16402
+.loop_sve2_addavg_8x\h\():403
+ sub w12, w12, #1404
+ ld1b {z0.b}, p0/z, x0405
+ ld1b {z1.b}, p0/z, x1406
+ add x0, x0, x3, lsl #1407
+ add x1, x1, x4, lsl #1408
+ ld1b {z2.b}, p0/z, x0409
+ ld1b {z3.b}, p0/z, x1410
+ add x0, x0, x3, lsl #1411
+ add x1, x1, x4, lsl #1412
+ add z0.h, p0/m, z0.h, z1.h413
+ add z2.h, p0/m, z2.h, z3.h414
+ sqrshrnb z0.b, z0.h, #7415
+ add z0.b, z0.b, #0x80416
+ sqrshrnb z2.b, z2.h, #7417
+ add z2.b, z2.b, #0x80418
+ st1b {z0.h}, p0, x2419
+ add x2, x2, x5420
+ st1b {z2.h}, p0, x2421
+ add x2, x2, x5422
+ cbnz w12, .loop_sve2_addavg_8x\h423
+ ret424
+endfunc425
+.endm426
+427
+addAvg_8xN_sve2 2428
+addAvg_8xN_sve2 4429
+addAvg_8xN_sve2 6430
+addAvg_8xN_sve2 8431
+addAvg_8xN_sve2 12432
+addAvg_8xN_sve2 16433
+addAvg_8xN1_sve2 32434
+addAvg_8xN1_sve2 64435
+436
+.macro addAvg_12xN_sve2 h437
+function PFX(addAvg_12x\h\()_sve2)438
+ mov w12, #\h439
+ rdvl x9, #1440
+ cmp x9, #16441
+ bgt .vl_gt_16_addAvg_12x\h442
+ ptrue p0.b, vl16443
+ ptrue p1.b, vl8444
+.loop_sve2_addavg_12x\h\():445
+ sub w12, w12, #1446
+ ld1b {z0.b}, p0/z, x0447
+ ld1b {z1.b}, p0/z, x1448
+ ld1b {z2.b}, p1/z, x0, #1, mul vl449
+ ld1b {z3.b}, p1/z, x1, #1, mul vl450
+ add x0, x0, x3, lsl #1451
+ add x1, x1, x4, lsl #1452
+ add z0.h, p0/m, z0.h, z1.h453
+ add z2.h, p1/m, z2.h, z3.h454
+ sqrshrnb z0.b, z0.h, #7455
+ add z0.b, z0.b, #0x80456
+ sqrshrnb z2.b, z2.h, #7457
+ add z2.b, z2.b, #0x80458
+ st1b {z0.h}, p0, x2459
+ st1b {z2.h}, p1, x2, #1, mul vl460
+ add x2, x2, x5461
+ cbnz w12, .loop_sve2_addavg_12x\h462
+ ret463
+.vl_gt_16_addAvg_12x\h\():464
+ mov x10, #24465
+ mov x11, #0466
+ whilelt p0.b, x11, x10467
+.loop_sve2_gt_16_addavg_12x\h\():468
+ sub w12, w12, #1469
+ ld1b {z0.b}, p0/z, x0470
+ ld1b {z1.b}, p0/z, x1471
+ add x0, x0, x3, lsl #1472
+ add x1, x1, x4, lsl #1473
+ add z0.h, p0/m, z0.h, z1.h474
+ sqrshrnb z0.b, z0.h, #7475
+ add z0.b, z0.b, #0x80476
+ sqrshrnb z2.b, z2.h, #7477
+ add z2.b, z2.b, #0x80478
+ st1b {z0.h}, p0, x2479
+ add x2, x2, x5480
+ cbnz w12, .loop_sve2_gt_16_addavg_12x\h481
+ ret482
+endfunc483
+.endm484
+485
+addAvg_12xN_sve2 16486
+addAvg_12xN_sve2 32487
+488
+.macro addAvg_16xN_sve2 h489
+function PFX(addAvg_16x\h\()_sve2)490
+ mov w12, #\h491
+ rdvl x9, #1492
+ cmp x9, #16493
+ bgt .vl_gt_16_addAvg_16x\h494
+ ptrue p0.b, vl16495
+.loop_eq_16_sve2_addavg_16x\h\():496
+ sub w12, w12, #1497
+ ld1b {z0.b}, p0/z, x0498
+ ld1b {z1.b}, p0/z, x1499
+ ld1b {z2.b}, p0/z, x0, #1, mul vl500
+ ld1b {z3.b}, p0/z, x1, #1, mul vl501
+ add x0, x0, x3, lsl #1502
+ add x1, x1, x4, lsl #1503
+ add z0.h, p0/m, z0.h, z1.h504
+ add z2.h, p0/m, z2.h, z3.h505
+ sqrshrnb z0.b, z0.h, #7506
+ add z0.b, z0.b, #0x80507
+ sqrshrnb z2.b, z2.h, #7508
+ add z2.b, z2.b, #0x80509
+ st1b {z0.h}, p0, x2510
+ st1b {z2.h}, p0, x2, #1, mul vl511
+ add x2, x2, x5512
+ cbnz w12, .loop_eq_16_sve2_addavg_16x\h513
+ ret514
+.vl_gt_16_addAvg_16x\h\():515
+ cmp x9, #32516
+ bgt .vl_gt_32_addAvg_16x\h517
+ ptrue p0.b, vl32518
+.loop_gt_16_sve2_addavg_16x\h\():519
+ sub w12, w12, #1520
+ ld1b {z0.b}, p0/z, x0521
+ ld1b {z1.b}, p0/z, x1522
+ add x0, x0, x3, lsl #1523
+ add x1, x1, x4, lsl #1524
+ add z0.h, p0/m, z0.h, z1.h525
+ sqrshrnb z0.b, z0.h, #7526
+ add z0.b, z0.b, #0x80527
+ st1b {z0.h}, p1, x2528
+ add x2, x2, x5529
+ cbnz w12, .loop_gt_16_sve2_addavg_16x\h530
+ ret531
+.vl_gt_32_addAvg_16x\h\():532
+ mov x10, #48533
+ mov x11, #0534
+ whilelt p0.b, x11, x10535
+.loop_gt_32_sve2_addavg_16x\h\():536
+ sub w12, w12, #1537
+ ld1b {z0.b}, p0/z, x0538
+ add x0, x0, x3, lsl #1539
+ add x1, x1, x4, lsl #1540
+ add z0.h, p0/m, z0.h, z1.h541
+ sqrshrnb z0.b, z0.h, #7542
+ add z0.b, z0.b, #0x80543
+ st1b {z0.h}, p0, x2544
+ add x2, x2, x5545
+ cbnz w12, .loop_gt_32_sve2_addavg_16x\h546
+ ret547
+endfunc548
+.endm549
+550
+addAvg_16xN_sve2 4551
+addAvg_16xN_sve2 8552
+addAvg_16xN_sve2 12553
+addAvg_16xN_sve2 16554
+addAvg_16xN_sve2 24555
+addAvg_16xN_sve2 32556
+addAvg_16xN_sve2 64557
+558
+.macro addAvg_24xN_sve2 h559
+function PFX(addAvg_24x\h\()_sve2)560
+ mov w12, #\h561
+ rdvl x9, #1562
+ cmp x9, #16563
+ bgt .vl_gt_16_addAvg_24x\h564
+ addAvg_start565
+.loop_eq_16_sve2_addavg_24x\h\():566
+ sub w12, w12, #1567
+ ld1 {v0.16b-v2.16b}, x0, x3568
+ ld1 {v3.16b-v5.16b}, x1, x4569
+ addavg_1 v0, v3570
+ addavg_1 v1, v4571
+ addavg_1 v2, v5572
+ sqxtun v0.8b, v0.8h573
+ sqxtun v1.8b, v1.8h574
+ sqxtun v2.8b, v2.8h575
+ st1 {v0.8b-v2.8b}, x2, x5576
+ cbnz w12, .loop_eq_16_sve2_addavg_24x\h577
+ ret578
+.vl_gt_16_addAvg_24x\h\():579
+ cmp x9, #48580
+ bgt .vl_gt_48_addAvg_24x\h581
+ ptrue p0.b, vl32582
+ ptrue p1.b, vl16583
+.loop_gt_16_sve2_addavg_24x\h\():584
+ sub w12, w12, #1585
+ ld1b {z0.b}, p0/z, x0586
+ ld1b {z1.b}, p1/z, x0, #1, mul vl587
+ ld1b {z2.b}, p0/z, x1588
+ ld1b {z3.b}, p1/z, x1, #1, mul vl589
+ add x0, x0, x3, lsl #1590
+ add x1, x1, x4, lsl #1591
+ add z0.h, p0/m, z0.h, z2.h592
+ add z1.h, p1/m, z1.h, z3.h593
+ sqrshrnb z0.b, z0.h, #7594
+ add z0.b, z0.b, #0x80595
+ sqrshrnb z1.b, z1.h, #7596
+ add z1.b, z1.b, #0x80597
+ st1b {z0.h}, p0, x2598
+ st1b {z1.h}, p1, x2, #1, mul vl599
+ add x2, x2, x5600
+ cbnz w12, .loop_gt_16_sve2_addavg_24x\h601
+ ret602
+.vl_gt_48_addAvg_24x\h\():603
+ mov x10, #48604
+ mov x11, #0605
+ whilelt p0.b, x11, x10606
+.loop_gt_48_sve2_addavg_24x\h\():607
+ sub w12, w12, #1608
+ ld1b {z0.b}, p0/z, x0609
+ ld1b {z2.b}, p0/z, x1610
+ add x0, x0, x3, lsl #1611
+ add x1, x1, x4, lsl #1612
+ add z0.h, p0/m, z0.h, z2.h613
+ sqrshrnb z0.b, z0.h, #7614
+ add z0.b, z0.b, #0x80615
+ st1b {z0.h}, p0, x2616
+ add x2, x2, x5617
+ cbnz w12, .loop_gt_48_sve2_addavg_24x\h618
+ ret619
+endfunc620
+.endm621
+622
+addAvg_24xN_sve2 32623
+addAvg_24xN_sve2 64624
+625
+.macro addAvg_32xN_sve2 h626
+function PFX(addAvg_32x\h\()_sve2)627
+ mov w12, #\h628
+ rdvl x9, #1629
+ cmp x9, #16630
+ bgt .vl_gt_16_addAvg_32x\h631
+ ptrue p0.b, vl16632
+.loop_eq_16_sve2_addavg_32x\h\():633
+ sub w12, w12, #1634
+ ld1b {z0.b}, p0/z, x0635
+ ld1b {z1.b}, p0/z, x0, #1, mul vl636
+ ld1b {z2.b}, p0/z, x0, #2, mul vl637
+ ld1b {z3.b}, p0/z, x0, #3, mul vl638
+ ld1b {z4.b}, p0/z, x1639
+ ld1b {z5.b}, p0/z, x1, #1, mul vl640
+ ld1b {z6.b}, p0/z, x1, #2, mul vl641
+ ld1b {z7.b}, p0/z, x1, #3, mul vl642
+ add x0, x0, x3, lsl #1643
+ add x1, x1, x4, lsl #1644
+ add z0.h, p0/m, z0.h, z4.h645
+ add z1.h, p0/m, z1.h, z5.h646
+ add z2.h, p0/m, z2.h, z6.h647
+ add z3.h, p0/m, z3.h, z7.h648
+ sqrshrnb z0.b, z0.h, #7649
+ add z0.b, z0.b, #0x80650
+ sqrshrnb z1.b, z1.h, #7651
+ add z1.b, z1.b, #0x80652
+ sqrshrnb z2.b, z2.h, #7653
+ add z2.b, z2.b, #0x80654
+ sqrshrnb z3.b, z3.h, #7655
+ add z3.b, z3.b, #0x80656
+ st1b {z0.h}, p0, x2657
+ st1b {z1.h}, p0, x2, #1, mul vl658
+ st1b {z2.h}, p0, x2, #2, mul vl659
+ st1b {z3.h}, p0, x2, #3, mul vl660
+ add x2, x2, x5661
+ cbnz w12, .loop_eq_16_sve2_addavg_32x\h662
+ ret663
+.vl_gt_16_addAvg_32x\h\():664
+ cmp x9, #48665
+ bgt .vl_gt_48_addAvg_32x\h666
+ ptrue p0.b, vl32667
+.loop_gt_eq_32_sve2_addavg_32x\h\():668
+ sub w12, w12, #1669
+ ld1b {z0.b}, p0/z, x0670
+ ld1b {z1.b}, p0/z, x0, #1, mul vl671
+ ld1b {z2.b}, p0/z, x1672
+ ld1b {z3.b}, p0/z, x1, #1, mul vl673
+ add x0, x0, x3, lsl #1674
+ add x1, x1, x4, lsl #1675
+ add z0.h, p0/m, z0.h, z2.h676
+ add z1.h, p0/m, z1.h, z3.h677
+ sqrshrnb z0.b, z0.h, #7678
+ add z1.b, z1.b, #0x80679
+ sqrshrnb z1.b, z1.h, #7680
+ add z0.b, z0.b, #0x80681
+ st1b {z0.h}, p0, x2682
+ st1b {z1.h}, p0, x2, #1, mul vl683
+ add x2, x2, x5684
+ cbnz w12, .loop_gt_eq_32_sve2_addavg_32x\h685
+ ret686
+.vl_gt_48_addAvg_32x\h\():687
+ ptrue p0.b, vl64688
+.loop_eq_64_sve2_addavg_32x\h\():689
+ sub w12, w12, #1690
+ ld1b {z0.b}, p0/z, x0691
+ ld1b {z1.b}, p0/z, x1692
+ add x0, x0, x3, lsl #1693
+ add x1, x1, x4, lsl #1694
+ add z0.h, p0/m, z0.h, z1.h695
+ sqrshrnb z0.b, z0.h, #7696
+ add z0.b, z0.b, #0x80697
+ st1b {z0.h}, p0, x2698
+ add x2, x2, x5699
+ cbnz w12, .loop_eq_64_sve2_addavg_32x\h700
+ ret701
+endfunc702
+.endm703
+704
+addAvg_32xN_sve2 8705
+addAvg_32xN_sve2 16706
+addAvg_32xN_sve2 24707
+addAvg_32xN_sve2 32708
+addAvg_32xN_sve2 48709
+addAvg_32xN_sve2 64710
+711
+function PFX(addAvg_48x64_sve2)712
+ mov w12, #64713
+ rdvl x9, #1714
+ cmp x9, #16715
+ bgt .vl_gt_16_addAvg_48x64716
+ addAvg_start717
+ sub x3, x3, #64718
+ sub x4, x4, #64719
+.loop_eq_16_sve2_addavg_48x64:720
+ sub w12, w12, #1721
+ ld1 {v0.8h-v3.8h}, x0, #64722
+ ld1 {v4.8h-v7.8h}, x1, #64723
+ ld1 {v20.8h-v21.8h}, x0, x3724
+ ld1 {v22.8h-v23.8h}, x1, x4725
+ addavg_1 v0, v4726
+ addavg_1 v1, v5727
+ addavg_1 v2, v6728
+ addavg_1 v3, v7729
+ addavg_1 v20, v22730
+ addavg_1 v21, v23731
+ sqxtun v0.8b, v0.8h732
+ sqxtun2 v0.16b, v1.8h733
+ sqxtun v1.8b, v2.8h734
+ sqxtun2 v1.16b, v3.8h735
+ sqxtun v2.8b, v20.8h736
+ sqxtun2 v2.16b, v21.8h737
+ st1 {v0.16b-v2.16b}, x2, x5738
+ cbnz w12, .loop_eq_16_sve2_addavg_48x64739
+ ret740
+.vl_gt_16_addAvg_48x64:741
+ cmp x9, #48742
+ bgt .vl_gt_48_addAvg_48x64743
+ ptrue p0.b, vl32744
+.loop_gt_eq_32_sve2_addavg_48x64:745
+ sub w12, w12, #1746
+ ld1b {z0.b}, p0/z, x0747
+ ld1b {z1.b}, p0/z, x0, #1, mul vl748
+ ld1b {z2.b}, p0/z, x0, #2, mul vl749
+ ld1b {z4.b}, p0/z, x1750
+ ld1b {z5.b}, p0/z, x1, #1, mul vl751
+ ld1b {z6.b}, p0/z, x1, #2, mul vl752
+ add x0, x0, x3, lsl #1753
+ add x1, x1, x4, lsl #1754
+ add z0.h, p0/m, z0.h, z4.h755
+ add z1.h, p0/m, z1.h, z5.h756
+ add z2.h, p0/m, z2.h, z6.h757
+ sqrshrnb z0.b, z0.h, #7758
+ add z0.b, z0.b, #0x80759
+ sqrshrnb z1.b, z1.h, #7760
+ add z1.b, z1.b, #0x80761
+ sqrshrnb z2.b, z2.h, #7762
+ add z2.b, z2.b, #0x80763
+ st1b {z0.h}, p0, x2764
+ st1b {z1.h}, p0, x2, #1, mul vl765
+ st1b {z2.h}, p0, x2, #2, mul vl766
+ add x2, x2, x5767
+ cbnz w12, .loop_gt_eq_32_sve2_addavg_48x64768
+ ret769
+.vl_gt_48_addAvg_48x64:770
+ cmp x9, #112771
+ bgt .vl_gt_112_addAvg_48x64772
+ ptrue p0.b, vl64773
+ ptrue p1.b, vl32774
+.loop_gt_48_sve2_addavg_48x64:775
+ sub w12, w12, #1776
+ ld1b {z0.b}, p0/z, x0777
+ ld1b {z1.b}, p1/z, x0, #1, mul vl778
+ ld1b {z4.b}, p0/z, x1779
+ ld1b {z5.b}, p1/z, x1, #1, mul vl780
+ add x0, x0, x3, lsl #1781
+ add x1, x1, x4, lsl #1782
+ add z0.h, p0/m, z0.h, z4.h783
+ add z1.h, p1/m, z1.h, z5.h784
+ sqrshrnb z0.b, z0.h, #7785
+ add z0.b, z0.b, #0x80786
+ sqrshrnb z1.b, z1.h, #7787
+ add z1.b, z1.b, #0x80788
+ st1b {z0.h}, p0, x2789
+ st1b {z1.h}, p1, x2, #1, mul vl790
+ add x2, x2, x5791
+ cbnz w12, .loop_gt_48_sve2_addavg_48x64792
+ ret793
+.vl_gt_112_addAvg_48x64:794
+ mov x10, #96795
+ mov x11, #0796
+ whilelt p0.b, x11, x10797
+.loop_gt_112_sve2_addavg_48x64:798
+ sub w12, w12, #1799
+ ld1b {z0.b}, p0/z, x0800
+ ld1b {z4.b}, p0/z, x1801
+ add x0, x0, x3, lsl #1802
+ add x1, x1, x4, lsl #1803
+ add z0.h, p0/m, z0.h, z4.h804
+ sqrshrnb z0.b, z0.h, #7805
+ add z0.b, z0.b, #0x80806
+ st1b {z0.h}, p0, x2807
+ add x2, x2, x5808
+ cbnz w12, .loop_gt_112_sve2_addavg_48x64809
+ ret810
+endfunc811
+812
+.macro addAvg_64xN_sve2 h813
+function PFX(addAvg_64x\h\()_sve2)814
+ mov w12, #\h815
+ rdvl x9, #1816
+ cmp x9, #16817
+ bgt .vl_gt_16_addAvg_64x\h818
+ addAvg_start819
+ sub x3, x3, #64820
+ sub x4, x4, #64821
+.loop_eq_16_sve2_addavg_64x\h\():822
+ sub w12, w12, #1823
+ ld1 {v0.8h-v3.8h}, x0, #64824
+ ld1 {v4.8h-v7.8h}, x1, #64825
+ ld1 {v20.8h-v23.8h}, x0, x3826
+ ld1 {v24.8h-v27.8h}, x1, x4827
+ addavg_1 v0, v4828
+ addavg_1 v1, v5829
+ addavg_1 v2, v6830
+ addavg_1 v3, v7831
+ addavg_1 v20, v24832
+ addavg_1 v21, v25833
+ addavg_1 v22, v26834
+ addavg_1 v23, v27835
+ sqxtun v0.8b, v0.8h836
+ sqxtun2 v0.16b, v1.8h837
+ sqxtun v1.8b, v2.8h838
+ sqxtun2 v1.16b, v3.8h839
+ sqxtun v2.8b, v20.8h840
+ sqxtun2 v2.16b, v21.8h841
+ sqxtun v3.8b, v22.8h842
+ sqxtun2 v3.16b, v23.8h843
+ st1 {v0.16b-v3.16b}, x2, x5844
+ cbnz w12, .loop_eq_16_sve2_addavg_64x\h845
+ ret846
+.vl_gt_16_addAvg_64x\h\():847
+ cmp x9, #48848
+ bgt .vl_gt_48_addAvg_64x\h849
+ ptrue p0.b, vl32850
+.loop_gt_eq_32_sve2_addavg_64x\h\():851
+ sub w12, w12, #1852
+ ld1b {z0.b}, p0/z, x0853
+ ld1b {z1.b}, p0/z, x0, #1, mul vl854
+ ld1b {z2.b}, p0/z, x0, #2, mul vl855
+ ld1b {z3.b}, p0/z, x0, #3, mul vl856
+ ld1b {z4.b}, p0/z, x1857
+ ld1b {z5.b}, p0/z, x1, #1, mul vl858
+ ld1b {z6.b}, p0/z, x1, #2, mul vl859
+ ld1b {z7.b}, p0/z, x1, #3, mul vl860
+ add x0, x0, x3, lsl #1861
+ add x1, x1, x4, lsl #1862
+ add z0.h, p0/m, z0.h, z4.h863
+ add z1.h, p0/m, z1.h, z5.h864
+ add z2.h, p0/m, z2.h, z6.h865
+ add z3.h, p0/m, z3.h, z7.h866
+ sqrshrnb z0.b, z0.h, #7867
+ add z0.b, z0.b, #0x80868
+ sqrshrnb z1.b, z1.h, #7869
+ add z1.b, z1.b, #0x80870
+ sqrshrnb z2.b, z2.h, #7871
+ add z2.b, z2.b, #0x80872
+ sqrshrnb z3.b, z3.h, #7873
+ add z3.b, z3.b, #0x80874
+ st1b {z0.h}, p0, x2875
+ st1b {z1.h}, p0, x2, #1, mul vl876
+ st1b {z2.h}, p0, x2, #2, mul vl877
+ st1b {z3.h}, p0, x2, #3, mul vl878
+ add x2, x2, x5879
+ cbnz w12, .loop_gt_eq_32_sve2_addavg_64x\h880
+ ret881
+.vl_gt_48_addAvg_64x\h\():882
+ cmp x9, #112883
+ bgt .vl_gt_112_addAvg_64x\h884
+ ptrue p0.b, vl64885
+.loop_gt_eq_48_sve2_addavg_64x\h\():886
+ sub w12, w12, #1887
+ ld1b {z0.b}, p0/z, x0888
+ ld1b {z1.b}, p0/z, x0, #1, mul vl889
+ ld1b {z4.b}, p0/z, x1890
+ ld1b {z5.b}, p0/z, x1, #1, mul vl891
+ add x0, x0, x3, lsl #1892
+ add x1, x1, x4, lsl #1893
+ add z0.h, p0/m, z0.h, z4.h894
+ add z1.h, p0/m, z1.h, z5.h895
+ sqrshrnb z0.b, z0.h, #7896
+ add z0.b, z0.b, #0x80897
+ sqrshrnb z1.b, z1.h, #7898
+ add z1.b, z1.b, #0x80899
+ st1b {z0.h}, p0, x2900
+ st1b {z1.h}, p0, x2, #1, mul vl901
+ add x2, x2, x5902
+ cbnz w12, .loop_gt_eq_48_sve2_addavg_64x\h903
+ ret904
+.vl_gt_112_addAvg_64x\h\():905
+ ptrue p0.b, vl128906
+.loop_gt_eq_128_sve2_addavg_64x\h\():907
+ sub w12, w12, #1908
+ ld1b {z0.b}, p0/z, x0909
+ ld1b {z4.b}, p0/z, x1910
+ add x0, x0, x3, lsl #1911
+ add x1, x1, x4, lsl #1912
+ add z0.h, p0/m, z0.h, z4.h913
+ sqrshrnb z0.b, z0.h, #7914
+ add z0.b, z0.b, #0x80915
+ st1b {z0.h}, p0, x2916
+ add x2, x2, x5917
+ cbnz w12, .loop_gt_eq_128_sve2_addavg_64x\h918
+ ret919
+endfunc920
+.endm921
+922
+addAvg_64xN_sve2 16923
+addAvg_64xN_sve2 32924
+addAvg_64xN_sve2 48925
+addAvg_64xN_sve2 64926
x265_3.5.tar.gz/source/common/aarch64/mc-a.S -> x265_3.6.tar.gz/source/common/aarch64/mc-a.S
Changed
534
1
2
/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
+ * Copyright (C) 2020-2021 MulticoreWare, Inc5
*6
* Authors: Hongbin Liu <liuhongbin1@huawei.com>7
+ * Sebastian Pop <spop@amazon.com>8
*9
* This program is free software; you can redistribute it and/or modify10
* it under the terms of the GNU General Public License as published by11
12
*****************************************************************************/13
14
#include "asm.S"15
+#include "mc-a-common.S"16
17
+#ifdef __APPLE__18
+.section __RODATA,__rodata19
+#else20
.section .rodata21
+#endif22
23
.align 424
25
.text26
27
.macro pixel_avg_pp_4xN_neon h28
-function x265_pixel_avg_pp_4x\h\()_neon29
+function PFX(pixel_avg_pp_4x\h\()_neon)30
.rept \h31
ld1 {v0.s}0, x2, x332
ld1 {v1.s}0, x4, x533
34
pixel_avg_pp_4xN_neon 1635
36
.macro pixel_avg_pp_8xN_neon h37
-function x265_pixel_avg_pp_8x\h\()_neon38
+function PFX(pixel_avg_pp_8x\h\()_neon)39
.rept \h40
ld1 {v0.8b}, x2, x341
ld1 {v1.8b}, x4, x542
43
pixel_avg_pp_8xN_neon 844
pixel_avg_pp_8xN_neon 1645
pixel_avg_pp_8xN_neon 3246
+47
+function PFX(pixel_avg_pp_12x16_neon)48
+ sub x1, x1, #449
+ sub x3, x3, #450
+ sub x5, x5, #451
+.rept 1652
+ ld1 {v0.s}0, x2, #453
+ ld1 {v1.8b}, x2, x354
+ ld1 {v2.s}0, x4, #455
+ ld1 {v3.8b}, x4, x556
+ urhadd v4.8b, v0.8b, v2.8b57
+ urhadd v5.8b, v1.8b, v3.8b58
+ st1 {v4.s}0, x0, #459
+ st1 {v5.8b}, x0, x160
+.endr61
+ ret62
+endfunc63
+64
+.macro pixel_avg_pp_16xN_neon h65
+function PFX(pixel_avg_pp_16x\h\()_neon)66
+.rept \h67
+ ld1 {v0.16b}, x2, x368
+ ld1 {v1.16b}, x4, x569
+ urhadd v2.16b, v0.16b, v1.16b70
+ st1 {v2.16b}, x0, x171
+.endr72
+ ret73
+endfunc74
+.endm75
+76
+pixel_avg_pp_16xN_neon 477
+pixel_avg_pp_16xN_neon 878
+pixel_avg_pp_16xN_neon 1279
+pixel_avg_pp_16xN_neon 1680
+pixel_avg_pp_16xN_neon 3281
+82
+function PFX(pixel_avg_pp_16x64_neon)83
+ mov w12, #884
+.lpavg_16x64:85
+ sub w12, w12, #186
+.rept 887
+ ld1 {v0.16b}, x2, x388
+ ld1 {v1.16b}, x4, x589
+ urhadd v2.16b, v0.16b, v1.16b90
+ st1 {v2.16b}, x0, x191
+.endr92
+ cbnz w12, .lpavg_16x6493
+ ret94
+endfunc95
+96
+function PFX(pixel_avg_pp_24x32_neon)97
+ sub x1, x1, #1698
+ sub x3, x3, #1699
+ sub x5, x5, #16100
+ mov w12, #4101
+.lpavg_24x32:102
+ sub w12, w12, #1103
+.rept 8104
+ ld1 {v0.16b}, x2, #16105
+ ld1 {v1.8b}, x2, x3106
+ ld1 {v2.16b}, x4, #16107
+ ld1 {v3.8b}, x4, x5108
+ urhadd v0.16b, v0.16b, v2.16b109
+ urhadd v1.8b, v1.8b, v3.8b110
+ st1 {v0.16b}, x0, #16111
+ st1 {v1.8b}, x0, x1112
+.endr113
+ cbnz w12, .lpavg_24x32114
+ ret115
+endfunc116
+117
+.macro pixel_avg_pp_32xN_neon h118
+function PFX(pixel_avg_pp_32x\h\()_neon)119
+.rept \h120
+ ld1 {v0.16b-v1.16b}, x2, x3121
+ ld1 {v2.16b-v3.16b}, x4, x5122
+ urhadd v0.16b, v0.16b, v2.16b123
+ urhadd v1.16b, v1.16b, v3.16b124
+ st1 {v0.16b-v1.16b}, x0, x1125
+.endr126
+ ret127
+endfunc128
+.endm129
+130
+pixel_avg_pp_32xN_neon 8131
+pixel_avg_pp_32xN_neon 16132
+pixel_avg_pp_32xN_neon 24133
+134
+.macro pixel_avg_pp_32xN1_neon h135
+function PFX(pixel_avg_pp_32x\h\()_neon)136
+ mov w12, #\h / 8137
+.lpavg_32x\h\():138
+ sub w12, w12, #1139
+.rept 8140
+ ld1 {v0.16b-v1.16b}, x2, x3141
+ ld1 {v2.16b-v3.16b}, x4, x5142
+ urhadd v0.16b, v0.16b, v2.16b143
+ urhadd v1.16b, v1.16b, v3.16b144
+ st1 {v0.16b-v1.16b}, x0, x1145
+.endr146
+ cbnz w12, .lpavg_32x\h147
+ ret148
+endfunc149
+.endm150
+151
+pixel_avg_pp_32xN1_neon 32152
+pixel_avg_pp_32xN1_neon 64153
+154
+function PFX(pixel_avg_pp_48x64_neon)155
+ mov w12, #8156
+.lpavg_48x64:157
+ sub w12, w12, #1158
+.rept 8159
+ ld1 {v0.16b-v2.16b}, x2, x3160
+ ld1 {v3.16b-v5.16b}, x4, x5161
+ urhadd v0.16b, v0.16b, v3.16b162
+ urhadd v1.16b, v1.16b, v4.16b163
+ urhadd v2.16b, v2.16b, v5.16b164
+ st1 {v0.16b-v2.16b}, x0, x1165
+.endr166
+ cbnz w12, .lpavg_48x64167
+ ret168
+endfunc169
+170
+.macro pixel_avg_pp_64xN_neon h171
+function PFX(pixel_avg_pp_64x\h\()_neon)172
+ mov w12, #\h / 4173
+.lpavg_64x\h\():174
+ sub w12, w12, #1175
+.rept 4176
+ ld1 {v0.16b-v3.16b}, x2, x3177
+ ld1 {v4.16b-v7.16b}, x4, x5178
+ urhadd v0.16b, v0.16b, v4.16b179
+ urhadd v1.16b, v1.16b, v5.16b180
+ urhadd v2.16b, v2.16b, v6.16b181
+ urhadd v3.16b, v3.16b, v7.16b182
+ st1 {v0.16b-v3.16b}, x0, x1183
+.endr184
+ cbnz w12, .lpavg_64x\h185
+ ret186
+endfunc187
+.endm188
+189
+pixel_avg_pp_64xN_neon 16190
+pixel_avg_pp_64xN_neon 32191
+pixel_avg_pp_64xN_neon 48192
+pixel_avg_pp_64xN_neon 64193
+194
+// void addAvg(const int16_t* src0, const int16_t* src1, pixel* dst, intptr_t src0Stride, intptr_t src1Stride, intptr_t dstStride)195
+.macro addAvg_2xN h196
+function PFX(addAvg_2x\h\()_neon)197
+ addAvg_start198
+.rept \h / 2199
+ ldr w10, x0200
+ ldr w11, x1201
+ add x0, x0, x3202
+ add x1, x1, x4203
+ ldr w12, x0204
+ ldr w13, x1205
+ add x0, x0, x3206
+ add x1, x1, x4207
+ dup v0.2s, w10208
+ dup v1.2s, w11209
+ dup v2.2s, w12210
+ dup v3.2s, w13211
+ add v0.4h, v0.4h, v1.4h212
+ add v2.4h, v2.4h, v3.4h213
+ saddl v0.4s, v0.4h, v30.4h214
+ saddl v2.4s, v2.4h, v30.4h215
+ shrn v0.4h, v0.4s, #7216
+ shrn2 v0.8h, v2.4s, #7217
+ sqxtun v0.8b, v0.8h218
+ st1 {v0.h}0, x2, x5219
+ st1 {v0.h}2, x2, x5220
+.endr221
+ ret222
+endfunc223
+.endm224
+225
+addAvg_2xN 4226
+addAvg_2xN 8227
+addAvg_2xN 16228
+229
+.macro addAvg_4xN h230
+function PFX(addAvg_4x\h\()_neon)231
+ addAvg_start232
+.rept \h / 2233
+ ld1 {v0.8b}, x0, x3234
+ ld1 {v1.8b}, x1, x4235
+ ld1 {v2.8b}, x0, x3236
+ ld1 {v3.8b}, x1, x4237
+ add v0.4h, v0.4h, v1.4h238
+ add v2.4h, v2.4h, v3.4h239
+ saddl v0.4s, v0.4h, v30.4h240
+ saddl v2.4s, v2.4h, v30.4h241
+ shrn v0.4h, v0.4s, #7242
+ shrn2 v0.8h, v2.4s, #7243
+ sqxtun v0.8b, v0.8h244
+ st1 {v0.s}0, x2, x5245
+ st1 {v0.s}1, x2, x5246
+.endr247
+ ret248
+endfunc249
+.endm250
+251
+addAvg_4xN 2252
+addAvg_4xN 4253
+addAvg_4xN 8254
+addAvg_4xN 16255
+addAvg_4xN 32256
+257
+.macro addAvg_6xN h258
+function PFX(addAvg_6x\h\()_neon)259
+ addAvg_start260
+ mov w12, #\h / 2261
+ sub x5, x5, #4262
+.loop_addavg_6x\h:263
+ sub w12, w12, #1264
+ ld1 {v0.16b}, x0, x3265
+ ld1 {v1.16b}, x1, x4266
+ ld1 {v2.16b}, x0, x3267
+ ld1 {v3.16b}, x1, x4268
+ add v0.8h, v0.8h, v1.8h269
+ add v2.8h, v2.8h, v3.8h270
+ saddl v16.4s, v0.4h, v30.4h271
+ saddl2 v17.4s, v0.8h, v30.8h272
+ saddl v18.4s, v2.4h, v30.4h273
+ saddl2 v19.4s, v2.8h, v30.8h274
+ shrn v0.4h, v16.4s, #7275
+ shrn2 v0.8h, v17.4s, #7276
+ shrn v1.4h, v18.4s, #7277
+ shrn2 v1.8h, v19.4s, #7278
+ sqxtun v0.8b, v0.8h279
+ sqxtun v1.8b, v1.8h280
+ str s0, x2, #4281
+ st1 {v0.h}2, x2, x5282
+ str s1, x2, #4283
+ st1 {v1.h}2, x2, x5284
+ cbnz w12, .loop_addavg_6x\h285
+ ret286
+endfunc287
+.endm288
+289
+addAvg_6xN 8290
+addAvg_6xN 16291
+292
+.macro addAvg_8xN h293
+function PFX(addAvg_8x\h\()_neon)294
+ addAvg_start295
+.rept \h / 2296
+ ld1 {v0.16b}, x0, x3297
+ ld1 {v1.16b}, x1, x4298
+ ld1 {v2.16b}, x0, x3299
+ ld1 {v3.16b}, x1, x4300
+ add v0.8h, v0.8h, v1.8h301
+ add v2.8h, v2.8h, v3.8h302
+ saddl v16.4s, v0.4h, v30.4h303
+ saddl2 v17.4s, v0.8h, v30.8h304
+ saddl v18.4s, v2.4h, v30.4h305
+ saddl2 v19.4s, v2.8h, v30.8h306
+ shrn v0.4h, v16.4s, #7307
+ shrn2 v0.8h, v17.4s, #7308
+ shrn v1.4h, v18.4s, #7309
+ shrn2 v1.8h, v19.4s, #7310
+ sqxtun v0.8b, v0.8h311
+ sqxtun v1.8b, v1.8h312
+ st1 {v0.8b}, x2, x5313
+ st1 {v1.8b}, x2, x5314
+.endr315
+ ret316
+endfunc317
+.endm318
+319
+.macro addAvg_8xN1 h320
+function PFX(addAvg_8x\h\()_neon)321
+ addAvg_start322
+ mov w12, #\h / 2323
+.loop_addavg_8x\h:324
+ sub w12, w12, #1325
+ ld1 {v0.16b}, x0, x3326
+ ld1 {v1.16b}, x1, x4327
+ ld1 {v2.16b}, x0, x3328
+ ld1 {v3.16b}, x1, x4329
+ add v0.8h, v0.8h, v1.8h330
+ add v2.8h, v2.8h, v3.8h331
+ saddl v16.4s, v0.4h, v30.4h332
+ saddl2 v17.4s, v0.8h, v30.8h333
+ saddl v18.4s, v2.4h, v30.4h334
+ saddl2 v19.4s, v2.8h, v30.8h335
+ shrn v0.4h, v16.4s, #7336
+ shrn2 v0.8h, v17.4s, #7337
+ shrn v1.4h, v18.4s, #7338
+ shrn2 v1.8h, v19.4s, #7339
+ sqxtun v0.8b, v0.8h340
+ sqxtun v1.8b, v1.8h341
+ st1 {v0.8b}, x2, x5342
+ st1 {v1.8b}, x2, x5343
+ cbnz w12, .loop_addavg_8x\h344
+ ret345
+endfunc346
+.endm347
+348
+addAvg_8xN 2349
+addAvg_8xN 4350
+addAvg_8xN 6351
+addAvg_8xN 8352
+addAvg_8xN 12353
+addAvg_8xN 16354
+addAvg_8xN1 32355
+addAvg_8xN1 64356
+357
+.macro addAvg_12xN h358
+function PFX(addAvg_12x\h\()_neon)359
+ addAvg_start360
+ sub x3, x3, #16361
+ sub x4, x4, #16362
+ sub x5, x5, #8363
+ mov w12, #\h364
+.loop_addAvg_12X\h\():365
+ sub w12, w12, #1366
+ ld1 {v0.16b}, x0, #16367
+ ld1 {v1.16b}, x1, #16368
+ ld1 {v2.8b}, x0, x3369
+ ld1 {v3.8b}, x1, x4370
+ add v0.8h, v0.8h, v1.8h371
+ add v2.4h, v2.4h, v3.4h372
+ saddl v16.4s, v0.4h, v30.4h373
+ saddl2 v17.4s, v0.8h, v30.8h374
+ saddl v18.4s, v2.4h, v30.4h375
+ shrn v0.4h, v16.4s, #7376
+ shrn2 v0.8h, v17.4s, #7377
+ shrn v1.4h, v18.4s, #7378
+ sqxtun v0.8b, v0.8h379
+ sqxtun v1.8b, v1.8h380
+ st1 {v0.8b}, x2, #8381
+ st1 {v1.s}0, x2, x5382
+ cbnz w12, .loop_addAvg_12X\h383
+ ret384
+endfunc385
+.endm386
+387
+addAvg_12xN 16388
+addAvg_12xN 32389
+390
+.macro addAvg_16xN h391
+function PFX(addAvg_16x\h\()_neon)392
+ addAvg_start393
+ mov w12, #\h394
+.loop_addavg_16x\h:395
+ sub w12, w12, #1396
+ ld1 {v0.8h-v1.8h}, x0, x3397
+ ld1 {v2.8h-v3.8h}, x1, x4398
+ addavg_1 v0, v2399
+ addavg_1 v1, v3400
+ sqxtun v0.8b, v0.8h401
+ sqxtun2 v0.16b, v1.8h402
+ st1 {v0.16b}, x2, x5403
+ cbnz w12, .loop_addavg_16x\h404
+ ret405
+endfunc406
+.endm407
+408
+addAvg_16xN 4409
+addAvg_16xN 8410
+addAvg_16xN 12411
+addAvg_16xN 16412
+addAvg_16xN 24413
+addAvg_16xN 32414
+addAvg_16xN 64415
+416
+.macro addAvg_24xN h417
+function PFX(addAvg_24x\h\()_neon)418
+ addAvg_start419
+ mov w12, #\h420
+.loop_addavg_24x\h\():421
+ sub w12, w12, #1422
+ ld1 {v0.16b-v2.16b}, x0, x3423
+ ld1 {v3.16b-v5.16b}, x1, x4424
+ addavg_1 v0, v3425
+ addavg_1 v1, v4426
+ addavg_1 v2, v5427
+ sqxtun v0.8b, v0.8h428
+ sqxtun v1.8b, v1.8h429
+ sqxtun v2.8b, v2.8h430
+ st1 {v0.8b-v2.8b}, x2, x5431
+ cbnz w12, .loop_addavg_24x\h432
+ ret433
+endfunc434
+.endm435
+436
+addAvg_24xN 32437
+addAvg_24xN 64438
+439
+.macro addAvg_32xN h440
+function PFX(addAvg_32x\h\()_neon)441
+ addAvg_start442
+ mov w12, #\h443
+.loop_addavg_32x\h\():444
+ sub w12, w12, #1445
+ ld1 {v0.8h-v3.8h}, x0, x3446
+ ld1 {v4.8h-v7.8h}, x1, x4447
+ addavg_1 v0, v4448
+ addavg_1 v1, v5449
+ addavg_1 v2, v6450
+ addavg_1 v3, v7451
+ sqxtun v0.8b, v0.8h452
+ sqxtun v1.8b, v1.8h453
+ sqxtun v2.8b, v2.8h454
+ sqxtun v3.8b, v3.8h455
+ st1 {v0.8b-v3.8b}, x2, x5456
+ cbnz w12, .loop_addavg_32x\h457
+ ret458
+endfunc459
+.endm460
+461
+addAvg_32xN 8462
+addAvg_32xN 16463
+addAvg_32xN 24464
+addAvg_32xN 32465
+addAvg_32xN 48466
+addAvg_32xN 64467
+468
+function PFX(addAvg_48x64_neon)469
+ addAvg_start470
+ sub x3, x3, #64471
+ sub x4, x4, #64472
+ mov w12, #64473
+.loop_addavg_48x64:474
+ sub w12, w12, #1475
+ ld1 {v0.8h-v3.8h}, x0, #64476
+ ld1 {v4.8h-v7.8h}, x1, #64477
+ ld1 {v20.8h-v21.8h}, x0, x3478
+ ld1 {v22.8h-v23.8h}, x1, x4479
+ addavg_1 v0, v4480
+ addavg_1 v1, v5481
+ addavg_1 v2, v6482
+ addavg_1 v3, v7483
+ addavg_1 v20, v22484
+ addavg_1 v21, v23485
+ sqxtun v0.8b, v0.8h486
+ sqxtun2 v0.16b, v1.8h487
+ sqxtun v1.8b, v2.8h488
+ sqxtun2 v1.16b, v3.8h489
+ sqxtun v2.8b, v20.8h490
+ sqxtun2 v2.16b, v21.8h491
+ st1 {v0.16b-v2.16b}, x2, x5492
+ cbnz w12, .loop_addavg_48x64493
+ ret494
+endfunc495
+496
+.macro addAvg_64xN h497
+function PFX(addAvg_64x\h\()_neon)498
+ addAvg_start499
+ mov w12, #\h500
+ sub x3, x3, #64501
+ sub x4, x4, #64502
+.loop_addavg_64x\h\():503
+ sub w12, w12, #1504
+ ld1 {v0.8h-v3.8h}, x0, #64505
+ ld1 {v4.8h-v7.8h}, x1, #64506
+ ld1 {v20.8h-v23.8h}, x0, x3507
+ ld1 {v24.8h-v27.8h}, x1, x4508
+ addavg_1 v0, v4509
+ addavg_1 v1, v5510
+ addavg_1 v2, v6511
+ addavg_1 v3, v7512
+ addavg_1 v20, v24513
+ addavg_1 v21, v25514
+ addavg_1 v22, v26515
+ addavg_1 v23, v27516
+ sqxtun v0.8b, v0.8h517
+ sqxtun2 v0.16b, v1.8h518
+ sqxtun v1.8b, v2.8h519
+ sqxtun2 v1.16b, v3.8h520
+ sqxtun v2.8b, v20.8h521
+ sqxtun2 v2.16b, v21.8h522
+ sqxtun v3.8b, v22.8h523
+ sqxtun2 v3.16b, v23.8h524
+ st1 {v0.16b-v3.16b}, x2, x5525
+ cbnz w12, .loop_addavg_64x\h526
+ ret527
+endfunc528
+.endm529
+530
+addAvg_64xN 16531
+addAvg_64xN 32532
+addAvg_64xN 48533
+addAvg_64xN 64534
x265_3.6.tar.gz/source/common/aarch64/p2s-common.S
Added
104
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+.arch armv8-a29
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+#if HIGH_BIT_DEPTH39
+# if BIT_DEPTH == 1040
+# define P2S_SHIFT 441
+# elif BIT_DEPTH == 1242
+# define P2S_SHIFT 243
+# endif44
+.macro p2s_start45
+ add x3, x3, x346
+ add x1, x1, x147
+ movi v31.8h, #0xe0, lsl #848
+.endm49
+50
+#else // if !HIGH_BIT_DEPTH51
+# define P2S_SHIFT 652
+.macro p2s_start53
+ add x3, x3, x354
+ movi v31.8h, #0xe0, lsl #855
+.endm56
+#endif // HIGH_BIT_DEPTH57
+58
+.macro p2s_2x259
+#if HIGH_BIT_DEPTH60
+ ld1 {v0.s}0, x0, x161
+ ld1 {v0.s}1, x0, x162
+ shl v3.8h, v0.8h, #P2S_SHIFT63
+#else64
+ ldrh w10, x065
+ add x0, x0, x166
+ ldrh w11, x067
+ orr w10, w10, w11, lsl #1668
+ add x0, x0, x169
+ dup v0.4s, w1070
+ ushll v3.8h, v0.8b, #P2S_SHIFT71
+#endif72
+ add v3.8h, v3.8h, v31.8h73
+ st1 {v3.s}0, x2, x374
+ st1 {v3.s}1, x2, x375
+.endm76
+77
+.macro p2s_6x278
+#if HIGH_BIT_DEPTH79
+ ld1 {v0.d}0, x0, #880
+ ld1 {v1.s}0, x0, x181
+ ld1 {v0.d}1, x0, #882
+ ld1 {v1.s}1, x0, x183
+ shl v3.8h, v0.8h, #P2S_SHIFT84
+ shl v4.8h, v1.8h, #P2S_SHIFT85
+#else86
+ ldr s0, x087
+ ldrh w10, x0, #488
+ add x0, x0, x189
+ ld1 {v0.s}1, x090
+ ldrh w11, x0, #491
+ add x0, x0, x192
+ orr w10, w10, w11, lsl #1693
+ dup v1.4s, w1094
+ ushll v3.8h, v0.8b, #P2S_SHIFT95
+ ushll v4.8h, v1.8b, #P2S_SHIFT96
+#endif97
+ add v3.8h, v3.8h, v31.8h98
+ add v4.8h, v4.8h, v31.8h99
+ st1 {v3.d}0, x2, #8100
+ st1 {v4.s}0, x2, x3101
+ st1 {v3.d}1, x2, #8102
+ st1 {v4.s}1, x2, x3103
+.endm104
x265_3.6.tar.gz/source/common/aarch64/p2s-sve.S
Added
447
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "p2s-common.S"27
+28
+.arch armv8-a+sve29
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+#if HIGH_BIT_DEPTH41
+# if BIT_DEPTH == 1042
+# define P2S_SHIFT 443
+# elif BIT_DEPTH == 1244
+# define P2S_SHIFT 245
+# endif46
+47
+.macro p2s_start_sve48
+ add x3, x3, x349
+ add x1, x1, x150
+ mov z31.h, #0xe0, lsl #851
+.endm52
+53
+#else // if !HIGH_BIT_DEPTH54
+# define P2S_SHIFT 655
+.macro p2s_start_sve56
+ add x3, x3, x357
+ mov z31.h, #0xe0, lsl #858
+.endm59
+60
+#endif // HIGH_BIT_DEPTH61
+62
+// filterPixelToShort(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride)63
+.macro p2s_2xN_sve h64
+function PFX(filterPixelToShort_2x\h\()_sve)65
+ p2s_start_sve66
+.rept \h / 267
+ p2s_2x268
+.endr69
+ ret70
+endfunc71
+.endm72
+73
+p2s_2xN_sve 474
+p2s_2xN_sve 875
+p2s_2xN_sve 1676
+77
+.macro p2s_6xN_sve h78
+function PFX(filterPixelToShort_6x\h\()_sve)79
+ p2s_start_sve80
+ sub x3, x3, #881
+#if HIGH_BIT_DEPTH82
+ sub x1, x1, #883
+#endif84
+.rept \h / 285
+ p2s_6x286
+.endr87
+ ret88
+endfunc89
+.endm90
+91
+p2s_6xN_sve 892
+p2s_6xN_sve 1693
+94
+function PFX(filterPixelToShort_4x2_sve)95
+ p2s_start_sve96
+#if HIGH_BIT_DEPTH97
+ ptrue p0.h, vl898
+ index z1.d, #0, x199
+ index z2.d, #0, x3100
+ ld1d {z3.d}, p0/z, x0, z1.d101
+ lsl z3.h, p0/m, z3.h, #P2S_SHIFT102
+ add z3.h, p0/m, z3.h, z31.h103
+ st1d {z3.d}, p0, x2, z2.d104
+#else105
+ ptrue p0.h, vl4106
+ ld1b {z0.h}, p0/z, x0107
+ add x0, x0, x1108
+ ld1b {z1.h}, p0/z, x0109
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT110
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT111
+ add z0.h, p0/m, z0.h, z31.h112
+ add z1.h, p0/m, z1.h, z31.h113
+ st1h {z0.h}, p0, x2114
+ add x2, x2, x3115
+ st1h {z1.h}, p0, x2116
+#endif117
+ ret118
+endfunc119
+120
+121
+.macro p2s_8xN_sve h122
+function PFX(filterPixelToShort_8x\h\()_sve)123
+ p2s_start_sve124
+ ptrue p0.h, vl8125
+.rept \h126
+#if HIGH_BIT_DEPTH127
+ ld1d {z0.d}, p0/z, x0128
+ add x0, x0, x1129
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT130
+ add z0.h, p0/m, z0.h, z31.h131
+ st1h {z0.h}, p0, x2132
+ add x2, x2, x3133
+#else134
+ ld1b {z0.h}, p0/z, x0135
+ add x0, x0, x1136
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT137
+ add z0.h, p0/m, z0.h, z31.h138
+ st1h {z0.h}, p0, x2139
+ add x2, x2, x3140
+#endif141
+.endr142
+ ret143
+endfunc144
+.endm145
+146
+p2s_8xN_sve 2147
+148
+.macro p2s_32xN_sve h149
+function PFX(filterPixelToShort_32x\h\()_sve)150
+#if HIGH_BIT_DEPTH151
+ p2s_start_sve152
+ rdvl x9, #1153
+ cmp x9, #16154
+ bgt .vl_gt_16_filterPixelToShort_high_32x\h155
+ ptrue p0.h, vl8156
+.rept \h157
+ ld1h {z0.h}, p0/z, x0158
+ ld1h {z1.h}, p0/z, x0, #1, mul vl159
+ ld1h {z2.h}, p0/z, x0, #2, mul vl160
+ ld1h {z3.h}, p0/z, x0, #3, mul vl161
+ add x0, x0, x1162
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT163
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT164
+ lsl z2.h, p0/m, z2.h, #P2S_SHIFT165
+ lsl z3.h, p0/m, z3.h, #P2S_SHIFT166
+ add z0.h, p0/m, z0.h, z31.h167
+ add z1.h, p0/m, z1.h, z31.h168
+ add z2.h, p0/m, z2.h, z31.h169
+ add z3.h, p0/m, z3.h, z31.h170
+ st1h {z0.h}, p0, x2171
+ st1h {z1.h}, p0, x2, #1, mul vl172
+ st1h {z2.h}, p0, x2, #2, mul vl173
+ st1h {z3.h}, p0, x2, #3, mul vl174
+ add x2, x2, x3175
+.endr176
+ ret177
+.vl_gt_16_filterPixelToShort_high_32x\h\():178
+ cmp x9, #48179
+ bgt .vl_gt_48_filterPixelToShort_high_32x\h180
+ ptrue p0.h, vl16181
+.rept \h182
+ ld1h {z0.h}, p0/z, x0183
+ ld1h {z1.h}, p0/z, x0, #1, mul vl184
+ add x0, x0, x1185
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT186
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT187
+ add z0.h, p0/m, z0.h, z31.h188
+ add z1.h, p0/m, z1.h, z31.h189
+ st1h {z0.h}, p0, x2190
+ st1h {z1.h}, p0, x2, #1, mul vl191
+ add x2, x2, x3192
+.endr193
+ ret194
+.vl_gt_48_filterPixelToShort_high_32x\h\():195
+ ptrue p0.h, vl32196
+.rept \h197
+ ld1h {z0.h}, p0/z, x0198
+ add x0, x0, x1199
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT200
+ add z0.h, p0/m, z0.h, z31.h201
+ st1h {z0.h}, p0, x2202
+ add x2, x2, x3203
+.endr204
+ ret205
+#else206
+ p2s_start207
+ mov x9, #\h208
+.loop_filter_sve_P2S_32x\h:209
+ sub x9, x9, #1210
+ ld1 {v0.16b-v1.16b}, x0, x1211
+ ushll v22.8h, v0.8b, #P2S_SHIFT212
+ ushll2 v23.8h, v0.16b, #P2S_SHIFT213
+ ushll v24.8h, v1.8b, #P2S_SHIFT214
+ ushll2 v25.8h, v1.16b, #P2S_SHIFT215
+ add v22.8h, v22.8h, v31.8h216
+ add v23.8h, v23.8h, v31.8h217
+ add v24.8h, v24.8h, v31.8h218
+ add v25.8h, v25.8h, v31.8h219
+ st1 {v22.16b-v25.16b}, x2, x3220
+ cbnz x9, .loop_filter_sve_P2S_32x\h221
+ ret222
+#endif223
+endfunc224
+.endm225
+226
+p2s_32xN_sve 8227
+p2s_32xN_sve 16228
+p2s_32xN_sve 24229
+p2s_32xN_sve 32230
+p2s_32xN_sve 48231
+p2s_32xN_sve 64232
+233
+.macro p2s_64xN_sve h234
+function PFX(filterPixelToShort_64x\h\()_sve)235
+#if HIGH_BIT_DEPTH236
+ p2s_start_sve237
+ rdvl x9, #1238
+ cmp x9, #16239
+ bgt .vl_gt_16_filterPixelToShort_high_64x\h240
+ ptrue p0.h, vl8241
+.rept \h242
+ ld1h {z0.h}, p0/z, x0243
+ ld1h {z1.h}, p0/z, x0, #1, mul vl244
+ ld1h {z2.h}, p0/z, x0, #2, mul vl245
+ ld1h {z3.h}, p0/z, x0, #3, mul vl246
+ ld1h {z4.h}, p0/z, x0, #4, mul vl247
+ ld1h {z5.h}, p0/z, x0, #5, mul vl248
+ ld1h {z6.h}, p0/z, x0, #6, mul vl249
+ ld1h {z7.h}, p0/z, x0, #7, mul vl250
+ add x0, x0, x1251
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT252
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT253
+ lsl z2.h, p0/m, z2.h, #P2S_SHIFT254
+ lsl z3.h, p0/m, z3.h, #P2S_SHIFT255
+ lsl z4.h, p0/m, z4.h, #P2S_SHIFT256
+ lsl z5.h, p0/m, z5.h, #P2S_SHIFT257
+ lsl z6.h, p0/m, z6.h, #P2S_SHIFT258
+ lsl z7.h, p0/m, z7.h, #P2S_SHIFT259
+ add z0.h, p0/m, z0.h, z31.h260
+ add z1.h, p0/m, z1.h, z31.h261
+ add z2.h, p0/m, z2.h, z31.h262
+ add z3.h, p0/m, z3.h, z31.h263
+ add z4.h, p0/m, z4.h, z31.h264
+ add z5.h, p0/m, z5.h, z31.h265
+ add z6.h, p0/m, z6.h, z31.h266
+ add z7.h, p0/m, z7.h, z31.h267
+ st1h {z0.h}, p0, x2268
+ st1h {z1.h}, p0, x2, #1, mul vl269
+ st1h {z2.h}, p0, x2, #2, mul vl270
+ st1h {z3.h}, p0, x2, #3, mul vl271
+ st1h {z4.h}, p0, x2, #4, mul vl272
+ st1h {z5.h}, p0, x2, #5, mul vl273
+ st1h {z6.h}, p0, x2, #6, mul vl274
+ st1h {z7.h}, p0, x2, #7, mul vl275
+ add x2, x2, x3276
+.endr277
+ ret278
+.vl_gt_16_filterPixelToShort_high_64x\h\():279
+ cmp x9, #48280
+ bgt .vl_gt_48_filterPixelToShort_high_64x\h281
+ ptrue p0.h, vl16282
+.rept \h283
+ ld1h {z0.h}, p0/z, x0284
+ ld1h {z1.h}, p0/z, x0, #1, mul vl285
+ ld1h {z2.h}, p0/z, x0, #2, mul vl286
+ ld1h {z3.h}, p0/z, x0, #3, mul vl287
+ add x0, x0, x1288
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT289
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT290
+ lsl z2.h, p0/m, z2.h, #P2S_SHIFT291
+ lsl z3.h, p0/m, z3.h, #P2S_SHIFT292
+ add z0.h, p0/m, z0.h, z31.h293
+ add z1.h, p0/m, z1.h, z31.h294
+ add z2.h, p0/m, z2.h, z31.h295
+ add z3.h, p0/m, z3.h, z31.h296
+ st1h {z0.h}, p0, x2297
+ st1h {z1.h}, p0, x2, #1, mul vl298
+ st1h {z2.h}, p0, x2, #2, mul vl299
+ st1h {z3.h}, p0, x2, #3, mul vl300
+ add x2, x2, x3301
+.endr302
+ ret303
+.vl_gt_48_filterPixelToShort_high_64x\h\():304
+ cmp x9, #112305
+ bgt .vl_gt_112_filterPixelToShort_high_64x\h306
+ ptrue p0.h, vl32307
+.rept \h308
+ ld1h {z0.h}, p0/z, x0309
+ ld1h {z1.h}, p0/z, x0, #1, mul vl310
+ add x0, x0, x1311
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT312
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT313
+ add z0.h, p0/m, z0.h, z31.h314
+ add z1.h, p0/m, z1.h, z31.h315
+ st1h {z0.h}, p0, x2316
+ st1h {z1.h}, p0, x2, #1, mul vl317
+ add x2, x2, x3318
+.endr319
+ ret320
+.vl_gt_112_filterPixelToShort_high_64x\h\():321
+ ptrue p0.h, vl64322
+.rept \h323
+ ld1h {z0.h}, p0/z, x0324
+ add x0, x0, x1325
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT326
+ add z0.h, p0/m, z0.h, z31.h327
+ st1h {z0.h}, p0, x2328
+ add x2, x2, x3329
+.endr330
+ ret331
+#else332
+ p2s_start333
+ sub x3, x3, #64334
+ mov x9, #\h335
+.loop_filter_sve_P2S_64x\h:336
+ sub x9, x9, #1337
+ ld1 {v0.16b-v3.16b}, x0, x1338
+ ushll v16.8h, v0.8b, #P2S_SHIFT339
+ ushll2 v17.8h, v0.16b, #P2S_SHIFT340
+ ushll v18.8h, v1.8b, #P2S_SHIFT341
+ ushll2 v19.8h, v1.16b, #P2S_SHIFT342
+ ushll v20.8h, v2.8b, #P2S_SHIFT343
+ ushll2 v21.8h, v2.16b, #P2S_SHIFT344
+ ushll v22.8h, v3.8b, #P2S_SHIFT345
+ ushll2 v23.8h, v3.16b, #P2S_SHIFT346
+ add v16.8h, v16.8h, v31.8h347
+ add v17.8h, v17.8h, v31.8h348
+ add v18.8h, v18.8h, v31.8h349
+ add v19.8h, v19.8h, v31.8h350
+ add v20.8h, v20.8h, v31.8h351
+ add v21.8h, v21.8h, v31.8h352
+ add v22.8h, v22.8h, v31.8h353
+ add v23.8h, v23.8h, v31.8h354
+ st1 {v16.16b-v19.16b}, x2, #64355
+ st1 {v20.16b-v23.16b}, x2, x3356
+ cbnz x9, .loop_filter_sve_P2S_64x\h357
+ ret358
+#endif359
+endfunc360
+.endm361
+362
+p2s_64xN_sve 16363
+p2s_64xN_sve 32364
+p2s_64xN_sve 48365
+p2s_64xN_sve 64366
+367
+function PFX(filterPixelToShort_48x64_sve)368
+#if HIGH_BIT_DEPTH369
+ p2s_start_sve370
+ rdvl x9, #1371
+ cmp x9, #16372
+ bgt .vl_gt_16_filterPixelToShort_high_48x64373
+ ptrue p0.h, vl8374
+.rept 64375
+ ld1h {z0.h}, p0/z, x0376
+ ld1h {z1.h}, p0/z, x0, #1, mul vl377
+ ld1h {z2.h}, p0/z, x0, #2, mul vl378
+ ld1h {z3.h}, p0/z, x0, #3, mul vl379
+ ld1h {z4.h}, p0/z, x0, #4, mul vl380
+ ld1h {z5.h}, p0/z, x0, #5, mul vl381
+ add x0, x0, x1382
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT383
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT384
+ lsl z2.h, p0/m, z2.h, #P2S_SHIFT385
+ lsl z3.h, p0/m, z3.h, #P2S_SHIFT386
+ lsl z4.h, p0/m, z4.h, #P2S_SHIFT387
+ lsl z5.h, p0/m, z5.h, #P2S_SHIFT388
+ add z0.h, p0/m, z0.h, z31.h389
+ add z1.h, p0/m, z1.h, z31.h390
+ add z2.h, p0/m, z2.h, z31.h391
+ add z3.h, p0/m, z3.h, z31.h392
+ add z4.h, p0/m, z4.h, z31.h393
+ add z5.h, p0/m, z5.h, z31.h394
+ st1h {z0.h}, p0, x2395
+ st1h {z1.h}, p0, x2, #1, mul vl396
+ st1h {z2.h}, p0, x2, #2, mul vl397
+ st1h {z3.h}, p0, x2, #3, mul vl398
+ st1h {z4.h}, p0, x2, #4, mul vl399
+ st1h {z5.h}, p0, x2, #5, mul vl400
+ add x2, x2, x3401
+.endr402
+ ret403
+.vl_gt_16_filterPixelToShort_high_48x64:404
+ ptrue p0.h, vl16405
+.rept 64406
+ ld1h {z0.h}, p0/z, x0407
+ ld1h {z1.h}, p0/z, x0, #1, mul vl408
+ ld1h {z2.h}, p0/z, x0, #2, mul vl409
+ add x0, x0, x1410
+ lsl z0.h, p0/m, z0.h, #P2S_SHIFT411
+ lsl z1.h, p0/m, z1.h, #P2S_SHIFT412
+ lsl z2.h, p0/m, z2.h, #P2S_SHIFT413
+ add z0.h, p0/m, z0.h, z31.h414
+ add z1.h, p0/m, z1.h, z31.h415
+ add z2.h, p0/m, z2.h, z31.h416
+ st1h {z0.h}, p0, x2417
+ st1h {z1.h}, p0, x2, #1, mul vl418
+ st1h {z2.h}, p0, x2, #2, mul vl419
+ add x2, x2, x3420
+.endr421
+ ret422
+#else423
+ p2s_start424
+ sub x3, x3, #64425
+ mov x9, #64426
+.loop_filterP2S_sve_48x64:427
+ sub x9, x9, #1428
+ ld1 {v0.16b-v2.16b}, x0, x1429
+ ushll v16.8h, v0.8b, #P2S_SHIFT430
+ ushll2 v17.8h, v0.16b, #P2S_SHIFT431
+ ushll v18.8h, v1.8b, #P2S_SHIFT432
+ ushll2 v19.8h, v1.16b, #P2S_SHIFT433
+ ushll v20.8h, v2.8b, #P2S_SHIFT434
+ ushll2 v21.8h, v2.16b, #P2S_SHIFT435
+ add v16.8h, v16.8h, v31.8h436
+ add v17.8h, v17.8h, v31.8h437
+ add v18.8h, v18.8h, v31.8h438
+ add v19.8h, v19.8h, v31.8h439
+ add v20.8h, v20.8h, v31.8h440
+ add v21.8h, v21.8h, v31.8h441
+ st1 {v16.16b-v19.16b}, x2, #64442
+ st1 {v20.16b-v21.16b}, x2, x3443
+ cbnz x9, .loop_filterP2S_sve_48x64444
+ ret445
+#endif446
+endfunc447
x265_3.6.tar.gz/source/common/aarch64/p2s.S
Added
388
1
2
+/*****************************************************************************3
+ * Copyright (C) 2021 MulticoreWare, Inc4
+ *5
+ * Authors: Sebastian Pop <spop@amazon.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm.S"26
+#include "p2s-common.S"27
+28
+#ifdef __APPLE__29
+.section __RODATA,__rodata30
+#else31
+.section .rodata32
+#endif33
+34
+.align 435
+36
+.text37
+38
+// filterPixelToShort(const pixel* src, intptr_t srcStride, int16_t* dst, intptr_t dstStride)39
+.macro p2s_2xN h40
+function PFX(filterPixelToShort_2x\h\()_neon)41
+ p2s_start42
+.rept \h / 243
+ p2s_2x244
+.endr45
+ ret46
+endfunc47
+.endm48
+49
+p2s_2xN 450
+p2s_2xN 851
+p2s_2xN 1652
+53
+.macro p2s_6xN h54
+function PFX(filterPixelToShort_6x\h\()_neon)55
+ p2s_start56
+ sub x3, x3, #857
+#if HIGH_BIT_DEPTH58
+ sub x1, x1, #859
+#endif60
+.rept \h / 261
+ p2s_6x262
+.endr63
+ ret64
+endfunc65
+.endm66
+67
+p2s_6xN 868
+p2s_6xN 1669
+70
+function PFX(filterPixelToShort_4x2_neon)71
+ p2s_start72
+#if HIGH_BIT_DEPTH73
+ ld1 {v0.d}0, x0, x174
+ ld1 {v0.d}1, x0, x175
+ shl v3.8h, v0.8h, #P2S_SHIFT76
+#else77
+ ld1 {v0.s}0, x0, x178
+ ld1 {v0.s}1, x0, x179
+ ushll v3.8h, v0.8b, #P2S_SHIFT80
+#endif81
+ add v3.8h, v3.8h, v31.8h82
+ st1 {v3.d}0, x2, x383
+ st1 {v3.d}1, x2, x384
+ ret85
+endfunc86
+87
+function PFX(filterPixelToShort_4x4_neon)88
+ p2s_start89
+#if HIGH_BIT_DEPTH90
+ ld1 {v0.d}0, x0, x191
+ ld1 {v0.d}1, x0, x192
+ shl v3.8h, v0.8h, #P2S_SHIFT93
+#else94
+ ld1 {v0.s}0, x0, x195
+ ld1 {v0.s}1, x0, x196
+ ushll v3.8h, v0.8b, #P2S_SHIFT97
+#endif98
+ add v3.8h, v3.8h, v31.8h99
+ st1 {v3.d}0, x2, x3100
+ st1 {v3.d}1, x2, x3101
+#if HIGH_BIT_DEPTH102
+ ld1 {v1.d}0, x0, x1103
+ ld1 {v1.d}1, x0, x1104
+ shl v4.8h, v1.8h, #P2S_SHIFT105
+#else106
+ ld1 {v1.s}0, x0, x1107
+ ld1 {v1.s}1, x0, x1108
+ ushll v4.8h, v1.8b, #P2S_SHIFT109
+#endif110
+ add v4.8h, v4.8h, v31.8h111
+ st1 {v4.d}0, x2, x3112
+ st1 {v4.d}1, x2, x3113
+ ret114
+endfunc115
+116
+.macro p2s_4xN h117
+function PFX(filterPixelToShort_4x\h\()_neon)118
+ p2s_start119
+.rept \h / 2120
+#if HIGH_BIT_DEPTH121
+ ld1 {v0.16b}, x0, x1122
+ shl v0.8h, v0.8h, #P2S_SHIFT123
+#else124
+ ld1 {v0.8b}, x0, x1125
+ ushll v0.8h, v0.8b, #P2S_SHIFT126
+#endif127
+ add v2.4h, v0.4h, v31.4h128
+ st1 {v2.4h}, x2, x3129
+#if HIGH_BIT_DEPTH130
+ ld1 {v1.16b}, x0, x1131
+ shl v1.8h, v1.8h, #P2S_SHIFT132
+#else133
+ ld1 {v1.8b}, x0, x1134
+ ushll v1.8h, v1.8b, #P2S_SHIFT135
+#endif136
+ add v3.4h, v1.4h, v31.4h137
+ st1 {v3.4h}, x2, x3138
+.endr139
+ ret140
+endfunc141
+.endm142
+143
+p2s_4xN 8144
+p2s_4xN 16145
+p2s_4xN 32146
+147
+.macro p2s_8xN h148
+function PFX(filterPixelToShort_8x\h\()_neon)149
+ p2s_start150
+.rept \h / 2151
+#if HIGH_BIT_DEPTH152
+ ld1 {v0.16b}, x0, x1153
+ ld1 {v1.16b}, x0, x1154
+ shl v0.8h, v0.8h, #P2S_SHIFT155
+ shl v1.8h, v1.8h, #P2S_SHIFT156
+#else157
+ ld1 {v0.8b}, x0, x1158
+ ld1 {v1.8b}, x0, x1159
+ ushll v0.8h, v0.8b, #P2S_SHIFT160
+ ushll v1.8h, v1.8b, #P2S_SHIFT161
+#endif162
+ add v2.8h, v0.8h, v31.8h163
+ st1 {v2.8h}, x2, x3164
+ add v3.8h, v1.8h, v31.8h165
+ st1 {v3.8h}, x2, x3166
+.endr167
+ ret168
+endfunc169
+.endm170
+171
+p2s_8xN 2172
+p2s_8xN 4173
+p2s_8xN 6174
+p2s_8xN 8175
+p2s_8xN 12176
+p2s_8xN 16177
+p2s_8xN 32178
+p2s_8xN 64179
+180
+.macro p2s_12xN h181
+function PFX(filterPixelToShort_12x\h\()_neon)182
+ p2s_start183
+ sub x3, x3, #16184
+.rept \h185
+#if HIGH_BIT_DEPTH186
+ ld1 {v0.16b-v1.16b}, x0, x1187
+ shl v2.8h, v0.8h, #P2S_SHIFT188
+ shl v3.8h, v1.8h, #P2S_SHIFT189
+#else190
+ ld1 {v0.16b}, x0, x1191
+ ushll v2.8h, v0.8b, #P2S_SHIFT192
+ ushll2 v3.8h, v0.16b, #P2S_SHIFT193
+#endif194
+ add v2.8h, v2.8h, v31.8h195
+ add v3.8h, v3.8h, v31.8h196
+ st1 {v2.16b}, x2, #16197
+ st1 {v3.8b}, x2, x3198
+.endr199
+ ret200
+endfunc201
+.endm202
+203
+p2s_12xN 16204
+p2s_12xN 32205
+206
+.macro p2s_16xN h207
+function PFX(filterPixelToShort_16x\h\()_neon)208
+ p2s_start209
+.rept \h210
+#if HIGH_BIT_DEPTH211
+ ld1 {v0.16b-v1.16b}, x0, x1212
+ shl v2.8h, v0.8h, #P2S_SHIFT213
+ shl v3.8h, v1.8h, #P2S_SHIFT214
+#else215
+ ld1 {v0.16b}, x0, x1216
+ ushll v2.8h, v0.8b, #P2S_SHIFT217
+ ushll2 v3.8h, v0.16b, #P2S_SHIFT218
+#endif219
+ add v2.8h, v2.8h, v31.8h220
+ add v3.8h, v3.8h, v31.8h221
+ st1 {v2.16b-v3.16b}, x2, x3222
+.endr223
+ ret224
+endfunc225
+.endm226
+227
+p2s_16xN 4228
+p2s_16xN 8229
+p2s_16xN 12230
+p2s_16xN 16231
+p2s_16xN 24232
+p2s_16xN 32233
+p2s_16xN 64234
+235
+.macro p2s_24xN h236
+function PFX(filterPixelToShort_24x\h\()_neon)237
+ p2s_start238
+.rept \h239
+#if HIGH_BIT_DEPTH240
+ ld1 {v0.16b-v2.16b}, x0, x1241
+ shl v3.8h, v0.8h, #P2S_SHIFT242
+ shl v4.8h, v1.8h, #P2S_SHIFT243
+ shl v5.8h, v2.8h, #P2S_SHIFT244
+#else245
+ ld1 {v0.8b-v2.8b}, x0, x1246
+ ushll v3.8h, v0.8b, #P2S_SHIFT247
+ ushll v4.8h, v1.8b, #P2S_SHIFT248
+ ushll v5.8h, v2.8b, #P2S_SHIFT249
+#endif250
+ add v3.8h, v3.8h, v31.8h251
+ add v4.8h, v4.8h, v31.8h252
+ add v5.8h, v5.8h, v31.8h253
+ st1 {v3.16b-v5.16b}, x2, x3254
+.endr255
+ ret256
+endfunc257
+.endm258
+259
+p2s_24xN 32260
+p2s_24xN 64261
+262
+.macro p2s_32xN h263
+function PFX(filterPixelToShort_32x\h\()_neon)264
+ p2s_start265
+ mov x9, #\h266
+.loop_filterP2S_32x\h:267
+ sub x9, x9, #1268
+#if HIGH_BIT_DEPTH269
+ ld1 {v0.16b-v3.16b}, x0, x1270
+ shl v22.8h, v0.8h, #P2S_SHIFT271
+ shl v23.8h, v1.8h, #P2S_SHIFT272
+ shl v24.8h, v2.8h, #P2S_SHIFT273
+ shl v25.8h, v3.8h, #P2S_SHIFT274
+#else275
+ ld1 {v0.16b-v1.16b}, x0, x1276
+ ushll v22.8h, v0.8b, #P2S_SHIFT277
+ ushll2 v23.8h, v0.16b, #P2S_SHIFT278
+ ushll v24.8h, v1.8b, #P2S_SHIFT279
+ ushll2 v25.8h, v1.16b, #P2S_SHIFT280
+#endif281
+ add v22.8h, v22.8h, v31.8h282
+ add v23.8h, v23.8h, v31.8h283
+ add v24.8h, v24.8h, v31.8h284
+ add v25.8h, v25.8h, v31.8h285
+ st1 {v22.16b-v25.16b}, x2, x3286
+ cbnz x9, .loop_filterP2S_32x\h287
+ ret288
+endfunc289
+.endm290
+291
+p2s_32xN 8292
+p2s_32xN 16293
+p2s_32xN 24294
+p2s_32xN 32295
+p2s_32xN 48296
+p2s_32xN 64297
+298
+.macro p2s_64xN h299
+function PFX(filterPixelToShort_64x\h\()_neon)300
+ p2s_start301
+#if HIGH_BIT_DEPTH302
+ sub x1, x1, #64303
+#endif304
+ sub x3, x3, #64305
+ mov x9, #\h306
+.loop_filterP2S_64x\h:307
+ sub x9, x9, #1308
+#if HIGH_BIT_DEPTH309
+ ld1 {v0.16b-v3.16b}, x0, #64310
+ ld1 {v4.16b-v7.16b}, x0, x1311
+ shl v16.8h, v0.8h, #P2S_SHIFT312
+ shl v17.8h, v1.8h, #P2S_SHIFT313
+ shl v18.8h, v2.8h, #P2S_SHIFT314
+ shl v19.8h, v3.8h, #P2S_SHIFT315
+ shl v20.8h, v4.8h, #P2S_SHIFT316
+ shl v21.8h, v5.8h, #P2S_SHIFT317
+ shl v22.8h, v6.8h, #P2S_SHIFT318
+ shl v23.8h, v7.8h, #P2S_SHIFT319
+#else320
+ ld1 {v0.16b-v3.16b}, x0, x1321
+ ushll v16.8h, v0.8b, #P2S_SHIFT322
+ ushll2 v17.8h, v0.16b, #P2S_SHIFT323
+ ushll v18.8h, v1.8b, #P2S_SHIFT324
+ ushll2 v19.8h, v1.16b, #P2S_SHIFT325
+ ushll v20.8h, v2.8b, #P2S_SHIFT326
+ ushll2 v21.8h, v2.16b, #P2S_SHIFT327
+ ushll v22.8h, v3.8b, #P2S_SHIFT328
+ ushll2 v23.8h, v3.16b, #P2S_SHIFT329
+#endif330
+ add v16.8h, v16.8h, v31.8h331
+ add v17.8h, v17.8h, v31.8h332
+ add v18.8h, v18.8h, v31.8h333
+ add v19.8h, v19.8h, v31.8h334
+ add v20.8h, v20.8h, v31.8h335
+ add v21.8h, v21.8h, v31.8h336
+ add v22.8h, v22.8h, v31.8h337
+ add v23.8h, v23.8h, v31.8h338
+ st1 {v16.16b-v19.16b}, x2, #64339
+ st1 {v20.16b-v23.16b}, x2, x3340
+ cbnz x9, .loop_filterP2S_64x\h341
+ ret342
+endfunc343
+.endm344
+345
+p2s_64xN 16346
+p2s_64xN 32347
+p2s_64xN 48348
+p2s_64xN 64349
+350
+function PFX(filterPixelToShort_48x64_neon)351
+ p2s_start352
+#if HIGH_BIT_DEPTH353
+ sub x1, x1, #64354
+#endif355
+ sub x3, x3, #64356
+ mov x9, #64357
+.loop_filterP2S_48x64:358
+ sub x9, x9, #1359
+#if HIGH_BIT_DEPTH360
+ ld1 {v0.16b-v3.16b}, x0, #64361
+ ld1 {v4.16b-v5.16b}, x0, x1362
+ shl v16.8h, v0.8h, #P2S_SHIFT363
+ shl v17.8h, v1.8h, #P2S_SHIFT364
+ shl v18.8h, v2.8h, #P2S_SHIFT365
+ shl v19.8h, v3.8h, #P2S_SHIFT366
+ shl v20.8h, v4.8h, #P2S_SHIFT367
+ shl v21.8h, v5.8h, #P2S_SHIFT368
+#else369
+ ld1 {v0.16b-v2.16b}, x0, x1370
+ ushll v16.8h, v0.8b, #P2S_SHIFT371
+ ushll2 v17.8h, v0.16b, #P2S_SHIFT372
+ ushll v18.8h, v1.8b, #P2S_SHIFT373
+ ushll2 v19.8h, v1.16b, #P2S_SHIFT374
+ ushll v20.8h, v2.8b, #P2S_SHIFT375
+ ushll2 v21.8h, v2.16b, #P2S_SHIFT376
+#endif377
+ add v16.8h, v16.8h, v31.8h378
+ add v17.8h, v17.8h, v31.8h379
+ add v18.8h, v18.8h, v31.8h380
+ add v19.8h, v19.8h, v31.8h381
+ add v20.8h, v20.8h, v31.8h382
+ add v21.8h, v21.8h, v31.8h383
+ st1 {v16.16b-v19.16b}, x2, #64384
+ st1 {v20.16b-v21.16b}, x2, x3385
+ cbnz x9, .loop_filterP2S_48x64386
+ ret387
+endfunc388
x265_3.6.tar.gz/source/common/aarch64/pixel-prim.cpp
Added
2061
1
2
+#include "common.h"3
+#include "slicetype.h" // LOWRES_COST_MASK4
+#include "primitives.h"5
+#include "x265.h"6
+7
+#include "pixel-prim.h"8
+#include "arm64-utils.h"9
+#if HAVE_NEON10
+11
+#include <arm_neon.h>12
+13
+using namespace X265_NS;14
+15
+16
+17
+namespace18
+{19
+20
+21
+/* SATD SA8D variants - based on x264 */22
+static inline void SUMSUB_AB(int16x8_t &sum, int16x8_t &sub, const int16x8_t a, const int16x8_t b)23
+{24
+ sum = vaddq_s16(a, b);25
+ sub = vsubq_s16(a, b);26
+}27
+28
+static inline void transpose_8h(int16x8_t &t1, int16x8_t &t2, const int16x8_t s1, const int16x8_t s2)29
+{30
+ t1 = vtrn1q_s16(s1, s2);31
+ t2 = vtrn2q_s16(s1, s2);32
+}33
+34
+static inline void transpose_4s(int16x8_t &t1, int16x8_t &t2, const int16x8_t s1, const int16x8_t s2)35
+{36
+ t1 = vtrn1q_s32(s1, s2);37
+ t2 = vtrn2q_s32(s1, s2);38
+}39
+40
+#if (X265_DEPTH <= 10)41
+static inline void transpose_2d(int16x8_t &t1, int16x8_t &t2, const int16x8_t s1, const int16x8_t s2)42
+{43
+ t1 = vtrn1q_s64(s1, s2);44
+ t2 = vtrn2q_s64(s1, s2);45
+}46
+#endif47
+48
+49
+static inline void SUMSUB_ABCD(int16x8_t &s1, int16x8_t &d1, int16x8_t &s2, int16x8_t &d2,50
+ int16x8_t a, int16x8_t b, int16x8_t c, int16x8_t d)51
+{52
+ SUMSUB_AB(s1, d1, a, b);53
+ SUMSUB_AB(s2, d2, c, d);54
+}55
+56
+static inline void HADAMARD4_V(int16x8_t &r1, int16x8_t &r2, int16x8_t &r3, int16x8_t &r4,57
+ int16x8_t &t1, int16x8_t &t2, int16x8_t &t3, int16x8_t &t4)58
+{59
+ SUMSUB_ABCD(t1, t2, t3, t4, r1, r2, r3, r4);60
+ SUMSUB_ABCD(r1, r3, r2, r4, t1, t3, t2, t4);61
+}62
+63
+64
+static int _satd_4x8_8x4_end_neon(int16x8_t v0, int16x8_t v1, int16x8_t v2, int16x8_t v3)65
+66
+{67
+68
+ int16x8_t v4, v5, v6, v7, v16, v17, v18, v19;69
+70
+71
+ SUMSUB_AB(v16, v17, v0, v1);72
+ SUMSUB_AB(v18, v19, v2, v3);73
+74
+ SUMSUB_AB(v4 , v6 , v16, v18);75
+ SUMSUB_AB(v5 , v7 , v17, v19);76
+77
+ v0 = vtrn1q_s16(v4, v5);78
+ v1 = vtrn2q_s16(v4, v5);79
+ v2 = vtrn1q_s16(v6, v7);80
+ v3 = vtrn2q_s16(v6, v7);81
+82
+ SUMSUB_AB(v16, v17, v0, v1);83
+ SUMSUB_AB(v18, v19, v2, v3);84
+85
+ v0 = vtrn1q_s32(v16, v18);86
+ v1 = vtrn2q_s32(v16, v18);87
+ v2 = vtrn1q_s32(v17, v19);88
+ v3 = vtrn2q_s32(v17, v19);89
+90
+ v0 = vabsq_s16(v0);91
+ v1 = vabsq_s16(v1);92
+ v2 = vabsq_s16(v2);93
+ v3 = vabsq_s16(v3);94
+95
+ v0 = vmaxq_u16(v0, v1);96
+ v1 = vmaxq_u16(v2, v3);97
+98
+ v0 = vaddq_u16(v0, v1);99
+ return vaddlvq_u16(v0);100
+}101
+102
+static inline int _satd_4x4_neon(int16x8_t v0, int16x8_t v1)103
+{104
+ int16x8_t v2, v3;105
+ SUMSUB_AB(v2, v3, v0, v1);106
+107
+ v0 = vzip1q_s64(v2, v3);108
+ v1 = vzip2q_s64(v2, v3);109
+ SUMSUB_AB(v2, v3, v0, v1);110
+111
+ v0 = vtrn1q_s16(v2, v3);112
+ v1 = vtrn2q_s16(v2, v3);113
+ SUMSUB_AB(v2, v3, v0, v1);114
+115
+ v0 = vtrn1q_s32(v2, v3);116
+ v1 = vtrn2q_s32(v2, v3);117
+118
+ v0 = vabsq_s16(v0);119
+ v1 = vabsq_s16(v1);120
+ v0 = vmaxq_u16(v0, v1);121
+122
+ return vaddlvq_s16(v0);123
+}124
+125
+static void _satd_8x4v_8x8h_neon(int16x8_t &v0, int16x8_t &v1, int16x8_t &v2, int16x8_t &v3, int16x8_t &v20,126
+ int16x8_t &v21, int16x8_t &v22, int16x8_t &v23)127
+{128
+ int16x8_t v16, v17, v18, v19, v4, v5, v6, v7;129
+130
+ SUMSUB_AB(v16, v18, v0, v2);131
+ SUMSUB_AB(v17, v19, v1, v3);132
+133
+ HADAMARD4_V(v20, v21, v22, v23, v0, v1, v2, v3);134
+135
+ transpose_8h(v0, v1, v16, v17);136
+ transpose_8h(v2, v3, v18, v19);137
+ transpose_8h(v4, v5, v20, v21);138
+ transpose_8h(v6, v7, v22, v23);139
+140
+ SUMSUB_AB(v16, v17, v0, v1);141
+ SUMSUB_AB(v18, v19, v2, v3);142
+ SUMSUB_AB(v20, v21, v4, v5);143
+ SUMSUB_AB(v22, v23, v6, v7);144
+145
+ transpose_4s(v0, v2, v16, v18);146
+ transpose_4s(v1, v3, v17, v19);147
+ transpose_4s(v4, v6, v20, v22);148
+ transpose_4s(v5, v7, v21, v23);149
+150
+ v0 = vabsq_s16(v0);151
+ v1 = vabsq_s16(v1);152
+ v2 = vabsq_s16(v2);153
+ v3 = vabsq_s16(v3);154
+ v4 = vabsq_s16(v4);155
+ v5 = vabsq_s16(v5);156
+ v6 = vabsq_s16(v6);157
+ v7 = vabsq_s16(v7);158
+159
+ v0 = vmaxq_u16(v0, v2);160
+ v1 = vmaxq_u16(v1, v3);161
+ v2 = vmaxq_u16(v4, v6);162
+ v3 = vmaxq_u16(v5, v7);163
+164
+}165
+166
+#if HIGH_BIT_DEPTH167
+168
+#if (X265_DEPTH > 10)169
+static inline void transpose_2d(int32x4_t &t1, int32x4_t &t2, const int32x4_t s1, const int32x4_t s2)170
+{171
+ t1 = vtrn1q_s64(s1, s2);172
+ t2 = vtrn2q_s64(s1, s2);173
+}174
+175
+static inline void ISUMSUB_AB(int32x4_t &sum, int32x4_t &sub, const int32x4_t a, const int32x4_t b)176
+{177
+ sum = vaddq_s32(a, b);178
+ sub = vsubq_s32(a, b);179
+}180
+181
+static inline void ISUMSUB_AB_FROM_INT16(int32x4_t &suml, int32x4_t &sumh, int32x4_t &subl, int32x4_t &subh,182
+ const int16x8_t a, const int16x8_t b)183
+{184
+ suml = vaddl_s16(vget_low_s16(a), vget_low_s16(b));185
+ sumh = vaddl_high_s16(a, b);186
+ subl = vsubl_s16(vget_low_s16(a), vget_low_s16(b));187
+ subh = vsubl_high_s16(a, b);188
+}189
+190
+#endif191
+192
+static inline void _sub_8x8_fly(const uint16_t *pix1, intptr_t stride_pix1, const uint16_t *pix2, intptr_t stride_pix2,193
+ int16x8_t &v0, int16x8_t &v1, int16x8_t &v2, int16x8_t &v3,194
+ int16x8_t &v20, int16x8_t &v21, int16x8_t &v22, int16x8_t &v23)195
+{196
+ uint16x8_t r0, r1, r2, r3;197
+ uint16x8_t t0, t1, t2, t3;198
+ int16x8_t v16, v17;199
+ int16x8_t v18, v19;200
+201
+ r0 = *(uint16x8_t *)(pix1 + 0 * stride_pix1);202
+ r1 = *(uint16x8_t *)(pix1 + 1 * stride_pix1);203
+ r2 = *(uint16x8_t *)(pix1 + 2 * stride_pix1);204
+ r3 = *(uint16x8_t *)(pix1 + 3 * stride_pix1);205
+206
+ t0 = *(uint16x8_t *)(pix2 + 0 * stride_pix2);207
+ t1 = *(uint16x8_t *)(pix2 + 1 * stride_pix2);208
+ t2 = *(uint16x8_t *)(pix2 + 2 * stride_pix2);209
+ t3 = *(uint16x8_t *)(pix2 + 3 * stride_pix2);210
+211
+ v16 = vsubq_u16(r0, t0);212
+ v17 = vsubq_u16(r1, t1);213
+ v18 = vsubq_u16(r2, t2);214
+ v19 = vsubq_u16(r3, t3);215
+216
+ r0 = *(uint16x8_t *)(pix1 + 4 * stride_pix1);217
+ r1 = *(uint16x8_t *)(pix1 + 5 * stride_pix1);218
+ r2 = *(uint16x8_t *)(pix1 + 6 * stride_pix1);219
+ r3 = *(uint16x8_t *)(pix1 + 7 * stride_pix1);220
+221
+ t0 = *(uint16x8_t *)(pix2 + 4 * stride_pix2);222
+ t1 = *(uint16x8_t *)(pix2 + 5 * stride_pix2);223
+ t2 = *(uint16x8_t *)(pix2 + 6 * stride_pix2);224
+ t3 = *(uint16x8_t *)(pix2 + 7 * stride_pix2);225
+226
+ v20 = vsubq_u16(r0, t0);227
+ v21 = vsubq_u16(r1, t1);228
+ v22 = vsubq_u16(r2, t2);229
+ v23 = vsubq_u16(r3, t3);230
+231
+ SUMSUB_AB(v0, v1, v16, v17);232
+ SUMSUB_AB(v2, v3, v18, v19);233
+234
+}235
+236
+237
+238
+239
+static void _satd_16x4_neon(const uint16_t *pix1, intptr_t stride_pix1, const uint16_t *pix2, intptr_t stride_pix2,240
+ int16x8_t &v0, int16x8_t &v1, int16x8_t &v2, int16x8_t &v3)241
+{242
+ uint8x16_t r0, r1, r2, r3;243
+ uint8x16_t t0, t1, t2, t3;244
+ int16x8_t v16, v17, v20, v21;245
+ int16x8_t v18, v19, v22, v23;246
+247
+ r0 = *(int16x8_t *)(pix1 + 0 * stride_pix1);248
+ r1 = *(int16x8_t *)(pix1 + 1 * stride_pix1);249
+ r2 = *(int16x8_t *)(pix1 + 2 * stride_pix1);250
+ r3 = *(int16x8_t *)(pix1 + 3 * stride_pix1);251
+252
+ t0 = *(int16x8_t *)(pix2 + 0 * stride_pix2);253
+ t1 = *(int16x8_t *)(pix2 + 1 * stride_pix2);254
+ t2 = *(int16x8_t *)(pix2 + 2 * stride_pix2);255
+ t3 = *(int16x8_t *)(pix2 + 3 * stride_pix2);256
+257
+258
+ v16 = vsubq_u16((r0), (t0));259
+ v17 = vsubq_u16((r1), (t1));260
+ v18 = vsubq_u16((r2), (t2));261
+ v19 = vsubq_u16((r3), (t3));262
+263
+ r0 = *(int16x8_t *)(pix1 + 0 * stride_pix1 + 8);264
+ r1 = *(int16x8_t *)(pix1 + 1 * stride_pix1 + 8);265
+ r2 = *(int16x8_t *)(pix1 + 2 * stride_pix1 + 8);266
+ r3 = *(int16x8_t *)(pix1 + 3 * stride_pix1 + 8);267
+268
+ t0 = *(int16x8_t *)(pix2 + 0 * stride_pix2 + 8);269
+ t1 = *(int16x8_t *)(pix2 + 1 * stride_pix2 + 8);270
+ t2 = *(int16x8_t *)(pix2 + 2 * stride_pix2 + 8);271
+ t3 = *(int16x8_t *)(pix2 + 3 * stride_pix2 + 8);272
+273
+274
+ v20 = vsubq_u16(r0, t0);275
+ v21 = vsubq_u16(r1, t1);276
+ v22 = vsubq_u16(r2, t2);277
+ v23 = vsubq_u16(r3, t3);278
+279
+ SUMSUB_AB(v0, v1, v16, v17);280
+ SUMSUB_AB(v2, v3, v18, v19);281
+282
+ _satd_8x4v_8x8h_neon(v0, v1, v2, v3, v20, v21, v22, v23);283
+284
+}285
+286
+287
+int pixel_satd_4x4_neon(const uint16_t *pix1, intptr_t stride_pix1, const uint16_t *pix2, intptr_t stride_pix2)288
+{289
+ uint64x2_t t0, t1, r0, r1;290
+ t00 = *(uint64_t *)(pix1 + 0 * stride_pix1);291
+ t10 = *(uint64_t *)(pix1 + 1 * stride_pix1);292
+ t01 = *(uint64_t *)(pix1 + 2 * stride_pix1);293
+ t11 = *(uint64_t *)(pix1 + 3 * stride_pix1);294
+295
+ r00 = *(uint64_t *)(pix2 + 0 * stride_pix1);296
+ r10 = *(uint64_t *)(pix2 + 1 * stride_pix2);297
+ r01 = *(uint64_t *)(pix2 + 2 * stride_pix2);298
+ r11 = *(uint64_t *)(pix2 + 3 * stride_pix2);299
+300
+ return _satd_4x4_neon(vsubq_u16(t0, r0), vsubq_u16(r1, t1));301
+}302
+303
+304
+305
+306
+307
+308
+int pixel_satd_8x4_neon(const uint16_t *pix1, intptr_t stride_pix1, const uint16_t *pix2, intptr_t stride_pix2)309
+{310
+ uint16x8_t i0, i1, i2, i3, i4, i5, i6, i7;311
+312
+ i0 = *(uint16x8_t *)(pix1 + 0 * stride_pix1);313
+ i1 = *(uint16x8_t *)(pix2 + 0 * stride_pix2);314
+ i2 = *(uint16x8_t *)(pix1 + 1 * stride_pix1);315
+ i3 = *(uint16x8_t *)(pix2 + 1 * stride_pix2);316
+ i4 = *(uint16x8_t *)(pix1 + 2 * stride_pix1);317
+ i5 = *(uint16x8_t *)(pix2 + 2 * stride_pix2);318
+ i6 = *(uint16x8_t *)(pix1 + 3 * stride_pix1);319
+ i7 = *(uint16x8_t *)(pix2 + 3 * stride_pix2);320
+321
+ int16x8_t v0 = vsubq_u16(i0, i1);322
+ int16x8_t v1 = vsubq_u16(i2, i3);323
+ int16x8_t v2 = vsubq_u16(i4, i5);324
+ int16x8_t v3 = vsubq_u16(i6, i7);325
+326
+ return _satd_4x8_8x4_end_neon(v0, v1, v2, v3);327
+}328
+329
+330
+int pixel_satd_16x16_neon(const uint16_t *pix1, intptr_t stride_pix1, const uint16_t *pix2, intptr_t stride_pix2)331
+{332
+ int32x4_t v30 = vdupq_n_u32(0), v31 = vdupq_n_u32(0);333
+ int16x8_t v0, v1, v2, v3;334
+ for (int offset = 0; offset <= 12; offset += 4) {335
+ _satd_16x4_neon(pix1 + offset * stride_pix1, stride_pix1, pix2 + offset * stride_pix2, stride_pix2, v0, v1, v2, v3);336
+ v30 = vpadalq_u16(v30, v0);337
+ v30 = vpadalq_u16(v30, v1);338
+ v31 = vpadalq_u16(v31, v2);339
+ v31 = vpadalq_u16(v31, v3);340
+ }341
+ return vaddvq_s32(vaddq_s32(v30, v31));342
+343
+}344
+345
+#else //HIGH_BIT_DEPTH346
+347
+static void _satd_16x4_neon(const uint8_t *pix1, intptr_t stride_pix1, const uint8_t *pix2, intptr_t stride_pix2,348
+ int16x8_t &v0, int16x8_t &v1, int16x8_t &v2, int16x8_t &v3)349
+{350
+ uint8x16_t r0, r1, r2, r3;351
+ uint8x16_t t0, t1, t2, t3;352
+ int16x8_t v16, v17, v20, v21;353
+ int16x8_t v18, v19, v22, v23;354
+355
+ r0 = *(uint8x16_t *)(pix1 + 0 * stride_pix1);356
+ r1 = *(uint8x16_t *)(pix1 + 1 * stride_pix1);357
+ r2 = *(uint8x16_t *)(pix1 + 2 * stride_pix1);358
+ r3 = *(uint8x16_t *)(pix1 + 3 * stride_pix1);359
+360
+ t0 = *(uint8x16_t *)(pix2 + 0 * stride_pix2);361
+ t1 = *(uint8x16_t *)(pix2 + 1 * stride_pix2);362
+ t2 = *(uint8x16_t *)(pix2 + 2 * stride_pix2);363
+ t3 = *(uint8x16_t *)(pix2 + 3 * stride_pix2);364
+365
+366
+367
+ v16 = vsubl_u8(vget_low_u8(r0), vget_low_u8(t0));368
+ v20 = vsubl_high_u8(r0, t0);369
+ v17 = vsubl_u8(vget_low_u8(r1), vget_low_u8(t1));370
+ v21 = vsubl_high_u8(r1, t1);371
+ v18 = vsubl_u8(vget_low_u8(r2), vget_low_u8(t2));372
+ v22 = vsubl_high_u8(r2, t2);373
+ v19 = vsubl_u8(vget_low_u8(r3), vget_low_u8(t3));374
+ v23 = vsubl_high_u8(r3, t3);375
+376
+ SUMSUB_AB(v0, v1, v16, v17);377
+ SUMSUB_AB(v2, v3, v18, v19);378
+379
+ _satd_8x4v_8x8h_neon(v0, v1, v2, v3, v20, v21, v22, v23);380
+381
+}382
+383
+384
+static inline void _sub_8x8_fly(const uint8_t *pix1, intptr_t stride_pix1, const uint8_t *pix2, intptr_t stride_pix2,385
+ int16x8_t &v0, int16x8_t &v1, int16x8_t &v2, int16x8_t &v3,386
+ int16x8_t &v20, int16x8_t &v21, int16x8_t &v22, int16x8_t &v23)387
+{388
+ uint8x8_t r0, r1, r2, r3;389
+ uint8x8_t t0, t1, t2, t3;390
+ int16x8_t v16, v17;391
+ int16x8_t v18, v19;392
+393
+ r0 = *(uint8x8_t *)(pix1 + 0 * stride_pix1);394
+ r1 = *(uint8x8_t *)(pix1 + 1 * stride_pix1);395
+ r2 = *(uint8x8_t *)(pix1 + 2 * stride_pix1);396
+ r3 = *(uint8x8_t *)(pix1 + 3 * stride_pix1);397
+398
+ t0 = *(uint8x8_t *)(pix2 + 0 * stride_pix2);399
+ t1 = *(uint8x8_t *)(pix2 + 1 * stride_pix2);400
+ t2 = *(uint8x8_t *)(pix2 + 2 * stride_pix2);401
+ t3 = *(uint8x8_t *)(pix2 + 3 * stride_pix2);402
+403
+ v16 = vsubl_u8(r0, t0);404
+ v17 = vsubl_u8(r1, t1);405
+ v18 = vsubl_u8(r2, t2);406
+ v19 = vsubl_u8(r3, t3);407
+408
+ r0 = *(uint8x8_t *)(pix1 + 4 * stride_pix1);409
+ r1 = *(uint8x8_t *)(pix1 + 5 * stride_pix1);410
+ r2 = *(uint8x8_t *)(pix1 + 6 * stride_pix1);411
+ r3 = *(uint8x8_t *)(pix1 + 7 * stride_pix1);412
+413
+ t0 = *(uint8x8_t *)(pix2 + 4 * stride_pix2);414
+ t1 = *(uint8x8_t *)(pix2 + 5 * stride_pix2);415
+ t2 = *(uint8x8_t *)(pix2 + 6 * stride_pix2);416
+ t3 = *(uint8x8_t *)(pix2 + 7 * stride_pix2);417
+418
+ v20 = vsubl_u8(r0, t0);419
+ v21 = vsubl_u8(r1, t1);420
+ v22 = vsubl_u8(r2, t2);421
+ v23 = vsubl_u8(r3, t3);422
+423
+424
+ SUMSUB_AB(v0, v1, v16, v17);425
+ SUMSUB_AB(v2, v3, v18, v19);426
+427
+}428
+429
+int pixel_satd_4x4_neon(const uint8_t *pix1, intptr_t stride_pix1, const uint8_t *pix2, intptr_t stride_pix2)430
+{431
+ uint32x2_t t0, t1, r0, r1;432
+ t00 = *(uint32_t *)(pix1 + 0 * stride_pix1);433
+ t10 = *(uint32_t *)(pix1 + 1 * stride_pix1);434
+ t01 = *(uint32_t *)(pix1 + 2 * stride_pix1);435
+ t11 = *(uint32_t *)(pix1 + 3 * stride_pix1);436
+437
+ r00 = *(uint32_t *)(pix2 + 0 * stride_pix1);438
+ r10 = *(uint32_t *)(pix2 + 1 * stride_pix2);439
+ r01 = *(uint32_t *)(pix2 + 2 * stride_pix2);440
+ r11 = *(uint32_t *)(pix2 + 3 * stride_pix2);441
+442
+ return _satd_4x4_neon(vsubl_u8(t0, r0), vsubl_u8(r1, t1));443
+}444
+445
+446
+int pixel_satd_8x4_neon(const uint8_t *pix1, intptr_t stride_pix1, const uint8_t *pix2, intptr_t stride_pix2)447
+{448
+ uint8x8_t i0, i1, i2, i3, i4, i5, i6, i7;449
+450
+ i0 = *(uint8x8_t *)(pix1 + 0 * stride_pix1);451
+ i1 = *(uint8x8_t *)(pix2 + 0 * stride_pix2);452
+ i2 = *(uint8x8_t *)(pix1 + 1 * stride_pix1);453
+ i3 = *(uint8x8_t *)(pix2 + 1 * stride_pix2);454
+ i4 = *(uint8x8_t *)(pix1 + 2 * stride_pix1);455
+ i5 = *(uint8x8_t *)(pix2 + 2 * stride_pix2);456
+ i6 = *(uint8x8_t *)(pix1 + 3 * stride_pix1);457
+ i7 = *(uint8x8_t *)(pix2 + 3 * stride_pix2);458
+459
+ int16x8_t v0 = vsubl_u8(i0, i1);460
+ int16x8_t v1 = vsubl_u8(i2, i3);461
+ int16x8_t v2 = vsubl_u8(i4, i5);462
+ int16x8_t v3 = vsubl_u8(i6, i7);463
+464
+ return _satd_4x8_8x4_end_neon(v0, v1, v2, v3);465
+}466
+467
+int pixel_satd_16x16_neon(const uint8_t *pix1, intptr_t stride_pix1, const uint8_t *pix2, intptr_t stride_pix2)468
+{469
+ int16x8_t v30, v31;470
+ int16x8_t v0, v1, v2, v3;471
+472
+ _satd_16x4_neon(pix1, stride_pix1, pix2, stride_pix2, v0, v1, v2, v3);473
+ v30 = vaddq_s16(v0, v1);474
+ v31 = vaddq_s16(v2, v3);475
+476
+ _satd_16x4_neon(pix1 + 4 * stride_pix1, stride_pix1, pix2 + 4 * stride_pix2, stride_pix2, v0, v1, v2, v3);477
+ v0 = vaddq_s16(v0, v1);478
+ v1 = vaddq_s16(v2, v3);479
+ v30 = vaddq_s16(v30, v0);480
+ v31 = vaddq_s16(v31, v1);481
+482
+ _satd_16x4_neon(pix1 + 8 * stride_pix1, stride_pix1, pix2 + 8 * stride_pix2, stride_pix2, v0, v1, v2, v3);483
+ v0 = vaddq_s16(v0, v1);484
+ v1 = vaddq_s16(v2, v3);485
+ v30 = vaddq_s16(v30, v0);486
+ v31 = vaddq_s16(v31, v1);487
+488
+ _satd_16x4_neon(pix1 + 12 * stride_pix1, stride_pix1, pix2 + 12 * stride_pix2, stride_pix2, v0, v1, v2, v3);489
+ v0 = vaddq_s16(v0, v1);490
+ v1 = vaddq_s16(v2, v3);491
+ v30 = vaddq_s16(v30, v0);492
+ v31 = vaddq_s16(v31, v1);493
+494
+ int32x4_t sum0 = vpaddlq_u16(v30);495
+ int32x4_t sum1 = vpaddlq_u16(v31);496
+ sum0 = vaddq_s32(sum0, sum1);497
+ return vaddvq_s32(sum0);498
+499
+}500
+#endif //HIGH_BIT_DEPTH501
+502
+503
+static inline void _sa8d_8x8_neon_end(int16x8_t &v0, int16x8_t &v1, int16x8_t v2, int16x8_t v3,504
+ int16x8_t v20, int16x8_t v21, int16x8_t v22, int16x8_t v23)505
+{506
+ int16x8_t v16, v17, v18, v19;507
+ int16x8_t v4, v5, v6, v7;508
+509
+ SUMSUB_AB(v16, v18, v0, v2);510
+ SUMSUB_AB(v17, v19, v1, v3);511
+512
+ HADAMARD4_V(v20, v21, v22, v23, v0, v1, v2, v3);513
+514
+ SUMSUB_AB(v0, v16, v16, v20);515
+ SUMSUB_AB(v1, v17, v17, v21);516
+ SUMSUB_AB(v2, v18, v18, v22);517
+ SUMSUB_AB(v3, v19, v19, v23);518
+519
+ transpose_8h(v20, v21, v16, v17);520
+ transpose_8h(v4, v5, v0, v1);521
+ transpose_8h(v22, v23, v18, v19);522
+ transpose_8h(v6, v7, v2, v3);523
+524
+#if (X265_DEPTH <= 10)525
+526
+ int16x8_t v24, v25;527
+528
+ SUMSUB_AB(v2, v3, v20, v21);529
+ SUMSUB_AB(v24, v25, v4, v5);530
+ SUMSUB_AB(v0, v1, v22, v23);531
+ SUMSUB_AB(v4, v5, v6, v7);532
+533
+ transpose_4s(v20, v22, v2, v0);534
+ transpose_4s(v21, v23, v3, v1);535
+ transpose_4s(v16, v18, v24, v4);536
+ transpose_4s(v17, v19, v25, v5);537
+538
+ SUMSUB_AB(v0, v2, v20, v22);539
+ SUMSUB_AB(v1, v3, v21, v23);540
+ SUMSUB_AB(v4, v6, v16, v18);541
+ SUMSUB_AB(v5, v7, v17, v19);542
+543
+ transpose_2d(v16, v20, v0, v4);544
+ transpose_2d(v17, v21, v1, v5);545
+ transpose_2d(v18, v22, v2, v6);546
+ transpose_2d(v19, v23, v3, v7);547
+548
+549
+ v16 = vabsq_s16(v16);550
+ v17 = vabsq_s16(v17);551
+ v18 = vabsq_s16(v18);552
+ v19 = vabsq_s16(v19);553
+ v20 = vabsq_s16(v20);554
+ v21 = vabsq_s16(v21);555
+ v22 = vabsq_s16(v22);556
+ v23 = vabsq_s16(v23);557
+558
+ v16 = vmaxq_u16(v16, v20);559
+ v17 = vmaxq_u16(v17, v21);560
+ v18 = vmaxq_u16(v18, v22);561
+ v19 = vmaxq_u16(v19, v23);562
+563
+#if HIGH_BIT_DEPTH564
+ v0 = vpaddlq_u16(v16);565
+ v1 = vpaddlq_u16(v17);566
+ v0 = vpadalq_u16(v0, v18);567
+ v1 = vpadalq_u16(v1, v19);568
+569
+#else //HIGH_BIT_DEPTH570
+571
+ v0 = vaddq_u16(v16, v17);572
+ v1 = vaddq_u16(v18, v19);573
+574
+#endif //HIGH_BIT_DEPTH575
+576
+#else // HIGH_BIT_DEPTH 12 bit only, switching math to int32, each int16x8 is up-convreted to 2 int32x4 (low and high)577
+578
+ int32x4_t v2l, v2h, v3l, v3h, v24l, v24h, v25l, v25h, v0l, v0h, v1l, v1h;579
+ int32x4_t v22l, v22h, v23l, v23h;580
+ int32x4_t v4l, v4h, v5l, v5h;581
+ int32x4_t v6l, v6h, v7l, v7h;582
+ int32x4_t v16l, v16h, v17l, v17h;583
+ int32x4_t v18l, v18h, v19l, v19h;584
+ int32x4_t v20l, v20h, v21l, v21h;585
+586
+ ISUMSUB_AB_FROM_INT16(v2l, v2h, v3l, v3h, v20, v21);587
+ ISUMSUB_AB_FROM_INT16(v24l, v24h, v25l, v25h, v4, v5);588
+589
+ v22l = vmovl_s16(vget_low_s16(v22));590
+ v22h = vmovl_high_s16(v22);591
+ v23l = vmovl_s16(vget_low_s16(v23));592
+ v23h = vmovl_high_s16(v23);593
+594
+ ISUMSUB_AB(v0l, v1l, v22l, v23l);595
+ ISUMSUB_AB(v0h, v1h, v22h, v23h);596
+597
+ v6l = vmovl_s16(vget_low_s16(v6));598
+ v6h = vmovl_high_s16(v6);599
+ v7l = vmovl_s16(vget_low_s16(v7));600
+ v7h = vmovl_high_s16(v7);601
+602
+ ISUMSUB_AB(v4l, v5l, v6l, v7l);603
+ ISUMSUB_AB(v4h, v5h, v6h, v7h);604
+605
+ transpose_2d(v20l, v22l, v2l, v0l);606
+ transpose_2d(v21l, v23l, v3l, v1l);607
+ transpose_2d(v16l, v18l, v24l, v4l);608
+ transpose_2d(v17l, v19l, v25l, v5l);609
+610
+ transpose_2d(v20h, v22h, v2h, v0h);611
+ transpose_2d(v21h, v23h, v3h, v1h);612
+ transpose_2d(v16h, v18h, v24h, v4h);613
+ transpose_2d(v17h, v19h, v25h, v5h);614
+615
+ ISUMSUB_AB(v0l, v2l, v20l, v22l);616
+ ISUMSUB_AB(v1l, v3l, v21l, v23l);617
+ ISUMSUB_AB(v4l, v6l, v16l, v18l);618
+ ISUMSUB_AB(v5l, v7l, v17l, v19l);619
+620
+ ISUMSUB_AB(v0h, v2h, v20h, v22h);621
+ ISUMSUB_AB(v1h, v3h, v21h, v23h);622
+ ISUMSUB_AB(v4h, v6h, v16h, v18h);623
+ ISUMSUB_AB(v5h, v7h, v17h, v19h);624
+625
+ v16l = v0l;626
+ v16h = v4l;627
+ v20l = v0h;628
+ v20h = v4h;629
+630
+ v17l = v1l;631
+ v17h = v5l;632
+ v21l = v1h;633
+ v21h = v5h;634
+635
+ v18l = v2l;636
+ v18h = v6l;637
+ v22l = v2h;638
+ v22h = v6h;639
+640
+ v19l = v3l;641
+ v19h = v7l;642
+ v23l = v3h;643
+ v23h = v7h;644
+645
+ v16l = vabsq_s32(v16l);646
+ v17l = vabsq_s32(v17l);647
+ v18l = vabsq_s32(v18l);648
+ v19l = vabsq_s32(v19l);649
+ v20l = vabsq_s32(v20l);650
+ v21l = vabsq_s32(v21l);651
+ v22l = vabsq_s32(v22l);652
+ v23l = vabsq_s32(v23l);653
+654
+ v16h = vabsq_s32(v16h);655
+ v17h = vabsq_s32(v17h);656
+ v18h = vabsq_s32(v18h);657
+ v19h = vabsq_s32(v19h);658
+ v20h = vabsq_s32(v20h);659
+ v21h = vabsq_s32(v21h);660
+ v22h = vabsq_s32(v22h);661
+ v23h = vabsq_s32(v23h);662
+663
+ v16l = vmaxq_u32(v16l, v20l);664
+ v17l = vmaxq_u32(v17l, v21l);665
+ v18l = vmaxq_u32(v18l, v22l);666
+ v19l = vmaxq_u32(v19l, v23l);667
+668
+ v16h = vmaxq_u32(v16h, v20h);669
+ v17h = vmaxq_u32(v17h, v21h);670
+ v18h = vmaxq_u32(v18h, v22h);671
+ v19h = vmaxq_u32(v19h, v23h);672
+673
+ v16l = vaddq_u32(v16l, v16h);674
+ v17l = vaddq_u32(v17l, v17h);675
+ v18l = vaddq_u32(v18l, v18h);676
+ v19l = vaddq_u32(v19l, v19h);677
+678
+ v0 = vaddq_u32(v16l, v17l);679
+ v1 = vaddq_u32(v18l, v19l);680
+681
+682
+#endif683
+684
+}685
+686
+687
+688
+static inline void _satd_8x8_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2,689
+ int16x8_t &v0, int16x8_t &v1, int16x8_t &v2, int16x8_t &v3)690
+{691
+692
+ int16x8_t v20, v21, v22, v23;693
+ _sub_8x8_fly(pix1, stride_pix1, pix2, stride_pix2, v0, v1, v2, v3, v20, v21, v22, v23);694
+ _satd_8x4v_8x8h_neon(v0, v1, v2, v3, v20, v21, v22, v23);695
+696
+}697
+698
+699
+700
+int pixel_satd_8x8_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2)701
+{702
+ int16x8_t v30, v31;703
+ int16x8_t v0, v1, v2, v3;704
+705
+ _satd_8x8_neon(pix1, stride_pix1, pix2, stride_pix2, v0, v1, v2, v3);706
+#if !(HIGH_BIT_DEPTH)707
+ v30 = vaddq_u16(v0, v1);708
+ v31 = vaddq_u16(v2, v3);709
+710
+ uint16x8_t sum = vaddq_u16(v30, v31);711
+ return vaddvq_s32(vpaddlq_u16(sum));712
+#else713
+714
+ v30 = vaddq_u16(v0, v1);715
+ v31 = vaddq_u16(v2, v3);716
+717
+ int32x4_t sum = vpaddlq_u16(v30);718
+ sum = vpadalq_u16(sum, v31);719
+ return vaddvq_s32(sum);720
+#endif721
+}722
+723
+724
+int pixel_sa8d_8x8_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2)725
+{726
+ int16x8_t v0, v1, v2, v3;727
+ int16x8_t v20, v21, v22, v23;728
+729
+ _sub_8x8_fly(pix1, stride_pix1, pix2, stride_pix2, v0, v1, v2, v3, v20, v21, v22, v23);730
+ _sa8d_8x8_neon_end(v0, v1, v2, v3, v20, v21, v22, v23);731
+732
+#if HIGH_BIT_DEPTH733
+ int32x4_t s = vaddq_u32(v0, v1);734
+ return (vaddvq_u32(s) + 1) >> 1;735
+#else736
+ return (vaddlvq_s16(vaddq_u16(v0, v1)) + 1) >> 1;737
+#endif738
+}739
+740
+741
+742
+743
+744
+int pixel_sa8d_16x16_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2)745
+{746
+ int16x8_t v0, v1, v2, v3;747
+ int16x8_t v20, v21, v22, v23;748
+ int32x4_t v30, v31;749
+750
+ _sub_8x8_fly(pix1, stride_pix1, pix2, stride_pix2, v0, v1, v2, v3, v20, v21, v22, v23);751
+ _sa8d_8x8_neon_end(v0, v1, v2, v3, v20, v21, v22, v23);752
+753
+#if !(HIGH_BIT_DEPTH)754
+ v30 = vpaddlq_u16(v0);755
+ v31 = vpaddlq_u16(v1);756
+#else757
+ v30 = vaddq_s32(v0, v1);758
+#endif759
+760
+ _sub_8x8_fly(pix1 + 8, stride_pix1, pix2 + 8, stride_pix2, v0, v1, v2, v3, v20, v21, v22, v23);761
+ _sa8d_8x8_neon_end(v0, v1, v2, v3, v20, v21, v22, v23);762
+763
+#if !(HIGH_BIT_DEPTH)764
+ v30 = vpadalq_u16(v30, v0);765
+ v31 = vpadalq_u16(v31, v1);766
+#else767
+ v31 = vaddq_s32(v0, v1);768
+#endif769
+770
+771
+ _sub_8x8_fly(pix1 + 8 * stride_pix1, stride_pix1, pix2 + 8 * stride_pix2, stride_pix2, v0, v1, v2, v3, v20, v21, v22,772
+ v23);773
+ _sa8d_8x8_neon_end(v0, v1, v2, v3, v20, v21, v22, v23);774
+775
+#if !(HIGH_BIT_DEPTH)776
+ v30 = vpadalq_u16(v30, v0);777
+ v31 = vpadalq_u16(v31, v1);778
+#else779
+ v30 = vaddq_s32(v30, v0);780
+ v31 = vaddq_s32(v31, v1);781
+#endif782
+783
+ _sub_8x8_fly(pix1 + 8 * stride_pix1 + 8, stride_pix1, pix2 + 8 * stride_pix2 + 8, stride_pix2, v0, v1, v2, v3, v20, v21,784
+ v22, v23);785
+ _sa8d_8x8_neon_end(v0, v1, v2, v3, v20, v21, v22, v23);786
+787
+#if !(HIGH_BIT_DEPTH)788
+ v30 = vpadalq_u16(v30, v0);789
+ v31 = vpadalq_u16(v31, v1);790
+#else791
+ v30 = vaddq_s32(v30, v0);792
+ v31 = vaddq_s32(v31, v1);793
+#endif794
+795
+ v30 = vaddq_u32(v30, v31);796
+797
+ return (vaddvq_u32(v30) + 1) >> 1;798
+}799
+800
+801
+802
+803
+804
+805
+806
+807
+template<int size>808
+void blockfill_s_neon(int16_t *dst, intptr_t dstride, int16_t val)809
+{810
+ for (int y = 0; y < size; y++)811
+ {812
+ int x = 0;813
+ int16x8_t v = vdupq_n_s16(val);814
+ for (; (x + 8) <= size; x += 8)815
+ {816
+ *(int16x8_t *)&dsty * dstride + x = v;817
+ }818
+ for (; x < size; x++)819
+ {820
+ dsty * dstride + x = val;821
+ }822
+ }823
+}824
+825
+template<int lx, int ly>826
+int sad_pp_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2)827
+{828
+ int sum = 0;829
+830
+831
+ for (int y = 0; y < ly; y++)832
+ {833
+#if HIGH_BIT_DEPTH834
+ int x = 0;835
+ uint16x8_t vsum16_1 = vdupq_n_u16(0);836
+ for (; (x + 8) <= lx; x += 8)837
+ {838
+ uint16x8_t p1 = *(uint16x8_t *)&pix1x;839
+ uint16x8_t p2 = *(uint16x8_t *)&pix2x;840
+ vsum16_1 = vabaq_s16(vsum16_1, p1, p2);841
+842
+ }843
+ if (lx & 4)844
+ {845
+ uint16x4_t p1 = *(uint16x4_t *)&pix1x;846
+ uint16x4_t p2 = *(uint16x4_t *)&pix2x;847
+ sum += vaddlv_s16(vaba_s16(vdup_n_s16(0), p1, p2));848
+ x += 4;849
+ }850
+ if (lx >= 4)851
+ {852
+ sum += vaddlvq_s16(vsum16_1);853
+ }854
+855
+#else856
+857
+ int x = 0;858
+ uint16x8_t vsum16_1 = vdupq_n_u16(0);859
+ uint16x8_t vsum16_2 = vdupq_n_u16(0);860
+861
+ for (; (x + 16) <= lx; x += 16)862
+ {863
+ uint8x16_t p1 = *(uint8x16_t *)&pix1x;864
+ uint8x16_t p2 = *(uint8x16_t *)&pix2x;865
+ vsum16_1 = vabal_u8(vsum16_1, vget_low_u8(p1), vget_low_u8(p2));866
+ vsum16_2 = vabal_high_u8(vsum16_2, p1, p2);867
+ }868
+ if (lx & 8)869
+ {870
+ uint8x8_t p1 = *(uint8x8_t *)&pix1x;871
+ uint8x8_t p2 = *(uint8x8_t *)&pix2x;872
+ vsum16_1 = vabal_u8(vsum16_1, p1, p2);873
+ x += 8;874
+ }875
+ if (lx & 4)876
+ {877
+ uint32x2_t p1 = vdup_n_u32(0);878
+ p10 = *(uint32_t *)&pix1x;879
+ uint32x2_t p2 = vdup_n_u32(0);880
+ p20 = *(uint32_t *)&pix2x;881
+ vsum16_1 = vabal_u8(vsum16_1, p1, p2);882
+ x += 4;883
+ }884
+ if (lx >= 16)885
+ {886
+ vsum16_1 = vaddq_u16(vsum16_1, vsum16_2);887
+ }888
+ if (lx >= 4)889
+ {890
+ sum += vaddvq_u16(vsum16_1);891
+ }892
+893
+#endif894
+ if (lx & 3) for (; x < lx; x++)895
+ {896
+ sum += abs(pix1x - pix2x);897
+ }898
+899
+ pix1 += stride_pix1;900
+ pix2 += stride_pix2;901
+ }902
+903
+ return sum;904
+}905
+906
+template<int lx, int ly>907
+void sad_x3_neon(const pixel *pix1, const pixel *pix2, const pixel *pix3, const pixel *pix4, intptr_t frefstride,908
+ int32_t *res)909
+{910
+ res0 = 0;911
+ res1 = 0;912
+ res2 = 0;913
+ for (int y = 0; y < ly; y++)914
+ {915
+ int x = 0;916
+ uint16x8_t vsum16_0 = vdupq_n_u16(0);917
+ uint16x8_t vsum16_1 = vdupq_n_u16(0);918
+ uint16x8_t vsum16_2 = vdupq_n_u16(0);919
+#if HIGH_BIT_DEPTH920
+ for (; (x + 8) <= lx; x += 8)921
+ {922
+ uint16x8_t p1 = *(uint16x8_t *)&pix1x;923
+ uint16x8_t p2 = *(uint16x8_t *)&pix2x;924
+ uint16x8_t p3 = *(uint16x8_t *)&pix3x;925
+ uint16x8_t p4 = *(uint16x8_t *)&pix4x;926
+ vsum16_0 = vabaq_s16(vsum16_0, p1, p2);927
+ vsum16_1 = vabaq_s16(vsum16_1, p1, p3);928
+ vsum16_2 = vabaq_s16(vsum16_2, p1, p4);929
+930
+ }931
+ if (lx & 4)932
+ {933
+ uint16x4_t p1 = *(uint16x4_t *)&pix1x;934
+ uint16x4_t p2 = *(uint16x4_t *)&pix2x;935
+ uint16x4_t p3 = *(uint16x4_t *)&pix3x;936
+ uint16x4_t p4 = *(uint16x4_t *)&pix4x;937
+ res0 += vaddlv_s16(vaba_s16(vdup_n_s16(0), p1, p2));938
+ res1 += vaddlv_s16(vaba_s16(vdup_n_s16(0), p1, p3));939
+ res2 += vaddlv_s16(vaba_s16(vdup_n_s16(0), p1, p4));940
+ x += 4;941
+ }942
+ if (lx >= 4)943
+ {944
+ res0 += vaddlvq_s16(vsum16_0);945
+ res1 += vaddlvq_s16(vsum16_1);946
+ res2 += vaddlvq_s16(vsum16_2);947
+ }948
+#else949
+950
+ for (; (x + 16) <= lx; x += 16)951
+ {952
+ uint8x16_t p1 = *(uint8x16_t *)&pix1x;953
+ uint8x16_t p2 = *(uint8x16_t *)&pix2x;954
+ uint8x16_t p3 = *(uint8x16_t *)&pix3x;955
+ uint8x16_t p4 = *(uint8x16_t *)&pix4x;956
+ vsum16_0 = vabal_u8(vsum16_0, vget_low_u8(p1), vget_low_u8(p2));957
+ vsum16_0 = vabal_high_u8(vsum16_0, p1, p2);958
+ vsum16_1 = vabal_u8(vsum16_1, vget_low_u8(p1), vget_low_u8(p3));959
+ vsum16_1 = vabal_high_u8(vsum16_1, p1, p3);960
+ vsum16_2 = vabal_u8(vsum16_2, vget_low_u8(p1), vget_low_u8(p4));961
+ vsum16_2 = vabal_high_u8(vsum16_2, p1, p4);962
+ }963
+ if (lx & 8)964
+ {965
+ uint8x8_t p1 = *(uint8x8_t *)&pix1x;966
+ uint8x8_t p2 = *(uint8x8_t *)&pix2x;967
+ uint8x8_t p3 = *(uint8x8_t *)&pix3x;968
+ uint8x8_t p4 = *(uint8x8_t *)&pix4x;969
+ vsum16_0 = vabal_u8(vsum16_0, p1, p2);970
+ vsum16_1 = vabal_u8(vsum16_1, p1, p3);971
+ vsum16_2 = vabal_u8(vsum16_2, p1, p4);972
+ x += 8;973
+ }974
+ if (lx & 4)975
+ {976
+ uint32x2_t p1 = vdup_n_u32(0);977
+ p10 = *(uint32_t *)&pix1x;978
+ uint32x2_t p2 = vdup_n_u32(0);979
+ p20 = *(uint32_t *)&pix2x;980
+ uint32x2_t p3 = vdup_n_u32(0);981
+ p30 = *(uint32_t *)&pix3x;982
+ uint32x2_t p4 = vdup_n_u32(0);983
+ p40 = *(uint32_t *)&pix4x;984
+ vsum16_0 = vabal_u8(vsum16_0, p1, p2);985
+ vsum16_1 = vabal_u8(vsum16_1, p1, p3);986
+ vsum16_2 = vabal_u8(vsum16_2, p1, p4);987
+ x += 4;988
+ }989
+ if (lx >= 4)990
+ {991
+ res0 += vaddvq_u16(vsum16_0);992
+ res1 += vaddvq_u16(vsum16_1);993
+ res2 += vaddvq_u16(vsum16_2);994
+ }995
+996
+#endif997
+ if (lx & 3) for (; x < lx; x++)998
+ {999
+ res0 += abs(pix1x - pix2x);1000
+ res1 += abs(pix1x - pix3x);1001
+ res2 += abs(pix1x - pix4x);1002
+ }1003
+1004
+ pix1 += FENC_STRIDE;1005
+ pix2 += frefstride;1006
+ pix3 += frefstride;1007
+ pix4 += frefstride;1008
+ }1009
+}1010
+1011
+template<int lx, int ly>1012
+void sad_x4_neon(const pixel *pix1, const pixel *pix2, const pixel *pix3, const pixel *pix4, const pixel *pix5,1013
+ intptr_t frefstride, int32_t *res)1014
+{1015
+ int32x4_t result = {0};1016
+ for (int y = 0; y < ly; y++)1017
+ {1018
+ int x = 0;1019
+ uint16x8_t vsum16_0 = vdupq_n_u16(0);1020
+ uint16x8_t vsum16_1 = vdupq_n_u16(0);1021
+ uint16x8_t vsum16_2 = vdupq_n_u16(0);1022
+ uint16x8_t vsum16_3 = vdupq_n_u16(0);1023
+#if HIGH_BIT_DEPTH1024
+ for (; (x + 16) <= lx; x += 16)1025
+ {1026
+ uint16x8x2_t p1 = vld1q_u16_x2(&pix1x);1027
+ uint16x8x2_t p2 = vld1q_u16_x2(&pix2x);1028
+ uint16x8x2_t p3 = vld1q_u16_x2(&pix3x);1029
+ uint16x8x2_t p4 = vld1q_u16_x2(&pix4x);1030
+ uint16x8x2_t p5 = vld1q_u16_x2(&pix5x);1031
+ vsum16_0 = vabaq_s16(vsum16_0, p1.val0, p2.val0);1032
+ vsum16_1 = vabaq_s16(vsum16_1, p1.val0, p3.val0);1033
+ vsum16_2 = vabaq_s16(vsum16_2, p1.val0, p4.val0);1034
+ vsum16_3 = vabaq_s16(vsum16_3, p1.val0, p5.val0);1035
+ vsum16_0 = vabaq_s16(vsum16_0, p1.val1, p2.val1);1036
+ vsum16_1 = vabaq_s16(vsum16_1, p1.val1, p3.val1);1037
+ vsum16_2 = vabaq_s16(vsum16_2, p1.val1, p4.val1);1038
+ vsum16_3 = vabaq_s16(vsum16_3, p1.val1, p5.val1);1039
+ }1040
+ if (lx & 8)1041
+ {1042
+ uint16x8_t p1 = *(uint16x8_t *)&pix1x;1043
+ uint16x8_t p2 = *(uint16x8_t *)&pix2x;1044
+ uint16x8_t p3 = *(uint16x8_t *)&pix3x;1045
+ uint16x8_t p4 = *(uint16x8_t *)&pix4x;1046
+ uint16x8_t p5 = *(uint16x8_t *)&pix5x;1047
+ vsum16_0 = vabaq_s16(vsum16_0, p1, p2);1048
+ vsum16_1 = vabaq_s16(vsum16_1, p1, p3);1049
+ vsum16_2 = vabaq_s16(vsum16_2, p1, p4);1050
+ vsum16_3 = vabaq_s16(vsum16_3, p1, p5);1051
+ x += 8;1052
+ }1053
+ if (lx & 4)1054
+ {1055
+ /* This is equivalent to getting the absolute difference of pix1x with each of1056
+ * pix2 - pix5, then summing across the vector (4 values each) and adding the1057
+ * result to result. */1058
+ uint16x8_t p1 = vreinterpretq_s16_u64(1059
+ vld1q_dup_u64((uint64_t *)&pix1x));1060
+ uint16x8_t p2_3 = vcombine_s16(*(uint16x4_t *)&pix2x, *(uint16x4_t *)&pix3x);1061
+ uint16x8_t p4_5 = vcombine_s16(*(uint16x4_t *)&pix4x, *(uint16x4_t *)&pix5x);1062
+1063
+ uint16x8_t a = vabdq_u16(p1, p2_3);1064
+ uint16x8_t b = vabdq_u16(p1, p4_5);1065
+1066
+ result = vpadalq_s16(result, vpaddq_s16(a, b));1067
+ x += 4;1068
+ }1069
+ if (lx >= 4)1070
+ {1071
+ /* This is equivalent to adding across each of the sum vectors and then adding1072
+ * to result. */1073
+ uint16x8_t a = vpaddq_s16(vsum16_0, vsum16_1);1074
+ uint16x8_t b = vpaddq_s16(vsum16_2, vsum16_3);1075
+ uint16x8_t c = vpaddq_s16(a, b);1076
+ result = vpadalq_s16(result, c);1077
+ }1078
+1079
+#else1080
+1081
+ for (; (x + 16) <= lx; x += 16)1082
+ {1083
+ uint8x16_t p1 = *(uint8x16_t *)&pix1x;1084
+ uint8x16_t p2 = *(uint8x16_t *)&pix2x;1085
+ uint8x16_t p3 = *(uint8x16_t *)&pix3x;1086
+ uint8x16_t p4 = *(uint8x16_t *)&pix4x;1087
+ uint8x16_t p5 = *(uint8x16_t *)&pix5x;1088
+ vsum16_0 = vabal_u8(vsum16_0, vget_low_u8(p1), vget_low_u8(p2));1089
+ vsum16_0 = vabal_high_u8(vsum16_0, p1, p2);1090
+ vsum16_1 = vabal_u8(vsum16_1, vget_low_u8(p1), vget_low_u8(p3));1091
+ vsum16_1 = vabal_high_u8(vsum16_1, p1, p3);1092
+ vsum16_2 = vabal_u8(vsum16_2, vget_low_u8(p1), vget_low_u8(p4));1093
+ vsum16_2 = vabal_high_u8(vsum16_2, p1, p4);1094
+ vsum16_3 = vabal_u8(vsum16_3, vget_low_u8(p1), vget_low_u8(p5));1095
+ vsum16_3 = vabal_high_u8(vsum16_3, p1, p5);1096
+ }1097
+ if (lx & 8)1098
+ {1099
+ uint8x8_t p1 = *(uint8x8_t *)&pix1x;1100
+ uint8x8_t p2 = *(uint8x8_t *)&pix2x;1101
+ uint8x8_t p3 = *(uint8x8_t *)&pix3x;1102
+ uint8x8_t p4 = *(uint8x8_t *)&pix4x;1103
+ uint8x8_t p5 = *(uint8x8_t *)&pix5x;1104
+ vsum16_0 = vabal_u8(vsum16_0, p1, p2);1105
+ vsum16_1 = vabal_u8(vsum16_1, p1, p3);1106
+ vsum16_2 = vabal_u8(vsum16_2, p1, p4);1107
+ vsum16_3 = vabal_u8(vsum16_3, p1, p5);1108
+ x += 8;1109
+ }1110
+ if (lx & 4)1111
+ {1112
+ uint8x16_t p1 = vreinterpretq_u32_u8(1113
+ vld1q_dup_u32((uint32_t *)&pix1x));1114
+1115
+ uint32x4_t p_x4;1116
+ p_x4 = vld1q_lane_u32((uint32_t *)&pix2x, p_x4, 0);1117
+ p_x4 = vld1q_lane_u32((uint32_t *)&pix3x, p_x4, 1);1118
+ p_x4 = vld1q_lane_u32((uint32_t *)&pix4x, p_x4, 2);1119
+ p_x4 = vld1q_lane_u32((uint32_t *)&pix5x, p_x4, 3);1120
+1121
+ uint16x8_t sum = vabdl_u8(vget_low_u8(p1), vget_low_u8(p_x4));1122
+ uint16x8_t sum2 = vabdl_high_u8(p1, p_x4);1123
+1124
+ uint16x8_t a = vpaddq_u16(sum, sum2);1125
+ result = vpadalq_u16(result, a);1126
+ }1127
+ if (lx >= 4)1128
+ {1129
+ result0 += vaddvq_u16(vsum16_0);1130
+ result1 += vaddvq_u16(vsum16_1);1131
+ result2 += vaddvq_u16(vsum16_2);1132
+ result3 += vaddvq_u16(vsum16_3);1133
+ }1134
+1135
+#endif1136
+ if (lx & 3) for (; x < lx; x++)1137
+ {1138
+ result0 += abs(pix1x - pix2x);1139
+ result1 += abs(pix1x - pix3x);1140
+ result2 += abs(pix1x - pix4x);1141
+ result3 += abs(pix1x - pix5x);1142
+ }1143
+1144
+ pix1 += FENC_STRIDE;1145
+ pix2 += frefstride;1146
+ pix3 += frefstride;1147
+ pix4 += frefstride;1148
+ pix5 += frefstride;1149
+ }1150
+ vst1q_s32(res, result);1151
+}1152
+1153
+1154
+template<int lx, int ly, class T1, class T2>1155
+sse_t sse_neon(const T1 *pix1, intptr_t stride_pix1, const T2 *pix2, intptr_t stride_pix2)1156
+{1157
+ sse_t sum = 0;1158
+1159
+ int32x4_t vsum1 = vdupq_n_s32(0);1160
+ int32x4_t vsum2 = vdupq_n_s32(0);1161
+ for (int y = 0; y < ly; y++)1162
+ {1163
+ int x = 0;1164
+ for (; (x + 8) <= lx; x += 8)1165
+ {1166
+ int16x8_t tmp;1167
+ if (sizeof(T1) == 2 && sizeof(T2) == 2)1168
+ {1169
+ tmp = vsubq_s16(*(int16x8_t *)&pix1x, *(int16x8_t *)&pix2x);1170
+ }1171
+ else if (sizeof(T1) == 1 && sizeof(T2) == 1)1172
+ {1173
+ tmp = vsubl_u8(*(uint8x8_t *)&pix1x, *(uint8x8_t *)&pix2x);1174
+ }1175
+ else1176
+ {1177
+ X265_CHECK(false, "unsupported sse");1178
+ }1179
+ vsum1 = vmlal_s16(vsum1, vget_low_s16(tmp), vget_low_s16(tmp));1180
+ vsum2 = vmlal_high_s16(vsum2, tmp, tmp);1181
+ }1182
+ for (; x < lx; x++)1183
+ {1184
+ int tmp = pix1x - pix2x;1185
+ sum += (tmp * tmp);1186
+ }1187
+1188
+ if (sizeof(T1) == 2 && sizeof(T2) == 2)1189
+ {1190
+ int32x4_t vsum = vaddq_u32(vsum1, vsum2);;1191
+ sum += vaddvq_u32(vsum);1192
+ vsum1 = vsum2 = vdupq_n_u16(0);1193
+ }1194
+1195
+ pix1 += stride_pix1;1196
+ pix2 += stride_pix2;1197
+ }1198
+ int32x4_t vsum = vaddq_u32(vsum1, vsum2);1199
+1200
+ return sum + vaddvq_u32(vsum);1201
+}1202
+1203
+1204
+template<int bx, int by>1205
+void blockcopy_ps_neon(int16_t *a, intptr_t stridea, const pixel *b, intptr_t strideb)1206
+{1207
+ for (int y = 0; y < by; y++)1208
+ {1209
+ int x = 0;1210
+ for (; (x + 8) <= bx; x += 8)1211
+ {1212
+#if HIGH_BIT_DEPTH1213
+ *(int16x8_t *)&ax = *(int16x8_t *)&bx;1214
+#else1215
+ *(int16x8_t *)&ax = vmovl_u8(*(int8x8_t *)&bx);1216
+#endif1217
+ }1218
+ for (; x < bx; x++)1219
+ {1220
+ ax = (int16_t)bx;1221
+ }1222
+1223
+ a += stridea;1224
+ b += strideb;1225
+ }1226
+}1227
+1228
+1229
+template<int bx, int by>1230
+void blockcopy_pp_neon(pixel *a, intptr_t stridea, const pixel *b, intptr_t strideb)1231
+{1232
+ for (int y = 0; y < by; y++)1233
+ {1234
+ int x = 0;1235
+#if HIGH_BIT_DEPTH1236
+ for (; (x + 8) <= bx; x += 8)1237
+ {1238
+ *(int16x8_t *)&ax = *(int16x8_t *)&bx;1239
+ }1240
+ if (bx & 4)1241
+ {1242
+ *(uint64_t *)&ax = *(uint64_t *)&bx;1243
+ x += 4;1244
+ }1245
+#else1246
+ for (; (x + 16) <= bx; x += 16)1247
+ {1248
+ *(uint8x16_t *)&ax = *(uint8x16_t *)&bx;1249
+ }1250
+ if (bx & 8)1251
+ {1252
+ *(uint8x8_t *)&ax = *(uint8x8_t *)&bx;1253
+ x += 8;1254
+ }1255
+ if (bx & 4)1256
+ {1257
+ *(uint32_t *)&ax = *(uint32_t *)&bx;1258
+ x += 4;1259
+ }1260
+#endif1261
+ for (; x < bx; x++)1262
+ {1263
+ ax = bx;1264
+ }1265
+1266
+ a += stridea;1267
+ b += strideb;1268
+ }1269
+}1270
+1271
+1272
+template<int bx, int by>1273
+void pixel_sub_ps_neon(int16_t *a, intptr_t dstride, const pixel *b0, const pixel *b1, intptr_t sstride0,1274
+ intptr_t sstride1)1275
+{1276
+ for (int y = 0; y < by; y++)1277
+ {1278
+ int x = 0;1279
+ for (; (x + 8) <= bx; x += 8)1280
+ {1281
+#if HIGH_BIT_DEPTH1282
+ *(int16x8_t *)&ax = vsubq_s16(*(int16x8_t *)&b0x, *(int16x8_t *)&b1x);1283
+#else1284
+ *(int16x8_t *)&ax = vsubl_u8(*(uint8x8_t *)&b0x, *(uint8x8_t *)&b1x);1285
+#endif1286
+ }1287
+ for (; x < bx; x++)1288
+ {1289
+ ax = (int16_t)(b0x - b1x);1290
+ }1291
+1292
+ b0 += sstride0;1293
+ b1 += sstride1;1294
+ a += dstride;1295
+ }1296
+}1297
+1298
+template<int bx, int by>1299
+void pixel_add_ps_neon(pixel *a, intptr_t dstride, const pixel *b0, const int16_t *b1, intptr_t sstride0,1300
+ intptr_t sstride1)1301
+{1302
+ for (int y = 0; y < by; y++)1303
+ {1304
+ int x = 0;1305
+ for (; (x + 8) <= bx; x += 8)1306
+ {1307
+ int16x8_t t;1308
+ int16x8_t b1e = *(int16x8_t *)&b1x;1309
+ int16x8_t b0e;1310
+#if HIGH_BIT_DEPTH1311
+ b0e = *(int16x8_t *)&b0x;1312
+ t = vaddq_s16(b0e, b1e);1313
+ t = vminq_s16(t, vdupq_n_s16((1 << X265_DEPTH) - 1));1314
+ t = vmaxq_s16(t, vdupq_n_s16(0));1315
+ *(int16x8_t *)&ax = t;1316
+#else1317
+ b0e = vmovl_u8(*(uint8x8_t *)&b0x);1318
+ t = vaddq_s16(b0e, b1e);1319
+ *(uint8x8_t *)&ax = vqmovun_s16(t);1320
+#endif1321
+ }1322
+ for (; x < bx; x++)1323
+ {1324
+ ax = (int16_t)x265_clip(b0x + b1x);1325
+ }1326
+1327
+ b0 += sstride0;1328
+ b1 += sstride1;1329
+ a += dstride;1330
+ }1331
+}1332
+1333
+template<int bx, int by>1334
+void addAvg_neon(const int16_t *src0, const int16_t *src1, pixel *dst, intptr_t src0Stride, intptr_t src1Stride,1335
+ intptr_t dstStride)1336
+{1337
+1338
+ const int shiftNum = IF_INTERNAL_PREC + 1 - X265_DEPTH;1339
+ const int offset = (1 << (shiftNum - 1)) + 2 * IF_INTERNAL_OFFS;1340
+1341
+ const int32x4_t addon = vdupq_n_s32(offset);1342
+ for (int y = 0; y < by; y++)1343
+ {1344
+ int x = 0;1345
+1346
+ for (; (x + 8) <= bx; x += 8)1347
+ {1348
+ int16x8_t in0 = *(int16x8_t *)&src0x;1349
+ int16x8_t in1 = *(int16x8_t *)&src1x;1350
+ int32x4_t t1 = vaddl_s16(vget_low_s16(in0), vget_low_s16(in1));1351
+ int32x4_t t2 = vaddl_high_s16(in0, in1);1352
+ t1 = vaddq_s32(t1, addon);1353
+ t2 = vaddq_s32(t2, addon);1354
+ t1 = vshrq_n_s32(t1, shiftNum);1355
+ t2 = vshrq_n_s32(t2, shiftNum);1356
+ int16x8_t t = vuzp1q_s16(t1, t2);1357
+#if HIGH_BIT_DEPTH1358
+ t = vminq_s16(t, vdupq_n_s16((1 << X265_DEPTH) - 1));1359
+ t = vmaxq_s16(t, vdupq_n_s16(0));1360
+ *(int16x8_t *)&dstx = t;1361
+#else1362
+ *(uint8x8_t *)&dstx = vqmovun_s16(t);1363
+#endif1364
+ }1365
+ for (; x < bx; x += 2)1366
+ {1367
+ dstx + 0 = x265_clip((src0x + 0 + src1x + 0 + offset) >> shiftNum);1368
+ dstx + 1 = x265_clip((src0x + 1 + src1x + 1 + offset) >> shiftNum);1369
+ }1370
+1371
+ src0 += src0Stride;1372
+ src1 += src1Stride;1373
+ dst += dstStride;1374
+ }1375
+}1376
+1377
+template<int lx, int ly>1378
+void pixelavg_pp_neon(pixel *dst, intptr_t dstride, const pixel *src0, intptr_t sstride0, const pixel *src1,1379
+ intptr_t sstride1, int)1380
+{1381
+ for (int y = 0; y < ly; y++)1382
+ {1383
+ int x = 0;1384
+ for (; (x + 8) <= lx; x += 8)1385
+ {1386
+#if HIGH_BIT_DEPTH1387
+ uint16x8_t in0 = *(uint16x8_t *)&src0x;1388
+ uint16x8_t in1 = *(uint16x8_t *)&src1x;1389
+ uint16x8_t t = vrhaddq_u16(in0, in1);1390
+ *(uint16x8_t *)&dstx = t;1391
+#else1392
+ int16x8_t in0 = vmovl_u8(*(uint8x8_t *)&src0x);1393
+ int16x8_t in1 = vmovl_u8(*(uint8x8_t *)&src1x);1394
+ int16x8_t t = vrhaddq_s16(in0, in1);1395
+ *(uint8x8_t *)&dstx = vmovn_u16(t);1396
+#endif1397
+ }1398
+ for (; x < lx; x++)1399
+ {1400
+ dstx = (src0x + src1x + 1) >> 1;1401
+ }1402
+1403
+ src0 += sstride0;1404
+ src1 += sstride1;1405
+ dst += dstride;1406
+ }1407
+}1408
+1409
+1410
+template<int size>1411
+void cpy1Dto2D_shl_neon(int16_t *dst, const int16_t *src, intptr_t dstStride, int shift)1412
+{1413
+ X265_CHECK((((intptr_t)dst | (dstStride * sizeof(*dst))) & 15) == 0 || size == 4, "dst alignment error\n");1414
+ X265_CHECK(((intptr_t)src & 15) == 0, "src alignment error\n");1415
+ X265_CHECK(shift >= 0, "invalid shift\n");1416
+1417
+ for (int i = 0; i < size; i++)1418
+ {1419
+ int j = 0;1420
+ for (; (j + 8) <= size; j += 8)1421
+ {1422
+ *(int16x8_t *)&dstj = vshlq_s16(*(int16x8_t *)&srcj, vdupq_n_s16(shift));1423
+ }1424
+ for (; j < size; j++)1425
+ {1426
+ dstj = srcj << shift;1427
+ }1428
+ src += size;1429
+ dst += dstStride;1430
+ }1431
+}1432
+1433
+1434
+template<int size>1435
+uint64_t pixel_var_neon(const uint8_t *pix, intptr_t i_stride)1436
+{1437
+ uint32_t sum = 0, sqr = 0;1438
+1439
+ int32x4_t vsqr = vdupq_n_s32(0);1440
+ for (int y = 0; y < size; y++)1441
+ {1442
+ int x = 0;1443
+ int16x8_t vsum = vdupq_n_s16(0);1444
+ for (; (x + 8) <= size; x += 8)1445
+ {1446
+ int16x8_t in;1447
+ in = vmovl_u8(*(uint8x8_t *)&pixx);1448
+ vsum = vaddq_u16(vsum, in);1449
+ vsqr = vmlal_s16(vsqr, vget_low_s16(in), vget_low_s16(in));1450
+ vsqr = vmlal_high_s16(vsqr, in, in);1451
+ }1452
+ for (; x < size; x++)1453
+ {1454
+ sum += pixx;1455
+ sqr += pixx * pixx;1456
+ }1457
+ sum += vaddvq_s16(vsum);1458
+1459
+ pix += i_stride;1460
+ }1461
+ sqr += vaddvq_u32(vsqr);1462
+ return sum + ((uint64_t)sqr << 32);1463
+}1464
+1465
+template<int blockSize>1466
+void getResidual_neon(const pixel *fenc, const pixel *pred, int16_t *residual, intptr_t stride)1467
+{1468
+ for (int y = 0; y < blockSize; y++)1469
+ {1470
+ int x = 0;1471
+ for (; (x + 8) < blockSize; x += 8)1472
+ {1473
+ int16x8_t vfenc, vpred;1474
+#if HIGH_BIT_DEPTH1475
+ vfenc = *(int16x8_t *)&fencx;1476
+ vpred = *(int16x8_t *)&predx;1477
+#else1478
+ vfenc = vmovl_u8(*(uint8x8_t *)&fencx);1479
+ vpred = vmovl_u8(*(uint8x8_t *)&predx);1480
+#endif1481
+ *(int16x8_t *)&residualx = vsubq_s16(vfenc, vpred);1482
+ }1483
+ for (; x < blockSize; x++)1484
+ {1485
+ residualx = static_cast<int16_t>(fencx) - static_cast<int16_t>(predx);1486
+ }1487
+ fenc += stride;1488
+ residual += stride;1489
+ pred += stride;1490
+ }1491
+}1492
+1493
+template<int size>1494
+int psyCost_pp_neon(const pixel *source, intptr_t sstride, const pixel *recon, intptr_t rstride)1495
+{1496
+ static pixel zeroBuf8 /* = { 0 } */;1497
+1498
+ if (size)1499
+ {1500
+ int dim = 1 << (size + 2);1501
+ uint32_t totEnergy = 0;1502
+ for (int i = 0; i < dim; i += 8)1503
+ {1504
+ for (int j = 0; j < dim; j += 8)1505
+ {1506
+ /* AC energy, measured by sa8d (AC + DC) minus SAD (DC) */1507
+ int sourceEnergy = pixel_sa8d_8x8_neon(source + i * sstride + j, sstride, zeroBuf, 0) -1508
+ (sad_pp_neon<8, 8>(source + i * sstride + j, sstride, zeroBuf, 0) >> 2);1509
+ int reconEnergy = pixel_sa8d_8x8_neon(recon + i * rstride + j, rstride, zeroBuf, 0) -1510
+ (sad_pp_neon<8, 8>(recon + i * rstride + j, rstride, zeroBuf, 0) >> 2);1511
+1512
+ totEnergy += abs(sourceEnergy - reconEnergy);1513
+ }1514
+ }1515
+ return totEnergy;1516
+ }1517
+ else1518
+ {1519
+ /* 4x4 is too small for sa8d */1520
+ int sourceEnergy = pixel_satd_4x4_neon(source, sstride, zeroBuf, 0) - (sad_pp_neon<4, 4>(source, sstride, zeroBuf,1521
+ 0) >> 2);1522
+ int reconEnergy = pixel_satd_4x4_neon(recon, rstride, zeroBuf, 0) - (sad_pp_neon<4, 4>(recon, rstride, zeroBuf,1523
+ 0) >> 2);1524
+ return abs(sourceEnergy - reconEnergy);1525
+ }1526
+}1527
+1528
+1529
+template<int w, int h>1530
+// Calculate sa8d in blocks of 8x81531
+int sa8d8(const pixel *pix1, intptr_t i_pix1, const pixel *pix2, intptr_t i_pix2)1532
+{1533
+ int cost = 0;1534
+1535
+ for (int y = 0; y < h; y += 8)1536
+ for (int x = 0; x < w; x += 8)1537
+ {1538
+ cost += pixel_sa8d_8x8_neon(pix1 + i_pix1 * y + x, i_pix1, pix2 + i_pix2 * y + x, i_pix2);1539
+ }1540
+1541
+ return cost;1542
+}1543
+1544
+template<int w, int h>1545
+// Calculate sa8d in blocks of 16x161546
+int sa8d16(const pixel *pix1, intptr_t i_pix1, const pixel *pix2, intptr_t i_pix2)1547
+{1548
+ int cost = 0;1549
+1550
+ for (int y = 0; y < h; y += 16)1551
+ for (int x = 0; x < w; x += 16)1552
+ {1553
+ cost += pixel_sa8d_16x16_neon(pix1 + i_pix1 * y + x, i_pix1, pix2 + i_pix2 * y + x, i_pix2);1554
+ }1555
+1556
+ return cost;1557
+}1558
+1559
+template<int size>1560
+void cpy2Dto1D_shl_neon(int16_t *dst, const int16_t *src, intptr_t srcStride, int shift)1561
+{1562
+ X265_CHECK(((intptr_t)dst & 15) == 0, "dst alignment error\n");1563
+ X265_CHECK((((intptr_t)src | (srcStride * sizeof(*src))) & 15) == 0 || size == 4, "src alignment error\n");1564
+ X265_CHECK(shift >= 0, "invalid shift\n");1565
+1566
+ for (int i = 0; i < size; i++)1567
+ {1568
+ for (int j = 0; j < size; j++)1569
+ {1570
+ dstj = srcj << shift;1571
+ }1572
+1573
+ src += srcStride;1574
+ dst += size;1575
+ }1576
+}1577
+1578
+1579
+template<int w, int h>1580
+// calculate satd in blocks of 4x41581
+int satd4_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2)1582
+{1583
+ int satd = 0;1584
+1585
+ for (int row = 0; row < h; row += 4)1586
+ for (int col = 0; col < w; col += 4)1587
+ satd += pixel_satd_4x4_neon(pix1 + row * stride_pix1 + col, stride_pix1,1588
+ pix2 + row * stride_pix2 + col, stride_pix2);1589
+1590
+ return satd;1591
+}1592
+1593
+template<int w, int h>1594
+// calculate satd in blocks of 8x41595
+int satd8_neon(const pixel *pix1, intptr_t stride_pix1, const pixel *pix2, intptr_t stride_pix2)1596
+{1597
+ int satd = 0;1598
+1599
+ if (((w | h) & 15) == 0)1600
+ {1601
+ for (int row = 0; row < h; row += 16)1602
+ for (int col = 0; col < w; col += 16)1603
+ satd += pixel_satd_16x16_neon(pix1 + row * stride_pix1 + col, stride_pix1,1604
+ pix2 + row * stride_pix2 + col, stride_pix2);1605
+1606
+ }1607
+ else if (((w | h) & 7) == 0)1608
+ {1609
+ for (int row = 0; row < h; row += 8)1610
+ for (int col = 0; col < w; col += 8)1611
+ satd += pixel_satd_8x8_neon(pix1 + row * stride_pix1 + col, stride_pix1,1612
+ pix2 + row * stride_pix2 + col, stride_pix2);1613
+1614
+ }1615
+ else1616
+ {1617
+ for (int row = 0; row < h; row += 4)1618
+ for (int col = 0; col < w; col += 8)1619
+ satd += pixel_satd_8x4_neon(pix1 + row * stride_pix1 + col, stride_pix1,1620
+ pix2 + row * stride_pix2 + col, stride_pix2);1621
+ }1622
+1623
+ return satd;1624
+}1625
+1626
+1627
+template<int blockSize>1628
+void transpose_neon(pixel *dst, const pixel *src, intptr_t stride)1629
+{1630
+ for (int k = 0; k < blockSize; k++)1631
+ for (int l = 0; l < blockSize; l++)1632
+ {1633
+ dstk * blockSize + l = srcl * stride + k;1634
+ }1635
+}1636
+1637
+1638
+template<>1639
+void transpose_neon<8>(pixel *dst, const pixel *src, intptr_t stride)1640
+{1641
+ transpose8x8(dst, src, 8, stride);1642
+}1643
+1644
+template<>1645
+void transpose_neon<16>(pixel *dst, const pixel *src, intptr_t stride)1646
+{1647
+ transpose16x16(dst, src, 16, stride);1648
+}1649
+1650
+template<>1651
+void transpose_neon<32>(pixel *dst, const pixel *src, intptr_t stride)1652
+{1653
+ transpose32x32(dst, src, 32, stride);1654
+}1655
+1656
+1657
+template<>1658
+void transpose_neon<64>(pixel *dst, const pixel *src, intptr_t stride)1659
+{1660
+ transpose32x32(dst, src, 64, stride);1661
+ transpose32x32(dst + 32 * 64 + 32, src + 32 * stride + 32, 64, stride);1662
+ transpose32x32(dst + 32 * 64, src + 32, 64, stride);1663
+ transpose32x32(dst + 32, src + 32 * stride, 64, stride);1664
+}1665
+1666
+1667
+template<int size>1668
+sse_t pixel_ssd_s_neon(const int16_t *a, intptr_t dstride)1669
+{1670
+ sse_t sum = 0;1671
+1672
+1673
+ int32x4_t vsum = vdupq_n_s32(0);1674
+1675
+ for (int y = 0; y < size; y++)1676
+ {1677
+ int x = 0;1678
+1679
+ for (; (x + 8) <= size; x += 8)1680
+ {1681
+ int16x8_t in = *(int16x8_t *)&ax;1682
+ vsum = vmlal_s16(vsum, vget_low_s16(in), vget_low_s16(in));1683
+ vsum = vmlal_high_s16(vsum, (in), (in));1684
+ }1685
+ for (; x < size; x++)1686
+ {1687
+ sum += ax * ax;1688
+ }1689
+1690
+ a += dstride;1691
+ }1692
+ return sum + vaddvq_s32(vsum);1693
+}1694
+1695
+1696
+};1697
+1698
+1699
+1700
+1701
+namespace X265_NS1702
+{1703
+1704
+1705
+void setupPixelPrimitives_neon(EncoderPrimitives &p)1706
+{1707
+#define LUMA_PU(W, H) \1708
+ p.puLUMA_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1709
+ p.puLUMA_ ## W ## x ## H.addAvgNONALIGNED = addAvg_neon<W, H>; \1710
+ p.puLUMA_ ## W ## x ## H.addAvgALIGNED = addAvg_neon<W, H>; \1711
+ p.puLUMA_ ## W ## x ## H.sad = sad_pp_neon<W, H>; \1712
+ p.puLUMA_ ## W ## x ## H.sad_x3 = sad_x3_neon<W, H>; \1713
+ p.puLUMA_ ## W ## x ## H.sad_x4 = sad_x4_neon<W, H>; \1714
+ p.puLUMA_ ## W ## x ## H.pixelavg_ppNONALIGNED = pixelavg_pp_neon<W, H>; \1715
+ p.puLUMA_ ## W ## x ## H.pixelavg_ppALIGNED = pixelavg_pp_neon<W, H>;1716
+1717
+#if !(HIGH_BIT_DEPTH)1718
+#define LUMA_PU_S(W, H) \1719
+ p.puLUMA_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1720
+ p.puLUMA_ ## W ## x ## H.addAvgNONALIGNED = addAvg_neon<W, H>; \1721
+ p.puLUMA_ ## W ## x ## H.addAvgALIGNED = addAvg_neon<W, H>;1722
+#else // !(HIGH_BIT_DEPTH)1723
+#define LUMA_PU_S(W, H) \1724
+ p.puLUMA_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1725
+ p.puLUMA_ ## W ## x ## H.addAvgNONALIGNED = addAvg_neon<W, H>; \1726
+ p.puLUMA_ ## W ## x ## H.addAvgALIGNED = addAvg_neon<W, H>; \1727
+ p.puLUMA_ ## W ## x ## H.sad_x3 = sad_x3_neon<W, H>; \1728
+ p.puLUMA_ ## W ## x ## H.sad_x4 = sad_x4_neon<W, H>; \1729
+ p.puLUMA_ ## W ## x ## H.pixelavg_ppNONALIGNED = pixelavg_pp_neon<W, H>; \1730
+ p.puLUMA_ ## W ## x ## H.pixelavg_ppALIGNED = pixelavg_pp_neon<W, H>;1731
+#endif // !(HIGH_BIT_DEPTH)1732
+1733
+#define LUMA_CU(W, H) \1734
+ p.cuBLOCK_ ## W ## x ## H.sub_ps = pixel_sub_ps_neon<W, H>; \1735
+ p.cuBLOCK_ ## W ## x ## H.add_psNONALIGNED = pixel_add_ps_neon<W, H>; \1736
+ p.cuBLOCK_ ## W ## x ## H.add_psALIGNED = pixel_add_ps_neon<W, H>; \1737
+ p.cuBLOCK_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1738
+ p.cuBLOCK_ ## W ## x ## H.copy_ps = blockcopy_ps_neon<W, H>; \1739
+ p.cuBLOCK_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1740
+ p.cuBLOCK_ ## W ## x ## H.cpy2Dto1D_shl = cpy2Dto1D_shl_neon<W>; \1741
+ p.cuBLOCK_ ## W ## x ## H.cpy1Dto2D_shlNONALIGNED = cpy1Dto2D_shl_neon<W>; \1742
+ p.cuBLOCK_ ## W ## x ## H.cpy1Dto2D_shlALIGNED = cpy1Dto2D_shl_neon<W>; \1743
+ p.cuBLOCK_ ## W ## x ## H.psy_cost_pp = psyCost_pp_neon<BLOCK_ ## W ## x ## H>; \1744
+ p.cuBLOCK_ ## W ## x ## H.transpose = transpose_neon<W>;1745
+1746
+1747
+ LUMA_PU_S(4, 4);1748
+ LUMA_PU_S(8, 8);1749
+ LUMA_PU(16, 16);1750
+ LUMA_PU(32, 32);1751
+ LUMA_PU(64, 64);1752
+ LUMA_PU_S(4, 8);1753
+ LUMA_PU_S(8, 4);1754
+ LUMA_PU(16, 8);1755
+ LUMA_PU_S(8, 16);1756
+ LUMA_PU(16, 12);1757
+ LUMA_PU(12, 16);1758
+ LUMA_PU(16, 4);1759
+ LUMA_PU_S(4, 16);1760
+ LUMA_PU(32, 16);1761
+ LUMA_PU(16, 32);1762
+ LUMA_PU(32, 24);1763
+ LUMA_PU(24, 32);1764
+ LUMA_PU(32, 8);1765
+ LUMA_PU_S(8, 32);1766
+ LUMA_PU(64, 32);1767
+ LUMA_PU(32, 64);1768
+ LUMA_PU(64, 48);1769
+ LUMA_PU(48, 64);1770
+ LUMA_PU(64, 16);1771
+ LUMA_PU(16, 64);1772
+ 1773
+#if defined(__APPLE__)1774
+ p.puLUMA_4x4.sad = sad_pp_neon<4, 4>;1775
+ p.puLUMA_4x8.sad = sad_pp_neon<4, 8>;1776
+ p.puLUMA_4x16.sad = sad_pp_neon<4, 16>;1777
+#endif // defined(__APPLE__)1778
+ p.puLUMA_8x4.sad = sad_pp_neon<8, 4>;1779
+ p.puLUMA_8x8.sad = sad_pp_neon<8, 8>;1780
+ p.puLUMA_8x16.sad = sad_pp_neon<8, 16>;1781
+ p.puLUMA_8x32.sad = sad_pp_neon<8, 32>;1782
+1783
+#if !(HIGH_BIT_DEPTH)1784
+ p.puLUMA_4x4.sad_x3 = sad_x3_neon<4, 4>;1785
+ p.puLUMA_4x4.sad_x4 = sad_x4_neon<4, 4>;1786
+ p.puLUMA_4x8.sad_x3 = sad_x3_neon<4, 8>;1787
+ p.puLUMA_4x8.sad_x4 = sad_x4_neon<4, 8>;1788
+ p.puLUMA_4x16.sad_x3 = sad_x3_neon<4, 16>;1789
+ p.puLUMA_4x16.sad_x4 = sad_x4_neon<4, 16>;1790
+#endif // !(HIGH_BIT_DEPTH)1791
+1792
+ p.puLUMA_4x4.satd = pixel_satd_4x4_neon;1793
+ p.puLUMA_8x4.satd = pixel_satd_8x4_neon;1794
+ 1795
+ p.puLUMA_8x8.satd = satd8_neon<8, 8>;1796
+ p.puLUMA_16x16.satd = satd8_neon<16, 16>;1797
+ p.puLUMA_16x8.satd = satd8_neon<16, 8>;1798
+ p.puLUMA_8x16.satd = satd8_neon<8, 16>;1799
+ p.puLUMA_16x12.satd = satd8_neon<16, 12>;1800
+ p.puLUMA_16x4.satd = satd8_neon<16, 4>;1801
+ p.puLUMA_32x32.satd = satd8_neon<32, 32>;1802
+ p.puLUMA_32x16.satd = satd8_neon<32, 16>;1803
+ p.puLUMA_16x32.satd = satd8_neon<16, 32>;1804
+ p.puLUMA_32x24.satd = satd8_neon<32, 24>;1805
+ p.puLUMA_24x32.satd = satd8_neon<24, 32>;1806
+ p.puLUMA_32x8.satd = satd8_neon<32, 8>;1807
+ p.puLUMA_8x32.satd = satd8_neon<8, 32>;1808
+ p.puLUMA_64x64.satd = satd8_neon<64, 64>;1809
+ p.puLUMA_64x32.satd = satd8_neon<64, 32>;1810
+ p.puLUMA_32x64.satd = satd8_neon<32, 64>;1811
+ p.puLUMA_64x48.satd = satd8_neon<64, 48>;1812
+ p.puLUMA_48x64.satd = satd8_neon<48, 64>;1813
+ p.puLUMA_64x16.satd = satd8_neon<64, 16>;1814
+ p.puLUMA_16x64.satd = satd8_neon<16, 64>;1815
+1816
+#if HIGH_BIT_DEPTH1817
+ p.puLUMA_4x8.satd = satd4_neon<4, 8>;1818
+ p.puLUMA_4x16.satd = satd4_neon<4, 16>;1819
+#endif // HIGH_BIT_DEPTH1820
+1821
+#if !defined(__APPLE__) || HIGH_BIT_DEPTH1822
+ p.puLUMA_12x16.satd = satd4_neon<12, 16>;1823
+#endif // !defined(__APPLE__)1824
+1825
+1826
+ LUMA_CU(4, 4);1827
+ LUMA_CU(8, 8);1828
+ LUMA_CU(16, 16);1829
+ LUMA_CU(32, 32);1830
+ LUMA_CU(64, 64);1831
+ 1832
+#if !(HIGH_BIT_DEPTH)1833
+ p.cuBLOCK_8x8.var = pixel_var_neon<8>;1834
+ p.cuBLOCK_16x16.var = pixel_var_neon<16>;1835
+#if defined(__APPLE__)1836
+ p.cuBLOCK_32x32.var = pixel_var_neon<32>;1837
+ p.cuBLOCK_64x64.var = pixel_var_neon<64>;1838
+#endif // defined(__APPLE__)1839
+#endif // !(HIGH_BIT_DEPTH)1840
+1841
+ p.cuBLOCK_16x16.blockfill_sNONALIGNED = blockfill_s_neon<16>; 1842
+ p.cuBLOCK_16x16.blockfill_sALIGNED = blockfill_s_neon<16>;1843
+ p.cuBLOCK_32x32.blockfill_sNONALIGNED = blockfill_s_neon<32>; 1844
+ p.cuBLOCK_32x32.blockfill_sALIGNED = blockfill_s_neon<32>;1845
+ p.cuBLOCK_64x64.blockfill_sNONALIGNED = blockfill_s_neon<64>; 1846
+ p.cuBLOCK_64x64.blockfill_sALIGNED = blockfill_s_neon<64>;1847
+1848
+1849
+ p.cuBLOCK_4x4.calcresidualNONALIGNED = getResidual_neon<4>;1850
+ p.cuBLOCK_4x4.calcresidualALIGNED = getResidual_neon<4>;1851
+ p.cuBLOCK_8x8.calcresidualNONALIGNED = getResidual_neon<8>;1852
+ p.cuBLOCK_8x8.calcresidualALIGNED = getResidual_neon<8>;1853
+ p.cuBLOCK_16x16.calcresidualNONALIGNED = getResidual_neon<16>;1854
+ p.cuBLOCK_16x16.calcresidualALIGNED = getResidual_neon<16>;1855
+ 1856
+#if defined(__APPLE__)1857
+ p.cuBLOCK_32x32.calcresidualNONALIGNED = getResidual_neon<32>;1858
+ p.cuBLOCK_32x32.calcresidualALIGNED = getResidual_neon<32>;1859
+#endif // defined(__APPLE__)1860
+1861
+ p.cuBLOCK_4x4.sa8d = pixel_satd_4x4_neon;1862
+ p.cuBLOCK_8x8.sa8d = pixel_sa8d_8x8_neon;1863
+ p.cuBLOCK_16x16.sa8d = pixel_sa8d_16x16_neon;1864
+ p.cuBLOCK_32x32.sa8d = sa8d16<32, 32>;1865
+ p.cuBLOCK_64x64.sa8d = sa8d16<64, 64>;1866
+1867
+1868
+#define CHROMA_PU_420(W, H) \1869
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.addAvgNONALIGNED = addAvg_neon<W, H>; \1870
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.addAvgALIGNED = addAvg_neon<W, H>; \1871
+ p.chromaX265_CSP_I420.puCHROMA_420_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1872
+1873
+1874
+ CHROMA_PU_420(4, 4);1875
+ CHROMA_PU_420(8, 8);1876
+ CHROMA_PU_420(16, 16);1877
+ CHROMA_PU_420(32, 32);1878
+ CHROMA_PU_420(4, 2);1879
+ CHROMA_PU_420(8, 4);1880
+ CHROMA_PU_420(4, 8);1881
+ CHROMA_PU_420(8, 6);1882
+ CHROMA_PU_420(6, 8);1883
+ CHROMA_PU_420(8, 2);1884
+ CHROMA_PU_420(2, 8);1885
+ CHROMA_PU_420(16, 8);1886
+ CHROMA_PU_420(8, 16);1887
+ CHROMA_PU_420(16, 12);1888
+ CHROMA_PU_420(12, 16);1889
+ CHROMA_PU_420(16, 4);1890
+ CHROMA_PU_420(4, 16);1891
+ CHROMA_PU_420(32, 16);1892
+ CHROMA_PU_420(16, 32);1893
+ CHROMA_PU_420(32, 24);1894
+ CHROMA_PU_420(24, 32);1895
+ CHROMA_PU_420(32, 8);1896
+ CHROMA_PU_420(8, 32);1897
+1898
+1899
+1900
+ p.chromaX265_CSP_I420.puCHROMA_420_2x2.satd = NULL;1901
+ p.chromaX265_CSP_I420.puCHROMA_420_4x4.satd = pixel_satd_4x4_neon;1902
+ p.chromaX265_CSP_I420.puCHROMA_420_8x8.satd = satd8_neon<8, 8>;1903
+ p.chromaX265_CSP_I420.puCHROMA_420_16x16.satd = satd8_neon<16, 16>;1904
+ p.chromaX265_CSP_I420.puCHROMA_420_32x32.satd = satd8_neon<32, 32>;1905
+1906
+ p.chromaX265_CSP_I420.puCHROMA_420_4x2.satd = NULL;1907
+ p.chromaX265_CSP_I420.puCHROMA_420_2x4.satd = NULL;1908
+ p.chromaX265_CSP_I420.puCHROMA_420_8x4.satd = pixel_satd_8x4_neon;1909
+ p.chromaX265_CSP_I420.puCHROMA_420_16x8.satd = satd8_neon<16, 8>;1910
+ p.chromaX265_CSP_I420.puCHROMA_420_8x16.satd = satd8_neon<8, 16>;1911
+ p.chromaX265_CSP_I420.puCHROMA_420_32x16.satd = satd8_neon<32, 16>;1912
+ p.chromaX265_CSP_I420.puCHROMA_420_16x32.satd = satd8_neon<16, 32>;1913
+1914
+ p.chromaX265_CSP_I420.puCHROMA_420_8x6.satd = NULL;1915
+ p.chromaX265_CSP_I420.puCHROMA_420_6x8.satd = NULL;1916
+ p.chromaX265_CSP_I420.puCHROMA_420_8x2.satd = NULL;1917
+ p.chromaX265_CSP_I420.puCHROMA_420_2x8.satd = NULL;1918
+ p.chromaX265_CSP_I420.puCHROMA_420_16x12.satd = satd4_neon<16, 12>;1919
+ p.chromaX265_CSP_I420.puCHROMA_420_16x4.satd = satd4_neon<16, 4>;1920
+ p.chromaX265_CSP_I420.puCHROMA_420_32x24.satd = satd8_neon<32, 24>;1921
+ p.chromaX265_CSP_I420.puCHROMA_420_24x32.satd = satd8_neon<24, 32>;1922
+ p.chromaX265_CSP_I420.puCHROMA_420_32x8.satd = satd8_neon<32, 8>;1923
+ p.chromaX265_CSP_I420.puCHROMA_420_8x32.satd = satd8_neon<8, 32>;1924
+ 1925
+#if HIGH_BIT_DEPTH1926
+ p.chromaX265_CSP_I420.puCHROMA_420_4x8.satd = satd4_neon<4, 8>;1927
+ p.chromaX265_CSP_I420.puCHROMA_420_4x16.satd = satd4_neon<4, 16>;1928
+#endif // HIGH_BIT_DEPTH1929
+1930
+#if !defined(__APPLE__) || HIGH_BIT_DEPTH1931
+ p.chromaX265_CSP_I420.puCHROMA_420_12x16.satd = satd4_neon<12, 16>;1932
+#endif // !defined(__APPLE__)1933
+1934
+1935
+#define CHROMA_CU_420(W, H) \1936
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.sse_pp = sse_neon<W, H, pixel, pixel>; \1937
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1938
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.copy_ps = blockcopy_ps_neon<W, H>; \1939
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.sub_ps = pixel_sub_ps_neon<W, H>; \1940
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.add_psNONALIGNED = pixel_add_ps_neon<W, H>; \1941
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.add_psALIGNED = pixel_add_ps_neon<W, H>;1942
+ 1943
+#define CHROMA_CU_S_420(W, H) \1944
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1945
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.copy_ps = blockcopy_ps_neon<W, H>; \1946
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.sub_ps = pixel_sub_ps_neon<W, H>; \1947
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.add_psNONALIGNED = pixel_add_ps_neon<W, H>; \1948
+ p.chromaX265_CSP_I420.cuBLOCK_420_ ## W ## x ## H.add_psALIGNED = pixel_add_ps_neon<W, H>;1949
+1950
+1951
+ CHROMA_CU_S_420(4, 4)1952
+ CHROMA_CU_420(8, 8)1953
+ CHROMA_CU_420(16, 16)1954
+ CHROMA_CU_420(32, 32)1955
+1956
+1957
+ p.chromaX265_CSP_I420.cuBLOCK_8x8.sa8d = p.chromaX265_CSP_I420.puCHROMA_420_4x4.satd;1958
+ p.chromaX265_CSP_I420.cuBLOCK_16x16.sa8d = sa8d8<8, 8>;1959
+ p.chromaX265_CSP_I420.cuBLOCK_32x32.sa8d = sa8d16<16, 16>;1960
+ p.chromaX265_CSP_I420.cuBLOCK_64x64.sa8d = sa8d16<32, 32>;1961
+1962
+1963
+#define CHROMA_PU_422(W, H) \1964
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.addAvgNONALIGNED = addAvg_neon<W, H>; \1965
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.addAvgALIGNED = addAvg_neon<W, H>; \1966
+ p.chromaX265_CSP_I422.puCHROMA_422_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \1967
+1968
+1969
+ CHROMA_PU_422(4, 8);1970
+ CHROMA_PU_422(8, 16);1971
+ CHROMA_PU_422(16, 32);1972
+ CHROMA_PU_422(32, 64);1973
+ CHROMA_PU_422(4, 4);1974
+ CHROMA_PU_422(2, 8);1975
+ CHROMA_PU_422(8, 8);1976
+ CHROMA_PU_422(4, 16);1977
+ CHROMA_PU_422(8, 12);1978
+ CHROMA_PU_422(6, 16);1979
+ CHROMA_PU_422(8, 4);1980
+ CHROMA_PU_422(2, 16);1981
+ CHROMA_PU_422(16, 16);1982
+ CHROMA_PU_422(8, 32);1983
+ CHROMA_PU_422(16, 24);1984
+ CHROMA_PU_422(12, 32);1985
+ CHROMA_PU_422(16, 8);1986
+ CHROMA_PU_422(4, 32);1987
+ CHROMA_PU_422(32, 32);1988
+ CHROMA_PU_422(16, 64);1989
+ CHROMA_PU_422(32, 48);1990
+ CHROMA_PU_422(24, 64);1991
+ CHROMA_PU_422(32, 16);1992
+ CHROMA_PU_422(8, 64);1993
+1994
+1995
+ p.chromaX265_CSP_I422.puCHROMA_422_2x4.satd = NULL;1996
+ p.chromaX265_CSP_I422.puCHROMA_422_8x16.satd = satd8_neon<8, 16>;1997
+ p.chromaX265_CSP_I422.puCHROMA_422_16x32.satd = satd8_neon<16, 32>;1998
+ p.chromaX265_CSP_I422.puCHROMA_422_32x64.satd = satd8_neon<32, 64>;1999
+ p.chromaX265_CSP_I422.puCHROMA_422_4x4.satd = pixel_satd_4x4_neon;2000
+ p.chromaX265_CSP_I422.puCHROMA_422_2x8.satd = NULL;2001
+ p.chromaX265_CSP_I422.puCHROMA_422_8x8.satd = satd8_neon<8, 8>;2002
+ p.chromaX265_CSP_I422.puCHROMA_422_16x16.satd = satd8_neon<16, 16>;2003
+ p.chromaX265_CSP_I422.puCHROMA_422_8x32.satd = satd8_neon<8, 32>;2004
+ p.chromaX265_CSP_I422.puCHROMA_422_32x32.satd = satd8_neon<32, 32>;2005
+ p.chromaX265_CSP_I422.puCHROMA_422_16x64.satd = satd8_neon<16, 64>;2006
+ p.chromaX265_CSP_I422.puCHROMA_422_6x16.satd = NULL;2007
+ p.chromaX265_CSP_I422.puCHROMA_422_8x4.satd = satd4_neon<8, 4>;2008
+ p.chromaX265_CSP_I422.puCHROMA_422_2x16.satd = NULL;2009
+ p.chromaX265_CSP_I422.puCHROMA_422_16x8.satd = satd8_neon<16, 8>;2010
+ p.chromaX265_CSP_I422.puCHROMA_422_32x16.satd = satd8_neon<32, 16>;2011
+ 2012
+ p.chromaX265_CSP_I422.puCHROMA_422_8x12.satd = satd4_neon<8, 12>;2013
+ p.chromaX265_CSP_I422.puCHROMA_422_8x64.satd = satd8_neon<8, 64>;2014
+ p.chromaX265_CSP_I422.puCHROMA_422_12x32.satd = satd4_neon<12, 32>;2015
+ p.chromaX265_CSP_I422.puCHROMA_422_16x24.satd = satd8_neon<16, 24>;2016
+ p.chromaX265_CSP_I422.puCHROMA_422_24x64.satd = satd8_neon<24, 64>;2017
+ p.chromaX265_CSP_I422.puCHROMA_422_32x48.satd = satd8_neon<32, 48>;2018
+2019
+#if HIGH_BIT_DEPTH2020
+ p.chromaX265_CSP_I422.puCHROMA_422_4x8.satd = satd4_neon<4, 8>;2021
+ p.chromaX265_CSP_I422.puCHROMA_422_4x16.satd = satd4_neon<4, 16>;2022
+ p.chromaX265_CSP_I422.puCHROMA_422_4x32.satd = satd4_neon<4, 32>;2023
+#endif // HIGH_BIT_DEPTH2024
+2025
+2026
+#define CHROMA_CU_422(W, H) \2027
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.sse_pp = sse_neon<W, H, pixel, pixel>; \2028
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \2029
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.copy_ps = blockcopy_ps_neon<W, H>; \2030
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.sub_ps = pixel_sub_ps_neon<W, H>; \2031
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.add_psNONALIGNED = pixel_add_ps_neon<W, H>; \2032
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.add_psALIGNED = pixel_add_ps_neon<W, H>;2033
+2034
+#define CHROMA_CU_S_422(W, H) \2035
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.copy_pp = blockcopy_pp_neon<W, H>; \2036
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.copy_ps = blockcopy_ps_neon<W, H>; \2037
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.sub_ps = pixel_sub_ps_neon<W, H>; \2038
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.add_psNONALIGNED = pixel_add_ps_neon<W, H>; \2039
+ p.chromaX265_CSP_I422.cuBLOCK_422_ ## W ## x ## H.add_psALIGNED = pixel_add_ps_neon<W, H>;2040
+ 2041
+ 2042
+ CHROMA_CU_S_422(4, 8)2043
+ CHROMA_CU_422(8, 16)2044
+ CHROMA_CU_422(16, 32)2045
+ CHROMA_CU_422(32, 64)2046
+2047
+ p.chromaX265_CSP_I422.cuBLOCK_8x8.sa8d = p.chromaX265_CSP_I422.puCHROMA_422_4x8.satd;2048
+ p.chromaX265_CSP_I422.cuBLOCK_16x16.sa8d = sa8d8<8, 16>;2049
+ p.chromaX265_CSP_I422.cuBLOCK_32x32.sa8d = sa8d16<16, 32>;2050
+ p.chromaX265_CSP_I422.cuBLOCK_64x64.sa8d = sa8d16<32, 64>;2051
+2052
+2053
+}2054
+2055
+2056
+}2057
+2058
+2059
+#endif2060
+2061
x265_3.6.tar.gz/source/common/aarch64/pixel-prim.h
Added
25
1
2
+#ifndef PIXEL_PRIM_NEON_H__3
+#define PIXEL_PRIM_NEON_H__4
+5
+#include "common.h"6
+#include "slicetype.h" // LOWRES_COST_MASK7
+#include "primitives.h"8
+#include "x265.h"9
+10
+11
+12
+namespace X265_NS13
+{14
+15
+16
+17
+void setupPixelPrimitives_neon(EncoderPrimitives &p);18
+19
+20
+}21
+22
+23
+#endif24
+25
x265_3.6.tar.gz/source/common/aarch64/pixel-util-common.S
Added
86
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+.arch armv8-a29
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.macro pixel_var_start39
+ movi v0.16b, #040
+ movi v1.16b, #041
+ movi v2.16b, #042
+ movi v3.16b, #043
+.endm44
+45
+.macro pixel_var_1 v46
+ uaddw v0.8h, v0.8h, \v\().8b47
+ umull v30.8h, \v\().8b, \v\().8b48
+ uaddw2 v1.8h, v1.8h, \v\().16b49
+ umull2 v31.8h, \v\().16b, \v\().16b50
+ uadalp v2.4s, v30.8h51
+ uadalp v3.4s, v31.8h52
+.endm53
+54
+.macro pixel_var_end55
+ uaddlv s0, v0.8h56
+ uaddlv s1, v1.8h57
+ add v2.4s, v2.4s, v3.4s58
+ fadd s0, s0, s159
+ uaddlv d2, v2.4s60
+ fmov w0, s061
+ fmov x2, d262
+ orr x0, x0, x2, lsl #3263
+.endm64
+65
+.macro ssimDist_start66
+ movi v0.16b, #067
+ movi v1.16b, #068
+.endm69
+70
+.macro ssimDist_end71
+ uaddlv d0, v0.4s72
+ uaddlv d1, v1.4s73
+ str d0, x674
+ str d1, x475
+.endm76
+77
+.macro normFact_start78
+ movi v0.16b, #079
+.endm80
+81
+.macro normFact_end82
+ uaddlv d0, v0.4s83
+ str d0, x384
+.endm85
+86
x265_3.6.tar.gz/source/common/aarch64/pixel-util-sve.S
Added
375
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "pixel-util-common.S"27
+28
+.arch armv8-a+sve29
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+function PFX(pixel_sub_ps_8x16_sve)41
+ lsl x1, x1, #142
+ ptrue p0.h, vl843
+.rept 844
+ ld1b {z0.h}, p0/z, x245
+ ld1b {z1.h}, p0/z, x346
+ add x2, x2, x447
+ add x3, x3, x548
+ ld1b {z2.h}, p0/z, x249
+ ld1b {z3.h}, p0/z, x350
+ add x2, x2, x451
+ add x3, x3, x552
+ sub z4.h, z0.h, z1.h53
+ sub z5.h, z2.h, z3.h54
+ st1 {v4.8h}, x0, x155
+ st1 {v5.8h}, x0, x156
+.endr57
+ ret58
+endfunc59
+60
+//******* satd *******61
+.macro satd_4x4_sve62
+ ld1b {z0.h}, p0/z, x063
+ ld1b {z2.h}, p0/z, x264
+ add x0, x0, x165
+ add x2, x2, x366
+ ld1b {z1.h}, p0/z, x067
+ ld1b {z3.h}, p0/z, x268
+ add x0, x0, x169
+ add x2, x2, x370
+ ld1b {z4.h}, p0/z, x071
+ ld1b {z6.h}, p0/z, x272
+ add x0, x0, x173
+ add x2, x2, x374
+ ld1b {z5.h}, p0/z, x075
+ ld1b {z7.h}, p0/z, x276
+ add x0, x0, x177
+ add x2, x2, x378
+79
+ sub z0.h, z0.h, z2.h80
+ sub z1.h, z1.h, z3.h81
+ sub z2.h, z4.h, z6.h82
+ sub z3.h, z5.h, z7.h83
+84
+ add z4.h, z0.h, z2.h85
+ add z5.h, z1.h, z3.h86
+ sub z6.h, z0.h, z2.h87
+ sub z7.h, z1.h, z3.h88
+89
+ add z0.h, z4.h, z5.h90
+ sub z1.h, z4.h, z5.h91
+92
+ add z2.h, z6.h, z7.h93
+ sub z3.h, z6.h, z7.h94
+95
+ trn1 z4.h, z0.h, z2.h96
+ trn2 z5.h, z0.h, z2.h97
+98
+ trn1 z6.h, z1.h, z3.h99
+ trn2 z7.h, z1.h, z3.h100
+101
+ add z0.h, z4.h, z5.h102
+ sub z1.h, z4.h, z5.h103
+104
+ add z2.h, z6.h, z7.h105
+ sub z3.h, z6.h, z7.h106
+107
+ trn1 z4.s, z0.s, z1.s108
+ trn2 z5.s, z0.s, z1.s109
+110
+ trn1 z6.s, z2.s, z3.s111
+ trn2 z7.s, z2.s, z3.s112
+113
+ abs z4.h, p0/m, z4.h114
+ abs z5.h, p0/m, z5.h115
+ abs z6.h, p0/m, z6.h116
+ abs z7.h, p0/m, z7.h117
+118
+ smax z4.h, p0/m, z4.h, z5.h119
+ smax z6.h, p0/m, z6.h, z7.h120
+121
+ add z0.h, z4.h, z6.h122
+123
+ uaddlp v0.2s, v0.4h124
+ uaddlp v0.1d, v0.2s125
+.endm126
+127
+// int satd_4x4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)128
+function PFX(pixel_satd_4x4_sve)129
+ ptrue p0.h, vl4130
+ satd_4x4_sve131
+ fmov x0, d0132
+ ret133
+endfunc134
+135
+function PFX(pixel_satd_8x4_sve)136
+ ptrue p0.h, vl4137
+ mov x4, x0138
+ mov x5, x2139
+ satd_4x4_sve140
+ add x0, x4, #4141
+ add x2, x5, #4142
+ umov x6, v0.d0143
+ satd_4x4_sve144
+ umov x0, v0.d0145
+ add x0, x0, x6146
+ ret147
+endfunc148
+149
+function PFX(pixel_satd_8x12_sve)150
+ ptrue p0.h, vl4151
+ mov x4, x0152
+ mov x5, x2153
+ mov x7, #0154
+ satd_4x4_sve155
+ umov x6, v0.d0156
+ add x7, x7, x6157
+ add x0, x4, #4158
+ add x2, x5, #4159
+ satd_4x4_sve160
+ umov x6, v0.d0161
+ add x7, x7, x6162
+.rept 2163
+ sub x0, x0, #4164
+ sub x2, x2, #4165
+ mov x4, x0166
+ mov x5, x2167
+ satd_4x4_sve168
+ umov x6, v0.d0169
+ add x7, x7, x6170
+ add x0, x4, #4171
+ add x2, x5, #4172
+ satd_4x4_sve173
+ umov x6, v0.d0174
+ add x7, x7, x6175
+.endr176
+ mov x0, x7177
+ ret178
+endfunc179
+180
+.macro LOAD_DIFF_16x4_sve v0 v1 v2 v3 v4 v5 v6 v7181
+ mov x11, #8 // in order to consider CPUs whose vector size is greater than 128 bits182
+ ld1b {z0.h}, p0/z, x0183
+ ld1b {z1.h}, p0/z, x0, x11184
+ ld1b {z2.h}, p0/z, x2185
+ ld1b {z3.h}, p0/z, x2, x11186
+ add x0, x0, x1187
+ add x2, x2, x3188
+ ld1b {z4.h}, p0/z, x0189
+ ld1b {z5.h}, p0/z, x0, x11190
+ ld1b {z6.h}, p0/z, x2191
+ ld1b {z7.h}, p0/z, x2, x11192
+ add x0, x0, x1193
+ add x2, x2, x3194
+ ld1b {z29.h}, p0/z, x0195
+ ld1b {z9.h}, p0/z, x0, x11196
+ ld1b {z10.h}, p0/z, x2197
+ ld1b {z11.h}, p0/z, x2, x11198
+ add x0, x0, x1199
+ add x2, x2, x3200
+ ld1b {z12.h}, p0/z, x0201
+ ld1b {z13.h}, p0/z, x0, x11202
+ ld1b {z14.h}, p0/z, x2203
+ ld1b {z15.h}, p0/z, x2, x11204
+ add x0, x0, x1205
+ add x2, x2, x3206
+207
+ sub \v0\().h, z0.h, z2.h208
+ sub \v4\().h, z1.h, z3.h209
+ sub \v1\().h, z4.h, z6.h210
+ sub \v5\().h, z5.h, z7.h211
+ sub \v2\().h, z29.h, z10.h212
+ sub \v6\().h, z9.h, z11.h213
+ sub \v3\().h, z12.h, z14.h214
+ sub \v7\().h, z13.h, z15.h215
+.endm216
+217
+// one vertical hadamard pass and two horizontal218
+function PFX(satd_8x4v_8x8h_sve), export=0219
+ HADAMARD4_V z16.h, z18.h, z17.h, z19.h, z0.h, z2.h, z1.h, z3.h220
+ HADAMARD4_V z20.h, z21.h, z22.h, z23.h, z0.h, z1.h, z2.h, z3.h221
+ trn4 z0.h, z1.h, z2.h, z3.h, z16.h, z17.h, z18.h, z19.h222
+ trn4 z4.h, z5.h, z6.h, z7.h, z20.h, z21.h, z22.h, z23.h223
+ SUMSUB_ABCD z16.h, z17.h, z18.h, z19.h, z0.h, z1.h, z2.h, z3.h224
+ SUMSUB_ABCD z20.h, z21.h, z22.h, z23.h, z4.h, z5.h, z6.h, z7.h225
+ trn4 z0.s, z2.s, z1.s, z3.s, z16.s, z18.s, z17.s, z19.s226
+ trn4 z4.s, z6.s, z5.s, z7.s, z20.s, z22.s, z21.s, z23.s227
+ ABS8_SVE z0.h, z1.h, z2.h, z3.h, z4.h, z5.h, z6.h, z7.h, p0228
+ smax z0.h, p0/m, z0.h, z2.h229
+ smax z1.h, p0/m, z1.h, z3.h230
+ smax z4.h, p0/m, z4.h, z6.h231
+ smax z5.h, p0/m, z5.h, z7.h232
+ ret233
+endfunc234
+235
+function PFX(satd_16x4_sve), export=0236
+ LOAD_DIFF_16x4_sve z16, z17, z18, z19, z20, z21, z22, z23237
+ b PFX(satd_8x4v_8x8h_sve)238
+endfunc239
+240
+.macro pixel_satd_32x8_sve241
+ mov x4, x0242
+ mov x5, x2243
+.rept 2244
+ bl PFX(satd_16x4_sve)245
+ add z30.h, z30.h, z0.h246
+ add z31.h, z31.h, z1.h247
+ add z30.h, z30.h, z4.h248
+ add z31.h, z31.h, z5.h249
+.endr250
+ add x0, x4, #16251
+ add x2, x5, #16252
+.rept 2253
+ bl PFX(satd_16x4_sve)254
+ add z30.h, z30.h, z0.h255
+ add z31.h, z31.h, z1.h256
+ add z30.h, z30.h, z4.h257
+ add z31.h, z31.h, z5.h258
+.endr259
+.endm260
+261
+.macro satd_32x16_sve262
+ movi v30.2d, #0263
+ movi v31.2d, #0264
+ pixel_satd_32x8_sve265
+ sub x0, x0, #16266
+ sub x2, x2, #16267
+ pixel_satd_32x8_sve268
+ add z0.h, z30.h, z31.h269
+ uaddlv s0, v0.8h270
+ mov w6, v0.s0271
+.endm272
+273
+function PFX(pixel_satd_32x16_sve)274
+ ptrue p0.h, vl8275
+ mov x10, x30276
+ satd_32x16_sve277
+ mov x0, x6278
+ ret x10279
+endfunc280
+281
+function PFX(pixel_satd_32x32_sve)282
+ ptrue p0.h, vl8283
+ mov x10, x30284
+ mov x7, #0285
+ satd_32x16_sve286
+ sub x0, x0, #16287
+ sub x2, x2, #16288
+ add x7, x7, x6289
+ satd_32x16_sve290
+ add x0, x7, x6291
+ ret x10292
+endfunc293
+294
+.macro satd_64x16_sve295
+ mov x8, x0296
+ mov x9, x2297
+ satd_32x16_sve298
+ add x7, x7, x6299
+ add x0, x8, #32300
+ add x2, x9, #32301
+ satd_32x16_sve302
+ add x7, x7, x6303
+.endm304
+305
+function PFX(pixel_satd_64x48_sve)306
+ ptrue p0.h, vl8307
+ mov x10, x30308
+ mov x7, #0309
+.rept 2310
+ satd_64x16_sve311
+ sub x0, x0, #48312
+ sub x2, x2, #48313
+.endr314
+ satd_64x16_sve315
+ mov x0, x7316
+ ret x10317
+endfunc318
+319
+/********* ssim ***********/320
+// uint32_t quant_c(const int16_t* coef, const int32_t* quantCoeff, int32_t* deltaU, int16_t* qCoef, int qBits, int add, int numCoeff)321
+// No need to fully use sve instructions for this function322
+function PFX(quant_sve)323
+ mov w9, #1324
+ lsl w9, w9, w4325
+ mov z0.s, w9326
+ neg w9, w4327
+ mov z1.s, w9328
+ add w9, w9, #8329
+ mov z2.s, w9330
+ mov z3.s, w5331
+332
+ lsr w6, w6, #2333
+ eor z4.d, z4.d, z4.d334
+ eor w10, w10, w10335
+ eor z17.d, z17.d, z17.d336
+337
+.loop_quant_sve:338
+ ld1 {v18.4h}, x0, #8339
+ ld1 {v7.4s}, x1, #16340
+ sxtl v6.4s, v18.4h341
+342
+ cmlt v5.4s, v6.4s, #0343
+344
+ abs v6.4s, v6.4s345
+346
+347
+ mul v6.4s, v6.4s, v7.4s348
+349
+ add v7.4s, v6.4s, v3.4s350
+ sshl v7.4s, v7.4s, v1.4s351
+352
+ mls v6.4s, v7.4s, v0.s0353
+ sshl v16.4s, v6.4s, v2.4s354
+ st1 {v16.4s}, x2, #16355
+356
+ // numsig357
+ cmeq v16.4s, v7.4s, v17.4s358
+ add v4.4s, v4.4s, v16.4s359
+ add w10, w10, #4360
+361
+ // level *= sign362
+ eor z16.d, z7.d, z5.d363
+ sub v16.4s, v16.4s, v5.4s364
+ sqxtn v5.4h, v16.4s365
+ st1 {v5.4h}, x3, #8366
+367
+ subs w6, w6, #1368
+ b.ne .loop_quant_sve369
+370
+ addv s4, v4.4s371
+ mov w9, v4.s0372
+ add w0, w10, w9373
+ ret374
+endfunc375
x265_3.6.tar.gz/source/common/aarch64/pixel-util-sve2.S
Added
1688
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "pixel-util-common.S"27
+28
+.arch armv8-a+sve229
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+// uint64_t pixel_var(const pixel* pix, intptr_t i_stride)41
+function PFX(pixel_var_8x8_sve2)42
+ ptrue p0.h, vl843
+ ld1b {z0.h}, p0/z, x044
+ add x0, x0, x145
+ mul z31.h, z0.h, z0.h46
+ uaddlp v1.4s, v31.8h47
+.rept 748
+ ld1b {z4.h}, p0/z, x049
+ add x0, x0, x150
+ add z0.h, z0.h, z4.h51
+ mul z31.h, z4.h, z4.h52
+ uadalp z1.s, p0/m, z31.h53
+.endr54
+ uaddlv s0, v0.8h55
+ uaddlv d1, v1.4s56
+ fmov w0, s057
+ fmov x1, d158
+ orr x0, x0, x1, lsl #3259
+ ret60
+endfunc61
+62
+function PFX(pixel_var_16x16_sve2)63
+ rdvl x9, #164
+ cmp x9, #1665
+ bgt .vl_gt_16_pixel_var_16x1666
+ pixel_var_start67
+ mov w12, #1668
+.loop_var_16_sve2:69
+ sub w12, w12, #170
+ ld1 {v4.16b}, x0, x171
+ pixel_var_1 v472
+ cbnz w12, .loop_var_16_sve273
+ pixel_var_end74
+ ret75
+.vl_gt_16_pixel_var_16x16:76
+ ptrue p0.h, vl1677
+ mov z0.d, #078
+.rept 1679
+ ld1b {z4.h}, p0/z, x080
+ add x0, x0, x181
+ add z0.h, z0.h, z4.h82
+ mul z30.h, z4.h, z4.h83
+ uadalp z1.s, p0/m, z30.h84
+.endr85
+ uaddv d0, p0, z0.h86
+ uaddv d1, p0, z1.s87
+ fmov w0, s088
+ fmov x1, d189
+ orr x0, x0, x1, lsl #3290
+ ret91
+endfunc92
+93
+function PFX(pixel_var_32x32_sve2)94
+ rdvl x9, #195
+ cmp x9, #1696
+ bgt .vl_gt_16_pixel_var_32x3297
+ pixel_var_start98
+ mov w12, #3299
+.loop_var_32_sve2:100
+ sub w12, w12, #1101
+ ld1 {v4.16b-v5.16b}, x0, x1102
+ pixel_var_1 v4103
+ pixel_var_1 v5104
+ cbnz w12, .loop_var_32_sve2105
+ pixel_var_end106
+ ret107
+.vl_gt_16_pixel_var_32x32:108
+ cmp x9, #48109
+ bgt .vl_gt_48_pixel_var_32x32110
+ ptrue p0.b, vl32111
+ mov z0.d, #0112
+ mov z1.d, #0113
+.rept 32114
+ ld1b {z4.b}, p0/z, x0115
+ add x0, x0, x1116
+ uaddwb z0.h, z0.h, z4.b117
+ uaddwt z0.h, z0.h, z4.b118
+ umullb z28.h, z4.b, z4.b119
+ umullt z29.h, z4.b, z4.b120
+ uadalp z1.s, p0/m, z28.h121
+ uadalp z1.s, p0/m, z29.h122
+.endr123
+ uaddv d0, p0, z0.h124
+ uaddv d1, p0, z1.s125
+ fmov w0, s0126
+ fmov x1, d1127
+ orr x0, x0, x1, lsl #32128
+ ret129
+.vl_gt_48_pixel_var_32x32:130
+ ptrue p0.h, vl32131
+ mov z0.d, #0132
+ mov z1.d, #0133
+.rept 32134
+ ld1b {z4.h}, p0/z, x0135
+ add x0, x0, x1136
+ add z0.h, z0.h, z4.h137
+ mul z28.h, z4.h, z4.h138
+ uadalp z1.s, p0/m, z28.h139
+.endr140
+ uaddv d0, p0, z0.h141
+ uaddv d1, p0, z1.s142
+ fmov w0, s0143
+ fmov x1, d1144
+ orr x0, x0, x1, lsl #32145
+ ret146
+endfunc147
+148
+function PFX(pixel_var_64x64_sve2)149
+ rdvl x9, #1150
+ cmp x9, #16151
+ bgt .vl_gt_16_pixel_var_64x64152
+ pixel_var_start153
+ mov w12, #64154
+.loop_var_64_sve2:155
+ sub w12, w12, #1156
+ ld1 {v4.16b-v7.16b}, x0, x1157
+ pixel_var_1 v4158
+ pixel_var_1 v5159
+ pixel_var_1 v6160
+ pixel_var_1 v7161
+ cbnz w12, .loop_var_64_sve2162
+ pixel_var_end163
+ ret164
+.vl_gt_16_pixel_var_64x64:165
+ cmp x9, #48166
+ bgt .vl_gt_48_pixel_var_64x64167
+ ptrue p0.b, vl32168
+ mov z0.d, #0169
+ mov z2.d, #0170
+.rept 64171
+ ld1b {z4.b}, p0/z, x0172
+ ld1b {z5.b}, p0/z, x0, #1, mul vl173
+ add x0, x0, x1174
+ uaddwb z0.h, z0.h, z4.b175
+ uaddwt z0.h, z0.h, z4.b176
+ uaddwb z0.h, z0.h, z5.b177
+ uaddwt z0.h, z0.h, z5.b178
+ umullb z24.h, z4.b, z4.b179
+ umullt z25.h, z4.b, z4.b180
+ umullb z26.h, z5.b, z5.b181
+ umullt z27.h, z5.b, z5.b182
+ uadalp z2.s, p0/m, z24.h183
+ uadalp z2.s, p0/m, z25.h184
+ uadalp z2.s, p0/m, z26.h185
+ uadalp z2.s, p0/m, z27.h186
+.endr187
+ uaddv d0, p0, z0.h188
+ uaddv d1, p0, z2.s189
+ fmov w0, s0190
+ fmov x1, d1191
+ orr x0, x0, x1, lsl #32192
+ ret193
+.vl_gt_48_pixel_var_64x64:194
+ cmp x9, #112195
+ bgt .vl_gt_112_pixel_var_64x64196
+ ptrue p0.b, vl64197
+ mov z0.d, #0198
+ mov z1.d, #0199
+.rept 64200
+ ld1b {z4.b}, p0/z, x0201
+ add x0, x0, x1202
+ uaddwb z0.h, z0.h, z4.b203
+ uaddwt z0.h, z0.h, z4.b204
+ umullb z24.h, z4.b, z4.b205
+ umullt z25.h, z4.b, z4.b206
+ uadalp z2.s, p0/m, z24.h207
+ uadalp z2.s, p0/m, z25.h208
+.endr209
+ uaddv d0, p0, z0.h210
+ uaddv d1, p0, z2.s211
+ fmov w0, s0212
+ fmov x1, d1213
+ orr x0, x0, x1, lsl #32214
+ ret215
+.vl_gt_112_pixel_var_64x64:216
+ ptrue p0.h, vl64217
+ mov z0.d, #0218
+ mov z1.d, #0219
+.rept 64220
+ ld1b {z4.h}, p0/z, x0221
+ add x0, x0, x1222
+ add z0.h, z0.h, z4.h223
+ mul z24.h, z4.h, z4.h224
+ uadalp z1.s, p0/m, z24.h225
+.endr226
+ uaddv d0, p0, z0.h227
+ uaddv d1, p0, z1.s228
+ fmov w0, s0229
+ fmov x1, d1230
+ orr x0, x0, x1, lsl #32231
+ ret232
+endfunc233
+234
+function PFX(getResidual16_sve2)235
+ rdvl x9, #1236
+ cmp x9, #16237
+ bgt .vl_gt_16_getResidual16238
+ lsl x4, x3, #1239
+.rept 8240
+ ld1 {v0.16b}, x0, x3241
+ ld1 {v1.16b}, x1, x3242
+ ld1 {v2.16b}, x0, x3243
+ ld1 {v3.16b}, x1, x3244
+ usubl v4.8h, v0.8b, v1.8b245
+ usubl2 v5.8h, v0.16b, v1.16b246
+ usubl v6.8h, v2.8b, v3.8b247
+ usubl2 v7.8h, v2.16b, v3.16b248
+ st1 {v4.8h-v5.8h}, x2, x4249
+ st1 {v6.8h-v7.8h}, x2, x4250
+.endr251
+ ret252
+.vl_gt_16_getResidual16:253
+ ptrue p0.h, vl16254
+.rept 16255
+ ld1b {z0.h}, p0/z, x0256
+ ld1b {z2.h}, p0/z, x1257
+ add x0, x0, x3258
+ add x1, x1, x3259
+ sub z4.h, z0.h, z2.h260
+ st1h {z4.h}, p0, x2261
+ add x2, x2, x3, lsl #1262
+.endr263
+ ret264
+endfunc265
+266
+function PFX(getResidual32_sve2)267
+ rdvl x9, #1268
+ cmp x9, #16269
+ bgt .vl_gt_16_getResidual32270
+ lsl x4, x3, #1271
+ mov w12, #4272
+.loop_residual_32:273
+ sub w12, w12, #1274
+.rept 4275
+ ld1 {v0.16b-v1.16b}, x0, x3276
+ ld1 {v2.16b-v3.16b}, x1, x3277
+ ld1 {v4.16b-v5.16b}, x0, x3278
+ ld1 {v6.16b-v7.16b}, x1, x3279
+ usubl v16.8h, v0.8b, v2.8b280
+ usubl2 v17.8h, v0.16b, v2.16b281
+ usubl v18.8h, v1.8b, v3.8b282
+ usubl2 v19.8h, v1.16b, v3.16b283
+ usubl v20.8h, v4.8b, v6.8b284
+ usubl2 v21.8h, v4.16b, v6.16b285
+ usubl v22.8h, v5.8b, v7.8b286
+ usubl2 v23.8h, v5.16b, v7.16b287
+ st1 {v16.8h-v19.8h}, x2, x4288
+ st1 {v20.8h-v23.8h}, x2, x4289
+.endr290
+ cbnz w12, .loop_residual_32291
+ ret292
+.vl_gt_16_getResidual32:293
+ cmp x9, #48294
+ bgt .vl_gt_48_getResidual32295
+ ptrue p0.b, vl32296
+.rept 32297
+ ld1b {z0.b}, p0/z, x0298
+ ld1b {z2.b}, p0/z, x1299
+ add x0, x0, x3300
+ add x1, x1, x3301
+ usublb z4.h, z0.b, z2.b302
+ usublt z5.h, z0.b, z2.b303
+ st2h {z4.h, z5.h}, p0, x2304
+ add x2, x2, x3, lsl #1305
+.endr306
+ ret307
+.vl_gt_48_getResidual32:308
+ ptrue p0.h, vl32309
+.rept 32310
+ ld1b {z0.h}, p0/z, x0311
+ ld1b {z4.h}, p0/z, x1312
+ add x0, x0, x3313
+ add x1, x1, x3314
+ sub z8.h, z0.h, z4.h315
+ st1h {z8.h}, p0, x2316
+ add x2, x2, x3, lsl #1317
+.endr318
+ ret319
+endfunc320
+321
+function PFX(pixel_sub_ps_32x32_sve2)322
+ rdvl x9, #1323
+ cmp x9, #16324
+ bgt .vl_gt_16_pixel_sub_ps_32x32325
+ lsl x1, x1, #1326
+ mov w12, #4327
+.loop_sub_ps_32_sve2:328
+ sub w12, w12, #1329
+.rept 4330
+ ld1 {v0.16b-v1.16b}, x2, x4331
+ ld1 {v2.16b-v3.16b}, x3, x5332
+ ld1 {v4.16b-v5.16b}, x2, x4333
+ ld1 {v6.16b-v7.16b}, x3, x5334
+ usubl v16.8h, v0.8b, v2.8b335
+ usubl2 v17.8h, v0.16b, v2.16b336
+ usubl v18.8h, v1.8b, v3.8b337
+ usubl2 v19.8h, v1.16b, v3.16b338
+ usubl v20.8h, v4.8b, v6.8b339
+ usubl2 v21.8h, v4.16b, v6.16b340
+ usubl v22.8h, v5.8b, v7.8b341
+ usubl2 v23.8h, v5.16b, v7.16b342
+ st1 {v16.8h-v19.8h}, x0, x1343
+ st1 {v20.8h-v23.8h}, x0, x1344
+.endr345
+ cbnz w12, .loop_sub_ps_32_sve2346
+ ret347
+.vl_gt_16_pixel_sub_ps_32x32:348
+ cmp x9, #48349
+ bgt .vl_gt_48_pixel_sub_ps_32x32350
+ ptrue p0.b, vl32351
+ mov w12, #8352
+.vl_gt_16_loop_sub_ps_32_sve2:353
+ sub w12, w12, #1354
+.rept 4355
+ ld1b {z0.b}, p0/z, x2356
+ ld1b {z2.b}, p0/z, x3357
+ add x2, x2, x4358
+ add x3, x3, x5359
+ usublb z16.h, z0.b, z2.b360
+ usublt z17.h, z0.b, z2.b361
+ st2h {z16.h, z17.h}, p0, x0362
+ add x0, x0, x1, lsl #1363
+.endr364
+ cbnz w12, .vl_gt_16_loop_sub_ps_32_sve2365
+ ret366
+.vl_gt_48_pixel_sub_ps_32x32:367
+ ptrue p0.h, vl32368
+ mov w12, #8369
+.vl_gt_48_loop_sub_ps_32_sve2:370
+ sub w12, w12, #1371
+.rept 4372
+ ld1b {z0.h}, p0/z, x2373
+ ld1b {z4.h}, p0/z, x3374
+ add x2, x2, x4375
+ add x3, x3, x5376
+ sub z8.h, z0.h, z4.h377
+ st1h {z8.h}, p0, x0378
+ add x0, x0, x1, lsl #1379
+.endr380
+ cbnz w12, .vl_gt_48_loop_sub_ps_32_sve2381
+ ret382
+endfunc383
+384
+function PFX(pixel_sub_ps_64x64_sve2)385
+ rdvl x9, #1386
+ cmp x9, #16387
+ bgt .vl_gt_16_pixel_sub_ps_64x64388
+ lsl x1, x1, #1389
+ sub x1, x1, #64390
+ mov w12, #16391
+.loop_sub_ps_64_sve2:392
+ sub w12, w12, #1393
+.rept 4394
+ ld1 {v0.16b-v3.16b}, x2, x4395
+ ld1 {v4.16b-v7.16b}, x3, x5396
+ usubl v16.8h, v0.8b, v4.8b397
+ usubl2 v17.8h, v0.16b, v4.16b398
+ usubl v18.8h, v1.8b, v5.8b399
+ usubl2 v19.8h, v1.16b, v5.16b400
+ usubl v20.8h, v2.8b, v6.8b401
+ usubl2 v21.8h, v2.16b, v6.16b402
+ usubl v22.8h, v3.8b, v7.8b403
+ usubl2 v23.8h, v3.16b, v7.16b404
+ st1 {v16.8h-v19.8h}, x0, #64405
+ st1 {v20.8h-v23.8h}, x0, x1406
+.endr407
+ cbnz w12, .loop_sub_ps_64_sve2408
+ ret409
+.vl_gt_16_pixel_sub_ps_64x64:410
+ rdvl x9, #1411
+ cmp x9, #16412
+ bgt .vl_gt_16_pixel_sub_ps_64x64413
+ ptrue p0.b, vl32414
+ mov w12, #16415
+.vl_gt_16_loop_sub_ps_64_sve2:416
+ sub w12, w12, #1417
+.rept 4418
+ ld1b {z0.b}, p0/z, x2419
+ ld1b {z1.b}, p0/z, x2, #1, mul vl420
+ ld1b {z4.b}, p0/z, x3421
+ ld1b {z5.b}, p0/z, x3, #1, mul vl422
+ add x2, x2, x4423
+ add x3, x3, x5424
+ usublb z16.h, z0.b, z4.b425
+ usublt z17.h, z0.b, z4.b426
+ usublb z18.h, z1.b, z5.b427
+ usublt z19.h, z1.b, z5.b428
+ st2h {z16.h, z17.h}, p0, x0429
+ st2h {z18.h, z19.h}, p0, x0, #2, mul vl430
+ add x0, x0, x1, lsl #1431
+.endr432
+ cbnz w12, .vl_gt_16_loop_sub_ps_64_sve2433
+ ret434
+.vl_gt_48_pixel_sub_ps_64x64:435
+ cmp x9, #112436
+ bgt .vl_gt_112_pixel_sub_ps_64x64437
+ ptrue p0.b, vl64438
+ mov w12, #16439
+.vl_gt_48_loop_sub_ps_64_sve2:440
+ sub w12, w12, #1441
+.rept 4442
+ ld1b {z0.b}, p0/z, x2443
+ ld1b {z4.b}, p0/z, x3444
+ add x2, x2, x4445
+ add x3, x3, x5446
+ usublb z16.h, z0.b, z4.b447
+ usublt z17.h, z0.b, z4.b448
+ st2h {z16.h, z17.h}, p0, x0449
+ add x0, x0, x1, lsl #1450
+.endr451
+ cbnz w12, .vl_gt_48_loop_sub_ps_64_sve2452
+ ret453
+.vl_gt_112_pixel_sub_ps_64x64:454
+ ptrue p0.h, vl64455
+ mov w12, #16456
+.vl_gt_112_loop_sub_ps_64_sve2:457
+ sub w12, w12, #1458
+.rept 4459
+ ld1b {z0.h}, p0/z, x2460
+ ld1b {z8.h}, p0/z, x3461
+ add x2, x2, x4462
+ add x3, x3, x5463
+ sub z16.h, z0.h, z8.h464
+ st1h {z16.h}, p0, x0465
+ add x0, x0, x1, lsl #1466
+.endr467
+ cbnz w12, .vl_gt_112_loop_sub_ps_64_sve2468
+ ret469
+endfunc470
+471
+function PFX(pixel_sub_ps_32x64_sve2)472
+ rdvl x9, #1473
+ cmp x9, #16474
+ bgt .vl_gt_16_pixel_sub_ps_32x64475
+ lsl x1, x1, #1476
+ mov w12, #8477
+.loop_sub_ps_32x64_sve2:478
+ sub w12, w12, #1479
+.rept 4480
+ ld1 {v0.16b-v1.16b}, x2, x4481
+ ld1 {v2.16b-v3.16b}, x3, x5482
+ ld1 {v4.16b-v5.16b}, x2, x4483
+ ld1 {v6.16b-v7.16b}, x3, x5484
+ usubl v16.8h, v0.8b, v2.8b485
+ usubl2 v17.8h, v0.16b, v2.16b486
+ usubl v18.8h, v1.8b, v3.8b487
+ usubl2 v19.8h, v1.16b, v3.16b488
+ usubl v20.8h, v4.8b, v6.8b489
+ usubl2 v21.8h, v4.16b, v6.16b490
+ usubl v22.8h, v5.8b, v7.8b491
+ usubl2 v23.8h, v5.16b, v7.16b492
+ st1 {v16.8h-v19.8h}, x0, x1493
+ st1 {v20.8h-v23.8h}, x0, x1494
+.endr495
+ cbnz w12, .loop_sub_ps_32x64_sve2496
+ ret497
+.vl_gt_16_pixel_sub_ps_32x64:498
+ cmp x9, #48499
+ bgt .vl_gt_48_pixel_sub_ps_32x64500
+ ptrue p0.b, vl32501
+ mov w12, #8502
+.vl_gt_16_loop_sub_ps_32x64_sve2:503
+ sub w12, w12, #1504
+.rept 8505
+ ld1b {z0.b}, p0/z, x2506
+ ld1b {z2.b}, p0/z, x3507
+ add x2, x2, x4508
+ add x3, x3, x5509
+ usublb z16.h, z0.b, z2.b510
+ usublt z17.h, z0.b, z2.b511
+ st2h {z16.h, z17.h}, p0, x0512
+ add x0, x0, x1, lsl #1513
+.endr514
+ cbnz w12, .vl_gt_16_loop_sub_ps_32x64_sve2515
+ ret516
+.vl_gt_48_pixel_sub_ps_32x64:517
+ ptrue p0.h, vl32518
+ mov w12, #8519
+.vl_gt_48_loop_sub_ps_32x64_sve2:520
+ sub w12, w12, #1521
+.rept 8522
+ ld1b {z0.h}, p0/z, x2523
+ ld1b {z4.h}, p0/z, x3524
+ add x2, x2, x4525
+ add x3, x3, x5526
+ sub z8.h, z0.h, z4.h527
+ st1h {z8.h}, p0, x0528
+ add x0, x0, x1, lsl #1529
+.endr530
+ cbnz w12, .vl_gt_48_loop_sub_ps_32x64_sve2531
+ ret532
+endfunc533
+534
+function PFX(pixel_add_ps_4x4_sve2)535
+ ptrue p0.h, vl8536
+ ptrue p1.h, vl4537
+.rept 4538
+ ld1b {z0.h}, p0/z, x2539
+ ld1h {z2.h}, p1/z, x3540
+ add x2, x2, x4541
+ add x3, x3, x5, lsl #1542
+ add z4.h, z0.h, z2.h543
+ sqxtunb z4.b, z4.h544
+ st1b {z4.h}, p1, x0545
+ add x0, x0, x1546
+.endr547
+ ret548
+endfunc549
+550
+function PFX(pixel_add_ps_8x8_sve2)551
+ ptrue p0.h, vl8552
+.rept 8553
+ ld1b {z0.h}, p0/z, x2554
+ ld1h {z2.h}, p0/z, x3555
+ add x2, x2, x4556
+ add x3, x3, x5, lsl #1557
+ add z4.h, z0.h, z2.h558
+ sqxtunb z4.b, z4.h559
+ st1b {z4.h}, p0, x0560
+ add x0, x0, x1561
+.endr562
+ ret563
+endfunc564
+565
+.macro pixel_add_ps_16xN_sve2 h566
+function PFX(pixel_add_ps_16x\h\()_sve2)567
+ rdvl x9, #1568
+ cmp x9, #16569
+ bgt .vl_gt_16_pixel_add_ps_16x\h570
+ ptrue p0.b, vl16571
+.rept \h572
+ ld1b {z0.h}, p0/z, x2573
+ ld1b {z1.h}, p0/z, x2, #1, mul vl574
+ ld1h {z2.h}, p0/z, x3575
+ ld1h {z3.h}, p0/z, x3, #1, mul vl576
+ add x2, x2, x4577
+ add x3, x3, x5, lsl #1578
+ add z24.h, z0.h, z2.h579
+ add z25.h, z1.h, z3.h580
+ sqxtunb z6.b, z24.h581
+ sqxtunb z7.b, z25.h582
+ st1b {z6.h}, p0, x0583
+ st1b {z7.h}, p0, x0, #1, mul vl584
+ add x0, x0, x1585
+.endr586
+ ret587
+.vl_gt_16_pixel_add_ps_16x\h\():588
+ ptrue p0.b, vl32589
+.rept \h590
+ ld1b {z0.h}, p0/z, x2591
+ ld1h {z2.h}, p0/z, x3592
+ add x2, x2, x4593
+ add x3, x3, x5, lsl #1594
+ add z24.h, z0.h, z2.h595
+ sqxtunb z6.b, z24.h596
+ st1b {z6.h}, p0, x0597
+ add x0, x0, x1598
+.endr599
+ ret600
+endfunc601
+.endm602
+603
+pixel_add_ps_16xN_sve2 16604
+pixel_add_ps_16xN_sve2 32605
+606
+.macro pixel_add_ps_32xN_sve2 h607
+ function PFX(pixel_add_ps_32x\h\()_sve2)608
+ rdvl x9, #1609
+ cmp x9, #16610
+ bgt .vl_gt_16_pixel_add_ps_32x\h611
+ lsl x5, x5, #1612
+ mov w12, #\h / 4613
+.loop_add_ps__sve2_32x\h\():614
+ sub w12, w12, #1615
+.rept 4616
+ ld1 {v0.16b-v1.16b}, x2, x4617
+ ld1 {v16.8h-v19.8h}, x3, x5618
+ uxtl v4.8h, v0.8b619
+ uxtl2 v5.8h, v0.16b620
+ uxtl v6.8h, v1.8b621
+ uxtl2 v7.8h, v1.16b622
+ add v24.8h, v4.8h, v16.8h623
+ add v25.8h, v5.8h, v17.8h624
+ add v26.8h, v6.8h, v18.8h625
+ add v27.8h, v7.8h, v19.8h626
+ sqxtun v4.8b, v24.8h627
+ sqxtun2 v4.16b, v25.8h628
+ sqxtun v5.8b, v26.8h629
+ sqxtun2 v5.16b, v27.8h630
+ st1 {v4.16b-v5.16b}, x0, x1631
+.endr632
+ cbnz w12, .loop_add_ps__sve2_32x\h633
+ ret634
+.vl_gt_16_pixel_add_ps_32x\h\():635
+ cmp x9, #48636
+ bgt .vl_gt_48_pixel_add_ps_32x\h637
+ ptrue p0.b, vl32638
+.rept \h639
+ ld1b {z0.h}, p0/z, x2640
+ ld1b {z1.h}, p0/z, x2, #1, mul vl641
+ ld1h {z4.h}, p0/z, x3642
+ ld1h {z5.h}, p0/z, x3, #1, mul vl643
+ add x2, x2, x4644
+ add x3, x3, x5, lsl #1645
+ add z24.h, z0.h, z4.h646
+ add z25.h, z1.h, z5.h647
+ sqxtunb z6.b, z24.h648
+ sqxtunb z7.b, z25.h649
+ st1b {z6.h}, p0, x0650
+ st1b {z7.h}, p0, x0, #1, mul vl651
+ add x0, x0, x1652
+.endr653
+ ret654
+.vl_gt_48_pixel_add_ps_32x\h\():655
+ ptrue p0.b, vl64656
+.rept \h657
+ ld1b {z0.h}, p0/z, x2658
+ ld1h {z4.h}, p0/z, x3659
+ add x2, x2, x4660
+ add x3, x3, x5, lsl #1661
+ add z24.h, z0.h, z4.h662
+ sqxtunb z6.b, z24.h663
+ st1b {z6.h}, p0, x0664
+ add x0, x0, x1665
+.endr666
+ ret667
+endfunc668
+.endm669
+670
+pixel_add_ps_32xN_sve2 32671
+pixel_add_ps_32xN_sve2 64672
+673
+function PFX(pixel_add_ps_64x64_sve2)674
+ rdvl x9, #1675
+ cmp x9, #16676
+ bgt .vl_gt_16_pixel_add_ps_64x64677
+ ptrue p0.b, vl16678
+.rept 64679
+ ld1b {z0.h}, p0/z, x2680
+ ld1b {z1.h}, p0/z, x2, #1, mul vl681
+ ld1b {z2.h}, p0/z, x2, #2, mul vl682
+ ld1b {z3.h}, p0/z, x2, #3, mul vl683
+ ld1b {z4.h}, p0/z, x2, #4 ,mul vl684
+ ld1b {z5.h}, p0/z, x2, #5, mul vl685
+ ld1b {z6.h}, p0/z, x2, #6, mul vl686
+ ld1b {z7.h}, p0/z, x2, #7, mul vl687
+ ld1h {z8.h}, p0/z, x3688
+ ld1h {z9.h}, p0/z, x3, #1, mul vl689
+ ld1h {z10.h}, p0/z, x3, #2, mul vl690
+ ld1h {z11.h}, p0/z, x3, #3, mul vl691
+ ld1h {z12.h}, p0/z, x3, #4, mul vl692
+ ld1h {z13.h}, p0/z, x3, #5, mul vl693
+ ld1h {z14.h}, p0/z, x3, #6, mul vl694
+ ld1h {z15.h}, p0/z, x3, #7, mul vl695
+ add x2, x2, x4696
+ add x3, x3, x5, lsl #1697
+ add z24.h, z0.h, z8.h698
+ add z25.h, z1.h, z9.h699
+ add z26.h, z2.h, z10.h700
+ add z27.h, z3.h, z11.h701
+ add z28.h, z4.h, z12.h702
+ add z29.h, z5.h, z13.h703
+ add z30.h, z6.h, z14.h704
+ add z31.h, z7.h, z15.h705
+ sqxtunb z6.b, z24.h706
+ sqxtunb z7.b, z25.h707
+ sqxtunb z8.b, z26.h708
+ sqxtunb z9.b, z27.h709
+ sqxtunb z10.b, z28.h710
+ sqxtunb z11.b, z29.h711
+ sqxtunb z12.b, z30.h712
+ sqxtunb z13.b, z31.h713
+ st1b {z6.h}, p0, x0714
+ st1b {z7.h}, p0, x0, #1, mul vl715
+ st1b {z8.h}, p0, x0, #2, mul vl716
+ st1b {z9.h}, p0, x0, #3, mul vl717
+ st1b {z10.h}, p0, x0, #4, mul vl718
+ st1b {z11.h}, p0, x0, #5, mul vl719
+ st1b {z12.h}, p0, x0, #6, mul vl720
+ st1b {z13.h}, p0, x0, #7, mul vl721
+ add x0, x0, x1722
+.endr723
+ ret724
+.vl_gt_16_pixel_add_ps_64x64:725
+ cmp x9, #48726
+ bgt .vl_gt_48_pixel_add_ps_64x64727
+ ptrue p0.b, vl32728
+.rept 64729
+ ld1b {z0.h}, p0/z, x2730
+ ld1b {z1.h}, p0/z, x2, #1, mul vl731
+ ld1b {z2.h}, p0/z, x2, #2, mul vl732
+ ld1b {z3.h}, p0/z, x2, #3, mul vl733
+ ld1h {z8.h}, p0/z, x3734
+ ld1h {z9.h}, p0/z, x3, #1, mul vl735
+ ld1h {z10.h}, p0/z, x3, #2, mul vl736
+ ld1h {z11.h}, p0/z, x3, #3, mul vl737
+ add x2, x2, x4738
+ add x3, x3, x5, lsl #1739
+ add z24.h, z0.h, z8.h740
+ add z25.h, z1.h, z9.h741
+ add z26.h, z2.h, z10.h742
+ add z27.h, z3.h, z11.h743
+ sqxtunb z6.b, z24.h744
+ sqxtunb z7.b, z25.h745
+ sqxtunb z8.b, z26.h746
+ sqxtunb z9.b, z27.h747
+ st1b {z6.h}, p0, x0748
+ st1b {z7.h}, p0, x0, #1, mul vl749
+ st1b {z8.h}, p0, x0, #2, mul vl750
+ st1b {z9.h}, p0, x0, #3, mul vl751
+ add x0, x0, x1752
+.endr753
+ ret754
+.vl_gt_48_pixel_add_ps_64x64:755
+ cmp x9, #112756
+ bgt .vl_gt_112_pixel_add_ps_64x64757
+ ptrue p0.b, vl64758
+.rept 64759
+ ld1b {z0.h}, p0/z, x2760
+ ld1b {z1.h}, p0/z, x2, #1, mul vl761
+ ld1h {z8.h}, p0/z, x3762
+ ld1h {z9.h}, p0/z, x3, #1, mul vl763
+ add x2, x2, x4764
+ add x3, x3, x5, lsl #1765
+ add z24.h, z0.h, z8.h766
+ add z25.h, z1.h, z9.h767
+ sqxtunb z6.b, z24.h768
+ sqxtunb z7.b, z25.h769
+ st1b {z6.h}, p0, x0770
+ st1b {z7.h}, p0, x0, #1, mul vl771
+ add x0, x0, x1772
+.endr773
+ ret774
+.vl_gt_112_pixel_add_ps_64x64:775
+ ptrue p0.b, vl128776
+.rept 64777
+ ld1b {z0.h}, p0/z, x2778
+ ld1h {z8.h}, p0/z, x3779
+ add x2, x2, x4780
+ add x3, x3, x5, lsl #1781
+ add z24.h, z0.h, z8.h782
+ sqxtunb z6.b, z24.h783
+ st1b {z6.h}, p0, x0784
+ add x0, x0, x1785
+.endr786
+ ret787
+endfunc788
+789
+// Chroma add_ps790
+function PFX(pixel_add_ps_4x8_sve2)791
+ ptrue p0.h,vl4792
+.rept 8793
+ ld1b {z0.h}, p0/z, x2794
+ ld1h {z2.h}, p0/z, x3795
+ add x2, x2, x4796
+ add x3, x3, x5, lsl #1797
+ add z4.h, z0.h, z2.h798
+ sqxtunb z4.b, z4.h799
+ st1b {z4.h}, p0, x0800
+ add x0, x0, x1801
+.endr802
+ ret803
+endfunc804
+805
+function PFX(pixel_add_ps_8x16_sve2)806
+ ptrue p0.h,vl8807
+.rept 16808
+ ld1b {z0.h}, p0/z, x2809
+ ld1h {z2.h}, p0/z, x3810
+ add x2, x2, x4811
+ add x3, x3, x5, lsl #1812
+ add z4.h, z0.h, z2.h813
+ sqxtunb z4.b, z4.h814
+ st1b {z4.h}, p0, x0815
+ add x0, x0, x1816
+.endr817
+ ret818
+endfunc819
+820
+// void scale1D_128to64(pixel *dst, const pixel *src)821
+function PFX(scale1D_128to64_sve2)822
+ rdvl x9, #1823
+ cmp x9, #16824
+ bgt .vl_gt_16_scale1D_128to64825
+ ptrue p0.b, vl16826
+.rept 2827
+ ld2b {z0.b, z1.b}, p0/z, x1828
+ ld2b {z2.b, z3.b}, p0/z, x1, #2, mul vl829
+ ld2b {z4.b, z5.b}, p0/z, x1, #4, mul vl830
+ ld2b {z6.b, z7.b}, p0/z, x1, #6, mul vl831
+ add x1, x1, #128832
+ urhadd z0.b, p0/m, z0.b, z1.b833
+ urhadd z2.b, p0/m, z2.b, z3.b834
+ urhadd z4.b, p0/m, z4.b, z5.b835
+ urhadd z6.b, p0/m, z6.b, z7.b836
+ st1b {z0.b}, p0, x0837
+ st1b {z2.b}, p0, x0, #1, mul vl838
+ st1b {z4.b}, p0, x0, #2, mul vl839
+ st1b {z6.b}, p0, x0, #3, mul vl840
+ add x0, x0, #64841
+.endr842
+ ret843
+.vl_gt_16_scale1D_128to64:844
+ cmp x9, #48845
+ bgt .vl_gt_48_scale1D_128to64846
+ ptrue p0.b, vl32847
+.rept 2848
+ ld2b {z0.b, z1.b}, p0/z, x1849
+ ld2b {z2.b, z3.b}, p0/z, x1, #2, mul vl850
+ add x1, x1, #128851
+ urhadd z0.b, p0/m, z0.b, z1.b852
+ urhadd z2.b, p0/m, z2.b, z3.b853
+ st1b {z0.b}, p0, x0854
+ st1b {z2.b}, p0, x0, #1, mul vl855
+ add x0, x0, #64856
+.endr857
+ ret858
+.vl_gt_48_scale1D_128to64:859
+ ptrue p0.b, vl64860
+.rept 2861
+ ld2b {z0.b, z1.b}, p0/z, x1862
+ add x1, x1, #128863
+ urhadd z0.b, p0/m, z0.b, z1.b864
+ st1b {z0.b}, p0, x0865
+ add x0, x0, #64866
+.endr867
+ ret868
+endfunc869
+870
+/***** dequant_scaling*****/871
+// void dequant_scaling_c(const int16_t* quantCoef, const int32_t* deQuantCoef, int16_t* coef, int num, int per, int shift)872
+function PFX(dequant_scaling_sve2)873
+ ptrue p0.h, vl8874
+ add x5, x5, #4 // shift + 4875
+ lsr x3, x3, #3 // num / 8876
+ cmp x5, x4877
+ blt .dequant_skip_sve2878
+879
+ mov x12, #1880
+ sub x6, x5, x4 // shift - per881
+ sub x6, x6, #1 // shift - per - 1882
+ lsl x6, x12, x6 // 1 << shift - per - 1 (add)883
+ mov z0.s, w6884
+ sub x7, x4, x5 // per - shift885
+ mov z3.s, w7886
+887
+.dequant_loop1_sve2:888
+ ld1h {z19.h}, p0/z, x0889
+ ld1w {z2.s}, p0/z, x1890
+ add x1, x1, #16891
+ ld1w {z20.s}, p0/z, x1892
+ add x0, x0, #16893
+ add x1, x1, #16894
+895
+ sub x3, x3, #1896
+ sunpklo z1.s, z19.h897
+ sunpkhi z19.s, z19.h898
+899
+ mul z1.s, z1.s, z2.s // quantCoef * deQuantCoef900
+ mul z19.s, z19.s, z20.s901
+ add z1.s, z1.s, z0.s // quantCoef * deQuantCoef + add902
+ add z19.s, z19.s, z0.s903
+904
+ // No equivalent instructions in SVE2 for sshl905
+ // as sqshl has double latency906
+ sshl v1.4s, v1.4s, v3.4s907
+ sshl v19.4s, v19.4s, v3.4s908
+909
+ sqxtnb z16.h, z1.s910
+ sqxtnb z17.h, z19.s911
+ st1h {z16.s}, p0, x2912
+ st1h {z17.s}, p0, x2, #1, mul vl913
+ add x2, x2, #16914
+ cbnz x3, .dequant_loop1_sve2915
+ ret916
+917
+.dequant_skip_sve2:918
+ sub x6, x4, x5 // per - shift919
+ mov z0.h, w6920
+921
+.dequant_loop2_sve2:922
+ ld1h {z19.h}, p0/z, x0923
+ ld1w {z2.s}, p0/z, x1924
+ add x1, x1, #16925
+ ld1w {z20.s}, p0/z, x1926
+ add x0, x0, #16927
+ add x1, x1, #16928
+929
+930
+ sub x3, x3, #1931
+ sunpklo z1.s, z19.h932
+ sunpkhi z19.s, z19.h933
+934
+ mul z1.s, z1.s, z2.s // quantCoef * deQuantCoef935
+ mul z19.s, z19.s, z20.s936
+937
+ // Keeping NEON instructions here in order to have938
+ // one sqshl later939
+ sqxtn v16.4h, v1.4s // x265_clip3940
+ sqxtn2 v16.8h, v19.4s941
+942
+ sqshl z16.h, p0/m, z16.h, z0.h // coefQ << per - shift943
+ st1h {z16.h}, p0, x2944
+ add x2, x2, #16945
+ cbnz x3, .dequant_loop2_sve2946
+ ret947
+endfunc948
+949
+// void dequant_normal_c(const int16_t* quantCoef, int16_t* coef, int num, int scale, int shift)950
+function PFX(dequant_normal_sve2)951
+ lsr w2, w2, #4 // num / 16952
+ neg w4, w4953
+ mov z0.h, w3954
+ mov z1.s, w4955
+ rdvl x9, #1956
+ cmp x9, #16957
+ bgt .vl_gt_16_dequant_normal_sve2958
+.dqn_loop1_sve2:959
+ ld1 {v2.8h, v3.8h}, x0, #32960
+ smull v16.4s, v2.4h, v0.4h961
+ smull2 v17.4s, v2.8h, v0.8h962
+ smull v18.4s, v3.4h, v0.4h963
+ smull2 v19.4s, v3.8h, v0.8h964
+965
+ srshl v16.4s, v16.4s, v1.4s966
+ srshl v17.4s, v17.4s, v1.4s967
+ srshl v18.4s, v18.4s, v1.4s968
+ srshl v19.4s, v19.4s, v1.4s969
+970
+ sqxtn v2.4h, v16.4s971
+ sqxtn2 v2.8h, v17.4s972
+ sqxtn v3.4h, v18.4s973
+ sqxtn2 v3.8h, v19.4s974
+975
+ sub w2, w2, #1976
+ st1 {v2.8h, v3.8h}, x1, #32977
+ cbnz w2, .dqn_loop1_sve2978
+ ret979
+.vl_gt_16_dequant_normal_sve2:980
+ ptrue p0.h, vl16981
+.gt_16_dqn_loop1_sve2:982
+ ld1h {z2.h}, p0/z, x0983
+ add x0, x0, #32984
+ smullb z16.s, z2.h, z0.h985
+ smullt z17.s, z2.h, z0.h986
+987
+ srshl z16.s, p0/m, z16.s, z1.s988
+ srshl z17.s, p0/m, z17.s, z1.s989
+990
+ sqxtnb z2.h, z16.s991
+ sqxtnt z2.h, z17.s992
+ 993
+ sub w2, w2, #1994
+ st1h {z2.h}, p0, x1995
+ add x1, x1, #32996
+ cbnz w2, .gt_16_dqn_loop1_sve2997
+ ret998
+999
+endfunc1000
+1001
+// void ssim_4x4x2_core(const pixel* pix1, intptr_t stride1, const pixel* pix2, intptr_t stride2, int sums24)1002
+function PFX(ssim_4x4x2_core_sve2)1003
+ ptrue p0.b, vl161004
+ movi v30.2d, #01005
+ movi v31.2d, #01006
+1007
+ ld1b {z0.h}, p0/z, x01008
+ add x0, x0, x11009
+ ld1b {z1.h}, p0/z, x01010
+ add x0, x0, x11011
+ ld1b {z2.h}, p0/z, x01012
+ add x0, x0, x11013
+ ld1b {z3.h}, p0/z, x01014
+ add x0, x0, x11015
+1016
+ ld1b {z4.h}, p0/z, x21017
+ add x2, x2, x31018
+ ld1b {z5.h}, p0/z, x21019
+ add x2, x2, x31020
+ ld1b {z6.h}, p0/z, x21021
+ add x2, x2, x31022
+ ld1b {z7.h}, p0/z, x21023
+ add x2, x2, x31024
+1025
+ mul z16.h, z0.h, z0.h1026
+ mul z17.h, z1.h, z1.h1027
+ mul z18.h, z2.h, z2.h1028
+ uaddlp v30.4s, v16.8h1029
+1030
+ mul z19.h, z3.h, z3.h1031
+ mul z20.h, z4.h, z4.h1032
+ mul z21.h, z5.h, z5.h1033
+ uadalp v30.4s, v17.8h1034
+1035
+ mul z22.h, z6.h, z6.h1036
+ mul z23.h, z7.h, z7.h1037
+ mul z24.h, z0.h, z4.h1038
+ uadalp v30.4s, v18.8h1039
+1040
+ mul z25.h, z1.h, z5.h1041
+ mul z26.h, z2.h, z6.h1042
+ mul z27.h, z3.h, z7.h1043
+ uadalp v30.4s, v19.8h1044
+1045
+ add z28.h, z0.h, z1.h1046
+ add z29.h, z4.h, z5.h1047
+ uadalp v30.4s, v20.8h1048
+ uaddlp v31.4s, v24.8h1049
+1050
+ add z28.h, z28.h, z2.h1051
+ add z29.h, z29.h, z6.h1052
+ uadalp v30.4s, v21.8h1053
+ uadalp v31.4s, v25.8h1054
+1055
+ add z28.h, z28.h, z3.h1056
+ add z29.h, z29.h, z7.h1057
+ uadalp v30.4s, v22.8h1058
+ uadalp v31.4s, v26.8h1059
+1060
+ // Better use NEON instructions here1061
+ uaddlp v28.4s, v28.8h1062
+ uaddlp v29.4s, v29.8h1063
+ uadalp v30.4s, v23.8h1064
+ uadalp v31.4s, v27.8h1065
+1066
+ addp v28.4s, v28.4s, v28.4s1067
+ addp v29.4s, v29.4s, v29.4s1068
+ addp v30.4s, v30.4s, v30.4s1069
+ addp v31.4s, v31.4s, v31.4s1070
+1071
+ st4 {v28.2s, v29.2s, v30.2s, v31.2s}, x41072
+ ret1073
+endfunc1074
+1075
+// void ssimDist_c(const pixel* fenc, uint32_t fStride, const pixel* recon, intptr_t rstride, uint64_t *ssBlock, int shift, uint64_t *ac_k)1076
+.macro ssimDist_start_sve21077
+ mov z0.d, #01078
+ mov z1.d, #01079
+.endm1080
+1081
+.macro ssimDist_1_sve2 z0 z1 z2 z31082
+ sub z16.s, \z0\().s, \z2\().s1083
+ sub z17.s, \z1\().s, \z3\().s1084
+ mul z18.s, \z0\().s, \z0\().s1085
+ mul z19.s, \z1\().s, \z1\().s1086
+ mul z20.s, z16.s, z16.s1087
+ mul z21.s, z17.s, z17.s1088
+ add z0.s, z0.s, z18.s1089
+ add z0.s, z0.s, z19.s1090
+ add z1.s, z1.s, z20.s1091
+ add z1.s, z1.s, z21.s1092
+.endm1093
+1094
+.macro ssimDist_end_sve21095
+ uaddv d0, p0, z0.s1096
+ uaddv d1, p0, z1.s1097
+ str d0, x61098
+ str d1, x41099
+.endm1100
+1101
+function PFX(ssimDist4_sve2)1102
+ ssimDist_start1103
+ ptrue p0.s, vl41104
+.rept 41105
+ ld1b {z4.s}, p0/z, x01106
+ add x0, x0, x11107
+ ld1b {z5.s}, p0/z, x21108
+ add x2, x2, x31109
+ sub z2.s, z4.s, z5.s1110
+ mul z3.s, z4.s, z4.s1111
+ mul z2.s, z2.s, z2.s1112
+ add z0.s, z0.s, z3.s1113
+ add z1.s, z1.s, z2.s1114
+.endr1115
+ ssimDist_end1116
+ ret1117
+endfunc1118
+1119
+function PFX(ssimDist8_sve2)1120
+ rdvl x9, #11121
+ cmp x9, #161122
+ bgt .vl_gt_16_ssimDist81123
+ ssimDist_start1124
+ ptrue p0.s, vl41125
+.rept 81126
+ ld1b {z4.s}, p0/z, x01127
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1128
+ add x0, x0, x11129
+ ld1b {z6.s}, p0/z, x21130
+ ld1b {z7.s}, p0/z, x2, #1, mul vl1131
+ add x2, x2, x31132
+ ssimDist_1_sve2 z4, z5, z6, z71133
+.endr1134
+ ssimDist_end1135
+ ret1136
+.vl_gt_16_ssimDist8:1137
+ ssimDist_start_sve21138
+ ptrue p0.s, vl81139
+.rept 81140
+ ld1b {z4.s}, p0/z, x01141
+ add x0, x0, x11142
+ ld1b {z6.s}, p0/z, x21143
+ add x2, x2, x31144
+ sub z20.s, z4.s, z6.s1145
+ mul z16.s, z4.s, z4.s1146
+ mul z18.s, z20.s, z20.s1147
+ add z0.s, z0.s, z16.s1148
+ add z1.s, z1.s, z18.s1149
+.endr1150
+ ssimDist_end_sve21151
+ ret1152
+endfunc1153
+1154
+function PFX(ssimDist16_sve2)1155
+ mov w12, #161156
+ rdvl x9, #11157
+ cmp x9, #161158
+ bgt .vl_gt_16_ssimDist161159
+ ssimDist_start1160
+ ptrue p0.s, vl41161
+.loop_ssimDist16_sve2:1162
+ sub w12, w12, #11163
+ ld1b {z4.s}, p0/z, x01164
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1165
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1166
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1167
+ add x0, x0, x11168
+ ld1b {z8.s}, p0/z, x21169
+ ld1b {z9.s}, p0/z, x2, #1, mul vl1170
+ ld1b {z10.s}, p0/z, x2, #2, mul vl1171
+ ld1b {z11.s}, p0/z, x2, #3, mul vl1172
+ add x2, x2, x31173
+ ssimDist_1_sve2 z4, z5, z8, z91174
+ ssimDist_1_sve2 z6, z7, z10, z111175
+ cbnz w12, .loop_ssimDist16_sve21176
+ ssimDist_end1177
+ ret1178
+.vl_gt_16_ssimDist16:1179
+ cmp x9, #481180
+ bgt .vl_gt_48_ssimDist161181
+ ssimDist_start_sve21182
+ ptrue p0.s, vl81183
+.vl_gt_16_loop_ssimDist16_sve2:1184
+ sub w12, w12, #11185
+ ld1b {z4.s}, p0/z, x01186
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1187
+ add x0, x0, x11188
+ ld1b {z8.s}, p0/z, x21189
+ ld1b {z9.s}, p0/z, x2, #1, mul vl1190
+ add x2, x2, x31191
+ ssimDist_1_sve2 z4, z5, z8, z91192
+ cbnz w12, .vl_gt_16_loop_ssimDist16_sve21193
+ ssimDist_end_sve21194
+ ret1195
+.vl_gt_48_ssimDist16:1196
+ ssimDist_start_sve21197
+ ptrue p0.s, vl161198
+.vl_gt_48_loop_ssimDist16_sve2:1199
+ sub w12, w12, #11200
+ ld1b {z4.s}, p0/z, x01201
+ add x0, x0, x11202
+ ld1b {z8.s}, p0/z, x21203
+ add x2, x2, x31204
+ sub z20.s, z4.s, z8.s1205
+ mul z16.s, z4.s, z4.s1206
+ mul z18.s, z20.s, z20.s1207
+ add z0.s, z0.s, z16.s1208
+ add z1.s, z1.s, z18.s1209
+ cbnz w12, .vl_gt_48_loop_ssimDist16_sve21210
+ ssimDist_end_sve21211
+ ret1212
+endfunc1213
+1214
+function PFX(ssimDist32_sve2)1215
+ mov w12, #321216
+ rdvl x9, #11217
+ cmp x9, #161218
+ bgt .vl_gt_16_ssimDist321219
+ ssimDist_start1220
+ ptrue p0.s, vl41221
+.loop_ssimDist32_sve2:1222
+ sub w12, w12, #11223
+ ld1b {z2.s}, p0/z, x01224
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1225
+ ld1b {z4.s}, p0/z, x0, #2, mul vl1226
+ ld1b {z5.s}, p0/z, x0, #3, mul vl1227
+ ld1b {z6.s}, p0/z, x0, #4, mul vl1228
+ ld1b {z7.s}, p0/z, x0, #5, mul vl1229
+ ld1b {z8.s}, p0/z, x0, #6, mul vl1230
+ ld1b {z9.s}, p0/z, x0, #7, mul vl1231
+ add x0, x0, x11232
+ ld1b {z10.s}, p0/z, x21233
+ ld1b {z11.s}, p0/z, x2, #1, mul vl1234
+ ld1b {z12.s}, p0/z, x2, #2, mul vl1235
+ ld1b {z13.s}, p0/z, x2, #3, mul vl1236
+ ld1b {z14.s}, p0/z, x2, #4, mul vl1237
+ ld1b {z15.s}, p0/z, x2, #5, mul vl1238
+ ld1b {z30.s}, p0/z, x2, #6, mul vl1239
+ ld1b {z31.s}, p0/z, x2, #7, mul vl1240
+ add x2, x2, x31241
+ ssimDist_1_sve2 z2, z3, z10, z111242
+ ssimDist_1_sve2 z4, z5, z12, z131243
+ ssimDist_1_sve2 z6, z7, z14, z151244
+ ssimDist_1_sve2 z8, z9, z30, z311245
+ cbnz w12, .loop_ssimDist32_sve21246
+ ssimDist_end1247
+ ret1248
+.vl_gt_16_ssimDist32:1249
+ cmp x9, #481250
+ bgt .vl_gt_48_ssimDist321251
+ ssimDist_start_sve21252
+ ptrue p0.s, vl81253
+.vl_gt_16_loop_ssimDist32_sve2:1254
+ sub w12, w12, #11255
+ ld1b {z2.s}, p0/z, x01256
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1257
+ ld1b {z4.s}, p0/z, x0, #2, mul vl1258
+ ld1b {z5.s}, p0/z, x0, #3, mul vl1259
+ add x0, x0, x11260
+ ld1b {z10.s}, p0/z, x21261
+ ld1b {z11.s}, p0/z, x2, #1, mul vl1262
+ ld1b {z12.s}, p0/z, x2, #2, mul vl1263
+ ld1b {z13.s}, p0/z, x2, #3, mul vl1264
+ add x2, x2, x31265
+ ssimDist_1_sve2 z2, z3, z10, z111266
+ ssimDist_1_sve2 z4, z5, z12, z131267
+ cbnz w12, .vl_gt_16_loop_ssimDist32_sve21268
+ ssimDist_end_sve21269
+ ret1270
+.vl_gt_48_ssimDist32:1271
+ cmp x9, #1121272
+ bgt .vl_gt_112_ssimDist321273
+ ssimDist_start_sve21274
+ ptrue p0.s, vl161275
+.vl_gt_48_loop_ssimDist32_sve2:1276
+ sub w12, w12, #11277
+ ld1b {z2.s}, p0/z, x01278
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1279
+ add x0, x0, x11280
+ ld1b {z10.s}, p0/z, x21281
+ ld1b {z11.s}, p0/z, x2, #1, mul vl1282
+ add x2, x2, x31283
+ ssimDist_1_sve2 z2, z3, z10, z111284
+ cbnz w12, .vl_gt_48_loop_ssimDist32_sve21285
+ ssimDist_end_sve21286
+ ret1287
+.vl_gt_112_ssimDist32:1288
+ ssimDist_start_sve21289
+ ptrue p0.s, vl321290
+.vl_gt_112_loop_ssimDist32_sve2:1291
+ sub w12, w12, #11292
+ ld1b {z2.s}, p0/z, x01293
+ add x0, x0, x11294
+ ld1b {z10.s}, p0/z, x21295
+ add x2, x2, x31296
+ sub z20.s, z2.s, z10.s1297
+ mul z16.s, z2.s, z2.s1298
+ mul z18.s, z20.s, z20.s1299
+ add z0.s, z0.s, z16.s1300
+ add z1.s, z1.s, z18.s1301
+ cbnz w12, .vl_gt_112_loop_ssimDist32_sve21302
+ ssimDist_end_sve21303
+ ret1304
+endfunc1305
+1306
+function PFX(ssimDist64_sve2)1307
+ mov w12, #641308
+ rdvl x9, #11309
+ cmp x9, #161310
+ bgt .vl_gt_16_ssimDist641311
+ ssimDist_start1312
+ ptrue p0.s, vl41313
+.loop_ssimDist64_sve2:1314
+ sub w12, w12, #11315
+ ld1b {z2.s}, p0/z, x01316
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1317
+ ld1b {z4.s}, p0/z, x0, #2, mul vl1318
+ ld1b {z5.s}, p0/z, x0, #3, mul vl1319
+ ld1b {z6.s}, p0/z, x0, #4, mul vl1320
+ ld1b {z7.s}, p0/z, x0, #5, mul vl1321
+ ld1b {z8.s}, p0/z, x0, #6, mul vl1322
+ ld1b {z9.s}, p0/z, x0, #7, mul vl1323
+ ld1b {z23.s}, p0/z, x21324
+ ld1b {z24.s}, p0/z, x2, #1, mul vl1325
+ ld1b {z25.s}, p0/z, x2, #2, mul vl1326
+ ld1b {z26.s}, p0/z, x2, #3, mul vl1327
+ ld1b {z27.s}, p0/z, x2, #4, mul vl1328
+ ld1b {z28.s}, p0/z, x2, #5, mul vl1329
+ ld1b {z29.s}, p0/z, x2, #6, mul vl1330
+ ld1b {z30.s}, p0/z, x2, #7, mul vl1331
+ ssimDist_1_sve2 z2, z3, z23, z241332
+ ssimDist_1_sve2 z4, z5, z25, z261333
+ ssimDist_1_sve2 z6, z7, z27, z281334
+ ssimDist_1_sve2 z8, z9, z29, z301335
+ mov x4, x01336
+ mov x5, x21337
+ add x4, x4, #321338
+ add x5, x5, #321339
+ ld1b {z2.s}, p0/z, x41340
+ ld1b {z3.s}, p0/z, x4, #1, mul vl1341
+ ld1b {z4.s}, p0/z, x4, #2, mul vl1342
+ ld1b {z5.s}, p0/z, x4, #3, mul vl1343
+ ld1b {z6.s}, p0/z, x4, #4, mul vl1344
+ ld1b {z7.s}, p0/z, x4, #5, mul vl1345
+ ld1b {z8.s}, p0/z, x4, #6, mul vl1346
+ ld1b {z9.s}, p0/z, x4, #7, mul vl1347
+ ld1b {z23.s}, p0/z, x51348
+ ld1b {z24.s}, p0/z, x5, #1, mul vl1349
+ ld1b {z25.s}, p0/z, x5, #2, mul vl1350
+ ld1b {z26.s}, p0/z, x5, #3, mul vl1351
+ ld1b {z27.s}, p0/z, x5, #4, mul vl1352
+ ld1b {z28.s}, p0/z, x5, #5, mul vl1353
+ ld1b {z29.s}, p0/z, x5, #6, mul vl1354
+ ld1b {z30.s}, p0/z, x5, #7, mul vl1355
+ ssimDist_1_sve2 z2, z3, z23, z241356
+ ssimDist_1_sve2 z4, z5, z25, z261357
+ ssimDist_1_sve2 z6, z7, z27, z281358
+ ssimDist_1_sve2 z8, z9, z29, z301359
+ add x0, x0, x11360
+ add x2, x2, x31361
+ cbnz w12, .loop_ssimDist64_sve21362
+ ssimDist_end1363
+ ret1364
+.vl_gt_16_ssimDist64:1365
+ cmp x9, #481366
+ bgt .vl_gt_48_ssimDist641367
+ ssimDist_start_sve21368
+ ptrue p0.s, vl81369
+.vl_gt_16_loop_ssimDist64_sve2:1370
+ sub w12, w12, #11371
+ ld1b {z2.s}, p0/z, x01372
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1373
+ ld1b {z4.s}, p0/z, x0, #2, mul vl1374
+ ld1b {z5.s}, p0/z, x0, #3, mul vl1375
+ ld1b {z6.s}, p0/z, x0, #4, mul vl1376
+ ld1b {z7.s}, p0/z, x0, #5, mul vl1377
+ ld1b {z8.s}, p0/z, x0, #6, mul vl1378
+ ld1b {z9.s}, p0/z, x0, #7, mul vl1379
+ ld1b {z23.s}, p0/z, x21380
+ ld1b {z24.s}, p0/z, x2, #1, mul vl1381
+ ld1b {z25.s}, p0/z, x2, #2, mul vl1382
+ ld1b {z26.s}, p0/z, x2, #3, mul vl1383
+ ld1b {z27.s}, p0/z, x2, #4, mul vl1384
+ ld1b {z28.s}, p0/z, x2, #5, mul vl1385
+ ld1b {z29.s}, p0/z, x2, #6, mul vl1386
+ ld1b {z30.s}, p0/z, x2, #7, mul vl1387
+ ssimDist_1_sve2 z2, z3, z23, z241388
+ ssimDist_1_sve2 z4, z5, z25, z261389
+ ssimDist_1_sve2 z6, z7, z27, z281390
+ ssimDist_1_sve2 z8, z9, z29, z301391
+ add x0, x0, x11392
+ add x2, x2, x31393
+ cbnz w12, .vl_gt_16_loop_ssimDist64_sve21394
+ ssimDist_end_sve21395
+ ret1396
+.vl_gt_48_ssimDist64:1397
+ cmp x9, #1121398
+ bgt .vl_gt_112_ssimDist641399
+ ssimDist_start_sve21400
+ ptrue p0.s, vl161401
+.vl_gt_48_loop_ssimDist64_sve2:1402
+ sub w12, w12, #11403
+ ld1b {z2.s}, p0/z, x01404
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1405
+ ld1b {z4.s}, p0/z, x0, #2, mul vl1406
+ ld1b {z5.s}, p0/z, x0, #3, mul vl1407
+ ld1b {z23.s}, p0/z, x21408
+ ld1b {z24.s}, p0/z, x2, #1, mul vl1409
+ ld1b {z25.s}, p0/z, x2, #2, mul vl1410
+ ld1b {z26.s}, p0/z, x2, #3, mul vl1411
+ ssimDist_1_sve2 z2, z3, z23, z241412
+ ssimDist_1_sve2 z4, z5, z25, z261413
+ add x0, x0, x11414
+ add x2, x2, x31415
+ cbnz w12, .vl_gt_48_loop_ssimDist64_sve21416
+ ssimDist_end_sve21417
+ ret1418
+.vl_gt_112_ssimDist64:1419
+ ssimDist_start_sve21420
+ ptrue p0.s, vl321421
+.vl_gt_112_loop_ssimDist64_sve2:1422
+ sub w12, w12, #11423
+ ld1b {z2.s}, p0/z, x01424
+ ld1b {z3.s}, p0/z, x0, #1, mul vl1425
+ ld1b {z23.s}, p0/z, x21426
+ ld1b {z24.s}, p0/z, x2, #1, mul vl1427
+ ssimDist_1_sve2 z2, z3, z23, z241428
+ add x0, x0, x11429
+ add x2, x2, x31430
+ cbnz w12, .vl_gt_112_loop_ssimDist64_sve21431
+ ssimDist_end_sve21432
+ ret1433
+endfunc1434
+1435
+// void normFact_c(const pixel* src, uint32_t blockSize, int shift, uint64_t *z_k)1436
+.macro normFact_start_sve21437
+ mov z0.d, #01438
+.endm1439
+1440
+.macro normFact_1_sve2 z0, z11441
+ mul z16.s, \z0\().s, \z0\().s1442
+ mul z17.s, \z1\().s, \z1\().s1443
+ add z0.s, z0.s, z16.s1444
+ add z0.s, z0.s, z17.s1445
+.endm1446
+1447
+.macro normFact_end_sve21448
+ uaddv d0, p0, z0.s1449
+ str d0, x31450
+.endm1451
+1452
+function PFX(normFact8_sve2)1453
+ rdvl x9, #11454
+ cmp x9, #161455
+ bgt .vl_gt_16_normFact81456
+ normFact_start1457
+ ptrue p0.s, vl41458
+.rept 81459
+ ld1b {z4.s}, p0/z, x01460
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1461
+ add x0, x0, x11462
+ normFact_1_sve2 z4, z51463
+.endr1464
+ normFact_end1465
+ ret1466
+.vl_gt_16_normFact8:1467
+ normFact_start_sve21468
+ ptrue p0.s, vl81469
+.rept 81470
+ ld1b {z4.s}, p0/z, x01471
+ add x0, x0, x11472
+ mul z16.s, z4.s, z4.s1473
+ add z0.s, z0.s, z16.s1474
+.endr1475
+ normFact_end_sve21476
+ ret1477
+endfunc1478
+1479
+function PFX(normFact16_sve2)1480
+ mov w12, #161481
+ rdvl x9, #11482
+ cmp x9, #161483
+ bgt .vl_gt_16_normFact161484
+ normFact_start1485
+ ptrue p0.s, vl41486
+.loop_normFact16_sve2:1487
+ sub w12, w12, #11488
+ ld1b {z4.s}, p0/z, x01489
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1490
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1491
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1492
+ add x0, x0, x11493
+ normFact_1_sve2 z4, z51494
+ normFact_1_sve2 z6, z71495
+ cbnz w12, .loop_normFact16_sve21496
+ normFact_end1497
+ ret1498
+.vl_gt_16_normFact16:1499
+ cmp x9, #481500
+ bgt .vl_gt_48_normFact161501
+ normFact_start_sve21502
+ ptrue p0.s, vl81503
+.vl_gt_16_loop_normFact16_sve2:1504
+ sub w12, w12, #11505
+ ld1b {z4.s}, p0/z, x01506
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1507
+ add x0, x0, x11508
+ normFact_1_sve2 z4, z51509
+ cbnz w12, .vl_gt_16_loop_normFact16_sve21510
+ normFact_end_sve21511
+ ret1512
+.vl_gt_48_normFact16:1513
+ normFact_start_sve21514
+ ptrue p0.s, vl161515
+.vl_gt_48_loop_normFact16_sve2:1516
+ sub w12, w12, #11517
+ ld1b {z4.s}, p0/z, x01518
+ add x0, x0, x11519
+ mul z16.s, z4.s, z4.s1520
+ add z0.s, z0.s, z16.s1521
+ cbnz w12, .vl_gt_48_loop_normFact16_sve21522
+ normFact_end_sve21523
+ ret1524
+endfunc1525
+1526
+function PFX(normFact32_sve2)1527
+ mov w12, #321528
+ rdvl x9, #11529
+ cmp x9, #161530
+ bgt .vl_gt_16_normFact321531
+ normFact_start1532
+ ptrue p0.s, vl41533
+.loop_normFact32_sve2:1534
+ sub w12, w12, #11535
+ ld1b {z4.s}, p0/z, x01536
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1537
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1538
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1539
+ ld1b {z8.s}, p0/z, x0, #4, mul vl1540
+ ld1b {z9.s}, p0/z, x0, #5, mul vl1541
+ ld1b {z10.s}, p0/z, x0, #6, mul vl1542
+ ld1b {z11.s}, p0/z, x0, #7, mul vl1543
+ add x0, x0, x11544
+ normFact_1_sve2 z4, z51545
+ normFact_1_sve2 z6, z71546
+ normFact_1_sve2 z8, z91547
+ normFact_1_sve2 z10, z111548
+ cbnz w12, .loop_normFact32_sve21549
+ normFact_end1550
+ ret1551
+.vl_gt_16_normFact32:1552
+ cmp x9, #481553
+ bgt .vl_gt_48_normFact321554
+ normFact_start_sve21555
+ ptrue p0.s, vl81556
+.vl_gt_16_loop_normFact32_sve2:1557
+ sub w12, w12, #11558
+ ld1b {z4.s}, p0/z, x01559
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1560
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1561
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1562
+ add x0, x0, x11563
+ normFact_1_sve2 z4, z51564
+ normFact_1_sve2 z6, z71565
+ cbnz w12, .vl_gt_16_loop_normFact32_sve21566
+ normFact_end_sve21567
+ ret1568
+.vl_gt_48_normFact32:1569
+ cmp x9, #1121570
+ bgt .vl_gt_112_normFact321571
+ normFact_start_sve21572
+ ptrue p0.s, vl161573
+.vl_gt_48_loop_normFact32_sve2:1574
+ sub w12, w12, #11575
+ ld1b {z4.s}, p0/z, x01576
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1577
+ add x0, x0, x11578
+ normFact_1_sve2 z4, z51579
+ cbnz w12, .vl_gt_48_loop_normFact32_sve21580
+ normFact_end_sve21581
+ ret1582
+.vl_gt_112_normFact32:1583
+ normFact_start_sve21584
+ ptrue p0.s, vl321585
+.vl_gt_112_loop_normFact32_sve2:1586
+ sub w12, w12, #11587
+ ld1b {z4.s}, p0/z, x01588
+ add x0, x0, x11589
+ mul z16.s, z4.s, z4.s1590
+ add z0.s, z0.s, z16.s1591
+ cbnz w12, .vl_gt_112_loop_normFact32_sve21592
+ normFact_end_sve21593
+ ret1594
+endfunc1595
+1596
+function PFX(normFact64_sve2)1597
+ mov w12, #641598
+ rdvl x9, #11599
+ cmp x9, #161600
+ bgt .vl_gt_16_normFact641601
+ normFact_start1602
+ ptrue p0.s, vl41603
+.loop_normFact64_sve2:1604
+ sub w12, w12, #11605
+ ld1b {z4.s}, p0/z, x01606
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1607
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1608
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1609
+ ld1b {z8.s}, p0/z, x0, #4, mul vl1610
+ ld1b {z9.s}, p0/z, x0, #5, mul vl1611
+ ld1b {z10.s}, p0/z, x0, #6, mul vl1612
+ ld1b {z11.s}, p0/z, x0, #7, mul vl1613
+ normFact_1_sve2 z4, z51614
+ normFact_1_sve2 z6, z71615
+ normFact_1_sve2 z8, z91616
+ normFact_1_sve2 z10, z111617
+ mov x2, x01618
+ add x2, x2, #321619
+ ld1b {z4.s}, p0/z, x21620
+ ld1b {z5.s}, p0/z, x2, #1, mul vl1621
+ ld1b {z6.s}, p0/z, x2, #2, mul vl1622
+ ld1b {z7.s}, p0/z, x2, #3, mul vl1623
+ ld1b {z8.s}, p0/z, x2, #4, mul vl1624
+ ld1b {z9.s}, p0/z, x2, #5, mul vl1625
+ ld1b {z10.s}, p0/z, x2, #6, mul vl1626
+ ld1b {z11.s}, p0/z, x2, #7, mul vl1627
+ normFact_1_sve2 z4, z51628
+ normFact_1_sve2 z6, z71629
+ normFact_1_sve2 z8, z91630
+ normFact_1_sve2 z10, z111631
+ add x0, x0, x11632
+ cbnz w12, .loop_normFact64_sve21633
+ normFact_end1634
+ ret1635
+.vl_gt_16_normFact64:1636
+ cmp x9, #481637
+ bgt .vl_gt_48_normFact641638
+ normFact_start_sve21639
+ ptrue p0.s, vl81640
+.vl_gt_16_loop_normFact64_sve2:1641
+ sub w12, w12, #11642
+ ld1b {z4.s}, p0/z, x01643
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1644
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1645
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1646
+ ld1b {z8.s}, p0/z, x0, #4, mul vl1647
+ ld1b {z9.s}, p0/z, x0, #5, mul vl1648
+ ld1b {z10.s}, p0/z, x0, #6, mul vl1649
+ ld1b {z11.s}, p0/z, x0, #7, mul vl1650
+ normFact_1_sve2 z4, z51651
+ normFact_1_sve2 z6, z71652
+ normFact_1_sve2 z8, z91653
+ normFact_1_sve2 z10, z111654
+ add x0, x0, x11655
+ cbnz w12, .vl_gt_16_loop_normFact64_sve21656
+ normFact_end_sve21657
+ ret1658
+.vl_gt_48_normFact64:1659
+ cmp x9, #1121660
+ bgt .vl_gt_112_normFact641661
+ normFact_start_sve21662
+ ptrue p0.s, vl161663
+.vl_gt_48_loop_normFact64_sve2:1664
+ sub w12, w12, #11665
+ ld1b {z4.s}, p0/z, x01666
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1667
+ ld1b {z6.s}, p0/z, x0, #2, mul vl1668
+ ld1b {z7.s}, p0/z, x0, #3, mul vl1669
+ normFact_1_sve2 z4, z51670
+ normFact_1_sve2 z6, z71671
+ add x0, x0, x11672
+ cbnz w12, .vl_gt_48_loop_normFact64_sve21673
+ normFact_end_sve21674
+ ret1675
+.vl_gt_112_normFact64:1676
+ normFact_start_sve21677
+ ptrue p0.s, vl321678
+.vl_gt_112_loop_normFact64_sve2:1679
+ sub w12, w12, #11680
+ ld1b {z4.s}, p0/z, x01681
+ ld1b {z5.s}, p0/z, x0, #1, mul vl1682
+ normFact_1_sve2 z4, z51683
+ add x0, x0, x11684
+ cbnz w12, .vl_gt_112_loop_normFact64_sve21685
+ normFact_end_sve21686
+ ret1687
+endfunc1688
x265_3.5.tar.gz/source/common/aarch64/pixel-util.S -> x265_3.6.tar.gz/source/common/aarch64/pixel-util.S
Changed
2419
1
2
/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
+ * Copyright (C) 2020-2021 MulticoreWare, Inc5
*6
* Authors: Yimeng Su <yimeng.su@huawei.com>7
* Hongbin Liu <liuhongbin1@huawei.com>8
+ * Sebastian Pop <spop@amazon.com>9
*10
* This program is free software; you can redistribute it and/or modify11
* it under the terms of the GNU General Public License as published by12
13
*****************************************************************************/14
15
#include "asm.S"16
+#include "pixel-util-common.S"17
18
+#ifdef __APPLE__19
+.section __RODATA,__rodata20
+#else21
.section .rodata22
+#endif23
24
.align 425
26
.text27
28
+// uint64_t pixel_var(const pixel* pix, intptr_t i_stride)29
+function PFX(pixel_var_8x8_neon)30
+ ld1 {v4.8b}, x0, x1 // pixx31
+ uxtl v0.8h, v4.8b // sum = pixx32
+ umull v1.8h, v4.8b, v4.8b33
+ uaddlp v1.4s, v1.8h // sqr = pixx * pixx34
+35
+.rept 736
+ ld1 {v4.8b}, x0, x1 // pixx37
+ umull v31.8h, v4.8b, v4.8b38
+ uaddw v0.8h, v0.8h, v4.8b // sum += pixx39
+ uadalp v1.4s, v31.8h // sqr += pixx * pixx40
+.endr41
+ uaddlv s0, v0.8h42
+ uaddlv d1, v1.4s43
+ fmov w0, s044
+ fmov x1, d145
+ orr x0, x0, x1, lsl #32 // return sum + ((uint64_t)sqr << 32);46
+ ret47
+endfunc48
+49
+function PFX(pixel_var_16x16_neon)50
+ pixel_var_start51
+ mov w12, #1652
+.loop_var_16:53
+ sub w12, w12, #154
+ ld1 {v4.16b}, x0, x155
+ pixel_var_1 v456
+ cbnz w12, .loop_var_1657
+ pixel_var_end58
+ ret59
+endfunc60
+61
+function PFX(pixel_var_32x32_neon)62
+ pixel_var_start63
+ mov w12, #3264
+.loop_var_32:65
+ sub w12, w12, #166
+ ld1 {v4.16b-v5.16b}, x0, x167
+ pixel_var_1 v468
+ pixel_var_1 v569
+ cbnz w12, .loop_var_3270
+ pixel_var_end71
+ ret72
+endfunc73
+74
+function PFX(pixel_var_64x64_neon)75
+ pixel_var_start76
+ mov w12, #6477
+.loop_var_64:78
+ sub w12, w12, #179
+ ld1 {v4.16b-v7.16b}, x0, x180
+ pixel_var_1 v481
+ pixel_var_1 v582
+ pixel_var_1 v683
+ pixel_var_1 v784
+ cbnz w12, .loop_var_6485
+ pixel_var_end86
+ ret87
+endfunc88
+89
+// void getResidual4_neon(const pixel* fenc, const pixel* pred, int16_t* residual, intptr_t stride)90
+function PFX(getResidual4_neon)91
+ lsl x4, x3, #192
+.rept 293
+ ld1 {v0.8b}, x0, x394
+ ld1 {v1.8b}, x1, x395
+ ld1 {v2.8b}, x0, x396
+ ld1 {v3.8b}, x1, x397
+ usubl v4.8h, v0.8b, v1.8b98
+ usubl v5.8h, v2.8b, v3.8b99
+ st1 {v4.8b}, x2, x4100
+ st1 {v5.8b}, x2, x4101
+.endr102
+ ret103
+endfunc104
+105
+function PFX(getResidual8_neon)106
+ lsl x4, x3, #1107
+.rept 4108
+ ld1 {v0.8b}, x0, x3109
+ ld1 {v1.8b}, x1, x3110
+ ld1 {v2.8b}, x0, x3111
+ ld1 {v3.8b}, x1, x3112
+ usubl v4.8h, v0.8b, v1.8b113
+ usubl v5.8h, v2.8b, v3.8b114
+ st1 {v4.16b}, x2, x4115
+ st1 {v5.16b}, x2, x4116
+.endr117
+ ret118
+endfunc119
+120
+function PFX(getResidual16_neon)121
+ lsl x4, x3, #1122
+.rept 8123
+ ld1 {v0.16b}, x0, x3124
+ ld1 {v1.16b}, x1, x3125
+ ld1 {v2.16b}, x0, x3126
+ ld1 {v3.16b}, x1, x3127
+ usubl v4.8h, v0.8b, v1.8b128
+ usubl2 v5.8h, v0.16b, v1.16b129
+ usubl v6.8h, v2.8b, v3.8b130
+ usubl2 v7.8h, v2.16b, v3.16b131
+ st1 {v4.8h-v5.8h}, x2, x4132
+ st1 {v6.8h-v7.8h}, x2, x4133
+.endr134
+ ret135
+endfunc136
+137
+function PFX(getResidual32_neon)138
+ lsl x4, x3, #1139
+ mov w12, #4140
+.loop_residual_32:141
+ sub w12, w12, #1142
+.rept 4143
+ ld1 {v0.16b-v1.16b}, x0, x3144
+ ld1 {v2.16b-v3.16b}, x1, x3145
+ ld1 {v4.16b-v5.16b}, x0, x3146
+ ld1 {v6.16b-v7.16b}, x1, x3147
+ usubl v16.8h, v0.8b, v2.8b148
+ usubl2 v17.8h, v0.16b, v2.16b149
+ usubl v18.8h, v1.8b, v3.8b150
+ usubl2 v19.8h, v1.16b, v3.16b151
+ usubl v20.8h, v4.8b, v6.8b152
+ usubl2 v21.8h, v4.16b, v6.16b153
+ usubl v22.8h, v5.8b, v7.8b154
+ usubl2 v23.8h, v5.16b, v7.16b155
+ st1 {v16.8h-v19.8h}, x2, x4156
+ st1 {v20.8h-v23.8h}, x2, x4157
+.endr158
+ cbnz w12, .loop_residual_32159
+ ret160
+endfunc161
+162
+// void pixel_sub_ps_neon(int16_t* a, intptr_t dstride, const pixel* b0, const pixel* b1, intptr_t sstride0, intptr_t sstride1)163
+function PFX(pixel_sub_ps_4x4_neon)164
+ lsl x1, x1, #1165
+.rept 2166
+ ld1 {v0.8b}, x2, x4167
+ ld1 {v1.8b}, x3, x5168
+ ld1 {v2.8b}, x2, x4169
+ ld1 {v3.8b}, x3, x5170
+ usubl v4.8h, v0.8b, v1.8b171
+ usubl v5.8h, v2.8b, v3.8b172
+ st1 {v4.4h}, x0, x1173
+ st1 {v5.4h}, x0, x1174
+.endr175
+ ret176
+endfunc177
+178
+function PFX(pixel_sub_ps_8x8_neon)179
+ lsl x1, x1, #1180
+.rept 4181
+ ld1 {v0.8b}, x2, x4182
+ ld1 {v1.8b}, x3, x5183
+ ld1 {v2.8b}, x2, x4184
+ ld1 {v3.8b}, x3, x5185
+ usubl v4.8h, v0.8b, v1.8b186
+ usubl v5.8h, v2.8b, v3.8b187
+ st1 {v4.8h}, x0, x1188
+ st1 {v5.8h}, x0, x1189
+.endr190
+ ret191
+endfunc192
+193
+function PFX(pixel_sub_ps_16x16_neon)194
+ lsl x1, x1, #1195
+.rept 8196
+ ld1 {v0.16b}, x2, x4197
+ ld1 {v1.16b}, x3, x5198
+ ld1 {v2.16b}, x2, x4199
+ ld1 {v3.16b}, x3, x5200
+ usubl v4.8h, v0.8b, v1.8b201
+ usubl2 v5.8h, v0.16b, v1.16b202
+ usubl v6.8h, v2.8b, v3.8b203
+ usubl2 v7.8h, v2.16b, v3.16b204
+ st1 {v4.8h-v5.8h}, x0, x1205
+ st1 {v6.8h-v7.8h}, x0, x1206
+.endr207
+ ret208
+endfunc209
+210
+function PFX(pixel_sub_ps_32x32_neon)211
+ lsl x1, x1, #1212
+ mov w12, #4213
+.loop_sub_ps_32:214
+ sub w12, w12, #1215
+.rept 4216
+ ld1 {v0.16b-v1.16b}, x2, x4217
+ ld1 {v2.16b-v3.16b}, x3, x5218
+ ld1 {v4.16b-v5.16b}, x2, x4219
+ ld1 {v6.16b-v7.16b}, x3, x5220
+ usubl v16.8h, v0.8b, v2.8b221
+ usubl2 v17.8h, v0.16b, v2.16b222
+ usubl v18.8h, v1.8b, v3.8b223
+ usubl2 v19.8h, v1.16b, v3.16b224
+ usubl v20.8h, v4.8b, v6.8b225
+ usubl2 v21.8h, v4.16b, v6.16b226
+ usubl v22.8h, v5.8b, v7.8b227
+ usubl2 v23.8h, v5.16b, v7.16b228
+ st1 {v16.8h-v19.8h}, x0, x1229
+ st1 {v20.8h-v23.8h}, x0, x1230
+.endr231
+ cbnz w12, .loop_sub_ps_32232
+ ret233
+endfunc234
+235
+function PFX(pixel_sub_ps_64x64_neon)236
+ lsl x1, x1, #1237
+ sub x1, x1, #64238
+ mov w12, #16239
+.loop_sub_ps_64:240
+ sub w12, w12, #1241
+.rept 4242
+ ld1 {v0.16b-v3.16b}, x2, x4243
+ ld1 {v4.16b-v7.16b}, x3, x5244
+ usubl v16.8h, v0.8b, v4.8b245
+ usubl2 v17.8h, v0.16b, v4.16b246
+ usubl v18.8h, v1.8b, v5.8b247
+ usubl2 v19.8h, v1.16b, v5.16b248
+ usubl v20.8h, v2.8b, v6.8b249
+ usubl2 v21.8h, v2.16b, v6.16b250
+ usubl v22.8h, v3.8b, v7.8b251
+ usubl2 v23.8h, v3.16b, v7.16b252
+ st1 {v16.8h-v19.8h}, x0, #64253
+ st1 {v20.8h-v23.8h}, x0, x1254
+.endr255
+ cbnz w12, .loop_sub_ps_64256
+ ret257
+endfunc258
+259
+// chroma sub_ps260
+function PFX(pixel_sub_ps_4x8_neon)261
+ lsl x1, x1, #1262
+.rept 4263
+ ld1 {v0.8b}, x2, x4264
+ ld1 {v1.8b}, x3, x5265
+ ld1 {v2.8b}, x2, x4266
+ ld1 {v3.8b}, x3, x5267
+ usubl v4.8h, v0.8b, v1.8b268
+ usubl v5.8h, v2.8b, v3.8b269
+ st1 {v4.4h}, x0, x1270
+ st1 {v5.4h}, x0, x1271
+.endr272
+ ret273
+endfunc274
+275
+function PFX(pixel_sub_ps_8x16_neon)276
+ lsl x1, x1, #1277
+.rept 8278
+ ld1 {v0.8b}, x2, x4279
+ ld1 {v1.8b}, x3, x5280
+ ld1 {v2.8b}, x2, x4281
+ ld1 {v3.8b}, x3, x5282
+ usubl v4.8h, v0.8b, v1.8b283
+ usubl v5.8h, v2.8b, v3.8b284
+ st1 {v4.8h}, x0, x1285
+ st1 {v5.8h}, x0, x1286
+.endr287
+ ret288
+endfunc289
+290
+function PFX(pixel_sub_ps_16x32_neon)291
+ lsl x1, x1, #1292
+.rept 16293
+ ld1 {v0.16b}, x2, x4294
+ ld1 {v1.16b}, x3, x5295
+ ld1 {v2.16b}, x2, x4296
+ ld1 {v3.16b}, x3, x5297
+ usubl v4.8h, v0.8b, v1.8b298
+ usubl2 v5.8h, v0.16b, v1.16b299
+ usubl v6.8h, v2.8b, v3.8b300
+ usubl2 v7.8h, v2.16b, v3.16b301
+ st1 {v4.8h-v5.8h}, x0, x1302
+ st1 {v6.8h-v7.8h}, x0, x1303
+.endr304
+ ret305
+endfunc306
+307
+function PFX(pixel_sub_ps_32x64_neon)308
+ lsl x1, x1, #1309
+ mov w12, #8310
+.loop_sub_ps_32x64:311
+ sub w12, w12, #1312
+.rept 4313
+ ld1 {v0.16b-v1.16b}, x2, x4314
+ ld1 {v2.16b-v3.16b}, x3, x5315
+ ld1 {v4.16b-v5.16b}, x2, x4316
+ ld1 {v6.16b-v7.16b}, x3, x5317
+ usubl v16.8h, v0.8b, v2.8b318
+ usubl2 v17.8h, v0.16b, v2.16b319
+ usubl v18.8h, v1.8b, v3.8b320
+ usubl2 v19.8h, v1.16b, v3.16b321
+ usubl v20.8h, v4.8b, v6.8b322
+ usubl2 v21.8h, v4.16b, v6.16b323
+ usubl v22.8h, v5.8b, v7.8b324
+ usubl2 v23.8h, v5.16b, v7.16b325
+ st1 {v16.8h-v19.8h}, x0, x1326
+ st1 {v20.8h-v23.8h}, x0, x1327
+.endr328
+ cbnz w12, .loop_sub_ps_32x64329
+ ret330
+endfunc331
+332
+// void x265_pixel_add_ps_neon(pixel* a, intptr_t dstride, const pixel* b0, const int16_t* b1, intptr_t sstride0, intptr_t sstride1);333
+function PFX(pixel_add_ps_4x4_neon)334
+ lsl x5, x5, #1335
+.rept 2336
+ ld1 {v0.8b}, x2, x4337
+ ld1 {v1.8b}, x2, x4338
+ ld1 {v2.4h}, x3, x5339
+ ld1 {v3.4h}, x3, x5340
+ uxtl v0.8h, v0.8b341
+ uxtl v1.8h, v1.8b342
+ add v4.8h, v0.8h, v2.8h343
+ add v5.8h, v1.8h, v3.8h344
+ sqxtun v4.8b, v4.8h345
+ sqxtun v5.8b, v5.8h346
+ st1 {v4.s}0, x0, x1347
+ st1 {v5.s}0, x0, x1348
+.endr349
+ ret350
+endfunc351
+352
+function PFX(pixel_add_ps_8x8_neon)353
+ lsl x5, x5, #1354
+.rept 4355
+ ld1 {v0.8b}, x2, x4356
+ ld1 {v1.8b}, x2, x4357
+ ld1 {v2.8h}, x3, x5358
+ ld1 {v3.8h}, x3, x5359
+ uxtl v0.8h, v0.8b360
+ uxtl v1.8h, v1.8b361
+ add v4.8h, v0.8h, v2.8h362
+ add v5.8h, v1.8h, v3.8h363
+ sqxtun v4.8b, v4.8h364
+ sqxtun v5.8b, v5.8h365
+ st1 {v4.8b}, x0, x1366
+ st1 {v5.8b}, x0, x1367
+.endr368
+ ret369
+endfunc370
+371
+.macro pixel_add_ps_16xN_neon h372
+function PFX(pixel_add_ps_16x\h\()_neon)373
+ lsl x5, x5, #1374
+ mov w12, #\h / 8375
+.loop_add_ps_16x\h\():376
+ sub w12, w12, #1377
+.rept 4378
+ ld1 {v0.16b}, x2, x4379
+ ld1 {v1.16b}, x2, x4380
+ ld1 {v16.8h-v17.8h}, x3, x5381
+ ld1 {v18.8h-v19.8h}, x3, x5382
+ uxtl v4.8h, v0.8b383
+ uxtl2 v5.8h, v0.16b384
+ uxtl v6.8h, v1.8b385
+ uxtl2 v7.8h, v1.16b386
+ add v24.8h, v4.8h, v16.8h387
+ add v25.8h, v5.8h, v17.8h388
+ add v26.8h, v6.8h, v18.8h389
+ add v27.8h, v7.8h, v19.8h390
+ sqxtun v4.8b, v24.8h391
+ sqxtun2 v4.16b, v25.8h392
+ sqxtun v5.8b, v26.8h393
+ sqxtun2 v5.16b, v27.8h394
+ st1 {v4.16b}, x0, x1395
+ st1 {v5.16b}, x0, x1396
+.endr397
+ cbnz w12, .loop_add_ps_16x\h398
+ ret399
+endfunc400
+.endm401
+402
+pixel_add_ps_16xN_neon 16403
+pixel_add_ps_16xN_neon 32404
+405
+.macro pixel_add_ps_32xN_neon h406
+ function PFX(pixel_add_ps_32x\h\()_neon)407
+ lsl x5, x5, #1408
+ mov w12, #\h / 4409
+.loop_add_ps_32x\h\():410
+ sub w12, w12, #1411
+.rept 4412
+ ld1 {v0.16b-v1.16b}, x2, x4413
+ ld1 {v16.8h-v19.8h}, x3, x5414
+ uxtl v4.8h, v0.8b415
+ uxtl2 v5.8h, v0.16b416
+ uxtl v6.8h, v1.8b417
+ uxtl2 v7.8h, v1.16b418
+ add v24.8h, v4.8h, v16.8h419
+ add v25.8h, v5.8h, v17.8h420
+ add v26.8h, v6.8h, v18.8h421
+ add v27.8h, v7.8h, v19.8h422
+ sqxtun v4.8b, v24.8h423
+ sqxtun2 v4.16b, v25.8h424
+ sqxtun v5.8b, v26.8h425
+ sqxtun2 v5.16b, v27.8h426
+ st1 {v4.16b-v5.16b}, x0, x1427
+.endr428
+ cbnz w12, .loop_add_ps_32x\h429
+ ret430
+endfunc431
+.endm432
+433
+pixel_add_ps_32xN_neon 32434
+pixel_add_ps_32xN_neon 64435
+436
+function PFX(pixel_add_ps_64x64_neon)437
+ lsl x5, x5, #1438
+ sub x5, x5, #64439
+ mov w12, #32440
+.loop_add_ps_64x64:441
+ sub w12, w12, #1442
+.rept 2443
+ ld1 {v0.16b-v3.16b}, x2, x4444
+ ld1 {v16.8h-v19.8h}, x3, #64445
+ ld1 {v20.8h-v23.8h}, x3, x5446
+ uxtl v4.8h, v0.8b447
+ uxtl2 v5.8h, v0.16b448
+ uxtl v6.8h, v1.8b449
+ uxtl2 v7.8h, v1.16b450
+ uxtl v24.8h, v2.8b451
+ uxtl2 v25.8h, v2.16b452
+ uxtl v26.8h, v3.8b453
+ uxtl2 v27.8h, v3.16b454
+ add v0.8h, v4.8h, v16.8h455
+ add v1.8h, v5.8h, v17.8h456
+ add v2.8h, v6.8h, v18.8h457
+ add v3.8h, v7.8h, v19.8h458
+ add v4.8h, v24.8h, v20.8h459
+ add v5.8h, v25.8h, v21.8h460
+ add v6.8h, v26.8h, v22.8h461
+ add v7.8h, v27.8h, v23.8h462
+ sqxtun v0.8b, v0.8h463
+ sqxtun2 v0.16b, v1.8h464
+ sqxtun v1.8b, v2.8h465
+ sqxtun2 v1.16b, v3.8h466
+ sqxtun v2.8b, v4.8h467
+ sqxtun2 v2.16b, v5.8h468
+ sqxtun v3.8b, v6.8h469
+ sqxtun2 v3.16b, v7.8h470
+ st1 {v0.16b-v3.16b}, x0, x1471
+.endr472
+ cbnz w12, .loop_add_ps_64x64473
+ ret474
+endfunc475
+476
+// Chroma add_ps477
+function PFX(pixel_add_ps_4x8_neon)478
+ lsl x5, x5, #1479
+.rept 4480
+ ld1 {v0.8b}, x2, x4481
+ ld1 {v1.8b}, x2, x4482
+ ld1 {v2.4h}, x3, x5483
+ ld1 {v3.4h}, x3, x5484
+ uxtl v0.8h, v0.8b485
+ uxtl v1.8h, v1.8b486
+ add v4.8h, v0.8h, v2.8h487
+ add v5.8h, v1.8h, v3.8h488
+ sqxtun v4.8b, v4.8h489
+ sqxtun v5.8b, v5.8h490
+ st1 {v4.s}0, x0, x1491
+ st1 {v5.s}0, x0, x1492
+.endr493
+ ret494
+endfunc495
+496
+function PFX(pixel_add_ps_8x16_neon)497
+ lsl x5, x5, #1498
+.rept 8499
+ ld1 {v0.8b}, x2, x4500
+ ld1 {v1.8b}, x2, x4501
+ ld1 {v2.8h}, x3, x5502
+ ld1 {v3.8h}, x3, x5503
+ uxtl v0.8h, v0.8b504
+ uxtl v1.8h, v1.8b505
+ add v4.8h, v0.8h, v2.8h506
+ add v5.8h, v1.8h, v3.8h507
+ sqxtun v4.8b, v4.8h508
+ sqxtun v5.8b, v5.8h509
+ st1 {v4.8b}, x0, x1510
+ st1 {v5.8b}, x0, x1511
+.endr512
+ ret513
+endfunc514
+515
+// void scale1D_128to64(pixel *dst, const pixel *src)516
+function PFX(scale1D_128to64_neon)517
+.rept 2518
+ ld2 {v0.16b, v1.16b}, x1, #32519
+ ld2 {v2.16b, v3.16b}, x1, #32520
+ ld2 {v4.16b, v5.16b}, x1, #32521
+ ld2 {v6.16b, v7.16b}, x1, #32522
+ urhadd v0.16b, v0.16b, v1.16b523
+ urhadd v1.16b, v2.16b, v3.16b524
+ urhadd v2.16b, v4.16b, v5.16b525
+ urhadd v3.16b, v6.16b, v7.16b526
+ st1 {v0.16b-v3.16b}, x0, #64527
+.endr528
+ ret529
+endfunc530
+531
+.macro scale2D_1 v0, v1532
+ uaddlp \v0\().8h, \v0\().16b533
+ uaddlp \v1\().8h, \v1\().16b534
+ add \v0\().8h, \v0\().8h, \v1\().8h535
+.endm536
+537
+// void scale2D_64to32(pixel* dst, const pixel* src, intptr_t stride)538
+function PFX(scale2D_64to32_neon)539
+ mov w12, #32540
+.loop_scale2D:541
+ ld1 {v0.16b-v3.16b}, x1, x2542
+ sub w12, w12, #1543
+ ld1 {v4.16b-v7.16b}, x1, x2544
+ scale2D_1 v0, v4545
+ scale2D_1 v1, v5546
+ scale2D_1 v2, v6547
+ scale2D_1 v3, v7548
+ uqrshrn v0.8b, v0.8h, #2549
+ uqrshrn2 v0.16b, v1.8h, #2550
+ uqrshrn v1.8b, v2.8h, #2551
+ uqrshrn2 v1.16b, v3.8h, #2552
+ st1 {v0.16b-v1.16b}, x0, #32553
+ cbnz w12, .loop_scale2D554
+ ret555
+endfunc556
+557
+// void planecopy_cp_c(const uint8_t* src, intptr_t srcStride, pixel* dst, intptr_t dstStride, int width, int height, int shift)558
+function PFX(pixel_planecopy_cp_neon)559
+ dup v2.16b, w6560
+ sub x5, x5, #1561
+.loop_h:562
+ mov x6, x0563
+ mov x12, x2564
+ mov x7, #0565
+.loop_w:566
+ ldr q0, x6, #16567
+ ushl v0.16b, v0.16b, v2.16b568
+ str q0, x12, #16569
+ add x7, x7, #16570
+ cmp x7, x4571
+ blt .loop_w572
+573
+ add x0, x0, x1574
+ add x2, x2, x3575
+ sub x5, x5, #1576
+ cbnz x5, .loop_h577
+578
+// handle last row579
+ mov x5, x4580
+ lsr x5, x5, #3581
+.loopW8:582
+ ldr d0, x0, #8583
+ ushl v0.8b, v0.8b, v2.8b584
+ str d0, x2, #8585
+ sub x4, x4, #8586
+ sub x5, x5, #1587
+ cbnz x5, .loopW8588
+589
+ mov x5, #8590
+ sub x5, x5, x4591
+ sub x0, x0, x5592
+ sub x2, x2, x5593
+ ldr d0, x0594
+ ushl v0.8b, v0.8b, v2.8b595
+ str d0, x2596
+ ret597
+endfunc598
+599
+//******* satd *******600
+.macro satd_4x4_neon601
+ ld1 {v0.s}0, x0, x1602
+ ld1 {v0.s}1, x0, x1603
+ ld1 {v1.s}0, x2, x3604
+ ld1 {v1.s}1, x2, x3605
+ ld1 {v2.s}0, x0, x1606
+ ld1 {v2.s}1, x0, x1607
+ ld1 {v3.s}0, x2, x3608
+ ld1 {v3.s}1, x2, x3609
+610
+ usubl v4.8h, v0.8b, v1.8b611
+ usubl v5.8h, v2.8b, v3.8b612
+613
+ add v6.8h, v4.8h, v5.8h614
+ sub v7.8h, v4.8h, v5.8h615
+616
+ mov v4.d0, v6.d1617
+ add v0.4h, v6.4h, v4.4h618
+ sub v2.4h, v6.4h, v4.4h619
+620
+ mov v5.d0, v7.d1621
+ add v1.4h, v7.4h, v5.4h622
+ sub v3.4h, v7.4h, v5.4h623
+624
+ trn1 v4.4h, v0.4h, v1.4h625
+ trn2 v5.4h, v0.4h, v1.4h626
+627
+ trn1 v6.4h, v2.4h, v3.4h628
+ trn2 v7.4h, v2.4h, v3.4h629
+630
+ add v0.4h, v4.4h, v5.4h631
+ sub v1.4h, v4.4h, v5.4h632
+633
+ add v2.4h, v6.4h, v7.4h634
+ sub v3.4h, v6.4h, v7.4h635
+636
+ trn1 v4.2s, v0.2s, v1.2s637
+ trn2 v5.2s, v0.2s, v1.2s638
+639
+ trn1 v6.2s, v2.2s, v3.2s640
+ trn2 v7.2s, v2.2s, v3.2s641
+642
+ abs v4.4h, v4.4h643
+ abs v5.4h, v5.4h644
+ abs v6.4h, v6.4h645
+ abs v7.4h, v7.4h646
+647
+ smax v1.4h, v4.4h, v5.4h648
+ smax v2.4h, v6.4h, v7.4h649
+650
+ add v0.4h, v1.4h, v2.4h651
+ uaddlp v0.2s, v0.4h652
+ uaddlp v0.1d, v0.2s653
+.endm654
+655
+// int satd_4x4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)656
+function PFX(pixel_satd_4x4_neon)657
+ satd_4x4_neon658
+ fmov x0, d0659
+ ret660
+endfunc661
+662
.macro x265_satd_4x8_8x4_end_neon663
add v0.8h, v4.8h, v6.8h664
add v1.8h, v5.8h, v7.8h665
666
.endm667
668
.macro pixel_satd_4x8_neon669
- ld1r {v1.2s}, x2, x3670
+ ld1r {v1.2s}, x2, x3671
ld1r {v0.2s}, x0, x1672
ld1r {v3.2s}, x2, x3673
ld1r {v2.2s}, x0, x1674
675
sub v5.8h, v0.8h, v1.8h676
ld1 {v6.s}1, x0, x1677
usubl v3.8h, v6.8b, v7.8b678
- add v6.8h, v2.8h, v3.8h679
- sub v7.8h, v2.8h, v3.8h680
+ add v6.8h, v2.8h, v3.8h681
+ sub v7.8h, v2.8h, v3.8h682
x265_satd_4x8_8x4_end_neon683
.endm684
685
-// template<int w, int h>686
-// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)687
-function x265_pixel_satd_4x8_neon688
- pixel_satd_4x8_neon689
- mov w0, v0.s0690
- ret691
+// template<int w, int h>692
+// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)693
+function PFX(pixel_satd_4x8_neon)694
+ pixel_satd_4x8_neon695
+ mov w0, v0.s0696
+ ret697
+endfunc698
+699
+function PFX(pixel_satd_4x16_neon)700
+ mov w4, #0701
+ pixel_satd_4x8_neon702
+ mov w5, v0.s0703
+ add w4, w4, w5704
+ pixel_satd_4x8_neon705
+ mov w5, v0.s0706
+ add w0, w5, w4707
+ ret708
+endfunc709
+710
+function PFX(pixel_satd_4x32_neon)711
+ mov w4, #0712
+.rept 4713
+ pixel_satd_4x8_neon714
+ mov w5, v0.s0715
+ add w4, w4, w5716
+.endr717
+ mov w0, w4718
+ ret719
+endfunc720
+721
+function PFX(pixel_satd_12x16_neon)722
+ mov x4, x0723
+ mov x5, x2724
+ mov w7, #0725
+ pixel_satd_4x8_neon726
+ mov w6, v0.s0727
+ add w7, w7, w6728
+ pixel_satd_4x8_neon729
+ mov w6, v0.s0730
+ add w7, w7, w6731
+732
+ add x0, x4, #4733
+ add x2, x5, #4734
+ pixel_satd_4x8_neon735
+ mov w6, v0.s0736
+ add w7, w7, w6737
+ pixel_satd_4x8_neon738
+ mov w6, v0.s0739
+ add w7, w7, w6740
+741
+ add x0, x4, #8742
+ add x2, x5, #8743
+ pixel_satd_4x8_neon744
+ mov w6, v0.s0745
+ add w7, w7, w6746
+ pixel_satd_4x8_neon747
+ mov w6, v0.s0748
+ add w0, w7, w6749
+ ret750
+endfunc751
+752
+function PFX(pixel_satd_12x32_neon)753
+ mov x4, x0754
+ mov x5, x2755
+ mov w7, #0756
+.rept 4757
+ pixel_satd_4x8_neon758
+ mov w6, v0.s0759
+ add w7, w7, w6760
+.endr761
+762
+ add x0, x4, #4763
+ add x2, x5, #4764
+.rept 4765
+ pixel_satd_4x8_neon766
+ mov w6, v0.s0767
+ add w7, w7, w6768
+.endr769
+770
+ add x0, x4, #8771
+ add x2, x5, #8772
+.rept 4773
+ pixel_satd_4x8_neon774
+ mov w6, v0.s0775
+ add w7, w7, w6776
+.endr777
+778
+ mov w0, w7779
+ ret780
+endfunc781
+782
+function PFX(pixel_satd_8x4_neon)783
+ mov x4, x0784
+ mov x5, x2785
+ satd_4x4_neon786
+ add x0, x4, #4787
+ add x2, x5, #4788
+ umov x6, v0.d0789
+ satd_4x4_neon790
+ umov x0, v0.d0791
+ add x0, x0, x6792
+ ret793
+endfunc794
+795
+.macro LOAD_DIFF_8x4 v0 v1 v2 v3796
+ ld1 {v0.8b}, x0, x1797
+ ld1 {v1.8b}, x2, x3798
+ ld1 {v2.8b}, x0, x1799
+ ld1 {v3.8b}, x2, x3800
+ ld1 {v4.8b}, x0, x1801
+ ld1 {v5.8b}, x2, x3802
+ ld1 {v6.8b}, x0, x1803
+ ld1 {v7.8b}, x2, x3804
+ usubl \v0, v0.8b, v1.8b805
+ usubl \v1, v2.8b, v3.8b806
+ usubl \v2, v4.8b, v5.8b807
+ usubl \v3, v6.8b, v7.8b808
+.endm809
+810
+.macro LOAD_DIFF_16x4 v0 v1 v2 v3 v4 v5 v6 v7811
+ ld1 {v0.16b}, x0, x1812
+ ld1 {v1.16b}, x2, x3813
+ ld1 {v2.16b}, x0, x1814
+ ld1 {v3.16b}, x2, x3815
+ ld1 {v4.16b}, x0, x1816
+ ld1 {v5.16b}, x2, x3817
+ ld1 {v6.16b}, x0, x1818
+ ld1 {v7.16b}, x2, x3819
+ usubl \v0, v0.8b, v1.8b820
+ usubl \v1, v2.8b, v3.8b821
+ usubl \v2, v4.8b, v5.8b822
+ usubl \v3, v6.8b, v7.8b823
+ usubl2 \v4, v0.16b, v1.16b824
+ usubl2 \v5, v2.16b, v3.16b825
+ usubl2 \v6, v4.16b, v5.16b826
+ usubl2 \v7, v6.16b, v7.16b827
+.endm828
+829
+function PFX(satd_16x4_neon), export=0830
+ LOAD_DIFF_16x4 v16.8h, v17.8h, v18.8h, v19.8h, v20.8h, v21.8h, v22.8h, v23.8h831
+ b PFX(satd_8x4v_8x8h_neon)832
+endfunc833
+834
+function PFX(satd_8x8_neon), export=0835
+ LOAD_DIFF_8x4 v16.8h, v17.8h, v18.8h, v19.8h836
+ LOAD_DIFF_8x4 v20.8h, v21.8h, v22.8h, v23.8h837
+ b PFX(satd_8x4v_8x8h_neon)838
+endfunc839
+840
+// one vertical hadamard pass and two horizontal841
+function PFX(satd_8x4v_8x8h_neon), export=0842
+ HADAMARD4_V v16.8h, v18.8h, v17.8h, v19.8h, v0.8h, v2.8h, v1.8h, v3.8h843
+ HADAMARD4_V v20.8h, v21.8h, v22.8h, v23.8h, v0.8h, v1.8h, v2.8h, v3.8h844
+ trn4 v0.8h, v1.8h, v2.8h, v3.8h, v16.8h, v17.8h, v18.8h, v19.8h845
+ trn4 v4.8h, v5.8h, v6.8h, v7.8h, v20.8h, v21.8h, v22.8h, v23.8h846
+ SUMSUB_ABCD v16.8h, v17.8h, v18.8h, v19.8h, v0.8h, v1.8h, v2.8h, v3.8h847
+ SUMSUB_ABCD v20.8h, v21.8h, v22.8h, v23.8h, v4.8h, v5.8h, v6.8h, v7.8h848
+ trn4 v0.4s, v2.4s, v1.4s, v3.4s, v16.4s, v18.4s, v17.4s, v19.4s849
+ trn4 v4.4s, v6.4s, v5.4s, v7.4s, v20.4s, v22.4s, v21.4s, v23.4s850
+ ABS8 v0.8h, v1.8h, v2.8h, v3.8h, v4.8h, v5.8h, v6.8h, v7.8h851
+ smax v0.8h, v0.8h, v2.8h852
+ smax v1.8h, v1.8h, v3.8h853
+ smax v2.8h, v4.8h, v6.8h854
+ smax v3.8h, v5.8h, v7.8h855
+ ret856
+endfunc857
+858
+function PFX(pixel_satd_8x8_neon)859
+ mov x10, x30860
+ bl PFX(satd_8x8_neon)861
+ add v0.8h, v0.8h, v1.8h862
+ add v1.8h, v2.8h, v3.8h863
+ add v0.8h, v0.8h, v1.8h864
+ uaddlv s0, v0.8h865
+ mov w0, v0.s0866
+ ret x10867
+endfunc868
+869
+function PFX(pixel_satd_8x12_neon)870
+ mov x4, x0871
+ mov x5, x2872
+ mov x7, #0873
+ satd_4x4_neon874
+ umov x6, v0.d0875
+ add x7, x7, x6876
+ add x0, x4, #4877
+ add x2, x5, #4878
+ satd_4x4_neon879
+ umov x6, v0.d0880
+ add x7, x7, x6881
+.rept 2882
+ sub x0, x0, #4883
+ sub x2, x2, #4884
+ mov x4, x0885
+ mov x5, x2886
+ satd_4x4_neon887
+ umov x6, v0.d0888
+ add x7, x7, x6889
+ add x0, x4, #4890
+ add x2, x5, #4891
+ satd_4x4_neon892
+ umov x6, v0.d0893
+ add x7, x7, x6894
+.endr895
+ mov x0, x7896
+ ret897
+endfunc898
+899
+function PFX(pixel_satd_8x16_neon)900
+ mov x10, x30901
+ bl PFX(satd_8x8_neon)902
+ add v30.8h, v0.8h, v1.8h903
+ add v31.8h, v2.8h, v3.8h904
+ bl PFX(satd_8x8_neon)905
+ add v30.8h, v30.8h, v0.8h906
+ add v31.8h, v31.8h, v1.8h907
+ add v30.8h, v30.8h, v2.8h908
+ add v31.8h, v31.8h, v3.8h909
+ add v0.8h, v30.8h, v31.8h910
+ uaddlv s0, v0.8h911
+ mov w0, v0.s0912
+ ret x10913
+endfunc914
+915
+function PFX(pixel_satd_8x32_neon)916
+ mov x10, x30917
+ bl PFX(satd_8x8_neon)918
+ add v30.8h, v0.8h, v1.8h919
+ add v31.8h, v2.8h, v3.8h920
+.rept 3921
+ bl PFX(satd_8x8_neon)922
+ add v30.8h, v30.8h, v0.8h923
+ add v31.8h, v31.8h, v1.8h924
+ add v30.8h, v30.8h, v2.8h925
+ add v31.8h, v31.8h, v3.8h926
+.endr927
+ add v0.8h, v30.8h, v31.8h928
+ uaddlv s0, v0.8h929
+ mov w0, v0.s0930
+ ret x10931
+endfunc932
+933
+function PFX(pixel_satd_8x64_neon)934
+ mov x10, x30935
+ bl PFX(satd_8x8_neon)936
+ add v30.8h, v0.8h, v1.8h937
+ add v31.8h, v2.8h, v3.8h938
+.rept 7939
+ bl PFX(satd_8x8_neon)940
+ add v30.8h, v30.8h, v0.8h941
+ add v31.8h, v31.8h, v1.8h942
+ add v30.8h, v30.8h, v2.8h943
+ add v31.8h, v31.8h, v3.8h944
+.endr945
+ add v0.8h, v30.8h, v31.8h946
+ uaddlv s0, v0.8h947
+ mov w0, v0.s0948
+ ret x10949
+endfunc950
+951
+function PFX(pixel_satd_16x4_neon)952
+ mov x10, x30953
+ bl PFX(satd_16x4_neon)954
+ add v30.8h, v0.8h, v1.8h955
+ add v31.8h, v2.8h, v3.8h956
+ add v0.8h, v30.8h, v31.8h957
+ uaddlv s0, v0.8h958
+ mov w0, v0.s0959
+ ret x10960
+endfunc961
+962
+function PFX(pixel_satd_16x8_neon)963
+ mov x10, x30964
+ bl PFX(satd_16x4_neon)965
+ add v30.8h, v0.8h, v1.8h966
+ add v31.8h, v2.8h, v3.8h967
+ bl PFX(satd_16x4_neon)968
+ add v30.8h, v30.8h, v0.8h969
+ add v31.8h, v31.8h, v1.8h970
+ add v30.8h, v30.8h, v2.8h971
+ add v31.8h, v31.8h, v3.8h972
+ add v0.8h, v30.8h, v31.8h973
+ uaddlv s0, v0.8h974
+ mov w0, v0.s0975
+ ret x10976
+endfunc977
+978
+function PFX(pixel_satd_16x12_neon)979
+ mov x10, x30980
+ bl PFX(satd_16x4_neon)981
+ add v30.8h, v0.8h, v1.8h982
+ add v31.8h, v2.8h, v3.8h983
+.rept 2984
+ bl PFX(satd_16x4_neon)985
+ add v30.8h, v30.8h, v0.8h986
+ add v31.8h, v31.8h, v1.8h987
+ add v30.8h, v30.8h, v2.8h988
+ add v31.8h, v31.8h, v3.8h989
+.endr990
+ add v0.8h, v30.8h, v31.8h991
+ uaddlv s0, v0.8h992
+ mov w0, v0.s0993
+ ret x10994
+endfunc995
+996
+function PFX(pixel_satd_16x16_neon)997
+ mov x10, x30998
+ bl PFX(satd_16x4_neon)999
+ add v30.8h, v0.8h, v1.8h1000
+ add v31.8h, v2.8h, v3.8h1001
+.rept 31002
+ bl PFX(satd_16x4_neon)1003
+ add v30.8h, v30.8h, v0.8h1004
+ add v31.8h, v31.8h, v1.8h1005
+ add v30.8h, v30.8h, v2.8h1006
+ add v31.8h, v31.8h, v3.8h1007
+.endr1008
+ add v0.8h, v30.8h, v31.8h1009
+ uaddlv s0, v0.8h1010
+ mov w0, v0.s01011
+ ret x101012
+endfunc1013
+1014
+function PFX(pixel_satd_16x24_neon)1015
+ mov x10, x301016
+ bl PFX(satd_16x4_neon)1017
+ add v30.8h, v0.8h, v1.8h1018
+ add v31.8h, v2.8h, v3.8h1019
+.rept 51020
+ bl PFX(satd_16x4_neon)1021
+ add v30.8h, v30.8h, v0.8h1022
+ add v31.8h, v31.8h, v1.8h1023
+ add v30.8h, v30.8h, v2.8h1024
+ add v31.8h, v31.8h, v3.8h1025
+.endr1026
+ add v0.8h, v30.8h, v31.8h1027
+ uaddlv s0, v0.8h1028
+ mov w0, v0.s01029
+ ret x101030
+endfunc1031
+1032
+.macro pixel_satd_16x32_neon1033
+ bl PFX(satd_16x4_neon)1034
+ add v30.8h, v0.8h, v1.8h1035
+ add v31.8h, v2.8h, v3.8h1036
+.rept 71037
+ bl PFX(satd_16x4_neon)1038
+ add v30.8h, v30.8h, v0.8h1039
+ add v31.8h, v31.8h, v1.8h1040
+ add v30.8h, v30.8h, v2.8h1041
+ add v31.8h, v31.8h, v3.8h1042
+.endr1043
+.endm1044
+1045
+function PFX(pixel_satd_16x32_neon)1046
+ mov x10, x301047
+ pixel_satd_16x32_neon1048
+ add v0.8h, v30.8h, v31.8h1049
+ uaddlv s0, v0.8h1050
+ mov w0, v0.s01051
+ ret x101052
endfunc1053
1054
-// template<int w, int h>1055
-// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)1056
-function x265_pixel_satd_4x16_neon1057
- eor w4, w4, w41058
- pixel_satd_4x8_neon1059
- mov w5, v0.s01060
- add w4, w4, w51061
- pixel_satd_4x8_neon1062
- mov w5, v0.s01063
- add w0, w5, w41064
- ret1065
+function PFX(pixel_satd_16x64_neon)1066
+ mov x10, x301067
+ bl PFX(satd_16x4_neon)1068
+ add v30.8h, v0.8h, v1.8h1069
+ add v31.8h, v2.8h, v3.8h1070
+.rept 151071
+ bl PFX(satd_16x4_neon)1072
+ add v30.8h, v30.8h, v0.8h1073
+ add v31.8h, v31.8h, v1.8h1074
+ add v30.8h, v30.8h, v2.8h1075
+ add v31.8h, v31.8h, v3.8h1076
+.endr1077
+ add v0.8h, v30.8h, v31.8h1078
+ uaddlv s0, v0.8h1079
+ mov w0, v0.s01080
+ ret x101081
endfunc1082
1083
-// template<int w, int h>1084
-// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)1085
-function x265_pixel_satd_4x32_neon1086
- eor w4, w4, w41087
+function PFX(pixel_satd_24x32_neon)1088
+ mov x10, x301089
+ mov x7, #01090
+ mov x4, x01091
+ mov x5, x21092
+.rept 31093
+ movi v30.8h, #01094
+ movi v31.8h, #01095
.rept 41096
- pixel_satd_4x8_neon1097
- mov w5, v0.s01098
- add w4, w4, w51099
+ bl PFX(satd_8x8_neon)1100
+ add v30.8h, v30.8h, v0.8h1101
+ add v31.8h, v31.8h, v1.8h1102
+ add v30.8h, v30.8h, v2.8h1103
+ add v31.8h, v31.8h, v3.8h1104
.endr1105
- mov w0, w41106
- ret1107
+ add v0.8h, v30.8h, v31.8h1108
+ uaddlv s0, v0.8h1109
+ mov w6, v0.s01110
+ add x7, x7, x61111
+ add x4, x4, #81112
+ add x5, x5, #81113
+ mov x0, x41114
+ mov x2, x51115
+.endr1116
+ mov x0, x71117
+ ret x101118
endfunc1119
1120
-// template<int w, int h>1121
-// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)1122
-function x265_pixel_satd_12x16_neon1123
+function PFX(pixel_satd_24x64_neon)1124
+ mov x10, x301125
+ mov x7, #01126
mov x4, x01127
mov x5, x21128
- eor w7, w7, w71129
- pixel_satd_4x8_neon1130
+.rept 31131
+ movi v30.8h, #01132
+ movi v31.8h, #01133
+.rept 41134
+ bl PFX(satd_8x8_neon)1135
+ add v30.8h, v30.8h, v0.8h1136
+ add v31.8h, v31.8h, v1.8h1137
+ add v30.8h, v30.8h, v2.8h1138
+ add v31.8h, v31.8h, v3.8h1139
+.endr1140
+ add v0.8h, v30.8h, v31.8h1141
+ uaddlv s0, v0.8h1142
mov w6, v0.s01143
- add w7, w7, w61144
- pixel_satd_4x8_neon1145
+ add x7, x7, x61146
+ add x4, x4, #81147
+ add x5, x5, #81148
+ mov x0, x41149
+ mov x2, x51150
+.endr1151
+ sub x4, x4, #241152
+ sub x5, x5, #241153
+ add x0, x4, x1, lsl #51154
+ add x2, x5, x3, lsl #51155
+ mov x4, x01156
+ mov x5, x21157
+.rept 31158
+ movi v30.8h, #01159
+ movi v31.8h, #01160
+.rept 41161
+ bl PFX(satd_8x8_neon)1162
+ add v30.8h, v30.8h, v0.8h1163
+ add v31.8h, v31.8h, v1.8h1164
+ add v30.8h, v30.8h, v2.8h1165
+ add v31.8h, v31.8h, v3.8h1166
+.endr1167
+ add v0.8h, v30.8h, v31.8h1168
+ uaddlv s0, v0.8h1169
mov w6, v0.s01170
- add w7, w7, w61171
+ add x7, x7, x61172
+ add x4, x4, #81173
+ add x5, x5, #81174
+ mov x0, x41175
+ mov x2, x51176
+.endr1177
+ mov x0, x71178
+ ret x101179
+endfunc1180
1181
- add x0, x4, #41182
- add x2, x5, #41183
- pixel_satd_4x8_neon1184
- mov w6, v0.s01185
- add w7, w7, w61186
- pixel_satd_4x8_neon1187
- mov w6, v0.s01188
- add w7, w7, w61189
+.macro pixel_satd_32x81190
+ mov x4, x01191
+ mov x5, x21192
+.rept 21193
+ bl PFX(satd_16x4_neon)1194
+ add v30.8h, v30.8h, v0.8h1195
+ add v31.8h, v31.8h, v1.8h1196
+ add v30.8h, v30.8h, v2.8h1197
+ add v31.8h, v31.8h, v3.8h1198
+.endr1199
+ add x0, x4, #161200
+ add x2, x5, #161201
+.rept 21202
+ bl PFX(satd_16x4_neon)1203
+ add v30.8h, v30.8h, v0.8h1204
+ add v31.8h, v31.8h, v1.8h1205
+ add v30.8h, v30.8h, v2.8h1206
+ add v31.8h, v31.8h, v3.8h1207
+.endr1208
+.endm1209
1210
- add x0, x4, #81211
- add x2, x5, #81212
- pixel_satd_4x8_neon1213
- mov w6, v0.s01214
- add w7, w7, w61215
- pixel_satd_4x8_neon1216
+.macro satd_32x16_neon1217
+ movi v30.8h, #01218
+ movi v31.8h, #01219
+ pixel_satd_32x81220
+ sub x0, x0, #161221
+ sub x2, x2, #161222
+ pixel_satd_32x81223
+ add v0.8h, v30.8h, v31.8h1224
+ uaddlv s0, v0.8h1225
mov w6, v0.s01226
- add w0, w7, w61227
- ret1228
-endfunc1229
+.endm1230
1231
-// template<int w, int h>1232
-// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)1233
-function x265_pixel_satd_12x32_neon1234
+.macro satd_64x16_neon1235
+ mov x8, x01236
+ mov x9, x21237
+ satd_32x16_neon1238
+ add x7, x7, x61239
+ add x0, x8, #321240
+ add x2, x9, #321241
+ satd_32x16_neon1242
+ add x7, x7, x61243
+.endm1244
+1245
+function PFX(pixel_satd_32x8_neon)1246
+ mov x10, x301247
+ mov x7, #01248
mov x4, x01249
mov x5, x21250
- eor w7, w7, w71251
-.rept 41252
- pixel_satd_4x8_neon1253
- mov w6, v0.s01254
- add w7, w7, w61255
+ movi v30.8h, #01256
+ movi v31.8h, #01257
+ pixel_satd_32x81258
+ add v0.8h, v30.8h, v31.8h1259
+ uaddlv s0, v0.8h1260
+ mov w0, v0.s01261
+ ret x101262
+endfunc1263
+1264
+function PFX(pixel_satd_32x16_neon)1265
+ mov x10, x301266
+ satd_32x16_neon1267
+ mov x0, x61268
+ ret x101269
+endfunc1270
+1271
+function PFX(pixel_satd_32x24_neon)1272
+ mov x10, x301273
+ satd_32x16_neon1274
+ movi v30.8h, #01275
+ movi v31.8h, #01276
+ sub x0, x0, #161277
+ sub x2, x2, #161278
+ pixel_satd_32x81279
+ add v0.8h, v30.8h, v31.8h1280
+ uaddlv s0, v0.8h1281
+ mov w0, v0.s01282
+ add x0, x0, x61283
+ ret x101284
+endfunc1285
+1286
+function PFX(pixel_satd_32x32_neon)1287
+ mov x10, x301288
+ mov x7, #01289
+ satd_32x16_neon1290
+ sub x0, x0, #161291
+ sub x2, x2, #161292
+ add x7, x7, x61293
+ satd_32x16_neon1294
+ add x0, x7, x61295
+ ret x101296
+endfunc1297
+1298
+function PFX(pixel_satd_32x48_neon)1299
+ mov x10, x301300
+ mov x7, #01301
+.rept 21302
+ satd_32x16_neon1303
+ sub x0, x0, #161304
+ sub x2, x2, #161305
+ add x7, x7, x61306
.endr1307
+ satd_32x16_neon1308
+ add x0, x7, x61309
+ ret x101310
+endfunc1311
1312
- add x0, x4, #41313
- add x2, x5, #41314
-.rept 41315
- pixel_satd_4x8_neon1316
- mov w6, v0.s01317
- add w7, w7, w61318
+function PFX(pixel_satd_32x64_neon)1319
+ mov x10, x301320
+ mov x7, #01321
+.rept 31322
+ satd_32x16_neon1323
+ sub x0, x0, #161324
+ sub x2, x2, #161325
+ add x7, x7, x61326
.endr1327
+ satd_32x16_neon1328
+ add x0, x7, x61329
+ ret x101330
+endfunc1331
1332
- add x0, x4, #81333
- add x2, x5, #81334
-.rept 41335
- pixel_satd_4x8_neon1336
- mov w6, v0.s01337
- add w7, w7, w61338
+function PFX(pixel_satd_64x16_neon)1339
+ mov x10, x301340
+ mov x7, #01341
+ satd_64x16_neon1342
+ mov x0, x71343
+ ret x101344
+endfunc1345
+1346
+function PFX(pixel_satd_64x32_neon)1347
+ mov x10, x301348
+ mov x7, #01349
+ satd_64x16_neon1350
+ sub x0, x0, #481351
+ sub x2, x2, #481352
+ satd_64x16_neon1353
+ mov x0, x71354
+ ret x101355
+endfunc1356
+1357
+function PFX(pixel_satd_64x48_neon)1358
+ mov x10, x301359
+ mov x7, #01360
+.rept 21361
+ satd_64x16_neon1362
+ sub x0, x0, #481363
+ sub x2, x2, #481364
.endr1365
+ satd_64x16_neon1366
+ mov x0, x71367
+ ret x101368
+endfunc1369
1370
- mov w0, w71371
+function PFX(pixel_satd_64x64_neon)1372
+ mov x10, x301373
+ mov x7, #01374
+.rept 31375
+ satd_64x16_neon1376
+ sub x0, x0, #481377
+ sub x2, x2, #481378
+.endr1379
+ satd_64x16_neon1380
+ mov x0, x71381
+ ret x101382
+endfunc1383
+1384
+function PFX(pixel_satd_48x64_neon)1385
+ mov x10, x301386
+ mov x7, #01387
+ mov x8, x01388
+ mov x9, x21389
+.rept 31390
+ satd_32x16_neon1391
+ sub x0, x0, #161392
+ sub x2, x2, #161393
+ add x7, x7, x61394
+.endr1395
+ satd_32x16_neon1396
+ add x7, x7, x61397
+1398
+ add x0, x8, #321399
+ add x2, x9, #321400
+ pixel_satd_16x32_neon1401
+ add v0.8h, v30.8h, v31.8h1402
+ uaddlv s0, v0.8h1403
+ mov w6, v0.s01404
+ add x7, x7, x61405
+1406
+ movi v30.8h, #01407
+ movi v31.8h, #01408
+ pixel_satd_16x32_neon1409
+ add v0.8h, v30.8h, v31.8h1410
+ uaddlv s0, v0.8h1411
+ mov w6, v0.s01412
+ add x0, x7, x61413
+ ret x101414
+endfunc1415
+1416
+function PFX(sa8d_8x8_neon), export=01417
+ LOAD_DIFF_8x4 v16.8h, v17.8h, v18.8h, v19.8h1418
+ LOAD_DIFF_8x4 v20.8h, v21.8h, v22.8h, v23.8h1419
+ HADAMARD4_V v16.8h, v18.8h, v17.8h, v19.8h, v0.8h, v2.8h, v1.8h, v3.8h1420
+ HADAMARD4_V v20.8h, v21.8h, v22.8h, v23.8h, v0.8h, v1.8h, v2.8h, v3.8h1421
+ SUMSUB_ABCD v0.8h, v16.8h, v1.8h, v17.8h, v16.8h, v20.8h, v17.8h, v21.8h1422
+ SUMSUB_ABCD v2.8h, v18.8h, v3.8h, v19.8h, v18.8h, v22.8h, v19.8h, v23.8h1423
+ trn4 v4.8h, v5.8h, v6.8h, v7.8h, v0.8h, v1.8h, v2.8h, v3.8h1424
+ trn4 v20.8h, v21.8h, v22.8h, v23.8h, v16.8h, v17.8h, v18.8h, v19.8h1425
+ SUMSUB_ABCD v2.8h, v3.8h, v24.8h, v25.8h, v20.8h, v21.8h, v4.8h, v5.8h1426
+ SUMSUB_ABCD v0.8h, v1.8h, v4.8h, v5.8h, v22.8h, v23.8h, v6.8h, v7.8h1427
+ trn4 v20.4s, v22.4s, v21.4s, v23.4s, v2.4s, v0.4s, v3.4s, v1.4s1428
+ trn4 v16.4s, v18.4s, v17.4s, v19.4s, v24.4s, v4.4s, v25.4s, v5.4s1429
+ SUMSUB_ABCD v0.8h, v2.8h, v1.8h, v3.8h, v20.8h, v22.8h, v21.8h, v23.8h1430
+ SUMSUB_ABCD v4.8h, v6.8h, v5.8h, v7.8h, v16.8h, v18.8h, v17.8h, v19.8h1431
+ trn4 v16.2d, v20.2d, v17.2d, v21.2d, v0.2d, v4.2d, v1.2d, v5.2d1432
+ trn4 v18.2d, v22.2d, v19.2d, v23.2d, v2.2d, v6.2d, v3.2d, v7.2d1433
+ ABS8 v16.8h, v17.8h, v18.8h, v19.8h, v20.8h, v21.8h, v22.8h, v23.8h1434
+ smax v16.8h, v16.8h, v20.8h1435
+ smax v17.8h, v17.8h, v21.8h1436
+ smax v18.8h, v18.8h, v22.8h1437
+ smax v19.8h, v19.8h, v23.8h1438
+ add v0.8h, v16.8h, v17.8h1439
+ add v1.8h, v18.8h, v19.8h1440
ret1441
endfunc1442
1443
-// template<int w, int h>1444
-// int satd4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)1445
-function x265_pixel_satd_8x8_neon1446
- eor w4, w4, w41447
- mov x6, x01448
- mov x7, x21449
- pixel_satd_4x8_neon1450
- mov w5, v0.s01451
- add w4, w4, w51452
- add x0, x6, #41453
- add x2, x7, #41454
- pixel_satd_4x8_neon1455
+function PFX(pixel_sa8d_8x8_neon)1456
+ mov x10, x301457
+ bl PFX(sa8d_8x8_neon)1458
+ add v0.8h, v0.8h, v1.8h1459
+ uaddlv s0, v0.8h1460
+ mov w0, v0.s01461
+ add w0, w0, #11462
+ lsr w0, w0, #11463
+ ret x101464
+endfunc1465
+1466
+function PFX(pixel_sa8d_8x16_neon)1467
+ mov x10, x301468
+ bl PFX(sa8d_8x8_neon)1469
+ add v0.8h, v0.8h, v1.8h1470
+ uaddlv s0, v0.8h1471
mov w5, v0.s01472
+ add w5, w5, #11473
+ lsr w5, w5, #11474
+ bl PFX(sa8d_8x8_neon)1475
+ add v0.8h, v0.8h, v1.8h1476
+ uaddlv s0, v0.8h1477
+ mov w4, v0.s01478
+ add w4, w4, #11479
+ lsr w4, w4, #11480
+ add w0, w4, w51481
+ ret x101482
+endfunc1483
+1484
+.macro sa8d_16x16 reg1485
+ bl PFX(sa8d_8x8_neon)1486
+ uaddlp v30.4s, v0.8h1487
+ uaddlp v31.4s, v1.8h1488
+ bl PFX(sa8d_8x8_neon)1489
+ uadalp v30.4s, v0.8h1490
+ uadalp v31.4s, v1.8h1491
+ sub x0, x0, x1, lsl #41492
+ sub x2, x2, x3, lsl #41493
+ add x0, x0, #81494
+ add x2, x2, #81495
+ bl PFX(sa8d_8x8_neon)1496
+ uadalp v30.4s, v0.8h1497
+ uadalp v31.4s, v1.8h1498
+ bl PFX(sa8d_8x8_neon)1499
+ uadalp v30.4s, v0.8h1500
+ uadalp v31.4s, v1.8h1501
+ add v0.4s, v30.4s, v31.4s1502
+ addv s0, v0.4s1503
+ mov \reg, v0.s01504
+ add \reg, \reg, #11505
+ lsr \reg, \reg, #11506
+.endm1507
+1508
+function PFX(pixel_sa8d_16x16_neon)1509
+ mov x10, x301510
+ sa8d_16x16 w01511
+ ret x101512
+endfunc1513
+1514
+function PFX(pixel_sa8d_16x32_neon)1515
+ mov x10, x301516
+ sa8d_16x16 w41517
+ sub x0, x0, #81518
+ sub x2, x2, #81519
+ sa8d_16x16 w51520
add w0, w4, w51521
+ ret x101522
+endfunc1523
+1524
+function PFX(pixel_sa8d_32x32_neon)1525
+ mov x10, x301526
+ sa8d_16x16 w41527
+ sub x0, x0, x1, lsl #41528
+ sub x2, x2, x3, lsl #41529
+ add x0, x0, #81530
+ add x2, x2, #81531
+ sa8d_16x16 w51532
+ sub x0, x0, #241533
+ sub x2, x2, #241534
+ sa8d_16x16 w61535
+ sub x0, x0, x1, lsl #41536
+ sub x2, x2, x3, lsl #41537
+ add x0, x0, #81538
+ add x2, x2, #81539
+ sa8d_16x16 w71540
+ add w4, w4, w51541
+ add w6, w6, w71542
+ add w0, w4, w61543
+ ret x101544
+endfunc1545
+1546
+function PFX(pixel_sa8d_32x64_neon)1547
+ mov x10, x301548
+ mov w11, #41549
+ mov w9, #01550
+.loop_sa8d_32:1551
+ sub w11, w11, #11552
+ sa8d_16x16 w41553
+ sub x0, x0, x1, lsl #41554
+ sub x2, x2, x3, lsl #41555
+ add x0, x0, #81556
+ add x2, x2, #81557
+ sa8d_16x16 w51558
+ add w4, w4, w51559
+ add w9, w9, w41560
+ sub x0, x0, #241561
+ sub x2, x2, #241562
+ cbnz w11, .loop_sa8d_321563
+ mov w0, w91564
+ ret x101565
+endfunc1566
+1567
+function PFX(pixel_sa8d_64x64_neon)1568
+ mov x10, x301569
+ mov w11, #41570
+ mov w9, #01571
+.loop_sa8d_64:1572
+ sub w11, w11, #11573
+ sa8d_16x16 w41574
+ sub x0, x0, x1, lsl #41575
+ sub x2, x2, x3, lsl #41576
+ add x0, x0, #81577
+ add x2, x2, #81578
+ sa8d_16x16 w51579
+ sub x0, x0, x1, lsl #41580
+ sub x2, x2, x3, lsl #41581
+ add x0, x0, #81582
+ add x2, x2, #81583
+ sa8d_16x16 w61584
+ sub x0, x0, x1, lsl #41585
+ sub x2, x2, x3, lsl #41586
+ add x0, x0, #81587
+ add x2, x2, #81588
+ sa8d_16x16 w71589
+ add w4, w4, w51590
+ add w6, w6, w71591
+ add w8, w4, w61592
+ add w9, w9, w81593
+1594
+ sub x0, x0, #561595
+ sub x2, x2, #561596
+ cbnz w11, .loop_sa8d_641597
+ mov w0, w91598
+ ret x101599
+endfunc1600
+1601
+/***** dequant_scaling*****/1602
+// void dequant_scaling_c(const int16_t* quantCoef, const int32_t* deQuantCoef, int16_t* coef, int num, int per, int shift)1603
+function PFX(dequant_scaling_neon)1604
+ add x5, x5, #4 // shift + 41605
+ lsr x3, x3, #3 // num / 81606
+ cmp x5, x41607
+ blt .dequant_skip1608
+1609
+ mov x12, #11610
+ sub x6, x5, x4 // shift - per1611
+ sub x6, x6, #1 // shift - per - 11612
+ lsl x6, x12, x6 // 1 << shift - per - 1 (add)1613
+ dup v0.4s, w61614
+ sub x7, x4, x5 // per - shift1615
+ dup v3.4s, w71616
+1617
+.dequant_loop1:1618
+ ld1 {v19.8h}, x0, #16 // quantCoef1619
+ ld1 {v2.4s}, x1, #16 // deQuantCoef1620
+ ld1 {v20.4s}, x1, #161621
+ sub x3, x3, #11622
+ sxtl v1.4s, v19.4h1623
+ sxtl2 v19.4s, v19.8h1624
+1625
+ mul v1.4s, v1.4s, v2.4s // quantCoef * deQuantCoef1626
+ mul v19.4s, v19.4s, v20.4s1627
+ add v1.4s, v1.4s, v0.4s // quantCoef * deQuantCoef + add1628
+ add v19.4s, v19.4s, v0.4s1629
+1630
+ sshl v1.4s, v1.4s, v3.4s1631
+ sshl v19.4s, v19.4s, v3.4s1632
+ sqxtn v16.4h, v1.4s // x265_clip31633
+ sqxtn2 v16.8h, v19.4s1634
+ st1 {v16.8h}, x2, #161635
+ cbnz x3, .dequant_loop11636
+ ret1637
+1638
+.dequant_skip:1639
+ sub x6, x4, x5 // per - shift1640
+ dup v0.8h, w61641
+1642
+.dequant_loop2:1643
+ ld1 {v19.8h}, x0, #16 // quantCoef1644
+ ld1 {v2.4s}, x1, #16 // deQuantCoef1645
+ ld1 {v20.4s}, x1, #161646
+ sub x3, x3, #11647
+ sxtl v1.4s, v19.4h1648
+ sxtl2 v19.4s, v19.8h1649
+1650
+ mul v1.4s, v1.4s, v2.4s // quantCoef * deQuantCoef1651
+ mul v19.4s, v19.4s, v20.4s1652
+ sqxtn v16.4h, v1.4s // x265_clip31653
+ sqxtn2 v16.8h, v19.4s1654
+1655
+ sqshl v16.8h, v16.8h, v0.8h // coefQ << per - shift1656
+ st1 {v16.8h}, x2, #161657
+ cbnz x3, .dequant_loop21658
+ ret1659
+endfunc1660
+1661
+// void dequant_normal_c(const int16_t* quantCoef, int16_t* coef, int num, int scale, int shift)1662
+function PFX(dequant_normal_neon)1663
+ lsr w2, w2, #4 // num / 161664
+ neg w4, w41665
+ dup v0.8h, w31666
+ dup v1.4s, w41667
+1668
+.dqn_loop1:1669
+ ld1 {v2.8h, v3.8h}, x0, #321670
+ smull v16.4s, v2.4h, v0.4h1671
+ smull2 v17.4s, v2.8h, v0.8h1672
+ smull v18.4s, v3.4h, v0.4h1673
+ smull2 v19.4s, v3.8h, v0.8h1674
+1675
+ srshl v16.4s, v16.4s, v1.4s1676
+ srshl v17.4s, v17.4s, v1.4s1677
+ srshl v18.4s, v18.4s, v1.4s1678
+ srshl v19.4s, v19.4s, v1.4s1679
+1680
+ sqxtn v2.4h, v16.4s1681
+ sqxtn2 v2.8h, v17.4s1682
+ sqxtn v3.4h, v18.4s1683
+ sqxtn2 v3.8h, v19.4s1684
+1685
+ sub w2, w2, #11686
+ st1 {v2.8h, v3.8h}, x1, #321687
+ cbnz w2, .dqn_loop11688
+ ret1689
+endfunc1690
+1691
+/********* ssim ***********/1692
+// void ssim_4x4x2_core(const pixel* pix1, intptr_t stride1, const pixel* pix2, intptr_t stride2, int sums24)1693
+function PFX(ssim_4x4x2_core_neon)1694
+ ld1 {v0.8b}, x0, x11695
+ ld1 {v1.8b}, x0, x11696
+ ld1 {v2.8b}, x0, x11697
+ ld1 {v3.8b}, x0, x11698
+1699
+ ld1 {v4.8b}, x2, x31700
+ ld1 {v5.8b}, x2, x31701
+ ld1 {v6.8b}, x2, x31702
+ ld1 {v7.8b}, x2, x31703
+1704
+ umull v16.8h, v0.8b, v0.8b1705
+ umull v17.8h, v1.8b, v1.8b1706
+ umull v18.8h, v2.8b, v2.8b1707
+ uaddlp v30.4s, v16.8h1708
+ umull v19.8h, v3.8b, v3.8b1709
+ umull v20.8h, v4.8b, v4.8b1710
+ umull v21.8h, v5.8b, v5.8b1711
+ uadalp v30.4s, v17.8h1712
+ umull v22.8h, v6.8b, v6.8b1713
+ umull v23.8h, v7.8b, v7.8b1714
+1715
+ umull v24.8h, v0.8b, v4.8b1716
+ uadalp v30.4s, v18.8h1717
+ umull v25.8h, v1.8b, v5.8b1718
+ umull v26.8h, v2.8b, v6.8b1719
+ umull v27.8h, v3.8b, v7.8b1720
+ uadalp v30.4s, v19.8h1721
+1722
+ uaddl v28.8h, v0.8b, v1.8b1723
+ uaddl v29.8h, v4.8b, v5.8b1724
+ uadalp v30.4s, v20.8h1725
+ uaddlp v31.4s, v24.8h1726
+1727
+ uaddw v28.8h, v28.8h, v2.8b1728
+ uaddw v29.8h, v29.8h, v6.8b1729
+ uadalp v30.4s, v21.8h1730
+ uadalp v31.4s, v25.8h1731
+1732
+ uaddw v28.8h, v28.8h, v3.8b1733
+ uaddw v29.8h, v29.8h, v7.8b1734
+ uadalp v30.4s, v22.8h1735
+ uadalp v31.4s, v26.8h1736
+1737
+ uaddlp v28.4s, v28.8h1738
+ uaddlp v29.4s, v29.8h1739
+ uadalp v30.4s, v23.8h1740
+ uadalp v31.4s, v27.8h1741
+1742
+ addp v28.4s, v28.4s, v28.4s1743
+ addp v29.4s, v29.4s, v29.4s1744
+ addp v30.4s, v30.4s, v30.4s1745
+ addp v31.4s, v31.4s, v31.4s1746
+1747
+ st4 {v28.2s, v29.2s, v30.2s, v31.2s}, x41748
ret1749
endfunc1750
1751
// int psyCost_pp(const pixel* source, intptr_t sstride, const pixel* recon, intptr_t rstride)1752
-function x265_psyCost_4x4_neon1753
+function PFX(psyCost_4x4_neon)1754
ld1r {v4.2s}, x0, x11755
ld1r {v5.2s}, x0, x11756
ld1 {v4.s}1, x0, x11757
1758
endfunc1759
1760
// uint32_t quant_c(const int16_t* coef, const int32_t* quantCoeff, int32_t* deltaU, int16_t* qCoef, int qBits, int add, int numCoeff)1761
-function x265_quant_neon1762
+function PFX(quant_neon)1763
mov w9, #11764
lsl w9, w9, w41765
dup v0.2s, w91766
1767
ret1768
endfunc1769
1770
-.macro satd_4x4_neon1771
- ld1 {v1.s}0, x2, x31772
- ld1 {v0.s}0, x0, x11773
- ld1 {v3.s}0, x2, x31774
- ld1 {v2.s}0, x0, x11775
+// uint32_t nquant_c(const int16_t* coef, const int32_t* quantCoeff, int16_t* qCoef, int qBits, int add, int numCoeff)1776
+function PFX(nquant_neon)1777
+ neg x12, x31778
+ dup v0.4s, w12 // q0= -qbits1779
+ dup v1.4s, w4 // add1780
1781
- ld1 {v1.s}1, x2, x31782
- ld1 {v0.s}1, x0, x11783
- ld1 {v3.s}1, x2, x31784
- ld1 {v2.s}1, x0, x11785
+ lsr w5, w5, #21786
+ movi v4.4s, #0 // v4= accumulate numsig1787
+ mov x4, #01788
+ movi v22.4s, #01789
1790
- usubl v4.8h, v0.8b, v1.8b1791
- usubl v5.8h, v2.8b, v3.8b1792
+.loop_nquant:1793
+ ld1 {v16.4h}, x0, #81794
+ sub w5, w5, #11795
+ sxtl v19.4s, v16.4h // v19 = coefblockpos1796
1797
- add v6.8h, v4.8h, v5.8h1798
- sub v7.8h, v4.8h, v5.8h1799
+ cmlt v18.4s, v19.4s, #0 // v18 = sign1800
1801
- mov v4.d0, v6.d11802
- add v0.8h, v6.8h, v4.8h1803
- sub v2.8h, v6.8h, v4.8h1804
+ abs v19.4s, v19.4s // v19 = level=abs(coefblockpos)1805
+ ld1 {v20.4s}, x1, #16 // v20 = quantCoeffblockpos1806
+ mul v19.4s, v19.4s, v20.4s // v19 = tmplevel = abs(level) * quantCoeffblockpos;1807
1808
- mov v5.d0, v7.d11809
- add v1.8h, v7.8h, v5.8h1810
- sub v3.8h, v7.8h, v5.8h1811
+ add v20.4s, v19.4s, v1.4s // v20 = tmplevel+add1812
+ sshl v20.4s, v20.4s, v0.4s // v20 = level =(tmplevel+add) >> qbits1813
1814
- trn1 v4.4h, v0.4h, v1.4h1815
- trn2 v5.4h, v0.4h, v1.4h1816
+ // numsig1817
+ cmeq v21.4s, v20.4s, v22.4s1818
+ add v4.4s, v4.4s, v21.4s1819
+ add x4, x4, #41820
1821
- trn1 v6.4h, v2.4h, v3.4h1822
- trn2 v7.4h, v2.4h, v3.4h1823
+ eor v21.16b, v20.16b, v18.16b1824
+ sub v21.4s, v21.4s, v18.4s1825
+ sqxtn v16.4h, v21.4s1826
+ abs v17.4h, v16.4h1827
+ st1 {v17.4h}, x2, #81828
1829
- add v0.4h, v4.4h, v5.4h1830
- sub v1.4h, v4.4h, v5.4h1831
+ cbnz w5, .loop_nquant1832
1833
- add v2.4h, v6.4h, v7.4h1834
- sub v3.4h, v6.4h, v7.4h1835
+ uaddlv d4, v4.4s1836
+ fmov x12, d41837
+ add x0, x4, x121838
+ ret1839
+endfunc1840
1841
- trn1 v4.2s, v0.2s, v1.2s1842
- trn2 v5.2s, v0.2s, v1.2s1843
+// void ssimDist_c(const pixel* fenc, uint32_t fStride, const pixel* recon, intptr_t rstride, uint64_t *ssBlock, int shift, uint64_t *ac_k)1844
+.macro ssimDist_1 v4 v51845
+ sub v20.8h, \v4\().8h, \v5\().8h1846
+ smull v16.4s, \v4\().4h, \v4\().4h1847
+ smull2 v17.4s, \v4\().8h, \v4\().8h1848
+ smull v18.4s, v20.4h, v20.4h1849
+ smull2 v19.4s, v20.8h, v20.8h1850
+ add v0.4s, v0.4s, v16.4s1851
+ add v0.4s, v0.4s, v17.4s1852
+ add v1.4s, v1.4s, v18.4s1853
+ add v1.4s, v1.4s, v19.4s1854
+.endm1855
1856
- trn1 v6.2s, v2.2s, v3.2s1857
- trn2 v7.2s, v2.2s, v3.2s1858
+function PFX(ssimDist4_neon)1859
+ ssimDist_start1860
+.rept 41861
+ ld1 {v4.s}0, x0, x11862
+ ld1 {v5.s}0, x2, x31863
+ uxtl v4.8h, v4.8b1864
+ uxtl v5.8h, v5.8b1865
+ sub v2.4h, v4.4h, v5.4h1866
+ smull v3.4s, v4.4h, v4.4h1867
+ smull v2.4s, v2.4h, v2.4h1868
+ add v0.4s, v0.4s, v3.4s1869
+ add v1.4s, v1.4s, v2.4s1870
+.endr1871
+ ssimDist_end1872
+ ret1873
+endfunc1874
1875
- abs v4.4h, v4.4h1876
- abs v5.4h, v5.4h1877
- abs v6.4h, v6.4h1878
- abs v7.4h, v7.4h1879
+function PFX(ssimDist8_neon)1880
+ ssimDist_start1881
+.rept 81882
+ ld1 {v4.8b}, x0, x11883
+ ld1 {v5.8b}, x2, x31884
+ uxtl v4.8h, v4.8b1885
+ uxtl v5.8h, v5.8b1886
+ ssimDist_1 v4, v51887
+.endr1888
+ ssimDist_end1889
+ ret1890
+endfunc1891
1892
- smax v1.4h, v4.4h, v5.4h1893
- smax v2.4h, v6.4h, v7.4h1894
+function PFX(ssimDist16_neon)1895
+ mov w12, #161896
+ ssimDist_start1897
+.loop_ssimDist16:1898
+ sub w12, w12, #11899
+ ld1 {v4.16b}, x0, x11900
+ ld1 {v5.16b}, x2, x31901
+ uxtl v6.8h, v4.8b1902
+ uxtl v7.8h, v5.8b1903
+ uxtl2 v4.8h, v4.16b1904
+ uxtl2 v5.8h, v5.16b1905
+ ssimDist_1 v6, v71906
+ ssimDist_1 v4, v51907
+ cbnz w12, .loop_ssimDist161908
+ ssimDist_end1909
+ ret1910
+endfunc1911
1912
- add v0.4h, v1.4h, v2.4h1913
- uaddlp v0.2s, v0.4h1914
- uaddlp v0.1d, v0.2s1915
+function PFX(ssimDist32_neon)1916
+ mov w12, #321917
+ ssimDist_start1918
+.loop_ssimDist32:1919
+ sub w12, w12, #11920
+ ld1 {v4.16b-v5.16b}, x0, x11921
+ ld1 {v6.16b-v7.16b}, x2, x31922
+ uxtl v21.8h, v4.8b1923
+ uxtl v22.8h, v6.8b1924
+ uxtl v23.8h, v5.8b1925
+ uxtl v24.8h, v7.8b1926
+ uxtl2 v25.8h, v4.16b1927
+ uxtl2 v26.8h, v6.16b1928
+ uxtl2 v27.8h, v5.16b1929
+ uxtl2 v28.8h, v7.16b1930
+ ssimDist_1 v21, v221931
+ ssimDist_1 v23, v241932
+ ssimDist_1 v25, v261933
+ ssimDist_1 v27, v281934
+ cbnz w12, .loop_ssimDist321935
+ ssimDist_end1936
+ ret1937
+endfunc1938
+1939
+function PFX(ssimDist64_neon)1940
+ mov w12, #641941
+ ssimDist_start1942
+.loop_ssimDist64:1943
+ sub w12, w12, #11944
+ ld1 {v4.16b-v7.16b}, x0, x11945
+ ld1 {v16.16b-v19.16b}, x2, x31946
+ uxtl v21.8h, v4.8b1947
+ uxtl v22.8h, v16.8b1948
+ uxtl v23.8h, v5.8b1949
+ uxtl v24.8h, v17.8b1950
+ uxtl2 v25.8h, v4.16b1951
+ uxtl2 v26.8h, v16.16b1952
+ uxtl2 v27.8h, v5.16b1953
+ uxtl2 v28.8h, v17.16b1954
+ ssimDist_1 v21, v221955
+ ssimDist_1 v23, v241956
+ ssimDist_1 v25, v261957
+ ssimDist_1 v27, v281958
+ uxtl v21.8h, v6.8b1959
+ uxtl v22.8h, v18.8b1960
+ uxtl v23.8h, v7.8b1961
+ uxtl v24.8h, v19.8b1962
+ uxtl2 v25.8h, v6.16b1963
+ uxtl2 v26.8h, v18.16b1964
+ uxtl2 v27.8h, v7.16b1965
+ uxtl2 v28.8h, v19.16b1966
+ ssimDist_1 v21, v221967
+ ssimDist_1 v23, v241968
+ ssimDist_1 v25, v261969
+ ssimDist_1 v27, v281970
+ cbnz w12, .loop_ssimDist641971
+ ssimDist_end1972
+ ret1973
+endfunc1974
+1975
+// void normFact_c(const pixel* src, uint32_t blockSize, int shift, uint64_t *z_k)1976
+1977
+.macro normFact_1 v41978
+ smull v16.4s, \v4\().4h, \v4\().4h1979
+ smull2 v17.4s, \v4\().8h, \v4\().8h1980
+ add v0.4s, v0.4s, v16.4s1981
+ add v0.4s, v0.4s, v17.4s1982
.endm1983
1984
-// int satd_4x4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)1985
-function x265_pixel_satd_4x4_neon1986
- satd_4x4_neon1987
- umov x0, v0.d01988
+function PFX(normFact8_neon)1989
+ normFact_start1990
+.rept 81991
+ ld1 {v4.8b}, x0, x11992
+ uxtl v4.8h, v4.8b1993
+ normFact_1 v41994
+.endr1995
+ normFact_end1996
ret1997
endfunc1998
1999
-// int satd_8x4(const pixel* pix1, intptr_t stride_pix1, const pixel* pix2, intptr_t stride_pix2)2000
-function x265_pixel_satd_8x4_neon2001
- mov x4, x02002
- mov x5, x22003
- satd_4x4_neon2004
- add x0, x4, #42005
- add x2, x5, #42006
- umov x6, v0.d02007
- satd_4x4_neon2008
- umov x0, v0.d02009
- add x0, x0, x62010
+function PFX(normFact16_neon)2011
+ mov w12, #162012
+ normFact_start2013
+.loop_normFact16:2014
+ sub w12, w12, #12015
+ ld1 {v4.16b}, x0, x12016
+ uxtl v5.8h, v4.8b2017
+ uxtl2 v4.8h, v4.16b2018
+ normFact_1 v52019
+ normFact_1 v42020
+ cbnz w12, .loop_normFact162021
+ normFact_end2022
+ ret2023
+endfunc2024
+2025
+function PFX(normFact32_neon)2026
+ mov w12, #322027
+ normFact_start2028
+.loop_normFact32:2029
+ sub w12, w12, #12030
+ ld1 {v4.16b-v5.16b}, x0, x12031
+ uxtl v6.8h, v4.8b2032
+ uxtl2 v4.8h, v4.16b2033
+ uxtl v7.8h, v5.8b2034
+ uxtl2 v5.8h, v5.16b2035
+ normFact_1 v42036
+ normFact_1 v52037
+ normFact_1 v62038
+ normFact_1 v72039
+ cbnz w12, .loop_normFact322040
+ normFact_end2041
+ ret2042
+endfunc2043
+2044
+function PFX(normFact64_neon)2045
+ mov w12, #642046
+ normFact_start2047
+.loop_normFact64:2048
+ sub w12, w12, #12049
+ ld1 {v4.16b-v7.16b}, x0, x12050
+ uxtl v26.8h, v4.8b2051
+ uxtl2 v24.8h, v4.16b2052
+ uxtl v27.8h, v5.8b2053
+ uxtl2 v25.8h, v5.16b2054
+ normFact_1 v242055
+ normFact_1 v252056
+ normFact_1 v262057
+ normFact_1 v272058
+ uxtl v26.8h, v6.8b2059
+ uxtl2 v24.8h, v6.16b2060
+ uxtl v27.8h, v7.8b2061
+ uxtl2 v25.8h, v7.16b2062
+ normFact_1 v242063
+ normFact_1 v252064
+ normFact_1 v262065
+ normFact_1 v272066
+ cbnz w12, .loop_normFact642067
+ normFact_end2068
+ ret2069
+endfunc2070
+2071
+// void weight_pp_c(const pixel* src, pixel* dst, intptr_t stride, int width, int height, int w0, int round, int shift, int offset)2072
+function PFX(weight_pp_neon)2073
+ sub x2, x2, x32074
+ ldr w9, sp // offset2075
+ lsl w5, w5, #6 // w0 << correction2076
+2077
+ // count trailing zeros in w5 and compare against shift right amount.2078
+ rbit w10, w52079
+ clz w10, w102080
+ cmp w10, w72081
+ b.lt .unfoldedShift2082
+2083
+ // shift right only removes trailing zeros: hoist LSR out of the loop.2084
+ lsr w10, w5, w7 // w0 << correction >> shift2085
+ dup v25.16b, w102086
+ lsr w6, w6, w7 // round >> shift2087
+ add w6, w6, w9 // round >> shift + offset2088
+ dup v26.8h, w62089
+2090
+ // Check arithmetic range.2091
+ mov w11, #2552092
+ madd w11, w11, w10, w62093
+ add w11, w11, w92094
+ lsr w11, w11, #162095
+ cbnz w11, .widenTo32Bit2096
+2097
+ // 16-bit arithmetic is enough.2098
+.loopHpp:2099
+ mov x12, x32100
+.loopWpp:2101
+ ldr q0, x0, #162102
+ sub x12, x12, #162103
+ umull v1.8h, v0.8b, v25.8b // val *= w0 << correction >> shift2104
+ umull2 v2.8h, v0.16b, v25.16b2105
+ add v1.8h, v1.8h, v26.8h // val += round >> shift + offset2106
+ add v2.8h, v2.8h, v26.8h2107
+ sqxtun v0.8b, v1.8h // val = x265_clip(val)2108
+ sqxtun2 v0.16b, v2.8h2109
+ str q0, x1, #162110
+ cbnz x12, .loopWpp2111
+ add x1, x1, x22112
+ add x0, x0, x22113
+ sub x4, x4, #12114
+ cbnz x4, .loopHpp2115
+ ret2116
+2117
+ // 32-bit arithmetic is needed.2118
+.widenTo32Bit:2119
+.loopHpp32:2120
+ mov x12, x32121
+.loopWpp32:2122
+ ldr d0, x0, #82123
+ sub x12, x12, #82124
+ uxtl v0.8h, v0.8b2125
+ umull v1.4s, v0.4h, v25.4h // val *= w0 << correction >> shift2126
+ umull2 v2.4s, v0.8h, v25.8h2127
+ add v1.4s, v1.4s, v26.4s // val += round >> shift + offset2128
+ add v2.4s, v2.4s, v26.4s2129
+ sqxtn v0.4h, v1.4s // val = x265_clip(val)2130
+ sqxtn2 v0.8h, v2.4s2131
+ sqxtun v0.8b, v0.8h2132
+ str d0, x1, #82133
+ cbnz x12, .loopWpp322134
+ add x1, x1, x22135
+ add x0, x0, x22136
+ sub x4, x4, #12137
+ cbnz x4, .loopHpp322138
+ ret2139
+2140
+ // The shift right cannot be moved out of the loop.2141
+.unfoldedShift:2142
+ dup v25.8h, w5 // w0 << correction2143
+ dup v26.4s, w6 // round2144
+ neg w7, w7 // -shift2145
+ dup v27.4s, w72146
+ dup v29.4s, w9 // offset2147
+.loopHppUS:2148
+ mov x12, x32149
+.loopWppUS:2150
+ ldr d0, x0, #82151
+ sub x12, x12, #82152
+ uxtl v0.8h, v0.8b2153
+ umull v1.4s, v0.4h, v25.4h // val *= w02154
+ umull2 v2.4s, v0.8h, v25.8h2155
+ add v1.4s, v1.4s, v26.4s // val += round2156
+ add v2.4s, v2.4s, v26.4s2157
+ sshl v1.4s, v1.4s, v27.4s // val >>= shift2158
+ sshl v2.4s, v2.4s, v27.4s2159
+ add v1.4s, v1.4s, v29.4s // val += offset2160
+ add v2.4s, v2.4s, v29.4s2161
+ sqxtn v0.4h, v1.4s // val = x265_clip(val)2162
+ sqxtn2 v0.8h, v2.4s2163
+ sqxtun v0.8b, v0.8h2164
+ str d0, x1, #82165
+ cbnz x12, .loopWppUS2166
+ add x1, x1, x22167
+ add x0, x0, x22168
+ sub x4, x4, #12169
+ cbnz x4, .loopHppUS2170
+ ret2171
+endfunc2172
+2173
+// int scanPosLast(2174
+// const uint16_t *scan, // x02175
+// const coeff_t *coeff, // x12176
+// uint16_t *coeffSign, // x22177
+// uint16_t *coeffFlag, // x32178
+// uint8_t *coeffNum, // x42179
+// int numSig, // x52180
+// const uint16_t* scanCG4x4, // x62181
+// const int trSize) // x72182
+function PFX(scanPosLast_neon)2183
+ // convert unit of Stride(trSize) to int16_t2184
+ add x7, x7, x72185
+2186
+ // load scan table and convert to Byte2187
+ ldp q0, q1, x62188
+ xtn v0.8b, v0.8h2189
+ xtn2 v0.16b, v1.8h // v0 - Zigzag scan table2190
+2191
+ movrel x10, g_SPL_and_mask2192
+ ldr q28, x10 // v28 = mask for pmovmskb2193
+ movi v31.16b, #0 // v31 = {0, ..., 0}2194
+ add x10, x7, x7 // 2*x72195
+ add x11, x10, x7 // 3*x72196
+ add x9, x4, #1 // CG count2197
+2198
+.loop_spl:2199
+ // position of current CG2200
+ ldrh w6, x0, #322201
+ add x6, x1, x6, lsl #12202
+2203
+ // loading current CG2204
+ ldr d2, x62205
+ ldr d3, x6, x72206
+ ldr d4, x6, x102207
+ ldr d5, x6, x112208
+ mov v2.d1, v3.d02209
+ mov v4.d1, v5.d02210
+ sqxtn v2.8b, v2.8h2211
+ sqxtn2 v2.16b, v4.8h2212
+2213
+ // Zigzag2214
+ tbl v3.16b, {v2.16b}, v0.16b2215
+2216
+ // get sign2217
+ cmhi v5.16b, v3.16b, v31.16b // v5 = non-zero2218
+ cmlt v3.16b, v3.16b, #0 // v3 = negative2219
+2220
+ // val - w13 = pmovmskb(v3)2221
+ and v3.16b, v3.16b, v28.16b2222
+ mov d4, v3.d12223
+ addv b23, v3.8b2224
+ addv b24, v4.8b2225
+ mov v23.b1, v24.b02226
+ fmov w13, s232227
+2228
+ // mask - w15 = pmovmskb(v5)2229
+ and v5.16b, v5.16b, v28.16b2230
+ mov d6, v5.d12231
+ addv b25, v5.8b2232
+ addv b26, v6.8b2233
+ mov v25.b1, v26.b02234
+ fmov w15, s252235
+2236
+ // coeffFlag = reverse_bit(w15) in 16-bit2237
+ rbit w12, w152238
+ lsr w12, w12, #162239
+ fmov s30, w122240
+ strh w12, x3, #22241
+2242
+ // accelerate by preparing w13 = w13 & w152243
+ and w13, w13, w152244
+ mov x14, xzr2245
+.loop_spl_1:2246
+ cbz w15, .pext_end2247
+ clz w6, w152248
+ lsl w13, w13, w62249
+ lsl w15, w15, w62250
+ extr w14, w14, w13, #312251
+ bfm w15, wzr, #1, #02252
+ b .loop_spl_12253
+.pext_end:2254
+ strh w14, x2, #22255
+2256
+ // compute coeffNum = popcount(coeffFlag)2257
+ cnt v30.8b, v30.8b2258
+ addp v30.8b, v30.8b, v30.8b2259
+ fmov w6, s302260
+ sub x5, x5, x62261
+ strb w6, x4, #12262
+2263
+ cbnz x5, .loop_spl2264
+2265
+ // count trailing zeros2266
+ rbit w13, w122267
+ clz w13, w132268
+ lsr w12, w12, w132269
+ strh w12, x3, #-22270
+2271
+ // get last pos2272
+ sub x9, x4, x92273
+ lsl x0, x9, #42274
+ eor w13, w13, #152275
+ add x0, x0, x132276
+ ret2277
+endfunc2278
+2279
+// uint32_t costCoeffNxN(2280
+// uint16_t *scan, // x02281
+// coeff_t *coeff, // x12282
+// intptr_t trSize, // x22283
+// uint16_t *absCoeff, // x32284
+// uint8_t *tabSigCtx, // x42285
+// uint16_t scanFlagMask, // x52286
+// uint8_t *baseCtx, // x62287
+// int offset, // x72288
+// int scanPosSigOff, // sp2289
+// int subPosBase) // sp + 82290
+function PFX(costCoeffNxN_neon)2291
+ // abs(coeff)2292
+ add x2, x2, x22293
+ ld1 {v1.d}0, x1, x22294
+ ld1 {v1.d}1, x1, x22295
+ ld1 {v2.d}0, x1, x22296
+ ld1 {v2.d}1, x1, x22297
+ abs v1.8h, v1.8h2298
+ abs v2.8h, v2.8h2299
+2300
+ // WARNING: beyond-bound read here!2301
+ // loading scan table2302
+ ldr w2, sp2303
+ eor w15, w2, #152304
+ add x1, x0, x15, lsl #12305
+ ldp q20, q21, x12306
+ uzp1 v20.16b, v20.16b, v21.16b2307
+ movi v21.16b, #152308
+ eor v0.16b, v20.16b, v21.16b2309
+2310
+ // reorder coeff2311
+ uzp1 v22.16b, v1.16b, v2.16b2312
+ uzp2 v23.16b, v1.16b, v2.16b2313
+ tbl v24.16b, {v22.16b}, v0.16b2314
+ tbl v25.16b, {v23.16b}, v0.16b2315
+ zip1 v2.16b, v24.16b, v25.16b2316
+ zip2 v3.16b, v24.16b, v25.16b2317
+2318
+ // loading tabSigCtx (+offset)2319
+ ldr q1, x42320
+ tbl v1.16b, {v1.16b}, v0.16b2321
+ dup v4.16b, w72322
+ movi v5.16b, #02323
+ tbl v4.16b, {v4.16b}, v5.16b2324
+ add v1.16b, v1.16b, v4.16b2325
+2326
+ // register mapping2327
+ // x0 - sum2328
+ // x1 - entropyStateBits2329
+ // v1 - sigCtx2330
+ // {v3,v2} - abs(coeff)2331
+ // x2 - scanPosSigOff2332
+ // x3 - absCoeff2333
+ // x4 - numNonZero2334
+ // x5 - scanFlagMask2335
+ // x6 - baseCtx2336
+ mov x0, #02337
+ movrel x1, PFX_C(entropyStateBits)2338
+ mov x4, #02339
+ mov x11, #02340
+ movi v31.16b, #02341
+ cbz x2, .idx_zero2342
+.loop_ccnn:2343
+// {2344
+// const uint32_t cnt = tabSigCtxblkPos + offset + posOffset;2345
+// ctxSig = cnt & posZeroMask;2346
+// const uint32_t mstate = baseCtxctxSig;2347
+// const uint32_t mps = mstate & 1;2348
+// const uint32_t stateBits = x265_entropyStateBitsmstate ^ sig;2349
+// uint32_t nextState = (stateBits >> 24) + mps;2350
+// if ((mstate ^ sig) == 1)2351
+// nextState = sig;2352
+// baseCtxctxSig = (uint8_t)nextState;2353
+// sum += stateBits;2354
+// }2355
+// absCoeffnumNonZero = tmpCoeffblkPos;2356
+// numNonZero += sig;2357
+// scanPosSigOff--;2358
+2359
+ add x13, x3, x4, lsl #12360
+ sub x2, x2, #12361
+ str h2, x13 // absCoeffnumNonZero = tmpCoeffblkPos2362
+ fmov w14, s1 // x14 = ctxSig2363
+ uxtb w14, w142364
+ ubfx w11, w5, #0, #1 // x11 = sig2365
+ lsr x5, x5, #12366
+ add x4, x4, x11 // numNonZero += sig2367
+ ext v1.16b, v1.16b, v31.16b, #12368
+ ext v2.16b, v2.16b, v3.16b, #22369
+ ext v3.16b, v3.16b, v31.16b, #22370
+ ldrb w9, x6, x14 // mstate = baseCtxctxSig2371
+ and w10, w9, #1 // mps = mstate & 12372
+ eor w9, w9, w11 // x9 = mstate ^ sig2373
+ add x12, x1, x9, lsl #22374
+ ldr w13, x122375
+ add w0, w0, w13 // sum += x265_entropyStateBitsmstate ^ sig2376
+ ldrb w13, x12, #32377
+ add w10, w10, w13 // nextState = (stateBits >> 24) + mps2378
+ cmp w9, #12379
+ csel w10, w11, w10, eq2380
+ strb w10, x6, x142381
+ cbnz x2, .loop_ccnn2382
+.idx_zero:2383
+2384
+ add x13, x3, x4, lsl #12385
+ add x4, x4, x152386
+ str h2, x13 // absCoeffnumNonZero = tmpCoeffblkPos2387
+2388
+ ldr x9, sp, #8 // subPosBase2389
+ uxth w9, w92390
+ cmp w9, #02391
+ cset x2, eq2392
+ add x4, x4, x22393
+ cbz x4, .exit_ccnn2394
+2395
+ sub w2, w2, #12396
+ uxtb w2, w22397
+ fmov w3, s12398
+ and w2, w2, w32399
+2400
+ ldrb w3, x6, x2 // mstate = baseCtxctxSig2401
+ eor w4, w5, w3 // x5 = mstate ^ sig2402
+ and w3, w3, #1 // mps = mstate & 12403
+ add x1, x1, x4, lsl #22404
+ ldr w11, x12405
+ ldrb w12, x1, #32406
+ add w0, w0, w11 // sum += x265_entropyStateBitsmstate ^ sig2407
+ add w3, w3, w12 // nextState = (stateBits >> 24) + mps2408
+ cmp w4, #12409
+ csel w3, w5, w3, eq2410
+ strb w3, x6, x22411
+.exit_ccnn:2412
+ ubfx w0, w0, #0, #242413
ret2414
endfunc2415
+2416
+const g_SPL_and_mask, align=82417
+.byte 0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x40, 0x80, 0x1, 0x2, 0x4, 0x8, 0x10, 0x20, 0x40, 0x802418
+endconst2419
x265_3.6.tar.gz/source/common/aarch64/sad-a-common.S
Added
516
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+#include "asm.S"29
+30
+.arch armv8-a31
+32
+#ifdef __APPLE__33
+.section __RODATA,__rodata34
+#else35
+.section .rodata36
+#endif37
+38
+.align 439
+40
+.macro SAD_START_4 f41
+ ld1 {v0.s}0, x0, x142
+ ld1 {v0.s}1, x0, x143
+ ld1 {v1.s}0, x2, x344
+ ld1 {v1.s}1, x2, x345
+ \f v16.8h, v0.8b, v1.8b46
+.endm47
+48
+.macro SAD_4 h49
+.rept \h / 2 - 150
+ SAD_START_4 uabal51
+.endr52
+.endm53
+54
+.macro SAD_START_8 f55
+ ld1 {v0.8b}, x0, x156
+ ld1 {v1.8b}, x2, x357
+ ld1 {v2.8b}, x0, x158
+ ld1 {v3.8b}, x2, x359
+ \f v16.8h, v0.8b, v1.8b60
+ \f v17.8h, v2.8b, v3.8b61
+.endm62
+63
+.macro SAD_8 h64
+.rept \h / 2 - 165
+ SAD_START_8 uabal66
+.endr67
+.endm68
+69
+.macro SAD_START_16 f70
+ ld1 {v0.16b}, x0, x171
+ ld1 {v1.16b}, x2, x372
+ ld1 {v2.16b}, x0, x173
+ ld1 {v3.16b}, x2, x374
+ \f v16.8h, v0.8b, v1.8b75
+ \f\()2 v17.8h, v0.16b, v1.16b76
+ uabal v16.8h, v2.8b, v3.8b77
+ uabal2 v17.8h, v2.16b, v3.16b78
+.endm79
+80
+.macro SAD_16 h81
+.rept \h / 2 - 182
+ SAD_START_16 uabal83
+.endr84
+.endm85
+86
+.macro SAD_START_3287
+ movi v16.16b, #088
+ movi v17.16b, #089
+ movi v18.16b, #090
+ movi v19.16b, #091
+.endm92
+93
+.macro SAD_3294
+ ld1 {v0.16b-v1.16b}, x0, x195
+ ld1 {v2.16b-v3.16b}, x2, x396
+ ld1 {v4.16b-v5.16b}, x0, x197
+ ld1 {v6.16b-v7.16b}, x2, x398
+ uabal v16.8h, v0.8b, v2.8b99
+ uabal2 v17.8h, v0.16b, v2.16b100
+ uabal v18.8h, v1.8b, v3.8b101
+ uabal2 v19.8h, v1.16b, v3.16b102
+ uabal v16.8h, v4.8b, v6.8b103
+ uabal2 v17.8h, v4.16b, v6.16b104
+ uabal v18.8h, v5.8b, v7.8b105
+ uabal2 v19.8h, v5.16b, v7.16b106
+.endm107
+108
+.macro SAD_END_32109
+ add v16.8h, v16.8h, v17.8h110
+ add v17.8h, v18.8h, v19.8h111
+ add v16.8h, v16.8h, v17.8h112
+ uaddlv s0, v16.8h113
+ fmov w0, s0114
+ ret115
+.endm116
+117
+.macro SAD_START_64118
+ movi v16.16b, #0119
+ movi v17.16b, #0120
+ movi v18.16b, #0121
+ movi v19.16b, #0122
+ movi v20.16b, #0123
+ movi v21.16b, #0124
+ movi v22.16b, #0125
+ movi v23.16b, #0126
+.endm127
+128
+.macro SAD_64129
+ ld1 {v0.16b-v3.16b}, x0, x1130
+ ld1 {v4.16b-v7.16b}, x2, x3131
+ ld1 {v24.16b-v27.16b}, x0, x1132
+ ld1 {v28.16b-v31.16b}, x2, x3133
+ uabal v16.8h, v0.8b, v4.8b134
+ uabal2 v17.8h, v0.16b, v4.16b135
+ uabal v18.8h, v1.8b, v5.8b136
+ uabal2 v19.8h, v1.16b, v5.16b137
+ uabal v20.8h, v2.8b, v6.8b138
+ uabal2 v21.8h, v2.16b, v6.16b139
+ uabal v22.8h, v3.8b, v7.8b140
+ uabal2 v23.8h, v3.16b, v7.16b141
+142
+ uabal v16.8h, v24.8b, v28.8b143
+ uabal2 v17.8h, v24.16b, v28.16b144
+ uabal v18.8h, v25.8b, v29.8b145
+ uabal2 v19.8h, v25.16b, v29.16b146
+ uabal v20.8h, v26.8b, v30.8b147
+ uabal2 v21.8h, v26.16b, v30.16b148
+ uabal v22.8h, v27.8b, v31.8b149
+ uabal2 v23.8h, v27.16b, v31.16b150
+.endm151
+152
+.macro SAD_END_64153
+ add v16.8h, v16.8h, v17.8h154
+ add v17.8h, v18.8h, v19.8h155
+ add v16.8h, v16.8h, v17.8h156
+ uaddlp v16.4s, v16.8h157
+ add v18.8h, v20.8h, v21.8h158
+ add v19.8h, v22.8h, v23.8h159
+ add v17.8h, v18.8h, v19.8h160
+ uaddlp v17.4s, v17.8h161
+ add v16.4s, v16.4s, v17.4s162
+ uaddlv d0, v16.4s163
+ fmov x0, d0164
+ ret165
+.endm166
+167
+.macro SAD_START_12168
+ movrel x12, sad12_mask169
+ ld1 {v31.16b}, x12170
+ movi v16.16b, #0171
+ movi v17.16b, #0172
+.endm173
+174
+.macro SAD_12175
+ ld1 {v0.16b}, x0, x1176
+ and v0.16b, v0.16b, v31.16b177
+ ld1 {v1.16b}, x2, x3178
+ and v1.16b, v1.16b, v31.16b179
+ ld1 {v2.16b}, x0, x1180
+ and v2.16b, v2.16b, v31.16b181
+ ld1 {v3.16b}, x2, x3182
+ and v3.16b, v3.16b, v31.16b183
+ uabal v16.8h, v0.8b, v1.8b184
+ uabal2 v17.8h, v0.16b, v1.16b185
+ uabal v16.8h, v2.8b, v3.8b186
+ uabal2 v17.8h, v2.16b, v3.16b187
+.endm188
+189
+.macro SAD_END_12190
+ add v16.8h, v16.8h, v17.8h191
+ uaddlv s0, v16.8h192
+ fmov w0, s0193
+ ret194
+.endm195
+196
+.macro SAD_START_24197
+ movi v16.16b, #0198
+ movi v17.16b, #0199
+ movi v18.16b, #0200
+ sub x1, x1, #16201
+ sub x3, x3, #16202
+.endm203
+204
+.macro SAD_24205
+ ld1 {v0.16b}, x0, #16206
+ ld1 {v1.8b}, x0, x1207
+ ld1 {v2.16b}, x2, #16208
+ ld1 {v3.8b}, x2, x3209
+ ld1 {v4.16b}, x0, #16210
+ ld1 {v5.8b}, x0, x1211
+ ld1 {v6.16b}, x2, #16212
+ ld1 {v7.8b}, x2, x3213
+ uabal v16.8h, v0.8b, v2.8b214
+ uabal2 v17.8h, v0.16b, v2.16b215
+ uabal v18.8h, v1.8b, v3.8b216
+ uabal v16.8h, v4.8b, v6.8b217
+ uabal2 v17.8h, v4.16b, v6.16b218
+ uabal v18.8h, v5.8b, v7.8b219
+.endm220
+221
+.macro SAD_END_24222
+ add v16.8h, v16.8h, v17.8h223
+ add v16.8h, v16.8h, v18.8h224
+ uaddlv s0, v16.8h225
+ fmov w0, s0226
+ ret227
+.endm228
+229
+.macro SAD_START_48230
+ movi v16.16b, #0231
+ movi v17.16b, #0232
+ movi v18.16b, #0233
+ movi v19.16b, #0234
+ movi v20.16b, #0235
+ movi v21.16b, #0236
+.endm237
+238
+.macro SAD_48239
+ ld1 {v0.16b-v2.16b}, x0, x1240
+ ld1 {v4.16b-v6.16b}, x2, x3241
+ ld1 {v24.16b-v26.16b}, x0, x1242
+ ld1 {v28.16b-v30.16b}, x2, x3243
+ uabal v16.8h, v0.8b, v4.8b244
+ uabal2 v17.8h, v0.16b, v4.16b245
+ uabal v18.8h, v1.8b, v5.8b246
+ uabal2 v19.8h, v1.16b, v5.16b247
+ uabal v20.8h, v2.8b, v6.8b248
+ uabal2 v21.8h, v2.16b, v6.16b249
+250
+ uabal v16.8h, v24.8b, v28.8b251
+ uabal2 v17.8h, v24.16b, v28.16b252
+ uabal v18.8h, v25.8b, v29.8b253
+ uabal2 v19.8h, v25.16b, v29.16b254
+ uabal v20.8h, v26.8b, v30.8b255
+ uabal2 v21.8h, v26.16b, v30.16b256
+.endm257
+258
+.macro SAD_END_48259
+ add v16.8h, v16.8h, v17.8h260
+ add v17.8h, v18.8h, v19.8h261
+ add v16.8h, v16.8h, v17.8h262
+ uaddlv s0, v16.8h263
+ fmov w0, s0264
+ add v18.8h, v20.8h, v21.8h265
+ uaddlv s1, v18.8h266
+ fmov w1, s1267
+ add w0, w0, w1268
+ ret269
+.endm270
+271
+.macro SAD_X_START_4 h, x, f272
+ ld1 {v0.s}0, x0, x9273
+ ld1 {v0.s}1, x0, x9274
+ ld1 {v1.s}0, x1, x5275
+ ld1 {v1.s}1, x1, x5276
+ ld1 {v2.s}0, x2, x5277
+ ld1 {v2.s}1, x2, x5278
+ ld1 {v3.s}0, x3, x5279
+ ld1 {v3.s}1, x3, x5280
+ \f v16.8h, v0.8b, v1.8b281
+ \f v17.8h, v0.8b, v2.8b282
+ \f v18.8h, v0.8b, v3.8b283
+.if \x == 4284
+ ld1 {v4.s}0, x4, x5285
+ ld1 {v4.s}1, x4, x5286
+ \f v19.8h, v0.8b, v4.8b287
+.endif288
+.endm289
+290
+.macro SAD_X_4 h, x291
+.rept \h/2 - 1292
+ SAD_X_START_4 \h, \x, uabal293
+.endr294
+.endm295
+296
+.macro SAD_X_END_4 x297
+ uaddlv s0, v16.8h298
+ uaddlv s1, v17.8h299
+ uaddlv s2, v18.8h300
+ stp s0, s1, x6301
+.if \x == 3302
+ str s2, x6, #8303
+.elseif \x == 4304
+ uaddlv s3, v19.8h305
+ stp s2, s3, x6, #8306
+.endif307
+ ret308
+.endm309
+310
+.macro SAD_X_START_8 h, x, f311
+ ld1 {v0.8b}, x0, x9312
+ ld1 {v1.8b}, x1, x5313
+ ld1 {v2.8b}, x2, x5314
+ ld1 {v3.8b}, x3, x5315
+ \f v16.8h, v0.8b, v1.8b316
+ \f v17.8h, v0.8b, v2.8b317
+ \f v18.8h, v0.8b, v3.8b318
+.if \x == 4319
+ ld1 {v4.8b}, x4, x5320
+ \f v19.8h, v0.8b, v4.8b321
+.endif322
+.endm323
+324
+.macro SAD_X_8 h x325
+.rept \h - 1326
+ SAD_X_START_8 \h, \x, uabal327
+.endr328
+.endm329
+330
+.macro SAD_X_END_8 x331
+ SAD_X_END_4 \x332
+.endm333
+334
+.macro SAD_X_START_12 h, x, f335
+ ld1 {v0.16b}, x0, x9336
+ and v0.16b, v0.16b, v31.16b337
+ ld1 {v1.16b}, x1, x5338
+ and v1.16b, v1.16b, v31.16b339
+ ld1 {v2.16b}, x2, x5340
+ and v2.16b, v2.16b, v31.16b341
+ ld1 {v3.16b}, x3, x5342
+ and v3.16b, v3.16b, v31.16b343
+ \f v16.8h, v1.8b, v0.8b344
+ \f\()2 v20.8h, v1.16b, v0.16b345
+ \f v17.8h, v2.8b, v0.8b346
+ \f\()2 v21.8h, v2.16b, v0.16b347
+ \f v18.8h, v3.8b, v0.8b348
+ \f\()2 v22.8h, v3.16b, v0.16b349
+.if \x == 4350
+ ld1 {v4.16b}, x4, x5351
+ and v4.16b, v4.16b, v31.16b352
+ \f v19.8h, v4.8b, v0.8b353
+ \f\()2 v23.8h, v4.16b, v0.16b354
+.endif355
+.endm356
+357
+.macro SAD_X_12 h x358
+.rept \h - 1359
+ SAD_X_START_12 \h, \x, uabal360
+.endr361
+.endm362
+363
+.macro SAD_X_END_12 x364
+ SAD_X_END_16 \x365
+.endm366
+367
+.macro SAD_X_START_16 h, x, f368
+ ld1 {v0.16b}, x0, x9369
+ ld1 {v1.16b}, x1, x5370
+ ld1 {v2.16b}, x2, x5371
+ ld1 {v3.16b}, x3, x5372
+ \f v16.8h, v1.8b, v0.8b373
+ \f\()2 v20.8h, v1.16b, v0.16b374
+ \f v17.8h, v2.8b, v0.8b375
+ \f\()2 v21.8h, v2.16b, v0.16b376
+ \f v18.8h, v3.8b, v0.8b377
+ \f\()2 v22.8h, v3.16b, v0.16b378
+.if \x == 4379
+ ld1 {v4.16b}, x4, x5380
+ \f v19.8h, v4.8b, v0.8b381
+ \f\()2 v23.8h, v4.16b, v0.16b382
+.endif383
+.endm384
+385
+.macro SAD_X_16 h x386
+.rept \h - 1387
+ SAD_X_START_16 \h, \x, uabal388
+.endr389
+.endm390
+391
+.macro SAD_X_END_16 x392
+ add v16.8h, v16.8h, v20.8h393
+ add v17.8h, v17.8h, v21.8h394
+ add v18.8h, v18.8h, v22.8h395
+.if \x == 4396
+ add v19.8h, v19.8h, v23.8h397
+.endif398
+399
+ SAD_X_END_4 \x400
+.endm401
+402
+.macro SAD_X_START_24 x403
+ SAD_X_START_32 \x404
+ sub x5, x5, #16405
+ sub x9, x9, #16406
+.endm407
+408
+.macro SAD_X_24 base v1 v2409
+ ld1 {v0.16b}, \base , #16410
+ ld1 {v1.8b}, \base , x5411
+ uabal \v1\().8h, v0.8b, v6.8b412
+ uabal \v1\().8h, v1.8b, v7.8b413
+ uabal2 \v2\().8h, v0.16b, v6.16b414
+.endm415
+416
+.macro SAD_X_END_24 x417
+ SAD_X_END_16 \x418
+.endm419
+420
+.macro SAD_X_START_32 x421
+ movi v16.16b, #0422
+ movi v17.16b, #0423
+ movi v18.16b, #0424
+ movi v20.16b, #0425
+ movi v21.16b, #0426
+ movi v22.16b, #0427
+.if \x == 4428
+ movi v19.16b, #0429
+ movi v23.16b, #0430
+.endif431
+.endm432
+433
+.macro SAD_X_32 base v1 v2434
+ ld1 {v0.16b-v1.16b}, \base , x5435
+ uabal \v1\().8h, v0.8b, v6.8b436
+ uabal \v1\().8h, v1.8b, v7.8b437
+ uabal2 \v2\().8h, v0.16b, v6.16b438
+ uabal2 \v2\().8h, v1.16b, v7.16b439
+.endm440
+441
+.macro SAD_X_END_32 x442
+ SAD_X_END_16 \x443
+.endm444
+445
+.macro SAD_X_START_48 x446
+ SAD_X_START_32 \x447
+.endm448
+449
+.macro SAD_X_48 x1 v1 v2450
+ ld1 {v0.16b-v2.16b}, \x1 , x5451
+ uabal \v1\().8h, v0.8b, v4.8b452
+ uabal \v1\().8h, v1.8b, v5.8b453
+ uabal \v1\().8h, v2.8b, v6.8b454
+ uabal2 \v2\().8h, v0.16b, v4.16b455
+ uabal2 \v2\().8h, v1.16b, v5.16b456
+ uabal2 \v2\().8h, v2.16b, v6.16b457
+.endm458
+459
+.macro SAD_X_END_48 x460
+ SAD_X_END_64 \x461
+.endm462
+463
+.macro SAD_X_START_64 x464
+ SAD_X_START_32 \x465
+.endm466
+467
+.macro SAD_X_64 x1 v1 v2468
+ ld1 {v0.16b-v3.16b}, \x1 , x5469
+ uabal \v1\().8h, v0.8b, v4.8b470
+ uabal \v1\().8h, v1.8b, v5.8b471
+ uabal \v1\().8h, v2.8b, v6.8b472
+ uabal \v1\().8h, v3.8b, v7.8b473
+ uabal2 \v2\().8h, v0.16b, v4.16b474
+ uabal2 \v2\().8h, v1.16b, v5.16b475
+ uabal2 \v2\().8h, v2.16b, v6.16b476
+ uabal2 \v2\().8h, v3.16b, v7.16b477
+.endm478
+479
+.macro SAD_X_END_64 x480
+ uaddlp v16.4s, v16.8h481
+ uaddlp v17.4s, v17.8h482
+ uaddlp v18.4s, v18.8h483
+ uaddlp v20.4s, v20.8h484
+ uaddlp v21.4s, v21.8h485
+ uaddlp v22.4s, v22.8h486
+ add v16.4s, v16.4s, v20.4s487
+ add v17.4s, v17.4s, v21.4s488
+ add v18.4s, v18.4s, v22.4s489
+ trn2 v20.2d, v16.2d, v16.2d490
+ trn2 v21.2d, v17.2d, v17.2d491
+ trn2 v22.2d, v18.2d, v18.2d492
+ add v16.2s, v16.2s, v20.2s493
+ add v17.2s, v17.2s, v21.2s494
+ add v18.2s, v18.2s, v22.2s495
+ uaddlp v16.1d, v16.2s496
+ uaddlp v17.1d, v17.2s497
+ uaddlp v18.1d, v18.2s498
+ stp s16, s17, x6, #8499
+.if \x == 3500
+ str s18, x6501
+.elseif \x == 4502
+ uaddlp v19.4s, v19.8h503
+ uaddlp v23.4s, v23.8h504
+ add v19.4s, v19.4s, v23.4s505
+ trn2 v23.2d, v19.2d, v19.2d506
+ add v19.2s, v19.2s, v23.2s507
+ uaddlp v19.1d, v19.2s508
+ stp s18, s19, x6509
+.endif510
+ ret511
+.endm512
+513
+const sad12_mask, align=8514
+.byte 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 0, 0, 0, 0515
+endconst516
x265_3.6.tar.gz/source/common/aarch64/sad-a-sve2.S
Added
513
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "sad-a-common.S"27
+28
+.arch armv8-a+sve229
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+.macro SAD_SVE2_16 h41
+ mov z16.d, #042
+ ptrue p0.h, vl1643
+.rept \h44
+ ld1b {z0.h}, p0/z, x045
+ ld1b {z2.h}, p0/z, x246
+ add x0, x0, x147
+ add x2, x2, x348
+ uaba z16.h, z0.h, z2.h49
+.endr50
+ uaddv d0, p0, z16.h51
+ fmov w0, s052
+ ret53
+.endm54
+55
+.macro SAD_SVE2_32 h56
+ ptrue p0.b, vl3257
+.rept \h58
+ ld1b {z0.b}, p0/z, x059
+ ld1b {z4.b}, p0/z, x260
+ add x0, x0, x161
+ add x2, x2, x362
+ uabalb z16.h, z0.b, z4.b63
+ uabalt z16.h, z0.b, z4.b64
+.endr65
+ uaddv d0, p0, z16.h66
+ fmov w0, s067
+ ret68
+.endm69
+70
+.macro SAD_SVE2_64 h71
+ cmp x9, #4872
+ bgt .vl_gt_48_pixel_sad_64x\h73
+ mov z16.d, #074
+ mov z17.d, #075
+ mov z18.d, #076
+ mov z19.d, #077
+ ptrue p0.b, vl3278
+.rept \h79
+ ld1b {z0.b}, p0/z, x080
+ ld1b {z1.b}, p0/z, x0, #1, mul vl81
+ ld1b {z4.b}, p0/z, x282
+ ld1b {z5.b}, p0/z, x2, #1, mul vl83
+ add x0, x0, x184
+ add x2, x2, x385
+ uabalb z16.h, z0.b, z4.b86
+ uabalt z17.h, z0.b, z4.b87
+ uabalb z18.h, z1.b, z5.b88
+ uabalt z19.h, z1.b, z5.b89
+.endr90
+ add z16.h, z16.h, z17.h91
+ add z17.h, z18.h, z19.h92
+ add z16.h, z16.h, z17.h93
+ uadalp z24.s, p0/m, z16.h94
+ uaddv d5, p0, z24.s95
+ fmov x0, d596
+ ret97
+.vl_gt_48_pixel_sad_64x\h\():98
+ mov z16.d, #099
+ mov z17.d, #0100
+ mov z24.d, #0101
+ ptrue p0.b, vl64102
+.rept \h103
+ ld1b {z0.b}, p0/z, x0104
+ ld1b {z4.b}, p0/z, x2105
+ add x0, x0, x1106
+ add x2, x2, x3107
+ uabalb z16.h, z0.b, z4.b108
+ uabalt z17.h, z0.b, z4.b109
+.endr110
+ add z16.h, z16.h, z17.h111
+ uadalp z24.s, p0/m, z16.h112
+ uaddv d5, p0, z24.s113
+ fmov x0, d5114
+ ret115
+.endm116
+117
+.macro SAD_SVE2_24 h118
+ mov z16.d, #0119
+ mov x10, #24120
+ mov x11, #0121
+ whilelt p0.b, x11, x10122
+.rept \h123
+ ld1b {z0.b}, p0/z, x0124
+ ld1b {z8.b}, p0/z, x2125
+ add x0, x0, x1126
+ add x2, x2, x3127
+ uabalb z16.h, z0.b, z8.b128
+ uabalt z16.h, z0.b, z8.b129
+.endr130
+ uaddv d5, p0, z16.h131
+ fmov w0, s5132
+ ret133
+.endm134
+135
+.macro SAD_SVE2_48 h136
+ cmp x9, #48137
+ bgt .vl_gt_48_pixel_sad_48x\h138
+ mov z16.d, #0139
+ mov z17.d, #0140
+ mov z18.d, #0141
+ mov z19.d, #0142
+ ptrue p0.b, vl32143
+ ptrue p1.b, vl16144
+.rept \h145
+ ld1b {z0.b}, p0/z, x0146
+ ld1b {z1.b}, p1/z, x0, #1, mul vl147
+ ld1b {z8.b}, p0/z, x2148
+ ld1b {z9.b}, p1/z, x2, #1, mul vl149
+ add x0, x0, x1150
+ add x2, x2, x3151
+ uabalb z16.h, z0.b, z8.b152
+ uabalt z17.h, z0.b, z8.b153
+ uabalb z18.h, z1.b, z9.b154
+ uabalt z19.h, z1.b, z9.b155
+.endr156
+ add z16.h, z16.h, z17.h157
+ add z17.h, z18.h, z19.h158
+ add z16.h, z16.h, z17.h159
+ uaddv d5, p0, z16.h160
+ fmov w0, s5161
+ ret162
+.vl_gt_48_pixel_sad_48x\h\():163
+ mov z16.d, #0164
+ mov z17.d, #0165
+ mov x10, #48166
+ mov x11, #0167
+ whilelt p0.b, x11, x10168
+.rept \h169
+ ld1b {z0.b}, p0/z, x0170
+ ld1b {z8.b}, p0/z, x2171
+ add x0, x0, x1172
+ add x2, x2, x3173
+ uabalb z16.h, z0.b, z8.b174
+ uabalt z17.h, z0.b, z8.b175
+.endr176
+ add z16.h, z16.h, z17.h177
+ uaddv d5, p0, z16.h178
+ fmov w0, s5179
+ ret180
+.endm181
+182
+// Fully unrolled.183
+.macro SAD_FUNC_SVE2 w, h184
+function PFX(pixel_sad_\w\()x\h\()_sve2)185
+ rdvl x9, #1186
+ cmp x9, #16187
+ bgt .vl_gt_16_pixel_sad_\w\()x\h188
+ SAD_START_\w uabdl189
+ SAD_\w \h190
+.if \w > 4191
+ add v16.8h, v16.8h, v17.8h192
+.endif193
+ uaddlv s0, v16.8h194
+ fmov w0, s0195
+ ret196
+.vl_gt_16_pixel_sad_\w\()x\h\():197
+.if \w == 4 || \w == 8 || \w == 12198
+ SAD_START_\w uabdl199
+ SAD_\w \h200
+.if \w > 4201
+ add v16.8h, v16.8h, v17.8h202
+.endif203
+ uaddlv s0, v16.8h204
+ fmov w0, s0205
+ ret206
+.else207
+ SAD_SVE2_\w \h208
+.endif209
+endfunc210
+.endm211
+212
+// Loop unrolled 4.213
+.macro SAD_FUNC_LOOP_SVE2 w, h214
+function PFX(pixel_sad_\w\()x\h\()_sve2)215
+ rdvl x9, #1216
+ cmp x9, #16217
+ bgt .vl_gt_16_pixel_sad_loop_\w\()x\h218
+ SAD_START_\w219
+220
+ mov w9, #\h/8221
+.loop_sve2_\w\()x\h:222
+ sub w9, w9, #1223
+.rept 4224
+ SAD_\w225
+.endr226
+ cbnz w9, .loop_sve2_\w\()x\h227
+228
+ SAD_END_\w229
+230
+.vl_gt_16_pixel_sad_loop_\w\()x\h\():231
+.if \w == 4 || \w == 8 || \w == 12232
+ SAD_START_\w233
+234
+ mov w9, #\h/8235
+.loop_sve2_loop_\w\()x\h:236
+ sub w9, w9, #1237
+.rept 4238
+ SAD_\w239
+.endr240
+ cbnz w9, .loop_sve2_loop_\w\()x\h241
+242
+ SAD_END_\w243
+.else244
+ SAD_SVE2_\w \h245
+.endif246
+endfunc247
+.endm248
+249
+SAD_FUNC_SVE2 4, 4250
+SAD_FUNC_SVE2 4, 8251
+SAD_FUNC_SVE2 4, 16252
+SAD_FUNC_SVE2 8, 4253
+SAD_FUNC_SVE2 8, 8254
+SAD_FUNC_SVE2 8, 16255
+SAD_FUNC_SVE2 8, 32256
+SAD_FUNC_SVE2 16, 4257
+SAD_FUNC_SVE2 16, 8258
+SAD_FUNC_SVE2 16, 12259
+SAD_FUNC_SVE2 16, 16260
+SAD_FUNC_SVE2 16, 32261
+SAD_FUNC_SVE2 16, 64262
+263
+SAD_FUNC_LOOP_SVE2 32, 8264
+SAD_FUNC_LOOP_SVE2 32, 16265
+SAD_FUNC_LOOP_SVE2 32, 24266
+SAD_FUNC_LOOP_SVE2 32, 32267
+SAD_FUNC_LOOP_SVE2 32, 64268
+SAD_FUNC_LOOP_SVE2 64, 16269
+SAD_FUNC_LOOP_SVE2 64, 32270
+SAD_FUNC_LOOP_SVE2 64, 48271
+SAD_FUNC_LOOP_SVE2 64, 64272
+SAD_FUNC_LOOP_SVE2 12, 16273
+SAD_FUNC_LOOP_SVE2 24, 32274
+SAD_FUNC_LOOP_SVE2 48, 64275
+276
+// SAD_X3 and SAD_X4 code start277
+278
+.macro SAD_X_SVE2_24_INNER_GT_16 base z279
+ ld1b {z4.b}, p0/z, \base 280
+ add \base, \base, x5281
+ uabalb \z\().h, z4.b, z0.b282
+ uabalt \z\().h, z4.b, z0.b283
+.endm284
+285
+.macro SAD_X_SVE2_24 h x286
+ mov z20.d, #0287
+ mov z21.d, #0288
+ mov z22.d, #0289
+ mov z23.d, #0290
+ mov x10, #24291
+ mov x11, #0292
+ whilelt p0.b, x11, x10293
+.rept \h294
+ ld1b {z0.b}, p0/z, x0295
+ add x0, x0, x9296
+ SAD_X_SVE2_24_INNER_GT_16 x1, z20297
+ SAD_X_SVE2_24_INNER_GT_16 x2, z21298
+ SAD_X_SVE2_24_INNER_GT_16 x3, z22299
+.if \x == 4300
+ SAD_X_SVE2_24_INNER_GT_16 x4, z23301
+.endif302
+.endr303
+ uaddlv s0, v20.8h304
+ uaddlv s1, v21.8h305
+ uaddlv s2, v22.8h306
+ stp s0, s1, x6307
+.if \x == 3308
+ str s2, x6, #8309
+.elseif \x == 4310
+ uaddv d0, p0, z20.h311
+ uaddv d1, p0, z21.h312
+ uaddv d2, p0, z22.h313
+ stp s2, s3, x6, #8314
+.endif315
+ ret316
+.endm317
+318
+.macro SAD_X_SVE2_32_INNER_GT_16 base z319
+ ld1b {z4.b}, p0/z, \base 320
+ add \base, \base, x5321
+ uabalb \z\().h, z4.b, z0.b322
+ uabalt \z\().h, z4.b, z0.b323
+.endm324
+325
+.macro SAD_X_SVE2_32 h x326
+ mov z20.d, #0327
+ mov z21.d, #0328
+ mov z22.d, #0329
+ mov z23.d, #0330
+ ptrue p0.b, vl32331
+.rept \h332
+ ld1b {z0.b}, p0/z, x0333
+ add x0, x0, x9334
+ SAD_X_SVE2_32_INNER_GT_16 x1, z20335
+ SAD_X_SVE2_32_INNER_GT_16 x2, z21336
+ SAD_X_SVE2_32_INNER_GT_16 x3, z22337
+.if \x == 4338
+ SAD_X_SVE2_32_INNER_GT_16 x4, z23339
+.endif340
+.endr341
+ uaddv d0, p0, z20.h342
+ uaddv d1, p0, z21.h343
+ uaddv d2, p0, z22.h344
+ stp s0, s1, x6345
+.if \x == 3346
+ str s2, x6, #8347
+.elseif \x == 4348
+ uaddv d3, p0, z23.h349
+ stp s2, s3, x6, #8350
+.endif351
+ ret352
+.endm353
+354
+// static void x264_pixel_sad_x3_##size(pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, intptr_t i_stride, int scores3)355
+// static void x264_pixel_sad_x4_##size(pixel *fenc, pixel *pix0, pixel *pix1,pixel *pix2, pixel *pix3, intptr_t i_stride, int scores4)356
+.macro SAD_X_FUNC_SVE2 x, w, h357
+function PFX(sad_x\x\()_\w\()x\h\()_sve2)358
+ mov x9, #FENC_STRIDE359
+360
+// Make function arguments for x == 3 look like x == 4.361
+.if \x == 3362
+ mov x6, x5363
+ mov x5, x4364
+.endif365
+ rdvl x11, #1366
+ cmp x11, #16367
+ bgt .vl_gt_16_sad_x\x\()_\w\()x\h368
+.if \w == 12369
+ movrel x12, sad12_mask370
+ ld1 {v31.16b}, x12371
+.endif372
+373
+ SAD_X_START_\w \h, \x, uabdl374
+ SAD_X_\w \h, \x375
+ SAD_X_END_\w \x376
+.vl_gt_16_sad_x\x\()_\w\()x\h\():377
+.if \w == 24 || \w == 32378
+ SAD_X_SVE2_\w \h, \x379
+.else380
+.if \w == 12381
+ movrel x12, sad12_mask382
+ ld1 {v31.16b}, x12383
+.endif384
+385
+ SAD_X_START_\w \h, \x, uabdl386
+ SAD_X_\w \h, \x387
+ SAD_X_END_\w \x388
+.endif389
+endfunc390
+.endm391
+392
+.macro SAD_X_LOOP_SVE2 x, w, h393
+function PFX(sad_x\x\()_\w\()x\h\()_sve2)394
+ mov x9, #FENC_STRIDE395
+396
+// Make function arguments for x == 3 look like x == 4.397
+.if \x == 3398
+ mov x6, x5399
+ mov x5, x4400
+.endif401
+ rdvl x11, #1402
+ cmp x11, #16403
+ bgt .vl_gt_16_sad_x_loop_\x\()_\w\()x\h404
+ SAD_X_START_\w \x405
+ mov w12, #\h/4406
+.loop_sad_sve2_x\x\()_\w\()x\h:407
+ sub w12, w12, #1408
+ .rept 4409
+ .if \w == 24410
+ ld1 {v6.16b}, x0, #16411
+ ld1 {v7.8b}, x0, x9412
+ .elseif \w == 32413
+ ld1 {v6.16b-v7.16b}, x0, x9414
+ .elseif \w == 48415
+ ld1 {v4.16b-v6.16b}, x0, x9416
+ .elseif \w == 64417
+ ld1 {v4.16b-v7.16b}, x0, x9418
+ .endif419
+ SAD_X_\w x1, v16, v20420
+ SAD_X_\w x2, v17, v21421
+ SAD_X_\w x3, v18, v22422
+ .if \x == 4423
+ SAD_X_\w x4, v19, v23424
+ .endif425
+ .endr426
+ cbnz w12, .loop_sad_sve2_x\x\()_\w\()x\h427
+ SAD_X_END_\w \x428
+.vl_gt_16_sad_x_loop_\x\()_\w\()x\h\():429
+.if \w == 24 || \w == 32430
+ SAD_X_SVE2_\w \h, \x431
+ ret432
+.else433
+ SAD_X_START_\w \x434
+ mov w12, #\h/4435
+.loop_sad_sve2_gt_16_x\x\()_\w\()x\h:436
+ sub w12, w12, #1437
+ .rept 4438
+ .if \w == 24439
+ ld1 {v6.16b}, x0, #16440
+ ld1 {v7.8b}, x0, x9441
+ .elseif \w == 32442
+ ld1 {v6.16b-v7.16b}, x0, x9443
+ .elseif \w == 48444
+ ld1 {v4.16b-v6.16b}, x0, x9445
+ .elseif \w == 64446
+ ld1 {v4.16b-v7.16b}, x0, x9447
+ .endif448
+ SAD_X_\w x1, v16, v20449
+ SAD_X_\w x2, v17, v21450
+ SAD_X_\w x3, v18, v22451
+ .if \x == 4452
+ SAD_X_\w x4, v19, v23453
+ .endif454
+ .endr455
+ cbnz w12, .loop_sad_sve2_gt_16_x\x\()_\w\()x\h456
+ SAD_X_END_\w \x457
+.endif458
+endfunc459
+.endm460
+461
+462
+SAD_X_FUNC_SVE2 3, 4, 4463
+SAD_X_FUNC_SVE2 3, 4, 8464
+SAD_X_FUNC_SVE2 3, 4, 16465
+SAD_X_FUNC_SVE2 3, 8, 4466
+SAD_X_FUNC_SVE2 3, 8, 8467
+SAD_X_FUNC_SVE2 3, 8, 16468
+SAD_X_FUNC_SVE2 3, 8, 32469
+SAD_X_FUNC_SVE2 3, 12, 16470
+SAD_X_FUNC_SVE2 3, 16, 4471
+SAD_X_FUNC_SVE2 3, 16, 8472
+SAD_X_FUNC_SVE2 3, 16, 12473
+SAD_X_FUNC_SVE2 3, 16, 16474
+SAD_X_FUNC_SVE2 3, 16, 32475
+SAD_X_FUNC_SVE2 3, 16, 64476
+SAD_X_LOOP_SVE2 3, 24, 32477
+SAD_X_LOOP_SVE2 3, 32, 8478
+SAD_X_LOOP_SVE2 3, 32, 16479
+SAD_X_LOOP_SVE2 3, 32, 24480
+SAD_X_LOOP_SVE2 3, 32, 32481
+SAD_X_LOOP_SVE2 3, 32, 64482
+SAD_X_LOOP_SVE2 3, 48, 64483
+SAD_X_LOOP_SVE2 3, 64, 16484
+SAD_X_LOOP_SVE2 3, 64, 32485
+SAD_X_LOOP_SVE2 3, 64, 48486
+SAD_X_LOOP_SVE2 3, 64, 64487
+488
+SAD_X_FUNC_SVE2 4, 4, 4489
+SAD_X_FUNC_SVE2 4, 4, 8490
+SAD_X_FUNC_SVE2 4, 4, 16491
+SAD_X_FUNC_SVE2 4, 8, 4492
+SAD_X_FUNC_SVE2 4, 8, 8493
+SAD_X_FUNC_SVE2 4, 8, 16494
+SAD_X_FUNC_SVE2 4, 8, 32495
+SAD_X_FUNC_SVE2 4, 12, 16496
+SAD_X_FUNC_SVE2 4, 16, 4497
+SAD_X_FUNC_SVE2 4, 16, 8498
+SAD_X_FUNC_SVE2 4, 16, 12499
+SAD_X_FUNC_SVE2 4, 16, 16500
+SAD_X_FUNC_SVE2 4, 16, 32501
+SAD_X_FUNC_SVE2 4, 16, 64502
+SAD_X_LOOP_SVE2 4, 24, 32503
+SAD_X_LOOP_SVE2 4, 32, 8504
+SAD_X_LOOP_SVE2 4, 32, 16505
+SAD_X_LOOP_SVE2 4, 32, 24506
+SAD_X_LOOP_SVE2 4, 32, 32507
+SAD_X_LOOP_SVE2 4, 32, 64508
+SAD_X_LOOP_SVE2 4, 48, 64509
+SAD_X_LOOP_SVE2 4, 64, 16510
+SAD_X_LOOP_SVE2 4, 64, 32511
+SAD_X_LOOP_SVE2 4, 64, 48512
+SAD_X_LOOP_SVE2 4, 64, 64513
x265_3.5.tar.gz/source/common/aarch64/sad-a.S -> x265_3.6.tar.gz/source/common/aarch64/sad-a.S
Changed
256
1
2
/*****************************************************************************3
- * Copyright (C) 2020 MulticoreWare, Inc4
+ * Copyright (C) 2020-2021 MulticoreWare, Inc5
*6
* Authors: Hongbin Liu <liuhongbin1@huawei.com>7
+ * Sebastian Pop <spop@amazon.com>8
*9
* This program is free software; you can redistribute it and/or modify10
* it under the terms of the GNU General Public License as published by11
12
*****************************************************************************/13
14
#include "asm.S"15
+#include "sad-a-common.S"16
17
+#ifdef __APPLE__18
+.section __RODATA,__rodata19
+#else20
.section .rodata21
+#endif22
23
.align 424
25
.text26
27
-.macro SAD_X_START_8 x28
- ld1 {v0.8b}, x0, x929
-.if \x == 330
- ld1 {v1.8b}, x1, x431
- ld1 {v2.8b}, x2, x432
- ld1 {v3.8b}, x3, x433
-.elseif \x == 434
- ld1 {v1.8b}, x1, x535
- ld1 {v2.8b}, x2, x536
- ld1 {v3.8b}, x3, x537
- ld1 {v4.8b}, x4, x538
-.endif39
- uabdl v16.8h, v0.8b, v1.8b40
- uabdl v17.8h, v0.8b, v2.8b41
- uabdl v18.8h, v0.8b, v3.8b42
-.if \x == 443
- uabdl v19.8h, v0.8b, v4.8b44
+// Fully unrolled.45
+.macro SAD_FUNC w, h46
+function PFX(pixel_sad_\w\()x\h\()_neon)47
+ SAD_START_\w uabdl48
+ SAD_\w \h49
+.if \w > 450
+ add v16.8h, v16.8h, v17.8h51
.endif52
+ uaddlv s0, v16.8h53
+ fmov w0, s054
+ ret55
+endfunc56
+.endm57
+58
+// Loop unrolled 4.59
+.macro SAD_FUNC_LOOP w, h60
+function PFX(pixel_sad_\w\()x\h\()_neon)61
+ SAD_START_\w62
+63
+ mov w9, #\h/864
+.loop_\w\()x\h:65
+ sub w9, w9, #166
+.rept 467
+ SAD_\w68
+.endr69
+ cbnz w9, .loop_\w\()x\h70
+71
+ SAD_END_\w72
+endfunc73
.endm74
75
-.macro SAD_X_8 x76
- ld1 {v0.8b}, x0, x977
+SAD_FUNC 4, 478
+SAD_FUNC 4, 879
+SAD_FUNC 4, 1680
+SAD_FUNC 8, 481
+SAD_FUNC 8, 882
+SAD_FUNC 8, 1683
+SAD_FUNC 8, 3284
+SAD_FUNC 16, 485
+SAD_FUNC 16, 886
+SAD_FUNC 16, 1287
+SAD_FUNC 16, 1688
+SAD_FUNC 16, 3289
+SAD_FUNC 16, 6490
+91
+SAD_FUNC_LOOP 32, 892
+SAD_FUNC_LOOP 32, 1693
+SAD_FUNC_LOOP 32, 2494
+SAD_FUNC_LOOP 32, 3295
+SAD_FUNC_LOOP 32, 6496
+SAD_FUNC_LOOP 64, 1697
+SAD_FUNC_LOOP 64, 3298
+SAD_FUNC_LOOP 64, 4899
+SAD_FUNC_LOOP 64, 64100
+SAD_FUNC_LOOP 12, 16101
+SAD_FUNC_LOOP 24, 32102
+SAD_FUNC_LOOP 48, 64103
+104
+// SAD_X3 and SAD_X4 code start105
+106
+// static void x264_pixel_sad_x3_##size(pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, intptr_t i_stride, int scores3)107
+// static void x264_pixel_sad_x4_##size(pixel *fenc, pixel *pix0, pixel *pix1,pixel *pix2, pixel *pix3, intptr_t i_stride, int scores4)108
+.macro SAD_X_FUNC x, w, h109
+function PFX(sad_x\x\()_\w\()x\h\()_neon)110
+ mov x9, #FENC_STRIDE111
+112
+// Make function arguments for x == 3 look like x == 4.113
.if \x == 3114
- ld1 {v1.8b}, x1, x4115
- ld1 {v2.8b}, x2, x4116
- ld1 {v3.8b}, x3, x4117
-.elseif \x == 4118
- ld1 {v1.8b}, x1, x5119
- ld1 {v2.8b}, x2, x5120
- ld1 {v3.8b}, x3, x5121
- ld1 {v4.8b}, x4, x5122
+ mov x6, x5123
+ mov x5, x4124
.endif125
- uabal v16.8h, v0.8b, v1.8b126
- uabal v17.8h, v0.8b, v2.8b127
- uabal v18.8h, v0.8b, v3.8b128
-.if \x == 4129
- uabal v19.8h, v0.8b, v4.8b130
+131
+.if \w == 12132
+ movrel x12, sad12_mask133
+ ld1 {v31.16b}, x12134
.endif135
+136
+ SAD_X_START_\w \h, \x, uabdl137
+ SAD_X_\w \h, \x138
+ SAD_X_END_\w \x139
+endfunc140
.endm141
142
-.macro SAD_X_8xN x, h143
-function x265_sad_x\x\()_8x\h\()_neon144
+.macro SAD_X_LOOP x, w, h145
+function PFX(sad_x\x\()_\w\()x\h\()_neon)146
mov x9, #FENC_STRIDE147
- SAD_X_START_8 \x148
-.rept \h - 1149
- SAD_X_8 \x150
-.endr151
- uaddlv s0, v16.8h152
- uaddlv s1, v17.8h153
- uaddlv s2, v18.8h154
-.if \x == 4155
- uaddlv s3, v19.8h156
-.endif157
158
+// Make function arguments for x == 3 look like x == 4.159
.if \x == 3160
- stp s0, s1, x5161
- str s2, x5, #8162
-.elseif \x == 4163
- stp s0, s1, x6164
- stp s2, s3, x6, #8165
+ mov x6, x5166
+ mov x5, x4167
.endif168
- ret169
+ SAD_X_START_\w \x170
+ mov w12, #\h/4171
+.loop_sad_x\x\()_\w\()x\h:172
+ sub w12, w12, #1173
+ .rept 4174
+ .if \w == 24175
+ ld1 {v6.16b}, x0, #16176
+ ld1 {v7.8b}, x0, x9177
+ .elseif \w == 32178
+ ld1 {v6.16b-v7.16b}, x0, x9179
+ .elseif \w == 48180
+ ld1 {v4.16b-v6.16b}, x0, x9181
+ .elseif \w == 64182
+ ld1 {v4.16b-v7.16b}, x0, x9183
+ .endif184
+ SAD_X_\w x1, v16, v20185
+ SAD_X_\w x2, v17, v21186
+ SAD_X_\w x3, v18, v22187
+ .if \x == 4188
+ SAD_X_\w x4, v19, v23189
+ .endif190
+ .endr191
+ cbnz w12, .loop_sad_x\x\()_\w\()x\h192
+ SAD_X_END_\w \x193
endfunc194
.endm195
196
-SAD_X_8xN 3 4197
-SAD_X_8xN 3 8198
-SAD_X_8xN 3 16199
-SAD_X_8xN 3 32200
201
-SAD_X_8xN 4 4202
-SAD_X_8xN 4 8203
-SAD_X_8xN 4 16204
-SAD_X_8xN 4 32205
+SAD_X_FUNC 3, 4, 4206
+SAD_X_FUNC 3, 4, 8207
+SAD_X_FUNC 3, 4, 16208
+SAD_X_FUNC 3, 8, 4209
+SAD_X_FUNC 3, 8, 8210
+SAD_X_FUNC 3, 8, 16211
+SAD_X_FUNC 3, 8, 32212
+SAD_X_FUNC 3, 12, 16213
+SAD_X_FUNC 3, 16, 4214
+SAD_X_FUNC 3, 16, 8215
+SAD_X_FUNC 3, 16, 12216
+SAD_X_FUNC 3, 16, 16217
+SAD_X_FUNC 3, 16, 32218
+SAD_X_FUNC 3, 16, 64219
+SAD_X_LOOP 3, 24, 32220
+SAD_X_LOOP 3, 32, 8221
+SAD_X_LOOP 3, 32, 16222
+SAD_X_LOOP 3, 32, 24223
+SAD_X_LOOP 3, 32, 32224
+SAD_X_LOOP 3, 32, 64225
+SAD_X_LOOP 3, 48, 64226
+SAD_X_LOOP 3, 64, 16227
+SAD_X_LOOP 3, 64, 32228
+SAD_X_LOOP 3, 64, 48229
+SAD_X_LOOP 3, 64, 64230
+231
+SAD_X_FUNC 4, 4, 4232
+SAD_X_FUNC 4, 4, 8233
+SAD_X_FUNC 4, 4, 16234
+SAD_X_FUNC 4, 8, 4235
+SAD_X_FUNC 4, 8, 8236
+SAD_X_FUNC 4, 8, 16237
+SAD_X_FUNC 4, 8, 32238
+SAD_X_FUNC 4, 12, 16239
+SAD_X_FUNC 4, 16, 4240
+SAD_X_FUNC 4, 16, 8241
+SAD_X_FUNC 4, 16, 12242
+SAD_X_FUNC 4, 16, 16243
+SAD_X_FUNC 4, 16, 32244
+SAD_X_FUNC 4, 16, 64245
+SAD_X_LOOP 4, 24, 32246
+SAD_X_LOOP 4, 32, 8247
+SAD_X_LOOP 4, 32, 16248
+SAD_X_LOOP 4, 32, 24249
+SAD_X_LOOP 4, 32, 32250
+SAD_X_LOOP 4, 32, 64251
+SAD_X_LOOP 4, 48, 64252
+SAD_X_LOOP 4, 64, 16253
+SAD_X_LOOP 4, 64, 32254
+SAD_X_LOOP 4, 64, 48255
+SAD_X_LOOP 4, 64, 64256
x265_3.6.tar.gz/source/common/aarch64/ssd-a-common.S
Added
39
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+// This file contains the macros written using NEON instruction set26
+// that are also used by the SVE2 functions27
+28
+#include "asm.S"29
+30
+.arch armv8-a31
+32
+.macro ret_v0_w033
+ trn2 v1.2d, v0.2d, v0.2d34
+ add v0.2s, v0.2s, v1.2s35
+ addp v0.2s, v0.2s, v0.2s36
+ fmov w0, s037
+ ret38
+.endm39
x265_3.6.tar.gz/source/common/aarch64/ssd-a-sve.S
Added
80
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+27
+.arch armv8-a+sve28
+29
+#ifdef __APPLE__30
+.section __RODATA,__rodata31
+#else32
+.section .rodata33
+#endif34
+35
+.align 436
+37
+.text38
+39
+function PFX(pixel_sse_pp_4x4_sve)40
+ ptrue p0.s, vl441
+ ld1b {z0.s}, p0/z, x042
+ ld1b {z17.s}, p0/z, x243
+ add x0, x0, x144
+ add x2, x2, x345
+ sub z0.s, p0/m, z0.s, z17.s46
+ mul z0.s, p0/m, z0.s, z0.s47
+.rept 348
+ ld1b {z16.s}, p0/z, x049
+ ld1b {z17.s}, p0/z, x250
+ add x0, x0, x151
+ add x2, x2, x352
+ sub z16.s, p0/m, z16.s, z17.s53
+ mla z0.s, p0/m, z16.s, z16.s54
+.endr55
+ uaddv d0, p0, z0.s56
+ fmov w0, s057
+ ret58
+endfunc59
+60
+function PFX(pixel_sse_pp_4x8_sve)61
+ ptrue p0.s, vl462
+ ld1b {z0.s}, p0/z, x063
+ ld1b {z17.s}, p0/z, x264
+ add x0, x0, x165
+ add x2, x2, x366
+ sub z0.s, p0/m, z0.s, z17.s67
+ mul z0.s, p0/m, z0.s, z0.s68
+.rept 769
+ ld1b {z16.s}, p0/z, x070
+ ld1b {z17.s}, p0/z, x271
+ add x0, x0, x172
+ add x2, x2, x373
+ sub z16.s, p0/m, z16.s, z17.s74
+ mla z0.s, p0/m, z16.s, z16.s75
+.endr76
+ uaddv d0, p0, z0.s77
+ fmov w0, s078
+ ret79
+endfunc80
x265_3.6.tar.gz/source/common/aarch64/ssd-a-sve2.S
Added
889
1
2
+/*****************************************************************************3
+ * Copyright (C) 2022-2023 MulticoreWare, Inc4
+ *5
+ * Authors: David Chen <david.chen@myais.com.cn>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm-sve.S"26
+#include "ssd-a-common.S"27
+28
+.arch armv8-a+sve229
+30
+#ifdef __APPLE__31
+.section __RODATA,__rodata32
+#else33
+.section .rodata34
+#endif35
+36
+.align 437
+38
+.text39
+40
+function PFX(pixel_sse_pp_32x32_sve2)41
+ rdvl x9, #142
+ cmp x9, #1643
+ bgt .vl_gt_16_pixel_sse_pp_32x3244
+ mov w12, #845
+ movi v0.16b, #046
+ movi v1.16b, #047
+.loop_sse_pp_32_sve2:48
+ sub w12, w12, #149
+.rept 450
+ ld1 {v16.16b,v17.16b}, x0, x151
+ ld1 {v18.16b,v19.16b}, x2, x352
+ usubl v2.8h, v16.8b, v18.8b53
+ usubl2 v3.8h, v16.16b, v18.16b54
+ usubl v4.8h, v17.8b, v19.8b55
+ usubl2 v5.8h, v17.16b, v19.16b56
+ smlal v0.4s, v2.4h, v2.4h57
+ smlal2 v1.4s, v2.8h, v2.8h58
+ smlal v0.4s, v3.4h, v3.4h59
+ smlal2 v1.4s, v3.8h, v3.8h60
+ smlal v0.4s, v4.4h, v4.4h61
+ smlal2 v1.4s, v4.8h, v4.8h62
+ smlal v0.4s, v5.4h, v5.4h63
+ smlal2 v1.4s, v5.8h, v5.8h64
+.endr65
+ cbnz w12, .loop_sse_pp_32_sve266
+ add v0.4s, v0.4s, v1.4s67
+ ret_v0_w068
+.vl_gt_16_pixel_sse_pp_32x32:69
+ ptrue p0.b, vl3270
+ ld1b {z16.b}, p0/z, x071
+ ld1b {z18.b}, p0/z, x272
+ add x0, x0, x173
+ add x2, x2, x374
+ usublb z1.h, z16.b, z18.b75
+ usublt z2.h, z16.b, z18.b76
+ smullb z0.s, z1.h, z1.h77
+ smlalt z0.s, z1.h, z1.h78
+ smlalb z0.s, z2.h, z2.h79
+ smlalt z0.s, z2.h, z2.h80
+.rept 3181
+ ld1b {z16.b}, p0/z, x082
+ ld1b {z18.b}, p0/z, x283
+ add x0, x0, x184
+ add x2, x2, x385
+ usublb z1.h, z16.b, z18.b86
+ usublt z2.h, z16.b, z18.b87
+ smullb z0.s, z1.h, z1.h88
+ smlalt z0.s, z1.h, z1.h89
+ smlalb z0.s, z2.h, z2.h90
+ smlalt z0.s, z2.h, z2.h91
+.endr92
+ uaddv d3, p0, z0.s93
+ fmov w0, s394
+ ret95
+endfunc96
+97
+function PFX(pixel_sse_pp_32x64_sve2)98
+ rdvl x9, #199
+ cmp x9, #16100
+ bgt .vl_gt_16_pixel_sse_pp_32x64101
+ ptrue p0.b, vl16102
+ ld1b {z16.b}, p0/z, x0103
+ ld1b {z17.b}, p0/z, x0, #1, mul vl104
+ ld1b {z18.b}, p0/z, x2105
+ ld1b {z19.b}, p0/z, x2, #1, mul vl106
+ add x0, x0, x1107
+ add x2, x2, x3108
+ usublb z1.h, z16.b, z18.b109
+ usublt z2.h, z16.b, z18.b110
+ usublb z3.h, z17.b, z19.b111
+ usublt z4.h, z17.b, z19.b112
+ smullb z20.s, z1.h, z1.h113
+ smullt z21.s, z1.h, z1.h114
+ smlalb z20.s, z2.h, z2.h115
+ smlalt z21.s, z2.h, z2.h116
+ smlalb z20.s, z3.h, z3.h117
+ smlalt z21.s, z3.h, z3.h118
+ smlalb z20.s, z4.h, z4.h119
+ smlalt z21.s, z4.h, z4.h120
+.rept 63121
+ ld1b {z16.b}, p0/z, x0122
+ ld1b {z17.b}, p0/z, x0, #1, mul vl123
+ ld1b {z18.b}, p0/z, x2124
+ ld1b {z19.b}, p0/z, x2, #1, mul vl125
+ add x0, x0, x1126
+ add x2, x2, x3127
+ usublb z1.h, z16.b, z18.b128
+ usublt z2.h, z16.b, z18.b129
+ usublb z3.h, z17.b, z19.b130
+ usublt z4.h, z17.b, z19.b131
+ smlalb z20.s, z1.h, z1.h132
+ smlalt z21.s, z1.h, z1.h133
+ smlalb z20.s, z2.h, z2.h134
+ smlalt z21.s, z2.h, z2.h135
+ smlalb z20.s, z3.h, z3.h136
+ smlalt z21.s, z3.h, z3.h137
+ smlalb z20.s, z4.h, z4.h138
+ smlalt z21.s, z4.h, z4.h139
+.endr140
+ uaddv d3, p0, z20.s141
+ fmov w0, s3142
+ uaddv d4, p0, z21.s143
+ fmov w1, s4144
+ add w0, w0, w1145
+ ret146
+.vl_gt_16_pixel_sse_pp_32x64:147
+ ptrue p0.b, vl32148
+ ld1b {z16.b}, p0/z, x0149
+ ld1b {z18.b}, p0/z, x2150
+ add x0, x0, x1151
+ add x2, x2, x3152
+ usublb z1.h, z16.b, z18.b153
+ usublt z2.h, z16.b, z18.b154
+ smullb z20.s, z1.h, z1.h155
+ smullt z21.s, z1.h, z1.h156
+ smlalb z20.s, z2.h, z2.h157
+ smlalt z21.s, z2.h, z2.h158
+.rept 63159
+ ld1b {z16.b}, p0/z, x0160
+ ld1b {z18.b}, p0/z, x2161
+ add x0, x0, x1162
+ add x2, x2, x3163
+ usublb z1.h, z16.b, z18.b164
+ usublt z2.h, z16.b, z18.b165
+ smlalb z20.s, z1.h, z1.h166
+ smlalt z21.s, z1.h, z1.h167
+ smlalb z20.s, z2.h, z2.h168
+ smlalt z21.s, z2.h, z2.h169
+.endr170
+ uaddv d3, p0, z20.s171
+ fmov w0, s3172
+ uaddv d4, p0, z21.s173
+ fmov w1, s4174
+ add w0, w0, w1175
+ ret176
+endfunc177
+178
+function PFX(pixel_sse_pp_64x64_sve2)179
+ rdvl x9, #1180
+ cmp x9, #16181
+ bgt .vl_gt_16_pixel_sse_pp_64x64182
+ mov w12, #16183
+ movi v0.16b, #0184
+ movi v1.16b, #0185
+186
+.loop_sse_pp_64_sve2:187
+ sub w12, w12, #1188
+.rept 4189
+ ld1 {v16.16b-v19.16b}, x0, x1190
+ ld1 {v20.16b-v23.16b}, x2, x3191
+192
+ usubl v2.8h, v16.8b, v20.8b193
+ usubl2 v3.8h, v16.16b, v20.16b194
+ usubl v4.8h, v17.8b, v21.8b195
+ usubl2 v5.8h, v17.16b, v21.16b196
+ smlal v0.4s, v2.4h, v2.4h197
+ smlal2 v1.4s, v2.8h, v2.8h198
+ smlal v0.4s, v3.4h, v3.4h199
+ smlal2 v1.4s, v3.8h, v3.8h200
+ smlal v0.4s, v4.4h, v4.4h201
+ smlal2 v1.4s, v4.8h, v4.8h202
+ smlal v0.4s, v5.4h, v5.4h203
+ smlal2 v1.4s, v5.8h, v5.8h204
+205
+ usubl v2.8h, v18.8b, v22.8b206
+ usubl2 v3.8h, v18.16b, v22.16b207
+ usubl v4.8h, v19.8b, v23.8b208
+ usubl2 v5.8h, v19.16b, v23.16b209
+ smlal v0.4s, v2.4h, v2.4h210
+ smlal2 v1.4s, v2.8h, v2.8h211
+ smlal v0.4s, v3.4h, v3.4h212
+ smlal2 v1.4s, v3.8h, v3.8h213
+ smlal v0.4s, v4.4h, v4.4h214
+ smlal2 v1.4s, v4.8h, v4.8h215
+ smlal v0.4s, v5.4h, v5.4h216
+ smlal2 v1.4s, v5.8h, v5.8h217
+.endr218
+ cbnz w12, .loop_sse_pp_64_sve2219
+ add v0.4s, v0.4s, v1.4s220
+ ret_v0_w0221
+.vl_gt_16_pixel_sse_pp_64x64:222
+ cmp x9, #48223
+ bgt .vl_gt_48_pixel_sse_pp_64x64224
+ ptrue p0.b, vl32225
+ ld1b {z16.b}, p0/z, x0226
+ ld1b {z17.b}, p0/z, x0, #1, mul vl227
+ ld1b {z20.b}, p0/z, x2228
+ ld1b {z21.b}, p0/z, x2, #1, mul vl229
+ add x0, x0, x1230
+ add x2, x2, x3231
+ usublb z1.h, z16.b, z20.b232
+ usublt z2.h, z16.b, z20.b233
+ usublb z3.h, z17.b, z21.b234
+ usublt z4.h, z17.b, z21.b235
+ smullb z24.s, z1.h, z1.h236
+ smullt z25.s, z1.h, z1.h237
+ smlalb z24.s, z2.h, z2.h238
+ smlalt z25.s, z2.h, z2.h239
+ smlalb z24.s, z3.h, z3.h240
+ smlalt z25.s, z3.h, z3.h241
+ smlalb z24.s, z4.h, z4.h242
+ smlalt z25.s, z4.h, z4.h243
+.rept 63244
+ ld1b {z16.b}, p0/z, x0245
+ ld1b {z17.b}, p0/z, x0, #1, mul vl246
+ ld1b {z20.b}, p0/z, x2247
+ ld1b {z21.b}, p0/z, x2, #1, mul vl248
+ add x0, x0, x1249
+ add x2, x2, x3250
+ usublb z1.h, z16.b, z20.b251
+ usublt z2.h, z16.b, z20.b252
+ usublb z3.h, z17.b, z21.b253
+ usublt z4.h, z17.b, z21.b254
+ smlalb z24.s, z1.h, z1.h255
+ smlalt z25.s, z1.h, z1.h256
+ smlalb z24.s, z2.h, z2.h257
+ smlalt z25.s, z2.h, z2.h258
+ smlalb z24.s, z3.h, z3.h259
+ smlalt z25.s, z3.h, z3.h260
+ smlalb z24.s, z4.h, z4.h261
+ smlalt z25.s, z4.h, z4.h262
+.endr263
+ uaddv d3, p0, z24.s264
+ fmov w0, s3265
+ uaddv d4, p0, z25.s266
+ fmov w1, s4267
+ add w0, w0, w1268
+ ret269
+.vl_gt_48_pixel_sse_pp_64x64:270
+ ptrue p0.b, vl64271
+ ld1b {z16.b}, p0/z, x0272
+ ld1b {z20.b}, p0/z, x2273
+ add x0, x0, x1274
+ add x2, x2, x3275
+ usublb z1.h, z16.b, z20.b276
+ usublt z2.h, z16.b, z20.b277
+ smullb z24.s, z1.h, z1.h278
+ smullt z25.s, z1.h, z1.h279
+ smlalb z24.s, z2.h, z2.h280
+ smlalt z25.s, z2.h, z2.h281
+.rept 63282
+ ld1b {z16.b}, p0/z, x0283
+ ld1b {z20.b}, p0/z, x2284
+ add x0, x0, x1285
+ add x2, x2, x3286
+ usublb z1.h, z16.b, z20.b287
+ usublt z2.h, z16.b, z20.b288
+ smlalb z24.s, z1.h, z1.h289
+ smlalt z25.s, z1.h, z1.h290
+ smlalb z24.s, z2.h, z2.h291
+ smlalt z25.s, z2.h, z2.h292
+.endr293
+ uaddv d3, p0, z24.s294
+ fmov w0, s3295
+ uaddv d4, p0, z25.s296
+ fmov w1, s4297
+ add w0, w0, w1298
+ ret299
+endfunc300
+301
+function PFX(pixel_sse_ss_4x4_sve2)302
+ ptrue p0.b, vl8303
+ ld1b {z16.b}, p0/z, x0304
+ ld1b {z17.b}, p0/z, x2305
+ add x0, x0, x1, lsl #1306
+ add x2, x2, x3, lsl #1307
+ sub z1.h, z16.h, z17.h308
+ smullb z3.s, z1.h, z1.h309
+ smullt z4.s, z1.h, z1.h310
+.rept 3311
+ ld1b {z16.b}, p0/z, x0312
+ ld1b {z17.b}, p0/z, x2313
+ add x0, x0, x1, lsl #1314
+ add x2, x2, x3, lsl #1315
+ sub z1.h, z16.h, z17.h316
+ smlalb z3.s, z1.h, z1.h317
+ smlalt z4.s, z1.h, z1.h318
+.endr319
+ uaddv d3, p0, z3.s320
+ fmov w0, s3321
+ uaddv d4, p0, z4.s322
+ fmov w1, s4323
+ add w0, w0, w1324
+ ret325
+endfunc326
+327
+function PFX(pixel_sse_ss_8x8_sve2)328
+ ptrue p0.b, vl16329
+ ld1b {z16.b}, p0/z, x0330
+ ld1b {z17.b}, p0/z, x2331
+ add x0, x0, x1, lsl #1332
+ add x2, x2, x3, lsl #1333
+ sub z1.h, z16.h, z17.h334
+ smullb z3.s, z1.h, z1.h335
+ smullt z4.s, z1.h, z1.h336
+.rept 7337
+ ld1b {z16.b}, p0/z, x0338
+ ld1b {z17.b}, p0/z, x2339
+ add x0, x0, x1, lsl #1340
+ add x2, x2, x3, lsl #1341
+ sub z1.h, z16.h, z17.h342
+ smlalb z3.s, z1.h, z1.h343
+ smlalt z4.s, z1.h, z1.h344
+.endr345
+ uaddv d3, p0, z3.s346
+ fmov w0, s3347
+ uaddv d4, p0, z4.s348
+ fmov w1, s4349
+ add w0, w0, w1350
+ ret351
+endfunc352
+353
+function PFX(pixel_sse_ss_16x16_sve2)354
+ rdvl x9, #1355
+ cmp x9, #16356
+ bgt .vl_gt_16_pixel_sse_ss_16x16357
+ ptrue p0.b, vl16358
+ ld1b {z16.b}, p0/z, x0359
+ ld1b {z17.b}, p0/z, x0, #1, mul vl360
+ ld1b {z18.b}, p0/z, x2361
+ ld1b {z19.b}, p0/z, x2, #1, mul vl362
+ add x0, x0, x1, lsl #1363
+ add x2, x2, x3, lsl #1364
+ sub z1.h, z16.h, z18.h365
+ sub z2.h, z17.h, z19.h366
+ smullb z3.s, z1.h, z1.h367
+ smullt z4.s, z1.h, z1.h368
+ smlalb z3.s, z2.h, z2.h369
+ smlalt z4.s, z2.h, z2.h370
+.rept 15371
+ ld1b {z16.b}, p0/z, x0372
+ ld1b {z17.b}, p0/z, x0, #1, mul vl373
+ ld1b {z18.b}, p0/z, x2374
+ ld1b {z19.b}, p0/z, x2, #1, mul vl375
+ add x0, x0, x1, lsl #1376
+ add x2, x2, x3, lsl #1377
+ sub z1.h, z16.h, z18.h378
+ sub z2.h, z17.h, z19.h379
+ smlalb z3.s, z1.h, z1.h380
+ smlalt z4.s, z1.h, z1.h381
+ smlalb z3.s, z2.h, z2.h382
+ smlalt z4.s, z2.h, z2.h383
+.endr384
+ uaddv d3, p0, z3.s385
+ fmov w0, s3386
+ uaddv d4, p0, z4.s387
+ fmov w1, s4388
+ add w0, w0, w1389
+ ret390
+.vl_gt_16_pixel_sse_ss_16x16:391
+ ptrue p0.b, vl32392
+ ld1b {z16.b}, p0/z, x0393
+ ld1b {z18.b}, p0/z, x2394
+ add x0, x0, x1, lsl #1395
+ add x2, x2, x3, lsl #1396
+ sub z1.h, z16.h, z18.h397
+ smullb z3.s, z1.h, z1.h398
+ smullt z4.s, z1.h, z1.h399
+.rept 15400
+ ld1b {z16.b}, p0/z, x0401
+ ld1b {z18.b}, p0/z, x2402
+ add x0, x0, x1, lsl #1403
+ add x2, x2, x3, lsl #1404
+ sub z1.h, z16.h, z18.h405
+ smlalb z3.s, z1.h, z1.h406
+ smlalt z4.s, z1.h, z1.h407
+.endr408
+ uaddv d3, p0, z3.s409
+ fmov w0, s3410
+ uaddv d4, p0, z4.s411
+ fmov w1, s4412
+ add w0, w0, w1413
+ ret414
+endfunc415
+416
+function PFX(pixel_sse_ss_32x32_sve2)417
+ rdvl x9, #1418
+ cmp x9, #16419
+ bgt .vl_gt_16_pixel_sse_ss_32x32420
+ ptrue p0.b, vl16421
+ ld1b {z16.b}, p0/z, x0422
+ ld1b {z17.b}, p0/z, x0, #1, mul vl423
+ ld1b {z18.b}, p0/z, x0, #2, mul vl424
+ ld1b {z19.b}, p0/z, x0, #3, mul vl425
+ ld1b {z20.b}, p0/z, x2426
+ ld1b {z21.b}, p0/z, x2, #1, mul vl427
+ ld1b {z22.b}, p0/z, x2, #2, mul vl428
+ ld1b {z23.b}, p0/z, x2, #3, mul vl429
+ add x0, x0, x1, lsl #1430
+ add x2, x2, x3, lsl #1431
+ sub z1.h, z16.h, z20.h432
+ sub z2.h, z17.h, z21.h433
+ sub z3.h, z18.h, z22.h434
+ sub z4.h, z19.h, z23.h435
+ smullb z5.s, z1.h, z1.h436
+ smullt z6.s, z1.h, z1.h437
+ smlalb z5.s, z2.h, z2.h438
+ smlalt z6.s, z2.h, z2.h439
+ smlalb z5.s, z3.h, z3.h440
+ smlalt z6.s, z3.h, z3.h441
+ smlalb z5.s, z4.h, z4.h442
+ smlalt z6.s, z4.h, z4.h443
+.rept 31444
+ ld1b {z16.b}, p0/z, x0445
+ ld1b {z17.b}, p0/z, x0, #1, mul vl446
+ ld1b {z18.b}, p0/z, x0, #2, mul vl447
+ ld1b {z19.b}, p0/z, x0, #3, mul vl448
+ ld1b {z20.b}, p0/z, x2449
+ ld1b {z21.b}, p0/z, x2, #1, mul vl450
+ ld1b {z22.b}, p0/z, x2, #2, mul vl451
+ ld1b {z23.b}, p0/z, x2, #3, mul vl452
+ add x0, x0, x1, lsl #1453
+ add x2, x2, x3, lsl #1454
+ sub z1.h, z16.h, z20.h455
+ sub z2.h, z17.h, z21.h456
+ sub z3.h, z18.h, z22.h457
+ sub z4.h, z19.h, z23.h458
+ smlalb z5.s, z1.h, z1.h459
+ smlalt z6.s, z1.h, z1.h460
+ smlalb z5.s, z2.h, z2.h461
+ smlalt z6.s, z2.h, z2.h462
+ smlalb z5.s, z3.h, z3.h463
+ smlalt z6.s, z3.h, z3.h464
+ smlalb z5.s, z4.h, z4.h465
+ smlalt z6.s, z4.h, z4.h466
+.endr467
+ uaddv d3, p0, z5.s468
+ fmov w0, s3469
+ uaddv d4, p0, z6.s470
+ fmov w1, s4471
+ add w0, w0, w1472
+ ret473
+.vl_gt_16_pixel_sse_ss_32x32:474
+ cmp x9, #48475
+ bgt .vl_gt_48_pixel_sse_ss_32x32476
+ ptrue p0.b, vl32477
+ ld1b {z16.b}, p0/z, x0478
+ ld1b {z17.b}, p0/z, x0, #1, mul vl479
+ ld1b {z20.b}, p0/z, x2480
+ ld1b {z21.b}, p0/z, x2, #1, mul vl481
+ add x0, x0, x1, lsl #1482
+ add x2, x2, x3, lsl #1483
+ sub z1.h, z16.h, z20.h484
+ sub z2.h, z17.h, z21.h485
+ smullb z5.s, z1.h, z1.h486
+ smullt z6.s, z1.h, z1.h487
+ smlalb z5.s, z2.h, z2.h488
+ smlalt z6.s, z2.h, z2.h489
+.rept 31490
+ ld1b {z16.b}, p0/z, x0491
+ ld1b {z17.b}, p0/z, x0, #1, mul vl492
+ ld1b {z20.b}, p0/z, x2493
+ ld1b {z21.b}, p0/z, x2, #1, mul vl494
+ add x0, x0, x1, lsl #1495
+ add x2, x2, x3, lsl #1496
+ sub z1.h, z16.h, z20.h497
+ sub z2.h, z17.h, z21.h498
+ smlalb z5.s, z1.h, z1.h499
+ smlalt z6.s, z1.h, z1.h500
+ smlalb z5.s, z2.h, z2.h501
+ smlalt z6.s, z2.h, z2.h502
+.endr503
+ uaddv d3, p0, z5.s504
+ fmov w0, s3505
+ uaddv d4, p0, z6.s506
+ fmov w1, s4507
+ add w0, w0, w1508
+ ret509
+.vl_gt_48_pixel_sse_ss_32x32:510
+ ptrue p0.b, vl64511
+ ld1b {z16.b}, p0/z, x0512
+ ld1b {z20.b}, p0/z, x2513
+ add x0, x0, x1, lsl #1514
+ add x2, x2, x3, lsl #1515
+ sub z1.h, z16.h, z20.h516
+ smullb z5.s, z1.h, z1.h517
+ smullt z6.s, z1.h, z1.h518
+.rept 31519
+ ld1b {z16.b}, p0/z, x0520
+ ld1b {z20.b}, p0/z, x2521
+ add x0, x0, x1, lsl #1522
+ add x2, x2, x3, lsl #1523
+ sub z1.h, z16.h, z20.h524
+ smlalb z5.s, z1.h, z1.h525
+ smlalt z6.s, z1.h, z1.h526
+.endr527
+ uaddv d3, p0, z5.s528
+ fmov w0, s3529
+ uaddv d4, p0, z6.s530
+ fmov w1, s4531
+ add w0, w0, w1532
+ ret533
+endfunc534
+535
+function PFX(pixel_sse_ss_64x64_sve2)536
+ rdvl x9, #1537
+ cmp x9, #16538
+ bgt .vl_gt_16_pixel_sse_ss_64x64539
+ ptrue p0.b, vl16540
+ ld1b {z24.b}, p0/z, x0541
+ ld1b {z25.b}, p0/z, x0, #1, mul vl542
+ ld1b {z26.b}, p0/z, x0, #2, mul vl543
+ ld1b {z27.b}, p0/z, x0, #3, mul vl544
+ ld1b {z28.b}, p0/z, x2545
+ ld1b {z29.b}, p0/z, x2, #1, mul vl546
+ ld1b {z30.b}, p0/z, x2, #2, mul vl547
+ ld1b {z31.b}, p0/z, x2, #3, mul vl548
+ sub z0.h, z24.h, z28.h549
+ sub z1.h, z25.h, z29.h550
+ sub z2.h, z26.h, z30.h551
+ sub z3.h, z27.h, z31.h552
+ smullb z5.s, z0.h, z0.h553
+ smullt z6.s, z0.h, z0.h554
+ smlalb z5.s, z1.h, z1.h555
+ smlalt z6.s, z1.h, z1.h556
+ smlalb z5.s, z2.h, z2.h557
+ smlalt z6.s, z2.h, z2.h558
+ smlalb z5.s, z3.h, z3.h559
+ smlalt z6.s, z3.h, z3.h560
+ ld1b {z24.b}, p0/z, x0, #4, mul vl561
+ ld1b {z25.b}, p0/z, x0, #5, mul vl562
+ ld1b {z26.b}, p0/z, x0, #6, mul vl563
+ ld1b {z27.b}, p0/z, x0, #7, mul vl564
+ ld1b {z28.b}, p0/z, x2, #4, mul vl565
+ ld1b {z29.b}, p0/z, x2, #5, mul vl566
+ ld1b {z30.b}, p0/z, x2, #6, mul vl567
+ ld1b {z31.b}, p0/z, x2, #7, mul vl568
+ sub z0.h, z24.h, z28.h569
+ sub z1.h, z25.h, z29.h570
+ sub z2.h, z26.h, z30.h571
+ sub z3.h, z27.h, z31.h572
+ smlalb z5.s, z0.h, z0.h573
+ smlalt z6.s, z0.h, z0.h574
+ smlalb z5.s, z1.h, z1.h575
+ smlalt z6.s, z1.h, z1.h576
+ smlalb z5.s, z2.h, z2.h577
+ smlalt z6.s, z2.h, z2.h578
+ smlalb z5.s, z3.h, z3.h579
+ smlalt z6.s, z3.h, z3.h580
+ add x0, x0, x1, lsl #1581
+ add x2, x2, x3, lsl #1582
+.rept 63583
+ ld1b {z24.b}, p0/z, x0584
+ ld1b {z25.b}, p0/z, x0, #1, mul vl585
+ ld1b {z26.b}, p0/z, x0, #2, mul vl586
+ ld1b {z27.b}, p0/z, x0, #3, mul vl587
+ ld1b {z28.b}, p0/z, x2588
+ ld1b {z29.b}, p0/z, x2, #1, mul vl589
+ ld1b {z30.b}, p0/z, x2, #2, mul vl590
+ ld1b {z31.b}, p0/z, x2, #3, mul vl591
+ sub z0.h, z24.h, z28.h592
+ sub z1.h, z25.h, z29.h593
+ sub z2.h, z26.h, z30.h594
+ sub z3.h, z27.h, z31.h595
+ smlalb z5.s, z0.h, z0.h596
+ smlalt z6.s, z0.h, z0.h597
+ smlalb z5.s, z1.h, z1.h598
+ smlalt z6.s, z1.h, z1.h599
+ smlalb z5.s, z2.h, z2.h600
+ smlalt z6.s, z2.h, z2.h601
+ smlalb z5.s, z3.h, z3.h602
+ smlalt z6.s, z3.h, z3.h603
+ ld1b {z24.b}, p0/z, x0, #4, mul vl604
+ ld1b {z25.b}, p0/z, x0, #5, mul vl605
+ ld1b {z26.b}, p0/z, x0, #6, mul vl606
+ ld1b {z27.b}, p0/z, x0, #7, mul vl607
+ ld1b {z28.b}, p0/z, x2, #4, mul vl608
+ ld1b {z29.b}, p0/z, x2, #5, mul vl609
+ ld1b {z30.b}, p0/z, x2, #6, mul vl610
+ ld1b {z31.b}, p0/z, x2, #7, mul vl611
+ sub z0.h, z24.h, z28.h612
+ sub z1.h, z25.h, z29.h613
+ sub z2.h, z26.h, z30.h614
+ sub z3.h, z27.h, z31.h615
+ smlalb z5.s, z0.h, z0.h616
+ smlalt z6.s, z0.h, z0.h617
+ smlalb z5.s, z1.h, z1.h618
+ smlalt z6.s, z1.h, z1.h619
+ smlalb z5.s, z2.h, z2.h620
+ smlalt z6.s, z2.h, z2.h621
+ smlalb z5.s, z3.h, z3.h622
+ smlalt z6.s, z3.h, z3.h623
+ add x0, x0, x1, lsl #1624
+ add x2, x2, x3, lsl #1625
+.endr626
+ uaddv d3, p0, z5.s627
+ fmov w0, s3628
+ uaddv d4, p0, z6.s629
+ fmov w1, s4630
+ add w0, w0, w1631
+ ret632
+.vl_gt_16_pixel_sse_ss_64x64:633
+ cmp x9, #48634
+ bgt .vl_gt_48_pixel_sse_ss_64x64635
+ ptrue p0.b, vl32636
+ ld1b {z24.b}, p0/z, x0637
+ ld1b {z25.b}, p0/z, x0, #1, mul vl638
+ ld1b {z28.b}, p0/z, x2639
+ ld1b {z29.b}, p0/z, x2, #1, mul vl640
+ sub z0.h, z24.h, z28.h641
+ sub z1.h, z25.h, z29.h642
+ smullb z5.s, z0.h, z0.h643
+ smullt z6.s, z0.h, z0.h644
+ smlalb z5.s, z1.h, z1.h645
+ smlalt z6.s, z1.h, z1.h646
+ ld1b {z24.b}, p0/z, x0, #1, mul vl647
+ ld1b {z25.b}, p0/z, x0, #2, mul vl648
+ ld1b {z28.b}, p0/z, x2, #1, mul vl649
+ ld1b {z29.b}, p0/z, x2, #2, mul vl650
+ sub z0.h, z24.h, z28.h651
+ sub z1.h, z25.h, z29.h652
+ smlalb z5.s, z0.h, z0.h653
+ smlalt z6.s, z0.h, z0.h654
+ smlalb z5.s, z1.h, z1.h655
+ smlalt z6.s, z1.h, z1.h656
+ add x0, x0, x1, lsl #1657
+ add x2, x2, x3, lsl #1658
+.rept 63659
+ ld1b {z24.b}, p0/z, x0660
+ ld1b {z25.b}, p0/z, x0, #1, mul vl661
+ ld1b {z28.b}, p0/z, x2662
+ ld1b {z29.b}, p0/z, x2, #1, mul vl663
+ sub z0.h, z24.h, z28.h664
+ sub z1.h, z25.h, z29.h665
+ smlalb z5.s, z0.h, z0.h666
+ smlalt z6.s, z0.h, z0.h667
+ smlalb z5.s, z1.h, z1.h668
+ smlalt z6.s, z1.h, z1.h669
+ ld1b {z24.b}, p0/z, x0, #1, mul vl670
+ ld1b {z25.b}, p0/z, x0, #2, mul vl671
+ ld1b {z28.b}, p0/z, x2, #1, mul vl672
+ ld1b {z29.b}, p0/z, x2, #2, mul vl673
+ sub z0.h, z24.h, z28.h674
+ sub z1.h, z25.h, z29.h675
+ smlalb z5.s, z0.h, z0.h676
+ smlalt z6.s, z0.h, z0.h677
+ smlalb z5.s, z1.h, z1.h678
+ smlalt z6.s, z1.h, z1.h679
+ add x0, x0, x1, lsl #1680
+ add x2, x2, x3, lsl #1681
+.endr682
+ uaddv d3, p0, z5.s683
+ fmov w0, s3684
+ uaddv d4, p0, z6.s685
+ fmov w1, s4686
+ add w0, w0, w1687
+ ret688
+.vl_gt_48_pixel_sse_ss_64x64:689
+ cmp x9, #112690
+ bgt .vl_gt_112_pixel_sse_ss_64x64691
+ ptrue p0.b, vl64692
+ ld1b {z24.b}, p0/z, x0693
+ ld1b {z28.b}, p0/z, x2694
+ sub z0.h, z24.h, z28.h695
+ smullb z5.s, z0.h, z0.h696
+ smullt z6.s, z0.h, z0.h697
+ ld1b {z24.b}, p0/z, x0, #1, mul vl698
+ ld1b {z28.b}, p0/z, x2, #1, mul vl699
+ sub z0.h, z24.h, z28.h700
+ smlalb z5.s, z0.h, z0.h701
+ smlalt z6.s, z0.h, z0.h702
+ add x0, x0, x1, lsl #1703
+ add x2, x2, x3, lsl #1704
+.rept 63705
+ ld1b {z24.b}, p0/z, x0706
+ ld1b {z28.b}, p0/z, x2707
+ sub z0.h, z24.h, z28.h708
+ smlalb z5.s, z0.h, z0.h709
+ smlalt z6.s, z0.h, z0.h710
+ ld1b {z24.b}, p0/z, x0, #1, mul vl711
+ ld1b {z28.b}, p0/z, x2, #1, mul vl712
+ sub z0.h, z24.h, z28.h713
+ smlalb z5.s, z0.h, z0.h714
+ smlalt z6.s, z0.h, z0.h715
+ add x0, x0, x1, lsl #1716
+ add x2, x2, x3, lsl #1717
+.endr718
+ uaddv d3, p0, z5.s719
+ fmov w0, s3720
+ uaddv d4, p0, z6.s721
+ fmov w1, s4722
+ add w0, w0, w1723
+ ret724
+.vl_gt_112_pixel_sse_ss_64x64:725
+ ptrue p0.b, vl128726
+ ld1b {z24.b}, p0/z, x0727
+ ld1b {z28.b}, p0/z, x2728
+ sub z0.h, z24.h, z28.h729
+ smullb z5.s, z0.h, z0.h730
+ smullt z6.s, z0.h, z0.h731
+ add x0, x0, x1, lsl #1732
+ add x2, x2, x3, lsl #1733
+.rept 63734
+ ld1b {z24.b}, p0/z, x0735
+ ld1b {z28.b}, p0/z, x2736
+ sub z0.h, z24.h, z28.h737
+ smlalb z5.s, z0.h, z0.h738
+ smlalt z6.s, z0.h, z0.h739
+ add x0, x0, x1, lsl #1740
+ add x2, x2, x3, lsl #1741
+.endr742
+ uaddv d3, p0, z5.s743
+ fmov w0, s3744
+ uaddv d4, p0, z6.s745
+ fmov w1, s4746
+ add w0, w0, w1747
+ ret748
+endfunc749
+750
+function PFX(pixel_ssd_s_4x4_sve2)751
+ ptrue p0.b, vl8752
+ ld1b {z16.b}, p0/z, x0753
+ add x0, x0, x1, lsl #1754
+ smullb z0.s, z16.h, z16.h755
+ smlalt z0.s, z16.h, z16.h756
+.rept 3757
+ ld1b {z16.b}, p0/z, x0758
+ add x0, x0, x1, lsl #1759
+ smlalb z0.s, z16.h, z16.h760
+ smlalt z0.s, z16.h, z16.h761
+.endr762
+ uaddv d3, p0, z0.s763
+ fmov w0, s3764
+ ret765
+endfunc766
+767
+function PFX(pixel_ssd_s_8x8_sve2)768
+ ptrue p0.b, vl16769
+ ld1b {z16.b}, p0/z, x0770
+ add x0, x0, x1, lsl #1771
+ smullb z0.s, z16.h, z16.h772
+ smlalt z0.s, z16.h, z16.h773
+.rept 7774
+ ld1b {z16.b}, p0/z, x0775
+ add x0, x0, x1, lsl #1776
+ smlalb z0.s, z16.h, z16.h777
+ smlalt z0.s, z16.h, z16.h778
+.endr779
+ uaddv d3, p0, z0.s780
+ fmov w0, s3781
+ ret782
+endfunc783
+784
+function PFX(pixel_ssd_s_16x16_sve2)785
+ rdvl x9, #1786
+ cmp x9, #16787
+ bgt .vl_gt_16_pixel_ssd_s_16x16788
+ add x1, x1, x1789
+ mov w12, #4790
+ movi v0.16b, #0791
+ movi v1.16b, #0792
+.loop_ssd_s_16_sve2:793
+ sub w12, w12, #1794
+.rept 2795
+ ld1 {v4.16b,v5.16b}, x0, x1796
+ ld1 {v6.16b,v7.16b}, x0, x1797
+ smlal v0.4s, v4.4h, v4.4h798
+ smlal2 v1.4s, v4.8h, v4.8h799
+ smlal v0.4s, v5.4h, v5.4h800
+ smlal2 v1.4s, v5.8h, v5.8h801
+ smlal v0.4s, v6.4h, v6.4h802
+ smlal2 v1.4s, v6.8h, v6.8h803
+ smlal v0.4s, v7.4h, v7.4h804
+ smlal2 v1.4s, v7.8h, v7.8h805
+.endr806
+ cbnz w12, .loop_ssd_s_16_sve2807
+ add v0.4s, v0.4s, v1.4s808
+ ret_v0_w0809
+.vl_gt_16_pixel_ssd_s_16x16:810
+ ptrue p0.b, vl32811
+ ld1b {z16.b}, p0/z, x0812
+ add x0, x0, x1, lsl #1813
+ smullb z0.s, z16.h, z16.h814
+ smlalt z0.s, z16.h, z16.h815
+.rept 15816
+ ld1b {z16.b}, p0/z, x0817
+ add x0, x0, x1, lsl #1818
+ smlalb z0.s, z16.h, z16.h819
+ smlalt z0.s, z16.h, z16.h820
+.endr821
+ uaddv d3, p0, z0.s822
+ fmov w0, s3823
+ ret824
+endfunc825
+826
+function PFX(pixel_ssd_s_32x32_sve2)827
+ rdvl x9, #1828
+ cmp x9, #16829
+ bgt .vl_gt_16_pixel_ssd_s_32x32830
+ add x1, x1, x1831
+ mov w12, #8832
+ movi v0.16b, #0833
+ movi v1.16b, #0834
+.loop_ssd_s_32:835
+ sub w12, w12, #1836
+.rept 4837
+ ld1 {v4.16b-v7.16b}, x0, x1838
+ smlal v0.4s, v4.4h, v4.4h839
+ smlal2 v1.4s, v4.8h, v4.8h840
+ smlal v0.4s, v5.4h, v5.4h841
+ smlal2 v1.4s, v5.8h, v5.8h842
+ smlal v0.4s, v6.4h, v6.4h843
+ smlal2 v1.4s, v6.8h, v6.8h844
+ smlal v0.4s, v7.4h, v7.4h845
+ smlal2 v1.4s, v7.8h, v7.8h846
+.endr847
+ cbnz w12, .loop_ssd_s_32848
+ add v0.4s, v0.4s, v1.4s849
+ ret_v0_w0850
+.vl_gt_16_pixel_ssd_s_32x32:851
+ cmp x9, #48852
+ bgt .vl_gt_48_pixel_ssd_s_32x32853
+ ptrue p0.b, vl32854
+ ld1b {z16.b}, p0/z, x0855
+ ld1b {z17.b}, p0/z, x0, #1, mul vl856
+ add x0, x0, x1, lsl #1857
+ smullb z0.s, z16.h, z16.h858
+ smlalt z0.s, z16.h, z16.h859
+ smlalb z0.s, z17.h, z17.h860
+ smlalt z0.s, z17.h, z17.h861
+.rept 31862
+ ld1b {z16.b}, p0/z, x0863
+ ld1b {z17.b}, p0/z, x0, #1, mul vl864
+ add x0, x0, x1, lsl #1865
+ smlalb z0.s, z16.h, z16.h866
+ smlalt z0.s, z16.h, z16.h867
+ smlalb z0.s, z17.h, z17.h868
+ smlalt z0.s, z17.h, z17.h869
+.endr870
+ uaddv d3, p0, z0.s871
+ fmov w0, s3872
+ ret873
+.vl_gt_48_pixel_ssd_s_32x32:874
+ ptrue p0.b, vl64875
+ ld1b {z16.b}, p0/z, x0876
+ add x0, x0, x1, lsl #1877
+ smullb z0.s, z16.h, z16.h878
+ smlalt z0.s, z16.h, z16.h879
+.rept 31880
+ ld1b {z16.b}, p0/z, x0881
+ add x0, x0, x1, lsl #1882
+ smlalb z0.s, z16.h, z16.h883
+ smlalt z0.s, z16.h, z16.h884
+.endr885
+ uaddv d3, p0, z0.s886
+ fmov w0, s3887
+ ret888
+endfunc889
x265_3.6.tar.gz/source/common/aarch64/ssd-a.S
Added
478
1
2
+/*****************************************************************************3
+ * Copyright (C) 2021 MulticoreWare, Inc4
+ *5
+ * Authors: Sebastian Pop <spop@amazon.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com.23
+ *****************************************************************************/24
+25
+#include "asm.S"26
+#include "ssd-a-common.S"27
+28
+#ifdef __APPLE__29
+.section __RODATA,__rodata30
+#else31
+.section .rodata32
+#endif33
+34
+.align 435
+36
+.text37
+38
+function PFX(pixel_sse_pp_4x4_neon)39
+ ld1 {v16.s}0, x0, x140
+ ld1 {v17.s}0, x2, x341
+ ld1 {v18.s}0, x0, x142
+ ld1 {v19.s}0, x2, x343
+ ld1 {v20.s}0, x0, x144
+ ld1 {v21.s}0, x2, x345
+ ld1 {v22.s}0, x0, x146
+ ld1 {v23.s}0, x2, x347
+48
+ usubl v1.8h, v16.8b, v17.8b49
+ usubl v2.8h, v18.8b, v19.8b50
+ usubl v3.8h, v20.8b, v21.8b51
+ usubl v4.8h, v22.8b, v23.8b52
+53
+ smull v0.4s, v1.4h, v1.4h54
+ smlal v0.4s, v2.4h, v2.4h55
+ smlal v0.4s, v3.4h, v3.4h56
+ smlal v0.4s, v4.4h, v4.4h57
+ ret_v0_w058
+endfunc59
+60
+function PFX(pixel_sse_pp_4x8_neon)61
+ ld1 {v16.s}0, x0, x162
+ ld1 {v17.s}0, x2, x363
+ usubl v1.8h, v16.8b, v17.8b64
+ ld1 {v16.s}0, x0, x165
+ ld1 {v17.s}0, x2, x366
+ smull v0.4s, v1.4h, v1.4h67
+.rept 668
+ usubl v1.8h, v16.8b, v17.8b69
+ ld1 {v16.s}0, x0, x170
+ smlal v0.4s, v1.4h, v1.4h71
+ ld1 {v17.s}0, x2, x372
+.endr73
+ usubl v1.8h, v16.8b, v17.8b74
+ smlal v0.4s, v1.4h, v1.4h75
+ ret_v0_w076
+endfunc77
+78
+function PFX(pixel_sse_pp_8x8_neon)79
+ ld1 {v16.8b}, x0, x180
+ ld1 {v17.8b}, x2, x381
+ usubl v1.8h, v16.8b, v17.8b82
+ ld1 {v16.8b}, x0, x183
+ smull v0.4s, v1.4h, v1.4h84
+ smlal2 v0.4s, v1.8h, v1.8h85
+ ld1 {v17.8b}, x2, x386
+87
+.rept 688
+ usubl v1.8h, v16.8b, v17.8b89
+ ld1 {v16.8b}, x0, x190
+ smlal v0.4s, v1.4h, v1.4h91
+ smlal2 v0.4s, v1.8h, v1.8h92
+ ld1 {v17.8b}, x2, x393
+.endr94
+ usubl v1.8h, v16.8b, v17.8b95
+ smlal v0.4s, v1.4h, v1.4h96
+ smlal2 v0.4s, v1.8h, v1.8h97
+ ret_v0_w098
+endfunc99
+100
+function PFX(pixel_sse_pp_8x16_neon)101
+ ld1 {v16.8b}, x0, x1102
+ ld1 {v17.8b}, x2, x3103
+ usubl v1.8h, v16.8b, v17.8b104
+ ld1 {v16.8b}, x0, x1105
+ smull v0.4s, v1.4h, v1.4h106
+ smlal2 v0.4s, v1.8h, v1.8h107
+ ld1 {v17.8b}, x2, x3108
+109
+.rept 14110
+ usubl v1.8h, v16.8b, v17.8b111
+ ld1 {v16.8b}, x0, x1112
+ smlal v0.4s, v1.4h, v1.4h113
+ smlal2 v0.4s, v1.8h, v1.8h114
+ ld1 {v17.8b}, x2, x3115
+.endr116
+ usubl v1.8h, v16.8b, v17.8b117
+ smlal v0.4s, v1.4h, v1.4h118
+ smlal2 v0.4s, v1.8h, v1.8h119
+ ret_v0_w0120
+endfunc121
+122
+.macro sse_pp_16xN h123
+function PFX(pixel_sse_pp_16x\h\()_neon)124
+ ld1 {v16.16b}, x0, x1125
+ ld1 {v17.16b}, x2, x3126
+ usubl v1.8h, v16.8b, v17.8b127
+ usubl2 v2.8h, v16.16b, v17.16b128
+ ld1 {v16.16b}, x0, x1129
+ ld1 {v17.16b}, x2, x3130
+ smull v0.4s, v1.4h, v1.4h131
+ smlal2 v0.4s, v1.8h, v1.8h132
+ smlal v0.4s, v2.4h, v2.4h133
+ smlal2 v0.4s, v2.8h, v2.8h134
+.rept \h - 2135
+ usubl v1.8h, v16.8b, v17.8b136
+ usubl2 v2.8h, v16.16b, v17.16b137
+ ld1 {v16.16b}, x0, x1138
+ smlal v0.4s, v1.4h, v1.4h139
+ smlal2 v0.4s, v1.8h, v1.8h140
+ ld1 {v17.16b}, x2, x3141
+ smlal v0.4s, v2.4h, v2.4h142
+ smlal2 v0.4s, v2.8h, v2.8h143
+.endr144
+ usubl v1.8h, v16.8b, v17.8b145
+ usubl2 v2.8h, v16.16b, v17.16b146
+ smlal v0.4s, v1.4h, v1.4h147
+ smlal2 v0.4s, v1.8h, v1.8h148
+ smlal v0.4s, v2.4h, v2.4h149
+ smlal2 v0.4s, v2.8h, v2.8h150
+ ret_v0_w0151
+endfunc152
+.endm153
+154
+sse_pp_16xN 16155
+sse_pp_16xN 32156
+157
+function PFX(pixel_sse_pp_32x32_neon)158
+ mov w12, #8159
+ movi v0.16b, #0160
+ movi v1.16b, #0161
+.loop_sse_pp_32:162
+ sub w12, w12, #1163
+.rept 4164
+ ld1 {v16.16b,v17.16b}, x0, x1165
+ ld1 {v18.16b,v19.16b}, x2, x3166
+ usubl v2.8h, v16.8b, v18.8b167
+ usubl2 v3.8h, v16.16b, v18.16b168
+ usubl v4.8h, v17.8b, v19.8b169
+ usubl2 v5.8h, v17.16b, v19.16b170
+ smlal v0.4s, v2.4h, v2.4h171
+ smlal2 v1.4s, v2.8h, v2.8h172
+ smlal v0.4s, v3.4h, v3.4h173
+ smlal2 v1.4s, v3.8h, v3.8h174
+ smlal v0.4s, v4.4h, v4.4h175
+ smlal2 v1.4s, v4.8h, v4.8h176
+ smlal v0.4s, v5.4h, v5.4h177
+ smlal2 v1.4s, v5.8h, v5.8h178
+.endr179
+ cbnz w12, .loop_sse_pp_32180
+ add v0.4s, v0.4s, v1.4s181
+ ret_v0_w0182
+endfunc183
+184
+function PFX(pixel_sse_pp_32x64_neon)185
+ mov w12, #16186
+ movi v0.16b, #0187
+ movi v1.16b, #0188
+.loop_sse_pp_32x64:189
+ sub w12, w12, #1190
+.rept 4191
+ ld1 {v16.16b,v17.16b}, x0, x1192
+ ld1 {v18.16b,v19.16b}, x2, x3193
+ usubl v2.8h, v16.8b, v18.8b194
+ usubl2 v3.8h, v16.16b, v18.16b195
+ usubl v4.8h, v17.8b, v19.8b196
+ usubl2 v5.8h, v17.16b, v19.16b197
+ smlal v0.4s, v2.4h, v2.4h198
+ smlal2 v1.4s, v2.8h, v2.8h199
+ smlal v0.4s, v3.4h, v3.4h200
+ smlal2 v1.4s, v3.8h, v3.8h201
+ smlal v0.4s, v4.4h, v4.4h202
+ smlal2 v1.4s, v4.8h, v4.8h203
+ smlal v0.4s, v5.4h, v5.4h204
+ smlal2 v1.4s, v5.8h, v5.8h205
+.endr206
+ cbnz w12, .loop_sse_pp_32x64207
+ add v0.4s, v0.4s, v1.4s208
+ ret_v0_w0209
+endfunc210
+211
+function PFX(pixel_sse_pp_64x64_neon)212
+ mov w12, #16213
+ movi v0.16b, #0214
+ movi v1.16b, #0215
+216
+.loop_sse_pp_64:217
+ sub w12, w12, #1218
+.rept 4219
+ ld1 {v16.16b-v19.16b}, x0, x1220
+ ld1 {v20.16b-v23.16b}, x2, x3221
+222
+ usubl v2.8h, v16.8b, v20.8b223
+ usubl2 v3.8h, v16.16b, v20.16b224
+ usubl v4.8h, v17.8b, v21.8b225
+ usubl2 v5.8h, v17.16b, v21.16b226
+ smlal v0.4s, v2.4h, v2.4h227
+ smlal2 v1.4s, v2.8h, v2.8h228
+ smlal v0.4s, v3.4h, v3.4h229
+ smlal2 v1.4s, v3.8h, v3.8h230
+ smlal v0.4s, v4.4h, v4.4h231
+ smlal2 v1.4s, v4.8h, v4.8h232
+ smlal v0.4s, v5.4h, v5.4h233
+ smlal2 v1.4s, v5.8h, v5.8h234
+235
+ usubl v2.8h, v18.8b, v22.8b236
+ usubl2 v3.8h, v18.16b, v22.16b237
+ usubl v4.8h, v19.8b, v23.8b238
+ usubl2 v5.8h, v19.16b, v23.16b239
+ smlal v0.4s, v2.4h, v2.4h240
+ smlal2 v1.4s, v2.8h, v2.8h241
+ smlal v0.4s, v3.4h, v3.4h242
+ smlal2 v1.4s, v3.8h, v3.8h243
+ smlal v0.4s, v4.4h, v4.4h244
+ smlal2 v1.4s, v4.8h, v4.8h245
+ smlal v0.4s, v5.4h, v5.4h246
+ smlal2 v1.4s, v5.8h, v5.8h247
+.endr248
+ cbnz w12, .loop_sse_pp_64249
+ add v0.4s, v0.4s, v1.4s250
+ ret_v0_w0251
+endfunc252
+253
+function PFX(pixel_sse_ss_4x4_neon)254
+ add x1, x1, x1255
+ add x3, x3, x3256
+ ld1 {v16.8b}, x0, x1257
+ ld1 {v17.8b}, x2, x3258
+ sub v2.4h, v16.4h, v17.4h259
+ ld1 {v16.8b}, x0, x1260
+ ld1 {v17.8b}, x2, x3261
+ smull v0.4s, v2.4h, v2.4h262
+ sub v2.4h, v16.4h, v17.4h263
+ ld1 {v16.8b}, x0, x1264
+ ld1 {v17.8b}, x2, x3265
+ smlal v0.4s, v2.4h, v2.4h266
+ sub v2.4h, v16.4h, v17.4h267
+ ld1 {v16.8b}, x0, x1268
+ smlal v0.4s, v2.4h, v2.4h269
+ ld1 {v17.8b}, x2, x3270
+ sub v2.4h, v16.4h, v17.4h271
+ smlal v0.4s, v2.4h, v2.4h272
+ ret_v0_w0273
+endfunc274
+275
+function PFX(pixel_sse_ss_8x8_neon)276
+ add x1, x1, x1277
+ add x3, x3, x3278
+ ld1 {v16.16b}, x0, x1279
+ ld1 {v17.16b}, x2, x3280
+ sub v2.8h, v16.8h, v17.8h281
+ ld1 {v16.16b}, x0, x1282
+ ld1 {v17.16b}, x2, x3283
+ smull v0.4s, v2.4h, v2.4h284
+ smull2 v1.4s, v2.8h, v2.8h285
+ sub v2.8h, v16.8h, v17.8h286
+.rept 6287
+ ld1 {v16.16b}, x0, x1288
+ ld1 {v17.16b}, x2, x3289
+ smlal v0.4s, v2.4h, v2.4h290
+ smlal2 v1.4s, v2.8h, v2.8h291
+ sub v2.8h, v16.8h, v17.8h292
+.endr293
+ smlal v0.4s, v2.4h, v2.4h294
+ smlal2 v1.4s, v2.8h, v2.8h295
+ add v0.4s, v0.4s, v1.4s296
+ ret_v0_w0297
+endfunc298
+299
+function PFX(pixel_sse_ss_16x16_neon)300
+ add x1, x1, x1301
+ add x3, x3, x3302
+ mov w12, #4303
+ movi v0.16b, #0304
+ movi v1.16b, #0305
+.loop_sse_ss_16:306
+ sub w12, w12, #1307
+.rept 4308
+ ld1 {v16.16b, v17.16b}, x0, x1309
+ ld1 {v18.16b, v19.16b}, x2, x3310
+ sub v2.8h, v16.8h, v18.8h311
+ sub v3.8h, v17.8h, v19.8h312
+ smlal v0.4s, v2.4h, v2.4h313
+ smlal2 v1.4s, v2.8h, v2.8h314
+ smlal v0.4s, v3.4h, v3.4h315
+ smlal2 v1.4s, v3.8h, v3.8h316
+.endr317
+ cbnz w12, .loop_sse_ss_16318
+ add v0.4s, v0.4s, v1.4s319
+ ret_v0_w0320
+endfunc321
+322
+function PFX(pixel_sse_ss_32x32_neon)323
+ add x1, x1, x1324
+ add x3, x3, x3325
+326
+ mov w12, #8327
+ movi v0.16b, #0328
+ movi v1.16b, #0329
+.loop_sse_ss_32:330
+ sub w12, w12, #1331
+.rept 4332
+ ld1 {v16.16b-v19.16b}, x0, x1333
+ ld1 {v20.16b-v23.16b}, x2, x3334
+ sub v2.8h, v16.8h, v20.8h335
+ sub v3.8h, v17.8h, v21.8h336
+ sub v4.8h, v18.8h, v22.8h337
+ sub v5.8h, v19.8h, v23.8h338
+ smlal v0.4s, v2.4h, v2.4h339
+ smlal2 v1.4s, v2.8h, v2.8h340
+ smlal v0.4s, v3.4h, v3.4h341
+ smlal2 v1.4s, v3.8h, v3.8h342
+ smlal v0.4s, v4.4h, v4.4h343
+ smlal2 v1.4s, v4.8h, v4.8h344
+ smlal v0.4s, v5.4h, v5.4h345
+ smlal2 v1.4s, v5.8h, v5.8h346
+.endr347
+ cbnz w12, .loop_sse_ss_32348
+ add v0.4s, v0.4s, v1.4s349
+ ret_v0_w0350
+endfunc351
+352
+function PFX(pixel_sse_ss_64x64_neon)353
+ add x1, x1, x1354
+ add x3, x3, x3355
+ sub x1, x1, #64356
+ sub x3, x3, #64357
+358
+ mov w12, #32359
+ movi v0.16b, #0360
+ movi v1.16b, #0361
+.loop_sse_ss_64:362
+ sub w12, w12, #1363
+.rept 2364
+ ld1 {v16.16b-v19.16b}, x0, #64365
+ ld1 {v20.16b-v23.16b}, x2, #64366
+ sub v2.8h, v16.8h, v20.8h367
+ sub v3.8h, v17.8h, v21.8h368
+ sub v4.8h, v18.8h, v22.8h369
+ sub v5.8h, v19.8h, v23.8h370
+ ld1 {v16.16b-v19.16b}, x0, x1371
+ ld1 {v20.16b-v23.16b}, x2, x3372
+ smlal v0.4s, v2.4h, v2.4h373
+ smlal2 v1.4s, v2.8h, v2.8h374
+ smlal v0.4s, v3.4h, v3.4h375
+ smlal2 v1.4s, v3.8h, v3.8h376
+ smlal v0.4s, v4.4h, v4.4h377
+ smlal2 v1.4s, v4.8h, v4.8h378
+ smlal v0.4s, v5.4h, v5.4h379
+ smlal2 v1.4s, v5.8h, v5.8h380
+ sub v2.8h, v16.8h, v20.8h381
+ sub v3.8h, v17.8h, v21.8h382
+ sub v4.8h, v18.8h, v22.8h383
+ sub v5.8h, v19.8h, v23.8h384
+ smlal v0.4s, v2.4h, v2.4h385
+ smlal2 v1.4s, v2.8h, v2.8h386
+ smlal v0.4s, v3.4h, v3.4h387
+ smlal2 v1.4s, v3.8h, v3.8h388
+ smlal v0.4s, v4.4h, v4.4h389
+ smlal2 v1.4s, v4.8h, v4.8h390
+ smlal v0.4s, v5.4h, v5.4h391
+ smlal2 v1.4s, v5.8h, v5.8h392
+.endr393
+ cbnz w12, .loop_sse_ss_64394
+ add v0.4s, v0.4s, v1.4s395
+ ret_v0_w0396
+endfunc397
+398
+function PFX(pixel_ssd_s_4x4_neon)399
+ add x1, x1, x1400
+ ld1 {v4.8b}, x0, x1401
+ ld1 {v5.8b}, x0, x1402
+ ld1 {v6.8b}, x0, x1403
+ ld1 {v7.8b}, x0404
+ smull v0.4s, v4.4h, v4.4h405
+ smull v1.4s, v5.4h, v5.4h406
+ smlal v0.4s, v6.4h, v6.4h407
+ smlal v1.4s, v7.4h, v7.4h408
+ add v0.4s, v0.4s, v1.4s409
+ ret_v0_w0410
+endfunc411
+412
+function PFX(pixel_ssd_s_8x8_neon)413
+ add x1, x1, x1414
+ ld1 {v4.16b}, x0, x1415
+ ld1 {v5.16b}, x0, x1416
+ smull v0.4s, v4.4h, v4.4h417
+ smull2 v1.4s, v4.8h, v4.8h418
+ smlal v0.4s, v5.4h, v5.4h419
+ smlal2 v1.4s, v5.8h, v5.8h420
+.rept 3421
+ ld1 {v4.16b}, x0, x1422
+ ld1 {v5.16b}, x0, x1423
+ smlal v0.4s, v4.4h, v4.4h424
+ smlal2 v1.4s, v4.8h, v4.8h425
+ smlal v0.4s, v5.4h, v5.4h426
+ smlal2 v1.4s, v5.8h, v5.8h427
+.endr428
+ add v0.4s, v0.4s, v1.4s429
+ ret_v0_w0430
+endfunc431
+432
+function PFX(pixel_ssd_s_16x16_neon)433
+ add x1, x1, x1434
+ mov w12, #4435
+ movi v0.16b, #0436
+ movi v1.16b, #0437
+.loop_ssd_s_16:438
+ sub w12, w12, #1439
+.rept 2440
+ ld1 {v4.16b,v5.16b}, x0, x1441
+ ld1 {v6.16b,v7.16b}, x0, x1442
+ smlal v0.4s, v4.4h, v4.4h443
+ smlal2 v1.4s, v4.8h, v4.8h444
+ smlal v0.4s, v5.4h, v5.4h445
+ smlal2 v1.4s, v5.8h, v5.8h446
+ smlal v0.4s, v6.4h, v6.4h447
+ smlal2 v1.4s, v6.8h, v6.8h448
+ smlal v0.4s, v7.4h, v7.4h449
+ smlal2 v1.4s, v7.8h, v7.8h450
+.endr451
+ cbnz w12, .loop_ssd_s_16452
+ add v0.4s, v0.4s, v1.4s453
+ ret_v0_w0454
+endfunc455
+456
+function PFX(pixel_ssd_s_32x32_neon)457
+ add x1, x1, x1458
+ mov w12, #8459
+ movi v0.16b, #0460
+ movi v1.16b, #0461
+.loop_ssd_s_32:462
+ sub w12, w12, #1463
+.rept 4464
+ ld1 {v4.16b-v7.16b}, x0, x1465
+ smlal v0.4s, v4.4h, v4.4h466
+ smlal2 v1.4s, v4.8h, v4.8h467
+ smlal v0.4s, v5.4h, v5.4h468
+ smlal2 v1.4s, v5.8h, v5.8h469
+ smlal v0.4s, v6.4h, v6.4h470
+ smlal2 v1.4s, v6.8h, v6.8h471
+ smlal v0.4s, v7.4h, v7.4h472
+ smlal2 v1.4s, v7.8h, v7.8h473
+.endr474
+ cbnz w12, .loop_ssd_s_32475
+ add v0.4s, v0.4s, v1.4s476
+ ret_v0_w0477
+endfunc478
x265_3.5.tar.gz/source/common/common.h -> x265_3.6.tar.gz/source/common/common.h
Changed
51
1
2
typedef uint64_t pixel4;3
typedef int64_t ssum2_t;4
#define SHIFT_TO_BITPLANE 95
-#define HISTOGRAM_BINS 10246
#else7
typedef uint8_t pixel;8
typedef uint16_t sum_t;9
10
typedef uint32_t pixel4;11
typedef int32_t ssum2_t; // Signed sum12
#define SHIFT_TO_BITPLANE 713
-#define HISTOGRAM_BINS 25614
#endif // if HIGH_BIT_DEPTH15
16
#if X265_DEPTH < 1017
18
19
#define MIN_QPSCALE 0.2124999999999999920
#define MAX_MAX_QPSCALE 615.4657423447710021
+#define FRAME_BRIGHTNESS_THRESHOLD 50.0 // Min % of pixels in a frame, that are above BRIGHTNESS_THRESHOLD for it to be considered a bright frame22
+#define FRAME_EDGE_THRESHOLD 10.0 // Min % of edge pixels in a frame, for it to be considered to have high edge density23
24
25
template<typename T>26
27
#define FILLER_OVERHEAD (NAL_TYPE_OVERHEAD + START_CODE_OVERHEAD + 1)28
29
#define MAX_NUM_DYN_REFINE (NUM_CU_DEPTH * X265_REFINE_INTER_LEVELS)30
+#define X265_BYTE 831
+32
+#define MAX_MCSTF_TEMPORAL_WINDOW_LENGTH 833
34
namespace X265_NS {35
36
37
#define x265_unlink(fileName) unlink(fileName)38
#define x265_rename(oldName, newName) rename(oldName, newName)39
#endif40
+/* Close a file */41
+#define x265_fclose(file) if (file != NULL) fclose(file); file=NULL;42
+#define x265_fread(val, size, readSize, fileOffset,errorMessage)\43
+ if (fread(val, size, readSize, fileOffset) != readSize)\44
+ {\45
+ x265_log(NULL, X265_LOG_ERROR, errorMessage); \46
+ return; \47
+ }48
int x265_exp2fix8(double x);49
50
double x265_ssim2dB(double ssim);51
x265_3.5.tar.gz/source/common/cpu.cpp -> x265_3.6.tar.gz/source/common/cpu.cpp
Changed
58
1
2
* Steve Borho <steve@borho.org>3
* Hongbin Liu <liuhongbin1@huawei.com>4
* Yimeng Su <yimeng.su@huawei.com>5
+ * Josh Dekker <josh@itanimul.li>6
+ * Jean-Baptiste Kempf <jb@videolan.org>7
*8
* This program is free software; you can redistribute it and/or modify9
* it under the terms of the GNU General Public License as published by10
11
{ "NEON", X265_CPU_NEON },12
{ "FastNeonMRC", X265_CPU_FAST_NEON_MRC },13
14
+#elif X265_ARCH_ARM6415
+ { "NEON", X265_CPU_NEON },16
+#if defined(HAVE_SVE)17
+ { "SVE", X265_CPU_SVE },18
+#endif19
+#if defined(HAVE_SVE2)20
+ { "SVE2", X265_CPU_SVE2 },21
+#endif22
#elif X265_ARCH_POWER823
{ "Altivec", X265_CPU_ALTIVEC },24
25
26
flags |= PFX(cpu_fast_neon_mrc_test)() ? X265_CPU_FAST_NEON_MRC : 0;27
#endif28
// TODO: write dual issue test? currently it's A8 (dual issue) vs. A9 (fast mrc)29
-#elif X265_ARCH_ARM6430
- flags |= X265_CPU_NEON;31
#endif // if HAVE_ARMV632
return flags;33
}34
35
+#elif X265_ARCH_ARM6436
+37
+uint32_t cpu_detect(bool benableavx512)38
+{39
+ int flags = 0;40
+41
+ #if defined(HAVE_SVE2)42
+ flags |= X265_CPU_SVE2;43
+ flags |= X265_CPU_SVE;44
+ flags |= X265_CPU_NEON;45
+ #elif defined(HAVE_SVE)46
+ flags |= X265_CPU_SVE;47
+ flags |= X265_CPU_NEON;48
+ #elif HAVE_NEON49
+ flags |= X265_CPU_NEON;50
+ #endif51
+ 52
+ return flags;53
+}54
+55
#elif X265_ARCH_POWER856
57
uint32_t cpu_detect(bool benableavx512)58
x265_3.5.tar.gz/source/common/frame.cpp -> x265_3.6.tar.gz/source/common/frame.cpp
Changed
102
1
2
m_edgeBitPlane = NULL;3
m_edgeBitPic = NULL;4
m_isInsideWindow = 0;5
+6
+ // mcstf7
+ m_isSubSampled = NULL;8
+ m_mcstf = NULL;9
+ m_refPicCnt0 = 0;10
+ m_refPicCnt1 = 0;11
+ m_nextMCSTF = NULL;12
+ m_prevMCSTF = NULL;13
+14
+ m_tempLayer = 0;15
+ m_sameLayerRefPic = false;16
}17
18
bool Frame::create(x265_param *param, float* quantOffsets)19
{20
m_fencPic = new PicYuv;21
m_param = param;22
+23
+ if (m_param->bEnableTemporalFilter)24
+ {25
+ m_mcstf = new TemporalFilter;26
+ m_mcstf->init(param);27
+28
+ m_fencPicSubsampled2 = new PicYuv;29
+ m_fencPicSubsampled4 = new PicYuv;30
+31
+ if (!m_fencPicSubsampled2->createScaledPicYUV(param, 2))32
+ return false;33
+ if (!m_fencPicSubsampled4->createScaledPicYUV(param, 4))34
+ return false;35
+36
+ CHECKED_MALLOC_ZERO(m_isSubSampled, int, 1);37
+ }38
+39
CHECKED_MALLOC_ZERO(m_rcData, RcStats, 1);40
41
if (param->bCTUInfo)42
43
return false;44
}45
46
+bool Frame::createSubSample()47
+{48
+49
+ m_fencPicSubsampled2 = new PicYuv;50
+ m_fencPicSubsampled4 = new PicYuv;51
+52
+ if (!m_fencPicSubsampled2->createScaledPicYUV(m_param, 2))53
+ return false;54
+ if (!m_fencPicSubsampled4->createScaledPicYUV(m_param, 4))55
+ return false;56
+ CHECKED_MALLOC_ZERO(m_isSubSampled, int, 1);57
+ return true;58
+fail:59
+ return false;60
+}61
+62
bool Frame::allocEncodeData(x265_param *param, const SPS& sps)63
{64
m_encData = new FrameData;65
66
m_fencPic = NULL;67
}68
69
+ if (m_param->bEnableTemporalFilter)70
+ {71
+72
+ if (m_fencPicSubsampled2)73
+ {74
+ m_fencPicSubsampled2->destroy();75
+ delete m_fencPicSubsampled2;76
+ m_fencPicSubsampled2 = NULL;77
+ }78
+79
+ if (m_fencPicSubsampled4)80
+ {81
+ m_fencPicSubsampled4->destroy();82
+ delete m_fencPicSubsampled4;83
+ m_fencPicSubsampled4 = NULL;84
+ }85
+ delete m_mcstf;86
+ X265_FREE(m_isSubSampled);87
+ }88
+89
if (m_reconPic)90
{91
m_reconPic->destroy();92
93
X265_FREE(m_addOnPrevChange);94
m_addOnPrevChange = NULL;95
}96
- m_lowres.destroy();97
+98
+ m_lowres.destroy(m_param);99
X265_FREE(m_rcData);100
101
if (m_param->bDynamicRefine)102
x265_3.5.tar.gz/source/common/frame.h -> x265_3.6.tar.gz/source/common/frame.h
Changed
60
1
2
#include "common.h"3
#include "lowres.h"4
#include "threading.h"5
+#include "temporalfilter.h"6
7
namespace X265_NS {8
// private namespace9
10
double count4;11
double offset4;12
double bufferFillFinal;13
+ int64_t currentSatd;14
};15
16
class Frame17
18
19
/* Data associated with x265_picture */20
PicYuv* m_fencPic;21
+ PicYuv* m_fencPicSubsampled2;22
+ PicYuv* m_fencPicSubsampled4;23
+24
int m_poc;25
int m_encodeOrder;26
+ int m_gopOffset;27
int64_t m_pts; // user provided presentation time stamp28
int64_t m_reorderedPts;29
int64_t m_dts;30
31
bool m_classifyFrame;32
int m_fieldNum;33
34
+ /*MCSTF*/35
+ TemporalFilter* m_mcstf;36
+ int m_refPicCnt2;37
+ Frame* m_nextMCSTF; // PicList doubly linked list pointers38
+ Frame* m_prevMCSTF;39
+ int* m_isSubSampled;40
+41
/* aq-mode 4 : Gaussian, edge and theta frames for edge information */42
pixel* m_edgePic;43
pixel* m_gaussianPic;44
45
46
int m_isInsideWindow;47
48
+ /*Frame's temporal layer info*/49
+ uint8_t m_tempLayer;50
+ int8_t m_gopId;51
+ bool m_sameLayerRefPic;52
+53
Frame();54
55
bool create(x265_param *param, float* quantOffsets);56
+ bool createSubSample();57
bool allocEncodeData(x265_param *param, const SPS& sps);58
void reinit(const SPS& sps);59
void destroy();60
x265_3.5.tar.gz/source/common/framedata.cpp -> x265_3.6.tar.gz/source/common/framedata.cpp
Changed
10
1
2
}3
else4
return false;5
- CHECKED_MALLOC_ZERO(m_cuStat, RCStatCU, sps.numCUsInFrame);6
+ CHECKED_MALLOC_ZERO(m_cuStat, RCStatCU, sps.numCUsInFrame + 1);7
CHECKED_MALLOC(m_rowStat, RCStatRow, sps.numCuInHeight);8
reinit(sps);9
10
x265_3.5.tar.gz/source/common/lowres.cpp -> x265_3.6.tar.gz/source/common/lowres.cpp
Changed
154
1
2
3
using namespace X265_NS;4
5
+/*6
+ * Down Sample input picture7
+ */8
+static9
+void frame_lowres_core(const pixel* src0, pixel* dst0,10
+ intptr_t src_stride, intptr_t dst_stride, int width, int height)11
+{12
+ for (int y = 0; y < height; y++)13
+ {14
+ const pixel* src1 = src0 + src_stride;15
+ for (int x = 0; x < width; x++)16
+ {17
+ // slower than naive bilinear, but matches asm18
+#define FILTER(a, b, c, d) ((((a + b + 1) >> 1) + ((c + d + 1) >> 1) + 1) >> 1)19
+ dst0x = FILTER(src02 * x, src12 * x, src02 * x + 1, src12 * x + 1);20
+#undef FILTER21
+ }22
+ src0 += src_stride * 2;23
+ dst0 += dst_stride;24
+ }25
+}26
+27
bool PicQPAdaptationLayer::create(uint32_t width, uint32_t height, uint32_t partWidth, uint32_t partHeight, uint32_t numAQPartInWidthExt, uint32_t numAQPartInHeightExt)28
{29
aqPartWidth = partWidth;30
31
32
size_t planesize = lumaStride * (lines + 2 * origPic->m_lumaMarginY);33
size_t padoffset = lumaStride * origPic->m_lumaMarginY + origPic->m_lumaMarginX;34
- if (!!param->rc.aqMode || !!param->rc.hevcAq || !!param->bAQMotion)35
+ if (!!param->rc.aqMode || !!param->rc.hevcAq || !!param->bAQMotion || !!param->bEnableWeightedPred || !!param->bEnableWeightedBiPred)36
{37
CHECKED_MALLOC_ZERO(qpAqOffset, double, cuCountFullRes);38
CHECKED_MALLOC_ZERO(invQscaleFactor, int, cuCountFullRes);39
40
}41
}42
43
+ if (param->bHistBasedSceneCut)44
+ {45
+ quarterSampleLowResWidth = widthFullRes / 4;46
+ quarterSampleLowResHeight = heightFullRes / 4;47
+ quarterSampleLowResOriginX = 16;48
+ quarterSampleLowResOriginY = 16;49
+ quarterSampleLowResStrideY = quarterSampleLowResWidth + 2 * quarterSampleLowResOriginY;50
+51
+ size_t quarterSampleLowResPlanesize = quarterSampleLowResStrideY * (quarterSampleLowResHeight + 2 * quarterSampleLowResOriginX);52
+ /* allocate quarter sampled lowres buffers */53
+ CHECKED_MALLOC_ZERO(quarterSampleLowResBuffer, pixel, quarterSampleLowResPlanesize);54
+55
+ // Allocate memory for Histograms56
+ picHistogram = X265_MALLOC(uint32_t***, NUMBER_OF_SEGMENTS_IN_WIDTH * sizeof(uint32_t***));57
+ picHistogram0 = X265_MALLOC(uint32_t**, NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);58
+ for (uint32_t wd = 1; wd < NUMBER_OF_SEGMENTS_IN_WIDTH; wd++) {59
+ picHistogramwd = picHistogram0 + wd * NUMBER_OF_SEGMENTS_IN_HEIGHT;60
+ }61
+62
+ for (uint32_t regionInPictureWidthIndex = 0; regionInPictureWidthIndex < NUMBER_OF_SEGMENTS_IN_WIDTH; regionInPictureWidthIndex++)63
+ {64
+ for (uint32_t regionInPictureHeightIndex = 0; regionInPictureHeightIndex < NUMBER_OF_SEGMENTS_IN_HEIGHT; regionInPictureHeightIndex++)65
+ {66
+ picHistogramregionInPictureWidthIndexregionInPictureHeightIndex = X265_MALLOC(uint32_t*, NUMBER_OF_SEGMENTS_IN_WIDTH *sizeof(uint32_t*));67
+ picHistogramregionInPictureWidthIndexregionInPictureHeightIndex0 = X265_MALLOC(uint32_t, 3 * HISTOGRAM_NUMBER_OF_BINS * sizeof(uint32_t));68
+ for (uint32_t wd = 1; wd < 3; wd++) {69
+ picHistogramregionInPictureWidthIndexregionInPictureHeightIndexwd = picHistogramregionInPictureWidthIndexregionInPictureHeightIndex0 + wd * HISTOGRAM_NUMBER_OF_BINS;70
+ }71
+ }72
+ }73
+ }74
+75
return true;76
77
fail:78
return false;79
}80
81
-void Lowres::destroy()82
+void Lowres::destroy(x265_param* param)83
{84
X265_FREE(buffer0);85
if(bEnableHME)86
87
X265_FREE(invQscaleFactor8x8);88
X265_FREE(edgeInclined);89
X265_FREE(qpAqMotionOffset);90
- X265_FREE(blockVariance);91
+ if (param->bDynamicRefine || param->bEnableFades)92
+ X265_FREE(blockVariance);93
if (maxAQDepth > 0)94
{95
for (uint32_t d = 0; d < 4; d++)96
97
98
delete pAQLayer;99
}100
+101
+ // Histograms102
+ if (param->bHistBasedSceneCut)103
+ {104
+ for (uint32_t segmentInFrameWidthIdx = 0; segmentInFrameWidthIdx < NUMBER_OF_SEGMENTS_IN_WIDTH; segmentInFrameWidthIdx++)105
+ {106
+ if (picHistogramsegmentInFrameWidthIdx)107
+ {108
+ for (uint32_t segmentInFrameHeightIdx = 0; segmentInFrameHeightIdx < NUMBER_OF_SEGMENTS_IN_HEIGHT; segmentInFrameHeightIdx++)109
+ {110
+ if (picHistogramsegmentInFrameWidthIdxsegmentInFrameHeightIdx)111
+ X265_FREE(picHistogramsegmentInFrameWidthIdxsegmentInFrameHeightIdx0);112
+ X265_FREE(picHistogramsegmentInFrameWidthIdxsegmentInFrameHeightIdx);113
+ }114
+ }115
+ }116
+ if (picHistogram)117
+ X265_FREE(picHistogram0);118
+ X265_FREE(picHistogram);119
+120
+ X265_FREE(quarterSampleLowResBuffer);121
+122
+ }123
}124
// (re) initialize lowres state125
void Lowres::init(PicYuv *origPic, int poc)126
127
indB = 0;128
memset(costEst, -1, sizeof(costEst));129
memset(weightedCostDelta, 0, sizeof(weightedCostDelta));130
- interPCostPercDiff = 0.0;131
- intraCostPercDiff = 0.0;132
- m_bIsMaxThres = false;133
- m_bIsHardScenecut = false;134
135
if (qpAqOffset && invQscaleFactor)136
memset(costEstAq, -1, sizeof(costEstAq));137
138
}139
140
fpelPlane0 = lowresPlane0;141
+142
+ if (origPic->m_param->bHistBasedSceneCut)143
+ {144
+ // Quarter Sampled Input Picture Formation145
+ // TO DO: Replace with ASM function146
+ frame_lowres_core(147
+ lowresPlane0,148
+ quarterSampleLowResBuffer + quarterSampleLowResOriginX + quarterSampleLowResOriginY * quarterSampleLowResStrideY,149
+ lumaStride,150
+ quarterSampleLowResStrideY,151
+ widthFullRes / 4, heightFullRes / 4);152
+ }153
}154
x265_3.5.tar.gz/source/common/lowres.h -> x265_3.6.tar.gz/source/common/lowres.h
Changed
73
1
2
namespace X265_NS {3
// private namespace4
5
+#define HISTOGRAM_NUMBER_OF_BINS 2566
+#define NUMBER_OF_SEGMENTS_IN_WIDTH 47
+#define NUMBER_OF_SEGMENTS_IN_HEIGHT 48
+9
struct ReferencePlanes10
{11
ReferencePlanes() { memset(this, 0, sizeof(ReferencePlanes)); }12
13
14
int frameNum; // Presentation frame number15
int sliceType; // Slice type decided by lookahead16
+ int sliceTypeReq; // Slice type required as per the QP file17
int width; // width of lowres frame in pixels18
int lines; // height of lowres frame in pixel lines19
int leadingBframes; // number of leading B frames for P or I20
21
double* qpAqOffset; // AQ QP offset values for each 16x16 CU22
double* qpCuTreeOffset; // cuTree QP offset values for each 16x16 CU23
double* qpAqMotionOffset;24
- int* invQscaleFactor; // qScale values for qp Aq Offsets25
+ int* invQscaleFactor; // qScale values for qp Aq Offsets26
int* invQscaleFactor8x8; // temporary buffer for qg-size 827
uint32_t* blockVariance;28
uint64_t wp_ssd3; // This is different than SSDY, this is sum(pixel^2) - sum(pixel)^2 for entire frame29
uint64_t wp_sum3;30
double frameVariance;31
- int* edgeInclined;32
+ int* edgeInclined;33
34
35
/* cutree intermediate data */36
37
uint32_t heightFullRes;38
uint32_t m_maxCUSize;39
uint32_t m_qgSize;40
- 41
+42
uint16_t* propagateCost;43
double weightedCostDeltaX265_BFRAME_MAX + 2;44
ReferencePlanes weightedRefX265_BFRAME_MAX + 2;45
+46
/* For hist-based scenecut */47
- bool m_bIsMaxThres;48
- double interPCostPercDiff;49
- double intraCostPercDiff;50
- bool m_bIsHardScenecut;51
+ int quarterSampleLowResWidth; // width of 1/4 lowres frame in pixels52
+ int quarterSampleLowResHeight; // height of 1/4 lowres frame in pixels53
+ int quarterSampleLowResStrideY;54
+ int quarterSampleLowResOriginX;55
+ int quarterSampleLowResOriginY;56
+ pixel *quarterSampleLowResBuffer;57
+ bool bHistScenecutAnalyzed;58
+59
+ uint16_t picAvgVariance;60
+ uint16_t picAvgVarianceCb;61
+ uint16_t picAvgVarianceCr;62
+63
+ uint32_t ****picHistogram;64
+ uint64_t averageIntensityPerSegmentNUMBER_OF_SEGMENTS_IN_WIDTHNUMBER_OF_SEGMENTS_IN_HEIGHT3;65
+ uint8_t averageIntensity3;66
67
bool create(x265_param* param, PicYuv *origPic, uint32_t qgSize);68
- void destroy();69
+ void destroy(x265_param* param);70
void init(PicYuv *origPic, int poc);71
};72
}73
x265_3.5.tar.gz/source/common/mv.h -> x265_3.6.tar.gz/source/common/mv.h
Changed
10
1
2
{3
return x >= _min.x && x <= _max.x && y >= _min.y && y <= _max.y;4
}5
+6
+ void set(int32_t _x, int32_t _y) { x = _x; y = _y; }7
};8
}9
10
x265_3.5.tar.gz/source/common/param.cpp -> x265_3.6.tar.gz/source/common/param.cpp
Changed
668
1
2
param->bAnnexB = 1;3
param->bRepeatHeaders = 0;4
param->bEnableAccessUnitDelimiters = 0;5
+ param->bEnableEndOfBitstream = 0;6
+ param->bEnableEndOfSequence = 0;7
param->bEmitHRDSEI = 0;8
param->bEmitInfoSEI = 1;9
param->bEmitHDRSEI = 0; /*Deprecated*/10
11
param->keyframeMax = 250;12
param->gopLookahead = 0;13
param->bOpenGOP = 1;14
+ param->craNal = 0;15
param->bframes = 4;16
param->lookaheadDepth = 20;17
param->bFrameAdaptive = X265_B_ADAPT_TRELLIS;18
param->bBPyramid = 1;19
param->scenecutThreshold = 40; /* Magic number pulled in from x264 */20
- param->edgeTransitionThreshold = 0.03;21
param->bHistBasedSceneCut = 0;22
param->lookaheadSlices = 8;23
param->lookaheadThreads = 0;24
25
param->bEnableHRDConcatFlag = 0;26
param->bEnableFades = 0;27
param->bEnableSceneCutAwareQp = 0;28
- param->fwdScenecutWindow = 500;29
- param->fwdRefQpDelta = 5;30
- param->fwdNonRefQpDelta = param->fwdRefQpDelta + (SLICE_TYPE_DELTA * param->fwdRefQpDelta);31
- param->bwdScenecutWindow = 100;32
- param->bwdRefQpDelta = -1;33
- param->bwdNonRefQpDelta = -1;34
+ param->fwdMaxScenecutWindow = 1200;35
+ param->bwdMaxScenecutWindow = 600;36
+ for (int i = 0; i < 6; i++)37
+ {38
+ int deltas6 = { 5, 4, 3, 2, 1, 0 };39
+40
+ param->fwdScenecutWindowi = 200;41
+ param->fwdRefQpDeltai = deltasi;42
+ param->fwdNonRefQpDeltai = param->fwdRefQpDeltai + (SLICE_TYPE_DELTA * param->fwdRefQpDeltai);43
+44
+ param->bwdScenecutWindowi = 100;45
+ param->bwdRefQpDeltai = -1;46
+ param->bwdNonRefQpDeltai = -1;47
+ }48
49
/* Intra Coding Tools */50
param->bEnableConstrainedIntra = 0;51
52
param->rc.rfConstantMin = 0;53
param->rc.bStatRead = 0;54
param->rc.bStatWrite = 0;55
+ param->rc.dataShareMode = X265_SHARE_MODE_FILE;56
param->rc.statFileName = NULL;57
+ param->rc.sharedMemName = NULL;58
+ param->rc.bEncFocusedFramesOnly = 0;59
param->rc.complexityBlur = 20;60
param->rc.qblur = 0.5;61
param->rc.zoneCount = 0;62
63
param->maxLuma = PIXEL_MAX;64
param->log2MaxPocLsb = 8;65
param->maxSlices = 1;66
+ param->videoSignalTypePreset = NULL;67
68
/*Conformance window*/69
param->confWinRightOffset = 0;70
71
param->bEnableSvtHevc = 0;72
param->svtHevcParam = NULL;73
74
+ /* MCSTF */75
+ param->bEnableTemporalFilter = 0;76
+ param->temporalFilterStrength = 0.95;77
+78
#ifdef SVT_HEVC79
param->svtHevcParam = svtParam;80
svt_param_default(param);81
#endif82
+ /* Film grain characteristics model filename */83
+ param->filmGrain = NULL;84
+ param->bEnableSBRC = 0;85
}86
87
int x265_param_default_preset(x265_param* param, const char* preset, const char* tune)88
89
#define atof(str) x265_atof(str, bError)90
#define atobool(str) (x265_atobool(str, bError))91
92
+int x265_scenecut_aware_qp_param_parse(x265_param* p, const char* name, const char* value)93
+{94
+ bool bError = false;95
+ char nameBuf64;96
+ if (!name)97
+ return X265_PARAM_BAD_NAME;98
+ // skip -- prefix if provided99
+ if (name0 == '-' && name1 == '-')100
+ name += 2;101
+ // s/_/-/g102
+ if (strlen(name) + 1 < sizeof(nameBuf) && strchr(name, '_'))103
+ {104
+ char *c;105
+ strcpy(nameBuf, name);106
+ while ((c = strchr(nameBuf, '_')) != 0)107
+ *c = '-';108
+ name = nameBuf;109
+ }110
+ if (!value)111
+ value = "true";112
+ else if (value0 == '=')113
+ value++;114
+#define OPT(STR) else if (!strcmp(name, STR))115
+ if (0);116
+ OPT("scenecut-aware-qp") p->bEnableSceneCutAwareQp = x265_atoi(value, bError);117
+ OPT("masking-strength") bError = parseMaskingStrength(p, value);118
+ else119
+ return X265_PARAM_BAD_NAME;120
+#undef OPT121
+ return bError ? X265_PARAM_BAD_VALUE : 0;122
+}123
+124
+125
+/* internal versions of string-to-int with additional error checking */126
+#undef atoi127
+#undef atof128
+#define atoi(str) x265_atoi(str, bError)129
+#define atof(str) x265_atof(str, bError)130
+#define atobool(str) (x265_atobool(str, bError))131
+132
int x265_zone_param_parse(x265_param* p, const char* name, const char* value)133
{134
bool bError = false;135
136
{137
bError = false;138
p->scenecutThreshold = atoi(value);139
- p->bHistBasedSceneCut = 0;140
}141
}142
- OPT("temporal-layers") p->bEnableTemporalSubLayers = atobool(value);143
+ OPT("temporal-layers") p->bEnableTemporalSubLayers = atoi(value);144
OPT("keyint") p->keyframeMax = atoi(value);145
OPT("min-keyint") p->keyframeMin = atoi(value);146
OPT("rc-lookahead") p->lookaheadDepth = atoi(value);147
148
int pass = x265_clip3(0, 3, atoi(value));149
p->rc.bStatWrite = pass & 1;150
p->rc.bStatRead = pass & 2;151
+ p->rc.dataShareMode = X265_SHARE_MODE_FILE;152
}153
OPT("stats") p->rc.statFileName = strdup(value);154
OPT("scaling-list") p->scalingLists = strdup(value);155
156
OPT("opt-ref-list-length-pps") p->bOptRefListLengthPPS = atobool(value);157
OPT("multi-pass-opt-rps") p->bMultiPassOptRPS = atobool(value);158
OPT("scenecut-bias") p->scenecutBias = atof(value);159
- OPT("hist-scenecut")160
- {161
- p->bHistBasedSceneCut = atobool(value);162
- if (bError)163
- {164
- bError = false;165
- p->bHistBasedSceneCut = 0;166
- }167
- if (p->bHistBasedSceneCut)168
- {169
- bError = false;170
- p->scenecutThreshold = 0;171
- }172
- }173
- OPT("hist-threshold") p->edgeTransitionThreshold = atof(value);174
+ OPT("hist-scenecut") p->bHistBasedSceneCut = atobool(value);175
OPT("rskip-edge-threshold") p->edgeVarThreshold = atoi(value)/100.0f;176
OPT("lookahead-threads") p->lookaheadThreads = atoi(value);177
OPT("opt-cu-delta-qp") p->bOptCUDeltaQP = atobool(value);178
179
OPT("multi-pass-opt-distortion") p->analysisMultiPassDistortion = atobool(value);180
OPT("aq-motion") p->bAQMotion = atobool(value);181
OPT("dynamic-rd") p->dynamicRd = atof(value);182
+ OPT("cra-nal") p->craNal = atobool(value);183
OPT("analysis-reuse-level")184
{185
p->analysisReuseLevel = atoi(value);186
187
}188
OPT("fades") p->bEnableFades = atobool(value);189
OPT("scenecut-aware-qp") p->bEnableSceneCutAwareQp = atoi(value);190
- OPT("masking-strength")191
- {192
- int window1;193
- double refQpDelta1, nonRefQpDelta1;194
-195
- if (p->bEnableSceneCutAwareQp == FORWARD)196
- {197
- if (3 == sscanf(value, "%d,%lf,%lf", &window1, &refQpDelta1, &nonRefQpDelta1))198
- {199
- if (window1 > 0)200
- p->fwdScenecutWindow = window1;201
- if (refQpDelta1 > 0)202
- p->fwdRefQpDelta = refQpDelta1;203
- if (nonRefQpDelta1 > 0)204
- p->fwdNonRefQpDelta = nonRefQpDelta1;205
- }206
- else207
- {208
- x265_log(NULL, X265_LOG_ERROR, "Specify all the necessary offsets for masking-strength \n");209
- bError = true;210
- }211
- }212
- else if (p->bEnableSceneCutAwareQp == BACKWARD)213
- {214
- if (3 == sscanf(value, "%d,%lf,%lf", &window1, &refQpDelta1, &nonRefQpDelta1))215
- {216
- if (window1 > 0)217
- p->bwdScenecutWindow = window1;218
- if (refQpDelta1 > 0)219
- p->bwdRefQpDelta = refQpDelta1;220
- if (nonRefQpDelta1 > 0)221
- p->bwdNonRefQpDelta = nonRefQpDelta1;222
- }223
- else224
- {225
- x265_log(NULL, X265_LOG_ERROR, "Specify all the necessary offsets for masking-strength \n");226
- bError = true;227
- }228
- }229
- else if (p->bEnableSceneCutAwareQp == BI_DIRECTIONAL)230
- {231
- int window2;232
- double refQpDelta2, nonRefQpDelta2;233
- if (6 == sscanf(value, "%d,%lf,%lf,%d,%lf,%lf", &window1, &refQpDelta1, &nonRefQpDelta1, &window2, &refQpDelta2, &nonRefQpDelta2))234
- {235
- if (window1 > 0)236
- p->fwdScenecutWindow = window1;237
- if (refQpDelta1 > 0)238
- p->fwdRefQpDelta = refQpDelta1;239
- if (nonRefQpDelta1 > 0)240
- p->fwdNonRefQpDelta = nonRefQpDelta1;241
- if (window2 > 0)242
- p->bwdScenecutWindow = window2;243
- if (refQpDelta2 > 0)244
- p->bwdRefQpDelta = refQpDelta2;245
- if (nonRefQpDelta2 > 0)246
- p->bwdNonRefQpDelta = nonRefQpDelta2;247
- }248
- else249
- {250
- x265_log(NULL, X265_LOG_ERROR, "Specify all the necessary offsets for masking-strength \n");251
- bError = true;252
- }253
- }254
- }255
+ OPT("masking-strength") bError |= parseMaskingStrength(p, value);256
OPT("field") p->bField = atobool( value );257
OPT("cll") p->bEmitCLL = atobool(value);258
OPT("frame-dup") p->bEnableFrameDuplication = atobool(value);259
260
OPT("vbv-live-multi-pass") p->bliveVBV2pass = atobool(value);261
OPT("min-vbv-fullness") p->minVbvFullness = atof(value);262
OPT("max-vbv-fullness") p->maxVbvFullness = atof(value);263
+ OPT("video-signal-type-preset") p->videoSignalTypePreset = strdup(value);264
+ OPT("eob") p->bEnableEndOfBitstream = atobool(value);265
+ OPT("eos") p->bEnableEndOfSequence = atobool(value);266
+ /* Film grain characterstics model filename */267
+ OPT("film-grain") p->filmGrain = (char* )value;268
+ OPT("mcstf") p->bEnableTemporalFilter = atobool(value);269
+ OPT("sbrc") p->bEnableSBRC = atobool(value);270
else271
return X265_PARAM_BAD_NAME;272
}273
274
"scenecutThreshold must be greater than 0");275
CHECK(param->scenecutBias < 0 || 100 < param->scenecutBias,276
"scenecut-bias must be between 0 and 100");277
- CHECK(param->edgeTransitionThreshold < 0.0 || 1.0 < param->edgeTransitionThreshold,278
- "hist-threshold must be between 0.0 and 1.0");279
CHECK(param->radl < 0 || param->radl > param->bframes,280
"radl must be between 0 and bframes");281
CHECK(param->rdPenalty < 0 || param->rdPenalty > 2,282
283
"Invalid refine-ctu-distortion value, must be either 0 or 1");284
CHECK(param->maxAUSizeFactor < 0.5 || param->maxAUSizeFactor > 1.0,285
"Supported factor for controlling max AU size is from 0.5 to 1");286
- CHECK((param->dolbyProfile != 0) && (param->dolbyProfile != 50) && (param->dolbyProfile != 81) && (param->dolbyProfile != 82),287
- "Unsupported Dolby Vision profile, only profile 5, profile 8.1 and profile 8.2 enabled");288
+ CHECK((param->dolbyProfile != 0) && (param->dolbyProfile != 50) && (param->dolbyProfile != 81) && (param->dolbyProfile != 82) && (param->dolbyProfile != 84),289
+ "Unsupported Dolby Vision profile, only profile 5, profile 8.1, profile 8.2 and profile 8.4 enabled");290
CHECK(param->dupThreshold < 1 || 99 < param->dupThreshold,291
"Invalid frame-duplication threshold. Value must be between 1 and 99.");292
if (param->dolbyProfile)293
{294
CHECK((param->rc.vbvMaxBitrate <= 0 || param->rc.vbvBufferSize <= 0), "Dolby Vision requires VBV settings to enable HRD.\n");295
- CHECK((param->internalBitDepth != 10), "Dolby Vision profile - 5, profile - 8.1 and profile - 8.2 is Main10 only\n");296
- CHECK((param->internalCsp != X265_CSP_I420), "Dolby Vision profile - 5, profile - 8.1 and profile - 8.2 requires YCbCr 4:2:0 color space\n");297
+ CHECK((param->internalBitDepth != 10), "Dolby Vision profile - 5, profile - 8.1, profile - 8.2 and profile - 8.4 are Main10 only\n");298
+ CHECK((param->internalCsp != X265_CSP_I420), "Dolby Vision profile - 5, profile - 8.1, profile - 8.2 and profile - 8.4 requires YCbCr 4:2:0 color space\n");299
if (param->dolbyProfile == 81)300
CHECK(!(param->masteringDisplayColorVolume), "Dolby Vision profile - 8.1 requires Mastering display color volume information\n");301
}302
303
{304
CHECK(param->bEnableSceneCutAwareQp < 0 || param->bEnableSceneCutAwareQp > 3,305
"Invalid masking direction. Value must be between 0 and 3(inclusive)");306
- CHECK(param->fwdScenecutWindow < 0 || param->fwdScenecutWindow > 1000,307
- "Invalid forward scenecut Window duration. Value must be between 0 and 1000(inclusive)");308
- CHECK(param->fwdRefQpDelta < 0 || param->fwdRefQpDelta > 10,309
- "Invalid fwdRefQpDelta value. Value must be between 0 and 10 (inclusive)");310
- CHECK(param->fwdNonRefQpDelta < 0 || param->fwdNonRefQpDelta > 10,311
- "Invalid fwdNonRefQpDelta value. Value must be between 0 and 10 (inclusive)");312
-313
- CHECK(param->bwdScenecutWindow < 0 || param->bwdScenecutWindow > 1000,314
- "Invalid backward scenecut Window duration. Value must be between 0 and 1000(inclusive)");315
- CHECK(param->bwdRefQpDelta < -1 || param->bwdRefQpDelta > 10,316
- "Invalid bwdRefQpDelta value. Value must be between 0 and 10 (inclusive)");317
- CHECK(param->bwdNonRefQpDelta < -1 || param->bwdNonRefQpDelta > 10,318
- "Invalid bwdNonRefQpDelta value. Value must be between 0 and 10 (inclusive)");319
+ for (int i = 0; i < 6; i++)320
+ {321
+ CHECK(param->fwdScenecutWindowi < 0 || param->fwdScenecutWindowi > 1000,322
+ "Invalid forward scenecut Window duration. Value must be between 0 and 1000(inclusive)");323
+ CHECK(param->fwdRefQpDeltai < 0 || param->fwdRefQpDeltai > 20,324
+ "Invalid fwdRefQpDelta value. Value must be between 0 and 20 (inclusive)");325
+ CHECK(param->fwdNonRefQpDeltai < 0 || param->fwdNonRefQpDeltai > 20,326
+ "Invalid fwdNonRefQpDelta value. Value must be between 0 and 20 (inclusive)");327
+328
+ CHECK(param->bwdScenecutWindowi < 0 || param->bwdScenecutWindowi > 1000,329
+ "Invalid backward scenecut Window duration. Value must be between 0 and 1000(inclusive)");330
+ CHECK(param->bwdRefQpDeltai < -1 || param->bwdRefQpDeltai > 20,331
+ "Invalid bwdRefQpDelta value. Value must be between 0 and 20 (inclusive)");332
+ CHECK(param->bwdNonRefQpDeltai < -1 || param->bwdNonRefQpDeltai > 20,333
+ "Invalid bwdNonRefQpDelta value. Value must be between 0 and 20 (inclusive)");334
+ }335
}336
}337
if (param->bEnableHME)338
339
param->bSingleSeiNal = 0;340
x265_log(param, X265_LOG_WARNING, "None of the SEI messages are enabled. Disabling Single SEI NAL\n");341
}342
+ if (param->bEnableTemporalFilter && (param->frameNumThreads > 1))343
+ {344
+ param->bEnableTemporalFilter = 0;345
+ x265_log(param, X265_LOG_WARNING, "MCSTF can be enabled with frame thread = 1 only. Disabling MCSTF\n");346
+ }347
CHECK(param->confWinRightOffset < 0, "Conformance Window Right Offset must be 0 or greater");348
CHECK(param->confWinBottomOffset < 0, "Conformance Window Bottom Offset must be 0 or greater");349
CHECK(param->decoderVbvMaxRate < 0, "Invalid Decoder Vbv Maxrate. Value can not be less than zero");350
351
x265_log(param, X265_LOG_WARNING, "Live VBV enabled without VBV settings.Disabling live VBV in 2 pass\n");352
}353
}354
+ CHECK(param->rc.dataShareMode != X265_SHARE_MODE_FILE && param->rc.dataShareMode != X265_SHARE_MODE_SHAREDMEM, "Invalid data share mode. It must be one of the X265_DATA_SHARE_MODES enum values\n" );355
return check_failed;356
}357
358
359
x265_log(param, X265_LOG_INFO, "Keyframe min / max / scenecut / bias : %d / %d / %d / %.2lf \n",360
param->keyframeMin, param->keyframeMax, param->scenecutThreshold, param->scenecutBias * 100);361
else if (param->bHistBasedSceneCut && param->keyframeMax != INT_MAX) 362
- x265_log(param, X265_LOG_INFO, "Keyframe min / max / scenecut / edge threshold : %d / %d / %d / %.2lf\n",363
- param->keyframeMin, param->keyframeMax, param->bHistBasedSceneCut, param->edgeTransitionThreshold);364
+ x265_log(param, X265_LOG_INFO, "Keyframe min / max / scenecut : %d / %d / %d\n",365
+ param->keyframeMin, param->keyframeMax, param->bHistBasedSceneCut);366
else if (param->keyframeMax == INT_MAX)367
x265_log(param, X265_LOG_INFO, "Keyframe min / max / scenecut : disabled\n");368
369
370
bufSize += strlen(p->numaPools);371
if (p->masteringDisplayColorVolume)372
bufSize += strlen(p->masteringDisplayColorVolume);373
+ if (p->videoSignalTypePreset)374
+ bufSize += strlen(p->videoSignalTypePreset);375
376
buf = s = X265_MALLOC(char, bufSize);377
if (!buf)378
379
BOOL(p->bRepeatHeaders, "repeat-headers");380
BOOL(p->bAnnexB, "annexb");381
BOOL(p->bEnableAccessUnitDelimiters, "aud");382
+ BOOL(p->bEnableEndOfBitstream, "eob");383
+ BOOL(p->bEnableEndOfSequence, "eos");384
BOOL(p->bEmitHRDSEI, "hrd");385
BOOL(p->bEmitInfoSEI, "info");386
s += sprintf(s, " hash=%d", p->decodedPictureHashSEI);387
- BOOL(p->bEnableTemporalSubLayers, "temporal-layers");388
+ s += sprintf(s, " temporal-layers=%d", p->bEnableTemporalSubLayers);389
BOOL(p->bOpenGOP, "open-gop");390
s += sprintf(s, " min-keyint=%d", p->keyframeMin);391
s += sprintf(s, " keyint=%d", p->keyframeMax);392
393
s += sprintf(s, " rc-lookahead=%d", p->lookaheadDepth);394
s += sprintf(s, " lookahead-slices=%d", p->lookaheadSlices);395
s += sprintf(s, " scenecut=%d", p->scenecutThreshold);396
- s += sprintf(s, " hist-scenecut=%d", p->bHistBasedSceneCut);397
+ BOOL(p->bHistBasedSceneCut, "hist-scenecut");398
s += sprintf(s, " radl=%d", p->radl);399
BOOL(p->bEnableHRDConcatFlag, "splice");400
BOOL(p->bIntraRefresh, "intra-refresh");401
402
BOOL(p->bOptRefListLengthPPS, "opt-ref-list-length-pps");403
BOOL(p->bMultiPassOptRPS, "multi-pass-opt-rps");404
s += sprintf(s, " scenecut-bias=%.2f", p->scenecutBias);405
- s += sprintf(s, " hist-threshold=%.2f", p->edgeTransitionThreshold);406
BOOL(p->bOptCUDeltaQP, "opt-cu-delta-qp");407
BOOL(p->bAQMotion, "aq-motion");408
BOOL(p->bEmitHDR10SEI, "hdr10");409
410
s += sprintf(s, " qp-adaptation-range=%.2f", p->rc.qpAdaptationRange);411
s += sprintf(s, " scenecut-aware-qp=%d", p->bEnableSceneCutAwareQp);412
if (p->bEnableSceneCutAwareQp)413
- s += sprintf(s, " fwd-scenecut-window=%d fwd-ref-qp-delta=%f fwd-nonref-qp-delta=%f bwd-scenecut-window=%d bwd-ref-qp-delta=%f bwd-nonref-qp-delta=%f", p->fwdScenecutWindow, p->fwdRefQpDelta, p->fwdNonRefQpDelta, p->bwdScenecutWindow, p->bwdRefQpDelta, p->bwdNonRefQpDelta);414
+ s += sprintf(s, " fwd-scenecut-window=%d fwd-ref-qp-delta=%f fwd-nonref-qp-delta=%f bwd-scenecut-window=%d bwd-ref-qp-delta=%f bwd-nonref-qp-delta=%f", p->fwdMaxScenecutWindow, p->fwdRefQpDelta0, p->fwdNonRefQpDelta0, p->bwdMaxScenecutWindow, p->bwdRefQpDelta0, p->bwdNonRefQpDelta0);415
s += sprintf(s, "conformance-window-offsets right=%d bottom=%d", p->confWinRightOffset, p->confWinBottomOffset);416
s += sprintf(s, " decoder-max-rate=%d", p->decoderVbvMaxRate);417
BOOL(p->bliveVBV2pass, "vbv-live-multi-pass");418
+ if (p->filmGrain)419
+ s += sprintf(s, " film-grain=%s", p->filmGrain); // Film grain characteristics model filename420
+ BOOL(p->bEnableTemporalFilter, "mcstf");421
+ BOOL(p->bEnableSBRC, "sbrc");422
#undef BOOL423
return buf;424
}425
426
return false;427
}428
429
+bool parseMaskingStrength(x265_param* p, const char* value)430
+{431
+ bool bError = false;432
+ int window16;433
+ double refQpDelta16, nonRefQpDelta16;434
+ if (p->bEnableSceneCutAwareQp == FORWARD)435
+ {436
+ if (3 == sscanf(value, "%d,%lf,%lf", &window10, &refQpDelta10, &nonRefQpDelta10))437
+ {438
+ if (window10 > 0)439
+ p->fwdMaxScenecutWindow = window10;440
+ if (refQpDelta10 > 0)441
+ p->fwdRefQpDelta0 = refQpDelta10;442
+ if (nonRefQpDelta10 > 0)443
+ p->fwdNonRefQpDelta0 = nonRefQpDelta10;444
+445
+ p->fwdScenecutWindow0 = p->fwdMaxScenecutWindow / 6;446
+ for (int i = 1; i < 6; i++)447
+ {448
+ p->fwdScenecutWindowi = p->fwdMaxScenecutWindow / 6;449
+ p->fwdRefQpDeltai = p->fwdRefQpDeltai - 1 - (0.15 * p->fwdRefQpDeltai - 1);450
+ p->fwdNonRefQpDeltai = p->fwdNonRefQpDeltai - 1 - (0.15 * p->fwdNonRefQpDeltai - 1);451
+ }452
+ }453
+ else if (18 == sscanf(value, "%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf"454
+ , &window10, &refQpDelta10, &nonRefQpDelta10, &window11, &refQpDelta11, &nonRefQpDelta11455
+ , &window12, &refQpDelta12, &nonRefQpDelta12, &window13, &refQpDelta13, &nonRefQpDelta13456
+ , &window14, &refQpDelta14, &nonRefQpDelta14, &window15, &refQpDelta15, &nonRefQpDelta15))457
+ {458
+ p->fwdMaxScenecutWindow = 0;459
+ for (int i = 0; i < 6; i++)460
+ {461
+ p->fwdScenecutWindowi = window1i;462
+ p->fwdRefQpDeltai = refQpDelta1i;463
+ p->fwdNonRefQpDeltai = nonRefQpDelta1i;464
+ p->fwdMaxScenecutWindow += p->fwdScenecutWindowi;465
+ }466
+ }467
+ else468
+ {469
+ x265_log(NULL, X265_LOG_ERROR, "Specify all the necessary offsets for masking-strength \n");470
+ bError = true;471
+ }472
+ }473
+ else if (p->bEnableSceneCutAwareQp == BACKWARD)474
+ {475
+ if (3 == sscanf(value, "%d,%lf,%lf", &window10, &refQpDelta10, &nonRefQpDelta10))476
+ {477
+ if (window10 > 0)478
+ p->bwdMaxScenecutWindow = window10;479
+ if (refQpDelta10 > 0)480
+ p->bwdRefQpDelta0 = refQpDelta10;481
+ if (nonRefQpDelta10 > 0)482
+ p->bwdNonRefQpDelta0 = nonRefQpDelta10;483
+484
+ p->bwdScenecutWindow0 = p->bwdMaxScenecutWindow / 6;485
+ for (int i = 1; i < 6; i++)486
+ {487
+ p->bwdScenecutWindowi = p->bwdMaxScenecutWindow / 6;488
+ p->bwdRefQpDeltai = p->bwdRefQpDeltai - 1 - (0.15 * p->bwdRefQpDeltai - 1);489
+ p->bwdNonRefQpDeltai = p->bwdNonRefQpDeltai - 1 - (0.15 * p->bwdNonRefQpDeltai - 1);490
+ }491
+ }492
+ else if (18 == sscanf(value, "%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf"493
+ , &window10, &refQpDelta10, &nonRefQpDelta10, &window11, &refQpDelta11, &nonRefQpDelta11494
+ , &window12, &refQpDelta12, &nonRefQpDelta12, &window13, &refQpDelta13, &nonRefQpDelta13495
+ , &window14, &refQpDelta14, &nonRefQpDelta14, &window15, &refQpDelta15, &nonRefQpDelta15))496
+ {497
+ p->bwdMaxScenecutWindow = 0;498
+ for (int i = 0; i < 6; i++)499
+ {500
+ p->bwdScenecutWindowi = window1i;501
+ p->bwdRefQpDeltai = refQpDelta1i;502
+ p->bwdNonRefQpDeltai = nonRefQpDelta1i;503
+ p->bwdMaxScenecutWindow += p->bwdScenecutWindowi;504
+ }505
+ }506
+ else507
+ {508
+ x265_log(NULL, X265_LOG_ERROR, "Specify all the necessary offsets for masking-strength \n");509
+ bError = true;510
+ }511
+ }512
+ else if (p->bEnableSceneCutAwareQp == BI_DIRECTIONAL)513
+ {514
+ int window26;515
+ double refQpDelta26, nonRefQpDelta26;516
+ if (6 == sscanf(value, "%d,%lf,%lf,%d,%lf,%lf", &window10, &refQpDelta10, &nonRefQpDelta10, &window20, &refQpDelta20, &nonRefQpDelta20))517
+ {518
+ if (window10 > 0)519
+ p->fwdMaxScenecutWindow = window10;520
+ if (refQpDelta10 > 0)521
+ p->fwdRefQpDelta0 = refQpDelta10;522
+ if (nonRefQpDelta10 > 0)523
+ p->fwdNonRefQpDelta0 = nonRefQpDelta10;524
+ if (window20 > 0)525
+ p->bwdMaxScenecutWindow = window20;526
+ if (refQpDelta20 > 0)527
+ p->bwdRefQpDelta0 = refQpDelta20;528
+ if (nonRefQpDelta20 > 0)529
+ p->bwdNonRefQpDelta0 = nonRefQpDelta20;530
+531
+ p->fwdScenecutWindow0 = p->fwdMaxScenecutWindow / 6;532
+ p->bwdScenecutWindow0 = p->bwdMaxScenecutWindow / 6;533
+ for (int i = 1; i < 6; i++)534
+ {535
+ p->fwdScenecutWindowi = p->fwdMaxScenecutWindow / 6;536
+ p->bwdScenecutWindowi = p->bwdMaxScenecutWindow / 6;537
+ p->fwdRefQpDeltai = p->fwdRefQpDeltai - 1 - (0.15 * p->fwdRefQpDeltai - 1);538
+ p->fwdNonRefQpDeltai = p->fwdNonRefQpDeltai - 1 - (0.15 * p->fwdNonRefQpDeltai - 1);539
+ p->bwdRefQpDeltai = p->bwdRefQpDeltai - 1 - (0.15 * p->bwdRefQpDeltai - 1);540
+ p->bwdNonRefQpDeltai = p->bwdNonRefQpDeltai - 1 - (0.15 * p->bwdNonRefQpDeltai - 1);541
+ }542
+ }543
+ else if (36 == sscanf(value, "%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf,%d,%lf,%lf"544
+ , &window10, &refQpDelta10, &nonRefQpDelta10, &window11, &refQpDelta11, &nonRefQpDelta11545
+ , &window12, &refQpDelta12, &nonRefQpDelta12, &window13, &refQpDelta13, &nonRefQpDelta13546
+ , &window14, &refQpDelta14, &nonRefQpDelta14, &window15, &refQpDelta15, &nonRefQpDelta15547
+ , &window20, &refQpDelta20, &nonRefQpDelta20, &window21, &refQpDelta21, &nonRefQpDelta21548
+ , &window22, &refQpDelta22, &nonRefQpDelta22, &window23, &refQpDelta23, &nonRefQpDelta23549
+ , &window24, &refQpDelta24, &nonRefQpDelta24, &window25, &refQpDelta25, &nonRefQpDelta25))550
+ {551
+ p->fwdMaxScenecutWindow = 0;552
+ p->bwdMaxScenecutWindow = 0;553
+ for (int i = 0; i < 6; i++)554
+ {555
+ p->fwdScenecutWindowi = window1i;556
+ p->fwdRefQpDeltai = refQpDelta1i;557
+ p->fwdNonRefQpDeltai = nonRefQpDelta1i;558
+ p->bwdScenecutWindowi = window2i;559
+ p->bwdRefQpDeltai = refQpDelta2i;560
+ p->bwdNonRefQpDeltai = nonRefQpDelta2i;561
+ p->fwdMaxScenecutWindow += p->fwdScenecutWindowi;562
+ p->bwdMaxScenecutWindow += p->bwdScenecutWindowi;563
+ }564
+ }565
+ else566
+ {567
+ x265_log(NULL, X265_LOG_ERROR, "Specify all the necessary offsets for masking-strength \n");568
+ bError = true;569
+ }570
+ }571
+ return bError;572
+}573
+574
void x265_copy_params(x265_param* dst, x265_param* src)575
{576
dst->cpuid = src->cpuid;577
578
dst->bRepeatHeaders = src->bRepeatHeaders;579
dst->bAnnexB = src->bAnnexB;580
dst->bEnableAccessUnitDelimiters = src->bEnableAccessUnitDelimiters;581
+ dst->bEnableEndOfBitstream = src->bEnableEndOfBitstream;582
+ dst->bEnableEndOfSequence = src->bEnableEndOfSequence;583
dst->bEmitInfoSEI = src->bEmitInfoSEI;584
dst->decodedPictureHashSEI = src->decodedPictureHashSEI;585
dst->bEnableTemporalSubLayers = src->bEnableTemporalSubLayers;586
dst->bOpenGOP = src->bOpenGOP;587
+ dst->craNal = src->craNal;588
dst->keyframeMax = src->keyframeMax;589
dst->keyframeMin = src->keyframeMin;590
dst->bframes = src->bframes;591
592
dst->rc.rfConstantMin = src->rc.rfConstantMin;593
dst->rc.bStatWrite = src->rc.bStatWrite;594
dst->rc.bStatRead = src->rc.bStatRead;595
+ dst->rc.dataShareMode = src->rc.dataShareMode;596
if (src->rc.statFileName) dst->rc.statFileName=strdup(src->rc.statFileName);597
else dst->rc.statFileName = NULL;598
+ if (src->rc.sharedMemName) dst->rc.sharedMemName = strdup(src->rc.sharedMemName);599
+ else dst->rc.sharedMemName = NULL;600
dst->rc.qblur = src->rc.qblur;601
dst->rc.complexityBlur = src->rc.complexityBlur;602
dst->rc.bEnableSlowFirstPass = src->rc.bEnableSlowFirstPass;603
604
dst->rc.zonefileCount = src->rc.zonefileCount;605
dst->reconfigWindowSize = src->reconfigWindowSize;606
dst->bResetZoneConfig = src->bResetZoneConfig;607
+ dst->bNoResetZoneConfig = src->bNoResetZoneConfig;608
dst->decoderVbvMaxRate = src->decoderVbvMaxRate;609
610
if (src->rc.zonefileCount && src->rc.zones && src->bResetZoneConfig)611
612
for (int i = 0; i < src->rc.zonefileCount; i++)613
{614
dst->rc.zonesi.startFrame = src->rc.zonesi.startFrame;615
+ dst->rc.zones0.keyframeMax = src->rc.zones0.keyframeMax;616
memcpy(dst->rc.zonesi.zoneParam, src->rc.zonesi.zoneParam, sizeof(x265_param));617
}618
}619
620
dst->bOptRefListLengthPPS = src->bOptRefListLengthPPS;621
dst->bMultiPassOptRPS = src->bMultiPassOptRPS;622
dst->scenecutBias = src->scenecutBias;623
- dst->edgeTransitionThreshold = src->edgeTransitionThreshold;624
dst->gopLookahead = src->lookaheadDepth;625
dst->bOptCUDeltaQP = src->bOptCUDeltaQP;626
dst->analysisMultiPassDistortion = src->analysisMultiPassDistortion;627
628
dst->bEnableSvtHevc = src->bEnableSvtHevc;629
dst->bEnableFades = src->bEnableFades;630
dst->bEnableSceneCutAwareQp = src->bEnableSceneCutAwareQp;631
- dst->fwdScenecutWindow = src->fwdScenecutWindow;632
- dst->fwdRefQpDelta = src->fwdRefQpDelta;633
- dst->fwdNonRefQpDelta = src->fwdNonRefQpDelta;634
- dst->bwdScenecutWindow = src->bwdScenecutWindow;635
- dst->bwdRefQpDelta = src->bwdRefQpDelta;636
- dst->bwdNonRefQpDelta = src->bwdNonRefQpDelta;637
+ dst->fwdMaxScenecutWindow = src->fwdMaxScenecutWindow;638
+ dst->bwdMaxScenecutWindow = src->bwdMaxScenecutWindow;639
+ for (int i = 0; i < 6; i++)640
+ {641
+ dst->fwdScenecutWindowi = src->fwdScenecutWindowi;642
+ dst->fwdRefQpDeltai = src->fwdRefQpDeltai;643
+ dst->fwdNonRefQpDeltai = src->fwdNonRefQpDeltai;644
+ dst->bwdScenecutWindowi = src->bwdScenecutWindowi;645
+ dst->bwdRefQpDeltai = src->bwdRefQpDeltai;646
+ dst->bwdNonRefQpDeltai = src->bwdNonRefQpDeltai;647
+ }648
dst->bField = src->bField;649
-650
+ dst->bEnableTemporalFilter = src->bEnableTemporalFilter;651
+ dst->temporalFilterStrength = src->temporalFilterStrength;652
dst->confWinRightOffset = src->confWinRightOffset;653
dst->confWinBottomOffset = src->confWinBottomOffset;654
dst->bliveVBV2pass = src->bliveVBV2pass;655
+656
+ if (src->videoSignalTypePreset) dst->videoSignalTypePreset = strdup(src->videoSignalTypePreset);657
+ else dst->videoSignalTypePreset = NULL;658
#ifdef SVT_HEVC659
memcpy(dst->svtHevcParam, src->svtHevcParam, sizeof(EB_H265_ENC_CONFIGURATION));660
#endif661
+ /* Film grain */662
+ if (src->filmGrain)663
+ dst->filmGrain = src->filmGrain;664
+ dst->bEnableSBRC = src->bEnableSBRC;665
}666
667
#ifdef SVT_HEVC668
x265_3.5.tar.gz/source/common/param.h -> x265_3.6.tar.gz/source/common/param.h
Changed
17
1
2
void getParamAspectRatio(x265_param *p, int& width, int& height);3
bool parseLambdaFile(x265_param *param);4
void x265_copy_params(x265_param* dst, x265_param* src);5
+bool parseMaskingStrength(x265_param* p, const char* value);6
7
/* this table is kept internal to avoid confusion, since log level indices start at -1 */8
static const char * const logLevelNames = { "none", "error", "warning", "info", "debug", "full", 0 };9
10
int x265_param_default_preset(x265_param *, const char *preset, const char *tune);11
int x265_param_apply_profile(x265_param *, const char *profile);12
int x265_param_parse(x265_param *p, const char *name, const char *value);13
+int x265_scenecut_aware_qp_param_parse(x265_param* p, const char* name, const char* value);14
int x265_zone_param_parse(x265_param* p, const char* name, const char* value);15
#define PARAM_NS X265_NS16
#endif17
x265_3.5.tar.gz/source/common/piclist.cpp -> x265_3.6.tar.gz/source/common/piclist.cpp
Changed
134
1
2
m_count++;3
}4
5
+void PicList::pushFrontMCSTF(Frame& curFrame)6
+{7
+ X265_CHECK(!curFrame.m_nextMCSTF && !curFrame.m_nextMCSTF, "piclist: picture already in OPB list\n"); // ensure frame is not in a list8
+ curFrame.m_nextMCSTF = m_start;9
+ curFrame.m_prevMCSTF = NULL;10
+11
+ if (m_count)12
+ {13
+ m_start->m_prevMCSTF = &curFrame;14
+ m_start = &curFrame;15
+ }16
+ else17
+ {18
+ m_start = m_end = &curFrame;19
+ }20
+ m_count++;21
+22
+}23
+24
void PicList::pushBack(Frame& curFrame)25
{26
X265_CHECK(!curFrame.m_next && !curFrame.m_prev, "piclist: picture already in list\n"); // ensure frame is not in a list27
28
m_count++;29
}30
31
+void PicList::pushBackMCSTF(Frame& curFrame)32
+{33
+ X265_CHECK(!curFrame.m_nextMCSTF && !curFrame.m_prevMCSTF, "piclist: picture already in OPB list\n"); // ensure frame is not in a list34
+ curFrame.m_nextMCSTF = NULL;35
+ curFrame.m_prevMCSTF = m_end;36
+37
+ if (m_count)38
+ {39
+ m_end->m_nextMCSTF = &curFrame;40
+ m_end = &curFrame;41
+ }42
+ else43
+ {44
+ m_start = m_end = &curFrame;45
+ }46
+ m_count++;47
+}48
+49
Frame *PicList::popFront()50
{51
if (m_start)52
53
return curFrame;54
}55
56
+Frame* PicList::getPOCMCSTF(int poc)57
+{58
+ Frame *curFrame = m_start;59
+ while (curFrame && curFrame->m_poc != poc)60
+ curFrame = curFrame->m_nextMCSTF;61
+ return curFrame;62
+}63
+64
Frame *PicList::popBack()65
{66
if (m_end)67
68
return NULL;69
}70
71
+Frame *PicList::popBackMCSTF()72
+{73
+ if (m_end)74
+ {75
+ Frame* temp = m_end;76
+ m_count--;77
+78
+ if (m_count)79
+ {80
+ m_end = m_end->m_prevMCSTF;81
+ m_end->m_nextMCSTF = NULL;82
+ }83
+ else84
+ {85
+ m_start = m_end = NULL;86
+ }87
+ temp->m_nextMCSTF = temp->m_prevMCSTF = NULL;88
+ return temp;89
+ }90
+ else91
+ return NULL;92
+}93
+94
Frame* PicList::getCurFrame(void)95
{96
Frame *curFrame = m_start;97
98
99
curFrame.m_next = curFrame.m_prev = NULL;100
}101
+102
+void PicList::removeMCSTF(Frame& curFrame)103
+{104
+#if _DEBUG105
+ Frame *tmp = m_start;106
+ while (tmp && tmp != &curFrame)107
+ {108
+ tmp = tmp->m_nextMCSTF;109
+ }110
+111
+ X265_CHECK(tmp == &curFrame, "framelist: pic being removed was not in list\n"); // verify pic is in this list112
+#endif113
+114
+ m_count--;115
+ if (m_count)116
+ {117
+ if (m_start == &curFrame)118
+ m_start = curFrame.m_nextMCSTF;119
+ if (m_end == &curFrame)120
+ m_end = curFrame.m_prevMCSTF;121
+122
+ if (curFrame.m_nextMCSTF)123
+ curFrame.m_nextMCSTF->m_prevMCSTF = curFrame.m_prevMCSTF;124
+ if (curFrame.m_prevMCSTF)125
+ curFrame.m_prevMCSTF->m_nextMCSTF = curFrame.m_nextMCSTF;126
+ }127
+ else128
+ {129
+ m_start = m_end = NULL;130
+ }131
+132
+ curFrame.m_nextMCSTF = curFrame.m_prevMCSTF = NULL;133
+}134
x265_3.5.tar.gz/source/common/piclist.h -> x265_3.6.tar.gz/source/common/piclist.h
Changed
33
1
2
3
/** Push picture to end of the list */4
void pushBack(Frame& pic);5
+ void pushBackMCSTF(Frame& pic);6
7
/** Push picture to beginning of the list */8
void pushFront(Frame& pic);9
+ void pushFrontMCSTF(Frame& pic);10
11
/** Pop picture from end of the list */12
Frame* popBack();13
+ Frame* popBackMCSTF();14
15
/** Pop picture from beginning of the list */16
Frame* popFront();17
18
/** Find frame with specified POC */19
Frame* getPOC(int poc);20
+ /* Find next MCSTF frame with specified POC */21
+ Frame* getPOCMCSTF(int poc);22
23
/** Get the current Frame from the list **/24
Frame* getCurFrame(void);25
26
/** Remove picture from list */27
void remove(Frame& pic);28
+ /* Remove MCSTF picture from list */29
+ void removeMCSTF(Frame& pic);30
31
Frame* first() { return m_start; }32
33
x265_3.5.tar.gz/source/common/picyuv.cpp -> x265_3.6.tar.gz/source/common/picyuv.cpp
Changed
60
1
2
return false;3
}4
5
+/*Copy pixels from the picture buffer of a frame to picture buffer of another frame*/6
+void PicYuv::copyFromFrame(PicYuv* source)7
+{8
+ uint32_t numCuInHeight = (m_picHeight + m_param->maxCUSize - 1) / m_param->maxCUSize;9
+10
+ int maxHeight = numCuInHeight * m_param->maxCUSize;11
+ memcpy(m_picBuf0, source->m_picBuf0, sizeof(pixel)* m_stride * (maxHeight + (m_lumaMarginY * 2)));12
+ m_picOrg0 = m_picBuf0 + m_lumaMarginY * m_stride + m_lumaMarginX;13
+14
+ if (m_picCsp != X265_CSP_I400)15
+ {16
+ memcpy(m_picBuf1, source->m_picBuf1, sizeof(pixel)* m_strideC * ((maxHeight >> m_vChromaShift) + (m_chromaMarginY * 2)));17
+ memcpy(m_picBuf2, source->m_picBuf2, sizeof(pixel)* m_strideC * ((maxHeight >> m_vChromaShift) + (m_chromaMarginY * 2)));18
+19
+ m_picOrg1 = m_picBuf1 + m_chromaMarginY * m_strideC + m_chromaMarginX;20
+ m_picOrg2 = m_picBuf2 + m_chromaMarginY * m_strideC + m_chromaMarginX;21
+ }22
+ else23
+ {24
+ m_picBuf1 = m_picBuf2 = NULL;25
+ m_picOrg1 = m_picOrg2 = NULL;26
+ }27
+}28
+29
+bool PicYuv::createScaledPicYUV(x265_param* param, uint8_t scaleFactor)30
+{31
+ m_param = param;32
+ m_picWidth = m_param->sourceWidth / scaleFactor;33
+ m_picHeight = m_param->sourceHeight / scaleFactor;34
+35
+ m_picCsp = m_param->internalCsp;36
+ m_hChromaShift = CHROMA_H_SHIFT(m_picCsp);37
+ m_vChromaShift = CHROMA_V_SHIFT(m_picCsp);38
+39
+ uint32_t numCuInWidth = (m_picWidth + param->maxCUSize - 1) / param->maxCUSize;40
+ uint32_t numCuInHeight = (m_picHeight + param->maxCUSize - 1) / param->maxCUSize;41
+42
+ m_lumaMarginX = 128; // search margin for L0 and L1 ME in horizontal direction43
+ m_lumaMarginY = 128; // search margin for L0 and L1 ME in vertical direction44
+ m_stride = (numCuInWidth * param->maxCUSize) + (m_lumaMarginX << 1);45
+46
+ int maxHeight = numCuInHeight * param->maxCUSize;47
+ CHECKED_MALLOC_ZERO(m_picBuf0, pixel, m_stride * (maxHeight + (m_lumaMarginY * 2)));48
+ m_picOrg0 = m_picBuf0 + m_lumaMarginY * m_stride + m_lumaMarginX;49
+ m_picBuf1 = m_picBuf2 = NULL;50
+ m_picOrg1 = m_picOrg2 = NULL;51
+ return true;52
+53
+fail:54
+ return false;55
+}56
+57
int PicYuv::getLumaBufLen(uint32_t picWidth, uint32_t picHeight, uint32_t picCsp)58
{59
m_picWidth = picWidth;60
x265_3.5.tar.gz/source/common/picyuv.h -> x265_3.6.tar.gz/source/common/picyuv.h
Changed
15
1
2
PicYuv();3
4
bool create(x265_param* param, bool picAlloc = true, pixel *pixelbuf = NULL);5
+ bool createScaledPicYUV(x265_param* param, uint8_t scaleFactor);6
bool createOffsets(const SPS& sps);7
void destroy();8
int getLumaBufLen(uint32_t picWidth, uint32_t picHeight, uint32_t picCsp);9
10
void copyFromPicture(const x265_picture&, const x265_param& param, int padx, int pady);11
+ void copyFromFrame(PicYuv* source);12
13
intptr_t getChromaAddrOffset(uint32_t ctuAddr, uint32_t absPartIdx) const { return m_cuOffsetCctuAddr + m_buOffsetCabsPartIdx; }14
15
x265_3.5.tar.gz/source/common/pixel.cpp -> x265_3.6.tar.gz/source/common/pixel.cpp
Changed
51
1
2
{3
int satd = 0;4
5
-#if ENABLE_ASSEMBLY && X265_ARCH_ARM646
+#if ENABLE_ASSEMBLY && X265_ARCH_ARM64 && !HIGH_BIT_DEPTH7
pixelcmp_t satd_4x4 = x265_pixel_satd_4x4_neon;8
#endif9
10
11
{12
int satd = 0;13
14
-#if ENABLE_ASSEMBLY && X265_ARCH_ARM6415
+#if ENABLE_ASSEMBLY && X265_ARCH_ARM64 && !HIGH_BIT_DEPTH16
pixelcmp_t satd_8x4 = x265_pixel_satd_8x4_neon;17
#endif18
19
20
}21
}22
23
+static24
+void frame_subsample_luma(const pixel* src0, pixel* dst0, intptr_t src_stride, intptr_t dst_stride, int width, int height)25
+{26
+ for (int y = 0; y < height; y++, src0 += 2 * src_stride, dst0 += dst_stride)27
+ {28
+ const pixel *inRow = src0;29
+ const pixel *inRowBelow = src0 + src_stride;30
+ pixel *target = dst0;31
+ for (int x = 0; x < width; x++)32
+ {33
+ targetx = (((inRow0 + inRowBelow0 + 1) >> 1) + ((inRow1 + inRowBelow1 + 1) >> 1) + 1) >> 1;34
+ inRow += 2;35
+ inRowBelow += 2;36
+ }37
+ }38
+}39
+40
/* structural similarity metric */41
static void ssim_4x4x2_core(const pixel* pix1, intptr_t stride1, const pixel* pix2, intptr_t stride2, int sums24)42
{43
44
p.cuBLOCK_16x16.normFact = normFact_c;45
p.cuBLOCK_32x32.normFact = normFact_c;46
p.cuBLOCK_64x64.normFact = normFact_c;47
+ /* SubSample Luma*/48
+ p.frameSubSampleLuma = frame_subsample_luma;49
}50
}51
x265_3.5.tar.gz/source/common/ppc/intrapred_altivec.cpp -> x265_3.6.tar.gz/source/common/ppc/intrapred_altivec.cpp
Changed
10
1
2
#include <assert.h>3
#include <math.h>4
#include <cmath>5
-#include <linux/types.h>6
+#include <sys/types.h>7
#include <stdlib.h>8
#include <stdio.h>9
#include <stdint.h>10
x265_3.5.tar.gz/source/common/primitives.h -> x265_3.6.tar.gz/source/common/primitives.h
Changed
28
1
2
typedef void(*psyRdoQuant_t2)(int16_t *m_resiDctCoeff, int16_t *m_fencDctCoeff, int64_t *costUncoded, int64_t *totalUncodedCost, int64_t *totalRdCost, int64_t *psyScale, uint32_t blkPos);3
typedef void(*ssimDistortion_t)(const pixel *fenc, uint32_t fStride, const pixel *recon, intptr_t rstride, uint64_t *ssBlock, int shift, uint64_t *ac_k);4
typedef void(*normFactor_t)(const pixel *src, uint32_t blockSize, int shift, uint64_t *z_k);5
+/* SubSampling Luma */6
+typedef void (*downscaleluma_t)(const pixel* src0, pixel* dstf, intptr_t src_stride, intptr_t dst_stride, int width, int height);7
/* Function pointers to optimized encoder primitives. Each pointer can reference8
* either an assembly routine, a SIMD intrinsic primitive, or a C function */9
struct EncoderPrimitives10
11
12
downscale_t frameInitLowres;13
downscale_t frameInitLowerRes;14
+ /* Sub Sample Luma */15
+ downscaleluma_t frameSubSampleLuma;16
cutree_propagate_cost propagateCost;17
cutree_fix8_unpack fix8Unpack;18
cutree_fix8_pack fix8Pack;19
20
21
#if ENABLE_ASSEMBLY && X265_ARCH_ARM6422
extern "C" {23
-#include "aarch64/pixel-util.h"24
+#include "aarch64/fun-decls.h"25
}26
#endif27
28
x265_3.6.tar.gz/source/common/ringmem.cpp
Added
359
1
2
+/*****************************************************************************3
+ * Copyright (C) 2013-2017 MulticoreWare, Inc4
+ *5
+ * Authors: liwei <liwei@multicorewareinc.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com23
+ *****************************************************************************/24
+25
+#include "ringmem.h"26
+27
+#ifndef _WIN3228
+#include <sys/mman.h>29
+#endif ////< _WIN3230
+31
+#ifdef _WIN3232
+#define X265_SHARED_MEM_NAME "Local\\_x265_shr_mem_"33
+#define X265_SEMAPHORE_RINGMEM_WRITER_NAME "_x265_semW_"34
+#define X265_SEMAPHORE_RINGMEM_READER_NAME "_x265_semR_"35
+#else /* POSIX / pthreads */36
+#define X265_SHARED_MEM_NAME "/tmp/_x265_shr_mem_"37
+#define X265_SEMAPHORE_RINGMEM_WRITER_NAME "/tmp/_x265_semW_"38
+#define X265_SEMAPHORE_RINGMEM_READER_NAME "/tmp/_x265_semR_"39
+#endif40
+41
+#define RINGMEM_ALLIGNMENT 6442
+43
+namespace X265_NS {44
+ RingMem::RingMem() 45
+ : m_initialized(false)46
+ , m_protectRW(false)47
+ , m_itemSize(0)48
+ , m_itemCnt(0)49
+ , m_dataPool(NULL)50
+ , m_shrMem(NULL)51
+#ifdef _WIN3252
+ , m_handle(NULL)53
+#else //_WIN3254
+ , m_filepath(NULL)55
+#endif //_WIN3256
+ , m_writeSem(NULL)57
+ , m_readSem(NULL)58
+ {59
+ }60
+61
+62
+ RingMem::~RingMem()63
+ {64
+ }65
+66
+ bool RingMem::skipRead(int32_t cnt) {67
+ if (!m_initialized)68
+ {69
+ return false;70
+ }71
+72
+ if (m_protectRW)73
+ {74
+ for (int i = 0; i < cnt; i++)75
+ {76
+ m_readSem->take();77
+ }78
+ }79
+ 80
+ ATOMIC_ADD(&m_shrMem->m_read, cnt);81
+82
+ if (m_protectRW)83
+ {84
+ m_writeSem->give(cnt);85
+ }86
+87
+ return true;88
+ }89
+90
+ bool RingMem::skipWrite(int32_t cnt) {91
+ if (!m_initialized)92
+ {93
+ return false;94
+ }95
+96
+ if (m_protectRW)97
+ {98
+ for (int i = 0; i < cnt; i++)99
+ {100
+ m_writeSem->take();101
+ }102
+ }103
+104
+ ATOMIC_ADD(&m_shrMem->m_write, cnt);105
+106
+ if (m_protectRW)107
+ {108
+ m_readSem->give(cnt);109
+ }110
+111
+ return true;112
+ }113
+114
+ ///< initialize115
+ bool RingMem::init(int32_t itemSize, int32_t itemCnt, const char *name, bool protectRW)116
+ {117
+ ///< check parameters118
+ if (itemSize <= 0 || itemCnt <= 0 || NULL == name)119
+ {120
+ ///< invalid parameters 121
+ return false;122
+ }123
+124
+ if (!m_initialized)125
+ {126
+ ///< formating names127
+ char nameBufMAX_SHR_NAME_LEN = { 0 };128
+129
+ ///< shared memory name130
+ snprintf(nameBuf, sizeof(nameBuf) - 1, "%s%s", X265_SHARED_MEM_NAME, name);131
+132
+ ///< create or open shared memory133
+ bool newCreated = false;134
+135
+ ///< calculate the size of the shared memory136
+ int32_t shrMemSize = (itemSize * itemCnt + sizeof(ShrMemCtrl) + RINGMEM_ALLIGNMENT - 1) & ~(RINGMEM_ALLIGNMENT - 1);137
+138
+#ifdef _WIN32139
+ HANDLE h = OpenFileMappingA(FILE_MAP_WRITE | FILE_MAP_READ, FALSE, nameBuf);140
+ if (!h)141
+ {142
+ h = CreateFileMappingA(INVALID_HANDLE_VALUE, NULL, PAGE_READWRITE, 0, shrMemSize, nameBuf);143
+144
+ if (!h)145
+ {146
+ return false;147
+ }148
+149
+ newCreated = true;150
+ }151
+152
+ void *pool = MapViewOfFile(h, FILE_MAP_ALL_ACCESS, 0, 0, 0);153
+154
+ ///< should not close the handle here, otherwise the OpenFileMapping would fail155
+ //CloseHandle(h);156
+ m_handle = h;157
+158
+ if (!pool)159
+ {160
+ return false;161
+ }162
+163
+#else /* POSIX / pthreads */164
+ mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH;165
+ int flag = O_RDWR;166
+ int shrfd = -1;167
+ if ((shrfd = open(nameBuf, flag, mode)) < 0)168
+ {169
+ flag |= O_CREAT;170
+ 171
+ shrfd = open(nameBuf, flag, mode);172
+ if (shrfd < 0)173
+ {174
+ return false;175
+ }176
+ newCreated = true;177
+178
+ lseek(shrfd, shrMemSize - 1, SEEK_SET);179
+180
+ if (-1 == write(shrfd, "\0", 1))181
+ {182
+ close(shrfd);183
+ return false;184
+ }185
+186
+ if (lseek(shrfd, 0, SEEK_END) < shrMemSize)187
+ {188
+ close(shrfd);189
+ return false;190
+ }191
+ }192
+193
+ void *pool = mmap(0,194
+ shrMemSize,195
+ PROT_READ | PROT_WRITE,196
+ MAP_SHARED,197
+ shrfd,198
+ 0);199
+200
+ close(shrfd);201
+ if (pool == MAP_FAILED)202
+ { 203
+ return false;204
+ }205
+206
+ m_filepath = strdup(nameBuf);207
+#endif ///< _WIN32208
+209
+ if (newCreated)210
+ {211
+ memset(pool, 0, shrMemSize);212
+ }213
+ 214
+ m_shrMem = reinterpret_cast<ShrMemCtrl *>(pool);215
+ m_dataPool = reinterpret_cast<uint8_t *>(pool) + sizeof(ShrMemCtrl);216
+ m_itemSize = itemSize;217
+ m_itemCnt = itemCnt;218
+ m_initialized = true;219
+220
+ if (protectRW)221
+ {222
+ m_protectRW = true;223
+ m_writeSem = new NamedSemaphore();224
+ if (!m_writeSem)225
+ {226
+ release();227
+ return false;228
+ }229
+230
+ ///< shared memory name231
+ snprintf(nameBuf, sizeof(nameBuf) - 1, "%s%s", X265_SEMAPHORE_RINGMEM_WRITER_NAME, name);232
+ if (!m_writeSem->create(nameBuf, m_itemCnt, m_itemCnt))233
+ {234
+ release();235
+ return false;236
+ }237
+238
+ m_readSem = new NamedSemaphore();239
+ if (!m_readSem)240
+ {241
+ release();242
+ return false;243
+ }244
+245
+ ///< shared memory name246
+ snprintf(nameBuf, sizeof(nameBuf) - 1, "%s%s", X265_SEMAPHORE_RINGMEM_READER_NAME, name);247
+ if (!m_readSem->create(nameBuf, 0, m_itemCnt))248
+ {249
+ release();250
+ return false;251
+ }252
+ }253
+ }254
+255
+ return true;256
+ }257
+ ///< finalize258
+ void RingMem::release()259
+ {260
+ if (m_initialized)261
+ {262
+ m_initialized = false;263
+264
+ if (m_shrMem)265
+ {266
+#ifdef _WIN32267
+ UnmapViewOfFile(m_shrMem);268
+ CloseHandle(m_handle);269
+ m_handle = NULL;270
+#else /* POSIX / pthreads */271
+ int32_t shrMemSize = (m_itemSize * m_itemCnt + sizeof(ShrMemCtrl) + RINGMEM_ALLIGNMENT - 1) & (~RINGMEM_ALLIGNMENT - 1);272
+ munmap(m_shrMem, shrMemSize);273
+ unlink(m_filepath);274
+ free(m_filepath);275
+ m_filepath = NULL;276
+#endif ///< _WIN32277
+ m_shrMem = NULL;278
+ m_dataPool = NULL;279
+ m_itemSize = 0;280
+ m_itemCnt = 0;281
+ }282
+ 283
+ if (m_protectRW)284
+ {285
+ m_protectRW = false;286
+ if (m_writeSem)287
+ {288
+ m_writeSem->release();289
+290
+ delete m_writeSem;291
+ m_writeSem = NULL;292
+ }293
+294
+ if (m_readSem)295
+ {296
+ m_readSem->release();297
+298
+ delete m_readSem;299
+ m_readSem = NULL;300
+ }301
+ }302
+303
+ }304
+ }305
+306
+ ///< data read307
+ bool RingMem::readNext(void* dst, fnRWSharedData callback)308
+ {309
+ if (!m_initialized || !callback || !dst)310
+ {311
+ return false;312
+ }313
+314
+ if (m_protectRW)315
+ {316
+ if (!m_readSem->take())317
+ {318
+ return false;319
+ }320
+ }321
+322
+ int32_t index = ATOMIC_ADD(&m_shrMem->m_read, 1) % m_itemCnt;323
+ (*callback)(dst, reinterpret_cast<uint8_t *>(m_dataPool) + index * m_itemSize, m_itemSize);324
+325
+ if (m_protectRW)326
+ {327
+ m_writeSem->give(1);328
+ }329
+330
+ return true;331
+ }332
+ ///< data write333
+ bool RingMem::writeData(void *data, fnRWSharedData callback)334
+ {335
+ if (!m_initialized || !data || !callback)336
+ {337
+ return false;338
+ }339
+340
+ if (m_protectRW)341
+ {342
+ if (!m_writeSem->take())343
+ {344
+ return false;345
+ }346
+ }347
+348
+ int32_t index = ATOMIC_ADD(&m_shrMem->m_write, 1) % m_itemCnt;349
+ (*callback)(reinterpret_cast<uint8_t *>(m_dataPool) + index * m_itemSize, data, m_itemSize);350
+351
+ if (m_protectRW)352
+ {353
+ m_readSem->give(1);354
+ }355
+356
+ return true;357
+ }358
+}359
x265_3.6.tar.gz/source/common/ringmem.h
Added
92
1
2
+/*****************************************************************************3
+ * Copyright (C) 2013-2017 MulticoreWare, Inc4
+ *5
+ * Authors: liwei <liwei@multicorewareinc.com>6
+ *7
+ * This program is free software; you can redistribute it and/or modify8
+ * it under the terms of the GNU General Public License as published by9
+ * the Free Software Foundation; either version 2 of the License, or10
+ * (at your option) any later version.11
+ *12
+ * This program is distributed in the hope that it will be useful,13
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of14
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+ * GNU General Public License for more details.16
+ *17
+ * You should have received a copy of the GNU General Public License18
+ * along with this program; if not, write to the Free Software19
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+ *21
+ * This program is also available under a commercial proprietary license.22
+ * For more information, contact us at license @ x265.com23
+ *****************************************************************************/24
+25
+#ifndef X265_RINGMEM_H26
+#define X265_RINGMEM_H27
+28
+#include "common.h"29
+#include "threading.h"30
+31
+#if _MSC_VER32
+#define snprintf _snprintf33
+#define strdup _strdup34
+#endif35
+36
+namespace X265_NS {37
+38
+#define MAX_SHR_NAME_LEN 25639
+40
+ class RingMem {41
+ public:42
+ RingMem();43
+ ~RingMem();44
+45
+ bool skipRead(int32_t cnt);46
+47
+ bool skipWrite(int32_t cnt);48
+49
+ ///< initialize50
+ ///< protectRW: if use the semaphore the protect the write and read operation.51
+ bool init(int32_t itemSize, int32_t itemCnt, const char *name, bool protectRW = false);52
+ ///< finalize53
+ void release();54
+55
+ typedef void(*fnRWSharedData)(void *dst, void *src, int32_t size);56
+57
+ ///< data read58
+ bool readNext(void* dst, fnRWSharedData callback);59
+ ///< data write60
+ bool writeData(void *data, fnRWSharedData callback);61
+62
+ private: 63
+ bool m_initialized;64
+ bool m_protectRW;65
+66
+ int32_t m_itemSize;67
+ int32_t m_itemCnt;68
+ ///< data pool69
+ void *m_dataPool;70
+ typedef struct {71
+ ///< index to write72
+ int32_t m_write;73
+ ///< index to read74
+ int32_t m_read;75
+ 76
+ }ShrMemCtrl;77
+78
+ ShrMemCtrl *m_shrMem;79
+#ifdef _WIN3280
+ void *m_handle;81
+#else // _WIN3282
+ char *m_filepath;83
+#endif // _WIN3284
+85
+ ///< Semaphores86
+ NamedSemaphore *m_writeSem;87
+ NamedSemaphore *m_readSem;88
+ };89
+};90
+91
+#endif // ifndef X265_RINGMEM_H92
x265_3.5.tar.gz/source/common/slice.h -> x265_3.6.tar.gz/source/common/slice.h
Changed
35
1
2
HRDInfo hrdParameters;3
ProfileTierLevel ptl;4
uint32_t maxTempSubLayers;5
- uint32_t numReorderPics;6
- uint32_t maxDecPicBuffering;7
- uint32_t maxLatencyIncrease;8
+ uint32_t numReorderPicsMAX_T_LAYERS;9
+ uint32_t maxDecPicBufferingMAX_T_LAYERS;10
+ uint32_t maxLatencyIncreaseMAX_T_LAYERS;11
};12
13
struct Window14
15
uint32_t maxAMPDepth;16
17
uint32_t maxTempSubLayers; // max number of Temporal Sub layers18
- uint32_t maxDecPicBuffering; // these are dups of VPS values19
- uint32_t maxLatencyIncrease;20
- int numReorderPics;21
+ uint32_t maxDecPicBufferingMAX_T_LAYERS; // these are dups of VPS values22
+ uint32_t maxLatencyIncreaseMAX_T_LAYERS;23
+ int numReorderPicsMAX_T_LAYERS;24
25
RPS spsrpsMAX_NUM_SHORT_TERM_RPS;26
int spsrpsNum;27
28
int m_iNumRPSInSPS;29
const x265_param *m_param;30
int m_fieldNum;31
+ Frame* m_mcstfRefFrameList2MAX_MCSTF_TEMPORAL_WINDOW_LENGTH;32
33
Slice()34
{35
x265_3.6.tar.gz/source/common/temporalfilter.cpp
Added
1019
1
2
+/*****************************************************************************3
+* Copyright (C) 2013-2021 MulticoreWare, Inc4
+*5
+ * Authors: Ashok Kumar Mishra <ashok@multicorewareinc.com>6
+ *7
+* This program is free software; you can redistribute it and/or modify8
+* it under the terms of the GNU General Public License as published by9
+* the Free Software Foundation; either version 2 of the License, or10
+* (at your option) any later version.11
+*12
+* This program is distributed in the hope that it will be useful,13
+* but WITHOUT ANY WARRANTY; without even the implied warranty of14
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+* GNU General Public License for more details.16
+*17
+* You should have received a copy of the GNU General Public License18
+* along with this program; if not, write to the Free Software19
+* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+*21
+* This program is also available under a commercial proprietary license.22
+* For more information, contact us at license @ x265.com.23
+*****************************************************************************/24
+#include "common.h"25
+#include "temporalfilter.h"26
+#include "primitives.h"27
+28
+#include "frame.h"29
+#include "slice.h"30
+#include "framedata.h"31
+#include "analysis.h"32
+33
+using namespace X265_NS;34
+35
+void OrigPicBuffer::addPicture(Frame* inFrame)36
+{37
+ m_mcstfPicList.pushFrontMCSTF(*inFrame);38
+}39
+40
+void OrigPicBuffer::addEncPicture(Frame* inFrame)41
+{42
+ m_mcstfOrigPicFreeList.pushFrontMCSTF(*inFrame);43
+}44
+45
+void OrigPicBuffer::addEncPictureToPicList(Frame* inFrame)46
+{47
+ m_mcstfOrigPicList.pushFrontMCSTF(*inFrame);48
+}49
+50
+OrigPicBuffer::~OrigPicBuffer()51
+{52
+ while (!m_mcstfOrigPicList.empty())53
+ {54
+ Frame* curFrame = m_mcstfOrigPicList.popBackMCSTF();55
+ curFrame->destroy();56
+ delete curFrame;57
+ }58
+59
+ while (!m_mcstfOrigPicFreeList.empty())60
+ {61
+ Frame* curFrame = m_mcstfOrigPicFreeList.popBackMCSTF();62
+ curFrame->destroy();63
+ delete curFrame;64
+ }65
+}66
+67
+void OrigPicBuffer::setOrigPicList(Frame* inFrame, int frameCnt)68
+{69
+ Slice* slice = inFrame->m_encData->m_slice;70
+ uint8_t j = 0;71
+ for (int iterPOC = (inFrame->m_poc - inFrame->m_mcstf->m_range);72
+ iterPOC <= (inFrame->m_poc + inFrame->m_mcstf->m_range); iterPOC++)73
+ {74
+ if (iterPOC != inFrame->m_poc)75
+ {76
+ if (iterPOC < 0)77
+ continue;78
+ if (iterPOC >= frameCnt)79
+ break;80
+81
+ Frame *iterFrame = m_mcstfPicList.getPOCMCSTF(iterPOC);82
+ X265_CHECK(iterFrame, "Reference frame not found in OPB");83
+ if (iterFrame != NULL)84
+ {85
+ slice->m_mcstfRefFrameList1j = iterFrame;86
+ iterFrame->m_refPicCnt1--;87
+ }88
+89
+ iterFrame = m_mcstfOrigPicList.getPOCMCSTF(iterPOC);90
+ if (iterFrame != NULL)91
+ {92
+93
+ slice->m_mcstfRefFrameList1j = iterFrame;94
+95
+ iterFrame->m_refPicCnt1--;96
+ Frame *cFrame = m_mcstfOrigPicList.getPOCMCSTF(inFrame->m_poc);97
+ X265_CHECK(cFrame, "Reference frame not found in encoded OPB");98
+ cFrame->m_refPicCnt1--;99
+ }100
+ j++;101
+ }102
+ }103
+}104
+105
+void OrigPicBuffer::recycleOrigPicList()106
+{107
+ Frame *iterFrame = m_mcstfPicList.first();108
+109
+ while (iterFrame)110
+ {111
+ Frame *curFrame = iterFrame;112
+ iterFrame = iterFrame->m_nextMCSTF;113
+ if (!curFrame->m_refPicCnt1)114
+ {115
+ m_mcstfPicList.removeMCSTF(*curFrame);116
+ iterFrame = m_mcstfPicList.first();117
+ }118
+ }119
+120
+ iterFrame = m_mcstfOrigPicList.first();121
+122
+ while (iterFrame)123
+ {124
+ Frame *curFrame = iterFrame;125
+ iterFrame = iterFrame->m_nextMCSTF;126
+ if (!curFrame->m_refPicCnt1)127
+ {128
+ m_mcstfOrigPicList.removeMCSTF(*curFrame);129
+ *curFrame->m_isSubSampled = false;130
+ m_mcstfOrigPicFreeList.pushFrontMCSTF(*curFrame);131
+ iterFrame = m_mcstfOrigPicList.first();132
+ }133
+ }134
+}135
+136
+void OrigPicBuffer::addPictureToFreelist(Frame* inFrame)137
+{138
+ m_mcstfOrigPicFreeList.pushBack(*inFrame);139
+}140
+141
+TemporalFilter::TemporalFilter()142
+{143
+ m_sourceWidth = 0;144
+ m_sourceHeight = 0,145
+ m_QP = 0;146
+ m_sliceTypeConfig = 3;147
+ m_numRef = 0;148
+ m_useSADinME = 1;149
+150
+ m_range = 2;151
+ m_chromaFactor = 0.55;152
+ m_sigmaMultiplier = 9.0;153
+ m_sigmaZeroPoint = 10.0;154
+ m_motionVectorFactor = 16;155
+}156
+157
+void TemporalFilter::init(const x265_param* param)158
+{159
+ m_param = param;160
+ m_bitDepth = param->internalBitDepth;161
+ m_sourceWidth = param->sourceWidth;162
+ m_sourceHeight = param->sourceHeight;163
+ m_internalCsp = param->internalCsp;164
+ m_numComponents = (m_internalCsp != X265_CSP_I400) ? MAX_NUM_COMPONENT : 1;165
+166
+ m_metld = new MotionEstimatorTLD;167
+168
+ predPUYuv.create(FENC_STRIDE, X265_CSP_I400);169
+}170
+171
+int TemporalFilter::createRefPicInfo(TemporalFilterRefPicInfo* refFrame, x265_param* param)172
+{173
+ CHECKED_MALLOC_ZERO(refFrame->mvs, MV, sizeof(MV)* ((m_sourceWidth ) / 4) * ((m_sourceHeight ) / 4));174
+ refFrame->mvsStride = m_sourceWidth / 4;175
+ CHECKED_MALLOC_ZERO(refFrame->mvs0, MV, sizeof(MV)* ((m_sourceWidth ) / 16) * ((m_sourceHeight ) / 16));176
+ refFrame->mvsStride0 = m_sourceWidth / 16;177
+ CHECKED_MALLOC_ZERO(refFrame->mvs1, MV, sizeof(MV)* ((m_sourceWidth ) / 16) * ((m_sourceHeight ) / 16));178
+ refFrame->mvsStride1 = m_sourceWidth / 16;179
+ CHECKED_MALLOC_ZERO(refFrame->mvs2, MV, sizeof(MV)* ((m_sourceWidth ) / 16)*((m_sourceHeight ) / 16));180
+ refFrame->mvsStride2 = m_sourceWidth / 16;181
+182
+ CHECKED_MALLOC_ZERO(refFrame->noise, int, sizeof(int) * ((m_sourceWidth) / 4) * ((m_sourceHeight) / 4));183
+ CHECKED_MALLOC_ZERO(refFrame->error, int, sizeof(int) * ((m_sourceWidth) / 4) * ((m_sourceHeight) / 4));184
+185
+ refFrame->slicetype = X265_TYPE_AUTO;186
+187
+ refFrame->compensatedPic = new PicYuv;188
+ refFrame->compensatedPic->create(param, true);189
+190
+ return 1;191
+fail:192
+ return 0;193
+}194
+195
+int TemporalFilter::motionErrorLumaSAD(196
+ PicYuv *orig,197
+ PicYuv *buffer,198
+ int x,199
+ int y,200
+ int dx,201
+ int dy,202
+ int bs,203
+ int besterror)204
+{205
+206
+ pixel* origOrigin = orig->m_picOrg0;207
+ intptr_t origStride = orig->m_stride;208
+ pixel *buffOrigin = buffer->m_picOrg0;209
+ intptr_t buffStride = buffer->m_stride;210
+ int error = 0;// dx * 10 + dy * 10;211
+ if (((dx | dy) & 0xF) == 0)212
+ {213
+ dx /= m_motionVectorFactor;214
+ dy /= m_motionVectorFactor;215
+216
+ const pixel* bufferRowStart = buffOrigin + (y + dy) * buffStride + (x + dx);217
+#if 0218
+ const pixel* origRowStart = origOrigin + y *origStride + x;219
+220
+ for (int y1 = 0; y1 < bs; y1++)221
+ {222
+ for (int x1 = 0; x1 < bs; x1++)223
+ {224
+ int diff = origRowStartx1 - bufferRowStartx1;225
+ error += abs(diff);226
+ }227
+228
+ origRowStart += origStride;229
+ bufferRowStart += buffStride;230
+ }231
+#else232
+ int partEnum = partitionFromSizes(bs, bs);233
+ /* copy PU block into cache */234
+ primitives.pupartEnum.copy_pp(predPUYuv.m_buf0, FENC_STRIDE, bufferRowStart, buffStride);235
+236
+ error = m_metld->me.bufSAD(predPUYuv.m_buf0, FENC_STRIDE);237
+#endif238
+ if (error > besterror)239
+ {240
+ return error;241
+ }242
+ }243
+ else244
+ {245
+ const int *xFilter = s_interpolationFilterdx & 0xF;246
+ const int *yFilter = s_interpolationFilterdy & 0xF;247
+ int tempArray64 + 864;248
+249
+ int iSum, iBase;250
+ for (int y1 = 1; y1 < bs + 7; y1++)251
+ {252
+ const int yOffset = y + y1 + (dy >> 4) - 3;253
+ const pixel *sourceRow = buffOrigin + (yOffset)*buffStride + 0;254
+ for (int x1 = 0; x1 < bs; x1++)255
+ {256
+ iSum = 0;257
+ iBase = x + x1 + (dx >> 4) - 3;258
+ const pixel *rowStart = sourceRow + iBase;259
+260
+ iSum += xFilter1 * rowStart1;261
+ iSum += xFilter2 * rowStart2;262
+ iSum += xFilter3 * rowStart3;263
+ iSum += xFilter4 * rowStart4;264
+ iSum += xFilter5 * rowStart5;265
+ iSum += xFilter6 * rowStart6;266
+267
+ tempArrayy1x1 = iSum;268
+ }269
+ }270
+271
+ const pixel maxSampleValue = (1 << m_bitDepth) - 1;272
+ for (int y1 = 0; y1 < bs; y1++)273
+ {274
+ const pixel *origRow = origOrigin + (y + y1)*origStride + 0;275
+ for (int x1 = 0; x1 < bs; x1++)276
+ {277
+ iSum = 0;278
+ iSum += yFilter1 * tempArrayy1 + 1x1;279
+ iSum += yFilter2 * tempArrayy1 + 2x1;280
+ iSum += yFilter3 * tempArrayy1 + 3x1;281
+ iSum += yFilter4 * tempArrayy1 + 4x1;282
+ iSum += yFilter5 * tempArrayy1 + 5x1;283
+ iSum += yFilter6 * tempArrayy1 + 6x1;284
+285
+ iSum = (iSum + (1 << 11)) >> 12;286
+ iSum = iSum < 0 ? 0 : (iSum > maxSampleValue ? maxSampleValue : iSum);287
+288
+ error += abs(iSum - origRowx + x1);289
+ }290
+ if (error > besterror)291
+ {292
+ return error;293
+ }294
+ }295
+ }296
+ return error;297
+}298
+299
+int TemporalFilter::motionErrorLumaSSD(300
+ PicYuv *orig,301
+ PicYuv *buffer,302
+ int x,303
+ int y,304
+ int dx,305
+ int dy,306
+ int bs,307
+ int besterror)308
+{309
+310
+ pixel* origOrigin = orig->m_picOrg0;311
+ intptr_t origStride = orig->m_stride;312
+ pixel *buffOrigin = buffer->m_picOrg0;313
+ intptr_t buffStride = buffer->m_stride;314
+ int error = 0;// dx * 10 + dy * 10;315
+ if (((dx | dy) & 0xF) == 0)316
+ {317
+ dx /= m_motionVectorFactor;318
+ dy /= m_motionVectorFactor;319
+320
+ const pixel* bufferRowStart = buffOrigin + (y + dy) * buffStride + (x + dx);321
+#if 0322
+ const pixel* origRowStart = origOrigin + y * origStride + x;323
+324
+ for (int y1 = 0; y1 < bs; y1++)325
+ {326
+ for (int x1 = 0; x1 < bs; x1++)327
+ {328
+ int diff = origRowStartx1 - bufferRowStartx1;329
+ error += diff * diff;330
+ }331
+332
+ origRowStart += origStride;333
+ bufferRowStart += buffStride;334
+ }335
+#else336
+ int partEnum = partitionFromSizes(bs, bs);337
+ /* copy PU block into cache */338
+ primitives.pupartEnum.copy_pp(predPUYuv.m_buf0, FENC_STRIDE, bufferRowStart, buffStride);339
+340
+ error = (int)primitives.cupartEnum.sse_pp(m_metld->me.fencPUYuv.m_buf0, FENC_STRIDE, predPUYuv.m_buf0, FENC_STRIDE);341
+342
+#endif343
+ if (error > besterror)344
+ {345
+ return error;346
+ }347
+ }348
+ else349
+ {350
+ const int *xFilter = s_interpolationFilterdx & 0xF;351
+ const int *yFilter = s_interpolationFilterdy & 0xF;352
+ int tempArray64 + 864;353
+354
+ int iSum, iBase;355
+ for (int y1 = 1; y1 < bs + 7; y1++)356
+ {357
+ const int yOffset = y + y1 + (dy >> 4) - 3;358
+ const pixel *sourceRow = buffOrigin + (yOffset)*buffStride + 0;359
+ for (int x1 = 0; x1 < bs; x1++)360
+ {361
+ iSum = 0;362
+ iBase = x + x1 + (dx >> 4) - 3;363
+ const pixel *rowStart = sourceRow + iBase;364
+365
+ iSum += xFilter1 * rowStart1;366
+ iSum += xFilter2 * rowStart2;367
+ iSum += xFilter3 * rowStart3;368
+ iSum += xFilter4 * rowStart4;369
+ iSum += xFilter5 * rowStart5;370
+ iSum += xFilter6 * rowStart6;371
+372
+ tempArrayy1x1 = iSum;373
+ }374
+ }375
+376
+ const pixel maxSampleValue = (1 << m_bitDepth) - 1;377
+ for (int y1 = 0; y1 < bs; y1++)378
+ {379
+ const pixel *origRow = origOrigin + (y + y1)*origStride + 0;380
+ for (int x1 = 0; x1 < bs; x1++)381
+ {382
+ iSum = 0;383
+ iSum += yFilter1 * tempArrayy1 + 1x1;384
+ iSum += yFilter2 * tempArrayy1 + 2x1;385
+ iSum += yFilter3 * tempArrayy1 + 3x1;386
+ iSum += yFilter4 * tempArrayy1 + 4x1;387
+ iSum += yFilter5 * tempArrayy1 + 5x1;388
+ iSum += yFilter6 * tempArrayy1 + 6x1;389
+390
+ iSum = (iSum + (1 << 11)) >> 12;391
+ iSum = iSum < 0 ? 0 : (iSum > maxSampleValue ? maxSampleValue : iSum);392
+393
+ error += (iSum - origRowx + x1) * (iSum - origRowx + x1);394
+ }395
+ if (error > besterror)396
+ {397
+ return error;398
+ }399
+ }400
+ }401
+ return error;402
+}403
+404
+void TemporalFilter::applyMotion(MV *mvs, uint32_t mvsStride, PicYuv *input, PicYuv *output)405
+{406
+ static const int lumaBlockSize = 8;407
+ int srcStride = 0;408
+ int dstStride = 0;409
+ int csx = 0, csy = 0;410
+ for (int c = 0; c < m_numComponents; c++)411
+ {412
+ const pixel maxValue = (1 << X265_DEPTH) - 1;413
+414
+ const pixel *pSrcImage = input->m_picOrgc;415
+ pixel *pDstImage = output->m_picOrgc;416
+417
+ if (!c)418
+ {419
+ srcStride = (int)input->m_stride;420
+ dstStride = (int)output->m_stride;421
+ }422
+ else423
+ {424
+ srcStride = (int)input->m_strideC;425
+ dstStride = (int)output->m_strideC;426
+ csx = CHROMA_H_SHIFT(m_internalCsp);427
+ csy = CHROMA_V_SHIFT(m_internalCsp);428
+ }429
+ const int blockSizeX = lumaBlockSize >> csx;430
+ const int blockSizeY = lumaBlockSize >> csy;431
+ const int height = input->m_picHeight >> csy;432
+ const int width = input->m_picWidth >> csx;433
+434
+ for (int y = 0, blockNumY = 0; y + blockSizeY <= height; y += blockSizeY, blockNumY++)435
+ {436
+ for (int x = 0, blockNumX = 0; x + blockSizeX <= width; x += blockSizeX, blockNumX++)437
+ {438
+ int mvIdx = blockNumY * mvsStride + blockNumX;439
+ const MV &mv = mvsmvIdx;440
+ const int dx = mv.x >> csx;441
+ const int dy = mv.y >> csy;442
+ const int xInt = mv.x >> (4 + csx);443
+ const int yInt = mv.y >> (4 + csy);444
+445
+ const int *xFilter = s_interpolationFilterdx & 0xf;446
+ const int *yFilter = s_interpolationFilterdy & 0xf; // will add 6 bit.447
+ const int numFilterTaps = 7;448
+ const int centreTapOffset = 3;449
+450
+ int tempArraylumaBlockSize + numFilterTapslumaBlockSize;451
+452
+ for (int by = 1; by < blockSizeY + numFilterTaps; by++)453
+ {454
+ const int yOffset = y + by + yInt - centreTapOffset;455
+ const pixel *sourceRow = pSrcImage + yOffset * srcStride;456
+ for (int bx = 0; bx < blockSizeX; bx++)457
+ {458
+ int iBase = x + bx + xInt - centreTapOffset;459
+ const pixel *rowStart = sourceRow + iBase;460
+461
+ int iSum = 0;462
+ iSum += xFilter1 * rowStart1;463
+ iSum += xFilter2 * rowStart2;464
+ iSum += xFilter3 * rowStart3;465
+ iSum += xFilter4 * rowStart4;466
+ iSum += xFilter5 * rowStart5;467
+ iSum += xFilter6 * rowStart6;468
+469
+ tempArraybybx = iSum;470
+ }471
+ }472
+473
+ pixel *pDstRow = pDstImage + y * dstStride;474
+ for (int by = 0; by < blockSizeY; by++, pDstRow += dstStride)475
+ {476
+ pixel *pDstPel = pDstRow + x;477
+ for (int bx = 0; bx < blockSizeX; bx++, pDstPel++)478
+ {479
+ int iSum = 0;480
+481
+ iSum += yFilter1 * tempArrayby + 1bx;482
+ iSum += yFilter2 * tempArrayby + 2bx;483
+ iSum += yFilter3 * tempArrayby + 3bx;484
+ iSum += yFilter4 * tempArrayby + 4bx;485
+ iSum += yFilter5 * tempArrayby + 5bx;486
+ iSum += yFilter6 * tempArrayby + 6bx;487
+488
+ iSum = (iSum + (1 << 11)) >> 12;489
+ iSum = iSum < 0 ? 0 : (iSum > maxValue ? maxValue : iSum);490
+ *pDstPel = (pixel)iSum;491
+ }492
+ }493
+ }494
+ }495
+ }496
+}497
+498
+void TemporalFilter::bilateralFilter(Frame* frame,499
+ TemporalFilterRefPicInfo* m_mcstfRefList,500
+ double overallStrength)501
+{502
+503
+ const int numRefs = frame->m_mcstf->m_numRef;504
+505
+ for (int i = 0; i < numRefs; i++)506
+ {507
+ TemporalFilterRefPicInfo *ref = &m_mcstfRefListi;508
+ applyMotion(m_mcstfRefListi.mvs, m_mcstfRefListi.mvsStride, m_mcstfRefListi.picBuffer, ref->compensatedPic);509
+ }510
+511
+ int refStrengthRow = 2;512
+ if (numRefs == m_range * 2)513
+ {514
+ refStrengthRow = 0;515
+ }516
+ else if (numRefs == m_range)517
+ {518
+ refStrengthRow = 1;519
+ }520
+521
+ const double lumaSigmaSq = (m_QP - m_sigmaZeroPoint) * (m_QP - m_sigmaZeroPoint) * m_sigmaMultiplier;522
+ const double chromaSigmaSq = 30 * 30;523
+524
+ PicYuv* orgPic = frame->m_fencPic;525
+526
+ for (int c = 0; c < m_numComponents; c++)527
+ {528
+ int height, width;529
+ pixel *srcPelRow = NULL;530
+ intptr_t srcStride, correctedPicsStride = 0;531
+532
+ if (!c)533
+ {534
+ height = orgPic->m_picHeight;535
+ width = orgPic->m_picWidth;536
+ srcPelRow = orgPic->m_picOrgc;537
+ srcStride = orgPic->m_stride;538
+ }539
+ else540
+ {541
+ int csx = CHROMA_H_SHIFT(m_internalCsp);542
+ int csy = CHROMA_V_SHIFT(m_internalCsp);543
+544
+ height = orgPic->m_picHeight >> csy;545
+ width = orgPic->m_picWidth >> csx;546
+ srcPelRow = orgPic->m_picOrgc;547
+ srcStride = (int)orgPic->m_strideC;548
+ }549
+550
+ const double sigmaSq = (!c) ? lumaSigmaSq : chromaSigmaSq;551
+ const double weightScaling = overallStrength * ( (!c) ? 0.4 : m_chromaFactor);552
+553
+ const double maxSampleValue = (1 << m_bitDepth) - 1;554
+ const double bitDepthDiffWeighting = 1024.0 / (maxSampleValue + 1);555
+556
+ const int blkSize = (!c) ? 8 : 4;557
+558
+ for (int y = 0; y < height; y++, srcPelRow += srcStride)559
+ {560
+ pixel *srcPel = srcPelRow;561
+562
+ for (int x = 0; x < width; x++, srcPel++)563
+ {564
+ const int orgVal = (int)*srcPel;565
+ double temporalWeightSum = 1.0;566
+ double newVal = (double)orgVal;567
+568
+ if ((y % blkSize == 0) && (x % blkSize == 0))569
+ {570
+ for (int i = 0; i < numRefs; i++)571
+ {572
+ TemporalFilterRefPicInfo *refPicInfo = &m_mcstfRefListi;573
+574
+ if (!c)575
+ correctedPicsStride = refPicInfo->compensatedPic->m_stride;576
+ else577
+ correctedPicsStride = refPicInfo->compensatedPic->m_strideC;578
+579
+ double variance = 0, diffsum = 0;580
+ for (int y1 = 0; y1 < blkSize - 1; y1++)581
+ {582
+ for (int x1 = 0; x1 < blkSize - 1; x1++)583
+ {584
+ int pix = *(srcPel + x1);585
+ int pixR = *(srcPel + x1 + 1);586
+ int pixD = *(srcPel + x1 + srcStride);587
+588
+ int ref = *(refPicInfo->compensatedPic->m_picOrgc + ((y + y1) * correctedPicsStride + x + x1));589
+ int refR = *(refPicInfo->compensatedPic->m_picOrgc + ((y + y1) * correctedPicsStride + x + x1 + 1));590
+ int refD = *(refPicInfo->compensatedPic->m_picOrgc + ((y + y1 + 1) * correctedPicsStride + x + x1));591
+592
+ int diff = pix - ref;593
+ int diffR = pixR - refR;594
+ int diffD = pixD - refD;595
+596
+ variance += diff * diff;597
+ diffsum += (diffR - diff) * (diffR - diff);598
+ diffsum += (diffD - diff) * (diffD - diff);599
+ }600
+ }601
+602
+ refPicInfo->noise(y / blkSize) * refPicInfo->mvsStride + (x / blkSize) = (int)round((300 * variance + 50) / (10 * diffsum + 50));603
+ }604
+ }605
+606
+ double minError = 9999999;607
+ for (int i = 0; i < numRefs; i++)608
+ {609
+ TemporalFilterRefPicInfo *refPicInfo = &m_mcstfRefListi;610
+ minError = X265_MIN(minError, (double)refPicInfo->error(y / blkSize) * refPicInfo->mvsStride + (x / blkSize));611
+ }612
+613
+ for (int i = 0; i < numRefs; i++)614
+ {615
+ TemporalFilterRefPicInfo *refPicInfo = &m_mcstfRefListi;616
+617
+ const int error = refPicInfo->error(y / blkSize) * refPicInfo->mvsStride + (x / blkSize);618
+ const int noise = refPicInfo->noise(y / blkSize) * refPicInfo->mvsStride + (x / blkSize);619
+620
+ const pixel *pCorrectedPelPtr = refPicInfo->compensatedPic->m_picOrgc + (y * correctedPicsStride + x);621
+ const int refVal = (int)*pCorrectedPelPtr;622
+ double diff = (double)(refVal - orgVal);623
+ diff *= bitDepthDiffWeighting;624
+ double diffSq = diff * diff;625
+626
+ const int index = X265_MIN(3, std::abs(refPicInfo->origOffset) - 1);627
+ double ww = 1, sw = 1;628
+ ww *= (noise < 25) ? 1 : 1.2;629
+ sw *= (noise < 25) ? 1.3 : 0.8;630
+ ww *= (error < 50) ? 1.2 : ((error > 100) ? 0.8 : 1);631
+ sw *= (error < 50) ? 1.3 : 1;632
+ ww *= ((minError + 1) / (error + 1));633
+ const double weight = weightScaling * s_refStrengthsrefStrengthRowindex * ww * exp(-diffSq / (2 * sw * sigmaSq));634
+635
+ newVal += weight * refVal;636
+ temporalWeightSum += weight;637
+ }638
+ newVal /= temporalWeightSum;639
+ double sampleVal = round(newVal);640
+ sampleVal = (sampleVal < 0 ? 0 : (sampleVal > maxSampleValue ? maxSampleValue : sampleVal));641
+ *srcPel = (pixel)sampleVal;642
+ }643
+ }644
+ }645
+}646
+647
+void TemporalFilter::motionEstimationLuma(MV *mvs, uint32_t mvStride, PicYuv *orig, PicYuv *buffer, int blockSize,648
+ MV *previous, uint32_t prevMvStride, int factor)649
+{650
+651
+ int range = 5;652
+653
+654
+ const int stepSize = blockSize;655
+656
+ const int origWidth = orig->m_picWidth;657
+ const int origHeight = orig->m_picHeight;658
+659
+ int error;660
+661
+ for (int blockY = 0; blockY + blockSize <= origHeight; blockY += stepSize)662
+ {663
+ for (int blockX = 0; blockX + blockSize <= origWidth; blockX += stepSize)664
+ {665
+ const intptr_t pelOffset = blockY * orig->m_stride + blockX;666
+ m_metld->me.setSourcePU(orig->m_picOrg0, orig->m_stride, pelOffset, blockSize, blockSize, X265_HEX_SEARCH, 1);667
+668
+669
+ MV best(0, 0);670
+ int leastError = INT_MAX;671
+672
+ if (previous == NULL)673
+ {674
+ range = 8;675
+ }676
+ else677
+ {678
+679
+ for (int py = -1; py <= 1; py++)680
+ {681
+ int testy = blockY / (2 * blockSize) + py;682
+683
+ for (int px = -1; px <= 1; px++)684
+ {685
+686
+ int testx = blockX / (2 * blockSize) + px;687
+ if ((testx >= 0) && (testx < origWidth / (2 * blockSize)) && (testy >= 0) && (testy < origHeight / (2 * blockSize)))688
+ {689
+ int mvIdx = testy * prevMvStride + testx;690
+ MV old = previousmvIdx;691
+692
+ if (m_useSADinME)693
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, old.x * factor, old.y * factor, blockSize, leastError);694
+ else695
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, old.x * factor, old.y * factor, blockSize, leastError);696
+697
+ if (error < leastError)698
+ {699
+ best.set(old.x * factor, old.y * factor);700
+ leastError = error;701
+ }702
+ }703
+ }704
+ }705
+706
+ if (m_useSADinME)707
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, 0, 0, blockSize, leastError);708
+ else709
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, 0, 0, blockSize, leastError);710
+711
+ if (error < leastError)712
+ {713
+ best.set(0, 0);714
+ leastError = error;715
+ }716
+717
+ }718
+719
+ MV prevBest = best;720
+ for (int y2 = prevBest.y / m_motionVectorFactor - range; y2 <= prevBest.y / m_motionVectorFactor + range; y2++)721
+ {722
+ for (int x2 = prevBest.x / m_motionVectorFactor - range; x2 <= prevBest.x / m_motionVectorFactor + range; x2++)723
+ {724
+ if (m_useSADinME)725
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, blockSize, leastError);726
+ else727
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, blockSize, leastError);728
+ if (error < leastError)729
+ {730
+ best.set(x2 * m_motionVectorFactor, y2 * m_motionVectorFactor);731
+ leastError = error;732
+ }733
+ }734
+ }735
+736
+ if (blockY > 0)737
+ {738
+ int idx = ((blockY - stepSize) / stepSize) * mvStride + (blockX / stepSize);739
+ MV aboveMV = mvsidx;740
+741
+ if (m_useSADinME)742
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, aboveMV.x, aboveMV.y, blockSize, leastError);743
+ else744
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, aboveMV.x, aboveMV.y, blockSize, leastError);745
+746
+ if (error < leastError)747
+ {748
+ best.set(aboveMV.x, aboveMV.y);749
+ leastError = error;750
+ }751
+ }752
+753
+ if (blockX > 0)754
+ {755
+ int idx = ((blockY / stepSize) * mvStride + (blockX - stepSize) / stepSize);756
+ MV leftMV = mvsidx;757
+758
+ if (m_useSADinME)759
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, leftMV.x, leftMV.y, blockSize, leastError);760
+ else761
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, leftMV.x, leftMV.y, blockSize, leastError);762
+763
+ if (error < leastError)764
+ {765
+ best.set(leftMV.x, leftMV.y);766
+ leastError = error;767
+ }768
+ }769
+770
+ // calculate average771
+ double avg = 0.0;772
+ for (int x1 = 0; x1 < blockSize; x1++)773
+ {774
+ for (int y1 = 0; y1 < blockSize; y1++)775
+ {776
+ avg = avg + *(orig->m_picOrg0 + (blockX + x1 + orig->m_stride * (blockY + y1)));777
+ }778
+ }779
+ avg = avg / (blockSize * blockSize);780
+781
+ // calculate variance782
+ double variance = 0;783
+ for (int x1 = 0; x1 < blockSize; x1++)784
+ {785
+ for (int y1 = 0; y1 < blockSize; y1++)786
+ {787
+ int pix = *(orig->m_picOrg0 + (blockX + x1 + orig->m_stride * (blockY + y1)));788
+ variance = variance + (pix - avg) * (pix - avg);789
+ }790
+ }791
+792
+ leastError = (int)(20 * ((leastError + 5.0) / (variance + 5.0)) + (leastError / (blockSize * blockSize)) / 50);793
+794
+ int mvIdx = (blockY / stepSize) * mvStride + (blockX / stepSize);795
+ mvsmvIdx = best;796
+ }797
+ }798
+}799
+800
+801
+void TemporalFilter::motionEstimationLumaDoubleRes(MV *mvs, uint32_t mvStride, PicYuv *orig, PicYuv *buffer, int blockSize,802
+ MV *previous, uint32_t prevMvStride, int factor, int* minError)803
+{804
+805
+ int range = 0;806
+807
+808
+ const int stepSize = blockSize;809
+810
+ const int origWidth = orig->m_picWidth;811
+ const int origHeight = orig->m_picHeight;812
+813
+ int error;814
+815
+ for (int blockY = 0; blockY + blockSize <= origHeight; blockY += stepSize)816
+ {817
+ for (int blockX = 0; blockX + blockSize <= origWidth; blockX += stepSize)818
+ {819
+820
+ const intptr_t pelOffset = blockY * orig->m_stride + blockX;821
+ m_metld->me.setSourcePU(orig->m_picOrg0, orig->m_stride, pelOffset, blockSize, blockSize, X265_HEX_SEARCH, 1);822
+823
+ MV best(0, 0);824
+ int leastError = INT_MAX;825
+826
+ if (previous == NULL)827
+ {828
+ range = 8;829
+ }830
+ else831
+ {832
+833
+ for (int py = -1; py <= 1; py++)834
+ {835
+ int testy = blockY / (2 * blockSize) + py;836
+837
+ for (int px = -1; px <= 1; px++)838
+ {839
+840
+ int testx = blockX / (2 * blockSize) + px;841
+ if ((testx >= 0) && (testx < origWidth / (2 * blockSize)) && (testy >= 0) && (testy < origHeight / (2 * blockSize)))842
+ {843
+ int mvIdx = testy * prevMvStride + testx;844
+ MV old = previousmvIdx;845
+846
+ if (m_useSADinME)847
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, old.x * factor, old.y * factor, blockSize, leastError);848
+ else849
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, old.x * factor, old.y * factor, blockSize, leastError);850
+851
+ if (error < leastError)852
+ {853
+ best.set(old.x * factor, old.y * factor);854
+ leastError = error;855
+ }856
+ }857
+ }858
+ }859
+860
+ if (m_useSADinME)861
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, 0, 0, blockSize, leastError);862
+ else863
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, 0, 0, blockSize, leastError);864
+865
+ if (error < leastError)866
+ {867
+ best.set(0, 0);868
+ leastError = error;869
+ }870
+871
+ }872
+873
+ MV prevBest = best;874
+ for (int y2 = prevBest.y / m_motionVectorFactor - range; y2 <= prevBest.y / m_motionVectorFactor + range; y2++)875
+ {876
+ for (int x2 = prevBest.x / m_motionVectorFactor - range; x2 <= prevBest.x / m_motionVectorFactor + range; x2++)877
+ {878
+ if (m_useSADinME)879
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, blockSize, leastError);880
+ else881
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, x2 * m_motionVectorFactor, y2 * m_motionVectorFactor, blockSize, leastError);882
+883
+ if (error < leastError)884
+ {885
+ best.set(x2 * m_motionVectorFactor, y2 * m_motionVectorFactor);886
+ leastError = error;887
+ }888
+ }889
+ }890
+891
+ prevBest = best;892
+ int doubleRange = 3 * 4;893
+ for (int y2 = prevBest.y - doubleRange; y2 <= prevBest.y + doubleRange; y2 += 4)894
+ {895
+ for (int x2 = prevBest.x - doubleRange; x2 <= prevBest.x + doubleRange; x2 += 4)896
+ {897
+ if (m_useSADinME)898
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, x2, y2, blockSize, leastError);899
+ else900
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, x2, y2, blockSize, leastError);901
+902
+ if (error < leastError)903
+ {904
+ best.set(x2, y2);905
+ leastError = error;906
+ }907
+ }908
+ }909
+910
+ prevBest = best;911
+ doubleRange = 3;912
+ for (int y2 = prevBest.y - doubleRange; y2 <= prevBest.y + doubleRange; y2++)913
+ {914
+ for (int x2 = prevBest.x - doubleRange; x2 <= prevBest.x + doubleRange; x2++)915
+ {916
+ if (m_useSADinME)917
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, x2, y2, blockSize, leastError);918
+ else919
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, x2, y2, blockSize, leastError);920
+921
+ if (error < leastError)922
+ {923
+ best.set(x2, y2);924
+ leastError = error;925
+ }926
+ }927
+ }928
+929
+930
+ if (blockY > 0)931
+ {932
+ int idx = ((blockY - stepSize) / stepSize) * mvStride + (blockX / stepSize);933
+ MV aboveMV = mvsidx;934
+935
+ if (m_useSADinME)936
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, aboveMV.x, aboveMV.y, blockSize, leastError);937
+ else938
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, aboveMV.x, aboveMV.y, blockSize, leastError);939
+940
+ if (error < leastError)941
+ {942
+ best.set(aboveMV.x, aboveMV.y);943
+ leastError = error;944
+ }945
+ }946
+947
+ if (blockX > 0)948
+ {949
+ int idx = ((blockY / stepSize) * mvStride + (blockX - stepSize) / stepSize);950
+ MV leftMV = mvsidx;951
+952
+ if (m_useSADinME)953
+ error = motionErrorLumaSAD(orig, buffer, blockX, blockY, leftMV.x, leftMV.y, blockSize, leastError);954
+ else955
+ error = motionErrorLumaSSD(orig, buffer, blockX, blockY, leftMV.x, leftMV.y, blockSize, leastError);956
+957
+ if (error < leastError)958
+ {959
+ best.set(leftMV.x, leftMV.y);960
+ leastError = error;961
+ }962
+ }963
+964
+ // calculate average965
+ double avg = 0.0;966
+ for (int x1 = 0; x1 < blockSize; x1++)967
+ {968
+ for (int y1 = 0; y1 < blockSize; y1++)969
+ {970
+ avg = avg + *(orig->m_picOrg0 + (blockX + x1 + orig->m_stride * (blockY + y1)));971
+ }972
+ }973
+ avg = avg / (blockSize * blockSize);974
+975
+ // calculate variance976
+ double variance = 0;977
+ for (int x1 = 0; x1 < blockSize; x1++)978
+ {979
+ for (int y1 = 0; y1 < blockSize; y1++)980
+ {981
+ int pix = *(orig->m_picOrg0 + (blockX + x1 + orig->m_stride * (blockY + y1)));982
+ variance = variance + (pix - avg) * (pix - avg);983
+ }984
+ }985
+986
+ leastError = (int)(20 * ((leastError + 5.0) / (variance + 5.0)) + (leastError / (blockSize * blockSize)) / 50);987
+988
+ int mvIdx = (blockY / stepSize) * mvStride + (blockX / stepSize);989
+ mvsmvIdx = best;990
+ minErrormvIdx = leastError;991
+ }992
+ }993
+}994
+995
+void TemporalFilter::destroyRefPicInfo(TemporalFilterRefPicInfo* curFrame)996
+{997
+ if (curFrame)998
+ {999
+ if (curFrame->compensatedPic)1000
+ {1001
+ curFrame->compensatedPic->destroy();1002
+ delete curFrame->compensatedPic;1003
+ }1004
+1005
+ if (curFrame->mvs)1006
+ X265_FREE(curFrame->mvs);1007
+ if (curFrame->mvs0)1008
+ X265_FREE(curFrame->mvs0);1009
+ if (curFrame->mvs1)1010
+ X265_FREE(curFrame->mvs1);1011
+ if (curFrame->mvs2)1012
+ X265_FREE(curFrame->mvs2);1013
+ if (curFrame->noise)1014
+ X265_FREE(curFrame->noise);1015
+ if (curFrame->error)1016
+ X265_FREE(curFrame->error);1017
+ }1018
+}1019
x265_3.6.tar.gz/source/common/temporalfilter.h
Added
187
1
2
+/*****************************************************************************3
+* Copyright (C) 2013-2021 MulticoreWare, Inc4
+*5
+ * Authors: Ashok Kumar Mishra <ashok@multicorewareinc.com>6
+ *7
+* This program is free software; you can redistribute it and/or modify8
+* it under the terms of the GNU General Public License as published by9
+* the Free Software Foundation; either version 2 of the License, or10
+* (at your option) any later version.11
+*12
+* This program is distributed in the hope that it will be useful,13
+* but WITHOUT ANY WARRANTY; without even the implied warranty of14
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the15
+* GNU General Public License for more details.16
+*17
+* You should have received a copy of the GNU General Public License18
+* along with this program; if not, write to the Free Software19
+* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.20
+*21
+* This program is also available under a commercial proprietary license.22
+* For more information, contact us at license @ x265.com.23
+*****************************************************************************/24
+25
+#ifndef X265_TEMPORAL_FILTER_H26
+#define X265_TEMPORAL_FILTER_H27
+28
+#include "x265.h"29
+#include "picyuv.h"30
+#include "mv.h"31
+#include "piclist.h"32
+#include "yuv.h"33
+#include "motion.h"34
+35
+const int s_interpolationFilter168 =36
+{37
+ { 0, 0, 0, 64, 0, 0, 0, 0 }, //038
+ { 0, 1, -3, 64, 4, -2, 0, 0 }, //1 -->-->39
+ { 0, 1, -6, 62, 9, -3, 1, 0 }, //2 -->40
+ { 0, 2, -8, 60, 14, -5, 1, 0 }, //3 -->-->41
+ { 0, 2, -9, 57, 19, -7, 2, 0 }, //442
+ { 0, 3, -10, 53, 24, -8, 2, 0 }, //5 -->-->43
+ { 0, 3, -11, 50, 29, -9, 2, 0 }, //6 -->44
+ { 0, 3, -11, 44, 35, -10, 3, 0 }, //7 -->-->45
+ { 0, 1, -7, 38, 38, -7, 1, 0 }, //846
+ { 0, 3, -10, 35, 44, -11, 3, 0 }, //9 -->-->47
+ { 0, 2, -9, 29, 50, -11, 3, 0 }, //10-->48
+ { 0, 2, -8, 24, 53, -10, 3, 0 }, //11-->-->49
+ { 0, 2, -7, 19, 57, -9, 2, 0 }, //1250
+ { 0, 1, -5, 14, 60, -8, 2, 0 }, //13-->-->51
+ { 0, 1, -3, 9, 62, -6, 1, 0 }, //14-->52
+ { 0, 0, -2, 4, 64, -3, 1, 0 } //15-->-->53
+};54
+55
+const double s_refStrengths34 =56
+{ // abs(POC offset)57
+ // 1, 2 3 458
+ {0.85, 0.57, 0.41, 0.33}, // m_range * 259
+ {1.13, 0.97, 0.81, 0.57}, // m_range60
+ {0.30, 0.30, 0.30, 0.30} // otherwise61
+};62
+63
+namespace X265_NS {64
+ class OrigPicBuffer65
+ {66
+ public:67
+ PicList m_mcstfPicList;68
+ PicList m_mcstfOrigPicFreeList;69
+ PicList m_mcstfOrigPicList;70
+71
+ ~OrigPicBuffer();72
+ void addPicture(Frame*);73
+ void addEncPicture(Frame*);74
+ void setOrigPicList(Frame*, int);75
+ void recycleOrigPicList();76
+ void addPictureToFreelist(Frame*);77
+ void addEncPictureToPicList(Frame*);78
+ };79
+80
+ struct MotionEstimatorTLD81
+ {82
+ MotionEstimate me;83
+84
+ MotionEstimatorTLD()85
+ {86
+ me.init(X265_CSP_I400);87
+ me.setQP(X265_LOOKAHEAD_QP);88
+ }89
+90
+ ~MotionEstimatorTLD() {}91
+ };92
+93
+ struct TemporalFilterRefPicInfo94
+ {95
+ PicYuv* picBuffer;96
+ PicYuv* picBufferSubSampled2;97
+ PicYuv* picBufferSubSampled4;98
+ MV* mvs;99
+ MV* mvs0;100
+ MV* mvs1;101
+ MV* mvs2;102
+ uint32_t mvsStride;103
+ uint32_t mvsStride0;104
+ uint32_t mvsStride1;105
+ uint32_t mvsStride2;106
+ int* error;107
+ int* noise;108
+109
+ int16_t origOffset;110
+ bool isFilteredFrame;111
+ PicYuv* compensatedPic;112
+113
+ int* isSubsampled;114
+115
+ int slicetype;116
+ };117
+118
+ class TemporalFilter119
+ {120
+ public:121
+ TemporalFilter();122
+ ~TemporalFilter() {}123
+124
+ void init(const x265_param* param);125
+126
+ //private:127
+ // Private static member variables128
+ const x265_param *m_param;129
+ int32_t m_bitDepth;130
+ int m_range;131
+ uint8_t m_numRef;132
+ double m_chromaFactor;133
+ double m_sigmaMultiplier;134
+ double m_sigmaZeroPoint;135
+ int m_motionVectorFactor;136
+ int m_padding;137
+138
+ // Private member variables139
+140
+ int m_sourceWidth;141
+ int m_sourceHeight;142
+ int m_QP;143
+144
+ int m_internalCsp;145
+ int m_numComponents;146
+ uint8_t m_sliceTypeConfig;147
+148
+ MotionEstimatorTLD* m_metld;149
+ Yuv predPUYuv;150
+ int m_useSADinME;151
+152
+ int createRefPicInfo(TemporalFilterRefPicInfo* refFrame, x265_param* param);153
+154
+ void bilateralFilter(Frame* frame, TemporalFilterRefPicInfo* mctfRefList, double overallStrength);155
+156
+ void motionEstimationLuma(MV *mvs, uint32_t mvStride, PicYuv *orig, PicYuv *buffer, int bs,157
+ MV *previous = 0, uint32_t prevmvStride = 0, int factor = 1);158
+159
+ void motionEstimationLumaDoubleRes(MV *mvs, uint32_t mvStride, PicYuv *orig, PicYuv *buffer, int blockSize,160
+ MV *previous, uint32_t prevMvStride, int factor, int* minError);161
+162
+ int motionErrorLumaSSD(PicYuv *orig,163
+ PicYuv *buffer,164
+ int x,165
+ int y,166
+ int dx,167
+ int dy,168
+ int bs,169
+ int besterror = 8 * 8 * 1024 * 1024);170
+171
+ int motionErrorLumaSAD(PicYuv *orig,172
+ PicYuv *buffer,173
+ int x,174
+ int y,175
+ int dx,176
+ int dy,177
+ int bs,178
+ int besterror = 8 * 8 * 1024 * 1024);179
+180
+ void destroyRefPicInfo(TemporalFilterRefPicInfo* curFrame);181
+182
+ void applyMotion(MV *mvs, uint32_t mvsStride, PicYuv *input, PicYuv *output);183
+184
+ };185
+}186
+#endif187
x265_3.5.tar.gz/source/common/threading.h -> x265_3.6.tar.gz/source/common/threading.h
Changed
340
1
2
*3
* Authors: Steve Borho <steve@borho.org>4
* Min Chen <chenm003@163.com>5
+ liwei <liwei@multicorewareinc.com>6
*7
* This program is free software; you can redistribute it and/or modify8
* it under the terms of the GNU General Public License as published by9
10
int m_val;11
};12
13
+class NamedSemaphore14
+{15
+public:16
+ NamedSemaphore() : m_sem(NULL)17
+ {18
+ }19
+20
+ ~NamedSemaphore()21
+ {22
+ }23
+24
+ bool create(const char* name, const int initcnt, const int maxcnt)25
+ {26
+ if(!m_sem)27
+ {28
+ m_sem = CreateSemaphoreA(NULL, initcnt, maxcnt, name);29
+ }30
+ return m_sem != NULL;31
+ }32
+33
+ bool give(const int32_t cnt)34
+ {35
+ return ReleaseSemaphore(m_sem, (LONG)cnt, NULL) != FALSE;36
+ }37
+38
+ bool take(const uint32_t time_out = INFINITE)39
+ {40
+ int32_t rt = WaitForSingleObject(m_sem, time_out);41
+ return rt != WAIT_TIMEOUT && rt != WAIT_FAILED;42
+ }43
+44
+ void release()45
+ {46
+ CloseHandle(m_sem);47
+ m_sem = NULL;48
+ }49
+50
+private:51
+ HANDLE m_sem;52
+};53
+54
#else /* POSIX / pthreads */55
56
typedef pthread_t ThreadHandle;57
58
int m_val;59
};60
61
+#define TIMEOUT_INFINITE 0xFFFFFFFF62
+63
+class NamedSemaphore64
+{65
+public:66
+ NamedSemaphore() 67
+ : m_sem(NULL)68
+#ifndef __APPLE__69
+ , m_name(NULL)70
+#endif //__APPLE__71
+ {72
+ }73
+74
+ ~NamedSemaphore()75
+ {76
+ }77
+78
+ bool create(const char* name, const int initcnt, const int maxcnt)79
+ {80
+ bool ret = false;81
+82
+ if (initcnt >= maxcnt)83
+ {84
+ return false;85
+ }86
+87
+#ifdef __APPLE__88
+ do89
+ {90
+ int32_t pshared = name != NULL ? PTHREAD_PROCESS_SHARED : PTHREAD_PROCESS_PRIVATE;91
+92
+ m_sem = (mac_sem_t *)malloc(sizeof(mac_sem_t));93
+ if (!m_sem)94
+ {95
+ break;96
+ }97
+98
+ if (pthread_mutexattr_init(&m_sem->mutexAttr))99
+ {100
+ break;101
+ }102
+103
+ if (pthread_mutexattr_setpshared(&m_sem->mutexAttr, pshared))104
+ {105
+ break;106
+ }107
+108
+ if (pthread_condattr_init(&m_sem->condAttr))109
+ {110
+ break;111
+ }112
+113
+ if (pthread_condattr_setpshared(&m_sem->condAttr, pshared))114
+ {115
+ break;116
+ }117
+118
+ if (pthread_mutex_init(&m_sem->mutex, &m_sem->mutexAttr))119
+ {120
+ break;121
+ }122
+123
+ if (pthread_cond_init(&m_sem->cond, &m_sem->condAttr))124
+ {125
+ break;126
+ }127
+128
+ m_sem->curCnt = initcnt;129
+ m_sem->maxCnt = maxcnt;130
+131
+ ret = true;132
+ } while (0);133
+ 134
+ if (!ret)135
+ {136
+ release();137
+ }138
+139
+#else //__APPLE__140
+ m_sem = sem_open(name, O_CREAT | O_EXCL, 0666, initcnt);141
+ if (m_sem != SEM_FAILED) 142
+ {143
+ m_name = strdup(name);144
+ ret = true;145
+ }146
+ else 147
+ {148
+ if (EEXIST == errno) 149
+ {150
+ m_sem = sem_open(name, 0);151
+ if (m_sem != SEM_FAILED) 152
+ {153
+ m_name = strdup(name);154
+ ret = true;155
+ }156
+ }157
+ }158
+#endif //__APPLE__159
+160
+ return ret;161
+ }162
+163
+ bool give(const int32_t cnt)164
+ {165
+ if (!m_sem)166
+ {167
+ return false;168
+ }169
+170
+#ifdef __APPLE__171
+ if (pthread_mutex_lock(&m_sem->mutex))172
+ {173
+ return false;174
+ }175
+176
+ int oldCnt = m_sem->curCnt;177
+ m_sem->curCnt += cnt;178
+ if (m_sem->curCnt > m_sem->maxCnt)179
+ {180
+ m_sem->curCnt = m_sem->maxCnt;181
+ }182
+183
+ bool ret = true;184
+ if (!oldCnt)185
+ {186
+ ret = 0 == pthread_cond_broadcast(&m_sem->cond);187
+ }188
+189
+ if (pthread_mutex_unlock(&m_sem->mutex))190
+ {191
+ return false;192
+ }193
+194
+ return ret;195
+#else //__APPLE__196
+ int ret = 0;197
+ int32_t curCnt = cnt;198
+ while (curCnt-- && !ret) {199
+ ret = sem_post(m_sem);200
+ }201
+202
+ return 0 == ret;203
+#endif //_APPLE__204
+ }205
+206
+ bool take(const uint32_t time_out = TIMEOUT_INFINITE)207
+ {208
+ if (!m_sem)209
+ {210
+ return false;211
+ }212
+213
+#ifdef __APPLE__214
+215
+ if (pthread_mutex_lock(&m_sem->mutex))216
+ {217
+ return false;218
+ }219
+220
+ bool ret = true;221
+ if (TIMEOUT_INFINITE == time_out) 222
+ {223
+ if (!m_sem->curCnt)224
+ {225
+ if (pthread_cond_wait(&m_sem->cond, &m_sem->mutex))226
+ {227
+ ret = false;228
+ } 229
+ }230
+231
+ if (m_sem->curCnt && ret)232
+ {233
+ m_sem->curCnt--;234
+ }235
+ }236
+ else237
+ {238
+ if (0 == time_out)239
+ {240
+ if (m_sem->curCnt)241
+ {242
+ m_sem->curCnt--;243
+ }244
+ else245
+ {246
+ ret = false;247
+ }248
+ }249
+ else250
+ {251
+ if (!m_sem->curCnt)252
+ {253
+ struct timespec ts;254
+ ts.tv_sec = time_out / 1000L;255
+ ts.tv_nsec = (time_out * 1000000L) - ts.tv_sec * 1000 * 1000 * 1000;256
+257
+ if (pthread_cond_timedwait(&m_sem->cond, &m_sem->mutex, &ts))258
+ {259
+ ret = false;260
+ }261
+ }262
+263
+ if (m_sem->curCnt && ret)264
+ {265
+ m_sem->curCnt--;266
+ }267
+ }268
+ }269
+270
+ if (pthread_mutex_unlock(&m_sem->mutex))271
+ {272
+ return false;273
+ }274
+275
+ return ret;276
+#else //__APPLE__277
+ if (TIMEOUT_INFINITE == time_out) 278
+ {279
+ return 0 == sem_wait(m_sem);280
+ }281
+ else 282
+ {283
+ if (0 == time_out)284
+ {285
+ return 0 == sem_trywait(m_sem);286
+ }287
+ else288
+ {289
+ struct timespec ts;290
+ ts.tv_sec = time_out / 1000L;291
+ ts.tv_nsec = (time_out * 1000000L) - ts.tv_sec * 1000 * 1000 * 1000;292
+ return 0 == sem_timedwait(m_sem, &ts);293
+ }294
+ }295
+#endif //_APPLE__296
+ }297
+298
+ void release()299
+ {300
+ if (m_sem)301
+ {302
+#ifdef __APPLE__303
+ pthread_condattr_destroy(&m_sem->condAttr);304
+ pthread_mutexattr_destroy(&m_sem->mutexAttr);305
+ pthread_mutex_destroy(&m_sem->mutex);306
+ pthread_cond_destroy(&m_sem->cond);307
+ free(m_sem);308
+ m_sem = NULL;309
+#else //__APPLE__310
+ sem_close(m_sem);311
+ sem_unlink(m_name);312
+ m_sem = NULL;313
+ free(m_name);314
+ m_name = NULL;315
+#endif //__APPLE__316
+ }317
+ }318
+319
+private:320
+#ifdef __APPLE__321
+ typedef struct322
+ {323
+ pthread_mutex_t mutex;324
+ pthread_cond_t cond;325
+ pthread_mutexattr_t mutexAttr;326
+ pthread_condattr_t condAttr;327
+ uint32_t curCnt;328
+ uint32_t maxCnt;329
+ }mac_sem_t;330
+ mac_sem_t *m_sem;331
+#else // __APPLE__332
+ sem_t *m_sem;333
+ char *m_name;334
+#endif // __APPLE_335
+};336
+337
#endif // ifdef _WIN32338
339
class ScopedLock340
x265_3.5.tar.gz/source/common/threadpool.cpp -> x265_3.6.tar.gz/source/common/threadpool.cpp
Changed
10
1
2
/* limit threads based on param->numaPools3
* For windows because threads can't be allocated to live across sockets4
* changing the default behavior to be per-socket pools -- FIXME */5
-#if defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN76
+#if defined(_WIN32_WINNT) && _WIN32_WINNT >= _WIN32_WINNT_WIN7 || HAVE_LIBNUMA7
if (!p->numaPools || (strcmp(p->numaPools, "NULL") == 0 || strcmp(p->numaPools, "*") == 0 || strcmp(p->numaPools, "") == 0))8
{9
char poolString50 = "";10
x265_3.5.tar.gz/source/common/version.cpp -> x265_3.6.tar.gz/source/common/version.cpp
Changed
10
1
2
#define ONOS "Unk-OS"3
#endif4
5
-#if X86_646
+#if defined(_LP64) || defined(_WIN64)7
#define BITS "64 bit"8
#else9
#define BITS "32 bit"10
x265_3.5.tar.gz/source/common/x86/asm-primitives.cpp -> x265_3.6.tar.gz/source/common/x86/asm-primitives.cpp
Changed
85
1
2
3
p.frameInitLowres = PFX(frame_init_lowres_core_sse2);4
p.frameInitLowerRes = PFX(frame_init_lowres_core_sse2);5
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_sse2);6
// TODO: the planecopy_sp is really planecopy_SC now, must be fix it 7
//p.planecopy_sp = PFX(downShift_16_sse2);8
p.planecopy_sp_shl = PFX(upShift_16_sse2);9
10
{11
ASSIGN2(p.scale1D_128to64, scale1D_128to64_ssse3);12
p.scale2D_64to32 = PFX(scale2D_64to32_ssse3);13
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_ssse3);14
15
// p.puLUMA_4x4.satd = p.cuBLOCK_4x4.sa8d = PFX(pixel_satd_4x4_ssse3); this one is broken16
ALL_LUMA_PU(satd, pixel_satd, ssse3);17
18
p.puLUMA_64x48.copy_pp = (copy_pp_t)PFX(blockcopy_ss_64x48_avx);19
p.puLUMA_64x64.copy_pp = (copy_pp_t)PFX(blockcopy_ss_64x64_avx);20
p.propagateCost = PFX(mbtree_propagate_cost_avx);21
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_avx);22
}23
if (cpuMask & X265_CPU_XOP)24
{25
26
LUMA_VAR(xop);27
p.frameInitLowres = PFX(frame_init_lowres_core_xop);28
p.frameInitLowerRes = PFX(frame_init_lowres_core_xop);29
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_xop);30
}31
if (cpuMask & X265_CPU_AVX2)32
{33
34
35
p.frameInitLowres = PFX(frame_init_lowres_core_avx2);36
p.frameInitLowerRes = PFX(frame_init_lowres_core_avx2);37
+38
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_avx2);39
+40
p.propagateCost = PFX(mbtree_propagate_cost_avx2);41
p.fix8Unpack = PFX(cutree_fix8_unpack_avx2);42
p.fix8Pack = PFX(cutree_fix8_pack_avx2);43
44
//p.frameInitLowres = PFX(frame_init_lowres_core_mmx2);45
p.frameInitLowres = PFX(frame_init_lowres_core_sse2);46
p.frameInitLowerRes = PFX(frame_init_lowres_core_sse2);47
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_sse2);48
49
ALL_LUMA_TU(blockfill_sNONALIGNED, blockfill_s, sse2);50
ALL_LUMA_TU(blockfill_sALIGNED, blockfill_s, sse2);51
52
ASSIGN2(p.scale1D_128to64, scale1D_128to64_ssse3);53
p.scale2D_64to32 = PFX(scale2D_64to32_ssse3);54
55
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_ssse3);56
+57
ASSIGN2(p.puLUMA_8x4.convert_p2s, filterPixelToShort_8x4_ssse3);58
ASSIGN2(p.puLUMA_8x8.convert_p2s, filterPixelToShort_8x8_ssse3);59
ASSIGN2(p.puLUMA_8x16.convert_p2s, filterPixelToShort_8x16_ssse3);60
61
p.frameInitLowres = PFX(frame_init_lowres_core_avx);62
p.frameInitLowerRes = PFX(frame_init_lowres_core_avx);63
p.propagateCost = PFX(mbtree_propagate_cost_avx);64
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_avx);65
}66
if (cpuMask & X265_CPU_XOP)67
{68
69
p.cuBLOCK_16x16.sse_pp = PFX(pixel_ssd_16x16_xop);70
p.frameInitLowres = PFX(frame_init_lowres_core_xop);71
p.frameInitLowerRes = PFX(frame_init_lowres_core_xop);72
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_xop);73
74
}75
#if X86_6476
77
p.saoCuStatsE2 = PFX(saoCuStatsE2_avx2);78
p.saoCuStatsE3 = PFX(saoCuStatsE3_avx2);79
80
+ p.frameSubSampleLuma = PFX(frame_subsample_luma_avx2);81
+82
if (cpuMask & X265_CPU_BMI2)83
{84
p.scanPosLast = PFX(scanPosLast_avx2_bmi2);85
x265_3.5.tar.gz/source/common/x86/const-a.asm -> x265_3.6.tar.gz/source/common/x86/const-a.asm
Changed
10
1
2
const pw_2000, times 16 dw 0x20003
const pw_8000, times 8 dw 0x80004
const pw_3fff, times 16 dw 0x3fff5
-const pw_32_0, times 4 dw 32,6
+const pw_32_0, times 4 dw 327
times 4 dw 08
const pw_pixel_max, times 16 dw ((1 << BIT_DEPTH)-1)9
10
x265_3.5.tar.gz/source/common/x86/h-ipfilter8.asm -> x265_3.6.tar.gz/source/common/x86/h-ipfilter8.asm
Changed
20
1
2
ALIGN 323
interp4_hps_shuf: times 2 db 0, 1, 2, 3, 1, 2, 3, 4, 8, 9, 10, 11, 9, 10, 11, 124
5
+ALIGN 326
+const interp_4tap_8x8_horiz_shuf, dd 0, 4, 1, 5, 2, 6, 3, 77
+8
SECTION .text9
10
cextern pw_111
12
13
RET14
15
-ALIGN 3216
-const interp_4tap_8x8_horiz_shuf, dd 0, 4, 1, 5, 2, 6, 3, 717
18
%macro FILTER_H4_w6 319
movu %1, srcq - 120
x265_3.5.tar.gz/source/common/x86/mc-a2.asm -> x265_3.6.tar.gz/source/common/x86/mc-a2.asm
Changed
264
1
2
FRAME_INIT_LOWRES3
%endif4
5
+%macro SUBSAMPLEFILT8x4 76
+ mova %3, r0+%77
+ mova %4, r0+r2+%78
+ pavgb %3, %49
+ pavgb %4, r0+r2*2+%710
+ PALIGNR %1, %3, 1, m611
+ PALIGNR %2, %4, 1, m612
+%if cpuflag(xop)13
+ pavgb %1, %314
+ pavgb %2, %415
+%else16
+ pavgb %1, %317
+ pavgb %2, %418
+ psrlw %5, %1, 819
+ psrlw %6, %2, 820
+ pand %1, m721
+ pand %2, m722
+%endif23
+%endmacro24
+25
+%macro SUBSAMPLEFILT32x4U 126
+ movu m1, r0+r227
+ pavgb m0, m1, r028
+ movu m3, r0+r2+129
+ pavgb m2, m3, r0+130
+ pavgb m1, r0+r2*231
+ pavgb m3, r0+r2*2+132
+ pavgb m0, m233
+ pavgb m1, m334
+35
+ movu m3, r0+r2+mmsize36
+ pavgb m2, m3, r0+mmsize37
+ movu m5, r0+r2+1+mmsize38
+ pavgb m4, m5, r0+1+mmsize39
+ pavgb m2, m440
+41
+ pshufb m0, m742
+ pshufb m2, m743
+ punpcklqdq m0, m0, m244
+ vpermq m0, m0, q312045
+ movu %1, m046
+%endmacro47
+48
+%macro SUBSAMPLEFILT16x2 349
+ mova m3, r0+%3+mmsize50
+ mova m2, r0+%351
+ pavgb m3, r0+%3+r2+mmsize52
+ pavgb m2, r0+%3+r253
+ PALIGNR %1, m3, 1, m654
+ pavgb %1, m355
+ PALIGNR m3, m2, 1, m656
+ pavgb m3, m257
+%if cpuflag(xop)58
+ vpperm m3, m3, %1, m659
+%else60
+ pand m3, m761
+ pand %1, m762
+ packuswb m3, %163
+%endif64
+ mova %2, m365
+ mova %1, m266
+%endmacro67
+68
+%macro SUBSAMPLEFILT8x2U 269
+ mova m2, r0+%270
+ pavgb m2, r0+%2+r271
+ mova m0, r0+%2+172
+ pavgb m0, r0+%2+r2+173
+ pavgb m1, m374
+ pavgb m0, m275
+ pand m1, m776
+ pand m0, m777
+ packuswb m0, m178
+ mova %1, m079
+%endmacro80
+81
+%macro SUBSAMPLEFILT8xU 282
+ mova m3, r0+%2+883
+ mova m2, r0+%284
+ pavgw m3, r0+%2+r2+885
+ pavgw m2, r0+%2+r286
+ movu m1, r0+%2+1087
+ movu m0, r0+%2+288
+ pavgw m1, r0+%2+r2+1089
+ pavgw m0, r0+%2+r2+290
+ pavgw m1, m391
+ pavgw m0, m292
+ psrld m3, m1, 1693
+ pand m1, m794
+ pand m0, m795
+ packssdw m0, m196
+ movu %1, m097
+%endmacro98
+99
+%macro SUBSAMPLEFILT8xA 3100
+ movu m3, r0+%3+mmsize101
+ movu m2, r0+%3102
+ pavgw m3, r0+%3+r2+mmsize103
+ pavgw m2, r0+%3+r2104
+ PALIGNR %1, m3, 2, m6105
+ pavgw %1, m3106
+ PALIGNR m3, m2, 2, m6107
+ pavgw m3, m2108
+%if cpuflag(xop)109
+ vpperm m3, m3, %1, m6110
+%else111
+ pand m3, m7112
+ pand %1, m7113
+ packssdw m3, %1114
+%endif115
+%if cpuflag(avx2)116
+ vpermq m3, m3, q3120117
+%endif118
+ movu %2, m3119
+ movu %1, m2120
+%endmacro121
+122
+;-----------------------------------------------------------------------------123
+; void frame_subsample_luma( uint8_t *src0, uint8_t *dst0,124
+; intptr_t src_stride, intptr_t dst_stride, int width, int height )125
+;-----------------------------------------------------------------------------126
+127
+%macro FRAME_SUBSAMPLE_LUMA 0128
+cglobal frame_subsample_luma, 6,7,(12-4*(BIT_DEPTH/9)) ; 8 for HIGH_BIT_DEPTH, 12 otherwise129
+%if HIGH_BIT_DEPTH130
+ shl dword r3m, 1131
+ FIX_STRIDES r2132
+ shl dword r4m, 1133
+%endif134
+%if mmsize >= 16135
+ add dword r4m, mmsize-1136
+ and dword r4m, ~(mmsize-1)137
+%endif138
+ ; src += 2*(height-1)*stride + 2*width139
+ mov r6d, r5m140
+ dec r6d141
+ imul r6d, r2d142
+ add r6d, r4m143
+ lea r0, r0+r6*2144
+ ; dst += (height-1)*stride + width145
+ mov r6d, r5m146
+ dec r6d147
+ imul r6d, r3m148
+ add r6d, r4m149
+ add r1, r6150
+ ; gap = stride - width151
+ mov r6d, r3m152
+ sub r6d, r4m153
+ PUSH r6154
+ %define dst_gap rsp+gprsize155
+ mov r6d, r2d156
+ sub r6d, r4m157
+ shl r6d, 1158
+ PUSH r6159
+ %define src_gap rsp160
+%if HIGH_BIT_DEPTH161
+%if cpuflag(xop)162
+ mova m6, deinterleave_shuf32a163
+ mova m7, deinterleave_shuf32b164
+%else165
+ pcmpeqw m7, m7166
+ psrld m7, 16167
+%endif168
+.vloop:169
+ mov r6d, r4m170
+%ifnidn cpuname, mmx2171
+ movu m0, r0172
+ movu m1, r0+r2173
+ pavgw m0, m1174
+ pavgw m1, r0+r2*2175
+%endif176
+.hloop:177
+ sub r0, mmsize*2178
+ sub r1, mmsize179
+%ifidn cpuname, mmx2180
+ SUBSAMPLEFILT8xU r1, 0181
+%else182
+ SUBSAMPLEFILT8xA m0, r1, 0183
+%endif184
+ sub r6d, mmsize185
+ jg .hloop186
+%else ; !HIGH_BIT_DEPTH187
+%if cpuflag(avx2)188
+ mova m7, deinterleave_shuf189
+%elif cpuflag(xop)190
+ mova m6, deinterleave_shuf32a191
+ mova m7, deinterleave_shuf32b192
+%else193
+ pcmpeqb m7, m7194
+ psrlw m7, 8195
+%endif196
+.vloop:197
+ mov r6d, r4m198
+%ifnidn cpuname, mmx2199
+%if mmsize <= 16200
+ mova m0, r0201
+ mova m1, r0+r2202
+ pavgb m0, m1203
+ pavgb m1, r0+r2*2204
+%endif205
+%endif206
+.hloop:207
+ sub r0, mmsize*2208
+ sub r1, mmsize209
+%if mmsize==32210
+ SUBSAMPLEFILT32x4U r1211
+%elifdef m8212
+ SUBSAMPLEFILT8x4 m0, m1, m2, m3, m10, m11, mmsize213
+ mova m8, m0214
+ mova m9, m1215
+ SUBSAMPLEFILT8x4 m2, m3, m0, m1, m4, m5, 0216
+%if cpuflag(xop)217
+ vpperm m4, m2, m8, m7218
+ vpperm m2, m2, m8, m6219
+%else220
+ packuswb m2, m8221
+%endif222
+ mova r1, m2223
+%elifidn cpuname, mmx2224
+ SUBSAMPLEFILT8x2U r1, 0225
+%else226
+ SUBSAMPLEFILT16x2 m0, r1, 0227
+%endif228
+ sub r6d, mmsize229
+ jg .hloop230
+%endif ; HIGH_BIT_DEPTH231
+.skip:232
+ mov r3, dst_gap233
+ sub r0, src_gap234
+ sub r1, r3235
+ dec dword r5m236
+ jg .vloop237
+ ADD rsp, 2*gprsize238
+ emms239
+ RET240
+%endmacro ; FRAME_SUBSAMPLE_LUMA241
+242
+INIT_MMX mmx2243
+FRAME_SUBSAMPLE_LUMA244
+%if ARCH_X86_64 == 0245
+INIT_MMX cache32, mmx2246
+FRAME_SUBSAMPLE_LUMA247
+%endif248
+INIT_XMM sse2249
+FRAME_SUBSAMPLE_LUMA250
+INIT_XMM ssse3251
+FRAME_SUBSAMPLE_LUMA252
+INIT_XMM avx253
+FRAME_SUBSAMPLE_LUMA254
+INIT_XMM xop255
+FRAME_SUBSAMPLE_LUMA256
+%if ARCH_X86_64 == 1257
+INIT_YMM avx2258
+FRAME_SUBSAMPLE_LUMA259
+%endif260
+261
;-----------------------------------------------------------------------------262
; void mbtree_propagate_cost( int *dst, uint16_t *propagate_in, int32_t *intra_costs,263
; uint16_t *inter_costs, int32_t *inv_qscales, double *fps_factor, int len )264
x265_3.5.tar.gz/source/common/x86/mc.h -> x265_3.6.tar.gz/source/common/x86/mc.h
Changed
19
1
2
3
#undef LOWRES4
5
+#define SUBSAMPLELUMA(cpu) \6
+ void PFX(frame_subsample_luma_ ## cpu)(const pixel* src0, pixel* dst0, intptr_t src_stride, intptr_t dst_stride, int width, int height);7
+SUBSAMPLELUMA(mmx2)8
+SUBSAMPLELUMA(sse2)9
+SUBSAMPLELUMA(ssse3)10
+SUBSAMPLELUMA(avx)11
+SUBSAMPLELUMA(avx2)12
+SUBSAMPLELUMA(xop)13
+14
+#undef SUBSAMPLELUMA15
+16
#define PROPAGATE_COST(cpu) \17
void PFX(mbtree_propagate_cost_ ## cpu)(int* dst, const uint16_t* propagateIn, const int32_t* intraCosts, \18
const uint16_t* interCosts, const int32_t* invQscales, const double* fpsFactor, int len);19
x265_3.5.tar.gz/source/common/x86/x86inc.asm -> x265_3.6.tar.gz/source/common/x86/x86inc.asm
Changed
96
1
2
%endif3
%endmacro4
5
-%macro DEFINE_ARGS_INTERNAL 3+6
- %ifnum %27
- DEFINE_ARGS %38
- %elif %1 == 49
- DEFINE_ARGS %210
- %elif %1 > 411
- DEFINE_ARGS %2, %312
- %endif13
-%endmacro14
-15
%if WIN64 ; Windows x64 ;=================================================16
17
DECLARE_REG 0, rcx18
19
DECLARE_REG 13, R12, 11220
DECLARE_REG 14, R13, 12021
22
-%macro PROLOGUE 2-5+ 0 ; #args, #regs, #xmm_regs, stack_size, arg_names...23
+%macro PROLOGUE 2-5+ 0, 0 ; #args, #regs, #xmm_regs, stack_size, arg_names...24
%assign num_args %125
%assign regs_used %226
ASSERT regs_used >= num_args27
28
WIN64_SPILL_XMM %329
%endif30
LOAD_IF_USED 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 1431
- DEFINE_ARGS_INTERNAL %0, %4, %532
+ %if %0 > 433
+ %ifnum %434
+ DEFINE_ARGS %535
+ %else36
+ DEFINE_ARGS %4, %537
+ %endif38
+ %elifnnum %439
+ DEFINE_ARGS %440
+ %endif41
%endmacro42
43
%macro WIN64_PUSH_XMM 044
45
DECLARE_REG 13, R12, 6446
DECLARE_REG 14, R13, 7247
48
-%macro PROLOGUE 2-5+ 0; #args, #regs, #xmm_regs, stack_size, arg_names...49
+%macro PROLOGUE 2-5+ 0, 0 ; #args, #regs, #xmm_regs, stack_size, arg_names...50
%assign num_args %151
%assign regs_used %252
%assign xmm_regs_used %353
54
PUSH_IF_USED 9, 10, 11, 12, 13, 1455
ALLOC_STACK %456
LOAD_IF_USED 6, 7, 8, 9, 10, 11, 12, 13, 1457
- DEFINE_ARGS_INTERNAL %0, %4, %558
+ %if %0 > 459
+ %ifnum %460
+ DEFINE_ARGS %561
+ %else62
+ DEFINE_ARGS %4, %563
+ %endif64
+ %elifnnum %465
+ DEFINE_ARGS %466
+ %endif67
%endmacro68
69
%define has_epilogue regs_used > 9 || stack_size > 0 || vzeroupper_required70
71
72
DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 1473
74
-%macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, stack_size, arg_names...75
+%macro PROLOGUE 2-5+ 0, 0 ; #args, #regs, #xmm_regs, stack_size, arg_names...76
%assign num_args %177
%assign regs_used %278
ASSERT regs_used >= num_args79
80
PUSH_IF_USED 3, 4, 5, 681
ALLOC_STACK %482
LOAD_IF_USED 0, 1, 2, 3, 4, 5, 683
- DEFINE_ARGS_INTERNAL %0, %4, %584
+ %if %0 > 485
+ %ifnum %486
+ DEFINE_ARGS %587
+ %else88
+ DEFINE_ARGS %4, %589
+ %endif90
+ %elifnnum %491
+ DEFINE_ARGS %492
+ %endif93
%endmacro94
95
%define has_epilogue regs_used > 3 || stack_size > 0 || vzeroupper_required96
x265_3.5.tar.gz/source/common/x86/x86util.asm -> x265_3.6.tar.gz/source/common/x86/x86util.asm
Changed
13
1
2
%elif %1==23
%if mmsize==84
SBUTTERFLY dq, %3, %4, %55
- %else6
+ %elif %0==67
TRANS q, ORDER, %3, %4, %5, %68
+ %else9
+ TRANS q, ORDER, %3, %4, %510
%endif11
%elif %1==412
SBUTTERFLY qdq, %3, %4, %513
x265_3.5.tar.gz/source/encoder/analysis.cpp -> x265_3.6.tar.gz/source/encoder/analysis.cpp
Changed
10
1
2
qp += distortionData->offsetctu.m_cuAddr;3
}4
5
- if (m_param->analysisLoadReuseLevel == 10 && m_param->rc.cuTree)6
+ if (m_param->analysisLoadReuseLevel >= 2 && m_param->rc.cuTree)7
{8
int cuIdx = (ctu.m_cuAddr * ctu.m_numPartitions) + cuGeom.absPartIdx;9
if (ctu.m_slice->m_sliceType == I_SLICE)10
x265_3.5.tar.gz/source/encoder/api.cpp -> x265_3.6.tar.gz/source/encoder/api.cpp
Changed
50
1
2
memcpy(zoneParam, param, sizeof(x265_param));3
for (int i = 0; i < param->rc.zonefileCount; i++)4
{5
- param->rc.zonesi.startFrame = -1;6
encoder->configureZone(zoneParam, param->rc.zonesi.zoneParam);7
}8
9
10
if (numEncoded < 0)11
encoder->m_aborted = true;12
13
+ if ((!encoder->m_numDelayedPic && !numEncoded) && (encoder->m_param->bEnableEndOfSequence || encoder->m_param->bEnableEndOfBitstream))14
+ {15
+ Bitstream bs;16
+ encoder->getEndNalUnits(encoder->m_nalList, bs);17
+ *pp_nal = &encoder->m_nalList.m_nal0;18
+ if (pi_nal) *pi_nal = encoder->m_nalList.m_numNal;19
+ }20
+21
return numEncoded;22
}23
24
25
&PARAM_NS::x265_param_free,26
&PARAM_NS::x265_param_default,27
&PARAM_NS::x265_param_parse,28
+ &PARAM_NS::x265_scenecut_aware_qp_param_parse,29
&PARAM_NS::x265_param_apply_profile,30
&PARAM_NS::x265_param_default_preset,31
&x265_picture_alloc,32
33
if (param->csvLogLevel)34
{35
fprintf(csvfp, "Encode Order, Type, POC, QP, Bits, Scenecut, ");36
+ if (!!param->bEnableTemporalSubLayers)37
+ fprintf(csvfp, "Temporal Sub Layer ID, ");38
if (param->csvLogLevel >= 2)39
fprintf(csvfp, "I/P cost ratio, ");40
if (param->rc.rateControlMode == X265_RC_CRF)41
42
const x265_frame_stats* frameStats = &pic->frameData;43
fprintf(param->csvfpt, "%d, %c-SLICE, %4d, %2.2lf, %10d, %d,", frameStats->encoderOrder, frameStats->sliceType, frameStats->poc,44
frameStats->qp, (int)frameStats->bits, frameStats->bScenecut);45
+ if (!!param->bEnableTemporalSubLayers)46
+ fprintf(param->csvfpt, "%d,", frameStats->tLayer);47
if (param->csvLogLevel >= 2)48
fprintf(param->csvfpt, "%.2f,", frameStats->ipCostRatio);49
if (param->rc.rateControlMode == X265_RC_CRF)50
x265_3.5.tar.gz/source/encoder/dpb.cpp -> x265_3.6.tar.gz/source/encoder/dpb.cpp
Changed
258
1
2
{3
Frame *curFrame = iterFrame;4
iterFrame = iterFrame->m_next;5
- if (!curFrame->m_encData->m_bHasReferences && !curFrame->m_countRefEncoders)6
+ bool isMCSTFReferenced = false;7
+8
+ if (curFrame->m_param->bEnableTemporalFilter)9
+ isMCSTFReferenced =!!(curFrame->m_refPicCnt1);10
+11
+ if (!curFrame->m_encData->m_bHasReferences && !curFrame->m_countRefEncoders && !isMCSTFReferenced)12
{13
curFrame->m_bChromaExtended = false;14
15
+ if (curFrame->m_param->bEnableTemporalFilter)16
+ *curFrame->m_isSubSampled = false;17
+18
// Reset column counter19
X265_CHECK(curFrame->m_reconRowFlag != NULL, "curFrame->m_reconRowFlag check failure");20
X265_CHECK(curFrame->m_reconColCount != NULL, "curFrame->m_reconColCount check failure");21
22
{23
newFrame->m_encData->m_bHasReferences = false;24
25
+ newFrame->m_tempLayer = (newFrame->m_param->bEnableTemporalSubLayers && !m_bTemporalSublayer) ? 1 : newFrame->m_tempLayer;26
// Adjust NAL type for unreferenced B frames (change from _R "referenced"27
// to _N "non-referenced" NAL unit type)28
switch (slice->m_nalUnitType)29
{30
case NAL_UNIT_CODED_SLICE_TRAIL_R:31
- slice->m_nalUnitType = m_bTemporalSublayer ? NAL_UNIT_CODED_SLICE_TSA_N : NAL_UNIT_CODED_SLICE_TRAIL_N;32
+ slice->m_nalUnitType = newFrame->m_param->bEnableTemporalSubLayers ? NAL_UNIT_CODED_SLICE_TSA_N : NAL_UNIT_CODED_SLICE_TRAIL_N;33
break;34
case NAL_UNIT_CODED_SLICE_RADL_R:35
slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_RADL_N;36
37
38
m_picList.pushFront(*newFrame);39
40
+ if (m_bTemporalSublayer && getTemporalLayerNonReferenceFlag())41
+ {42
+ switch (slice->m_nalUnitType)43
+ {44
+ case NAL_UNIT_CODED_SLICE_TRAIL_R:45
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_TRAIL_N;46
+ break;47
+ case NAL_UNIT_CODED_SLICE_RADL_R:48
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_RADL_N;49
+ break;50
+ case NAL_UNIT_CODED_SLICE_RASL_R:51
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_RASL_N;52
+ break;53
+ default:54
+ break;55
+ }56
+ }57
// Do decoding refresh marking if any58
decodingRefreshMarking(pocCurr, slice->m_nalUnitType);59
60
- computeRPS(pocCurr, slice->isIRAP(), &slice->m_rps, slice->m_sps->maxDecPicBuffering);61
-62
+ computeRPS(pocCurr, newFrame->m_tempLayer, slice->isIRAP(), &slice->m_rps, slice->m_sps->maxDecPicBufferingnewFrame->m_tempLayer);63
+ bool isTSAPic = ((slice->m_nalUnitType == 2) || (slice->m_nalUnitType == 3)) ? true : false;64
// Mark pictures in m_piclist as unreferenced if they are not included in RPS65
- applyReferencePictureSet(&slice->m_rps, pocCurr);66
+ applyReferencePictureSet(&slice->m_rps, pocCurr, newFrame->m_tempLayer, isTSAPic);67
+68
+69
+ if (m_bTemporalSublayer && newFrame->m_tempLayer > 070
+ && !(slice->m_nalUnitType == NAL_UNIT_CODED_SLICE_RADL_N // Check if not a leading picture71
+ || slice->m_nalUnitType == NAL_UNIT_CODED_SLICE_RADL_R72
+ || slice->m_nalUnitType == NAL_UNIT_CODED_SLICE_RASL_N73
+ || slice->m_nalUnitType == NAL_UNIT_CODED_SLICE_RASL_R)74
+ )75
+ {76
+ if (isTemporalLayerSwitchingPoint(pocCurr, newFrame->m_tempLayer) || (slice->m_sps->maxTempSubLayers == 1))77
+ {78
+ if (getTemporalLayerNonReferenceFlag())79
+ {80
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_TSA_N;81
+ }82
+ else83
+ {84
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_TSA_R;85
+ }86
+ }87
+ else if (isStepwiseTemporalLayerSwitchingPoint(&slice->m_rps, pocCurr, newFrame->m_tempLayer))88
+ {89
+ bool isSTSA = true;90
+ int id = newFrame->m_gopOffset % x265_gop_ra_lengthnewFrame->m_gopId;91
+ for (int ii = id; (ii < x265_gop_ra_lengthnewFrame->m_gopId && isSTSA == true); ii++)92
+ {93
+ int tempIdRef = x265_gop_ranewFrame->m_gopIdii.layer;94
+ if (tempIdRef == newFrame->m_tempLayer)95
+ {96
+ for (int jj = 0; jj < slice->m_rps.numberOfPositivePictures + slice->m_rps.numberOfNegativePictures; jj++)97
+ {98
+ if (slice->m_rps.bUsedjj)99
+ {100
+ int refPoc = x265_gop_ranewFrame->m_gopIdii.poc_offset + slice->m_rps.deltaPOCjj;101
+ int kk = 0;102
+ for (kk = 0; kk < x265_gop_ra_lengthnewFrame->m_gopId; kk++)103
+ {104
+ if (x265_gop_ranewFrame->m_gopIdkk.poc_offset == refPoc)105
+ {106
+ break;107
+ }108
+ }109
+ if (x265_gop_ranewFrame->m_gopIdkk.layer >= newFrame->m_tempLayer)110
+ {111
+ isSTSA = false;112
+ break;113
+ }114
+ }115
+ }116
+ }117
+ }118
+ if (isSTSA == true)119
+ {120
+ if (getTemporalLayerNonReferenceFlag())121
+ {122
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_STSA_N;123
+ }124
+ else125
+ {126
+ slice->m_nalUnitType = NAL_UNIT_CODED_SLICE_STSA_R;127
+ }128
+ }129
+ }130
+ }131
132
if (slice->m_sliceType != I_SLICE)133
slice->m_numRefIdx0 = x265_clip3(1, newFrame->m_param->maxNumReferences, slice->m_rps.numberOfNegativePictures);134
135
}136
}137
138
-void DPB::computeRPS(int curPoc, bool isRAP, RPS * rps, unsigned int maxDecPicBuffer)139
+void DPB::computeRPS(int curPoc, int tempId, bool isRAP, RPS * rps, unsigned int maxDecPicBuffer)140
{141
unsigned int poci = 0, numNeg = 0, numPos = 0;142
143
144
{145
if ((iterPic->m_poc != curPoc) && iterPic->m_encData->m_bHasReferences)146
{147
- if ((m_lastIDR >= curPoc) || (m_lastIDR <= iterPic->m_poc))148
+ if ((!m_bTemporalSublayer || (iterPic->m_tempLayer <= tempId)) && ((m_lastIDR >= curPoc) || (m_lastIDR <= iterPic->m_poc)))149
{150
rps->pocpoci = iterPic->m_poc;151
rps->deltaPOCpoci = rps->pocpoci - curPoc;152
153
rps->sortDeltaPOC();154
}155
156
+bool DPB::getTemporalLayerNonReferenceFlag()157
+{158
+ Frame* curFrame = m_picList.first();159
+ if (curFrame->m_encData->m_bHasReferences)160
+ {161
+ curFrame->m_sameLayerRefPic = true;162
+ return false;163
+ }164
+ else165
+ return true;166
+}167
+168
/* Marking reference pictures when an IDR/CRA is encountered. */169
void DPB::decodingRefreshMarking(int pocCurr, NalUnitType nalUnitType)170
{171
172
}173
174
/** Function for applying picture marking based on the Reference Picture Set */175
-void DPB::applyReferencePictureSet(RPS *rps, int curPoc)176
+void DPB::applyReferencePictureSet(RPS *rps, int curPoc, int tempId, bool isTSAPicture)177
{178
// loop through all pictures in the reference picture buffer179
Frame* iterFrame = m_picList.first();180
181
}182
if (!referenced)183
iterFrame->m_encData->m_bHasReferences = false;184
+185
+ if (m_bTemporalSublayer)186
+ {187
+ //check that pictures of higher temporal layers are not used188
+ assert(referenced == 0 || iterFrame->m_encData->m_bHasReferences == false || iterFrame->m_tempLayer <= tempId);189
+190
+ //check that pictures of higher or equal temporal layer are not in the RPS if the current picture is a TSA picture191
+ if (isTSAPicture)192
+ {193
+ assert(referenced == 0 || iterFrame->m_tempLayer < tempId);194
+ }195
+ //check that pictures marked as temporal layer non-reference pictures are not used for reference196
+ if (iterFrame->m_tempLayer == tempId)197
+ {198
+ assert(referenced == 0 || iterFrame->m_sameLayerRefPic == true);199
+ }200
+ }201
+ }202
+ iterFrame = iterFrame->m_next;203
+ }204
+}205
+206
+bool DPB::isTemporalLayerSwitchingPoint(int curPoc, int tempId)207
+{208
+ // loop through all pictures in the reference picture buffer209
+ Frame* iterFrame = m_picList.first();210
+ while (iterFrame)211
+ {212
+ if (iterFrame->m_poc != curPoc && iterFrame->m_encData->m_bHasReferences)213
+ {214
+ if (iterFrame->m_tempLayer >= tempId)215
+ {216
+ return false;217
+ }218
+ }219
+ iterFrame = iterFrame->m_next;220
+ }221
+ return true;222
+}223
+224
+bool DPB::isStepwiseTemporalLayerSwitchingPoint(RPS *rps, int curPoc, int tempId)225
+{226
+ // loop through all pictures in the reference picture buffer227
+ Frame* iterFrame = m_picList.first();228
+ while (iterFrame)229
+ {230
+ if (iterFrame->m_poc != curPoc && iterFrame->m_encData->m_bHasReferences)231
+ {232
+ for (int i = 0; i < rps->numberOfPositivePictures + rps->numberOfNegativePictures; i++)233
+ {234
+ if ((iterFrame->m_poc == curPoc + rps->deltaPOCi) && rps->bUsedi)235
+ {236
+ if (iterFrame->m_tempLayer >= tempId)237
+ {238
+ return false;239
+ }240
+ }241
+ }242
}243
iterFrame = iterFrame->m_next;244
}245
+ return true;246
}247
248
/* deciding the nal_unit_type */249
250
if (!curPOC)251
return NAL_UNIT_CODED_SLICE_IDR_N_LP;252
if (bIsKeyFrame)253
- return m_bOpenGOP ? NAL_UNIT_CODED_SLICE_CRA : m_bhasLeadingPicture ? NAL_UNIT_CODED_SLICE_IDR_W_RADL : NAL_UNIT_CODED_SLICE_IDR_N_LP;254
+ return (m_bOpenGOP || m_craNal) ? NAL_UNIT_CODED_SLICE_CRA : m_bhasLeadingPicture ? NAL_UNIT_CODED_SLICE_IDR_W_RADL : NAL_UNIT_CODED_SLICE_IDR_N_LP;255
if (m_pocCRA && curPOC < m_pocCRA)256
// All leading pictures are being marked as TFD pictures here since257
// current encoder uses all reference pictures while encoding leading258
x265_3.5.tar.gz/source/encoder/dpb.h -> x265_3.6.tar.gz/source/encoder/dpb.h
Changed
35
1
2
int m_lastIDR;3
int m_pocCRA;4
int m_bOpenGOP;5
+ int m_craNal;6
int m_bhasLeadingPicture;7
bool m_bRefreshPending;8
bool m_bTemporalSublayer;9
10
m_bRefreshPending = false;11
m_frameDataFreeList = NULL;12
m_bOpenGOP = param->bOpenGOP;13
- m_bTemporalSublayer = !!param->bEnableTemporalSubLayers;14
+ m_craNal = param->craNal;15
+ m_bTemporalSublayer = (param->bEnableTemporalSubLayers > 2);16
}17
18
~DPB();19
20
21
protected:22
23
- void computeRPS(int curPoc, bool isRAP, RPS * rps, unsigned int maxDecPicBuffer);24
+ void computeRPS(int curPoc,int tempId, bool isRAP, RPS * rps, unsigned int maxDecPicBuffer);25
26
- void applyReferencePictureSet(RPS *rps, int curPoc);27
+ void applyReferencePictureSet(RPS *rps, int curPoc, int tempId, bool isTSAPicture);28
+ bool getTemporalLayerNonReferenceFlag();29
void decodingRefreshMarking(int pocCurr, NalUnitType nalUnitType);30
+ bool isTemporalLayerSwitchingPoint(int curPoc, int tempId);31
+ bool isStepwiseTemporalLayerSwitchingPoint(RPS *rps, int curPoc, int tempId);32
33
NalUnitType getNalUnitType(int curPoc, bool bIsKeyFrame);34
};35
x265_3.5.tar.gz/source/encoder/encoder.cpp -> x265_3.6.tar.gz/source/encoder/encoder.cpp
Changed
1237
1
2
{3
{ 1, 1, 1, 1, 1, 5, 1, 2, 2, 2, 50 },4
{ 1, 1, 1, 1, 1, 5, 0, 16, 9, 9, 81 },5
- { 1, 1, 1, 1, 1, 5, 0, 1, 1, 1, 82 }6
+ { 1, 1, 1, 1, 1, 5, 0, 1, 1, 1, 82 },7
+ { 1, 1, 1, 1, 1, 5, 0, 18, 9, 9, 84 }8
+};9
+10
+typedef struct11
+{12
+ int bEnableVideoSignalTypePresentFlag;13
+ int bEnableColorDescriptionPresentFlag;14
+ int bEnableChromaLocInfoPresentFlag;15
+ int colorPrimaries;16
+ int transferCharacteristics;17
+ int matrixCoeffs;18
+ int bEnableVideoFullRangeFlag;19
+ int chromaSampleLocTypeTopField;20
+ int chromaSampleLocTypeBottomField;21
+ const char* systemId;22
+}VideoSignalTypePresets;23
+24
+VideoSignalTypePresets vstPresets =25
+{26
+ {1, 1, 1, 6, 6, 6, 0, 0, 0, "BT601_525"},27
+ {1, 1, 1, 5, 6, 5, 0, 0, 0, "BT601_626"},28
+ {1, 1, 1, 1, 1, 1, 0, 0, 0, "BT709_YCC"},29
+ {1, 1, 0, 1, 1, 0, 0, 0, 0, "BT709_RGB"},30
+ {1, 1, 1, 9, 14, 1, 0, 2, 2, "BT2020_YCC_NCL"},31
+ {1, 1, 0, 9, 16, 9, 0, 0, 0, "BT2020_RGB"},32
+ {1, 1, 1, 9, 16, 9, 0, 2, 2, "BT2100_PQ_YCC"},33
+ {1, 1, 1, 9, 16, 14, 0, 2, 2, "BT2100_PQ_ICTCP"},34
+ {1, 1, 0, 9, 16, 0, 0, 0, 0, "BT2100_PQ_RGB"},35
+ {1, 1, 1, 9, 18, 9, 0, 2, 2, "BT2100_HLG_YCC"},36
+ {1, 1, 0, 9, 18, 0, 0, 0, 0, "BT2100_HLG_RGB"},37
+ {1, 1, 0, 1, 1, 0, 1, 0, 0, "FR709_RGB"},38
+ {1, 1, 0, 9, 14, 0, 1, 0, 0, "FR2020_RGB"},39
+ {1, 1, 1, 12, 1, 6, 1, 1, 1, "FRP3D65_YCC"}40
};41
}42
43
44
m_threadPool = NULL;45
m_analysisFileIn = NULL;46
m_analysisFileOut = NULL;47
+ m_filmGrainIn = NULL;48
m_naluFile = NULL;49
m_offsetEmergency = NULL;50
m_iFrameNum = 0;51
52
m_prevTonemapPayload.payload = NULL;53
m_startPoint = 0;54
m_saveCTUSize = 0;55
- m_edgePic = NULL;56
- m_edgeHistThreshold = 0;57
- m_chromaHistThreshold = 0.0;58
- m_scaledEdgeThreshold = 0.0;59
- m_scaledChromaThreshold = 0.0;60
m_zoneIndex = 0;61
+ m_origPicBuffer = 0;62
}63
64
inline char *strcatFilename(const char *input, const char *suffix)65
66
}67
}68
69
- if (m_param->bHistBasedSceneCut)70
- {71
- m_planeSizes0 = (m_param->sourceWidth >> x265_cli_cspsp->internalCsp.width0) * (m_param->sourceHeight >> x265_cli_cspsm_param->internalCsp.height0);72
- uint32_t pixelbytes = m_param->internalBitDepth > 8 ? 2 : 1;73
- m_edgePic = X265_MALLOC(pixel, m_planeSizes0 * pixelbytes);74
- m_edgeHistThreshold = m_param->edgeTransitionThreshold;75
- m_chromaHistThreshold = x265_min(m_edgeHistThreshold * 10.0, MAX_SCENECUT_THRESHOLD);76
- m_scaledEdgeThreshold = x265_min(m_edgeHistThreshold * SCENECUT_STRENGTH_FACTOR, MAX_SCENECUT_THRESHOLD);77
- m_scaledChromaThreshold = x265_min(m_chromaHistThreshold * SCENECUT_STRENGTH_FACTOR, MAX_SCENECUT_THRESHOLD);78
- if (m_param->sourceBitDepth != m_param->internalBitDepth)79
- {80
- int size = m_param->sourceWidth * m_param->sourceHeight;81
- int hshift = CHROMA_H_SHIFT(m_param->internalCsp);82
- int vshift = CHROMA_V_SHIFT(m_param->internalCsp);83
- int widthC = m_param->sourceWidth >> hshift;84
- int heightC = m_param->sourceHeight >> vshift;85
-86
- m_inputPic0 = X265_MALLOC(pixel, size);87
- if (m_param->internalCsp != X265_CSP_I400)88
- {89
- for (int j = 1; j < 3; j++)90
- {91
- m_inputPicj = X265_MALLOC(pixel, widthC * heightC);92
- }93
- }94
- }95
- }96
-97
// Do not allow WPP if only one row or fewer than 3 columns, it is pointless and unstable98
if (rows == 1 || cols < 3)99
{100
101
lookAheadThreadPooli.start();102
m_lookahead->m_numPools = pools;103
m_dpb = new DPB(m_param);104
+105
+ if (m_param->bEnableTemporalFilter)106
+ m_origPicBuffer = new OrigPicBuffer();107
+108
m_rateControl = new RateControl(*m_param, this);109
if (!m_param->bResetZoneConfig)110
{111
112
}113
}114
}115
+ if (m_param->filmGrain)116
+ {117
+ m_filmGrainIn = x265_fopen(m_param->filmGrain, "rb");118
+ if (!m_filmGrainIn)119
+ {120
+ x265_log_file(NULL, X265_LOG_ERROR, "Failed to open film grain characteristics binary file %s\n", m_param->filmGrain);121
+ }122
+ }123
+124
m_bZeroLatency = !m_param->bframes && !m_param->lookaheadDepth && m_param->frameNumThreads == 1 && m_param->maxSlices == 1;125
m_aborted |= parseLambdaFile(m_param);126
127
128
}129
}130
131
- if (m_param->bHistBasedSceneCut)132
- {133
- if (m_edgePic != NULL)134
- {135
- X265_FREE_ZERO(m_edgePic);136
- }137
-138
- if (m_param->sourceBitDepth != m_param->internalBitDepth)139
- {140
- X265_FREE_ZERO(m_inputPic0);141
- if (m_param->internalCsp != X265_CSP_I400)142
- {143
- for (int i = 1; i < 3; i++)144
- {145
- X265_FREE_ZERO(m_inputPici);146
- }147
- }148
- }149
- }150
-151
for (int i = 0; i < m_param->frameNumThreads; i++)152
{153
if (m_frameEncoderi)154
155
delete zoneReadCount;156
delete zoneWriteCount;157
}158
+159
+ if (m_param->bEnableTemporalFilter)160
+ delete m_origPicBuffer;161
+162
if (m_rateControl)163
{164
m_rateControl->destroy();165
166
}167
if (m_naluFile)168
fclose(m_naluFile);169
+ if (m_filmGrainIn)170
+ x265_fclose(m_filmGrainIn);171
172
#ifdef SVT_HEVC173
X265_FREE(m_svtAppData);174
175
/* release string arguments that were strdup'd */176
free((char*)m_param->rc.lambdaFileName);177
free((char*)m_param->rc.statFileName);178
+ free((char*)m_param->rc.sharedMemName);179
free((char*)m_param->analysisReuseFileName);180
free((char*)m_param->scalingLists);181
free((char*)m_param->csvfn);182
183
free((char*)m_param->toneMapFile);184
free((char*)m_param->analysisSave);185
free((char*)m_param->analysisLoad);186
+ free((char*)m_param->videoSignalTypePreset);187
PARAM_NS::x265_param_free(m_param);188
}189
}190
191
dest->planes2 = (char*)dest->planes1 + src->stride1 * (src->height >> x265_cli_cspssrc->colorSpace.height1);192
}193
194
-bool Encoder::computeHistograms(x265_picture *pic)195
+bool Encoder::isFilterThisframe(uint8_t sliceTypeConfig, int curSliceType)196
{197
- pixel *src = NULL, *planeV = NULL, *planeU = NULL;198
- uint32_t widthC, heightC;199
- int hshift, vshift;200
-201
- hshift = CHROMA_H_SHIFT(pic->colorSpace);202
- vshift = CHROMA_V_SHIFT(pic->colorSpace);203
- widthC = pic->width >> hshift;204
- heightC = pic->height >> vshift;205
-206
- if (pic->bitDepth == X265_DEPTH)207
+ uint8_t newSliceType = 0;208
+ switch (curSliceType)209
{210
- src = (pixel*)pic->planes0;211
- if (m_param->internalCsp != X265_CSP_I400)212
- {213
- planeU = (pixel*)pic->planes1;214
- planeV = (pixel*)pic->planes2;215
- }216
- }217
- else if (pic->bitDepth == 8 && X265_DEPTH > 8)218
- {219
- int shift = (X265_DEPTH - 8);220
- uint8_t *yChar, *uChar, *vChar;221
-222
- yChar = (uint8_t*)pic->planes0;223
- primitives.planecopy_cp(yChar, pic->stride0 / sizeof(*yChar), m_inputPic0, pic->stride0 / sizeof(*yChar), pic->width, pic->height, shift);224
- src = m_inputPic0;225
- if (m_param->internalCsp != X265_CSP_I400)226
- {227
- uChar = (uint8_t*)pic->planes1;228
- vChar = (uint8_t*)pic->planes2;229
- primitives.planecopy_cp(uChar, pic->stride1 / sizeof(*uChar), m_inputPic1, pic->stride1 / sizeof(*uChar), widthC, heightC, shift);230
- primitives.planecopy_cp(vChar, pic->stride2 / sizeof(*vChar), m_inputPic2, pic->stride2 / sizeof(*vChar), widthC, heightC, shift);231
- planeU = m_inputPic1;232
- planeV = m_inputPic2;233
- }234
- }235
- else236
- {237
- uint16_t *yShort, *uShort, *vShort;238
- /* mask off bits that are supposed to be zero */239
- uint16_t mask = (1 << X265_DEPTH) - 1;240
- int shift = abs(pic->bitDepth - X265_DEPTH);241
-242
- yShort = (uint16_t*)pic->planes0;243
- uShort = (uint16_t*)pic->planes1;244
- vShort = (uint16_t*)pic->planes2;245
-246
- if (pic->bitDepth > X265_DEPTH)247
- {248
- /* shift right and mask pixels to final size */249
- primitives.planecopy_sp(yShort, pic->stride0 / sizeof(*yShort), m_inputPic0, pic->stride0 / sizeof(*yShort), pic->width, pic->height, shift, mask);250
- if (m_param->internalCsp != X265_CSP_I400)251
- {252
- primitives.planecopy_sp(uShort, pic->stride1 / sizeof(*uShort), m_inputPic1, pic->stride1 / sizeof(*uShort), widthC, heightC, shift, mask);253
- primitives.planecopy_sp(vShort, pic->stride2 / sizeof(*vShort), m_inputPic2, pic->stride2 / sizeof(*vShort), widthC, heightC, shift, mask);254
- }255
- }256
- else /* Case for (pic.bitDepth < X265_DEPTH) */257
- {258
- /* shift left and mask pixels to final size */259
- primitives.planecopy_sp_shl(yShort, pic->stride0 / sizeof(*yShort), m_inputPic0, pic->stride0 / sizeof(*yShort), pic->width, pic->height, shift, mask);260
- if (m_param->internalCsp != X265_CSP_I400)261
- {262
- primitives.planecopy_sp_shl(uShort, pic->stride1 / sizeof(*uShort), m_inputPic1, pic->stride1 / sizeof(*uShort), widthC, heightC, shift, mask);263
- primitives.planecopy_sp_shl(vShort, pic->stride2 / sizeof(*vShort), m_inputPic2, pic->stride2 / sizeof(*vShort), widthC, heightC, shift, mask);264
- }265
- }266
-267
- src = m_inputPic0;268
- planeU = m_inputPic1;269
- planeV = m_inputPic2;270
- }271
-272
- size_t bufSize = sizeof(pixel) * m_planeSizes0;273
- memset(m_edgePic, 0, bufSize);274
-275
- if (!computeEdge(m_edgePic, src, NULL, pic->width, pic->height, pic->width, false, 1))276
- {277
- x265_log(m_param, X265_LOG_ERROR, "Failed to compute edge!");278
- return false;279
- }280
-281
- pixel pixelVal;282
- int32_t *edgeHist = m_curEdgeHist;283
- memset(edgeHist, 0, EDGE_BINS * sizeof(int32_t));284
- for (uint32_t i = 0; i < m_planeSizes0; i++)285
- {286
- if (m_edgePici)287
- edgeHist1++;288
- else289
- edgeHist0++;290
- }291
-292
- /* Y Histogram Calculation */293
- int32_t *yHist = m_curYUVHist0;294
- memset(yHist, 0, HISTOGRAM_BINS * sizeof(int32_t));295
- for (uint32_t i = 0; i < m_planeSizes0; i++)296
- {297
- pixelVal = srci;298
- yHistpixelVal++;299
+ case 1: newSliceType |= 1 << 0;300
+ break;301
+ case 2: newSliceType |= 1 << 0;302
+ break;303
+ case 3: newSliceType |= 1 << 1;304
+ break;305
+ case 4: newSliceType |= 1 << 2;306
+ break;307
+ case 5: newSliceType |= 1 << 3;308
+ break;309
+ default: return 0;310
}311
+ return ((sliceTypeConfig & newSliceType) != 0);312
+}313
314
- if (pic->colorSpace != X265_CSP_I400)315
- {316
- /* U Histogram Calculation */317
- int32_t *uHist = m_curYUVHist1;318
- memset(uHist, 0, sizeof(m_curYUVHist1));319
- for (uint32_t i = 0; i < m_planeSizes1; i++)320
- {321
- pixelVal = planeUi;322
- uHistpixelVal++;323
- }324
+inline int enqueueRefFrame(FrameEncoder* curframeEncoder, Frame* iterFrame, Frame* curFrame, bool isPreFiltered, int16_t i)325
+{326
+ TemporalFilterRefPicInfo* dest = &curframeEncoder->m_mcstfRefListcurFrame->m_mcstf->m_numRef;327
+ dest->picBuffer = iterFrame->m_fencPic;328
+ dest->picBufferSubSampled2 = iterFrame->m_fencPicSubsampled2;329
+ dest->picBufferSubSampled4 = iterFrame->m_fencPicSubsampled4;330
+ dest->isFilteredFrame = isPreFiltered;331
+ dest->isSubsampled = iterFrame->m_isSubSampled;332
+ dest->origOffset = i;333
+ curFrame->m_mcstf->m_numRef++;334
335
- /* V Histogram Calculation */336
- pixelVal = 0;337
- int32_t *vHist = m_curYUVHist2;338
- memset(vHist, 0, sizeof(m_curYUVHist2));339
- for (uint32_t i = 0; i < m_planeSizes2; i++)340
- {341
- pixelVal = planeVi;342
- vHistpixelVal++;343
- }344
- }345
- return true;346
+ return 1;347
}348
349
-void Encoder::computeHistogramSAD(double *normalizedMaxUVSad, double *normalizedEdgeSad, int curPoc)350
+bool Encoder::generateMcstfRef(Frame* frameEnc, FrameEncoder* currEncoder)351
{352
+ frameEnc->m_mcstf->m_numRef = 0;353
354
- if (curPoc == 0)355
- { /* first frame is scenecut by default no sad computation for the same. */356
- *normalizedMaxUVSad = 0.0;357
- *normalizedEdgeSad = 0.0;358
- }359
- else360
+ for (int iterPOC = (frameEnc->m_poc - frameEnc->m_mcstf->m_range);361
+ iterPOC <= (frameEnc->m_poc + frameEnc->m_mcstf->m_range); iterPOC++)362
{363
- /* compute sum of absolute differences of histogram bins of chroma and luma edge response between the current and prev pictures. */364
- int32_t edgeHistSad = 0;365
- int32_t uHistSad = 0;366
- int32_t vHistSad = 0;367
- double normalizedUSad = 0.0;368
- double normalizedVSad = 0.0;369
-370
- for (int j = 0; j < HISTOGRAM_BINS; j++)371
+ bool isFound = false;372
+ if (iterPOC != frameEnc->m_poc)373
{374
- if (j < 2)375
+ //search for the reference frame in the Original Picture Buffer376
+ if (!isFound)377
{378
- edgeHistSad += abs(m_curEdgeHistj - m_prevEdgeHistj);379
- }380
- uHistSad += abs(m_curYUVHist1j - m_prevYUVHist1j);381
- vHistSad += abs(m_curYUVHist2j - m_prevYUVHist2j);382
- }383
- *normalizedEdgeSad = normalizeRange(edgeHistSad, 0, 2 * m_planeSizes0, 0.0, 1.0);384
- normalizedUSad = normalizeRange(uHistSad, 0, 2 * m_planeSizes1, 0.0, 1.0);385
- normalizedVSad = normalizeRange(vHistSad, 0, 2 * m_planeSizes2, 0.0, 1.0);386
- *normalizedMaxUVSad = x265_max(normalizedUSad, normalizedVSad);387
- }388
-389
- /* store histograms of previous frame for reference */390
- memcpy(m_prevEdgeHist, m_curEdgeHist, sizeof(m_curEdgeHist));391
- memcpy(m_prevYUVHist, m_curYUVHist, sizeof(m_curYUVHist));392
-}393
+ for (int j = 0; j < (2 * frameEnc->m_mcstf->m_range); j++)394
+ {395
+ if (iterPOC < 0)396
+ continue;397
+ if (iterPOC >= m_pocLast)398
+ {399
400
-double Encoder::normalizeRange(int32_t value, int32_t minValue, int32_t maxValue, double rangeStart, double rangeEnd)401
-{402
- return (double)(value - minValue) * (rangeEnd - rangeStart) / (maxValue - minValue) + rangeStart;403
-}404
+ TemporalFilter* mcstf = frameEnc->m_mcstf;405
+ while (mcstf->m_numRef)406
+ {407
+ memset(currEncoder->m_mcstfRefListmcstf->m_numRef.mvs0, 0, sizeof(MV) * ((mcstf->m_sourceWidth / 16) * (mcstf->m_sourceHeight / 16)));408
+ memset(currEncoder->m_mcstfRefListmcstf->m_numRef.mvs1, 0, sizeof(MV) * ((mcstf->m_sourceWidth / 16) * (mcstf->m_sourceHeight / 16)));409
+ memset(currEncoder->m_mcstfRefListmcstf->m_numRef.mvs2, 0, sizeof(MV) * ((mcstf->m_sourceWidth / 16) * (mcstf->m_sourceHeight / 16)));410
+ memset(currEncoder->m_mcstfRefListmcstf->m_numRef.mvs, 0, sizeof(MV) * ((mcstf->m_sourceWidth / 4) * (mcstf->m_sourceHeight / 4)));411
+ memset(currEncoder->m_mcstfRefListmcstf->m_numRef.noise, 0, sizeof(int) * ((mcstf->m_sourceWidth / 4) * (mcstf->m_sourceHeight / 4)));412
+ memset(currEncoder->m_mcstfRefListmcstf->m_numRef.error, 0, sizeof(int) * ((mcstf->m_sourceWidth / 4) * (mcstf->m_sourceHeight / 4)));413
414
-void Encoder::findSceneCuts(x265_picture *pic, bool& bDup, double maxUVSad, double edgeSad, bool& isMaxThres, bool& isHardSC)415
-{416
- double minEdgeT = m_edgeHistThreshold * MIN_EDGE_FACTOR;417
- double minChromaT = minEdgeT * SCENECUT_CHROMA_FACTOR;418
- double maxEdgeT = m_edgeHistThreshold * MAX_EDGE_FACTOR;419
- double maxChromaT = maxEdgeT * SCENECUT_CHROMA_FACTOR;420
- pic->frameData.bScenecut = false;421
+ mcstf->m_numRef--;422
+ }423
424
- if (pic->poc == 0)425
- {426
- /* for first frame */427
- pic->frameData.bScenecut = false;428
- bDup = false;429
- }430
- else431
- {432
- if (edgeSad == 0.0 && maxUVSad == 0.0)433
- {434
- bDup = true;435
- }436
- else if (edgeSad < minEdgeT && maxUVSad < minChromaT)437
- {438
- pic->frameData.bScenecut = false;439
- }440
- else if (edgeSad > maxEdgeT && maxUVSad > maxChromaT)441
- {442
- pic->frameData.bScenecut = true;443
- isMaxThres = true;444
- isHardSC = true;445
- }446
- else if (edgeSad > m_scaledEdgeThreshold || maxUVSad >= m_scaledChromaThreshold447
- || (edgeSad > m_edgeHistThreshold && maxUVSad >= m_chromaHistThreshold))448
- {449
- pic->frameData.bScenecut = true;450
- bDup = false;451
- if (edgeSad > m_scaledEdgeThreshold || maxUVSad >= m_scaledChromaThreshold)452
- isHardSC = true;453
+ break;454
+ }455
+ Frame* iterFrame = frameEnc->m_encData->m_slice->m_mcstfRefFrameList1j;456
+ if (iterFrame->m_poc == iterPOC)457
+ {458
+ if (!enqueueRefFrame(currEncoder, iterFrame, frameEnc, false, (int16_t)(iterPOC - frameEnc->m_poc)))459
+ {460
+ return false;461
+ };462
+ break;463
+ }464
+ }465
+ }466
}467
}468
+469
+ return true;470
}471
472
/**473
474
const x265_picture* inputPic = NULL;475
static int written = 0, read = 0;476
bool dontRead = false;477
- bool bdropFrame = false;478
bool dropflag = false;479
- bool isMaxThres = false;480
- bool isHardSC = false;481
482
if (m_exportedPic)483
{484
if (!m_param->bUseAnalysisFile && m_param->analysisSave)485
x265_free_analysis_data(m_param, &m_exportedPic->m_analysisData);486
+487
ATOMIC_DEC(&m_exportedPic->m_countRefEncoders);488
+489
m_exportedPic = NULL;490
m_dpb->recycleUnreferenced();491
+492
+ if (m_param->bEnableTemporalFilter)493
+ m_origPicBuffer->recycleOrigPicList();494
}495
+496
if ((pic_in && (!m_param->chunkEnd || (m_encodedFrameNum < m_param->chunkEnd))) || (m_param->bEnableFrameDuplication && !pic_in && (read < written)))497
{498
- if (m_param->bHistBasedSceneCut && pic_in)499
- {500
- x265_picture *pic = (x265_picture *) pic_in;501
-502
- if (pic->poc == 0)503
- {504
- /* for entire encode compute the chroma plane sizes only once */505
- for (int i = 1; i < x265_cli_cspsm_param->internalCsp.planes; i++)506
- m_planeSizesi = (pic->width >> x265_cli_cspsm_param->internalCsp.widthi) * (pic->height >> x265_cli_cspsm_param->internalCsp.heighti);507
- }508
-509
- if (computeHistograms(pic))510
- {511
- double maxUVSad = 0.0, edgeSad = 0.0;512
- computeHistogramSAD(&maxUVSad, &edgeSad, pic_in->poc);513
- findSceneCuts(pic, bdropFrame, maxUVSad, edgeSad, isMaxThres, isHardSC);514
- }515
- }516
-517
if ((m_param->bEnableFrameDuplication && !pic_in && (read < written)))518
dontRead = true;519
else520
521
written++;522
}523
524
- if (m_param->bEnableFrameDuplication && m_param->bHistBasedSceneCut)525
- {526
- if (!bdropFrame && m_dupBuffer1->dupPic->frameData.bScenecut == false)527
- {528
- psnrWeight = ComputePSNR(m_dupBuffer0->dupPic, m_dupBuffer1->dupPic, m_param);529
- if (psnrWeight >= m_param->dupThreshold)530
- dropflag = true;531
- }532
- else533
- {534
- dropflag = true;535
- }536
- }537
- else if (m_param->bEnableFrameDuplication)538
+ if (m_param->bEnableFrameDuplication)539
{540
psnrWeight = ComputePSNR(m_dupBuffer0->dupPic, m_dupBuffer1->dupPic, m_param);541
if (psnrWeight >= m_param->dupThreshold)542
543
}544
}545
}546
- if (m_param->recursionSkipMode == EDGE_BASED_RSKIP && m_param->bHistBasedSceneCut)547
- {548
- pixel* src = m_edgePic;549
- primitives.planecopy_pp_shr(src, inFrame->m_fencPic->m_picWidth, inFrame->m_edgeBitPic, inFrame->m_fencPic->m_stride,550
- inFrame->m_fencPic->m_picWidth, inFrame->m_fencPic->m_picHeight, 0);551
- }552
}553
else554
{555
556
inFrame->m_lowres.satdCost = (int64_t)-1;557
inFrame->m_lowresInit = false;558
inFrame->m_isInsideWindow = 0;559
+ inFrame->m_tempLayer = 0;560
+ inFrame->m_sameLayerRefPic = 0;561
}562
563
/* Copy input picture into a Frame and PicYuv, send to lookahead */564
565
inFrame->m_poc = ++m_pocLast;566
inFrame->m_userData = inputPic->userData;567
inFrame->m_pts = inputPic->pts;568
- if (m_param->bHistBasedSceneCut)569
- {570
- inFrame->m_lowres.bScenecut = (inputPic->frameData.bScenecut == 1) ? true : false;571
- inFrame->m_lowres.m_bIsMaxThres = isMaxThres;572
- if (m_param->radl && m_param->keyframeMax != m_param->keyframeMin)573
- inFrame->m_lowres.m_bIsHardScenecut = isHardSC;574
- }575
576
if ((m_param->bEnableSceneCutAwareQp & BACKWARD) && m_param->rc.bStatRead)577
{578
579
rcEntry = &(m_rateControl->m_rce2PassinFrame->m_poc);580
if(rcEntry->scenecut)581
{582
- int backwardWindow = X265_MIN(int((m_param->bwdScenecutWindow / 1000.0) * (m_param->fpsNum / m_param->fpsDenom)), p->lookaheadDepth);583
+ int backwardWindow = X265_MIN(int((m_param->bwdMaxScenecutWindow / 1000.0) * (m_param->fpsNum / m_param->fpsDenom)), p->lookaheadDepth);584
for (int i = 1; i <= backwardWindow; i++)585
{586
int frameNum = inFrame->m_poc - i;587
588
}589
}590
}591
- if (m_param->bHistBasedSceneCut && m_param->analysisSave)592
- {593
- memcpy(inFrame->m_analysisData.edgeHist, m_curEdgeHist, EDGE_BINS * sizeof(int32_t));594
- memcpy(inFrame->m_analysisData.yuvHist0, m_curYUVHist0, HISTOGRAM_BINS *sizeof(int32_t));595
- if (inputPic->colorSpace != X265_CSP_I400)596
- {597
- memcpy(inFrame->m_analysisData.yuvHist1, m_curYUVHist1, HISTOGRAM_BINS * sizeof(int32_t));598
- memcpy(inFrame->m_analysisData.yuvHist2, m_curYUVHist2, HISTOGRAM_BINS * sizeof(int32_t));599
- }600
- }601
+602
inFrame->m_forceqp = inputPic->forceqp;603
inFrame->m_param = (m_reconfigure || m_reconfigureRc) ? m_latestParam : m_param;604
inFrame->m_picStruct = inputPic->picStruct;605
606
}607
608
/* Use the frame types from the first pass, if available */609
- int sliceType = (m_param->rc.bStatRead) ? m_rateControl->rateControlSliceType(inFrame->m_poc) : inputPic->sliceType;610
+ int sliceType = (m_param->rc.bStatRead) ? m_rateControl->rateControlSliceType(inFrame->m_poc) : X265_TYPE_AUTO;611
+ inFrame->m_lowres.sliceTypeReq = inputPic->sliceType;612
613
/* In analysisSave mode, x265_analysis_data is allocated in inputPic and inFrame points to this */614
/* Load analysis data before lookahead->addPicture, since sliceType has been decided */615
616
if (m_reconfigureRc)617
inFrame->m_reconfigureRc = true;618
619
+ if (m_param->bEnableTemporalFilter)620
+ {621
+ if (!m_pocLast)622
+ {623
+ /*One shot allocation of frames in OriginalPictureBuffer*/624
+ int numFramesinOPB = X265_MAX(m_param->bframes, (inFrame->m_mcstf->m_range << 1)) + 1;625
+ for (int i = 0; i < numFramesinOPB; i++)626
+ {627
+ Frame* dupFrame = new Frame;628
+ if (!(dupFrame->create(m_param, pic_in->quantOffsets)))629
+ {630
+ m_aborted = true;631
+ x265_log(m_param, X265_LOG_ERROR, "Memory allocation failure, aborting encode\n");632
+ fflush(stderr);633
+ dupFrame->destroy();634
+ delete dupFrame;635
+ return -1;636
+ }637
+ else638
+ {639
+ if (m_sps.cuOffsetY)640
+ {641
+ dupFrame->m_fencPic->m_cuOffsetC = m_sps.cuOffsetC;642
+ dupFrame->m_fencPic->m_buOffsetC = m_sps.buOffsetC;643
+ dupFrame->m_fencPic->m_cuOffsetY = m_sps.cuOffsetY;644
+ dupFrame->m_fencPic->m_buOffsetY = m_sps.buOffsetY;645
+ if (m_param->internalCsp != X265_CSP_I400)646
+ {647
+ dupFrame->m_fencPic->m_cuOffsetC = m_sps.cuOffsetC;648
+ dupFrame->m_fencPic->m_buOffsetC = m_sps.buOffsetC;649
+ }650
+ m_origPicBuffer->addEncPicture(dupFrame);651
+ }652
+ }653
+ }654
+ }655
+656
+ inFrame->m_refPicCnt1 = 2 * inFrame->m_mcstf->m_range + 1;657
+ if (inFrame->m_poc < inFrame->m_mcstf->m_range)658
+ inFrame->m_refPicCnt1 -= (uint8_t)(inFrame->m_mcstf->m_range - inFrame->m_poc);659
+ if (m_param->totalFrames && (inFrame->m_poc >= (m_param->totalFrames - inFrame->m_mcstf->m_range)))660
+ inFrame->m_refPicCnt1 -= (uint8_t)(inFrame->m_poc + inFrame->m_mcstf->m_range - m_param->totalFrames + 1);661
+662
+ //Extend full-res original picture border663
+ PicYuv *orig = inFrame->m_fencPic;664
+ extendPicBorder(orig->m_picOrg0, orig->m_stride, orig->m_picWidth, orig->m_picHeight, orig->m_lumaMarginX, orig->m_lumaMarginY);665
+ extendPicBorder(orig->m_picOrg1, orig->m_strideC, orig->m_picWidth >> orig->m_hChromaShift, orig->m_picHeight >> orig->m_vChromaShift, orig->m_chromaMarginX, orig->m_chromaMarginY);666
+ extendPicBorder(orig->m_picOrg2, orig->m_strideC, orig->m_picWidth >> orig->m_hChromaShift, orig->m_picHeight >> orig->m_vChromaShift, orig->m_chromaMarginX, orig->m_chromaMarginY);667
+668
+ //TODO: Add subsampling here if required669
+ m_origPicBuffer->addPicture(inFrame);670
+ }671
+672
m_lookahead->addPicture(*inFrame, sliceType);673
m_numDelayedPic++;674
}675
676
pic_out->bitDepth = X265_DEPTH;677
pic_out->userData = outFrame->m_userData;678
pic_out->colorSpace = m_param->internalCsp;679
+ pic_out->frameData.tLayer = outFrame->m_tempLayer;680
frameData = &(pic_out->frameData);681
682
pic_out->pts = outFrame->m_pts;683
684
pic_out->analysisData.poc = pic_out->poc;685
pic_out->analysisData.sliceType = pic_out->sliceType;686
pic_out->analysisData.bScenecut = outFrame->m_lowres.bScenecut;687
- if (m_param->bHistBasedSceneCut)688
- {689
- memcpy(pic_out->analysisData.edgeHist, outFrame->m_analysisData.edgeHist, EDGE_BINS * sizeof(int32_t));690
- memcpy(pic_out->analysisData.yuvHist0, outFrame->m_analysisData.yuvHist0, HISTOGRAM_BINS * sizeof(int32_t));691
- if (pic_out->colorSpace != X265_CSP_I400)692
- {693
- memcpy(pic_out->analysisData.yuvHist1, outFrame->m_analysisData.yuvHist1, HISTOGRAM_BINS * sizeof(int32_t));694
- memcpy(pic_out->analysisData.yuvHist2, outFrame->m_analysisData.yuvHist2, HISTOGRAM_BINS * sizeof(int32_t));695
- }696
- }697
pic_out->analysisData.satdCost = outFrame->m_lowres.satdCost;698
pic_out->analysisData.numCUsInFrame = outFrame->m_analysisData.numCUsInFrame;699
pic_out->analysisData.numPartitions = outFrame->m_analysisData.numPartitions;700
701
if (m_rateControl->writeRateControlFrameStats(outFrame, &curEncoder->m_rce))702
m_aborted = true;703
if (pic_out)704
- { 705
+ {706
/* m_rcData is allocated for every frame */707
pic_out->rcData = outFrame->m_rcData;708
outFrame->m_rcData->qpaRc = outFrame->m_encData->m_avgQpRc;709
710
outFrame->m_rcData->iCuCount = outFrame->m_encData->m_frameStats.percent8x8Intra * m_rateControl->m_ncu;711
outFrame->m_rcData->pCuCount = outFrame->m_encData->m_frameStats.percent8x8Inter * m_rateControl->m_ncu;712
outFrame->m_rcData->skipCuCount = outFrame->m_encData->m_frameStats.percent8x8Skip * m_rateControl->m_ncu;713
+ outFrame->m_rcData->currentSatd = curEncoder->m_rce.coeffBits;714
+ }715
+716
+ if (m_param->bEnableTemporalFilter)717
+ {718
+ Frame *curFrame = m_origPicBuffer->m_mcstfPicList.getPOCMCSTF(outFrame->m_poc);719
+ X265_CHECK(curFrame, "Outframe not found in DPB's mcstfPicList");720
+ curFrame->m_refPicCnt0--;721
+ curFrame->m_refPicCnt1--;722
+ curFrame = m_origPicBuffer->m_mcstfOrigPicList.getPOCMCSTF(outFrame->m_poc);723
+ X265_CHECK(curFrame, "Outframe not found in OPB's mcstfOrigPicList");724
+ curFrame->m_refPicCnt1--;725
}726
727
/* Allow this frame to be recycled if no frame encoders are using it for reference */728
729
{730
ATOMIC_DEC(&outFrame->m_countRefEncoders);731
m_dpb->recycleUnreferenced();732
+ if (m_param->bEnableTemporalFilter)733
+ m_origPicBuffer->recycleOrigPicList();734
}735
else736
m_exportedPic = outFrame;737
738
m_rateControl->m_lastScenecut = frameEnc->m_poc;739
else740
{741
- int maxWindowSize = int((m_param->fwdScenecutWindow / 1000.0) * (m_param->fpsNum / m_param->fpsDenom) + 0.5);742
+ int maxWindowSize = int((m_param->fwdMaxScenecutWindow / 1000.0) * (m_param->fpsNum / m_param->fpsDenom) + 0.5);743
if (frameEnc->m_poc > (m_rateControl->m_lastScenecut + maxWindowSize))744
m_rateControl->m_lastScenecut = frameEnc->m_poc;745
}746
747
analysis->numPartitions = m_param->num4x4Partitions;748
x265_alloc_analysis_data(m_param, analysis);749
}750
+ if (m_param->bEnableTemporalSubLayers > 2)751
+ {752
+ //Re-assign temporalid if the current frame is at the end of encode or when I slice is encountered753
+ if ((frameEnc->m_poc == (m_param->totalFrames - 1)) || (frameEnc->m_lowres.sliceType == X265_TYPE_I) || (frameEnc->m_lowres.sliceType == X265_TYPE_IDR))754
+ {755
+ frameEnc->m_tempLayer = (int8_t)0;756
+ }757
+ }758
/* determine references, setup RPS, etc */759
m_dpb->prepareEncode(frameEnc);760
+761
+ if (m_param->bEnableTemporalFilter)762
+ {763
+ X265_CHECK(!m_origPicBuffer->m_mcstfOrigPicFreeList.empty(), "Frames not available in Encoded OPB");764
+765
+ Frame *dupFrame = m_origPicBuffer->m_mcstfOrigPicFreeList.popBackMCSTF();766
+ dupFrame->m_fencPic->copyFromFrame(frameEnc->m_fencPic);767
+ dupFrame->m_poc = frameEnc->m_poc;768
+ dupFrame->m_encodeOrder = frameEnc->m_encodeOrder;769
+ dupFrame->m_refPicCnt1 = 2 * dupFrame->m_mcstf->m_range + 1;770
+771
+ if (dupFrame->m_poc < dupFrame->m_mcstf->m_range)772
+ dupFrame->m_refPicCnt1 -= (uint8_t)(dupFrame->m_mcstf->m_range - dupFrame->m_poc);773
+ if (m_param->totalFrames && (dupFrame->m_poc >= (m_param->totalFrames - dupFrame->m_mcstf->m_range)))774
+ dupFrame->m_refPicCnt1 -= (uint8_t)(dupFrame->m_poc + dupFrame->m_mcstf->m_range - m_param->totalFrames + 1);775
+776
+ m_origPicBuffer->addEncPictureToPicList(dupFrame);777
+ m_origPicBuffer->setOrigPicList(frameEnc, m_pocLast);778
+ }779
+780
if (!!m_param->selectiveSAO)781
{782
Slice* slice = frameEnc->m_encData->m_slice;783
784
785
if (m_param->rc.rateControlMode != X265_RC_CQP)786
m_lookahead->getEstimatedPictureCost(frameEnc);787
+788
if (m_param->bIntraRefresh)789
calcRefreshInterval(frameEnc);790
791
+ // Generate MCSTF References and perform HME792
+ if (m_param->bEnableTemporalFilter && isFilterThisframe(frameEnc->m_mcstf->m_sliceTypeConfig, frameEnc->m_lowres.sliceType))793
+ {794
+795
+ if (!generateMcstfRef(frameEnc, curEncoder))796
+ {797
+ m_aborted = true;798
+ x265_log(m_param, X265_LOG_ERROR, "Failed to initialize MCSTFReferencePicInfo at POC %d\n", frameEnc->m_poc);799
+ fflush(stderr);800
+ return -1;801
+ }802
+803
+804
+ if (!*frameEnc->m_isSubSampled)805
+ {806
+ primitives.frameSubSampleLuma((const pixel *)frameEnc->m_fencPic->m_picOrg0,frameEnc->m_fencPicSubsampled2->m_picOrg0, frameEnc->m_fencPic->m_stride, frameEnc->m_fencPicSubsampled2->m_stride, frameEnc->m_fencPicSubsampled2->m_picWidth, frameEnc->m_fencPicSubsampled2->m_picHeight);807
+ extendPicBorder(frameEnc->m_fencPicSubsampled2->m_picOrg0, frameEnc->m_fencPicSubsampled2->m_stride, frameEnc->m_fencPicSubsampled2->m_picWidth, frameEnc->m_fencPicSubsampled2->m_picHeight, frameEnc->m_fencPicSubsampled2->m_lumaMarginX, frameEnc->m_fencPicSubsampled2->m_lumaMarginY);808
+ primitives.frameSubSampleLuma((const pixel *)frameEnc->m_fencPicSubsampled2->m_picOrg0,frameEnc->m_fencPicSubsampled4->m_picOrg0, frameEnc->m_fencPicSubsampled2->m_stride, frameEnc->m_fencPicSubsampled4->m_stride, frameEnc->m_fencPicSubsampled4->m_picWidth, frameEnc->m_fencPicSubsampled4->m_picHeight);809
+ extendPicBorder(frameEnc->m_fencPicSubsampled4->m_picOrg0, frameEnc->m_fencPicSubsampled4->m_stride, frameEnc->m_fencPicSubsampled4->m_picWidth, frameEnc->m_fencPicSubsampled4->m_picHeight, frameEnc->m_fencPicSubsampled4->m_lumaMarginX, frameEnc->m_fencPicSubsampled4->m_lumaMarginY);810
+ *frameEnc->m_isSubSampled = true;811
+ }812
+813
+ for (uint8_t i = 1; i <= frameEnc->m_mcstf->m_numRef; i++)814
+ {815
+ TemporalFilterRefPicInfo *ref = &curEncoder->m_mcstfRefListi - 1;816
+ if (!*ref->isSubsampled)817
+ {818
+ primitives.frameSubSampleLuma((const pixel *)ref->picBuffer->m_picOrg0, ref->picBufferSubSampled2->m_picOrg0, ref->picBuffer->m_stride, ref->picBufferSubSampled2->m_stride, ref->picBufferSubSampled2->m_picWidth, ref->picBufferSubSampled2->m_picHeight);819
+ extendPicBorder(ref->picBufferSubSampled2->m_picOrg0, ref->picBufferSubSampled2->m_stride, ref->picBufferSubSampled2->m_picWidth, ref->picBufferSubSampled2->m_picHeight, ref->picBufferSubSampled2->m_lumaMarginX, ref->picBufferSubSampled2->m_lumaMarginY);820
+ primitives.frameSubSampleLuma((const pixel *)ref->picBufferSubSampled2->m_picOrg0,ref->picBufferSubSampled4->m_picOrg0, ref->picBufferSubSampled2->m_stride, ref->picBufferSubSampled4->m_stride, ref->picBufferSubSampled4->m_picWidth, ref->picBufferSubSampled4->m_picHeight);821
+ extendPicBorder(ref->picBufferSubSampled4->m_picOrg0, ref->picBufferSubSampled4->m_stride, ref->picBufferSubSampled4->m_picWidth, ref->picBufferSubSampled4->m_picHeight, ref->picBufferSubSampled4->m_lumaMarginX, ref->picBufferSubSampled4->m_lumaMarginY);822
+ *ref->isSubsampled = true;823
+ }824
+ }825
+826
+ for (uint8_t i = 1; i <= frameEnc->m_mcstf->m_numRef; i++)827
+ {828
+ TemporalFilterRefPicInfo *ref = &curEncoder->m_mcstfRefListi - 1;829
+830
+ curEncoder->m_frameEncTF->motionEstimationLuma(ref->mvs0, ref->mvsStride0, frameEnc->m_fencPicSubsampled4, ref->picBufferSubSampled4, 16);831
+ curEncoder->m_frameEncTF->motionEstimationLuma(ref->mvs1, ref->mvsStride1, frameEnc->m_fencPicSubsampled2, ref->picBufferSubSampled2, 16, ref->mvs0, ref->mvsStride0, 2);832
+ curEncoder->m_frameEncTF->motionEstimationLuma(ref->mvs2, ref->mvsStride2, frameEnc->m_fencPic, ref->picBuffer, 16, ref->mvs1, ref->mvsStride1, 2);833
+ curEncoder->m_frameEncTF->motionEstimationLumaDoubleRes(ref->mvs, ref->mvsStride, frameEnc->m_fencPic, ref->picBuffer, 8, ref->mvs2, ref->mvsStride2, 1, ref->error);834
+ }835
+836
+ for (int i = 0; i < frameEnc->m_mcstf->m_numRef; i++)837
+ {838
+ TemporalFilterRefPicInfo *ref = &curEncoder->m_mcstfRefListi;839
+ ref->slicetype = m_lookahead->findSliceType(frameEnc->m_poc + ref->origOffset);840
+ Frame* dpbframePtr = m_dpb->m_picList.getPOC(frameEnc->m_poc + ref->origOffset);841
+ if (dpbframePtr != NULL)842
+ {843
+ if (dpbframePtr->m_encData->m_slice->m_sliceType == B_SLICE)844
+ ref->slicetype = X265_TYPE_B;845
+ else if (dpbframePtr->m_encData->m_slice->m_sliceType == P_SLICE)846
+ ref->slicetype = X265_TYPE_P;847
+ else848
+ ref->slicetype = X265_TYPE_I;849
+ }850
+ }851
+ }852
+853
/* Allow FrameEncoder::compressFrame() to start in the frame encoder thread */854
if (!curEncoder->startCompressFrame(frameEnc))855
m_aborted = true;856
857
encParam->dynamicRd = param->dynamicRd;858
encParam->bEnableTransformSkip = param->bEnableTransformSkip;859
encParam->bEnableAMP = param->bEnableAMP;860
-861
+ if (param->confWinBottomOffset == 0 && param->confWinRightOffset == 0)862
+ {863
+ encParam->confWinBottomOffset = param->confWinBottomOffset;864
+ encParam->confWinRightOffset = param->confWinRightOffset;865
+ }866
/* Resignal changes in params in Parameter Sets */867
m_sps.maxAMPDepth = (m_sps.bUseAMP = param->bEnableAMP && param->bEnableAMP) ? param->maxCUDepth : 0;868
m_pps.bTransformSkipEnabled = param->bEnableTransformSkip ? 1 : 0;869
870
(float)100.0 * m_numLumaWPBiFrames / m_analyzeB.m_numPics,871
(float)100.0 * m_numChromaWPBiFrames / m_analyzeB.m_numPics);872
}873
- int pWithB = 0;874
- for (int i = 0; i <= m_param->bframes; i++)875
- pWithB += m_lookahead->m_histogrami;876
877
- if (pWithB)878
- {879
- int p = 0;880
- for (int i = 0; i <= m_param->bframes; i++)881
- p += sprintf(buffer + p, "%.1f%% ", 100. * m_lookahead->m_histogrami / pWithB);882
-883
- x265_log(m_param, X265_LOG_INFO, "consecutive B-frames: %s\n", buffer);884
- }885
if (m_param->bLossless)886
{887
float frameSize = (float)(m_param->sourceWidth - m_sps.conformanceWindow.rightOffset) *888
889
}890
}891
892
+void Encoder::getEndNalUnits(NALList& list, Bitstream& bs)893
+{894
+ NALList nalList;895
+ bs.resetBits();896
+897
+ if (m_param->bEnableEndOfSequence)898
+ nalList.serialize(NAL_UNIT_EOS, bs);899
+ if (m_param->bEnableEndOfBitstream)900
+ nalList.serialize(NAL_UNIT_EOB, bs);901
+902
+ list.takeContents(nalList);903
+}904
+905
void Encoder::initVPS(VPS *vps)906
{907
/* Note that much of the VPS is initialized by determineLevel() */908
909
sps->bUseAMP = m_param->bEnableAMP;910
sps->maxAMPDepth = m_param->bEnableAMP ? m_param->maxCUDepth : 0;911
912
- sps->maxTempSubLayers = m_param->bEnableTemporalSubLayers ? 2 : 1;913
- sps->maxDecPicBuffering = m_vps.maxDecPicBuffering;914
- sps->numReorderPics = m_vps.numReorderPics;915
- sps->maxLatencyIncrease = m_vps.maxLatencyIncrease = m_param->bframes;916
+ sps->maxTempSubLayers = m_vps.maxTempSubLayers;// Getting the value from the user917
+918
+ for(uint8_t i = 0; i < sps->maxTempSubLayers; i++)919
+ {920
+ sps->maxDecPicBufferingi = m_vps.maxDecPicBufferingi;921
+ sps->numReorderPicsi = m_vps.numReorderPicsi;922
+ sps->maxLatencyIncreasei = m_vps.maxLatencyIncreasei = m_param->bframes;923
+ }924
925
sps->bUseStrongIntraSmoothing = m_param->bEnableStrongIntraSmoothing;926
sps->bTemporalMVPEnabled = m_param->bEnableTemporalMvp;927
928
p->rc.aqMode = X265_AQ_NONE;929
p->rc.hevcAq = 0;930
}931
+ if (p->rc.aqMode == 0 && p->rc.cuTree)932
+ {933
+ p->rc.aqMode = X265_AQ_VARIANCE;934
+ p->rc.aqStrength = 0;935
+ }936
p->radl = zone->radl;937
}938
memcpy(zone, p, sizeof(x265_param));939
940
p->crQpOffset = 3;941
}942
943
+void Encoder::configureVideoSignalTypePreset(x265_param* p)944
+{945
+ char systemId20 = {};946
+ char colorVolume20 = {};947
+ sscanf(p->videoSignalTypePreset, "%^::%s", systemId, colorVolume);948
+ uint32_t sysId = 0;949
+ while (strcmp(vstPresetssysId.systemId, systemId))950
+ {951
+ if (sysId + 1 == sizeof(vstPresets) / sizeof(vstPresets0))952
+ {953
+ x265_log(NULL, X265_LOG_ERROR, "Incorrect system-id, aborting\n");954
+ m_aborted = true;955
+ break;956
+ }957
+ sysId++;958
+ }959
+960
+ p->vui.bEnableVideoSignalTypePresentFlag = vstPresetssysId.bEnableVideoSignalTypePresentFlag;961
+ p->vui.bEnableColorDescriptionPresentFlag = vstPresetssysId.bEnableColorDescriptionPresentFlag;962
+ p->vui.bEnableChromaLocInfoPresentFlag = vstPresetssysId.bEnableChromaLocInfoPresentFlag;963
+ p->vui.colorPrimaries = vstPresetssysId.colorPrimaries;964
+ p->vui.transferCharacteristics = vstPresetssysId.transferCharacteristics;965
+ p->vui.matrixCoeffs = vstPresetssysId.matrixCoeffs;966
+ p->vui.bEnableVideoFullRangeFlag = vstPresetssysId.bEnableVideoFullRangeFlag;967
+ p->vui.chromaSampleLocTypeTopField = vstPresetssysId.chromaSampleLocTypeTopField;968
+ p->vui.chromaSampleLocTypeBottomField = vstPresetssysId.chromaSampleLocTypeBottomField;969
+970
+ if (colorVolume0 != '\0')971
+ {972
+ if (!strcmp(systemId, "BT2100_PQ_YCC") || !strcmp(systemId, "BT2100_PQ_ICTCP") || !strcmp(systemId, "BT2100_PQ_RGB"))973
+ {974
+ p->bEmitHDR10SEI = 1;975
+ if (!strcmp(colorVolume, "P3D65x1000n0005"))976
+ {977
+ p->masteringDisplayColorVolume = strdup("G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(10000000,5)");978
+ }979
+ else if (!strcmp(colorVolume, "P3D65x4000n005"))980
+ {981
+ p->masteringDisplayColorVolume = strdup("G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(40000000,50)");982
+ }983
+ else if (!strcmp(colorVolume, "BT2100x108n0005"))984
+ {985
+ p->masteringDisplayColorVolume = strdup("G(8500,39850)B(6550,2300)R(34000,146000)WP(15635,16450)L(10000000,1)");986
+ }987
+ else988
+ {989
+ x265_log(NULL, X265_LOG_ERROR, "Incorrect color-volume, aborting\n");990
+ m_aborted = true;991
+ }992
+ }993
+ else994
+ {995
+ x265_log(NULL, X265_LOG_ERROR, "Color-volume is not supported with the given system-id, aborting\n");996
+ m_aborted = true;997
+ }998
+ }999
+1000
+}1001
+1002
void Encoder::configure(x265_param *p)1003
{1004
this->m_param = p;1005
1006
if (!p->rdoqLevel)1007
p->psyRdoq = 0;1008
1009
+ if (p->craNal && p->keyframeMax > 1)1010
+ {1011
+ x265_log_file(NULL, X265_LOG_ERROR, " --cra-nal works only with keyint 1, but given keyint = %s\n", p->keyframeMax);1012
+ m_aborted = true;1013
+ }1014
+1015
/* Disable features which are not supported by the current RD level */1016
if (p->rdLevel < 3)1017
{1018
1019
p->limitReferences = 0;1020
}1021
1022
- if (p->bEnableTemporalSubLayers && !p->bframes)1023
+ if ((p->bEnableTemporalSubLayers > 2) && !p->bframes)1024
{1025
x265_log(p, X265_LOG_WARNING, "B frames not enabled, temporal sublayer disabled\n");1026
p->bEnableTemporalSubLayers = 0;1027
}1028
1029
+ if (!!p->bEnableTemporalSubLayers && p->bEnableTemporalSubLayers < 2)1030
+ {1031
+ p->bEnableTemporalSubLayers = 0;1032
+ x265_log(p, X265_LOG_WARNING, "No support for temporal sublayers less than 2; Disabling temporal layers\n");1033
+ }1034
+1035
+ if (p->bEnableTemporalSubLayers > 5)1036
+ {1037
+ p->bEnableTemporalSubLayers = 5;1038
+ x265_log(p, X265_LOG_WARNING, "No support for temporal sublayers more than 5; Reducing the temporal sublayers to 5\n");1039
+ }1040
+1041
+ // Assign number of B frames for temporal layers1042
+ if (p->bEnableTemporalSubLayers > 2)1043
+ p->bframes = x265_temporal_layer_bframesp->bEnableTemporalSubLayers - 1;1044
+1045
+ if (p->bEnableTemporalSubLayers > 2)1046
+ {1047
+ if (p->bFrameAdaptive)1048
+ {1049
+ x265_log(p, X265_LOG_WARNING, "Disabling adaptive B-frame placement to support temporal sub-layers\n");1050
+ p->bFrameAdaptive = 0;1051
+ }1052
+ }1053
+1054
m_bframeDelay = p->bframes ? (p->bBPyramid ? 2 : 1) : 0;1055
1056
p->bFrameBias = X265_MIN(X265_MAX(-90, p->bFrameBias), 100);1057
1058
p->rc.bStatRead = 0;1059
}1060
1061
+ if ((p->rc.bStatWrite || p->rc.bStatRead) && p->rc.dataShareMode != X265_SHARE_MODE_FILE && p->rc.dataShareMode != X265_SHARE_MODE_SHAREDMEM)1062
+ {1063
+ p->rc.dataShareMode = X265_SHARE_MODE_FILE;1064
+ }1065
+1066
+ if (!p->rc.bStatRead || p->rc.rateControlMode != X265_RC_CRF)1067
+ {1068
+ p->rc.bEncFocusedFramesOnly = 0;1069
+ }1070
+1071
/* some options make no sense if others are disabled */1072
p->bSaoNonDeblocked &= p->bEnableSAO;1073
p->bEnableTSkipFast &= p->bEnableTransformSkip;1074
1075
}1076
}1077
1078
+ if (p->videoSignalTypePreset) // Default disabled.1079
+ configureVideoSignalTypePreset(p);1080
+1081
if (m_param->toneMapFile || p->bHDR10Opt || p->bEmitHDR10SEI)1082
{1083
if (!p->bRepeatHeaders)1084
1085
m_param->searchRange = m_param->hmeRange2;1086
}1087
1088
- if (p->bHistBasedSceneCut && !p->edgeTransitionThreshold)1089
- {1090
- p->edgeTransitionThreshold = 0.03;1091
- x265_log(p, X265_LOG_WARNING, "using default threshold %.2lf for scene cut detection\n", p->edgeTransitionThreshold);1092
- }1093
+ if (p->bEnableSBRC && (p->rc.rateControlMode != X265_RC_CRF || (p->rc.vbvBufferSize == 0 || p->rc.vbvMaxBitrate == 0)))1094
+ {1095
+ x265_log(p, X265_LOG_WARNING, "SBRC can be enabled only with CRF+VBV mode. Disabling SBRC\n");1096
+ p->bEnableSBRC = 0;1097
+ }1098
1099
+ if (p->bEnableSBRC)1100
+ {1101
+ p->rc.ipFactor = p->rc.ipFactor * X265_IPRATIO_STRENGTH;1102
+ if (p->bOpenGOP)1103
+ {1104
+ x265_log(p, X265_LOG_WARNING, "Segment based RateControl requires closed gop structure. Enabling closed GOP.\n");1105
+ p->bOpenGOP = 0;1106
+ }1107
+ if (p->keyframeMax != p->keyframeMin)1108
+ {1109
+ x265_log(p, X265_LOG_WARNING, "Segment based RateControl requires fixed gop length. Force set min-keyint equal to keyint.\n");1110
+ p->keyframeMin = p->keyframeMax;1111
+ }1112
+ }1113
}1114
1115
void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, const x265_picture* picIn, int paramBytes)1116
1117
analysis->frameRecordSize = frameRecordSize;1118
X265_FREAD(&analysis->sliceType, sizeof(int), 1, m_analysisFileIn, &(picData->sliceType));1119
X265_FREAD(&analysis->bScenecut, sizeof(int), 1, m_analysisFileIn, &(picData->bScenecut));1120
- if (m_param->bHistBasedSceneCut)1121
- {1122
- X265_FREAD(&analysis->edgeHist, sizeof(int32_t), EDGE_BINS, m_analysisFileIn, &m_curEdgeHist);1123
- X265_FREAD(&analysis->yuvHist0, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileIn, &m_curYUVHist0);1124
- if (m_param->internalCsp != X265_CSP_I400)1125
- {1126
- X265_FREAD(&analysis->yuvHist1, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileIn, &m_curYUVHist1);1127
- X265_FREAD(&analysis->yuvHist2, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileIn, &m_curYUVHist2);1128
- }1129
- }1130
X265_FREAD(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFileIn, &(picData->satdCost));1131
X265_FREAD(&numCUsLoad, sizeof(int), 1, m_analysisFileIn, &(picData->numCUsInFrame));1132
X265_FREAD(&analysis->numPartitions, sizeof(int), 1, m_analysisFileIn, &(picData->numPartitions));1133
1134
analysis->frameRecordSize = frameRecordSize;1135
X265_FREAD(&analysis->sliceType, sizeof(int), 1, m_analysisFileIn, &(picData->sliceType));1136
X265_FREAD(&analysis->bScenecut, sizeof(int), 1, m_analysisFileIn, &(picData->bScenecut));1137
- if (m_param->bHistBasedSceneCut)1138
- {1139
- X265_FREAD(&analysis->edgeHist, sizeof(int32_t), EDGE_BINS, m_analysisFileIn, &m_curEdgeHist);1140
- X265_FREAD(&analysis->yuvHist0, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileIn, &m_curYUVHist0);1141
- if (m_param->internalCsp != X265_CSP_I400)1142
- {1143
- X265_FREAD(&analysis->yuvHist1, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileIn, &m_curYUVHist1);1144
- X265_FREAD(&analysis->yuvHist2, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileIn, &m_curYUVHist2);1145
- }1146
- }1147
X265_FREAD(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFileIn, &(picData->satdCost));1148
X265_FREAD(&analysis->numCUsInFrame, sizeof(int), 1, m_analysisFileIn, &(picData->numCUsInFrame));1149
X265_FREAD(&analysis->numPartitions, sizeof(int), 1, m_analysisFileIn, &(picData->numPartitions));1150
1151
1152
if (analysis->sliceType == X265_TYPE_IDR || analysis->sliceType == X265_TYPE_I)1153
{1154
- if (m_param->analysisLoadReuseLevel < 2)1155
- return;1156
+ if (m_param->analysisLoadReuseLevel < 2)1157
+ {1158
+ /* Restore to the current encode's numPartitions and numCUsInFrame */1159
+ analysis->numPartitions = m_param->num4x4Partitions;1160
+ analysis->numCUsInFrame = cuLoc.heightInCU * cuLoc.widthInCU;1161
+ analysis->numCuInHeight = cuLoc.heightInCU;1162
+ return;1163
+ }1164
1165
uint8_t *tempBuf = NULL, *depthBuf = NULL, *modeBuf = NULL, *partSizes = NULL;1166
int8_t *cuQPBuf = NULL;1167
1168
uint32_t numDir = analysis->sliceType == X265_TYPE_P ? 1 : 2;1169
uint32_t numPlanes = m_param->internalCsp == X265_CSP_I400 ? 1 : 3;1170
X265_FREAD((WeightParam*)analysis->wt, sizeof(WeightParam), numPlanes * numDir, m_analysisFileIn, (picIn->analysisData.wt));1171
- if (m_param->analysisLoadReuseLevel < 2)1172
- return;1173
+ if (m_param->analysisLoadReuseLevel < 2)1174
+ {1175
+ /* Restore to the current encode's numPartitions and numCUsInFrame */1176
+ analysis->numPartitions = m_param->num4x4Partitions;1177
+ analysis->numCUsInFrame = cuLoc.heightInCU * cuLoc.widthInCU;1178
+ analysis->numCuInHeight = cuLoc.heightInCU;1179
+ return;1180
+ }1181
1182
uint8_t *tempBuf = NULL, *depthBuf = NULL, *modeBuf = NULL, *partSize = NULL, *mergeFlag = NULL;1183
uint8_t *interDir = NULL, *chromaDir = NULL, *mvpIdx2;1184
1185
1186
int bcutree;1187
X265_FREAD(&bcutree, sizeof(int), 1, m_analysisFileIn, &(saveParam->cuTree));1188
- if (loadLevel == 10 && m_param->rc.cuTree && (!bcutree || saveLevel < 2))1189
+ if (loadLevel >= 2 && m_param->rc.cuTree && (!bcutree || saveLevel < 2))1190
{1191
x265_log(NULL, X265_LOG_ERROR, "Error reading cu-tree info. Disabling cutree offsets. \n");1192
m_param->rc.cuTree = 0;1193
1194
distortionData->highDistortionCtuCount++;1195
}1196
}1197
+1198
void Encoder::readAnalysisFile(x265_analysis_data* analysis, int curPoc, int sliceType)1199
{1200
1201
1202
/* calculate frameRecordSize */1203
analysis->frameRecordSize = sizeof(analysis->frameRecordSize) + sizeof(depthBytes) + sizeof(analysis->poc) + sizeof(analysis->sliceType) +1204
sizeof(analysis->numCUsInFrame) + sizeof(analysis->numPartitions) + sizeof(analysis->bScenecut) + sizeof(analysis->satdCost);1205
- if (m_param->bHistBasedSceneCut)1206
- {1207
- analysis->frameRecordSize += sizeof(analysis->edgeHist);1208
- analysis->frameRecordSize += sizeof(int32_t) * HISTOGRAM_BINS;1209
- if (m_param->internalCsp != X265_CSP_I400)1210
- {1211
- analysis->frameRecordSize += sizeof(int32_t) * HISTOGRAM_BINS;1212
- analysis->frameRecordSize += sizeof(int32_t) * HISTOGRAM_BINS;1213
- }1214
- }1215
-1216
if (analysis->sliceType > X265_TYPE_I)1217
{1218
numDir = (analysis->sliceType == X265_TYPE_P) ? 1 : 2;1219
1220
X265_FWRITE(&analysis->poc, sizeof(int), 1, m_analysisFileOut);1221
X265_FWRITE(&analysis->sliceType, sizeof(int), 1, m_analysisFileOut);1222
X265_FWRITE(&analysis->bScenecut, sizeof(int), 1, m_analysisFileOut);1223
- if (m_param->bHistBasedSceneCut)1224
- {1225
- X265_FWRITE(&analysis->edgeHist, sizeof(int32_t), EDGE_BINS, m_analysisFileOut);1226
- X265_FWRITE(&analysis->yuvHist0, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileOut);1227
- if (m_param->internalCsp != X265_CSP_I400)1228
- {1229
- X265_FWRITE(&analysis->yuvHist1, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileOut);1230
- X265_FWRITE(&analysis->yuvHist2, sizeof(int32_t), HISTOGRAM_BINS, m_analysisFileOut);1231
- }1232
- }1233
-1234
X265_FWRITE(&analysis->satdCost, sizeof(int64_t), 1, m_analysisFileOut);1235
X265_FWRITE(&analysis->numCUsInFrame, sizeof(int), 1, m_analysisFileOut);1236
X265_FWRITE(&analysis->numPartitions, sizeof(int), 1, m_analysisFileOut);1237
x265_3.5.tar.gz/source/encoder/encoder.h -> x265_3.6.tar.gz/source/encoder/encoder.h
Changed
72
1
2
#include "nal.h"3
#include "framedata.h"4
#include "svt.h"5
+#include "temporalfilter.h"6
#ifdef ENABLE_HDR10_PLUS7
#include "dynamicHDR10/hdr10plus.h"8
#endif9
10
int m_bToneMap; // Enables tone-mapping11
int m_enableNal;12
13
- /* For histogram based scene-cut detection */14
- pixel* m_edgePic;15
- pixel* m_inputPic3;16
- int32_t m_curYUVHist3HISTOGRAM_BINS;17
- int32_t m_prevYUVHist3HISTOGRAM_BINS;18
- int32_t m_curEdgeHist2;19
- int32_t m_prevEdgeHist2;20
- uint32_t m_planeSizes3;21
- double m_edgeHistThreshold;22
- double m_chromaHistThreshold;23
- double m_scaledEdgeThreshold;24
- double m_scaledChromaThreshold;25
-26
#ifdef ENABLE_HDR10_PLUS27
const hdr10plus_api *m_hdr10plus_api;28
uint8_t **m_cim;29
30
31
ThreadSafeInteger* zoneReadCount;32
ThreadSafeInteger* zoneWriteCount;33
+ /* Film grain model file */34
+ FILE* m_filmGrainIn;35
+ OrigPicBuffer* m_origPicBuffer;36
37
Encoder();38
~Encoder()39
40
41
void getStreamHeaders(NALList& list, Entropy& sbacCoder, Bitstream& bs);42
43
+ void getEndNalUnits(NALList& list, Bitstream& bs);44
+45
void fetchStats(x265_stats* stats, size_t statsSizeBytes);46
47
void printSummary();48
49
50
void copyPicture(x265_picture *dest, const x265_picture *src);51
52
- bool computeHistograms(x265_picture *pic);53
- void computeHistogramSAD(double *maxUVNormalizedSAD, double *edgeNormalizedSAD, int curPoc);54
- double normalizeRange(int32_t value, int32_t minValue, int32_t maxValue, double rangeStart, double rangeEnd);55
- void findSceneCuts(x265_picture *pic, bool& bDup, double m_maxUVSADVal, double m_edgeSADVal, bool& isMaxThres, bool& isHardSC);56
-57
void initRefIdx();58
void analyseRefIdx(int *numRefIdx);59
void updateRefIdx();60
61
62
void configureDolbyVisionParams(x265_param* p);63
64
+ void configureVideoSignalTypePreset(x265_param* p);65
+66
+ bool isFilterThisframe(uint8_t sliceTypeConfig, int curSliceType);67
+ bool generateMcstfRef(Frame* frameEnc, FrameEncoder* currEncoder);68
+69
protected:70
71
void initVPS(VPS *vps);72
x265_3.5.tar.gz/source/encoder/entropy.cpp -> x265_3.6.tar.gz/source/encoder/entropy.cpp
Changed
41
1
2
3
for (uint32_t i = 0; i < vps.maxTempSubLayers; i++)4
{5
- WRITE_UVLC(vps.maxDecPicBuffering - 1, "vps_max_dec_pic_buffering_minus1i");6
- WRITE_UVLC(vps.numReorderPics, "vps_num_reorder_picsi");7
- WRITE_UVLC(vps.maxLatencyIncrease + 1, "vps_max_latency_increase_plus1i");8
+ WRITE_UVLC(vps.maxDecPicBufferingi - 1, "vps_max_dec_pic_buffering_minus1i");9
+ WRITE_UVLC(vps.numReorderPicsi, "vps_num_reorder_picsi");10
+ WRITE_UVLC(vps.maxLatencyIncreasei + 1, "vps_max_latency_increase_plus1i");11
}12
13
WRITE_CODE(0, 6, "vps_max_nuh_reserved_zero_layer_id");14
15
16
for (uint32_t i = 0; i < sps.maxTempSubLayers; i++)17
{18
- WRITE_UVLC(sps.maxDecPicBuffering - 1, "sps_max_dec_pic_buffering_minus1i");19
- WRITE_UVLC(sps.numReorderPics, "sps_num_reorder_picsi");20
- WRITE_UVLC(sps.maxLatencyIncrease + 1, "sps_max_latency_increase_plus1i");21
+ WRITE_UVLC(sps.maxDecPicBufferingi - 1, "sps_max_dec_pic_buffering_minus1i");22
+ WRITE_UVLC(sps.numReorderPicsi, "sps_num_reorder_picsi");23
+ WRITE_UVLC(sps.maxLatencyIncreasei + 1, "sps_max_latency_increase_plus1i");24
}25
26
WRITE_UVLC(sps.log2MinCodingBlockSize - 3, "log2_min_coding_block_size_minus3");27
28
29
if (maxTempSubLayers > 1)30
{31
- WRITE_FLAG(0, "sub_layer_profile_present_flagi");32
- WRITE_FLAG(0, "sub_layer_level_present_flagi");33
+ for(int i = 0; i < maxTempSubLayers - 1; i++)34
+ {35
+ WRITE_FLAG(0, "sub_layer_profile_present_flagi");36
+ WRITE_FLAG(0, "sub_layer_level_present_flagi");37
+ }38
for (int i = maxTempSubLayers - 1; i < 8 ; i++)39
WRITE_CODE(0, 2, "reserved_zero_2bits");40
}41
x265_3.5.tar.gz/source/encoder/frameencoder.cpp -> x265_3.6.tar.gz/source/encoder/frameencoder.cpp
Changed
200
1
2
#include "common.h"3
#include "slicetype.h"4
#include "nal.h"5
+#include "temporalfilter.h"6
7
namespace X265_NS {8
void weightAnalyse(Slice& slice, Frame& frame, x265_param& param);9
10
delete m_rce.picTimingSEI;11
delete m_rce.hrdTiming;12
}13
+14
+ if (m_param->bEnableTemporalFilter)15
+ {16
+ delete m_frameEncTF->m_metld;17
+18
+ for (int i = 0; i < (m_frameEncTF->m_range << 1); i++)19
+ m_frameEncTF->destroyRefPicInfo(&m_mcstfRefListi);20
+21
+ delete m_frameEncTF;22
+ }23
}24
25
bool FrameEncoder::init(Encoder *top, int numRows, int numCols)26
27
m_sliceAddrBits = (uint16_t)(tmp + 1);28
}29
30
+ if (m_param->bEnableTemporalFilter)31
+ {32
+ m_frameEncTF = new TemporalFilter();33
+ if (m_frameEncTF)34
+ m_frameEncTF->init(m_param);35
+36
+ for (int i = 0; i < (m_frameEncTF->m_range << 1); i++)37
+ ok &= !!m_frameEncTF->createRefPicInfo(&m_mcstfRefListi, m_param);38
+ }39
+40
return ok;41
}42
43
44
m_ssimCnt = 0;45
memset(&(m_frame->m_encData->m_frameStats), 0, sizeof(m_frame->m_encData->m_frameStats));46
47
- if (!m_param->bHistBasedSceneCut && m_param->rc.aqMode != X265_AQ_EDGE && m_param->recursionSkipMode == EDGE_BASED_RSKIP)48
+ if (m_param->rc.aqMode != X265_AQ_EDGE && m_param->recursionSkipMode == EDGE_BASED_RSKIP)49
{50
int height = m_frame->m_fencPic->m_picHeight;51
int width = m_frame->m_fencPic->m_picWidth;52
53
* unit) */54
Slice* slice = m_frame->m_encData->m_slice;55
56
+ if (m_param->bEnableEndOfSequence && m_frame->m_lowres.sliceType == X265_TYPE_IDR && m_frame->m_poc)57
+ {58
+ m_bs.resetBits();59
+ m_nalList.serialize(NAL_UNIT_EOS, m_bs);60
+ }61
+62
if (m_param->bEnableAccessUnitDelimiters && (m_frame->m_poc || m_param->bRepeatHeaders))63
{64
m_bs.resetBits();65
66
int qp = m_top->m_rateControl->rateControlStart(m_frame, &m_rce, m_top);67
m_rce.newQp = qp;68
69
+ if (m_param->bEnableTemporalFilter)70
+ {71
+ m_frameEncTF->m_QP = qp;72
+ m_frameEncTF->bilateralFilter(m_frame, m_mcstfRefList, m_param->temporalFilterStrength);73
+ }74
+75
if (m_nr)76
{77
if (qp > QP_MAX_SPEC && m_frame->m_param->rc.vbvBufferSize)78
79
// wait after removal of the access unit with the most recent80
// buffering period SEI message81
sei->m_auCpbRemovalDelay = X265_MIN(X265_MAX(1, m_rce.encodeOrder - prevBPSEI), (1 << hrd->cpbRemovalDelayLength));82
- sei->m_picDpbOutputDelay = slice->m_sps->numReorderPics + poc - m_rce.encodeOrder;83
+ sei->m_picDpbOutputDelay = slice->m_sps->numReorderPicsm_frame->m_tempLayer + poc - m_rce.encodeOrder;84
}85
86
sei->writeSEImessages(m_bs, *slice->m_sps, NAL_UNIT_PREFIX_SEI, m_nalList, m_param->bSingleSeiNal);87
88
m_seiAlternativeTC.m_preferredTransferCharacteristics = m_param->preferredTransferCharacteristics;89
m_seiAlternativeTC.writeSEImessages(m_bs, *slice->m_sps, NAL_UNIT_PREFIX_SEI, m_nalList, m_param->bSingleSeiNal);90
}91
-92
+ /* Write Film grain characteristics if present */93
+ if (this->m_top->m_filmGrainIn)94
+ {95
+ FilmGrainCharacteristics m_filmGrain;96
+ /* Read the Film grain model file */97
+ readModel(&m_filmGrain, this->m_top->m_filmGrainIn);98
+ m_filmGrain.writeSEImessages(m_bs, *slice->m_sps, NAL_UNIT_PREFIX_SEI, m_nalList, m_param->bSingleSeiNal);99
+ }100
/* Write user SEI */101
for (int i = 0; i < m_frame->m_userSEI.numPayloads; i++)102
{103
104
if (m_param->bDynamicRefine && m_top->m_startPoint <= m_frame->m_encodeOrder) //Avoid collecting data that will not be used by future frames.105
collectDynDataFrame();106
107
+ if (m_param->bEnableTemporalFilter && m_top->isFilterThisframe(m_frame->m_mcstf->m_sliceTypeConfig, m_frame->m_lowres.sliceType))108
+ {109
+ //Reset the MCSTF context in Frame Encoder and Frame110
+ for (int i = 0; i < (m_frameEncTF->m_range << 1); i++)111
+ {112
+ memset(m_mcstfRefListi.mvs0, 0, sizeof(MV) * ((m_param->sourceWidth / 16) * (m_param->sourceHeight / 16)));113
+ memset(m_mcstfRefListi.mvs1, 0, sizeof(MV) * ((m_param->sourceWidth / 16) * (m_param->sourceHeight / 16)));114
+ memset(m_mcstfRefListi.mvs2, 0, sizeof(MV) * ((m_param->sourceWidth / 16) * (m_param->sourceHeight / 16)));115
+ memset(m_mcstfRefListi.mvs, 0, sizeof(MV) * ((m_param->sourceWidth / 4) * (m_param->sourceHeight / 4)));116
+ memset(m_mcstfRefListi.noise, 0, sizeof(int) * ((m_param->sourceWidth / 4) * (m_param->sourceHeight / 4)));117
+ memset(m_mcstfRefListi.error, 0, sizeof(int) * ((m_param->sourceWidth / 4) * (m_param->sourceHeight / 4)));118
+119
+ m_frame->m_mcstf->m_numRef = 0;120
+ }121
+ }122
+123
+124
if (m_param->rc.bStatWrite)125
{126
int totalI = 0, totalP = 0, totalSkip = 0;127
128
129
m_bs.writeByteAlignment();130
131
- m_nalList.serialize(slice->m_nalUnitType, m_bs);132
+ m_nalList.serialize(slice->m_nalUnitType, m_bs, (!!m_param->bEnableTemporalSubLayers ? m_frame->m_tempLayer + 1 : (1 + (slice->m_nalUnitType == NAL_UNIT_CODED_SLICE_TSA_N))));133
}134
}135
else136
137
m_entropyCoder.codeSliceHeaderWPPEntryPoints(m_substreamSizes, (slice->m_sps->numCuInHeight - 1), maxStreamSize);138
m_bs.writeByteAlignment();139
140
- m_nalList.serialize(slice->m_nalUnitType, m_bs);141
+ m_nalList.serialize(slice->m_nalUnitType, m_bs, (!!m_param->bEnableTemporalSubLayers ? m_frame->m_tempLayer + 1 : (1 + (slice->m_nalUnitType == NAL_UNIT_CODED_SLICE_TSA_N))));142
}143
144
if (m_param->decodedPictureHashSEI)145
146
m_nr->nrOffsetDenoisecat0 = 0;147
}148
}149
+150
+void FrameEncoder::readModel(FilmGrainCharacteristics* m_filmGrain, FILE* filmgrain)151
+{152
+ char const* errorMessage = "Error reading FilmGrain characteristics\n";153
+ FilmGrain m_fg;154
+ x265_fread((char* )&m_fg, sizeof(bool) * 3 + sizeof(uint8_t), 1, filmgrain, errorMessage);155
+ m_filmGrain->m_filmGrainCharacteristicsCancelFlag = m_fg.m_filmGrainCharacteristicsCancelFlag;156
+ m_filmGrain->m_filmGrainCharacteristicsPersistenceFlag = m_fg.m_filmGrainCharacteristicsPersistenceFlag;157
+ m_filmGrain->m_filmGrainModelId = m_fg.m_filmGrainModelId;158
+ m_filmGrain->m_separateColourDescriptionPresentFlag = m_fg.m_separateColourDescriptionPresentFlag;159
+ if (m_filmGrain->m_separateColourDescriptionPresentFlag)160
+ {161
+ ColourDescription m_clr;162
+ x265_fread((char* )&m_clr, sizeof(bool) + sizeof(uint8_t) * 5, 1, filmgrain, errorMessage);163
+ m_filmGrain->m_filmGrainBitDepthLumaMinus8 = m_clr.m_filmGrainBitDepthLumaMinus8;164
+ m_filmGrain->m_filmGrainBitDepthChromaMinus8 = m_clr.m_filmGrainBitDepthChromaMinus8;165
+ m_filmGrain->m_filmGrainFullRangeFlag = m_clr.m_filmGrainFullRangeFlag;166
+ m_filmGrain->m_filmGrainColourPrimaries = m_clr.m_filmGrainColourPrimaries;167
+ m_filmGrain->m_filmGrainTransferCharacteristics = m_clr.m_filmGrainTransferCharacteristics;168
+ m_filmGrain->m_filmGrainMatrixCoeffs = m_clr.m_filmGrainMatrixCoeffs;169
+ }170
+ FGPresent m_present;171
+ x265_fread((char* )&m_present, sizeof(bool) * 3 + sizeof(uint8_t) * 2, 1, filmgrain, errorMessage);172
+ m_filmGrain->m_blendingModeId = m_present.m_blendingModeId;173
+ m_filmGrain->m_log2ScaleFactor = m_present.m_log2ScaleFactor;174
+ m_filmGrain->m_compModel0.bPresentFlag = m_present.m_presentFlag0;175
+ m_filmGrain->m_compModel1.bPresentFlag = m_present.m_presentFlag1;176
+ m_filmGrain->m_compModel2.bPresentFlag = m_present.m_presentFlag2;177
+ for (int i = 0; i < MAX_NUM_COMPONENT; i++)178
+ {179
+ if (m_filmGrain->m_compModeli.bPresentFlag)180
+ {181
+ x265_fread((char* )(&m_filmGrain->m_compModeli.m_filmGrainNumIntensityIntervalMinus1), sizeof(uint8_t), 1, filmgrain, errorMessage);182
+ x265_fread((char* )(&m_filmGrain->m_compModeli.numModelValues), sizeof(uint8_t), 1, filmgrain, errorMessage);183
+ m_filmGrain->m_compModeli.intensityValues = (FilmGrainCharacteristics::CompModelIntensityValues* ) malloc(sizeof(FilmGrainCharacteristics::CompModelIntensityValues) * (m_filmGrain->m_compModeli.m_filmGrainNumIntensityIntervalMinus1+1)) ;184
+ for (int j = 0; j <= m_filmGrain->m_compModeli.m_filmGrainNumIntensityIntervalMinus1; j++)185
+ {186
+ x265_fread((char* )(&m_filmGrain->m_compModeli.intensityValuesj.intensityIntervalLowerBound), sizeof(uint8_t), 1, filmgrain, errorMessage);187
+ x265_fread((char* )(&m_filmGrain->m_compModeli.intensityValuesj.intensityIntervalUpperBound), sizeof(uint8_t), 1, filmgrain, errorMessage);188
+ m_filmGrain->m_compModeli.intensityValuesj.compModelValue = (int* ) malloc(sizeof(int) * (m_filmGrain->m_compModeli.numModelValues));189
+ for (int k = 0; k < m_filmGrain->m_compModeli.numModelValues; k++)190
+ {191
+ x265_fread((char* )(&m_filmGrain->m_compModeli.intensityValuesj.compModelValuek), sizeof(int), 1, filmgrain, errorMessage);192
+ }193
+ }194
+ }195
+ }196
+}197
#if ENABLE_LIBVMAF198
void FrameEncoder::vmafFrameLevelScore()199
{200
x265_3.5.tar.gz/source/encoder/frameencoder.h -> x265_3.6.tar.gz/source/encoder/frameencoder.h
Changed
63
1
2
#include "ratecontrol.h"3
#include "reference.h"4
#include "nal.h"5
+#include "temporalfilter.h"6
7
namespace X265_NS {8
// private x265 namespace9
10
}11
};12
13
+/*Film grain characteristics*/14
+struct FilmGrain15
+{16
+ bool m_filmGrainCharacteristicsCancelFlag;17
+ bool m_filmGrainCharacteristicsPersistenceFlag;18
+ bool m_separateColourDescriptionPresentFlag;19
+ uint8_t m_filmGrainModelId;20
+ uint8_t m_blendingModeId;21
+ uint8_t m_log2ScaleFactor;22
+};23
+24
+struct ColourDescription25
+{26
+ bool m_filmGrainFullRangeFlag;27
+ uint8_t m_filmGrainBitDepthLumaMinus8;28
+ uint8_t m_filmGrainBitDepthChromaMinus8;29
+ uint8_t m_filmGrainColourPrimaries;30
+ uint8_t m_filmGrainTransferCharacteristics;31
+ uint8_t m_filmGrainMatrixCoeffs;32
+};33
+34
+struct FGPresent35
+{36
+ uint8_t m_blendingModeId;37
+ uint8_t m_log2ScaleFactor;38
+ bool m_presentFlag3;39
+};40
+41
// Manages the wave-front processing of a single encoding frame42
class FrameEncoder : public WaveFront, public Thread43
{44
45
FrameFilter m_frameFilter;46
NALList m_nalList;47
48
+ // initialization for mcstf49
+ TemporalFilter* m_frameEncTF;50
+ TemporalFilterRefPicInfo m_mcstfRefListMAX_MCSTF_TEMPORAL_WINDOW_LENGTH;51
+52
class WeightAnalysis : public BondedTaskGroup53
{54
public:55
56
void collectDynDataFrame();57
void computeAvgTrainingData();58
void collectDynDataRow(CUData& ctu, FrameStats* rowStats); 59
+ void readModel(FilmGrainCharacteristics* m_filmGrain, FILE* filmgrain);60
};61
}62
63
x265_3.5.tar.gz/source/encoder/level.cpp -> x265_3.6.tar.gz/source/encoder/level.cpp
Changed
86
1
2
* for intra-only profiles (vps.ptl.intraConstraintFlag) */3
vps.ptl.lowerBitRateConstraintFlag = true;4
5
- vps.maxTempSubLayers = param.bEnableTemporalSubLayers ? 2 : 1;6
+ vps.maxTempSubLayers = !!param.bEnableTemporalSubLayers ? param.bEnableTemporalSubLayers : 1;7
8
if (param.internalCsp == X265_CSP_I420 && param.internalBitDepth <= 10)9
{10
11
12
/* The value of sps_max_dec_pic_buffering_minus1 HighestTid + 1 shall be less than13
* or equal to MaxDpbSize */14
- if (vps.maxDecPicBuffering > maxDpbSize)15
+ if (vps.maxDecPicBufferingvps.maxTempSubLayers - 1 > maxDpbSize)16
continue;17
18
/* For level 5 and higher levels, the value of CtbSizeY shall be equal to 32 or 64 */19
20
}21
22
/* The value of NumPocTotalCurr shall be less than or equal to 8 */23
- int numPocTotalCurr = param.maxNumReferences + vps.numReorderPics;24
- if (numPocTotalCurr > 8)25
+ int numPocTotalCurr = param.maxNumReferences + vps.numReorderPicsvps.maxTempSubLayers - 1;26
+ if (numPocTotalCurr > 10)27
{28
x265_log(¶m, X265_LOG_WARNING, "level %s detected, but NumPocTotalCurr (total references) is non-compliant\n", levelsi.name);29
vps.ptl.profileIdc = Profile::NONE;30
31
* circumstances it will be quite noisy */32
bool enforceLevel(x265_param& param, VPS& vps)33
{34
- vps.numReorderPics = (param.bBPyramid && param.bframes > 1) ? 2 : !!param.bframes;35
- vps.maxDecPicBuffering = X265_MIN(MAX_NUM_REF, X265_MAX(vps.numReorderPics + 2, (uint32_t)param.maxNumReferences) + 1);36
+ vps.maxTempSubLayers = !!param.bEnableTemporalSubLayers ? param.bEnableTemporalSubLayers : 1;37
+ for (uint32_t i = 0; i < vps.maxTempSubLayers; i++)38
+ {39
+ vps.numReorderPicsi = (i == 0) ? ((param.bBPyramid && param.bframes > 1) ? 2 : !!param.bframes) : i;40
+ vps.maxDecPicBufferingi = X265_MIN(MAX_NUM_REF, X265_MAX(vps.numReorderPicsi + 2, (uint32_t)param.maxNumReferences) + 1);41
+ }42
43
+ if (!!param.bEnableTemporalSubLayers)44
+ {45
+ for (int i = 0; i < MAX_T_LAYERS - 1; i++)46
+ {47
+ // a lower layer can not have higher value of numReorderPics than a higher layer48
+ if (vps.numReorderPicsi + 1 < vps.numReorderPicsi)49
+ {50
+ vps.numReorderPicsi + 1 = vps.numReorderPicsi;51
+ }52
+ // the value of numReorderPicsi shall be in the range of 0 to maxDecPicBufferingi - 1, inclusive53
+ if (vps.numReorderPicsi > vps.maxDecPicBufferingi - 1)54
+ {55
+ vps.maxDecPicBufferingi = vps.numReorderPicsi + 1;56
+ }57
+ // a lower layer can not have higher value of maxDecPicBuffering than a higher layer58
+ if (vps.maxDecPicBufferingi + 1 < vps.maxDecPicBufferingi)59
+ {60
+ vps.maxDecPicBufferingi + 1 = vps.maxDecPicBufferingi;61
+ }62
+ }63
+64
+ // the value of numReorderPicsi shall be in the range of 0 to maxDecPicBuffering i - 1, inclusive65
+ if (vps.numReorderPicsMAX_T_LAYERS - 1 > vps.maxDecPicBufferingMAX_T_LAYERS - 1 - 1)66
+ {67
+ vps.maxDecPicBufferingMAX_T_LAYERS - 1 = vps.numReorderPicsMAX_T_LAYERS - 1 + 1;68
+ }69
+ }70
/* no level specified by user, just auto-detect from the configuration */71
if (param.levelIdc <= 0)72
return true;73
74
}75
76
int savedRefCount = param.maxNumReferences;77
- while (vps.maxDecPicBuffering > maxDpbSize && param.maxNumReferences > 1)78
+ while (vps.maxDecPicBufferingvps.maxTempSubLayers - 1 > maxDpbSize && param.maxNumReferences > 1)79
{80
param.maxNumReferences--;81
- vps.maxDecPicBuffering = X265_MIN(MAX_NUM_REF, X265_MAX(vps.numReorderPics + 1, (uint32_t)param.maxNumReferences) + 1);82
+ vps.maxDecPicBufferingvps.maxTempSubLayers - 1 = X265_MIN(MAX_NUM_REF, X265_MAX(vps.numReorderPicsvps.maxTempSubLayers - 1 + 1, (uint32_t)param.maxNumReferences) + 1);83
}84
if (param.maxNumReferences != savedRefCount)85
x265_log(¶m, X265_LOG_WARNING, "Lowering max references to %d to meet level requirement\n", param.maxNumReferences);86
x265_3.5.tar.gz/source/encoder/motion.cpp -> x265_3.6.tar.gz/source/encoder/motion.cpp
Changed
33
1
2
X265_CHECK(!bChromaSATD, "chroma distortion measurements impossible in this code path\n");3
}4
5
+/* Called by lookahead, luma only, no use of PicYuv */6
+void MotionEstimate::setSourcePU(pixel *fencY, intptr_t stride, intptr_t offset, int pwidth, int pheight, const int method, const int refine)7
+{8
+ partEnum = partitionFromSizes(pwidth, pheight);9
+ X265_CHECK(LUMA_4x4 != partEnum, "4x4 inter partition detected!\n");10
+ sad = primitives.pupartEnum.sad;11
+ ads = primitives.pupartEnum.ads;12
+ satd = primitives.pupartEnum.satd;13
+ sad_x3 = primitives.pupartEnum.sad_x3;14
+ sad_x4 = primitives.pupartEnum.sad_x4;15
+16
+17
+ blockwidth = pwidth;18
+ blockOffset = offset;19
+ absPartIdx = ctuAddr = -1;20
+21
+ /* Search params */22
+ searchMethod = method;23
+ subpelRefine = refine;24
+25
+ /* copy PU block into cache */26
+ primitives.pupartEnum.copy_pp(fencPUYuv.m_buf0, FENC_STRIDE, fencY + offset, stride);27
+ X265_CHECK(!bChromaSATD, "chroma distortion measurements impossible in this code path\n");28
+}29
+30
/* Called by Search::predInterSearch() or --pme equivalent, chroma residual might be considered */31
void MotionEstimate::setSourcePU(const Yuv& srcFencYuv, int _ctuAddr, int cuPartIdx, int puPartIdx, int pwidth, int pheight, const int method, const int refine, bool bChroma)32
{33
x265_3.5.tar.gz/source/encoder/motion.h -> x265_3.6.tar.gz/source/encoder/motion.h
Changed
10
1
2
void init(int csp);3
4
/* Methods called at slice setup */5
-6
+ void setSourcePU(pixel *fencY, intptr_t stride, intptr_t offset, int pwidth, int pheight, const int searchMethod, const int subpelRefine);7
void setSourcePU(pixel *fencY, intptr_t stride, intptr_t offset, int pwidth, int pheight, const int searchMethod, const int searchL0, const int searchL1, const int subpelRefine);8
void setSourcePU(const Yuv& srcFencYuv, int ctuAddr, int cuPartIdx, int puPartIdx, int pwidth, int pheight, const int searchMethod, const int subpelRefine, bool bChroma);9
10
x265_3.5.tar.gz/source/encoder/nal.cpp -> x265_3.6.tar.gz/source/encoder/nal.cpp
Changed
19
1
2
other.m_buffer = X265_MALLOC(uint8_t, m_allocSize);3
}4
5
-void NALList::serialize(NalUnitType nalUnitType, const Bitstream& bs)6
+void NALList::serialize(NalUnitType nalUnitType, const Bitstream& bs, uint8_t temporalID)7
{8
static const char startCodePrefix = { 0, 0, 0, 1 };9
10
11
* nuh_reserved_zero_6bits 6-bits12
* nuh_temporal_id_plus1 3-bits */13
outbytes++ = (uint8_t)nalUnitType << 1;14
- outbytes++ = 1 + (nalUnitType == NAL_UNIT_CODED_SLICE_TSA_N);15
+ outbytes++ = temporalID;16
17
/* 7.4.1 ...18
* Within the NAL unit, the following three-byte sequences shall not occur at19
x265_3.5.tar.gz/source/encoder/nal.h -> x265_3.6.tar.gz/source/encoder/nal.h
Changed
10
1
2
3
void takeContents(NALList& other);4
5
- void serialize(NalUnitType nalUnitType, const Bitstream& bs);6
+ void serialize(NalUnitType nalUnitType, const Bitstream& bs, uint8_t temporalID = 1);7
8
uint32_t serializeSubstreams(uint32_t* streamSizeBytes, uint32_t streamCount, const Bitstream* streams);9
};10
x265_3.5.tar.gz/source/encoder/ratecontrol.cpp -> x265_3.6.tar.gz/source/encoder/ratecontrol.cpp
Changed
1457
1
2
#define BR_SHIFT 63
#define CPB_SHIFT 44
5
+#define SHARED_DATA_ALIGNMENT 4 ///< 4btye, 32bit6
+#define CUTREE_SHARED_MEM_NAME "cutree"7
+#define GOP_CNT_CU_TREE 38
+9
using namespace X265_NS;10
11
/* Amortize the partial cost of I frames over the next N frames */12
13
return output;14
}15
16
+typedef struct CUTreeSharedDataItem17
+{18
+ uint8_t *type;19
+ uint16_t *stats;20
+}CUTreeSharedDataItem;21
+22
+void static ReadSharedCUTreeData(void *dst, void *src, int32_t size)23
+{24
+ CUTreeSharedDataItem *statsDst = reinterpret_cast<CUTreeSharedDataItem *>(dst);25
+ uint8_t *typeSrc = reinterpret_cast<uint8_t *>(src);26
+ *statsDst->type = *typeSrc;27
+28
+ ///< for memory alignment, the type will take 32bit in the shared memory29
+ int32_t offset = (sizeof(*statsDst->type) + SHARED_DATA_ALIGNMENT - 1) & ~(SHARED_DATA_ALIGNMENT - 1);30
+ uint16_t *statsSrc = reinterpret_cast<uint16_t *>(typeSrc + offset);31
+ memcpy(statsDst->stats, statsSrc, size - offset);32
+}33
+34
+void static WriteSharedCUTreeData(void *dst, void *src, int32_t size)35
+{36
+ CUTreeSharedDataItem *statsSrc = reinterpret_cast<CUTreeSharedDataItem *>(src);37
+ uint8_t *typeDst = reinterpret_cast<uint8_t *>(dst);38
+ *typeDst = *statsSrc->type;39
+40
+ ///< for memory alignment, the type will take 32bit in the shared memory41
+ int32_t offset = (sizeof(*statsSrc->type) + SHARED_DATA_ALIGNMENT - 1) & ~(SHARED_DATA_ALIGNMENT - 1);42
+ uint16_t *statsDst = reinterpret_cast<uint16_t *>(typeDst + offset);43
+ memcpy(statsDst, statsSrc->stats, size - offset);44
+}45
+46
+47
inline double qScale2bits(RateControlEntry *rce, double qScale)48
{49
if (qScale < 0.1)50
51
m_lastAbrResetPoc = -1;52
m_statFileOut = NULL;53
m_cutreeStatFileOut = m_cutreeStatFileIn = NULL;54
+ m_cutreeShrMem = NULL;55
m_rce2Pass = NULL;56
m_encOrder = NULL;57
m_lastBsliceSatdCost = 0;58
59
m_initVbv = false;60
m_singleFrameVbv = 0;61
m_rateTolerance = 1.0;62
+ m_encodedSegmentBits = 0;63
+ m_segDur = 0;64
65
if (m_param->rc.vbvBufferSize)66
{67
68
m_cuTreeStats.qpBufferi = NULL;69
}70
71
-bool RateControl::init(const SPS& sps)72
+bool RateControl::initCUTreeSharedMem()73
{74
- if (m_isVbv && !m_initVbv)75
- {76
- /* We don't support changing the ABR bitrate right now,77
- * so if the stream starts as CBR, keep it CBR. */78
- if (m_param->rc.vbvBufferSize < (int)(m_param->rc.vbvMaxBitrate / m_fps))79
+ if (!m_cutreeShrMem) {80
+ m_cutreeShrMem = new RingMem();81
+ if (!m_cutreeShrMem)82
{83
- m_param->rc.vbvBufferSize = (int)(m_param->rc.vbvMaxBitrate / m_fps);84
- x265_log(m_param, X265_LOG_WARNING, "VBV buffer size cannot be smaller than one frame, using %d kbit\n",85
- m_param->rc.vbvBufferSize);86
+ return false;87
}88
- int vbvBufferSize = m_param->rc.vbvBufferSize * 1000;89
- int vbvMaxBitrate = m_param->rc.vbvMaxBitrate * 1000;90
91
- if (m_param->bEmitHRDSEI && !m_param->decoderVbvMaxRate)92
+ ///< now cutree data form at most 3 gops would be stored in the shared memory at the same time93
+ int32_t itemSize = (sizeof(uint8_t) + SHARED_DATA_ALIGNMENT - 1) & ~(SHARED_DATA_ALIGNMENT - 1);94
+ if (m_param->rc.qgSize == 8)95
{96
- const HRDInfo* hrd = &sps.vuiParameters.hrdParameters;97
- vbvBufferSize = hrd->cpbSizeValue << (hrd->cpbSizeScale + CPB_SHIFT);98
- vbvMaxBitrate = hrd->bitRateValue << (hrd->bitRateScale + BR_SHIFT);99
+ itemSize += sizeof(uint16_t) * m_ncu * 4;100
}101
- m_bufferRate = vbvMaxBitrate / m_fps;102
- m_vbvMaxRate = vbvMaxBitrate;103
- m_bufferSize = vbvBufferSize;104
- m_singleFrameVbv = m_bufferRate * 1.1 > m_bufferSize;105
+ else106
+ {107
+ itemSize += sizeof(uint16_t) * m_ncu;108
+ }109
+110
+ int32_t itemCnt = X265_MIN(m_param->keyframeMax, (int)(m_fps + 0.5));111
+ itemCnt *= GOP_CNT_CU_TREE;112
113
- if (m_param->rc.vbvBufferInit > 1.)114
- m_param->rc.vbvBufferInit = x265_clip3(0.0, 1.0, m_param->rc.vbvBufferInit / m_param->rc.vbvBufferSize);115
- if (m_param->vbvBufferEnd > 1.)116
- m_param->vbvBufferEnd = x265_clip3(0.0, 1.0, m_param->vbvBufferEnd / m_param->rc.vbvBufferSize);117
- if (m_param->vbvEndFrameAdjust > 1.)118
- m_param->vbvEndFrameAdjust = x265_clip3(0.0, 1.0, m_param->vbvEndFrameAdjust);119
- m_param->rc.vbvBufferInit = x265_clip3(0.0, 1.0, X265_MAX(m_param->rc.vbvBufferInit, m_bufferRate / m_bufferSize));120
- m_bufferFillFinal = m_bufferSize * m_param->rc.vbvBufferInit;121
- m_bufferFillActual = m_bufferFillFinal;122
- m_bufferExcess = 0;123
- m_minBufferFill = m_param->minVbvFullness / 100;124
- m_maxBufferFill = 1 - (m_param->maxVbvFullness / 100);125
- m_initVbv = true;126
+ char shrnameMAX_SHR_NAME_LEN = { 0 };127
+ strcpy(shrname, m_param->rc.sharedMemName);128
+ strcat(shrname, CUTREE_SHARED_MEM_NAME);129
+130
+ if (!m_cutreeShrMem->init(itemSize, itemCnt, shrname))131
+ {132
+ return false;133
+ }134
}135
136
+ return true;137
+}138
+139
+void RateControl::initVBV(const SPS& sps)140
+{141
+ /* We don't support changing the ABR bitrate right now,142
+ * so if the stream starts as CBR, keep it CBR. */143
+ if (m_param->rc.vbvBufferSize < (int)(m_param->rc.vbvMaxBitrate / m_fps))144
+ {145
+ m_param->rc.vbvBufferSize = (int)(m_param->rc.vbvMaxBitrate / m_fps);146
+ x265_log(m_param, X265_LOG_WARNING, "VBV buffer size cannot be smaller than one frame, using %d kbit\n",147
+ m_param->rc.vbvBufferSize);148
+ }149
+ int vbvBufferSize = m_param->rc.vbvBufferSize * 1000;150
+ int vbvMaxBitrate = m_param->rc.vbvMaxBitrate * 1000;151
+152
+ if (m_param->bEmitHRDSEI && !m_param->decoderVbvMaxRate)153
+ {154
+ const HRDInfo* hrd = &sps.vuiParameters.hrdParameters;155
+ vbvBufferSize = hrd->cpbSizeValue << (hrd->cpbSizeScale + CPB_SHIFT);156
+ vbvMaxBitrate = hrd->bitRateValue << (hrd->bitRateScale + BR_SHIFT);157
+ }158
+ m_bufferRate = vbvMaxBitrate / m_fps;159
+ m_vbvMaxRate = vbvMaxBitrate;160
+ m_bufferSize = vbvBufferSize;161
+ m_singleFrameVbv = m_bufferRate * 1.1 > m_bufferSize;162
+163
+ if (m_param->rc.vbvBufferInit > 1.)164
+ m_param->rc.vbvBufferInit = x265_clip3(0.0, 1.0, m_param->rc.vbvBufferInit / m_param->rc.vbvBufferSize);165
+ if (m_param->vbvBufferEnd > 1.)166
+ m_param->vbvBufferEnd = x265_clip3(0.0, 1.0, m_param->vbvBufferEnd / m_param->rc.vbvBufferSize);167
+ if (m_param->vbvEndFrameAdjust > 1.)168
+ m_param->vbvEndFrameAdjust = x265_clip3(0.0, 1.0, m_param->vbvEndFrameAdjust);169
+ m_param->rc.vbvBufferInit = x265_clip3(0.0, 1.0, X265_MAX(m_param->rc.vbvBufferInit, m_bufferRate / m_bufferSize));170
+ m_bufferFillFinal = m_bufferSize * m_param->rc.vbvBufferInit;171
+ m_bufferFillActual = m_bufferFillFinal;172
+ m_bufferExcess = 0;173
+ m_minBufferFill = m_param->minVbvFullness / 100;174
+ m_maxBufferFill = 1 - (m_param->maxVbvFullness / 100);175
+ m_initVbv = true;176
+}177
+178
+bool RateControl::init(const SPS& sps)179
+{180
+ if (m_isVbv && !m_initVbv)181
+ initVBV(sps);182
+183
if (!m_param->bResetZoneConfig && (m_relativeComplexity == NULL))184
{185
m_relativeComplexity = X265_MALLOC(double, m_param->reconfigWindowSize);186
187
188
m_totalBits = 0;189
m_encodedBits = 0;190
+ m_encodedSegmentBits = 0;191
m_framesDone = 0;192
+ m_segDur = 0;193
m_residualCost = 0;194
m_partialResidualCost = 0;195
m_amortizeFraction = 0.85;196
197
/* Load stat file and init 2pass algo */198
if (m_param->rc.bStatRead)199
{200
- m_expectedBitsSum = 0;201
- char *p, *statsIn, *statsBuf;202
- /* read 1st pass stats */203
- statsIn = statsBuf = x265_slurp_file(fileName);204
- if (!statsBuf)205
- return false;206
- if (m_param->rc.cuTree)207
+ if (X265_SHARE_MODE_FILE == m_param->rc.dataShareMode)208
{209
- char *tmpFile = strcatFilename(fileName, ".cutree");210
- if (!tmpFile)211
+ m_expectedBitsSum = 0;212
+ char *p, *statsIn, *statsBuf;213
+ /* read 1st pass stats */214
+ statsIn = statsBuf = x265_slurp_file(fileName);215
+ if (!statsBuf)216
return false;217
- m_cutreeStatFileIn = x265_fopen(tmpFile, "rb");218
- X265_FREE(tmpFile);219
- if (!m_cutreeStatFileIn)220
+ if (m_param->rc.cuTree)221
{222
- x265_log_file(m_param, X265_LOG_ERROR, "can't open stats file %s.cutree\n", fileName);223
- return false;224
+ char *tmpFile = strcatFilename(fileName, ".cutree");225
+ if (!tmpFile)226
+ return false;227
+ m_cutreeStatFileIn = x265_fopen(tmpFile, "rb");228
+ X265_FREE(tmpFile);229
+ if (!m_cutreeStatFileIn)230
+ {231
+ x265_log_file(m_param, X265_LOG_ERROR, "can't open stats file %s.cutree\n", fileName);232
+ return false;233
+ }234
}235
- }236
237
- /* check whether 1st pass options were compatible with current options */238
- if (strncmp(statsBuf, "#options:", 9))239
- {240
- x265_log(m_param, X265_LOG_ERROR,"options list in stats file not valid\n");241
- return false;242
- }243
- {244
- int i, j, m;245
- uint32_t k , l;246
- bool bErr = false;247
- char *opts = statsBuf;248
- statsIn = strchr(statsBuf, '\n');249
- if (!statsIn)250
- {251
- x265_log(m_param, X265_LOG_ERROR, "Malformed stats file\n");252
- return false;253
- }254
- *statsIn = '\0';255
- statsIn++;256
- if ((p = strstr(opts, " input-res=")) == 0 || sscanf(p, " input-res=%dx%d", &i, &j) != 2)257
- {258
- x265_log(m_param, X265_LOG_ERROR, "Resolution specified in stats file not valid\n");259
- return false;260
- }261
- if ((p = strstr(opts, " fps=")) == 0 || sscanf(p, " fps=%u/%u", &k, &l) != 2)262
- {263
- x265_log(m_param, X265_LOG_ERROR, "fps specified in stats file not valid\n");264
- return false;265
- }266
- if (((p = strstr(opts, " vbv-maxrate=")) == 0 || sscanf(p, " vbv-maxrate=%d", &m) != 1) && m_param->rc.rateControlMode == X265_RC_CRF)267
- {268
- x265_log(m_param, X265_LOG_ERROR, "Constant rate-factor is incompatible with 2pass without vbv-maxrate in the previous pass\n");269
- return false;270
- }271
- if (k != m_param->fpsNum || l != m_param->fpsDenom)272
+ /* check whether 1st pass options were compatible with current options */273
+ if (strncmp(statsBuf, "#options:", 9))274
{275
- x265_log(m_param, X265_LOG_ERROR, "fps mismatch with 1st pass (%u/%u vs %u/%u)\n",276
- m_param->fpsNum, m_param->fpsDenom, k, l);277
+ x265_log(m_param, X265_LOG_ERROR, "options list in stats file not valid\n");278
return false;279
}280
- if (m_param->analysisMultiPassRefine)281
{282
- p = strstr(opts, "ref=");283
- sscanf(p, "ref=%d", &i);284
- if (i > m_param->maxNumReferences)285
+ int i, j, m;286
+ uint32_t k, l;287
+ bool bErr = false;288
+ char *opts = statsBuf;289
+ statsIn = strchr(statsBuf, '\n');290
+ if (!statsIn)291
{292
- x265_log(m_param, X265_LOG_ERROR, "maxNumReferences cannot be less than 1st pass (%d vs %d)\n",293
- i, m_param->maxNumReferences);294
+ x265_log(m_param, X265_LOG_ERROR, "Malformed stats file\n");295
return false;296
}297
- }298
- if (m_param->analysisMultiPassRefine || m_param->analysisMultiPassDistortion)299
- {300
- p = strstr(opts, "ctu=");301
- sscanf(p, "ctu=%u", &k);302
- if (k != m_param->maxCUSize)303
+ *statsIn = '\0';304
+ statsIn++;305
+ if ((p = strstr(opts, " input-res=")) == 0 || sscanf(p, " input-res=%dx%d", &i, &j) != 2)306
{307
- x265_log(m_param, X265_LOG_ERROR, "maxCUSize mismatch with 1st pass (%u vs %u)\n",308
- k, m_param->maxCUSize);309
+ x265_log(m_param, X265_LOG_ERROR, "Resolution specified in stats file not valid\n");310
return false;311
}312
+ if ((p = strstr(opts, " fps=")) == 0 || sscanf(p, " fps=%u/%u", &k, &l) != 2)313
+ {314
+ x265_log(m_param, X265_LOG_ERROR, "fps specified in stats file not valid\n");315
+ return false;316
+ }317
+ if (((p = strstr(opts, " vbv-maxrate=")) == 0 || sscanf(p, " vbv-maxrate=%d", &m) != 1) && m_param->rc.rateControlMode == X265_RC_CRF)318
+ {319
+ x265_log(m_param, X265_LOG_ERROR, "Constant rate-factor is incompatible with 2pass without vbv-maxrate in the previous pass\n");320
+ return false;321
+ }322
+ if (k != m_param->fpsNum || l != m_param->fpsDenom)323
+ {324
+ x265_log(m_param, X265_LOG_ERROR, "fps mismatch with 1st pass (%u/%u vs %u/%u)\n",325
+ m_param->fpsNum, m_param->fpsDenom, k, l);326
+ return false;327
+ }328
+ if (m_param->analysisMultiPassRefine)329
+ {330
+ p = strstr(opts, "ref=");331
+ sscanf(p, "ref=%d", &i);332
+ if (i > m_param->maxNumReferences)333
+ {334
+ x265_log(m_param, X265_LOG_ERROR, "maxNumReferences cannot be less than 1st pass (%d vs %d)\n",335
+ i, m_param->maxNumReferences);336
+ return false;337
+ }338
+ }339
+ if (m_param->analysisMultiPassRefine || m_param->analysisMultiPassDistortion)340
+ {341
+ p = strstr(opts, "ctu=");342
+ sscanf(p, "ctu=%u", &k);343
+ if (k != m_param->maxCUSize)344
+ {345
+ x265_log(m_param, X265_LOG_ERROR, "maxCUSize mismatch with 1st pass (%u vs %u)\n",346
+ k, m_param->maxCUSize);347
+ return false;348
+ }349
+ }350
+ CMP_OPT_FIRST_PASS("bitdepth", m_param->internalBitDepth);351
+ CMP_OPT_FIRST_PASS("weightp", m_param->bEnableWeightedPred);352
+ CMP_OPT_FIRST_PASS("bframes", m_param->bframes);353
+ CMP_OPT_FIRST_PASS("b-pyramid", m_param->bBPyramid);354
+ CMP_OPT_FIRST_PASS("open-gop", m_param->bOpenGOP);355
+ CMP_OPT_FIRST_PASS(" keyint", m_param->keyframeMax);356
+ CMP_OPT_FIRST_PASS("scenecut", m_param->scenecutThreshold);357
+ CMP_OPT_FIRST_PASS("intra-refresh", m_param->bIntraRefresh);358
+ CMP_OPT_FIRST_PASS("frame-dup", m_param->bEnableFrameDuplication);359
+ if (m_param->bMultiPassOptRPS)360
+ {361
+ CMP_OPT_FIRST_PASS("multi-pass-opt-rps", m_param->bMultiPassOptRPS);362
+ CMP_OPT_FIRST_PASS("repeat-headers", m_param->bRepeatHeaders);363
+ CMP_OPT_FIRST_PASS("min-keyint", m_param->keyframeMin);364
+ }365
+366
+ if ((p = strstr(opts, "b-adapt=")) != 0 && sscanf(p, "b-adapt=%d", &i) && i >= X265_B_ADAPT_NONE && i <= X265_B_ADAPT_TRELLIS)367
+ {368
+ m_param->bFrameAdaptive = i;369
+ }370
+ else if (m_param->bframes)371
+ {372
+ x265_log(m_param, X265_LOG_ERROR, "b-adapt method specified in stats file not valid\n");373
+ return false;374
+ }375
+376
+ if ((p = strstr(opts, "rc-lookahead=")) != 0 && sscanf(p, "rc-lookahead=%d", &i))377
+ m_param->lookaheadDepth = i;378
}379
- CMP_OPT_FIRST_PASS("bitdepth", m_param->internalBitDepth);380
- CMP_OPT_FIRST_PASS("weightp", m_param->bEnableWeightedPred);381
- CMP_OPT_FIRST_PASS("bframes", m_param->bframes);382
- CMP_OPT_FIRST_PASS("b-pyramid", m_param->bBPyramid);383
- CMP_OPT_FIRST_PASS("open-gop", m_param->bOpenGOP);384
- CMP_OPT_FIRST_PASS(" keyint", m_param->keyframeMax);385
- CMP_OPT_FIRST_PASS("scenecut", m_param->scenecutThreshold);386
- CMP_OPT_FIRST_PASS("intra-refresh", m_param->bIntraRefresh);387
- CMP_OPT_FIRST_PASS("frame-dup", m_param->bEnableFrameDuplication);388
- if (m_param->bMultiPassOptRPS)389
+ /* find number of pics */390
+ p = statsIn;391
+ int numEntries;392
+ for (numEntries = -1; p; numEntries++)393
+ p = strchr(p + 1, ';');394
+ if (!numEntries)395
{396
- CMP_OPT_FIRST_PASS("multi-pass-opt-rps", m_param->bMultiPassOptRPS);397
- CMP_OPT_FIRST_PASS("repeat-headers", m_param->bRepeatHeaders);398
- CMP_OPT_FIRST_PASS("min-keyint", m_param->keyframeMin);399
+ x265_log(m_param, X265_LOG_ERROR, "empty stats file\n");400
+ return false;401
}402
+ m_numEntries = numEntries;403
404
- if ((p = strstr(opts, "b-adapt=")) != 0 && sscanf(p, "b-adapt=%d", &i) && i >= X265_B_ADAPT_NONE && i <= X265_B_ADAPT_TRELLIS)405
+ if (m_param->totalFrames < m_numEntries && m_param->totalFrames > 0)406
{407
- m_param->bFrameAdaptive = i;408
+ x265_log(m_param, X265_LOG_WARNING, "2nd pass has fewer frames than 1st pass (%d vs %d)\n",409
+ m_param->totalFrames, m_numEntries);410
}411
- else if (m_param->bframes)412
+ if (m_param->totalFrames > m_numEntries && !m_param->bEnableFrameDuplication)413
{414
- x265_log(m_param, X265_LOG_ERROR, "b-adapt method specified in stats file not valid\n");415
+ x265_log(m_param, X265_LOG_ERROR, "2nd pass has more frames than 1st pass (%d vs %d)\n",416
+ m_param->totalFrames, m_numEntries);417
return false;418
}419
420
- if ((p = strstr(opts, "rc-lookahead=")) != 0 && sscanf(p, "rc-lookahead=%d", &i))421
- m_param->lookaheadDepth = i;422
- }423
- /* find number of pics */424
- p = statsIn;425
- int numEntries;426
- for (numEntries = -1; p; numEntries++)427
- p = strchr(p + 1, ';');428
- if (!numEntries)429
- {430
- x265_log(m_param, X265_LOG_ERROR, "empty stats file\n");431
- return false;432
- }433
- m_numEntries = numEntries;434
-435
- if (m_param->totalFrames < m_numEntries && m_param->totalFrames > 0)436
- {437
- x265_log(m_param, X265_LOG_WARNING, "2nd pass has fewer frames than 1st pass (%d vs %d)\n",438
- m_param->totalFrames, m_numEntries);439
- }440
- if (m_param->totalFrames > m_numEntries && !m_param->bEnableFrameDuplication)441
- {442
- x265_log(m_param, X265_LOG_ERROR, "2nd pass has more frames than 1st pass (%d vs %d)\n",443
- m_param->totalFrames, m_numEntries);444
- return false;445
- }446
-447
- m_rce2Pass = X265_MALLOC(RateControlEntry, m_numEntries);448
- if (!m_rce2Pass)449
- {450
- x265_log(m_param, X265_LOG_ERROR, "Rce Entries for 2 pass cannot be allocated\n");451
- return false;452
- }453
- m_encOrder = X265_MALLOC(int, m_numEntries);454
- if (!m_encOrder)455
- {456
- x265_log(m_param, X265_LOG_ERROR, "Encode order for 2 pass cannot be allocated\n");457
- return false;458
- }459
- /* init all to skipped p frames */460
- for (int i = 0; i < m_numEntries; i++)461
- {462
- RateControlEntry *rce = &m_rce2Passi;463
- rce->sliceType = P_SLICE;464
- rce->qScale = rce->newQScale = x265_qp2qScale(20);465
- rce->miscBits = m_ncu + 10;466
- rce->newQp = 0;467
- }468
- /* read stats */469
- p = statsIn;470
- double totalQpAq = 0;471
- for (int i = 0; i < m_numEntries; i++)472
- {473
- RateControlEntry *rce, *rcePocOrder;474
- int frameNumber;475
- int encodeOrder;476
- char picType;477
- int e;478
- char *next;479
- double qpRc, qpAq, qNoVbv, qRceq;480
- next = strstr(p, ";");481
- if (next)482
- *next++ = 0;483
- e = sscanf(p, " in:%d out:%d", &frameNumber, &encodeOrder);484
- if (frameNumber < 0 || frameNumber >= m_numEntries)485
+ m_rce2Pass = X265_MALLOC(RateControlEntry, m_numEntries);486
+ if (!m_rce2Pass)487
{488
- x265_log(m_param, X265_LOG_ERROR, "bad frame number (%d) at stats line %d\n", frameNumber, i);489
+ x265_log(m_param, X265_LOG_ERROR, "Rce Entries for 2 pass cannot be allocated\n");490
return false;491
}492
- rce = &m_rce2PassencodeOrder;493
- rcePocOrder = &m_rce2PassframeNumber;494
- m_encOrderframeNumber = encodeOrder;495
- if (!m_param->bMultiPassOptRPS)496
- {497
- int scenecut = 0;498
- e += sscanf(p, " in:%*d out:%*d type:%c q:%lf q-aq:%lf q-noVbv:%lf q-Rceq:%lf tex:%d mv:%d misc:%d icu:%lf pcu:%lf scu:%lf sc:%d",499
- &picType, &qpRc, &qpAq, &qNoVbv, &qRceq, &rce->coeffBits,500
- &rce->mvBits, &rce->miscBits, &rce->iCuCount, &rce->pCuCount,501
- &rce->skipCuCount, &scenecut);502
- rcePocOrder->scenecut = scenecut != 0;503
+ m_encOrder = X265_MALLOC(int, m_numEntries);504
+ if (!m_encOrder)505
+ {506
+ x265_log(m_param, X265_LOG_ERROR, "Encode order for 2 pass cannot be allocated\n");507
+ return false;508
}509
- else510
+ /* init all to skipped p frames */511
+ for (int i = 0; i < m_numEntries; i++)512
{513
- char deltaPOC128;514
- char bUsed40;515
- memset(deltaPOC, 0, sizeof(deltaPOC));516
- memset(bUsed, 0, sizeof(bUsed));517
- e += sscanf(p, " in:%*d out:%*d type:%c q:%lf q-aq:%lf q-noVbv:%lf q-Rceq:%lf tex:%d mv:%d misc:%d icu:%lf pcu:%lf scu:%lf nump:%d numnegp:%d numposp:%d deltapoc:%s bused:%s",518
- &picType, &qpRc, &qpAq, &qNoVbv, &qRceq, &rce->coeffBits,519
- &rce->mvBits, &rce->miscBits, &rce->iCuCount, &rce->pCuCount,520
- &rce->skipCuCount, &rce->rpsData.numberOfPictures, &rce->rpsData.numberOfNegativePictures, &rce->rpsData.numberOfPositivePictures, deltaPOC, bUsed);521
- splitdeltaPOC(deltaPOC, rce);522
- splitbUsed(bUsed, rce);523
- rce->rpsIdx = -1;524
- }525
- rce->keptAsRef = true;526
- rce->isIdr = false;527
- if (picType == 'b' || picType == 'p')528
- rce->keptAsRef = false;529
- if (picType == 'I')530
- rce->isIdr = true;531
- if (picType == 'I' || picType == 'i')532
- rce->sliceType = I_SLICE;533
- else if (picType == 'P' || picType == 'p')534
+ RateControlEntry *rce = &m_rce2Passi;535
rce->sliceType = P_SLICE;536
- else if (picType == 'B' || picType == 'b')537
- rce->sliceType = B_SLICE;538
- else539
- e = -1;540
- if (e < 10)541
+ rce->qScale = rce->newQScale = x265_qp2qScale(20);542
+ rce->miscBits = m_ncu + 10;543
+ rce->newQp = 0;544
+ }545
+ /* read stats */546
+ p = statsIn;547
+ double totalQpAq = 0;548
+ for (int i = 0; i < m_numEntries; i++)549
+ {550
+ RateControlEntry *rce, *rcePocOrder;551
+ int frameNumber;552
+ int encodeOrder;553
+ char picType;554
+ int e;555
+ char *next;556
+ double qpRc, qpAq, qNoVbv, qRceq;557
+ next = strstr(p, ";");558
+ if (next)559
+ *next++ = 0;560
+ e = sscanf(p, " in:%d out:%d", &frameNumber, &encodeOrder);561
+ if (frameNumber < 0 || frameNumber >= m_numEntries)562
+ {563
+ x265_log(m_param, X265_LOG_ERROR, "bad frame number (%d) at stats line %d\n", frameNumber, i);564
+ return false;565
+ }566
+ rce = &m_rce2PassencodeOrder;567
+ rcePocOrder = &m_rce2PassframeNumber;568
+ m_encOrderframeNumber = encodeOrder;569
+ if (!m_param->bMultiPassOptRPS)570
+ {571
+ int scenecut = 0;572
+ e += sscanf(p, " in:%*d out:%*d type:%c q:%lf q-aq:%lf q-noVbv:%lf q-Rceq:%lf tex:%d mv:%d misc:%d icu:%lf pcu:%lf scu:%lf sc:%d",573
+ &picType, &qpRc, &qpAq, &qNoVbv, &qRceq, &rce->coeffBits,574
+ &rce->mvBits, &rce->miscBits, &rce->iCuCount, &rce->pCuCount,575
+ &rce->skipCuCount, &scenecut);576
+ rcePocOrder->scenecut = scenecut != 0;577
+ }578
+ else579
+ {580
+ char deltaPOC128;581
+ char bUsed40;582
+ memset(deltaPOC, 0, sizeof(deltaPOC));583
+ memset(bUsed, 0, sizeof(bUsed));584
+ e += sscanf(p, " in:%*d out:%*d type:%c q:%lf q-aq:%lf q-noVbv:%lf q-Rceq:%lf tex:%d mv:%d misc:%d icu:%lf pcu:%lf scu:%lf nump:%d numnegp:%d numposp:%d deltapoc:%s bused:%s",585
+ &picType, &qpRc, &qpAq, &qNoVbv, &qRceq, &rce->coeffBits,586
+ &rce->mvBits, &rce->miscBits, &rce->iCuCount, &rce->pCuCount,587
+ &rce->skipCuCount, &rce->rpsData.numberOfPictures, &rce->rpsData.numberOfNegativePictures, &rce->rpsData.numberOfPositivePictures, deltaPOC, bUsed);588
+ splitdeltaPOC(deltaPOC, rce);589
+ splitbUsed(bUsed, rce);590
+ rce->rpsIdx = -1;591
+ }592
+ rce->keptAsRef = true;593
+ rce->isIdr = false;594
+ if (picType == 'b' || picType == 'p')595
+ rce->keptAsRef = false;596
+ if (picType == 'I')597
+ rce->isIdr = true;598
+ if (picType == 'I' || picType == 'i')599
+ rce->sliceType = I_SLICE;600
+ else if (picType == 'P' || picType == 'p')601
+ rce->sliceType = P_SLICE;602
+ else if (picType == 'B' || picType == 'b')603
+ rce->sliceType = B_SLICE;604
+ else605
+ e = -1;606
+ if (e < 10)607
+ {608
+ x265_log(m_param, X265_LOG_ERROR, "statistics are damaged at line %d, parser out=%d\n", i, e);609
+ return false;610
+ }611
+ rce->qScale = rce->newQScale = x265_qp2qScale(qpRc);612
+ totalQpAq += qpAq;613
+ rce->qpNoVbv = qNoVbv;614
+ rce->qpaRc = qpRc;615
+ rce->qpAq = qpAq;616
+ rce->qRceq = qRceq;617
+ p = next;618
+ }619
+ X265_FREE(statsBuf);620
+ if (m_param->rc.rateControlMode != X265_RC_CQP)621
+ {622
+ m_start = 0;623
+ m_isQpModified = true;624
+ if (!initPass2())625
+ return false;626
+ } /* else we're using constant quant, so no need to run the bitrate allocation */627
+ }628
+ else // X265_SHARE_MODE_SHAREDMEM == m_param->rc.dataShareMode629
+ {630
+ if (m_param->rc.cuTree)631
{632
- x265_log(m_param, X265_LOG_ERROR, "statistics are damaged at line %d, parser out=%d\n", i, e);633
- return false;634
+ if (!initCUTreeSharedMem())635
+ {636
+ return false;637
+ }638
}639
- rce->qScale = rce->newQScale = x265_qp2qScale(qpRc);640
- totalQpAq += qpAq;641
- rce->qpNoVbv = qNoVbv;642
- rce->qpaRc = qpRc;643
- rce->qpAq = qpAq;644
- rce->qRceq = qRceq;645
- p = next;646
- }647
- X265_FREE(statsBuf);648
- if (m_param->rc.rateControlMode != X265_RC_CQP)649
- {650
- m_start = 0;651
- m_isQpModified = true;652
- if (!initPass2())653
- return false;654
- } /* else we're using constant quant, so no need to run the bitrate allocation */655
+ }656
}657
/* Open output file */658
/* If input and output files are the same, output to a temp file659
660
X265_FREE(p);661
if (m_param->rc.cuTree && !m_param->rc.bStatRead)662
{663
- statFileTmpname = strcatFilename(fileName, ".cutree.temp");664
- if (!statFileTmpname)665
- return false;666
- m_cutreeStatFileOut = x265_fopen(statFileTmpname, "wb");667
- X265_FREE(statFileTmpname);668
- if (!m_cutreeStatFileOut)669
+ if (X265_SHARE_MODE_FILE == m_param->rc.dataShareMode)670
{671
- x265_log_file(m_param, X265_LOG_ERROR, "can't open mbtree stats file %s.cutree.temp\n", fileName);672
- return false;673
+ statFileTmpname = strcatFilename(fileName, ".cutree.temp");674
+ if (!statFileTmpname)675
+ return false;676
+ m_cutreeStatFileOut = x265_fopen(statFileTmpname, "wb");677
+ X265_FREE(statFileTmpname);678
+ if (!m_cutreeStatFileOut)679
+ {680
+ x265_log_file(m_param, X265_LOG_ERROR, "can't open mbtree stats file %s.cutree.temp\n", fileName);681
+ return false;682
+ }683
+ }684
+ else // X265_SHARE_MODE_SHAREDMEM == m_param->rc.dataShareMode685
+ {686
+ if (!initCUTreeSharedMem())687
+ {688
+ return false;689
+ }690
}691
}692
}693
- if (m_param->rc.cuTree)694
+ if (m_param->rc.cuTree && !m_cuTreeStats.qpBuffer0)695
{696
if (m_param->rc.qgSize == 8)697
{698
699
return true;700
}701
702
+void RateControl::skipCUTreeSharedMemRead(int32_t cnt)703
+{704
+ m_cutreeShrMem->skipRead(cnt);705
+}706
void RateControl::reconfigureRC()707
{708
if (m_isVbv)709
710
711
TimingInfo *time = &sps.vuiParameters.timingInfo;712
int maxCpbOutputDelay = (int)(X265_MIN(m_param->keyframeMax * MAX_DURATION * time->timeScale / time->numUnitsInTick, INT_MAX));713
- int maxDpbOutputDelay = (int)(sps.maxDecPicBuffering * MAX_DURATION * time->timeScale / time->numUnitsInTick);714
+ int maxDpbOutputDelay = (int)(sps.maxDecPicBufferingsps.maxTempSubLayers - 1 * MAX_DURATION * time->timeScale / time->numUnitsInTick);715
int maxDelay = (int)(90000.0 * cpbSizeUnscale / bitRateUnscale + 0.5);716
717
hrd->initialCpbRemovalDelayLength = 2 + x265_clip3(4, 22, 32 - calcLength(maxDelay));718
719
{720
uint64_t allConstBits = 0, allCodedBits = 0;721
uint64_t allAvailableBits = uint64_t(m_param->rc.bitrate * 1000. * m_numEntries * m_frameDuration);722
- int startIndex, framesCount, endIndex;723
+ int startIndex, endIndex;724
int fps = X265_MIN(m_param->keyframeMax, (int)(m_fps + 0.5));725
- startIndex = endIndex = framesCount = 0;726
- int diffQp = 0;727
+ int distance = fps << 1;728
+ distance = distance > m_param->keyframeMax ? (m_param->keyframeMax << 1) : m_param->keyframeMax;729
+ startIndex = endIndex = 0;730
double targetBits = 0;731
double expectedBits = 0;732
- for (startIndex = m_start, endIndex = m_start; endIndex < m_numEntries; endIndex++)733
+ double targetBits2 = 0;734
+ double expectedBits2 = 0;735
+ double cpxSum = 0;736
+ double cpxSum2 = 0;737
+738
+ if (m_param->rc.rateControlMode == X265_RC_ABR)739
{740
- allConstBits += m_rce2PassendIndex.miscBits;741
- allCodedBits += m_rce2PassendIndex.coeffBits + m_rce2PassendIndex.mvBits;742
- if (m_param->rc.rateControlMode == X265_RC_CRF)743
+ for (endIndex = m_start; endIndex < m_numEntries; endIndex++)744
{745
- framesCount = endIndex - startIndex + 1;746
- diffQp += int (m_rce2PassendIndex.qpaRc - m_rce2PassendIndex.qpNoVbv);747
- if (framesCount > fps)748
- diffQp -= int (m_rce2PassendIndex - fps.qpaRc - m_rce2PassendIndex - fps.qpNoVbv);749
- if (framesCount >= fps)750
- {751
- if (diffQp >= 1)752
- {753
- if (!m_isQpModified && endIndex > fps)754
- {755
- double factor = 2;756
- double step = 0;757
- if (endIndex + fps >= m_numEntries)758
- {759
- m_start = endIndex - (endIndex % fps);760
- return true;761
- }762
- for (int start = endIndex + 1; start <= endIndex + fps && start < m_numEntries; start++)763
- {764
- RateControlEntry *rce = &m_rce2Passstart;765
- targetBits += qScale2bits(rce, x265_qp2qScale(rce->qpNoVbv));766
- expectedBits += qScale2bits(rce, rce->qScale);767
- }768
- if (expectedBits < 0.95 * targetBits)769
- {770
- m_isQpModified = true;771
- m_isGopReEncoded = true;772
- while (endIndex + fps < m_numEntries)773
- {774
- step = pow(2, factor / 6.0);775
- expectedBits = 0;776
- for (int start = endIndex + 1; start <= endIndex + fps; start++)777
- {778
- RateControlEntry *rce = &m_rce2Passstart;779
- rce->newQScale = rce->qScale / step;780
- X265_CHECK(rce->newQScale >= 0, "new Qscale is negative\n");781
- expectedBits += qScale2bits(rce, rce->newQScale);782
- rce->newQp = x265_qScale2qp(rce->newQScale);783
- }784
- if (expectedBits >= targetBits && step > 1)785
- factor *= 0.90;786
- else787
- break;788
- }789
-790
- if (m_isVbv && endIndex + fps < m_numEntries)791
- if (!vbv2Pass((uint64_t)targetBits, endIndex + fps, endIndex + 1))792
- return false;793
-794
- targetBits = 0;795
- expectedBits = 0;796
-797
- for (int start = endIndex - fps + 1; start <= endIndex; start++)798
- {799
- RateControlEntry *rce = &m_rce2Passstart;800
- targetBits += qScale2bits(rce, x265_qp2qScale(rce->qpNoVbv));801
- }802
- while (1)803
- {804
- step = pow(2, factor / 6.0);805
- expectedBits = 0;806
- for (int start = endIndex - fps + 1; start <= endIndex; start++)807
- {808
- RateControlEntry *rce = &m_rce2Passstart;809
- rce->newQScale = rce->qScale * step;810
- X265_CHECK(rce->newQScale >= 0, "new Qscale is negative\n");811
- expectedBits += qScale2bits(rce, rce->newQScale);812
- rce->newQp = x265_qScale2qp(rce->newQScale);813
- }814
- if (expectedBits > targetBits && step > 1)815
- factor *= 1.1;816
- else817
- break;818
- }819
- if (m_isVbv)820
- if (!vbv2Pass((uint64_t)targetBits, endIndex, endIndex - fps + 1))821
- return false;822
- diffQp = 0;823
- m_reencode = endIndex - fps + 1;824
- endIndex = endIndex + fps;825
- startIndex = endIndex + 1;826
- m_start = startIndex;827
- targetBits = expectedBits = 0;828
- }829
- else830
- targetBits = expectedBits = 0;831
- }832
- }833
- else834
- m_isQpModified = false;835
- }836
+ allConstBits += m_rce2PassendIndex.miscBits;837
+ allCodedBits += m_rce2PassendIndex.coeffBits + m_rce2PassendIndex.mvBits;838
}839
- }840
841
- if (m_param->rc.rateControlMode == X265_RC_ABR)842
- {843
if (allAvailableBits < allConstBits)844
{845
x265_log(m_param, X265_LOG_ERROR, "requested bitrate is too low. estimated minimum is %d kbps\n",846
- (int)(allConstBits * m_fps / framesCount * 1000.));847
+ (int)(allConstBits * m_fps / (m_numEntries - m_start) * 1000.));848
return false;849
}850
if (!analyseABR2Pass(allAvailableBits))851
return false;852
+853
+ return true;854
+ }855
+856
+ if (m_isQpModified)857
+ {858
+ return true;859
+ }860
+861
+ if (m_start + (fps << 1) > m_numEntries)862
+ {863
+ return true;864
+ }865
+866
+ for (startIndex = m_start, endIndex = m_numEntries - 1; startIndex < endIndex; startIndex++, endIndex--)867
+ {868
+ cpxSum += m_rce2PassstartIndex.qScale / m_rce2PassstartIndex.coeffBits;869
+ cpxSum2 += m_rce2PassendIndex.qScale / m_rce2PassendIndex.coeffBits;870
+871
+ RateControlEntry *rce = &m_rce2PassstartIndex;872
+ targetBits += qScale2bits(rce, x265_qp2qScale(rce->qpNoVbv));873
+ expectedBits += qScale2bits(rce, rce->qScale);874
+875
+ rce = &m_rce2PassendIndex;876
+ targetBits2 += qScale2bits(rce, x265_qp2qScale(rce->qpNoVbv));877
+ expectedBits2 += qScale2bits(rce, rce->qScale);878
}879
880
- m_start = X265_MAX(m_start, endIndex - fps);881
+ if (expectedBits < 0.95 * targetBits || expectedBits2 < 0.95 * targetBits2)882
+ {883
+ if (cpxSum / cpxSum2 < 0.95 || cpxSum2 / cpxSum < 0.95)884
+ {885
+ m_isQpModified = true;886
+ m_isGopReEncoded = true;887
+888
+ m_shortTermCplxSum = 0;889
+ m_shortTermCplxCount = 0;890
+ m_framesDone = m_start;891
+892
+ for (startIndex = m_start; startIndex < m_numEntries; startIndex++)893
+ {894
+ m_shortTermCplxSum *= 0.5;895
+ m_shortTermCplxCount *= 0.5;896
+ m_shortTermCplxSum += m_rce2PassstartIndex.currentSatd / (CLIP_DURATION(m_frameDuration) / BASE_FRAME_DURATION);897
+ m_shortTermCplxCount++;898
+ }899
+900
+ m_bufferFill = m_rce2Passm_start - 1.bufferFill;901
+ m_bufferFillFinal = m_rce2Passm_start - 1.bufferFillFinal;902
+ m_bufferFillActual = m_rce2Passm_start - 1.bufferFillActual;903
+904
+ m_reencode = m_start;905
+ m_start = m_numEntries;906
+ }907
+ else908
+ {909
+910
+ m_isQpModified = false;911
+ m_isGopReEncoded = false;912
+ }913
+ }914
+ else915
+ {916
+917
+ m_isQpModified = false;918
+ m_isGopReEncoded = false;919
+ }920
+921
+ m_start = X265_MAX(m_start, m_numEntries - distance + m_param->keyframeMax);922
923
return true;924
}925
926
m_predType = getPredictorType(curFrame->m_lowres.sliceType, m_sliceType);927
rce->poc = m_curSlice->m_poc;928
929
+ if (m_param->bEnableSBRC)930
+ {931
+ if (rce->poc == 0 || (m_framesDone % m_param->keyframeMax == 0))932
+ {933
+ //Reset SBRC buffer934
+ m_encodedSegmentBits = 0;935
+ m_segDur = 0;936
+ }937
+ }938
+939
if (!m_param->bResetZoneConfig && (rce->encodeOrder % m_param->reconfigWindowSize == 0))940
{941
int index = m_zoneBufferIdx % m_param->rc.zonefileCount;942
943
{944
m_param = m_param->rc.zonesi.zoneParam;945
reconfigureRC();946
- init(*m_curSlice->m_sps);947
+ if (!m_param->bNoResetZoneConfig)948
+ init(*m_curSlice->m_sps);949
}950
}951
}952
953
rce->frameSizeMaximum *= m_param->maxAUSizeFactor;954
}955
}956
+957
+ ///< regenerate the qp958
if (!m_isAbr && m_2pass && m_param->rc.rateControlMode == X265_RC_CRF)959
{960
- rce->qpPrev = x265_qScale2qp(rce->qScale);961
- rce->qScale = rce->newQScale;962
- rce->qpaRc = curEncData.m_avgQpRc = curEncData.m_avgQpAq = x265_qScale2qp(rce->newQScale);963
- m_qp = int(rce->qpaRc + 0.5);964
- rce->frameSizePlanned = qScale2bits(rce, rce->qScale);965
- m_framesDone++;966
- return m_qp;967
+ if (!m_param->rc.bEncFocusedFramesOnly)968
+ {969
+ rce->qpPrev = x265_qScale2qp(rce->qScale);970
+ if (m_param->bEnableSceneCutAwareQp)971
+ {972
+ double lqmin = m_lminm_sliceType;973
+ double lqmax = m_lmaxm_sliceType;974
+ if (m_param->bEnableSceneCutAwareQp & FORWARD)975
+ rce->newQScale = forwardMasking(curFrame, rce->newQScale);976
+ if (m_param->bEnableSceneCutAwareQp & BACKWARD)977
+ rce->newQScale = backwardMasking(curFrame, rce->newQScale);978
+ rce->newQScale = x265_clip3(lqmin, lqmax, rce->newQScale);979
+ }980
+ rce->qScale = rce->newQScale;981
+ rce->qpaRc = curEncData.m_avgQpRc = curEncData.m_avgQpAq = x265_qScale2qp(rce->newQScale);982
+ m_qp = int(rce->qpaRc + 0.5);983
+ rce->frameSizePlanned = qScale2bits(rce, rce->qScale);984
+ m_framesDone++;985
+ return m_qp;986
+ }987
+ else988
+ { 989
+ int index = m_encOrderrce->poc;990
+ index++;991
+ double totalDuration = m_frameDuration;992
+ for (int j = 0; totalDuration < 1.0 && index < m_numEntries; j++)993
+ {994
+ switch (m_rce2Passindex.sliceType)995
+ {996
+ case B_SLICE:997
+ curFrame->m_lowres.plannedTypej = m_rce2Passindex.keptAsRef ? X265_TYPE_BREF : X265_TYPE_B;998
+ break;999
+ case P_SLICE:1000
+ curFrame->m_lowres.plannedTypej = X265_TYPE_P;1001
+ break;1002
+ case I_SLICE:1003
+ curFrame->m_lowres.plannedTypej = m_param->bOpenGOP ? X265_TYPE_I : X265_TYPE_IDR;1004
+ break;1005
+ default:1006
+ break;1007
+ }1008
+1009
+ curFrame->m_lowres.plannedSatdj = m_rce2Passindex.currentSatd;1010
+ totalDuration += m_frameDuration;1011
+ index++;1012
+ }1013
+ }1014
}1015
1016
if (m_isAbr || m_2pass) // ABR,CRF1017
1018
{1019
m_cuTreeStats.qpBufPos++;1020
1021
- if (!fread(&type, 1, 1, m_cutreeStatFileIn))1022
- goto fail;1023
- if (fread(m_cuTreeStats.qpBufferm_cuTreeStats.qpBufPos, sizeof(uint16_t), ncu, m_cutreeStatFileIn) != (size_t)ncu)1024
- goto fail;1025
+ if (X265_SHARE_MODE_FILE == m_param->rc.dataShareMode)1026
+ {1027
+ if (!fread(&type, 1, 1, m_cutreeStatFileIn))1028
+ goto fail;1029
+ if (fread(m_cuTreeStats.qpBufferm_cuTreeStats.qpBufPos, sizeof(uint16_t), ncu, m_cutreeStatFileIn) != (size_t)ncu)1030
+ goto fail;1031
+ }1032
+ else // X265_SHARE_MODE_SHAREDMEM == m_param->rc.dataShareMode1033
+ {1034
+ if (!m_cutreeShrMem)1035
+ {1036
+ goto fail;1037
+ }1038
+1039
+ CUTreeSharedDataItem shrItem;1040
+ shrItem.type = &type;1041
+ shrItem.stats = m_cuTreeStats.qpBufferm_cuTreeStats.qpBufPos;1042
+ m_cutreeShrMem->readNext(&shrItem, ReadSharedCUTreeData);1043
+ }1044
1045
if (type != sliceTypeActual && m_cuTreeStats.qpBufPos == 1)1046
{1047
1048
m_sliderPos++;1049
}1050
1051
- if (m_sliceType == B_SLICE)1052
+ if((!m_param->bEnableSBRC && m_sliceType == B_SLICE) || (m_param->bEnableSBRC && !IS_REFERENCED(curFrame)))1053
{1054
/* B-frames don't have independent rate control, but rather get the1055
* average QP of the two adjacent P-frames + an offset */1056
1057
double minScenecutQscale =x265_qp2qScale(ABR_SCENECUT_INIT_QP_MIN); 1058
m_lastQScaleForP_SLICE = X265_MAX(minScenecutQscale, m_lastQScaleForP_SLICE);1059
}1060
+1061
double qScale = x265_qp2qScale(q);1062
rce->qpNoVbv = q;1063
+1064
+ if (m_param->bEnableSBRC)1065
+ {1066
+ qScale = tuneQscaleForSBRC(curFrame, qScale);1067
+ rce->qpNoVbv = x265_qScale2qp(qScale);1068
+ }1069
+1070
double lmin = 0, lmax = 0;1071
if (m_isGrainEnabled && m_isFirstMiniGop)1072
{1073
1074
qScale = x265_clip3(lqmin, lqmax, qScale);1075
}1076
1077
- if (!m_2pass || m_param->bliveVBV2pass)1078
+ if (!m_2pass || m_param->bliveVBV2pass || (m_2pass && m_param->rc.rateControlMode == X265_RC_CRF && m_param->rc.bEncFocusedFramesOnly))1079
{1080
/* clip qp to permissible range after vbv-lookahead estimation to avoid possible 1081
* mispredictions by initial frame size predictors */1082
1083
else1084
{1085
double abrBuffer = 2 * m_rateTolerance * m_bitrate;1086
- if (m_2pass)1087
+ if (m_2pass && (m_param->rc.rateControlMode != X265_RC_CRF || !m_param->rc.bEncFocusedFramesOnly))1088
{1089
double lmin = m_lminm_sliceType;1090
double lmax = m_lmaxm_sliceType;1091
1092
1093
if (m_param->rc.rateControlMode == X265_RC_CRF)1094
{1095
+ if (m_param->bEnableSBRC)1096
+ {1097
+ double rfConstant = m_param->rc.rfConstant;1098
+ if (m_currentSatd < rce->movingAvgSum)1099
+ rfConstant += 2;1100
+ double ipOffset = (curFrame->m_lowres.bScenecut ? m_ipOffset : m_ipOffset / 2.0);1101
+ rfConstant = (rce->sliceType == I_SLICE ? rfConstant - ipOffset :1102
+ (rce->sliceType == B_SLICE ? rfConstant + m_pbOffset : rfConstant));1103
+ double mbtree_offset = m_param->rc.cuTree ? (1.0 - m_param->rc.qCompress) * 13.5 : 0;1104
+ double qComp = (m_param->rc.cuTree && !m_param->rc.hevcAq) ? 0.99 : m_param->rc.qCompress;1105
+ m_rateFactorConstant = pow(m_currentSatd, 1.0 - qComp) /1106
+ x265_qp2qScale(rfConstant + mbtree_offset);1107
+ }1108
q = getQScale(rce, m_rateFactorConstant);1109
x265_zone* zone = getZone();1110
if (zone)1111
1112
}1113
double tunedQScale = tuneAbrQScaleFromFeedback(initialQScale);1114
overflow = tunedQScale / initialQScale;1115
- q = !m_partialResidualFrames? tunedQScale : initialQScale;1116
+ q = !m_partialResidualFrames ? tunedQScale : initialQScale;1117
bool isEncodeEnd = (m_param->totalFrames && 1118
m_framesDone > 0.75 * m_param->totalFrames) ? 1 : 0;1119
bool isEncodeBeg = m_framesDone < (int)(m_fps + 0.5);1120
1121
q = X265_MAX(minScenecutQscale, q);1122
m_lastQScaleForP_SLICE = X265_MAX(minScenecutQscale, m_lastQScaleForP_SLICE);1123
}1124
+ if (m_param->bEnableSBRC)1125
+ q = tuneQscaleForSBRC(curFrame, q);1126
+1127
rce->qpNoVbv = x265_qScale2qp(q);1128
if (m_sliceType == P_SLICE)1129
{1130
1131
return (p->coeff * var + p->offset) / (q * p->count);1132
}1133
1134
+double RateControl::tuneQscaleForSBRC(Frame* curFrame, double q)1135
+{1136
+ int depth = 0;1137
+ int framesDoneInSeg = m_framesDone % m_param->keyframeMax;1138
+ if (framesDoneInSeg + m_param->lookaheadDepth <= m_param->keyframeMax)1139
+ depth = m_param->lookaheadDepth;1140
+ else1141
+ depth = m_param->keyframeMax - framesDoneInSeg;1142
+ for (int iterations = 0; iterations < 1000; iterations++)1143
+ {1144
+ double totalDuration = m_segDur;1145
+ double frameBitsTotal = m_encodedSegmentBits + predictSize(&m_predm_predType, q, (double)m_currentSatd);1146
+ for (int i = 0; i < depth; i++)1147
+ {1148
+ int type = curFrame->m_lowres.plannedTypei;1149
+ if (type == X265_TYPE_AUTO)1150
+ break;1151
+ int64_t satd = curFrame->m_lowres.plannedSatdi >> (X265_DEPTH - 8);1152
+ type = IS_X265_TYPE_I(curFrame->m_lowres.plannedTypei) ? I_SLICE : IS_X265_TYPE_B(curFrame->m_lowres.plannedTypei) ? B_SLICE : P_SLICE;1153
+ int predType = getPredictorType(curFrame->m_lowres.plannedTypei, type);1154
+ double curBits = predictSize(&m_predpredType, q, (double)satd);1155
+ frameBitsTotal += curBits;1156
+ totalDuration += m_frameDuration;1157
+ }1158
+ //Check for segment buffer overflow and adjust QP accordingly1159
+ double segDur = m_param->keyframeMax / m_fps;1160
+ double allowedSize = m_vbvMaxRate * segDur;1161
+ double remDur = segDur - totalDuration;1162
+ double remainingBits = frameBitsTotal / totalDuration * remDur;1163
+ if (frameBitsTotal + remainingBits > 0.9 * allowedSize)1164
+ q = q * 1.01;1165
+ else1166
+ break;1167
+ }1168
+ return q;1169
+}1170
+1171
double RateControl::clipQscale(Frame* curFrame, RateControlEntry* rce, double q)1172
{1173
// B-frames are not directly subject to VBV,1174
1175
{1176
finalDur = x265_clip3(0.4, 1.0, totalDuration);1177
}1178
- targetFill = X265_MIN(m_bufferFill + totalDuration * m_vbvMaxRate * 0.5, m_bufferSize * (1 - m_minBufferFill * finalDur));1179
+ targetFill = X265_MIN(m_bufferFill + totalDuration * m_vbvMaxRate * 0.5, m_bufferSize * (m_minBufferFill * finalDur));1180
if (bufferFillCur < targetFill)1181
{1182
q *= 1.01;1183
1184
1185
if (m_param->rc.aqMode || m_isVbv || m_param->bAQMotion || bEnableDistOffset)1186
{1187
- if (m_isVbv && !(m_2pass && m_param->rc.rateControlMode == X265_RC_CRF))1188
+ if (m_isVbv && !(m_2pass && m_param->rc.rateControlMode == X265_RC_CRF && !m_param->rc.bEncFocusedFramesOnly))1189
{1190
double avgQpRc = 0;1191
/* determine avg QP decided by VBV rate control */1192
1193
if (m_param->rc.rateControlMode == X265_RC_CRF)1194
{1195
double crfVal, qpRef = curEncData.m_avgQpRc;1196
+1197
bool is2passCrfChange = false;1198
- if (m_2pass)1199
+ if (m_2pass && !m_param->rc.bEncFocusedFramesOnly)1200
{1201
if (fabs(curEncData.m_avgQpRc - rce->qpPrev) > 0.1)1202
{1203
1204
m_wantedBitsWindow += m_frameDuration * m_bitrate;1205
m_totalBits += bits - rce->rowTotalBits;1206
m_encodedBits += actualBits;1207
+ m_encodedSegmentBits += actualBits;1208
+ m_segDur += m_frameDuration;1209
int pos = m_sliderPos - m_param->frameNumThreads;1210
if (pos >= 0)1211
m_encodedBitsWindowpos % s_slidingWindowFrames = actualBits;1212
1213
{1214
uint8_t sliceType = (uint8_t)rce->sliceType;1215
primitives.fix8Pack(m_cuTreeStats.qpBuffer0, curFrame->m_lowres.qpCuTreeOffset, ncu);1216
- if (fwrite(&sliceType, 1, 1, m_cutreeStatFileOut) < 1)1217
- goto writeFailure;1218
- if (fwrite(m_cuTreeStats.qpBuffer0, sizeof(uint16_t), ncu, m_cutreeStatFileOut) < (size_t)ncu)1219
- goto writeFailure;1220
+1221
+ if (X265_SHARE_MODE_FILE == m_param->rc.dataShareMode)1222
+ {1223
+ if (fwrite(&sliceType, 1, 1, m_cutreeStatFileOut) < 1)1224
+ goto writeFailure;1225
+ if (fwrite(m_cuTreeStats.qpBuffer0, sizeof(uint16_t), ncu, m_cutreeStatFileOut) < (size_t)ncu)1226
+ goto writeFailure;1227
+ }1228
+ else // X265_SHARE_MODE_SHAREDMEM == m_param->rc.dataShareMode1229
+ {1230
+ if (!m_cutreeShrMem)1231
+ {1232
+ goto writeFailure;1233
+ }1234
+1235
+ CUTreeSharedDataItem shrItem;1236
+ shrItem.type = &sliceType;1237
+ shrItem.stats = m_cuTreeStats.qpBuffer0;1238
+ m_cutreeShrMem->writeData(&shrItem, WriteSharedCUTreeData);1239
+ } 1240
}1241
return 0;1242
1243
1244
if (m_cutreeStatFileIn)1245
fclose(m_cutreeStatFileIn);1246
1247
+ if (m_cutreeShrMem)1248
+ {1249
+ m_cutreeShrMem->release();1250
+ delete m_cutreeShrMem;1251
+ m_cutreeShrMem = NULL;1252
+ }1253
+1254
X265_FREE(m_rce2Pass);1255
X265_FREE(m_encOrder);1256
for (int i = 0; i < 2; i++)1257
1258
double RateControl::forwardMasking(Frame* curFrame, double q)1259
{1260
double qp = x265_qScale2qp(q);1261
- uint32_t maxWindowSize = uint32_t((m_param->fwdScenecutWindow / 1000.0) * (m_param->fpsNum / m_param->fpsDenom) + 0.5);1262
- uint32_t windowSize = maxWindowSize / 3;1263
+ uint32_t maxWindowSize = uint32_t((m_param->fwdMaxScenecutWindow / 1000.0) * (m_param->fpsNum / m_param->fpsDenom) + 0.5);1264
+ uint32_t windowSize6, prevWindow = 0;1265
int lastScenecut = m_top->m_rateControl->m_lastScenecut;1266
- int lastIFrame = m_top->m_rateControl->m_lastScenecutAwareIFrame;1267
- double fwdRefQpDelta = double(m_param->fwdRefQpDelta);1268
- double fwdNonRefQpDelta = double(m_param->fwdNonRefQpDelta);1269
- double sliceTypeDelta = SLICE_TYPE_DELTA * fwdRefQpDelta;1270
+1271
+ double fwdRefQpDelta6, fwdNonRefQpDelta6, sliceTypeDelta6;1272
+ for (int i = 0; i < 6; i++)1273
+ {1274
+ windowSizei = prevWindow + (uint32_t((m_param->fwdScenecutWindowi / 1000.0) * (m_param->fpsNum / m_param->fpsDenom) + 0.5));1275
+ fwdRefQpDeltai = double(m_param->fwdRefQpDeltai);1276
+ fwdNonRefQpDeltai = double(m_param->fwdNonRefQpDeltai);1277
+ sliceTypeDeltai = SLICE_TYPE_DELTA * fwdRefQpDeltai;1278
+ prevWindow = windowSizei;1279
+ }1280
+1281
1282
//Check whether the current frame is within the forward window1283
if (curFrame->m_poc > lastScenecut && curFrame->m_poc <= (lastScenecut + int(maxWindowSize)))1284
1285
}1286
else if (curFrame->m_lowres.sliceType == X265_TYPE_P)1287
{1288
- if (!(lastIFrame > lastScenecut && lastIFrame <= (lastScenecut + int(maxWindowSize))1289
- && curFrame->m_poc >= lastIFrame))1290
- {1291
- //Add offsets corresponding to the window in which the P-frame occurs1292
- if (curFrame->m_poc <= (lastScenecut + int(windowSize)))1293
- qp += WINDOW1_DELTA * (fwdRefQpDelta - sliceTypeDelta);1294
- else if (((curFrame->m_poc) > (lastScenecut + int(windowSize))) && ((curFrame->m_poc) <= (lastScenecut + 2 * int(windowSize))))1295
- qp += WINDOW2_DELTA * (fwdRefQpDelta - sliceTypeDelta);1296
- else if (curFrame->m_poc > lastScenecut + 2 * int(windowSize))1297
- qp += WINDOW3_DELTA * (fwdRefQpDelta - sliceTypeDelta);1298
- }1299
+ //Add offsets corresponding to the window in which the P-frame occurs1300
+ if (curFrame->m_poc <= (lastScenecut + int(windowSize0)))1301
+ qp += fwdRefQpDelta0 - sliceTypeDelta0;1302
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize0))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize1))))1303
+ qp += fwdRefQpDelta1 - sliceTypeDelta1;1304
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize1))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize2))))1305
+ qp += fwdRefQpDelta2 - sliceTypeDelta2;1306
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize2))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize3))))1307
+ qp += fwdRefQpDelta3 - sliceTypeDelta3;1308
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize3))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize4))))1309
+ qp += fwdRefQpDelta4 - sliceTypeDelta4;1310
+ else if (curFrame->m_poc > lastScenecut + int(windowSize4))1311
+ qp += fwdRefQpDelta5 - sliceTypeDelta5;1312
}1313
else if (curFrame->m_lowres.sliceType == X265_TYPE_BREF)1314
{1315
- if (!(lastIFrame > lastScenecut && lastIFrame <= (lastScenecut + int(maxWindowSize))1316
- && curFrame->m_poc >= lastIFrame))1317
- {1318
- //Add offsets corresponding to the window in which the B-frame occurs1319
- if (curFrame->m_poc <= (lastScenecut + int(windowSize)))1320
- qp += WINDOW1_DELTA * fwdRefQpDelta;1321
- else if (((curFrame->m_poc) > (lastScenecut + int(windowSize))) && ((curFrame->m_poc) <= (lastScenecut + 2 * int(windowSize))))1322
- qp += WINDOW2_DELTA * fwdRefQpDelta;1323
- else if (curFrame->m_poc > lastScenecut + 2 * int(windowSize))1324
- qp += WINDOW3_DELTA * fwdRefQpDelta;1325
- }1326
+ //Add offsets corresponding to the window in which the B-frame occurs1327
+ if (curFrame->m_poc <= (lastScenecut + int(windowSize0)))1328
+ qp += fwdRefQpDelta0;1329
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize0))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize1))))1330
+ qp += fwdRefQpDelta1;1331
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize1))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize2))))1332
+ qp += fwdRefQpDelta2;1333
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize2))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize3))))1334
+ qp += fwdRefQpDelta3;1335
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize3))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize4))))1336
+ qp += fwdRefQpDelta4;1337
+ else if (curFrame->m_poc > lastScenecut + int(windowSize4))1338
+ qp += fwdRefQpDelta5;1339
}1340
else if (curFrame->m_lowres.sliceType == X265_TYPE_B)1341
{1342
- if (!(lastIFrame > lastScenecut && lastIFrame <= (lastScenecut + int(maxWindowSize))1343
- && curFrame->m_poc >= lastIFrame))1344
- {1345
- //Add offsets corresponding to the window in which the b-frame occurs1346
- if (curFrame->m_poc <= (lastScenecut + int(windowSize)))1347
- qp += WINDOW1_DELTA * fwdNonRefQpDelta;1348
- else if (((curFrame->m_poc) > (lastScenecut + int(windowSize))) && ((curFrame->m_poc) <= (lastScenecut + 2 * int(windowSize))))1349
- qp += WINDOW2_DELTA * fwdNonRefQpDelta;1350
- else if (curFrame->m_poc > lastScenecut + 2 * int(windowSize))1351
- qp += WINDOW3_DELTA * fwdNonRefQpDelta;1352
- }1353
+ //Add offsets corresponding to the window in which the b-frame occurs1354
+ if (curFrame->m_poc <= (lastScenecut + int(windowSize0)))1355
+ qp += fwdNonRefQpDelta0;1356
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize0))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize1))))1357
+ qp += fwdNonRefQpDelta1;1358
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize1))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize2))))1359
+ qp += fwdNonRefQpDelta2;1360
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize2))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize3))))1361
+ qp += fwdNonRefQpDelta3;1362
+ else if (((curFrame->m_poc) > (lastScenecut + int(windowSize3))) && ((curFrame->m_poc) <= (lastScenecut + int(windowSize4))))1363
+ qp += fwdNonRefQpDelta4;1364
+ else if (curFrame->m_poc > lastScenecut + int(windowSize4))1365
+ qp += fwdNonRefQpDelta5;1366
}1367
}1368
1369
1370
double RateControl::backwardMasking(Frame* curFrame, double q)1371
{1372
double qp = x265_qScale2qp(q);1373
- double fwdRefQpDelta = double(m_param->fwdRefQpDelta);1374
- double bwdRefQpDelta = double(m_param->bwdRefQpDelta);1375
- double bwdNonRefQpDelta = double(m_param->bwdNonRefQpDelta);1376
+ uint32_t windowSize6, prevWindow = 0;1377
+ int lastScenecut = m_top->m_rateControl->m_lastScenecut;1378
1379
- if (curFrame->m_isInsideWindow == BACKWARD_WINDOW)1380
+ double bwdRefQpDelta6, bwdNonRefQpDelta6, sliceTypeDelta6;1381
+ for (int i = 0; i < 6; i++)1382
{1383
- if (bwdRefQpDelta < 0)1384
- bwdRefQpDelta = WINDOW3_DELTA * fwdRefQpDelta;1385
- double sliceTypeDelta = SLICE_TYPE_DELTA * bwdRefQpDelta;1386
- if (bwdNonRefQpDelta < 0)1387
- bwdNonRefQpDelta = bwdRefQpDelta + sliceTypeDelta;1388
+ windowSizei = prevWindow + (uint32_t((m_param->bwdScenecutWindowi / 1000.0) * (m_param->fpsNum / m_param->fpsDenom) + 0.5));1389
+ prevWindow = windowSizei;1390
+ bwdRefQpDeltai = double(m_param->bwdRefQpDeltai);1391
+ bwdNonRefQpDeltai = double(m_param->bwdNonRefQpDeltai);1392
+1393
+ if (bwdRefQpDeltai < 0)1394
+ bwdRefQpDeltai = BWD_WINDOW_DELTA * m_param->fwdRefQpDeltai;1395
+ sliceTypeDeltai = SLICE_TYPE_DELTA * bwdRefQpDeltai;1396
+1397
+ if (bwdNonRefQpDeltai < 0)1398
+ bwdNonRefQpDeltai = bwdRefQpDeltai + sliceTypeDeltai;1399
+ }1400
1401
+ if (curFrame->m_isInsideWindow == BACKWARD_WINDOW)1402
+ {1403
if (curFrame->m_lowres.sliceType == X265_TYPE_P)1404
- qp += bwdRefQpDelta - sliceTypeDelta;1405
+ {1406
+ //Add offsets corresponding to the window in which the P-frame occurs1407
+ if (curFrame->m_poc >= (lastScenecut - int(windowSize0)))1408
+ qp += bwdRefQpDelta0 - sliceTypeDelta0;1409
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize0))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize1))))1410
+ qp += bwdRefQpDelta1 - sliceTypeDelta1;1411
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize1))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize2))))1412
+ qp += bwdRefQpDelta2 - sliceTypeDelta2;1413
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize2))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize3))))1414
+ qp += bwdRefQpDelta3 - sliceTypeDelta3;1415
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize3))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize4))))1416
+ qp += bwdRefQpDelta4 - sliceTypeDelta4;1417
+ else if (curFrame->m_poc < lastScenecut - int(windowSize4))1418
+ qp += bwdRefQpDelta5 - sliceTypeDelta5;1419
+ }1420
else if (curFrame->m_lowres.sliceType == X265_TYPE_BREF)1421
- qp += bwdRefQpDelta;1422
+ {1423
+ //Add offsets corresponding to the window in which the B-frame occurs1424
+ if (curFrame->m_poc >= (lastScenecut - int(windowSize0)))1425
+ qp += bwdRefQpDelta0;1426
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize0))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize1))))1427
+ qp += bwdRefQpDelta1;1428
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize1))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize2))))1429
+ qp += bwdRefQpDelta2;1430
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize2))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize3))))1431
+ qp += bwdRefQpDelta3;1432
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize3))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize4))))1433
+ qp += bwdRefQpDelta4;1434
+ else if (curFrame->m_poc < lastScenecut - int(windowSize4))1435
+ qp += bwdRefQpDelta5;1436
+ }1437
else if (curFrame->m_lowres.sliceType == X265_TYPE_B)1438
- qp += bwdNonRefQpDelta;1439
+ {1440
+ //Add offsets corresponding to the window in which the b-frame occurs1441
+ if (curFrame->m_poc >= (lastScenecut - int(windowSize0)))1442
+ qp += bwdNonRefQpDelta0;1443
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize0))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize1))))1444
+ qp += bwdNonRefQpDelta1;1445
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize1))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize2))))1446
+ qp += bwdNonRefQpDelta2;1447
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize2))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize3))))1448
+ qp += bwdNonRefQpDelta3;1449
+ else if (((curFrame->m_poc) < (lastScenecut - int(windowSize3))) && ((curFrame->m_poc) >= (lastScenecut - int(windowSize4))))1450
+ qp += bwdNonRefQpDelta4;1451
+ else if (curFrame->m_poc < lastScenecut - int(windowSize4))1452
+ qp += bwdNonRefQpDelta5;1453
+ }1454
}1455
1456
return x265_qp2qScale(qp);1457
x265_3.5.tar.gz/source/encoder/ratecontrol.h -> x265_3.6.tar.gz/source/encoder/ratecontrol.h
Changed
90
1
2
3
#include "common.h"4
#include "sei.h"5
+#include "ringmem.h"6
7
namespace X265_NS {8
// encoder namespace9
10
#define MIN_AMORTIZE_FRACTION 0.211
#define CLIP_DURATION(f) x265_clip3(MIN_FRAME_DURATION, MAX_FRAME_DURATION, f)12
13
-/*Scenecut Aware QP*/14
-#define WINDOW1_DELTA 1.0 /* The offset for the frames coming in the window-1*/15
-#define WINDOW2_DELTA 0.7 /* The offset for the frames coming in the window-2*/16
-#define WINDOW3_DELTA 0.4 /* The offset for the frames coming in the window-3*/17
-18
struct Predictor19
{20
double coeffMin;21
22
Predictor rowPreds32;23
Predictor* rowPred2;24
25
+ int64_t currentSatd;26
int64_t lastSatd; /* Contains the picture cost of the previous frame, required for resetAbr and VBV */27
int64_t leadingNoBSatd;28
int64_t rowTotalBits; /* update cplxrsum and totalbits at the end of 2 rows */29
30
double rowCplxrSum;31
double qpNoVbv;32
double bufferFill;33
+ double bufferFillFinal;34
+ double bufferFillActual;35
double targetFill;36
bool vbvEndAdj;37
double frameDuration;38
39
double m_qCompress;40
int64_t m_totalBits; /* total bits used for already encoded frames (after ammortization) */41
int64_t m_encodedBits; /* bits used for encoded frames (without ammortization) */42
+ int64_t m_encodedSegmentBits; /* bits used for encoded frames in a segment*/43
+ double m_segDur;44
double m_fps;45
int64_t m_satdCostWindow50;46
int64_t m_encodedBitsWindow50;47
48
FILE* m_statFileOut;49
FILE* m_cutreeStatFileOut;50
FILE* m_cutreeStatFileIn;51
+ ///< store the cutree data in memory instead of file52
+ RingMem *m_cutreeShrMem;53
double m_lastAccumPNorm;54
double m_expectedBitsSum; /* sum of qscale2bits after rceq, ratefactor, and overflow, only includes finished frames */55
int64_t m_predictedBits;56
57
RateControl(x265_param& p, Encoder *enc);58
bool init(const SPS& sps);59
void initHRD(SPS& sps);60
+ void initVBV(const SPS& sps);61
void reconfigureRC();62
63
void setFinalFrameCount(int count);64
65
int writeRateControlFrameStats(Frame* curFrame, RateControlEntry* rce);66
bool initPass2();67
68
+ bool initCUTreeSharedMem();69
+ void skipCUTreeSharedMemRead(int32_t cnt);70
+71
double forwardMasking(Frame* curFrame, double q);72
double backwardMasking(Frame* curFrame, double q);73
74
75
double rateEstimateQscale(Frame* pic, RateControlEntry *rce); // main logic for calculating QP based on ABR76
double tuneAbrQScaleFromFeedback(double qScale);77
double tuneQScaleForZone(RateControlEntry *rce, double qScale); // Tune qScale to adhere to zone budget78
+ double tuneQscaleForSBRC(Frame* curFrame, double q); // Tune qScale to adhere to segment budget79
void accumPQpUpdate();80
81
int getPredictorType(int lowresSliceType, int sliceType);82
83
double tuneQScaleForGrain(double rcOverflow);84
void splitdeltaPOC(char deltapoc, RateControlEntry *rce);85
void splitbUsed(char deltapoc, RateControlEntry *rce);86
+ void checkAndResetCRF(RateControlEntry* rce);87
};88
}89
#endif // ifndef X265_RATECONTROL_H90
x265_3.5.tar.gz/source/encoder/sei.cpp -> x265_3.6.tar.gz/source/encoder/sei.cpp
Changed
10
1
2
{3
if (nalUnitType != NAL_UNIT_UNSPECIFIED)4
bs.writeByteAlignment();5
- list.serialize(nalUnitType, bs);6
+ list.serialize(nalUnitType, bs, (1 + (nalUnitType == NAL_UNIT_CODED_SLICE_TSA_N)));7
}8
}9
10
x265_3.5.tar.gz/source/encoder/sei.h -> x265_3.6.tar.gz/source/encoder/sei.h
Changed
103
1
2
}3
};4
5
+/* Film grain characteristics */6
+class FilmGrainCharacteristics : public SEI7
+{8
+ public:9
+10
+ FilmGrainCharacteristics()11
+ {12
+ m_payloadType = FILM_GRAIN_CHARACTERISTICS;13
+ m_payloadSize = 0;14
+ }15
+16
+ struct CompModelIntensityValues17
+ {18
+ uint8_t intensityIntervalLowerBound;19
+ uint8_t intensityIntervalUpperBound;20
+ int* compModelValue;21
+ };22
+23
+ struct CompModel24
+ {25
+ bool bPresentFlag;26
+ uint8_t numModelValues;27
+ uint8_t m_filmGrainNumIntensityIntervalMinus1;28
+ CompModelIntensityValues* intensityValues;29
+ };30
+31
+ CompModel m_compModelMAX_NUM_COMPONENT;32
+ bool m_filmGrainCharacteristicsPersistenceFlag;33
+ bool m_filmGrainCharacteristicsCancelFlag;34
+ bool m_separateColourDescriptionPresentFlag;35
+ bool m_filmGrainFullRangeFlag;36
+ uint8_t m_filmGrainModelId;37
+ uint8_t m_blendingModeId;38
+ uint8_t m_log2ScaleFactor;39
+ uint8_t m_filmGrainBitDepthLumaMinus8;40
+ uint8_t m_filmGrainBitDepthChromaMinus8;41
+ uint8_t m_filmGrainColourPrimaries;42
+ uint8_t m_filmGrainTransferCharacteristics;43
+ uint8_t m_filmGrainMatrixCoeffs;44
+45
+ void writeSEI(const SPS&)46
+ {47
+ WRITE_FLAG(m_filmGrainCharacteristicsCancelFlag, "film_grain_characteristics_cancel_flag");48
+49
+ if (!m_filmGrainCharacteristicsCancelFlag)50
+ {51
+ WRITE_CODE(m_filmGrainModelId, 2, "film_grain_model_id");52
+ WRITE_FLAG(m_separateColourDescriptionPresentFlag, "separate_colour_description_present_flag");53
+ if (m_separateColourDescriptionPresentFlag)54
+ {55
+ WRITE_CODE(m_filmGrainBitDepthLumaMinus8, 3, "film_grain_bit_depth_luma_minus8");56
+ WRITE_CODE(m_filmGrainBitDepthChromaMinus8, 3, "film_grain_bit_depth_chroma_minus8");57
+ WRITE_FLAG(m_filmGrainFullRangeFlag, "film_grain_full_range_flag");58
+ WRITE_CODE(m_filmGrainColourPrimaries, X265_BYTE, "film_grain_colour_primaries");59
+ WRITE_CODE(m_filmGrainTransferCharacteristics, X265_BYTE, "film_grain_transfer_characteristics");60
+ WRITE_CODE(m_filmGrainMatrixCoeffs, X265_BYTE, "film_grain_matrix_coeffs");61
+ }62
+ WRITE_CODE(m_blendingModeId, 2, "blending_mode_id");63
+ WRITE_CODE(m_log2ScaleFactor, 4, "log2_scale_factor");64
+ for (uint8_t c = 0; c < 3; c++)65
+ {66
+ WRITE_FLAG(m_compModelc.bPresentFlag && m_compModelc.m_filmGrainNumIntensityIntervalMinus1 + 1 > 0 && m_compModelc.numModelValues > 0, "comp_model_present_flagc");67
+ }68
+ for (uint8_t c = 0; c < 3; c++)69
+ {70
+ if (m_compModelc.bPresentFlag && m_compModelc.m_filmGrainNumIntensityIntervalMinus1 + 1 > 0 && m_compModelc.numModelValues > 0)71
+ {72
+ assert(m_compModelc.m_filmGrainNumIntensityIntervalMinus1 + 1 <= 256);73
+ assert(m_compModelc.numModelValues <= X265_BYTE);74
+ WRITE_CODE(m_compModelc.m_filmGrainNumIntensityIntervalMinus1 , X265_BYTE, "num_intensity_intervals_minus1c");75
+ WRITE_CODE(m_compModelc.numModelValues - 1, 3, "num_model_values_minus1c");76
+ for (uint8_t interval = 0; interval < m_compModelc.m_filmGrainNumIntensityIntervalMinus1 + 1; interval++)77
+ {78
+ WRITE_CODE(m_compModelc.intensityValuesinterval.intensityIntervalLowerBound, X265_BYTE, "intensity_interval_lower_boundci");79
+ WRITE_CODE(m_compModelc.intensityValuesinterval.intensityIntervalUpperBound, X265_BYTE, "intensity_interval_upper_boundci");80
+ for (uint8_t j = 0; j < m_compModelc.numModelValues; j++)81
+ {82
+ WRITE_SVLC(m_compModelc.intensityValuesinterval.compModelValuej,"comp_model_valueci");83
+ }84
+ }85
+ }86
+ }87
+ WRITE_FLAG(m_filmGrainCharacteristicsPersistenceFlag, "film_grain_characteristics_persistence_flag");88
+ }89
+ if (m_bitIf->getNumberOfWrittenBits() % X265_BYTE != 0)90
+ {91
+ WRITE_FLAG(1, "payload_bit_equal_to_one");92
+ while (m_bitIf->getNumberOfWrittenBits() % X265_BYTE != 0)93
+ {94
+ WRITE_FLAG(0, "payload_bit_equal_to_zero");95
+ }96
+ }97
+ }98
+};99
+100
static const uint32_t ISO_IEC_11578_LEN = 16;101
102
class SEIuserDataUnregistered : public SEI103
x265_3.5.tar.gz/source/encoder/slicetype.cpp -> x265_3.6.tar.gz/source/encoder/slicetype.cpp
Changed
1444
1
2
3
namespace X265_NS {4
5
+uint32_t acEnergyVarHist(uint64_t sum_ssd, int shift)6
+{7
+ uint32_t sum = (uint32_t)sum_ssd;8
+ uint32_t ssd = (uint32_t)(sum_ssd >> 32);9
+10
+ return ssd - ((uint64_t)sum * sum >> shift);11
+}12
+13
bool computeEdge(pixel* edgePic, pixel* refPic, pixel* edgeTheta, intptr_t stride, int height, int width, bool bcalcTheta, pixel whitePixel)14
{15
intptr_t rowOne = 0, rowTwo = 0, rowThree = 0, colOne = 0, colTwo = 0, colThree = 0;16
17
{18
for (int colNum = 0; colNum < width; colNum++)19
{20
- if ((rowNum >= 2) && (colNum >= 2) && (rowNum != height - 2) && (colNum != width - 2)) //Ignoring the border pixels of the picture21
+ if ((rowNum >= 2) && (colNum >= 2) && (rowNum < height - 2) && (colNum < width - 2)) //Ignoring the border pixels of the picture22
{23
/* 5x5 Gaussian filter24
2 4 5 4 225
26
if (param->rc.aqMode == X265_AQ_EDGE)27
edgeFilter(curFrame, param);28
29
- if (param->rc.aqMode == X265_AQ_EDGE && !param->bHistBasedSceneCut && param->recursionSkipMode == EDGE_BASED_RSKIP)30
+ if (param->rc.aqMode == X265_AQ_EDGE && param->recursionSkipMode == EDGE_BASED_RSKIP)31
{32
pixel* src = curFrame->m_edgePic + curFrame->m_fencPic->m_lumaMarginY * curFrame->m_fencPic->m_stride + curFrame->m_fencPic->m_lumaMarginX;33
primitives.planecopy_pp_shr(src, curFrame->m_fencPic->m_stride, curFrame->m_edgeBitPic,34
35
m_countPreLookahead = 0;36
#endif37
38
- memset(m_histogram, 0, sizeof(m_histogram));39
+ m_accHistDiffRunningAvgCb = X265_MALLOC(uint32_t*, NUMBER_OF_SEGMENTS_IN_WIDTH * sizeof(uint32_t*));40
+ m_accHistDiffRunningAvgCb0 = X265_MALLOC(uint32_t, NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);41
+ memset(m_accHistDiffRunningAvgCb0, 0, sizeof(uint32_t) * NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);42
+ for (uint32_t w = 1; w < NUMBER_OF_SEGMENTS_IN_WIDTH; w++) {43
+ m_accHistDiffRunningAvgCbw = m_accHistDiffRunningAvgCb0 + w * NUMBER_OF_SEGMENTS_IN_HEIGHT;44
+ }45
+46
+ m_accHistDiffRunningAvgCr = X265_MALLOC(uint32_t*, NUMBER_OF_SEGMENTS_IN_WIDTH * sizeof(uint32_t*));47
+ m_accHistDiffRunningAvgCr0 = X265_MALLOC(uint32_t, NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);48
+ memset(m_accHistDiffRunningAvgCr0, 0, sizeof(uint32_t) * NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);49
+ for (uint32_t w = 1; w < NUMBER_OF_SEGMENTS_IN_WIDTH; w++) {50
+ m_accHistDiffRunningAvgCrw = m_accHistDiffRunningAvgCr0 + w * NUMBER_OF_SEGMENTS_IN_HEIGHT;51
+ }52
+53
+ m_accHistDiffRunningAvg = X265_MALLOC(uint32_t*, NUMBER_OF_SEGMENTS_IN_WIDTH * sizeof(uint32_t*));54
+ m_accHistDiffRunningAvg0 = X265_MALLOC(uint32_t, NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);55
+ memset(m_accHistDiffRunningAvg0, 0, sizeof(uint32_t) * NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT);56
+ for (uint32_t w = 1; w < NUMBER_OF_SEGMENTS_IN_WIDTH; w++) {57
+ m_accHistDiffRunningAvgw = m_accHistDiffRunningAvg0 + w * NUMBER_OF_SEGMENTS_IN_HEIGHT;58
+ }59
+60
+ m_resetRunningAvg = true;61
+62
+ m_segmentCountThreshold = (uint32_t)(((float)((NUMBER_OF_SEGMENTS_IN_WIDTH * NUMBER_OF_SEGMENTS_IN_HEIGHT) * 50) / 100) + 0.5);63
+64
+ if (m_param->bEnableTemporalSubLayers > 2)65
+ {66
+ switch (m_param->bEnableTemporalSubLayers)67
+ {68
+ case 3:69
+ m_gopId = 0;70
+ break;71
+ case 4:72
+ m_gopId = 1;73
+ break;74
+ case 5:75
+ m_gopId = 2;76
+ break;77
+ default:78
+ break;79
+ }80
+ }81
}82
83
#if DETAILED_CU_STATS84
85
m_pooli.stopWorkers();86
}87
}88
+89
void Lookahead::destroy()90
{91
// these two queues will be empty unless the encode was aborted92
93
default:94
return;95
}96
- if (!m_param->analysisLoad || !m_param->bDisableLookahead)97
+ if (!curFrame->m_param->analysisLoad || !curFrame->m_param->bDisableLookahead)98
{99
X265_CHECK(curFrame->m_lowres.costEstb - p0p1 - b > 0, "Slice cost not estimated\n")100
101
- if (m_param->rc.cuTree && !m_param->rc.bStatRead)102
+ if (curFrame->m_param->rc.cuTree && !curFrame->m_param->rc.bStatRead)103
/* update row satds based on cutree offsets */104
curFrame->m_lowres.satdCost = frameCostRecalculate(frames, p0, p1, b);105
- else if (!m_param->analysisLoad || m_param->scaleFactor || m_param->bAnalysisType == HEVC_INFO)106
+ else if (!curFrame->m_param->analysisLoad || curFrame->m_param->scaleFactor || curFrame->m_param->bAnalysisType == HEVC_INFO)107
{108
- if (m_param->rc.aqMode)109
+ if (curFrame->m_param->rc.aqMode)110
curFrame->m_lowres.satdCost = curFrame->m_lowres.costEstAqb - p0p1 - b;111
else112
curFrame->m_lowres.satdCost = curFrame->m_lowres.costEstb - p0p1 - b;113
}114
- if (m_param->rc.vbvBufferSize && m_param->rc.vbvMaxBitrate)115
+ if (curFrame->m_param->rc.vbvBufferSize && curFrame->m_param->rc.vbvMaxBitrate)116
{117
/* aggregate lowres row satds to CTU resolution */118
curFrame->m_lowres.lowresCostForRc = curFrame->m_lowres.lowresCostsb - p0p1 - b;119
uint32_t lowresRow = 0, lowresCol = 0, lowresCuIdx = 0, sum = 0, intraSum = 0;120
- uint32_t scale = m_param->maxCUSize / (2 * X265_LOWRES_CU_SIZE);121
- uint32_t numCuInHeight = (m_param->sourceHeight + m_param->maxCUSize - 1) / m_param->maxCUSize;122
+ uint32_t scale = curFrame->m_param->maxCUSize / (2 * X265_LOWRES_CU_SIZE);123
+ uint32_t numCuInHeight = (curFrame->m_param->sourceHeight + curFrame->m_param->maxCUSize - 1) / curFrame->m_param->maxCUSize;124
uint32_t widthInLowresCu = (uint32_t)m_8x8Width, heightInLowresCu = (uint32_t)m_8x8Height;125
double *qp_offset = 0;126
/* Factor in qpoffsets based on Aq/Cutree in CU costs */127
- if (m_param->rc.aqMode || m_param->bAQMotion)128
- qp_offset = (framesb->sliceType == X265_TYPE_B || !m_param->rc.cuTree) ? framesb->qpAqOffset : framesb->qpCuTreeOffset;129
+ if (curFrame->m_param->rc.aqMode || curFrame->m_param->bAQMotion)130
+ qp_offset = (framesb->sliceType == X265_TYPE_B || !curFrame->m_param->rc.cuTree) ? framesb->qpAqOffset : framesb->qpCuTreeOffset;131
132
for (uint32_t row = 0; row < numCuInHeight; row++)133
{134
135
if (qp_offset)136
{137
double qpOffset;138
- if (m_param->rc.qgSize == 8)139
+ if (curFrame->m_param->rc.qgSize == 8)140
qpOffset = (qp_offsetlowresCol * 2 + lowresRow * widthInLowresCu * 4 +141
qp_offsetlowresCol * 2 + lowresRow * widthInLowresCu * 4 + 1 +142
qp_offsetlowresCol * 2 + lowresRow * widthInLowresCu * 4 + curFrame->m_lowres.maxBlocksInRowFullRes +143
144
int32_t intraCuCost = curFrame->m_lowres.intraCostlowresCuIdx;145
curFrame->m_lowres.intraCostlowresCuIdx = (intraCuCost * x265_exp2fix8(qpOffset) + 128) >> 8;146
}147
- if (m_param->bIntraRefresh && slice->m_sliceType == X265_TYPE_P)148
+ if (curFrame->m_param->bIntraRefresh && slice->m_sliceType == X265_TYPE_P)149
for (uint32_t x = curFrame->m_encData->m_pir.pirStartCol; x <= curFrame->m_encData->m_pir.pirEndCol; x++)150
diff += curFrame->m_lowres.intraCostlowresCuIdx - lowresCuCost;151
curFrame->m_lowres.lowresCostForRclowresCuIdx = lowresCuCost;152
153
}154
}155
156
+uint32_t LookaheadTLD::calcVariance(pixel* inpSrc, intptr_t stride, intptr_t blockOffset, uint32_t plane)157
+{158
+ pixel* src = inpSrc + blockOffset;159
+160
+ uint32_t var;161
+ if (!plane)162
+ var = acEnergyVarHist(primitives.cuBLOCK_8x8.var(src, stride), 6);163
+ else164
+ var = acEnergyVarHist(primitives.cuBLOCK_4x4.var(src, stride), 4);165
+166
+ x265_emms();167
+ return var;168
+}169
+170
+/*171
+** Compute Block and Picture Variance, Block Mean for all blocks in the picture172
+*/173
+void LookaheadTLD::computePictureStatistics(Frame *curFrame)174
+{175
+ int maxCol = curFrame->m_fencPic->m_picWidth;176
+ int maxRow = curFrame->m_fencPic->m_picHeight;177
+ intptr_t inpStride = curFrame->m_fencPic->m_stride;178
+179
+ // Variance180
+ uint64_t picTotVariance = 0;181
+ uint32_t variance;182
+183
+ uint64_t blockXY = 0;184
+ pixel* src = curFrame->m_fencPic->m_picOrg0;185
+186
+ for (int blockY = 0; blockY < maxRow; blockY += 8)187
+ {188
+ uint64_t rowVariance = 0;189
+ for (int blockX = 0; blockX < maxCol; blockX += 8)190
+ {191
+ intptr_t blockOffsetLuma = blockX + (blockY * inpStride);192
+193
+ variance = calcVariance(194
+ src,195
+ inpStride,196
+ blockOffsetLuma, 0);197
+198
+ rowVariance += variance;199
+ blockXY++;200
+ }201
+ picTotVariance += (uint16_t)(rowVariance / maxCol);202
+ }203
+204
+ curFrame->m_lowres.picAvgVariance = (uint16_t)(picTotVariance / maxRow);205
+206
+ // Collect chroma variance207
+ int hShift = curFrame->m_fencPic->m_hChromaShift;208
+ int vShift = curFrame->m_fencPic->m_vChromaShift;209
+210
+ int maxColChroma = curFrame->m_fencPic->m_picWidth >> hShift;211
+ int maxRowChroma = curFrame->m_fencPic->m_picHeight >> vShift;212
+ intptr_t cStride = curFrame->m_fencPic->m_strideC;213
+214
+ pixel* srcCb = curFrame->m_fencPic->m_picOrg1;215
+216
+ picTotVariance = 0;217
+ for (int blockY = 0; blockY < maxRowChroma; blockY += 4)218
+ {219
+ uint64_t rowVariance = 0;220
+ for (int blockX = 0; blockX < maxColChroma; blockX += 4)221
+ {222
+ intptr_t blockOffsetChroma = blockX + blockY * cStride;223
+224
+ variance = calcVariance(225
+ srcCb,226
+ cStride,227
+ blockOffsetChroma, 1);228
+229
+ rowVariance += variance;230
+ blockXY++;231
+ }232
+ picTotVariance += (uint16_t)(rowVariance / maxColChroma);233
+ }234
+235
+ curFrame->m_lowres.picAvgVarianceCb = (uint16_t)(picTotVariance / maxRowChroma);236
+237
+238
+ pixel* srcCr = curFrame->m_fencPic->m_picOrg2;239
+240
+ picTotVariance = 0;241
+ for (int blockY = 0; blockY < maxRowChroma; blockY += 4)242
+ {243
+ uint64_t rowVariance = 0;244
+ for (int blockX = 0; blockX < maxColChroma; blockX += 4)245
+ {246
+ intptr_t blockOffsetChroma = blockX + blockY * cStride;247
+248
+ variance = calcVariance(249
+ srcCr,250
+ cStride,251
+ blockOffsetChroma, 2);252
+253
+ rowVariance += variance;254
+ blockXY++;255
+ }256
+ picTotVariance += (uint16_t)(rowVariance / maxColChroma);257
+ }258
+259
+ curFrame->m_lowres.picAvgVarianceCr = (uint16_t)(picTotVariance / maxRowChroma);260
+}261
+262
+/*263
+* Compute histogram of n-bins for the input264
+*/265
+void LookaheadTLD::calculateHistogram(266
+ pixel *inputSrc,267
+ uint32_t inputWidth,268
+ uint32_t inputHeight,269
+ intptr_t stride,270
+ uint8_t dsFactor,271
+ uint32_t *histogram,272
+ uint64_t *sum)273
+274
+{275
+ *sum = 0;276
+277
+ for (uint32_t verticalIdx = 0; verticalIdx < inputHeight; verticalIdx += dsFactor)278
+ {279
+ for (uint32_t horizontalIdx = 0; horizontalIdx < inputWidth; horizontalIdx += dsFactor)280
+ {281
+ ++(histograminputSrchorizontalIdx);282
+ *sum += inputSrchorizontalIdx;283
+ }284
+ inputSrc += (stride << (dsFactor >> 1));285
+ }286
+287
+ return;288
+}289
+290
+/*291
+* Compute histogram bins and chroma pixel intensity *292
+*/293
+void LookaheadTLD::computeIntensityHistogramBinsChroma(294
+ Frame *curFrame,295
+ uint64_t *sumAverageIntensityCb,296
+ uint64_t *sumAverageIntensityCr)297
+{298
+ uint64_t sum;299
+ uint8_t dsFactor = 4;300
+301
+ uint32_t segmentWidth = curFrame->m_lowres.widthFullRes / NUMBER_OF_SEGMENTS_IN_WIDTH;302
+ uint32_t segmentHeight = curFrame->m_lowres.heightFullRes / NUMBER_OF_SEGMENTS_IN_HEIGHT;303
+304
+ for (uint32_t segmentInFrameWidthIndex = 0; segmentInFrameWidthIndex < NUMBER_OF_SEGMENTS_IN_WIDTH; segmentInFrameWidthIndex++)305
+ {306
+ for (uint32_t segmentInFrameHeightIndex = 0; segmentInFrameHeightIndex < NUMBER_OF_SEGMENTS_IN_HEIGHT; segmentInFrameHeightIndex++)307
+ {308
+ // Initialize bins to 1309
+ for (uint32_t cuIndex = 0; cuIndex < 256; cuIndex++) {310
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex1cuIndex = 1;311
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex2cuIndex = 1;312
+ }313
+314
+ uint32_t segmentWidthOffset = (segmentInFrameWidthIndex == NUMBER_OF_SEGMENTS_IN_WIDTH - 1) ?315
+ curFrame->m_lowres.widthFullRes - (NUMBER_OF_SEGMENTS_IN_WIDTH * segmentWidth) : 0;316
+317
+ uint32_t segmentHeightOffset = (segmentInFrameHeightIndex == NUMBER_OF_SEGMENTS_IN_HEIGHT - 1) ?318
+ curFrame->m_lowres.heightFullRes - (NUMBER_OF_SEGMENTS_IN_HEIGHT * segmentHeight) : 0;319
+320
+321
+ // U Histogram322
+ calculateHistogram(323
+ curFrame->m_fencPic->m_picOrg1 + ((segmentInFrameWidthIndex * segmentWidth) >> 1) + (((segmentInFrameHeightIndex * segmentHeight) >> 1) * curFrame->m_fencPic->m_strideC),324
+ (segmentWidth + segmentWidthOffset) >> 1,325
+ (segmentHeight + segmentHeightOffset) >> 1,326
+ curFrame->m_fencPic->m_strideC,327
+ dsFactor,328
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex1,329
+ &sum);330
+331
+ sum = (sum << dsFactor);332
+ *sumAverageIntensityCb += sum;333
+ curFrame->m_lowres.averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex1 =334
+ (uint8_t)((sum + (((segmentWidth + segmentWidthOffset) * (segmentHeight + segmentHeightOffset)) >> 3)) / (((segmentWidth + segmentWidthOffset) * (segmentHeight + segmentHeightOffset)) >> 2));335
+336
+ for (uint16_t histogramBin = 0; histogramBin < HISTOGRAM_NUMBER_OF_BINS; histogramBin++) {337
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex1histogramBin =338
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex1histogramBin << dsFactor;339
+ }340
+341
+ // V Histogram342
+ calculateHistogram(343
+ curFrame->m_fencPic->m_picOrg2 + ((segmentInFrameWidthIndex * segmentWidth) >> 1) + (((segmentInFrameHeightIndex * segmentHeight) >> 1) * curFrame->m_fencPic->m_strideC),344
+ (segmentWidth + segmentWidthOffset) >> 1,345
+ (segmentHeight + segmentHeightOffset) >> 1,346
+ curFrame->m_fencPic->m_strideC,347
+ dsFactor,348
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex2,349
+ &sum);350
+351
+ sum = (sum << dsFactor);352
+ *sumAverageIntensityCr += sum;353
+ curFrame->m_lowres.averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex2 =354
+ (uint8_t)((sum + (((segmentWidth + segmentWidthOffset) * (segmentHeight + segmentHeightOffset)) >> 3)) / (((segmentWidth + segmentHeightOffset) * (segmentHeight + segmentHeightOffset)) >> 2));355
+356
+ for (uint16_t histogramBin = 0; histogramBin < HISTOGRAM_NUMBER_OF_BINS; histogramBin++) {357
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex2histogramBin =358
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex2histogramBin << dsFactor;359
+ }360
+ }361
+ }362
+ return;363
+364
+}365
+366
+/*367
+* Compute histogram bins and luma pixel intensity *368
+*/369
+void LookaheadTLD::computeIntensityHistogramBinsLuma(370
+ Frame *curFrame,371
+ uint64_t *sumAvgIntensityTotalSegmentsLuma)372
+{373
+ uint64_t sum;374
+375
+ uint32_t segmentWidth = curFrame->m_lowres.quarterSampleLowResWidth / NUMBER_OF_SEGMENTS_IN_WIDTH;376
+ uint32_t segmentHeight = curFrame->m_lowres.quarterSampleLowResHeight / NUMBER_OF_SEGMENTS_IN_HEIGHT;377
+378
+ for (uint32_t segmentInFrameWidthIndex = 0; segmentInFrameWidthIndex < NUMBER_OF_SEGMENTS_IN_WIDTH; segmentInFrameWidthIndex++)379
+ {380
+ for (uint32_t segmentInFrameHeightIndex = 0; segmentInFrameHeightIndex < NUMBER_OF_SEGMENTS_IN_HEIGHT; segmentInFrameHeightIndex++)381
+ {382
+ // Initialize bins to 1383
+ for (uint32_t cuIndex = 0; cuIndex < 256; cuIndex++) {384
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex0cuIndex = 1;385
+ }386
+387
+ uint32_t segmentWidthOffset = (segmentInFrameWidthIndex == NUMBER_OF_SEGMENTS_IN_WIDTH - 1) ?388
+ curFrame->m_lowres.quarterSampleLowResWidth - (NUMBER_OF_SEGMENTS_IN_WIDTH * segmentWidth) : 0;389
+390
+ uint32_t segmentHeightOffset = (segmentInFrameHeightIndex == NUMBER_OF_SEGMENTS_IN_HEIGHT - 1) ?391
+ curFrame->m_lowres.quarterSampleLowResHeight - (NUMBER_OF_SEGMENTS_IN_HEIGHT * segmentHeight) : 0;392
+393
+ // Y Histogram394
+ calculateHistogram(395
+ curFrame->m_lowres.quarterSampleLowResBuffer + (curFrame->m_lowres.quarterSampleLowResOriginX + segmentInFrameWidthIndex * segmentWidth) + ((curFrame->m_lowres.quarterSampleLowResOriginY + segmentInFrameHeightIndex * segmentHeight) * curFrame->m_lowres.quarterSampleLowResStrideY),396
+ segmentWidth + segmentWidthOffset,397
+ segmentHeight + segmentHeightOffset,398
+ curFrame->m_lowres.quarterSampleLowResStrideY,399
+ 1,400
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex0,401
+ &sum);402
+403
+ curFrame->m_lowres.averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0 = (uint8_t)((sum + (((segmentWidth + segmentWidthOffset)*(segmentWidth + segmentHeightOffset)) >> 1)) / ((segmentWidth + segmentWidthOffset)*(segmentHeight + segmentHeightOffset)));404
+ (*sumAvgIntensityTotalSegmentsLuma) += (sum << 4);405
+ for (uint32_t histogramBin = 0; histogramBin < HISTOGRAM_NUMBER_OF_BINS; histogramBin++)406
+ {407
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex0histogramBin =408
+ curFrame->m_lowres.picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex0histogramBin << 4;409
+ }410
+ }411
+ }412
+}413
+414
+void LookaheadTLD::collectPictureStatistics(Frame *curFrame)415
+{416
+417
+ uint64_t sumAverageIntensityCb = 0;418
+ uint64_t sumAverageIntensityCr = 0;419
+ uint64_t sumAverageIntensity = 0;420
+421
+ // Histogram bins for Luma422
+ computeIntensityHistogramBinsLuma(423
+ curFrame,424
+ &sumAverageIntensity);425
+426
+ // Histogram bins for Chroma427
+ computeIntensityHistogramBinsChroma(428
+ curFrame,429
+ &sumAverageIntensityCb,430
+ &sumAverageIntensityCr);431
+432
+ curFrame->m_lowres.averageIntensity0 = (uint8_t)((sumAverageIntensity + ((curFrame->m_lowres.widthFullRes * curFrame->m_lowres.heightFullRes) >> 1)) / (curFrame->m_lowres.widthFullRes * curFrame->m_lowres.heightFullRes));433
+ curFrame->m_lowres.averageIntensity1 = (uint8_t)((sumAverageIntensityCb + ((curFrame->m_lowres.widthFullRes * curFrame->m_lowres.heightFullRes) >> 3)) / ((curFrame->m_lowres.widthFullRes * curFrame->m_lowres.heightFullRes) >> 2));434
+ curFrame->m_lowres.averageIntensity2 = (uint8_t)((sumAverageIntensityCr + ((curFrame->m_lowres.widthFullRes * curFrame->m_lowres.heightFullRes) >> 3)) / ((curFrame->m_lowres.widthFullRes * curFrame->m_lowres.heightFullRes) >> 2));435
+436
+ computePictureStatistics(curFrame);437
+438
+ curFrame->m_lowres.bHistScenecutAnalyzed = false;439
+}440
+441
void PreLookaheadGroup::processTasks(int workerThreadID)442
{443
if (workerThreadID < 0)444
445
preFrame->m_lowres.init(preFrame->m_fencPic, preFrame->m_poc);446
if (m_lookahead.m_bAdaptiveQuant)447
tld.calcAdaptiveQuantFrame(preFrame, m_lookahead.m_param);448
+449
+ if (m_lookahead.m_param->bHistBasedSceneCut)450
+ tld.collectPictureStatistics(preFrame);451
+452
tld.lowresIntraEstimate(preFrame->m_lowres, m_lookahead.m_param->rc.qgSize);453
preFrame->m_lowresInit = true;454
455
456
m_lock.release();457
}458
459
+460
+void Lookahead::placeBref(Frame** frames, int start, int end, int num, int *brefs)461
+{462
+ int avg = (start + end) / 2;463
+ if (m_param->bEnableTemporalSubLayers < 2)464
+ {465
+ (*framesavg).m_lowres.sliceType = X265_TYPE_BREF;466
+ (*brefs)++;467
+ return;468
+ }469
+ else470
+ {471
+ if (num <= 2)472
+ return;473
+ else474
+ {475
+ (*framesavg).m_lowres.sliceType = X265_TYPE_BREF;476
+ (*brefs)++;477
+ placeBref(frames, start, avg, avg - start, brefs);478
+ placeBref(frames, avg + 1, end, end - avg, brefs);479
+ return;480
+ }481
+ }482
+}483
+484
+485
+void Lookahead::compCostBref(Lowres **frames, int start, int end, int num)486
+{487
+ CostEstimateGroup estGroup(*this, frames);488
+ int avg = (start + end) / 2;489
+ if (num <= 2)490
+ {491
+ for (int i = start; i < end; i++)492
+ {493
+ estGroup.singleCost(start, end + 1, i + 1);494
+ }495
+ return;496
+ }497
+ else498
+ {499
+ estGroup.singleCost(start, end + 1, avg + 1);500
+ compCostBref(frames, start, avg, avg - start);501
+ compCostBref(frames, avg + 1, end, end - avg);502
+ return;503
+ }504
+}505
+506
/* called by API thread or worker thread with inputQueueLock acquired */507
void Lookahead::slicetypeDecide()508
{509
510
ScopedLock lock(m_inputLock);511
512
Frame *curFrame = m_inputQueue.first();513
+ if (m_param->bResetZoneConfig)514
+ {515
+ for (int i = 0; i < m_param->rc.zonefileCount; i++)516
+ {517
+ if (m_param->rc.zonesi.startFrame == curFrame->m_poc)518
+ m_param = m_param->rc.zonesi.zoneParam;519
+ int nextZoneStart = m_param->rc.zonesi.startFrame;520
+ nextZoneStart += nextZoneStart ? m_param->rc.zonesi.zoneParam->radl : 0;521
+ if (nextZoneStart < curFrame->m_poc + maxSearch && curFrame->m_poc < nextZoneStart)522
+ maxSearch = nextZoneStart - curFrame->m_poc;523
+ }524
+ }525
int j;526
for (j = 0; j < m_param->bframes + 2; j++)527
{528
529
m_param->rc.cuTree || m_param->scenecutThreshold || m_param->bHistBasedSceneCut ||530
(m_param->lookaheadDepth && m_param->rc.vbvBufferSize)))531
{532
- if(!m_param->rc.bStatRead)533
+ if (!m_param->rc.bStatRead)534
slicetypeAnalyse(frames, false);535
bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;536
if ((m_param->analysisLoad && m_param->scaleFactor && bIsVbv) || m_param->bliveVBV2pass)537
538
{539
Lowres& frm = listbframes->m_lowres;540
541
+ if (frm.sliceTypeReq != X265_TYPE_AUTO && frm.sliceTypeReq != frm.sliceType)542
+ frm.sliceType = frm.sliceTypeReq;543
if (frm.sliceType == X265_TYPE_BREF && !m_param->bBPyramid && brefs == m_param->bBPyramid)544
{545
frm.sliceType = X265_TYPE_B;546
547
}548
if (frm.sliceType == X265_TYPE_IDR && frm.bScenecut && isClosedGopRadl)549
{550
- if (!m_param->bHistBasedSceneCut || (m_param->bHistBasedSceneCut && frm.m_bIsHardScenecut))551
- {552
- for (int i = bframes; i < bframes + m_param->radl; i++)553
- listi->m_lowres.sliceType = X265_TYPE_B;554
- list(bframes + m_param->radl)->m_lowres.sliceType = X265_TYPE_IDR;555
- }556
+ for (int i = bframes; i < bframes + m_param->radl; i++)557
+ listi->m_lowres.sliceType = X265_TYPE_B;558
+ list(bframes + m_param->radl)->m_lowres.sliceType = X265_TYPE_IDR;559
}560
if (frm.sliceType == X265_TYPE_IDR)561
{562
563
break;564
}565
}566
- if (bframes)567
- listbframes - 1->m_lowres.bLastMiniGopBFrame = true;568
- listbframes->m_lowres.leadingBframes = bframes;569
- m_lastNonB = &listbframes->m_lowres;570
- m_histogrambframes++;571
-572
- /* insert a bref into the sequence */573
- if (m_param->bBPyramid && bframes > 1 && !brefs)574
- {575
- listbframes / 2->m_lowres.sliceType = X265_TYPE_BREF;576
- brefs++;577
- }578
- /* calculate the frame costs ahead of time for estimateFrameCost while we still have lowres */579
- if (m_param->rc.rateControlMode != X265_RC_CQP)580
- {581
- int p0, p1, b;582
- /* For zero latency tuning, calculate frame cost to be used later in RC */583
- if (!maxSearch)584
+585
+ if (m_param->bEnableTemporalSubLayers > 2)586
+ {587
+ //Split the partial mini GOP into sub mini GOPs when temporal sub layers are enabled588
+ if (bframes < m_param->bframes)589
{590
- for (int i = 0; i <= bframes; i++)591
- framesi + 1 = &listi->m_lowres;592
- }593
+ int leftOver = bframes + 1;594
+ int8_t gopId = m_gopId - 1;595
+ int gopLen = x265_gop_ra_lengthgopId;596
+ int listReset = 0;597
598
- /* estimate new non-B cost */599
- p1 = b = bframes + 1;600
- p0 = (IS_X265_TYPE_I(framesbframes + 1->sliceType)) ? b : 0;601
+ m_outputLock.acquire();602
603
- CostEstimateGroup estGroup(*this, frames);604
+ while ((gopId >= 0) && (leftOver > 3))605
+ {606
+ if (leftOver < gopLen)607
+ {608
+ gopId = gopId - 1;609
+ gopLen = x265_gop_ra_lengthgopId;610
+ continue;611
+ }612
+ else613
+ {614
+ int newbFrames = listReset + gopLen - 1;615
+ //Re-assign GOP616
+ listnewbFrames->m_lowres.sliceType = IS_X265_TYPE_I(listnewbFrames->m_lowres.sliceType) ? listnewbFrames->m_lowres.sliceType : X265_TYPE_P;617
+ if (newbFrames)618
+ listnewbFrames - 1->m_lowres.bLastMiniGopBFrame = true;619
+ listnewbFrames->m_lowres.leadingBframes = newbFrames;620
+ m_lastNonB = &listnewbFrames->m_lowres;621
+622
+ /* insert a bref into the sequence */623
+ if (m_param->bBPyramid && newbFrames)624
+ {625
+ placeBref(list, listReset, newbFrames, newbFrames + 1, &brefs);626
+ }627
+ if (m_param->rc.rateControlMode != X265_RC_CQP)628
+ {629
+ int p0, p1, b;630
+ /* For zero latency tuning, calculate frame cost to be used later in RC */631
+ if (!maxSearch)632
+ {633
+ for (int i = listReset; i <= newbFrames; i++)634
+ framesi + 1 = &listlistReset + i->m_lowres;635
+ }636
637
- estGroup.singleCost(p0, p1, b);638
+ /* estimate new non-B cost */639
+ p1 = b = newbFrames + 1;640
+ p0 = (IS_X265_TYPE_I(framesnewbFrames + 1->sliceType)) ? b : listReset;641
642
- if (bframes)643
+ CostEstimateGroup estGroup(*this, frames);644
+645
+ estGroup.singleCost(p0, p1, b);646
+647
+ if (newbFrames)648
+ compCostBref(frames, listReset, newbFrames, newbFrames + 1);649
+ }650
+651
+ m_inputLock.acquire();652
+ /* dequeue all frames from inputQueue that are about to be enqueued653
+ * in the output queue. The order is important because Frame can654
+ * only be in one list at a time */655
+ int64_t ptsX265_BFRAME_MAX + 1;656
+ for (int i = 0; i < gopLen; i++)657
+ {658
+ Frame *curFrame;659
+ curFrame = m_inputQueue.popFront();660
+ ptsi = curFrame->m_pts;661
+ maxSearch--;662
+ }663
+ m_inputLock.release();664
+665
+ int idx = 0;666
+ /* add non-B to output queue */667
+ listnewbFrames->m_reorderedPts = ptsidx++;668
+ listnewbFrames->m_gopOffset = 0;669
+ listnewbFrames->m_gopId = gopId;670
+ listnewbFrames->m_tempLayer = x265_gop_ragopId0.layer;671
+ m_outputQueue.pushBack(*listnewbFrames);672
+673
+ /* add B frames to output queue */674
+ int i = 1, j = 1;675
+ while (i < gopLen)676
+ {677
+ int offset = listReset + (x265_gop_ragopIdj.poc_offset - 1);678
+ if (!listoffset || offset == newbFrames)679
+ continue;680
+681
+ // Assign gop offset and temporal layer of frames682
+ listoffset->m_gopOffset = j;683
+ listbframes->m_gopId = gopId;684
+ listoffset->m_tempLayer = x265_gop_ragopIdj++.layer;685
+686
+ listoffset->m_reorderedPts = ptsidx++;687
+ m_outputQueue.pushBack(*listoffset);688
+ i++;689
+ }690
+691
+ listReset += gopLen;692
+ leftOver = leftOver - gopLen;693
+ gopId -= 1;694
+ gopLen = (gopId >= 0) ? x265_gop_ra_lengthgopId : 0;695
+ }696
+ }697
+698
+ if (leftOver > 0 && leftOver < 4)699
+ {700
+ int64_t ptsX265_BFRAME_MAX + 1;701
+ int idx = 0;702
+703
+ int newbFrames = listReset + leftOver - 1;704
+ listnewbFrames->m_lowres.sliceType = IS_X265_TYPE_I(listnewbFrames->m_lowres.sliceType) ? listnewbFrames->m_lowres.sliceType : X265_TYPE_P;705
+ if (newbFrames)706
+ listnewbFrames - 1->m_lowres.bLastMiniGopBFrame = true;707
+ listnewbFrames->m_lowres.leadingBframes = newbFrames;708
+ m_lastNonB = &listnewbFrames->m_lowres;709
+710
+ /* insert a bref into the sequence */711
+ if (m_param->bBPyramid && (newbFrames- listReset) > 1)712
+ placeBref(list, listReset, newbFrames, newbFrames + 1, &brefs);713
+714
+ if (m_param->rc.rateControlMode != X265_RC_CQP)715
+ {716
+ int p0, p1, b;717
+ /* For zero latency tuning, calculate frame cost to be used later in RC */718
+ if (!maxSearch)719
+ {720
+ for (int i = listReset; i <= newbFrames; i++)721
+ framesi + 1 = &listlistReset + i->m_lowres;722
+ }723
+724
+ /* estimate new non-B cost */725
+ p1 = b = newbFrames + 1;726
+ p0 = (IS_X265_TYPE_I(framesnewbFrames + 1->sliceType)) ? b : listReset;727
+728
+ CostEstimateGroup estGroup(*this, frames);729
+730
+ estGroup.singleCost(p0, p1, b);731
+732
+ if (newbFrames)733
+ compCostBref(frames, listReset, newbFrames, newbFrames + 1);734
+ }735
+736
+ m_inputLock.acquire();737
+ /* dequeue all frames from inputQueue that are about to be enqueued738
+ * in the output queue. The order is important because Frame can739
+ * only be in one list at a time */740
+ for (int i = 0; i < leftOver; i++)741
+ {742
+ Frame *curFrame;743
+ curFrame = m_inputQueue.popFront();744
+ ptsi = curFrame->m_pts;745
+ maxSearch--;746
+ }747
+ m_inputLock.release();748
+749
+ m_lastNonB = &listnewbFrames->m_lowres;750
+ listnewbFrames->m_reorderedPts = ptsidx++;751
+ listnewbFrames->m_gopOffset = 0;752
+ listnewbFrames->m_gopId = -1;753
+ listnewbFrames->m_tempLayer = 0;754
+ m_outputQueue.pushBack(*listnewbFrames);755
+ if (brefs)756
+ {757
+ for (int i = listReset; i < newbFrames; i++)758
+ {759
+ if (listi->m_lowres.sliceType == X265_TYPE_BREF)760
+ {761
+ listi->m_reorderedPts = ptsidx++;762
+ listi->m_gopOffset = 0;763
+ listi->m_gopId = -1;764
+ listi->m_tempLayer = 0;765
+ m_outputQueue.pushBack(*listi);766
+ }767
+ }768
+ }769
+770
+ /* add B frames to output queue */771
+ for (int i = listReset; i < newbFrames; i++)772
+ {773
+ /* push all the B frames into output queue except B-ref, which already pushed into output queue */774
+ if (listi->m_lowres.sliceType != X265_TYPE_BREF)775
+ {776
+ listi->m_reorderedPts = ptsidx++;777
+ listi->m_gopOffset = 0;778
+ listi->m_gopId = -1;779
+ listi->m_tempLayer = 1;780
+ m_outputQueue.pushBack(*listi);781
+ }782
+ }783
+ }784
+ }785
+ else786
+ // Fill the complete mini GOP when temporal sub layers are enabled787
{788
- p0 = 0; // last nonb789
- bool isp0available = framesbframes + 1->sliceType == X265_TYPE_IDR ? false : true;790
791
- for (b = 1; b <= bframes; b++)792
+ listbframes - 1->m_lowres.bLastMiniGopBFrame = true;793
+ listbframes->m_lowres.leadingBframes = bframes;794
+ m_lastNonB = &listbframes->m_lowres;795
+796
+ /* insert a bref into the sequence */797
+ if (m_param->bBPyramid && !brefs)798
{799
- if (!isp0available)800
- p0 = b;801
+ placeBref(list, 0, bframes, bframes + 1, &brefs);802
+ }803
804
- if (framesb->sliceType == X265_TYPE_B)805
- for (p1 = b; framesp1->sliceType == X265_TYPE_B; p1++)806
- ; // find new nonb or bref807
- else808
- p1 = bframes + 1;809
+ /* calculate the frame costs ahead of time for estimateFrameCost while we still have lowres */810
+ if (m_param->rc.rateControlMode != X265_RC_CQP)811
+ {812
+ int p0, p1, b;813
+ /* For zero latency tuning, calculate frame cost to be used later in RC */814
+ if (!maxSearch)815
+ {816
+ for (int i = 0; i <= bframes; i++)817
+ framesi + 1 = &listi->m_lowres;818
+ }819
820
+ /* estimate new non-B cost */821
+ p1 = b = bframes + 1;822
+ p0 = (IS_X265_TYPE_I(framesbframes + 1->sliceType)) ? b : 0;823
+824
+ CostEstimateGroup estGroup(*this, frames);825
estGroup.singleCost(p0, p1, b);826
827
- if (framesb->sliceType == X265_TYPE_BREF)828
+ compCostBref(frames, 0, bframes, bframes + 1);829
+ }830
+831
+ m_inputLock.acquire();832
+ /* dequeue all frames from inputQueue that are about to be enqueued833
+ * in the output queue. The order is important because Frame can834
+ * only be in one list at a time */835
+ int64_t ptsX265_BFRAME_MAX + 1;836
+ for (int i = 0; i <= bframes; i++)837
+ {838
+ Frame *curFrame;839
+ curFrame = m_inputQueue.popFront();840
+ ptsi = curFrame->m_pts;841
+ maxSearch--;842
+ }843
+ m_inputLock.release();844
+845
+ m_outputLock.acquire();846
+847
+ int idx = 0;848
+ /* add non-B to output queue */849
+ listbframes->m_reorderedPts = ptsidx++;850
+ listbframes->m_gopOffset = 0;851
+ listbframes->m_gopId = m_gopId;852
+ listbframes->m_tempLayer = x265_gop_ram_gopId0.layer;853
+ m_outputQueue.pushBack(*listbframes);854
+855
+ int i = 1, j = 1;856
+ while (i <= bframes)857
+ {858
+ int offset = x265_gop_ram_gopIdj.poc_offset - 1;859
+ if (!listoffset || offset == bframes)860
+ continue;861
+862
+ // Assign gop offset and temporal layer of frames863
+ listoffset->m_gopOffset = j;864
+ listoffset->m_gopId = m_gopId;865
+ listoffset->m_tempLayer = x265_gop_ram_gopIdj++.layer;866
+867
+ /* add B frames to output queue */868
+ listoffset->m_reorderedPts = ptsidx++;869
+ m_outputQueue.pushBack(*listoffset);870
+ i++;871
+ }872
+ }873
+874
+ bool isKeyFrameAnalyse = (m_param->rc.cuTree || (m_param->rc.vbvBufferSize && m_param->lookaheadDepth));875
+ if (isKeyFrameAnalyse && IS_X265_TYPE_I(m_lastNonB->sliceType))876
+ {877
+ m_inputLock.acquire();878
+ Frame *curFrame = m_inputQueue.first();879
+ frames0 = m_lastNonB;880
+ int j;881
+ for (j = 0; j < maxSearch; j++)882
+ {883
+ framesj + 1 = &curFrame->m_lowres;884
+ curFrame = curFrame->m_next;885
+ }886
+ m_inputLock.release();887
+888
+ framesj + 1 = NULL;889
+ if (!m_param->rc.bStatRead)890
+ slicetypeAnalyse(frames, true);891
+ bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;892
+ if ((m_param->analysisLoad && m_param->scaleFactor && bIsVbv) || m_param->bliveVBV2pass)893
+ {894
+ int numFrames;895
+ for (numFrames = 0; numFrames < maxSearch; numFrames++)896
{897
- p0 = b;898
- isp0available = true;899
+ Lowres *fenc = framesnumFrames + 1;900
+ if (!fenc)901
+ break;902
}903
+ vbvLookahead(frames, numFrames, true);904
}905
}906
- }907
908
- m_inputLock.acquire();909
- /* dequeue all frames from inputQueue that are about to be enqueued910
- * in the output queue. The order is important because Frame can911
- * only be in one list at a time */912
- int64_t ptsX265_BFRAME_MAX + 1;913
- for (int i = 0; i <= bframes; i++)914
- {915
- Frame *curFrame;916
- curFrame = m_inputQueue.popFront();917
- ptsi = curFrame->m_pts;918
- maxSearch--;919
- }920
- m_inputLock.release();921
922
- m_outputLock.acquire();923
- /* add non-B to output queue */924
- int idx = 0;925
- listbframes->m_reorderedPts = ptsidx++;926
- m_outputQueue.pushBack(*listbframes);927
- /* Add B-ref frame next to P frame in output queue, the B-ref encode before non B-ref frame */928
- if (brefs)929
+ m_outputLock.release();930
+ }931
+ else932
{933
- for (int i = 0; i < bframes; i++)934
+935
+ if (bframes)936
+ listbframes - 1->m_lowres.bLastMiniGopBFrame = true;937
+ listbframes->m_lowres.leadingBframes = bframes;938
+ m_lastNonB = &listbframes->m_lowres;939
+940
+ /* insert a bref into the sequence */941
+ if (m_param->bBPyramid && bframes > 1 && !brefs)942
{943
- if (listi->m_lowres.sliceType == X265_TYPE_BREF)944
+ placeBref(list, 0, bframes, bframes + 1, &brefs);945
+ }946
+ /* calculate the frame costs ahead of time for estimateFrameCost while we still have lowres */947
+ if (m_param->rc.rateControlMode != X265_RC_CQP)948
+ {949
+ int p0, p1, b;950
+ /* For zero latency tuning, calculate frame cost to be used later in RC */951
+ if (!maxSearch)952
{953
- listi->m_reorderedPts = ptsidx++;954
- m_outputQueue.pushBack(*listi);955
+ for (int i = 0; i <= bframes; i++)956
+ framesi + 1 = &listi->m_lowres;957
+ }958
+959
+ /* estimate new non-B cost */960
+ p1 = b = bframes + 1;961
+ p0 = (IS_X265_TYPE_I(framesbframes + 1->sliceType)) ? b : 0;962
+963
+ CostEstimateGroup estGroup(*this, frames);964
+ estGroup.singleCost(p0, p1, b);965
+966
+ if (m_param->bEnableTemporalSubLayers > 1 && bframes)967
+ {968
+ compCostBref(frames, 0, bframes, bframes + 1);969
+ }970
+ else971
+ {972
+ if (bframes)973
+ {974
+ p0 = 0; // last nonb975
+ bool isp0available = framesbframes + 1->sliceType == X265_TYPE_IDR ? false : true;976
+977
+ for (b = 1; b <= bframes; b++)978
+ {979
+ if (!isp0available)980
+ p0 = b;981
+982
+ if (framesb->sliceType == X265_TYPE_B)983
+ for (p1 = b; framesp1->sliceType == X265_TYPE_B; p1++)984
+ ; // find new nonb or bref985
+ else986
+ p1 = bframes + 1;987
+988
+ estGroup.singleCost(p0, p1, b);989
+990
+ if (framesb->sliceType == X265_TYPE_BREF)991
+ {992
+ p0 = b;993
+ isp0available = true;994
+ }995
+ }996
+ }997
}998
}999
- }1000
1001
- /* add B frames to output queue */1002
- for (int i = 0; i < bframes; i++)1003
- {1004
- /* push all the B frames into output queue except B-ref, which already pushed into output queue */1005
- if (listi->m_lowres.sliceType != X265_TYPE_BREF)1006
+ m_inputLock.acquire();1007
+ /* dequeue all frames from inputQueue that are about to be enqueued1008
+ * in the output queue. The order is important because Frame can1009
+ * only be in one list at a time */1010
+ int64_t ptsX265_BFRAME_MAX + 1;1011
+ for (int i = 0; i <= bframes; i++)1012
+ {1013
+ Frame *curFrame;1014
+ curFrame = m_inputQueue.popFront();1015
+ ptsi = curFrame->m_pts;1016
+ maxSearch--;1017
+ }1018
+ m_inputLock.release();1019
+1020
+ m_outputLock.acquire();1021
+1022
+ /* add non-B to output queue */1023
+ int idx = 0;1024
+ listbframes->m_reorderedPts = ptsidx++;1025
+ m_outputQueue.pushBack(*listbframes);1026
+1027
+ /* Add B-ref frame next to P frame in output queue, the B-ref encode before non B-ref frame */1028
+ if (brefs)1029
{1030
- listi->m_reorderedPts = ptsidx++;1031
- m_outputQueue.pushBack(*listi);1032
+ for (int i = 0; i < bframes; i++)1033
+ {1034
+ if (listi->m_lowres.sliceType == X265_TYPE_BREF)1035
+ {1036
+ listi->m_reorderedPts = ptsidx++;1037
+ m_outputQueue.pushBack(*listi);1038
+ }1039
+ }1040
}1041
- }1042
1043
- bool isKeyFrameAnalyse = (m_param->rc.cuTree || (m_param->rc.vbvBufferSize && m_param->lookaheadDepth));1044
- if (isKeyFrameAnalyse && IS_X265_TYPE_I(m_lastNonB->sliceType))1045
- {1046
- m_inputLock.acquire();1047
- Frame *curFrame = m_inputQueue.first();1048
- frames0 = m_lastNonB;1049
- int j;1050
- for (j = 0; j < maxSearch; j++)1051
+ /* add B frames to output queue */1052
+ for (int i = 0; i < bframes; i++)1053
{1054
- framesj + 1 = &curFrame->m_lowres;1055
- curFrame = curFrame->m_next;1056
+ /* push all the B frames into output queue except B-ref, which already pushed into output queue */1057
+ if (listi->m_lowres.sliceType != X265_TYPE_BREF)1058
+ {1059
+ listi->m_reorderedPts = ptsidx++;1060
+ m_outputQueue.pushBack(*listi);1061
+ }1062
}1063
- m_inputLock.release();1064
1065
- framesj + 1 = NULL;1066
- if (!m_param->rc.bStatRead)1067
- slicetypeAnalyse(frames, true);1068
- bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;1069
- if ((m_param->analysisLoad && m_param->scaleFactor && bIsVbv) || m_param->bliveVBV2pass)1070
+1071
+ bool isKeyFrameAnalyse = (m_param->rc.cuTree || (m_param->rc.vbvBufferSize && m_param->lookaheadDepth));1072
+ if (isKeyFrameAnalyse && IS_X265_TYPE_I(m_lastNonB->sliceType))1073
{1074
- int numFrames;1075
- for (numFrames = 0; numFrames < maxSearch; numFrames++)1076
+ m_inputLock.acquire();1077
+ Frame *curFrame = m_inputQueue.first();1078
+ frames0 = m_lastNonB;1079
+ int j;1080
+ for (j = 0; j < maxSearch; j++)1081
+ {1082
+ framesj + 1 = &curFrame->m_lowres;1083
+ curFrame = curFrame->m_next;1084
+ }1085
+ m_inputLock.release();1086
+1087
+ framesj + 1 = NULL;1088
+ if (!m_param->rc.bStatRead)1089
+ slicetypeAnalyse(frames, true);1090
+ bool bIsVbv = m_param->rc.vbvBufferSize > 0 && m_param->rc.vbvMaxBitrate > 0;1091
+ if ((m_param->analysisLoad && m_param->scaleFactor && bIsVbv) || m_param->bliveVBV2pass)1092
{1093
- Lowres *fenc = framesnumFrames + 1;1094
- if (!fenc)1095
- break;1096
+ int numFrames;1097
+ for (numFrames = 0; numFrames < maxSearch; numFrames++)1098
+ {1099
+ Lowres *fenc = framesnumFrames + 1;1100
+ if (!fenc)1101
+ break;1102
+ }1103
+ vbvLookahead(frames, numFrames, true);1104
}1105
- vbvLookahead(frames, numFrames, true);1106
}1107
+1108
+ m_outputLock.release();1109
}1110
- m_outputLock.release();1111
}1112
1113
void Lookahead::vbvLookahead(Lowres **frames, int numFrames, int keyframe)1114
1115
nextZoneStart += (i + 1 < m_param->rc.zonefileCount) ? m_param->rc.zonesi + 1.startFrame + m_param->rc.zonesi + 1.zoneParam->radl : m_param->totalFrames;1116
if (curZoneStart <= frames0->frameNum && nextZoneStart > frames0->frameNum)1117
m_param->keyframeMax = nextZoneStart - curZoneStart;1118
+ if (m_param->rc.zonesm_param->rc.zonefileCount - 1.startFrame <= frames0->frameNum && nextZoneStart == 0)1119
+ m_param->keyframeMax = m_param->rc.zones0.keyframeMax;1120
}1121
}1122
int keylimit = m_param->keyframeMax;1123
1124
int numAnalyzed = numFrames;1125
bool isScenecut = false;1126
1127
- /* Temporal computations for scenecut detection */1128
if (m_param->bHistBasedSceneCut)1129
- {1130
- for (int i = numFrames - 1; i > 0; i--)1131
- {1132
- if (framesi->interPCostPercDiff > 0.0)1133
- continue;1134
- int64_t interCost = framesi->costEst10;1135
- int64_t intraCost = framesi->costEst00;1136
- if (interCost < 0 || intraCost < 0)1137
- continue;1138
- int times = 0;1139
- double averagePcost = 0.0, averageIcost = 0.0;1140
- for (int j = i - 1; j >= 0 && times < 5; j--, times++)1141
- {1142
- if (framesj->costEst00 > 0 && framesj->costEst10 > 0)1143
- {1144
- averageIcost += framesj->costEst00;1145
- averagePcost += framesj->costEst10;1146
- }1147
- else1148
- times--;1149
- }1150
- if (times)1151
- {1152
- averageIcost = averageIcost / times;1153
- averagePcost = averagePcost / times;1154
- framesi->interPCostPercDiff = abs(interCost - averagePcost) / X265_MIN(interCost, averagePcost) * 100;1155
- framesi->intraCostPercDiff = abs(intraCost - averageIcost) / X265_MIN(intraCost, averageIcost) * 100;1156
- }1157
- }1158
- }1159
-1160
- /* When scenecut threshold is set, use scenecut detection for I frame placements */1161
- if (!m_param->bHistBasedSceneCut || (m_param->bHistBasedSceneCut && frames1->bScenecut))1162
+ isScenecut = histBasedScenecut(frames, 0, 1, origNumFrames);1163
+ else1164
isScenecut = scenecut(frames, 0, 1, true, origNumFrames);1165
1166
- if (isScenecut && (m_param->bHistBasedSceneCut || m_param->scenecutThreshold))1167
+ /* When scenecut threshold is set, use scenecut detection for I frame placements */1168
+ if (m_param->scenecutThreshold && isScenecut)1169
{1170
frames1->sliceType = X265_TYPE_I;1171
return;1172
1173
m_extendGopBoundary = false;1174
for (int i = m_param->bframes + 1; i < origNumFrames; i += m_param->bframes + 1)1175
{1176
- if (!m_param->bHistBasedSceneCut || (m_param->bHistBasedSceneCut && framesi + 1->bScenecut))1177
- scenecut(frames, i, i + 1, true, origNumFrames);1178
+ scenecut(frames, i, i + 1, true, origNumFrames);1179
1180
for (int j = i + 1; j <= X265_MIN(i + m_param->bframes + 1, origNumFrames); j++)1181
{1182
1183
{1184
for (int j = 1; j < numBFrames + 1; j++)1185
{1186
- bool isNextScenecut = false;1187
- if (!m_param->bHistBasedSceneCut || (m_param->bHistBasedSceneCut && framesj + 1->bScenecut))1188
- isNextScenecut = scenecut(frames, j, j + 1, false, origNumFrames);1189
- if (isNextScenecut || (bForceRADL && framesj->frameNum == preRADL))1190
+ if (scenecut(frames, j, j + 1, false, origNumFrames) ||1191
+ (bForceRADL && (framesj->frameNum == preRADL)))1192
{1193
framesj->sliceType = X265_TYPE_P;1194
numAnalyzed = j;1195
1196
/* Where A and B are scenes: AAAAAABBBAAAAAA1197
* If BBB is shorter than (maxp1-p0), it is detected as a flash1198
* and not considered a scenecut. */1199
+1200
for (int cp1 = p1; cp1 <= maxp1; cp1++)1201
{1202
- if (!scenecutInternal(frames, p0, cp1, false) && !m_param->bHistBasedSceneCut)1203
+ if (!scenecutInternal(frames, p0, cp1, false))1204
{1205
/* Any frame in between p0 and cur_p1 cannot be a real scenecut. */1206
for (int i = cp1; i > p0; i--)1207
1208
noScenecuts = false;1209
}1210
}1211
- else if ((m_param->bHistBasedSceneCut && framescp1->m_bIsMaxThres) || scenecutInternal(frames, cp1 - 1, cp1, false))1212
+ else if (scenecutInternal(frames, cp1 - 1, cp1, false))1213
{1214
/* If current frame is a Scenecut from p0 frame as well as Scenecut from1215
* preceeding frame, mark it as a Scenecut */1216
1217
1218
if (!framesp1->bScenecut)1219
return false;1220
- /* Check only scene transitions if max threshold */1221
- if (m_param->bHistBasedSceneCut && framesp1->m_bIsMaxThres)1222
- return framesp1->bScenecut;1223
1224
return scenecutInternal(frames, p0, p1, bRealScenecut);1225
}1226
1227
/* magic numbers pulled out of thin air */1228
float threshMin = (float)(threshMax * 0.25);1229
double bias = m_param->scenecutBias;1230
- if (m_param->bHistBasedSceneCut)1231
- {1232
- double minT = TEMPORAL_SCENECUT_THRESHOLD * (1 + m_param->edgeTransitionThreshold);1233
- if (frame->interPCostPercDiff > minT || frame->intraCostPercDiff > minT)1234
- {1235
- if (bRealScenecut && frame->bScenecut)1236
- x265_log(m_param, X265_LOG_DEBUG, "scene cut at %d \n", frame->frameNum);1237
- return frame->bScenecut;1238
- }1239
- else1240
- return false;1241
- }1242
- else if (bRealScenecut)1243
+1244
+ if (bRealScenecut)1245
{1246
if (m_param->keyframeMin == m_param->keyframeMax)1247
threshMin = threshMax;1248
1249
return res;1250
}1251
1252
+bool Lookahead::detectHistBasedSceneChange(Lowres **frames, int p0, int p1, int p2)1253
+{1254
+ bool isAbruptChange;1255
+ bool isSceneChange;1256
+1257
+ Lowres *previousFrame = framesp0;1258
+ Lowres *currentFrame = framesp1;1259
+ Lowres *futureFrame = framesp2;1260
+1261
+ currentFrame->bHistScenecutAnalyzed = true;1262
+1263
+ uint32_t **accHistDiffRunningAvgCb = m_accHistDiffRunningAvgCb;1264
+ uint32_t **accHistDiffRunningAvgCr = m_accHistDiffRunningAvgCr;1265
+ uint32_t **accHistDiffRunningAvg = m_accHistDiffRunningAvg;1266
+1267
+ uint8_t absIntDiffFuturePast = 0;1268
+ uint8_t absIntDiffFuturePresent = 0;1269
+ uint8_t absIntDiffPresentPast = 0;1270
+1271
+ uint32_t abruptChangeCount = 0;1272
+ uint32_t sceneChangeCount = 0;1273
+1274
+ uint32_t segmentWidth = frames1->widthFullRes / NUMBER_OF_SEGMENTS_IN_WIDTH;1275
+ uint32_t segmentHeight = frames1->heightFullRes / NUMBER_OF_SEGMENTS_IN_HEIGHT;1276
+1277
+ for (uint32_t segmentInFrameWidthIndex = 0; segmentInFrameWidthIndex < NUMBER_OF_SEGMENTS_IN_WIDTH; segmentInFrameWidthIndex++)1278
+ {1279
+ for (uint32_t segmentInFrameHeightIndex = 0; segmentInFrameHeightIndex < NUMBER_OF_SEGMENTS_IN_HEIGHT; segmentInFrameHeightIndex++)1280
+ {1281
+ isAbruptChange = false;1282
+ isSceneChange = false;1283
+1284
+ // accumulative absolute histogram differences between the past and current frame1285
+ uint32_t accHistDiff = 0;1286
+ uint32_t accHistDiffCb = 0;1287
+ uint32_t accHistDiffCr = 0;1288
+1289
+ uint32_t segmentWidthOffset = (segmentInFrameWidthIndex == NUMBER_OF_SEGMENTS_IN_WIDTH - 1) ?1290
+ frames1->widthFullRes - (NUMBER_OF_SEGMENTS_IN_WIDTH * segmentWidth) : 0;1291
+1292
+ uint32_t segmentHeightOffset = (segmentInFrameHeightIndex == NUMBER_OF_SEGMENTS_IN_HEIGHT - 1) ?1293
+ frames1->heightFullRes - (NUMBER_OF_SEGMENTS_IN_HEIGHT * segmentHeight) : 0;1294
+1295
+ segmentWidth += segmentWidthOffset;1296
+ segmentHeight += segmentHeightOffset;1297
+1298
+ uint32_t segmentThreshHold = (1299
+ ((X265_ABS((int64_t)currentFrame->picAvgVariance - (int64_t)previousFrame->picAvgVariance)) > PICTURE_DIFF_VARIANCE_TH) &&1300
+ (currentFrame->picAvgVariance > PICTURE_VARIANCE_TH || previousFrame->picAvgVariance > PICTURE_VARIANCE_TH)) ?1301
+ HIGH_VAR_SCENE_CHANGE_TH * NUM64x64INPIC(segmentWidth, segmentHeight) : LOW_VAR_SCENE_CHANGE_TH * NUM64x64INPIC(segmentWidth, segmentHeight);1302
+1303
+ uint32_t segmentThreshHoldCb = (1304
+ ((X265_ABS((int64_t)currentFrame->picAvgVarianceCb - (int64_t)previousFrame->picAvgVarianceCb)) > PICTURE_DIFF_VARIANCE_CHROMA_TH) &&1305
+ (currentFrame->picAvgVarianceCb > PICTURE_VARIANCE_CHROMA_TH || previousFrame->picAvgVarianceCb > PICTURE_VARIANCE_CHROMA_TH)) ?1306
+ HIGH_VAR_SCENE_CHANGE_CHROMA_TH * NUM64x64INPIC(segmentWidth, segmentHeight) : LOW_VAR_SCENE_CHANGE_CHROMA_TH * NUM64x64INPIC(segmentWidth, segmentHeight);1307
+1308
+ uint32_t segmentThreshHoldCr = (1309
+ ((X265_ABS((int64_t)currentFrame->picAvgVarianceCr - (int64_t)previousFrame->picAvgVarianceCr)) > PICTURE_DIFF_VARIANCE_CHROMA_TH) &&1310
+ (currentFrame->picAvgVarianceCr > PICTURE_VARIANCE_CHROMA_TH || previousFrame->picAvgVarianceCr > PICTURE_VARIANCE_CHROMA_TH)) ?1311
+ HIGH_VAR_SCENE_CHANGE_CHROMA_TH * NUM64x64INPIC(segmentWidth, segmentHeight) : LOW_VAR_SCENE_CHANGE_CHROMA_TH * NUM64x64INPIC(segmentWidth, segmentHeight);1312
+1313
+ for (uint32_t bin = 0; bin < HISTOGRAM_NUMBER_OF_BINS; ++bin) {1314
+ accHistDiff += X265_ABS((int32_t)currentFrame->picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex0bin - (int32_t)previousFrame->picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex0bin);1315
+ accHistDiffCb += X265_ABS((int32_t)currentFrame->picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex1bin - (int32_t)previousFrame->picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex1bin);1316
+ accHistDiffCr += X265_ABS((int32_t)currentFrame->picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex2bin - (int32_t)previousFrame->picHistogramsegmentInFrameWidthIndexsegmentInFrameHeightIndex2bin);1317
+ }1318
+1319
+ if (m_resetRunningAvg) {1320
+ accHistDiffRunningAvgsegmentInFrameWidthIndexsegmentInFrameHeightIndex = accHistDiff;1321
+ accHistDiffRunningAvgCbsegmentInFrameWidthIndexsegmentInFrameHeightIndex = accHistDiffCb;1322
+ accHistDiffRunningAvgCrsegmentInFrameWidthIndexsegmentInFrameHeightIndex = accHistDiffCr;1323
+ }1324
+1325
+ // difference between accumulative absolute histogram differences and the running average at the current frame.1326
+ uint32_t accHistDiffError = X265_ABS((int32_t)accHistDiffRunningAvgsegmentInFrameWidthIndexsegmentInFrameHeightIndex - (int32_t)accHistDiff);1327
+ uint32_t accHistDiffErrorCb = X265_ABS((int32_t)accHistDiffRunningAvgCbsegmentInFrameWidthIndexsegmentInFrameHeightIndex - (int32_t)accHistDiffCb);1328
+ uint32_t accHistDiffErrorCr = X265_ABS((int32_t)accHistDiffRunningAvgCrsegmentInFrameWidthIndexsegmentInFrameHeightIndex - (int32_t)accHistDiffCr);1329
+1330
+ if ((accHistDiffError > segmentThreshHold && accHistDiff >= accHistDiffError) ||1331
+ (accHistDiffErrorCb > segmentThreshHoldCb && accHistDiffCb >= accHistDiffErrorCb) ||1332
+ (accHistDiffErrorCr > segmentThreshHoldCr && accHistDiffCr >= accHistDiffErrorCr)) {1333
+1334
+ isAbruptChange = true;1335
+ }1336
+1337
+ if (isAbruptChange)1338
+ {1339
+ absIntDiffFuturePast = (uint8_t)X265_ABS((int16_t)futureFrame->averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0 - (int16_t)previousFrame->averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0);1340
+ absIntDiffFuturePresent = (uint8_t)X265_ABS((int16_t)futureFrame->averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0 - (int16_t)currentFrame->averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0);1341
+ absIntDiffPresentPast = (uint8_t)X265_ABS((int16_t)currentFrame->averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0 - (int16_t)previousFrame->averageIntensityPerSegmentsegmentInFrameWidthIndexsegmentInFrameHeightIndex0);1342
+1343
+ if (absIntDiffFuturePresent >= FLASH_TH * absIntDiffFuturePast && absIntDiffPresentPast >= FLASH_TH * absIntDiffFuturePast) {1344
+ x265_log(m_param, X265_LOG_DEBUG, "Flash in frame# %i , %i, %i, %i\n", currentFrame->frameNum, absIntDiffFuturePast, absIntDiffFuturePresent, absIntDiffPresentPast);1345
+ }1346
+ else if (absIntDiffFuturePresent < FADE_TH && absIntDiffPresentPast < FADE_TH) {1347
+ x265_log(m_param, X265_LOG_DEBUG, "Fade in frame# %i , %i, %i, %i\n", currentFrame->frameNum, absIntDiffFuturePast, absIntDiffFuturePresent, absIntDiffPresentPast);1348
+ }1349
+ else if (X265_ABS(absIntDiffFuturePresent - absIntDiffPresentPast) < INTENSITY_CHANGE_TH && absIntDiffFuturePresent + absIntDiffPresentPast >= absIntDiffFuturePast) {1350
+ x265_log(m_param, X265_LOG_DEBUG, "Intensity Change in frame# %i , %i, %i, %i\n", currentFrame->frameNum, absIntDiffFuturePast, absIntDiffFuturePresent, absIntDiffPresentPast);1351
+ }1352
+ else {1353
+ isSceneChange = true;1354
+ x265_log(m_param, X265_LOG_DEBUG, "Scene change in frame# %i , %i, %i, %i\n", currentFrame->frameNum, absIntDiffFuturePast, absIntDiffFuturePresent, absIntDiffPresentPast);1355
+ }1356
+1357
+ }1358
+ else {1359
+ accHistDiffRunningAvgsegmentInFrameWidthIndexsegmentInFrameHeightIndex = (3 * accHistDiffRunningAvgsegmentInFrameWidthIndexsegmentInFrameHeightIndex + accHistDiff) / 4;1360
+ }1361
+1362
+ abruptChangeCount += isAbruptChange;1363
+ sceneChangeCount += isSceneChange;1364
+ }1365
+ }1366
+1367
+ if (abruptChangeCount >= m_segmentCountThreshold) {1368
+ m_resetRunningAvg = true;1369
+ }1370
+ else {1371
+ m_resetRunningAvg = false;1372
+ }1373
+1374
+ if ((sceneChangeCount >= m_segmentCountThreshold)) {1375
+ x265_log(m_param, X265_LOG_DEBUG, "Scene Change in Pic Number# %i\n", currentFrame->frameNum);1376
+1377
+ return true;1378
+ }1379
+ else {1380
+ return false;1381
+ }1382
+1383
+}1384
+1385
+bool Lookahead::histBasedScenecut(Lowres **frames, int p0, int p1, int numFrames)1386
+{1387
+ /* Only do analysis during a normal scenecut check. */1388
+ if (m_param->bframes)1389
+ {1390
+ int origmaxp1 = p0 + 1;1391
+ /* Look ahead to avoid coding short flashes as scenecuts. */1392
+ origmaxp1 += m_param->bframes;1393
+ int maxp1 = X265_MIN(origmaxp1, numFrames);1394
+1395
+ for (int cp1 = p0; cp1 < maxp1; cp1++)1396
+ {1397
+ if (framescp1 + 1->bHistScenecutAnalyzed == true)1398
+ continue;1399
+1400
+ if (framescp1 + 2 != NULL && detectHistBasedSceneChange(frames, cp1, cp1 + 1, cp1 + 2))1401
+ {1402
+ /* If current frame is a Scenecut from p0 frame as well as Scenecut from1403
+ * preceeding frame, mark it as a Scenecut */1404
+ framescp1+1->bScenecut = true;1405
+ }1406
+ }1407
+1408
+ }1409
+1410
+ return framesp1->bScenecut;1411
+}1412
+1413
void Lookahead::slicetypePath(Lowres **frames, int length, char(*best_paths)X265_LOOKAHEAD_MAX + 1)1414
{1415
char paths2X265_LOOKAHEAD_MAX + 1;1416
1417
memcpy(best_pathslength % (X265_BFRAME_MAX + 1), pathsidx ^ 1, length);1418
}1419
1420
+// Find slicetype of the frame with poc # in lookahead buffer1421
+int Lookahead::findSliceType(int poc)1422
+{1423
+ int out_slicetype = X265_TYPE_AUTO;1424
+ if (m_filled)1425
+ {1426
+ m_outputLock.acquire();1427
+ Frame* out = m_outputQueue.first();1428
+ while (out != NULL) {1429
+ if (poc == out->m_poc)1430
+ {1431
+ out_slicetype = out->m_lowres.sliceType;1432
+ break;1433
+ }1434
+ out = out->m_next;1435
+ }1436
+ m_outputLock.release();1437
+ }1438
+ return out_slicetype;1439
+}1440
+1441
int64_t Lookahead::slicetypePathCost(Lowres **frames, char *path, int64_t threshold)1442
{1443
int64_t cost = 0;1444
x265_3.5.tar.gz/source/encoder/slicetype.h -> x265_3.6.tar.gz/source/encoder/slicetype.h
Changed
110
1
2
#define EDGE_INCLINATION 453
#define TEMPORAL_SCENECUT_THRESHOLD 504
5
+#define X265_ABS(a) (((a) < 0) ? (-(a)) : (a))6
+7
+#define PICTURE_DIFF_VARIANCE_TH 3908
+#define PICTURE_VARIANCE_TH 15009
+#define LOW_VAR_SCENE_CHANGE_TH 225010
+#define HIGH_VAR_SCENE_CHANGE_TH 350011
+12
+#define PICTURE_DIFF_VARIANCE_CHROMA_TH 1013
+#define PICTURE_VARIANCE_CHROMA_TH 2014
+#define LOW_VAR_SCENE_CHANGE_CHROMA_TH 2250/415
+#define HIGH_VAR_SCENE_CHANGE_CHROMA_TH 3500/416
+17
+#define FLASH_TH 1.518
+#define FADE_TH 419
+#define INTENSITY_CHANGE_TH 420
+21
+#define NUM64x64INPIC(w,h) ((w*h)>> (MAX_LOG2_CU_SIZE<<1))22
+23
#if HIGH_BIT_DEPTH24
#define EDGE_THRESHOLD 1023.025
#else26
27
28
~LookaheadTLD() { X265_FREE(wbuffer0); }29
30
+ void collectPictureStatistics(Frame *curFrame);31
+ void computeIntensityHistogramBinsLuma(Frame *curFrame, uint64_t *sumAvgIntensityTotalSegmentsLuma);32
+33
+ void computeIntensityHistogramBinsChroma(34
+ Frame *curFrame,35
+ uint64_t *sumAverageIntensityCb,36
+ uint64_t *sumAverageIntensityCr);37
+38
+ void calculateHistogram(39
+ pixel *inputSrc,40
+ uint32_t inputWidth,41
+ uint32_t inputHeight,42
+ intptr_t stride,43
+ uint8_t dsFactor,44
+ uint32_t *histogram,45
+ uint64_t *sum);46
+47
+ void computePictureStatistics(Frame *curFrame);48
+49
+ uint32_t calcVariance(pixel* src, intptr_t stride, intptr_t blockOffset, uint32_t plane);50
+51
void calcAdaptiveQuantFrame(Frame *curFrame, x265_param* param);52
+ void calcFrameSegment(Frame *curFrame);53
void lowresIntraEstimate(Lowres& fenc, uint32_t qgSize);54
55
void weightsAnalyse(Lowres& fenc, Lowres& ref);56
57
58
/* pre-lookahead */59
int m_fullQueueSize;60
- int m_histogramX265_BFRAME_MAX + 1;61
int m_lastKeyframe;62
int m_8x8Width;63
int m_8x8Height;64
65
bool m_isFadeIn;66
uint64_t m_fadeCount;67
int m_fadeStart;68
+69
+ uint32_t **m_accHistDiffRunningAvgCb;70
+ uint32_t **m_accHistDiffRunningAvgCr;71
+ uint32_t **m_accHistDiffRunningAvg;72
+73
+ bool m_resetRunningAvg;74
+ uint32_t m_segmentCountThreshold;75
+76
+ int8_t m_gopId;77
+78
Lookahead(x265_param *param, ThreadPool *pool);79
#if DETAILED_CU_STATS80
int64_t m_slicetypeDecideElapsedTime;81
82
83
void getEstimatedPictureCost(Frame *pic);84
void setLookaheadQueue();85
+ int findSliceType(int poc);86
87
protected:88
89
90
/* called by slicetypeAnalyse() to make slice decisions */91
bool scenecut(Lowres **frames, int p0, int p1, bool bRealScenecut, int numFrames);92
bool scenecutInternal(Lowres **frames, int p0, int p1, bool bRealScenecut);93
+94
+ bool histBasedScenecut(Lowres **frames, int p0, int p1, int numFrames);95
+ bool detectHistBasedSceneChange(Lowres **frames, int p0, int p1, int p2);96
+97
void slicetypePath(Lowres **frames, int length, char(*best_paths)X265_LOOKAHEAD_MAX + 1);98
int64_t slicetypePathCost(Lowres **frames, char *path, int64_t threshold);99
int64_t vbvFrameCost(Lowres **frames, int p0, int p1, int b);100
101
102
/* called by getEstimatedPictureCost() to finalize cuTree costs */103
int64_t frameCostRecalculate(Lowres **frames, int p0, int p1, int b);104
+ /*Compute index for positioning B-Ref frames*/105
+ void placeBref(Frame** frames, int start, int end, int num, int *brefs);106
+ void compCostBref(Lowres **frame, int start, int end, int num);107
};108
109
class PreLookaheadGroup : public BondedTaskGroup110
x265_3.5.tar.gz/source/output/output.cpp -> x265_3.6.tar.gz/source/output/output.cpp
Changed
19
1
2
3
using namespace X265_NS;4
5
-ReconFile* ReconFile::open(const char *fname, int width, int height, uint32_t bitdepth, uint32_t fpsNum, uint32_t fpsDenom, int csp)6
+ReconFile* ReconFile::open(const char *fname, int width, int height, uint32_t bitdepth, uint32_t fpsNum, uint32_t fpsDenom, int csp, int sourceBitDepth)7
{8
const char * s = strrchr(fname, '.');9
10
if (s && !strcmp(s, ".y4m"))11
- return new Y4MOutput(fname, width, height, fpsNum, fpsDenom, csp);12
+ return new Y4MOutput(fname, width, height, bitdepth, fpsNum, fpsDenom, csp, sourceBitDepth);13
else14
- return new YUVOutput(fname, width, height, bitdepth, csp);15
+ return new YUVOutput(fname, width, height, bitdepth, csp, sourceBitDepth);16
}17
18
OutputFile* OutputFile::open(const char *fname, InputFileInfo& inputInfo)19
x265_3.5.tar.gz/source/output/output.h -> x265_3.6.tar.gz/source/output/output.h
Changed
10
1
2
ReconFile() {}3
4
static ReconFile* open(const char *fname, int width, int height, uint32_t bitdepth,5
- uint32_t fpsNum, uint32_t fpsDenom, int csp);6
+ uint32_t fpsNum, uint32_t fpsDenom, int csp, int sourceBitDepth);7
8
virtual bool isFail() const = 0;9
10
x265_3.5.tar.gz/source/output/y4m.cpp -> x265_3.6.tar.gz/source/output/y4m.cpp
Changed
145
1
2
using namespace X265_NS;3
using namespace std;4
5
-Y4MOutput::Y4MOutput(const char *filename, int w, int h, uint32_t fpsNum, uint32_t fpsDenom, int csp)6
+Y4MOutput::Y4MOutput(const char* filename, int w, int h, uint32_t bitdepth, uint32_t fpsNum, uint32_t fpsDenom, int csp, int inputdepth)7
: width(w)8
, height(h)9
+ , bitDepth(bitdepth)10
, colorSpace(csp)11
, frameSize(0)12
+ , inputDepth(inputdepth)13
{14
ofs.open(filename, ios::binary | ios::out);15
buf = new charwidth;16
17
18
if (ofs)19
{20
- ofs << "YUV4MPEG2 W" << width << " H" << height << " F" << fpsNum << ":" << fpsDenom << " Ip" << " C" << cf << "\n";21
+ if (bitDepth == 10)22
+ ofs << "YUV4MPEG2 W" << width << " H" << height << " F" << fpsNum << ":" << fpsDenom << " Ip" << " C" << cf << "p10" << " XYSCSS = " << cf << "P10" << "\n";23
+ else if (bitDepth == 12)24
+ ofs << "YUV4MPEG2 W" << width << " H" << height << " F" << fpsNum << ":" << fpsDenom << " Ip" << " C" << cf << "p12" << " XYSCSS = " << cf << "P12" << "\n";25
+ else26
+ ofs << "YUV4MPEG2 W" << width << " H" << height << " F" << fpsNum << ":" << fpsDenom << " Ip" << " C" << cf << "\n";27
+28
header = ofs.tellp();29
}30
31
32
bool Y4MOutput::writePicture(const x265_picture& pic)33
{34
std::ofstream::pos_type outPicPos = header;35
- outPicPos += (uint64_t)pic.poc * (6 + frameSize);36
+ if (pic.bitDepth > 8)37
+ outPicPos += (uint64_t)(pic.poc * (6 + frameSize * 2));38
+ else39
+ outPicPos += (uint64_t)pic.poc * (6 + frameSize);40
ofs.seekp(outPicPos);41
ofs << "FRAME\n";42
43
-#if HIGH_BIT_DEPTH44
- if (pic.bitDepth > 8 && pic.poc == 0)45
- x265_log(NULL, X265_LOG_WARNING, "y4m: down-shifting reconstructed pixels to 8 bits\n");46
-#else47
- if (pic.bitDepth > 8 && pic.poc == 0)48
- x265_log(NULL, X265_LOG_WARNING, "y4m: forcing reconstructed pixels to 8 bits\n");49
-#endif50
+ if (inputDepth > 8)51
+ {52
+ if (pic.bitDepth == 8 && pic.poc == 0)53
+ x265_log(NULL, X265_LOG_WARNING, "y4m: down-shifting reconstructed pixels to 8 bits\n");54
+ }55
56
X265_CHECK(pic.colorSpace == colorSpace, "invalid chroma subsampling\n");57
58
-#if HIGH_BIT_DEPTH59
-60
- // encoder gave us short pixels, downshift, then write61
- X265_CHECK(pic.bitDepth > 8, "invalid bit depth\n");62
- int shift = pic.bitDepth - 8;63
- for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)64
+ if (inputDepth > 8)//if HIGH_BIT_DEPTH65
{66
- uint16_t *src = (uint16_t*)pic.planesi;67
- for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)68
+ if (pic.bitDepth == 8)69
{70
- for (int w = 0; w < width >> x265_cli_cspscolorSpace.widthi; w++)71
- bufw = (char)(srcw >> shift);72
-73
- ofs.write(buf, width >> x265_cli_cspscolorSpace.widthi);74
- src += pic.stridei / sizeof(*src);75
+ // encoder gave us short pixels, downshift, then write76
+ X265_CHECK(pic.bitDepth == 8, "invalid bit depth\n");77
+ int shift = pic.bitDepth - 8;78
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)79
+ {80
+ char *src = (char*)pic.planesi;81
+ for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)82
+ {83
+ for (int w = 0; w < width >> x265_cli_cspscolorSpace.widthi; w++)84
+ bufw = (char)(srcw >> shift);85
+86
+ ofs.write(buf, width >> x265_cli_cspscolorSpace.widthi);87
+ src += pic.stridei / sizeof(*src);88
+ }89
+ }90
+ }91
+ else92
+ {93
+ X265_CHECK(pic.bitDepth > 8, "invalid bit depth\n");94
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)95
+ {96
+ uint16_t *src = (uint16_t*)pic.planesi;97
+ for (int h = 0; h < (height * 1) >> x265_cli_cspscolorSpace.heighti; h++)98
+ {99
+ ofs.write((const char*)src, (width * 2) >> x265_cli_cspscolorSpace.widthi);100
+ src += pic.stridei / sizeof(*src);101
+ }102
+ }103
}104
}105
-106
-#else // if HIGH_BIT_DEPTH107
-108
- X265_CHECK(pic.bitDepth == 8, "invalid bit depth\n");109
- for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)110
+ else if (inputDepth == 8 && pic.bitDepth > 8)111
{112
- char *src = (char*)pic.planesi;113
- for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)114
+ X265_CHECK(pic.bitDepth > 8, "invalid bit depth\n");115
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)116
{117
- ofs.write(src, width >> x265_cli_cspscolorSpace.widthi);118
- src += pic.stridei / sizeof(*src);119
+ uint16_t* src = (uint16_t*)pic.planesi;120
+ for (int h = 0; h < (height * 1) >> x265_cli_cspscolorSpace.heighti; h++)121
+ {122
+ ofs.write((const char*)src, (width * 2) >> x265_cli_cspscolorSpace.widthi);123
+ src += pic.stridei / sizeof(*src);124
+ }125
+ }126
+ }127
+ else128
+ {129
+ X265_CHECK(pic.bitDepth == 8, "invalid bit depth\n");130
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)131
+ {132
+ char *src = (char*)pic.planesi;133
+ for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)134
+ {135
+ ofs.write(src, width >> x265_cli_cspscolorSpace.widthi);136
+ src += pic.stridei / sizeof(*src);137
+ }138
}139
}140
-141
-#endif // if HIGH_BIT_DEPTH142
143
return true;144
}145
x265_3.5.tar.gz/source/output/y4m.h -> x265_3.6.tar.gz/source/output/y4m.h
Changed
25
1
2
3
int height;4
5
+ uint32_t bitDepth;6
+7
int colorSpace;8
9
uint32_t frameSize;10
11
+ int inputDepth;12
+13
std::ofstream ofs;14
15
std::ofstream::pos_type header;16
17
18
public:19
20
- Y4MOutput(const char *filename, int width, int height, uint32_t fpsNum, uint32_t fpsDenom, int csp);21
+ Y4MOutput(const char *filename, int width, int height, uint32_t bitdepth, uint32_t fpsNum, uint32_t fpsDenom, int csp, int inputDepth);22
23
virtual ~Y4MOutput();24
25
x265_3.5.tar.gz/source/output/yuv.cpp -> x265_3.6.tar.gz/source/output/yuv.cpp
Changed
107
1
2
using namespace X265_NS;3
using namespace std;4
5
-YUVOutput::YUVOutput(const char *filename, int w, int h, uint32_t d, int csp)6
+YUVOutput::YUVOutput(const char *filename, int w, int h, uint32_t d, int csp, int inputdepth)7
: width(w)8
, height(h)9
, depth(d)10
, colorSpace(csp)11
, frameSize(0)12
+ , inputDepth(inputdepth)13
{14
ofs.open(filename, ios::binary | ios::out);15
buf = new charwidth;16
17
X265_CHECK(pic.colorSpace == colorSpace, "invalid chroma subsampling\n");18
X265_CHECK(pic.bitDepth == (int)depth, "invalid bit depth\n");19
20
-#if HIGH_BIT_DEPTH21
- if (depth == 8)22
+ if (inputDepth > 8)23
{24
- int shift = pic.bitDepth - 8;25
- ofs.seekp((std::streamoff)fileOffset);26
- for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)27
- {28
- uint16_t *src = (uint16_t*)pic.planesi;29
- for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)30
- {31
- for (int w = 0; w < width >> x265_cli_cspscolorSpace.widthi; w++)32
- bufw = (char)(srcw >> shift);33
+ if (depth == 8)34
+ {35
+ int shift = pic.bitDepth - 8;36
+ ofs.seekp((std::streamoff)fileOffset);37
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)38
+ {39
+ uint16_t *src = (uint16_t*)pic.planesi;40
+ for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)41
+ {42
+ for (int w = 0; w < width >> x265_cli_cspscolorSpace.widthi; w++)43
+ bufw = (char)(srcw >> shift);44
45
- ofs.write(buf, width >> x265_cli_cspscolorSpace.widthi);46
- src += pic.stridei / sizeof(*src);47
- }48
- }49
+ ofs.write(buf, width >> x265_cli_cspscolorSpace.widthi);50
+ src += pic.stridei / sizeof(*src);51
+ }52
+ }53
+ }54
+ else55
+ {56
+ ofs.seekp((std::streamoff)(fileOffset * 2));57
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)58
+ {59
+ uint16_t *src = (uint16_t*)pic.planesi;60
+ for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)61
+ {62
+ ofs.write((const char*)src, (width * 2) >> x265_cli_cspscolorSpace.widthi);63
+ src += pic.stridei / sizeof(*src);64
+ }65
+ }66
+ }67
}68
else69
{70
- ofs.seekp((std::streamoff)(fileOffset * 2));71
- for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)72
- {73
- uint16_t *src = (uint16_t*)pic.planesi;74
- for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)75
- {76
- ofs.write((const char*)src, (width * 2) >> x265_cli_cspscolorSpace.widthi);77
- src += pic.stridei / sizeof(*src);78
- }79
- }80
+ ofs.seekp((std::streamoff)fileOffset);81
+ for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)82
+ {83
+ char *src = (char*)pic.planesi;84
+ for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)85
+ {86
+ ofs.write(src, width >> x265_cli_cspscolorSpace.widthi);87
+ src += pic.stridei / sizeof(*src);88
+ }89
+ }90
}91
-#else // if HIGH_BIT_DEPTH92
- ofs.seekp((std::streamoff)fileOffset);93
- for (int i = 0; i < x265_cli_cspscolorSpace.planes; i++)94
- {95
- char *src = (char*)pic.planesi;96
- for (int h = 0; h < height >> x265_cli_cspscolorSpace.heighti; h++)97
- {98
- ofs.write(src, width >> x265_cli_cspscolorSpace.widthi);99
- src += pic.stridei / sizeof(*src);100
- }101
- }102
-103
-#endif // if HIGH_BIT_DEPTH104
105
return true;106
}107
x265_3.5.tar.gz/source/output/yuv.h -> x265_3.6.tar.gz/source/output/yuv.h
Changed
18
1
2
3
uint32_t frameSize;4
5
+ int inputDepth;6
+7
char *buf;8
9
std::ofstream ofs;10
11
public:12
13
- YUVOutput(const char *filename, int width, int height, uint32_t bitdepth, int csp);14
+ YUVOutput(const char *filename, int width, int height, uint32_t bitdepth, int csp, int inputDepth);15
16
virtual ~YUVOutput();17
18
x265_3.5.tar.gz/source/test/CMakeLists.txt -> x265_3.6.tar.gz/source/test/CMakeLists.txt
Changed
24
1
2
3
# add ARM assembly files4
if(ARM OR CROSS_COMPILE_ARM)5
- if(NOT ARM64)6
- enable_language(ASM)7
- set(NASM_SRC checkasm-arm.S)8
- add_custom_command(9
- OUTPUT checkasm-arm.obj10
- COMMAND ${CMAKE_CXX_COMPILER}11
- ARGS ${NASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/checkasm-arm.S -o checkasm-arm.obj12
- DEPENDS checkasm-arm.S)13
- endif()14
+ enable_language(ASM)15
+ set(NASM_SRC checkasm-arm.S)16
+ add_custom_command(17
+ OUTPUT checkasm-arm.obj18
+ COMMAND ${CMAKE_CXX_COMPILER}19
+ ARGS ${NASM_FLAGS} ${CMAKE_CURRENT_SOURCE_DIR}/checkasm-arm.S -o checkasm-arm.obj20
+ DEPENDS checkasm-arm.S)21
endif(ARM OR CROSS_COMPILE_ARM)22
23
# add PowerPC assembly files24
x265_3.5.tar.gz/source/test/pixelharness.cpp -> x265_3.6.tar.gz/source/test/pixelharness.cpp
Changed
63
1
2
return true;3
}4
5
+bool PixelHarness::check_downscaleluma_t(downscaleluma_t ref, downscaleluma_t opt)6
+{7
+ ALIGN_VAR_16(pixel, ref_destf32 * 32);8
+ ALIGN_VAR_16(pixel, opt_destf32 * 32);9
+10
+ intptr_t src_stride = 64;11
+ intptr_t dst_stride = 32;12
+ int bx = 32;13
+ int by = 32;14
+ int j = 0;15
+ for (int i = 0; i < ITERS; i++)16
+ {17
+ int index = i % TEST_CASES;18
+ ref(pixel_test_buffindex + j, ref_destf, src_stride, dst_stride, bx, by);19
+ checked(opt, pixel_test_buffindex + j, opt_destf, src_stride, dst_stride, bx, by);20
+21
+ if (memcmp(ref_destf, opt_destf, 32 * 32 * sizeof(pixel)))22
+ return false;23
+24
+ reportfail();25
+ j += INCR;26
+ }27
+28
+ return true;29
+}30
+31
bool PixelHarness::check_cpy2Dto1D_shl_t(cpy2Dto1D_shl_t ref, cpy2Dto1D_shl_t opt)32
{33
ALIGN_VAR_16(int16_t, ref_dest64 * 64);34
35
}36
}37
38
+ if (opt.frameSubSampleLuma)39
+ {40
+ if (!check_downscaleluma_t(ref.frameSubSampleLuma, opt.frameSubSampleLuma))41
+ {42
+ printf("SubSample Luma failed!\n");43
+ return false;44
+ }45
+ }46
+47
if (opt.scale1D_128to64NONALIGNED)48
{49
if (!check_scale1D_pp(ref.scale1D_128to64NONALIGNED, opt.scale1D_128to64NONALIGNED))50
51
REPORT_SPEEDUP(opt.frameInitLowres, ref.frameInitLowres, pbuf2, pbuf1, pbuf2, pbuf3, pbuf4, 64, 64, 64, 64);52
}53
54
+ if (opt.frameSubSampleLuma)55
+ {56
+ HEADER0("downscaleluma");57
+ REPORT_SPEEDUP(opt.frameSubSampleLuma, ref.frameSubSampleLuma, pbuf2, pbuf1, 64, 64, 64, 64);58
+ }59
+60
if (opt.scale1D_128to64NONALIGNED)61
{62
HEADER0("scale1D_128to64");63
x265_3.5.tar.gz/source/test/pixelharness.h -> x265_3.6.tar.gz/source/test/pixelharness.h
Changed
9
1
2
bool check_integral_inith(integralh_t ref, integralh_t opt);3
bool check_ssimDist(ssimDistortion_t ref, ssimDistortion_t opt);4
bool check_normFact(normFactor_t ref, normFactor_t opt, int block);5
+ bool check_downscaleluma_t(downscaleluma_t ref, downscaleluma_t opt);6
7
public:8
9
x265_3.5.tar.gz/source/test/rate-control-tests.txt -> x265_3.6.tar.gz/source/test/rate-control-tests.txt
Changed
10
1
2
112_1920x1080_25.yuv,--preset ultrafast --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 15000 --hrd --strict-cbr3
Traffic_4096x2048_30.yuv,--preset superfast --bitrate 20000 --vbv-maxrate 20000 --vbv-bufsize 20000 --repeat-headers --strict-cbr4
Traffic_4096x2048_30.yuv,--preset faster --bitrate 8000 --vbv-maxrate 8000 --vbv-bufsize 6000 --aud --repeat-headers --no-open-gop --hrd --pmode --pme5
-News-4k.y4m,--preset veryfast --bitrate 3000 --vbv-maxrate 5000 --vbv-bufsize 5000 --repeat-headers --temporal-layers6
+News-4k.y4m,--preset veryfast --bitrate 3000 --vbv-maxrate 5000 --vbv-bufsize 5000 --repeat-headers --temporal-layers 37
NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset medium --bitrate 18000 --vbv-bufsize 20000 --vbv-maxrate 18000 --strict-cbr8
NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset medium --bitrate 8000 --vbv-bufsize 12000 --vbv-maxrate 10000 --tune grain9
big_buck_bunny_360p24.y4m,--preset medium --bitrate 400 --vbv-bufsize 600 --vbv-maxrate 600 --aud --hrd --tune fast-decode10
x265_3.5.tar.gz/source/test/regression-tests.txt -> x265_3.6.tar.gz/source/test/regression-tests.txt
Changed
91
1
2
BasketballDrive_1920x1080_50.y4m,--preset faster --aq-strength 2 --merange 190 --slices 33
BasketballDrive_1920x1080_50.y4m,--preset medium --ctu 16 --max-tu-size 8 --subme 7 --qg-size 16 --cu-lossless --tu-inter-depth 3 --limit-tu 14
BasketballDrive_1920x1080_50.y4m,--preset medium --keyint -1 --nr-inter 100 -F4 --no-sao5
-BasketballDrive_1920x1080_50.y4m,--preset medium --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 2 --bitrate 7000 --limit-modes::--preset medium --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 2 --bitrate 7000 --limit-modes6
+BasketballDrive_1920x1080_50.y4m,--preset medium --analysis-save x265_analysis.dat --analysis-save-reuse-level 2 --bitrate 7000 --limit-modes::--preset medium --analysis-load x265_analysis.dat --analysis-load-reuse-level 2 --bitrate 7000 --limit-modes7
BasketballDrive_1920x1080_50.y4m,--preset slow --nr-intra 100 -F4 --aq-strength 3 --qg-size 16 --limit-refs 18
BasketballDrive_1920x1080_50.y4m,--preset slower --lossless --chromaloc 3 --subme 0 --limit-tu 49
-BasketballDrive_1920x1080_50.y4m,--preset slower --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --bitrate 7000 --limit-tu 0::--preset slower --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --bitrate 7000 --limit-tu 010
+BasketballDrive_1920x1080_50.y4m,--preset slower --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --bitrate 7000 --limit-tu 0::--preset slower --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --bitrate 7000 --limit-tu 011
BasketballDrive_1920x1080_50.y4m,--preset veryslow --crf 4 --cu-lossless --pmode --limit-refs 1 --aq-mode 3 --limit-tu 312
-BasketballDrive_1920x1080_50.y4m,--preset veryslow --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --crf 18 --tskip-fast --limit-tu 2::--preset veryslow --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --crf 18 --tskip-fast --limit-tu 213
+BasketballDrive_1920x1080_50.y4m,--preset veryslow --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --crf 18 --tskip-fast --limit-tu 2::--preset veryslow --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --crf 18 --tskip-fast --limit-tu 214
BasketballDrive_1920x1080_50.y4m,--preset veryslow --recon-y4m-exec "ffplay -i pipe:0 -autoexit"15
Coastguard-4k.y4m,--preset ultrafast --recon-y4m-exec "ffplay -i pipe:0 -autoexit"16
Coastguard-4k.y4m,--preset superfast --tune grain --overscan=crop17
18
Coastguard-4k.y4m,--preset slow --tune psnr --cbqpoffs -1 --crqpoffs 1 --limit-refs 119
CrowdRun_1920x1080_50_10bit_422.yuv,--preset ultrafast --weightp --tune zerolatency --qg-size 1620
CrowdRun_1920x1080_50_10bit_422.yuv,--preset superfast --weightp --no-wpp --sao21
-CrowdRun_1920x1080_50_10bit_422.yuv,--preset veryfast --temporal-layers --tune grain22
+CrowdRun_1920x1080_50_10bit_422.yuv,--preset veryfast --temporal-layers 2 --tune grain23
CrowdRun_1920x1080_50_10bit_422.yuv,--preset faster --max-tu-size 4 --min-cu-size 3224
CrowdRun_1920x1080_50_10bit_422.yuv,--preset fast --aq-mode 0 --sar 2 --range full25
CrowdRun_1920x1080_50_10bit_422.yuv,--preset medium --no-wpp --no-cutree --no-strong-intra-smoothing --limit-refs 126
27
CrowdRun_1920x1080_50_10bit_422.yuv,--preset slower --tune ssim --tune fastdecode --limit-refs 228
CrowdRun_1920x1080_50_10bit_444.yuv,--preset ultrafast --weightp --no-wpp --no-open-gop29
CrowdRun_1920x1080_50_10bit_444.yuv,--preset superfast --weightp --dither --no-psy-rd30
-CrowdRun_1920x1080_50_10bit_444.yuv,--preset veryfast --temporal-layers --repeat-headers --limit-refs 231
+CrowdRun_1920x1080_50_10bit_444.yuv,--preset veryfast --temporal-layers 2 --repeat-headers --limit-refs 232
CrowdRun_1920x1080_50_10bit_444.yuv,--preset medium --dither --keyint -1 --rdoq-level 1 --limit-modes33
CrowdRun_1920x1080_50_10bit_444.yuv,--preset veryslow --tskip --tskip-fast --no-scenecut --limit-tu 134
CrowdRun_1920x1080_50_10bit_444.yuv,--preset veryslow --aq-mode 3 --aq-strength 1.5 --aq-motion --bitrate 500035
36
CrowdRun_1920x1080_50_10bit_444.yuv,--preset veryslow --hevc-aq --no-cutree --qg-size 1637
DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset superfast --weightp --qg-size 1638
DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset medium --tune psnr --bframes 16 --limit-modes39
-DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset slow --temporal-layers --no-psy-rd --qg-size 32 --limit-refs 0 --cu-lossless40
+DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset slow --temporal-layers 2 --no-psy-rd --qg-size 32 --limit-refs 0 --cu-lossless41
DucksAndLegs_1920x1080_60_10bit_444.yuv,--preset veryfast --weightp --nr-intra 1000 -F442
DucksAndLegs_1920x1080_60_10bit_444.yuv,--preset medium --nr-inter 500 -F4 --no-psy-rdoq43
DucksAndLegs_1920x1080_60_10bit_444.yuv,--preset slower --no-weightp --rdoq-level 0 --limit-refs 3 --tu-inter-depth 4 --limit-tu 344
-DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset fast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 1::--preset fast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 145
+DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset fast --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 1::--preset fast --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --bitrate 3000 --early-skip --tu-inter-depth 3 --limit-tu 146
FourPeople_1280x720_60.y4m,--preset superfast --no-wpp --lookahead-slices 247
FourPeople_1280x720_60.y4m,--preset veryfast --aq-mode 2 --aq-strength 1.5 --qg-size 848
FourPeople_1280x720_60.y4m,--preset medium --qp 38 --no-psy-rd49
50
ducks_take_off_420_1_720p50.y4m,--preset medium --selective-sao 4 --sao --crf 2051
Traffic_4096x2048_30p.y4m, --preset medium --frame-dup --dup-threshold 60 --hrd --bitrate 10000 --vbv-bufsize 15000 --vbv-maxrate 1200052
Kimono1_1920x1080_24_400.yuv,--preset superfast --qp 28 --zones 0,139,q=3253
-sintel_trailer_2k_1920x1080_24.yuv, --preset medium --hist-scenecut --hist-threshold 0.02 --frame-dup --dup-threshold 60 --hrd --bitrate 10000 --vbv-bufsize 15000 --vbv-maxrate 1200054
-sintel_trailer_2k_1920x1080_24.yuv, --preset medium --hist-scenecut --hist-threshold 0.0255
-sintel_trailer_2k_1920x1080_24.yuv, --preset ultrafast --hist-scenecut --hist-threshold 0.0256
crowd_run_1920x1080_50.yuv, --preset faster --ctu 32 --rskip 2 --rskip-edge-threshold 557
crowd_run_1920x1080_50.yuv, --preset fast --ctu 64 --rskip 2 --rskip-edge-threshold 5 --aq-mode 458
-crowd_run_1920x1080_50.yuv, --preset slow --ctu 32 --rskip 2 --rskip-edge-threshold 5 --hist-scenecut --hist-threshold 0.159
-crowd_run_1920x1080_50.yuv, --preset slower --ctu 16 --rskip 2 --rskip-edge-threshold 5 --hist-scenecut --hist-threshold 0.1 --aq-mode 460
+crowd_run_1920x1080_50.yuv, --preset ultrafast --video-signal-type-preset BT2100_PQ_YCC:BT2100x108n000561
+crowd_run_1920x1080_50.yuv, --preset ultrafast --eob --eos62
63
# Main12 intraCost overflow bug test64
720p50_parkrun_ter.y4m,--preset medium65
66
67
#scaled save/load test68
crowd_run_1080p50.y4m,--preset ultrafast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 1 --scale-factor 2 --crf 26 --vbv-maxrate 8000 --vbv-bufsize 8000::crowd_run_2160p50.y4m, --preset ultrafast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 1 --scale-factor 2 --crf 26 --vbv-maxrate 12000 --vbv-bufsize 12000 69
-crowd_run_1080p50.y4m,--preset superfast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 5000 --vbv-bufsize 5000::crowd_run_2160p50.y4m, --preset superfast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 10000 --vbv-bufsize 10000 70
-crowd_run_1080p50.y4m,--preset fast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --scale-factor 2 --qp 18::crowd_run_2160p50.y4m, --preset fast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --scale-factor 2 --qp 1871
+crowd_run_1080p50.y4m,--preset superfast --analysis-save x265_analysis.dat --analysis-save-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 5000 --vbv-bufsize 5000::crowd_run_2160p50.y4m, --preset superfast --analysis-load x265_analysis.dat --analysis-load-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 10000 --vbv-bufsize 10000 72
+crowd_run_1080p50.y4m,--preset fast --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --scale-factor 2 --qp 18::crowd_run_2160p50.y4m, --preset fast --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --scale-factor 2 --qp 1873
crowd_run_1080p50.y4m,--preset medium --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 5000 --vbv-maxrate 5000 --vbv-bufsize 5000 --early-skip --tu-inter-depth 3::crowd_run_2160p50.y4m, --preset medium --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 10000 --early-skip --tu-inter-depth 3 --refine-intra 4 --dynamic-refine::crowd_run_2160p50.y4m, --preset medium --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 10000 --early-skip --tu-inter-depth 3 --refine-intra 3 --refine-inter 374
-RaceHorses_416x240_30.y4m,--preset slow --no-cutree --ctu 16 --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --scale-factor 2 --crf 22 --vbv-maxrate 1000 --vbv-bufsize 1000::RaceHorses_832x480_30.y4m, --preset slow --no-cutree --ctu 32 --analysis-load x265_analysis.dat --analysis-save x265_analysis_2.dat --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --crf 16 --vbv-maxrate 4000 --vbv-bufsize 4000 --refine-intra 0 --refine-inter 1::RaceHorses_1664x960_30.y4m,--preset slow --no-cutree --ctu 64 --analysis-load x265_analysis_2.dat --analysis-load-reuse-level 10 --scale-factor 2 --crf 12 --vbv-maxrate 7000 --vbv-bufsize 7000 --refine-intra 2 --refine-inter 275
+RaceHorses_416x240_30.y4m,--preset slow --ctu 16 --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --scale-factor 2 --crf 22 --vbv-maxrate 1000 --vbv-bufsize 1000::RaceHorses_832x480_30.y4m, --preset slow --ctu 32 --analysis-load x265_analysis.dat --analysis-save x265_analysis_2.dat --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --crf 16 --vbv-maxrate 4000 --vbv-bufsize 4000 --refine-intra 0 --refine-inter 1::RaceHorses_1664x960_30.y4m,--preset slow --ctu 64 --analysis-load x265_analysis_2.dat --analysis-load-reuse-level 10 --scale-factor 2 --crf 12 --vbv-maxrate 7000 --vbv-bufsize 7000 --refine-intra 2 --refine-inter 276
ElFunete_960x540_60.yuv,--colorprim bt709 --transfer bt709 --chromaloc 2 --aud --repeat-headers --no-opt-qp-pps --no-opt-ref-list-length-pps --wpp --no-interlace --sar 1:1 --min-keyint 60 --no-open-gop --rc-lookahead 180 --bframes 5 --b-intra --ref 4 --cbqpoffs -2 --crqpoffs -2 --lookahead-threads 0 --weightb --qg-size 8 --me star --preset veryslow --frame-threads 1 --b-adapt 2 --aq-mode 3 --rd 6 --pools 15 --colormatrix bt709 --keyint 120 --high-tier --ctu 64 --tune psnr --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500 --analysis-save-reuse-level 10 --analysis-save elfuente_960x540.dat --scale-factor 2::ElFunete_1920x1080_60.yuv,--colorprim bt709 --transfer bt709 --chromaloc 2 --aud --repeat-headers --no-opt-qp-pps --no-opt-ref-list-length-pps --wpp --no-interlace --sar 1:1 --min-keyint 60 --no-open-gop --rc-lookahead 180 --bframes 5 --b-intra --ref 4 --cbqpoffs -2 --crqpoffs -2 --lookahead-threads 0 --weightb --qg-size 8 --me star --preset veryslow --frame-threads 1 --b-adapt 2 --aq-mode 3 --rd 6 --pools 15 --colormatrix bt709 --keyint 120 --high-tier --ctu 64 --tune psnr --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500 --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --analysis-save elfuente_1920x1080.dat --limit-tu 0 --scale-factor 2 --analysis-load elfuente_960x540.dat --refine-intra 4 --refine-inter 2::ElFuente_3840x2160_60.yuv,--colorprim bt709 --transfer bt709 --chromaloc 2 --aud --repeat-headers --no-opt-qp-pps --no-opt-ref-list-length-pps --wpp --no-interlace --sar 1:1 --min-keyint 60 --no-open-gop --rc-lookahead 180 --bframes 5 --b-intra --ref 4 --cbqpoffs -2 --crqpoffs -2 --lookahead-threads 0 --weightb --qg-size 8 --me star --preset veryslow --frame-threads 1 --b-adapt 2 --aq-mode 3 --rd 6 --pools 15 --colormatrix bt709 --keyint 120 --high-tier --ctu 64 --tune=psnr --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 49000 --analysis-load-reuse-level 10 --limit-tu 0 --scale-factor 2 --analysis-load elfuente_1920x1080.dat --refine-intra 4 --refine-inter 277
#save/load with ctu distortion refinement78
CrowdRun_1920x1080_50_10bit_422.yuv,--no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --refine-ctu-distortion 1 --bitrate 7000::--no-cutree --analysis-load x265_analysis.dat --refine-ctu-distortion 1 --bitrate 7000 --analysis-load-reuse-level 579
#segment encoding80
BasketballDrive_1920x1080_50.y4m, --preset ultrafast --no-open-gop --chunk-start 100 --chunk-end 20081
82
+#Test FG SEI message addition83
+#OldTownCross_1920x1080_50_10bit_422.yuv,--preset slower --tune grain --film-grain "OldTownCross_1920x1080_50_10bit_422.bin"84
+#RaceHorses_416x240_30_10bit.yuv,--preset ultrafast --signhide --colormatrix bt709 --film-grain "RaceHorses_416x240_30_10bit.bin"85
+86
+#Temporal layers tests87
+ducks_take_off_420_720p50.y4m,--preset slow --temporal-layers 3 --b-adapt 088
+parkrun_ter_720p50.y4m,--preset medium --temporal-layers 4 --b-adapt 089
+BasketballDrive_1920x1080_50.y4m, --preset medium --no-open-gop --keyint 50 --min-keyint 50 --temporal-layers 5 --b-adapt 090
# vim: tw=20091
x265_3.5.tar.gz/source/test/save-load-tests.txt -> x265_3.6.tar.gz/source/test/save-load-tests.txt
Changed
16
1
2
# not auto-detected.3
crowd_run_1080p50.y4m, --preset ultrafast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 1 --scale-factor 2 --crf 26 --vbv-maxrate 8000 --vbv-bufsize 8000::crowd_run_2160p50.y4m, --preset ultrafast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 1 --scale-factor 2 --crf 26 --vbv-maxrate 12000 --vbv-bufsize 120004
crowd_run_540p50.y4m, --preset ultrafast --no-cutree --analysis-save x265_analysis.dat --scale-factor 2 --crf 26 --vbv-maxrate 8000 --vbv-bufsize 8000::crowd_run_1080p50.y4m, --preset ultrafast --no-cutree --analysis-load x265_analysis.dat --scale-factor 2 --crf 26 --vbv-maxrate 12000 --vbv-bufsize 120005
-crowd_run_1080p50.y4m, --preset superfast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 5000 --vbv-bufsize 5000::crowd_run_2160p50.y4m, --preset superfast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 10000 --vbv-bufsize 100006
-crowd_run_1080p50.y4m, --preset fast --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --scale-factor 2 --qp 18::crowd_run_2160p50.y4m, --preset fast --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --scale-factor 2 --qp 187
-crowd_run_1080p50.y4m, --preset medium --no-cutree --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 5000 --vbv-maxrate 5000 --vbv-bufsize 5000 --early-skip --tu-inter-depth 3::crowd_run_2160p50.y4m, --preset medium --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 10000 --early-skip --tu-inter-depth 3 --refine-intra 4 --dynamic-refine::crowd_run_2160p50.y4m, --preset medium --no-cutree --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 10000 --early-skip --tu-inter-depth 3 --refine-intra 3 --refine-inter 38
+crowd_run_1080p50.y4m, --preset superfast --analysis-save x265_analysis.dat --analysis-save-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 5000 --vbv-bufsize 5000::crowd_run_2160p50.y4m, --preset superfast --analysis-load x265_analysis.dat --analysis-load-reuse-level 2 --scale-factor 2 --crf 22 --vbv-maxrate 10000 --vbv-bufsize 100009
+crowd_run_1080p50.y4m, --preset fast --analysis-save x265_analysis.dat --analysis-save-reuse-level 5 --scale-factor 2 --qp 18::crowd_run_2160p50.y4m, --preset fast --analysis-load x265_analysis.dat --analysis-load-reuse-level 5 --scale-factor 2 --qp 1810
+crowd_run_1080p50.y4m, --preset medium --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 5000 --vbv-maxrate 5000 --vbv-bufsize 5000 --early-skip --tu-inter-depth 3::crowd_run_2160p50.y4m, --preset medium --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 10000 --early-skip --tu-inter-depth 3 --refine-intra 4 --dynamic-refine::crowd_run_2160p50.y4m, --preset medium --analysis-load x265_analysis.dat --analysis-load-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-maxrate 10000 --vbv-bufsize 10000 --early-skip --tu-inter-depth 3 --refine-intra 3 --refine-inter 311
RaceHorses_416x240_30.y4m, --preset slow --no-cutree --ctu 16 --analysis-save x265_analysis.dat --analysis-save-reuse-level 10 --scale-factor 2 --crf 22 --vbv-maxrate 1000 --vbv-bufsize 1000::RaceHorses_832x480_30.y4m, --preset slow --no-cutree --ctu 32 --analysis-load x265_analysis.dat --analysis-save x265_analysis_2.dat --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --crf 16 --vbv-maxrate 4000 --vbv-bufsize 4000 --refine-intra 0 --refine-inter 1::RaceHorses_1664x960_30.y4m, --preset slow --no-cutree --ctu 64 --analysis-load x265_analysis_2.dat --analysis-load-reuse-level 10 --scale-factor 2 --crf 12 --vbv-maxrate 7000 --vbv-bufsize 7000 --refine-intra 2 --refine-inter 212
-crowd_run_540p50.y4m, --preset veryslow --no-cutree --analysis-save x265_analysis_540.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 5000 --vbv-bufsize 15000 --vbv-maxrate 9000::crowd_run_1080p50.y4m, --preset veryslow --no-cutree --analysis-save x265_analysis_1080.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500::crowd_run_1080p50.y4m, --preset veryslow --no-cutree --analysis-save x265_analysis_1080.dat --analysis-load x265_analysis_540.dat --refine-intra 4 --dynamic-refine --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500::crowd_run_2160p50.y4m, --preset veryslow --no-cutree --analysis-save x265_analysis_2160.dat --analysis-load x265_analysis_1080.dat --refine-intra 3 --dynamic-refine --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 49000::crowd_run_2160p50.y4m, --preset veryslow --no-cutree --analysis-load x265_analysis_2160.dat --refine-intra 2 --dynamic-refine --analysis-load-reuse-level 10 --scale-factor 1 --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 4900013
+crowd_run_540p50.y4m, --preset veryslow --analysis-save x265_analysis_540.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 5000 --vbv-bufsize 15000 --vbv-maxrate 9000::crowd_run_1080p50.y4m, --preset veryslow --analysis-save x265_analysis_1080.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500::crowd_run_1080p50.y4m, --preset veryslow --analysis-save x265_analysis_1080.dat --analysis-load x265_analysis_540.dat --refine-intra 4 --dynamic-refine --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500::crowd_run_2160p50.y4m, --preset veryslow --analysis-save x265_analysis_2160.dat --analysis-load x265_analysis_1080.dat --refine-intra 3 --dynamic-refine --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 49000::crowd_run_2160p50.y4m, --preset veryslow --analysis-load x265_analysis_2160.dat --refine-intra 2 --dynamic-refine --analysis-load-reuse-level 10 --scale-factor 1 --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 4900014
crowd_run_540p50.y4m, --preset medium --no-cutree --analysis-save x265_analysis_540.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 5000 --vbv-bufsize 15000 --vbv-maxrate 9000::crowd_run_1080p50.y4m, --preset medium --no-cutree --analysis-save x265_analysis_1080.dat --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500::crowd_run_1080p50.y4m, --preset medium --no-cutree --analysis-save x265_analysis_1080.dat --analysis-load x265_analysis_540.dat --refine-intra 4 --dynamic-refine --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 10000 --vbv-bufsize 30000 --vbv-maxrate 17500::crowd_run_2160p50.y4m, --preset medium --no-cutree --analysis-save x265_analysis_2160.dat --analysis-load x265_analysis_1080.dat --refine-intra 3 --dynamic-refine --analysis-load-reuse-level 10 --analysis-save-reuse-level 10 --scale-factor 2 --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 49000::crowd_run_2160p50.y4m, --preset medium --no-cutree --analysis-load x265_analysis_2160.dat --refine-intra 2 --dynamic-refine --analysis-load-reuse-level 10 --scale-factor 1 --bitrate 24000 --vbv-bufsize 84000 --vbv-maxrate 4900015
News-4k.y4m, --preset medium --analysis-save x265_analysis_fdup.dat --frame-dup --hrd --bitrate 10000 --vbv-bufsize 15000 --vbv-maxrate 12000::News-4k.y4m, --analysis-load x265_analysis_fdup.dat --frame-dup --hrd --bitrate 10000 --vbv-bufsize 15000 --vbv-maxrate 1200016
x265_3.5.tar.gz/source/test/smoke-tests.txt -> x265_3.6.tar.gz/source/test/smoke-tests.txt
Changed
9
1
2
# Main12 intraCost overflow bug test3
720p50_parkrun_ter.y4m,--preset medium4
720p50_parkrun_ter.y4m,--preset=fast --hevc-aq --no-cutree5
+# Test FG SEI message addition6
+# CrowdRun_1920x1080_50_10bit_444.yuv,--preset=ultrafast --weightp --keyint -1 --film-grain "CrowdRun_1920x1080_50_10bit_444.bin"7
+# DucksAndLegs_1920x1080_60_10bit_422.yuv,--preset=veryfast --min-cu 16 --film-grain "DucksAndLegs_1920x1080_60_10bit_422.bin"8
+# NebutaFestival_2560x1600_60_10bit_crop.yuv,--preset=superfast --bitrate 10000 --sao --limit-sao --cll --max-cll "1000,400" --film-grain "NebutaFestival_2560x1600_60_10bit_crop.bin"9
x265_3.5.tar.gz/source/test/testbench.cpp -> x265_3.6.tar.gz/source/test/testbench.cpp
Changed
43
1
2
{ "AVX512", X265_CPU_AVX512 },3
{ "ARMv6", X265_CPU_ARMV6 },4
{ "NEON", X265_CPU_NEON },5
+ { "SVE2", X265_CPU_SVE2 },6
+ { "SVE", X265_CPU_SVE },7
{ "FastNeonMRC", X265_CPU_FAST_NEON_MRC },8
{ "", 0 },9
};10
11
12
EncoderPrimitives asmprim;13
memset(&asmprim, 0, sizeof(asmprim));14
- setupAssemblyPrimitives(asmprim, test_archi.flag);15
-16
-#if X265_ARCH_ARM6417
- /* Temporary workaround because luma_vsp assembly primitive has not been completed18
- * but interp_8tap_hv_pp_cpu uses mixed C primitive and assembly primitive.19
- * Otherwise, segment fault occurs. */20
- setupAliasCPrimitives(cprim, asmprim, test_archi.flag);21
-#endif22
23
+ setupAssemblyPrimitives(asmprim, test_archi.flag);24
setupAliasPrimitives(asmprim);25
memcpy(&primitives, &asmprim, sizeof(EncoderPrimitives));26
for (size_t h = 0; h < sizeof(harness) / sizeof(TestHarness*); h++)27
28
#if X265_ARCH_X8629
setupInstrinsicPrimitives(optprim, cpuid);30
#endif31
- setupAssemblyPrimitives(optprim, cpuid);32
33
-#if X265_ARCH_ARM6434
- /* Temporary workaround because luma_vsp assembly primitive has not been completed35
- * but interp_8tap_hv_pp_cpu uses mixed C primitive and assembly primitive.36
- * Otherwise, segment fault occurs. */37
- setupAliasCPrimitives(cprim, optprim, cpuid);38
-#endif39
+ setupAssemblyPrimitives(optprim, cpuid);40
41
/* Note that we do not setup aliases for performance tests, that would be42
* redundant. The testbench only verifies they are correctly aliased */43
x265_3.5.tar.gz/source/test/testharness.h -> x265_3.6.tar.gz/source/test/testharness.h
Changed
48
1
2
#include <x86intrin.h>3
#elif ( !defined(__APPLE__) && defined (__GNUC__) && defined(__ARM_NEON__))4
#include <arm_neon.h>5
-#elif defined(__GNUC__) && (!defined(__clang__) || __clang_major__ < 4)6
+#else7
/* fallback for older GCC/MinGW */8
static inline uint32_t __rdtsc(void)9
{10
11
#if X265_ARCH_X8612
asm volatile("rdtsc" : "=a" (a) ::"edx");13
#elif X265_ARCH_ARM14
-#if X265_ARCH_ARM6415
- asm volatile("mrs %0, cntvct_el0" : "=r"(a));16
-#else17
// TOD-DO: verify following inline asm to get cpu Timestamp Counter for ARM arch18
// asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r"(a));19
20
// TO-DO: replace clock() function with appropriate ARM cpu instructions21
a = clock();22
-#endif23
+#elif X265_ARCH_ARM6424
+ asm volatile("mrs %0, cntvct_el0" : "=r"(a));25
#endif26
return a;27
}28
29
x265_emms(); \30
float optperf = (10.0f * cycles / runs) / 4; \31
float refperf = (10.0f * refcycles / refruns) / 4; \32
- printf("\t%3.2fx ", refperf / optperf); \33
- printf("\t %-8.2lf \t %-8.2lf\n", optperf, refperf); \34
+ printf(" | \t%3.2fx | ", refperf / optperf); \35
+ printf("\t %-8.2lf | \t %-8.2lf\n", optperf, refperf); \36
}37
38
extern "C" {39
40
* needs an explicit asm check because it only sometimes crashes in normal use. */41
intptr_t PFX(checkasm_call)(intptr_t (*func)(), int *ok, ...);42
float PFX(checkasm_call_float)(float (*func)(), int *ok, ...);43
-#elif X265_ARCH_ARM == 044
+#elif (X265_ARCH_ARM == 0 && X265_ARCH_ARM64 == 0)45
#define PFX(stack_pagealign)(func, align) func()46
#endif47
48
x265_3.5.tar.gz/source/x265.cpp -> x265_3.6.tar.gz/source/x265.cpp
Changed
18
1
2
3
int ret = 0;4
5
+ if (cliopt0.scenecutAwareQpConfig)6
+ {7
+ if (!cliopt0.parseScenecutAwareQpConfig())8
+ {9
+ x265_log(NULL, X265_LOG_ERROR, "Unable to parse scenecut aware qp config file \n");10
+ fclose(cliopt0.scenecutAwareQpConfig);11
+ cliopt0.scenecutAwareQpConfig = NULL;12
+ }13
+ }14
+15
AbrEncoder* abrEnc = new AbrEncoder(cliopt, numEncodes, ret);16
int threadsActive = abrEnc->m_numActiveEncodes.get();17
while (threadsActive)18
x265_3.5.tar.gz/source/x265.h -> x265_3.6.tar.gz/source/x265.h
Changed
470
1
2
#define X265_H3
#include <stdint.h>4
#include <stdio.h>5
+#include <sys/stat.h>6
#include "x265_config.h"7
#ifdef __cplusplus8
extern "C" {9
10
NAL_UNIT_CODED_SLICE_TRAIL_N = 0,11
NAL_UNIT_CODED_SLICE_TRAIL_R,12
NAL_UNIT_CODED_SLICE_TSA_N,13
- NAL_UNIT_CODED_SLICE_TLA_R,14
+ NAL_UNIT_CODED_SLICE_TSA_R,15
NAL_UNIT_CODED_SLICE_STSA_N,16
NAL_UNIT_CODED_SLICE_STSA_R,17
NAL_UNIT_CODED_SLICE_RADL_N,18
19
double vmafFrameScore;20
double bufferFillFinal;21
double unclippedBufferFillFinal;22
+ uint8_t tLayer;23
} x265_frame_stats;24
25
typedef struct x265_ctu_info_t26
27
/* ARM */28
#define X265_CPU_ARMV6 0x000000129
#define X265_CPU_NEON 0x0000002 /* ARM NEON */30
+#define X265_CPU_SVE2 0x0000008 /* ARM SVE2 */31
+#define X265_CPU_SVE 0x0000010 /* ARM SVE2 */32
#define X265_CPU_FAST_NEON_MRC 0x0000004 /* Transfer from NEON to ARM register is fast (Cortex-A9) */33
34
/* IBM Power8 */35
36
#define SLICE_TYPE_DELTA 0.3 /* The offset decremented or incremented for P-frames or b-frames respectively*/37
#define BACKWARD_WINDOW 1 /* Scenecut window before a scenecut */38
#define FORWARD_WINDOW 2 /* Scenecut window after a scenecut */39
+#define BWD_WINDOW_DELTA 0.440
+41
+#define X265_MAX_GOP_CONFIG 342
+#define X265_MAX_GOP_LENGTH 1643
+#define MAX_T_LAYERS 744
+45
+#define X265_IPRATIO_STRENGTH 1.4346
47
typedef struct x265_cli_csp48
{49
50
typedef struct x265_zone51
{52
int startFrame, endFrame; /* range of frame numbers */53
+ int keyframeMax; /* it store the default/user defined keyframeMax value*/54
int bForceQp; /* whether to use qp vs bitrate factor */55
int qp;56
float bitrateFactor;57
58
59
static const x265_vmaf_commondata vcd = { { NULL, (char *)"/usr/local/share/model/vmaf_v0.6.1.pkl", NULL, NULL, 0, 0, 0, 0, 0, 0, 0, NULL, 0, 1, 0 } };60
61
+typedef struct x265_temporal_layer {62
+ int poc_offset; /* POC offset */63
+ int8_t layer; /* Current layer */64
+ int8_t qp_offset; /* QP offset */65
+} x265_temporal_layer;66
+67
+static const int8_t x265_temporal_layer_bframesMAX_T_LAYERS = {-1, -1, 3, 7, 15, -1, -1};68
+69
+static const int8_t x265_gop_ra_lengthX265_MAX_GOP_CONFIG = { 4, 8, 16};70
+static const x265_temporal_layer x265_gop_raX265_MAX_GOP_CONFIGX265_MAX_GOP_LENGTH = {71
+ {72
+ {73
+ 4,74
+ 0,75
+ 1,76
+ },77
+ {78
+ 2,79
+ 1,80
+ 5,81
+ },82
+ {83
+ 1,84
+ 2,85
+ 3,86
+ },87
+ {88
+ 3,89
+ 2,90
+ 5,91
+ },92
+ {93
+ -1,94
+ -1,95
+ -1,96
+ },97
+ {98
+ -1,99
+ -1,100
+ -1,101
+ },102
+ {103
+ -1,104
+ -1,105
+ -1,106
+ },107
+ {108
+ -1,109
+ -1,110
+ -1,111
+ },112
+ {113
+ -1,114
+ -1,115
+ -1,116
+ },117
+ {118
+ -1,119
+ -1,120
+ -1,121
+ },122
+ {123
+ -1,124
+ -1,125
+ -1,126
+ },127
+ {128
+ -1,129
+ -1,130
+ -1,131
+ },132
+ {133
+ -1,134
+ -1,135
+ -1,136
+ },137
+ {138
+ -1,139
+ -1,140
+ -1,141
+ },142
+ {143
+ -1,144
+ -1,145
+ -1,146
+ },147
+ {148
+ -1,149
+ -1,150
+ -1,151
+ }152
+ },153
+154
+ {155
+ {156
+ 8,157
+ 0,158
+ 1,159
+ },160
+ {161
+ 4,162
+ 1,163
+ 5,164
+ },165
+ {166
+ 2,167
+ 2,168
+ 4,169
+ },170
+ {171
+ 1,172
+ 3,173
+ 5,174
+ },175
+ {176
+ 3,177
+ 3,178
+ 2,179
+ },180
+ {181
+ 6,182
+ 2,183
+ 5,184
+ },185
+ {186
+ 5,187
+ 3,188
+ 4,189
+ },190
+ {191
+ 7,192
+ 3,193
+ 5,194
+ },195
+ {196
+ -1,197
+ -1,198
+ -1,199
+ },200
+ {201
+ -1,202
+ -1,203
+ -1,204
+ },205
+ {206
+ -1,207
+ -1,208
+ -1,209
+ },210
+ {211
+ -1,212
+ -1,213
+ -1,214
+ },215
+ {216
+ -1,217
+ -1,218
+ -1,219
+ },220
+ {221
+ -1,222
+ -1,223
+ -1,224
+ },225
+ {226
+ -1,227
+ -1,228
+ -1,229
+ },230
+ {231
+ -1,232
+ -1,233
+ -1,234
+ },235
+ },236
+ {237
+ {238
+ 16,239
+ 0,240
+ 1,241
+ },242
+ {243
+ 8,244
+ 1,245
+ 6,246
+ },247
+ {248
+ 4,249
+ 2,250
+ 5,251
+ },252
+ {253
+ 2,254
+ 3,255
+ 6,256
+ },257
+ {258
+ 1,259
+ 4,260
+ 4,261
+ },262
+ {263
+ 3,264
+ 4,265
+ 6,266
+ },267
+ {268
+ 6,269
+ 3,270
+ 5,271
+ },272
+ {273
+ 5,274
+ 4,275
+ 6,276
+ },277
+ {278
+ 7,279
+ 4,280
+ 1,281
+ },282
+ {283
+ 12,284
+ 2,285
+ 6,286
+ },287
+ {288
+ 10,289
+ 3,290
+ 5,291
+ },292
+ {293
+ 9,294
+ 4,295
+ 6,296
+ },297
+ {298
+ 11,299
+ 4,300
+ 4,301
+ },302
+ {303
+ 14,304
+ 3,305
+ 6,306
+ },307
+ {308
+ 13,309
+ 4,310
+ 5,311
+ },312
+ {313
+ 15,314
+ 4,315
+ 6,316
+ }317
+ }318
+};319
+320
+typedef enum321
+{322
+ X265_SHARE_MODE_FILE = 0,323
+ X265_SHARE_MODE_SHAREDMEM324
+}X265_DATA_SHARE_MODES;325
+326
/* x265 input parameters327
*328
* For version safety you may use x265_param_alloc/free() to manage the329
330
* performance impact, but the use case may preclude it. Default true */331
int bOpenGOP;332
333
+ /*Force nal type to CRA to all frames expect first frame. Default disabled*/334
+ int craNal;335
+336
/* Scene cuts closer together than this are coded as I, not IDR. */337
int keyframeMin;338
339
340
double rfConstantMin;341
342
/* Multi-pass encoding */343
- /* Enable writing the stats in a multi-pass encode to the stat output file */344
+ /* Enable writing the stats in a multi-pass encode to the stat output file/memory */345
int bStatWrite;346
347
- /* Enable loading data from the stat input file in a multi pass encode */348
+ /* Enable loading data from the stat input file/memory in a multi pass encode */349
int bStatRead;350
351
/* Filename of the 2pass output/input stats file, if unspecified the352
353
/* internally enable if tune grain is set */354
int bEnableConstVbv;355
356
+ /* if only the focused frames would be re-encode or not */357
+ int bEncFocusedFramesOnly;358
+359
+ /* Share the data with stats file or shared memory.360
+ It must be one of the X265_DATA_SHARE_MODES enum values361
+ Available if the bStatWrite or bStatRead is true.362
+ Use stats file by default.363
+ The stats file mode would be used among the encoders running in sequence.364
+ The shared memory mode could only be used among the encoders running in parallel.365
+ Now only the cutree data could be shared among shared memory. More data would be support in the future.*/366
+ int dataShareMode;367
+368
+ /* Unique shared memory name. Required if the shared memory mode enabled. NULL by default */369
+ const char* sharedMemName;370
+371
} rc;372
373
/*== Video Usability Information ==*/374
375
Default 1 (Enabled). API only. */376
int bResetZoneConfig;377
378
+ /*Flag to indicate rate-control history has not to be reset during zone reconfiguration.379
+ Default 0 (Disabled) */380
+ int bNoResetZoneConfig;381
+382
/* It reduces the bits spent on the inter-frames within the scenecutWindow before and / or after a scenecut383
* by increasing their QP in ratecontrol pass2 algorithm without any deterioration in visual quality.384
* 0 - Disabled (default).385
386
387
/* The duration(in milliseconds) for which there is a reduction in the bits spent on the inter-frames after a scenecut388
* by increasing their QP, when bEnableSceneCutAwareQp is 1 or 3. Default is 500ms.*/389
- int fwdScenecutWindow;390
+ int fwdMaxScenecutWindow;391
+ int fwdScenecutWindow6;392
393
/* The offset by which QP is incremented for inter-frames after a scenecut when bEnableSceneCutAwareQp is 1 or 3.394
* Default is +5. */395
- double fwdRefQpDelta;396
+ double fwdRefQpDelta6;397
398
/* The offset by which QP is incremented for non-referenced inter-frames after a scenecut when bEnableSceneCutAwareQp is 1 or 3. */399
- double fwdNonRefQpDelta;400
-401
- /* A genuine threshold used for histogram based scene cut detection.402
- * This threshold determines whether a frame is a scenecut or not403
- * when compared against the edge and chroma histogram sad values.404
- * Default 0.03. Range: Real number in the interval (0,1). */405
- double edgeTransitionThreshold;406
+ double fwdNonRefQpDelta6;407
408
/* Enables histogram based scenecut detection algorithm to detect scenecuts. Default disabled */409
int bHistBasedSceneCut;410
411
412
/* The duration(in milliseconds) for which there is a reduction in the bits spent on the inter-frames before a scenecut413
* by increasing their QP, when bEnableSceneCutAwareQp is 2 or 3. Default is 100ms.*/414
- int bwdScenecutWindow;415
+ int bwdMaxScenecutWindow;416
+ int bwdScenecutWindow6;417
418
/* The offset by which QP is incremented for inter-frames before a scenecut when bEnableSceneCutAwareQp is 2 or 3. */419
- double bwdRefQpDelta;420
+ double bwdRefQpDelta6;421
422
/* The offset by which QP is incremented for non-referenced inter-frames before a scenecut when bEnableSceneCutAwareQp is 2 or 3. */423
- double bwdNonRefQpDelta;424
+ double bwdNonRefQpDelta6;425
+426
+ /* Specify combinations of color primaries, transfer characteristics, color matrix,427
+ * range of luma and chroma signals, and chroma sample location. This has higher428
+ * precedence than individual VUI parameters. If any individual VUI option is specified429
+ * together with this, which changes the values set corresponding to the system-id430
+ * or color-volume, it will be discarded. */431
+ const char* videoSignalTypePreset;432
+433
+ /* Flag indicating whether the encoder should emit an End of Bitstream434
+ * NAL at the end of bitstream. Default false */435
+ int bEnableEndOfBitstream;436
+437
+ /* Flag indicating whether the encoder should emit an End of Sequence438
+ * NAL at the end of every Coded Video Sequence. Default false */439
+ int bEnableEndOfSequence;440
+441
+ /* Film Grain Characteristic file */442
+ char* filmGrain;443
+444
+ /*Motion compensated temporal filter*/445
+ int bEnableTemporalFilter;446
+ double temporalFilterStrength;447
+448
+ /*SBRC*/449
+ int bEnableSBRC;450
} x265_param;451
452
/* x265_param_alloc:453
454
455
int x265_zone_param_parse(x265_param* p, const char* name, const char* value);456
457
+int x265_scenecut_aware_qp_param_parse(x265_param* p, const char* name, const char* value);458
+459
static const char * const x265_profile_names = {460
/* HEVC v1 */461
"main", "main10", "mainstillpicture", /* alias */ "msp",462
463
void (*param_free)(x265_param*);464
void (*param_default)(x265_param*);465
int (*param_parse)(x265_param*, const char*, const char*);466
+ int (*scenecut_aware_qp_param_parse)(x265_param*, const char*, const char*);467
int (*param_apply_profile)(x265_param*, const char*);468
int (*param_default_preset)(x265_param*, const char*, const char *);469
x265_picture* (*picture_alloc)(void);470
x265_3.5.tar.gz/source/x265cli.cpp -> x265_3.6.tar.gz/source/x265cli.cpp
Changed
393
1
2
#include "x265cli.h"3
#include "svt.h"4
5
-#define START_CODE 0x000000016
-#define START_CODE_BYTES 47
+#define START_CODE 0x000000018
+#define START_CODE_BYTES 49
10
#ifdef __cplusplus11
namespace X265_NS {12
13
H0(" --rdpenalty <0..2> penalty for 32x32 intra TU in non-I slices. 0:disabled 1:RD-penalty 2:maximum. Default %d\n", param->rdPenalty);14
H0("\nSlice decision options:\n");15
H0(" --no-open-gop Enable open-GOP, allows I slices to be non-IDR. Default %s\n", OPT(param->bOpenGOP));16
+ H0(" --cra-nal Force nal type to CRA to all frames expect first frame, works only with keyint 1. Default %s\n", OPT(param->craNal));17
H0("-I/--keyint <integer> Max IDR period in frames. -1 for infinite-gop. Default %d\n", param->keyframeMax);18
H0("-i/--min-keyint <integer> Scenecuts closer together than this are coded as I, not IDR. Default: auto\n");19
H0(" --gop-lookahead <integer> Extends gop boundary if a scenecut is found within this from keyint boundary. Default 0\n");20
21
H1(" --scenecut-bias <0..100.0> Bias for scenecut detection. Default %.2f\n", param->scenecutBias);22
H0(" --hist-scenecut Enables histogram based scene-cut detection using histogram based algorithm.\n");23
H0(" --no-hist-scenecut Disables histogram based scene-cut detection using histogram based algorithm.\n");24
- H1(" --hist-threshold <0.0..1.0> Luma Edge histogram's Normalized SAD threshold for histogram based scenecut detection Default %.2f\n", param->edgeTransitionThreshold);25
H0(" --no-fades Enable detection and handling of fade-in regions. Default %s\n", OPT(param->bEnableFades));26
H1(" --scenecut-aware-qp <0..3> Enable increasing QP for frames inside the scenecut window around scenecut. Default %s\n", OPT(param->bEnableSceneCutAwareQp));27
H1(" 0 - Disabled\n");28
29
H1(" 2 - Backward masking\n");30
H1(" 3 - Bidirectional masking\n");31
H1(" --masking-strength <string> Comma separated values which specify the duration and offset for the QP increment for inter-frames when scenecut-aware-qp is enabled.\n");32
+ H1(" --scenecut-qp-config <file> File containing scenecut-aware-qp mode, window duration and offsets settings required for the masking. Works only with --pass 2\n");33
H0(" --radl <integer> Number of RADL pictures allowed in front of IDR. Default %d\n", param->radl);34
H0(" --intra-refresh Use Periodic Intra Refresh instead of IDR frames\n");35
H0(" --rc-lookahead <integer> Number of frames for frame-type lookahead (determines encoder latency) Default %d\n", param->lookaheadDepth);36
37
H0(" --aq-strength <float> Reduces blocking and blurring in flat and textured areas (0 to 3.0). Default %.2f\n", param->rc.aqStrength);38
H0(" --qp-adaptation-range <float> Delta QP range by QP adaptation based on a psycho-visual model (1.0 to 6.0). Default %.2f\n", param->rc.qpAdaptationRange);39
H0(" --no-aq-motion Block level QP adaptation based on the relative motion between the block and the frame. Default %s\n", OPT(param->bAQMotion));40
+ H1(" --no-sbrc Enables the segment based rate control. Default %s\n", OPT(param->bEnableSBRC));41
H0(" --qg-size <int> Specifies the size of the quantization group (64, 32, 16, 8). Default %d\n", param->rc.qgSize);42
H0(" --no-cutree Enable cutree for Adaptive Quantization. Default %s\n", OPT(param->rc.cuTree));43
H0(" --no-rc-grain Enable ratecontrol mode to handle grains specifically. turned on with tune grain. Default %s\n", OPT(param->rc.bEnableGrain));44
45
H1(" q=<integer> (force QP)\n");46
H1(" or b=<float> (bitrate multiplier)\n");47
H0(" --zonefile <filename> Zone file containing the zone boundaries and the parameters to be reconfigured.\n");48
+ H0(" --no-zonefile-rc-init This allow to use rate-control history across zones in zonefile.\n");49
H1(" --lambda-file <string> Specify a file containing replacement values for the lambda tables\n");50
H1(" MAX_MAX_QP+1 floats for lambda table, then again for lambda2 table\n");51
H1(" Blank lines and lines starting with hash(#) are ignored\n");52
53
H0(" --master-display <string> SMPTE ST 2086 master display color volume info SEI (HDR)\n");54
H0(" format: G(x,y)B(x,y)R(x,y)WP(x,y)L(max,min)\n");55
H0(" --max-cll <string> Specify content light level info SEI as \"cll,fall\" (HDR).\n");56
+ H0(" --video-signal-type-preset <string> Specify combinations of color primaries, transfer characteristics, color matrix, range of luma and chroma signals, and chroma sample location\n");57
+ H0(" format: <system-id>:<color-volume>\n");58
+ H0(" This has higher precedence than individual VUI parameters. If any individual VUI option is specified together with this,\n");59
+ H0(" which changes the values set corresponding to the system-id or color-volume, it will be discarded.\n");60
+ H0(" The color-volume can be used only with the system-id options BT2100_PQ_YCC, BT2100_PQ_ICTCP, and BT2100_PQ_RGB.\n");61
+ H0(" system-id options and their corresponding values:\n");62
+ H0(" BT601_525: --colorprim smpte170m --transfer smpte170m --colormatrix smpte170m --range limited --chromaloc 0\n");63
+ H0(" BT601_626: --colorprim bt470bg --transfer smpte170m --colormatrix bt470bg --range limited --chromaloc 0\n");64
+ H0(" BT709_YCC: --colorprim bt709 --transfer bt709 --colormatrix bt709 --range limited --chromaloc 0\n");65
+ H0(" BT709_RGB: --colorprim bt709 --transfer bt709 --colormatrix gbr --range limited\n");66
+ H0(" BT2020_YCC_NCL: --colorprim bt2020 --transfer bt2020-10 --colormatrix bt709 --range limited --chromaloc 2\n");67
+ H0(" BT2020_RGB: --colorprim bt2020 --transfer smpte2084 --colormatrix bt2020nc --range limited\n");68
+ H0(" BT2100_PQ_YCC: --colorprim bt2020 --transfer smpte2084 --colormatrix bt2020nc --range limited --chromaloc 2\n");69
+ H0(" BT2100_PQ_ICTCP: --colorprim bt2020 --transfer smpte2084 --colormatrix ictcp --range limited --chromaloc 2\n");70
+ H0(" BT2100_PQ_RGB: --colorprim bt2020 --transfer smpte2084 --colormatrix gbr --range limited\n");71
+ H0(" BT2100_HLG_YCC: --colorprim bt2020 --transfer arib-std-b67 --colormatrix bt2020nc --range limited --chromaloc 2\n");72
+ H0(" BT2100_HLG_RGB: --colorprim bt2020 --transfer arib-std-b67 --colormatrix gbr --range limited\n");73
+ H0(" FR709_RGB: --colorprim bt709 --transfer bt709 --colormatrix gbr --range full\n");74
+ H0(" FR2020_RGB: --colorprim bt2020 --transfer bt2020-10 --colormatrix gbr --range full\n");75
+ H0(" FRP3D65_YCC: --colorprim smpte432 --transfer bt709 --colormatrix smpte170m --range full --chromaloc 1\n");76
+ H0(" color-volume options and their corresponding values:\n");77
+ H0(" P3D65x1000n0005: --master-display G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(10000000,5)\n");78
+ H0(" P3D65x4000n005: --master-display G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(40000000,50)\n");79
+ H0(" BT2100x108n0005: --master-display G(8500,39850)B(6550,2300)R(34000,146000)WP(15635,16450)L(10000000,1)\n");80
H0(" --no-cll Emit content light level info SEI. Default %s\n", OPT(param->bEmitCLL));81
H0(" --no-hdr10 Control dumping of HDR10 SEI packet. If max-cll or master-display has non-zero values, this is enabled. Default %s\n", OPT(param->bEmitHDR10SEI));82
H0(" --no-hdr-opt Add luma and chroma offsets for HDR/WCG content. Default %s. Now deprecated.\n", OPT(param->bHDROpt));83
84
H0(" --no-repeat-headers Emit SPS and PPS headers at each keyframe. Default %s\n", OPT(param->bRepeatHeaders));85
H0(" --no-info Emit SEI identifying encoder and parameters. Default %s\n", OPT(param->bEmitInfoSEI));86
H0(" --no-hrd Enable HRD parameters signaling. Default %s\n", OPT(param->bEmitHRDSEI));87
- H0(" --no-idr-recovery-sei Emit recovery point infor SEI at each IDR frame \n");88
- H0(" --no-temporal-layers Enable a temporal sublayer for unreferenced B frames. Default %s\n", OPT(param->bEnableTemporalSubLayers));89
+ H0(" --no-idr-recovery-sei Emit recovery point infor SEI at each IDR frame \n");90
+ H0(" --temporal-layers Enable a temporal sublayer for unreferenced B frames. Default %s\n", OPT(param->bEnableTemporalSubLayers));91
H0(" --no-aud Emit access unit delimiters at the start of each access unit. Default %s\n", OPT(param->bEnableAccessUnitDelimiters));92
+ H0(" --no-eob Emit end of bitstream nal unit at the end of the bitstream. Default %s\n", OPT(param->bEnableEndOfBitstream));93
+ H0(" --no-eos Emit end of sequence nal unit at the end of every coded video sequence. Default %s\n", OPT(param->bEnableEndOfSequence));94
H1(" --hash <integer> Decoded Picture Hash SEI 0: disabled, 1: MD5, 2: CRC, 3: Checksum. Default %d\n", param->decodedPictureHashSEI);95
H0(" --atc-sei <integer> Emit the alternative transfer characteristics SEI message where the integer is the preferred transfer characteristics. Default disabled\n");96
H0(" --pic-struct <integer> Set the picture structure and emits it in the picture timing SEI message. Values in the range 0..12. See D.3.3 of the HEVC spec. for a detailed explanation.\n");97
98
H0(" --lowpass-dct Use low-pass subband dct approximation. Default %s\n", OPT(param->bLowPassDct));99
H0(" --no-frame-dup Enable Frame duplication. Default %s\n", OPT(param->bEnableFrameDuplication));100
H0(" --dup-threshold <integer> PSNR threshold for Frame duplication. Default %d\n", param->dupThreshold);101
+ H0(" --no-mcstf Enable GOP based temporal filter. Default %d\n", param->bEnableTemporalFilter);102
#ifdef SVT_HEVC103
H0(" --nosvt Enable SVT HEVC encoder %s\n", OPT(param->bEnableSvtHevc));104
H0(" --no-svt-hme Enable Hierarchial motion estimation(HME) in SVT HEVC encoder \n");105
106
H1(" 2 - unable to open encoder\n");107
H1(" 3 - unable to generate stream headers\n");108
H1(" 4 - encoder abort\n");109
+ H0("\nSEI Message Options\n");110
+ H0(" --film-grain <filename> File containing Film Grain Characteristics to be written as a SEI Message\n");111
+112
#undef OPT113
#undef H0114
#undef H1115
116
117
memcpy(globalParam->rc.zoneszonefileCount.zoneParam, globalParam, sizeof(x265_param));118
119
+ if (zonefileCount == 0)120
+ globalParam->rc.zoneszonefileCount.keyframeMax = globalParam->keyframeMax;121
+122
for (optind = 0;;)123
{124
int long_options_index = -1;125
126
return true;127
}128
}129
+ OPT("scenecut-qp-config")130
+ {131
+ this->scenecutAwareQpConfig = x265_fopen(optarg, "rb");132
+ if (!this->scenecutAwareQpConfig)133
+ x265_log_file(param, X265_LOG_ERROR, "%s scenecut aware qp config file not found or error in opening config file\n", optarg);134
+ }135
OPT("zonefile")136
{137
this->zoneFile = x265_fopen(optarg, "rb");138
if (!this->zoneFile)139
x265_log_file(param, X265_LOG_ERROR, "%s zone file not found or error in opening zone file\n", optarg);140
}141
+ OPT("no-zonefile-rc-init") this->param->bNoResetZoneConfig = true;142
OPT("fullhelp")143
{144
param->logLevel = X265_LOG_FULL;145
146
if (reconFileBitDepth == 0)147
reconFileBitDepth = param->internalBitDepth;148
this->recon = ReconFile::open(reconfn, param->sourceWidth, param->sourceHeight, reconFileBitDepth,149
- param->fpsNum, param->fpsDenom, param->internalCsp);150
+ param->fpsNum, param->fpsDenom, param->internalCsp, param->sourceBitDepth);151
if (this->recon->isFail())152
{153
x265_log(param, X265_LOG_WARNING, "unable to write reconstructed outputs file\n");154
155
param->rc.zones = X265_MALLOC(x265_zone, param->rc.zonefileCount);156
for (int i = 0; i < param->rc.zonefileCount; i++)157
{158
+ param->rc.zonesi.startFrame = -1;159
while (fgets(line, sizeof(line), zoneFile))160
{161
if (*line == '#' || (strcmp(line, "\r\n") == 0))162
163
return 1;164
}165
166
- /* Parse the RPU file and extract the RPU corresponding to the current picture167
- * and fill the rpu field of the input picture */168
- int CLIOptions::rpuParser(x265_picture * pic)169
- {170
- uint8_t byteVal;171
- uint32_t code = 0;172
- int bytesRead = 0;173
- pic->rpu.payloadSize = 0;174
-175
- if (!pic->pts)176
- {177
- while (bytesRead++ < 4 && fread(&byteVal, sizeof(uint8_t), 1, dolbyVisionRpu))178
- code = (code << 8) | byteVal;179
-180
- if (code != START_CODE)181
- {182
- x265_log(NULL, X265_LOG_ERROR, "Invalid Dolby Vision RPU startcode in POC %d\n", pic->pts);183
- return 1;184
- }185
- }186
-187
- bytesRead = 0;188
- while (fread(&byteVal, sizeof(uint8_t), 1, dolbyVisionRpu))189
- {190
- code = (code << 8) | byteVal;191
- if (bytesRead++ < 3)192
- continue;193
- if (bytesRead >= 1024)194
- {195
- x265_log(NULL, X265_LOG_ERROR, "Invalid Dolby Vision RPU size in POC %d\n", pic->pts);196
- return 1;197
- }198
-199
- if (code != START_CODE)200
- pic->rpu.payloadpic->rpu.payloadSize++ = (code >> (3 * 8)) & 0xFF;201
- else202
- return 0;203
- }204
-205
- int ShiftBytes = START_CODE_BYTES - (bytesRead - pic->rpu.payloadSize);206
- int bytesLeft = bytesRead - pic->rpu.payloadSize;207
- code = (code << ShiftBytes * 8);208
- for (int i = 0; i < bytesLeft; i++)209
- {210
- pic->rpu.payloadpic->rpu.payloadSize++ = (code >> (3 * 8)) & 0xFF;211
- code = (code << 8);212
- }213
- if (!pic->rpu.payloadSize)214
- x265_log(NULL, X265_LOG_WARNING, "Dolby Vision RPU not found for POC %d\n", pic->pts);215
- return 0;216
- }217
+ /* Parse the RPU file and extract the RPU corresponding to the current picture218
+ * and fill the rpu field of the input picture */219
+ int CLIOptions::rpuParser(x265_picture * pic)220
+ {221
+ uint8_t byteVal;222
+ uint32_t code = 0;223
+ int bytesRead = 0;224
+ pic->rpu.payloadSize = 0;225
+226
+ if (!pic->pts)227
+ {228
+ while (bytesRead++ < 4 && fread(&byteVal, sizeof(uint8_t), 1, dolbyVisionRpu))229
+ code = (code << 8) | byteVal;230
+231
+ if (code != START_CODE)232
+ {233
+ x265_log(NULL, X265_LOG_ERROR, "Invalid Dolby Vision RPU startcode in POC %d\n", pic->pts);234
+ return 1;235
+ }236
+ }237
+238
+ bytesRead = 0;239
+ while (fread(&byteVal, sizeof(uint8_t), 1, dolbyVisionRpu))240
+ {241
+ code = (code << 8) | byteVal;242
+ if (bytesRead++ < 3)243
+ continue;244
+ if (bytesRead >= 1024)245
+ {246
+ x265_log(NULL, X265_LOG_ERROR, "Invalid Dolby Vision RPU size in POC %d\n", pic->pts);247
+ return 1;248
+ }249
+250
+ if (code != START_CODE)251
+ pic->rpu.payloadpic->rpu.payloadSize++ = (code >> (3 * 8)) & 0xFF;252
+ else253
+ return 0;254
+ }255
+256
+ int ShiftBytes = START_CODE_BYTES - (bytesRead - pic->rpu.payloadSize);257
+ int bytesLeft = bytesRead - pic->rpu.payloadSize;258
+ code = (code << ShiftBytes * 8);259
+ for (int i = 0; i < bytesLeft; i++)260
+ {261
+ pic->rpu.payloadpic->rpu.payloadSize++ = (code >> (3 * 8)) & 0xFF;262
+ code = (code << 8);263
+ }264
+ if (!pic->rpu.payloadSize)265
+ x265_log(NULL, X265_LOG_WARNING, "Dolby Vision RPU not found for POC %d\n", pic->pts);266
+ return 0;267
+ }268
+269
+ bool CLIOptions::parseScenecutAwareQpConfig()270
+ {271
+ char line256;272
+ char* argLine;273
+ rewind(scenecutAwareQpConfig);274
+ while (fgets(line, sizeof(line), scenecutAwareQpConfig))275
+ {276
+ if (*line == '#' || (strcmp(line, "\r\n") == 0))277
+ continue;278
+ int index = (int)strcspn(line, "\r\n");279
+ lineindex = '\0';280
+ argLine = line;281
+ while (isspace((unsigned char)*argLine)) argLine++;282
+ char* start = strchr(argLine, '-');283
+ int argCount = 0;284
+ char **args = (char**)malloc(256 * sizeof(char *));285
+ //Adding a dummy string to avoid file parsing error286
+ argsargCount++ = (char *)"x265";287
+ char* token = strtok(start, " ");288
+ while (token)289
+ {290
+ argsargCount++ = token;291
+ token = strtok(NULL, " ");292
+ }293
+ argsargCount = NULL;294
+ CLIOptions cliopt;295
+ if (cliopt.parseScenecutAwareQpParam(argCount, args, param))296
+ {297
+ cliopt.destroy();298
+ if (cliopt.api)299
+ cliopt.api->param_free(cliopt.param);300
+ exit(1);301
+ }302
+ break;303
+ }304
+ return 1;305
+ }306
+ bool CLIOptions::parseScenecutAwareQpParam(int argc, char **argv, x265_param* globalParam)307
+ {308
+ bool bError = false;309
+ int bShowHelp = false;310
+ int outputBitDepth = 0;311
+ const char *profile = NULL;312
+ /* Presets are applied before all other options. */313
+ for (optind = 0;;)314
+ {315
+ int c = getopt_long(argc, argv, short_options, long_options, NULL);316
+ if (c == -1)317
+ break;318
+ else if (c == 'D')319
+ outputBitDepth = atoi(optarg);320
+ else if (c == 'P')321
+ profile = optarg;322
+ else if (c == '?')323
+ bShowHelp = true;324
+ }325
+ if (!outputBitDepth && profile)326
+ {327
+ /*try to derive the output bit depth from the requested profile*/328
+ if (strstr(profile, "10"))329
+ outputBitDepth = 10;330
+ else if (strstr(profile, "12"))331
+ outputBitDepth = 12;332
+ else333
+ outputBitDepth = 8;334
+ }335
+ api = x265_api_get(outputBitDepth);336
+ if (!api)337
+ {338
+ x265_log(NULL, X265_LOG_WARNING, "falling back to default bit-depth\n");339
+ api = x265_api_get(0);340
+ }341
+ if (bShowHelp)342
+ {343
+ printVersion(globalParam, api);344
+ showHelp(globalParam);345
+ }346
+ for (optind = 0;;)347
+ {348
+ int long_options_index = -1;349
+ int c = getopt_long(argc, argv, short_options, long_options, &long_options_index);350
+ if (c == -1)351
+ break;352
+ if (long_options_index < 0 && c > 0)353
+ {354
+ for (size_t i = 0; i < sizeof(long_options) / sizeof(long_options0); i++)355
+ {356
+ if (long_optionsi.val == c)357
+ {358
+ long_options_index = (int)i;359
+ break;360
+ }361
+ }362
+ if (long_options_index < 0)363
+ {364
+ /* getopt_long might have already printed an error message */365
+ if (c != 63)366
+ x265_log(NULL, X265_LOG_WARNING, "internal error: short option '%c' has no long option\n", c);367
+ return true;368
+ }369
+ }370
+ if (long_options_index < 0)371
+ {372
+ x265_log(NULL, X265_LOG_WARNING, "short option '%c' unrecognized\n", c);373
+ return true;374
+ }375
+ bError |= !!api->scenecut_aware_qp_param_parse(globalParam, long_optionslong_options_index.name, optarg);376
+ if (bError)377
+ {378
+ const char *name = long_options_index > 0 ? long_optionslong_options_index.name : argvoptind - 2;379
+ x265_log(NULL, X265_LOG_ERROR, "invalid argument: %s = %s\n", name, optarg);380
+ return true;381
+ }382
+ }383
+ if (optind < argc)384
+ {385
+ x265_log(param, X265_LOG_WARNING, "extra unused command arguments given <%s>\n", argvoptind);386
+ return true;387
+ }388
+ return false;389
+ }390
391
#ifdef __cplusplus392
}393
x265_3.5.tar.gz/source/x265cli.h -> x265_3.6.tar.gz/source/x265cli.h
Changed
104
1
2
{ "no-fast-intra", no_argument, NULL, 0 },3
{ "no-open-gop", no_argument, NULL, 0 },4
{ "open-gop", no_argument, NULL, 0 },5
+ { "cra-nal", no_argument, NULL, 0 },6
{ "keyint", required_argument, NULL, 'I' },7
{ "min-keyint", required_argument, NULL, 'i' },8
{ "gop-lookahead", required_argument, NULL, 0 },9
10
{ "scenecut-bias", required_argument, NULL, 0 },11
{ "hist-scenecut", no_argument, NULL, 0},12
{ "no-hist-scenecut", no_argument, NULL, 0},13
- { "hist-threshold", required_argument, NULL, 0},14
{ "fades", no_argument, NULL, 0 },15
{ "no-fades", no_argument, NULL, 0 },16
{ "scenecut-aware-qp", required_argument, NULL, 0 },17
18
{ "qp", required_argument, NULL, 'q' },19
{ "aq-mode", required_argument, NULL, 0 },20
{ "aq-strength", required_argument, NULL, 0 },21
+ { "sbrc", no_argument, NULL, 0 },22
+ { "no-sbrc", no_argument, NULL, 0 },23
{ "rc-grain", no_argument, NULL, 0 },24
{ "no-rc-grain", no_argument, NULL, 0 },25
{ "ipratio", required_argument, NULL, 0 },26
27
{ "crop-rect", required_argument, NULL, 0 }, /* DEPRECATED */28
{ "master-display", required_argument, NULL, 0 },29
{ "max-cll", required_argument, NULL, 0 },30
+ {"video-signal-type-preset", required_argument, NULL, 0 },31
{ "min-luma", required_argument, NULL, 0 },32
{ "max-luma", required_argument, NULL, 0 },33
{ "log2-max-poc-lsb", required_argument, NULL, 8 },34
35
{ "repeat-headers", no_argument, NULL, 0 },36
{ "aud", no_argument, NULL, 0 },37
{ "no-aud", no_argument, NULL, 0 },38
+ { "eob", no_argument, NULL, 0 },39
+ { "no-eob", no_argument, NULL, 0 },40
+ { "eos", no_argument, NULL, 0 },41
+ { "no-eos", no_argument, NULL, 0 },42
{ "info", no_argument, NULL, 0 },43
{ "no-info", no_argument, NULL, 0 },44
{ "zones", required_argument, NULL, 0 },45
{ "qpfile", required_argument, NULL, 0 },46
{ "zonefile", required_argument, NULL, 0 },47
+ { "no-zonefile-rc-init", no_argument, NULL, 0 },48
{ "lambda-file", required_argument, NULL, 0 },49
{ "b-intra", no_argument, NULL, 0 },50
{ "no-b-intra", no_argument, NULL, 0 },51
52
{ "dynamic-refine", no_argument, NULL, 0 },53
{ "no-dynamic-refine", no_argument, NULL, 0 },54
{ "strict-cbr", no_argument, NULL, 0 },55
- { "temporal-layers", no_argument, NULL, 0 },56
- { "no-temporal-layers", no_argument, NULL, 0 },57
+ { "temporal-layers", required_argument, NULL, 0 },58
{ "qg-size", required_argument, NULL, 0 },59
{ "recon-y4m-exec", required_argument, NULL, 0 },60
{ "analyze-src-pics", no_argument, NULL, 0 },61
62
{ "frame-dup", no_argument, NULL, 0 },63
{ "no-frame-dup", no_argument, NULL, 0 },64
{ "dup-threshold", required_argument, NULL, 0 },65
+ { "mcstf", no_argument, NULL, 0 },66
+ { "no-mcstf", no_argument, NULL, 0 },67
#ifdef SVT_HEVC68
{ "svt", no_argument, NULL, 0 },69
{ "no-svt", no_argument, NULL, 0 },70
71
{ "abr-ladder", required_argument, NULL, 0 },72
{ "min-vbv-fullness", required_argument, NULL, 0 },73
{ "max-vbv-fullness", required_argument, NULL, 0 },74
+ { "scenecut-qp-config", required_argument, NULL, 0 },75
+ { "film-grain", required_argument, NULL, 0 },76
{ 0, 0, 0, 0 },77
{ 0, 0, 0, 0 },78
{ 0, 0, 0, 0 },79
80
FILE* qpfile;81
FILE* zoneFile;82
FILE* dolbyVisionRpu; /* File containing Dolby Vision BL RPU metadata */83
+ FILE* scenecutAwareQpConfig; /* File containing scenecut aware frame quantization related CLI options */84
const char* reconPlayCmd;85
const x265_api* api;86
x265_param* param;87
88
qpfile = NULL;89
zoneFile = NULL;90
dolbyVisionRpu = NULL;91
+ scenecutAwareQpConfig = NULL;92
reconPlayCmd = NULL;93
api = NULL;94
param = NULL;95
96
bool parseQPFile(x265_picture &pic_org);97
bool parseZoneFile();98
int rpuParser(x265_picture * pic);99
+ bool parseScenecutAwareQpConfig();100
+ bool parseScenecutAwareQpParam(int argc, char **argv, x265_param* globalParam);101
};102
#ifdef __cplusplus103
}104
x265_3.5.tar.gz/x265Version.txt -> x265_3.6.tar.gz/x265Version.txt
Changed
8
1
2
#Attribute: Values3
-repositorychangeset: f0c1022b64
+repositorychangeset: aa7f602f75
releasetagdistance: 16
-releasetag: 3.57
+releasetag: 3.68
Refresh
No build results available
Refresh
No rpmlint results available
Login required, please
login
or
signup
in order to comment
Request History
Aloysius created request 10 months ago
Update to version 3.6
Aloysius accepted request 10 months ago