We truncated the diff of some files because they were too big.
If you want to see the full diff for every file, click here.
Changes of Revision 42
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
201
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
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
201
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
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
201
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
x265_3.5.tar.gz/source/abrEncApp.cpp -> x265_3.6.tar.gz/source/abrEncApp.cpp
Changed
201
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
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
201
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
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
201
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
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
201
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
x265_3.6.tar.gz/source/common/aarch64/blockcopy8.S
Added
201
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
x265_3.6.tar.gz/source/common/aarch64/dct-prim.cpp
Added
201
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
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
201
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
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
201
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
x265_3.6.tar.gz/source/common/aarch64/intrapred-prim.cpp
Added
201
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
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
201
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
x265_3.6.tar.gz/source/common/aarch64/ipfilter-sve2.S
Added
201
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
x265_3.6.tar.gz/source/common/aarch64/ipfilter.S
Added
201
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
x265_3.6.tar.gz/source/common/aarch64/loopfilter-prim.cpp
Added
201
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
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
201
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
x265_3.5.tar.gz/source/common/aarch64/mc-a.S -> x265_3.6.tar.gz/source/common/aarch64/mc-a.S
Changed
201
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
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
201
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
x265_3.6.tar.gz/source/common/aarch64/p2s.S
Added
201
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
x265_3.6.tar.gz/source/common/aarch64/pixel-prim.cpp
Added
201
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
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
201
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
x265_3.6.tar.gz/source/common/aarch64/pixel-util-sve2.S
Added
201
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
x265_3.5.tar.gz/source/common/aarch64/pixel-util.S -> x265_3.6.tar.gz/source/common/aarch64/pixel-util.S
Changed
201
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
x265_3.6.tar.gz/source/common/aarch64/sad-a-common.S
Added
201
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
x265_3.6.tar.gz/source/common/aarch64/sad-a-sve2.S
Added
201
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
x265_3.5.tar.gz/source/common/aarch64/sad-a.S -> x265_3.6.tar.gz/source/common/aarch64/sad-a.S
Changed
201
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
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
201
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
x265_3.6.tar.gz/source/common/aarch64/ssd-a.S
Added
201
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
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
201
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
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
201
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
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
201
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
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
201
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
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
201
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
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
201
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
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
201
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
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
201
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
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
201
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
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
201
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
x265_3.5.tar.gz/source/x265cli.cpp -> x265_3.6.tar.gz/source/x265cli.cpp
Changed
201
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
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