From 3a056bde4fbe821e3bc0f3d50e9576e76ba8b2e5 Mon Sep 17 00:00:00 2001
From: Adam Wieckowski <adam.wieckowski@hhi.fraunhofer.de>
Date: Fri, 12 Mar 2021 11:05:51 +0100
Subject: [PATCH] removed split parallelism
---
CMakeLists.txt | 18 +-
.../App/BitstreamExtractorApp/CMakeLists.txt | 20 --
source/App/DecoderAnalyserApp/CMakeLists.txt | 20 --
source/App/DecoderApp/CMakeLists.txt | 20 --
source/App/EncoderApp/CMakeLists.txt | 20 --
source/App/EncoderApp/EncApp.cpp | 4 -
source/App/EncoderApp/EncAppCfg.cpp | 24 +-
source/App/EncoderApp/EncAppCfg.h | 7 -
source/App/EncoderApp/encmain.cpp | 10 -
source/App/SEIRemovalApp/CMakeLists.txt | 20 --
source/App/StreamMergeApp/CMakeLists.txt | 20 --
source/App/SubpicMergeApp/CMakeLists.txt | 20 --
source/Lib/CommonAnalyserLib/CMakeLists.txt | 22 --
source/Lib/CommonLib/CMakeLists.txt | 22 --
source/Lib/CommonLib/CodingStructure.cpp | 25 --
source/Lib/CommonLib/CommonDef.h | 10 -
source/Lib/CommonLib/Contexts.h | 6 -
source/Lib/CommonLib/Picture.cpp | 240 ++------------
source/Lib/CommonLib/Picture.h | 50 ---
source/Lib/CommonLib/Quant.cpp | 8 -
source/Lib/CommonLib/Quant.h | 4 -
source/Lib/CommonLib/RdCost.cpp | 20 --
source/Lib/CommonLib/RdCost.h | 4 -
source/Lib/CommonLib/Reshape.h | 4 -
source/Lib/CommonLib/TrQuant.cpp | 7 -
source/Lib/CommonLib/TrQuant.h | 4 -
source/Lib/CommonLib/TypeDef.h | 49 ---
source/Lib/CommonLib/Unit.h | 15 -
source/Lib/DecoderAnalyserLib/CMakeLists.txt | 20 --
source/Lib/DecoderLib/CMakeLists.txt | 20 --
source/Lib/EncoderLib/CABACWriter.cpp | 4 -
source/Lib/EncoderLib/CMakeLists.txt | 20 --
source/Lib/EncoderLib/EncCfg.h | 11 -
source/Lib/EncoderLib/EncCu.cpp | 254 +--------------
source/Lib/EncoderLib/EncCu.h | 13 +-
source/Lib/EncoderLib/EncGOP.cpp | 7 +-
source/Lib/EncoderLib/EncLib.cpp | 149 ---------
source/Lib/EncoderLib/EncLib.h | 57 ----
source/Lib/EncoderLib/EncModeCtrl.cpp | 297 +-----------------
source/Lib/EncoderLib/EncModeCtrl.h | 69 ----
source/Lib/EncoderLib/EncReshape.cpp | 40 ---
source/Lib/EncoderLib/EncReshape.h | 4 -
source/Lib/EncoderLib/EncSlice.cpp | 37 +--
source/Lib/EncoderLib/InterSearch.cpp | 7 -
source/Lib/EncoderLib/InterSearch.h | 3 -
source/Lib/Utilities/CMakeLists.txt | 20 --
46 files changed, 46 insertions(+), 1679 deletions(-)
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 10a86aa00..8728b0903 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -52,22 +52,6 @@ set_property( GLOBAL PROPERTY USE_FOLDERS ON )
# Include a utility module providing functions, macros, and settings
include( ${CMAKE_SOURCE_DIR}/cmake/CMakeBuild/cmake/modules/BBuildEnv.cmake )
-# Enable multithreading
-bb_multithreading()
-
-find_package(OpenMP)
-
-if( OpenMP_FOUND )
- set( CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}" )
- set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}" )
- set( CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_EXE_LINKER_FLAGS}" )
-
- set( SET_ENABLE_SPLIT_PARALLELISM OFF CACHE BOOL "Set ENABLE_SPLIT_PARALLELISM as a compiler flag" )
- set( ENABLE_SPLIT_PARALLELISM OFF CACHE BOOL "If SET_ENABLE_SPLIT_PARALLELISM is on, it will be set to this value" )
- set( SET_ENABLE_WPP_PARALLELISM OFF CACHE BOOL "Set ENABLE_WPP_PARALLELISM as a compiler flag" )
- set( ENABLE_WPP_PARALLELISM OFF CACHE BOOL "If SET_ENABLE_WPP_PARALLELISM is on, it will be set to this value" )
-endif()
-
# Enable warnings for some generators and toolsets.
# bb_enable_warnings( gcc warnings-as-errors -Wno-sign-compare )
# bb_enable_warnings( gcc -Wno-unused-variable )
@@ -75,6 +59,8 @@ endif()
# for gcc 8.2:
bb_enable_warnings( gcc warnings-as-errors -Wno-sign-compare -Wno-class-memaccess)
+bb_multithreading()
+
if( XCODE )
bb_enable_warnings( clang warnings-as-errors
-Wno-deprecated-declarations
diff --git a/source/App/BitstreamExtractorApp/CMakeLists.txt b/source/App/BitstreamExtractorApp/CMakeLists.txt
index a299cf5cc..09f7da6cf 100644
--- a/source/App/BitstreamExtractorApp/CMakeLists.txt
+++ b/source/App/BitstreamExtractorApp/CMakeLists.txt
@@ -33,26 +33,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/App/DecoderAnalyserApp/CMakeLists.txt b/source/App/DecoderAnalyserApp/CMakeLists.txt
index ad272ca1f..968a70c86 100644
--- a/source/App/DecoderAnalyserApp/CMakeLists.txt
+++ b/source/App/DecoderAnalyserApp/CMakeLists.txt
@@ -35,26 +35,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/App/DecoderApp/CMakeLists.txt b/source/App/DecoderApp/CMakeLists.txt
index 4e71c5c1e..a3db4ff0d 100644
--- a/source/App/DecoderApp/CMakeLists.txt
+++ b/source/App/DecoderApp/CMakeLists.txt
@@ -33,26 +33,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/App/EncoderApp/CMakeLists.txt b/source/App/EncoderApp/CMakeLists.txt
index dd87e52d1..b9621bc37 100644
--- a/source/App/EncoderApp/CMakeLists.txt
+++ b/source/App/EncoderApp/CMakeLists.txt
@@ -35,26 +35,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/App/EncoderApp/EncApp.cpp b/source/App/EncoderApp/EncApp.cpp
index e9a2ca876..bd0cf1bb8 100644
--- a/source/App/EncoderApp/EncApp.cpp
+++ b/source/App/EncoderApp/EncApp.cpp
@@ -1041,10 +1041,6 @@ void EncApp::xInitLibCfg()
m_cEncLib.setStopAfterFFtoPOC ( m_stopAfterFFtoPOC );
m_cEncLib.setBs2ModPOCAndType ( m_bs2ModPOCAndType );
m_cEncLib.setDebugCTU ( m_debugCTU );
-#if ENABLE_SPLIT_PARALLELISM
- m_cEncLib.setNumSplitThreads ( m_numSplitThreads );
- m_cEncLib.setForceSingleSplitThread ( m_forceSplitSequential );
-#endif
m_cEncLib.setUseALF ( m_alf );
#if JVET_U0081
m_cEncLib.setALFStrengthLuma (m_alfStrengthLuma);
diff --git a/source/App/EncoderApp/EncAppCfg.cpp b/source/App/EncoderApp/EncAppCfg.cpp
index 24b14a55d..7b87ededf 100644
--- a/source/App/EncoderApp/EncAppCfg.cpp
+++ b/source/App/EncoderApp/EncAppCfg.cpp
@@ -1392,12 +1392,8 @@ bool EncAppCfg::parseCfg( int argc, char* argv[] )
("StopAfterFFtoPOC", m_stopAfterFFtoPOC, false, "If using fast forward to POC, after the POC of interest has been hit, stop further encoding.")
("ForceDecodeBitstream1", m_forceDecodeBitstream1, false, "force decoding of bitstream 1 - use this only if you are realy sure about what you are doing ")
("DecodeBitstream2ModPOCAndType", m_bs2ModPOCAndType, false, "Modify POC and NALU-type of second input bitstream, to use second BS as closing I-slice")
- ("NumSplitThreads", m_numSplitThreads, 1, "Number of threads used to parallelize splitting")
- ("ForceSingleSplitThread", m_forceSplitSequential, false, "Force single thread execution even if taking the parallelized path")
- ("NumWppThreads", m_numWppThreads, 1, "Number of threads used to run WPP-style parallelization")
- ("NumWppExtraLines", m_numWppExtraLines, 0, "Number of additional wpp lines to switch when threads are blocked")
+
("DebugCTU", m_debugCTU, -1, "If DebugBitstream is present, load frames up to this POC from this bitstream. Starting with DebugPOC-frame at CTUline containin debug CTU.")
- ("EnsureWppBitEqual", m_ensureWppBitEqual, false, "Ensure the results are equal to results with WPP-style parallelism, even if WPP is off")
( "ALF", m_alf, true, "Adaptive Loop Filter\n" )
#if JVET_U0081
("ALFStrengthLuma", m_alfStrengthLuma, 1.0, "Adaptive Loop Filter strength for luma. The parameter scales the magnitudes of the ALF filter coefficients for luma. Valid range is 0.0 <= ALFStrengthLuma <= 1.0")
@@ -2588,16 +2584,6 @@ bool EncAppCfg::xCheckParameter()
xConfirmPara( m_wrapAroundOffset % minCUSize != 0, "Wrap-around offset must be an integer multiple of the specified minimum CU size" );
}
-#if ENABLE_SPLIT_PARALLELISM
- xConfirmPara( m_numSplitThreads < 1, "Number of used threads cannot be smaller than 1" );
- xConfirmPara( m_numSplitThreads > PARL_SPLIT_MAX_NUM_THREADS, "Number of used threads cannot be higher than the number of actual jobs" );
-#else
- xConfirmPara( m_numSplitThreads != 1, "ENABLE_SPLIT_PARALLELISM is disabled, numSplitThreads has to be 1" );
-#endif
-
- xConfirmPara( m_numWppThreads != 1, "ENABLE_WPP_PARALLELISM is disabled, numWppThreads has to be 1" );
- xConfirmPara( m_ensureWppBitEqual, "ENABLE_WPP_PARALLELISM is disabled, cannot ensure being WPP bit-equal" );
-
#if SHARP_LUMA_DELTA_QP && ENABLE_QPA
xConfirmPara( m_bUsePerceptQPA && m_lumaLevelToDeltaQPMapping.mode >= 2, "QPA and SharpDeltaQP mode 2 cannot be used together" );
@@ -4123,14 +4109,6 @@ void EncAppCfg::xPrintParameter()
if( m_MIP ) msg(VERBOSE, "FastMIP:%d ", m_useFastMIP);
msg( VERBOSE, "FastLocalDualTree:%d ", m_fastLocalDualTreeMode );
- msg( VERBOSE, "NumSplitThreads:%d ", m_numSplitThreads );
- if( m_numSplitThreads > 1 )
- {
- msg( VERBOSE, "ForceSingleSplitThread:%d ", m_forceSplitSequential );
- }
- msg( VERBOSE, "NumWppThreads:%d+%d ", m_numWppThreads, m_numWppExtraLines );
- msg( VERBOSE, "EnsureWppBitEqual:%d ", m_ensureWppBitEqual );
-
if (m_resChangeInClvsEnabled)
{
msg( VERBOSE, "RPR:(%1.2lfx, %1.2lfx)|%d ", m_scalingRatioHor, m_scalingRatioVer, m_switchPocPeriod );
diff --git a/source/App/EncoderApp/EncAppCfg.h b/source/App/EncoderApp/EncAppCfg.h
index fb76daaae..3b97a4ee1 100644
--- a/source/App/EncoderApp/EncAppCfg.h
+++ b/source/App/EncoderApp/EncAppCfg.h
@@ -398,13 +398,6 @@ protected:
bool m_useFastMIP;
int m_fastLocalDualTreeMode;
-
- int m_numSplitThreads;
- bool m_forceSplitSequential;
- int m_numWppThreads;
- int m_numWppExtraLines;
- bool m_ensureWppBitEqual;
-
int m_log2MaxTbSize;
// coding tools (bit-depth)
int m_inputBitDepth [MAX_NUM_CHANNEL_TYPE]; ///< bit-depth of input file
diff --git a/source/App/EncoderApp/encmain.cpp b/source/App/EncoderApp/encmain.cpp
index 69b324878..74ef03119 100644
--- a/source/App/EncoderApp/encmain.cpp
+++ b/source/App/EncoderApp/encmain.cpp
@@ -101,16 +101,6 @@ int main(int argc, char* argv[])
#endif
#if ENABLE_TRACING
fprintf( stdout, "[ENABLE_TRACING] " );
-#endif
-#if ENABLE_SPLIT_PARALLELISM
- fprintf( stdout, "[SPLIT_PARALLEL (%d jobs)]", PARL_SPLIT_MAX_NUM_JOBS );
-#endif
-#if ENABLE_SPLIT_PARALLELISM
- const char* waitPolicy = getenv( "OMP_WAIT_POLICY" );
- const char* maxThLim = getenv( "OMP_THREAD_LIMIT" );
- fprintf( stdout, waitPolicy ? "[OMP: WAIT_POLICY=%s," : "[OMP: WAIT_POLICY=,", waitPolicy );
- fprintf( stdout, maxThLim ? "THREAD_LIMIT=%s" : "THREAD_LIMIT=", maxThLim );
- fprintf( stdout, "]" );
#endif
fprintf( stdout, "\n" );
diff --git a/source/App/SEIRemovalApp/CMakeLists.txt b/source/App/SEIRemovalApp/CMakeLists.txt
index cb783adff..ec566ecc0 100644
--- a/source/App/SEIRemovalApp/CMakeLists.txt
+++ b/source/App/SEIRemovalApp/CMakeLists.txt
@@ -33,26 +33,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/App/StreamMergeApp/CMakeLists.txt b/source/App/StreamMergeApp/CMakeLists.txt
index 77c53ece6..84ba9a57f 100644
--- a/source/App/StreamMergeApp/CMakeLists.txt
+++ b/source/App/StreamMergeApp/CMakeLists.txt
@@ -33,26 +33,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/App/SubpicMergeApp/CMakeLists.txt b/source/App/SubpicMergeApp/CMakeLists.txt
index dd8fac45f..8418bf9af 100644
--- a/source/App/SubpicMergeApp/CMakeLists.txt
+++ b/source/App/SubpicMergeApp/CMakeLists.txt
@@ -33,26 +33,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
if( CMAKE_COMPILER_IS_GNUCC AND BUILD_STATIC )
set( ADDITIONAL_LIBS ${ADDITIONAL_LIBS} -static -static-libgcc -static-libstdc++ )
target_compile_definitions( ${EXE_NAME} PUBLIC ENABLE_WPP_STATIC_LINK=1 )
diff --git a/source/Lib/CommonAnalyserLib/CMakeLists.txt b/source/Lib/CommonAnalyserLib/CMakeLists.txt
index e915720f2..84ca09922 100644
--- a/source/Lib/CommonAnalyserLib/CMakeLists.txt
+++ b/source/Lib/CommonAnalyserLib/CMakeLists.txt
@@ -64,28 +64,6 @@ if( SET_ENABLE_TRACING )
target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_TRACING=0 )
endif()
endif()
-
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
- target_include_directories( ${LIB_NAME} PUBLIC ${OpenMP_CXX_INCLUDE_DIRS} )
- target_link_libraries( ${LIB_NAME} ${OpenMP_CXX_LIBRARIES} )
-else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
target_include_directories( ${LIB_NAME} PUBLIC ../CommonLib/. ../CommonLib/.. ../CommonLib/x86 ../libmd5 )
target_link_libraries( ${LIB_NAME} Threads::Threads )
diff --git a/source/Lib/CommonLib/CMakeLists.txt b/source/Lib/CommonLib/CMakeLists.txt
index 6ae75c82b..6e110a0b6 100644
--- a/source/Lib/CommonLib/CMakeLists.txt
+++ b/source/Lib/CommonLib/CMakeLists.txt
@@ -62,28 +62,6 @@ if( SET_ENABLE_TRACING )
target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_TRACING=0 )
endif()
endif()
-
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
- target_include_directories( ${LIB_NAME} PUBLIC ${OpenMP_CXX_INCLUDE_DIRS} )
- target_link_libraries( ${LIB_NAME} ${OpenMP_CXX_LIBRARIES} )
-else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
target_include_directories( ${LIB_NAME} PUBLIC . .. ./x86 ../libmd5 )
target_link_libraries( ${LIB_NAME} Threads::Threads )
diff --git a/source/Lib/CommonLib/CodingStructure.cpp b/source/Lib/CommonLib/CodingStructure.cpp
index a62d548ee..dbfdc3ee3 100644
--- a/source/Lib/CommonLib/CodingStructure.cpp
+++ b/source/Lib/CommonLib/CodingStructure.cpp
@@ -549,10 +549,6 @@ CodingUnit& CodingStructure::addCU( const UnitArea &unit, const ChannelType chTy
if( prevCU )
{
prevCU->next = cu;
-#if ENABLE_SPLIT_PARALLELISM
-
- CHECK( prevCU->cacheId != cu->cacheId, "Inconsintent cacheId between previous and current CU" );
-#endif
}
cus.push_back( cu );
@@ -593,21 +589,12 @@ PredictionUnit& CodingStructure::addPU( const UnitArea &unit, const ChannelType
pu->cs = this;
pu->cu = m_isTuEnc ? cus[0] : getCU( unit.blocks[chType].pos(), chType );
pu->chType = chType;
-#if ENABLE_SPLIT_PARALLELISM
-
- CHECK( pu->cacheId != pu->cu->cacheId, "Inconsintent cacheId between the PU and assigned CU" );
- CHECK( pu->cu->firstPU != nullptr, "Without an RQT the firstPU should be null" );
-#endif
PredictionUnit *prevPU = m_numPUs > 0 ? pus.back() : nullptr;
if( prevPU && prevPU->cu == pu->cu )
{
prevPU->next = pu;
-#if ENABLE_SPLIT_PARALLELISM
-
- CHECK( prevPU->cacheId != pu->cacheId, "Inconsintent cacheId between previous and current PU" );
-#endif
}
pus.push_back( pu );
@@ -654,14 +641,6 @@ TransformUnit& CodingStructure::addTU( const UnitArea &unit, const ChannelType c
tu->cs = this;
tu->cu = m_isTuEnc ? cus[0] : getCU( unit.blocks[chType].pos(), chType );
tu->chType = chType;
-#if ENABLE_SPLIT_PARALLELISM
-
- if( tu->cu )
- {
- CHECK(tu->cacheId != tu->cu->cacheId, "Inconsintent cacheId between the TU and assigned CU");
- }
-#endif
-
TransformUnit *prevTU = m_numTUs > 0 ? tus.back() : nullptr;
@@ -669,10 +648,6 @@ TransformUnit& CodingStructure::addTU( const UnitArea &unit, const ChannelType c
{
prevTU->next = tu;
tu->prev = prevTU;
-#if ENABLE_SPLIT_PARALLELISM
-
- CHECK( prevTU->cacheId != tu->cacheId, "Inconsintent cacheId between previous and current TU" );
-#endif
}
tus.push_back( tu );
diff --git a/source/Lib/CommonLib/CommonDef.h b/source/Lib/CommonLib/CommonDef.h
index 8f0c64f42..5304de1a0 100644
--- a/source/Lib/CommonLib/CommonDef.h
+++ b/source/Lib/CommonLib/CommonDef.h
@@ -680,16 +680,6 @@ static inline int ceilLog2(uint32_t x)
#define _UNIT_AREA_AT(_a,_x,_y,_w,_h)
#endif
-#if ENABLE_SPLIT_PARALLELISM
-#include <omp.h>
-
-#define PARL_PARAM(DEF) , DEF
-#define PARL_PARAM0(DEF) DEF
-#else
-#define PARL_PARAM(DEF)
-#define PARL_PARAM0(DEF)
-#endif
-
static const uint32_t CCALF_CANDS_COEFF_NR = 8;
static const int CCALF_SMALL_TAB[CCALF_CANDS_COEFF_NR] = { 0, 1, 2, 4, 8, 16, 32, 64 };
diff --git a/source/Lib/CommonLib/Contexts.h b/source/Lib/CommonLib/Contexts.h
index 06ee292cb..a5e89abf8 100644
--- a/source/Lib/CommonLib/Contexts.h
+++ b/source/Lib/CommonLib/Contexts.h
@@ -438,12 +438,6 @@ private:
CtxStore<BinProbModel_Std> m_CtxStore_Std;
protected:
unsigned m_GRAdaptStats[RExt__GOLOMB_RICE_ADAPTATION_STATISTICS_SETS];
-#if ENABLE_SPLIT_PARALLELISM
-
-public:
- int64_t cacheId;
- bool cacheUsed;
-#endif
};
diff --git a/source/Lib/CommonLib/Picture.cpp b/source/Lib/CommonLib/Picture.cpp
index df80b30d9..3fc433bef 100644
--- a/source/Lib/CommonLib/Picture.cpp
+++ b/source/Lib/CommonLib/Picture.cpp
@@ -40,130 +40,6 @@
#include "ChromaFormat.h"
#include "CommonLib/InterpolationFilter.h"
-
-#if ENABLE_SPLIT_PARALLELISM
-
-int g_wppThreadId( 0 );
-#pragma omp threadprivate(g_wppThreadId)
-
-#if ENABLE_SPLIT_PARALLELISM
-int g_splitThreadId( 0 );
-#pragma omp threadprivate(g_splitThreadId)
-
-int g_splitJobId( 0 );
-#pragma omp threadprivate(g_splitJobId)
-#endif
-
-Scheduler::Scheduler() :
-#if ENABLE_SPLIT_PARALLELISM
- m_numSplitThreads( 1 )
-#endif
-{
-}
-
-Scheduler::~Scheduler()
-{
-}
-
-#if ENABLE_SPLIT_PARALLELISM
-unsigned Scheduler::getSplitDataId( int jobId ) const
-{
- if( m_numSplitThreads > 1 && m_hasParallelBuffer )
- {
- int splitJobId = jobId == CURR_THREAD_ID ? g_splitJobId : jobId;
-
- return ( g_wppThreadId * NUM_RESERVERD_SPLIT_JOBS ) + splitJobId;
- }
- else
- {
- return 0;
- }
-}
-
-unsigned Scheduler::getSplitPicId( int tId /*= CURR_THREAD_ID */ ) const
-{
- if( m_numSplitThreads > 1 && m_hasParallelBuffer )
- {
- int threadId = tId == CURR_THREAD_ID ? g_splitThreadId : tId;
-
- return ( g_wppThreadId * m_numSplitThreads ) + threadId;
- }
- else
- {
- return 0;
- }
-}
-
-unsigned Scheduler::getSplitJobId() const
-{
- if( m_numSplitThreads > 1 )
- {
- return g_splitJobId;
- }
- else
- {
- return 0;
- }
-}
-
-void Scheduler::setSplitJobId( const int jobId )
-{
- CHECK( g_splitJobId != 0 && jobId != 0, "Need to reset the jobId after usage!" );
- g_splitJobId = jobId;
-}
-
-void Scheduler::startParallel()
-{
- m_hasParallelBuffer = true;
-}
-
-void Scheduler::finishParallel()
-{
- m_hasParallelBuffer = false;
-}
-
-void Scheduler::setSplitThreadId( const int tId )
-{
- g_splitThreadId = tId == CURR_THREAD_ID ? omp_get_thread_num() : tId;
-}
-
-#endif
-
-
-
-unsigned Scheduler::getDataId() const
-{
-#if ENABLE_SPLIT_PARALLELISM
- if( m_numSplitThreads > 1 )
- {
- return getSplitDataId();
- }
-#endif
- return 0;
-}
-
-bool Scheduler::init( const int ctuYsize, const int ctuXsize, const int numWppThreadsRunning, const int numWppExtraLines, const int numSplitThreads )
-{
-#if ENABLE_SPLIT_PARALLELISM
- m_numSplitThreads = numSplitThreads;
-#endif
-
- return true;
-}
-
-
-int Scheduler::getNumPicInstances() const
-{
-#if !ENABLE_SPLIT_PARALLELISM
- return 1;
-#else
- return ( m_numSplitThreads > 1 ? m_numSplitThreads : 1 );
-#endif
-}
-
-#endif
-
-
// ---------------------------------------------------------------------------
// picture methods
// ---------------------------------------------------------------------------
@@ -225,39 +101,34 @@ void Picture::create( const ChromaFormat &_chromaFormat, const Size &size, const
void Picture::destroy()
{
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 0; jId < PARL_SPLIT_MAX_NUM_THREADS; jId++ )
-#endif
+ for (uint32_t t = 0; t < NUM_PIC_TYPES; t++)
{
- for (uint32_t t = 0; t < NUM_PIC_TYPES; t++)
- {
- M_BUFS(jId, t).destroy();
- }
- m_hashMap.clearAll();
- if (cs)
- {
- cs->destroy();
- delete cs;
- cs = nullptr;
- }
+ M_BUFS(jId, t).destroy();
+ }
+ m_hashMap.clearAll();
+ if (cs)
+ {
+ cs->destroy();
+ delete cs;
+ cs = nullptr;
+ }
- for (auto &ps: slices)
- {
- delete ps;
- }
- slices.clear();
+ for (auto &ps: slices)
+ {
+ delete ps;
+ }
+ slices.clear();
- for (auto &psei: SEIs)
- {
- delete psei;
- }
- SEIs.clear();
+ for (auto &psei: SEIs)
+ {
+ delete psei;
+ }
+ SEIs.clear();
- if (m_spliceIdx)
- {
- delete[] m_spliceIdx;
- m_spliceIdx = NULL;
- }
+ if (m_spliceIdx)
+ {
+ delete[] m_spliceIdx;
+ m_spliceIdx = NULL;
}
}
@@ -269,21 +140,8 @@ void Picture::createTempBuffers( const unsigned _maxCUSize )
const Area a = m_ctuArea.Y();
#endif
-#if ENABLE_SPLIT_PARALLELISM
- scheduler.startParallel();
-
- for( int jId = 0; jId < scheduler.getNumPicInstances(); jId++ )
-#endif
- {
- M_BUFS( jId, PIC_PREDICTION ).create( chromaFormat, a, _maxCUSize );
- M_BUFS( jId, PIC_RESIDUAL ).create( chromaFormat, a, _maxCUSize );
-#if ENABLE_SPLIT_PARALLELISM
- if (jId > 0)
- {
- M_BUFS(jId, PIC_RECONSTRUCTION).create(chromaFormat, Y(), _maxCUSize, margin, MEMORY_ALIGN_DEF_SIZE);
- }
-#endif
- }
+ M_BUFS( jId, PIC_PREDICTION ).create( chromaFormat, a, _maxCUSize );
+ M_BUFS( jId, PIC_RESIDUAL ).create( chromaFormat, a, _maxCUSize );
if (cs)
{
@@ -293,24 +151,11 @@ void Picture::createTempBuffers( const unsigned _maxCUSize )
void Picture::destroyTempBuffers()
{
-#if ENABLE_SPLIT_PARALLELISM
- scheduler.finishParallel();
-
- for( int jId = 0; jId < scheduler.getNumPicInstances(); jId++ )
-#endif
+ for (uint32_t t = 0; t < NUM_PIC_TYPES; t++)
{
- for (uint32_t t = 0; t < NUM_PIC_TYPES; t++)
+ if (t == PIC_RESIDUAL || t == PIC_PREDICTION)
{
- if (t == PIC_RESIDUAL || t == PIC_PREDICTION)
- {
- M_BUFS(jId, t).destroy();
- }
-#if ENABLE_SPLIT_PARALLELISM
- if (t == PIC_RECONSTRUCTION && jId > 0)
- {
- M_BUFS(jId, t).destroy();
- }
-#endif
+ M_BUFS(0, t).destroy();
}
}
@@ -470,23 +315,6 @@ void Picture::clearSliceBuffer()
slices.clear();
}
-#if ENABLE_SPLIT_PARALLELISM
-void Picture::finishParallelPart( const UnitArea& area )
-{
- const UnitArea clipdArea = clipArea( area, *this );
- const int sourceID = scheduler.getSplitPicId( 0 );
- CHECK( scheduler.getSplitJobId() > 0, "Finish-CU cannot be called from within a mode- or split-parallelized block!" );
-
- // distribute the reconstruction across all of the parallel workers
- for( int tId = 1; tId < scheduler.getNumSplitThreads(); tId++ )
- {
- const int destID = scheduler.getSplitPicId( tId );
-
- M_BUFS( destID, PIC_RECONSTRUCTION ).subBuf( clipdArea ).copyFrom( M_BUFS( sourceID, PIC_RECONSTRUCTION ).subBuf( clipdArea ) );
- }
-}
-#endif
-
const TFilterCoeff DownsamplingFilterSRC[8][16][12] =
{
{ // D = 1
@@ -1240,9 +1068,6 @@ PelBuf Picture::getBuf( const CompArea &blk, const PictureType &type )
return PelBuf();
}
-#if ENABLE_SPLIT_PARALLELISM
- const int jId = ( type == PIC_ORIGINAL || type == PIC_TRUE_ORIGINAL || type == PIC_ORIGINAL_INPUT || type == PIC_TRUE_ORIGINAL_INPUT ) ? 0 : scheduler.getSplitPicId();
-#endif
#if !KEEP_PRED_AND_RESI_SIGNALS
if( type == PIC_RESIDUAL || type == PIC_PREDICTION )
{
@@ -1264,10 +1089,6 @@ const CPelBuf Picture::getBuf( const CompArea &blk, const PictureType &type ) co
return PelBuf();
}
-#if ENABLE_SPLIT_PARALLELISM
- const int jId = ( type == PIC_ORIGINAL || type == PIC_TRUE_ORIGINAL ) ? 0 : scheduler.getSplitPicId();
-
-#endif
#if !KEEP_PRED_AND_RESI_SIGNALS
if( type == PIC_RESIDUAL || type == PIC_PREDICTION )
{
@@ -1308,9 +1129,6 @@ const CPelUnitBuf Picture::getBuf( const UnitArea &unit, const PictureType &type
Pel* Picture::getOrigin( const PictureType &type, const ComponentID compID ) const
{
-#if ENABLE_SPLIT_PARALLELISM
- const int jId = ( type == PIC_ORIGINAL || type == PIC_TRUE_ORIGINAL ) ? 0 : scheduler.getSplitPicId();
-#endif
return M_BUFS( jId, type ).getOrigin( compID );
}
diff --git a/source/Lib/CommonLib/Picture.h b/source/Lib/CommonLib/Picture.h
index 5506da1b4..2f985828e 100644
--- a/source/Lib/CommonLib/Picture.h
+++ b/source/Lib/CommonLib/Picture.h
@@ -50,49 +50,13 @@
#include "MCTS.h"
#include <deque>
-#if ENABLE_SPLIT_PARALLELISM
-
-#define CURR_THREAD_ID -1
-
-class Scheduler
-{
-public:
- Scheduler();
- ~Scheduler();
-
-#if ENABLE_SPLIT_PARALLELISM
- unsigned getSplitDataId( int jobId = CURR_THREAD_ID ) const;
- unsigned getSplitPicId ( int tId = CURR_THREAD_ID ) const;
- unsigned getSplitJobId () const;
- void setSplitJobId ( const int jobId );
- void startParallel ();
- void finishParallel();
- void setSplitThreadId( const int tId = CURR_THREAD_ID );
- unsigned getNumSplitThreads() const { return m_numSplitThreads; };
-#endif
- unsigned getDataId () const;
- bool init ( const int ctuYsize, const int ctuXsize, const int numWppThreadsRunning, const int numWppExtraLines, const int numSplitThreads );
- int getNumPicInstances() const;
-#if ENABLE_SPLIT_PARALLELISM
-
- int m_numSplitThreads;
- bool m_hasParallelBuffer;
-#endif
-};
-#endif
class SEI;
class AQpLayer;
typedef std::list<SEI*> SEIMessages;
-
-
-#if ENABLE_SPLIT_PARALLELISM
-#define M_BUFS(JID,PID) m_bufs[JID][PID]
-#else
#define M_BUFS(JID,PID) m_bufs[PID]
-#endif
struct Picture : public UnitArea
{
@@ -237,12 +201,7 @@ public:
bool interLayerRefPicFlag;
bool mixedNaluTypesInPicFlag;
-
-#if ENABLE_SPLIT_PARALLELISM
- PelStorage m_bufs[PARL_SPLIT_MAX_NUM_JOBS][NUM_PIC_TYPES];
-#else
PelStorage m_bufs[NUM_PIC_TYPES];
-#endif
const Picture* unscaledPic;
TComHash m_hashMap;
@@ -280,15 +239,6 @@ private:
UnitArea m_ctuArea;
#endif
-#if ENABLE_SPLIT_PARALLELISM
-public:
- void finishParallelPart ( const UnitArea& ctuArea );
-#endif
-#if ENABLE_SPLIT_PARALLELISM
-public:
- Scheduler scheduler;
-#endif
-
public:
SAOBlkParam *getSAO(int id = 0) { return &m_sao[id][0]; };
void resizeSAO(unsigned numEntries, int dstid) { m_sao[dstid].resize(numEntries); }
diff --git a/source/Lib/CommonLib/Quant.cpp b/source/Lib/CommonLib/Quant.cpp
index 1f7ae1f97..ce9f46528 100644
--- a/source/Lib/CommonLib/Quant.cpp
+++ b/source/Lib/CommonLib/Quant.cpp
@@ -501,14 +501,6 @@ void Quant::init( uint32_t uiMaxTrSize,
m_resetStore = true;
}
-#if ENABLE_SPLIT_PARALLELISM
-void Quant::copyState( const Quant& other )
-{
- m_dLambda = other.m_dLambda;
- memcpy( m_lambdas, other.m_lambdas, sizeof( m_lambdas ) );
-}
-#endif
-
/** set quantized matrix coefficient for encode
* \param scalingList quantized matrix address
* \param format chroma format
diff --git a/source/Lib/CommonLib/Quant.h b/source/Lib/CommonLib/Quant.h
index 7d7f51d94..b9b0d56c2 100644
--- a/source/Lib/CommonLib/Quant.h
+++ b/source/Lib/CommonLib/Quant.h
@@ -140,10 +140,6 @@ public:
// de-quantization
virtual void dequant ( const TransformUnit &tu, CoeffBuf &dstCoeff, const ComponentID &compID, const QpParam &cQP );
-#if ENABLE_SPLIT_PARALLELISM
- virtual void copyState ( const Quant& other );
-#endif
-
protected:
#if T0196_SELECTIVE_RDOQ
diff --git a/source/Lib/CommonLib/RdCost.cpp b/source/Lib/CommonLib/RdCost.cpp
index 4392eae24..996c01935 100644
--- a/source/Lib/CommonLib/RdCost.cpp
+++ b/source/Lib/CommonLib/RdCost.cpp
@@ -222,26 +222,6 @@ void RdCost::init()
m_pairCheck = 0;
}
-
-#if ENABLE_SPLIT_PARALLELISM
-
-void RdCost::copyState( const RdCost& other )
-{
- m_costMode = other.m_costMode;
- m_dLambda = other.m_dLambda;
- m_DistScale = other.m_DistScale;
- memcpy( m_distortionWeight, other.m_distortionWeight, sizeof( m_distortionWeight ) );
- m_mvPredictor = other.m_mvPredictor;
- m_motionLambda = other.m_motionLambda;
- m_iCostScale = other.m_iCostScale;
- m_dLambdaMotionSAD = other.m_dLambdaMotionSAD;
-#if WCG_EXT
- m_dLambda_unadjusted = other.m_dLambda_unadjusted ;
- m_DistScaleUnadjusted = other.m_DistScaleUnadjusted;
-#endif
-}
-#endif
-
void RdCost::setDistParam( DistParam &rcDP, const CPelBuf &org, const Pel* piRefY, int iRefStride, int bitDepth, ComponentID compID, int subShiftMode, int step, bool useHadamard )
{
rcDP.bitDepth = bitDepth;
diff --git a/source/Lib/CommonLib/RdCost.h b/source/Lib/CommonLib/RdCost.h
index 3f82dc0ee..e3ad4f06e 100644
--- a/source/Lib/CommonLib/RdCost.h
+++ b/source/Lib/CommonLib/RdCost.h
@@ -293,10 +293,6 @@ public:
return length;
}
-#if ENABLE_SPLIT_PARALLELISM
- void copyState( const RdCost& other );
-#endif
-
// for motion cost
static uint32_t xGetExpGolombNumberOfBits( int iVal )
{
diff --git a/source/Lib/CommonLib/Reshape.h b/source/Lib/CommonLib/Reshape.h
index e3b9c9f8f..2d28d9719 100644
--- a/source/Lib/CommonLib/Reshape.h
+++ b/source/Lib/CommonLib/Reshape.h
@@ -74,11 +74,7 @@ protected:
int m_vpduY;
public:
Reshape();
-#if ENABLE_SPLIT_PARALLELISM
- virtual ~Reshape();
-#else
~Reshape();
-#endif
void createDec(int bitDepth);
void destroy();
diff --git a/source/Lib/CommonLib/TrQuant.cpp b/source/Lib/CommonLib/TrQuant.cpp
index 006862181..59d4c22de 100644
--- a/source/Lib/CommonLib/TrQuant.cpp
+++ b/source/Lib/CommonLib/TrQuant.cpp
@@ -208,13 +208,6 @@ TrQuant::~TrQuant()
}
}
-#if ENABLE_SPLIT_PARALLELISM
-void TrQuant::copyState( const TrQuant& other )
-{
- m_quant->copyState( *other.m_quant );
-}
-#endif
-
void TrQuant::xDeQuant(const TransformUnit &tu,
CoeffBuf &dstCoeff,
const ComponentID &compID,
diff --git a/source/Lib/CommonLib/TrQuant.h b/source/Lib/CommonLib/TrQuant.h
index 6fcafb91e..f93ab67de 100644
--- a/source/Lib/CommonLib/TrQuant.h
+++ b/source/Lib/CommonLib/TrQuant.h
@@ -115,10 +115,6 @@ public:
void lambdaAdjustColorTrans(bool forward) { m_quant->lambdaAdjustColorTrans(forward); }
void resetStore() { m_quant->resetStore(); }
-#if ENABLE_SPLIT_PARALLELISM
- void copyState( const TrQuant& other );
-#endif
-
protected:
TCoeff m_tempCoeff[MAX_TB_SIZEY * MAX_TB_SIZEY];
diff --git a/source/Lib/CommonLib/TypeDef.h b/source/Lib/CommonLib/TypeDef.h
index 019a18f6e..67fceef08 100644
--- a/source/Lib/CommonLib/TypeDef.h
+++ b/source/Lib/CommonLib/TypeDef.h
@@ -101,17 +101,6 @@ typedef std::pair<int, int> TrCost;
#define JVET_O0756_CALCULATE_HDRMETRICS 1
#endif
-#ifndef ENABLE_SPLIT_PARALLELISM
-#define ENABLE_SPLIT_PARALLELISM 0
-#endif
-#if ENABLE_SPLIT_PARALLELISM
-#define PARL_SPLIT_MAX_NUM_JOBS 6 // number of parallel jobs that can be defined and need memory allocated
-#define NUM_RESERVERD_SPLIT_JOBS ( PARL_SPLIT_MAX_NUM_JOBS + 1 ) // number of all data structures including the merge thread (0)
-#define PARL_SPLIT_MAX_NUM_THREADS PARL_SPLIT_MAX_NUM_JOBS
-#define NUM_SPLIT_THREADS_IF_MSVC 4
-
-#endif
-
// clang-format on
// ====================================================================================================================
@@ -1218,20 +1207,8 @@ template<typename T>
class dynamic_cache
{
std::vector<T*> m_cache;
-#if ENABLE_SPLIT_PARALLELISM
- int64_t m_cacheId;
-#endif
public:
-
-#if ENABLE_SPLIT_PARALLELISM
- dynamic_cache()
- {
- static int cacheId = 0;
- m_cacheId = cacheId++;
- }
-
-#endif
~dynamic_cache()
{
deleteEntries();
@@ -1256,48 +1233,22 @@ public:
{
ret = m_cache.back();
m_cache.pop_back();
-#if ENABLE_SPLIT_PARALLELISM
- CHECK( ret->cacheId != m_cacheId, "Putting item into wrong cache!" );
- CHECK( !ret->cacheUsed, "Fetched an element that should've been in cache!!" );
-#endif
}
else
{
ret = new T;
}
-#if ENABLE_SPLIT_PARALLELISM
- ret->cacheId = m_cacheId;
- ret->cacheUsed = false;
-
-#endif
return ret;
}
void cache( T* el )
{
-#if ENABLE_SPLIT_PARALLELISM
- CHECK( el->cacheId != m_cacheId, "Putting item into wrong cache!" );
- CHECK( el->cacheUsed, "Putting cached item back into cache!" );
-
- el->cacheUsed = true;
-
-#endif
m_cache.push_back( el );
}
void cache( std::vector<T*>& vel )
{
-#if ENABLE_SPLIT_PARALLELISM
- for( auto el : vel )
- {
- CHECK( el->cacheId != m_cacheId, "Putting item into wrong cache!" );
- CHECK( el->cacheUsed, "Putting cached item back into cache!" );
-
- el->cacheUsed = true;
- }
-
-#endif
m_cache.insert( m_cache.end(), vel.begin(), vel.end() );
vel.clear();
}
diff --git a/source/Lib/CommonLib/Unit.h b/source/Lib/CommonLib/Unit.h
index 62395e292..430f6e8dc 100644
--- a/source/Lib/CommonLib/Unit.h
+++ b/source/Lib/CommonLib/Unit.h
@@ -350,11 +350,7 @@ struct CodingUnit : public UnitArea
TransformUnit *firstTU;
TransformUnit *lastTU;
-#if ENABLE_SPLIT_PARALLELISM
- int64_t cacheId;
- bool cacheUsed;
-#endif
const uint8_t getSbtIdx() const { assert( ( ( sbtInfo >> 0 ) & 0xf ) < NUMBER_SBT_IDX ); return ( sbtInfo >> 0 ) & 0xf; }
const uint8_t getSbtPos() const { return ( sbtInfo >> 4 ) & 0x3; }
void setSbtIdx( uint8_t idx ) { CHECK( idx >= NUMBER_SBT_IDX, "sbt_idx wrong" ); sbtInfo = ( idx << 0 ) + ( sbtInfo & 0xf0 ); }
@@ -434,12 +430,6 @@ struct PredictionUnit : public UnitArea, public IntraPredictionData, public Inte
const MotionInfo& getMotionInfo( const Position& pos ) const;
MotionBuf getMotionBuf();
CMotionBuf getMotionBuf() const;
-
-#if ENABLE_SPLIT_PARALLELISM
-
- int64_t cacheId;
- bool cacheUsed;
-#endif
};
// ---------------------------------------------------------------------------
@@ -490,11 +480,6 @@ struct TransformUnit : public UnitArea
Pel* getPLTIndex(const ComponentID id);
bool* getRunTypes(const ComponentID id);
-#if ENABLE_SPLIT_PARALLELISM
- int64_t cacheId;
- bool cacheUsed;
-
-#endif
private:
TCoeff *m_coeffs[ MAX_NUM_TBLOCKS ];
Pel *m_pcmbuf[ MAX_NUM_TBLOCKS ];
diff --git a/source/Lib/DecoderAnalyserLib/CMakeLists.txt b/source/Lib/DecoderAnalyserLib/CMakeLists.txt
index 4fbd3463e..dc940061d 100644
--- a/source/Lib/DecoderAnalyserLib/CMakeLists.txt
+++ b/source/Lib/DecoderAnalyserLib/CMakeLists.txt
@@ -29,26 +29,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
target_include_directories( ${LIB_NAME} PUBLIC ../DecoderLib )
target_link_libraries( ${LIB_NAME} CommonAnalyserLib Threads::Threads )
diff --git a/source/Lib/DecoderLib/CMakeLists.txt b/source/Lib/DecoderLib/CMakeLists.txt
index 23a3659a1..9c84ac8f0 100644
--- a/source/Lib/DecoderLib/CMakeLists.txt
+++ b/source/Lib/DecoderLib/CMakeLists.txt
@@ -28,26 +28,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
target_include_directories( ${LIB_NAME} PUBLIC . )
target_link_libraries( ${LIB_NAME} CommonLib Threads::Threads )
diff --git a/source/Lib/EncoderLib/CABACWriter.cpp b/source/Lib/EncoderLib/CABACWriter.cpp
index ea573ccf1..1b827be85 100644
--- a/source/Lib/EncoderLib/CABACWriter.cpp
+++ b/source/Lib/EncoderLib/CABACWriter.cpp
@@ -1757,11 +1757,7 @@ Pel CABACWriter::writePLTIndex(const CodingUnit& cu, uint32_t idx, PelBuf& palet
void CABACWriter::prediction_unit( const PredictionUnit& pu )
{
CHECK( pu.cu->treeType == TREE_C, "cannot be chroma CU" );
-#if ENABLE_SPLIT_PARALLELISM
- CHECK( pu.cacheUsed, "Processing a PU that should be in cache!" );
- CHECK( pu.cu->cacheUsed, "Processing a CU that should be in cache!" );
-#endif
if( pu.cu->skip )
{
CHECK( !pu.mergeFlag, "merge_flag must be true for skipped CUs" );
diff --git a/source/Lib/EncoderLib/CMakeLists.txt b/source/Lib/EncoderLib/CMakeLists.txt
index 2a50346f2..c82c024aa 100644
--- a/source/Lib/EncoderLib/CMakeLists.txt
+++ b/source/Lib/EncoderLib/CMakeLists.txt
@@ -32,26 +32,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
target_include_directories( ${LIB_NAME} PUBLIC . )
target_link_libraries( ${LIB_NAME} CommonLib Threads::Threads )
diff --git a/source/Lib/EncoderLib/EncCfg.h b/source/Lib/EncoderLib/EncCfg.h
index 1a81a3420..8981657ec 100644
--- a/source/Lib/EncoderLib/EncCfg.h
+++ b/source/Lib/EncoderLib/EncCfg.h
@@ -736,11 +736,6 @@ protected:
CfgVPSParameters m_cfgVPSParameters;
-#if ENABLE_SPLIT_PARALLELISM
- int m_numSplitThreads;
- bool m_forceSingleSplitThread;
-#endif
-
bool m_alf; ///< Adaptive Loop Filter
#if JVET_U0081
double m_alfStrengthLuma;
@@ -1938,12 +1933,6 @@ public:
void setDebugCTU( int i ) { m_debugCTU = i; }
int getDebugCTU() const { return m_debugCTU; }
-#if ENABLE_SPLIT_PARALLELISM
- void setNumSplitThreads( int n ) { m_numSplitThreads = n; }
- int getNumSplitThreads() const { return m_numSplitThreads; }
- void setForceSingleSplitThread( bool b ) { m_forceSingleSplitThread = b; }
- int getForceSingleSplitThread() const { return m_forceSingleSplitThread; }
-#endif
void setUseALF( bool b ) { m_alf = b; }
bool getUseALF() const { return m_alf; }
#if JVET_U0081
diff --git a/source/Lib/EncoderLib/EncCu.cpp b/source/Lib/EncoderLib/EncCu.cpp
index 1ae5af099..cac2ed29f 100644
--- a/source/Lib/EncoderLib/EncCu.cpp
+++ b/source/Lib/EncoderLib/EncCu.cpp
@@ -227,22 +227,18 @@ EncCu::~EncCu()
/** \param pcEncLib pointer of encoder class
*/
-void EncCu::init( EncLib* pcEncLib, const SPS& sps PARL_PARAM( const int tId ) )
+void EncCu::init( EncLib* pcEncLib, const SPS& sps )
{
m_pcEncCfg = pcEncLib;
- m_pcIntraSearch = pcEncLib->getIntraSearch( PARL_PARAM0( tId ) );
- m_pcInterSearch = pcEncLib->getInterSearch( PARL_PARAM0( tId ) );
- m_pcTrQuant = pcEncLib->getTrQuant( PARL_PARAM0( tId ) );
- m_pcRdCost = pcEncLib->getRdCost ( PARL_PARAM0( tId ) );
- m_CABACEstimator = pcEncLib->getCABACEncoder( PARL_PARAM0( tId ) )->getCABACEstimator( &sps );
+ m_pcIntraSearch = pcEncLib->getIntraSearch();
+ m_pcInterSearch = pcEncLib->getInterSearch();
+ m_pcTrQuant = pcEncLib->getTrQuant();
+ m_pcRdCost = pcEncLib->getRdCost ();
+ m_CABACEstimator = pcEncLib->getCABACEncoder()->getCABACEstimator( &sps );
m_CABACEstimator->setEncCu(this);
- m_CtxCache = pcEncLib->getCtxCache( PARL_PARAM0( tId ) );
+ m_CtxCache = pcEncLib->getCtxCache();
m_pcRateCtrl = pcEncLib->getRateCtrl();
m_pcSliceEncoder = pcEncLib->getSliceEncoder();
-#if ENABLE_SPLIT_PARALLELISM
- m_pcEncLib = pcEncLib;
- m_dataId = tId;
-#endif
m_pcLoopFilter = pcEncLib->getLoopFilter();
m_GeoCostList.init(GEO_NUM_PARTITION_MODE, m_pcEncCfg->getMaxNumGeoCand());
m_AFFBestSATDCost = MAX_DOUBLE;
@@ -267,39 +263,6 @@ void EncCu::compressCtu( CodingStructure& cs, const UnitArea& area, const unsign
cs.slice->m_mapPltCost[0].clear();
cs.slice->m_mapPltCost[1].clear();
-#if ENABLE_SPLIT_PARALLELISM
- if( m_pcEncCfg->getNumSplitThreads() > 1 )
- {
- for( int jId = 1; jId < NUM_RESERVERD_SPLIT_JOBS; jId++ )
- {
- EncCu* jobEncCu = m_pcEncLib->getCuEncoder( cs.picture->scheduler.getSplitDataId( jId ) );
- CacheBlkInfoCtrl* cacheCtrl = dynamic_cast< CacheBlkInfoCtrl* >( jobEncCu->m_modeCtrl );
-#if REUSE_CU_RESULTS
- BestEncInfoCache* bestCache = dynamic_cast< BestEncInfoCache* >( jobEncCu->m_modeCtrl );
-#endif
- SaveLoadEncInfoSbt *sbtCache = dynamic_cast< SaveLoadEncInfoSbt* >( jobEncCu->m_modeCtrl );
- if( cacheCtrl )
- {
- cacheCtrl->init( *cs.slice );
- }
-#if REUSE_CU_RESULTS
- if (bestCache)
- {
- bestCache->init(*cs.slice);
- }
-#endif
- if (sbtCache)
- {
- sbtCache->init(*cs.slice);
- }
- }
- }
-
-#if REUSE_CU_RESULTS
- if( auto* cacheCtrl = dynamic_cast<BestEncInfoCache*>( m_modeCtrl ) ) { cacheCtrl->tick(); }
-#endif
- if( auto* cacheCtrl = dynamic_cast<CacheBlkInfoCtrl*>( m_modeCtrl ) ) { cacheCtrl->tick(); }
-#endif
// init the partitioning manager
QTBTPartitioner partitioner;
partitioner.initCtu(area, CH_L, *cs.slice);
@@ -535,20 +498,7 @@ bool EncCu::xCheckBestMode( CodingStructure *&tempCS, CodingStructure *&bestCS,
void EncCu::xCompressCU( CodingStructure*& tempCS, CodingStructure*& bestCS, Partitioner& partitioner, double maxCostAllowed )
{
CHECK(maxCostAllowed < 0, "Wrong value of maxCostAllowed!");
-#if ENABLE_SPLIT_PARALLELISM
- CHECK( m_dataId != tempCS->picture->scheduler.getDataId(), "Working in the wrong dataId!" );
-
- if( m_pcEncCfg->getNumSplitThreads() != 1 && tempCS->picture->scheduler.getSplitJobId() == 0 )
- {
- if( m_modeCtrl->isParallelSplit( *tempCS, partitioner ) )
- {
- m_modeCtrl->setParallelSplit( true );
- xCompressCUParallel( tempCS, bestCS, partitioner );
- return;
- }
- }
-#endif
uint32_t compBegin;
uint32_t numComp;
bool jointPLT = false;
@@ -610,17 +560,6 @@ void EncCu::xCompressCU( CodingStructure*& tempCS, CodingStructure*& bestCS, Par
auto slsSbt = dynamic_cast<SaveLoadEncInfoSbt*>( m_modeCtrl );
int maxSLSize = sps.getUseSBT() ? tempCS->slice->getSPS()->getMaxTbSize() : MTS_INTER_MAX_CU_SIZE;
slsSbt->resetSaveloadSbt( maxSLSize );
-#if ENABLE_SPLIT_PARALLELISM
- CHECK( tempCS->picture->scheduler.getSplitJobId() != 0, "The SBT search reset need to happen in sequential region." );
- if (m_pcEncCfg->getNumSplitThreads() > 1)
- {
- for (int jId = 1; jId < NUM_RESERVERD_SPLIT_JOBS; jId++)
- {
- auto slsSbt = dynamic_cast<SaveLoadEncInfoSbt *>(m_pcEncLib->getCuEncoder(jId)->m_modeCtrl);
- slsSbt->resetSaveloadSbt(maxSLSize);
- }
- }
-#endif
}
m_sbtCostSave[0] = m_sbtCostSave[1] = MAX_DOUBLE;
@@ -722,9 +661,6 @@ void EncCu::xCompressCU( CodingStructure*& tempCS, CodingStructure*& bestCS, Par
#endif
))
{
-#if ENABLE_SPLIT_PARALLELISM
- CHECK( tempCS->picture->scheduler.getSplitJobId() > 0, "Changing lambda is only allowed in the master thread!" );
-#endif
if (currTestMode.qp >= 0)
{
updateLambda (&slice, currTestMode.qp,
@@ -915,17 +851,6 @@ void EncCu::xCompressCU( CodingStructure*& tempCS, CodingStructure*& bestCS, Par
//////////////////////////////////////////////////////////////////////////
// Finishing CU
-#if ENABLE_SPLIT_PARALLELISM
- if( bestCS->cus.empty() )
- {
- CHECK( bestCS->cost != MAX_DOUBLE, "Cost should be maximal if no encoding found" );
- CHECK( bestCS->picture->scheduler.getSplitJobId() == 0, "Should always get a result in serial case" );
-
- m_modeCtrl->finishCULevel( partitioner );
- return;
- }
-
-#endif
if( tempCS->cost == MAX_DOUBLE && bestCS->cost == MAX_DOUBLE )
{
//although some coding modes were planned to be tried in RDO, no coding mode actually finished encoding due to early termination
@@ -968,13 +893,6 @@ void EncCu::xCompressCU( CodingStructure*& tempCS, CodingStructure*& bestCS, Par
m_pcIntraSearch->saveCuAreaCostInSCIPU( Area( partitioner.currArea().lumaPos(), partitioner.currArea().lumaSize() ), bestCS->cost );
}
-#if ENABLE_SPLIT_PARALLELISM
- if( tempCS->picture->scheduler.getSplitJobId() == 0 && m_pcEncCfg->getNumSplitThreads() != 1 )
- {
- tempCS->picture->finishParallelPart( currCsArea );
- }
-
-#endif
if (bestCS->cus.size() == 1) // no partition
{
CHECK(bestCS->cus[0]->tileIdx != bestCS->pps->getTileIdx(bestCS->area.lumaPos()), "Wrong tile index!");
@@ -1063,164 +981,6 @@ void EncCu::updateLambda (Slice* slice, const int dQP,
}
#endif // SHARP_LUMA_DELTA_QP || ENABLE_QPA_SUB_CTU
-#if ENABLE_SPLIT_PARALLELISM
-//#undef DEBUG_PARALLEL_TIMINGS
-//#define DEBUG_PARALLEL_TIMINGS 1
-void EncCu::xCompressCUParallel( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner )
-{
- const unsigned wIdx = gp_sizeIdxInfo->idxFrom( partitioner.currArea().lwidth() );
- const unsigned hIdx = gp_sizeIdxInfo->idxFrom( partitioner.currArea().lheight() );
-
- Picture* picture = tempCS->picture;
-
- int numJobs = m_modeCtrl->getNumParallelJobs( *bestCS, partitioner );
-
- bool jobUsed [NUM_RESERVERD_SPLIT_JOBS];
- std::fill( jobUsed, jobUsed + NUM_RESERVERD_SPLIT_JOBS, false );
-
- const UnitArea currArea = CS::getArea( *tempCS, partitioner.currArea(), partitioner.chType );
- const bool doParallel = !m_pcEncCfg->getForceSingleSplitThread();
- omp_set_num_threads( m_pcEncCfg->getNumSplitThreads() );
-
-#pragma omp parallel for schedule(dynamic,1) if(doParallel)
- for( int jId = 1; jId <= numJobs; jId++ )
- {
- // thread start
- picture->scheduler.setSplitThreadId();
- picture->scheduler.setSplitJobId( jId );
-
- QTBTPartitioner jobPartitioner;
- EncCu* jobCuEnc = m_pcEncLib->getCuEncoder( picture->scheduler.getSplitDataId( jId ) );
- auto* jobBlkCache = dynamic_cast<CacheBlkInfoCtrl*>( jobCuEnc->m_modeCtrl );
-#if REUSE_CU_RESULTS
- auto* jobBestCache = dynamic_cast<BestEncInfoCache*>( jobCuEnc->m_modeCtrl );
-#endif
-
- jobPartitioner.copyState( partitioner );
- jobCuEnc ->copyState( this, jobPartitioner, currArea, true );
-
- if( jobBlkCache ) { jobBlkCache ->tick(); }
-#if REUSE_CU_RESULTS
- if( jobBestCache ) { jobBestCache->tick(); }
-
-#endif
- CodingStructure *&jobBest = jobCuEnc->m_pBestCS[wIdx][hIdx];
- CodingStructure *&jobTemp = jobCuEnc->m_pTempCS[wIdx][hIdx];
-
- jobUsed[jId] = true;
-
- jobCuEnc->xCompressCU( jobTemp, jobBest, jobPartitioner );
-
- picture->scheduler.setSplitJobId( 0 );
- // thread stop
- }
- picture->scheduler.setSplitThreadId( 0 );
-
- int bestJId = 0;
- double bestCost = bestCS->cost;
- for( int jId = 1; jId <= numJobs; jId++ )
- {
- EncCu* jobCuEnc = m_pcEncLib->getCuEncoder( picture->scheduler.getSplitDataId( jId ) );
-
- if( jobUsed[jId] && jobCuEnc->m_pBestCS[wIdx][hIdx]->cost < bestCost )
- {
- bestCost = jobCuEnc->m_pBestCS[wIdx][hIdx]->cost;
- bestJId = jId;
- }
- }
-
- if( bestJId > 0 )
- {
- copyState( m_pcEncLib->getCuEncoder( picture->scheduler.getSplitDataId( bestJId ) ), partitioner, currArea, false );
- m_CurrCtx->best = m_CABACEstimator->getCtx();
-
- tempCS = m_pTempCS[wIdx][hIdx];
- bestCS = m_pBestCS[wIdx][hIdx];
- }
-
- const int bitDepthY = tempCS->sps->getBitDepth( CH_L );
- const UnitArea clipdArea = clipArea( currArea, *picture );
-
- CHECK( calcCheckSum( picture->getRecoBuf( clipdArea.Y() ), bitDepthY ) != calcCheckSum( bestCS->getRecoBuf( clipdArea.Y() ), bitDepthY ), "Data copied incorrectly!" );
-
- picture->finishParallelPart( currArea );
-
- if( auto *blkCache = dynamic_cast<CacheBlkInfoCtrl*>( m_modeCtrl ) )
- {
- for( int jId = 1; jId <= numJobs; jId++ )
- {
- if( !jobUsed[jId] || jId == bestJId ) continue;
-
- auto *jobBlkCache = dynamic_cast<CacheBlkInfoCtrl*>( m_pcEncLib->getCuEncoder( picture->scheduler.getSplitDataId( jId ) )->m_modeCtrl );
- CHECK( !jobBlkCache, "If own mode controller has blk info cache capability so should all other mode controllers!" );
- blkCache->CacheBlkInfoCtrl::copyState( *jobBlkCache, partitioner.currArea() );
- }
-
- blkCache->tick();
- }
-#if REUSE_CU_RESULTS
-
- if( auto *blkCache = dynamic_cast<BestEncInfoCache*>( m_modeCtrl ) )
- {
- for( int jId = 1; jId <= numJobs; jId++ )
- {
- if( !jobUsed[jId] || jId == bestJId ) continue;
-
- auto *jobBlkCache = dynamic_cast<BestEncInfoCache*>( m_pcEncLib->getCuEncoder( picture->scheduler.getSplitDataId( jId ) )->m_modeCtrl );
- CHECK( !jobBlkCache, "If own mode controller has blk info cache capability so should all other mode controllers!" );
- blkCache->BestEncInfoCache::copyState( *jobBlkCache, partitioner.currArea() );
- }
-
- blkCache->tick();
- }
-#endif
-}
-
-void EncCu::copyState( EncCu* other, Partitioner& partitioner, const UnitArea& currArea, const bool isDist )
-{
- const unsigned wIdx = gp_sizeIdxInfo->idxFrom( partitioner.currArea().lwidth () );
- const unsigned hIdx = gp_sizeIdxInfo->idxFrom( partitioner.currArea().lheight() );
-
- if( isDist )
- {
- other->m_pBestCS[wIdx][hIdx]->initSubStructure( *m_pBestCS[wIdx][hIdx], partitioner.chType, partitioner.currArea(), false );
- other->m_pTempCS[wIdx][hIdx]->initSubStructure( *m_pTempCS[wIdx][hIdx], partitioner.chType, partitioner.currArea(), false );
- }
- else
- {
- CodingStructure* dst = m_pBestCS[wIdx][hIdx];
- const CodingStructure* src = other->m_pBestCS[wIdx][hIdx];
- bool keepResi = KEEP_PRED_AND_RESI_SIGNALS;
- bool keepPred = true;
-
- dst->useSubStructure( *src, partitioner.chType, currArea, keepPred, true, keepResi, keepResi, true );
-
- dst->cost = src->cost;
- dst->dist = src->dist;
- dst->fracBits = src->fracBits;
- dst->features = src->features;
- }
-
- if( isDist )
- {
- m_CurrCtx = m_CtxBuffer.data();
- }
-
- m_pcInterSearch->copyState( *other->m_pcInterSearch );
- m_modeCtrl ->copyState( *other->m_modeCtrl, partitioner.currArea() );
- m_pcRdCost ->copyState( *other->m_pcRdCost );
- m_pcTrQuant ->copyState( *other->m_pcTrQuant );
- if( m_pcEncCfg->getLmcs() )
- {
- EncReshape *encReshapeThis = dynamic_cast<EncReshape*>( m_pcReshape);
- EncReshape *encReshapeOther = dynamic_cast<EncReshape*>(other->m_pcReshape);
- encReshapeThis->copyState( *encReshapeOther );
- }
-
- m_CABACEstimator->getCtx() = other->m_CABACEstimator->getCtx();
-}
-#endif
-
void EncCu::xCheckModeSplit(CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &partitioner, const EncTestMode& encTestMode, const ModeType modeTypeParent, bool &skipInterPass )
{
const int qp = encTestMode.qp;
diff --git a/source/Lib/EncoderLib/EncCu.h b/source/Lib/EncoderLib/EncCu.h
index 8ae2aad32..b953c9b2f 100644
--- a/source/Lib/EncoderLib/EncCu.h
+++ b/source/Lib/EncoderLib/EncCu.h
@@ -161,10 +161,6 @@ private:
CtxPair* m_CurrCtx;
CtxCache* m_CtxCache;
-#if ENABLE_SPLIT_PARALLELISM
- int m_dataId;
-#endif
-
// Data : encoder control
int m_cuChromaQpOffsetIdxPlus1; // if 0, then cu_chroma_qp_offset_flag will be 0, otherwise cu_chroma_qp_offset_flag will be 1.
@@ -199,9 +195,6 @@ private:
int m_ctuIbcSearchRangeX;
int m_ctuIbcSearchRangeY;
-#if ENABLE_SPLIT_PARALLELISM
- EncLib* m_pcEncLib;
-#endif
int m_bestBcwIdx[2];
double m_bestBcwCost[2];
GeoMotionInfo m_GeoModeTest[GEO_MAX_NUM_CANDS];
@@ -215,7 +208,7 @@ private:
double m_sbtCostSave[2];
public:
/// copy parameters from encoder class
- void init ( EncLib* pcEncLib, const SPS& sps PARL_PARAM( const int jId = 0 ) );
+ void init ( EncLib* pcEncLib, const SPS& sps );
void setDecCuReshaperInEncCU(EncReshape* pcReshape, ChromaFormat chromaFormatIDC) { initDecCuReshaper((Reshape*) pcReshape, chromaFormatIDC); }
/// create internal buffers
@@ -248,10 +241,6 @@ protected:
Distortion getDistortionDb ( CodingStructure &cs, CPelBuf org, CPelBuf reco, ComponentID compID, const CompArea& compArea, bool afterDb );
void xCompressCU ( CodingStructure*& tempCS, CodingStructure*& bestCS, Partitioner& pm, double maxCostAllowed = MAX_DOUBLE );
-#if ENABLE_SPLIT_PARALLELISM
- void xCompressCUParallel ( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &pm );
- void copyState ( EncCu* other, Partitioner& pm, const UnitArea& currArea, const bool isDist );
-#endif
bool
xCheckBestMode ( CodingStructure *&tempCS, CodingStructure *&bestCS, Partitioner &pm, const EncTestMode& encTestmode );
diff --git a/source/Lib/EncoderLib/EncGOP.cpp b/source/Lib/EncoderLib/EncGOP.cpp
index 199d0ec8a..81f350b69 100644
--- a/source/Lib/EncoderLib/EncGOP.cpp
+++ b/source/Lib/EncoderLib/EncGOP.cpp
@@ -2110,9 +2110,6 @@ void EncGOP::compressGOP( int iPOCLast, int iNumPicRcvd, PicList& rcListPic,
m_pcSliceEncoder->create( picWidth, picHeight, chromaFormatIDC, maxCUWidth, maxCUHeight, maxTotalCUDepth );
-#if ENABLE_SPLIT_PARALLELISM
- pcPic->scheduler.init( pcPic->cs->pcv->heightInCtus, pcPic->cs->pcv->widthInCtus, 1 , 0 , m_pcCfg->getNumSplitThreads() );
-#endif
pcPic->createTempBuffers( pcPic->cs->pps->pcv->maxCUWidth );
pcPic->cs->createCoeffs((bool)pcPic->cs->sps->getPLTMode());
@@ -3010,7 +3007,7 @@ void EncGOP::compressGOP( int iPOCLast, int iNumPicRcvd, PicList& rcListPic,
m_pcSAO->SAOProcess( cs, sliceEnabled, pcSlice->getLambdas(),
#if ENABLE_QPA
- (m_pcCfg->getUsePerceptQPA() && !m_pcCfg->getUseRateCtrl() && pcSlice->getPPS()->getUseDQP() ? m_pcEncLib->getRdCost (PARL_PARAM0 (0))->getChromaWeight() : 0.0),
+ (m_pcCfg->getUsePerceptQPA() && !m_pcCfg->getUseRateCtrl() && pcSlice->getPPS()->getUseDQP() ? m_pcEncLib->getRdCost ()->getChromaWeight() : 0.0),
#endif
m_pcCfg->getTestSAODisableAtPictureLevel(), m_pcCfg->getSaoEncodingRate(), m_pcCfg->getSaoEncodingRateChroma(), m_pcCfg->getSaoCtuBoundary(), m_pcCfg->getSaoGreedyMergeEnc() );
//assign SAO slice header
@@ -3042,7 +3039,7 @@ void EncGOP::compressGOP( int iPOCLast, int iNumPicRcvd, PicList& rcListPic,
m_pcALF->initCABACEstimator(m_pcEncLib->getCABACEncoder(), m_pcEncLib->getCtxCache(), pcSlice, m_pcEncLib->getApsMap());
m_pcALF->ALFProcess(cs, pcSlice->getLambdas()
#if ENABLE_QPA
- , (m_pcCfg->getUsePerceptQPA() && !m_pcCfg->getUseRateCtrl() && pcSlice->getPPS()->getUseDQP() ? m_pcEncLib->getRdCost(PARL_PARAM0(0))->getChromaWeight() : 0.0)
+ , (m_pcCfg->getUsePerceptQPA() && !m_pcCfg->getUseRateCtrl() && pcSlice->getPPS()->getUseDQP() ? m_pcEncLib->getRdCost()->getChromaWeight() : 0.0)
#endif
, pcPic, uiNumSliceSegments
);
diff --git a/source/Lib/EncoderLib/EncLib.cpp b/source/Lib/EncoderLib/EncLib.cpp
index af03029af..72eb14452 100644
--- a/source/Lib/EncoderLib/EncLib.cpp
+++ b/source/Lib/EncoderLib/EncLib.cpp
@@ -43,9 +43,6 @@
#include "CommonLib/Picture.h"
#include "CommonLib/CommonDef.h"
#include "CommonLib/ChromaFormat.h"
-#if ENABLE_SPLIT_PARALLELISM
-#include <omp.h>
-#endif
#include "EncLibCommon.h"
#include "CommonLib/ProfileLevelTier.h"
@@ -102,28 +99,7 @@ void EncLib::create( const int layerId )
m_iPOCLast = m_compositeRefEnabled ? -2 : -1;
// create processing unit classes
m_cGOPEncoder. create( );
-#if ENABLE_SPLIT_PARALLELISM
-#if ENABLE_SPLIT_PARALLELISM
- m_numCuEncStacks = m_numSplitThreads == 1 ? 1 : NUM_RESERVERD_SPLIT_JOBS;
-#else
- m_numCuEncStacks = 1;
-#endif
-
- m_cCuEncoder = new EncCu [m_numCuEncStacks];
- m_cInterSearch = new InterSearch [m_numCuEncStacks];
- m_cIntraSearch = new IntraSearch [m_numCuEncStacks];
- m_cTrQuant = new TrQuant [m_numCuEncStacks];
- m_CABACEncoder = new CABACEncoder [m_numCuEncStacks];
- m_cRdCost = new RdCost [m_numCuEncStacks];
- m_CtxCache = new CtxCache [m_numCuEncStacks];
-
- for( int jId = 0; jId < m_numCuEncStacks; jId++ )
- {
- m_cCuEncoder[jId]. create( this );
- }
-#else
m_cCuEncoder. create( this );
-#endif
#if JVET_J0090_MEMORY_BANDWITH_MEASURE
m_cInterSearch.cacheAssign( &m_cacheModel );
#endif
@@ -135,19 +111,9 @@ void EncLib::create( const int layerId )
m_cLoopFilter.initEncPicYuvBuffer(m_chromaFormatIDC, Size(getSourceWidth(), getSourceHeight()), getMaxCUWidth());
}
-#if ENABLE_SPLIT_PARALLELISM
- m_cReshaper = new EncReshape[m_numCuEncStacks];
-#endif
if (m_lmcsEnabled)
{
-#if ENABLE_SPLIT_PARALLELISM
- for (int jId = 0; jId < m_numCuEncStacks; jId++)
- {
- m_cReshaper[jId].createEnc(getSourceWidth(), getSourceHeight(), m_maxCUWidth, m_maxCUHeight, m_bitDepth[COMPONENT_Y]);
- }
-#else
m_cReshaper.createEnc( getSourceWidth(), getSourceHeight(), m_maxCUWidth, m_maxCUHeight, m_bitDepth[COMPONENT_Y]);
-#endif
}
if ( m_RCEnableRateControl )
{
@@ -162,14 +128,7 @@ void EncLib::destroy ()
// destroy processing unit classes
m_cGOPEncoder. destroy();
m_cSliceEncoder. destroy();
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 0; jId < m_numCuEncStacks; jId++ )
- {
- m_cCuEncoder[jId].destroy();
- }
-#else
m_cCuEncoder. destroy();
-#endif
if( m_alf )
{
m_cEncALF.destroy();
@@ -178,34 +137,9 @@ void EncLib::destroy ()
m_cEncSAO. destroy();
m_cLoopFilter. destroy();
m_cRateCtrl. destroy();
-#if ENABLE_SPLIT_PARALLELISM
- for (int jId = 0; jId < m_numCuEncStacks; jId++)
- {
- m_cReshaper[jId]. destroy();
- }
-#else
m_cReshaper. destroy();
-#endif
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 0; jId < m_numCuEncStacks; jId++ )
- {
- m_cInterSearch[jId]. destroy();
- m_cIntraSearch[jId]. destroy();
- }
-#else
m_cInterSearch. destroy();
m_cIntraSearch. destroy();
-#endif
-
-#if ENABLE_SPLIT_PARALLELISM
- delete[] m_cCuEncoder;
- delete[] m_cInterSearch;
- delete[] m_cIntraSearch;
- delete[] m_cTrQuant;
- delete[] m_CABACEncoder;
- delete[] m_cRdCost;
- delete[] m_CtxCache;
-#endif
return;
}
@@ -235,13 +169,6 @@ void EncLib::init( bool isFieldCoding, AUWriterIf* auWriterIf )
xInitVPS( sps0 );
xInitOPI(m_opi);
xInitDCI(m_dci, sps0);
-#if ENABLE_SPLIT_PARALLELISM
- if( omp_get_dynamic() )
- {
- omp_set_dynamic( false );
- }
- omp_set_nested( true );
-#endif
if (getUseCompositeRef() || getDependentRAPIndicationSEIEnabled())
{
@@ -254,14 +181,7 @@ void EncLib::init( bool isFieldCoding, AUWriterIf* auWriterIf )
m_cRateCtrl.initHrdParam(sps0.getGeneralHrdParameters(), sps0.getOlsHrdParameters(), m_iFrameRate, m_RCInitialCpbFullness);
}
#endif
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 0; jId < m_numCuEncStacks; jId++ )
- {
- m_cRdCost[jId].setCostMode ( m_costMode );
- }
-#else
m_cRdCost.setCostMode ( m_costMode );
-#endif
// initialize PPS
pps0.setPicWidthInLumaSamples( m_iSourceWidth );
@@ -373,54 +293,6 @@ void EncLib::init( bool isFieldCoding, AUWriterIf* auWriterIf )
// initialize processing unit classes
m_cGOPEncoder. init( this );
m_cSliceEncoder.init( this, sps0 );
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 0; jId < m_numCuEncStacks; jId++ )
- {
- // precache a few objects
- for( int i = 0; i < 10; i++ )
- {
- auto x = m_CtxCache[jId].get();
- m_CtxCache[jId].cache( x );
- }
-
- m_cCuEncoder[jId].init( this, sps0, jId );
-
- // initialize transform & quantization class
- m_cTrQuant[jId].init( jId == 0 ? nullptr : m_cTrQuant[0].getQuant(),
- 1 << m_log2MaxTbSize,
-
- m_useRDOQ,
- m_useRDOQTS,
-#if T0196_SELECTIVE_RDOQ
- m_useSelectiveRDOQ,
-#endif
- true
- );
-
- // initialize encoder search class
- CABACWriter* cabacEstimator = m_CABACEncoder[jId].getCABACEstimator( &sps0 );
- m_cIntraSearch[jId].init( this,
- &m_cTrQuant[jId],
- &m_cRdCost[jId],
- cabacEstimator,
- getCtxCache( jId ), m_maxCUWidth, m_maxCUHeight, floorLog2(m_maxCUWidth) - m_log2MinCUSize
- , &m_cReshaper[jId]
- , sps0.getBitDepth(CHANNEL_TYPE_LUMA)
- );
- m_cInterSearch[jId].init( this,
- &m_cTrQuant[jId],
- m_iSearchRange,
- m_bipredSearchRange,
- m_motionEstimationSearchMethod,
- getUseCompositeRef(),
- m_maxCUWidth, m_maxCUHeight, floorLog2(m_maxCUWidth) - m_log2MinCUSize, &m_cRdCost[jId], cabacEstimator, getCtxCache( jId )
- , &m_cReshaper[jId]
- );
-
- // link temporary buffets from intra search with inter search to avoid unnecessary memory overhead
- m_cInterSearch[jId].setTempBuffers( m_cIntraSearch[jId].getSplitCSBuf(), m_cIntraSearch[jId].getFullCSBuf(), m_cIntraSearch[jId].getSaveCSBuf() );
- }
-#else // ENABLE_SPLIT_PARALLELISM || ENABLE_WPP_PARALLELISM
m_cCuEncoder. init( this, sps0 );
// initialize transform & quantization class
@@ -456,7 +328,6 @@ void EncLib::init( bool isFieldCoding, AUWriterIf* auWriterIf )
// link temporary buffets from intra search with inter search to avoid unneccessary memory overhead
m_cInterSearch.setTempBuffers( m_cIntraSearch.getSplitCSBuf(), m_cIntraSearch.getFullCSBuf(), m_cIntraSearch.getSaveCSBuf() );
-#endif // ENABLE_SPLIT_PARALLELISM || ENABLE_WPP_PARALLELISM
m_iMaxRefPicNum = 0;
@@ -507,26 +378,12 @@ void EncLib::xInitScalingLists( SPS &sps, APS &aps )
{
quant->setFlatScalingList(maxLog2TrDynamicRange, sps.getBitDepths());
quant->setUseScalingList(false);
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 1; jId < m_numCuEncStacks; jId++ )
- {
- getTrQuant( jId )->getQuant()->setFlatScalingList( maxLog2TrDynamicRange, sps.getBitDepths() );
- getTrQuant( jId )->getQuant()->setUseScalingList( false );
- }
-#endif
}
else if(getUseScalingListId() == SCALING_LIST_DEFAULT)
{
aps.getScalingList().setDefaultScalingList ();
quant->setScalingList( &( aps.getScalingList() ), maxLog2TrDynamicRange, sps.getBitDepths() );
quant->setUseScalingList(true);
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 1; jId < m_numCuEncStacks; jId++ )
- {
- getTrQuant( jId )->getQuant()->setUseScalingList( true );
- }
- sps.setDisableScalingMatrixForLfnstBlks(getDisableScalingMatrixForLfnstBlks());
-#endif
}
else if(getUseScalingListId() == SCALING_LIST_FILE_READ)
{
@@ -540,12 +397,6 @@ void EncLib::xInitScalingLists( SPS &sps, APS &aps )
aps.getScalingList().setChromaScalingListPresentFlag((sps.getChromaFormatIdc()!=CHROMA_400));
quant->setScalingList( &( aps.getScalingList() ), maxLog2TrDynamicRange, sps.getBitDepths() );
quant->setUseScalingList(true);
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 1; jId < m_numCuEncStacks; jId++ )
- {
- getTrQuant( jId )->getQuant()->setUseScalingList( true );
- }
-#endif
sps.setDisableScalingMatrixForLfnstBlks(getDisableScalingMatrixForLfnstBlks());
}
diff --git a/source/Lib/EncoderLib/EncLib.h b/source/Lib/EncoderLib/EncLib.h
index a87bdf133..1c05ed00b 100644
--- a/source/Lib/EncoderLib/EncLib.h
+++ b/source/Lib/EncoderLib/EncLib.h
@@ -79,65 +79,35 @@ private:
int m_layerId;
// encoder search
-#if ENABLE_SPLIT_PARALLELISM
- InterSearch *m_cInterSearch; ///< encoder search class
- IntraSearch *m_cIntraSearch; ///< encoder search class
-#else
InterSearch m_cInterSearch; ///< encoder search class
IntraSearch m_cIntraSearch; ///< encoder search class
-#endif
// coding tool
-#if ENABLE_SPLIT_PARALLELISM
- TrQuant *m_cTrQuant; ///< transform & quantization class
-#else
TrQuant m_cTrQuant; ///< transform & quantization class
-#endif
LoopFilter m_cLoopFilter; ///< deblocking filter class
EncSampleAdaptiveOffset m_cEncSAO; ///< sample adaptive offset class
EncAdaptiveLoopFilter m_cEncALF;
HLSWriter m_HLSWriter; ///< CAVLC encoder
-#if ENABLE_SPLIT_PARALLELISM
- CABACEncoder *m_CABACEncoder;
-#else
CABACEncoder m_CABACEncoder;
-#endif
-#if ENABLE_SPLIT_PARALLELISM
- EncReshape *m_cReshaper; ///< reshaper class
-#else
EncReshape m_cReshaper; ///< reshaper class
-#endif
// processing unit
EncGOP m_cGOPEncoder; ///< GOP encoder
EncSlice m_cSliceEncoder; ///< slice encoder
-#if ENABLE_SPLIT_PARALLELISM
- EncCu *m_cCuEncoder; ///< CU encoder
-#else
EncCu m_cCuEncoder; ///< CU encoder
-#endif
// SPS
ParameterSetMap<SPS>& m_spsMap; ///< SPS. This is the base value. This is copied to PicSym
ParameterSetMap<PPS>& m_ppsMap; ///< PPS. This is the base value. This is copied to PicSym
ParameterSetMap<APS>& m_apsMap; ///< APS. This is the base value. This is copied to PicSym
PicHeader m_picHeader; ///< picture header
// RD cost computation
-#if ENABLE_SPLIT_PARALLELISM
- RdCost *m_cRdCost; ///< RD cost computation class
- CtxCache *m_CtxCache; ///< buffer for temporarily stored context models
-#else
RdCost m_cRdCost; ///< RD cost computation class
CtxCache m_CtxCache; ///< buffer for temporarily stored context models
-#endif
// quality control
RateCtrl m_cRateCtrl; ///< Rate control class
AUWriterIf* m_AUWriterIf;
-#if ENABLE_SPLIT_PARALLELISM
- int m_numCuEncStacks;
-#endif
-
#if JVET_J0090_MEMORY_BANDWITH_MEASURE
CacheModel m_cacheModel;
#endif
@@ -195,40 +165,22 @@ public:
AUWriterIf* getAUWriterIf () { return m_AUWriterIf; }
PicList* getListPic () { return &m_cListPic; }
-#if ENABLE_SPLIT_PARALLELISM
- InterSearch* getInterSearch ( int jId = 0 ) { return &m_cInterSearch[jId]; }
- IntraSearch* getIntraSearch ( int jId = 0 ) { return &m_cIntraSearch[jId]; }
-
- TrQuant* getTrQuant ( int jId = 0 ) { return &m_cTrQuant[jId]; }
-#else
InterSearch* getInterSearch () { return &m_cInterSearch; }
IntraSearch* getIntraSearch () { return &m_cIntraSearch; }
TrQuant* getTrQuant () { return &m_cTrQuant; }
-#endif
LoopFilter* getLoopFilter () { return &m_cLoopFilter; }
EncSampleAdaptiveOffset* getSAO () { return &m_cEncSAO; }
EncAdaptiveLoopFilter* getALF () { return &m_cEncALF; }
EncGOP* getGOPEncoder () { return &m_cGOPEncoder; }
EncSlice* getSliceEncoder () { return &m_cSliceEncoder; }
EncHRD* getHRD () { return &m_encHRD; }
-#if ENABLE_SPLIT_PARALLELISM
- EncCu* getCuEncoder ( int jId = 0 ) { return &m_cCuEncoder[jId]; }
-#else
EncCu* getCuEncoder () { return &m_cCuEncoder; }
-#endif
HLSWriter* getHLSWriter () { return &m_HLSWriter; }
-#if ENABLE_SPLIT_PARALLELISM
- CABACEncoder* getCABACEncoder ( int jId = 0 ) { return &m_CABACEncoder[jId]; }
-
- RdCost* getRdCost ( int jId = 0 ) { return &m_cRdCost[jId]; }
- CtxCache* getCtxCache ( int jId = 0 ) { return &m_CtxCache[jId]; }
-#else
CABACEncoder* getCABACEncoder () { return &m_CABACEncoder; }
RdCost* getRdCost () { return &m_cRdCost; }
CtxCache* getCtxCache () { return &m_CtxCache; }
-#endif
RateCtrl* getRateCtrl () { return &m_cRateCtrl; }
@@ -242,16 +194,7 @@ public:
const PPS* getPPS( int Id ) { return m_ppsMap.getPS( Id); }
const APS* getAPS(int Id) { return m_apsMap.getPS(Id); }
-#if ENABLE_SPLIT_PARALLELISM
- void setNumCuEncStacks( int n ) { m_numCuEncStacks = n; }
- int getNumCuEncStacks() const { return m_numCuEncStacks; }
-#endif
-
-#if ENABLE_SPLIT_PARALLELISM
- EncReshape* getReshaper( int jId = 0 ) { return &m_cReshaper[jId]; }
-#else
EncReshape* getReshaper() { return &m_cReshaper; }
-#endif
ParameterSetMap<APS>* getApsMap() { return &m_apsMap; }
diff --git a/source/Lib/EncoderLib/EncModeCtrl.cpp b/source/Lib/EncoderLib/EncModeCtrl.cpp
index ed4f7116e..cc8335fa1 100644
--- a/source/Lib/EncoderLib/EncModeCtrl.cpp
+++ b/source/Lib/EncoderLib/EncModeCtrl.cpp
@@ -64,15 +64,6 @@ void EncModeCtrl::init( EncCfg *pCfg, RateCtrl *pRateCtrl, RdCost* pRdCost )
bool EncModeCtrl::tryModeMaster( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner )
{
-#if ENABLE_SPLIT_PARALLELISM
- if( m_ComprCUCtxList.back().isLevelSplitParallel )
- {
- if( !parallelJobSelector( encTestmode, cs, partitioner ) )
- {
- return false;
- }
- }
-#endif
return tryMode( encTestmode, cs, partitioner );
}
@@ -248,18 +239,6 @@ int EncModeCtrl::calculateLumaDQP( const CPelBuf& rcOrg )
}
#endif
-#if ENABLE_SPLIT_PARALLELISM
-void EncModeCtrl::copyState( const EncModeCtrl& other, const UnitArea& area )
-{
- m_slice = other.m_slice;
- m_fastDeltaQP = other.m_fastDeltaQP;
- m_lumaQPOffset = other.m_lumaQPOffset;
- m_runNextInParallel
- = other.m_runNextInParallel;
- m_ComprCUCtxList = other.m_ComprCUCtxList;
-}
-
-#endif
void CacheBlkInfoCtrl::create()
{
const unsigned numPos = MAX_CU_SIZE >> MIN_CU_LOG2;
@@ -354,71 +333,7 @@ void CacheBlkInfoCtrl::init( const Slice &slice )
}
m_slice_chblk = &slice;
-#if ENABLE_SPLIT_PARALLELISM
-
- m_currTemporalId = 0;
-#endif
}
-#if ENABLE_SPLIT_PARALLELISM
-
-void CacheBlkInfoCtrl::touch( const UnitArea& area )
-{
- CodedCUInfo& cuInfo = getBlkInfo( area );
- cuInfo.temporalId = m_currTemporalId;
-}
-
-void CacheBlkInfoCtrl::copyState( const CacheBlkInfoCtrl &other, const UnitArea& area )
-{
- m_slice_chblk = other.m_slice_chblk;
-
- m_currTemporalId = other.m_currTemporalId;
-
- if( m_slice_chblk->isIntra() ) return;
-
- const int cuSizeMask = m_slice_chblk->getSPS()->getMaxCUWidth() - 1;
-
- const int minPosX = ( area.lx() & cuSizeMask ) >> MIN_CU_LOG2;
- const int minPosY = ( area.ly() & cuSizeMask ) >> MIN_CU_LOG2;
- const int maxPosX = ( area.Y().bottomRight().x & cuSizeMask ) >> MIN_CU_LOG2;
- const int maxPosY = ( area.Y().bottomRight().y & cuSizeMask ) >> MIN_CU_LOG2;
-
- for( unsigned x = minPosX; x <= maxPosX; x++ )
- {
- for( unsigned y = minPosY; y <= maxPosY; y++ )
- {
- for( int wIdx = 0; wIdx < gp_sizeIdxInfo->numWidths(); wIdx++ )
- {
- const int width = gp_sizeIdxInfo->sizeFrom( wIdx );
-
- if( m_codedCUInfo[x][y][wIdx] && width <= area.lwidth() && x + ( width >> MIN_CU_LOG2 ) <= ( maxPosX + 1 ) )
- {
- for( int hIdx = 0; hIdx < gp_sizeIdxInfo->numHeights(); hIdx++ )
- {
- const int height = gp_sizeIdxInfo->sizeFrom( hIdx );
-
- if( gp_sizeIdxInfo->isCuSize( height ) && height <= area.lheight() && y + ( height >> MIN_CU_LOG2 ) <= ( maxPosY + 1 ) )
- {
- if( other.m_codedCUInfo[x][y][wIdx][hIdx]->temporalId > m_codedCUInfo[x][y][wIdx][hIdx]->temporalId )
- {
- *m_codedCUInfo[x][y][wIdx][hIdx] = *other.m_codedCUInfo[x][y][wIdx][hIdx];
- m_codedCUInfo[x][y][wIdx][hIdx]->temporalId = m_currTemporalId;
- }
- }
- else if( y + ( height >> MIN_CU_LOG2 ) > maxPosY + 1 )
- {
- break;;
- }
- }
- }
- else if( x + ( width >> MIN_CU_LOG2 ) > maxPosX + 1 )
- {
- break;
- }
- }
- }
- }
-}
-#endif
CodedCUInfo& CacheBlkInfoCtrl::getBlkInfo( const UnitArea& area )
{
@@ -461,10 +376,6 @@ void CacheBlkInfoCtrl::setMv( const UnitArea& area, const RefPicList refPicList,
m_codedCUInfo[idx1][idx2][idx3][idx4]->saveMv [refPicList][iRefIdx] = rMv;
m_codedCUInfo[idx1][idx2][idx3][idx4]->validMv[refPicList][iRefIdx] = true;
-#if ENABLE_SPLIT_PARALLELISM
-
- touch( area );
-#endif
}
bool CacheBlkInfoCtrl::getMv( const UnitArea& area, const RefPicList refPicList, const int iRefIdx, Mv& rMv ) const
@@ -563,13 +474,6 @@ bool SaveLoadEncInfoSbt::saveBestSbt( const UnitArea& area, const uint32_t curPu
return true;
}
-#if ENABLE_SPLIT_PARALLELISM
-void SaveLoadEncInfoSbt::copyState(const SaveLoadEncInfoSbt &other)
-{
- m_sliceSbt = other.m_sliceSbt;
-}
-#endif
-
void SaveLoadEncInfoSbt::resetSaveloadSbt( int maxSbtSize )
{
int numSizeIdx = gp_sizeIdxInfo->idxFrom( maxSbtSize ) - MIN_CU_LOG2 + 1;
@@ -852,10 +756,6 @@ void BestEncInfoCache::init( const Slice &slice )
}
}
}
-#if ENABLE_SPLIT_PARALLELISM
-
- m_currTemporalId = 0;
-#endif
}
bool BestEncInfoCache::setFromCs( const CodingStructure& cs, const Partitioner& partitioner )
@@ -993,75 +893,6 @@ bool BestEncInfoCache::setCsFrom( CodingStructure& cs, EncTestMode& testMode, co
return true;
}
-#if ENABLE_SPLIT_PARALLELISM
-void BestEncInfoCache::copyState(const BestEncInfoCache &other, const UnitArea &area)
-{
- m_slice_bencinf = other.m_slice_bencinf;
- m_currTemporalId = other.m_currTemporalId;
-
- if( m_slice_bencinf->isIntra() ) return;
-
- const int cuSizeMask = m_slice_bencinf->getSPS()->getMaxCUWidth() - 1;
-
- const int minPosX = ( area.lx() & cuSizeMask ) >> MIN_CU_LOG2;
- const int minPosY = ( area.ly() & cuSizeMask ) >> MIN_CU_LOG2;
- const int maxPosX = ( area.Y().bottomRight().x & cuSizeMask ) >> MIN_CU_LOG2;
- const int maxPosY = ( area.Y().bottomRight().y & cuSizeMask ) >> MIN_CU_LOG2;
-
- for( unsigned x = minPosX; x <= maxPosX; x++ )
- {
- for( unsigned y = minPosY; y <= maxPosY; y++ )
- {
- for( int wIdx = 0; wIdx < gp_sizeIdxInfo->numWidths(); wIdx++ )
- {
- const int width = gp_sizeIdxInfo->sizeFrom( wIdx );
-
- if( m_bestEncInfo[x][y][wIdx] && width <= area.lwidth() && x + ( width >> MIN_CU_LOG2 ) <= ( maxPosX + 1 ) )
- {
- for( int hIdx = 0; hIdx < gp_sizeIdxInfo->numHeights(); hIdx++ )
- {
- const int height = gp_sizeIdxInfo->sizeFrom( hIdx );
-
- if( gp_sizeIdxInfo->isCuSize( height ) && height <= area.lheight() && y + ( height >> MIN_CU_LOG2 ) <= ( maxPosY + 1 ) )
- {
- if( other.m_bestEncInfo[x][y][wIdx][hIdx]->temporalId > m_bestEncInfo[x][y][wIdx][hIdx]->temporalId )
- {
- m_bestEncInfo[x][y][wIdx][hIdx]->cu = other.m_bestEncInfo[x][y][wIdx][hIdx]->cu;
- m_bestEncInfo[x][y][wIdx][hIdx]->pu = other.m_bestEncInfo[x][y][wIdx][hIdx]->pu;
- m_bestEncInfo[x][y][wIdx][hIdx]->numTus = other.m_bestEncInfo[x][y][wIdx][hIdx]->numTus;
- m_bestEncInfo[x][y][wIdx][hIdx]->poc = other.m_bestEncInfo[x][y][wIdx][hIdx]->poc;
- m_bestEncInfo[x][y][wIdx][hIdx]->testMode = other.m_bestEncInfo[x][y][wIdx][hIdx]->testMode;
-
- for( int i = 0; i < m_bestEncInfo[x][y][wIdx][hIdx]->numTus; i++ )
- m_bestEncInfo[x][y][wIdx][hIdx]->tus[i] = other.m_bestEncInfo[x][y][wIdx][hIdx]->tus[i];
- }
- }
- else if( y + ( height >> MIN_CU_LOG2 ) > maxPosY + 1 )
- {
- break;;
- }
- }
- }
- else if( x + ( width >> MIN_CU_LOG2 ) > maxPosX + 1 )
- {
- break;
- }
- }
- }
- }
-}
-
-void BestEncInfoCache::touch(const UnitArea &area)
-{
- unsigned idx1, idx2, idx3, idx4;
- getAreaIdx(area.Y(), *m_slice_bencinf->getPPS()->pcv, idx1, idx2, idx3, idx4);
- BestEncodingInfo &encInfo = *m_bestEncInfo[idx1][idx2][idx3][idx4];
-
- encInfo.temporalId = m_currTemporalId;
-}
-
-#endif
-
#endif
static bool interHadActive( const ComprCUCtx& ctx )
@@ -1102,9 +933,6 @@ void EncModeCtrlMTnoRQT::initCTUEncoding( const Slice &slice )
CHECK( !m_ComprCUCtxList.empty(), "Mode list is not empty at the beginning of a CTU" );
m_slice = &slice;
-#if ENABLE_SPLIT_PARALLELISM
- m_runNextInParallel = false;
-#endif
if( m_pcEncCfg->getUseE0023FastEnc() )
{
@@ -1136,19 +964,6 @@ void EncModeCtrlMTnoRQT::initCULevel( Partitioner &partitioner, const CodingStru
m_ComprCUCtxList.push_back( ComprCUCtx( cs, minDepth, maxDepth, NUM_EXTRA_FEATURES ) );
-#if ENABLE_SPLIT_PARALLELISM
- if( m_runNextInParallel )
- {
- for( auto &level : m_ComprCUCtxList )
- {
- CHECK( level.isLevelSplitParallel, "Tring to parallelize a level within parallel execution!" );
- }
- CHECK( cs.picture->scheduler.getSplitJobId() == 0, "Trying to run a parallel level although jobId is 0!" );
- m_runNextInParallel = false;
- m_ComprCUCtxList.back().isLevelSplitParallel = true;
- }
-
-#endif
const CodingUnit* cuLeft = cs.getCU( cs.area.blocks[partitioner.chType].pos().offset( -1, 0 ), partitioner.chType );
const CodingUnit* cuAbove = cs.getCU( cs.area.blocks[partitioner.chType].pos().offset( 0, -1 ), partitioner.chType );
@@ -1746,9 +1561,6 @@ bool EncModeCtrlMTnoRQT::tryMode( const EncTestMode& encTestmode, const CodingSt
{
case CU_QUAD_SPLIT:
{
-#if ENABLE_SPLIT_PARALLELISM
- if( !cuECtx.isLevelSplitParallel )
-#endif
if( !cuECtx.get<bool>( QT_BEFORE_BT ) && bestCU )
{
unsigned maxBTD = cs.pcv->getMaxBtDepth( slice, partitioner.chType );
@@ -1971,12 +1783,7 @@ bool EncModeCtrlMTnoRQT::tryMode( const EncTestMode& encTestmode, const CodingSt
relatedCU.relatedCuIsValid = true;
}
}
-#if ENABLE_SPLIT_PARALLELISM
-#if REUSE_CU_RESULTS
- BestEncInfoCache::touch(partitioner.currArea());
-#endif
- CacheBlkInfoCtrl::touch(partitioner.currArea());
-#endif
+
cuECtx.set( IS_BEST_NOSPLIT_SKIP, bestCU->skip );
}
}
@@ -2117,105 +1924,3 @@ bool EncModeCtrlMTnoRQT::useModeResult( const EncTestMode& encTestmode, CodingSt
}
}
-#if ENABLE_SPLIT_PARALLELISM
-void EncModeCtrlMTnoRQT::copyState( const EncModeCtrl& other, const UnitArea& area )
-{
- const EncModeCtrlMTnoRQT* pOther = dynamic_cast<const EncModeCtrlMTnoRQT*>( &other );
-
- CHECK( !pOther, "Trying to copy state from a different type of controller" );
-
- this->EncModeCtrl ::copyState( *pOther, area );
- this->CacheBlkInfoCtrl ::copyState( *pOther, area );
-#if REUSE_CU_RESULTS
- this->BestEncInfoCache ::copyState( *pOther, area );
-#endif
- this->SaveLoadEncInfoSbt ::copyState( *pOther );
-
- m_skipThreshold = pOther->m_skipThreshold;
-}
-
-int EncModeCtrlMTnoRQT::getNumParallelJobs( const CodingStructure &cs, Partitioner& partitioner ) const
-{
- int numJobs = 0;
-
- if( partitioner.canSplit( CU_TRIH_SPLIT, cs ) )
- {
- numJobs = 6;
- }
- else if( partitioner.canSplit( CU_TRIV_SPLIT, cs ) )
- {
- numJobs = 5;
- }
- else if( partitioner.canSplit( CU_HORZ_SPLIT, cs ) )
- {
- numJobs = 4;
- }
- else if( partitioner.canSplit( CU_VERT_SPLIT, cs ) )
- {
- numJobs = 3;
- }
- else if( partitioner.canSplit( CU_QUAD_SPLIT, cs ) )
- {
- numJobs = 2;
- }
- else if( partitioner.canSplit( CU_DONT_SPLIT, cs ) )
- {
- numJobs = 1;
- }
-
- CHECK( numJobs >= NUM_RESERVERD_SPLIT_JOBS, "More jobs specified than allowed" );
-
- return numJobs;
-}
-
-bool EncModeCtrlMTnoRQT::isParallelSplit( const CodingStructure &cs, Partitioner& partitioner ) const
-{
- if( partitioner.getImplicitSplit( cs ) != CU_DONT_SPLIT || cs.picture->scheduler.getSplitJobId() != 0 ) return false;
- if( cs.pps->getUseDQP() && partitioner.currQgEnable() ) return false;
- const int numJobs = getNumParallelJobs( cs, partitioner );
- const int numPxl = partitioner.currArea().Y().area();
- const int parlAt = m_pcEncCfg->getNumSplitThreads() <= 3 ? 1024 : 256;
- if( cs.slice->isIntra() && numJobs > 2 && ( numPxl == parlAt || !partitioner.canSplit( CU_QUAD_SPLIT, cs ) ) ) return true;
- if( !cs.slice->isIntra() && numJobs > 1 && ( numPxl == parlAt || !partitioner.canSplit( CU_QUAD_SPLIT, cs ) ) ) return true;
- return false;
-}
-
-bool EncModeCtrlMTnoRQT::parallelJobSelector( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner ) const
-{
- // Job descriptors
- // - 1: all non-split modes
- // - 2: QT-split
- // - 3: all vertical modes but TT_V
- // - 4: all horizontal modes but TT_H
- // - 5: TT_V
- // - 6: TT_H
- switch( cs.picture->scheduler.getSplitJobId() )
- {
- case 1:
- // be sure to execute post dont split
- return !isModeSplit( encTestmode );
- break;
- case 2:
- return encTestmode.type == ETM_SPLIT_QT;
- break;
- case 3:
- return encTestmode.type == ETM_SPLIT_BT_V;
- break;
- case 4:
- return encTestmode.type == ETM_SPLIT_BT_H;
- break;
- case 5:
- return encTestmode.type == ETM_SPLIT_TT_V;
- break;
- case 6:
- return encTestmode.type == ETM_SPLIT_TT_H;
- break;
- default:
- THROW( "Unknown job-ID for parallelization of EncModeCtrlMTnoRQT: " << cs.picture->scheduler.getSplitJobId() );
- break;
- }
-}
-
-#endif
-
-
diff --git a/source/Lib/EncoderLib/EncModeCtrl.h b/source/Lib/EncoderLib/EncModeCtrl.h
index 740f1a5e2..b590cb932 100644
--- a/source/Lib/EncoderLib/EncModeCtrl.h
+++ b/source/Lib/EncoderLib/EncModeCtrl.h
@@ -194,10 +194,6 @@ struct ComprCUCtx
, skipSecondMTSPass
( false )
, interHad (std::numeric_limits<Distortion>::max())
-#if ENABLE_SPLIT_PARALLELISM
- , isLevelSplitParallel
- ( false )
-#endif
, bestCostWithoutSplitFlags( MAX_DOUBLE )
, bestCostMtsFirstPassNoIsp( MAX_DOUBLE )
, bestCostIsp ( MAX_DOUBLE )
@@ -245,9 +241,6 @@ struct ComprCUCtx
double bestMtsSize2Nx2N1stPass;
bool skipSecondMTSPass;
Distortion interHad;
-#if ENABLE_SPLIT_PARALLELISM
- bool isLevelSplitParallel;
-#endif
double bestCostWithoutSplitFlags;
double bestCostMtsFirstPassNoIsp;
double bestCostIsp;
@@ -286,9 +279,6 @@ protected:
#endif
bool m_fastDeltaQP;
static_vector<ComprCUCtx, ( MAX_CU_DEPTH << 2 )> m_ComprCUCtxList;
-#if ENABLE_SPLIT_PARALLELISM
- int m_runNextInParallel;
-#endif
InterSearch* m_pcInterSearch;
bool m_doPlt;
@@ -311,13 +301,6 @@ public:
virtual bool useModeResult ( const EncTestMode& encTestmode, CodingStructure*& tempCS, Partitioner& partitioner ) = 0;
virtual bool checkSkipOtherLfnst ( const EncTestMode& encTestmode, CodingStructure*& tempCS, Partitioner& partitioner ) = 0;
-#if ENABLE_SPLIT_PARALLELISM
- virtual void copyState ( const EncModeCtrl& other, const UnitArea& area );
- virtual int getNumParallelJobs ( const CodingStructure &cs, Partitioner& partitioner ) const { return 1; }
- virtual bool isParallelSplit ( const CodingStructure &cs, Partitioner& partitioner ) const { return false; }
- virtual bool parallelJobSelector ( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner ) const { return true; }
- void setParallelSplit ( bool val ) { m_runNextInParallel = val; }
-#endif
void init ( EncCfg *pCfg, RateCtrl *pRateCtrl, RdCost *pRdCost );
bool tryModeMaster ( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner );
@@ -395,13 +378,7 @@ struct SaveLoadStructSbt
class SaveLoadEncInfoSbt
{
protected:
-#if ENABLE_SPLIT_PARALLELISM
-public:
-#endif
void init( const Slice &slice );
-#if ENABLE_SPLIT_PARALLELISM
-protected:
-#endif
void create();
void destroy();
@@ -414,9 +391,6 @@ public:
void resetSaveloadSbt( int maxSbtSize );
uint16_t findBestSbt( const UnitArea& area, const uint32_t curPuSse );
bool saveBestSbt( const UnitArea& area, const uint32_t curPuSse, const uint8_t curPuSbt, const uint8_t curPuTrs );
-#if ENABLE_SPLIT_PARALLELISM
- void copyState(const SaveLoadEncInfoSbt& other);
-#endif
};
static const int MAX_STORED_CU_INFO_REFS = 4;
@@ -439,12 +413,6 @@ struct CodedCUInfo
double bestNonDCT2Cost;
bool relatedCuIsValid;
uint8_t bestISPIntraMode;
-
-#if ENABLE_SPLIT_PARALLELISM
-
- uint64_t
- temporalId;
-#endif
};
class CacheBlkInfoCtrl
@@ -460,21 +428,7 @@ protected:
void create ();
void destroy ();
-#if ENABLE_SPLIT_PARALLELISM
-public:
-#endif
void init ( const Slice &slice );
-#if ENABLE_SPLIT_PARALLELISM
-private:
- uint64_t
- m_currTemporalId;
-public:
- void tick () { m_currTemporalId++; CHECK( m_currTemporalId <= 0, "Problem with integer overflow!" ); }
- // mark the state of the blk as changed within the current temporal id
- void copyState( const CacheBlkInfoCtrl &other, const UnitArea& area );
-protected:
- void touch ( const UnitArea& area );
-#endif
CodedCUInfo& getBlkInfo( const UnitArea& area );
@@ -508,10 +462,6 @@ struct BestEncodingInfo
EncTestMode testMode;
int poc;
-
-#if ENABLE_SPLIT_PARALLELISM
- int64_t temporalId;
-#endif
};
class BestEncInfoCache
@@ -526,9 +476,6 @@ private:
bool *m_runType;
CodingStructure m_dummyCS;
XUCache m_dummyCache;
-#if ENABLE_SPLIT_PARALLELISM
- int64_t m_currTemporalId;
-#endif
protected:
@@ -537,19 +484,10 @@ protected:
bool setFromCs( const CodingStructure& cs, const Partitioner& partitioner );
bool isValid ( const CodingStructure &cs, const Partitioner &partitioner, int qp );
-
-#if ENABLE_SPLIT_PARALLELISM
- void touch ( const UnitArea& area );
-#endif
public:
BestEncInfoCache() : m_slice_bencinf( nullptr ), m_dummyCS( m_dummyCache.cuCache, m_dummyCache.puCache, m_dummyCache.tuCache ) {}
virtual ~BestEncInfoCache() {}
-
-#if ENABLE_SPLIT_PARALLELISM
- void copyState( const BestEncInfoCache &other, const UnitArea &area );
- void tick () { m_currTemporalId++; CHECK( m_currTemporalId <= 0, "Problem with integer overflow!" ); }
-#endif
void init ( const Slice &slice );
bool setCsFrom( CodingStructure& cs, EncTestMode& testMode, const Partitioner& partitioner ) const;
};
@@ -602,13 +540,6 @@ public:
virtual bool tryMode ( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner );
virtual bool useModeResult ( const EncTestMode& encTestmode, CodingStructure*& tempCS, Partitioner& partitioner );
-#if ENABLE_SPLIT_PARALLELISM
- virtual void copyState ( const EncModeCtrl& other, const UnitArea& area );
-
- virtual int getNumParallelJobs ( const CodingStructure &cs, Partitioner& partitioner ) const;
- virtual bool isParallelSplit ( const CodingStructure &cs, Partitioner& partitioner ) const;
- virtual bool parallelJobSelector( const EncTestMode& encTestmode, const CodingStructure &cs, Partitioner& partitioner ) const;
-#endif
virtual bool checkSkipOtherLfnst( const EncTestMode& encTestmode, CodingStructure*& tempCS, Partitioner& partitioner );
};
diff --git a/source/Lib/EncoderLib/EncReshape.cpp b/source/Lib/EncoderLib/EncReshape.cpp
index a4de78ad2..ac70e08d4 100644
--- a/source/Lib/EncoderLib/EncReshape.cpp
+++ b/source/Lib/EncoderLib/EncReshape.cpp
@@ -1396,45 +1396,5 @@ void EncReshape::adjustLmcsPivot()
}
}
}
-
-#if ENABLE_SPLIT_PARALLELISM
-void EncReshape::copyState(const EncReshape &other)
-{
- m_srcReshaped = other.m_srcReshaped;
- m_picWidth = other.m_picWidth;
- m_picHeight = other.m_picHeight;
- m_maxCUWidth = other.m_maxCUWidth;
- m_maxCUHeight = other.m_maxCUHeight;
- m_widthInCtus = other.m_widthInCtus;
- m_heightInCtus = other.m_heightInCtus;
- m_numCtuInFrame = other.m_numCtuInFrame;
- m_exceedSTD = other.m_exceedSTD;
- m_binImportance = other.m_binImportance;
- m_tcase = other.m_tcase;
- m_rateAdpMode = other.m_rateAdpMode;
- m_useAdpCW = other.m_useAdpCW;
- m_initCWAnalyze = other.m_initCWAnalyze;
- m_reshapeCW = other.m_reshapeCW;
- memcpy( m_cwLumaWeight, other.m_cwLumaWeight, sizeof( m_cwLumaWeight ) );
- m_chromaWeight = other.m_chromaWeight;
- m_chromaAdj = other.m_chromaAdj;
-
- m_sliceReshapeInfo = other.m_sliceReshapeInfo;
- m_CTUFlag = other.m_CTUFlag;
- m_recReshaped = other.m_recReshaped;
- m_invLUT = other.m_invLUT;
- m_fwdLUT = other.m_fwdLUT;
- m_chromaAdjHelpLUT = other.m_chromaAdjHelpLUT;
- m_binCW = other.m_binCW;
- m_initCW = other.m_initCW;
- m_reshape = other.m_reshape;
- m_reshapePivot = other.m_reshapePivot;
- m_inputPivot = other.m_inputPivot;
- m_fwdScaleCoef = other.m_fwdScaleCoef;
- m_invScaleCoef = other.m_invScaleCoef;
- m_lumaBD = other.m_lumaBD;
- m_reshapeLUTSize = other.m_reshapeLUTSize;
-}
-#endif
//
//! \}
diff --git a/source/Lib/EncoderLib/EncReshape.h b/source/Lib/EncoderLib/EncReshape.h
index a16cd63e6..72efd26fe 100644
--- a/source/Lib/EncoderLib/EncReshape.h
+++ b/source/Lib/EncoderLib/EncReshape.h
@@ -115,10 +115,6 @@ public:
Pel * getWeightTable() { return m_cwLumaWeight; }
double getCWeight() { return m_chromaWeight; }
void adjustLmcsPivot();
-
-#if ENABLE_SPLIT_PARALLELISM
- void copyState(const EncReshape& other);
-#endif
};// END CLASS DEFINITION EncReshape
//! \}
diff --git a/source/Lib/EncoderLib/EncSlice.cpp b/source/Lib/EncoderLib/EncSlice.cpp
index 637c50754..e3d48195d 100644
--- a/source/Lib/EncoderLib/EncSlice.cpp
+++ b/source/Lib/EncoderLib/EncSlice.cpp
@@ -1323,14 +1323,6 @@ void EncSlice::compressSlice( Picture* pcPic, const bool bCompressEntireSlice, c
m_CABACEstimator->initCtxModels( *pcSlice );
-#if ENABLE_SPLIT_PARALLELISM
- for( int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++ )
- {
- CABACWriter* cw = m_pcLib->getCABACEncoder( jId )->getCABACEstimator( pcSlice->getSPS() );
- cw->initCtxModels( *pcSlice );
- }
-
-#endif
m_pcCuEncoder->getModeCtrl()->setFastDeltaQp(bFastDeltaQP);
@@ -1376,13 +1368,6 @@ void EncSlice::compressSlice( Picture* pcPic, const bool bCompressEntireSlice, c
(m_pcCfg->getBaseQP() >= 38) || (m_pcCfg->getSourceWidth() <= 512 && m_pcCfg->getSourceHeight() <= 320), m_adaptedLumaQP))
{
m_CABACEstimator->initCtxModels (*pcSlice);
-#if ENABLE_SPLIT_PARALLELISM
- for (int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++)
- {
- CABACWriter* cw = m_pcLib->getCABACEncoder (jId)->getCABACEstimator (pcSlice->getSPS());
- cw->initCtxModels (*pcSlice);
- }
-#endif
pcPic->m_prevQP[0] = pcPic->m_prevQP[1] = pcSlice->getSliceQp();
if (pcSlice->getFirstCtuRsAddrInSlice() == 0)
{
@@ -1506,12 +1491,9 @@ void EncSlice::encodeCtus( Picture* pcPic, const bool bCompressEntireSlice, cons
const int iQPIndex = pcSlice->getSliceQpBase();
#endif
-#if ENABLE_SPLIT_PARALLELISM
- const int dataId = 0;
-#endif
- CABACWriter* pCABACWriter = pEncLib->getCABACEncoder( PARL_PARAM0( dataId ) )->getCABACEstimator( pcSlice->getSPS() );
- TrQuant* pTrQuant = pEncLib->getTrQuant( PARL_PARAM0( dataId ) );
- RdCost* pRdCost = pEncLib->getRdCost( PARL_PARAM0( dataId ) );
+ CABACWriter* pCABACWriter = pEncLib->getCABACEncoder()->getCABACEstimator( pcSlice->getSPS() );
+ TrQuant* pTrQuant = pEncLib->getTrQuant();
+ RdCost* pRdCost = pEncLib->getRdCost();
EncCfg* pCfg = pEncLib;
RateCtrl* pRateCtrl = pEncLib->getRateCtrl();
pRdCost->setLosslessRDCost(pcSlice->isLossless());
@@ -1703,13 +1685,6 @@ void EncSlice::encodeCtus( Picture* pcPic, const bool bCompressEntireSlice, cons
if (pcSlice->getSPS()->getUseLmcs())
{
m_pcCuEncoder->setDecCuReshaperInEncCU(m_pcLib->getReshaper(), pcSlice->getSPS()->getChromaFormatIdc());
-
-#if ENABLE_SPLIT_PARALLELISM
- for (int jId = 1; jId < m_pcLib->getNumCuEncStacks(); jId++)
- {
- m_pcLib->getCuEncoder(jId)->setDecCuReshaperInEncCU(m_pcLib->getReshaper(jId), pcSlice->getSPS()->getChromaFormatIdc());
- }
-#endif
}
if( !cs.slice->isIntra() && pCfg->getMCTSEncConstraint() )
{
@@ -1728,13 +1703,7 @@ void EncSlice::encodeCtus( Picture* pcPic, const bool bCompressEntireSlice, cons
pCABACWriter->coding_tree_unit( cs, ctuArea, prevQP, ctuRsAddr, true, true );
const int numberOfWrittenBits = int( pCABACWriter->getEstFracBits() >> SCALE_BITS );
-#if ENABLE_SPLIT_PARALLELISM
-#pragma omp critical
-#endif
pcSlice->setSliceBits( ( uint32_t ) ( pcSlice->getSliceBits() + numberOfWrittenBits ) );
-#if ENABLE_SPLIT_PARALLELISM
-#pragma omp critical
-#endif
// Store probabilities of first CTU in line into buffer - used only if wavefront-parallel-processing is enabled.
if( cs.pps->ctuIsTileColBd( ctuXPosInCtus ) && pEncLib->getEntropyCodingSyncEnabledFlag() )
diff --git a/source/Lib/EncoderLib/InterSearch.cpp b/source/Lib/EncoderLib/InterSearch.cpp
index eecee45c2..e625a05db 100644
--- a/source/Lib/EncoderLib/InterSearch.cpp
+++ b/source/Lib/EncoderLib/InterSearch.cpp
@@ -177,13 +177,6 @@ void InterSearch::setTempBuffers( CodingStructure ****pSplitCS, CodingStructure
m_pSaveCS = pSaveCS;
}
-#if ENABLE_SPLIT_PARALLELISM
-void InterSearch::copyState( const InterSearch& other )
-{
- memcpy( m_aaiAdaptSR, other.m_aaiAdaptSR, sizeof( m_aaiAdaptSR ) );
-}
-#endif
-
InterSearch::~InterSearch()
{
if (m_isInitialized)
diff --git a/source/Lib/EncoderLib/InterSearch.h b/source/Lib/EncoderLib/InterSearch.h
index 59e96de5f..51800ad05 100644
--- a/source/Lib/EncoderLib/InterSearch.h
+++ b/source/Lib/EncoderLib/InterSearch.h
@@ -206,9 +206,6 @@ public:
void setTempBuffers (CodingStructure ****pSlitCS, CodingStructure ****pFullCS, CodingStructure **pSaveCS );
void resetCtuRecord () { m_ctuRecord.clear(); }
-#if ENABLE_SPLIT_PARALLELISM
- void copyState ( const InterSearch& other );
-#endif
void setAffineModeSelected ( bool flag) { m_affineModeSelected = flag; }
void resetAffineMVList() { m_affMVListIdx = 0; m_affMVListSize = 0; }
void savePrevAffMVInfo(int idx, AffineMVInfo &tmpMVInfo, bool& isSaved)
diff --git a/source/Lib/Utilities/CMakeLists.txt b/source/Lib/Utilities/CMakeLists.txt
index 83b05cf5d..5eaaad366 100644
--- a/source/Lib/Utilities/CMakeLists.txt
+++ b/source/Lib/Utilities/CMakeLists.txt
@@ -28,26 +28,6 @@ if( SET_ENABLE_TRACING )
endif()
endif()
-if( OpenMP_FOUND )
- if( SET_ENABLE_SPLIT_PARALLELISM )
- if( ENABLE_SPLIT_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- endif()
- endif()
- if( SET_ENABLE_WPP_PARALLELISM )
- if( ENABLE_WPP_PARALLELISM )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=1 )
- else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
- endif()
- endif()
-else()
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_SPLIT_PARALLELISM=0 )
- target_compile_definitions( ${LIB_NAME} PUBLIC ENABLE_WPP_PARALLELISM=0 )
-endif()
-
target_include_directories( ${LIB_NAME} PUBLIC . .. )
target_link_libraries( ${LIB_NAME} CommonLib Threads::Threads )
--
GitLab