15%+ speedup, RC improvements for high rates, various changes towards…

… 1.6.0 (#185)
fraunhoferhhi · Aug 25, 2022 · 3651187 · 3651187
1 parent 5765352
commit 3651187
Show file tree

Hide file tree

Showing 66 changed files with 3,072 additions and 2,747 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -10,10 +10,10 @@ if( NOT CMAKE_VERSION VERSION_LESS 3.13.0 )
 endif()
 
 # project name
-project( vvenc VERSION 1.5.0 )
+project( vvenc VERSION 1.6.0 )
 
 # set alternative version numbering for release candidates
-#set( PROJECT_VERSION_RC rc1 )
+set( PROJECT_VERSION_RC rc1 )
 if( PROJECT_VERSION_RC )
     set( PROJECT_VERSION "${PROJECT_VERSION}-rc${PROJECT_VERSION_RC}" )
 endif()
@@ -200,6 +200,11 @@ if( VVENC_ENABLE_LINK_TIME_OPT )
   set( CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELEASE ON )
   set( CMAKE_INTERPROCEDURAL_OPTIMIZATION_RELWITHDEBINFO ON )
   set( CMAKE_INTERPROCEDURAL_OPTIMIZATION_MINSIZEREL ON )
+
+  if( CMAKE_CXX_COMPILER_ID STREQUAL "GNU" )
+    # enable parallel link-time optimization for GCC
+    add_link_options( $<$<OR:$<CONFIG:Release>,$<CONFIG:RelWithDebInfo>,$<CONFIG:MinSizeRel>>:-flto=auto> )
+  endif()
 endif()
 
 if( VVCEncoderLib_ENABLE_ITT )

diff --git a/cfg/experimental/lowdelay_fast.cfg b/cfg/experimental/lowdelay_fast.cfg
@@ -31,7 +31,7 @@ SearchRange                   : 64          # (0: Search range is a Full frame)
 #MinSearchWindow               : 96          # Minimum motion search window size for the adaptive window ME
 BipredSearchRange             : 1           # Search range for bi-prediction refinement
 HadamardME                    : 1           # Use of hadamard measure for fractional ME
-FEN                           : 1           # Fast encoder decision
+FEN                           : 3           # Fast encoder decision
 FDM                           : 1           # Fast Decision for Merge RD cost
 
 #======== Quantization =============
@@ -67,15 +67,15 @@ DualITree                     : 1      # separate partitioning of luma and chrom
 MinQTLumaISlice               : 4
 MinQTChromaISliceInChromaSamples : 2   # minimum QT size in chroma samples for chroma separate tree
 MinQTNonISlice                : 4
-MaxMTTHierarchyDepth          : 0
-MaxMTTHierarchyDepthISliceL   : 1
-MaxMTTHierarchyDepthISliceC   : 1
+MaxMTTDepth                   : 0
+MaxMTTDepthI                  : 1
 MaxNumMergeCand               : 6      # Maximum number of merge candidates
 
 # Tools configuration
 Affine                        : 2      # Affine prediction: 0: disabled, 1: vtm, 2: fast
 ALF                           : 1      # Adpative Loop Filter: 0: disabled, 1: enabled
 ALFSpeed                      : 1      # ALF speed (skip filtering of non-referenced frames) [0-1]
+ALFUnitSize                   : 128    # ALF processing size 
 AllowDisFracMMVD              : 1      # Disable fractional MVD in MMVD mode adaptively
 BCW                           : 0      # Enable Generalized Bi-prediction(Bcw) 0: disabled, 1: enabled, 2: fast
 BIO                           : 1      # Bi-directional optical flow: 0: disabled, 1: enabled
@@ -95,7 +95,7 @@ LMCSOffset                    : 1      # chroma residual scaling offset
 LMCSSignalType                : 0      # Input signal type: 0:SDR, 1:HDR-PQ, 2:HDR-HLG
 LMCSUpdateCtrl                : 2      # LMCS model update control: 0:RA, 1:AI, 2:LDB/LDP
 MCTF                          : 2      # GOP based temporal filter (0:off, 1:filter all frames, 2:use SCC detection to disable for screen coded content)
-MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 3:fastest operation)
+MCTFSpeed                     : 3      # MCTF Fast Mode (0:best quality .. 4:fastest operation)
 MCTFFutureReference           : 0      # Enable referencing of future frames in the GOP based temporal filter. This is typically disabled for Low Delay configurations
 MIP                           : 0      # Matrix-based intra prediction: 0: disabled, 1: enabled
 FastMIP                       : 0      # Fast encoder search for MIP 0: disable, 1:vtm, 2-4: fast
@@ -138,5 +138,6 @@ IntraEstDecBit                : 2      # Test every eight intra directional mode
 NumIntraModesFullRD           : -1     # Number modes for full RD intra search
 ReduceIntraChromaModesFullRD  : 1      # Reduce modes for chroma full RD intra search
 FastTTSplit                   : 0      # Fast method for TT split
+ReduceFilterME                : 2      # Use reduced filter taps for subpel motion estimation (2: 4-tap, 1: 6-tap)
 
 
diff --git a/cfg/experimental/lowdelay_faster.cfg b/cfg/experimental/lowdelay_faster.cfg
@@ -31,7 +31,7 @@ SearchRange                   : 64          # (0: Search range is a Full frame)
 #MinSearchWindow               : 96          # Minimum motion search window size for the adaptive window ME
 BipredSearchRange             : 1           # Search range for bi-prediction refinement
 HadamardME                    : 1           # Use of hadamard measure for fractional ME
-FEN                           : 1           # Fast encoder decision
+FEN                           : 3           # Fast encoder decision
 FDM                           : 1           # Fast Decision for Merge RD cost
 
 #======== Quantization =============
@@ -67,15 +67,15 @@ DualITree                     : 1      # separate partitioning of luma and chrom
 MinQTLumaISlice               : 4
 MinQTChromaISliceInChromaSamples : 2   # minimum QT size in chroma samples for chroma separate tree
 MinQTNonISlice                : 4
-MaxMTTHierarchyDepth          : 0
-MaxMTTHierarchyDepthISliceL   : 0
-MaxMTTHierarchyDepthISliceC   : 0
+MaxMTTDepth                   : 0
+MaxMTTDepthI                  : 0
 MaxNumMergeCand               : 6      # Maximum number of merge candidates
 
 # Tools configuration
 Affine                        : 0      # Affine prediction: 0: disabled, 1: vtm, 2: fast
 ALF                           : 1      # Adpative Loop Filter: 0: disabled, 1: enabled
 ALFSpeed                      : 1      # ALF speed (skip filtering of non-referenced frames) [0-1]
+ALFUnitSize                   : 128    # ALF processing size 
 AllowDisFracMMVD              : 0      # Disable fractional MVD in MMVD mode adaptively
 BCW                           : 0      # Enable Generalized Bi-prediction(Bcw) 0: disabled, 1: enabled, 2: fast
 BIO                           : 0      # Bi-directional optical flow: 0: disabled, 1: enabled
@@ -95,7 +95,7 @@ LMCSOffset                    : 1      # chroma residual scaling offset
 LMCSSignalType                : 0      # Input signal type: 0:SDR, 1:HDR-PQ, 2:HDR-HLG
 LMCSUpdateCtrl                : 2      # LMCS model update control: 0:RA, 1:AI, 2:LDB/LDP
 MCTF                          : 2      # GOP based temporal filter (0:off, 1:filter all frames, 2:use SCC detection to disable for screen coded content)
-MCTFSpeed                     : 3      # MCTF Fast Mode (0:best quality .. 3:fastest operation)
+MCTFSpeed                     : 4      # MCTF Fast Mode (0:best quality .. 4:fastest operation)
 MCTFFutureReference           : 0      # Enable referencing of future frames in the GOP based temporal filter. This is typically disabled for Low Delay configurations
 MIP                           : 0      # Matrix-based intra prediction: 0: disabled, 1: enabled
 FastMIP                       : 0      # Fast encoder search for MIP 0: disable, 1:vtm, 2-4: fast
@@ -138,5 +138,6 @@ IntraEstDecBit                : 3      # Test every eight intra directional mode
 NumIntraModesFullRD           : 1      # Number modes for full RD intra search
 ReduceIntraChromaModesFullRD  : 1      # Reduce modes for chroma full RD intra search
 FastTTSplit                   : 0      # Fast method for TT split
+ReduceFilterME                : 2      # Use reduced filter taps for subpel motion estimation (2: 4-tap, 1: 6-tap)
 
 
diff --git a/cfg/experimental/lowdelay_medium.cfg b/cfg/experimental/lowdelay_medium.cfg
@@ -31,7 +31,7 @@ SearchRange                   : 64          # (0: Search range is a Full frame)
 #MinSearchWindow               : 96          # Minimum motion search window size for the adaptive window ME
 BipredSearchRange             : 4           # Search range for bi-prediction refinement
 HadamardME                    : 1           # Use of hadamard measure for fractional ME
-FEN                           : 1           # Fast encoder decision
+FEN                           : 3           # Fast encoder decision
 FDM                           : 1           # Fast Decision for Merge RD cost
 
 #======== Quantization =============
@@ -67,9 +67,8 @@ DualITree                     : 1      # separate partitioning of luma and chrom
 MinQTLumaISlice               : 8
 MinQTChromaISliceInChromaSamples : 4   # minimum QT size in chroma samples for chroma separate tree
 MinQTNonISlice                : 8
-MaxMTTHierarchyDepth          : 1
-MaxMTTHierarchyDepthISliceL   : 2
-MaxMTTHierarchyDepthISliceC   : 2
+MaxMTTDepth                   : 1
+MaxMTTDepthI                  : 2
 MaxNumMergeCand               : 6      # Maximum number of merge candidates
 
 # Tools configuration
@@ -95,7 +94,7 @@ LMCSOffset                    : 1      # chroma residual scaling offset
 LMCSSignalType                : 0      # Input signal type: 0:SDR, 1:HDR-PQ, 2:HDR-HLG
 LMCSUpdateCtrl                : 2      # LMCS model update control: 0:RA, 1:AI, 2:LDB/LDP
 MCTF                          : 2      # GOP based temporal filter (0:off, 1:filter all frames, 2:use SCC detection to disable for screen coded content)
-MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 3:fastest operation)
+MCTFSpeed                     : 2      # MCTF Fast Mode (0:best quality .. 4:fastest operation)
 MCTFFutureReference           : 0      # Enable referencing of future frames in the GOP based temporal filter. This is typically disabled for Low Delay configurations
 MIP                           : 1      # Matrix-based intra prediction: 0: disabled, 1: enabled
 FastMIP                       : 3      # Fast encoder search for MIP 0: disable, 1:vtm, 2-4: fast
@@ -130,13 +129,14 @@ FastHAD                       : 0      # Use fast sub-sampled hadamard for squar
 FastMrg                       : 2      # Fast methods for inter merge: 0: disabled, 1: vtm, 2: fast
 FastLocalDualTreeMode         : 2      # Fast intra pass coding for local dual-tree in intra coding region: 0: disable, 1: use threshold, 2: one intra mode only
 FastSubPel                    : 1      # Fast sub-pel ME: 0: disabled, 1: enabled
-FastIntraTools                : 0      # Speedup intra tools: LFNST, ISP, MTS
+FastIntraTools                : 1      # Speedup intra tools: LFNST, ISP, MTS
 FastInferMerge                : 0      # Fast method to skip Inter/Intra modes. 0: off, [1..4] speedups
 ECU                           : 0      # Early CU setting
 IntegerET                     : 0      # Enable early termination for integer motion search
 IntraEstDecBit                : 2      # Test every fourth intra directional mode in the first pass
 NumIntraModesFullRD           : -1     # Number modes for full RD intra search
 ReduceIntraChromaModesFullRD  : 1      # Reduce modes for chroma full RD intra search
 FastTTSplit                   : 0      # Fast method for TT split
+ReduceFilterME                : 2      # Use reduced filter taps for subpel motion estimation (2: 4-tap, 1: 6-tap)
 
 
diff --git a/cfg/experimental/lowdelay_slow.cfg b/cfg/experimental/lowdelay_slow.cfg
@@ -31,7 +31,7 @@ SearchRange                   : 64          # (0: Search range is a Full frame)
 #MinSearchWindow               : 96          # Minimum motion search window size for the adaptive window ME
 BipredSearchRange             : 4           # Search range for bi-prediction refinement
 HadamardME                    : 1           # Use of hadamard measure for fractional ME
-FEN                           : 1           # Fast encoder decision
+FEN                           : 3           # Fast encoder decision
 FDM                           : 1           # Fast Decision for Merge RD cost
 
 #======== Quantization =============
@@ -67,9 +67,8 @@ DualITree                     : 1      # separate partitioning of luma and chrom
 MinQTLumaISlice               : 8
 MinQTChromaISliceInChromaSamples : 4   # minimum QT size in chroma samples for chroma separate tree
 MinQTNonISlice                : 8
-MaxMTTHierarchyDepth          : 2
-MaxMTTHierarchyDepthISliceL   : 3
-MaxMTTHierarchyDepthISliceC   : 3
+MaxMTTDepth                   : 2
+MaxMTTDepthI                  : 3
 MaxNumMergeCand               : 6      # Maximum number of merge candidates
 
 # Tools configuration
@@ -95,7 +94,7 @@ LMCSOffset                    : 1      # chroma residual scaling offset
 LMCSSignalType                : 0      # Input signal type: 0:SDR, 1:HDR-PQ, 2:HDR-HLG
 LMCSUpdateCtrl                : 2      # LMCS model update control: 0:RA, 1:AI, 2:LDB/LDP
 MCTF                          : 2      # GOP based temporal filter (0:off, 1:filter all frames, 2:use SCC detection to disable for screen coded content)
-MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 3:fastest operation)
+MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 4:fastest operation)
 MCTFFutureReference           : 0      # Enable referencing of future frames in the GOP based temporal filter. This is typically disabled for Low Delay configurations
 MIP                           : 1      # Matrix-based intra prediction: 0: disabled, 1: enabled
 FastMIP                       : 0      # Fast encoder search for MIP 0: disable, 1:vtm, 2-4: fast
@@ -138,5 +137,6 @@ IntraEstDecBit                : 1      # Test every second intra directional mod
 NumIntraModesFullRD           : -1     # Number modes for full RD intra search
 ReduceIntraChromaModesFullRD  : 0      # Reduce modes for chroma full RD intra search
 FastTTSplit                   : 5      # Fast method for TT split
+ReduceFilterME                : 0      # Use reduced filter taps for subpel motion estimation (2: 4-tap, 1: 6-tap)
 
 
diff --git a/cfg/experimental/lowdelay_slower.cfg b/cfg/experimental/lowdelay_slower.cfg
@@ -31,7 +31,7 @@ SearchRange                   : 64          # (0: Search range is a Full frame)
 #MinSearchWindow               : 96          # Minimum motion search window size for the adaptive window ME
 BipredSearchRange             : 4           # Search range for bi-prediction refinement
 HadamardME                    : 1           # Use of hadamard measure for fractional ME
-FEN                           : 1           # Fast encoder decision
+FEN                           : 3           # Fast encoder decision
 FDM                           : 1           # Fast Decision for Merge RD cost
 
 #======== Quantization =============
@@ -67,9 +67,8 @@ DualITree                     : 1      # separate partitioning of luma and chrom
 MinQTLumaISlice               : 8
 MinQTChromaISliceInChromaSamples : 4   # minimum QT size in chroma samples for chroma separate tree
 MinQTNonISlice                : 8
-MaxMTTHierarchyDepth          : 3
-MaxMTTHierarchyDepthISliceL   : 3
-MaxMTTHierarchyDepthISliceC   : 3
+MaxMTTDepth                   : 3
+MaxMTTDepthI                  : 3
 MaxNumMergeCand               : 6      # Maximum number of merge candidates
 
 # Tools configuration
@@ -95,7 +94,7 @@ LMCSOffset                    : 1      # chroma residual scaling offset
 LMCSSignalType                : 0      # Input signal type: 0:SDR, 1:HDR-PQ, 2:HDR-HLG
 LMCSUpdateCtrl                : 2      # LMCS model update control: 0:RA, 1:AI, 2:LDB/LDP
 MCTF                          : 2      # GOP based temporal filter (0:off, 1:filter all frames, 2:use SCC detection to disable for screen coded content)
-MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 3:fastest operation)
+MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 4:fastest operation)
 MCTFFutureReference           : 0      # Enable referencing of future frames in the GOP based temporal filter. This is typically disabled for Low Delay configurations
 MIP                           : 1      # Matrix-based intra prediction: 0: disabled, 1: enabled
 FastMIP                       : 0      # Fast encoder search for MIP 0: disable, 1:vtm, 2-4: fast
@@ -138,5 +137,6 @@ IntraEstDecBit                : 1      # Test every second intra directional mod
 NumIntraModesFullRD           : -1     # Number modes for full RD intra search
 ReduceIntraChromaModesFullRD  : 0      # Reduce modes for chroma full RD intra search
 FastTTSplit                   : 1      # Fast method for TT split
+ReduceFilterME                : 0      # Use reduced filter taps for subpel motion estimation (2: 4-tap, 1: 6-tap)
 
 
diff --git a/cfg/randomaccess_fast.cfg b/cfg/randomaccess_fast.cfg
@@ -53,7 +53,7 @@ ASR                           : 1           # Adaptive motion search range
 MinSearchWindow               : 96          # Minimum motion search window size for the adaptive window ME
 BipredSearchRange             : 1           # Search range for bi-prediction refinement
 HadamardME                    : 1           # Use of hadamard measure for fractional ME
-FEN                           : 1           # Fast encoder decision
+FEN                           : 3           # Fast encoder decision
 FDM                           : 1           # Fast Decision for Merge RD cost
 
 #======== Quantization =============
@@ -89,15 +89,15 @@ DualITree                     : 1      # separate partitioning of luma and chrom
 MinQTLumaISlice               : 4
 MinQTChromaISliceInChromaSamples : 2   # minimum QT size in chroma samples for chroma separate tree
 MinQTNonISlice                : 4
-MaxMTTHierarchyDepth          : 0
-MaxMTTHierarchyDepthISliceL   : 1
-MaxMTTHierarchyDepthISliceC   : 1
+MaxMTTDepth                   : 0
+MaxMTTDepthI                  : 1
 MaxNumMergeCand               : 6      # Maximum number of merge candidates
 
 # Tools configuration
 Affine                        : 2      # Affine prediction: 0: disabled, 1: vtm, 2: fast
 ALF                           : 1      # Adpative Loop Filter: 0: disabled, 1: enabled
 ALFSpeed                      : 1      # ALF speed (skip filtering of non-referenced frames) [0-1]
+ALFUnitSize                   : 128    # ALF processing size 
 AllowDisFracMMVD              : 1      # Disable fractional MVD in MMVD mode adaptively
 BCW                           : 0      # Enable Generalized Bi-prediction(Bcw) 0: disabled, 1: enabled, 2: fast
 BIO                           : 1      # Bi-directional optical flow: 0: disabled, 1: enabled
@@ -117,7 +117,7 @@ LMCSOffset                    : 6      # chroma residual scaling offset
 LMCSSignalType                : 0      # Input signal type: 0:SDR, 1:HDR-PQ, 2:HDR-HLG
 LMCSUpdateCtrl                : 0      # LMCS model update control: 0:RA, 1:AI, 2:LDB/LDP
 MCTF                          : 2      # GOP based temporal filter (0:off, 1:filter all frames, 2:use SCC detection to disable for screen coded content)
-MCTFSpeed                     : 1      # MCTF Fast Mode (0:best quality .. 3:fastest operation)
+MCTFSpeed                     : 3      # MCTF Fast Mode (0:best quality .. 4:fastest operation)
 MCTFFutureReference           : 1      # Enable referencing of future frames in the GOP based temporal filter. This is typically disabled for Low Delay configurations
 MIP                           : 0      # Matrix-based intra prediction: 0: disabled, 1: enabled
 FastMIP                       : 0      # Fast encoder search for MIP 0: disable, 1:vtm, 2-4: fast
@@ -160,5 +160,6 @@ IntraEstDecBit                : 2      # Test every eight intra directional mode
 NumIntraModesFullRD           : -1     # Number modes for full RD intra search
 ReduceIntraChromaModesFullRD  : 1      # Reduce modes for chroma full RD intra search
 FastTTSplit                   : 0      # Fast method for TT split
+ReduceFilterME                : 2      # Use reduced filter taps for subpel motion estimation (2: 4-tap, 1: 6-tap)