prepare for 1.2.0 release

CMakeLists.txt

@@ -10,10 +10,10 @@ if( NOT CMAKE_VERSION VERSION_LESS 3.12.0 )
 endif()
 
 # project name
-project( vvdec VERSION 1.1.2 )
+project( vvdec VERSION 1.2.0 )
 
-set( VVDEC_ENABLE_X86_SIMD TRUE )
-set( VVDEC_ENABLE_ARM_SIMD FALSE )
+set( VVDEC_ENABLE_X86_SIMD TRUE  CACHE BOOL "enable x86 intrinsics" )
+set( VVDEC_ENABLE_ARM_SIMD FALSE CACHE BOOL "enable arm intrinsics" )
 
 if( APPLE )
   if( DEFINED CMAKE_OSX_ARCHITECTURES )
@@ -49,6 +49,37 @@ if( VVDEC_ENABLE_ARM_SIMD )
   set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DTARGET_SIMD_ARM" )
 endif()
 
+if( ${CMAKE_SYSTEM_NAME} STREQUAL "Emscripten" )
+  set( VVDEC_TARGET_WASM TRUE )
+  # make Emscripten output a html wrapper
+  #set( CMAKE_EXECUTABLE_SUFFIX  ".html" )
+
+  add_compile_definitions( TARGET_SIMD_WASM )
+  add_compile_options( -msimd128 ) # currently breaks running in nodejs (firefox & chrome work)
+
+  add_compile_options( -pthread )
+
+  add_link_options(
+    --bind
+    -sWASM_BIGINT
+    -sTOTAL_MEMORY=2000MB # this should be enough for FullHD decoding
+    -sINVOKE_RUN=0        # don't call main() automatically
+    -sUSE_PTHREADS
+    -sPROXY_TO_PTHREAD
+    #-sPTHREAD_POOL_SIZE=20
+    -sMINIFY_HTML=0
+    -sMODULARIZE
+    -sEXPORT_NAME=CreateVVdeC
+    -sEXPORTED_RUNTIME_METHODS=[ccall,cwrap,getValue,setValue,_malloc,callMain,FS]
+    -sEXPORTED_FUNCTIONS=@${CMAKE_CURRENT_SOURCE_DIR}/source/Lib/vvdec/wasm_exported_functions.json
+#    --shell-file=${CMAKE_CURRENT_SOURCE_DIR}/web_player/shell.html
+#    --preload-file=${CMAKE_CURRENT_SOURCE_DIR}/ext/bitstreams/AMVR_B_HHI_3/AMVR_B_HHI_3.bit@in.bit
+#    --pre-js=${CMAKE_CURRENT_SOURCE_DIR}/pre.js
+  )
+endif()
+
+
+
 list( APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/modules" )
 message( STATUS "CMAKE_MODULE_PATH: updating module path to: ${CMAKE_MODULE_PATH}" )
 

README.md

@@ -74,6 +74,25 @@ The above call only tests the sequences that are know to work. To run a test ove
 
     make test-all
 
+## Build WebAssembly using Emscripten
+
+Install the Emscripten emsdk as documented on the website https://emscripten.org and activate the latest version (tested with 2.0.25).
+
+Ensure the environment variables are set up correctly (e.g. run `source emsdk_env.sh` in the current shell).
+
+Configure the VVdeC project:
+
+    emcmake cmake -B build/wasm
+
+And build the project:
+
+    cmake --build build/wasm
+
+The produced output consists of the `vvdecapp.wasm` binary and the corresponding javascript helpers (`vvdecapp.js`, `vvdecapp.worker.js`).
+
+When importing vvdecapp.js, the function `CreateVVdeC()` creates an instance of the vvdec module, which exposes a similar API to the one defined in `include/vvdec/vvdec.h`.
+
+
 # Contributing
 
 Feel free to contribute. To do so:

cmake/modules/check_missing_intrinsics.cmake

@@ -18,6 +18,7 @@ macro( check_intrinsic symbol_name code )
    endif()
 endmacro()
 
+check_intrinsic( _mm_storeu_si16 "int16_t a = 0; __m128i x = _mm_setzero_si128(); _mm_storeu_si16( &a, x );" )
 check_intrinsic( _mm_storeu_si32 "int32_t a = 0; __m128i x = _mm_setzero_si128(); _mm_storeu_si32( &a, x );" )
 check_intrinsic( _mm_storeu_si64 "int64_t a = 0; __m128i x = _mm_setzero_si128(); _mm_storeu_si64( &a, x );" )
 check_intrinsic( _mm_loadu_si32  "int32_t a = 0; __m128i x = _mm_setzero_si128(); x = _mm_loadu_si32( &a );" )

include/vvdec/vvdec.h

@@ -389,6 +389,7 @@ typedef struct vvdecParams
   vvdecRPRUpscaling   upscaleOutput;     // do internal upscaling of rpl pictures to dest. resolution ( default: 0 )
   vvdecLogLevel       logLevel;          // verbosity level
   bool                verifyPictureHash; // verify picture, if digest is available, true: check hash in SEI messages if available, false: ignore SEI message
+  bool                removePadding;     // copy output pictures to new buffer to remove padding (stride==width) ( default: false )
   vvdecSIMD_Extension simd;              // set specific simd optimization (default: max. availalbe)
 } vvdecParams;
 

source/App/vvdecapp/CMakeLists.txt

@@ -25,6 +25,9 @@ target_compile_options( ${EXE_NAME} PRIVATE $<$<OR:$<CXX_COMPILER_ID:Clang>,$<CX
 target_link_libraries( ${EXE_NAME} Threads::Threads vvdec )
 target_include_directories( ${EXE_NAME} PRIVATE  ../../Lib/libmd5 )
 
+if( ${CMAKE_SYSTEM_NAME} STREQUAL "Emscripten" )
+  set_target_properties( ${EXE_NAME} PROPERTIES LINK_DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../../Lib/vvdec/wasm_exported_functions.json;" )
+endif()
 
 # example: place header files in different folders
 source_group( "Header Files"   FILES ${INC_FILES} )

source/App/vvdecapp/vvdecapp.cpp

@@ -375,6 +375,7 @@ int main( int argc, char* argv[] )
           if( pcFrame->picAttributes )
           {
             uiBitrate += pcFrame->picAttributes->bits;
+            (void)uiBitrate;
           }
 
 #if 0 // just sample code to retrieve sei messages

source/Lib/CommonLib/AdaptiveLoopFilter.cpp

@@ -664,7 +664,7 @@ void AdaptiveLoopFilter::filterCTU( const CPelUnitBuf&     srcBuf,
     if( chType < MAX_NUM_CHANNEL_TYPE && toChannelType( compID ) != chType )
       continue;
 #if ALF_FIX
-    if( !ctuEnableFlag[compIdx] && !slice->getTileGroupCcAlfEnabledFlag( compIdx-1 ) )
+    if( !ctuEnableFlag[compIdx] && ( compIdx == 0 || !slice->getTileGroupCcAlfEnabledFlag( compIdx-1 ) ) )
 #else
     if( !ctuEnableFlag[compIdx] )
 #endif

source/Lib/CommonLib/Buffer.cpp

@@ -193,33 +193,6 @@ void copyBufferCore( const char *src, ptrdiff_t srcStride, char *dst, ptrdiff_t
   }
 }
 
-template<int padSize>
-void paddingCore(Pel *ptr, ptrdiff_t stride, int width, int height)
-{
-  /*left and right padding*/
-  Pel *ptrTemp1 = ptr;
-  Pel *ptrTemp2 = ptr + (width - 1);
-  ptrdiff_t offset = 0;
-  for (int i = 0; i < height; i++)
-  {
-    offset = stride * i;
-    for (int j = 1; j <= padSize; j++)
-    {
-      *(ptrTemp1 - j + offset) = *(ptrTemp1 + offset);
-      *(ptrTemp2 + j + offset) = *(ptrTemp2 + offset);
-    }
-  }
-  /*Top and Bottom padding*/
-  int numBytes = (width + padSize + padSize) * sizeof(Pel);
-  ptrTemp1 = (ptr - padSize);
-  ptrTemp2 = (ptr + (stride * (height - 1)) - padSize);
-  for (int i = 1; i <= padSize; i++)
-  {
-    memcpy(ptrTemp1 - (i * stride), (ptrTemp1), numBytes);
-    memcpy(ptrTemp2 + (i * stride), (ptrTemp2), numBytes);
-  }
-}
-
 void applyLutCore( Pel* ptr, ptrdiff_t ptrStride, int width, int height, const Pel* lut )
 {
   //    const auto rsp_sgnl_op  = [=, &dst]( int ADDR ){ dst[ADDR] = lut[dst[ADDR]]; };
@@ -236,7 +209,6 @@ void applyLutCore( Pel* ptr, ptrdiff_t ptrStride, int width, int height, const P
 #undef RSP_SGNL_INC
 }
 
-
 void fillN_CuCore( CodingUnit** ptr, ptrdiff_t ptrStride, int width, int height, CodingUnit* cuPtr )
 {
   if( width == ptrStride )
@@ -358,13 +330,13 @@ PelBufferOps::PelBufferOps()
   wghtAvg8 = addWeightedAvgCore<Pel>;
 
   copyBuffer = copyBufferCore;
-  padding2 = paddingCore<2>;
-  padding1 = paddingCore<1>;
 
   transpose4x4 = transpose4x4Core<Pel>;
   transpose8x8 = transpose8x8Core<Pel>;
 
   applyLut = applyLutCore;
+  rspFwd   = nullptr;
+  rspBcw   = nullptr;
 
   fillN_CU = fillN_CuCore;
 
@@ -430,12 +402,6 @@ void AreaBuf<Pel>::rescaleBuf( const AreaBuf<const Pel>& beforeScaling, Componen
                              isLuma( compID ) ? 1 : horCollocatedChromaFlag, isLuma( compID ) ? 1 : verCollocatedChromaFlag );
 }
 
-template<>
-void AreaBuf<Pel>::rspSignal( const Pel* lut )
-{
-  g_pelBufOP.applyLut( buf, stride, width, height, lut );
-}
-
 template<>
 void AreaBuf<Pel>::scaleSignal(const int scale, const ClpRng& clpRng)
 {
@@ -716,7 +682,11 @@ void PelStorage::create( const ChromaFormat _chromaFormat, const Size& _size, co
       totalWidth = ( ( totalWidth + _alignment - 1 ) / _alignment ) * _alignment;
     }
 
+#if ENABLE_SIMD_OPT_INTER
+    uint32_t area = totalWidth * totalHeight + 1; // +1 for the extra Pel overread in prefetchPad_SSE, in case reading from the very bottom right of the picture
+#else
     uint32_t area = totalWidth * totalHeight;
+#endif
     CHECK( !area, "Trying to create a buffer with zero area" );
 
     m_origSi[i] = Size{ totalWidth, totalHeight };

source/Lib/CommonLib/Buffer.h

@@ -64,12 +64,11 @@ THE POSSIBILITY OF SUCH DAMAGE.
 namespace vvdec
 {
 
-#if ENABLE_SIMD_OPT_BUFFER
 struct PelBufferOps
 {
   PelBufferOps();
 
-#ifdef TARGET_SIMD_X86
+#if defined( TARGET_SIMD_X86 ) && ENABLE_SIMD_OPT_BUFFER
   void initPelBufOpsX86();
   template<X86_VEXT vext>
   void _initPelBufOpsX86();
@@ -85,11 +84,11 @@ struct PelBufferOps
   void ( *wghtAvg4 )      ( const Pel* src0, ptrdiff_t src0Stride, const Pel* src1, ptrdiff_t src1Stride, Pel *dst, ptrdiff_t dstStride, int width, int height, int shift, int offset, int w0, int w1, const ClpRng& clpRng );
   void ( *wghtAvg8 )      ( const Pel* src0, ptrdiff_t src0Stride, const Pel* src1, ptrdiff_t src1Stride, Pel *dst, ptrdiff_t dstStride, int width, int height, int shift, int offset, int w0, int w1, const ClpRng& clpRng );
   void ( *copyBuffer )    ( const char*src,  ptrdiff_t srcStride,        char* dst, ptrdiff_t  dstStride,                                int width, int height );
-  void ( *padding1 )      (       Pel *dst,  ptrdiff_t stride,                                                                           int width, int height );
-  void ( *padding2 )      (       Pel *dst,  ptrdiff_t stride,                                                                           int width, int height );
   void ( *transpose4x4 )  ( const Pel* src,  ptrdiff_t srcStride, Pel* dst, ptrdiff_t dstStride );
   void ( *transpose8x8 )  ( const Pel* src,  ptrdiff_t srcStride, Pel* dst, ptrdiff_t dstStride );
   void ( *applyLut )      (       Pel* ptr,  ptrdiff_t ptrStride, int width, int height, const Pel* lut );
+  void ( *rspFwd )        (       Pel* ptr,  ptrdiff_t ptrStride, int width, int height, const int bd,                   const Pel OrgCW,  const Pel* LmcsPivot, const Pel* ScaleCoeff, const Pel* InputPivot );
+  void ( *rspBcw )        (       Pel* ptr,  ptrdiff_t ptrStride, int width, int height, const int bd, const int minBin, const int maxBin, const Pel* LmcsPivot, const Pel* InvScCoeff, const Pel* InputPivot );
   void ( *fillN_CU )      (       CodingUnit** ptr, ptrdiff_t ptrStride, int width, int height, CodingUnit* cuPtr );
 
   void (*sampleRateConv)  ( const std::pair<int, int> scalingRatio, const std::pair<int, int> compScale,
@@ -99,7 +98,6 @@ struct PelBufferOps
                             const int afterScaleLeftOffset, const int afterScaleTopOffset,
                             const int bitDepth, const bool useLumaFilter, const bool horCollocatedPositionFlag, const bool verCollocatedPositionFlag );
 };
-#endif
 
 extern PelBufferOps g_pelBufOP;
 
@@ -146,8 +144,7 @@ struct AreaBuf : public Size
 
   void transposedFrom       ( const AreaBuf<const T> &other );
 
-  void rspSignal            ( const Pel *lut );
-  void scaleSignal          ( const int scale, const ClpRng& clpRng);
+  void scaleSignal          ( const int scale, const ClpRng& clpRng );
 
   void rescaleBuf           ( const AreaBuf<const T>& beforeScaling, const ComponentID compID, const std::pair<int, int> scalingRatio, const Window& confBefore, const Window& confAfter, const ChromaFormat chromaFormatIDC, const BitDepths& bitDepths, const bool horCollocatedChromaFlag = false, const bool verCollocatedChromaFlag = false );
 

source/Lib/CommonLib/CodingStructure.cpp

@@ -166,18 +166,22 @@ CodingUnit& CodingStructure::addCU( const UnitArea &unit, const ChannelType chTy
 
     if( i )
     {
-      m_predBufOffset += ( cuArea << 1 );
+      m_predBufOffset += (cuArea << 1);
     }
     else
     {
       m_predBufOffset += cuArea;
     }
 
     const ptrdiff_t  stride = ptrdiff_t( 1 ) << m_ctuWidthLog2[i];
-    const Area      &_blk   = cu-> blocks[i];
+    const Area&      _blk   = cu->blocks[i];
     const UnitScale  scale  = unitScale[i];
+    const int        sclX   = scale.scaleHor( _blk.x );
+    const int        sclY   = scale.scaleVer( _blk.y );
+    const int        sclW   = scale.scaleHor( _blk.width );
+    const int        sclH   = scale.scaleVer( _blk.height );
 
-    g_pelBufOP.fillN_CU( ctuData.cuPtr[i] + inCtuPos( _blk, ChannelType( i ) ), stride, scale.scaleHor( _blk.width ), scale.scaleVer( _blk.height ), cu );
+    g_pelBufOP.fillN_CU( ctuData.cuPtr[i] + ( sclX & m_ctuSizeMask[i] ) + ( ( sclY & m_ctuSizeMask[i] ) << m_ctuWidthLog2[i] ), stride, sclW, sclH, cu );
 
     if( i == chType )
     {
@@ -488,18 +492,27 @@ const CodingUnit* CodingStructure::getCURestricted( const Position &pos, const P
 {
   const int yshift     = pcv->maxCUWidthLog2 - getChannelTypeScaleY( _chType, area.chromaFormat );
   const int ydiff      = ( pos.y >> yshift ) - ( curPos.y >> yshift ); // ( a <= b ) ==> a - b <= 0
-  const CodingUnit* cu = ydiff <= 0 ? getCU( pos, _chType ) : nullptr;
   const int xshift     = pcv->maxCUWidthLog2 - getChannelTypeScaleX( _chType, area.chromaFormat );
   const int xdiff      = ( pos.x >> xshift ) - ( curPos.x >> xshift );
   const bool sameCTU   = !ydiff && !xdiff;
 
-  if( cu && ( sameCTU || ( cu->slice->getIndependentSliceIdx() == curSliceIdx && cu->tileIdx == curTileIdx ) ) )
+  const CodingUnit* cu = nullptr;
+
+  if( sameCTU )
   {
-    if( xdiff > 0 && sps->getEntropyCodingSyncEnabledFlag() )
-    {
-      return nullptr;
-    }
+    return getCU( pos, _chType );
+  }
+  else if( ydiff > 0 || xdiff > ( 1 - sps->getEntropyCodingSyncEnabledFlag() ) )
+  {
+    return nullptr;
+  }
+  else
+  {
+    cu = getCU( pos, _chType );
+  }
 
+  if( cu && cu->slice->getIndependentSliceIdx() == curSliceIdx && cu->tileIdx == curTileIdx )
+  {
     return cu;
   }
   else

source/Lib/CommonLib/CommonDef.h

@@ -56,10 +56,21 @@ THE POSSIBILITY OF SUCH DAMAGE.
 #include <iostream>
 #include <iomanip>
 #include <limits>
+#include <cmath>   // needed for std::log2()
 
 #include <functional>
 #include <mutex>
 
+#if defined( TARGET_SIMD_X86 )
+# ifdef _WIN32
+#  include <intrin.h>
+# elif defined( __GNUC__ ) && !defined( TARGET_SIMD_WASM )
+#  include <x86intrin.h>
+# elif defined( __GNUC__ )
+#  include <immintrin.h>
+# endif
+#endif // TARGET_SIMD_X86
+
 namespace vvdec
 {
 
@@ -346,7 +357,7 @@ static const int BIO_TEMP_BUFFER_SIZE         =                     (MAX_BDOF_AP
 static const int PROF_BORDER_EXT_W            =                     1;
 static const int PROF_BORDER_EXT_H            =                     1;
 static const int BCW_NUM =                                          5; ///< the number of weight options
-static const int BCW_DEFAULT =                                      ((uint8_t)(BCW_NUM >> 1)); ///< Default weighting index representing for w=0.5
+static const int BCW_DEFAULT =                                      0; ///< Default weighting index representing for w=0.5, in the internal domain
 static const int BCW_SIZE_CONSTRAINT =                            256; ///< disabling Bcw if cu size is smaller than 256
 static const int MAX_NUM_HMVP_CANDS =                              (MRG_MAX_NUM_CANDS-1); ///< maximum number of HMVP candidates to be stored and used in merge list
 static const int MAX_NUM_HMVP_AVMPCANDS =                          4; ///< maximum number of HMVP candidates to be used in AMVP list
@@ -558,42 +569,40 @@ template <typename ValueType> inline ValueType rightShift      (const ValueType
 template <typename ValueType> inline ValueType leftShift_round (const ValueType value, const int shift) { return (shift >= 0) ? ( value                                  << shift) : ((value + (ValueType(1) << (-shift - 1))) >> -shift); }
 template <typename ValueType> inline ValueType rightShift_round(const ValueType value, const int shift) { return (shift >= 0) ? ((value + (ValueType(1) << (shift - 1))) >> shift) : ( value                                   << -shift); }
 
-#ifdef TARGET_SIMD_X86
-#ifdef _WIN32
-}
-# include <intrin.h>
-namespace vvdec {
-static inline unsigned long _bit_scan_reverse( long a )
+#if defined( _WIN32 ) && defined( TARGET_SIMD_X86 )
+static inline unsigned int bit_scan_reverse( int a )
 {
   unsigned long idx = 0;
   _BitScanReverse( &idx, a );
   return idx;
 }
-#else
+#elif defined( __GNUC__ ) && defined( TARGET_SIMD_X86 ) && !defined( TARGET_SIMD_WASM )
+static inline unsigned int bit_scan_reverse( int a )
+{
+  return _bit_scan_reverse( a );
+}
+#elif defined( __GNUC__ )
+static inline unsigned int bit_scan_reverse( int a )
+{
+  return __builtin_clz( a ) ^ ( 8 * sizeof( a ) - 1 );
 }
-# include <x86intrin.h>
-namespace vvdec {
 #endif
 
-#endif
-#if ENABLE_SIMD_LOG2 && defined( TARGET_SIMD_X86 )
+#if ENABLE_SIMD_LOG2
 static inline int getLog2( long val )
 {
-  return _bit_scan_reverse( val );
+  return bit_scan_reverse( val );
 }
 #else
-}
-#include <cmath>
-namespace vvdec {
 extern int8_t g_aucLog2[MAX_CU_SIZE + 1];
 static inline int getLog2( long val )
 {
   CHECKD( g_aucLog2[2] != 1, "g_aucLog2[] has not been initialized yet." );
-  if( val > 0 && val < (int)sizeof(g_aucLog2) )
+  if( val > 0 && val < (int) sizeof( g_aucLog2 ) )
   {
     return g_aucLog2[val];
   }
-  return std::log2(val);
+  return std::log2( val );
 }
 #endif