Fix asymmetric chroma plane stride handling in encoder preprocessing

When preprocessing incoming I420 source frames, WelsMoveMemoryWrapper
previously read only the U chroma plane stride (iStride[1]) and used it as a
shared stride for copying both the U and V chroma planes. If an incoming
frame has distinct, asymmetric strides for the U and V chroma planes (for
example, when the U plane stride is larger than the V plane stride), copying
the V plane using the U plane's stride advances the source pointer incorrectly.

This change ensures correct memory advancement for each chroma plane by:
- Modifying WelsMoveMemory_c to accept separate destination and source
  strides for the U and V chroma planes (iDstStrideU, iDstStrideV,
  iSrcStrideU, iSrcStrideV).
- Updating WelsMoveMemoryWrapper to read and pass iStride[1] and iStride[2]
  separately from the input SSourcePicture.
- Updating internal callers (DownsamplePadding and Padding) to pass their
  respective U and V line sizes cleanly.
- Adding a unit test (CustomChromaPlaneStrides) to verify that distinct,
  asymmetric chroma plane strides are handled correctly.

Author: Erik Språng

codec/encoder/core/src/wels_preprocess.cpp

@@ -56,9 +56,11 @@ namespace WelsEnc {
 int32_t WelsInitScaledPic (SWelsSvcCodingParam* pParam,  Scaled_Picture*  pScaledPic, CMemoryAlign* pMemoryAlign);
 bool  JudgeNeedOfScaling (SWelsSvcCodingParam* pParam, Scaled_Picture* pScaledPic);
 void    FreeScaledPic (Scaled_Picture*  pScaledPic, CMemoryAlign* pMemoryAlign);
-void  WelsMoveMemory_c (uint8_t* pDstY, uint8_t* pDstU, uint8_t* pDstV,  int32_t iDstStrideY, int32_t iDstStrideUV,
-                        uint8_t* pSrcY, uint8_t* pSrcU, uint8_t* pSrcV, int32_t iSrcStrideY, int32_t iSrcStrideUV, int32_t iWidth,
-                        int32_t iHeight);
+void WelsMoveMemory_c(uint8_t* pDstY, uint8_t* pDstU, uint8_t* pDstV,
+                      int32_t iDstStrideY, int32_t iDstStrideU,
+                      int32_t iDstStrideV, uint8_t* pSrcY, uint8_t* pSrcU,
+                      uint8_t* pSrcV, int32_t iSrcStrideY, int32_t iSrcStrideU,
+                      int32_t iSrcStrideV, int32_t iWidth, int32_t iHeight);
 
 //******* table definition ***********************************************************************//
 const uint8_t g_kuiRefTemporalIdx[MAX_TEMPORAL_LEVEL][MAX_GOP_SIZE] = {
@@ -656,9 +658,11 @@ int32_t CWelsPreProcess::DownsamplePadding (SPicture* pSrc, SPicture* pDstPic,
     if (iSrcWidth != iShrinkWidth || iSrcHeight != iShrinkHeight) {
       iRet = m_pInterfaceVp->Process (iMethodIdx, &sSrcPixMap, &sDstPicMap);
     } else {
-      WelsMoveMemory_c (pDstPic->pData[0], pDstPic->pData[1], pDstPic->pData[2], pDstPic->iLineSize[0], pDstPic->iLineSize[1],
-                        pSrc->pData[0], pSrc->pData[1], pSrc->pData[2], pSrc->iLineSize[0], pSrc->iLineSize[1],
-                        iSrcWidth, iSrcHeight);
+      WelsMoveMemory_c(pDstPic->pData[0], pDstPic->pData[1], pDstPic->pData[2],
+                       pDstPic->iLineSize[0], pDstPic->iLineSize[1],
+                       pDstPic->iLineSize[2], pSrc->pData[0], pSrc->pData[1],
+                       pSrc->pData[2], pSrc->iLineSize[0], pSrc->iLineSize[1],
+                       pSrc->iLineSize[2], iSrcWidth, iSrcHeight);
     }
   } else {
     memcpy (&sDstPicMap, &sSrcPixMap, sizeof (sDstPicMap)); // confirmed_safe_unsafe_usage
@@ -1361,9 +1365,11 @@ void* WelsMemset (void* p, int32_t val, uint32_t uiSize) {
 }
 
 //i420_to_i420_c
-void  WelsMoveMemory_c (uint8_t* pDstY, uint8_t* pDstU, uint8_t* pDstV,  int32_t iDstStrideY, int32_t iDstStrideUV,
-                        uint8_t* pSrcY, uint8_t* pSrcU, uint8_t* pSrcV, int32_t iSrcStrideY, int32_t iSrcStrideUV, int32_t iWidth,
-                        int32_t iHeight) {
+void WelsMoveMemory_c(uint8_t* pDstY, uint8_t* pDstU, uint8_t* pDstV,
+                      int32_t iDstStrideY, int32_t iDstStrideU,
+                      int32_t iDstStrideV, uint8_t* pSrcY, uint8_t* pSrcU,
+                      uint8_t* pSrcV, int32_t iSrcStrideY, int32_t iSrcStrideU,
+                      int32_t iSrcStrideV, int32_t iWidth, int32_t iHeight) {
   int32_t   iWidth2 = iWidth >> 1;
   int32_t   iHeight2 = iHeight >> 1;
   int32_t   j;
@@ -1377,10 +1383,10 @@ void  WelsMoveMemory_c (uint8_t* pDstY, uint8_t* pDstU, uint8_t* pDstV,  int32_t
   for (j = iHeight2; j; j--) {
     WelsMemcpy (pDstU, pSrcU, iWidth2);
     WelsMemcpy (pDstV, pSrcV, iWidth2);
-    pDstU += iDstStrideUV;
-    pDstV += iDstStrideUV;
-    pSrcU += iSrcStrideUV;
-    pSrcV += iSrcStrideUV;
+    pDstU += iDstStrideU;
+    pDstV += iDstStrideV;
+    pSrcU += iSrcStrideU;
+    pSrcV += iSrcStrideV;
   }
 }
 
@@ -1413,13 +1419,15 @@ void  CWelsPreProcess::WelsMoveMemoryWrapper (SWelsSvcCodingParam* pSvcParam, SP
   uint8_t* pSrcU = kpSrc->pData[1] + iSrcOffset[1];
   uint8_t* pSrcV = kpSrc->pData[2] + iSrcOffset[2];
   const int32_t kiSrcStrideY = kpSrc->iStride[0];
-  const int32_t kiSrcStrideUV = kpSrc->iStride[1];
+  const int32_t kiSrcStrideU = kpSrc->iStride[1];
+  const int32_t kiSrcStrideV = kpSrc->iStride[2];
 
   uint8_t* pDstY = pDstPic->pData[0];
   uint8_t* pDstU = pDstPic->pData[1];
   uint8_t* pDstV = pDstPic->pData[2];
   const int32_t kiDstStrideY = pDstPic->iLineSize[0];
-  const int32_t kiDstStrideUV = pDstPic->iLineSize[1];
+  const int32_t kiDstStrideU = pDstPic->iLineSize[1];
+  const int32_t kiDstStrideV = pDstPic->iLineSize[2];
 
   if (pSrcY) {
     if (iSrcWidth <= 0 || iSrcHeight <= 0 || (iSrcWidth * iSrcHeight > (MAX_MBS_PER_FRAME << 8)))
@@ -1438,12 +1446,14 @@ void  CWelsPreProcess::WelsMoveMemoryWrapper (SWelsSvcCodingParam* pSvcParam, SP
       || (iSrcWidth & 1) || (iSrcHeight & 1)) {
   } else {
     //i420_to_i420_c
-    WelsMoveMemory_c (pDstY,  pDstU,  pDstV,  kiDstStrideY, kiDstStrideUV,
-                      pSrcY,  pSrcU,  pSrcV, kiSrcStrideY, kiSrcStrideUV, iSrcWidth, iSrcHeight);
+    WelsMoveMemory_c(pDstY, pDstU, pDstV, kiDstStrideY, kiDstStrideU,
+                     kiDstStrideV, pSrcY, pSrcU, pSrcV, kiSrcStrideY,
+                     kiSrcStrideU, kiSrcStrideV, iSrcWidth, iSrcHeight);
 
     //in VP Process
     if (kiTargetWidth > iSrcWidth || kiTargetHeight > iSrcHeight) {
-      Padding (pDstY, pDstU, pDstV, kiDstStrideY, kiDstStrideUV, iSrcWidth, kiTargetWidth, iSrcHeight, kiTargetHeight);
+      Padding(pDstY, pDstU, pDstV, kiDstStrideY, kiDstStrideU, iSrcWidth,
+              kiTargetWidth, iSrcHeight, kiTargetHeight);
     }
   }
 

meson.build

@@ -46,6 +46,7 @@ supported_arguments = cpp.get_supported_arguments([
 
 add_project_arguments(supported_arguments, language: 'cpp')
 
+
 deps = [dependency('threads')]
 c_args = []
 cpp_args = []

test/api/encoder_test.cpp

@@ -259,3 +259,56 @@ TEST_F(EncoderInitTest, VeryLargeSlices) {
     ASSERT_EQ(0, rv);
   }
 }
+
+// This test verifies that the encoder correctly handles frames with distinct,
+// asymmetric strides for the U and V chroma planes (e.g., when U stride is much
+// larger than V stride, or vice versa). It ensures that memory copy routines
+// advance each chroma plane pointer by its own stride without invalid memory
+// access.
+TEST_F(EncoderInitTest, CustomChromaPlaneStrides) {
+  SEncParamExt param;
+  encoder_->GetDefaultParams(&param);
+
+  param.iUsageType = CAMERA_VIDEO_REAL_TIME;
+  param.iPicWidth = 64;
+  param.iPicHeight = 64;
+  param.fMaxFrameRate = 30.0f;
+  param.iSpatialLayerNum = 1;
+  param.iRCMode = RC_OFF_MODE;
+
+  param.sSpatialLayers[0].iVideoWidth = param.iPicWidth;
+  param.sSpatialLayers[0].iVideoHeight = param.iPicHeight;
+  param.sSpatialLayers[0].fFrameRate = param.fMaxFrameRate;
+  param.sSpatialLayers[0].sSliceArgument.uiSliceMode = SM_SINGLE_SLICE;
+  param.sSpatialLayers[0].iDLayerQp = 24;
+
+  int rv = encoder_->InitializeExt(&param);
+  ASSERT_EQ(0, rv);
+
+  SFrameBSInfo info;
+  memset(&info, 0, sizeof(SFrameBSInfo));
+
+  // Configure distinct strides: U stride is much larger than V stride.
+  int strideY = 64;
+  int strideU = 4096;
+  int strideV = 32;
+
+  std::vector<uint8_t> bufY(strideY * param.iPicHeight, 128);
+  std::vector<uint8_t> bufU(strideU * (param.iPicHeight >> 1), 128);
+  std::vector<uint8_t> bufV(strideV * (param.iPicHeight >> 1), 128);
+
+  SSourcePicture pic;
+  memset(&pic, 0, sizeof(SSourcePicture));
+  pic.iPicWidth = param.iPicWidth;
+  pic.iPicHeight = param.iPicHeight;
+  pic.iColorFormat = videoFormatI420;
+  pic.iStride[0] = strideY;
+  pic.iStride[1] = strideU;
+  pic.iStride[2] = strideV;
+  pic.pData[0] = bufY.data();
+  pic.pData[1] = bufU.data();
+  pic.pData[2] = bufV.data();
+
+  rv = encoder_->EncodeFrame(&pic, &info);
+  ASSERT_EQ(0, rv);
+}