Use transposed buffers

Extend use of transposed buffers to enable SIMD for SAD, weightedAverage and linearTransform. This is already done in some parts of the ECM but this MR extend the usage It enables SIMD usage as some blocks are considered as a row instead of a column (please note that it changes only how the block is considered not the memory)

source/Lib/CommonLib/InterPrediction.cpp

@@ -8861,8 +8861,7 @@ bool InterPrediction::xAMLIBCGetCurBlkTemplate(PredictionUnit& pu, int nCurBlkWi
   if (m_bAMLTemplateAvailabe[0])
   {
     const Pel*    rec = recBuf.bufAt(pu.blocks[COMPONENT_Y].pos().offset(0, -AML_MERGE_TEMPLATE_SIZE));
-    PelBuf pcYBuf = PelBuf(m_acYuvCurAMLTemplate[0][0], nCurBlkWidth, AML_MERGE_TEMPLATE_SIZE);
-    Pel*   pcY = pcYBuf.bufAt(0, 0);
+    Pel*   pcY = m_acYuvCurAMLTemplate[0][0];
     for (int k = 0; k < nCurBlkWidth; k++)
     {
       for (int l = 0; l < AML_MERGE_TEMPLATE_SIZE; l++)
@@ -9140,8 +9139,7 @@ void InterPrediction::getIBCAMLRefTemplate(PredictionUnit &pu, int nCurBlkWidth,
 #endif
       if (numTemplate[0] + numTemplate[1] > 0)
       {
-        PelUnitBuf pcYBufTranspose(pu.chromaFormat, PelBuf(m_acYuvRefAMLTemplate[1][0], nCurBlkHeight, AML_MERGE_TEMPLATE_SIZE));
-        pcYBufTranspose.Y().linearTransform(scale, shift, offset, true, clpRng);
+        pcYBuf.Y().linearTransform(scale, shift, offset, true, clpRng);
       }
     }
     else
@@ -11542,8 +11540,7 @@ bool InterPrediction::xAMLGetCurBlkTemplate(PredictionUnit& pu, int nCurBlkWidth
   if (m_bAMLTemplateAvailabe[0])
   {
     const Pel*    rec = recBuf.bufAt(pu.blocks[COMPONENT_Y].pos().offset(0, -AML_MERGE_TEMPLATE_SIZE));
-    PelBuf pcYBuf = PelBuf(m_acYuvCurAMLTemplate[0][0], nCurBlkWidth, AML_MERGE_TEMPLATE_SIZE);
-    Pel*   pcY = pcYBuf.bufAt(0, 0);
+    Pel*   pcY = m_acYuvCurAMLTemplate[0][0];
     for (int k = 0; k < nCurBlkWidth; k++)
     {
       for (int l = 0; l < AML_MERGE_TEMPLATE_SIZE; l++)