[x265] [PATCH] TComYuv::copyPartToPartYuv, asm code intergration for blockcopy_ps

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1384788645 -19800
# Node ID 49a556cf22721d846a94e07c1933fcd092b898dd
# Parent  59646d515e79b4d0f9a3a72c77c7af17a83bf3d9
TComYuv::copyPartToPartYuv, asm code intergration for blockcopy_ps

diff -r 59646d515e79 -r 49a556cf2272 source/Lib/TLibCommon/TComYuv.cpp
--- a/source/Lib/TLibCommon/TComYuv.cpp	Mon Nov 18 20:53:29 2013 +0530
+++ b/source/Lib/TLibCommon/TComYuv.cpp	Mon Nov 18 21:00:45 2013 +0530
@@ -256,10 +256,12 @@
 
 void TComYuv::copyPartToPartYuv(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height, bool bLuma, bool bChroma)
 {
+    int part = partitionFromSizes(width, height);
+
     if (bLuma)
-        copyPartToPartLuma(dstPicYuv, partIdx, width, height);
+        copyPartToPartLuma(dstPicYuv, partIdx, part);
     if (bChroma)
-        copyPartToPartChroma(dstPicYuv, partIdx, width >> m_hChromaShift, height >> m_vChromaShift);
+        copyPartToPartChroma(dstPicYuv, partIdx, part);
 }
 
 void TComYuv::copyPartToPartLuma(TComYuv* dstPicYuv, uint32_t partIdx, uint32_t part)
@@ -275,7 +277,7 @@
     primitives.luma_copy_pp[part](dst, dststride, src, srcstride);
 }
 
-void TComYuv::copyPartToPartLuma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height)
+void TComYuv::copyPartToPartLuma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t part)
 {
     Pel* src = getLumaAddr(partIdx);
     int16_t* dst = dstPicYuv->getLumaAddr(partIdx);
@@ -283,7 +285,6 @@
     uint32_t  srcstride = getStride();
     uint32_t  dststride = dstPicYuv->m_width;
 
-    int part = partitionFromSizes(width, height);
     primitives.luma_copy_ps[part](dst, dststride, src, srcstride);
 }
 
@@ -303,7 +304,7 @@
     primitives.chroma_copy_pp[m_csp][part](dstV, dststride, srcV, srcstride);
 }
 
-void TComYuv::copyPartToPartChroma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height)
+void TComYuv::copyPartToPartChroma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t part)
 {
     Pel*   srcU = getCbAddr(partIdx);
     Pel*   srcV = getCrAddr(partIdx);
@@ -313,8 +314,8 @@
     uint32_t srcstride = getCStride();
     uint32_t dststride = dstPicYuv->m_cwidth;
 
-    primitives.blockcpy_sp(width, height, dstU, dststride, srcU, srcstride);
-    primitives.blockcpy_sp(width, height, dstV, dststride, srcV, srcstride);
+    primitives.chroma_copy_ps[m_csp][part](dstU, dststride, srcU, srcstride);
+    primitives.chroma_copy_ps[m_csp][part](dstV, dststride, srcV, srcstride);
 }
 
 void TComYuv::copyPartToPartChroma(TComYuv* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height, uint32_t chromaId)
diff -r 59646d515e79 -r 49a556cf2272 source/Lib/TLibCommon/TComYuv.h
--- a/source/Lib/TLibCommon/TComYuv.h	Mon Nov 18 20:53:29 2013 +0530
+++ b/source/Lib/TLibCommon/TComYuv.h	Mon Nov 18 21:00:45 2013 +0530
@@ -140,9 +140,9 @@
     void    copyPartToPartYuv(TComYuv* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height, bool bLuma = true, bool bChroma = true);
     void    copyPartToPartYuv(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height, bool bLuma = true, bool bChroma = true);
     void    copyPartToPartLuma(TComYuv* dstPicYuv, uint32_t partIdx, uint32_t part);
-    void    copyPartToPartLuma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height);
+    void    copyPartToPartLuma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t part);
     void    copyPartToPartChroma(TComYuv* dstPicYuv, uint32_t partIdx, uint32_t part);
-    void    copyPartToPartChroma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height);
+    void    copyPartToPartChroma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t part);
 
     void    copyPartToPartChroma(TComYuv* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height, uint32_t chromaId);
     void    copyPartToPartChroma(TShortYUV* dstPicYuv, uint32_t partIdx, uint32_t width, uint32_t height, uint32_t chromaId);
diff -r 59646d515e79 -r 49a556cf2272 source/Lib/TLibEncoder/TEncSearch.cpp
--- a/source/Lib/TLibEncoder/TEncSearch.cpp	Mon Nov 18 20:53:29 2013 +0530
+++ b/source/Lib/TLibEncoder/TEncSearch.cpp	Mon Nov 18 21:00:45 2013 +0530
@@ -1126,12 +1126,12 @@
     }
 
     //===== copy reconstruction =====
-    m_qtTempTransformSkipTComYuv.copyPartToPartLuma(&m_qtTempTComYuv[qtlayer], absPartIdx, 1 << trSizeLog2, 1 << trSizeLog2);
+    int part = partitionFromSizes(1 << trSizeLog2, 1 << trSizeLog2);
+    m_qtTempTransformSkipTComYuv.copyPartToPartLuma(&m_qtTempTComYuv[qtlayer], absPartIdx, part);
 
     if (!bLumaOnly && !bSkipChroma)
     {
-        uint32_t trSizeCLog2 = (bChromaSame ? trSizeLog2 : trSizeLog2 - 1);
-        m_qtTempTransformSkipTComYuv.copyPartToPartChroma(&m_qtTempTComYuv[qtlayer], absPartIdx, 1 << trSizeCLog2, 1 << trSizeCLog2);
+        m_qtTempTransformSkipTComYuv.copyPartToPartChroma(&m_qtTempTComYuv[qtlayer], absPartIdx, part);
     }
 
     uint32_t   zOrder           = cu->getZorderIdxInCU() + absPartIdx;