[x265] [PATCH] primitives: added C primitives for upShift/downShift input pixels

[murugan at multicorewareinc.com]

# HG changeset patch
# User Murugan Vairavel <murugan at multicorewareinc.com>
# Date 1394529456 -19800
#      Tue Mar 11 14:47:36 2014 +0530
# Node ID 6f60664fd99d54839cf8c71386d5419849e98b1f
# Parent  1417ba917211abb9f4a43ff44ca28ad9d9a0ab43
primitives: added C primitives for upShift/downShift input pixels

diff -r 1417ba917211 -r 6f60664fd99d source/Lib/TLibCommon/TComPicYuv.cpp
--- a/source/Lib/TLibCommon/TComPicYuv.cpp	Mon Mar 10 16:59:22 2014 +0530
+++ b/source/Lib/TLibCommon/TComPicYuv.cpp	Tue Mar 11 14:47:36 2014 +0530
@@ -193,30 +193,9 @@
         uint8_t *u = (uint8_t*)pic.planes[1];
         uint8_t *v = (uint8_t*)pic.planes[2];
 
-        for (int r = 0; r < height; r++)
-        {
-            for (int c = 0; c < width; c++)
-            {
-                Y[c] = ((pixel)y[c]) << 2;
-            }
-
-            Y += getStride();
-            y += pic.stride[0] / sizeof(*y);
-        }
-
-        for (int r = 0; r < height >> m_vChromaShift; r++)
-        {
-            for (int c = 0; c < width >> m_hChromaShift; c++)
-            {
-                U[c] = ((pixel)u[c]) << 2;
-                V[c] = ((pixel)v[c]) << 2;
-            }
-
-            U += getCStride();
-            V += getCStride();
-            u += pic.stride[1] / sizeof(*u);
-            v += pic.stride[2] / sizeof(*v);
-        }
+        primitives.upShift(y, pic.stride[0] / sizeof(*y), Y,getStride(), width, height);
+        primitives.upShift(u, pic.stride[1] / sizeof(*u), U,getStride(), width >> m_hChromaShift, height >> m_vChromaShift);
+        primitives.upShift(v, pic.stride[2] / sizeof(*v), V,getStride(), width >> m_hChromaShift, height >> m_vChromaShift);
     }
     else if (pic.bitDepth == 8)
     {
@@ -268,30 +247,9 @@
         int shift = X265_MAX(0, pic.bitDepth - X265_DEPTH);
 
         /* shift and mask pixels to final size */
-        for (int r = 0; r < height; r++)
-        {
-            for (int c = 0; c < width; c++)
-            {
-                Y[c] = (pixel)((y[c] >> shift) & mask);
-            }
-
-            Y += getStride();
-            y += pic.stride[0] / sizeof(*y);
-        }
-
-        for (int r = 0; r < height >> m_vChromaShift; r++)
-        {
-            for (int c = 0; c < width >> m_hChromaShift; c++)
-            {
-                U[c] = (pixel)((u[c] >> shift) & mask);
-                V[c] = (pixel)((v[c] >> shift) & mask);
-            }
-
-            U += getCStride();
-            V += getCStride();
-            u += pic.stride[1] / sizeof(*u);
-            v += pic.stride[2] / sizeof(*v);
-        }
+        primitives.downShift(y, pic.stride[0] / sizeof(*y), Y,getCStride(), width, height, shift, mask);
+        primitives.downShift(u, pic.stride[1] / sizeof(*u), U,getCStride(), width >> m_hChromaShift, height >> m_vChromaShift, shift, mask);
+        primitives.downShift(v, pic.stride[2] / sizeof(*v), V,getCStride(), width >> m_hChromaShift, height >> m_vChromaShift, shift, mask);
     }
 
     /* extend the right edge if width was not multiple of the minimum CU size */
diff -r 1417ba917211 -r 6f60664fd99d source/common/pixel.cpp
--- a/source/common/pixel.cpp	Mon Mar 10 16:59:22 2014 +0530
+++ b/source/common/pixel.cpp	Tue Mar 11 14:47:36 2014 +0530
@@ -852,6 +852,34 @@
         dst  += dstStride;
     }
 }
+
+void upShift(uint8_t *src, int srcStride, pixel *dst, int dstStride, int width, int height)
+{
+    for (int r = 0; r < height; r++)
+    {
+        for (int c = 0; c < width; c++)
+        {
+            dst[c] = ((pixel)src[c]) << 2;
+        }
+
+        dst += dstStride;
+        src += srcStride;
+    }
+}
+
+void downShift(uint16_t *src, int srcStride, pixel *dst, int dstStride, int width, int height, int shift, uint16_t mask)
+{
+    for (int r = 0; r < height; r++)
+    {
+        for (int c = 0; c < width; c++)
+        {
+            dst[c] = (pixel)((src[c] >> shift) & mask);
+        }
+
+        dst += dstStride;
+        src += srcStride;
+    }
+}
 }  // end anonymous namespace
 
 namespace x265 {
@@ -1099,5 +1127,7 @@
     p.var[BLOCK_32x32] = pixel_var<32>;
     p.var[BLOCK_64x64] = pixel_var<64>;
     p.plane_copy_deinterleave_c = plane_copy_deinterleave_chroma;
+    p.upShift = upShift;
+    p.downShift = downShift;
 }
 }
diff -r 1417ba917211 -r 6f60664fd99d source/common/primitives.h
--- a/source/common/primitives.h	Mon Mar 10 16:59:22 2014 +0530
+++ b/source/common/primitives.h	Tue Mar 11 14:47:36 2014 +0530
@@ -163,6 +163,8 @@
 typedef void (*addAvg_t)(int16_t* src0, int16_t* src1, pixel* dst, intptr_t src0Stride, intptr_t src1Stride, intptr_t dstStride);
 
 typedef void (*saoCuOrgE0_t)(pixel * rec, int8_t * offsetEo, int lcuWidth, int8_t signLeft);
+typedef void (*upShift_t) (uint8_t *src, int srcStride, pixel *dst, int dstStride, int width, int height);
+typedef void (*downShift_t) (uint16_t *src, int srcStride, pixel *dst, int dstStride, int width, int height, int shift, uint16_t mask);
 
 /* Define a structure containing function pointers to optimized encoder
  * primitives.  Each pointer can reference either an assembly routine,
@@ -233,6 +235,8 @@
     extendCURowBorder_t extendRowBorder;
     // sao primitives
     saoCuOrgE0_t      saoCuOrgE0;
+    upShift_t         upShift;
+    downShift_t       downShift;
 
     struct
     {