[x265] [PATCH 2 of 2] Fix for Testbench fail in weightpUni for Pixel input

[shazeb at multicorewareinc.com]

# HG changeset patch
# User Shazeb Nawaz Khan <shazeb at multicorewareinc.com>
# Date 1380886668 -19800
#      Fri Oct 04 17:07:48 2013 +0530
# Node ID 0bb2c80a73b7f36f9a491d750d78838bd866e893
# Parent  b54b030188bcb53f6c5c64fe99ac29a4d851eb78
Fix for Testbench fail in weightpUni for Pixel input

diff -r b54b030188bc -r 0bb2c80a73b7 source/common/vec/pixel.inc
--- a/source/common/vec/pixel.inc	Fri Oct 04 16:40:04 2013 +0530
+++ b/source/common/vec/pixel.inc	Fri Oct 04 17:07:48 2013 +0530
@@ -168,8 +168,8 @@
     p.blockfil_s[BLOCK_32x32] = blockfil_s_32;
 #endif
 
-    p.weightpUniPixel = weightUnidir<pixel>;
-    p.weightpUni = weightUnidir<int16_t>;
+    p.weightpUniPixel = weightUnidirPixel;
+    p.weightpUni = weightUnidir;
 
     p.calcresidual[BLOCK_4x4] = getResidual4;
     p.calcresidual[BLOCK_8x8] = getResidual8;
diff -r b54b030188bc -r 0bb2c80a73b7 source/common/vec/pixel8.inc
--- a/source/common/vec/pixel8.inc	Fri Oct 04 16:40:04 2013 +0530
+++ b/source/common/vec/pixel8.inc	Fri Oct 04 17:07:48 2013 +0530
@@ -717,8 +717,7 @@
     }
 }
 
-template <typename T>
-void weightUnidir(T *src, pixel *dst, intptr_t srcStride, intptr_t dstStride, int width, int height, int w0, int round, int shift, int offset)
+void weightUnidir(int16_t *src, pixel *dst, intptr_t srcStride, intptr_t dstStride, int width, int height, int w0, int round, int shift, int offset)
 {
     int x, y;
     Vec8s tmp;
@@ -745,3 +744,32 @@
         dst += dstStride;
     }
 }
+
+void weightUnidirPixel(pixel *src, pixel *dst, intptr_t srcStride, intptr_t dstStride, int width, int height, int w0, int round, int shift, int offset)
+{
+    int x, y;
+    Vec16uc tmp;
+
+    Vec4i vw0(w0), vsrc, iofs(IF_INTERNAL_OFFS), ofs(offset), vround(round), vdst;
+    for (y = height - 1; y >= 0; y--)
+    {
+        for (x = 0; x <= width - 4; x += 4)
+        {
+            tmp = load_partial(const_int(4), src + x);
+            // The intermediate results would outgrow 16 bits because internal offset is too high
+            vsrc = extend_low(extend_low(tmp));
+            vdst = ((vw0 * (vsrc + iofs) + vround) >> shift) + ofs;
+            store_partial(const_int(4), dst + x, compress_unsafe(compress_saturated(vdst, vdst), 0));
+        }
+
+        if (width > x)
+        {
+            tmp  = load_partial(const_int(4), src + x);
+            vsrc = extend_low(extend_low(tmp));
+            vdst = ((vw0 * (vsrc + iofs) + vround) >> shift) + ofs;
+            compress_unsafe(compress_saturated(vdst, vdst), 0).store_partial(2, dst + x);
+        }
+        src += srcStride;
+        dst += dstStride;
+    }
+}
diff -r b54b030188bc -r 0bb2c80a73b7 source/test/pixelharness.cpp
--- a/source/test/pixelharness.cpp	Fri Oct 04 16:40:04 2013 +0530
+++ b/source/test/pixelharness.cpp	Fri Oct 04 17:07:48 2013 +0530
@@ -371,8 +371,8 @@
     int offset = (rand() % 256) - 128;
     for (int i = 0; i < ITERS; i++)
     {
-        opt((pixel *)sbuf1 + j, opt_dest, 64, 64, width, height, w0, round, shift, offset);
-        ref((pixel *)sbuf1 + j, ref_dest, 64, 64, width, height, w0, round, shift, offset);
+        opt(pbuf1 + j, opt_dest, 64, 64, width, height, w0, round, shift, offset);
+        ref(pbuf1 + j, ref_dest, 64, 64, width, height, w0, round, shift, offset);
 
         if (memcmp(ref_dest, opt_dest, 64 * 64 * sizeof(pixel)))
             return false;
@@ -636,16 +636,7 @@
     {
         if (!check_weightpUni(ref.weightpUniPixel, opt.weightpUniPixel))
         {
-            printf("Weighted Prediction for Unidir failed!\n");
-            return false;
-        }
-    }
-
-    if (opt.weightpUniPixel)
-    {
-        if (!check_weightpUni(ref.weightpUniPixel, opt.weightpUniPixel))
-        {
-            printf("Weighted Prediction for Unidir failed!\n");
+            printf("Weighted Prediction for Unidir (Pixel) failed!\n");
             return false;
         }
     }
@@ -654,7 +645,7 @@
     {
         if (!check_weightpUni(ref.weightpUni, opt.weightpUni))
         {
-            printf("Weighted Prediction for Unidir failed!\n");
+            printf("Weighted Prediction for Unidir (int16_t) failed!\n");
             return false;
         }
     }