[x265] [PATCH] pixel: Replace weightUnidir vector class function with intrinsic

[yuvaraj at multicorewareinc.com]

# HG changeset patch
# User Yuvaraj Venkatesh <yuvaraj at multicorewareinc.com>
# Date 1381315230 -19800
#      Wed Oct 09 16:10:30 2013 +0530
# Node ID 1d3760e10f643954edb5dd8dd953c2511ff9a90f
# Parent  fc7fbdd18bc0d6d7f98180332e065d83c054fe02
pixel: Replace weightUnidir vector class function with intrinsic.

diff -r fc7fbdd18bc0 -r 1d3760e10f64 source/common/vec/pixel8.inc
--- a/source/common/vec/pixel8.inc	Wed Oct 09 00:00:10 2013 -0500
+++ b/source/common/vec/pixel8.inc	Wed Oct 09 16:10:30 2013 +0530
@@ -27,33 +27,45 @@
 
 /* intrinsics for when pixel type is uint8_t */
 
-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;
+void weightUnidir(short *src, pixel *dst, intptr_t srcStride, intptr_t dstStride, int width, int height, int w0, int round, int shift, int offset)
+{
+    __m128i w00, roundoff, ofs, fs, tmpsrc, tmpdst, tmp;
+    int x, y;
+
+    w00 = _mm_set1_epi32(w0);
+    ofs = _mm_set1_epi32(IF_INTERNAL_OFFS);
+    fs = _mm_set1_epi32(offset);
+    roundoff = _mm_set1_epi32(round);
+    for (y = height - 1; y >= 0; y--)
+    {
+        for (x = 0; x <= width - 4; x += 4)
+        {
+            tmpsrc = _mm_loadl_epi64((__m128i*)(src + x));
+            tmpsrc = _mm_unpacklo_epi16(tmpsrc, _mm_setzero_si128());
+            tmpdst = _mm_add_epi32(_mm_srai_epi32(_mm_add_epi32(_mm_mullo_epi32(w00, _mm_add_epi32(tmpsrc, ofs)), roundoff), shift), fs);
+            *(uint32_t*)(dst + x) = _mm_cvtsi128_si32(_mm_packus_epi16(_mm_packs_epi32(tmpdst, tmpdst), _mm_setzero_si128()));
+        }
+
+        if (width > x)
+        {
+            tmpsrc = _mm_loadl_epi64((__m128i*)(src + x));
+            tmpsrc = _mm_unpacklo_epi16(tmpsrc, _mm_setzero_si128());
+            tmpdst = _mm_add_epi32(_mm_srai_epi32(_mm_add_epi32(_mm_mullo_epi32(w00, _mm_add_epi32(tmpsrc, ofs)), roundoff), shift), fs);
+            tmp = _mm_packus_epi16(_mm_packs_epi32(tmpdst, tmpdst), _mm_setzero_si128());
+            union
+            {
+                int8_t  c[16];
+                int16_t s[8];
+            } u;
+
+            _mm_storeu_si128((__m128i*)u.c, tmp);
+            ((int16_t*)(dst + x))[0] = u.s[0];    //to store only first 16-bit from 128-bit to memory
+        }
+        src += srcStride;
+        dst += dstStride;
+    }
+}
 
-    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(8), src + x);
-            vsrc = 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(tmp);
-            vdst = ((vw0 * (vsrc + iofs) + vround) >> shift) + ofs;
-            compress_unsafe(compress_saturated(vdst, vdst), 0).store_partial(2, dst + x);
-        }
-        src += srcStride;
-        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)
 {