[x265] [PATCH] pixel: replace getResidual32 from vector class to intrinsic

[yuvaraj at multicorewareinc.com]

# HG changeset patch
# User Yuvaraj Venkatesh <yuvaraj at multicorewareinc.com>
# Date 1381223006 -19800
#      Tue Oct 08 14:33:26 2013 +0530
# Node ID 1a62566488b7ece9bbfb665e37ac402a08ce156e
# Parent  96e30370f4d96c7fed69f432027ed3be8e01dcf6
pixel: replace getResidual32 from vector class to intrinsic

diff -r 96e30370f4d9 -r 1a62566488b7 source/common/vec/pixel8.inc
--- a/source/common/vec/pixel8.inc	Tue Oct 08 14:16:23 2013 +0530
+++ b/source/common/vec/pixel8.inc	Tue Oct 08 14:33:26 2013 +0530
@@ -79,31 +79,37 @@
     RESIDUAL_16x4(12);
 }
 
-void getResidual32(pixel *fenc, pixel *pred, short *resi, int stride)
-{
-    Vec16uc f, p;
-    Vec8s r;
-
-    for (int y = 0; y < 32; y++)
-    {
-        f.load_a(fenc);
-        p.load_a(pred);
-        r = extend_low(f) - extend_low(p);
-        r.store(resi);
-        r = extend_high(f) - extend_high(p);
-        r.store(resi + 8);
-
-        f.load_a(fenc + 16);
-        p.load_a(pred + 16);
-        r = extend_low(f) - extend_low(p);
-        r.store(resi + 16);
-        r = extend_high(f) - extend_high(p);
-        r.store(resi + 24);
-
-        fenc += stride;
-        pred += stride;
-        resi += stride;
-    }
+void getResidual32(pixel *fenc, pixel *pred, short *resi, int stride)
+{
+    __m128i T00, T01, T02, T03, T04;
+
+#define RESIDUAL_32x4(BASE, OFFSET) \
+    T00 = _mm_load_si128((__m128i*)(fenc + OFFSET + (BASE + 0) * stride)); \
+    T01 = _mm_load_si128((__m128i*)(pred + OFFSET + (BASE + 0) * stride)); \
+    T02 = _mm_unpacklo_epi8(T00, _mm_setzero_si128()); \
+    T03 = _mm_unpacklo_epi8(T01, _mm_setzero_si128()); \
+    T04 = _mm_sub_epi16(T02, T03); \
+    _mm_store_si128((__m128i*)(resi + OFFSET + (BASE + 0) * stride), T04); \
+    T02 = _mm_unpackhi_epi8(T00, _mm_setzero_si128()); \
+    T03 = _mm_unpackhi_epi8(T01, _mm_setzero_si128()); \
+    T04 = _mm_sub_epi16(T02, T03); \
+    _mm_store_si128((__m128i*)(resi + 8 + OFFSET + (BASE + 0) * stride), T04); \
+    T00 = _mm_load_si128((__m128i*)(fenc + OFFSET + (BASE + 1) * stride)); \
+    T01 = _mm_load_si128((__m128i*)(pred + OFFSET + (BASE + 1) * stride)); \
+    T02 = _mm_unpacklo_epi8(T00, _mm_setzero_si128()); \
+    T03 = _mm_unpacklo_epi8(T01, _mm_setzero_si128()); \
+    T04 = _mm_sub_epi16(T02, T03); \
+    _mm_store_si128((__m128i*)(resi + OFFSET + (BASE + 1) * stride), T04); \
+    T02 = _mm_unpackhi_epi8(T00, _mm_setzero_si128()); \
+    T03 = _mm_unpackhi_epi8(T01, _mm_setzero_si128()); \
+    T04 = _mm_sub_epi16(T02, T03); \
+    _mm_store_si128((__m128i*)(resi + 8 + OFFSET + (BASE + 1) * stride), T04)
+
+    for (int i = 0; i < 32; i += 2)
+    {
+        RESIDUAL_32x4(i, 0);
+        RESIDUAL_32x4(i, 16);
+    }
 }
 
 void getResidual64(pixel *fenc, pixel *pred, short *resi, int stride)