[x265] [PATCH] asm code for blockcopy_ps_4x2

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1384173076 -19800
# Node ID 3b48733bee3327ae046c00ebc24877882e54c9ec
# Parent  afcf4952995ca2da024776b1795841cf45187d9c
asm code for blockcopy_ps_4x2

diff -r afcf4952995c -r 3b48733bee33 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Nov 11 17:50:45 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Nov 11 18:01:16 2013 +0530
@@ -455,6 +455,7 @@
         // until all partitions are coded and commit smaller patches, easier to
         // review.
 
+        p.chroma_copy_ps[CHROMA_4x2] = x265_blockcopy_ps_4x2_sse4;
         p.chroma_copy_ps[CHROMA_8x2] = x265_blockcopy_ps_8x2_sse4;
         p.chroma_copy_ps[CHROMA_8x4] = x265_blockcopy_ps_8x4_sse4;
         p.chroma_copy_ps[CHROMA_8x6] = x265_blockcopy_ps_8x6_sse4;
diff -r afcf4952995c -r 3b48733bee33 source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Mon Nov 11 17:50:45 2013 +0530
+++ b/source/common/x86/blockcopy8.asm	Mon Nov 11 18:01:16 2013 +0530
@@ -1657,6 +1657,25 @@
 
 BLOCKFILL_S_W32_H4 32, 32
 
+
+;-----------------------------------------------------------------------------
+; void blockcopy_ps_4x2(int16_t *dest, intptr_t destStride, pixel *src, intptr_t srcStride);
+;-----------------------------------------------------------------------------
+INIT_XMM sse4
+cglobal blockcopy_ps_4x2, 4, 4, 1, dest, destStride, src, srcStride
+
+add        r1,         r1
+
+movd       m0,         [r2]
+pmovzxbw   m0,         m0
+movh       [r0],       m0
+
+movd       m0,         [r2 + r3]
+pmovzxbw   m0,         m0
+movh       [r0 + r1],  m0
+
+RET
+
 ;-----------------------------------------------------------------------------
 ; void blockcopy_ps_8x2(int16_t *dest, intptr_t destStride, pixel *src, intptr_t srcStride);
 ;-----------------------------------------------------------------------------
diff -r afcf4952995c -r 3b48733bee33 source/common/x86/blockcopy8.h
--- a/source/common/x86/blockcopy8.h	Mon Nov 11 17:50:45 2013 +0530
+++ b/source/common/x86/blockcopy8.h	Mon Nov 11 18:01:16 2013 +0530
@@ -94,6 +94,7 @@
     void x265_blockcopy_ps_ ## W ## x ## H ## cpu(int16_t *dst, intptr_t dstStride, pixel *src, intptr_t srcStride);
 
 #define CHROMA_BLOCKCOPY_DEF_SSE4(cpu) \
+    SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(4, 2, cpu); \
     SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 2, cpu); \
     SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 4, cpu); \
     SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 6, cpu); \