[x265] [PATCH] asm code for luma blockcopy_ps_32x64

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1384183503 -19800
# Node ID afbc2f56225c1fab5b53a8c51b979814162a638b
# Parent  7a6dde5dd20ddba0ebea77e90830262d94d9b198
asm code for luma blockcopy_ps_32x64

diff -r 7a6dde5dd20d -r afbc2f56225c source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Nov 11 20:50:50 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Nov 11 20:55:03 2013 +0530
@@ -457,7 +457,7 @@
         // review.
 
         p.luma_copy_ps[LUMA_16x64] = x265_blockcopy_ps_16x64_sse4;
-
+        p.luma_copy_ps[LUMA_32x64] = x265_blockcopy_ps_32x64_sse4;
     }
     if (cpuMask & X265_CPU_AVX)
     {
diff -r 7a6dde5dd20d -r afbc2f56225c source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Mon Nov 11 20:50:50 2013 +0530
+++ b/source/common/x86/blockcopy8.asm	Mon Nov 11 20:55:03 2013 +0530
@@ -2226,3 +2226,4 @@
 BLOCKCOPY_PS_W32_H2 32, 16
 BLOCKCOPY_PS_W32_H2 32, 24
 BLOCKCOPY_PS_W32_H2 32, 32
+BLOCKCOPY_PS_W32_H2 32, 64
diff -r 7a6dde5dd20d -r afbc2f56225c source/common/x86/blockcopy8.h
--- a/source/common/x86/blockcopy8.h	Mon Nov 11 20:50:50 2013 +0530
+++ b/source/common/x86/blockcopy8.h	Mon Nov 11 20:55:03 2013 +0530
@@ -124,6 +124,7 @@
 
 #define LUMA_BLOCKCOPY_DEF_SSE4(cpu) \
     SETUP_LUMA_BLOCKCOPY_FUNC_SSE4(16, 64, cpu); \
+    SETUP_LUMA_BLOCKCOPY_FUNC_SSE4(32, 64, cpu);
 
 CHROMA_BLOCKCOPY_DEF_SSE4(_sse4);
 LUMA_BLOCKCOPY_DEF_SSE4(_sse4);