[x265] [PATCH] added asm function for luma blockcopy_ps_16x64

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1384183250 -19800
# Node ID 7a6dde5dd20ddba0ebea77e90830262d94d9b198
# Parent  5d12397bc1572588a5f69699d7910eaf5c096b06
added asm function for luma blockcopy_ps_16x64

diff -r 5d12397bc157 -r 7a6dde5dd20d source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Nov 11 20:35:55 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Nov 11 20:50:50 2013 +0530
@@ -451,6 +451,13 @@
         p.chroma_copy_sp[CHROMA_2x8] = x265_blockcopy_sp_2x8_sse4;
         p.chroma_copy_sp[CHROMA_6x8] = x265_blockcopy_sp_6x8_sse4;
 
+        // This function pointer initialization is temporary will be removed
+        // later with macro definitions.  It is used to avoid linker errors
+        // until all partitions are coded and commit smaller patches, easier to
+        // review.
+
+        p.luma_copy_ps[LUMA_16x64] = x265_blockcopy_ps_16x64_sse4;
+
     }
     if (cpuMask & X265_CPU_AVX)
     {
diff -r 5d12397bc157 -r 7a6dde5dd20d source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Mon Nov 11 20:35:55 2013 +0530
+++ b/source/common/x86/blockcopy8.asm	Mon Nov 11 20:50:50 2013 +0530
@@ -2132,7 +2132,7 @@
 BLOCKCOPY_PS_W16_H4 16, 12
 BLOCKCOPY_PS_W16_H4 16, 16
 BLOCKCOPY_PS_W16_H4 16, 32
-
+BLOCKCOPY_PS_W16_H4 16, 64
 
 ;-----------------------------------------------------------------------------
 ; void blockcopy_ps_%1x%2(int16_t *dest, intptr_t destStride, pixel *src, intptr_t srcStride);
diff -r 5d12397bc157 -r 7a6dde5dd20d source/common/x86/blockcopy8.h
--- a/source/common/x86/blockcopy8.h	Mon Nov 11 20:35:55 2013 +0530
+++ b/source/common/x86/blockcopy8.h	Mon Nov 11 20:50:50 2013 +0530
@@ -119,7 +119,14 @@
     SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(32, 24, cpu); \
     SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(32, 32, cpu);
 
+#define SETUP_LUMA_BLOCKCOPY_FUNC_SSE4(W, H, cpu) \
+    void x265_blockcopy_ps_ ## W ## x ## H ## cpu(int16_t *dst, intptr_t dstStride, pixel *src, intptr_t srcStride);
+
+#define LUMA_BLOCKCOPY_DEF_SSE4(cpu) \
+    SETUP_LUMA_BLOCKCOPY_FUNC_SSE4(16, 64, cpu); \
+
 CHROMA_BLOCKCOPY_DEF_SSE4(_sse4);
+LUMA_BLOCKCOPY_DEF_SSE4(_sse4);
 
 void x265_blockcopy_sp_2x4_sse4(pixel *a, intptr_t stridea, int16_t *b, intptr_t strideb);
 void x265_blockcopy_sp_2x8_sse4(pixel *a, intptr_t stridea, int16_t *b, intptr_t strideb);
@@ -137,5 +144,7 @@
 
 #undef SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4
 #undef CHROMA_BLOCKCOPY_DEF_SSE4
+#undef SETUP_LUMA_BLOCKCOPY_FUNC_SSE4
+#undef LUMA_BLOCKCOPY_DEF_SSE4
 
 #endif // ifndef X265_I386_PIXEL_H