[x265] [PATCH] asm code for blockcopy_ps, 8x6, 8x16 and 8x32

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1384162089 -19800
# Node ID 6da0a0291ed8d10dc3dfdb3df396cd1a8c74ceeb
# Parent  da0b44e67fe07caa7ed113ec4946a371d96801be
asm code for blockcopy_ps, 8x6, 8x16 and 8x32

diff -r da0b44e67fe0 -r 6da0a0291ed8 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Nov 11 14:36:21 2013 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Nov 11 14:58:09 2013 +0530
@@ -459,6 +459,9 @@
         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;
+        p.chroma_copy_ps[CHROMA_8x8] = x265_blockcopy_ps_8x8_sse4;
+        p.chroma_copy_ps[CHROMA_8x16] = x265_blockcopy_ps_8x16_sse4;
+        p.chroma_copy_ps[CHROMA_8x32] = x265_blockcopy_ps_8x32_sse4;
     }
     if (cpuMask & X265_CPU_AVX)
     {
diff -r da0b44e67fe0 -r 6da0a0291ed8 source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Mon Nov 11 14:36:21 2013 +0530
+++ b/source/common/x86/blockcopy8.asm	Mon Nov 11 14:58:09 2013 +0530
@@ -1743,3 +1743,46 @@
 movu       [r0 + r1],     m0
 
 RET
+
+;-----------------------------------------------------------------------------
+; void blockcopy_ps_%1x%2(int16_t *dest, intptr_t destStride, pixel *src, intptr_t srcStride);
+;-----------------------------------------------------------------------------
+%macro BLOCKCOPY_PS_W8_H4 2
+INIT_XMM sse4
+cglobal blockcopy_ps_%1x%2, 4, 5, 1, dest, destStride, src, srcStride
+
+add     r1,      r1
+mov    r4d,      %2/4
+
+.loop
+      movh       m0,            [r2]
+      pmovzxbw   m0,            m0
+      movu       [r0],          m0
+
+      movh       m0,            [r2 + r3]
+      pmovzxbw   m0,            m0
+      movu       [r0 + r1],     m0
+
+      movh       m0,            [r2 + 2 * r3]
+      pmovzxbw   m0,            m0
+      movu       [r0 + 2 * r1], m0
+
+      lea        r2,            [r2 + 2 * r3]
+      lea        r0,            [r0 + 2 * r1]
+
+      movh       m0,            [r2 + r3]
+      pmovzxbw   m0,            m0
+      movu       [r0 + r1],     m0
+
+      lea        r0,            [r0 + 2 * r1]
+      lea        r2,            [r2 + 2 * r3]
+
+      dec        r4d
+      jnz        .loop
+
+RET
+%endmacro
+
+BLOCKCOPY_PS_W8_H4  8,  8
+BLOCKCOPY_PS_W8_H4  8, 16
+BLOCKCOPY_PS_W8_H4  8, 32
diff -r da0b44e67fe0 -r 6da0a0291ed8 source/common/x86/blockcopy8.h
--- a/source/common/x86/blockcopy8.h	Mon Nov 11 14:36:21 2013 +0530
+++ b/source/common/x86/blockcopy8.h	Mon Nov 11 14:58:09 2013 +0530
@@ -96,7 +96,10 @@
 #define CHROMA_BLOCKCOPY_DEF_SSE4(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);
+    SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 6, cpu); \
+    SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 8, cpu); \
+    SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 16, cpu); \
+    SETUP_CHROMA_BLOCKCOPY_FUNC_SSE4(8, 32, cpu);
 
 CHROMA_BLOCKCOPY_DEF_SSE4(_sse4);