[x265] [PATCH] blockcopy_pp_12x32: SSE2 asm code optimization

[praveen at multicorewareinc.com]

# HG changeset patch
# User Praveen Tiwari
# Date 1423137543 -19800
# Node ID b10384b8c8a9a60fe37f4e5f3506673dcf00c004
# Parent  499eddf5c1e4dfcb8447d65cb0b48d633b3660a5
blockcopy_pp_12x32: SSE2 asm code optimization

diff -r 499eddf5c1e4 -r b10384b8c8a9 source/common/x86/blockcopy8.asm
--- a/source/common/x86/blockcopy8.asm	Thu Feb 05 16:48:36 2015 +0530
+++ b/source/common/x86/blockcopy8.asm	Thu Feb 05 17:29:03 2015 +0530
@@ -584,7 +584,55 @@
 
 BLOCKCOPY_PP_W12_H4 12, 16
 
-BLOCKCOPY_PP_W12_H4 12, 32
+;-----------------------------------------------------------------------------
+; void blockcopy_pp_12x32(pixel* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride)
+;-----------------------------------------------------------------------------
+INIT_XMM sse2
+cglobal blockcopy_pp_12x32, 4, 7, 8
+
+   lea     r5,      [3 * r1]
+   lea     r6,      [3 * r3]
+
+    movh    m0,     [r2]
+    movd    m1,     [r2 + 8]
+    movh    m2,     [r2 + r3]
+    movd    m3,     [r2 + r3 + 8]
+    movh    m4,     [r2 + 2 * r3]
+    movd    m5,     [r2 + 2 * r3 + 8]
+    movh    m6,     [r2 + r6]
+    movd    m7,     [r2 + r6 + 8]
+
+    movh    [r0],              m0
+    movd    [r0 + 8],          m1
+    movh    [r0 + r1],         m2
+    movd    [r0 + r1 + 8],     m3
+    movh    [r0 + 2 * r1],     m4
+    movd    [r0 + 2 * r1 + 8], m5
+    movh    [r0 + r5],         m6
+    movd    [r0 + r5 + 8],     m7
+
+    %rep 7
+    lea     r2,     [r2 + 4 * r3]
+    movh    m0,     [r2]
+    movd    m1,     [r2 + 8]
+    movh    m2,     [r2 + r3]
+    movd    m3,     [r2 + r3 + 8]
+    movh    m4,     [r2 + 2 * r3]
+    movd    m5,     [r2 + 2 * r3 + 8]
+    movh    m6,     [r2 + r6]
+    movd    m7,     [r2 + r6 + 8]
+
+    lea     r0,     [r0 + 4 * r1]
+    movh    [r0],              m0
+    movd    [r0 + 8],          m1
+    movh    [r0 + r1],         m2
+    movd    [r0 + r1 + 8],     m3
+    movh    [r0 + 2 * r1],     m4
+    movd    [r0 + 2 * r1 + 8], m5
+    movh    [r0 + r5],         m6
+    movd    [r0 + r5 + 8],     m7
+%endrep
+    RET
 
 ;-----------------------------------------------------------------------------
 ; void blockcopy_pp_16x4(pixel* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride)