[x265] [PATCH] asm: ssse3 8bpp code for chroma_p2s[8x6](4.74x) for i420

[rajesh at multicorewareinc.com]

# HG changeset patch
# User Rajesh Paulraj<rajesh at multicorewareinc.com>
# Date 1428313137 -19800
#      Mon Apr 06 15:08:57 2015 +0530
# Node ID e0e94b642a1f169658267ef17bea754a5af4a22d
# Parent  e7890da101691cc902ef6df66ee71351c613250e
asm: ssse3 8bpp code for chroma_p2s[8x6](4.74x) for i420

diff -r e7890da10169 -r e0e94b642a1f source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Mon Apr 06 15:06:06 2015 +0530
+++ b/source/common/x86/asm-primitives.cpp	Mon Apr 06 15:08:57 2015 +0530
@@ -1338,6 +1338,7 @@
 
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x2].p2s = x265_filterPixelToShort_8x2_ssse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x4].p2s = x265_filterPixelToShort_8x4_ssse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_8x6].p2s = x265_filterPixelToShort_8x6_ssse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].p2s = x265_filterPixelToShort_8x8_ssse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x16].p2s = x265_filterPixelToShort_8x16_ssse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_8x32].p2s = x265_filterPixelToShort_8x32_ssse3;
diff -r e7890da10169 -r e0e94b642a1f source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Mon Apr 06 15:06:06 2015 +0530
+++ b/source/common/x86/ipfilter8.asm	Mon Apr 06 15:08:57 2015 +0530
@@ -7893,6 +7893,56 @@
 ;-----------------------------------------------------------------------------
 ; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
 ;-----------------------------------------------------------------------------
+INIT_XMM ssse3
+cglobal filterPixelToShort_8x6, 3, 7, 5
+    mov         r3d, r3m
+    add         r3d, r3d
+    lea         r4, [r1 * 3]
+    lea         r5, [r1 * 5]
+    lea         r6, [r3 * 3]
+
+    ; load constant
+    mova        m3, [pb_128]
+    mova        m4, [tab_c_64_n64]
+
+    movh        m0, [r0]
+    punpcklbw   m0, m3
+    pmaddubsw   m0, m4
+
+    movh        m1, [r0 + r1]
+    punpcklbw   m1, m3
+    pmaddubsw   m1, m4
+
+    movh        m2, [r0 + r1 * 2]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 + r3 * 0], m0
+    movu        [r2 + r3 * 1], m1
+    movu        [r2 + r3 * 2], m2
+
+    movh        m0, [r0 + r4]
+    punpcklbw   m0, m3
+    pmaddubsw   m0, m4
+
+    movh        m1, [r0 + r1 * 4]
+    punpcklbw   m1, m3
+    pmaddubsw   m1, m4
+
+    movh        m2, [r0 + r5]
+    punpcklbw   m2, m3
+    pmaddubsw   m2, m4
+
+    movu        [r2 + r6 ], m0
+    movu        [r2 + r3 * 4], m1
+    lea         r2, [r2 + r3 * 4]
+    movu        [r2 + r3], m2
+
+    RET
+
+;-----------------------------------------------------------------------------
+; void filterPixelToShort(pixel *src, intptr_t srcStride, int16_t *dst, int16_t dstStride)
+;-----------------------------------------------------------------------------
 %macro P2S_H_16xN 1
 INIT_XMM ssse3
 cglobal filterPixelToShort_16x%1, 3, 7, 6
diff -r e7890da10169 -r e0e94b642a1f source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h	Mon Apr 06 15:06:06 2015 +0530
+++ b/source/common/x86/ipfilter8.h	Mon Apr 06 15:08:57 2015 +0530
@@ -578,7 +578,8 @@
     SETUP_CHROMA_P2S_FUNC_DEF(6, 8, cpu); 
 
 #define CHROMA_420_P2S_FILTERS_SSSE3(cpu) \
-    SETUP_CHROMA_P2S_FUNC_DEF(8, 2, cpu);
+    SETUP_CHROMA_P2S_FUNC_DEF(8, 2, cpu); \
+    SETUP_CHROMA_P2S_FUNC_DEF(8, 6, cpu);
 
 #define CHROMA_422_P2S_FILTERS_SSE4(cpu) \
     SETUP_CHROMA_P2S_FUNC_DEF(6, 16, cpu);