[x265] [PATCH 7 of 8] asm: interp_4tap_horiz_pp_4x16_sse3

[dtyx265 at gmail.com]

# HG changeset patch
# User David T Yuen <dtyx265 at gmail.com>
# Date 1429216044 25200
# Node ID 707a65570cb93e06bb743bb11a9178a92ed7bfb3
# Parent  c6cdd6f4f47b7fa125d34abad2f93af1bcc97cce
asm: interp_4tap_horiz_pp_4x16_sse3

This replaces c code.

64-bit

./test/TestBench --testbench interp | grep "chroma_hpp\[ 4x16\]"
chroma_hpp[ 4x16]	2.85x 	 3297.50  	 9392.67
chroma_hpp[ 4x16]	2.85x 	 3297.50  	 9392.49

32-bit

./test/TestBench --testbench interp | grep "chroma_hpp\[ 4x16\]"
chroma_hpp[ 4x16]	2.83x 	 3362.50  	 9499.97
chroma_hpp[ 4x16]	2.82x 	 3363.26  	 9500.97

diff -r c6cdd6f4f47b -r 707a65570cb9 source/common/x86/asm-primitives.cpp
--- a/source/common/x86/asm-primitives.cpp	Thu Apr 16 13:22:16 2015 -0700
+++ b/source/common/x86/asm-primitives.cpp	Thu Apr 16 13:27:24 2015 -0700
@@ -1361,10 +1361,12 @@
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x2].filter_hpp = x265_interp_4tap_horiz_pp_4x2_sse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x4].filter_hpp = x265_interp_4tap_horiz_pp_4x4_sse3;
         p.chroma[X265_CSP_I420].pu[CHROMA_420_4x8].filter_hpp = x265_interp_4tap_horiz_pp_4x8_sse3;
+        p.chroma[X265_CSP_I420].pu[CHROMA_420_4x16].filter_hpp = x265_interp_4tap_horiz_pp_4x16_sse3;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_2x8].filter_hpp = x265_interp_4tap_horiz_pp_2x8_sse3;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_2x16].filter_hpp = x265_interp_4tap_horiz_pp_2x16_sse3;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_4x4].filter_hpp = x265_interp_4tap_horiz_pp_4x4_sse3;
         p.chroma[X265_CSP_I422].pu[CHROMA_422_4x8].filter_hpp = x265_interp_4tap_horiz_pp_4x8_sse3;
+        p.chroma[X265_CSP_I422].pu[CHROMA_422_4x16].filter_hpp = x265_interp_4tap_horiz_pp_4x16_sse3;
     }
     if (cpuMask & X265_CPU_SSSE3)
     {
diff -r c6cdd6f4f47b -r 707a65570cb9 source/common/x86/ipfilter8.asm
--- a/source/common/x86/ipfilter8.asm	Thu Apr 16 13:22:16 2015 -0700
+++ b/source/common/x86/ipfilter8.asm	Thu Apr 16 13:27:24 2015 -0700
@@ -518,6 +518,32 @@
 
 RET
 
+;-----------------------------------------------------------------------------
+; void interp_4tap_horiz_pp_4x16(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
+;-----------------------------------------------------------------------------
+INIT_XMM sse3
+cglobal interp_4tap_horiz_pp_4x16, 4, 6, 7, src, srcstride, dst, dststride
+mov         r4d,        r4m
+
+%ifdef PIC
+lea         r5,         [tabw_ChromaCoeff]
+movddup     m4,         [r5 + r4 * 8]
+%else
+movddup     m4,         [tabw_ChromaCoeff + r4 * 8]
+%endif
+
+%assign x 1
+%rep 8
+FILTER_H4_w4_2_sse2
+%if x < 8
+lea         srcq,       [srcq + srcstrideq * 2]
+lea         dstq,       [dstq + dststrideq * 2]
+%endif
+%assign x x+1
+%endrep
+
+RET
+
 %macro FILTER_H4_w2_2 3
     movh        %2, [srcq - 1]
     pshufb      %2, %2, Tm0
diff -r c6cdd6f4f47b -r 707a65570cb9 source/common/x86/ipfilter8.h
--- a/source/common/x86/ipfilter8.h	Thu Apr 16 13:22:16 2015 -0700
+++ b/source/common/x86/ipfilter8.h	Thu Apr 16 13:27:24 2015 -0700
@@ -769,6 +769,7 @@
 void x265_interp_4tap_horiz_pp_4x2_sse3(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
 void x265_interp_4tap_horiz_pp_4x4_sse3(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
 void x265_interp_4tap_horiz_pp_4x8_sse3(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
+void x265_interp_4tap_horiz_pp_4x16_sse3(const pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx);
 #undef LUMA_FILTERS
 #undef LUMA_SP_FILTERS
 #undef LUMA_SS_FILTERS