<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><div>
<div></div>
<div id="divNeteaseMailCard"></div>I haven't haswell now, the code seem right, I will test it at next monday.<br>In 4x4, I suggest modify algorithm, eg:</div>
<div> </div>
<div>movd 0</div>
<div>pinsrd 1</div>
<div>pinsrd 2</div>
<div>pinsrd 3<br>pshufb -> [33 23 13 03 - 32 22 12 02 - 31 21 11 01 - 30 20 10 00]</div><pre><br>At 2014-11-27 19:24:04,"Divya Manivannan" <divya@multicorewareinc.com> wrote:</pre>
># HG changeset patch
># User Divya Manivannan <divya@multicorewareinc.com>
># Date 1417087358 -19800
># Thu Nov 27 16:52:38 2014 +0530
># Node ID 887d458dbe264afd5e8904c607c27a17309dc2e7
># Parent 973662a5e779d1760af93fc713606bc923155ec6
>asm: chroma_vpp[4x4] for color space i420 in avx2: improve 228c->186c
>
>diff -r 973662a5e779 -r 887d458dbe26 source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp Thu Nov 27 11:19:16 2014 +0530
>+++ b/source/common/x86/asm-primitives.cpp Thu Nov 27 16:52:38 2014 +0530
>@@ -1856,6 +1856,9 @@
> p.luma_vpp[LUMA_8x8] = x265_interp_8tap_vert_pp_8x8_avx2;
> p.luma_vpp[LUMA_8x16] = x265_interp_8tap_vert_pp_8x16_avx2;
> p.luma_vpp[LUMA_8x32] = x265_interp_8tap_vert_pp_8x32_avx2;
>+
>+ p.chroma[X265_CSP_I420].filter_vpp[CHROMA_4x4] = x265_interp_4tap_vert_pp_4x4_avx2;
>+> }
> #endif // if HIGH_BIT_DEPTH> }
>diff -r 973662a5e779 -r 887d458dbe26 source/common/x86/ipfilter8.asm
>--- a/source/common/x86/ipfilter8.asm Thu Nov 27 11:19:16 2014 +0530
>+++ b/source/common/x86/ipfilter8.asm Thu Nov 27 16:52:38 2014 +0530
>@@ -1901,6 +1901,56 @@> > RET>
>+INIT_YMM avx2
>+cglobal interp_4tap_vert_pp_4x4, 4, 6, 6
>+ mov r4d, r4m
>+ sub r0, r1
>+>+%ifdef PIC
>+ lea r5, [tab_ChromaCoeff]
>+ vpbroadcastd m0, [r5 + r4 * 4]
>+%else
>+ vpbroadcastd m0, [tab_ChromaCoeff + r4 * 4]
>+%endif>+
>+ lea r5, [r1 * 3]
>+ lea r4, [r3 * 3]
>+
>+ movd xm1, [r0] ; m1 = row0
>+ movd xm2, [r0 + r1] ; m2 = row1
>+ punpcklbw xm1, xm2 ; m1 = [13 03 12 02 11 01 10 00]
>+
>+ movd xm3, [r0 + r1 * 2] ; m3 = row2
>+ punpcklbw xm2, xm3 ; m2 = [23 13 22 12 21 11 20 10]
>+ movd xm4, [r0 + r5]
>+ punpcklbw xm3, xm4 ; m3 = [33 23 32 22 31 21 30 20]
>+ punpcklwd xm1, xm3 ; m1 = [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
>+
>+ lea r0, [r0 + r1 * 4]
>+ movd xm5, [r0] ; m5 = row4
>+ punpcklbw xm4, xm5 ; m4 = [43 33 42 32 41 31 40 30]
>+ punpcklwd xm2, xm4 ; m2 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10]
>+ vinserti128 m1, m1, xm2, 1 ; m1 = [43 33 21 13 42 32 22 12 41 31 21 11 40 30 20 10] - [33 23 13 03 32 22 12 02 31 21 11 01 30 20 10 00]
>+ movd xm2, [r0 + r1] ; m2 = row5
>+ punpcklbw xm5, xm2 ; m5 = [53 43 52 42 51 41 50 40]
>+ punpcklwd xm3, xm5 ; m3 = [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
>+ movd xm5, [r0 + r1 * 2] ; m5 = row6
>+ punpcklbw xm2, xm5 ; m2 = [63 53 62 52 61 51 60 50]
>+ punpcklwd xm4, xm2 ; m4 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30]
>+ vinserti128 m3, m3, xm4, 1 ; m3 = [63 53 43 33 62 52 42 32 61 51 41 31 60 50 40 30] - [53 43 44 23 52 42 32 22 51 41 31 21 50 40 30 20]
>+
>+ pmaddubsw m1, m0 ; m1 = DQWORD ROW[1 0]
>+ pmaddubsw m3, m0 ; m3 = DQWORD ROW[3 2]
>+ phaddw m1, m3 ; m1 = WORD ROW[3 1 2 0]
>+ pmulhrsw m1, [pw_512]
>+ vextracti128 xm2, m1, 1
>+ packuswb xm1, xm2
>+ movd [r2], xm1
>+ pextrd [r2 + r3], xm1, 2
>+ pextrd [r2 + r3 * 2], xm1, 1
>+ pextrd [r2 + r4], xm1, 3
>+ RET>+
> ;-----------------------------------------------------------------------------
> ; void interp_4tap_vert_pp_%1x%2(pixel *src, intptr_t srcStride, pixel *dst, intptr_t dstStride, int coeffIdx)
> ;-----------------------------------------------------------------------------
>diff -r 973662a5e779 -r 887d458dbe26 source/common/x86/ipfilter8.h
>--- a/source/common/x86/ipfilter8.h Thu Nov 27 11:19:16 2014 +0530
>+++ b/source/common/x86/ipfilter8.h Thu Nov 27 16:52:38 2014 +0530
>@@ -573,6 +573,7 @@
> SETUP_CHROMA_SS_FUNC_DEF(16, 64, cpu);
> > CHROMA_FILTERS(_sse4);>+CHROMA_FILTERS(_avx2);
> CHROMA_SP_FILTERS(_sse2);> CHROMA_SP_FILTERS_SSE4(_sse4);
> CHROMA_SS_FILTERS(_sse2);
>_______________________________________________
>x265-devel mailing list>x265-devel@videolan.org
>https://mailman.videolan.org/listinfo/x265-devel</div>