<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><div>code are right, just a little improve idea<br></div><pre><br>At 2015-10-23 16:28:35,dnyaneshwar@multicorewareinc.com wrote:
># HG changeset patch
># User Dnyaneshwar G <dnyaneshwar@multicorewareinc.com>
># Date 1445588852 -19800
># Fri Oct 23 13:57:32 2015 +0530
># Node ID 0fb5a67c2f5ea4f3fe1a7e0dcbc0c5c117dd6dfc
># Parent a7251c3e0ef810b95bb25be5371035208e36996d
>asm: fix intrapred_planar16x16 sse4 code for main12
>
>diff -r a7251c3e0ef8 -r 0fb5a67c2f5e source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp Thu Oct 22 09:12:28 2015 +0530
>+++ b/source/common/x86/asm-primitives.cpp Fri Oct 23 13:57:32 2015 +0530
>@@ -1145,8 +1145,9 @@
> p.cu[BLOCK_4x4].intra_pred[PLANAR_IDX] = PFX(intra_pred_planar4_sse4);
> p.cu[BLOCK_8x8].intra_pred[PLANAR_IDX] = PFX(intra_pred_planar8_sse4);
>
>+ p.cu[BLOCK_16x16].intra_pred[PLANAR_IDX] = PFX(intra_pred_planar16_sse4);
>+
> #if X265_DEPTH <= 10
>- p.cu[BLOCK_16x16].intra_pred[PLANAR_IDX] = PFX(intra_pred_planar16_sse4);
> p.cu[BLOCK_32x32].intra_pred[PLANAR_IDX] = PFX(intra_pred_planar32_sse4);
> #endif
> ALL_LUMA_TU_S(intra_pred[DC_IDX], intra_pred_dc, sse4);
>diff -r a7251c3e0ef8 -r 0fb5a67c2f5e source/common/x86/const-a.asm
>--- a/source/common/x86/const-a.asm Thu Oct 22 09:12:28 2015 +0530
>+++ b/source/common/x86/const-a.asm Fri Oct 23 13:57:32 2015 +0530
>@@ -122,6 +122,7 @@
> const pd_2, times 8 dd 2
> const pd_4, times 4 dd 4
> const pd_8, times 4 dd 8
>+const pd_15, times 8 dd 15
> const pd_16, times 8 dd 16
> const pd_31, times 4 dd 31
> const pd_32, times 8 dd 32
>@@ -136,7 +137,8 @@
> const pd_524416, times 4 dd 524416
> const pd_n32768, times 8 dd 0xffff8000
> const pd_n131072, times 4 dd 0xfffe0000
>-
>+const pd_planar16_mul0, times 1 dd 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
>+const pd_planar16_mul1, times 1 dd 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
> const trans8_shuf, times 1 dd 0, 4, 1, 5, 2, 6, 3, 7
>
> const popcnt_table
>diff -r a7251c3e0ef8 -r 0fb5a67c2f5e source/common/x86/intrapred16.asm
>--- a/source/common/x86/intrapred16.asm Thu Oct 22 09:12:28 2015 +0530
>+++ b/source/common/x86/intrapred16.asm Fri Oct 23 13:57:32 2015 +0530
>@@ -109,6 +109,7 @@
> cextern pw_16
> cextern pw_31
> cextern pw_32
>+cextern pd_15
> cextern pd_16
> cextern pd_31
> cextern pd_32
>@@ -123,6 +124,8 @@
> cextern pb_unpackwq1
> cextern pb_unpackwq2
> cextern pw_planar16_mul
>+cextern pd_planar16_mul0
>+cextern pd_planar16_mul1
> cextern pw_planar32_mul
>
> ;-----------------------------------------------------------------------------------
>@@ -2216,6 +2219,114 @@
> ; void intra_pred_planar(pixel* dst, intptr_t dstStride, pixel*srcPix, int, int filter)
> ;---------------------------------------------------------------------------------------
> INIT_XMM sse4
>+%if ARCH_X86_64 == 1 && BIT_DEPTH == 12
>+cglobal intra_pred_planar16, 3,5,12
>+ add r1d, r1d
>+
>+ pmovzxwd m2, [r2 + 2]
>+ pmovzxwd m7, [r2 + 10]
>+ pmovzxwd m10, [r2 + 18]
>+ pmovzxwd m0, [r2 + 26]
>+
>+ movzx r3d, word [r2 + 34] ; topRight = above[16]
>+ lea r4, [pd_planar16_mul1]
>+
>+ movd m3, r3d
>+ pshufd m3, m3, 0 ; topRight
>+
>+ pmulld m8, m3, [r4 + 3*mmsize] ; (x + 1) * topRight
>+ pmulld m4, m3, [r4 + 2*mmsize] ; (x + 1) * topRight
>+ pmulld m9, m3, [r4 + 1*mmsize] ; (x + 1) * topRight
>+ pmulld m3, m3, [r4 + 0*mmsize] ; (x + 1) * topRight</pre><pre>dword multiplication very slow, I suggest do it on [r4 + 0*mmsize] only, other can get from addition [m3*4]</pre><pre>
</pre></div>