<html>
  <head>
    <meta content="text/html; charset=windows-1252"
      http-equiv="Content-Type">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">On 03/04/2015 06:44 PM, chen wrote:<br>
    </div>
    <blockquote
      cite="mid:381449a2.318.14be7d1801f.Coremail.chenm003@163.com"
      type="cite">
      <div
        style="line-height:1.7;color:#000000;font-size:14px;font-family:arial">
        <div><br>
           </div>
        At 2015-03-05 10:03:59,dave <a class="moz-txt-link-rfc2396E" href="mailto:dtyx265@gmail.com"><dtyx265@gmail.com></a> wrote:<br>
        <blockquote id="isReplyContent" style="margin: 0px 0px 0px
          0.8ex; padding-left: 1ex; border-left-color: rgb(204, 204,
          204); border-left-width: 1px; border-left-style: solid;">
          <div class="moz-cite-prefix">On 03/04/2015 04:39 PM, chen
            wrote:<br>
          </div>
          <blockquote
            cite="mid:5d8c895d.f7.14be75f9949.Coremail.chenm003@163.com"
            type="cite">
            <div style="color: rgb(0, 0, 0); line-height: 1.7;
              font-family: arial; font-size: 14px;">
              <div> </div>
              <pre>At 2015-03-05 07:54:02,<a moz-do-not-send="true" class="moz-txt-link-abbreviated" href="mailto:dtyx265@gmail.com">dtyx265@gmail.com</a> wrote:
># HG changeset patch
># User David T Yuen <a moz-do-not-send="true" class="moz-txt-link-rfc2396E" href="mailto:dtyx265@gmail.com"><dtyx265@gmail.com></a>
># Date 1425512599 28800
># Node ID 16880e791046ef8470f8307b76aae57c3be573c1
># Parent  c53b456ad909eeab8d83f8e0817e641d174cc706
>asm:intra pred planar8 sse2
>
>This replaces c code for systems using ssse3 to sse2 processors
>The code is backported from intrapred planar8 sse4
>
>64-bit
>
>./test/TestBench --testbench intrapred | grep intra_planar_8x8
>intra_planar_8x8    3.34x    997.49          3330.22
>
>32-bit
>
>./test/TestBench --testbench intrapred | grep intra_planar_8x8
>intra_planar_8x8    3.87x    1052.48         4072.68
>
>diff -r c53b456ad909 -r 16880e791046 source/common/x86/asm-primitives.cpp
>--- a/source/common/x86/asm-primitives.cpp  Tue Mar 03 18:40:21 2015 -0800
>+++ b/source/common/x86/asm-primitives.cpp  Wed Mar 04 15:43:19 2015 -0800
>@@ -1210,6 +1210,7 @@
>         p.cu[BLOCK_32x32].intra_pred[DC_IDX] = x265_intra_pred_dc32_sse2;

>         p.cu[BLOCK_4x4].intra_pred[PLANAR_IDX] = x265_intra_pred_planar4_sse2;
>+        p.cu[BLOCK_8x8].intra_pred[PLANAR_IDX] = x265_intra_pred_planar8_sse2;

>         p.cu[BLOCK_4x4].calcresidual = x265_getResidual4_sse2;
>         p.cu[BLOCK_8x8].calcresidual = x265_getResidual8_sse2;
>diff -r c53b456ad909 -r 16880e791046 source/common/x86/intrapred.h
>--- a/source/common/x86/intrapred.h Tue Mar 03 18:40:21 2015 -0800
>+++ b/source/common/x86/intrapred.h Wed Mar 04 15:43:19 2015 -0800
>@@ -36,6 +36,7 @@
> void x265_intra_pred_dc32_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int filter);

> void x265_intra_pred_planar4_sse2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
>+void x265_intra_pred_planar8_sse2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
> void x265_intra_pred_planar4_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
> void x265_intra_pred_planar8_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
> void x265_intra_pred_planar16_sse4(pixel* dst, intptr_t dstStride, const pixel* srcPix, int, int);
>diff -r c53b456ad909 -r 16880e791046 source/common/x86/intrapred8.asm
>--- a/source/common/x86/intrapred8.asm      Tue Mar 03 18:40:21 2015 -0800
>+++ b/source/common/x86/intrapred8.asm      Wed Mar 04 15:43:19 2015 -0800
>@@ -124,6 +124,7 @@
> cextern pw_32
> cextern pw_257
> cextern pw_1024
>+cextern pw_00ff
> cextern pb_unpackbd1
> cextern multiL
> cextern multiH
>@@ -588,6 +589,63 @@
>     movd            [r0 + r1], m1
>     RET

>+;---------------------------------------------------------------------------------------
>+; void intra_pred_planar(pixel* dst, intptr_t dstStride, pixel*srcPix, int, int filter)
>+;---------------------------------------------------------------------------------------
>+INIT_XMM sse2
>+cglobal intra_pred_planar8, 3,3,6
>+    pxor            m0, m0
>+    movh            m1, [r2 + 1]
>+    punpcklbw       m1, m0
>+    movh            m2, [r2 + 17]
>+    punpcklbw       m2, m0
>+
>+    movd            m3, [r2 + 9]            ; topRight   = above[8];
>+    movd            m4, [r2 + 25]           ; bottomLeft = left[8];
>+
>+    pand            m3, [pw_00ff]
>+    pand            m4, [pw_00ff]
how about mov+and+movd? it is less cycles on documents

</pre>
            </div>
          </blockquote>
          <p> something like this?<br>
            <br>
                mov             r3d, [r2 + 9]<br>
                mov             r4d, [r2 + 25]<br>
                and             r3d, [pw_00ff]<br>
                and             r4d, [pw_00ff]<br>
                movd            m3, r3d<br>
                movd            m4, r4d</p>
          <p>[MC] why not movzx?<br>
          </p>
        </blockquote>
      </div>
    </blockquote>
    Better than mov+and+movd but not much<br>
    <br>
    ./test/TestBench --testbench intrapred | grep intra_planar_8x8<br>
    intra_planar_8x8    3.18x      1049.99       3335.82<br>
    <blockquote
      cite="mid:381449a2.318.14be7d1801f.Coremail.chenm003@163.com"
      type="cite">
      <div
        style="line-height:1.7;color:#000000;font-size:14px;font-family:arial">
        <blockquote id="isReplyContent" style="margin: 0px 0px 0px
          0.8ex; padding-left: 1ex; border-left-color: rgb(204, 204,
          204); border-left-width: 1px; border-left-style: solid;">
          <p>btw: and may use immediate constant 0xFF, otherthan memory
            constant<br>
          </p>
        </blockquote>
      </div>
    </blockquote>
    also slightly better..<br>
    <br>
    ./test/TestBench --testbench intrapred | grep intra_planar_8x8<br>
    intra_planar_8x8    3.18x      1049.99       3335.84<br>
    <br>
    I think it's the movement between general purpose and simd registers
    that's slowing it down.<br>
    <blockquote
      cite="mid:381449a2.318.14be7d1801f.Coremail.chenm003@163.com"
      type="cite">
      <div
        style="line-height:1.7;color:#000000;font-size:14px;font-family:arial">
        <blockquote id="isReplyContent" style="margin: 0px 0px 0px
          0.8ex; padding-left: 1ex; border-left-color: rgb(204, 204,
          204); border-left-width: 1px; border-left-style: solid;">
          <p> </p>
          <p><br>
            This performs slightly worse on my system<br>
            <br>
            ./test/TestBench --testbench intrapred | grep
            intra_planar_8x8<br>
            intra_planar_8x8    3.15x      1059.98       3335.38<br>
          </p>
          <blockquote
            cite="mid:5d8c895d.f7.14be75f9949.Coremail.chenm003@163.com"
            type="cite">
            <div style="color: rgb(0, 0, 0); line-height: 1.7;
              font-family: arial; font-size: 14px;">
              <pre>>+    pshuflw         m3, m3, 0x00
>+    pshuflw         m4, m4, 0x00
>+    pshufd          m3, m3, 0x44
>+    pshufd          m4, m4, 0x44
>+
>+    pmullw          m3, [multiL]            ; (x + 1) * topRight
>+    pmullw          m0, m1, [pw_planar8_1]  ; (blkSize - 1 - y) * above[x]
>+    paddw           m3, [pw_8]
>+    paddw           m3, m4
>+    paddw           m3, m0
>+    psubw           m4, m1
>+
>+%macro INTRA_PRED_PLANAR_8 1
>+%if (%1 < 4)
>+    pshuflw         m5, m2, 0x55 * %1
>+    pshufd          m5, m5, 0
>+%else
>+    pshufhw         m5, m2, 0x55 * (%1 - 4)
>+    pshufd          m5, m5, 0xAA
>+%endif
>+    pmullw          m5, [pw_planar8_0]
>+    paddw           m5, m3
>+    psraw           m5, 4
>+    packuswb        m5, m5
>+    movh            [r0], m5
>+%if (%1 < 7)
>+    paddw           m3, m4
>+    lea             r0, [r0 + r1]
>+%endif
>+%endmacro
>+
>+    INTRA_PRED_PLANAR_8 0
>+    INTRA_PRED_PLANAR_8 1
>+    INTRA_PRED_PLANAR_8 2
>+    INTRA_PRED_PLANAR_8 3
>+    INTRA_PRED_PLANAR_8 4
>+    INTRA_PRED_PLANAR_8 5
>+    INTRA_PRED_PLANAR_8 6
>+    INTRA_PRED_PLANAR_8 7
>+    RET
>+
> ;---------------------------------------------------------------------------------------------
> ; void intra_pred_dc(pixel* dst, intptr_t dstStride, pixel *srcPix, int dirMode, int bFilter)
> ;---------------------------------------------------------------------------------------------
>_______________________________________________
>x265-devel mailing list
><a moz-do-not-send="true" class="moz-txt-link-abbreviated" href="mailto:x265-devel@videolan.org">x265-devel@videolan.org</a>
><a moz-do-not-send="true" class="moz-txt-link-freetext" href="https://mailman.videolan.org/listinfo/x265-devel">https://mailman.videolan.org/listinfo/x265-devel</a>
</pre>
            </div>
            <br>
            <fieldset class="mimeAttachmentHeader"></fieldset>
            <br>
            <pre wrap="">_______________________________________________
x265-devel mailing list
<a moz-do-not-send="true" class="moz-txt-link-abbreviated" href="mailto:x265-devel@videolan.org">x265-devel@videolan.org</a>
<a moz-do-not-send="true" class="moz-txt-link-freetext" href="https://mailman.videolan.org/listinfo/x265-devel">https://mailman.videolan.org/listinfo/x265-devel</a>
</pre>
          </blockquote>
          <br>
        </blockquote>
      </div>
      <br>
      <fieldset class="mimeAttachmentHeader"></fieldset>
      <br>
      <pre wrap="">_______________________________________________
x265-devel mailing list
<a class="moz-txt-link-abbreviated" href="mailto:x265-devel@videolan.org">x265-devel@videolan.org</a>
<a class="moz-txt-link-freetext" href="https://mailman.videolan.org/listinfo/x265-devel">https://mailman.videolan.org/listinfo/x265-devel</a>
</pre>
    </blockquote>
    <br>
  </body>
</html>