<div dir="ltr">Hello Chen,<div><br></div><div>Thanks for your comments. </div><div>I will work on the changes you mentioned.</div><div><br></div><div>Thanks,<br>Aasaipriya </div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Dec 10, 2014 at 1:16 AM, chen <span dir="ltr"><<a href="mailto:chenm003@163.com" target="_blank">chenm003@163.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><div> </div><pre><br>At 2014-12-09 17:58:14,<a href="mailto:aasaipriya@multicorewareinc.com" target="_blank">aasaipriya@multicorewareinc.com</a> wrote:
># HG changeset patch
># User Aasaipriya Chandran <a href="mailto:aasaipriya@multicorewareinc.com%3E%3E#%C2%A0Date%C2%A01418119075%C2%A0-19800%3E%23%C2%A0%C2%A0%C2%A0%C2%A0%C2%A0%C2%A0Tue%C2%A0Dec%C2%A009%C2%A015:27:55%C2%A02014%C2%A0+0530%3E%23%C2%A0Node%C2%A0ID%C2%A0896b7cb19a57387bccb16150a197ee8c6911c416%3E%23%C2%A0Parent%C2%A0%C2%A082622d89fe00d58da4e61740ad80f5cd43b4a607%3Easm:%C2%A0chroma_hpp[32x32]%C2%A0for%C2%A0colorspace%C2%A0i420%C2%A0in%C2%A0avx2%C2%A0improve%C2%A06189c-%3E4104c%3E" target="_blank">aasaipriya@multicorewareinc.com>
># Date 1418119075 -19800
>#      Tue Dec 09 15:27:55 2014 +0530
># Node ID 896b7cb19a57387bccb16150a197ee8c6911c416
># Parent  82622d89fe00d58da4e61740ad80f5cd43b4a607
>asm: chroma_hpp[32x32] for colorspace i420 in avx2 improve 6189c->4104c
>
</a>diff -r 82622d89fe00 -r 896b7cb19a57 source/common/x86/ipfilter8.asm
>--- a/source/common/x86/ipfilter8.asm      Tue Dec 09 15:15:14 2014 +0530
>+++ b/source/common/x86/ipfilter8.asm      Tue Dec 09 15:27:55 2014 +0530
>@@ -1515,6 +1515,61 @@
>     jnz               .loop
>     RET

>+INIT_YMM avx2
>+cglobal interp_4tap_horiz_pp_32x32, 4,6,6
>+    mov             r4d, r4m
>+
>+%ifdef PIC
>+    lea               r5,           [tab_ChromaCoeff]
>+    vpbroadcastd      m0,           [r5 + r4 * 4]
>+%else
>+    vpbroadcastd      m0,           [tab_ChromaCoeff + r4 * 4]
>+%endif
>+
>+    movu              m1,           [tab_Tm]
</pre><pre>constant is aligned</pre><div><div class="h5"><pre> </pre><pre>>+    vpbroadcastd      m2,           [pw_1]
>+
>+    ; register map
>+    ; m0 - interpolate coeff
>+    ; m1 - shuffle order table
>+    ; m2 - constant word 1
>+
>+    sub               r0,           1
>+    mov               r4d,          32
>+
>+.loop:
>+    ; Row 0
>+    vbroadcasti128    m3,           [r0]                        ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
>+    pshufb            m3,           m1
>+    pmaddubsw         m3,           m0
>+    pmaddwd           m3,           m2
>+    vbroadcasti128    m4,           [r0 + 8]
>+    pshufb            m4,           m1
>+    pmaddubsw         m4,           m0
>+    pmaddwd           m4,           m2
>+    packssdw          m3,           m4
>+    pmulhrsw          m3,           [pw_512]
>+    vbroadcasti128    m4,           [r0 + 16]
>+    pshufb            m4,           m1
>+    pmaddubsw         m4,           m0
>+    pmaddwd           m4,           m2
>+    vbroadcasti128    m5,           [r0 + 24]
>+    pshufb            m5,           m1
>+    pmaddubsw         m5,           m0
>+    pmaddwd           m5,           m2
>+    packssdw          m4,           m5
>+    pmulhrsw          m4,           [pw_512]
</pre></div></div><pre>we have more free register to buffer constant</pre><span class=""><pre> </pre><pre>>+    packuswb          m3,           m4
>+    mova              m5,           [interp_4tap_8x8_horiz_shuf]
>+    vpermd            m3,           m5,      m3
</pre><pre>>+    vextracti128      xm4,          m3,      1
</pre></span><pre>we can reorder above instuction/algorithm to replace MOVA and PERMD by PERMQ.</pre><span class=""><pre> </pre><pre>>+    movu              [r2],         xm3
>+    movu              [r2 + 16],    xm4
</pre></span><pre>why you extract from high 128bits and write to continuous memory?</pre><span class=""><pre>>+    lea               r2,           [r2 + r3]
>+    lea               r0,           [r0 + r1]
>+    dec               r4d
>+    jnz               .loop
>+    RET
</pre></span></div><br>_______________________________________________<br>
x265-devel mailing list<br>
<a href="mailto:x265-devel@videolan.org">x265-devel@videolan.org</a><br>
<a href="https://mailman.videolan.org/listinfo/x265-devel" target="_blank">https://mailman.videolan.org/listinfo/x265-devel</a><br>
<br></blockquote></div><br></div>