<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><div> </div><pre><br>At 2014-09-11 21:38:32,praveen@multicorewareinc.com wrote:
># HG changeset patch
># User Praveen Tiwari
># Date 1410442704 -19800
># Node ID df74723eb9a3861f6bba7f33d09a37efe53932a4
># Parent 9241634204a12babf8a2a90dc4f776646a9ddfb3
>copy_cnt_16: avx2 asm code as per new interface, improved 514.32 cycles -> 313.66 cycles
>
>diff -r 9241634204a1 -r df74723eb9a3 source/common/x86/blockcopy8.asm
>--- a/source/common/x86/blockcopy8.asm Thu Sep 11 17:33:44 2014 +0530
>+++ b/source/common/x86/blockcopy8.asm Thu Sep 11 19:08:24 2014 +0530
>@@ -4161,67 +4161,46 @@
> INIT_YMM avx2
> cglobal copy_cnt_16, 3,5,5
> add r2d, r2d
>- lea r4, [r2 * 3]
>- mov r3d, 16/4
>- ; NOTE: xorpd is faster than pxor
>+ lea r3, [r2 * 3]
>+ mov r4d, 16/4
>+
>+ movu m3, [pb_1]
</pre><pre>movu?</pre><pre> </pre><pre>>+ movu [r0 + 32], m1
>+ packsswb m0, m1
</pre><pre>ident</pre><pre> </pre></div>