[x265] [PATCH] blockfill_s_8x8 sse2 asm code optimization
chen
chenm003 at 163.com
Mon Feb 2 12:09:44 CET 2015
At 2015-02-02 16:55:16,praveen at multicorewareinc.com wrote:
># HG changeset patch
># User Praveen Tiwari
># Date 1422867249 -19800
># Branch stable
># Node ID 2618352a21d5917ee8c1f79bcc159e858dd19daa
># Parent e2c958ff874e2bf8992ba22605e993530e8a2d8c
>blockfill_s_8x8 sse2 asm code optimization
>
>improved, 100.04c -> 90.05c
>
>diff -r e2c958ff874e -r 2618352a21d5 source/common/x86/blockcopy8.asm
>--- a/source/common/x86/blockcopy8.asm Sat Jan 31 13:48:34 2015 -0600
>+++ b/source/common/x86/blockcopy8.asm Mon Feb 02 14:24:09 2015 +0530
>@@ -1748,9 +1748,10 @@
> ; void blockfill_s_8x8(int16_t* dst, intptr_t dstride, int16_t val)
> ;-----------------------------------------------------------------------------
> INIT_XMM sse2
>-cglobal blockfill_s_8x8, 3, 3, 1, dst, dstStride, val
>+cglobal blockfill_s_8x8, 3, 4, 1, dst, dstStride, val
>
> add r1, r1
>+lea r3, [3 * r1]
>
> movd m0, r2d
> pshuflw m0, m0, 0
>@@ -1760,17 +1761,13 @@
> movu [r0 + r1], m0
> movu [r0 + 2 * r1], m0
>
>-lea r0, [r0 + 2 * r1]
>+movu [r0 + r3], m0
>+movu [r0 + 4 * r1], m0
>+
>+lea r0, [r0 + 4 * r1]
swap LEA and above movu, you will get less bytes on binary code
> movu [r0 + r1], m0
> movu [r0 + 2 * r1], m0
>-
>-lea r0, [r0 + 2 * r1]
>-movu [r0 + r1], m0
>-movu [r0 + 2 * r1], m0
>-
>-lea r0, [r0 + 2 * r1]
>-movu [r0 + r1], m0
>-
>+movu [r0 + r3], m0
> RET
>
> ;-----------------------------------------------------------------------------
>_______________________________________________
>x265-devel mailing list
>x265-devel at videolan.org
>https://mailman.videolan.org/listinfo/x265-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20150202/5f16aeb7/attachment.html>
More information about the x265-devel
mailing list