[x265] [PATCH] blockcopy_pp_6x16 asm code optimization: eliminated branch instructions and optimize register uses

[chen]

8x code size just less than 20 cycles?
In really world, it will be slower.

At 2015-02-05 19:18:43,praveen at multicorewareinc.com wrote:
># HG changeset patch
># User Praveen Tiwari
># Date 1423135116 -19800
># Node ID 499eddf5c1e4dfcb8447d65cb0b48d633b3660a5
># Parent  d6c384f3625984a57335c679dbc4fcf37fd4f072
>blockcopy_pp_6x16 asm code optimization: eliminated branch instructions and optimize register uses
>
>improved, 411.14c -> 396.59c
>
>diff -r d6c384f36259 -r 499eddf5c1e4 source/common/x86/blockcopy8.asm
>--- a/source/common/x86/blockcopy8.asm	Thu Feb 05 16:27:13 2015 +0530
>+++ b/source/common/x86/blockcopy8.asm	Thu Feb 05 16:48:36 2015 +0530
>@@ -305,9 +305,9 @@
> ; void blockcopy_pp_6x16(pixel* dst, intptr_t dstStride, const pixel* src, intptr_t srcStride)
> ;-----------------------------------------------------------------------------
> INIT_XMM sse2
>-cglobal blockcopy_pp_6x16, 4, 7, 2
>-    mov     r6d,    16/2
>-.loop:
>+cglobal blockcopy_pp_6x16, 4, 6, 2
>+
>+%rep 8
>     movd    m0,     [r2]
>     mov     r4w,    [r2 + 4]
>     movd    m1,     [r2 + r3]
>@@ -318,8 +318,7 @@
>     movd    [r0 + r1],      m1
>     mov     [r0 + r1 + 4],  r5w
>     lea     r0,     [r0 + r1 * 2]
>-    dec     r6d
>-    jnz     .loop
>+%endrep
>     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/20150205/1863e48d/attachment.html>