[x265] [PATCH] asm: optimize dct4

Steve Borho steve at borho.org
Tue Aug 26 19:09:31 CEST 2014


On 08/26, dnyaneshwar at multicorewareinc.com wrote:
> # HG changeset patch
> # User Dnyaneshwar G <dnyaneshwar at multicorewareinc.com>
> # Date 1409046621 -19800
> #      Tue Aug 26 15:20:21 2014 +0530
> # Node ID bbd5b3f269b095760d21877e94d67df8bd72f479
> # Parent  5acfb12ec5d17cc700e313fc99248e2408e5967b
> asm: optimize dct4

You need to provide more detail about how this is an optimization, at
the least the commit message should contain before/after numbers from
the testbench.

Better if it had mentioned replacing pshufd + pshufhw with pshufb

> diff -r 5acfb12ec5d1 -r bbd5b3f269b0 source/common/x86/dct8.asm
> --- a/source/common/x86/dct8.asm	Mon Aug 25 17:53:12 2014 +0900
> +++ b/source/common/x86/dct8.asm	Tue Aug 26 15:20:21 2014 +0530
> @@ -30,6 +30,8 @@
>  
>  SECTION_RODATA 32
>  
> +dct4_shuf:      db 0, 1, 2, 3, 8, 9, 10, 11, 6, 7, 4, 5, 14, 15, 12, 13
> +
>  tab_dct4:       times 4 dw 64, 64
>                  times 4 dw 83, 36
>                  times 4 dw 64, -64
> @@ -118,16 +120,14 @@
>      movh        m0, [r0 + 0 * r2]
>      movh        m1, [r0 + 1 * r2]
>      punpcklqdq  m0, m1
> -    pshufd      m0, m0, 0xD8
> -    pshufhw     m0, m0, 0xB1
> +    pshufb      m0, [dct4_shuf]
>  
>      lea         r0, [r0 + 2 * r2]
>      movh        m1, [r0]
>      movh        m2, [r0 + r2]
>      punpcklqdq  m1, m2
> -    pshufd      m1, m1, 0xD8
> -    pshufhw     m1, m1, 0xB1
>  
> +    pshufb      m1, [dct4_shuf]
>      punpcklqdq  m2, m0, m1
>      punpckhqdq  m0, m1
>  
> @@ -140,8 +140,7 @@
>      paddd       m3, m7
>      psrad       m3, DCT_SHIFT
>      packssdw    m0, m3
> -    pshufd      m0, m0, 0xD8
> -    pshufhw     m0, m0, 0xB1
> +    pshufb      m0, [dct4_shuf]
>      pmaddwd     m1, m6
>      paddd       m1, m7
>      psrad       m1, DCT_SHIFT
> @@ -149,9 +148,8 @@
>      paddd       m2, m7
>      psrad       m2, DCT_SHIFT
>      packssdw    m1, m2
> -    pshufd      m1, m1, 0xD8
> -    pshufhw     m1, m1, 0xB1
>  
> +    pshufb      m1, [dct4_shuf]
>      punpcklqdq  m2, m0, m1
>      punpckhqdq  m0, m1
>  
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel

-- 
Steve Borho


More information about the x265-devel mailing list