[x265] [PATCH 1 of 9] asm: intra_pred_ang4_17 improved by ~57% over SSE4

[Steve Borho]

On 04/03, praveen at multicorewareinc.com wrote:
> # HG changeset patch
> # User Praveen Tiwari <praveen at multicorewareinc.com>
> # Date 1428042839 -19800
> #      Fri Apr 03 12:03:59 2015 +0530
> # Node ID 5d3bda7126381312ad23bb7ceed46691bc4cb471
> # Parent  9a5fa67583feb6ffb7668f82632f7e93e5ec9415
> asm: intra_pred_ang4_17 improved by ~57% over SSE4

series queued

> AVX2:
> intra_ang_4x4[17]       11.06x   104.22          1152.57
> 
> SSE4:
> intra_ang_4x4[17]       4.70x    244.43          1148.92
> 
> diff -r 9a5fa67583fe -r 5d3bda712638 source/common/x86/asm-primitives.cpp
> --- a/source/common/x86/asm-primitives.cpp	Thu Apr 02 13:21:32 2015 -0500
> +++ b/source/common/x86/asm-primitives.cpp	Fri Apr 03 12:03:59 2015 +0530
> @@ -1633,6 +1633,7 @@
>          p.weight_sp = x265_weight_sp_avx2;
>  
>          // intra_pred functions
> +        p.cu[BLOCK_4x4].intra_pred[17] = x265_intra_pred_ang4_17_avx2;
>          p.cu[BLOCK_4x4].intra_pred[19] = x265_intra_pred_ang4_19_avx2;
>          p.cu[BLOCK_4x4].intra_pred[20] = x265_intra_pred_ang4_20_avx2;
>          p.cu[BLOCK_4x4].intra_pred[21] = x265_intra_pred_ang4_21_avx2;
> diff -r 9a5fa67583fe -r 5d3bda712638 source/common/x86/intrapred.h
> --- a/source/common/x86/intrapred.h	Thu Apr 02 13:21:32 2015 -0500
> +++ b/source/common/x86/intrapred.h	Fri Apr 03 12:03:59 2015 +0530
> @@ -184,6 +184,7 @@
>  DECL_ANG(32, 33, sse4);
>  
>  #undef DECL_ANG
> +void x265_intra_pred_ang4_17_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
>  void x265_intra_pred_ang4_19_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
>  void x265_intra_pred_ang4_20_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
>  void x265_intra_pred_ang4_21_avx2(pixel* dst, intptr_t dstStride, const pixel* srcPix, int dirMode, int bFilter);
> diff -r 9a5fa67583fe -r 5d3bda712638 source/common/x86/intrapred8.asm
> --- a/source/common/x86/intrapred8.asm	Thu Apr 02 13:21:32 2015 -0500
> +++ b/source/common/x86/intrapred8.asm	Fri Apr 03 12:03:59 2015 +0530
> @@ -515,6 +515,7 @@
>  intra_pred4_shuff2:      db 0, 1, 1, 2, 2, 3, 3, 4, 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5, 1, 2, 2, 3, 3, 4, 4, 5
>  intra_pred4_shuff31:     db 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5, 1, 2, 2, 3, 3, 4, 4, 5, 2, 3, 3, 4, 4, 5, 5, 6
>  intra_pred4_shuff33:     db 0, 1, 1, 2, 2, 3, 3, 4, 1, 2, 2, 3, 3, 4, 4, 5, 2, 3, 3, 4, 4, 5, 5, 6, 3, 4, 4, 5, 5, 6, 6, 7
> +intra_pred4_shuff17:     db 0, 9, 9, 10, 10, 11, 11, 12, 1, 0, 0, 9, 9, 10, 10, 11, 2, 1, 1, 0, 0, 9, 9, 10, 4, 2, 2, 1, 1, 0, 0, 9
>  intra_pred4_shuff19:     db 0, 1, 1, 2, 2, 3, 3, 4, 9, 0, 0, 1, 1, 2, 2, 3, 10, 9, 9, 0, 0, 1, 1, 2, 12, 10, 10, 9, 9, 0, 0, 1
>  intra_pred4_shuff20:     db 0, 1, 1, 2, 2, 3, 3, 4, 10, 0, 0, 1, 1, 2, 2, 3, 10, 0, 0, 1, 1, 2, 2, 3, 11, 10, 10, 0, 0, 1, 1, 2
>  intra_pred4_shuff21:     db 0, 1, 1, 2, 2, 3, 3, 4, 10, 0, 0, 1, 1, 2, 2, 3, 10, 0, 0, 1, 1, 2, 2, 3, 12, 10, 10, 0, 0, 1, 1, 2
> @@ -528,6 +529,7 @@
>  c_ang4_mode_31:          db 15, 17, 15, 17, 15, 17, 15, 17, 30, 2, 30, 2, 30, 2, 30, 2, 13, 19, 13, 19, 13, 19, 13, 19, 28, 4, 28, 4, 28, 4, 28, 4
>  c_ang4_mode_32:          db 11, 21, 11, 21, 11, 21, 11, 21, 22, 10, 22, 10, 22, 10, 22, 10, 1, 31, 1, 31, 1, 31, 1, 31, 12, 20, 12, 20, 12, 20, 12, 20
>  c_ang4_mode_33:          db 6, 26, 6, 26, 6, 26, 6, 26, 12, 20, 12, 20, 12, 20, 12, 20, 18, 14, 18, 14, 18, 14, 18, 14, 24, 8, 24, 8, 24, 8, 24, 8
> +c_ang4_mode_17:          db 26, 6, 26, 6, 26, 6, 26, 6, 20, 12, 20, 12, 20, 12, 20, 12, 14, 18, 14, 18, 14, 18, 14, 18, 8, 24, 8, 24, 8, 24, 8, 24
>  c_ang4_mode_19:          db 26, 6, 26, 6, 26, 6, 26, 6, 20, 12, 20, 12, 20, 12, 20, 12, 14, 18, 14, 18, 14, 18, 14, 18, 8, 24, 8, 24, 8, 24, 8, 24
>  c_ang4_mode_20:          db 21, 11, 21, 11, 21, 11, 21, 11, 10, 22, 10, 22, 10, 22, 10, 22, 31, 1, 31, 1, 31, 1, 31, 1, 20, 12, 20, 12, 20, 12, 20, 12
>  c_ang4_mode_21:          db 17, 15, 17, 15, 17, 15, 17, 15, 2, 30, 2, 30, 2, 30, 2, 30, 19, 13, 19, 13, 19, 13, 19, 13, 4, 28, 4, 28, 4, 28, 4, 28
> @@ -15860,6 +15862,24 @@
>      INTRA_PRED_STORE_4x4
>      RET
>  
> +INIT_YMM avx2
> +cglobal intra_pred_ang4_17, 3, 3, 1
> +    vbroadcasti128    m0, [r2]
> +    pshufb            m0, [intra_pred4_shuff17]
> +    pmaddubsw         m0, [c_ang4_mode_17]
> +    pmulhrsw          m0, [pw_1024]
> +    packuswb          m0, m0
> +
> +    vpermq            m0, m0, 00001000b
> +    pshufb            m0, [c_trans_4x4]
> +
> +    ;store
> +    movd              [r0], xm0
> +    pextrd            [r0 + r1], xm0, 1
> +    lea               r0, [r0 + 2 * r1]
> +    pextrd            [r0], xm0, 2
> +    pextrd            [r0 + r1], xm0, 3
> +    RET
>  
>  INIT_YMM avx2
>  cglobal intra_pred_ang4_19, 3, 3, 1
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> https://mailman.videolan.org/listinfo/x265-devel

-- 
Steve Borho