[x265] [PATCH] asm: luma_hps[24x32] avx2 - improved 11545c->6874c

Steve Borho steve at borho.org
Tue Mar 17 15:14:03 CET 2015


On 03/17, aasaipriya at multicorewareinc.com wrote:
> # HG changeset patch
> # User Aasaipriya Chandran <aasaipriya at multicorewareinc.com>
> # Date 1426567349 -19800
> #      Tue Mar 17 10:12:29 2015 +0530
> # Node ID 7f63d7a2d7e37723d3dadf99aac3aa691d961b31
> # Parent  4e7713b3baba922a60c267efa4032e37dc6bc521
> asm: luma_hps[24x32] avx2 - improved 11545c->6874c

Revision under test:
changeset:   9953:6ac1567810b4
user:        Aasaipriya Chandran <aasaipriya at multicorewareinc.com>
date:        Tue Mar 17 10:12:29 2015 +0530
summary:     asm: luma_hps[24x32] avx2 - improved 11545c->6874c


Most recent output changing commit:
changeset:   9868:1bed2e325efc
user:        Aarthi Thirumalai
date:        Fri Mar 13 21:57:56 2015 +0530
summary:     rc: fix bug in CRF caused by e0f834c69cb2


NOTE: Revision under test is not public or has uncommited changes.
No new golden outputs will be generated during this run, neither
will it create pass/fail files.

building gcc...
Running testbench for gcc...


** testbench failure reported for gcc:: No pastebin key configured.
Contents:
system   : zeppelin
hardware : Haswell dual core HT Mac OS X Yosemite
generator: Unix Makefiles
options  : checked static tests {}
version  : 6ac1567810b4

Using random seed 55083688 8bpp
Testing primitives: SSE2
Testing primitives: SSE3
Testing primitives: SSSE3
Testing primitives: SSE4
Testing primitives: AVX
Testing primitives: AVX2
SIGSEGV


> diff -r 4e7713b3baba -r 7f63d7a2d7e3 source/common/x86/asm-primitives.cpp
> --- a/source/common/x86/asm-primitives.cpp	Tue Mar 17 10:09:02 2015 +0530
> +++ b/source/common/x86/asm-primitives.cpp	Tue Mar 17 10:12:29 2015 +0530
> @@ -1582,6 +1582,7 @@
>          p.pu[LUMA_16x4].luma_hps = x265_interp_8tap_horiz_ps_16x4_avx2;
>          p.pu[LUMA_16x32].luma_hps = x265_interp_8tap_horiz_ps_16x32_avx2;
>          p.pu[LUMA_16x64].luma_hps = x265_interp_8tap_horiz_ps_16x64_avx2;
> +        p.pu[LUMA_24x32].luma_hps = x265_interp_8tap_horiz_ps_24x32_avx2;
>  
>          p.chroma[X265_CSP_I420].pu[CHROMA_420_8x8].filter_hpp = x265_interp_4tap_horiz_pp_8x8_avx2;
>          p.chroma[X265_CSP_I420].pu[CHROMA_420_4x4].filter_hpp = x265_interp_4tap_horiz_pp_4x4_avx2;
> diff -r 4e7713b3baba -r 7f63d7a2d7e3 source/common/x86/ipfilter8.asm
> --- a/source/common/x86/ipfilter8.asm	Tue Mar 17 10:09:02 2015 +0530
> +++ b/source/common/x86/ipfilter8.asm	Tue Mar 17 10:12:29 2015 +0530
> @@ -17428,3 +17428,84 @@
>      dec               r4d
>      jnz               .loop
>      RET
> +
> +INIT_YMM avx2
> +cglobal interp_8tap_horiz_ps_24x32, 6, 7, 8
> +%ifdef PIC
> +    lea                         r6,                [tab_LumaCoeff]
> +    vpbroadcastq                m0,                [r6 + r4 * 8]
> +%else
> +    vpbroadcastq                m0,                [tab_LumaCoeff + r4 * 8]
> +%endif
> +    mova                        m6,                [tab_Lm + 32]
> +    mova                        m1,                [tab_Lm]
> +    mov                         r4,                32                           ;height
> +    add                         r3d,               r3d
> +    vbroadcasti128              m2,                [pw_2000]
> +    vbroadcasti128              m7,                [pw_1]
> +
> +    ; register map
> +    ; m0      - interpolate coeff
> +    ; m1 , m6 - shuffle order table
> +    ; m2      - pw_2000
> +
> +    sub                         r0,                3
> +    test                        r5d,               r5d
> +    jz                          .loop
> +    lea                         r6,                [r1 * 3]                     ; r6 = (N / 2 - 1) * srcStride
> +    sub                         r0,                r6                           ; r0(src)-r6
> +    add                         r4,                7                            ; blkheight += N - 1  (7 - 1 = 6 ; since the last one row not in loop)
> +
> +.loop
> +    ; Row 0
> +    vbroadcasti128              m3,                [r0]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
> +    pshufb                      m4,                m3,             m6           ; row 0 (col 4 to 7)
> +    pshufb                      m3,                m1                           ; shuffled based on the col order tab_Lm row 0 (col 0 to 3)
> +    pmaddubsw                   m3,                m0
> +    pmaddubsw                   m4,                m0
> +    pmaddwd                     m3,                m7
> +    pmaddwd                     m4,                m7
> +    packssdw                    m3,                m4
> +    vbroadcasti128              m4,                [r0 + 8]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
> +    pshufb                      m5,                m4,            m6            ;row 1 (col 4 to 7)
> +    pshufb                      m4,                m1                           ;row 1 (col 0 to 3)
> +    pmaddubsw                   m4,                m0
> +    pmaddubsw                   m5,                m0
> +    pmaddwd                     m4,                m7
> +    pmaddwd                     m5,                m7
> +    packssdw                    m4,                m5
> +
> +    pmaddwd                     m3,                m7
> +    pmaddwd                     m4,                m7
> +    packssdw                    m3,                m4
> +
> +    mova                        m5,                [interp8_hps_shuf]
> +    vpermd                      m3,                m5,               m3
> +    psubw                       m3,                m2
> +
> +    movu                        [r2],              m3                          ;row 0
> +
> +    vbroadcasti128              m3,                [r0 + 16]                         ; [x x x x x A 9 8 7 6 5 4 3 2 1 0]
> +    pshufb                      m4,                m3,          m6
> +    pshufb                      m3,                m1
> +    pmaddubsw                   m3,                m0
> +    pmaddubsw                   m4,                m0
> +    pmaddwd                     m3,                m7
> +    pmaddwd                     m4,                m7
> +    packssdw                    m3,                m4
> +
> +    pmaddwd                     m3,                m7
> +    pmaddwd                     m4,                m7
> +    packssdw                    m3,                m4
> +
> +    mova                        m4,                [interp8_hps_shuf]
> +    vpermd                      m3,                m4,            m3
> +    psubw                       m3,                m2
> +
> +    movu                        [r2 + 32],              xm3
> +
> +    add                         r0,                r1
> +    add                         r2,                r3
> +    dec                         r4d
> +    jnz                         .loop
> +    RET
> _______________________________________________
> 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