[x265] [PATCH] asm: luma_hps[32x8, 32x16, 32x24, 32x32, 32x64] avx2 - improved 6143c->3210c, 9343c->4767c, 12806c->6737c, 16065c->8648c, 29073c->15496c

Aasaipriya Chandran aasaipriya at multicorewareinc.com
Mon Mar 16 12:04:58 CET 2015


Okay let me check and send it again.


On Mon, Mar 16, 2015 at 4:32 PM, Deepthi Nandakumar <
deepthi at multicorewareinc.com> wrote:

> This crashes on my system. Pls check.
>
> On Mon, Mar 16, 2015 at 11:03 AM, <aasaipriya at multicorewareinc.com> wrote:
>
>> # HG changeset patch
>> # User Aasaipriya Chandran <aasaipriya at multicorewareinc.com>
>> # Date 1426483989 -19800
>> #      Mon Mar 16 11:03:09 2015 +0530
>> # Node ID b71432c674193a204de05e8957574431eaba3610
>> # Parent  19f6ba4dd66147698aa96deb2f364f30a7d63a7e
>> asm: luma_hps[32x8, 32x16, 32x24, 32x32, 32x64] avx2 - improved
>> 6143c->3210c, 9343c->4767c, 12806c->6737c, 16065c->8648c, 29073c->15496c
>>
>> diff -r 19f6ba4dd661 -r b71432c67419 source/common/x86/asm-primitives.cpp
>> --- a/source/common/x86/asm-primitives.cpp      Mon Mar 16 10:47:09 2015
>> +0530
>> +++ b/source/common/x86/asm-primitives.cpp      Mon Mar 16 11:03:09 2015
>> +0530
>> @@ -1576,6 +1576,12 @@
>>          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_32x32].luma_hps = x265_interp_8tap_horiz_ps_32x32_avx2;
>> +        p.pu[LUMA_32x16].luma_hps = x265_interp_8tap_horiz_ps_32x16_avx2;
>> +        p.pu[LUMA_32x24].luma_hps = x265_interp_8tap_horiz_ps_32x24_avx2;
>> +        p.pu[LUMA_32x8].luma_hps = x265_interp_8tap_horiz_ps_32x8_avx2;
>> +        p.pu[LUMA_32x64].luma_hps = x265_interp_8tap_horiz_ps_32x64_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;
>>          p.chroma[X265_CSP_I420].pu[CHROMA_420_32x32].filter_hpp =
>> x265_interp_4tap_horiz_pp_32x32_avx2;
>> diff -r 19f6ba4dd661 -r b71432c67419 source/common/x86/ipfilter8.asm
>> --- a/source/common/x86/ipfilter8.asm   Mon Mar 16 10:47:09 2015 +0530
>> +++ b/source/common/x86/ipfilter8.asm   Mon Mar 16 11:03:09 2015 +0530
>> @@ -2038,7 +2038,101 @@
>>  IPFILTER_LUMA_PS_16x_AVX2 16 , 32
>>  IPFILTER_LUMA_PS_16x_AVX2 16 , 64
>>
>> -
>> +%macro IPFILTER_LUMA_PS_32xN_AVX2 2
>> +INIT_YMM avx2
>> +cglobal interp_8tap_horiz_ps_%1x%2, 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                         r4d,                %2
>>          ;height
>> +    add                         r3d,               r3d
>> +    vbroadcasti128              m2,                [pw_1]
>> +    mova                        m7,                [interp8_hps_shuf]
>> +
>> +    ; register map
>> +    ; m0      - interpolate coeff
>> +    ; m1 , m6 - shuffle order table
>> +    ; m2      - pw_1
>> +
>> +    sub                         r0,                3
>> +    test                        r5d,               r5d
>> +    jz                          .loop
>> +    lea                         r6,                [r1 * 3]
>>        ; r8 = (N / 2 - 1) * srcStride
>> +    sub                         r0,                r6
>> +    add                         r4d,                7
>> +
>> +.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,                m2
>> +    pmaddwd                     m4,                m2
>> +    packssdw                    m3,                m4
>> +
>> +
>> +    vbroadcasti128              m4,                [r0 + 8]
>> +    pshufb                      m5,                m4,             m6
>>        ;row 0 (col 12 to 15)
>> +    pshufb                      m4,                m1
>>        ;row 0 (col 8 to 11)
>> +    pmaddubsw                   m4,                m0
>> +    pmaddubsw                   m5,                m0
>> +    pmaddwd                     m4,                m2
>> +    pmaddwd                     m5,                m2
>> +    packssdw                    m4,                m5
>> +
>> +    pmaddwd                     m3,                m2
>> +    pmaddwd                     m4,                m2
>> +    packssdw                    m3,                m4
>> +    vpermd                      m3,                m7,             m3
>> +    psubw                       m3,                [pw_2000]
>> +
>> +    movu                        [r2],              m3
>>       ;row 0
>> +
>> +    vbroadcasti128              m3,                [r0 + 16]
>> +    pshufb                      m4,                m3,             m6
>>        ; row 0 (col 20 to 23)
>> +    pshufb                      m3,                m1
>>        ; row 0 (col 16 to 19)
>> +    pmaddubsw                   m3,                m0
>> +    pmaddubsw                   m4,                m0
>> +    pmaddwd                     m3,                m2
>> +    pmaddwd                     m4,                m2
>> +    packssdw                    m3,                m4
>> +
>> +    vbroadcasti128              m4,                [r0 + 24]
>> +    pshufb                      m5,                m4,             m6
>>         ;row 0 (col 28 to 31)
>> +    pshufb                      m4,                m1
>>        ;row 0 (col 24 to 27)
>> +    pmaddubsw                   m4,                m0
>> +    pmaddubsw                   m5,                m0
>> +    pmaddwd                     m4,                m2
>> +    pmaddwd                     m5,                m2
>> +    packssdw                    m4,                m5
>> +
>> +    pmaddwd                     m3,                m2
>> +    pmaddwd                     m4,                m2
>> +    packssdw                    m3,                m4
>> +    vpermd                      m3,                m7,             m3
>> +    psubw                       m3,                [pw_2000]
>> +
>> +    movu                        [r2 + 32],         m3
>>       ;row 0
>> +
>> +    add                         r0,                r1
>> +    add                         r2,                r3
>> +    dec                         r4d
>> +    jnz                         .loop
>> +    RET
>> +%endmacro
>> +
>> +IPFILTER_LUMA_PS_32xN_AVX2 32 , 32
>> +IPFILTER_LUMA_PS_32xN_AVX2 32 , 16
>> +IPFILTER_LUMA_PS_32xN_AVX2 32 , 24
>> +IPFILTER_LUMA_PS_32xN_AVX2 32 , 8
>> +IPFILTER_LUMA_PS_32xN_AVX2 32 , 64
>>
>>  ;--------------------------------------------------------------------------------------------------------------
>>  ; void interp_8tap_horiz_pp_%1x%2(pixel *src, intptr_t srcStride, pixel
>> *dst, intptr_t dstStride, int coeffIdx)
>>
>>  ;--------------------------------------------------------------------------------------------------------------
>> _______________________________________________
>> x265-devel mailing list
>> x265-devel at videolan.org
>> https://mailman.videolan.org/listinfo/x265-devel
>>
>
>
> _______________________________________________
> 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/20150316/cd3bba10/attachment-0001.html>


More information about the x265-devel mailing list