diff mbox series

[FFmpeg-devel,v2] swscale/x86/output.asm: add x86-optimized planer gbr yuv2anyX functions

Message ID 20211104042747.55-1-mindmark@gmail.com
State New
Headers show
Series [FFmpeg-devel,v2] swscale/x86/output.asm: add x86-optimized planer gbr yuv2anyX functions | expand

Checks

Context Check Description
andriy/make_x86 success Make finished
andriy/make_fate_x86 success Make fate finished
andriy/make_ppc success Make finished
andriy/make_fate_ppc success Make fate finished

Commit Message

Mark Reid Nov. 4, 2021, 4:27 a.m. UTC
From: Mark Reid <mindmark@gmail.com>

changes since v1:
* remove vex intruction on sse4 path
* some load/pack marcos use less intructions
* fixed some typos

yuv2gbrp_full_X_4_512_c: 12757.6
yuv2gbrp_full_X_4_512_sse2: 8946.6
yuv2gbrp_full_X_4_512_sse4: 5138.6
yuv2gbrp_full_X_4_512_avx2: 3889.6
yuv2gbrap_full_X_4_512_c: 15368.6
yuv2gbrap_full_X_4_512_sse2: 11916.1
yuv2gbrap_full_X_4_512_sse4: 6294.6
yuv2gbrap_full_X_4_512_avx2: 3477.1
yuv2gbrp9be_full_X_4_512_c: 14381.6
yuv2gbrp9be_full_X_4_512_sse2: 9139.1
yuv2gbrp9be_full_X_4_512_sse4: 5150.1
yuv2gbrp9be_full_X_4_512_avx2: 2834.6
yuv2gbrp9le_full_X_4_512_c: 12990.1
yuv2gbrp9le_full_X_4_512_sse2: 9118.1
yuv2gbrp9le_full_X_4_512_sse4: 5132.1
yuv2gbrp9le_full_X_4_512_avx2: 2833.1
yuv2gbrp10be_full_X_4_512_c: 14401.6
yuv2gbrp10be_full_X_4_512_sse2: 9133.1
yuv2gbrp10be_full_X_4_512_sse4: 5126.1
yuv2gbrp10be_full_X_4_512_avx2: 2837.6
yuv2gbrp10le_full_X_4_512_c: 12718.1
yuv2gbrp10le_full_X_4_512_sse2: 9106.1
yuv2gbrp10le_full_X_4_512_sse4: 5120.1
yuv2gbrp10le_full_X_4_512_avx2: 2826.1
yuv2gbrap10be_full_X_4_512_c: 18535.6
yuv2gbrap10be_full_X_4_512_sse2: 33617.6
yuv2gbrap10be_full_X_4_512_sse4: 6264.1
yuv2gbrap10be_full_X_4_512_avx2: 3422.1
yuv2gbrap10le_full_X_4_512_c: 16724.1
yuv2gbrap10le_full_X_4_512_sse2: 11787.1
yuv2gbrap10le_full_X_4_512_sse4: 6282.1
yuv2gbrap10le_full_X_4_512_avx2: 3441.6
yuv2gbrp12be_full_X_4_512_c: 13723.6
yuv2gbrp12be_full_X_4_512_sse2: 9128.1
yuv2gbrp12be_full_X_4_512_sse4: 7997.6
yuv2gbrp12be_full_X_4_512_avx2: 2844.1
yuv2gbrp12le_full_X_4_512_c: 12257.1
yuv2gbrp12le_full_X_4_512_sse2: 9107.6
yuv2gbrp12le_full_X_4_512_sse4: 5142.6
yuv2gbrp12le_full_X_4_512_avx2: 2837.6
yuv2gbrap12be_full_X_4_512_c: 18511.1
yuv2gbrap12be_full_X_4_512_sse2: 12156.6
yuv2gbrap12be_full_X_4_512_sse4: 6251.1
yuv2gbrap12be_full_X_4_512_avx2: 3444.6
yuv2gbrap12le_full_X_4_512_c: 16687.1
yuv2gbrap12le_full_X_4_512_sse2: 11785.1
yuv2gbrap12le_full_X_4_512_sse4: 6243.6
yuv2gbrap12le_full_X_4_512_avx2: 3446.1
yuv2gbrp14be_full_X_4_512_c: 13690.6
yuv2gbrp14be_full_X_4_512_sse2: 9120.6
yuv2gbrp14be_full_X_4_512_sse4: 5138.1
yuv2gbrp14be_full_X_4_512_avx2: 2843.1
yuv2gbrp14le_full_X_4_512_c: 14995.6
yuv2gbrp14le_full_X_4_512_sse2: 9119.1
yuv2gbrp14le_full_X_4_512_sse4: 5126.1
yuv2gbrp14le_full_X_4_512_avx2: 2843.1
yuv2gbrp16be_full_X_4_512_c: 12367.1
yuv2gbrp16be_full_X_4_512_sse2: 8233.6
yuv2gbrp16be_full_X_4_512_sse4: 4820.1
yuv2gbrp16be_full_X_4_512_avx2: 2666.6
yuv2gbrp16le_full_X_4_512_c: 10904.1
yuv2gbrp16le_full_X_4_512_sse2: 8214.1
yuv2gbrp16le_full_X_4_512_sse4: 4824.1
yuv2gbrp16le_full_X_4_512_avx2: 2629.1
yuv2gbrap16be_full_X_4_512_c: 26569.6
yuv2gbrap16be_full_X_4_512_sse2: 10884.1
yuv2gbrap16be_full_X_4_512_sse4: 5488.1
yuv2gbrap16be_full_X_4_512_avx2: 3272.1
yuv2gbrap16le_full_X_4_512_c: 14010.1
yuv2gbrap16le_full_X_4_512_sse2: 10562.1
yuv2gbrap16le_full_X_4_512_sse4: 5463.6
yuv2gbrap16le_full_X_4_512_avx2: 3255.1
yuv2gbrpf32be_full_X_4_512_c: 14524.1
yuv2gbrpf32be_full_X_4_512_sse2: 8552.6
yuv2gbrpf32be_full_X_4_512_sse4: 4636.1
yuv2gbrpf32be_full_X_4_512_avx2: 2474.6
yuv2gbrpf32le_full_X_4_512_c: 13060.6
yuv2gbrpf32le_full_X_4_512_sse2: 9682.6
yuv2gbrpf32le_full_X_4_512_sse4: 4298.1
yuv2gbrpf32le_full_X_4_512_avx2: 2453.1
yuv2gbrapf32be_full_X_4_512_c: 18629.6
yuv2gbrapf32be_full_X_4_512_sse2: 11363.1
yuv2gbrapf32be_full_X_4_512_sse4: 15201.6
yuv2gbrapf32be_full_X_4_512_avx2: 3727.1
yuv2gbrapf32le_full_X_4_512_c: 16677.6
yuv2gbrapf32le_full_X_4_512_sse2: 10221.6
yuv2gbrapf32le_full_X_4_512_sse4: 5693.6
yuv2gbrapf32le_full_X_4_512_avx2: 3656.6

---
 libswscale/x86/output.asm | 435 +++++++++++++++++++++++++++++++++++++-
 libswscale/x86/swscale.c  |  98 +++++++++
 tests/checkasm/Makefile   |   2 +-
 tests/checkasm/checkasm.c |   1 +
 tests/checkasm/checkasm.h |   1 +
 tests/checkasm/sw_gbrp.c  | 198 +++++++++++++++++
 tests/fate/checkasm.mak   |   1 +
 7 files changed, 734 insertions(+), 2 deletions(-)
 create mode 100644 tests/checkasm/sw_gbrp.c

--
2.29.2

Comments

Mark Reid Nov. 16, 2021, 6:58 p.m. UTC | #1
On Wed, Nov 3, 2021 at 9:28 PM <mindmark@gmail.com> wrote:

> From: Mark Reid <mindmark@gmail.com>
>
> changes since v1:
> * remove vex intruction on sse4 path
> * some load/pack marcos use less intructions
> * fixed some typos
>
> yuv2gbrp_full_X_4_512_c: 12757.6
> yuv2gbrp_full_X_4_512_sse2: 8946.6
> yuv2gbrp_full_X_4_512_sse4: 5138.6
> yuv2gbrp_full_X_4_512_avx2: 3889.6
> yuv2gbrap_full_X_4_512_c: 15368.6
> yuv2gbrap_full_X_4_512_sse2: 11916.1
> yuv2gbrap_full_X_4_512_sse4: 6294.6
> yuv2gbrap_full_X_4_512_avx2: 3477.1
> yuv2gbrp9be_full_X_4_512_c: 14381.6
> yuv2gbrp9be_full_X_4_512_sse2: 9139.1
> yuv2gbrp9be_full_X_4_512_sse4: 5150.1
> yuv2gbrp9be_full_X_4_512_avx2: 2834.6
> yuv2gbrp9le_full_X_4_512_c: 12990.1
> yuv2gbrp9le_full_X_4_512_sse2: 9118.1
> yuv2gbrp9le_full_X_4_512_sse4: 5132.1
> yuv2gbrp9le_full_X_4_512_avx2: 2833.1
> yuv2gbrp10be_full_X_4_512_c: 14401.6
> yuv2gbrp10be_full_X_4_512_sse2: 9133.1
> yuv2gbrp10be_full_X_4_512_sse4: 5126.1
> yuv2gbrp10be_full_X_4_512_avx2: 2837.6
> yuv2gbrp10le_full_X_4_512_c: 12718.1
> yuv2gbrp10le_full_X_4_512_sse2: 9106.1
> yuv2gbrp10le_full_X_4_512_sse4: 5120.1
> yuv2gbrp10le_full_X_4_512_avx2: 2826.1
> yuv2gbrap10be_full_X_4_512_c: 18535.6
> yuv2gbrap10be_full_X_4_512_sse2: 33617.6
> yuv2gbrap10be_full_X_4_512_sse4: 6264.1
> yuv2gbrap10be_full_X_4_512_avx2: 3422.1
> yuv2gbrap10le_full_X_4_512_c: 16724.1
> yuv2gbrap10le_full_X_4_512_sse2: 11787.1
> yuv2gbrap10le_full_X_4_512_sse4: 6282.1
> yuv2gbrap10le_full_X_4_512_avx2: 3441.6
> yuv2gbrp12be_full_X_4_512_c: 13723.6
> yuv2gbrp12be_full_X_4_512_sse2: 9128.1
> yuv2gbrp12be_full_X_4_512_sse4: 7997.6
> yuv2gbrp12be_full_X_4_512_avx2: 2844.1
> yuv2gbrp12le_full_X_4_512_c: 12257.1
> yuv2gbrp12le_full_X_4_512_sse2: 9107.6
> yuv2gbrp12le_full_X_4_512_sse4: 5142.6
> yuv2gbrp12le_full_X_4_512_avx2: 2837.6
> yuv2gbrap12be_full_X_4_512_c: 18511.1
> yuv2gbrap12be_full_X_4_512_sse2: 12156.6
> yuv2gbrap12be_full_X_4_512_sse4: 6251.1
> yuv2gbrap12be_full_X_4_512_avx2: 3444.6
> yuv2gbrap12le_full_X_4_512_c: 16687.1
> yuv2gbrap12le_full_X_4_512_sse2: 11785.1
> yuv2gbrap12le_full_X_4_512_sse4: 6243.6
> yuv2gbrap12le_full_X_4_512_avx2: 3446.1
> yuv2gbrp14be_full_X_4_512_c: 13690.6
> yuv2gbrp14be_full_X_4_512_sse2: 9120.6
> yuv2gbrp14be_full_X_4_512_sse4: 5138.1
> yuv2gbrp14be_full_X_4_512_avx2: 2843.1
> yuv2gbrp14le_full_X_4_512_c: 14995.6
> yuv2gbrp14le_full_X_4_512_sse2: 9119.1
> yuv2gbrp14le_full_X_4_512_sse4: 5126.1
> yuv2gbrp14le_full_X_4_512_avx2: 2843.1
> yuv2gbrp16be_full_X_4_512_c: 12367.1
> yuv2gbrp16be_full_X_4_512_sse2: 8233.6
> yuv2gbrp16be_full_X_4_512_sse4: 4820.1
> yuv2gbrp16be_full_X_4_512_avx2: 2666.6
> yuv2gbrp16le_full_X_4_512_c: 10904.1
> yuv2gbrp16le_full_X_4_512_sse2: 8214.1
> yuv2gbrp16le_full_X_4_512_sse4: 4824.1
> yuv2gbrp16le_full_X_4_512_avx2: 2629.1
> yuv2gbrap16be_full_X_4_512_c: 26569.6
> yuv2gbrap16be_full_X_4_512_sse2: 10884.1
> yuv2gbrap16be_full_X_4_512_sse4: 5488.1
> yuv2gbrap16be_full_X_4_512_avx2: 3272.1
> yuv2gbrap16le_full_X_4_512_c: 14010.1
> yuv2gbrap16le_full_X_4_512_sse2: 10562.1
> yuv2gbrap16le_full_X_4_512_sse4: 5463.6
> yuv2gbrap16le_full_X_4_512_avx2: 3255.1
> yuv2gbrpf32be_full_X_4_512_c: 14524.1
> yuv2gbrpf32be_full_X_4_512_sse2: 8552.6
> yuv2gbrpf32be_full_X_4_512_sse4: 4636.1
> yuv2gbrpf32be_full_X_4_512_avx2: 2474.6
> yuv2gbrpf32le_full_X_4_512_c: 13060.6
> yuv2gbrpf32le_full_X_4_512_sse2: 9682.6
> yuv2gbrpf32le_full_X_4_512_sse4: 4298.1
> yuv2gbrpf32le_full_X_4_512_avx2: 2453.1
> yuv2gbrapf32be_full_X_4_512_c: 18629.6
> yuv2gbrapf32be_full_X_4_512_sse2: 11363.1
> yuv2gbrapf32be_full_X_4_512_sse4: 15201.6
> yuv2gbrapf32be_full_X_4_512_avx2: 3727.1
> yuv2gbrapf32le_full_X_4_512_c: 16677.6
> yuv2gbrapf32le_full_X_4_512_sse2: 10221.6
> yuv2gbrapf32le_full_X_4_512_sse4: 5693.6
> yuv2gbrapf32le_full_X_4_512_avx2: 3656.6
>
> ---
>  libswscale/x86/output.asm | 435 +++++++++++++++++++++++++++++++++++++-
>  libswscale/x86/swscale.c  |  98 +++++++++
>  tests/checkasm/Makefile   |   2 +-
>  tests/checkasm/checkasm.c |   1 +
>  tests/checkasm/checkasm.h |   1 +
>  tests/checkasm/sw_gbrp.c  | 198 +++++++++++++++++
>  tests/fate/checkasm.mak   |   1 +
>  7 files changed, 734 insertions(+), 2 deletions(-)
>  create mode 100644 tests/checkasm/sw_gbrp.c
>
> diff --git a/libswscale/x86/output.asm b/libswscale/x86/output.asm
> index 52cf9f2c2e..ce0a2650b4 100644
> --- a/libswscale/x86/output.asm
> +++ b/libswscale/x86/output.asm
> @@ -38,7 +38,49 @@ pw_32:         times 8 dw 32
>  pd_255:        times 8 dd 255
>  pw_512:        times 8 dw 512
>  pw_1024:       times 8 dw 1024
> -
> +pd_65535_invf:             times 8 dd 0x37800080 ;1.0/65535.0
> +pd_yuv2gbrp16_start:       times 8 dd -0x40000000
> +pd_yuv2gbrp_y_start:       times 8 dd  (1 << 9)
> +pd_yuv2gbrp_uv_start:      times 8 dd  ((1 << 9) - (128 << 19))
> +pd_yuv2gbrp_a_start:       times 8 dd  (1 << 18)
> +pd_yuv2gbrp16_offset:      times 8 dd  0x10000  ;(1 << 16)
> +pd_yuv2gbrp16_round13:     times 8 dd  0x02000  ;(1 << 13)
> +pd_yuv2gbrp16_a_offset:    times 8 dd  0x20002000
> +pd_yuv2gbrp16_upper30:     times 8 dd  0x3FFFFFFF ;(1<<30) - 1
> +pd_yuv2gbrp16_upper27:     times 8 dd  0x07FFFFFF ;(1<<27) - 1
> +pd_yuv2gbrp16_upperC:      times 8 dd  0xC0000000
> +pb_pack_shuffle8:       db  0,  4,  8, 12, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                            0,  4,  8, 12, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1
> +pb_pack_shuffle16le:    db  0,  1,  4,  5, \
> +                            8,  9, 12, 13, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                            0,  1,  4,  5, \
> +                            8,  9, 12, 13
> +pb_pack_shuffle16be:    db  1,  0,  5,  4, \
> +                            9,  8, 13, 12, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                           -1, -1, -1, -1, \
> +                            1,  0,  5,  4, \
> +                            9,  8, 13, 12
> +pb_shuffle32be          db  3,  2,  1,  0, \
> +                            7,  6,  5,  4, \
> +                           11, 10,  9,  8, \
> +                           15, 14, 13, 12, \
> +                            3,  2,  1,  0, \
> +                            7,  6,  5,  4, \
> +                           11, 10,  9,  8, \
> +                           15, 14, 13, 12
>  yuv2nv12_shuffle_mask: times 2 db 0,  4,  8, 12, \
>                                   -1, -1, -1, -1, \
>                                   -1, -1, -1, -1, \
> @@ -549,3 +591,394 @@ yuv2nv12cX_fn yuv2nv12
>  yuv2nv12cX_fn yuv2nv21
>  %endif
>  %endif ; ARCH_X86_64
> +
>
> +;-----------------------------------------------------------------------------
> +; planar grb yuv2anyX functions
> +; void ff_yuv2<gbr_format>_full_X_<opt>(SwsContext *c, const int16_t
> *lumFilter,
> +;                                       const int16_t **lumSrcx, int
> lumFilterSize,
> +;                                       const int16_t *chrFilter, const
> int16_t **chrUSrcx,
> +;                                       const int16_t **chrVSrcx, int
> chrFilterSize,
> +;                                       const int16_t **alpSrcx, uint8_t
> **dest,
> +;                                       int dstW, int y)
>
> +;-----------------------------------------------------------------------------
> +
> +%if ARCH_X86_64
> +struc SwsContext
> +    .padding:           resb 40292 ; offsetof(SwsContext,
> yuv2rgb_y_offset)
> +    .yuv2rgb_y_offset:  resd 1
> +    .yuv2rgb_y_coeff:   resd 1
> +    .yuv2rgb_v2r_coeff: resd 1
> +    .yuv2rgb_v2g_coeff: resd 1
> +    .yuv2rgb_u2g_coeff: resd 1
> +    .yuv2rgb_u2b_coeff: resd 1
> +endstruc
> +
> +%define R m0
> +%define G m1
> +%define B m2
> +%define A m3
> +
> +%define Y m4
> +%define U m5
> +%define V m6
> +
> +; Clip a signed integer to an unsigned power of two range.
> +; av_clip_uintp2
> +; 1 - dest
> +; 2 - bit position to clip at
> +%macro CLIPP2 2
> +    ; (~a) >> 31 & ((1<<p) - 1);
> +    pcmpeqb m4, m4
> +    pxor m4, %1
> +    psrad m4, 31
> +    movu m5, [pd_yuv2gbrp16_upper%2]
> +    pand m4, m5
> +
> +    ; (a & ~((1<<p) - 1)) == 0
> +    pandn m5, %1
> +    pxor m6, m6
> +    pcmpeqd m5, m6
> +%if cpuflag(avx2)
> +    vpblendvb %1, m4, %1, m5
> +%else
> +    pxor %1, m4
> +    pand %1, m5
> +    pxor %1, m4
> +%endif
> +%endmacro
> +
> +; 1 - dest
> +; 2 - source
> +%macro LOAD16 2
> +    %if cpuflag(avx2)
> +        movu xm%1, %2
> +        vpmovsxwd m%1, xm%1
> +    %elif cpuflag(sse4)
> +        movsd m%1, %2
> +        pmovsxwd m%1, m%1
> +    %else
> +        movsd m%1, %2
> +        punpcklwd m%1, m%1
> +        psrad m%1, 16 ; sign extend
> +    %endif
> +%endmacro
> +
> +; 1 - dest
> +; 2 - source
> +; 3 - depth
> +%macro LOAD_PIXELS 3
> +    mov ptrq, [%2 + jq*8]
> +%if %3 >= 16
> +    movu m%1, [ptrq + xq*4]
> +%else
> +    LOAD16 %1, [ptrq + xq*2]
> +%endif
> +%endmacro
> +
> +; 1 - dest
> +; 2 - source
> +%macro STORE8 2
> +    mov ptrq, %1
> +    %if mmsize > 16
> +        pshufb m%2, [pb_pack_shuffle8]
> +        vextractf128 xm4, m%2, 1
> +        por xm%2, xm4
> +        movq [ptrq + xq], xm%2
> +    %else
> +        %if cpuflag(sse4)
> +            pshufb m%2, [pb_pack_shuffle8]
> +        %else
> +            psrldq m4, m%2, 3
> +            por m%2, m4
> +            psrldq m4, m%2, 6
> +            por m%2, m4
> +        %endif
> +        movd [ptrq + xq], m%2
> +    %endif
> +%endmacro
> +
> +; 1 - dest
> +; 2 - source
> +; 3 - is big endian
> +%macro STORE16 3
> +    mov ptrq, %1
> +    %if mmsize > 16
> +        %if %3 ; bigendian
> +            pshufb m%2, [pb_pack_shuffle16be]
> +        %else
> +            pshufb m%2, [pb_pack_shuffle16le]
> +        %endif
> +        vextractf128 xm4, m%2, 1
> +        por xm%2, xm4
> +        movu [ptrq + xq*2], xm%2
> +    %else
> +        %if cpuflag(sse4) && %3 ; bigendian
> +            pshufb m%2, [pb_pack_shuffle16be]
> +        %elif cpuflag(sse4)
> +            pshufb m%2, [pb_pack_shuffle16le]
> +        %else
> +            pshuflw m%2, m%2, (1 << 6 | 1 << 4 | 2 << 2 | 0 << 0)
> +            pshufhw m%2, m%2, (1 << 6 | 1 << 4 | 2 << 2 | 0 << 0)
> +            pshufd  m%2, m%2, (3 << 6 | 3 << 4 | 2 << 2 | 0 << 0)
> +            %if %3 ; bigendian
> +                psrlw  m4, m%2, 8
> +                psllw  m%2, 8
> +                por m%2, m4
> +            %endif
> +        %endif
> +        movq [ptrq + xq*2], m%2
> +    %endif
> +%endmacro
> +
> +%macro SWAP32 1
> +%if mmsize > 16 || cpuflag(sse4)
> +    pshufb m%1, [pb_shuffle32be]
> +%else
> +    psrlw  m4, m%1, 8
> +    psllw  m%1, 8
> +    por m%1, m4
> +    pshuflw m%1, m%1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0)
> +    pshufhw m%1, m%1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0)
> +%endif
> +%endmacro
> +
> +; 1 - dest
> +; 2 - source
> +; 3 - depth
> +; 4 - is big endian
> +%macro STORE_PIXELS 4
> +%if %3 > 16
> +    %if %4
> +        SWAP32 %2
> +    %endif
> +    mov ptrq, %1
> +    movu [ptrq + xq*4], m%2
> +%elif %3 > 8
> +    STORE16 %1, %2, %4
> +%else
> +    STORE8 %1, %2
> +%endif
> +%endmacro
> +
> +%macro PMULLO 3
> +%if cpuflag(sse4) || mmsize > 16
> +    pmulld %1, %2, %3
> +%else
> +    %ifidni %1, %2
> +    %else
> +        mova %1, %2
> +    %endif
> +    pshufd m7, %1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0) ; 0xb1
> +    pshufd m8, %3, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0) ; 0xb1
> +    pmuludq m7, m8
> +    pshufd  m7, m7, (3 << 6 | 1 << 4 | 2 << 2 | 0 << 0) ; 0xd8
> +    pmuludq %1, %3
> +    pshufd  %1, %1, (3 << 6 | 1 << 4 | 2 << 2 | 0 << 0) ; 0xd8
> +    punpckldq %1, m7
> +%endif
> +%endmacro
> +
> +; 1 - name
> +; 2 - depth
> +; 3 - has alpha
> +; 3 - is big endian
> +; 5 - is float
> +%macro yuv2gbrp_fn 5
> +%define DEPTH %2
> +%define HAS_ALPHA %3
> +%define IS_BE %4
> +%define FLOAT %5
> +%define SH (22 + 8 - DEPTH)
> +
> +%if DEPTH >= 16
> +    %define RGB_SHIFT 14
> +    %define A_SHIFT 14
> +%elif 22 != SH
> +    %define RGB_SHIFT SH
> +    %define A_SHIFT (SH-3)
> +%else
> +    %define RGB_SHIFT 22
> +    %define A_SHIFT 19
> +%endif
> +
> +%if DEPTH >= 16
> +    %define YUV_SHIFT 14
> +    %define Y_START  m9
> +    %define Y_ROUND [pd_yuv2gbrp16_round13]
> +    %define UV_START m9
> +    %define A_START  m9
> +    %define A_CLIP2P 30
> +%else
> +    %define YUV_SHIFT 10
> +    %define Y_START  [pd_yuv2gbrp_y_start]
> +    %define Y_ROUND  m9
> +    %define UV_START [pd_yuv2gbrp_uv_start]
> +    %define A_START  [pd_yuv2gbrp_a_start]
> +    %define A_CLIP2P 27
> +%endif
> +
> +cglobal yuv2%1_full_X, 12, 14, 16, ptr, lumFilter, lumSrcx,
> lumFilterSize, chrFilter, chrUSrcx, chrVSrcx, chrFilterSize, alpSrcx, dest,
> dstW, y, x, j
> +    VBROADCASTSS m10, dword [ptrq + SwsContext.yuv2rgb_y_offset]
> +    VBROADCASTSS m11, dword [ptrq + SwsContext.yuv2rgb_y_coeff]
> +    VBROADCASTSS m12, dword [ptrq + SwsContext.yuv2rgb_v2r_coeff]
> +    VBROADCASTSS m13, dword [ptrq + SwsContext.yuv2rgb_v2g_coeff]
> +    VBROADCASTSS m14, dword [ptrq + SwsContext.yuv2rgb_u2g_coeff]
> +    VBROADCASTSS m15, dword [ptrq + SwsContext.yuv2rgb_u2b_coeff]
> +
> +%if DEPTH >= 16
> +    movu m9, [pd_yuv2gbrp16_start]
> +%else
> +    mov xq, (1 << (SH-1))
> +    movq xm9, xq
> +    VBROADCASTSS m9, xm9
> +%endif
> +    xor xq, xq
> +
> +    %%loop_x:
> +        movu Y, Y_START
> +        movu U, UV_START
> +        movu V, UV_START
> +
> +        xor jq, jq
> +        %%loop_luma:
> +            movsx ptrd, word [lumFilterq + jq*2]
> +            movd xm0, ptrd
> +            VBROADCASTSS m0, xm0
> +            LOAD_PIXELS 1, lumSrcxq, DEPTH
> +            PMULLO m1, m1, m0
> +            paddd Y, m1
> +            inc jd
> +            cmp jd, lumFilterSized
> +            jl %%loop_luma
> +
> +%if HAS_ALPHA
> +        cmp alpSrcxq, 0
> +        je %%skip_alpha_load
> +            xor jq, jq
> +            movu A, A_START
> +            %%loop_alpha:
> +                movsx ptrd, word [lumFilterq + jq*2]
> +                movd xm0, ptrd
> +                VBROADCASTSS m0, xm0
> +                LOAD_PIXELS 1, alpSrcxq, DEPTH
> +                PMULLO m1, m1, m0
> +                paddd A, m1
> +                inc jd
> +                cmp jd, lumFilterSized
> +                jl %%loop_alpha
> +%if DEPTH >= 16
> +            psrad A, 1
> +            paddd A, [pd_yuv2gbrp16_a_offset]
> +%endif
> +        %%skip_alpha_load:
> +%endif
> +        xor jq, jq
> +        %%loop_chr:
> +            movsx ptrd, word [chrFilterq + jq*2]
> +            movd xm0, ptrd
> +            VBROADCASTSS m0, xm0
> +            LOAD_PIXELS 1, chrUSrcxq, DEPTH
> +            LOAD_PIXELS 2, chrVSrcxq, DEPTH
> +            PMULLO m1, m1, m0
> +            PMULLO m2, m2, m0
> +            paddd U, m1
> +            paddd V, m2
> +            inc jd
> +            cmp jd, chrFilterSized
> +            jl %%loop_chr
> +
> +        psrad Y, YUV_SHIFT
> +%if  DEPTH >= 16
> +        paddd Y, [pd_yuv2gbrp16_offset]
> +%endif
> +        psrad U, YUV_SHIFT
> +        psrad V, YUV_SHIFT
> +
> +        psubd  Y, m10    ; yuv2rgb_y_offset
> +        PMULLO Y, Y, m11 ; yuv2rgb_y_coeff
> +        paddd  Y, Y_ROUND
> +
> +        PMULLO R, V, m12 ; yuv2rgb_v2r_coeff
> +        PMULLO B, U, m15 ; yuv2rgb_u2b_coeff
> +
> +        PMULLO U, U, m14 ; yuv2rgb_u2g_coeff
> +        PMULLO V, V, m13 ; yuv2rgb_v2g_coeff
> +        paddd G, U, V
> +        paddd R, Y
> +        paddd G, Y
> +        paddd B, Y
> +
> +        CLIPP2 R, 30
> +        CLIPP2 G, 30
> +        CLIPP2 B, 30
> +
> +        psrad R, RGB_SHIFT
> +        psrad G, RGB_SHIFT
> +        psrad B, RGB_SHIFT
> +
> +%if FLOAT
> +        cvtdq2ps R, R
> +        cvtdq2ps G, G
> +        cvtdq2ps B, B
> +        mulps R, [pd_65535_invf]
> +        mulps G, [pd_65535_invf]
> +        mulps B, [pd_65535_invf]
> +%endif
> +        STORE_PIXELS [destq +  0], 1, DEPTH, IS_BE ; G
> +        STORE_PIXELS [destq +  8], 2, DEPTH, IS_BE ; B
> +        STORE_PIXELS [destq + 16], 0, DEPTH, IS_BE ; R
> +
> +%if HAS_ALPHA
> +        cmp alpSrcxq, 0
> +        je %%skip_alpha_store
> +            CLIPP2 A, A_CLIP2P
> +            psrad A, A_SHIFT
> +%if FLOAT
> +            cvtdq2ps A, A
> +            mulps A, [pd_65535_invf]
> +%endif
> +            STORE_PIXELS [destq + 24], 3, DEPTH, IS_BE
> +        %%skip_alpha_store:
> +%endif
> +        add xq, mmsize/4
> +        cmp xd, dstWd
> +        jl %%loop_x
> +
> +    RET
> +%endmacro
> +
> +%macro yuv2gbrp_fn_decl 2
> +INIT_%1 %2
> +yuv2gbrp_fn gbrp,        8, 0, 0, 0
> +yuv2gbrp_fn gbrap,       8, 1, 0, 0
> +yuv2gbrp_fn gbrp9le,     9, 0, 0, 0
> +yuv2gbrp_fn gbrp10le,   10, 0, 0, 0
> +yuv2gbrp_fn gbrap10le,  10, 1, 0, 0
> +yuv2gbrp_fn gbrp12le,   12, 0, 0, 0
> +yuv2gbrp_fn gbrap12le,  12, 1, 0, 0
> +yuv2gbrp_fn gbrp14le,   14, 0, 0, 0
> +yuv2gbrp_fn gbrp16le,   16, 0, 0, 0
> +yuv2gbrp_fn gbrap16le,  16, 1, 0, 0
> +yuv2gbrp_fn gbrpf32le,  32, 0, 0, 1
> +yuv2gbrp_fn gbrapf32le, 32, 1, 0, 1
> +
> +yuv2gbrp_fn gbrp9be,     9, 0, 1, 0
> +yuv2gbrp_fn gbrp10be,   10, 0, 1, 0
> +yuv2gbrp_fn gbrap10be,  10, 1, 1, 0
> +yuv2gbrp_fn gbrp12be,   12, 0, 1, 0
> +yuv2gbrp_fn gbrap12be,  12, 1, 1, 0
> +yuv2gbrp_fn gbrp14be,   14, 0, 1, 0
> +yuv2gbrp_fn gbrp16be,   16, 0, 1, 0
> +yuv2gbrp_fn gbrap16be,  16, 1, 1, 0
> +yuv2gbrp_fn gbrpf32be,  32, 0, 1, 1
> +yuv2gbrp_fn gbrapf32be, 32, 1, 1, 1
> +%endmacro
> +
> +yuv2gbrp_fn_decl XMM, sse2
> +yuv2gbrp_fn_decl XMM, sse4
> +
> +%if HAVE_AVX2_EXTERNAL
> +yuv2gbrp_fn_decl YMM, avx2
> +%endif
> +
> +%endif ; ARCH_X86_64
> \ No newline at end of file
> diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
> index 0848a31461..e8bc15b6ee 100644
> --- a/libswscale/x86/swscale.c
> +++ b/libswscale/x86/swscale.c
> @@ -350,6 +350,43 @@ void ff_yuv2 ## fmt ## cX_ ## opt(enum AVPixelFormat
> format, const uint8_t *dith
>
>  YUV2NV_DECL(nv12, avx2);
>  YUV2NV_DECL(nv21, avx2);
> +
> +#define YUV2GBRP_FN_DECL(fmt, opt)
>               \
> +void ff_yuv2##fmt##_full_X_ ##opt(SwsContext *c, const int16_t
> *lumFilter,           \
> +                                 const int16_t **lumSrcx, int
> lumFilterSize,         \
> +                                 const int16_t *chrFilter, const int16_t
> **chrUSrcx, \
> +                                 const int16_t **chrVSrcx, int
> chrFilterSize,        \
> +                                 const int16_t **alpSrcx, uint8_t
> **dest,            \
> +                                 int dstW, int y)
> +
> +#define YUV2GBRP_DECL(opt)     \
> +YUV2GBRP_FN_DECL(gbrp,       opt); \
> +YUV2GBRP_FN_DECL(gbrap,      opt); \
> +YUV2GBRP_FN_DECL(gbrp9le,    opt); \
> +YUV2GBRP_FN_DECL(gbrp10le,   opt); \
> +YUV2GBRP_FN_DECL(gbrap10le,  opt); \
> +YUV2GBRP_FN_DECL(gbrp12le,   opt); \
> +YUV2GBRP_FN_DECL(gbrap12le,  opt); \
> +YUV2GBRP_FN_DECL(gbrp14le,   opt); \
> +YUV2GBRP_FN_DECL(gbrp16le,   opt); \
> +YUV2GBRP_FN_DECL(gbrap16le,  opt); \
> +YUV2GBRP_FN_DECL(gbrpf32le,  opt); \
> +YUV2GBRP_FN_DECL(gbrapf32le, opt); \
> +YUV2GBRP_FN_DECL(gbrp9be,    opt); \
> +YUV2GBRP_FN_DECL(gbrp10be,   opt); \
> +YUV2GBRP_FN_DECL(gbrap10be,  opt); \
> +YUV2GBRP_FN_DECL(gbrp12be,   opt); \
> +YUV2GBRP_FN_DECL(gbrap12be,  opt); \
> +YUV2GBRP_FN_DECL(gbrp14be,   opt); \
> +YUV2GBRP_FN_DECL(gbrp16be,   opt); \
> +YUV2GBRP_FN_DECL(gbrap16be,  opt); \
> +YUV2GBRP_FN_DECL(gbrpf32be,  opt); \
> +YUV2GBRP_FN_DECL(gbrapf32be, opt);
> +
> +YUV2GBRP_DECL(sse2);
> +YUV2GBRP_DECL(sse4);
> +YUV2GBRP_DECL(avx2);
> +
>  #endif
>
>  av_cold void ff_sws_init_swscale_x86(SwsContext *c)
> @@ -582,5 +619,66 @@ switch(c->dstBpc){ \
>              break;
>          }
>      }
> +
> +    if(c->flags & SWS_FULL_CHR_H_INT) {
> +
> +        /* yuv2gbrp uses the SwsContext for yuv coefficients
> +           if struct offsets change the asm needs to be updated too */
> +        av_assert0(offsetof(SwsContext, yuv2rgb_y_offset) == 40292);
> +
> +#define YUV2ANYX_FUNC_CASE(fmt, name, opt)              \
> +        case fmt:                                       \
> +            c->yuv2anyX = ff_yuv2##name##_full_X_##opt; \
> +            break;
> +
> +#define YUV2ANYX_GBRAP_CASES(opt)                                  \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP,       gbrp,       opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP,      gbrap,      opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP9LE,    gbrp9le,    opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP10LE,   gbrp10le,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP10LE,  gbrap10le,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP12LE,   gbrp12le,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP12LE,  gbrap12le,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP14LE,   gbrp14le,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP16LE,   gbrp16le,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP16LE,  gbrap16le,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRPF32LE,  gbrpf32le,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAPF32LE, gbrapf32le, opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP9BE,    gbrp9be,    opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP10BE,   gbrp10be,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP10BE,  gbrap10be,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP12BE,   gbrp12be,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP12BE,  gbrap12be,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP14BE,   gbrp14be,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP16BE,   gbrp16be,   opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP16BE,  gbrap16be,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRPF32BE,  gbrpf32be,  opt) \
> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAPF32BE, gbrapf32be, opt)
> +
> +        if (EXTERNAL_SSE2(cpu_flags)) {
> +            switch (c->dstFormat) {
> +            YUV2ANYX_GBRAP_CASES(sse2)
> +            default:
> +                break;
> +            }
> +        }
> +
> +        if (EXTERNAL_SSE4(cpu_flags)) {
> +            switch (c->dstFormat) {
> +            YUV2ANYX_GBRAP_CASES(sse4)
> +            default:
> +                break;
> +            }
> +        }
> +
> +        if (EXTERNAL_AVX2_FAST(cpu_flags)) {
> +            switch (c->dstFormat) {
> +            YUV2ANYX_GBRAP_CASES(avx2)
> +            default:
> +                break;
> +            }
> +        }
> +    }
> +
>  #endif
>  }
> diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> index 4ef5fa87da..f768b1144e 100644
> --- a/tests/checkasm/Makefile
> +++ b/tests/checkasm/Makefile
> @@ -45,7 +45,7 @@ AVFILTEROBJS-$(CONFIG_NLMEANS_FILTER)    += vf_nlmeans.o
>  CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes)
>
>  # swscale tests
> -SWSCALEOBJS                             += sw_rgb.o sw_scale.o
> +SWSCALEOBJS                             += sw_gbrp.o sw_rgb.o sw_scale.o
>
>  CHECKASMOBJS-$(CONFIG_SWSCALE)  += $(SWSCALEOBJS)
>
> diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
> index b1353f7cbe..5fa8ac7bcc 100644
> --- a/tests/checkasm/checkasm.c
> +++ b/tests/checkasm/checkasm.c
> @@ -183,6 +183,7 @@ static const struct {
>      #endif
>  #endif
>  #if CONFIG_SWSCALE
> +    { "sw_gbrp", checkasm_check_sw_gbrp },
>      { "sw_rgb", checkasm_check_sw_rgb },
>      { "sw_scale", checkasm_check_sw_scale },
>  #endif
> diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
> index 68b0697d3e..c3192d8c23 100644
> --- a/tests/checkasm/checkasm.h
> +++ b/tests/checkasm/checkasm.h
> @@ -72,6 +72,7 @@ void checkasm_check_opusdsp(void);
>  void checkasm_check_pixblockdsp(void);
>  void checkasm_check_sbrdsp(void);
>  void checkasm_check_synth_filter(void);
> +void checkasm_check_sw_gbrp(void);
>  void checkasm_check_sw_rgb(void);
>  void checkasm_check_sw_scale(void);
>  void checkasm_check_utvideodsp(void);
> diff --git a/tests/checkasm/sw_gbrp.c b/tests/checkasm/sw_gbrp.c
> new file mode 100644
> index 0000000000..320948a71d
> --- /dev/null
> +++ b/tests/checkasm/sw_gbrp.c
> @@ -0,0 +1,198 @@
> +/*
> + *
> + * This file is part of FFmpeg.
> + *
> + * FFmpeg is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License as published by
> + * the Free Software Foundation; either version 2 of the License, or
> + * (at your option) any later version.
> + *
> + * FFmpeg is distributed in the hope that it will be useful,
> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> + * GNU General Public License for more details.
> + *
> + * You should have received a copy of the GNU General Public License along
> + * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
> + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
> + */
> +
> +#include <string.h>
> +
> +#include "libavutil/common.h"
> +#include "libavutil/intreadwrite.h"
> +#include "libavutil/mem_internal.h"
> +#include "libavutil/pixdesc.h"
> +
> +#include "libswscale/swscale.h"
> +#include "libswscale/swscale_internal.h"
> +
> +#include "checkasm.h"
> +
> +#define randomize_buffers(buf, size)      \
> +    do {                                  \
> +        int j;                            \
> +        for (j = 0; j < size; j+=4)       \
> +            AV_WN32(buf + j, rnd());      \
> +    } while (0)
> +
> +static const int planar_fmts[] = {
> +    AV_PIX_FMT_GBRP,
> +    AV_PIX_FMT_GBRP9BE,
> +    AV_PIX_FMT_GBRP9LE,
> +    AV_PIX_FMT_GBRP10BE,
> +    AV_PIX_FMT_GBRP10LE,
> +    AV_PIX_FMT_GBRP12BE,
> +    AV_PIX_FMT_GBRP12LE,
> +    AV_PIX_FMT_GBRP14BE,
> +    AV_PIX_FMT_GBRP14LE,
> +    AV_PIX_FMT_GBRAP,
> +    AV_PIX_FMT_GBRAP10BE,
> +    AV_PIX_FMT_GBRAP10LE,
> +    AV_PIX_FMT_GBRAP12BE,
> +    AV_PIX_FMT_GBRAP12LE,
> +    AV_PIX_FMT_GBRP16BE,
> +    AV_PIX_FMT_GBRP16LE,
> +    AV_PIX_FMT_GBRAP16BE,
> +    AV_PIX_FMT_GBRAP16LE,
> +    AV_PIX_FMT_GBRPF32BE,
> +    AV_PIX_FMT_GBRPF32LE,
> +    AV_PIX_FMT_GBRAPF32BE,
> +    AV_PIX_FMT_GBRAPF32LE
> +};
> +
> +static void check_output_yuv2gbrp(void)
> +{
> +    struct SwsContext *ctx;
> +    const AVPixFmtDescriptor *desc;
> +    int fmi, fsi, isi, i;
> +    int dstW, byte_size, luma_filter_size, chr_filter_size;
> +#define LARGEST_FILTER 16
> +#define FILTER_SIZES 4
> +    static const int filter_sizes[FILTER_SIZES] = {1, 4, 8, 16};
> +#define LARGEST_INPUT_SIZE 512
> +#define INPUT_SIZES 6
> +    static const int input_sizes[INPUT_SIZES] = {8, 24, 128, 144, 256,
> 512};
> +    uint8_t *dst0[4];
> +    uint8_t *dst1[4];
> +
> +    declare_func(void, void *c, const int16_t *lumFilter,
> +                       const int16_t **lumSrcx, int lumFilterSize,
> +                       const int16_t *chrFilter, const int16_t **chrUSrcx,
> +                       const int16_t **chrVSrcx, int chrFilterSize,
> +                       const int16_t **alpSrcx, uint8_t **dest,
> +                       int dstW, int y);
> +
> +    const int16_t *luma[LARGEST_FILTER];
> +    const int16_t *chru[LARGEST_FILTER];
> +    const int16_t *chrv[LARGEST_FILTER];
> +    const int16_t *alpha[LARGEST_FILTER];
> +
> +    LOCAL_ALIGNED_8(int16_t, luma_filter, [LARGEST_FILTER]);
> +    LOCAL_ALIGNED_8(int16_t, chr_filter, [LARGEST_FILTER]);
> +
> +    LOCAL_ALIGNED_8(int32_t, src_y, [LARGEST_FILTER *
> LARGEST_INPUT_SIZE]);
> +    LOCAL_ALIGNED_8(int32_t, src_u, [LARGEST_FILTER *
> LARGEST_INPUT_SIZE]);
> +    LOCAL_ALIGNED_8(int32_t, src_v, [LARGEST_FILTER *
> LARGEST_INPUT_SIZE]);
> +    LOCAL_ALIGNED_8(int32_t, src_a, [LARGEST_FILTER *
> LARGEST_INPUT_SIZE]);
> +
> +    LOCAL_ALIGNED_8(uint8_t, dst0_r, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +    LOCAL_ALIGNED_8(uint8_t, dst0_g, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +    LOCAL_ALIGNED_8(uint8_t, dst0_b, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +    LOCAL_ALIGNED_8(uint8_t, dst0_a, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +
> +    LOCAL_ALIGNED_8(uint8_t, dst1_r, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +    LOCAL_ALIGNED_8(uint8_t, dst1_g, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +    LOCAL_ALIGNED_8(uint8_t, dst1_b, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +    LOCAL_ALIGNED_8(uint8_t, dst1_a, [LARGEST_INPUT_SIZE *
> sizeof(int32_t)]);
> +
> +    randomize_buffers((uint8_t*)src_y, LARGEST_FILTER *
> LARGEST_INPUT_SIZE * sizeof(int32_t));
> +    randomize_buffers((uint8_t*)src_u, LARGEST_FILTER *
> LARGEST_INPUT_SIZE * sizeof(int32_t));
> +    randomize_buffers((uint8_t*)src_v, LARGEST_FILTER *
> LARGEST_INPUT_SIZE * sizeof(int32_t));
> +    randomize_buffers((uint8_t*)src_a, LARGEST_FILTER *
> LARGEST_INPUT_SIZE * sizeof(int32_t));
> +    randomize_buffers((uint8_t*)luma_filter, LARGEST_FILTER *
> sizeof(int16_t));
> +    randomize_buffers((uint8_t*)chr_filter, LARGEST_FILTER *
> sizeof(int16_t));
> +
> +    dst0[0] = (uint8_t*)dst0_g;
> +    dst0[1] = (uint8_t*)dst0_b;
> +    dst0[2] = (uint8_t*)dst0_r;
> +    dst0[3] = (uint8_t*)dst0_a;
> +
> +    dst1[0] = (uint8_t*)dst1_g;
> +    dst1[1] = (uint8_t*)dst1_b;
> +    dst1[2] = (uint8_t*)dst1_r;
> +    dst1[3] = (uint8_t*)dst1_a;
> +
> +    for (i = 0; i < LARGEST_FILTER; i++) {
> +        luma[i] =  (int16_t *)(src_y + i*LARGEST_INPUT_SIZE);
> +        chru[i] =  (int16_t *)(src_u + i*LARGEST_INPUT_SIZE);
> +        chrv[i] =  (int16_t *)(src_v + i*LARGEST_INPUT_SIZE);
> +        alpha[i] = (int16_t *)(src_a + i*LARGEST_INPUT_SIZE);
> +    }
> +
> +    ctx = sws_alloc_context();
> +    if (sws_init_context(ctx, NULL, NULL) < 0)
> +        fail();
> +
> +    ctx->flags |= SWS_FULL_CHR_H_INT;
> +    ctx->yuv2rgb_y_offset  = rnd();
> +    ctx->yuv2rgb_y_coeff   = rnd();
> +    ctx->yuv2rgb_v2r_coeff = rnd();
> +    ctx->yuv2rgb_v2g_coeff = rnd();
> +    ctx->yuv2rgb_u2g_coeff = rnd();
> +    ctx->yuv2rgb_u2b_coeff = rnd();
> +
> +    for (fmi = 0; fmi < FF_ARRAY_ELEMS(planar_fmts); fmi++) {
> +        for (fsi = 0; fsi < FILTER_SIZES; fsi++) {
> +            for (isi = 0; isi < INPUT_SIZES; isi++ ) {
> +                desc = av_pix_fmt_desc_get(planar_fmts[fmi]);
> +                ctx->dstFormat = planar_fmts[fmi];
> +
> +                dstW = input_sizes[isi];
> +                luma_filter_size = filter_sizes[fsi];
> +                chr_filter_size = filter_sizes[fsi];
> +
> +                if (desc->comp[0].depth > 16) {
> +                    byte_size = 4;
> +                } else if (desc->comp[0].depth > 8) {
> +                    byte_size = 2;
> +                } else {
> +                    byte_size = 1;
> +                }
> +
> +                ff_sws_init_scale(ctx);
> +                if (check_func(ctx->yuv2anyX, "yuv2%s_full_X_%d_%d",
> desc->name, luma_filter_size, dstW)) {
> +                    for (i = 0; i < 4; i ++) {
> +                        memset(dst0[i], 0xFF, LARGEST_INPUT_SIZE *
> sizeof(int32_t));
> +                        memset(dst1[i], 0xFF, LARGEST_INPUT_SIZE *
> sizeof(int32_t));
> +                    }
> +
> +                    call_ref(ctx, luma_filter, luma, luma_filter_size,
> +                             chr_filter, chru, chrv, chr_filter_size,
> +                             alpha, dst0, dstW, 0);
> +                    call_new(ctx, luma_filter, luma, luma_filter_size,
> +                             chr_filter, chru, chrv, chr_filter_size,
> +                             alpha, dst1, dstW, 0);
> +
> +                    if (memcmp(dst0[0], dst1[0], dstW * byte_size) ||
> +                        memcmp(dst0[1], dst1[1], dstW * byte_size) ||
> +                        memcmp(dst0[2], dst1[2], dstW * byte_size) ||
> +                        memcmp(dst0[3], dst1[3], dstW * byte_size) )
> +                        fail();
> +
> +                    bench_new(ctx, luma_filter, luma, luma_filter_size,
> +                              chr_filter, chru, chrv, chr_filter_size,
> +                              alpha, dst1, dstW, 0);
> +                }
> +            }
> +        }
> +    }
> +    sws_freeContext(ctx);
> +}
> +
> +
> +void checkasm_check_sw_gbrp(void)
> +{
> +    check_output_yuv2gbrp();
> +    report("output_yuv2gbrp");
> +}
> diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak
> index 6e7edbe655..6db8f09d12 100644
> --- a/tests/fate/checkasm.mak
> +++ b/tests/fate/checkasm.mak
> @@ -26,6 +26,7 @@ FATE_CHECKASM = fate-checkasm-aacpsdsp
>                 \
>                  fate-checkasm-pixblockdsp                               \
>                  fate-checkasm-sbrdsp                                    \
>                  fate-checkasm-synth_filter                              \
> +                fate-checkasm-sw_gbrp                                   \
>                  fate-checkasm-sw_rgb                                    \
>                  fate-checkasm-sw_scale                                  \
>                  fate-checkasm-utvideodsp                                \
> --
> 2.29.2
>
>
ping
Mark Reid Nov. 24, 2021, 6:57 p.m. UTC | #2
On Tue, Nov 16, 2021 at 10:58 AM Mark Reid <mindmark@gmail.com> wrote:

>
>
> On Wed, Nov 3, 2021 at 9:28 PM <mindmark@gmail.com> wrote:
>
>> From: Mark Reid <mindmark@gmail.com>
>>
>> changes since v1:
>> * remove vex intruction on sse4 path
>> * some load/pack marcos use less intructions
>> * fixed some typos
>>
>> yuv2gbrp_full_X_4_512_c: 12757.6
>> yuv2gbrp_full_X_4_512_sse2: 8946.6
>> yuv2gbrp_full_X_4_512_sse4: 5138.6
>> yuv2gbrp_full_X_4_512_avx2: 3889.6
>> yuv2gbrap_full_X_4_512_c: 15368.6
>> yuv2gbrap_full_X_4_512_sse2: 11916.1
>> yuv2gbrap_full_X_4_512_sse4: 6294.6
>> yuv2gbrap_full_X_4_512_avx2: 3477.1
>> yuv2gbrp9be_full_X_4_512_c: 14381.6
>> yuv2gbrp9be_full_X_4_512_sse2: 9139.1
>> yuv2gbrp9be_full_X_4_512_sse4: 5150.1
>> yuv2gbrp9be_full_X_4_512_avx2: 2834.6
>> yuv2gbrp9le_full_X_4_512_c: 12990.1
>> yuv2gbrp9le_full_X_4_512_sse2: 9118.1
>> yuv2gbrp9le_full_X_4_512_sse4: 5132.1
>> yuv2gbrp9le_full_X_4_512_avx2: 2833.1
>> yuv2gbrp10be_full_X_4_512_c: 14401.6
>> yuv2gbrp10be_full_X_4_512_sse2: 9133.1
>> yuv2gbrp10be_full_X_4_512_sse4: 5126.1
>> yuv2gbrp10be_full_X_4_512_avx2: 2837.6
>> yuv2gbrp10le_full_X_4_512_c: 12718.1
>> yuv2gbrp10le_full_X_4_512_sse2: 9106.1
>> yuv2gbrp10le_full_X_4_512_sse4: 5120.1
>> yuv2gbrp10le_full_X_4_512_avx2: 2826.1
>> yuv2gbrap10be_full_X_4_512_c: 18535.6
>> yuv2gbrap10be_full_X_4_512_sse2: 33617.6
>> yuv2gbrap10be_full_X_4_512_sse4: 6264.1
>> yuv2gbrap10be_full_X_4_512_avx2: 3422.1
>> yuv2gbrap10le_full_X_4_512_c: 16724.1
>> yuv2gbrap10le_full_X_4_512_sse2: 11787.1
>> yuv2gbrap10le_full_X_4_512_sse4: 6282.1
>> yuv2gbrap10le_full_X_4_512_avx2: 3441.6
>> yuv2gbrp12be_full_X_4_512_c: 13723.6
>> yuv2gbrp12be_full_X_4_512_sse2: 9128.1
>> yuv2gbrp12be_full_X_4_512_sse4: 7997.6
>> yuv2gbrp12be_full_X_4_512_avx2: 2844.1
>> yuv2gbrp12le_full_X_4_512_c: 12257.1
>> yuv2gbrp12le_full_X_4_512_sse2: 9107.6
>> yuv2gbrp12le_full_X_4_512_sse4: 5142.6
>> yuv2gbrp12le_full_X_4_512_avx2: 2837.6
>> yuv2gbrap12be_full_X_4_512_c: 18511.1
>> yuv2gbrap12be_full_X_4_512_sse2: 12156.6
>> yuv2gbrap12be_full_X_4_512_sse4: 6251.1
>> yuv2gbrap12be_full_X_4_512_avx2: 3444.6
>> yuv2gbrap12le_full_X_4_512_c: 16687.1
>> yuv2gbrap12le_full_X_4_512_sse2: 11785.1
>> yuv2gbrap12le_full_X_4_512_sse4: 6243.6
>> yuv2gbrap12le_full_X_4_512_avx2: 3446.1
>> yuv2gbrp14be_full_X_4_512_c: 13690.6
>> yuv2gbrp14be_full_X_4_512_sse2: 9120.6
>> yuv2gbrp14be_full_X_4_512_sse4: 5138.1
>> yuv2gbrp14be_full_X_4_512_avx2: 2843.1
>> yuv2gbrp14le_full_X_4_512_c: 14995.6
>> yuv2gbrp14le_full_X_4_512_sse2: 9119.1
>> yuv2gbrp14le_full_X_4_512_sse4: 5126.1
>> yuv2gbrp14le_full_X_4_512_avx2: 2843.1
>> yuv2gbrp16be_full_X_4_512_c: 12367.1
>> yuv2gbrp16be_full_X_4_512_sse2: 8233.6
>> yuv2gbrp16be_full_X_4_512_sse4: 4820.1
>> yuv2gbrp16be_full_X_4_512_avx2: 2666.6
>> yuv2gbrp16le_full_X_4_512_c: 10904.1
>> yuv2gbrp16le_full_X_4_512_sse2: 8214.1
>> yuv2gbrp16le_full_X_4_512_sse4: 4824.1
>> yuv2gbrp16le_full_X_4_512_avx2: 2629.1
>> yuv2gbrap16be_full_X_4_512_c: 26569.6
>> yuv2gbrap16be_full_X_4_512_sse2: 10884.1
>> yuv2gbrap16be_full_X_4_512_sse4: 5488.1
>> yuv2gbrap16be_full_X_4_512_avx2: 3272.1
>> yuv2gbrap16le_full_X_4_512_c: 14010.1
>> yuv2gbrap16le_full_X_4_512_sse2: 10562.1
>> yuv2gbrap16le_full_X_4_512_sse4: 5463.6
>> yuv2gbrap16le_full_X_4_512_avx2: 3255.1
>> yuv2gbrpf32be_full_X_4_512_c: 14524.1
>> yuv2gbrpf32be_full_X_4_512_sse2: 8552.6
>> yuv2gbrpf32be_full_X_4_512_sse4: 4636.1
>> yuv2gbrpf32be_full_X_4_512_avx2: 2474.6
>> yuv2gbrpf32le_full_X_4_512_c: 13060.6
>> yuv2gbrpf32le_full_X_4_512_sse2: 9682.6
>> yuv2gbrpf32le_full_X_4_512_sse4: 4298.1
>> yuv2gbrpf32le_full_X_4_512_avx2: 2453.1
>> yuv2gbrapf32be_full_X_4_512_c: 18629.6
>> yuv2gbrapf32be_full_X_4_512_sse2: 11363.1
>> yuv2gbrapf32be_full_X_4_512_sse4: 15201.6
>> yuv2gbrapf32be_full_X_4_512_avx2: 3727.1
>> yuv2gbrapf32le_full_X_4_512_c: 16677.6
>> yuv2gbrapf32le_full_X_4_512_sse2: 10221.6
>> yuv2gbrapf32le_full_X_4_512_sse4: 5693.6
>> yuv2gbrapf32le_full_X_4_512_avx2: 3656.6
>>
>> ---
>>  libswscale/x86/output.asm | 435 +++++++++++++++++++++++++++++++++++++-
>>  libswscale/x86/swscale.c  |  98 +++++++++
>>  tests/checkasm/Makefile   |   2 +-
>>  tests/checkasm/checkasm.c |   1 +
>>  tests/checkasm/checkasm.h |   1 +
>>  tests/checkasm/sw_gbrp.c  | 198 +++++++++++++++++
>>  tests/fate/checkasm.mak   |   1 +
>>  7 files changed, 734 insertions(+), 2 deletions(-)
>>  create mode 100644 tests/checkasm/sw_gbrp.c
>>
>> diff --git a/libswscale/x86/output.asm b/libswscale/x86/output.asm
>> index 52cf9f2c2e..ce0a2650b4 100644
>> --- a/libswscale/x86/output.asm
>> +++ b/libswscale/x86/output.asm
>> @@ -38,7 +38,49 @@ pw_32:         times 8 dw 32
>>  pd_255:        times 8 dd 255
>>  pw_512:        times 8 dw 512
>>  pw_1024:       times 8 dw 1024
>> -
>> +pd_65535_invf:             times 8 dd 0x37800080 ;1.0/65535.0
>> +pd_yuv2gbrp16_start:       times 8 dd -0x40000000
>> +pd_yuv2gbrp_y_start:       times 8 dd  (1 << 9)
>> +pd_yuv2gbrp_uv_start:      times 8 dd  ((1 << 9) - (128 << 19))
>> +pd_yuv2gbrp_a_start:       times 8 dd  (1 << 18)
>> +pd_yuv2gbrp16_offset:      times 8 dd  0x10000  ;(1 << 16)
>> +pd_yuv2gbrp16_round13:     times 8 dd  0x02000  ;(1 << 13)
>> +pd_yuv2gbrp16_a_offset:    times 8 dd  0x20002000
>> +pd_yuv2gbrp16_upper30:     times 8 dd  0x3FFFFFFF ;(1<<30) - 1
>> +pd_yuv2gbrp16_upper27:     times 8 dd  0x07FFFFFF ;(1<<27) - 1
>> +pd_yuv2gbrp16_upperC:      times 8 dd  0xC0000000
>> +pb_pack_shuffle8:       db  0,  4,  8, 12, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                            0,  4,  8, 12, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1
>> +pb_pack_shuffle16le:    db  0,  1,  4,  5, \
>> +                            8,  9, 12, 13, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                            0,  1,  4,  5, \
>> +                            8,  9, 12, 13
>> +pb_pack_shuffle16be:    db  1,  0,  5,  4, \
>> +                            9,  8, 13, 12, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                           -1, -1, -1, -1, \
>> +                            1,  0,  5,  4, \
>> +                            9,  8, 13, 12
>> +pb_shuffle32be          db  3,  2,  1,  0, \
>> +                            7,  6,  5,  4, \
>> +                           11, 10,  9,  8, \
>> +                           15, 14, 13, 12, \
>> +                            3,  2,  1,  0, \
>> +                            7,  6,  5,  4, \
>> +                           11, 10,  9,  8, \
>> +                           15, 14, 13, 12
>>  yuv2nv12_shuffle_mask: times 2 db 0,  4,  8, 12, \
>>                                   -1, -1, -1, -1, \
>>                                   -1, -1, -1, -1, \
>> @@ -549,3 +591,394 @@ yuv2nv12cX_fn yuv2nv12
>>  yuv2nv12cX_fn yuv2nv21
>>  %endif
>>  %endif ; ARCH_X86_64
>> +
>>
>> +;-----------------------------------------------------------------------------
>> +; planar grb yuv2anyX functions
>> +; void ff_yuv2<gbr_format>_full_X_<opt>(SwsContext *c, const int16_t
>> *lumFilter,
>> +;                                       const int16_t **lumSrcx, int
>> lumFilterSize,
>> +;                                       const int16_t *chrFilter, const
>> int16_t **chrUSrcx,
>> +;                                       const int16_t **chrVSrcx, int
>> chrFilterSize,
>> +;                                       const int16_t **alpSrcx, uint8_t
>> **dest,
>> +;                                       int dstW, int y)
>>
>> +;-----------------------------------------------------------------------------
>> +
>> +%if ARCH_X86_64
>> +struc SwsContext
>> +    .padding:           resb 40292 ; offsetof(SwsContext,
>> yuv2rgb_y_offset)
>> +    .yuv2rgb_y_offset:  resd 1
>> +    .yuv2rgb_y_coeff:   resd 1
>> +    .yuv2rgb_v2r_coeff: resd 1
>> +    .yuv2rgb_v2g_coeff: resd 1
>> +    .yuv2rgb_u2g_coeff: resd 1
>> +    .yuv2rgb_u2b_coeff: resd 1
>> +endstruc
>> +
>> +%define R m0
>> +%define G m1
>> +%define B m2
>> +%define A m3
>> +
>> +%define Y m4
>> +%define U m5
>> +%define V m6
>> +
>> +; Clip a signed integer to an unsigned power of two range.
>> +; av_clip_uintp2
>> +; 1 - dest
>> +; 2 - bit position to clip at
>> +%macro CLIPP2 2
>> +    ; (~a) >> 31 & ((1<<p) - 1);
>> +    pcmpeqb m4, m4
>> +    pxor m4, %1
>> +    psrad m4, 31
>> +    movu m5, [pd_yuv2gbrp16_upper%2]
>> +    pand m4, m5
>> +
>> +    ; (a & ~((1<<p) - 1)) == 0
>> +    pandn m5, %1
>> +    pxor m6, m6
>> +    pcmpeqd m5, m6
>> +%if cpuflag(avx2)
>> +    vpblendvb %1, m4, %1, m5
>> +%else
>> +    pxor %1, m4
>> +    pand %1, m5
>> +    pxor %1, m4
>> +%endif
>> +%endmacro
>> +
>> +; 1 - dest
>> +; 2 - source
>> +%macro LOAD16 2
>> +    %if cpuflag(avx2)
>> +        movu xm%1, %2
>> +        vpmovsxwd m%1, xm%1
>> +    %elif cpuflag(sse4)
>> +        movsd m%1, %2
>> +        pmovsxwd m%1, m%1
>> +    %else
>> +        movsd m%1, %2
>> +        punpcklwd m%1, m%1
>> +        psrad m%1, 16 ; sign extend
>> +    %endif
>> +%endmacro
>> +
>> +; 1 - dest
>> +; 2 - source
>> +; 3 - depth
>> +%macro LOAD_PIXELS 3
>> +    mov ptrq, [%2 + jq*8]
>> +%if %3 >= 16
>> +    movu m%1, [ptrq + xq*4]
>> +%else
>> +    LOAD16 %1, [ptrq + xq*2]
>> +%endif
>> +%endmacro
>> +
>> +; 1 - dest
>> +; 2 - source
>> +%macro STORE8 2
>> +    mov ptrq, %1
>> +    %if mmsize > 16
>> +        pshufb m%2, [pb_pack_shuffle8]
>> +        vextractf128 xm4, m%2, 1
>> +        por xm%2, xm4
>> +        movq [ptrq + xq], xm%2
>> +    %else
>> +        %if cpuflag(sse4)
>> +            pshufb m%2, [pb_pack_shuffle8]
>> +        %else
>> +            psrldq m4, m%2, 3
>> +            por m%2, m4
>> +            psrldq m4, m%2, 6
>> +            por m%2, m4
>> +        %endif
>> +        movd [ptrq + xq], m%2
>> +    %endif
>> +%endmacro
>> +
>> +; 1 - dest
>> +; 2 - source
>> +; 3 - is big endian
>> +%macro STORE16 3
>> +    mov ptrq, %1
>> +    %if mmsize > 16
>> +        %if %3 ; bigendian
>> +            pshufb m%2, [pb_pack_shuffle16be]
>> +        %else
>> +            pshufb m%2, [pb_pack_shuffle16le]
>> +        %endif
>> +        vextractf128 xm4, m%2, 1
>> +        por xm%2, xm4
>> +        movu [ptrq + xq*2], xm%2
>> +    %else
>> +        %if cpuflag(sse4) && %3 ; bigendian
>> +            pshufb m%2, [pb_pack_shuffle16be]
>> +        %elif cpuflag(sse4)
>> +            pshufb m%2, [pb_pack_shuffle16le]
>> +        %else
>> +            pshuflw m%2, m%2, (1 << 6 | 1 << 4 | 2 << 2 | 0 << 0)
>> +            pshufhw m%2, m%2, (1 << 6 | 1 << 4 | 2 << 2 | 0 << 0)
>> +            pshufd  m%2, m%2, (3 << 6 | 3 << 4 | 2 << 2 | 0 << 0)
>> +            %if %3 ; bigendian
>> +                psrlw  m4, m%2, 8
>> +                psllw  m%2, 8
>> +                por m%2, m4
>> +            %endif
>> +        %endif
>> +        movq [ptrq + xq*2], m%2
>> +    %endif
>> +%endmacro
>> +
>> +%macro SWAP32 1
>> +%if mmsize > 16 || cpuflag(sse4)
>> +    pshufb m%1, [pb_shuffle32be]
>> +%else
>> +    psrlw  m4, m%1, 8
>> +    psllw  m%1, 8
>> +    por m%1, m4
>> +    pshuflw m%1, m%1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0)
>> +    pshufhw m%1, m%1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0)
>> +%endif
>> +%endmacro
>> +
>> +; 1 - dest
>> +; 2 - source
>> +; 3 - depth
>> +; 4 - is big endian
>> +%macro STORE_PIXELS 4
>> +%if %3 > 16
>> +    %if %4
>> +        SWAP32 %2
>> +    %endif
>> +    mov ptrq, %1
>> +    movu [ptrq + xq*4], m%2
>> +%elif %3 > 8
>> +    STORE16 %1, %2, %4
>> +%else
>> +    STORE8 %1, %2
>> +%endif
>> +%endmacro
>> +
>> +%macro PMULLO 3
>> +%if cpuflag(sse4) || mmsize > 16
>> +    pmulld %1, %2, %3
>> +%else
>> +    %ifidni %1, %2
>> +    %else
>> +        mova %1, %2
>> +    %endif
>> +    pshufd m7, %1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0) ; 0xb1
>> +    pshufd m8, %3, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0) ; 0xb1
>> +    pmuludq m7, m8
>> +    pshufd  m7, m7, (3 << 6 | 1 << 4 | 2 << 2 | 0 << 0) ; 0xd8
>> +    pmuludq %1, %3
>> +    pshufd  %1, %1, (3 << 6 | 1 << 4 | 2 << 2 | 0 << 0) ; 0xd8
>> +    punpckldq %1, m7
>> +%endif
>> +%endmacro
>> +
>> +; 1 - name
>> +; 2 - depth
>> +; 3 - has alpha
>> +; 3 - is big endian
>> +; 5 - is float
>> +%macro yuv2gbrp_fn 5
>> +%define DEPTH %2
>> +%define HAS_ALPHA %3
>> +%define IS_BE %4
>> +%define FLOAT %5
>> +%define SH (22 + 8 - DEPTH)
>> +
>> +%if DEPTH >= 16
>> +    %define RGB_SHIFT 14
>> +    %define A_SHIFT 14
>> +%elif 22 != SH
>> +    %define RGB_SHIFT SH
>> +    %define A_SHIFT (SH-3)
>> +%else
>> +    %define RGB_SHIFT 22
>> +    %define A_SHIFT 19
>> +%endif
>> +
>> +%if DEPTH >= 16
>> +    %define YUV_SHIFT 14
>> +    %define Y_START  m9
>> +    %define Y_ROUND [pd_yuv2gbrp16_round13]
>> +    %define UV_START m9
>> +    %define A_START  m9
>> +    %define A_CLIP2P 30
>> +%else
>> +    %define YUV_SHIFT 10
>> +    %define Y_START  [pd_yuv2gbrp_y_start]
>> +    %define Y_ROUND  m9
>> +    %define UV_START [pd_yuv2gbrp_uv_start]
>> +    %define A_START  [pd_yuv2gbrp_a_start]
>> +    %define A_CLIP2P 27
>> +%endif
>> +
>> +cglobal yuv2%1_full_X, 12, 14, 16, ptr, lumFilter, lumSrcx,
>> lumFilterSize, chrFilter, chrUSrcx, chrVSrcx, chrFilterSize, alpSrcx, dest,
>> dstW, y, x, j
>> +    VBROADCASTSS m10, dword [ptrq + SwsContext.yuv2rgb_y_offset]
>> +    VBROADCASTSS m11, dword [ptrq + SwsContext.yuv2rgb_y_coeff]
>> +    VBROADCASTSS m12, dword [ptrq + SwsContext.yuv2rgb_v2r_coeff]
>> +    VBROADCASTSS m13, dword [ptrq + SwsContext.yuv2rgb_v2g_coeff]
>> +    VBROADCASTSS m14, dword [ptrq + SwsContext.yuv2rgb_u2g_coeff]
>> +    VBROADCASTSS m15, dword [ptrq + SwsContext.yuv2rgb_u2b_coeff]
>> +
>> +%if DEPTH >= 16
>> +    movu m9, [pd_yuv2gbrp16_start]
>> +%else
>> +    mov xq, (1 << (SH-1))
>> +    movq xm9, xq
>> +    VBROADCASTSS m9, xm9
>> +%endif
>> +    xor xq, xq
>> +
>> +    %%loop_x:
>> +        movu Y, Y_START
>> +        movu U, UV_START
>> +        movu V, UV_START
>> +
>> +        xor jq, jq
>> +        %%loop_luma:
>> +            movsx ptrd, word [lumFilterq + jq*2]
>> +            movd xm0, ptrd
>> +            VBROADCASTSS m0, xm0
>> +            LOAD_PIXELS 1, lumSrcxq, DEPTH
>> +            PMULLO m1, m1, m0
>> +            paddd Y, m1
>> +            inc jd
>> +            cmp jd, lumFilterSized
>> +            jl %%loop_luma
>> +
>> +%if HAS_ALPHA
>> +        cmp alpSrcxq, 0
>> +        je %%skip_alpha_load
>> +            xor jq, jq
>> +            movu A, A_START
>> +            %%loop_alpha:
>> +                movsx ptrd, word [lumFilterq + jq*2]
>> +                movd xm0, ptrd
>> +                VBROADCASTSS m0, xm0
>> +                LOAD_PIXELS 1, alpSrcxq, DEPTH
>> +                PMULLO m1, m1, m0
>> +                paddd A, m1
>> +                inc jd
>> +                cmp jd, lumFilterSized
>> +                jl %%loop_alpha
>> +%if DEPTH >= 16
>> +            psrad A, 1
>> +            paddd A, [pd_yuv2gbrp16_a_offset]
>> +%endif
>> +        %%skip_alpha_load:
>> +%endif
>> +        xor jq, jq
>> +        %%loop_chr:
>> +            movsx ptrd, word [chrFilterq + jq*2]
>> +            movd xm0, ptrd
>> +            VBROADCASTSS m0, xm0
>> +            LOAD_PIXELS 1, chrUSrcxq, DEPTH
>> +            LOAD_PIXELS 2, chrVSrcxq, DEPTH
>> +            PMULLO m1, m1, m0
>> +            PMULLO m2, m2, m0
>> +            paddd U, m1
>> +            paddd V, m2
>> +            inc jd
>> +            cmp jd, chrFilterSized
>> +            jl %%loop_chr
>> +
>> +        psrad Y, YUV_SHIFT
>> +%if  DEPTH >= 16
>> +        paddd Y, [pd_yuv2gbrp16_offset]
>> +%endif
>> +        psrad U, YUV_SHIFT
>> +        psrad V, YUV_SHIFT
>> +
>> +        psubd  Y, m10    ; yuv2rgb_y_offset
>> +        PMULLO Y, Y, m11 ; yuv2rgb_y_coeff
>> +        paddd  Y, Y_ROUND
>> +
>> +        PMULLO R, V, m12 ; yuv2rgb_v2r_coeff
>> +        PMULLO B, U, m15 ; yuv2rgb_u2b_coeff
>> +
>> +        PMULLO U, U, m14 ; yuv2rgb_u2g_coeff
>> +        PMULLO V, V, m13 ; yuv2rgb_v2g_coeff
>> +        paddd G, U, V
>> +        paddd R, Y
>> +        paddd G, Y
>> +        paddd B, Y
>> +
>> +        CLIPP2 R, 30
>> +        CLIPP2 G, 30
>> +        CLIPP2 B, 30
>> +
>> +        psrad R, RGB_SHIFT
>> +        psrad G, RGB_SHIFT
>> +        psrad B, RGB_SHIFT
>> +
>> +%if FLOAT
>> +        cvtdq2ps R, R
>> +        cvtdq2ps G, G
>> +        cvtdq2ps B, B
>> +        mulps R, [pd_65535_invf]
>> +        mulps G, [pd_65535_invf]
>> +        mulps B, [pd_65535_invf]
>> +%endif
>> +        STORE_PIXELS [destq +  0], 1, DEPTH, IS_BE ; G
>> +        STORE_PIXELS [destq +  8], 2, DEPTH, IS_BE ; B
>> +        STORE_PIXELS [destq + 16], 0, DEPTH, IS_BE ; R
>> +
>> +%if HAS_ALPHA
>> +        cmp alpSrcxq, 0
>> +        je %%skip_alpha_store
>> +            CLIPP2 A, A_CLIP2P
>> +            psrad A, A_SHIFT
>> +%if FLOAT
>> +            cvtdq2ps A, A
>> +            mulps A, [pd_65535_invf]
>> +%endif
>> +            STORE_PIXELS [destq + 24], 3, DEPTH, IS_BE
>> +        %%skip_alpha_store:
>> +%endif
>> +        add xq, mmsize/4
>> +        cmp xd, dstWd
>> +        jl %%loop_x
>> +
>> +    RET
>> +%endmacro
>> +
>> +%macro yuv2gbrp_fn_decl 2
>> +INIT_%1 %2
>> +yuv2gbrp_fn gbrp,        8, 0, 0, 0
>> +yuv2gbrp_fn gbrap,       8, 1, 0, 0
>> +yuv2gbrp_fn gbrp9le,     9, 0, 0, 0
>> +yuv2gbrp_fn gbrp10le,   10, 0, 0, 0
>> +yuv2gbrp_fn gbrap10le,  10, 1, 0, 0
>> +yuv2gbrp_fn gbrp12le,   12, 0, 0, 0
>> +yuv2gbrp_fn gbrap12le,  12, 1, 0, 0
>> +yuv2gbrp_fn gbrp14le,   14, 0, 0, 0
>> +yuv2gbrp_fn gbrp16le,   16, 0, 0, 0
>> +yuv2gbrp_fn gbrap16le,  16, 1, 0, 0
>> +yuv2gbrp_fn gbrpf32le,  32, 0, 0, 1
>> +yuv2gbrp_fn gbrapf32le, 32, 1, 0, 1
>> +
>> +yuv2gbrp_fn gbrp9be,     9, 0, 1, 0
>> +yuv2gbrp_fn gbrp10be,   10, 0, 1, 0
>> +yuv2gbrp_fn gbrap10be,  10, 1, 1, 0
>> +yuv2gbrp_fn gbrp12be,   12, 0, 1, 0
>> +yuv2gbrp_fn gbrap12be,  12, 1, 1, 0
>> +yuv2gbrp_fn gbrp14be,   14, 0, 1, 0
>> +yuv2gbrp_fn gbrp16be,   16, 0, 1, 0
>> +yuv2gbrp_fn gbrap16be,  16, 1, 1, 0
>> +yuv2gbrp_fn gbrpf32be,  32, 0, 1, 1
>> +yuv2gbrp_fn gbrapf32be, 32, 1, 1, 1
>> +%endmacro
>> +
>> +yuv2gbrp_fn_decl XMM, sse2
>> +yuv2gbrp_fn_decl XMM, sse4
>> +
>> +%if HAVE_AVX2_EXTERNAL
>> +yuv2gbrp_fn_decl YMM, avx2
>> +%endif
>> +
>> +%endif ; ARCH_X86_64
>> \ No newline at end of file
>> diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
>> index 0848a31461..e8bc15b6ee 100644
>> --- a/libswscale/x86/swscale.c
>> +++ b/libswscale/x86/swscale.c
>> @@ -350,6 +350,43 @@ void ff_yuv2 ## fmt ## cX_ ## opt(enum AVPixelFormat
>> format, const uint8_t *dith
>>
>>  YUV2NV_DECL(nv12, avx2);
>>  YUV2NV_DECL(nv21, avx2);
>> +
>> +#define YUV2GBRP_FN_DECL(fmt, opt)
>>                 \
>> +void ff_yuv2##fmt##_full_X_ ##opt(SwsContext *c, const int16_t
>> *lumFilter,           \
>> +                                 const int16_t **lumSrcx, int
>> lumFilterSize,         \
>> +                                 const int16_t *chrFilter, const int16_t
>> **chrUSrcx, \
>> +                                 const int16_t **chrVSrcx, int
>> chrFilterSize,        \
>> +                                 const int16_t **alpSrcx, uint8_t
>> **dest,            \
>> +                                 int dstW, int y)
>> +
>> +#define YUV2GBRP_DECL(opt)     \
>> +YUV2GBRP_FN_DECL(gbrp,       opt); \
>> +YUV2GBRP_FN_DECL(gbrap,      opt); \
>> +YUV2GBRP_FN_DECL(gbrp9le,    opt); \
>> +YUV2GBRP_FN_DECL(gbrp10le,   opt); \
>> +YUV2GBRP_FN_DECL(gbrap10le,  opt); \
>> +YUV2GBRP_FN_DECL(gbrp12le,   opt); \
>> +YUV2GBRP_FN_DECL(gbrap12le,  opt); \
>> +YUV2GBRP_FN_DECL(gbrp14le,   opt); \
>> +YUV2GBRP_FN_DECL(gbrp16le,   opt); \
>> +YUV2GBRP_FN_DECL(gbrap16le,  opt); \
>> +YUV2GBRP_FN_DECL(gbrpf32le,  opt); \
>> +YUV2GBRP_FN_DECL(gbrapf32le, opt); \
>> +YUV2GBRP_FN_DECL(gbrp9be,    opt); \
>> +YUV2GBRP_FN_DECL(gbrp10be,   opt); \
>> +YUV2GBRP_FN_DECL(gbrap10be,  opt); \
>> +YUV2GBRP_FN_DECL(gbrp12be,   opt); \
>> +YUV2GBRP_FN_DECL(gbrap12be,  opt); \
>> +YUV2GBRP_FN_DECL(gbrp14be,   opt); \
>> +YUV2GBRP_FN_DECL(gbrp16be,   opt); \
>> +YUV2GBRP_FN_DECL(gbrap16be,  opt); \
>> +YUV2GBRP_FN_DECL(gbrpf32be,  opt); \
>> +YUV2GBRP_FN_DECL(gbrapf32be, opt);
>> +
>> +YUV2GBRP_DECL(sse2);
>> +YUV2GBRP_DECL(sse4);
>> +YUV2GBRP_DECL(avx2);
>> +
>>  #endif
>>
>>  av_cold void ff_sws_init_swscale_x86(SwsContext *c)
>> @@ -582,5 +619,66 @@ switch(c->dstBpc){ \
>>              break;
>>          }
>>      }
>> +
>> +    if(c->flags & SWS_FULL_CHR_H_INT) {
>> +
>> +        /* yuv2gbrp uses the SwsContext for yuv coefficients
>> +           if struct offsets change the asm needs to be updated too */
>> +        av_assert0(offsetof(SwsContext, yuv2rgb_y_offset) == 40292);
>> +
>> +#define YUV2ANYX_FUNC_CASE(fmt, name, opt)              \
>> +        case fmt:                                       \
>> +            c->yuv2anyX = ff_yuv2##name##_full_X_##opt; \
>> +            break;
>> +
>> +#define YUV2ANYX_GBRAP_CASES(opt)                                  \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP,       gbrp,       opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP,      gbrap,      opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP9LE,    gbrp9le,    opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP10LE,   gbrp10le,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP10LE,  gbrap10le,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP12LE,   gbrp12le,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP12LE,  gbrap12le,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP14LE,   gbrp14le,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP16LE,   gbrp16le,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP16LE,  gbrap16le,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRPF32LE,  gbrpf32le,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAPF32LE, gbrapf32le, opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP9BE,    gbrp9be,    opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP10BE,   gbrp10be,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP10BE,  gbrap10be,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP12BE,   gbrp12be,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP12BE,  gbrap12be,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP14BE,   gbrp14be,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP16BE,   gbrp16be,   opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP16BE,  gbrap16be,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRPF32BE,  gbrpf32be,  opt) \
>> +        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAPF32BE, gbrapf32be, opt)
>> +
>> +        if (EXTERNAL_SSE2(cpu_flags)) {
>> +            switch (c->dstFormat) {
>> +            YUV2ANYX_GBRAP_CASES(sse2)
>> +            default:
>> +                break;
>> +            }
>> +        }
>> +
>> +        if (EXTERNAL_SSE4(cpu_flags)) {
>> +            switch (c->dstFormat) {
>> +            YUV2ANYX_GBRAP_CASES(sse4)
>> +            default:
>> +                break;
>> +            }
>> +        }
>> +
>> +        if (EXTERNAL_AVX2_FAST(cpu_flags)) {
>> +            switch (c->dstFormat) {
>> +            YUV2ANYX_GBRAP_CASES(avx2)
>> +            default:
>> +                break;
>> +            }
>> +        }
>> +    }
>> +
>>  #endif
>>  }
>> diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
>> index 4ef5fa87da..f768b1144e 100644
>> --- a/tests/checkasm/Makefile
>> +++ b/tests/checkasm/Makefile
>> @@ -45,7 +45,7 @@ AVFILTEROBJS-$(CONFIG_NLMEANS_FILTER)    += vf_nlmeans.o
>>  CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes)
>>
>>  # swscale tests
>> -SWSCALEOBJS                             += sw_rgb.o sw_scale.o
>> +SWSCALEOBJS                             += sw_gbrp.o sw_rgb.o sw_scale.o
>>
>>  CHECKASMOBJS-$(CONFIG_SWSCALE)  += $(SWSCALEOBJS)
>>
>> diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
>> index b1353f7cbe..5fa8ac7bcc 100644
>> --- a/tests/checkasm/checkasm.c
>> +++ b/tests/checkasm/checkasm.c
>> @@ -183,6 +183,7 @@ static const struct {
>>      #endif
>>  #endif
>>  #if CONFIG_SWSCALE
>> +    { "sw_gbrp", checkasm_check_sw_gbrp },
>>      { "sw_rgb", checkasm_check_sw_rgb },
>>      { "sw_scale", checkasm_check_sw_scale },
>>  #endif
>> diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
>> index 68b0697d3e..c3192d8c23 100644
>> --- a/tests/checkasm/checkasm.h
>> +++ b/tests/checkasm/checkasm.h
>> @@ -72,6 +72,7 @@ void checkasm_check_opusdsp(void);
>>  void checkasm_check_pixblockdsp(void);
>>  void checkasm_check_sbrdsp(void);
>>  void checkasm_check_synth_filter(void);
>> +void checkasm_check_sw_gbrp(void);
>>  void checkasm_check_sw_rgb(void);
>>  void checkasm_check_sw_scale(void);
>>  void checkasm_check_utvideodsp(void);
>> diff --git a/tests/checkasm/sw_gbrp.c b/tests/checkasm/sw_gbrp.c
>> new file mode 100644
>> index 0000000000..320948a71d
>> --- /dev/null
>> +++ b/tests/checkasm/sw_gbrp.c
>> @@ -0,0 +1,198 @@
>> +/*
>> + *
>> + * This file is part of FFmpeg.
>> + *
>> + * FFmpeg is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License as published by
>> + * the Free Software Foundation; either version 2 of the License, or
>> + * (at your option) any later version.
>> + *
>> + * FFmpeg is distributed in the hope that it will be useful,
>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
>> + * GNU General Public License for more details.
>> + *
>> + * You should have received a copy of the GNU General Public License
>> along
>> + * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
>> + * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
>> + */
>> +
>> +#include <string.h>
>> +
>> +#include "libavutil/common.h"
>> +#include "libavutil/intreadwrite.h"
>> +#include "libavutil/mem_internal.h"
>> +#include "libavutil/pixdesc.h"
>> +
>> +#include "libswscale/swscale.h"
>> +#include "libswscale/swscale_internal.h"
>> +
>> +#include "checkasm.h"
>> +
>> +#define randomize_buffers(buf, size)      \
>> +    do {                                  \
>> +        int j;                            \
>> +        for (j = 0; j < size; j+=4)       \
>> +            AV_WN32(buf + j, rnd());      \
>> +    } while (0)
>> +
>> +static const int planar_fmts[] = {
>> +    AV_PIX_FMT_GBRP,
>> +    AV_PIX_FMT_GBRP9BE,
>> +    AV_PIX_FMT_GBRP9LE,
>> +    AV_PIX_FMT_GBRP10BE,
>> +    AV_PIX_FMT_GBRP10LE,
>> +    AV_PIX_FMT_GBRP12BE,
>> +    AV_PIX_FMT_GBRP12LE,
>> +    AV_PIX_FMT_GBRP14BE,
>> +    AV_PIX_FMT_GBRP14LE,
>> +    AV_PIX_FMT_GBRAP,
>> +    AV_PIX_FMT_GBRAP10BE,
>> +    AV_PIX_FMT_GBRAP10LE,
>> +    AV_PIX_FMT_GBRAP12BE,
>> +    AV_PIX_FMT_GBRAP12LE,
>> +    AV_PIX_FMT_GBRP16BE,
>> +    AV_PIX_FMT_GBRP16LE,
>> +    AV_PIX_FMT_GBRAP16BE,
>> +    AV_PIX_FMT_GBRAP16LE,
>> +    AV_PIX_FMT_GBRPF32BE,
>> +    AV_PIX_FMT_GBRPF32LE,
>> +    AV_PIX_FMT_GBRAPF32BE,
>> +    AV_PIX_FMT_GBRAPF32LE
>> +};
>> +
>> +static void check_output_yuv2gbrp(void)
>> +{
>> +    struct SwsContext *ctx;
>> +    const AVPixFmtDescriptor *desc;
>> +    int fmi, fsi, isi, i;
>> +    int dstW, byte_size, luma_filter_size, chr_filter_size;
>> +#define LARGEST_FILTER 16
>> +#define FILTER_SIZES 4
>> +    static const int filter_sizes[FILTER_SIZES] = {1, 4, 8, 16};
>> +#define LARGEST_INPUT_SIZE 512
>> +#define INPUT_SIZES 6
>> +    static const int input_sizes[INPUT_SIZES] = {8, 24, 128, 144, 256,
>> 512};
>> +    uint8_t *dst0[4];
>> +    uint8_t *dst1[4];
>> +
>> +    declare_func(void, void *c, const int16_t *lumFilter,
>> +                       const int16_t **lumSrcx, int lumFilterSize,
>> +                       const int16_t *chrFilter, const int16_t
>> **chrUSrcx,
>> +                       const int16_t **chrVSrcx, int chrFilterSize,
>> +                       const int16_t **alpSrcx, uint8_t **dest,
>> +                       int dstW, int y);
>> +
>> +    const int16_t *luma[LARGEST_FILTER];
>> +    const int16_t *chru[LARGEST_FILTER];
>> +    const int16_t *chrv[LARGEST_FILTER];
>> +    const int16_t *alpha[LARGEST_FILTER];
>> +
>> +    LOCAL_ALIGNED_8(int16_t, luma_filter, [LARGEST_FILTER]);
>> +    LOCAL_ALIGNED_8(int16_t, chr_filter, [LARGEST_FILTER]);
>> +
>> +    LOCAL_ALIGNED_8(int32_t, src_y, [LARGEST_FILTER *
>> LARGEST_INPUT_SIZE]);
>> +    LOCAL_ALIGNED_8(int32_t, src_u, [LARGEST_FILTER *
>> LARGEST_INPUT_SIZE]);
>> +    LOCAL_ALIGNED_8(int32_t, src_v, [LARGEST_FILTER *
>> LARGEST_INPUT_SIZE]);
>> +    LOCAL_ALIGNED_8(int32_t, src_a, [LARGEST_FILTER *
>> LARGEST_INPUT_SIZE]);
>> +
>> +    LOCAL_ALIGNED_8(uint8_t, dst0_r, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +    LOCAL_ALIGNED_8(uint8_t, dst0_g, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +    LOCAL_ALIGNED_8(uint8_t, dst0_b, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +    LOCAL_ALIGNED_8(uint8_t, dst0_a, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +
>> +    LOCAL_ALIGNED_8(uint8_t, dst1_r, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +    LOCAL_ALIGNED_8(uint8_t, dst1_g, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +    LOCAL_ALIGNED_8(uint8_t, dst1_b, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +    LOCAL_ALIGNED_8(uint8_t, dst1_a, [LARGEST_INPUT_SIZE *
>> sizeof(int32_t)]);
>> +
>> +    randomize_buffers((uint8_t*)src_y, LARGEST_FILTER *
>> LARGEST_INPUT_SIZE * sizeof(int32_t));
>> +    randomize_buffers((uint8_t*)src_u, LARGEST_FILTER *
>> LARGEST_INPUT_SIZE * sizeof(int32_t));
>> +    randomize_buffers((uint8_t*)src_v, LARGEST_FILTER *
>> LARGEST_INPUT_SIZE * sizeof(int32_t));
>> +    randomize_buffers((uint8_t*)src_a, LARGEST_FILTER *
>> LARGEST_INPUT_SIZE * sizeof(int32_t));
>> +    randomize_buffers((uint8_t*)luma_filter, LARGEST_FILTER *
>> sizeof(int16_t));
>> +    randomize_buffers((uint8_t*)chr_filter, LARGEST_FILTER *
>> sizeof(int16_t));
>> +
>> +    dst0[0] = (uint8_t*)dst0_g;
>> +    dst0[1] = (uint8_t*)dst0_b;
>> +    dst0[2] = (uint8_t*)dst0_r;
>> +    dst0[3] = (uint8_t*)dst0_a;
>> +
>> +    dst1[0] = (uint8_t*)dst1_g;
>> +    dst1[1] = (uint8_t*)dst1_b;
>> +    dst1[2] = (uint8_t*)dst1_r;
>> +    dst1[3] = (uint8_t*)dst1_a;
>> +
>> +    for (i = 0; i < LARGEST_FILTER; i++) {
>> +        luma[i] =  (int16_t *)(src_y + i*LARGEST_INPUT_SIZE);
>> +        chru[i] =  (int16_t *)(src_u + i*LARGEST_INPUT_SIZE);
>> +        chrv[i] =  (int16_t *)(src_v + i*LARGEST_INPUT_SIZE);
>> +        alpha[i] = (int16_t *)(src_a + i*LARGEST_INPUT_SIZE);
>> +    }
>> +
>> +    ctx = sws_alloc_context();
>> +    if (sws_init_context(ctx, NULL, NULL) < 0)
>> +        fail();
>> +
>> +    ctx->flags |= SWS_FULL_CHR_H_INT;
>> +    ctx->yuv2rgb_y_offset  = rnd();
>> +    ctx->yuv2rgb_y_coeff   = rnd();
>> +    ctx->yuv2rgb_v2r_coeff = rnd();
>> +    ctx->yuv2rgb_v2g_coeff = rnd();
>> +    ctx->yuv2rgb_u2g_coeff = rnd();
>> +    ctx->yuv2rgb_u2b_coeff = rnd();
>> +
>> +    for (fmi = 0; fmi < FF_ARRAY_ELEMS(planar_fmts); fmi++) {
>> +        for (fsi = 0; fsi < FILTER_SIZES; fsi++) {
>> +            for (isi = 0; isi < INPUT_SIZES; isi++ ) {
>> +                desc = av_pix_fmt_desc_get(planar_fmts[fmi]);
>> +                ctx->dstFormat = planar_fmts[fmi];
>> +
>> +                dstW = input_sizes[isi];
>> +                luma_filter_size = filter_sizes[fsi];
>> +                chr_filter_size = filter_sizes[fsi];
>> +
>> +                if (desc->comp[0].depth > 16) {
>> +                    byte_size = 4;
>> +                } else if (desc->comp[0].depth > 8) {
>> +                    byte_size = 2;
>> +                } else {
>> +                    byte_size = 1;
>> +                }
>> +
>> +                ff_sws_init_scale(ctx);
>> +                if (check_func(ctx->yuv2anyX, "yuv2%s_full_X_%d_%d",
>> desc->name, luma_filter_size, dstW)) {
>> +                    for (i = 0; i < 4; i ++) {
>> +                        memset(dst0[i], 0xFF, LARGEST_INPUT_SIZE *
>> sizeof(int32_t));
>> +                        memset(dst1[i], 0xFF, LARGEST_INPUT_SIZE *
>> sizeof(int32_t));
>> +                    }
>> +
>> +                    call_ref(ctx, luma_filter, luma, luma_filter_size,
>> +                             chr_filter, chru, chrv, chr_filter_size,
>> +                             alpha, dst0, dstW, 0);
>> +                    call_new(ctx, luma_filter, luma, luma_filter_size,
>> +                             chr_filter, chru, chrv, chr_filter_size,
>> +                             alpha, dst1, dstW, 0);
>> +
>> +                    if (memcmp(dst0[0], dst1[0], dstW * byte_size) ||
>> +                        memcmp(dst0[1], dst1[1], dstW * byte_size) ||
>> +                        memcmp(dst0[2], dst1[2], dstW * byte_size) ||
>> +                        memcmp(dst0[3], dst1[3], dstW * byte_size) )
>> +                        fail();
>> +
>> +                    bench_new(ctx, luma_filter, luma, luma_filter_size,
>> +                              chr_filter, chru, chrv, chr_filter_size,
>> +                              alpha, dst1, dstW, 0);
>> +                }
>> +            }
>> +        }
>> +    }
>> +    sws_freeContext(ctx);
>> +}
>> +
>> +
>> +void checkasm_check_sw_gbrp(void)
>> +{
>> +    check_output_yuv2gbrp();
>> +    report("output_yuv2gbrp");
>> +}
>> diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak
>> index 6e7edbe655..6db8f09d12 100644
>> --- a/tests/fate/checkasm.mak
>> +++ b/tests/fate/checkasm.mak
>> @@ -26,6 +26,7 @@ FATE_CHECKASM = fate-checkasm-aacpsdsp
>>                 \
>>                  fate-checkasm-pixblockdsp                               \
>>                  fate-checkasm-sbrdsp                                    \
>>                  fate-checkasm-synth_filter                              \
>> +                fate-checkasm-sw_gbrp                                   \
>>                  fate-checkasm-sw_rgb                                    \
>>                  fate-checkasm-sw_scale                                  \
>>                  fate-checkasm-utvideodsp                                \
>> --
>> 2.29.2
>>
>>
> ping
>
>

there is a few small changes I want to make to this, I also have the input
side working so, I'll submit a new version including that
diff mbox series

Patch

diff --git a/libswscale/x86/output.asm b/libswscale/x86/output.asm
index 52cf9f2c2e..ce0a2650b4 100644
--- a/libswscale/x86/output.asm
+++ b/libswscale/x86/output.asm
@@ -38,7 +38,49 @@  pw_32:         times 8 dw 32
 pd_255:        times 8 dd 255
 pw_512:        times 8 dw 512
 pw_1024:       times 8 dw 1024
-
+pd_65535_invf:             times 8 dd 0x37800080 ;1.0/65535.0
+pd_yuv2gbrp16_start:       times 8 dd -0x40000000
+pd_yuv2gbrp_y_start:       times 8 dd  (1 << 9)
+pd_yuv2gbrp_uv_start:      times 8 dd  ((1 << 9) - (128 << 19))
+pd_yuv2gbrp_a_start:       times 8 dd  (1 << 18)
+pd_yuv2gbrp16_offset:      times 8 dd  0x10000  ;(1 << 16)
+pd_yuv2gbrp16_round13:     times 8 dd  0x02000  ;(1 << 13)
+pd_yuv2gbrp16_a_offset:    times 8 dd  0x20002000
+pd_yuv2gbrp16_upper30:     times 8 dd  0x3FFFFFFF ;(1<<30) - 1
+pd_yuv2gbrp16_upper27:     times 8 dd  0x07FFFFFF ;(1<<27) - 1
+pd_yuv2gbrp16_upperC:      times 8 dd  0xC0000000
+pb_pack_shuffle8:       db  0,  4,  8, 12, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                            0,  4,  8, 12, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1
+pb_pack_shuffle16le:    db  0,  1,  4,  5, \
+                            8,  9, 12, 13, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                            0,  1,  4,  5, \
+                            8,  9, 12, 13
+pb_pack_shuffle16be:    db  1,  0,  5,  4, \
+                            9,  8, 13, 12, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                           -1, -1, -1, -1, \
+                            1,  0,  5,  4, \
+                            9,  8, 13, 12
+pb_shuffle32be          db  3,  2,  1,  0, \
+                            7,  6,  5,  4, \
+                           11, 10,  9,  8, \
+                           15, 14, 13, 12, \
+                            3,  2,  1,  0, \
+                            7,  6,  5,  4, \
+                           11, 10,  9,  8, \
+                           15, 14, 13, 12
 yuv2nv12_shuffle_mask: times 2 db 0,  4,  8, 12, \
                                  -1, -1, -1, -1, \
                                  -1, -1, -1, -1, \
@@ -549,3 +591,394 @@  yuv2nv12cX_fn yuv2nv12
 yuv2nv12cX_fn yuv2nv21
 %endif
 %endif ; ARCH_X86_64
+
+;-----------------------------------------------------------------------------
+; planar grb yuv2anyX functions
+; void ff_yuv2<gbr_format>_full_X_<opt>(SwsContext *c, const int16_t *lumFilter,
+;                                       const int16_t **lumSrcx, int lumFilterSize,
+;                                       const int16_t *chrFilter, const int16_t **chrUSrcx,
+;                                       const int16_t **chrVSrcx, int chrFilterSize,
+;                                       const int16_t **alpSrcx, uint8_t **dest,
+;                                       int dstW, int y)
+;-----------------------------------------------------------------------------
+
+%if ARCH_X86_64
+struc SwsContext
+    .padding:           resb 40292 ; offsetof(SwsContext, yuv2rgb_y_offset)
+    .yuv2rgb_y_offset:  resd 1
+    .yuv2rgb_y_coeff:   resd 1
+    .yuv2rgb_v2r_coeff: resd 1
+    .yuv2rgb_v2g_coeff: resd 1
+    .yuv2rgb_u2g_coeff: resd 1
+    .yuv2rgb_u2b_coeff: resd 1
+endstruc
+
+%define R m0
+%define G m1
+%define B m2
+%define A m3
+
+%define Y m4
+%define U m5
+%define V m6
+
+; Clip a signed integer to an unsigned power of two range.
+; av_clip_uintp2
+; 1 - dest
+; 2 - bit position to clip at
+%macro CLIPP2 2
+    ; (~a) >> 31 & ((1<<p) - 1);
+    pcmpeqb m4, m4
+    pxor m4, %1
+    psrad m4, 31
+    movu m5, [pd_yuv2gbrp16_upper%2]
+    pand m4, m5
+
+    ; (a & ~((1<<p) - 1)) == 0
+    pandn m5, %1
+    pxor m6, m6
+    pcmpeqd m5, m6
+%if cpuflag(avx2)
+    vpblendvb %1, m4, %1, m5
+%else
+    pxor %1, m4
+    pand %1, m5
+    pxor %1, m4
+%endif
+%endmacro
+
+; 1 - dest
+; 2 - source
+%macro LOAD16 2
+    %if cpuflag(avx2)
+        movu xm%1, %2
+        vpmovsxwd m%1, xm%1
+    %elif cpuflag(sse4)
+        movsd m%1, %2
+        pmovsxwd m%1, m%1
+    %else
+        movsd m%1, %2
+        punpcklwd m%1, m%1
+        psrad m%1, 16 ; sign extend
+    %endif
+%endmacro
+
+; 1 - dest
+; 2 - source
+; 3 - depth
+%macro LOAD_PIXELS 3
+    mov ptrq, [%2 + jq*8]
+%if %3 >= 16
+    movu m%1, [ptrq + xq*4]
+%else
+    LOAD16 %1, [ptrq + xq*2]
+%endif
+%endmacro
+
+; 1 - dest
+; 2 - source
+%macro STORE8 2
+    mov ptrq, %1
+    %if mmsize > 16
+        pshufb m%2, [pb_pack_shuffle8]
+        vextractf128 xm4, m%2, 1
+        por xm%2, xm4
+        movq [ptrq + xq], xm%2
+    %else
+        %if cpuflag(sse4)
+            pshufb m%2, [pb_pack_shuffle8]
+        %else
+            psrldq m4, m%2, 3
+            por m%2, m4
+            psrldq m4, m%2, 6
+            por m%2, m4
+        %endif
+        movd [ptrq + xq], m%2
+    %endif
+%endmacro
+
+; 1 - dest
+; 2 - source
+; 3 - is big endian
+%macro STORE16 3
+    mov ptrq, %1
+    %if mmsize > 16
+        %if %3 ; bigendian
+            pshufb m%2, [pb_pack_shuffle16be]
+        %else
+            pshufb m%2, [pb_pack_shuffle16le]
+        %endif
+        vextractf128 xm4, m%2, 1
+        por xm%2, xm4
+        movu [ptrq + xq*2], xm%2
+    %else
+        %if cpuflag(sse4) && %3 ; bigendian
+            pshufb m%2, [pb_pack_shuffle16be]
+        %elif cpuflag(sse4)
+            pshufb m%2, [pb_pack_shuffle16le]
+        %else
+            pshuflw m%2, m%2, (1 << 6 | 1 << 4 | 2 << 2 | 0 << 0)
+            pshufhw m%2, m%2, (1 << 6 | 1 << 4 | 2 << 2 | 0 << 0)
+            pshufd  m%2, m%2, (3 << 6 | 3 << 4 | 2 << 2 | 0 << 0)
+            %if %3 ; bigendian
+                psrlw  m4, m%2, 8
+                psllw  m%2, 8
+                por m%2, m4
+            %endif
+        %endif
+        movq [ptrq + xq*2], m%2
+    %endif
+%endmacro
+
+%macro SWAP32 1
+%if mmsize > 16 || cpuflag(sse4)
+    pshufb m%1, [pb_shuffle32be]
+%else
+    psrlw  m4, m%1, 8
+    psllw  m%1, 8
+    por m%1, m4
+    pshuflw m%1, m%1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0)
+    pshufhw m%1, m%1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0)
+%endif
+%endmacro
+
+; 1 - dest
+; 2 - source
+; 3 - depth
+; 4 - is big endian
+%macro STORE_PIXELS 4
+%if %3 > 16
+    %if %4
+        SWAP32 %2
+    %endif
+    mov ptrq, %1
+    movu [ptrq + xq*4], m%2
+%elif %3 > 8
+    STORE16 %1, %2, %4
+%else
+    STORE8 %1, %2
+%endif
+%endmacro
+
+%macro PMULLO 3
+%if cpuflag(sse4) || mmsize > 16
+    pmulld %1, %2, %3
+%else
+    %ifidni %1, %2
+    %else
+        mova %1, %2
+    %endif
+    pshufd m7, %1, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0) ; 0xb1
+    pshufd m8, %3, (2 << 6 | 3 << 4 | 0 << 2 | 1 << 0) ; 0xb1
+    pmuludq m7, m8
+    pshufd  m7, m7, (3 << 6 | 1 << 4 | 2 << 2 | 0 << 0) ; 0xd8
+    pmuludq %1, %3
+    pshufd  %1, %1, (3 << 6 | 1 << 4 | 2 << 2 | 0 << 0) ; 0xd8
+    punpckldq %1, m7
+%endif
+%endmacro
+
+; 1 - name
+; 2 - depth
+; 3 - has alpha
+; 3 - is big endian
+; 5 - is float
+%macro yuv2gbrp_fn 5
+%define DEPTH %2
+%define HAS_ALPHA %3
+%define IS_BE %4
+%define FLOAT %5
+%define SH (22 + 8 - DEPTH)
+
+%if DEPTH >= 16
+    %define RGB_SHIFT 14
+    %define A_SHIFT 14
+%elif 22 != SH
+    %define RGB_SHIFT SH
+    %define A_SHIFT (SH-3)
+%else
+    %define RGB_SHIFT 22
+    %define A_SHIFT 19
+%endif
+
+%if DEPTH >= 16
+    %define YUV_SHIFT 14
+    %define Y_START  m9
+    %define Y_ROUND [pd_yuv2gbrp16_round13]
+    %define UV_START m9
+    %define A_START  m9
+    %define A_CLIP2P 30
+%else
+    %define YUV_SHIFT 10
+    %define Y_START  [pd_yuv2gbrp_y_start]
+    %define Y_ROUND  m9
+    %define UV_START [pd_yuv2gbrp_uv_start]
+    %define A_START  [pd_yuv2gbrp_a_start]
+    %define A_CLIP2P 27
+%endif
+
+cglobal yuv2%1_full_X, 12, 14, 16, ptr, lumFilter, lumSrcx, lumFilterSize, chrFilter, chrUSrcx, chrVSrcx, chrFilterSize, alpSrcx, dest, dstW, y, x, j
+    VBROADCASTSS m10, dword [ptrq + SwsContext.yuv2rgb_y_offset]
+    VBROADCASTSS m11, dword [ptrq + SwsContext.yuv2rgb_y_coeff]
+    VBROADCASTSS m12, dword [ptrq + SwsContext.yuv2rgb_v2r_coeff]
+    VBROADCASTSS m13, dword [ptrq + SwsContext.yuv2rgb_v2g_coeff]
+    VBROADCASTSS m14, dword [ptrq + SwsContext.yuv2rgb_u2g_coeff]
+    VBROADCASTSS m15, dword [ptrq + SwsContext.yuv2rgb_u2b_coeff]
+
+%if DEPTH >= 16
+    movu m9, [pd_yuv2gbrp16_start]
+%else
+    mov xq, (1 << (SH-1))
+    movq xm9, xq
+    VBROADCASTSS m9, xm9
+%endif
+    xor xq, xq
+
+    %%loop_x:
+        movu Y, Y_START
+        movu U, UV_START
+        movu V, UV_START
+
+        xor jq, jq
+        %%loop_luma:
+            movsx ptrd, word [lumFilterq + jq*2]
+            movd xm0, ptrd
+            VBROADCASTSS m0, xm0
+            LOAD_PIXELS 1, lumSrcxq, DEPTH
+            PMULLO m1, m1, m0
+            paddd Y, m1
+            inc jd
+            cmp jd, lumFilterSized
+            jl %%loop_luma
+
+%if HAS_ALPHA
+        cmp alpSrcxq, 0
+        je %%skip_alpha_load
+            xor jq, jq
+            movu A, A_START
+            %%loop_alpha:
+                movsx ptrd, word [lumFilterq + jq*2]
+                movd xm0, ptrd
+                VBROADCASTSS m0, xm0
+                LOAD_PIXELS 1, alpSrcxq, DEPTH
+                PMULLO m1, m1, m0
+                paddd A, m1
+                inc jd
+                cmp jd, lumFilterSized
+                jl %%loop_alpha
+%if DEPTH >= 16
+            psrad A, 1
+            paddd A, [pd_yuv2gbrp16_a_offset]
+%endif
+        %%skip_alpha_load:
+%endif
+        xor jq, jq
+        %%loop_chr:
+            movsx ptrd, word [chrFilterq + jq*2]
+            movd xm0, ptrd
+            VBROADCASTSS m0, xm0
+            LOAD_PIXELS 1, chrUSrcxq, DEPTH
+            LOAD_PIXELS 2, chrVSrcxq, DEPTH
+            PMULLO m1, m1, m0
+            PMULLO m2, m2, m0
+            paddd U, m1
+            paddd V, m2
+            inc jd
+            cmp jd, chrFilterSized
+            jl %%loop_chr
+
+        psrad Y, YUV_SHIFT
+%if  DEPTH >= 16
+        paddd Y, [pd_yuv2gbrp16_offset]
+%endif
+        psrad U, YUV_SHIFT
+        psrad V, YUV_SHIFT
+
+        psubd  Y, m10    ; yuv2rgb_y_offset
+        PMULLO Y, Y, m11 ; yuv2rgb_y_coeff
+        paddd  Y, Y_ROUND
+
+        PMULLO R, V, m12 ; yuv2rgb_v2r_coeff
+        PMULLO B, U, m15 ; yuv2rgb_u2b_coeff
+
+        PMULLO U, U, m14 ; yuv2rgb_u2g_coeff
+        PMULLO V, V, m13 ; yuv2rgb_v2g_coeff
+        paddd G, U, V
+        paddd R, Y
+        paddd G, Y
+        paddd B, Y
+
+        CLIPP2 R, 30
+        CLIPP2 G, 30
+        CLIPP2 B, 30
+
+        psrad R, RGB_SHIFT
+        psrad G, RGB_SHIFT
+        psrad B, RGB_SHIFT
+
+%if FLOAT
+        cvtdq2ps R, R
+        cvtdq2ps G, G
+        cvtdq2ps B, B
+        mulps R, [pd_65535_invf]
+        mulps G, [pd_65535_invf]
+        mulps B, [pd_65535_invf]
+%endif
+        STORE_PIXELS [destq +  0], 1, DEPTH, IS_BE ; G
+        STORE_PIXELS [destq +  8], 2, DEPTH, IS_BE ; B
+        STORE_PIXELS [destq + 16], 0, DEPTH, IS_BE ; R
+
+%if HAS_ALPHA
+        cmp alpSrcxq, 0
+        je %%skip_alpha_store
+            CLIPP2 A, A_CLIP2P
+            psrad A, A_SHIFT
+%if FLOAT
+            cvtdq2ps A, A
+            mulps A, [pd_65535_invf]
+%endif
+            STORE_PIXELS [destq + 24], 3, DEPTH, IS_BE
+        %%skip_alpha_store:
+%endif
+        add xq, mmsize/4
+        cmp xd, dstWd
+        jl %%loop_x
+
+    RET
+%endmacro
+
+%macro yuv2gbrp_fn_decl 2
+INIT_%1 %2
+yuv2gbrp_fn gbrp,        8, 0, 0, 0
+yuv2gbrp_fn gbrap,       8, 1, 0, 0
+yuv2gbrp_fn gbrp9le,     9, 0, 0, 0
+yuv2gbrp_fn gbrp10le,   10, 0, 0, 0
+yuv2gbrp_fn gbrap10le,  10, 1, 0, 0
+yuv2gbrp_fn gbrp12le,   12, 0, 0, 0
+yuv2gbrp_fn gbrap12le,  12, 1, 0, 0
+yuv2gbrp_fn gbrp14le,   14, 0, 0, 0
+yuv2gbrp_fn gbrp16le,   16, 0, 0, 0
+yuv2gbrp_fn gbrap16le,  16, 1, 0, 0
+yuv2gbrp_fn gbrpf32le,  32, 0, 0, 1
+yuv2gbrp_fn gbrapf32le, 32, 1, 0, 1
+
+yuv2gbrp_fn gbrp9be,     9, 0, 1, 0
+yuv2gbrp_fn gbrp10be,   10, 0, 1, 0
+yuv2gbrp_fn gbrap10be,  10, 1, 1, 0
+yuv2gbrp_fn gbrp12be,   12, 0, 1, 0
+yuv2gbrp_fn gbrap12be,  12, 1, 1, 0
+yuv2gbrp_fn gbrp14be,   14, 0, 1, 0
+yuv2gbrp_fn gbrp16be,   16, 0, 1, 0
+yuv2gbrp_fn gbrap16be,  16, 1, 1, 0
+yuv2gbrp_fn gbrpf32be,  32, 0, 1, 1
+yuv2gbrp_fn gbrapf32be, 32, 1, 1, 1
+%endmacro
+
+yuv2gbrp_fn_decl XMM, sse2
+yuv2gbrp_fn_decl XMM, sse4
+
+%if HAVE_AVX2_EXTERNAL
+yuv2gbrp_fn_decl YMM, avx2
+%endif
+
+%endif ; ARCH_X86_64
\ No newline at end of file
diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
index 0848a31461..e8bc15b6ee 100644
--- a/libswscale/x86/swscale.c
+++ b/libswscale/x86/swscale.c
@@ -350,6 +350,43 @@  void ff_yuv2 ## fmt ## cX_ ## opt(enum AVPixelFormat format, const uint8_t *dith

 YUV2NV_DECL(nv12, avx2);
 YUV2NV_DECL(nv21, avx2);
+
+#define YUV2GBRP_FN_DECL(fmt, opt)                                                      \
+void ff_yuv2##fmt##_full_X_ ##opt(SwsContext *c, const int16_t *lumFilter,           \
+                                 const int16_t **lumSrcx, int lumFilterSize,         \
+                                 const int16_t *chrFilter, const int16_t **chrUSrcx, \
+                                 const int16_t **chrVSrcx, int chrFilterSize,        \
+                                 const int16_t **alpSrcx, uint8_t **dest,            \
+                                 int dstW, int y)
+
+#define YUV2GBRP_DECL(opt)     \
+YUV2GBRP_FN_DECL(gbrp,       opt); \
+YUV2GBRP_FN_DECL(gbrap,      opt); \
+YUV2GBRP_FN_DECL(gbrp9le,    opt); \
+YUV2GBRP_FN_DECL(gbrp10le,   opt); \
+YUV2GBRP_FN_DECL(gbrap10le,  opt); \
+YUV2GBRP_FN_DECL(gbrp12le,   opt); \
+YUV2GBRP_FN_DECL(gbrap12le,  opt); \
+YUV2GBRP_FN_DECL(gbrp14le,   opt); \
+YUV2GBRP_FN_DECL(gbrp16le,   opt); \
+YUV2GBRP_FN_DECL(gbrap16le,  opt); \
+YUV2GBRP_FN_DECL(gbrpf32le,  opt); \
+YUV2GBRP_FN_DECL(gbrapf32le, opt); \
+YUV2GBRP_FN_DECL(gbrp9be,    opt); \
+YUV2GBRP_FN_DECL(gbrp10be,   opt); \
+YUV2GBRP_FN_DECL(gbrap10be,  opt); \
+YUV2GBRP_FN_DECL(gbrp12be,   opt); \
+YUV2GBRP_FN_DECL(gbrap12be,  opt); \
+YUV2GBRP_FN_DECL(gbrp14be,   opt); \
+YUV2GBRP_FN_DECL(gbrp16be,   opt); \
+YUV2GBRP_FN_DECL(gbrap16be,  opt); \
+YUV2GBRP_FN_DECL(gbrpf32be,  opt); \
+YUV2GBRP_FN_DECL(gbrapf32be, opt);
+
+YUV2GBRP_DECL(sse2);
+YUV2GBRP_DECL(sse4);
+YUV2GBRP_DECL(avx2);
+
 #endif

 av_cold void ff_sws_init_swscale_x86(SwsContext *c)
@@ -582,5 +619,66 @@  switch(c->dstBpc){ \
             break;
         }
     }
+
+    if(c->flags & SWS_FULL_CHR_H_INT) {
+
+        /* yuv2gbrp uses the SwsContext for yuv coefficients
+           if struct offsets change the asm needs to be updated too */
+        av_assert0(offsetof(SwsContext, yuv2rgb_y_offset) == 40292);
+
+#define YUV2ANYX_FUNC_CASE(fmt, name, opt)              \
+        case fmt:                                       \
+            c->yuv2anyX = ff_yuv2##name##_full_X_##opt; \
+            break;
+
+#define YUV2ANYX_GBRAP_CASES(opt)                                  \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP,       gbrp,       opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP,      gbrap,      opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP9LE,    gbrp9le,    opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP10LE,   gbrp10le,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP10LE,  gbrap10le,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP12LE,   gbrp12le,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP12LE,  gbrap12le,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP14LE,   gbrp14le,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP16LE,   gbrp16le,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP16LE,  gbrap16le,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRPF32LE,  gbrpf32le,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAPF32LE, gbrapf32le, opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP9BE,    gbrp9be,    opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP10BE,   gbrp10be,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP10BE,  gbrap10be,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP12BE,   gbrp12be,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP12BE,  gbrap12be,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP14BE,   gbrp14be,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRP16BE,   gbrp16be,   opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAP16BE,  gbrap16be,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRPF32BE,  gbrpf32be,  opt) \
+        YUV2ANYX_FUNC_CASE(AV_PIX_FMT_GBRAPF32BE, gbrapf32be, opt)
+
+        if (EXTERNAL_SSE2(cpu_flags)) {
+            switch (c->dstFormat) {
+            YUV2ANYX_GBRAP_CASES(sse2)
+            default:
+                break;
+            }
+        }
+
+        if (EXTERNAL_SSE4(cpu_flags)) {
+            switch (c->dstFormat) {
+            YUV2ANYX_GBRAP_CASES(sse4)
+            default:
+                break;
+            }
+        }
+
+        if (EXTERNAL_AVX2_FAST(cpu_flags)) {
+            switch (c->dstFormat) {
+            YUV2ANYX_GBRAP_CASES(avx2)
+            default:
+                break;
+            }
+        }
+    }
+
 #endif
 }
diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
index 4ef5fa87da..f768b1144e 100644
--- a/tests/checkasm/Makefile
+++ b/tests/checkasm/Makefile
@@ -45,7 +45,7 @@  AVFILTEROBJS-$(CONFIG_NLMEANS_FILTER)    += vf_nlmeans.o
 CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes)

 # swscale tests
-SWSCALEOBJS                             += sw_rgb.o sw_scale.o
+SWSCALEOBJS                             += sw_gbrp.o sw_rgb.o sw_scale.o

 CHECKASMOBJS-$(CONFIG_SWSCALE)  += $(SWSCALEOBJS)

diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index b1353f7cbe..5fa8ac7bcc 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -183,6 +183,7 @@  static const struct {
     #endif
 #endif
 #if CONFIG_SWSCALE
+    { "sw_gbrp", checkasm_check_sw_gbrp },
     { "sw_rgb", checkasm_check_sw_rgb },
     { "sw_scale", checkasm_check_sw_scale },
 #endif
diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
index 68b0697d3e..c3192d8c23 100644
--- a/tests/checkasm/checkasm.h
+++ b/tests/checkasm/checkasm.h
@@ -72,6 +72,7 @@  void checkasm_check_opusdsp(void);
 void checkasm_check_pixblockdsp(void);
 void checkasm_check_sbrdsp(void);
 void checkasm_check_synth_filter(void);
+void checkasm_check_sw_gbrp(void);
 void checkasm_check_sw_rgb(void);
 void checkasm_check_sw_scale(void);
 void checkasm_check_utvideodsp(void);
diff --git a/tests/checkasm/sw_gbrp.c b/tests/checkasm/sw_gbrp.c
new file mode 100644
index 0000000000..320948a71d
--- /dev/null
+++ b/tests/checkasm/sw_gbrp.c
@@ -0,0 +1,198 @@ 
+/*
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <string.h>
+
+#include "libavutil/common.h"
+#include "libavutil/intreadwrite.h"
+#include "libavutil/mem_internal.h"
+#include "libavutil/pixdesc.h"
+
+#include "libswscale/swscale.h"
+#include "libswscale/swscale_internal.h"
+
+#include "checkasm.h"
+
+#define randomize_buffers(buf, size)      \
+    do {                                  \
+        int j;                            \
+        for (j = 0; j < size; j+=4)       \
+            AV_WN32(buf + j, rnd());      \
+    } while (0)
+
+static const int planar_fmts[] = {
+    AV_PIX_FMT_GBRP,
+    AV_PIX_FMT_GBRP9BE,
+    AV_PIX_FMT_GBRP9LE,
+    AV_PIX_FMT_GBRP10BE,
+    AV_PIX_FMT_GBRP10LE,
+    AV_PIX_FMT_GBRP12BE,
+    AV_PIX_FMT_GBRP12LE,
+    AV_PIX_FMT_GBRP14BE,
+    AV_PIX_FMT_GBRP14LE,
+    AV_PIX_FMT_GBRAP,
+    AV_PIX_FMT_GBRAP10BE,
+    AV_PIX_FMT_GBRAP10LE,
+    AV_PIX_FMT_GBRAP12BE,
+    AV_PIX_FMT_GBRAP12LE,
+    AV_PIX_FMT_GBRP16BE,
+    AV_PIX_FMT_GBRP16LE,
+    AV_PIX_FMT_GBRAP16BE,
+    AV_PIX_FMT_GBRAP16LE,
+    AV_PIX_FMT_GBRPF32BE,
+    AV_PIX_FMT_GBRPF32LE,
+    AV_PIX_FMT_GBRAPF32BE,
+    AV_PIX_FMT_GBRAPF32LE
+};
+
+static void check_output_yuv2gbrp(void)
+{
+    struct SwsContext *ctx;
+    const AVPixFmtDescriptor *desc;
+    int fmi, fsi, isi, i;
+    int dstW, byte_size, luma_filter_size, chr_filter_size;
+#define LARGEST_FILTER 16
+#define FILTER_SIZES 4
+    static const int filter_sizes[FILTER_SIZES] = {1, 4, 8, 16};
+#define LARGEST_INPUT_SIZE 512
+#define INPUT_SIZES 6
+    static const int input_sizes[INPUT_SIZES] = {8, 24, 128, 144, 256, 512};
+    uint8_t *dst0[4];
+    uint8_t *dst1[4];
+
+    declare_func(void, void *c, const int16_t *lumFilter,
+                       const int16_t **lumSrcx, int lumFilterSize,
+                       const int16_t *chrFilter, const int16_t **chrUSrcx,
+                       const int16_t **chrVSrcx, int chrFilterSize,
+                       const int16_t **alpSrcx, uint8_t **dest,
+                       int dstW, int y);
+
+    const int16_t *luma[LARGEST_FILTER];
+    const int16_t *chru[LARGEST_FILTER];
+    const int16_t *chrv[LARGEST_FILTER];
+    const int16_t *alpha[LARGEST_FILTER];
+
+    LOCAL_ALIGNED_8(int16_t, luma_filter, [LARGEST_FILTER]);
+    LOCAL_ALIGNED_8(int16_t, chr_filter, [LARGEST_FILTER]);
+
+    LOCAL_ALIGNED_8(int32_t, src_y, [LARGEST_FILTER * LARGEST_INPUT_SIZE]);
+    LOCAL_ALIGNED_8(int32_t, src_u, [LARGEST_FILTER * LARGEST_INPUT_SIZE]);
+    LOCAL_ALIGNED_8(int32_t, src_v, [LARGEST_FILTER * LARGEST_INPUT_SIZE]);
+    LOCAL_ALIGNED_8(int32_t, src_a, [LARGEST_FILTER * LARGEST_INPUT_SIZE]);
+
+    LOCAL_ALIGNED_8(uint8_t, dst0_r, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+    LOCAL_ALIGNED_8(uint8_t, dst0_g, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+    LOCAL_ALIGNED_8(uint8_t, dst0_b, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+    LOCAL_ALIGNED_8(uint8_t, dst0_a, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+
+    LOCAL_ALIGNED_8(uint8_t, dst1_r, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+    LOCAL_ALIGNED_8(uint8_t, dst1_g, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+    LOCAL_ALIGNED_8(uint8_t, dst1_b, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+    LOCAL_ALIGNED_8(uint8_t, dst1_a, [LARGEST_INPUT_SIZE * sizeof(int32_t)]);
+
+    randomize_buffers((uint8_t*)src_y, LARGEST_FILTER * LARGEST_INPUT_SIZE * sizeof(int32_t));
+    randomize_buffers((uint8_t*)src_u, LARGEST_FILTER * LARGEST_INPUT_SIZE * sizeof(int32_t));
+    randomize_buffers((uint8_t*)src_v, LARGEST_FILTER * LARGEST_INPUT_SIZE * sizeof(int32_t));
+    randomize_buffers((uint8_t*)src_a, LARGEST_FILTER * LARGEST_INPUT_SIZE * sizeof(int32_t));
+    randomize_buffers((uint8_t*)luma_filter, LARGEST_FILTER * sizeof(int16_t));
+    randomize_buffers((uint8_t*)chr_filter, LARGEST_FILTER * sizeof(int16_t));
+
+    dst0[0] = (uint8_t*)dst0_g;
+    dst0[1] = (uint8_t*)dst0_b;
+    dst0[2] = (uint8_t*)dst0_r;
+    dst0[3] = (uint8_t*)dst0_a;
+
+    dst1[0] = (uint8_t*)dst1_g;
+    dst1[1] = (uint8_t*)dst1_b;
+    dst1[2] = (uint8_t*)dst1_r;
+    dst1[3] = (uint8_t*)dst1_a;
+
+    for (i = 0; i < LARGEST_FILTER; i++) {
+        luma[i] =  (int16_t *)(src_y + i*LARGEST_INPUT_SIZE);
+        chru[i] =  (int16_t *)(src_u + i*LARGEST_INPUT_SIZE);
+        chrv[i] =  (int16_t *)(src_v + i*LARGEST_INPUT_SIZE);
+        alpha[i] = (int16_t *)(src_a + i*LARGEST_INPUT_SIZE);
+    }
+
+    ctx = sws_alloc_context();
+    if (sws_init_context(ctx, NULL, NULL) < 0)
+        fail();
+
+    ctx->flags |= SWS_FULL_CHR_H_INT;
+    ctx->yuv2rgb_y_offset  = rnd();
+    ctx->yuv2rgb_y_coeff   = rnd();
+    ctx->yuv2rgb_v2r_coeff = rnd();
+    ctx->yuv2rgb_v2g_coeff = rnd();
+    ctx->yuv2rgb_u2g_coeff = rnd();
+    ctx->yuv2rgb_u2b_coeff = rnd();
+
+    for (fmi = 0; fmi < FF_ARRAY_ELEMS(planar_fmts); fmi++) {
+        for (fsi = 0; fsi < FILTER_SIZES; fsi++) {
+            for (isi = 0; isi < INPUT_SIZES; isi++ ) {
+                desc = av_pix_fmt_desc_get(planar_fmts[fmi]);
+                ctx->dstFormat = planar_fmts[fmi];
+
+                dstW = input_sizes[isi];
+                luma_filter_size = filter_sizes[fsi];
+                chr_filter_size = filter_sizes[fsi];
+
+                if (desc->comp[0].depth > 16) {
+                    byte_size = 4;
+                } else if (desc->comp[0].depth > 8) {
+                    byte_size = 2;
+                } else {
+                    byte_size = 1;
+                }
+
+                ff_sws_init_scale(ctx);
+                if (check_func(ctx->yuv2anyX, "yuv2%s_full_X_%d_%d", desc->name, luma_filter_size, dstW)) {
+                    for (i = 0; i < 4; i ++) {
+                        memset(dst0[i], 0xFF, LARGEST_INPUT_SIZE * sizeof(int32_t));
+                        memset(dst1[i], 0xFF, LARGEST_INPUT_SIZE * sizeof(int32_t));
+                    }
+
+                    call_ref(ctx, luma_filter, luma, luma_filter_size,
+                             chr_filter, chru, chrv, chr_filter_size,
+                             alpha, dst0, dstW, 0);
+                    call_new(ctx, luma_filter, luma, luma_filter_size,
+                             chr_filter, chru, chrv, chr_filter_size,
+                             alpha, dst1, dstW, 0);
+
+                    if (memcmp(dst0[0], dst1[0], dstW * byte_size) ||
+                        memcmp(dst0[1], dst1[1], dstW * byte_size) ||
+                        memcmp(dst0[2], dst1[2], dstW * byte_size) ||
+                        memcmp(dst0[3], dst1[3], dstW * byte_size) )
+                        fail();
+
+                    bench_new(ctx, luma_filter, luma, luma_filter_size,
+                              chr_filter, chru, chrv, chr_filter_size,
+                              alpha, dst1, dstW, 0);
+                }
+            }
+        }
+    }
+    sws_freeContext(ctx);
+}
+
+
+void checkasm_check_sw_gbrp(void)
+{
+    check_output_yuv2gbrp();
+    report("output_yuv2gbrp");
+}
diff --git a/tests/fate/checkasm.mak b/tests/fate/checkasm.mak
index 6e7edbe655..6db8f09d12 100644
--- a/tests/fate/checkasm.mak
+++ b/tests/fate/checkasm.mak
@@ -26,6 +26,7 @@  FATE_CHECKASM = fate-checkasm-aacpsdsp                                  \
                 fate-checkasm-pixblockdsp                               \
                 fate-checkasm-sbrdsp                                    \
                 fate-checkasm-synth_filter                              \
+                fate-checkasm-sw_gbrp                                   \
                 fate-checkasm-sw_rgb                                    \
                 fate-checkasm-sw_scale                                  \
                 fate-checkasm-utvideodsp                                \