[FFmpeg-devel] [PATCH 3/4] swscale/x86: add sse4 {lum, chr}ConvertRange

Ramiro Polla ramiro.polla at gmail.com
Fri Jun 7 20:38:26 EEST 2024


On Fri, Jun 7, 2024 at 4:05 PM Ramiro Polla <ramiro.polla at gmail.com> wrote:
>
> chrRangeFromJpeg_8_c: 19.9
> chrRangeFromJpeg_8_sse4: 16.2
> chrRangeFromJpeg_24_c: 60.7
> chrRangeFromJpeg_24_sse4: 28.9
> chrRangeFromJpeg_128_c: 325.7
> chrRangeFromJpeg_128_sse4: 160.2
> chrRangeFromJpeg_144_c: 364.2
> chrRangeFromJpeg_144_sse4: 194.9
> chrRangeFromJpeg_256_c: 630.7
> chrRangeFromJpeg_256_sse4: 337.4
> chrRangeFromJpeg_512_c: 1240.4
> chrRangeFromJpeg_512_sse4: 668.4
> chrRangeToJpeg_8_c: 37.7
> chrRangeToJpeg_8_sse4: 19.7
> chrRangeToJpeg_24_c: 114.7
> chrRangeToJpeg_24_sse4: 30.2
> chrRangeToJpeg_128_c: 636.4
> chrRangeToJpeg_128_sse4: 161.7
> chrRangeToJpeg_144_c: 715.7
> chrRangeToJpeg_144_sse4: 272.9
> chrRangeToJpeg_256_c: 1256.7
> chrRangeToJpeg_256_sse4: 341.9
> chrRangeToJpeg_512_c: 2498.7
> chrRangeToJpeg_512_sse4: 668.4
> lumRangeFromJpeg_8_c: 11.7
> lumRangeFromJpeg_8_sse4: 12.4
> lumRangeFromJpeg_24_c: 36.9
> lumRangeFromJpeg_24_sse4: 17.7
> lumRangeFromJpeg_128_c: 228.4
> lumRangeFromJpeg_128_sse4: 85.2
> lumRangeFromJpeg_144_c: 272.9
> lumRangeFromJpeg_144_sse4: 96.9
> lumRangeFromJpeg_256_c: 463.4
> lumRangeFromJpeg_256_sse4: 183.9
> lumRangeFromJpeg_512_c: 879.9
> lumRangeFromJpeg_512_sse4: 355.9
> lumRangeToJpeg_8_c: 17.7
> lumRangeToJpeg_8_sse4: 15.4
> lumRangeToJpeg_24_c: 56.2
> lumRangeToJpeg_24_sse4: 18.4
> lumRangeToJpeg_128_c: 331.4
> lumRangeToJpeg_128_sse4: 84.4
> lumRangeToJpeg_144_c: 375.2
> lumRangeToJpeg_144_sse4: 96.9
> lumRangeToJpeg_256_c: 649.7
> lumRangeToJpeg_256_sse4: 184.4
> lumRangeToJpeg_512_c: 1281.9
> lumRangeToJpeg_512_sse4: 355.9
> ---
>  libswscale/swscale_internal.h    |   1 +
>  libswscale/utils.c               |   2 +
>  libswscale/x86/Makefile          |   1 +
>  libswscale/x86/range_convert.asm | 100 +++++++++++++++++++++++++++++++
>  libswscale/x86/swscale.c         |  36 +++++++++++
>  5 files changed, 140 insertions(+)
>  create mode 100644 libswscale/x86/range_convert.asm
>
> diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
> index d4b0c3cee2..92f6105443 100644
> --- a/libswscale/swscale_internal.h
> +++ b/libswscale/swscale_internal.h
> @@ -698,6 +698,7 @@ void ff_updateMMXDitherTables(SwsContext *c, int dstY);
>
>  av_cold void ff_sws_init_range_convert(SwsContext *c);
>  av_cold void ff_sws_init_range_convert_loongarch(SwsContext *c);
> +av_cold void ff_sws_init_range_convert_x86(SwsContext *c);
>
>  SwsFunc ff_yuv2rgb_init_x86(SwsContext *c);
>  SwsFunc ff_yuv2rgb_init_ppc(SwsContext *c);
> diff --git a/libswscale/utils.c b/libswscale/utils.c
> index 476a24fea5..8dfa57b5ff 100644
> --- a/libswscale/utils.c
> +++ b/libswscale/utils.c
> @@ -1082,6 +1082,8 @@ int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
>          ff_sws_init_range_convert(c);
>  #if ARCH_LOONGARCH64
>          ff_sws_init_range_convert_loongarch(c);
> +#elif ARCH_X86
> +        ff_sws_init_range_convert_x86(c);
>  #endif
>      }
>
> diff --git a/libswscale/x86/Makefile b/libswscale/x86/Makefile
> index 68391494be..f00154941d 100644
> --- a/libswscale/x86/Makefile
> +++ b/libswscale/x86/Makefile
> @@ -12,6 +12,7 @@ X86ASM-OBJS                     += x86/input.o                          \
>                                     x86/output.o                         \
>                                     x86/scale.o                          \
>                                     x86/scale_avx2.o                          \
> +                                   x86/range_convert.o                  \
>                                     x86/rgb_2_rgb.o                      \
>                                     x86/yuv_2_rgb.o                      \
>                                     x86/yuv2yuvX.o                       \
> diff --git a/libswscale/x86/range_convert.asm b/libswscale/x86/range_convert.asm
> new file mode 100644
> index 0000000000..333265fb65
> --- /dev/null
> +++ b/libswscale/x86/range_convert.asm
> @@ -0,0 +1,100 @@
> +;******************************************************************************
> +;* Copyright (c) 2024 Ramiro Polla
> +;*
> +;* This file is part of FFmpeg.
> +;*
> +;* FFmpeg is free software; you can redistribute it and/or
> +;* modify it under the terms of the GNU Lesser General Public
> +;* License as published by the Free Software Foundation; either
> +;* version 2.1 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
> +;* Lesser General Public License for more details.
> +;*
> +;* You should have received a copy of the GNU Lesser 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 "libavutil/x86/x86util.asm"
> +
> +; NOTE: there is no need to clamp the input when converting to jpeg range
> +;       (like we do in the C code) because packssdw will saturate the output.
> +
> +;-----------------------------------------------------------------------------
> +; lumConvertRange
> +;
> +; void ff_lumRangeToJpeg_<opt>(int16_t *dst, int width);
> +; void ff_lumRangeFromJpeg_<opt>(int16_t *dst, int width);
> +;
> +;-----------------------------------------------------------------------------
> +
> +%macro LUMCONVERTRANGE 4
> +SECTION_RODATA
> +mult_%1:        times 4 dd %2
> +offset_%1:      times 4 dd %3
> +SECTION .text
> +cglobal %1, 2, 3, 3, dst, width, x
> +    movsxdifnidn widthq, widthd
> +    xor              xq, xq
> +    mova             m1, [mult_%1]
> +    mova             m2, [offset_%1]
> +.loop:
> +    pmovsxwd         m0, [dstq+xq*2]
> +    pmulld           m0, m1
> +    paddd            m0, m2
> +    psrad            m0, %4
> +    packssdw         m0, m0
> +    movh    [dstq+xq*2], m0
> +    add              xq, mmsize / 4
> +    cmp              xd, widthd
> +    jl .loop
> +    RET
> +%endmacro
> +
> +;-----------------------------------------------------------------------------
> +; chrConvertRange
> +;
> +; void ff_chrRangeToJpeg_<opt>(int16_t *dstU, int16_t *dstV, int width);
> +; void ff_chrRangeFromJpeg_<opt>(int16_t *dstU, int16_t *dstV, int width);
> +;
> +;-----------------------------------------------------------------------------
> +
> +%macro CHRCONVERTRANGE 4
> +SECTION_RODATA
> +mult_%1:        times 4 dd %2
> +offset_%1:      times 4 dd %3
> +SECTION .text
> +cglobal %1, 3, 4, 4, dstU, dstV, width, x
> +    movsxdifnidn widthq, widthd
> +    xor              xq, xq
> +    mova             m1, [mult_%1]
> +    mova             m2, [offset_%1]
> +.loop:
> +    pmovsxwd         m0, [dstUq+xq*2]
> +    pmulld           m0, m1
> +    paddd            m0, m2
> +    psrad            m0, %4
> +    packssdw         m0, m0
> +    movh   [dstUq+xq*2], m0
> +    pmovsxwd         m0, [dstVq+xq*2]
> +    pmulld           m0, m1
> +    paddd            m0, m2
> +    psrad            m0, %4
> +    packssdw         m0, m0
> +    movh   [dstVq+xq*2], m0
> +    add              xq, mmsize / 4
> +    cmp              xd, widthd
> +    jl .loop
> +    RET
> +%endmacro
> +
> +%if ARCH_X86_64
> +INIT_XMM sse4
> +LUMCONVERTRANGE lumRangeToJpeg,   19077, -39057361, 14
> +CHRCONVERTRANGE chrRangeToJpeg,    4663,  -9289992, 12
> +LUMCONVERTRANGE lumRangeFromJpeg, 14071,  33561947, 14
> +CHRCONVERTRANGE chrRangeFromJpeg,  1799,   4081085, 11
> +%endif
> diff --git a/libswscale/x86/swscale.c b/libswscale/x86/swscale.c
> index fff8bb4396..c5ddfb5605 100644
> --- a/libswscale/x86/swscale.c
> +++ b/libswscale/x86/swscale.c
> @@ -447,6 +447,38 @@ INPUT_PLANAR_RGB_UV_ALL_DECL(avx2);
>  INPUT_PLANAR_RGB_A_ALL_DECL(avx2);
>  #endif
>
> +#if ARCH_X86_64
> +#define RANGE_CONVERT_FUNCS(opt) do {                                       \
> +    if (c->dstBpc <= 14) {                                                  \
> +        if (c->srcRange) {                                                  \
> +            c->lumConvertRange = ff_lumRangeFromJpeg_ ##opt;                \
> +            c->chrConvertRange = ff_chrRangeFromJpeg_ ##opt;                \
> +        } else {                                                            \
> +            c->lumConvertRange = ff_lumRangeToJpeg_ ##opt;                  \
> +            c->chrConvertRange = ff_chrRangeToJpeg_ ##opt;                  \
> +        }                                                                   \
> +    }                                                                       \
> +} while (0)
> +
> +#define RANGE_CONVERT_FUNCS_DECL(opt)                                       \
> +void ff_lumRangeFromJpeg_ ##opt(int16_t *dst, int width);                   \
> +void ff_chrRangeFromJpeg_ ##opt(int16_t *dstU, int16_t *dstV, int width);   \
> +void ff_lumRangeToJpeg_ ##opt(int16_t *dst, int width);                     \
> +void ff_chrRangeToJpeg_ ##opt(int16_t *dstU, int16_t *dstV, int width);     \
> +
> +RANGE_CONVERT_FUNCS_DECL(sse4);
> +
> +av_cold void ff_sws_init_range_convert_x86(SwsContext *c)
> +{
> +    if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
> +        int cpu_flags = av_get_cpu_flags();
> +        if (EXTERNAL_SSE4(cpu_flags)) {
> +            RANGE_CONVERT_FUNCS(sse4);
> +        }
> +    }
> +}
> +#endif
> +
>  av_cold void ff_sws_init_swscale_x86(SwsContext *c)
>  {
>      int cpu_flags = av_get_cpu_flags();
> @@ -805,4 +837,8 @@ switch(c->dstBpc){ \
>      }
>
>  #endif
> +
> +#if ARCH_X86_64
> +    ff_sws_init_range_convert_x86(c);
> +#endif
>  }
> --
> 2.30.2
>

Attached version is a little bit different, moving the consts out of
the macro (so they can be reused by avx2) and processing twice the
amount of data per loop.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-swscale-x86-add-sse4-lum-chr-ConvertRange.patch
Type: text/x-patch
Size: 10363 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20240607/245e9583/attachment.bin>


More information about the ffmpeg-devel mailing list