From: 陈昊 <chenhao@loongson.cn>
To: "FFmpeg development discussions and patches" <ffmpeg-devel@ffmpeg.org>
Subject: Re: [FFmpeg-devel] [PATCH 1/3] swscale: [LA] Optimize range convert for yuvj420p.
Date: Sat, 16 Mar 2024 14:22:07 +0800 (GMT+08:00)
Message-ID: <13d6431d.17d3.18e45ecc173.Coremail.chenhao@loongson.cn> (raw)
In-Reply-To: <20240316030333.31269-2-yinshiyou-hf@loongson.cn>
LGTM
2024-03-16 11:03:31 "yinshiyou-hf@loongson.cn" <yinshiyou-hf@loongson.cn> 写道：
> ---
> libswscale/loongarch/swscale.S | 368 ++++++++++++++++++
> libswscale/loongarch/swscale_init_loongarch.c | 33 ++
> libswscale/loongarch/swscale_loongarch.h | 11 +
> libswscale/swscale_internal.h | 1 +
> libswscale/utils.c | 6 +-
> 5 files changed, 418 insertions(+), 1 deletion(-)
>
> diff --git a/libswscale/loongarch/swscale.S b/libswscale/loongarch/swscale.S
> index aa4c5cbe28..67b1bc834d 100644
> --- a/libswscale/loongarch/swscale.S
> +++ b/libswscale/loongarch/swscale.S
> @@ -1866,3 +1866,371 @@ function ff_hscale_16_to_19_sub_lsx
> ld.d s8, sp, 64
> addi.d sp, sp, 72
> endfunc
> +
> +function lumRangeFromJpeg_lsx
> + li.w t0, 14071
> + li.w t1, 33561947
> + vreplgr2vr.h vr0, t0
> + srli.w t2, a1, 3
> + andi t3, a1, 7
> + beqz t2, 2f
> +1:
> + vld vr1, a0, 0
> + vreplgr2vr.w vr2, t1
> + vreplgr2vr.w vr3, t1
> + vmaddwev.w.h vr2, vr0, vr1
> + vmaddwod.w.h vr3, vr0, vr1
> + vsrai.w vr2, vr2, 14
> + vsrai.w vr3, vr3, 14
> + vpackev.h vr1, vr3, vr2
> + vst vr1, a0, 0
> + addi.d a0, a0, 16
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + mul.w t4, t4, t0
> + add.w t4, t4, t1
> + srai.w t4, t4, 14
> + st.h t4, a0, 0
> + addi.d a0, a0, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function lumRangeFromJpeg_lasx
> + li.w t0, 14071
> + li.w t1, 33561947
> + xvreplgr2vr.h xr0, t0
> + srli.w t2, a1, 4
> + andi t3, a1, 15
> + beqz t2, 2f
> +1:
> + xvld xr1, a0, 0
> + xvreplgr2vr.w xr2, t1
> + xvreplgr2vr.w xr3, t1
> + xvmaddwev.w.h xr2, xr0, xr1
> + xvmaddwod.w.h xr3, xr0, xr1
> + xvsrai.w xr2, xr2, 14
> + xvsrai.w xr3, xr3, 14
> + xvpackev.h xr1, xr3, xr2
> + xvst xr1, a0, 0
> + addi.d a0, a0, 32
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + mul.w t4, t4, t0
> + add.w t4, t4, t1
> + srai.w t4, t4, 14
> + st.h t4, a0, 0
> + addi.d a0, a0, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function lumRangeToJpeg_lsx
> + li.w t0, 19077
> + li.w t1, -39057361
> + li.w t2, 30189
> + vreplgr2vr.h vr0, t0
> + vreplgr2vr.h vr4, t2
> + srli.w t2, a1, 3
> + andi t3, a1, 7
> + beqz t2, 2f
> +1:
> + vld vr1, a0, 0
> + vreplgr2vr.w vr2, t1
> + vreplgr2vr.w vr3, t1
> + vmin.h vr1, vr1, vr4
> + vmaddwev.w.h vr2, vr0, vr1
> + vmaddwod.w.h vr3, vr0, vr1
> + vsrai.w vr2, vr2, 14
> + vsrai.w vr3, vr3, 14
> + vpackev.h vr1, vr3, vr2
> + vst vr1, a0, 0
> + addi.d a0, a0, 16
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + vreplgr2vr.h vr1, t4
> + vmin.h vr1, vr1, vr4
> + vpickve2gr.h t4, vr1, 0
> + mul.w t4, t4, t0
> + add.w t4, t4, t1
> + srai.w t4, t4, 14
> + st.h t4, a0, 0
> + addi.d a0, a0, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function lumRangeToJpeg_lasx
> + li.w t0, 19077
> + li.w t1, -39057361
> + li.w t2, 30189
> + xvreplgr2vr.h xr0, t0
> + xvreplgr2vr.h xr4, t2
> + srli.w t2, a1, 4
> + andi t3, a1, 15
> + beqz t2, 2f
> +1:
> + xvld xr1, a0, 0
> + xvreplgr2vr.w xr2, t1
> + xvreplgr2vr.w xr3, t1
> + xvmin.h xr1, xr1, xr4
> + xvmaddwev.w.h xr2, xr0, xr1
> + xvmaddwod.w.h xr3, xr0, xr1
> + xvsrai.w xr2, xr2, 14
> + xvsrai.w xr3, xr3, 14
> + xvpackev.h xr1, xr3, xr2
> + xvst xr1, a0, 0
> + addi.d a0, a0, 32
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + vreplgr2vr.h vr1, t4
> + vmin.h vr1, vr1, vr4
> + vpickve2gr.h t4, vr1, 0
> + mul.w t4, t4, t0
> + add.w t4, t4, t1
> + srai.w t4, t4, 14
> + st.h t4, a0, 0
> + addi.d a0, a0, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function chrRangeFromJpeg_lsx
> + li.w t0, 1799
> + li.w t1, 4081085
> + vreplgr2vr.h vr0, t0
> + srli.w t2, a2, 3
> + andi t3, a2, 7
> + beqz t2, 2f
> +1:
> + vld vr1, a0, 0
> + vld vr2, a1, 0
> + vreplgr2vr.w vr3, t1
> + vreplgr2vr.w vr4, t1
> + vreplgr2vr.w vr5, t1
> + vreplgr2vr.w vr6, t1
> + vmaddwev.w.h vr3, vr0, vr1
> + vmaddwod.w.h vr4, vr0, vr1
> + vmaddwev.w.h vr5, vr0, vr2
> + vmaddwod.w.h vr6, vr0, vr2
> + vsrai.w vr3, vr3, 11
> + vsrai.w vr4, vr4, 11
> + vsrai.w vr5, vr5, 11
> + vsrai.w vr6, vr6, 11
> + vpackev.h vr1, vr4, vr3
> + vpackev.h vr2, vr6, vr5
> + vst vr1, a0, 0
> + vst vr2, a1, 0
> + addi.d a0, a0, 16
> + addi.d a1, a1, 16
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + ld.h t5, a1, 0
> + mul.w t4, t4, t0
> + mul.w t5, t5, t0
> + add.w t4, t4, t1
> + add.w t5, t5, t1
> + srai.w t4, t4, 11
> + srai.w t5, t5, 11
> + st.h t4, a0, 0
> + st.h t5, a1, 0
> + addi.d a0, a0, 2
> + addi.d a1, a1, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function chrRangeFromJpeg_lasx
> + li.w t0, 1799
> + li.w t1, 4081085
> + xvreplgr2vr.h xr0, t0
> + srli.w t2, a2, 4
> + andi t3, a2, 15
> + beqz t2, 2f
> +1:
> + xvld xr1, a0, 0
> + xvld xr2, a1, 0
> + xvreplgr2vr.w xr3, t1
> + xvreplgr2vr.w xr4, t1
> + xvreplgr2vr.w xr5, t1
> + xvreplgr2vr.w xr6, t1
> + xvmaddwev.w.h xr3, xr0, xr1
> + xvmaddwod.w.h xr4, xr0, xr1
> + xvmaddwev.w.h xr5, xr0, xr2
> + xvmaddwod.w.h xr6, xr0, xr2
> + xvsrai.w xr3, xr3, 11
> + xvsrai.w xr4, xr4, 11
> + xvsrai.w xr5, xr5, 11
> + xvsrai.w xr6, xr6, 11
> + xvpackev.h xr1, xr4, xr3
> + xvpackev.h xr2, xr6, xr5
> + xvst xr1, a0, 0
> + xvst xr2, a1, 0
> + addi.d a0, a0, 32
> + addi.d a1, a1, 32
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + ld.h t5, a1, 0
> + mul.w t4, t4, t0
> + mul.w t5, t5, t0
> + add.w t4, t4, t1
> + add.w t5, t5, t1
> + srai.w t4, t4, 11
> + srai.w t5, t5, 11
> + st.h t4, a0, 0
> + st.h t5, a1, 0
> + addi.d a0, a0, 2
> + addi.d a1, a1, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function chrRangeToJpeg_lsx
> + li.w t0, 4663
> + li.w t1, -9289992
> + li.w t2, 30775
> + vreplgr2vr.h vr0, t0
> + vreplgr2vr.h vr7, t2
> + srli.w t2, a2, 3
> + andi t3, a2, 7
> + beqz t2, 2f
> +1:
> + vld vr1, a0, 0
> + vld vr2, a1, 0
> + vreplgr2vr.w vr3, t1
> + vreplgr2vr.w vr4, t1
> + vreplgr2vr.w vr5, t1
> + vreplgr2vr.w vr6, t1
> + vmin.h vr1, vr1, vr7
> + vmin.h vr2, vr2, vr7
> + vmaddwev.w.h vr3, vr0, vr1
> + vmaddwod.w.h vr4, vr0, vr1
> + vmaddwev.w.h vr5, vr0, vr2
> + vmaddwod.w.h vr6, vr0, vr2
> + vsrai.w vr3, vr3, 12
> + vsrai.w vr4, vr4, 12
> + vsrai.w vr5, vr5, 12
> + vsrai.w vr6, vr6, 12
> + vpackev.h vr1, vr4, vr3
> + vpackev.h vr2, vr6, vr5
> + vst vr1, a0, 0
> + vst vr2, a1, 0
> + addi.d a0, a0, 16
> + addi.d a1, a1, 16
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + ld.h t5, a1, 0
> + vreplgr2vr.h vr1, t4
> + vreplgr2vr.h vr2, t5
> + vmin.h vr1, vr1, vr7
> + vmin.h vr2, vr2, vr7
> + vpickve2gr.h t4, vr1, 0
> + vpickve2gr.h t5, vr2, 0
> + mul.w t4, t4, t0
> + mul.w t5, t5, t0
> + add.w t4, t4, t1
> + add.w t5, t5, t1
> + srai.w t4, t4, 12
> + srai.w t5, t5, 12
> + st.h t4, a0, 0
> + st.h t5, a1, 0
> + addi.d a0, a0, 2
> + addi.d a1, a1, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> +
> +function chrRangeToJpeg_lasx
> + li.w t0, 4663
> + li.w t1, -9289992
> + li.w t2, 30775
> + xvreplgr2vr.h xr0, t0
> + xvreplgr2vr.h xr7, t2
> + srli.w t2, a2, 4
> + andi t3, a2, 15
> + beqz t2, 2f
> +1:
> + xvld xr1, a0, 0
> + xvld xr2, a1, 0
> + xvreplgr2vr.w xr3, t1
> + xvreplgr2vr.w xr4, t1
> + xvreplgr2vr.w xr5, t1
> + xvreplgr2vr.w xr6, t1
> + xvmin.h xr1, xr1, xr7
> + xvmin.h xr2, xr2, xr7
> + xvmaddwev.w.h xr3, xr0, xr1
> + xvmaddwod.w.h xr4, xr0, xr1
> + xvmaddwev.w.h xr5, xr0, xr2
> + xvmaddwod.w.h xr6, xr0, xr2
> + xvsrai.w xr3, xr3, 12
> + xvsrai.w xr4, xr4, 12
> + xvsrai.w xr5, xr5, 12
> + xvsrai.w xr6, xr6, 12
> + xvpackev.h xr1, xr4, xr3
> + xvpackev.h xr2, xr6, xr5
> + xvst xr1, a0, 0
> + xvst xr2, a1, 0
> + addi.d a0, a0, 32
> + addi.d a1, a1, 32
> + addi.d t2, t2, -1
> + bnez t2, 1b
> +2:
> + beqz t3, 4f
> +3:
> + ld.h t4, a0, 0
> + ld.h t5, a1, 0
> + vreplgr2vr.h vr1, t4
> + vreplgr2vr.h vr2, t5
> + vmin.h vr1, vr1, vr7
> + vmin.h vr2, vr2, vr7
> + vpickve2gr.h t4, vr1, 0
> + vpickve2gr.h t5, vr2, 0
> + mul.w t4, t4, t0
> + mul.w t5, t5, t0
> + add.w t4, t4, t1
> + add.w t5, t5, t1
> + srai.w t4, t4, 12
> + srai.w t5, t5, 12
> + st.h t4, a0, 0
> + st.h t5, a1, 0
> + addi.d a0, a0, 2
> + addi.d a1, a1, 2
> + addi.d t3, t3, -1
> + bnez t3, 3b
> +4:
> +endfunc
> diff --git a/libswscale/loongarch/swscale_init_loongarch.c b/libswscale/loongarch/swscale_init_loongarch.c
> index 53e4f970b6..6d2786c55f 100644
> --- a/libswscale/loongarch/swscale_init_loongarch.c
> +++ b/libswscale/loongarch/swscale_init_loongarch.c
> @@ -24,6 +24,38 @@
> #include "libswscale/rgb2rgb.h"
> #include "libavutil/loongarch/cpu.h"
>
> +av_cold void ff_sws_init_range_convert_loongarch(SwsContext *c)
> +{
> + int cpu_flags = av_get_cpu_flags();
> +
> + if (have_lsx(cpu_flags)) {
> + if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
> + if (c->dstBpc <= 14) {
> + if (c->srcRange) {
> + c->lumConvertRange = lumRangeFromJpeg_lsx;
> + c->chrConvertRange = chrRangeFromJpeg_lsx;
> + } else {
> + c->lumConvertRange = lumRangeToJpeg_lsx;
> + c->chrConvertRange = chrRangeToJpeg_lsx;
> + }
> + }
> + }
> + }
> + if (have_lasx(cpu_flags)) {
> + if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
> + if (c->dstBpc <= 14) {
> + if (c->srcRange) {
> + c->lumConvertRange = lumRangeFromJpeg_lasx;
> + c->chrConvertRange = chrRangeFromJpeg_lasx;
> + } else {
> + c->lumConvertRange = lumRangeToJpeg_lasx;
> + c->chrConvertRange = chrRangeToJpeg_lasx;
> + }
> + }
> + }
> + }
> +}
> +
> av_cold void ff_sws_init_swscale_loongarch(SwsContext *c)
> {
> int cpu_flags = av_get_cpu_flags();
> @@ -77,6 +109,7 @@ av_cold void ff_sws_init_swscale_loongarch(SwsContext *c)
> c->yuv2planeX = ff_yuv2planeX_8_lasx;
> }
> #endif // #if HAVE_LASX
> + ff_sws_init_range_convert_loongarch(c);
> }
>
> av_cold void rgb2rgb_init_loongarch(void)
> diff --git a/libswscale/loongarch/swscale_loongarch.h b/libswscale/loongarch/swscale_loongarch.h
> index 0514abae21..c96b085982 100644
> --- a/libswscale/loongarch/swscale_loongarch.h
> +++ b/libswscale/loongarch/swscale_loongarch.h
> @@ -50,6 +50,11 @@ void ff_hscale_16_to_19_sub_lsx(SwsContext *c, int16_t *_dst, int dstW,
> const uint8_t *_src, const int16_t *filter,
> const int32_t *filterPos, int filterSize, int sh);
>
> +void lumRangeFromJpeg_lsx(int16_t *dst, int width);
> +void chrRangeFromJpeg_lsx(int16_t *dstU, int16_t *dstV, int width);
> +void lumRangeToJpeg_lsx(int16_t *dst, int width);
> +void chrRangeToJpeg_lsx(int16_t *dstU, int16_t *dstV, int width);
> +
> void planar_rgb_to_uv_lsx(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *src[4],
> int width, int32_t *rgb2yuv, void *opq);
>
> @@ -97,6 +102,11 @@ void ff_hscale_16_to_15_lasx(SwsContext *c, int16_t *dst, int dstW,
> const uint8_t *_src, const int16_t *filter,
> const int32_t *filterPos, int filterSize);
>
> +void lumRangeFromJpeg_lasx(int16_t *dst, int width);
> +void chrRangeFromJpeg_lasx(int16_t *dstU, int16_t *dstV, int width);
> +void lumRangeToJpeg_lasx(int16_t *dst, int width);
> +void chrRangeToJpeg_lasx(int16_t *dstU, int16_t *dstV, int width);
> +
> void planar_rgb_to_uv_lasx(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *src[4],
> int width, int32_t *rgb2yuv, void *opq);
>
> @@ -130,6 +140,7 @@ void ff_yuv2planeX_8_lasx(const int16_t *filter, int filterSize,
> const uint8_t *dither, int offset);
>
> av_cold void ff_sws_init_output_lasx(SwsContext *c);
> +
> #endif // #if HAVE_LASX
>
> #endif /* SWSCALE_LOONGARCH_SWSCALE_LOONGARCH_H */
> diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
> index abeebbb002..0db581acf8 100644
> --- a/libswscale/swscale_internal.h
> +++ b/libswscale/swscale_internal.h
> @@ -695,6 +695,7 @@ void ff_yuv2rgb_init_tables_ppc(SwsContext *c, const int inv_table[4],
> 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);
>
> 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 ab8a68e241..47db65ef0e 100644
> --- a/libswscale/utils.c
> +++ b/libswscale/utils.c
> @@ -1049,8 +1049,12 @@ int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
> c->srcRange = srcRange;
> c->dstRange = dstRange;
>
> - if (need_reinit)
> + if (need_reinit) {
> ff_sws_init_range_convert(c);
> +#if ARCH_LOONGARCH64
> + ff_sws_init_range_convert_loongarch(c);
> +#endif
> + }
>
> c->dstFormatBpp = av_get_bits_per_pixel(desc_dst);
> c->srcFormatBpp = av_get_bits_per_pixel(desc_src);
> --
> 2.20.1
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
本邮件及其附件含有龙芯中科的商业秘密信息，仅限于发送给上面地址中列出的个人或群组。禁止任何其他人以任何形式使用（包括但不限于全部或部分地泄露、复制或散发）本邮件及其附件中的信息。如果您错收本邮件，请您立即电话或邮件通知发件人并删除本邮件。
This email and its attachments contain confidential information from Loongson Technology , which is intended only for the person or entity whose address is listed above. Any use of the information contained herein in any way (including, but not limited to, total or partial disclosure, reproduction or dissemination) by persons other than the intended recipient(s) is prohibited. If you receive this email in error, please notify the sender by phone or email immediately and delete it.
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
next prev parent reply other threads:[~2024-03-16 6:22 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-03-16 3:03 [FFmpeg-devel] Add optimization in swscale for LA Shiyou Yin
2024-03-16 3:03 ` [FFmpeg-devel] [PATCH 1/3] swscale: [LA] Optimize range convert for yuvj420p Shiyou Yin
2024-03-16 6:22 ` 陈昊 [this message]
2024-03-16 3:03 ` [FFmpeg-devel] [PATCH 2/3] swscale: [LA] Optimize yuv2plane1_8_c Shiyou Yin
2024-03-16 3:03 ` [FFmpeg-devel] [PATCH 3/3] swscale: [LA] Optimize swscale funcs in input.c Shiyou Yin
2024-03-26 3:11 ` [FFmpeg-devel] Add optimization in swscale for LA Shiyou Yin
2024-03-26 19:31 ` Michael Niedermayer
2024-04-09 12:19 ` Shiyou Yin
2024-04-11 21:54 ` Michael Niedermayer
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=13d6431d.17d3.18e45ecc173.Coremail.chenhao@loongson.cn \
--to=chenhao@loongson.cn \
--cc=ffmpeg-devel@ffmpeg.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git