* [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation [not found] <20240603071732.52523-1-quinkblack@foxmail.com> @ 2024-06-03 7:17 ` Zhao Zhili 2024-06-03 8:07 ` Martin Storsjö 0 siblings, 1 reply; 6+ messages in thread From: Zhao Zhili @ 2024-06-03 7:17 UTC (permalink / raw) To: ffmpeg-devel; +Cc: Zhao Zhili From: Zhao Zhili <zhilizhao@tencent.com> Test on Apple M1: rgb24_to_uv_1080_c: 7.0 rgb24_to_uv_1080_neon: 5.5 rgb24_to_uv_1280_c: 8.2 rgb24_to_uv_1280_neon: 6.0 rgb24_to_uv_4096_c: 26.2 rgb24_to_uv_4096_neon: 20.7 rgb24_to_uv_half_540_c: 6.5 rgb24_to_uv_half_540_neon: 3.0 rgb24_to_uv_half_640_c: 8.0 rgb24_to_uv_half_640_neon: 3.2 rgb24_to_uv_half_2048_c: 24.0 rgb24_to_uv_half_2048_neon: 11.2 rgb24_to_y_1080_c: 4.5 rgb24_to_y_1080_neon: 3.2 rgb24_to_y_1280_c: 5.5 rgb24_to_y_1280_neon: 3.7 rgb24_to_y_4096_c: 17.7 rgb24_to_y_4096_neon: 12.2 Signed-off-by: Zhao Zhili <zhilizhao@tencent.com> --- libswscale/aarch64/Makefile | 1 + libswscale/aarch64/input.S | 229 +++++++++++++++++++++++++++++++++++ libswscale/aarch64/swscale.c | 21 ++++ 3 files changed, 251 insertions(+) create mode 100644 libswscale/aarch64/input.S diff --git a/libswscale/aarch64/Makefile b/libswscale/aarch64/Makefile index da1d909561..adfd90a1b6 100644 --- a/libswscale/aarch64/Makefile +++ b/libswscale/aarch64/Makefile @@ -3,6 +3,7 @@ OBJS += aarch64/rgb2rgb.o \ aarch64/swscale_unscaled.o \ NEON-OBJS += aarch64/hscale.o \ + aarch64/input.o \ aarch64/output.o \ aarch64/rgb2rgb_neon.o \ aarch64/yuv2rgb_neon.o \ diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S new file mode 100644 index 0000000000..0a46475723 --- /dev/null +++ b/libswscale/aarch64/input.S @@ -0,0 +1,229 @@ +/* + * Copyright (c) 2024 Zhao Zhili <quinkblack@foxmail.com> + * + * 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/aarch64/asm.S" + +.macro rgb24_to_yuv_load_rgb, src + ld3.16b { v16, v17, v18 }, [\src] + ushll.8h v19, v16, #0 // v19: r + ushll.8h v20, v17, #0 // v20: g + ushll.8h v21, v18, #0 // v21: b + ushll2.8h v22, v16, #0 // v22: r + ushll2.8h v23, v17, #0 // v23: g + ushll2.8h v24, v18, #0 // v24: b +.endm + +.macro rgb24_to_yuv_product, r, g, b, dst1, dst2, dst, coef0, coef1, coef2, right_shift + mov.4s \dst1, v6 // dst1 = const_offset + mov.4s \dst2, v6 // dst2 = const_offset + smlal.4s \dst1, \coef0, \r // dst1 += rx * r + smlal2.4s \dst2, \coef0, \r // dst2 += rx * r + smlal.4s \dst1, \coef1, \g // dst1 += gx * g + smlal2.4s \dst2, \coef1, \g // dst2 += gx * g + smlal.4s \dst1, \coef2, \b // dst1 += bx * b + smlal2.4s \dst2, \coef2, \b // dst2 += bx * b + sqshrn.4h \dst, \dst1, \right_shift // dst_lower_half = dst1 >> right_shift + sqshrn2.8h \dst, \dst2, \right_shift // dst_higher_half = dst2 >> right_shift +.endm + +function ff_rgb24ToY_neon, export=1 + cmp w4, #0 // check width > 0 + b.le 4f + + ldp w10, w11, [x5], #8 // w10: ry, w11: gy + dup v0.8H, w10 + dup v1.8H, w11 + ldr w12, [x5] // w12: by + dup v2.8H, w12 + + mov w9, #256 // w9 = 1 << (RGB2YUV_SHIFT - 7) + movk w9, #8, lsl #16 // w9 += 32 << (RGB2YUV_SHIFT - 1) + dup.4s v6, w9 // w9: const_offset + + mov x2, #0 // w2: i + and w3, w4, #0xFFFFFFF0 // w3 = width / 16 * 16 + cbz w3, 3f +1: + rgb24_to_yuv_load_rgb x1 + rgb24_to_yuv_product v19, v20, v21, v25, v26, v16, v0, v1, v2, #9 + rgb24_to_yuv_product v22, v23, v24, v27, v28, v17, v0, v1, v2, #9 + stp q16, q17, [x0], #32 // store to dst + + add w2, w2, #16 // i += 16 + add x1, x1, #48 // src += 48 + cmp w2, w3 // i < (width / 16 * 16) + b.lt 1b + b 3f +2: + ldrb w13, [x1] // w13: r + ldrb w14, [x1, #1] // w14: g + ldrb w15, [x1, #2] // w15: b + + smaddl x13, w13, w10, x9 // x13 = ry * r + const_offset + smaddl x13, w14, w11, x13 // x13 += gy * g + smaddl x13, w15, w12, x13 // x13 += by * b + asr w13, w13, #9 // x13 >>= 9 + strh w13, [x0], #2 // store to dst + + add w2, w2, #1 // i++ + add x1, x1, #3 // src += 3 +3: + cmp w2, w4 // i < width + b.lt 2b +4: + ret +endfunc + +.macro rgb24_load_uv_coeff half + add x6, x6, #12 + + ldp w10, w11, [x6], #8 // w10: ru, w11: gu + dup v0.8H, w10 + dup v1.8H, w11 + + ldp w12, w13, [x6], #8 // w12: bu, w13: rv + dup v2.8H, w12 + dup v3.8H, w13 + + ldp w14, w15, [x6], #8 // w14: gv, w15: bv + dup v4.8H, w14 + dup v5.8H, w15 + + .if \half + mov w9, #512 + movk w9, #128, lsl #16 // w9: const_offset + .else + mov w9, #256 + movk w9, #64, lsl #16 // w9: const_offset + .endif + dup v6.4S, w9 +.endm + +function ff_rgb24ToUV_half_neon, export=1 + cmp w5, #0 // check width > 0 + b.le 4f + + rgb24_load_uv_coeff half=1 + + mov x9, #0 // x9: i + and w7, w5, #0xFFFFFFF8 // w7 = width / 8 * 8 + cbz w7, 3f +1: + ld3.16b { v16, v17, v18 }, [x3] + uaddlp v19.8H, v16.16B // v19: r + uaddlp v20.8H, v17.16B // v20: g + uaddlp v21.8H, v18.16B // v21: b + + rgb24_to_yuv_product v19, v20, v21, v22, v23, v16, v0, v1, v2, #10 + str q16, [x0], #16 // store dst_u + rgb24_to_yuv_product v19, v20, v21, v24, v25, v17, v3, v4, v5, #10 + str q17, [x1], #16 // store dst_v + + add w9, w9, #8 // i += 8 + add x3, x3, #48 // src += 48 + cmp w9, w7 // i < (width * 8 / 8) + b.lt 1b + b 3f +2: + ldrb w2, [x3] // w2: r1 + ldrb w4, [x3, #3] // w4: r2 + add w2, w2, w4 // w2 = r1 + r2 + + ldrb w4, [x3, #1] // w4: g1 + ldrb w7, [x3, #4] // w7: g2 + add w4, w4, w7 // w4 = g1 + g2 + + ldrb w7, [x3, #2] // w7: b1 + ldrb w8, [x3, #5] // w8: b2 + add w7, w7, w8 // w7 = b1 + b2 + + umov w8, v6.S[0] // dst_u = const_offset + smaddl x8, w2, w10, x8 // dst_u += ru * r + smaddl x8, w4, w11, x8 // dst_u += gu * g + smaddl x8, w7, w12, x8 // dst_u += bu * b + asr x8, x8, #10 // dst_u >>= 10 + strh w8, [x0], #2 // store dst_u + + umov w8, v6.S[0] // dst_v = const_offset + smaddl x8, w2, w13, x8 // dst_v += rv * r + smaddl x8, w4, w14, x8 // dst_v += gv * g + smaddl x8, w7, w15, x8 // dst_v += bv * b + asr x8, x8, #10 // dst_v >>= 10 + strh w8, [x1], #2 // store dst_v + + add w9, w9, #1 // i++ + add x3, x3, #6 // src += 6 +3: + cmp w9, w5 + b.lt 2b +4: + ret +endfunc + +function ff_rgb24ToUV_neon, export=1 + cmp w5, #0 // check width > 0 + b.le 4f + + rgb24_load_uv_coeff half=0 + + mov x2, #0 // w2: i + and w4, w5, #0xFFFFFFF0 // w4: width / 16 * 16 + cbz w4, 3f +1: + rgb24_to_yuv_load_rgb x3 + rgb24_to_yuv_product v19, v20, v21, v25, v26, v16, v0, v1, v2, #9 + rgb24_to_yuv_product v22, v23, v24, v27, v28, v17, v0, v1, v2, #9 + stp q16, q17, [x0], #32 // store to dst_u + rgb24_to_yuv_product v19, v20, v21, v25, v26, v16, v3, v4, v5, #9 + rgb24_to_yuv_product v22, v23, v24, v27, v28, v17, v3, v4, v5, #9 + stp q16, q17, [x1], #32 // store to dst_v + + add w2, w2, #16 // i += 16 + add x3, x3, #48 // src += 48 + cmp w2, w4 // i < (width / 16 * 16) + b.lt 1b + b 3f +2: + ldrb w16, [x3] // w16: r + ldrb w17, [x3, #1] // w17: g + ldrb w4, [x3, #2] // w4: b + + umov w7, v6.S[0] // w7 = const_offset + + smaddl x8, w16, w10, x7 // x8 = ru * r + const_offset + smaddl x8, w17, w11, x8 // x8 += gu * g + smaddl x8, w4, w12, x8 // x8 += bu * b + asr w8, w8, #9 // x8 >>= 9 + strh w8, [x0], #2 // store to dst_u + + smaddl x8, w16, w13, x7 // x8 = rv * r + const_offset + smaddl x8, w17, w14, x8 // x8 += gv * g + smaddl x8, w4, w15, x8 // x8 += bv * b + asr w8, w8, #9 // x8 >>= 9 + strh w8, [x1], #2 // store to dst_v + + add w2, w2, #1 // i++ + add x3, x3, #3 // src += 3 +3: + cmp w2, w5 // i < width + b.lt 2b +4: + ret +endfunc diff --git a/libswscale/aarch64/swscale.c b/libswscale/aarch64/swscale.c index bbd9719a44..5dd700a16b 100644 --- a/libswscale/aarch64/swscale.c +++ b/libswscale/aarch64/swscale.c @@ -201,6 +201,20 @@ void ff_yuv2plane1_8_neon( default: break; \ } +void ff_rgb24ToY_neon(uint8_t *_dst, const uint8_t *src, const uint8_t *unused1, + const uint8_t *unused2, int width, + uint32_t *rgb2yuv, void *opq); + +void ff_rgb24ToUV_neon(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *unused0, + const uint8_t *src1, + const uint8_t *src2, int width, uint32_t *rgb2yuv, + void *opq); + +void ff_rgb24ToUV_half_neon(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *unused0, + const uint8_t *src1, + const uint8_t *src2, int width, uint32_t *rgb2yuv, + void *opq); + av_cold void ff_sws_init_swscale_aarch64(SwsContext *c) { int cpu_flags = av_get_cpu_flags(); @@ -212,5 +226,12 @@ av_cold void ff_sws_init_swscale_aarch64(SwsContext *c) if (c->dstBpc == 8) { c->yuv2planeX = ff_yuv2planeX_8_neon; } + if (c->srcFormat == AV_PIX_FMT_RGB24) { + c->lumToYV12 = ff_rgb24ToY_neon; + if (c->chrSrcHSubSample) + c->chrToYV12 = ff_rgb24ToUV_half_neon; + else + c->chrToYV12 = ff_rgb24ToUV_neon; + } } } -- 2.42.0 _______________________________________________ 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". ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation 2024-06-03 7:17 ` [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation Zhao Zhili @ 2024-06-03 8:07 ` Martin Storsjö 2024-06-03 13:11 ` Zhao Zhili 0 siblings, 1 reply; 6+ messages in thread From: Martin Storsjö @ 2024-06-03 8:07 UTC (permalink / raw) To: FFmpeg development discussions and patches; +Cc: Zhao Zhili On Mon, 3 Jun 2024, Zhao Zhili wrote: > diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S > new file mode 100644 > index 0000000000..0a46475723 > --- /dev/null > +++ b/libswscale/aarch64/input.S > @@ -0,0 +1,229 @@ > +/* > + * Copyright (c) 2024 Zhao Zhili <quinkblack@foxmail.com> > + * > + * 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/aarch64/asm.S" > + > +.macro rgb24_to_yuv_load_rgb, src > + ld3.16b { v16, v17, v18 }, [\src] > + ushll.8h v19, v16, #0 // v19: r > + ushll.8h v20, v17, #0 // v20: g > + ushll.8h v21, v18, #0 // v21: b > + ushll2.8h v22, v16, #0 // v22: r > + ushll2.8h v23, v17, #0 // v23: g > + ushll2.8h v24, v18, #0 // v24: b Don't use this nonstandard, Apple specific aarch64 syntax. This was used by Apple tools at the start, when the proper standardized aarch64 syntax wasn't quite settled yet, and it is still accepted. (And apparently this is still the preferred form to disassemble things in, for apple platforms.) With this syntax, the assembly is rejected by GNU binutils and MSVC. > +function ff_rgb24ToY_neon, export=1 > + cmp w4, #0 // check width > 0 > + b.le 4f > + > + ldp w10, w11, [x5], #8 // w10: ry, w11: gy > + dup v0.8H, w10 > + dup v1.8H, w11 > + ldr w12, [x5] // w12: by > + dup v2.8H, w12 Don't use uppercase .8H for field layout configurations, we prefer to stick to all lowercase here - see 184103b3105f02f1189fa0047af4269e027dfbd6. The same goes for a number of places in this patch. > + add w9, w9, #1 // i++ > + add x3, x3, #6 // src += 6 > +3: > + cmp w9, w5 > + b.lt 2b > +4: Incorrect indentation for the cmp/b.lt instructions here. I have set up a bunch of github actions for testing aarch64 assembly - see https://github.com/mstorsjo/ffmpeg/commits/gha-aarch64. If you have a github account, grab a copy of this branch into your repo, add your own commits on top, and push to your fork (and if necessary, activate running the actions), then you should get a wide testing of your patches. See https://github.com/mstorsjo/FFmpeg/actions/runs/9346228714 for one example run of these actions with your patches. // Martin _______________________________________________ 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". ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation 2024-06-03 8:07 ` Martin Storsjö @ 2024-06-03 13:11 ` Zhao Zhili 2024-06-03 14:17 ` Rémi Denis-Courmont 0 siblings, 1 reply; 6+ messages in thread From: Zhao Zhili @ 2024-06-03 13:11 UTC (permalink / raw) To: FFmpeg development discussions and patches > On Jun 3, 2024, at 16:07, Martin Storsjö <martin@martin.st> wrote: > > On Mon, 3 Jun 2024, Zhao Zhili wrote: > >> diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S >> new file mode 100644 >> index 0000000000..0a46475723 >> --- /dev/null >> +++ b/libswscale/aarch64/input.S >> @@ -0,0 +1,229 @@ >> +/* >> + * Copyright (c) 2024 Zhao Zhili <quinkblack@foxmail.com> >> + * >> + * 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/aarch64/asm.S" >> + >> +.macro rgb24_to_yuv_load_rgb, src >> + ld3.16b { v16, v17, v18 }, [\src] >> + ushll.8h v19, v16, #0 // v19: r >> + ushll.8h v20, v17, #0 // v20: g >> + ushll.8h v21, v18, #0 // v21: b >> + ushll2.8h v22, v16, #0 // v22: r >> + ushll2.8h v23, v17, #0 // v23: g >> + ushll2.8h v24, v18, #0 // v24: b > > Don't use this nonstandard, Apple specific aarch64 syntax. This was used by Apple tools at the start, when the proper standardized aarch64 syntax wasn't quite settled yet, and it is still accepted. (And apparently this is still the preferred form to disassemble things in, for apple platforms.) > > With this syntax, the assembly is rejected by GNU binutils and MSVC. > >> +function ff_rgb24ToY_neon, export=1 >> + cmp w4, #0 // check width > 0 >> + b.le 4f >> + >> + ldp w10, w11, [x5], #8 // w10: ry, w11: gy >> + dup v0.8H, w10 >> + dup v1.8H, w11 >> + ldr w12, [x5] // w12: by >> + dup v2.8H, w12 > > Don't use uppercase .8H for field layout configurations, we prefer to stick to all lowercase here - see 184103b3105f02f1189fa0047af4269e027dfbd6. The same goes for a number of places in this patch. > >> + add w9, w9, #1 // i++ >> + add x3, x3, #6 // src += 6 >> +3: >> + cmp w9, w5 >> + b.lt 2b >> +4: > > Incorrect indentation for the cmp/b.lt instructions here. > > > I have set up a bunch of github actions for testing aarch64 assembly - see https://github.com/mstorsjo/ffmpeg/commits/gha-aarch64. If you have a github account, grab a copy of this branch into your repo, add your own commits on top, and push to your fork (and if necessary, activate running the actions), then you should get a wide testing of your patches. > > See https://github.com/mstorsjo/FFmpeg/actions/runs/9346228714 for one example run of these actions with your patches. Wow, it’s very helpful. This is the action result of the updated patch: https://github.com/quink-black/FFmpeg/actions/runs/9350348848 https://ffmpeg.org/pipermail/ffmpeg-devel/2024-June/328786.html The test still failed on x86, but success on all arm64 platform and longarch. I have tried to call rgb24ToY_c and ff_rgb24ToY_avx directly and compare the results, they don't match. I’m confused. > > // Martin > > _______________________________________________ > 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". _______________________________________________ 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". ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation 2024-06-03 13:11 ` Zhao Zhili @ 2024-06-03 14:17 ` Rémi Denis-Courmont 2024-06-03 16:14 ` Zhao Zhili 0 siblings, 1 reply; 6+ messages in thread From: Rémi Denis-Courmont @ 2024-06-03 14:17 UTC (permalink / raw) To: FFmpeg development discussions and patches Le maanantaina 3. kesäkuuta 2024, 16.11.15 EEST Zhao Zhili a écrit : > > See https://github.com/mstorsjo/FFmpeg/actions/runs/9346228714 for one > > example run of these actions with your patches. > Wow, it’s very helpful. This is the action result of the updated patch: > > https://github.com/quink-black/FFmpeg/actions/runs/9350348848 > > https://ffmpeg.org/pipermail/ffmpeg-devel/2024-June/328786.html > > The test still failed on x86, but success on all arm64 platform and > longarch. I have tried to call rgb24ToY_c and ff_rgb24ToY_avx > directly and compare the results, they don't match. I’m confused. As Martin write, some x86 code is imprecise, or even wrong. I would just disable the test on x86 until some x86 arcane magician fixes it. -- 雷米‧德尼-库尔蒙 http://www.remlab.net/ _______________________________________________ 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". ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation 2024-06-03 14:17 ` Rémi Denis-Courmont @ 2024-06-03 16:14 ` Zhao Zhili 2024-06-03 17:22 ` Ronald S. Bultje 0 siblings, 1 reply; 6+ messages in thread From: Zhao Zhili @ 2024-06-03 16:14 UTC (permalink / raw) To: FFmpeg development discussions and patches > On Jun 3, 2024, at 22:17, Rémi Denis-Courmont <remi@remlab.net> wrote: > > Le maanantaina 3. kesäkuuta 2024, 16.11.15 EEST Zhao Zhili a écrit : >>> See https://github.com/mstorsjo/FFmpeg/actions/runs/9346228714 for one >>> example run of these actions with your patches. >> Wow, it’s very helpful. This is the action result of the updated patch: >> >> https://github.com/quink-black/FFmpeg/actions/runs/9350348848 >> >> https://ffmpeg.org/pipermail/ffmpeg-devel/2024-June/328786.html >> >> The test still failed on x86, but success on all arm64 platform and >> longarch. I have tried to call rgb24ToY_c and ff_rgb24ToY_avx >> directly and compare the results, they don't match. I’m confused. > > As Martin write, some x86 code is imprecise, or even wrong. On x86: With the following command: ./ffmpeg -bitexact -cpuflags 0 -f lavfi -i testsrc -frames 1 -pix_fmt yuv420p -f framemd5 - 0, 0, 0, 1, 115200, d6b3abfc5280311c2758d5e4028c07b5 Without “-cpuflags 0” ./ffmpeg -bitexact -f lavfi -i testsrc -frames 1 -pix_fmt yuv420p -f framemd5 - 0, 0, 0, 1, 115200, 1d302ce90bd5b6eec681730cc0868be4 It's indeed non bitexact. On aarch64 with the neon implementation I can get the same result as "-cpuflags 0": ./ffmpeg -bitexact -f lavfi -i testsrc -frames 1 -pix_fmt yuv420p -f framemd5 - 0, 0, 0, 1, 115200, d6b3abfc5280311c2758d5e4028c07b5 Now I can disable the test for x86 and continue the work on aarch64. > > I would just disable the test on x86 until some x86 arcane magician fixes it. > > -- > 雷米‧德尼-库尔蒙 > http://www.remlab.net/ > > > > _______________________________________________ > 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". _______________________________________________ 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". ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation 2024-06-03 16:14 ` Zhao Zhili @ 2024-06-03 17:22 ` Ronald S. Bultje 0 siblings, 0 replies; 6+ messages in thread From: Ronald S. Bultje @ 2024-06-03 17:22 UTC (permalink / raw) To: FFmpeg development discussions and patches Hi, On Mon, Jun 3, 2024 at 12:14 PM Zhao Zhili <quinkblack@foxmail.com> wrote: > > > > On Jun 3, 2024, at 22:17, Rémi Denis-Courmont <remi@remlab.net> wrote: > > > > Le maanantaina 3. kesäkuuta 2024, 16.11.15 EEST Zhao Zhili a écrit : > >>> See https://github.com/mstorsjo/FFmpeg/actions/runs/9346228714 for one > >>> example run of these actions with your patches. > >> Wow, it’s very helpful. This is the action result of the updated patch: > >> > >> https://github.com/quink-black/FFmpeg/actions/runs/9350348848 > >> > >> https://ffmpeg.org/pipermail/ffmpeg-devel/2024-June/328786.html > >> > >> The test still failed on x86, but success on all arm64 platform and > >> longarch. I have tried to call rgb24ToY_c and ff_rgb24ToY_avx > >> directly and compare the results, they don't match. I’m confused. > > > > As Martin write, some x86 code is imprecise, or even wrong. > > On x86: > > With the following command: > ./ffmpeg -bitexact -cpuflags 0 -f lavfi -i testsrc -frames 1 -pix_fmt > yuv420p -f framemd5 - > 0, 0, 0, 1, 115200, > d6b3abfc5280311c2758d5e4028c07b5 > > Without “-cpuflags 0” > ./ffmpeg -bitexact -f lavfi -i testsrc -frames 1 -pix_fmt yuv420p -f > framemd5 - > 0, 0, 0, 1, 115200, > 1d302ce90bd5b6eec681730cc0868be4 > > It's indeed non bitexact. > > On aarch64 with the neon implementation I can get the same result as > "-cpuflags 0": > ./ffmpeg -bitexact -f lavfi -i testsrc -frames 1 -pix_fmt yuv420p -f > framemd5 - > 0, 0, 0, 1, 115200, > d6b3abfc5280311c2758d5e4028c07b5 > > Now I can disable the test for x86 and continue the work on aarch64. > Uhm, that's a bit hacky. I think things like -sws_flags +bitexact need to be after the -i argument and then it might work? Ronald _______________________________________________ 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". ^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2024-06-03 17:23 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
[not found] <20240603071732.52523-1-quinkblack@foxmail.com>
2024-06-03 7:17 ` [FFmpeg-devel] [PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation Zhao Zhili
2024-06-03 8:07 ` Martin Storsjö
2024-06-03 13:11 ` Zhao Zhili
2024-06-03 14:17 ` Rémi Denis-Courmont
2024-06-03 16:14 ` Zhao Zhili
2024-06-03 17:22 ` Ronald S. Bultje
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