August 2022 - ffmpeg-devel - lists.ffmpeg.org

Re: [FFmpeg-devel] [PATCH v2] avcodec/mips: Fix MMI macro replaces in HEVC Decoder
by 戚铁铮 07 Sep '22

07 Sep '22

At 2022/8/18 PM 7:01, "Qi Tiezheng" <qitiezheng(a)360.cn> wrote: > The latest commit of Loongson MMI macro replaces were incorrect. > It makes a mass of green tints on HEVC videos when playing. I've > compared it with the older MMI implementation, and found out that > several lines have been replaced by wrong macros. > > Signed-off-by: Qi Tiezheng <qitiezheng(a)360.cn> > --- > libavcodec/mips/hevcdsp_mmi.c | 16 ++++++++-------- > 1 file changed, 8 insertions(+), 8 deletions(-) > > diff --git a/libavcodec/mips/hevcdsp_mmi.c b/libavcodec/mips/hevcdsp_mmi.c > index 0ea88a7c08..1da56d3d87 100644 > --- a/libavcodec/mips/hevcdsp_mmi.c > +++ b/libavcodec/mips/hevcdsp_mmi.c > @@ -80,7 +80,7 @@ void ff_hevc_put_hevc_qpel_h##w##_8_mmi(int16_t *dst, const uint8_t *_src, \ > "paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \ > "paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \ > "paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \ > - MMI_ULDC1(%[ftmp3], %[dst], 0x00) \ > + MMI_USDC1(%[ftmp3], %[dst], 0x00) \ > \ > "daddi %[x], %[x], -0x01 \n\t" \ > PTR_ADDIU "%[src], %[src], 0x04 \n\t" \ > @@ -178,7 +178,7 @@ void ff_hevc_put_hevc_qpel_hv##w##_8_mmi(int16_t *dst, const uint8_t *_src,\ > "paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \ > "paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \ > "paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \ > - MMI_ULDC1(%[ftmp3], %[tmp], 0x00) \ > + MMI_USDC1(%[ftmp3], %[tmp], 0x00) \ > \ > "daddi %[x], %[x], -0x01 \n\t" \ > PTR_ADDIU "%[src], %[src], 0x04 \n\t" \ > @@ -690,10 +690,10 @@ void ff_hevc_put_hevc_epel_bi_hv##w##_8_mmi(uint8_t *_dst, \ > \ > "1: \n\t" \ > "2: \n\t" \ > - MMI_ULDC1(%[ftmp3], %[src], 0x00) \ > - MMI_ULDC1(%[ftmp4], %[src], 0x01) \ > - MMI_ULDC1(%[ftmp5], %[src], 0x02) \ > - MMI_ULDC1(%[ftmp6], %[src], 0x03) \ > + MMI_ULWC1(%[ftmp2], %[src], 0x00) \ > + MMI_ULWC1(%[ftmp3], %[src], 0x01) \ > + MMI_ULWC1(%[ftmp4], %[src], 0x02) \ > + MMI_ULWC1(%[ftmp5], %[src], 0x03) \ > "punpcklbh %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \ > "pmullh %[ftmp2], %[ftmp2], %[ftmp1] \n\t" \ > "punpcklbh %[ftmp3], %[ftmp3], %[ftmp0] \n\t" \ > @@ -707,7 +707,7 @@ void ff_hevc_put_hevc_epel_bi_hv##w##_8_mmi(uint8_t *_dst, \ > "paddh %[ftmp2], %[ftmp2], %[ftmp3] \n\t" \ > "paddh %[ftmp4], %[ftmp4], %[ftmp5] \n\t" \ > "paddh %[ftmp2], %[ftmp2], %[ftmp4] \n\t" \ > - MMI_ULDC1(%[ftmp2], %[tmp], 0x00) \ > + MMI_USDC1(%[ftmp2], %[tmp], 0x00) \ > \ > "daddi %[x], %[x], -0x01 \n\t" \ > PTR_ADDIU "%[src], %[src], 0x04 \n\t" \ > @@ -773,7 +773,7 @@ void ff_hevc_put_hevc_epel_bi_hv##w##_8_mmi(uint8_t *_dst, \ > "paddw %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \ > "psraw %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \ > "packsswh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \ > - MMI_ULDC1(%[ftmp4], %[tmp], 0x02) \ > + MMI_ULDC1(%[ftmp4], %[src2], 0x00) \ > "li %[rtmp0], 0x10 \n\t" \ > "dmtc1 %[rtmp0], %[ftmp8] \n\t" \ > "punpcklhw %[ftmp5], %[ftmp2], %[ftmp3] \n\t" \ > -- > 2.25.1 Sorry, I must use e-mail client because our e-mail server is Exchange not SMTP. The patch system seems cannot process UTF-8 Chinese characters correctly. I try sending it as attachment again.

3 3

[PATCH v2 1/4] fftools: Add support for dictionary options
by Thilo Borgmann 07 Sep '22

07 Sep '22

Hi, this is an updated and cleaned-up version of Jan's patchset discussed in [1], therefore v2... Comments from the previous thread are in and left-overs from the several options version removed. I'd especially appreciate any comments on 2/4, ffmpeg_opt.c:112ff which is pretty ugly as-is. Should fix #8329 and #6370. Thanks, Thilo

5 20

[PATCH v3] libavfilter/x86/vf_convolution: add sobel filter optimization and unit test with intel AVX512 VNNI
by bin.wang＠intel.com 07 Sep '22

07 Sep '22

From: bwang30 <bin.wang(a)intel.com> This commit enabled assembly code with intel AVX512 VNNI and added unit test for sobel filter sobel_c: 4537 sobel_avx512icl 2470 Signed-off-by: bwang30 <bin.wang(a)intel.com> --- libavfilter/convolution.h | 2 + libavfilter/vf_convolution.c | 23 ++++ libavfilter/x86/vf_convolution.asm | 162 ++++++++++++++++++++++++++ libavfilter/x86/vf_convolution_init.c | 18 +++ tests/checkasm/Makefile | 1 + tests/checkasm/checkasm.c | 3 + tests/checkasm/checkasm.h | 1 + tests/checkasm/vf_convolution.c | 103 ++++++++++++++++ 8 files changed, 313 insertions(+) create mode 100644 tests/checkasm/vf_convolution.c diff --git a/libavfilter/convolution.h b/libavfilter/convolution.h index 88aabe9a20..4520ad13e0 100644 --- a/libavfilter/convolution.h +++ b/libavfilter/convolution.h @@ -61,4 +61,6 @@ typedef struct ConvolutionContext { } ConvolutionContext; void ff_convolution_init_x86(ConvolutionContext *s); +void ff_sobel_init_x86(ConvolutionContext *s); +void ff_convolution_init(ConvolutionContext *s, const char *filter_name); #endif diff --git a/libavfilter/vf_convolution.c b/libavfilter/vf_convolution.c index 9a9c099e6d..0eeaaa9fc3 100644 --- a/libavfilter/vf_convolution.c +++ b/libavfilter/vf_convolution.c @@ -874,6 +874,9 @@ static int param_init(AVFilterContext *ctx) if (s->depth > 8) for (p = 0; p < s->nb_planes; p++) s->filter[p] = filter16_sobel; +#if CONFIG_CONVOLUTION_FILTER && ARCH_X86_64 + ff_sobel_init_x86(s); +#endif } else if (!strcmp(ctx->filter->name, "kirsch")) { if (s->depth > 8) for (p = 0; p < s->nb_planes; p++) @@ -887,6 +890,26 @@ static int param_init(AVFilterContext *ctx) return 0; } +void ff_convolution_init(ConvolutionContext *s, const char *filter_name) +{ + if (!strcmp(filter_name, "sobel")) { + for (int i = 0; i < 4; i++) { + s->filter[i] = filter_sobel; + s->copy[i] = !((1 << i) & s->planes); + s->size[i] = 3; + s->setup[i] = setup_3x3; + s->rdiv[i] = s->scale; + s->bias[i] = s->delta; + } + if (s->depth > 8) + for (int i = 0; i < 4; i++) + s->filter[i] = filter16_sobel; +#if CONFIG_CONVOLUTION_FILTER && ARCH_X86_64 + ff_sobel_init_x86(s); +#endif + } +} + static int config_input(AVFilterLink *inlink) { AVFilterContext *ctx = inlink->dst; diff --git a/libavfilter/x86/vf_convolution.asm b/libavfilter/x86/vf_convolution.asm index 754d4d1064..8ad2452418 100644 --- a/libavfilter/x86/vf_convolution.asm +++ b/libavfilter/x86/vf_convolution.asm @@ -22,6 +22,10 @@ SECTION_RODATA half: dd 0.5 +data_p1: dd 1 +data_n1: dd -1 +data_p2: dd 2 +data_n2: dd -2 SECTION .text @@ -154,3 +158,161 @@ cglobal filter_3x3, 4, 15, 7, dst, width, rdiv, bias, matrix, ptr, c0, c1, c2, c INIT_XMM sse4 FILTER_3X3 %endif + + +%macro SOBEL_MUL_16 3 + movd xmm2, [%2] + VPBROADCASTD m2, xmm2 + movdqu xmm3, [c%1q + xq] + vpmovzxbd m3, xmm3 + vpdpbusd m%3, m3, m2 +%endmacro + +%macro SOBEL_ADD_16 2 + movdqu xmm3, [c%1q + xq] + vpmovzxbd m3, xmm3 + vpaddd m%2, m3 +%endmacro + + +%macro SOBEL_MUL 2 + movzx ptrd, byte [c%1q + xq] + imul ptrd, [%2] + add rd, ptrd +%endmacro + +%macro SOBEL_ADD 1 + movzx ptrd, byte [c%1q + xq] + add rd, ptrd +%endmacro + +; void filter_sobel_avx512(uint8_t *dst, int width, +; float scale, float delta, const int *const matrix, +; const uint8_t *c[], int peak, int radius, +; int dstride, int stride) +%macro FILTER_SOBEL 0 +%if UNIX64 +cglobal filter_sobel, 4, 15, 7, dst, width, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x +%else +cglobal filter_sobel, 4, 15, 7, dst, width, rdiv, bias, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x +%endif +%if WIN64 + SWAP xmm0, xmm2 + SWAP xmm1, xmm3 + mov r2q, matrixmp + mov r3q, ptrmp + DEFINE_ARGS dst, width, matrix, ptr, c0, c1, c2, c3, c4, c5, c6, c7, c8, r, x +%endif + movsxdifnidn widthq, widthd + VBROADCASTSS m0, xmm0 + VBROADCASTSS m1, xmm1 + pxor m6, m6 + mov c0q, [ptrq + 0*gprsize] + mov c1q, [ptrq + 1*gprsize] + mov c2q, [ptrq + 2*gprsize] + mov c3q, [ptrq + 3*gprsize] + mov c4q, [ptrq + 4*gprsize] + mov c5q, [ptrq + 5*gprsize] + mov c6q, [ptrq + 6*gprsize] + mov c7q, [ptrq + 7*gprsize] + mov c8q, [ptrq + 8*gprsize] + + xor xq, xq + cmp widthq, mmsize/4 + jl .loop2 + + mov rq, widthq + and rq, mmsize/4-1 + sub widthq, rq + +.loop1: + pxor m4, m4 + pxor m5, m5 + + ;Gx + SOBEL_MUL_16 0, data_n1, 4 + SOBEL_MUL_16 1, data_n2, 4 + SOBEL_MUL_16 2, data_n1, 4 + SOBEL_ADD_16 6, 4 + SOBEL_MUL_16 7, data_p2, 4 + SOBEL_ADD_16 8, 4 + + cvtdq2ps m4, m4 + mulps m4, m4 + + ;Gy + SOBEL_MUL_16 0, data_n1, 5 + SOBEL_ADD_16 2, 5 + SOBEL_MUL_16 3, data_n2, 5 + SOBEL_MUL_16 5, data_p2, 5 + SOBEL_MUL_16 6, data_n1, 5 + SOBEL_ADD_16 8, 5 + + cvtdq2ps m5, m5 + VFMADD231PS m4, m5, m5 + + sqrtps m4, m4 + mulps m4, m0 ; sum *= scale + addps m4, m1 ; sum += delta + cvttps2dq m4, m4 + vpmovusdb xmm4, m4 + movdqu [dstq + xq], xmm4 + + add xq, mmsize/4 + cmp xq, widthq + jl .loop1 + + add widthq, rq + cmp xq, widthq + jge .end + +.loop2: + xor rd, rd + pxor m4, m4 + + ;Gx + SOBEL_MUL 0, data_n1 + SOBEL_MUL 1, data_n2 + SOBEL_MUL 2, data_n1 + SOBEL_ADD 6 + SOBEL_MUL 7, data_p2 + SOBEL_ADD 8 + + cvtsi2ss xmm4, rd + mulss xmm4, xmm4 + + xor rd, rd + ;Gy + SOBEL_MUL 0, data_n1 + SOBEL_ADD 2 + SOBEL_MUL 3, data_n2 + SOBEL_MUL 5, data_p2 + SOBEL_MUL 6, data_n1 + SOBEL_ADD 8 + + cvtsi2ss xmm5, rd + mulss xmm5, xmm5 ; b1 * b1 + addss xmm4, xmm5 + + sqrtps xmm4, xmm4 + mulss xmm4, xmm0 ; sum *= rdiv + addss xmm4, xmm1 ; sum += bias + cvttps2dq xmm4, xmm4 ; trunc to integer + packssdw xmm4, xmm4 + packuswb xmm4, xmm4 + movd rd, xmm4 + mov [dstq + xq], rb + + add xq, 1 + cmp xq, widthq + jl .loop2 +.end: + RET +%endmacro + +%if ARCH_X86_64 +%if HAVE_AVX512ICL_EXTERNAL +INIT_ZMM avx512icl +FILTER_SOBEL +%endif +%endif diff --git a/libavfilter/x86/vf_convolution_init.c b/libavfilter/x86/vf_convolution_init.c index b78a47d02b..52a3d28991 100644 --- a/libavfilter/x86/vf_convolution_init.c +++ b/libavfilter/x86/vf_convolution_init.c @@ -29,6 +29,11 @@ void ff_filter_3x3_sse4(uint8_t *dst, int width, const uint8_t *c[], int peak, int radius, int dstride, int stride, int size); +void ff_filter_sobel_avx512icl(uint8_t *dst, int width, + float scale, float delta, const int *const matrix, + const uint8_t *c[], int peak, int radius, + int dstride, int stride, int size); + av_cold void ff_convolution_init_x86(ConvolutionContext *s) { #if ARCH_X86_64 @@ -44,3 +49,16 @@ av_cold void ff_convolution_init_x86(ConvolutionContext *s) } #endif } + +av_cold void ff_sobel_init_x86(ConvolutionContext *s) +{ +#if ARCH_X86_64 + int cpu_flags = av_get_cpu_flags(); + for (int i = 0; i < s->nb_planes; i++) { + if (s->depth == 8) { + if (EXTERNAL_AVX512ICL(cpu_flags)) + s->filter[i] = ff_filter_sobel_avx512icl; + } + } +#endif +} diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile index 1ac170491b..4e91547fde 100644 --- a/tests/checkasm/Makefile +++ b/tests/checkasm/Makefile @@ -44,6 +44,7 @@ AVFILTEROBJS-$(CONFIG_GBLUR_FILTER) += vf_gblur.o AVFILTEROBJS-$(CONFIG_HFLIP_FILTER) += vf_hflip.o AVFILTEROBJS-$(CONFIG_THRESHOLD_FILTER) += vf_threshold.o AVFILTEROBJS-$(CONFIG_NLMEANS_FILTER) += vf_nlmeans.o +AVFILTEROBJS-$(CONFIG_CONVOLUTION_FILTER) += vf_convolution.o CHECKASMOBJS-$(CONFIG_AVFILTER) += $(AVFILTEROBJS-yes) diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c index e56fd3850e..ae5e9d1143 100644 --- a/tests/checkasm/checkasm.c +++ b/tests/checkasm/checkasm.c @@ -191,6 +191,9 @@ static const struct { #if CONFIG_THRESHOLD_FILTER { "vf_threshold", checkasm_check_vf_threshold }, #endif + #if CONFIG_CONVOLUTION_FILTER + { "vf_convolution", checkasm_check_vf_convolution }, + #endif #endif #if CONFIG_SWSCALE { "sw_gbrp", checkasm_check_sw_gbrp }, diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h index d7645d3730..71c722cf77 100644 --- a/tests/checkasm/checkasm.h +++ b/tests/checkasm/checkasm.h @@ -85,6 +85,7 @@ void checkasm_check_vf_eq(void); void checkasm_check_vf_gblur(void); void checkasm_check_vf_hflip(void); void checkasm_check_vf_threshold(void); +void checkasm_check_vf_convolution(void); void checkasm_check_vp8dsp(void); void checkasm_check_vp9dsp(void); void checkasm_check_videodsp(void); diff --git a/tests/checkasm/vf_convolution.c b/tests/checkasm/vf_convolution.c new file mode 100644 index 0000000000..a10da8b45e --- /dev/null +++ b/tests/checkasm/vf_convolution.c @@ -0,0 +1,103 @@ +/* + * 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 "checkasm.h" +#include "libavfilter/avfilter.h" +#include "libavfilter/convolution.h" +#include "libavutil/intreadwrite.h" +#include "libavutil/mem_internal.h" + +#define WIDTH 512 +#define HEIGHT 512 +#define SRC_STRIDE 512 +#define PIXELS (WIDTH * HEIGHT) + +#define randomize_buffers(buf, size) \ + do { \ + int j; \ + uint8_t *tmp_buf = (uint8_t *)buf;\ + for (j = 0; j< size; j++) \ + tmp_buf[j] = rnd() & 0xFF; \ + } while (0) + +static void check_sobel(const char * report_name) +{ + LOCAL_ALIGNED_32(uint8_t, src, [PIXELS]); + LOCAL_ALIGNED_32(uint8_t, dst_ref, [PIXELS]); + LOCAL_ALIGNED_32(uint8_t, dst_new, [PIXELS]); + const int height = WIDTH; + const int width = HEIGHT; + const int stride = SRC_STRIDE; + const int dstride = SRC_STRIDE; + int mode = 0; + const uint8_t *c[49]; + const int radius = 1; + const int bpc = 1; + const int step = mode == MATRIX_COLUMN ? 16 : 1; + const int slice_start = 0; + const int slice_end = height; + int y; + const int sizew = mode == MATRIX_COLUMN ? height : width; + float scale = 2; + float delta = 10; + + ConvolutionContext s; + + s.scale = scale; + s.delta = delta; + s.depth = 8; + s.nb_planes = 3; + ff_convolution_init(&s, "sobel"); + + declare_func(void, uint8_t *dst, int width, float scale, float delta, const int *const matrix, + const uint8_t *c[], int peak, int radius, int dstride, int stride, int size); + + memset(dst_ref, 0, PIXELS); + memset(dst_new, 0, PIXELS); + randomize_buffers(src, PIXELS); + + if (check_func(s.filter[0], "%s", report_name)) { + for (y = slice_start; y < slice_end; y += step) { + const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : radius * bpc; + const int yoff = mode == MATRIX_COLUMN ? radius * dstride : 0; + + s.setup[0](radius, c, src, stride, radius, width, y, height, bpc); + call_ref(dst_ref + yoff + xoff, sizew - 2 * radius, + scale, delta, NULL, c, 0, radius, + dstride, stride, slice_end - step); + call_new(dst_new + yoff + xoff, sizew - 2 * radius, + scale, delta, NULL, c, 0, radius, + dstride, stride, slice_end - step); + if (memcmp(dst_ref + yoff + xoff, dst_new + yoff + xoff, slice_end - step)) + fail(); + bench_new(dst_new + yoff + xoff, sizew - 2 * radius, + scale, delta, NULL, c, 0, radius, + dstride, stride, slice_end - step); + if (mode != MATRIX_COLUMN) + dst_ref += dstride; + } + } + +} + +void checkasm_check_vf_convolution(void) +{ + check_sobel("sobel"); + report("convolution:sobel"); +} -- 2.27.0

3 3

[PATCH 1/6] libavcodec/qsvenc: Add max_frame_size reset support to qsv
by Wenbin Chen 06 Sep '22

06 Sep '22

Signed-off-by: Wenbin Chen <wenbin.chen(a)intel.com> --- doc/encoders.texi | 4 ++++ libavcodec/qsvenc.c | 20 ++++++++++++++++++++ libavcodec/qsvenc.h | 2 ++ 3 files changed, 26 insertions(+) diff --git a/doc/encoders.texi b/doc/encoders.texi index 6d73f74196..69fa46f3ea 100644 --- a/doc/encoders.texi +++ b/doc/encoders.texi @@ -3344,6 +3344,10 @@ Following options can be used durning qsv encoding. @item @var{b_quant_offset} Supported in h264_qsv and hevc_qsv. Change these value to reset qsv codec's qp configuration. + +@item @var{max_frame_size} +Supported in h264_qsv and hevc_qsv. +Change this value to reset qsv codec's MaxFrameSize configuration. @end table @subsection H264 options diff --git a/libavcodec/qsvenc.c b/libavcodec/qsvenc.c index 4831640868..c911b81e7d 100644 --- a/libavcodec/qsvenc.c +++ b/libavcodec/qsvenc.c @@ -828,6 +828,7 @@ static int init_video_param(AVCodecContext *avctx, QSVEncContext *q) q->extco2.ExtBRC = q->extbrc ? MFX_CODINGOPTION_ON : MFX_CODINGOPTION_OFF; if (q->max_frame_size >= 0) q->extco2.MaxFrameSize = q->max_frame_size; + q->old_max_frame_size = q->max_frame_size; if (q->int_ref_type >= 0) q->extco2.IntRefType = q->int_ref_type; if (q->int_ref_cycle_size >= 0) @@ -1673,6 +1674,24 @@ static int update_qp(AVCodecContext *avctx, QSVEncContext *q) return updated; } +static int update_max_frame_size(AVCodecContext *avctx, QSVEncContext *q) +{ + int updated = 0; + + if (avctx->codec_id != AV_CODEC_ID_H264 && avctx->codec_id != AV_CODEC_ID_HEVC) + return 0; + + UPDATE_PARAM(q->old_max_frame_size, q->max_frame_size); + if (!updated) + return 0; + + q->extco2.MaxFrameSize = FFMAX(0, q->max_frame_size); + av_log(avctx, AV_LOG_DEBUG, + "Reset MaxFrameSize: %d;\n", q->extco2.MaxFrameSize); + + return updated; +} + static int update_parameters(AVCodecContext *avctx, QSVEncContext *q, const AVFrame *frame) { @@ -1682,6 +1701,7 @@ static int update_parameters(AVCodecContext *avctx, QSVEncContext *q, return 0; needReset = update_qp(avctx, q); + needReset |= update_max_frame_size(avctx, q); if (!needReset) return 0; diff --git a/libavcodec/qsvenc.h b/libavcodec/qsvenc.h index d35a1318f2..c452c5b806 100644 --- a/libavcodec/qsvenc.h +++ b/libavcodec/qsvenc.h @@ -235,6 +235,8 @@ typedef struct QSVEncContext { float old_i_quant_offset; float old_b_quant_factor; float old_b_quant_offset; + // This is used for max_frame_size reset + int old_max_frame_size; } QSVEncContext; int ff_qsv_enc_init(AVCodecContext *avctx, QSVEncContext *q); -- 2.32.0

3 7

[PATCH v1] avformat/imfdec: check if Asset/Id exists before trying to read it
by pal＠sandflow.com 05 Sep '22

05 Sep '22

From: Pierre-Anthony Lemieux <pal(a)palemieux.com> Fixes Coverity issue #1512406 --- libavformat/imfdec.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/libavformat/imfdec.c b/libavformat/imfdec.c index 5bbe7a53f8..fde91a6419 100644 --- a/libavformat/imfdec.c +++ b/libavformat/imfdec.c @@ -233,7 +233,12 @@ static int parse_imf_asset_map_from_xml_dom(AVFormatContext *s, asset = &(asset_map->assets[asset_map->asset_count]); - if (ff_imf_xml_read_uuid(ff_imf_xml_get_child_element_by_name(asset_element, "Id"), asset->uuid)) { + if (!(node = ff_imf_xml_get_child_element_by_name(asset_element, "Id"))) { + av_log(s, AV_LOG_ERROR, "Unable to parse asset map XML - missing Id node\n"); + return AVERROR_INVALIDDATA; + } + + if (ff_imf_xml_read_uuid(node, asset->uuid)) { av_log(s, AV_LOG_ERROR, "Could not parse UUID from asset in asset map.\n"); return AVERROR_INVALIDDATA; } -- 2.25.1

2 1

[PATCH 0/3] V5: VAAPI: Add high bit depth encode/decode support
by Philip Langdale 05 Sep '22

05 Sep '22

This changeset fills in support for the remaining high bit depth formats the VAAPI exposes. This requires adding more weird Microsoft pixel formats and then mapping them in the VAAPI code. I've also enabled hw mapping between VAAPI and vulkan for sufficiently simple formats. V5: Switched to using alpha-less variants of Y410 and Y412 (called XV30 and XV36 in libdrm) to avoid forcing the overhead of alpha handling in situations where it can't actually be used. Philip Langdale (3): lavu/pixfmt: Add P012, Y212, XV30, and XV36 formats lavc/vaapi: Add support for remaining 10/12bit profiles lavu/hwcontext_vulkan: support mapping VUYX, P012, and XV36 libavcodec/hevcdec.c | 8 +++ libavcodec/vaapi_decode.c | 13 +++++ libavcodec/vaapi_encode.c | 4 ++ libavcodec/vaapi_encode_h265.c | 4 ++ libavcodec/vaapi_encode_vp9.c | 1 + libavcodec/vaapi_hevc.c | 11 +++- libavcodec/vp9.c | 2 + libavutil/hwcontext_vaapi.c | 25 +++++++++ libavutil/hwcontext_vulkan.c | 10 ++++ libavutil/pixdesc.c | 95 +++++++++++++++++++++++++++++++- libavutil/pixfmt.h | 16 ++++++ tests/ref/fate/imgutils | 8 +++ tests/ref/fate/sws-pixdesc-query | 38 +++++++++++++ 13 files changed, 233 insertions(+), 2 deletions(-) -- 2.34.1

2 6

[PATCH v1] avformat/mxf: set stream frame rates for ST 422 essence containers
by pal＠sandflow.com 05 Sep '22

05 Sep '22

From: Pierre-Anthony Lemieux <pal(a)palemieux.com> The MXF demuxer does not currently set AVStream::avg_frame_rate and ::r_frame_rate when J2K essence is wrapped according to SMPTE ST 422. --- libavformat/mxfdec.c | 25 +++++++++++++++++++++++++ 1 file changed, 25 insertions(+) diff --git a/libavformat/mxfdec.c b/libavformat/mxfdec.c index feebff67aa..043e2e06ec 100644 --- a/libavformat/mxfdec.c +++ b/libavformat/mxfdec.c @@ -2140,6 +2140,13 @@ finish_decoding_index: return ret; } +static int mxf_is_st_422(const UID *essence_container_ul) { + static const uint8_t st_422_essence_container_ul[] = { 0x06,0x0e,0x2b,0x34,0x04,0x01,0x01,0x07,0x0d,0x01,0x03,0x01,0x02,0x0c }; + + return essence_container_ul && mxf_match_uid(*essence_container_ul, st_422_essence_container_ul, + sizeof(st_422_essence_container_ul)); +} + static int mxf_is_intra_only(MXFDescriptor *descriptor) { return mxf_get_codec_ul(mxf_intra_only_essence_container_uls, @@ -2892,6 +2899,24 @@ static int mxf_parse_structural_metadata(MXFContext *mxf) av_log(mxf->fc, AV_LOG_INFO, "Unknown frame layout type: %d\n", descriptor->frame_layout); } + if (mxf_is_st_422(essence_container_ul)) { + switch ((*essence_container_ul)[14]) { + case 2: /* Cn: Clip- wrapped Picture Element */ + case 3: /* I1: Interlaced Frame, 1 field/KLV */ + case 4: /* I2: Interlaced Frame, 2 fields/KLV */ + case 6: /* P1: Frame- wrapped Picture Element */ + st->avg_frame_rate = source_track->edit_rate; + st->r_frame_rate = st->avg_frame_rate; + break; + case 5: /* F1: Field-wrapped Picture Element */ + st->avg_frame_rate = av_mul_q(av_make_q(2, 1), source_track->edit_rate); + st->r_frame_rate = st->avg_frame_rate; + break; + default: + break; + } + } + if (st->codecpar->codec_id == AV_CODEC_ID_PRORES) { switch (descriptor->essence_codec_ul[14]) { case 1: st->codecpar->codec_tag = MKTAG('a','p','c','o'); break; -- 2.25.1

3 2

[PATCH 1/2] lavu/hwcontext_qsv: add support for AV_PIX_FMT_VUYX
by Xiang, Haihao 05 Sep '22

05 Sep '22

From: Haihao Xiang <haihao.xiang(a)intel.com> AV_PIX_FMT_VUYX is used in FFmpeg for 8bit 4:4:4 content on Intel HW, and MFX_FOURCC_AYUV is used in the SDK --- libavutil/hwcontext_qsv.c | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/libavutil/hwcontext_qsv.c b/libavutil/hwcontext_qsv.c index 510f422562..a3350eae0f 100644 --- a/libavutil/hwcontext_qsv.c +++ b/libavutil/hwcontext_qsv.c @@ -119,6 +119,8 @@ static const struct { MFX_FOURCC_YUY2 }, { AV_PIX_FMT_Y210, MFX_FOURCC_Y210 }, + { AV_PIX_FMT_VUYX, + MFX_FOURCC_AYUV }, #endif }; @@ -1502,6 +1504,12 @@ static int map_frame_to_surface(const AVFrame *frame, mfxFrameSurface1 *surface) surface->Data.U16 = (mfxU16 *)frame->data[0] + 1; surface->Data.V16 = (mfxU16 *)frame->data[0] + 3; break; + case AV_PIX_FMT_VUYX: + surface->Data.V = frame->data[0]; + surface->Data.U = frame->data[0] + 1; + surface->Data.Y = frame->data[0] + 2; + surface->Data.A = frame->data[0] + 3; + break; #endif default: return MFX_ERR_UNSUPPORTED; -- 2.17.1

3 6

[PATCH 0/5] me_cmp: Provide arm64 neon implementations
by Hubert Mazur 04 Sep '22

04 Sep '22

Provide optimized arm64 neon implementations for functions from motion estimation family. Hubert Mazur (5): lavc/aarch64: Add neon implementation for vsad16 lavc/aarch64: Add neon implementation of vsse16 lavc/aarch64: Add neon implementation for vsad_intra16 lavc/aarch64: Add neon implementation for vsse_intra16 lavc/aarch64: Provide neon implementation of nsse16 libavcodec/aarch64/me_cmp_init_aarch64.c | 30 ++ libavcodec/aarch64/me_cmp_neon.S | 431 +++++++++++++++++++++++ 2 files changed, 461 insertions(+) -- 2.34.1

2 11

[PATCH 0/3] avformat/avisynth: support frame properties
by Stephen Hutchinson 04 Sep '22

04 Sep '22

AviSynth+ 3.6.0 introduced support for frame properties, allowing various metadata to be passed between filters or read out by client programs. Using this, FFmpeg can read Color Range, Transfer Characteristics, Matrix, Color Primaries, Chroma Location, and Field Order information from AviSynth scripts. Reading frame properties through AviSynth's C interface was not possible until a few months ago, though, so client programs that use the C API need version 3.7.1 or higher to be able to take advantage of it. Incorporates a previous patch by emcodem that fixes setting field order on non-frameprop-aware versions of AviSynth. Stephen Hutchinson (2): avisynth: use AviSynth+'s frame properties to set various fields configure: check avisynth header version emcodem (1): avisynth: corrected interlace detection configure | 4 +- libavformat/avisynth.c | 266 ++++++++++++++++++++++++++++++++++++++--- 2 files changed, 253 insertions(+), 17 deletions(-) -- 2.32.0

6 34