[FFmpeg-devel] [PATCH 1/3] vp9_parser: Return stream properties
James Almer
jamrial at gmail.com
Tue Mar 19 02:40:57 EET 2019
On 3/18/2019 9:18 PM, Mark Thompson wrote:
> Rewrites the parser entirely, using CBS for header parsing. A new
> entrypoint to the CBS code is added to avoid any copy overhead.
> ---
> On 18/03/2019 03:22, Li, Zhong wrote:
>> If you think that case can be resolved for VP9 parser, could you please provide more details?
>
> Since all the infrastructure is already available it seemed easier to just do it than to try to explain it all needs to fit together.
>
> - Mark
>
>
> libavcodec/cbs_vp9.c | 62 ++++++++++++++++++++++
> libavcodec/cbs_vp9.h | 12 +++++
> libavcodec/vp9_parser.c | 113 +++++++++++++++++++++++++++-------------
> 3 files changed, 152 insertions(+), 35 deletions(-)
>
> diff --git a/libavcodec/cbs_vp9.c b/libavcodec/cbs_vp9.c
> index 0b5f137ed8..237416a06f 100644
> --- a/libavcodec/cbs_vp9.c
> +++ b/libavcodec/cbs_vp9.c
> @@ -690,3 +690,65 @@ const CodedBitstreamType ff_cbs_type_vp9 = {
>
> .close = &cbs_vp9_close,
> };
> +
> +int ff_cbs_vp9_parse_headers(CodedBitstreamContext *ctx,
> + VP9RawFrameHeader *header,
> + const uint8_t *data, size_t data_size)
> +{
> + GetBitContext gbc;
> + uint8_t superframe_header;
> + int err;
> +
> + if (data_size < 1)
> + return AVERROR_INVALIDDATA;
> +
> + superframe_header = data[data_size - 1];
> + if ((superframe_header & 0xe0) == 0xc0) {
> + VP9RawSuperframeIndex sfi;
> + size_t index_size, pos;
> + int i;
> +
> + index_size = 2 + (((superframe_header & 0x18) >> 3) + 1) *
> + ((superframe_header & 0x07) + 1);
> + if (index_size > data_size)
> + return AVERROR_INVALIDDATA;
> +
> + err = init_get_bits(&gbc, data + data_size - index_size,
> + 8 * index_size);
> + if (err < 0)
> + return err;
> +
> + err = cbs_vp9_read_superframe_index(ctx, &gbc, &sfi);
> + if (err < 0)
> + return err;
> +
> + pos = 0;
> + for (i = 0; i <= sfi.frames_in_superframe_minus_1; i++) {
> + if (pos + sfi.frame_sizes[i] + index_size > data_size)
> + return AVERROR_INVALIDDATA;
> +
> + err = init_get_bits(&gbc, data + pos,
> + 8 * sfi.frame_sizes[i]);
> + if (err < 0)
> + return err;
> +
> + memset(header, 0, sizeof(*header));
> + err = cbs_vp9_read_uncompressed_header(ctx, &gbc, header);
> + if (err < 0)
> + return err;
> +
> + pos += sfi.frame_sizes[i];
> + }
> +
> + } else {
> + err = init_get_bits(&gbc, data, 8 * data_size);
> + if (err < 0)
> + return err;
> +
> + err = cbs_vp9_read_uncompressed_header(ctx, &gbc, header);
> + if (err < 0)
> + return err;
> + }
> +
> + return 0;
> +}
> diff --git a/libavcodec/cbs_vp9.h b/libavcodec/cbs_vp9.h
> index 4c9b2f880d..0b7e8dd71b 100644
> --- a/libavcodec/cbs_vp9.h
> +++ b/libavcodec/cbs_vp9.h
> @@ -214,4 +214,16 @@ typedef struct CodedBitstreamVP9Context {
> } CodedBitstreamVP9Context;
>
>
> +/**
> + * Entrypoint for VP9 parser.
> + *
> + * Parses headers only in a VP9 frame, and does not require refcounting.
> + *
> + * The data may contain multiple frames in a superframe; all will be parsed
> + * but the returned information will be for the final frame.
> + */
> +int ff_cbs_vp9_parse_headers(CodedBitstreamContext *ctx,
> + VP9RawFrameHeader *header,
> + const uint8_t *data, size_t data_size);
> +
> #endif /* AVCODEC_CBS_VP9_H */
> diff --git a/libavcodec/vp9_parser.c b/libavcodec/vp9_parser.c
> index c957a75667..5dd4d8d434 100644
> --- a/libavcodec/vp9_parser.c
> +++ b/libavcodec/vp9_parser.c
> @@ -1,8 +1,5 @@
> /*
> - * VP9 compatible video decoder
> - *
> - * Copyright (C) 2013 Ronald S. Bultje <rsbultje gmail com>
> - * Copyright (C) 2013 Clément Bœsch <u pkh me>
> + * VP9 parser
You should add your copyright since you rewrote it from scratch.
> *
> * This file is part of FFmpeg.
> *
> @@ -21,50 +18,96 @@
> * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
> */
>
> -#include "libavutil/intreadwrite.h"
> -#include "libavcodec/get_bits.h"
> +#include "libavutil/avassert.h"
> +#include "cbs.h"
> +#include "cbs_vp9.h"
> #include "parser.h"
>
> -static int parse(AVCodecParserContext *ctx,
> - AVCodecContext *avctx,
> - const uint8_t **out_data, int *out_size,
> - const uint8_t *data, int size)
> +typedef struct VP9ParserContext {
> + CodedBitstreamContext *cbc;
> + VP9RawFrameHeader frame_header;
> +} VP9ParserContext;
> +
> +static const enum AVPixelFormat vp9_pix_fmts[3][2][2] = {
> + { // 8-bit.
> + { AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P },
> + { AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV420P },
> + },
> + { // 10-bit.
> + { AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV440P10 },
> + { AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV420P10 },
> + },
> + { // 12-bit.
> + { AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12 },
> + { AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV420P12 },
> + },
> +};
> +
> +static int vp9_parser_parse(AVCodecParserContext *ctx,
> + AVCodecContext *avctx,
> + const uint8_t **out_data, int *out_size,
> + const uint8_t *data, int size)
> {
> - GetBitContext gb;
> - int res, profile, keyframe;
> + VP9ParserContext *s = ctx->priv_data;
> + const CodedBitstreamVP9Context *vp9 = s->cbc->priv_data;
> + const VP9RawFrameHeader *fh;
> + int err;
>
> *out_data = data;
> *out_size = size;
Maybe set a few sane default values here for pict_type and key_frame,
like h264 and av1 parsers do.
>
> - if (!size || (res = init_get_bits8(&gb, data, size)) < 0)
> - return size; // parsers can't return errors
> - get_bits(&gb, 2); // frame marker
> - profile = get_bits1(&gb);
> - profile |= get_bits1(&gb) << 1;
> - if (profile == 3) profile += get_bits1(&gb);
> - if (profile > 3)
> - return size;
> -
> - avctx->profile = profile;
> -
> - if (get_bits1(&gb)) {
> - keyframe = 0;
> - } else {
> - keyframe = !get_bits1(&gb);
> - }
> + if (!size)
> + return 0;
>
> - if (!keyframe) {
> - ctx->pict_type = AV_PICTURE_TYPE_P;
> - ctx->key_frame = 0;
> - } else {
> - ctx->pict_type = AV_PICTURE_TYPE_I;
> - ctx->key_frame = 1;
> + s->cbc->log_ctx = avctx;
Adding a custom context like you suggested before may be a better idea
(I need to do the same for av1).
> +
> + err = ff_cbs_vp9_parse_headers(s->cbc, &s->frame_header, data, size);
> + if (err < 0) {
> + av_log(avctx, AV_LOG_WARNING, "Failed to parse VP9 frame headers.\n");
> + goto end;
> }
> + fh = &s->frame_header;
> +
> + avctx->profile = vp9->profile;
> + avctx->level = FF_LEVEL_UNKNOWN;
> +
> + ctx->width = ctx->coded_width = vp9->frame_width;
> + ctx->height = ctx->coded_height = vp9->frame_height;
> +
> + ctx->pict_type = fh->intra_only ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
> + ctx->key_frame = !fh->frame_type;
> +
> + ctx->picture_structure = AV_PICTURE_STRUCTURE_FRAME;
> +
> + av_assert0(vp9->bit_depth == 8 ||
> + vp9->bit_depth == 10 ||
> + vp9->bit_depth == 12);
> +
> + ctx->format = vp9_pix_fmts[(vp9->bit_depth - 8) / 2]
> + [vp9->subsampling_x][vp9->subsampling_y];
> +
> +end:
> + s->cbc->log_ctx = NULL;
>
> return size;
> }
>
> +static av_cold int vp9_parser_init(AVCodecParserContext *ctx)
> +{
> + VP9ParserContext *s = ctx->priv_data;
> + return ff_cbs_init(&s->cbc, AV_CODEC_ID_VP9, NULL);
> +}
> +
> +static av_cold void vp9_parser_close(AVCodecParserContext *ctx)
> +{
> + VP9ParserContext *s = ctx->priv_data;
> + ff_cbs_close(&s->cbc);
> +}
> +
> AVCodecParser ff_vp9_parser = {
> .codec_ids = { AV_CODEC_ID_VP9 },
> - .parser_parse = parse,
> + .priv_data_size = sizeof(VP9ParserContext),
> + .parser_init = vp9_parser_init,
> + .parser_close = vp9_parser_close,
> + .parser_parse = vp9_parser_parse,
> };
>
More information about the ffmpeg-devel
mailing list