ffmpeg源码跟踪笔记之avcodec_decode_video2

1、函数介绍

/**
 * Decode the video frame of size avpkt->size from avpkt->data into picture.
 * Some decoders may support multiple frames in a single AVPacket, such
 * decoders would then just decode the first frame.
 *
 * @warning The input buffer must be FF_INPUT_BUFFER_PADDING_SIZE larger than
 * the actual read bytes because some optimized bitstream readers read 32 or 64
 * bits at once and could read over the end.
 *
 * @warning The end of the input buffer buf should be set to 0 to ensure that
 * no overreading happens for damaged MPEG streams.
 *
 * @note Codecs which have the CODEC_CAP_DELAY capability set have a delay
 * between input and output, these need to be fed with avpkt->data=NULL,
 * avpkt->size=0 at the end to return the remaining frames.
 *
 * @note The AVCodecContext MUST have been opened with @ref avcodec_open2()
 * before packets may be fed to the decoder.
 *
 * @param avctx the codec context
 * @param[out] picture The AVFrame in which the decoded video frame will be stored.
 *             Use av_frame_alloc() to get an AVFrame. The codec will
 *             allocate memory for the actual bitmap by calling the
 *             AVCodecContext.get_buffer2() callback.
 *             When AVCodecContext.refcounted_frames is set to 1, the frame is
 *             reference counted and the returned reference belongs to the
 *             caller. The caller must release the frame using av_frame_unref()
 *             when the frame is no longer needed. The caller may safely write
 *             to the frame if av_frame_is_writable() returns 1.
 *             When AVCodecContext.refcounted_frames is set to 0, the returned
 *             reference belongs to the decoder and is valid only until the
 *             next call to this function or until closing or flushing the
 *             decoder. The caller may not write to it.
 *
 * @param[in] avpkt The input AVPacket containing the input buffer.
 *            You can create such packet with av_init_packet() and by then setting
 *            data and size, some decoders might in addition need other fields like
 *            flags&AV_PKT_FLAG_KEY. All decoders are designed to use the least
 *            fields possible.
 * @param[in,out] got_picture_ptr Zero if no frame could be decompressed, otherwise, it is nonzero.
 * @return On error a negative value is returned, otherwise the number of bytes
 * used or zero if no frame could be decompressed.
 */
int avcodec_decode_video2(AVCodecContext *avctx, AVFrame *picture,
                         int *got_picture_ptr,
                         const AVPacket *avpkt);

2、函数调用图

3、函数源码

int attribute_align_arg avcodec_decode_video2(AVCodecContext *avctx, AVFrame *picture,
                                              int *got_picture_ptr,
                                              const AVPacket *avpkt)
{
    AVCodecInternal *avci = avctx->internal;
    int ret;
    // copy to ensure we do not change avpkt
    AVPacket tmp = *avpkt;
    //如果连解码器都还没打开，就没得搞了
    if (!avctx->codec)
        return AVERROR(EINVAL);//排除非视频
    if (avctx->codec->type != AVMEDIA_TYPE_VIDEO) {
        av_log(avctx, AV_LOG_ERROR, "Invalid media type for video\n");
        return AVERROR(EINVAL);
    }
    //检查下宽、高
    *got_picture_ptr = 0;
    if ((avctx->coded_width || avctx->coded_height) && av_image_check_size(avctx->coded_width, avctx->coded_height, 0, avctx))
        return AVERROR(EINVAL);

    av_frame_unref(picture);

    if ((avctx->codec->capabilities & CODEC_CAP_DELAY) || avpkt->size || (avctx->active_thread_type & FF_THREAD_FRAME)) {
        int did_split = av_packet_split_side_data(&tmp);
        ret = apply_param_change(avctx, &tmp);
        if (ret < 0) {
            av_log(avctx, AV_LOG_ERROR, "Error applying parameter changes.\n");
            if (avctx->err_recognition & AV_EF_EXPLODE)
                goto fail;
        }

        avctx->internal->pkt = &tmp;
        if (HAVE_THREADS && avctx->active_thread_type & FF_THREAD_FRAME)
            ret = ff_thread_decode_frame(avctx, picture, got_picture_ptr,
                                         &tmp);
        else {
            ret = avctx->codec->decode(avctx, picture, got_picture_ptr,
                                       &tmp);
            picture->pkt_dts = avpkt->dts;

            if(!avctx->has_b_frames){
                av_frame_set_pkt_pos(picture, avpkt->pos);
            }
            //FIXME these should be under if(!avctx->has_b_frames)
            /* get_buffer is supposed to set frame parameters */
            if (!(avctx->codec->capabilities & CODEC_CAP_DR1)) {
                if (!picture->sample_aspect_ratio.num)    picture->sample_aspect_ratio = avctx->sample_aspect_ratio;
                if (!picture->width)                      picture->width               = avctx->width;
                if (!picture->height)                     picture->height              = avctx->height;
                if (picture->format == AV_PIX_FMT_NONE)   picture->format              = avctx->pix_fmt;
            }
        }
        add_metadata_from_side_data(avctx, picture);

fail:
        emms_c(); //needed to avoid an emms_c() call before every return;

        avctx->internal->pkt = NULL;
        if (did_split) {
            av_packet_free_side_data(&tmp);
            if(ret == tmp.size)
                ret = avpkt->size;
        }

        if (*got_picture_ptr) {
            if (!avctx->refcounted_frames) {
                int err = unrefcount_frame(avci, picture);
                if (err < 0)
                    return err;
            }

            avctx->frame_number++;
            av_frame_set_best_effort_timestamp(picture,
                                               guess_correct_pts(avctx,
                                                                 picture->pkt_pts,
                                                                 picture->pkt_dts));
        } else
            av_frame_unref(picture);
    } else
        ret = 0;

    /* many decoders assign whole AVFrames, thus overwriting extended_data;
     * make sure it's set correctly */
    av_assert0(!picture->extended_data || picture->extended_data == picture->data);

#if FF_API_AVCTX_TIMEBASE
    if (avctx->framerate.num > 0 && avctx->framerate.den > 0)
        avctx->time_base = av_inv_q(av_mul_q(avctx->framerate, (AVRational){avctx->ticks_per_frame, 1}));
#endif

    return ret;
}

4、简单解释

函数avcodec_decode_video2出乎意料的简单，大致流程如下：

1、检查参数，解码器，宽，高。

2、调用codec的decode函数指针解码。

3、调用av_frame_set_best_effort_timestamp设置时间戳。

函数avcodec_decode_video2之所以这么简单，是因为所有解码的工作都由codec的decode函数指针做了，看一下yuv decoder的定义：

AVCodec ff_rawvideo_decoder = {
    .name           = "rawvideo",
    .long_name      = NULL_IF_CONFIG_SMALL("raw video"),
    .type           = AVMEDIA_TYPE_VIDEO,
    .id             = AV_CODEC_ID_RAWVIDEO,
    .priv_data_size = sizeof(RawVideoContext),
    .init           = raw_init_decoder,
    .close          = raw_close_decoder,
    .decode         = raw_decode,
    .priv_class     = &rawdec_class,
    .capabilities   = CODEC_CAP_PARAM_CHANGE,
};

在看看decode指针指向的raw_decode函数：

static int raw_decode(AVCodecContext *avctx, void *data, int *got_frame,
                      AVPacket *avpkt)
{
    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(avctx->pix_fmt);
    RawVideoContext *context       = avctx->priv_data;
    const uint8_t *buf             = avpkt->data;
    int buf_size                   = avpkt->size;
    int linesize_align             = 4;
    int res, len;
    int need_copy;

    AVFrame   *frame   = data;
    AVPicture *picture = data;

    if ((avctx->bits_per_coded_sample == 4 || avctx->bits_per_coded_sample == 2) &&
        avctx->pix_fmt == AV_PIX_FMT_PAL8 &&
       (!avctx->codec_tag || avctx->codec_tag == MKTAG('r','a','w',' '))) {
        context->is_2_4_bpp = 1;
        context->frame_size = avpicture_get_size(avctx->pix_fmt,
                                                 FFALIGN(avctx->width, 16),
                                                 avctx->height);
    } else {
        context->is_lt_16bpp = av_get_bits_per_pixel(desc) == 16 && avctx->bits_per_coded_sample && avctx->bits_per_coded_sample < 16;
        context->frame_size = avpicture_get_size(avctx->pix_fmt, avctx->width,
                                                 avctx->height);
    }
    if (context->frame_size < 0)
        return context->frame_size;

    need_copy = !avpkt->buf || context->is_2_4_bpp || context->is_yuv2 || context->is_lt_16bpp;

    frame->pict_type        = AV_PICTURE_TYPE_I;
    frame->key_frame        = 1;

    res = ff_decode_frame_props(avctx, frame);
    if (res < 0)
        return res;

    av_frame_set_pkt_pos     (frame, avctx->internal->pkt->pos);
    av_frame_set_pkt_duration(frame, avctx->internal->pkt->duration);

    if (context->tff >= 0) {
        frame->interlaced_frame = 1;
        frame->top_field_first  = context->tff;
    }

    if ((res = av_image_check_size(avctx->width, avctx->height, 0, avctx)) < 0)
        return res;

    if (need_copy)
        frame->buf[0] = av_buffer_alloc(FFMAX(context->frame_size, buf_size));
    else
        frame->buf[0] = av_buffer_ref(avpkt->buf);
    if (!frame->buf[0])
        return AVERROR(ENOMEM);

    //2bpp and 4bpp raw in avi and mov (yes this is ugly ...)
    if (context->is_2_4_bpp) {
        int i;
        uint8_t *dst = frame->buf[0]->data;
        buf_size = context->frame_size - AVPALETTE_SIZE;
        if (avctx->bits_per_coded_sample == 4) {
            for (i = 0; 2 * i + 1 < buf_size && i<avpkt->size; i++) {
                dst[2 * i + 0] = buf[i] >> 4;
                dst[2 * i + 1] = buf[i] & 15;
            }
            linesize_align = 8;
        } else {
            av_assert0(avctx->bits_per_coded_sample == 2);
            for (i = 0; 4 * i + 3 < buf_size && i<avpkt->size; i++) {
                dst[4 * i + 0] = buf[i] >> 6;
                dst[4 * i + 1] = buf[i] >> 4 & 3;
                dst[4 * i + 2] = buf[i] >> 2 & 3;
                dst[4 * i + 3] = buf[i]      & 3;
            }
            linesize_align = 16;
        }
        buf = dst;
    } else if (context->is_lt_16bpp) {
        uint8_t *dst = frame->buf[0]->data;
        int packed = (avctx->codec_tag & 0xFFFFFF) == MKTAG('B','I','T', 0);
        int swap   =  avctx->codec_tag >> 24;

        if (packed && swap) {
            av_fast_padded_malloc(&context->bitstream_buf, &context->bitstream_buf_size, buf_size);
            if (!context->bitstream_buf)
                return AVERROR(ENOMEM);
            if (swap == 16)
                context->bbdsp.bswap16_buf(context->bitstream_buf, (const uint16_t*)buf, buf_size / 2);
            else if (swap == 32)
                context->bbdsp.bswap_buf(context->bitstream_buf, (const uint32_t*)buf, buf_size / 4);
            else
                return AVERROR_INVALIDDATA;
            buf = context->bitstream_buf;
        }

        if (desc->flags & AV_PIX_FMT_FLAG_BE)
            scale16be(avctx, dst, buf, buf_size, packed);
        else
            scale16le(avctx, dst, buf, buf_size, packed);

        buf = dst;
    } else if (need_copy) {
        memcpy(frame->buf[0]->data, buf, buf_size);
        buf = frame->buf[0]->data;
    }

    if (avctx->codec_tag == MKTAG('A', 'V', '1', 'x') ||
        avctx->codec_tag == MKTAG('A', 'V', 'u', 'p'))
        buf += buf_size - context->frame_size;

    len = context->frame_size - (avctx->pix_fmt==AV_PIX_FMT_PAL8 ? AVPALETTE_SIZE : 0);
    if (buf_size < len && (avctx->codec_tag & 0xFFFFFF) != MKTAG('B','I','T', 0)) {
        av_log(avctx, AV_LOG_ERROR, "Invalid buffer size, packet size %d < expected frame_size %d\n", buf_size, len);
        av_buffer_unref(&frame->buf[0]);
        return AVERROR(EINVAL);
    }

    if ((res = avpicture_fill(picture, buf, avctx->pix_fmt,
                              avctx->width, avctx->height)) < 0) {
        av_buffer_unref(&frame->buf[0]);
        return res;
    }

    if (avctx->pix_fmt == AV_PIX_FMT_PAL8) {
        const uint8_t *pal = av_packet_get_side_data(avpkt, AV_PKT_DATA_PALETTE,
                                                     NULL);

        if (pal) {
            av_buffer_unref(&context->palette);
            context->palette = av_buffer_alloc(AVPALETTE_SIZE);
            if (!context->palette) {
                av_buffer_unref(&frame->buf[0]);
                return AVERROR(ENOMEM);
            }
            memcpy(context->palette->data, pal, AVPALETTE_SIZE);
            frame->palette_has_changed = 1;
        }
    }

    if ((avctx->pix_fmt==AV_PIX_FMT_BGR24    ||
        avctx->pix_fmt==AV_PIX_FMT_GRAY8    ||
        avctx->pix_fmt==AV_PIX_FMT_RGB555LE ||
        avctx->pix_fmt==AV_PIX_FMT_RGB555BE ||
        avctx->pix_fmt==AV_PIX_FMT_RGB565LE ||
        avctx->pix_fmt==AV_PIX_FMT_MONOWHITE ||
        avctx->pix_fmt==AV_PIX_FMT_PAL8) &&
        FFALIGN(frame->linesize[0], linesize_align) * avctx->height <= buf_size)
        frame->linesize[0] = FFALIGN(frame->linesize[0], linesize_align);

    if (avctx->pix_fmt == AV_PIX_FMT_NV12 && avctx->codec_tag == MKTAG('N', 'V', '1', '2') &&
        FFALIGN(frame->linesize[0], linesize_align) * avctx->height +
        FFALIGN(frame->linesize[1], linesize_align) * ((avctx->height + 1) / 2) <= buf_size) {
        int la0 = FFALIGN(frame->linesize[0], linesize_align);
        frame->data[1] += (la0 - frame->linesize[0]) * avctx->height;
        frame->linesize[0] = la0;
        frame->linesize[1] = FFALIGN(frame->linesize[1], linesize_align);
    }

    if ((avctx->pix_fmt == AV_PIX_FMT_PAL8 && buf_size < context->frame_size) ||
        (desc->flags & AV_PIX_FMT_FLAG_PSEUDOPAL)) {
        frame->buf[1]  = av_buffer_ref(context->palette);
        if (!frame->buf[1]) {
            av_buffer_unref(&frame->buf[0]);
            return AVERROR(ENOMEM);
        }
        frame->data[1] = frame->buf[1]->data;
    }

    if (avctx->pix_fmt == AV_PIX_FMT_BGR24 &&
        ((frame->linesize[0] + 3) & ~3) * avctx->height <= buf_size)
        frame->linesize[0] = (frame->linesize[0] + 3) & ~3;

    if (context->flip)
        flip(avctx, picture);

    if (avctx->codec_tag == MKTAG('Y', 'V', '1', '2') ||
        avctx->codec_tag == MKTAG('Y', 'V', '1', '6') ||
        avctx->codec_tag == MKTAG('Y', 'V', '2', '4') ||
        avctx->codec_tag == MKTAG('Y', 'V', 'U', '9'))
        FFSWAP(uint8_t *, picture->data[1], picture->data[2]);

    if (avctx->codec_tag == AV_RL32("I420") && (avctx->width+1)*(avctx->height+1) * 3/2 == buf_size) {
        picture->data[1] = picture->data[1] +  (avctx->width+1)*(avctx->height+1) -avctx->width*avctx->height;
        picture->data[2] = picture->data[2] + ((avctx->width+1)*(avctx->height+1) -avctx->width*avctx->height)*5/4;
    }

    if (avctx->codec_tag == AV_RL32("yuv2") &&
        avctx->pix_fmt   == AV_PIX_FMT_YUYV422) {
        int x, y;
        uint8_t *line = picture->data[0];
        for (y = 0; y < avctx->height; y++) {
            for (x = 0; x < avctx->width; x++)
                line[2 * x + 1] ^= 0x80;
            line += picture->linesize[0];
        }
    }

    if (avctx->field_order > AV_FIELD_PROGRESSIVE) { /* we have interlaced material flagged in container */
        frame->interlaced_frame = 1;
        if (avctx->field_order == AV_FIELD_TT || avctx->field_order == AV_FIELD_TB)
            frame->top_field_first = 1;
    }

    *got_frame = 1;
    return buf_size;
}

可以看到所有工作都被这个函数做完了。

以后研究到编解码器那一层再详细分析各个解码器的解码工作。