FFmpeg源码剖析-框架：process_input()

process_input()函数位于ffmpeg.c

1. 函数概述

它的作用的从文件中读取一个packet,并解码；

2. 函数调用结构图

3. 代码分析

/* 
 * 它是在open_input_file()->add_input_stream()中初始化的，
 * Add all the streams from the given input file to the global
 * list of input streams. */
 * static void add_input_streams(OptionsContext *o, AVFormatContext *ic)
 * {
 *    for (i = 0; i < ic->nb_streams; i++) {
 *       InputStream *ist = av_mallocz(sizeof(*ist));
 *
 *       GROW_ARRAY(input_streams, nb_input_streams);
 *       input_streams[nb_input_streams - 1] = ist;
 * }
 */
InputStream **input_streams = NULL;


/*
 * Return
 * - 0 -- one packet was read and processed
 * - AVERROR(EAGAIN) -- no packets were available for selected file,
 *   this function should be called again
 * - AVERROR_EOF -- this function should not be called again
 */
static int process_input(int file_index)
{
    InputFile *ifile = input_files[file_index];
    AVFormatContext *is;
    InputStream *ist;
    AVPacket pkt;
    int ret, i, j;
    int64_t duration;
    int64_t pkt_dts;

    is  = ifile->ctx;
    /* 读取一个packet(可以是视频的一帧，音频的多帧) */
    ret = get_input_packet(ifile, &pkt);
  
    reset_eagain();
    
    ist = input_streams[ifile->ist_index + pkt.stream_index];

    ist->data_size += pkt.size;
    ist->nb_packets++;

    if(!ist->wrap_correction_done && is->start_time != AV_NOPTS_VALUE && ist->st->pts_wrap_bits < 64){
        int64_t stime, stime2;

        // Correcting starttime based on the enabled streams
        stime = av_rescale_q(is->start_time, AV_TIME_BASE_Q, ist->st->time_base);
        stime2= stime + (1ULLpts_wrap_bits);
        ist->wrap_correction_done = 1;

        if(stime2 > stime && pkt.dts != AV_NOPTS_VALUE && pkt.dts > stime + (1LL<<(ist->st->pts_wrap_bits-1))) {
            pkt.dts -= 1ULLpts_wrap_bits;
            ist->wrap_correction_done = 0;
        }
        if(stime2 > stime && pkt.pts != AV_NOPTS_VALUE && pkt.pts > stime + (1LL<<(ist->st->pts_wrap_bits-1))) {
            pkt.pts -= 1ULLpts_wrap_bits;
            ist->wrap_correction_done = 0;
        }
    }

    /* 将packet中的时间戳转换成ffmpeg内部的时间戳 */
    if (pkt.dts != AV_NOPTS_VALUE)
        pkt.dts += av_rescale_q(ifile->ts_offset, AV_TIME_BASE_Q, ist->st->time_base);
    if (pkt.pts != AV_NOPTS_VALUE)
        pkt.pts += av_rescale_q(ifile->ts_offset, AV_TIME_BASE_Q, ist->st->time_base);

    if (pkt.pts != AV_NOPTS_VALUE)
        pkt.pts *= ist->ts_scale;
    if (pkt.dts != AV_NOPTS_VALUE)
        pkt.dts *= ist->ts_scale;

    pkt_dts = av_rescale_q_rnd(pkt.dts, ist->st->time_base, AV_TIME_BASE_Q, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
    if ((ist->dec_ctx->codec_type == AVMEDIA_TYPE_VIDEO ||
         ist->dec_ctx->codec_type == AVMEDIA_TYPE_AUDIO) &&
        pkt_dts != AV_NOPTS_VALUE && ist->next_dts == AV_NOPTS_VALUE && !copy_ts
        && (is->iformat->flags & AVFMT_TS_DISCONT) && ifile->last_ts != AV_NOPTS_VALUE) {
        int64_t delta   = pkt_dts - ifile->last_ts;
        if (delta < -1LL*dts_delta_threshold*AV_TIME_BASE ||
            delta >  1LL*dts_delta_threshold*AV_TIME_BASE){
            ifile->ts_offset -= delta;
            av_log(NULL, AV_LOG_DEBUG,
                   "Inter stream timestamp discontinuity %"PRId64", new offset= %"PRId64"\n",
                   delta, ifile->ts_offset);
            pkt.dts -= av_rescale_q(delta, AV_TIME_BASE_Q, ist->st->time_base);
            if (pkt.pts != AV_NOPTS_VALUE)
                pkt.pts -= av_rescale_q(delta, AV_TIME_BASE_Q, ist->st->time_base);
        }
    }

    /* 计算出当前packet所包含的音频/视频帧在显示时要持续的时长 */
    duration = av_rescale_q(ifile->duration, ifile->time_base, ist->st->time_base);
    if (pkt.pts != AV_NOPTS_VALUE) {
        pkt.pts += duration;
        ist->max_pts = FFMAX(pkt.pts, ist->max_pts);
        ist->min_pts = FFMIN(pkt.pts, ist->min_pts);
    }
    if (pkt.dts != AV_NOPTS_VALUE)
        pkt.dts += duration;

    pkt_dts = av_rescale_q_rnd(pkt.dts, ist->st->time_base, AV_TIME_BASE_Q, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
    if ((ist->dec_ctx->codec_type == AVMEDIA_TYPE_VIDEO ||
         ist->dec_ctx->codec_type == AVMEDIA_TYPE_AUDIO) &&
         pkt_dts != AV_NOPTS_VALUE && ist->next_dts != AV_NOPTS_VALUE &&
        !copy_ts) {
        int64_t delta   = pkt_dts - ist->next_dts;
        if (is->iformat->flags & AVFMT_TS_DISCONT) {
            if (delta < -1LL*dts_delta_threshold*AV_TIME_BASE ||
                delta >  1LL*dts_delta_threshold*AV_TIME_BASE ||
                pkt_dts + AV_TIME_BASE/10 < FFMAX(ist->pts, ist->dts)) {
                ifile->ts_offset -= delta;
                av_log(NULL, AV_LOG_DEBUG,
                       "timestamp discontinuity %"PRId64", new offset= %"PRId64"\n",
                       delta, ifile->ts_offset);
                pkt.dts -= av_rescale_q(delta, AV_TIME_BASE_Q, ist->st->time_base);
                if (pkt.pts != AV_NOPTS_VALUE)
                    pkt.pts -= av_rescale_q(delta, AV_TIME_BASE_Q, ist->st->time_base);
            }
        } else {
            if ( delta < -1LL*dts_error_threshold*AV_TIME_BASE ||
                 delta >  1LL*dts_error_threshold*AV_TIME_BASE) {
                av_log(NULL, AV_LOG_WARNING, "DTS %"PRId64", next:%"PRId64" st:%d invalid dropping\n", pkt.dts, ist->next_dts, pkt.stream_index);
                pkt.dts = AV_NOPTS_VALUE;
            }
            if (pkt.pts != AV_NOPTS_VALUE){
                int64_t pkt_pts = av_rescale_q(pkt.pts, ist->st->time_base, AV_TIME_BASE_Q);
                delta   = pkt_pts - ist->next_dts;
                if ( delta < -1LL*dts_error_threshold*AV_TIME_BASE ||
                     delta >  1LL*dts_error_threshold*AV_TIME_BASE) {
                    av_log(NULL, AV_LOG_WARNING, "PTS %"PRId64", next:%"PRId64" invalid dropping st:%d\n", pkt.pts, ist->next_dts, pkt.stream_index);
                    pkt.pts = AV_NOPTS_VALUE;
                }
            }
        }
    }

    if (pkt.dts != AV_NOPTS_VALUE)
        ifile->last_ts = av_rescale_q(pkt.dts, ist->st->time_base, AV_TIME_BASE_Q);

    sub2video_heartbeat(ist, pkt.pts);

    /* 解码当前的这个packet */
    process_input_packet(ist, &pkt, 0);

discard_packet:
    av_packet_unref(&pkt);

    return 0;
}

4.1 process_input_packet()

/* pkt = NULL means EOF (needed to flush decoder buffers) */
static int process_input_packet(InputStream *ist, const AVPacket *pkt, int no_eof)
{
    int ret = 0, i;
    int repeating = 0;
    int eof_reached = 0;

    AVPacket avpkt;
    if (!ist->saw_first_ts) {
        ist->dts = ist->st->avg_frame_rate.num ? - ist->dec_ctx->has_b_frames * AV_TIME_BASE / av_q2d(ist->st->avg_frame_rate) : 0;
        ist->pts = 0;
        if (pkt && pkt->pts != AV_NOPTS_VALUE && !ist->decoding_needed) {
            ist->dts += av_rescale_q(pkt->pts, ist->st->time_base, AV_TIME_BASE_Q);
            ist->pts = ist->dts; //unused but better to set it to a value thats not totally wrong
        }
        ist->saw_first_ts = 1;
    }

    if (ist->next_dts == AV_NOPTS_VALUE)
        ist->next_dts = ist->dts;
    if (ist->next_pts == AV_NOPTS_VALUE)
        ist->next_pts = ist->pts;

    if (!pkt) {
        /* EOF handling */
        av_init_packet(&avpkt);
        avpkt.data = NULL;
        avpkt.size = 0;
    } else {
        avpkt = *pkt;
    }

    if (pkt && pkt->dts != AV_NOPTS_VALUE) {
        ist->next_dts = ist->dts = av_rescale_q(pkt->dts, ist->st->time_base, AV_TIME_BASE_Q);
        if (ist->dec_ctx->codec_type != AVMEDIA_TYPE_VIDEO || !ist->decoding_needed)
            ist->next_pts = ist->pts = ist->dts;
    }

    // while we have more to decode or while the decoder did output something on EOF
    while (ist->decoding_needed) {
        int duration = 0;
        int got_output = 0;

        ist->pts = ist->next_pts;
        ist->dts = ist->next_dts;
        
        /* 依据当前的packet是视频、音频还是数据，选择对应的解码器解码当前的packet */
        switch (ist->dec_ctx->codec_type) {
        case AVMEDIA_TYPE_AUDIO:
            ret = decode_audio    (ist, repeating ? NULL : &avpkt, &got_output);
            break;
        case AVMEDIA_TYPE_VIDEO:
            ret = decode_video    (ist, repeating ? NULL : &avpkt, &got_output, !pkt);
            if (!repeating || !pkt || got_output) {
                if (pkt && pkt->duration) {
                    duration = av_rescale_q(pkt->duration, ist->st->time_base, AV_TIME_BASE_Q);
                } else if(ist->dec_ctx->framerate.num != 0 && ist->dec_ctx->framerate.den != 0) {
                    int ticks= av_stream_get_parser(ist->st) ? av_stream_get_parser(ist->st)->repeat_pict+1 : ist->dec_ctx->ticks_per_frame;
                    duration = ((int64_t)AV_TIME_BASE *
                                    ist->dec_ctx->framerate.den * ticks) /
                                    ist->dec_ctx->framerate.num / ist->dec_ctx->ticks_per_frame;
                }

                if(ist->dts != AV_NOPTS_VALUE && duration) {
                    ist->next_dts += duration;
                }else
                    ist->next_dts = AV_NOPTS_VALUE;
            }

            if (got_output)
                ist->next_pts += duration; //FIXME the duration is not correct in some cases
            break;
        case AVMEDIA_TYPE_SUBTITLE:
            if (repeating)
                break;
            ret = transcode_subtitles(ist, &avpkt, &got_output);
            if (!pkt && ret >= 0)
                ret = AVERROR_EOF;
            break;
        default:
            return -1;
        }

        if (ret == AVERROR_EOF) {
            eof_reached = 1;
            break;
        }

        if (ret < 0) {
            av_log(NULL, AV_LOG_ERROR, "Error while decoding stream #%d:%d: %s\n",
                   ist->file_index, ist->st->index, av_err2str(ret));
            if (exit_on_error)
                exit_program(1);
            // Decoding might not terminate if we're draining the decoder, and
            // the decoder keeps returning an error.
            // This should probably be considered a libavcodec issue.
            // Sample: fate-vsynth1-dnxhd-720p-hr-lb
            if (!pkt)
                eof_reached = 1;
            break;
        }

        if (!got_output)
            break;

        // During draining, we might get multiple output frames in this loop.
        // ffmpeg.c does not drain the filter chain on configuration changes,
        // which means if we send multiple frames at once to the filters, and
        // one of those frames changes configuration, the buffered frames will
        // be lost. This can upset certain FATE tests.
        // Decode only 1 frame per call on EOF to appease these FATE tests.
        // The ideal solution would be to rewrite decoding to use the new
        // decoding API in a better way.
        if (!pkt)
            break;

        repeating = 1;
    }

    /* after flushing, send an EOF on all the filter inputs attached to the stream */
    /* except when looping we need to flush but not to send an EOF */
    if (!pkt && ist->decoding_needed && eof_reached && !no_eof) {
        int ret = send_filter_eof(ist);
        if (ret < 0) {
            av_log(NULL, AV_LOG_FATAL, "Error marking filters as finished\n");
            exit_program(1);
        }
    }

    /* handle stream copy */
    ...

    for (i = 0; pkt && i < nb_output_streams; i++) {
        OutputStream *ost = output_streams[i];

        if (!check_output_constraints(ist, ost) || ost->encoding_needed)
            continue;

        do_streamcopy(ist, ost, pkt);
    }
    return !eof_reached;
}