IJKPLAYER误配nobuffer选项增加首开时间案例

1 背景

    在使用IJKPLAYER二开时,为了降低直播时延,使能了ffmpegnobuffer选项,该选项也是ffplay所支持的。自测发现,该选项对直播时延的确有一定帮助,但会增加首开时间

    本文记录下此次踩坑经历。

2 问题现象

  • IJKPLAYER首开时延不稳定,好则1秒内,坏则若干秒;
  • 目标是优化到秒开。

3 原因分析

3.1 分析过程

    以下是原因分析过程:

  • 通过打印日志分析,发现耗时主要集中在avformat_open_inputavformat_find_stream_info耗时比较理想,多次测试均在几十ms内,而avformat_open_input则耗时短则300ms长则1秒+,具体原因是SRS3.0版本在信令交互后推流数据时有不稳定延时,此文暂不做此分析,专注分析nobuffer选项;
  • av_read_frame读取的首个video AVPacket不是IDR帧;
  • ffprobe -show_frames rtmp://ip/live/test发现每次过来的首个video帧一定是IDR帧;
  • 用ffmpeg的api写了一个简易的拉流程序,不设置任何选项,断点调试发现过来的首个vidoe的AVPacket必是IDR帧;

    由此,得出一个结论:

  • ffmpeg内部具体是avformat模块一定有丢包,而且很可能是由于设置了某个option选项导致的。 

    而后,经过逐一注释掉option选项,发现罪魁祸首就是使能了ffmpegnobuffer选项:

// 原代码使能了nobuffer选项,查明原因后注释掉此行代码    
//[options setFormatOptionValue:@"nobuffer" forKey:@"fflags"];

    在注释了以上这行代码之后,断点调试发现av_read_frame读取到的首个video的AVPacket必是IDR帧。 nobuffer选项导致的首开时间增加问题得以解决。

3.2 源码剖析

    在找到了原因之后,我们再来看看ffmpeg的源码,查看下若使能了nobuffer选项,底层到底干了些什么事情。

3.2.1 avformat_open_input

    ffmpegnobuffer选项是在调用avformat_open_input接口时通过最后1个options参数传入:

int avformat_open_input(AVFormatContext **ps, const char *filename,
                        const AVInputFormat *fmt, AVDictionary **options)

    展开avformat_open_input方法看看主要实现: 

int avformat_open_input(AVFormatContext **ps, const char *filename,
                        const AVInputFormat *fmt, AVDictionary **options)
{
    AVFormatContext *s = *ps;
    FFFormatContext *si;
    AVDictionary *tmp = NULL;
    ID3v2ExtraMeta *id3v2_extra_meta = NULL;
    int ret = 0;

    // 分配AVFormatContext实例,并为av_class成员赋值为av_format_context_class
    if (!s && !(s = avformat_alloc_context()))
        return AVERROR(ENOMEM);

    // 此处略去不相关代码
    ......

    // 根据传入的options参数,完成对AVFormatContext的flags变量赋值
    if ((ret = av_opt_set_dict(s, &tmp)) < 0)
        goto fail;

    // 此处略去不相关代码
    ......
}

     在avformat_alloc_context方法里注册av_format_context_class

AVFormatContext *avformat_alloc_context(void)
{
    FFFormatContext *const si = av_mallocz(sizeof(*si));
    AVFormatContext *s;

    if (!si)
        return NULL;

    s = &si->pub;
    // 注册了av_format_context_class
    s->av_class = &av_format_context_class;
    
    ......

    return s;
}

    而av_format_context_class的定义: 

static const AVClass av_format_context_class = {
    .class_name     = "AVFormatContext",
    .item_name      = format_to_name,
    .option         = avformat_options,
    .version        = LIBAVUTIL_VERSION_INT,
    .child_next     = format_child_next,
    .child_class_iterate = format_child_class_iterate,
    .category       = AV_CLASS_CATEGORY_MUXER,
    .get_category   = get_category,
};

      可以看到,av_format_context_class的option选项被赋值为avformat_options了,以下是avformat_options的详细定义:

// 计算x成员在AVFormatContext中的偏移量 
#define OFFSET(x) offsetof(AVFormatContext,x)
#define DEFAULT 0 //should be NAN but it does not work as it is not a constant in glibc as required by ANSI/ISO C
//these names are too long to be readable
#define E AV_OPT_FLAG_ENCODING_PARAM
#define D AV_OPT_FLAG_DECODING_PARAM

static const AVOption avformat_options[] = {
{"avioflags", NULL, OFFSET(avio_flags), AV_OPT_TYPE_FLAGS, {.i64 = DEFAULT }, INT_MIN, INT_MAX, D|E, "avioflags"},
{"direct", "reduce buffering", 0, AV_OPT_TYPE_CONST, {.i64 = AVIO_FLAG_DIRECT }, INT_MIN, INT_MAX, D|E, "avioflags"},
{"probesize", "set probing size", OFFSET(probesize), AV_OPT_TYPE_INT64, {.i64 = 5000000 }, 32, INT64_MAX, D},
{"formatprobesize", "number of bytes to probe file format", OFFSET(format_probesize), AV_OPT_TYPE_INT, {.i64 = PROBE_BUF_MAX}, 0, INT_MAX-1, D},
{"packetsize", "set packet size", OFFSET(packet_size), AV_OPT_TYPE_INT, {.i64 = DEFAULT }, 0, INT_MAX, E},
// fflags成员在AVFormatContext中的偏移量为OFFSET(flags)
{"fflags", NULL, OFFSET(flags), AV_OPT_TYPE_FLAGS, {.i64 = AVFMT_FLAG_AUTO_BSF }, INT_MIN, INT_MAX, D|E, "fflags"},
{"flush_packets", "reduce the latency by flushing out packets immediately", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_FLUSH_PACKETS }, INT_MIN, INT_MAX, E, "fflags"},
{"ignidx", "ignore index", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_IGNIDX }, INT_MIN, INT_MAX, D, "fflags"},
{"genpts", "generate pts", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_GENPTS }, INT_MIN, INT_MAX, D, "fflags"},
{"nofillin", "do not fill in missing values that can be exactly calculated", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_NOFILLIN }, INT_MIN, INT_MAX, D, "fflags"},
{"noparse", "disable AVParsers, this needs nofillin too", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_NOPARSE }, INT_MIN, INT_MAX, D, "fflags"},
{"igndts", "ignore dts", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_IGNDTS }, INT_MIN, INT_MAX, D, "fflags"},
{"discardcorrupt", "discard corrupted frames", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_DISCARD_CORRUPT }, INT_MIN, INT_MAX, D, "fflags"},
{"sortdts", "try to interleave outputted packets by dts", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_SORT_DTS }, INT_MIN, INT_MAX, D, "fflags"},
{"fastseek", "fast but inaccurate seeks", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_FAST_SEEK }, INT_MIN, INT_MAX, D, "fflags"},
// nobuffer配置 
{"nobuffer", "reduce the latency introduced by optional buffering", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_FLAG_NOBUFFER }, 0, INT_MAX, D, "fflags"},
{"bitexact", "do not write random/volatile data", 0, AV_OPT_TYPE_CONST, { .i64 = AVFMT_FLAG_BITEXACT }, 0, 0, E, "fflags" },
#if FF_API_LAVF_SHORTEST
{"shortest", "stop muxing with the shortest stream", 0, AV_OPT_TYPE_CONST, { .i64 = AVFMT_FLAG_SHORTEST }, 0, 0, E | AV_OPT_FLAG_DEPRECATED, "fflags" },
#endif
{"autobsf", "add needed bsfs automatically", 0, AV_OPT_TYPE_CONST, { .i64 = AVFMT_FLAG_AUTO_BSF }, 0, 0, E, "fflags" },
{"seek2any", "allow seeking to non-keyframes on demuxer level when supported", OFFSET(seek2any), AV_OPT_TYPE_BOOL, {.i64 = 0 }, 0, 1, D},
{"analyzeduration", "specify how many microseconds are analyzed to probe the input", OFFSET(max_analyze_duration), AV_OPT_TYPE_INT64, {.i64 = 0 }, 0, INT64_MAX, D},
{"cryptokey", "decryption key", OFFSET(key), AV_OPT_TYPE_BINARY, {.dbl = 0}, 0, 0, D},
{"indexmem", "max memory used for timestamp index (per stream)", OFFSET(max_index_size), AV_OPT_TYPE_INT, {.i64 = 1<<20 }, 0, INT_MAX, D},
{"rtbufsize", "max memory used for buffering real-time frames", OFFSET(max_picture_buffer), AV_OPT_TYPE_INT, {.i64 = 3041280 }, 0, INT_MAX, D}, /* defaults to 1s of 15fps 352x288 YUYV422 video */
{"fdebug", "print specific debug info", OFFSET(debug), AV_OPT_TYPE_FLAGS, {.i64 = DEFAULT }, 0, INT_MAX, E|D, "fdebug"},
{"ts", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_FDEBUG_TS }, INT_MIN, INT_MAX, E|D, "fdebug"},
{"max_delay", "maximum muxing or demuxing delay in microseconds", OFFSET(max_delay), AV_OPT_TYPE_INT, {.i64 = -1 }, -1, INT_MAX, E|D},
{"start_time_realtime", "wall-clock time when stream begins (PTS==0)", OFFSET(start_time_realtime), AV_OPT_TYPE_INT64, {.i64 = AV_NOPTS_VALUE}, INT64_MIN, INT64_MAX, E},
{"fpsprobesize", "number of frames used to probe fps", OFFSET(fps_probe_size), AV_OPT_TYPE_INT, {.i64 = -1}, -1, INT_MAX-1, D},
{"audio_preload", "microseconds by which audio packets should be interleaved earlier", OFFSET(audio_preload), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX-1, E},
{"chunk_duration", "microseconds for each chunk", OFFSET(max_chunk_duration), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX-1, E},
{"chunk_size", "size in bytes for each chunk", OFFSET(max_chunk_size), AV_OPT_TYPE_INT, {.i64 = 0}, 0, INT_MAX-1, E},
/* this is a crutch for avconv, since it cannot deal with identically named options in different contexts.
 * to be removed when avconv is fixed */
{"f_err_detect", "set error detection flags (deprecated; use err_detect, save via avconv)", OFFSET(error_recognition), AV_OPT_TYPE_FLAGS, {.i64 = AV_EF_CRCCHECK }, INT_MIN, INT_MAX, D, "err_detect"},
{"err_detect", "set error detection flags", OFFSET(error_recognition), AV_OPT_TYPE_FLAGS, {.i64 = AV_EF_CRCCHECK }, INT_MIN, INT_MAX, D, "err_detect"},
{"crccheck", "verify embedded CRCs", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_CRCCHECK }, INT_MIN, INT_MAX, D, "err_detect"},
{"bitstream", "detect bitstream specification deviations", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_BITSTREAM }, INT_MIN, INT_MAX, D, "err_detect"},
{"buffer", "detect improper bitstream length", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_BUFFER }, INT_MIN, INT_MAX, D, "err_detect"},
{"explode", "abort decoding on minor error detection", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_EXPLODE }, INT_MIN, INT_MAX, D, "err_detect"},
{"ignore_err", "ignore errors", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_IGNORE_ERR }, INT_MIN, INT_MAX, D, "err_detect"},
{"careful",    "consider things that violate the spec, are fast to check and have not been seen in the wild as errors", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_CAREFUL }, INT_MIN, INT_MAX, D, "err_detect"},
{"compliant",  "consider all spec non compliancies as errors", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_COMPLIANT | AV_EF_CAREFUL }, INT_MIN, INT_MAX, D, "err_detect"},
{"aggressive", "consider things that a sane encoder shouldn't do as an error", 0, AV_OPT_TYPE_CONST, {.i64 = AV_EF_AGGRESSIVE | AV_EF_COMPLIANT | AV_EF_CAREFUL}, INT_MIN, INT_MAX, D, "err_detect"},
{"use_wallclock_as_timestamps", "use wallclock as timestamps", OFFSET(use_wallclock_as_timestamps), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, D},
{"skip_initial_bytes", "set number of bytes to skip before reading header and frames", OFFSET(skip_initial_bytes), AV_OPT_TYPE_INT64, {.i64 = 0}, 0, INT64_MAX-1, D},
{"correct_ts_overflow", "correct single timestamp overflows", OFFSET(correct_ts_overflow), AV_OPT_TYPE_BOOL, {.i64 = 1}, 0, 1, D},
{"flush_packets", "enable flushing of the I/O context after each packet", OFFSET(flush_packets), AV_OPT_TYPE_INT, {.i64 = -1}, -1, 1, E},
{"metadata_header_padding", "set number of bytes to be written as padding in a metadata header", OFFSET(metadata_header_padding), AV_OPT_TYPE_INT, {.i64 = -1}, -1, INT_MAX, E},
{"output_ts_offset", "set output timestamp offset", OFFSET(output_ts_offset), AV_OPT_TYPE_DURATION, {.i64 = 0}, -INT64_MAX, INT64_MAX, E},
{"max_interleave_delta", "maximum buffering duration for interleaving", OFFSET(max_interleave_delta), AV_OPT_TYPE_INT64, { .i64 = 10000000 }, 0, INT64_MAX, E },
{"f_strict", "how strictly to follow the standards (deprecated; use strict, save via avconv)", OFFSET(strict_std_compliance), AV_OPT_TYPE_INT, {.i64 = DEFAULT }, INT_MIN, INT_MAX, D|E, "strict"},
{"strict", "how strictly to follow the standards", OFFSET(strict_std_compliance), AV_OPT_TYPE_INT, {.i64 = DEFAULT }, INT_MIN, INT_MAX, D|E, "strict"},
{"very", "strictly conform to a older more strict version of the spec or reference software", 0, AV_OPT_TYPE_CONST, {.i64 = FF_COMPLIANCE_VERY_STRICT }, INT_MIN, INT_MAX, D|E, "strict"},
{"strict", "strictly conform to all the things in the spec no matter what the consequences", 0, AV_OPT_TYPE_CONST, {.i64 = FF_COMPLIANCE_STRICT }, INT_MIN, INT_MAX, D|E, "strict"},
{"normal", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = FF_COMPLIANCE_NORMAL }, INT_MIN, INT_MAX, D|E, "strict"},
{"unofficial", "allow unofficial extensions", 0, AV_OPT_TYPE_CONST, {.i64 = FF_COMPLIANCE_UNOFFICIAL }, INT_MIN, INT_MAX, D|E, "strict"},
{"experimental", "allow non-standardized experimental variants", 0, AV_OPT_TYPE_CONST, {.i64 = FF_COMPLIANCE_EXPERIMENTAL }, INT_MIN, INT_MAX, D|E, "strict"},
{"max_ts_probe", "maximum number of packets to read while waiting for the first timestamp", OFFSET(max_ts_probe), AV_OPT_TYPE_INT, { .i64 = 50 }, 0, INT_MAX, D },
{"avoid_negative_ts", "shift timestamps so they start at 0", OFFSET(avoid_negative_ts), AV_OPT_TYPE_INT, {.i64 = -1}, -1, 2, E, "avoid_negative_ts"},
{"auto",              "enabled when required by target format",    0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_AVOID_NEG_TS_AUTO },              INT_MIN, INT_MAX, E, "avoid_negative_ts"},
{"disabled",          "do not change timestamps",                  0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_AVOID_NEG_TS_DISABLED },          INT_MIN, INT_MAX, E, "avoid_negative_ts"},
{"make_non_negative", "shift timestamps so they are non negative", 0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_AVOID_NEG_TS_MAKE_NON_NEGATIVE }, INT_MIN, INT_MAX, E, "avoid_negative_ts"},
{"make_zero",         "shift timestamps so they start at 0",       0, AV_OPT_TYPE_CONST, {.i64 = AVFMT_AVOID_NEG_TS_MAKE_ZERO },         INT_MIN, INT_MAX, E, "avoid_negative_ts"},
{"dump_separator", "set information dump field separator", OFFSET(dump_separator), AV_OPT_TYPE_STRING, {.str = ", "}, 0, 0, D|E},
{"codec_whitelist", "List of decoders that are allowed to be used", OFFSET(codec_whitelist), AV_OPT_TYPE_STRING, { .str = NULL },  0, 0, D },
{"format_whitelist", "List of demuxers that are allowed to be used", OFFSET(format_whitelist), AV_OPT_TYPE_STRING, { .str = NULL },  0, 0, D },
{"protocol_whitelist", "List of protocols that are allowed to be used", OFFSET(protocol_whitelist), AV_OPT_TYPE_STRING, { .str = NULL },  0, 0, D },
{"protocol_blacklist", "List of protocols that are not allowed to be used", OFFSET(protocol_blacklist), AV_OPT_TYPE_STRING, { .str = NULL },  0, 0, D },
{"max_streams", "maximum number of streams", OFFSET(max_streams), AV_OPT_TYPE_INT, { .i64 = 1000 }, 0, INT_MAX, D },
{"skip_estimate_duration_from_pts", "skip duration calculation in estimate_timings_from_pts", OFFSET(skip_estimate_duration_from_pts), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, D},
{"max_probe_packets", "Maximum number of packets to probe a codec", OFFSET(max_probe_packets), AV_OPT_TYPE_INT, { .i64 = 2500 }, 0, INT_MAX, D },
{NULL},
};

    AVFormatContext的定义如下: 

typedef struct AVFormatContext {
    /**
     * A class for logging and @ref avoptions. Set by avformat_alloc_context().
     * Exports (de)muxer private options if they exist.
     */
    const AVClass *av_class;

    /**
     * The input container format.
     *
     * Demuxing only, set by avformat_open_input().
     */
    const struct AVInputFormat *iformat;

    /**
     * The output container format.
     *
     * Muxing only, must be set by the caller before avformat_write_header().
     */
    const struct AVOutputFormat *oformat;

    /**
     * Format private data. This is an AVOptions-enabled struct
     * if and only if iformat/oformat.priv_class is not NULL.
     *
     * - muxing: set by avformat_write_header()
     * - demuxing: set by avformat_open_input()
     */
    void *priv_data;

    /**
     * I/O context.
     *
     * - demuxing: either set by the user before avformat_open_input() (then
     *             the user must close it manually) or set by avformat_open_input().
     * - muxing: set by the user before avformat_write_header(). The caller must
     *           take care of closing / freeing the IO context.
     *
     * Do NOT set this field if AVFMT_NOFILE flag is set in
     * iformat/oformat.flags. In such a case, the (de)muxer will handle
     * I/O in some other way and this field will be NULL.
     */
    AVIOContext *pb;

    /* stream info */
    /**
     * Flags signalling stream properties. A combination of AVFMTCTX_*.
     * Set by libavformat.
     */
    int ctx_flags;

    /**
     * Number of elements in AVFormatContext.streams.
     *
     * Set by avformat_new_stream(), must not be modified by any other code.
     */
    unsigned int nb_streams;
    /**
     * A list of all streams in the file. New streams are created with
     * avformat_new_stream().
     *
     * - demuxing: streams are created by libavformat in avformat_open_input().
     *             If AVFMTCTX_NOHEADER is set in ctx_flags, then new streams may also
     *             appear in av_read_frame().
     * - muxing: streams are created by the user before avformat_write_header().
     *
     * Freed by libavformat in avformat_free_context().
     */
    AVStream **streams;

    /**
     * input or output URL. Unlike the old filename field, this field has no
     * length restriction.
     *
     * - demuxing: set by avformat_open_input(), initialized to an empty
     *             string if url parameter was NULL in avformat_open_input().
     * - muxing: may be set by the caller before calling avformat_write_header()
     *           (or avformat_init_output() if that is called first) to a string
     *           which is freeable by av_free(). Set to an empty string if it
     *           was NULL in avformat_init_output().
     *
     * Freed by libavformat in avformat_free_context().
     */
    char *url;

    /**
     * Position of the first frame of the component, in
     * AV_TIME_BASE fractional seconds. NEVER set this value directly:
     * It is deduced from the AVStream values.
     *
     * Demuxing only, set by libavformat.
     */
    int64_t start_time;

    /**
     * Duration of the stream, in AV_TIME_BASE fractional
     * seconds. Only set this value if you know none of the individual stream
     * durations and also do not set any of them. This is deduced from the
     * AVStream values if not set.
     *
     * Demuxing only, set by libavformat.
     */
    int64_t duration;

    /**
     * Total stream bitrate in bit/s, 0 if not
     * available. Never set it directly if the file_size and the
     * duration are known as FFmpeg can compute it automatically.
     */
    int64_t bit_rate;

    unsigned int packet_size;
    int max_delay;

    /**
     * Flags modifying the (de)muxer behaviour. A combination of AVFMT_FLAG_*.
     * Set by the user before avformat_open_input() / avformat_write_header().
     */
    int flags;
#define AVFMT_FLAG_GENPTS       0x0001 ///< Generate missing pts even if it requires parsing future frames.
#define AVFMT_FLAG_IGNIDX       0x0002 ///< Ignore index.
#define AVFMT_FLAG_NONBLOCK     0x0004 ///< Do not block when reading packets from input.
#define AVFMT_FLAG_IGNDTS       0x0008 ///< Ignore DTS on frames that contain both DTS & PTS
#define AVFMT_FLAG_NOFILLIN     0x0010 ///< Do not infer any values from other values, just return what is stored in the container
#define AVFMT_FLAG_NOPARSE      0x0020 ///< Do not use AVParsers, you also must set AVFMT_FLAG_NOFILLIN as the fillin code works on frames and no parsing -> no frames. Also seeking to frames can not work if parsing to find frame boundaries has been disabled
#define AVFMT_FLAG_NOBUFFER     0x0040 ///< Do not buffer frames when possible
#define AVFMT_FLAG_CUSTOM_IO    0x0080 ///< The caller has supplied a custom AVIOContext, don't avio_close() it.
#define AVFMT_FLAG_DISCARD_CORRUPT  0x0100 ///< Discard frames marked corrupted
#define AVFMT_FLAG_FLUSH_PACKETS    0x0200 ///< Flush the AVIOContext every packet.
/**
 * When muxing, try to avoid writing any random/volatile data to the output.
 * This includes any random IDs, real-time timestamps/dates, muxer version, etc.
 *
 * This flag is mainly intended for testing.
 */
#define AVFMT_FLAG_BITEXACT         0x0400
#define AVFMT_FLAG_SORT_DTS    0x10000 ///< try to interleave outputted packets by dts (using this flag can slow demuxing down)
#define AVFMT_FLAG_FAST_SEEK   0x80000 ///< Enable fast, but inaccurate seeks for some formats
#if FF_API_LAVF_SHORTEST
#define AVFMT_FLAG_SHORTEST   0x100000 ///< Stop muxing when the shortest stream stops.
#endif
#define AVFMT_FLAG_AUTO_BSF   0x200000 ///< Add bitstream filters as requested by the muxer

    /**
     * Maximum number of bytes read from input in order to determine stream
     * properties. Used when reading the global header and in
     * avformat_find_stream_info().
     *
     * Demuxing only, set by the caller before avformat_open_input().
     *
     * @note this is \e not  used for determining the \ref AVInputFormat
     *       "input format"
     * @sa format_probesize
     */
    int64_t probesize;

    /**
     * Maximum duration (in AV_TIME_BASE units) of the data read
     * from input in avformat_find_stream_info().
     * Demuxing only, set by the caller before avformat_find_stream_info().
     * Can be set to 0 to let avformat choose using a heuristic.
     */
    int64_t max_analyze_duration;

    const uint8_t *key;
    int keylen;

    unsigned int nb_programs;
    AVProgram **programs;

    /**
     * Forced video codec_id.
     * Demuxing: Set by user.
     */
    enum AVCodecID video_codec_id;

    /**
     * Forced audio codec_id.
     * Demuxing: Set by user.
     */
    enum AVCodecID audio_codec_id;

    /**
     * Forced subtitle codec_id.
     * Demuxing: Set by user.
     */
    enum AVCodecID subtitle_codec_id;

    /**
     * Maximum amount of memory in bytes to use for the index of each stream.
     * If the index exceeds this size, entries will be discarded as
     * needed to maintain a smaller size. This can lead to slower or less
     * accurate seeking (depends on demuxer).
     * Demuxers for which a full in-memory index is mandatory will ignore
     * this.
     * - muxing: unused
     * - demuxing: set by user
     */
    unsigned int max_index_size;

    /**
     * Maximum amount of memory in bytes to use for buffering frames
     * obtained from realtime capture devices.
     */
    unsigned int max_picture_buffer;

    /**
     * Number of chapters in AVChapter array.
     * When muxing, chapters are normally written in the file header,
     * so nb_chapters should normally be initialized before write_header
     * is called. Some muxers (e.g. mov and mkv) can also write chapters
     * in the trailer.  To write chapters in the trailer, nb_chapters
     * must be zero when write_header is called and non-zero when
     * write_trailer is called.
     * - muxing: set by user
     * - demuxing: set by libavformat
     */
    unsigned int nb_chapters;
    AVChapter **chapters;

    /**
     * Metadata that applies to the whole file.
     *
     * - demuxing: set by libavformat in avformat_open_input()
     * - muxing: may be set by the caller before avformat_write_header()
     *
     * Freed by libavformat in avformat_free_context().
     */
    AVDictionary *metadata;

    /**
     * Start time of the stream in real world time, in microseconds
     * since the Unix epoch (00:00 1st January 1970). That is, pts=0 in the
     * stream was captured at this real world time.
     * - muxing: Set by the caller before avformat_write_header(). If set to
     *           either 0 or AV_NOPTS_VALUE, then the current wall-time will
     *           be used.
     * - demuxing: Set by libavformat. AV_NOPTS_VALUE if unknown. Note that
     *             the value may become known after some number of frames
     *             have been received.
     */
    int64_t start_time_realtime;

    /**
     * The number of frames used for determining the framerate in
     * avformat_find_stream_info().
     * Demuxing only, set by the caller before avformat_find_stream_info().
     */
    int fps_probe_size;

    /**
     * Error recognition; higher values will detect more errors but may
     * misdetect some more or less valid parts as errors.
     * Demuxing only, set by the caller before avformat_open_input().
     */
    int error_recognition;

    /**
     * Custom interrupt callbacks for the I/O layer.
     *
     * demuxing: set by the user before avformat_open_input().
     * muxing: set by the user before avformat_write_header()
     * (mainly useful for AVFMT_NOFILE formats). The callback
     * should also be passed to avio_open2() if it's used to
     * open the file.
     */
    AVIOInterruptCB interrupt_callback;

    /**
     * Flags to enable debugging.
     */
    int debug;
#define FF_FDEBUG_TS        0x0001

    /**
     * Maximum buffering duration for interleaving.
     *
     * To ensure all the streams are interleaved correctly,
     * av_interleaved_write_frame() will wait until it has at least one packet
     * for each stream before actually writing any packets to the output file.
     * When some streams are "sparse" (i.e. there are large gaps between
     * successive packets), this can result in excessive buffering.
     *
     * This field specifies the maximum difference between the timestamps of the
     * first and the last packet in the muxing queue, above which libavformat
     * will output a packet regardless of whether it has queued a packet for all
     * the streams.
     *
     * Muxing only, set by the caller before avformat_write_header().
     */
    int64_t max_interleave_delta;

    /**
     * Allow non-standard and experimental extension
     * @see AVCodecContext.strict_std_compliance
     */
    int strict_std_compliance;

    /**
     * Flags indicating events happening on the file, a combination of
     * AVFMT_EVENT_FLAG_*.
     *
     * - demuxing: may be set by the demuxer in avformat_open_input(),
     *   avformat_find_stream_info() and av_read_frame(). Flags must be cleared
     *   by the user once the event has been handled.
     * - muxing: may be set by the user after avformat_write_header() to
     *   indicate a user-triggered event.  The muxer will clear the flags for
     *   events it has handled in av_[interleaved]_write_frame().
     */
    int event_flags;
/**
 * - demuxing: the demuxer read new metadata from the file and updated
 *   AVFormatContext.metadata accordingly
 * - muxing: the user updated AVFormatContext.metadata and wishes the muxer to
 *   write it into the file
 */
#define AVFMT_EVENT_FLAG_METADATA_UPDATED 0x0001

    /**
     * Maximum number of packets to read while waiting for the first timestamp.
     * Decoding only.
     */
    int max_ts_probe;

    /**
     * Avoid negative timestamps during muxing.
     * Any value of the AVFMT_AVOID_NEG_TS_* constants.
     * Note, this works better when using av_interleaved_write_frame().
     * - muxing: Set by user
     * - demuxing: unused
     */
    int avoid_negative_ts;
#define AVFMT_AVOID_NEG_TS_AUTO             -1 ///< Enabled when required by target format
#define AVFMT_AVOID_NEG_TS_DISABLED          0 ///< Do not shift timestamps even when they are negative.
#define AVFMT_AVOID_NEG_TS_MAKE_NON_NEGATIVE 1 ///< Shift timestamps so they are non negative
#define AVFMT_AVOID_NEG_TS_MAKE_ZERO         2 ///< Shift timestamps so that they start at 0

    /**
     * Transport stream id.
     * This will be moved into demuxer private options. Thus no API/ABI compatibility
     */
    int ts_id;

    /**
     * Audio preload in microseconds.
     * Note, not all formats support this and unpredictable things may happen if it is used when not supported.
     * - encoding: Set by user
     * - decoding: unused
     */
    int audio_preload;

    /**
     * Max chunk time in microseconds.
     * Note, not all formats support this and unpredictable things may happen if it is used when not supported.
     * - encoding: Set by user
     * - decoding: unused
     */
    int max_chunk_duration;

    /**
     * Max chunk size in bytes
     * Note, not all formats support this and unpredictable things may happen if it is used when not supported.
     * - encoding: Set by user
     * - decoding: unused
     */
    int max_chunk_size;

    /**
     * forces the use of wallclock timestamps as pts/dts of packets
     * This has undefined results in the presence of B frames.
     * - encoding: unused
     * - decoding: Set by user
     */
    int use_wallclock_as_timestamps;

    /**
     * avio flags, used to force AVIO_FLAG_DIRECT.
     * - encoding: unused
     * - decoding: Set by user
     */
    int avio_flags;

    /**
     * The duration field can be estimated through various ways, and this field can be used
     * to know how the duration was estimated.
     * - encoding: unused
     * - decoding: Read by user
     */
    enum AVDurationEstimationMethod duration_estimation_method;

    /**
     * Skip initial bytes when opening stream
     * - encoding: unused
     * - decoding: Set by user
     */
    int64_t skip_initial_bytes;

    /**
     * Correct single timestamp overflows
     * - encoding: unused
     * - decoding: Set by user
     */
    unsigned int correct_ts_overflow;

    /**
     * Force seeking to any (also non key) frames.
     * - encoding: unused
     * - decoding: Set by user
     */
    int seek2any;

    /**
     * Flush the I/O context after each packet.
     * - encoding: Set by user
     * - decoding: unused
     */
    int flush_packets;

    /**
     * format probing score.
     * The maximal score is AVPROBE_SCORE_MAX, its set when the demuxer probes
     * the format.
     * - encoding: unused
     * - decoding: set by avformat, read by user
     */
    int probe_score;

    /**
     * Maximum number of bytes read from input in order to identify the
     * \ref AVInputFormat "input format". Only used when the format is not set
     * explicitly by the caller.
     *
     * Demuxing only, set by the caller before avformat_open_input().
     *
     * @sa probesize
     */
    int format_probesize;

    /**
     * ',' separated list of allowed decoders.
     * If NULL then all are allowed
     * - encoding: unused
     * - decoding: set by user
     */
    char *codec_whitelist;

    /**
     * ',' separated list of allowed demuxers.
     * If NULL then all are allowed
     * - encoding: unused
     * - decoding: set by user
     */
    char *format_whitelist;

    /**
     * IO repositioned flag.
     * This is set by avformat when the underlaying IO context read pointer
     * is repositioned, for example when doing byte based seeking.
     * Demuxers can use the flag to detect such changes.
     */
    int io_repositioned;

    /**
     * Forced video codec.
     * This allows forcing a specific decoder, even when there are multiple with
     * the same codec_id.
     * Demuxing: Set by user
     */
    const struct AVCodec *video_codec;

    /**
     * Forced audio codec.
     * This allows forcing a specific decoder, even when there are multiple with
     * the same codec_id.
     * Demuxing: Set by user
     */
    const struct AVCodec *audio_codec;

    /**
     * Forced subtitle codec.
     * This allows forcing a specific decoder, even when there are multiple with
     * the same codec_id.
     * Demuxing: Set by user
     */
    const struct AVCodec *subtitle_codec;

    /**
     * Forced data codec.
     * This allows forcing a specific decoder, even when there are multiple with
     * the same codec_id.
     * Demuxing: Set by user
     */
    const struct AVCodec *data_codec;

    /**
     * Number of bytes to be written as padding in a metadata header.
     * Demuxing: Unused.
     * Muxing: Set by user via av_format_set_metadata_header_padding.
     */
    int metadata_header_padding;

    /**
     * User data.
     * This is a place for some private data of the user.
     */
    void *opaque;

    /**
     * Callback used by devices to communicate with application.
     */
    av_format_control_message control_message_cb;

    /**
     * Output timestamp offset, in microseconds.
     * Muxing: set by user
     */
    int64_t output_ts_offset;

    /**
     * dump format separator.
     * can be ", " or "\n      " or anything else
     * - muxing: Set by user.
     * - demuxing: Set by user.
     */
    uint8_t *dump_separator;

    /**
     * Forced Data codec_id.
     * Demuxing: Set by user.
     */
    enum AVCodecID data_codec_id;

    /**
     * ',' separated list of allowed protocols.
     * - encoding: unused
     * - decoding: set by user
     */
    char *protocol_whitelist;

    /**
     * A callback for opening new IO streams.
     *
     * Whenever a muxer or a demuxer needs to open an IO stream (typically from
     * avformat_open_input() for demuxers, but for certain formats can happen at
     * other times as well), it will call this callback to obtain an IO context.
     *
     * @param s the format context
     * @param pb on success, the newly opened IO context should be returned here
     * @param url the url to open
     * @param flags a combination of AVIO_FLAG_*
     * @param options a dictionary of additional options, with the same
     *                semantics as in avio_open2()
     * @return 0 on success, a negative AVERROR code on failure
     *
     * @note Certain muxers and demuxers do nesting, i.e. they open one or more
     * additional internal format contexts. Thus the AVFormatContext pointer
     * passed to this callback may be different from the one facing the caller.
     * It will, however, have the same 'opaque' field.
     */
    int (*io_open)(struct AVFormatContext *s, AVIOContext **pb, const char *url,
                   int flags, AVDictionary **options);

#if FF_API_AVFORMAT_IO_CLOSE
    /**
     * A callback for closing the streams opened with AVFormatContext.io_open().
     *
     * @deprecated use io_close2
     */
    attribute_deprecated
    void (*io_close)(struct AVFormatContext *s, AVIOContext *pb);
#endif

    /**
     * ',' separated list of disallowed protocols.
     * - encoding: unused
     * - decoding: set by user
     */
    char *protocol_blacklist;

    /**
     * The maximum number of streams.
     * - encoding: unused
     * - decoding: set by user
     */
    int max_streams;

    /**
     * Skip duration calcuation in estimate_timings_from_pts.
     * - encoding: unused
     * - decoding: set by user
     */
    int skip_estimate_duration_from_pts;

    /**
     * Maximum number of packets that can be probed
     * - encoding: unused
     * - decoding: set by user
     */
    int max_probe_packets;

    /**
     * A callback for closing the streams opened with AVFormatContext.io_open().
     *
     * Using this is preferred over io_close, because this can return an error.
     * Therefore this callback is used instead of io_close by the generic
     * libavformat code if io_close is NULL or the default.
     *
     * @param s the format context
     * @param pb IO context to be closed and freed
     * @return 0 on success, a negative AVERROR code on failure
     */
    int (*io_close2)(struct AVFormatContext *s, AVIOContext *pb);
} AVFormatContext;

    最后,在以下代码里完成对AVFormatContext的flags的赋值: 

  • target_obj指针实际便是外围传入的AVFormatContext实例;

  • dst指针所指向的内容即是flags的值,为64,也即#define AVFMT_FLAG_NOBUFFER 0x0040 ///< Do not buffer frames when possible;
  • o变量的类型是AVOption,在此便是fflags选项的实例,o->offset便是flags变量在AVFormatContext中的偏移量;
  • 于是,AVFormatContext的flags便完成了对avformat_open_input的option传参的赋值,由nobuffer转为flags的整形位操作;

    拿到AVFormatContext的实例target_obj,并计算出fflags选项所对应的flags变量在AVFormatContext中的偏移量:

    拿到nobuffer选项AVOption实例:

    并给AVFormatContext中的flags变量赋值,为#define AVFMT_FLAG_NOBUFFER 0x0040,即64:

3.2.2 avformat_find_stream_info 

    最后在avformat_find_stream_info中查找音视频及字幕流时,若使能了#define AVFMT_FLAG_NOBUFFER 0x0040选项,则不缓存而丢包,包括IDR帧: 

int avformat_find_stream_info(AVFormatContext *ic, AVDictionary **options) {

    ......

    ret = read_frame_internal(ic, pkt1);
    if (ret == AVERROR(EAGAIN))
        continue;

    if (ret < 0) {
        /* EOF or error*/
        eof_reached = 1;
        break;
    }

    if (!(ic->flags & AVFMT_FLAG_NOBUFFER)) {
        // 若未使能nobuffer选项,即AVFMT_FLAG_NOBUFFER为假,则缓存在FFFormatContext的packet_buffer中
        ret = avpriv_packet_list_put(&si->packet_buffer,
                                     pkt1, NULL, 0);
        if (ret < 0)
            goto unref_then_goto_end;

        pkt = &si->packet_buffer.tail->pkt;
    } else {
        // 若使能了nobuffer选项,即AVFMT_FLAG_NOBUFFER为真,则丢包不缓存,包括IDR帧
        pkt = pkt1;
    }

    ......
}

3.3.3 av_read_frame 

    而av_read_frame读取AVPacket时:

int av_read_frame(AVFormatContext *s, AVPacket *pkt)
{
    FFFormatContext *const si = ffformatcontext(s);
    const int genpts = s->flags & AVFMT_FLAG_GENPTS;
    int eof = 0;
    int ret;
    AVStream *st;

    if (!genpts) {
        ret = si->packet_buffer.head
              ? avpriv_packet_list_get(&si->packet_buffer, pkt)
              : read_frame_internal(s, pkt);
        if (ret < 0)
            return ret;
        goto return_packet;
    }

    for (;;) {
        // 首先从FFFormatContext的packet_buffer队首读取
        PacketListEntry *pktl = si->packet_buffer.head;

        if (pktl) {
            AVPacket *next_pkt = &pktl->pkt;
            // 此处略去对pts和dts的处理逻辑代码
            ......
        }

        // 若FFFormatContext的packet_buffer为空,再从FFFormatContext的raw_packet_buffer中读取,
        // 若raw_packet_buffer为空,则从AVInputFormat => AVIOContext => URLContext => URLProtocol直接从网络中读取字节流
        ret = read_frame_internal(s, pkt);
        if (ret < 0) {
            if (pktl && ret != AVERROR(EAGAIN)) {
                eof = 1;
                continue;
            } else
                return ret;
        }

        // 最后将读取到的AVPacket包放入到FFFormatContext的packet_buffer队尾
        ret = avpriv_packet_list_put(&si->packet_buffer,
                                     pkt, NULL, 0);
        if (ret < 0) {
            av_packet_unref(pkt);
            return ret;
        }
    }

return_packet:
    st = s->streams[pkt->stream_index];
    if ((s->iformat->flags & AVFMT_GENERIC_INDEX) && pkt->flags & AV_PKT_FLAG_KEY) {
        ff_reduce_index(s, st->index);
        av_add_index_entry(st, pkt->pos, pkt->dts, 0, 0, AVINDEX_KEYFRAME);
    }

    if (is_relative(pkt->dts))
        pkt->dts -= RELATIVE_TS_BASE;
    if (is_relative(pkt->pts))
        pkt->pts -= RELATIVE_TS_BASE;

    return ret;
}

     于是,便出现了后续用av_read_frame读取video的AVPacket时,首个video包不是IDR帧的情况。

4 解决办法

    找到了原因,则解决办法就简单了,通过注释掉nobuffer选项代码即可:

    // ffmpeg的nobuffer选项虽可降低视频的时延,但会增加首开时延,要合理使用!
    //[options setFormatOptionValue:@"nobuffer" forKey:@"fflags"];
  • 26
    点赞
  • 25
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值