博客
关于我
强烈建议你试试无所不能的chatGPT,快点击我
FFmpeg-4.0 的filter机制的架构与实现.之三 Filter实现的源码分析
阅读量:2717 次
发布时间:2019-05-13

本文共 16686 字,大约阅读时间需要 55 分钟。

五、Filter实现的源码分析

5.1 滤镜的回调函数的调用流程,以单滤镜设置(如 -vf "delogo")为例

init()

query_format(); // 输入输出的格式查询: 列出滤镜支持的格式列表

config_input();

config_output();

for (;;) {

request_frame();

filter_frame();

}

uninit();

 

 

5.2 函数的调用堆栈

STEP1:  init();                           // 滤镜初始化    : 用于初始化滤镜#0  init (ctx=0x24dbe40) at libavfilter/vf_delogo.c:196#1  0x00000000004e9ad5 in avfilter_init_str (filter=0x24dbe40, args=0x24dbdc0 "x=10:y=10:w=80:h=20") at libavfilter/avfilter.c:1018#2  0x00000000004fb54a in create_filter (filt_ctx=0x7fffffffce20, ctx=0x24da800, index=
, name=
, args=0x24dbdc0 "x=10:y=10:w=80:h=20", log_ctx=0x24da800) at libavfilter/graphparser.c:149#3 0x00000000004fbee2 in parse_filter (graph=0x24da800, filters=0x23a5488 "", inputs=0x7fffffffd0b8, outputs=0x7fffffffd0b0) at libavfilter/graphparser.c:192#4 avfilter_graph_parse2 (graph=0x24da800, filters=0x23a5488 "", inputs=0x7fffffffd0b8, outputs=0x7fffffffd0b0) at libavfilter/graphparser.c:427#5 0x00000000004bc528 in configure_filtergraph (fg=0x23a5500) at fftools/ffmpeg_filter.c:1047#6 0x00000000004cb4ba in ifilter_send_frame (ifilter=0x23aabc0, frame=0x24b2480) at fftools/ffmpeg.c:2173#7 0x00000000004cb9b6 in send_frame_to_filters (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2254#8 decode_video (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2455#9 0x00000000004cc1ad in process_input_packet (ist=0x23adfc0, pkt=
, no_eof=0) at fftools/ffmpeg.c:2609#10 0x00000000004ce659 in process_input () at fftools/ffmpeg.c:4441#11 transcode_step () at fftools/ffmpeg.c:4561#12 0x00000000004d0e65 in transcode (argc=
, argv=
) at fftools/ffmpeg.c:4615#13 main (argc=
, argv=
) at fftools/ffmpeg.c:4847preinit(); // 滤镜预初始化: 用于分配空间和初始化子对象STEP2:  query_format(); // 输入输出的格式查询: 列出滤镜支持的格式列表#0 query_formats (ctx=0x24dbe40) at libavfilter/vf_delogo.c:188#1 0x00000000004ec8e6 in filter_query_formats (ctx=0x24dbe40) at libavfilter/avfiltergraph.c:326#2 0x00000000004ed014 in query_formats (graph=0x24da800, log_ctx=0x0) at libavfilter/avfiltergraph.c:454#3 0x00000000004edc51 in graph_config_formats (graphctx=0x24da800, log_ctx=0x0) at libavfilter/avfiltergraph.c:1163#4 avfilter_graph_config (graphctx=0x24da800, log_ctx=0x0) at libavfilter/avfiltergraph.c:1274#5 0x00000000004bcabb in configure_filtergraph (fg=0x23a5500) at fftools/ffmpeg_filter.c:1100#6 0x00000000004cb4ba in ifilter_send_frame (ifilter=0x23aabc0, frame=0x24b2480) at fftools/ffmpeg.c:2173#7 0x00000000004cb9b6 in send_frame_to_filters (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2254#8 decode_video (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2455#9 0x00000000004cc1ad in process_input_packet (ist=0x23adfc0, pkt=
, no_eof=0) at fftools/ffmpeg.c:2609#10 0x00000000004ce659 in process_input () at fftools/ffmpeg.c:4441#11 transcode_step () at fftools/ffmpeg.c:4561#12 0x00000000004d0e65 in transcode (argc=
, argv=
) at fftools/ffmpeg.c:4615#13 main (argc=
, argv=
) at fftools/ffmpeg.c:4847STEP3: 配置滤镜的输入、输出格式和参数   config_input();      config_output();#0 config_input (inlink=0x24dd980) at libavfilter/vf_delogo.c:231#1 0x00000000004ea930 in avfilter_config_links (filter=0x24dbe40) at libavfilter/avfilter.c:369#2 0x00000000004ea8b0 in avfilter_config_links (filter=0x24da900) at libavfilter/avfilter.c:307#3 0x00000000004ea8b0 in avfilter_config_links (filter=0x24dec40) at libavfilter/avfilter.c:307#4 0x00000000004ee922 in graph_config_links (graphctx=0x24da800, log_ctx=0x0) at libavfilter/avfiltergraph.c:261#5 avfilter_graph_config (graphctx=0x24da800, log_ctx=0x0) at libavfilter/avfiltergraph.c:1276#6 0x00000000004bcabb in configure_filtergraph (fg=0x23a5500) at fftools/ffmpeg_filter.c:1100#7 0x00000000004cb4ba in ifilter_send_frame (ifilter=0x23aabc0, frame=0x24b2480) at fftools/ffmpeg.c:2173#8 0x00000000004cb9b6 in send_frame_to_filters (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2254#9 decode_video (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2455#10 0x00000000004cc1ad in process_input_packet (ist=0x23adfc0, pkt=
, no_eof=0) at fftools/ffmpeg.c:2609#11 0x00000000004ce659 in process_input () at fftools/ffmpeg.c:4441#12 transcode_step () at fftools/ffmpeg.c:4561#13 0x00000000004d0e65 in transcode (argc=
, argv=
) at fftools/ffmpeg.c:4615#14 main (argc=
, argv=
) at fftools/ffmpeg.c:4847STEP4: 滤镜体对每一帧数据进行处理   filter_frame();#0 filter_frame (inlink=0x24dd980, in=0x24de480) at libavfilter/vf_delogo.c:245#1 0x00000000004eb92e in ff_filter_frame_framed (filter=
) at libavfilter/avfilter.c:1071#2 ff_filter_frame_to_filter (filter=
) at libavfilter/avfilter.c:1219#3 ff_filter_activate_default (filter=
) at libavfilter/avfilter.c:1268#4 ff_filter_activate (filter=
) at libavfilter/avfilter.c:1430#5 0x00000000004efad0 in push_frame (ctx=
, frame=0x24b2480, flags=
) at libavfilter/buffersrc.c:181#6 av_buffersrc_add_frame_internal (ctx=
, frame=0x24b2480, flags=
) at libavfilter/buffersrc.c:255#7 0x00000000004efc2f in av_buffersrc_add_frame_flags (ctx=0x24ddcc0, frame=
, flags=
) at libavfilter/buffersrc.c:164#8 0x00000000004cb39d in ifilter_send_frame (ifilter=0x23aabc0, frame=0x24b2480) at fftools/ffmpeg.c:2180#9 0x00000000004cb9b6 in send_frame_to_filters (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2254#10 decode_video (ist=0x23adfc0, pkt=
, got_output=
, duration_pts=0x7fffffffd570, eof=
, decode_failed=0x7fffffffd578) at fftools/ffmpeg.c:2455#11 0x00000000004cc1ad in process_input_packet (ist=0x23adfc0, pkt=
, no_eof=0) at fftools/ffmpeg.c:2609#12 0x00000000004ce659 in process_input () at fftools/ffmpeg.c:4441#13 transcode_step () at fftools/ffmpeg.c:4561#14 0x00000000004d0e65 in transcode (argc=
, argv=
) at fftools/ffmpeg.c:4615#15 main (argc=
, argv=
) at fftools/ffmpeg.c:4847STEP5:  uninit()

5.3 filter_frame()调用流程

5.3.1. decode_video //ffmpeg.c最初的源头,是ffmpeg.c的decode_video函数。 核心代码如下所示:static int decode_video(InputStream *ist, AVPacket *pkt, int *got_output){    AVFrame* decoded_frame, f;    //解码    ret = avcodec_decode_video2(ist->dec_ctx,                                decoded_frame, got_output, pkt);    //......    //将解码后的帧数据  送给滤镜    for (i = 0; i < ist->nb_filters; i++) {        f = decodec_frame;        ret = av_buffersrc_add_frame_flags(ist->filters[i]->filter, f, AV_BUFFERSRC_FLAG_PUSH);    }}可见,最重要做2件事,一个解码,一个送给滤镜。 送给哪个滤镜呢?InputStream *ist的nb_filters为1,其实就是指向名字为“buffer”的filter(源文件:buffersrc.c)。这个filter与其他filter不同的是,它是所有filter的第一个入口。解码完,都先给它,它再传递给下一个。为啥先给他呢?很简单,它是一个FIFO,缓存数据用的。5.3.2. av_buffersrc_add_frame_flags//buffersrc.c该函数直接走到av_buffersrc_add_frame_internal //buffersrc.c5.3.3. av_buffersrc_add_frame_internal //buffersrc.cstatic int av_buffersrc_add_frame_internal(AVFilterContext *ctx,                                           AVFrame *frame, int flags){    AVFrame *copy = NULL;    av_frame_ref(copy, frame);    // 将解码后的帧数据 frame 写FIFO    av_fifo_generic_write(s->fifo, ©, sizeof(copy), NULL);    //     if ((flags & AV_BUFFERSRC_FLAG_PUSH))        if ((ret = ctx->output_pads[0].request_frame(ctx->outputs[0])) < 0)            return ret;    return 0;}可见, 该函数把frame写到FIFO,然后调了自己的output_pads[0]的request_frame。5.3.4. request_frame //buffersrc.cstatic int request_frame(AVFilterLink *link){    BufferSourceContext *c = link->src->priv;    AVFrame *frame;    int ret;    //省略......    // 从FIFO中读取一帧数据     av_fifo_generic_read(c->fifo, &frame, sizeof(frame), NULL);    av_log(NULL, AV_LOG_WARNING, "request_frame, frame-pts %lld \n", frame->pts);    //这个link,是第一个link,链接当前的AVFilterContext和下一个AVFilterContext,也就用户设置的真正要用的滤镜代码    ret = ff_filter_frame(link, frame);    return ret;}可见它从FIFO读取一帧数据。然后调用ff_filter_frame。此时输入的link是第一个AVFilterLink。5.3.5. ff_filter_frame // avfilter.c该函数做了一些基本检查,走到ff_filter_frame_framed5.3.6. ff_filter_frame_framed //avfilter.c{    //定义一个函数指针filter_frame。所指向的函数,参数为AVFilterLink *, AVFrame *,返回值为int    int (*filter_frame)(AVFilterLink *, AVFrame *);    //下一个AVFilterContext,对本例来说,就是我们自己写的transform 滤镜,源码在vf_transform.c    AVFilterContext *dstctx = link->dst;       AVFilterPad *dst = link->dstpad;    AVFrame *out = NULL;    int ret;    //函数指针filter_frame,link->dstpad其实就是dstctx->input_pads,也就是transform滤镜定义的    if (!(filter_frame = dst->filter_frame))          filter_frame = default_filter_frame;    //省略300字    ret = filter_frame(link, out);    link->frame_count++;    ff_update_link_current_pts(link, pts);    return ret;}定义一个函数指针filter_frame。所指向的函数,必须是参数为(AVFilterLink , AVFrame ) ,返回值为int filter_frame = dst->filter_frame dst = link->dstpad,而link->dstpad就是transform过滤器定义的input_padsstatic const AVFilterPad avfilter_vf_transform_inputs[] = {    {        .name         = "default",        .type         = AVMEDIA_TYPE_VIDEO,        .filter_frame = filter_frame,    },    { NULL }};所以,filter_frame函数指针,指向的就是vf_transform.c实现的filter_frame函数。5.3.7. filter_frame //vf_transform.c,当然啦,ffmpeg定义的各种filter,比如vf_colorbalance.c,vf_scale.c等,也有这个函数,流程一样的static int filter_frame(AVFilterLink *link, AVFrame *in){    AVFilterContext *avctx = link->dst;             //第一个link的dst AVFilterContext,其实就是当前的filter的AVFilterContext    AVFilterLink *outlink = avctx->outputs[0];   //当前的AVFilterContext,outputs[0]指向第二个AVFilterLink    AVFrame *out;    //分配一个空的AVFrame。    out = ff_get_video_buffer(outlink, outlink->w, outlink->h);    if (!out) {        av_frame_free(&in);        return AVERROR(ENOMEM);    }    //分配的空buffer的参数和上一个基本一致,但修改宽高。当然啦,如果你愿意,不修改宽高,那就不需要下面2句。    av_frame_copy_props(out, in);    out->width  = outlink->w;    out->height = outlink->h;    out->format = outlink->format;    ThreadData td;    td.in = in;    td.out = out;    int res;    if(res = avctx->internal->execute(avctx, do_conversion, &td, NULL, FFMIN(outlink->h, avctx->graph->nb_threads))) {        return res;    }//启用一个子线程,执行比较耗时的变换。do_conversion是我们要做的变换。    av_frame_free(&in);        // 此时的ff_filter_frame,输入参数和前面buffersrc.c调用的已经不一样。outlink是第二个AVFilterLink,buffer也是做了变换的新的buffer    return ff_filter_frame(outlink, out);  }通过ff_get_video_buffer,分配一个空buffer,该buffer用于存储变换的结果,并会通过ff_filter_frame传递到下一个filter。 do_conversion是一个真正做变换的函数,但其实如果要做的处理并不耗时,也不一定要用另一个线程来处理。直接在该filter_frame做也行。 处理好的新的数据,放在out,调用ff_filter_frame,传递给下一个filter。注意,ff_filter_frame的oulink,对应上图的第二个AVFilterLink。5.3.8. 再次走进ff_filter_frame // avfilter.c如上已知,ff_filter_frame只做了一些基本检查,走到ff_filter_frame_framed。故而我们直接看ff_filter_frame_framedstatic int ff_filter_frame_framed(AVFilterLink *link, AVFrame *frame){    //定义一个函数指针filter_frame。所指向的函数,参数为AVFilterLink *, AVFrame *,返回值为int    int (*filter_frame)(AVFilterLink *, AVFrame *);     //下一个AVFilterContext,对本例来说,就是系统默认的第三个滤镜,名字叫"format",源码在vf_format.c    AVFilterContext *dstctx = link->dst;    AVFilterPad *dst = link->dstpad;    AVFrame *out = NULL;    int ret;    if (!(filter_frame = dst->filter_frame))     //vf_format.c没有实现filter函数,因为返回为空        filter_frame = default_filter_frame;    //所以函数会走到这,函数指针filter_frame 将指向default_filter_frame    //省略300字    ret = filter_frame(link, out);    link->frame_count++;    ff_update_link_current_pts(link, pts);    return ret;}如注释所说,由于vf_format.c没有实现filter函数,所以此时的filter_frame指针,指向的是defalut_filter_frame。5.3.9. default_filter_frame //avfilter.cstatic int default_filter_frame(AVFilterLink *link, AVFrame *frame){    //该函数没干啥,又调用ff_filter_frame了,第一个参数,换成第三个AVFilterLink了,第二个参数不变,frame默默的传递出去    return ff_filter_frame(link->dst->outputs[0], frame);}此时link->dst->outputs[0]对应上图第三个AVFilterLink。5.3.10. 第三次走进ff_filter_frame // avfilter.cstatic int ff_filter_frame_framed(AVFilterLink *link, AVFrame *frame){    //定义一个函数指针filter_frame。所指向的函数,参数为AVFilterLink *, AVFrame *,返回值为int    int (*filter_frame)(AVFilterLink *, AVFrame *);    //下一个AVFilterContext,对本例来说,就是系统默认的最后一个滤镜,名字叫"buffersink",源码在bufffersink.c    AVFilterContext *dstctx = link->dst;    AVFilterPad *dst = link->dstpad;    AVFrame *out = NULL;    int ret;    if (!(filter_frame = dst->filter_frame))   //指向buffersink.c实现的filte_frame函数        filter_frame = default_filter_frame;    //省略300字    ret = filter_frame(link, out);    link->frame_count++;    ff_update_link_current_pts(link, pts);    return ret;}此时的filter_frame指针,指向buffersink.c实现的filter_frame函数5.3.11. filter_frame //buffersink.cstatic int filter_frame(AVFilterLink *link, AVFrame *frame){    AVFilterContext *ctx = link->dst;    BufferSinkContext *buf = link->dst->priv;    int ret;    if ((ret = add_buffer_ref(ctx, frame)) < 0)        return ret;    //省略300字    return 0;}static int add_buffer_ref(AVFilterContext *ctx, AVFrame *ref){    BufferSinkContext *buf = ctx->priv;    /* cache frame */    //把buffer存到FIFO    av_fifo_generic_write(buf->fifo, &ref, FIFO_INIT_ELEMENT_SIZE, NULL);    return 0;}很清晰的看到,其实就是把buffer存到FIFO。至此,把filter_frame的来龙去脉搞清楚啦

 

5.4 filter之后,ffmpeg如何编码

当我们写了一个filter,把视频做处理后,ffmpeg是如何把它编码的呢?

通过研究,发现编码的源头函数是reap_filters(…),它会被transcode_step(…)函数调用。

5.4.1. reap_filters //ffmpeg.cstatic int reap_filters(int flush){    AVFrame *filtered_frame = NULL;//该指针将存储一个经过滤镜处理后的buffer,并送给encoder    int i;    /* Reap all buffers present in the buffer sinks */    for (i = 0; i < nb_output_streams; i++) {//一路video,一路audio,那么nb_output_streams = 2        OutputStream *ost = output_streams[i];        OutputFile    *of = output_files[ost->file_index];        AVFilterContext *filter;        AVCodecContext *enc = ost->enc_ctx;        int ret = 0;        if (!ost->filter)            continue;        filter = ost->filter->filter;//OutputStream的filter指针指向buffersink.c定义的AVFilterContext。也就是本文讨论的,最后一个AVFilterContext        if (!ost->filtered_frame && !(ost->filtered_frame = av_frame_alloc())) {            return AVERROR(ENOMEM);        }        filtered_frame = ost->filtered_frame;        while (1) {            double float_pts = AV_NOPTS_VALUE; // this is identical to filtered_frame.pts but with higher precision            //av_buffersink_get_frame_flags定义在buffersink.c,用于从FIFO读出一帧            ret = av_buffersink_get_frame_flags(filter, filtered_frame,                                               AV_BUFFERSINK_FLAG_NO_REQUEST);            if (ret < 0) {                //省略,检查ret                //如果ret<0,不是别的错误,那认为还没有数据,跳出循环                break;            }            switch (filter->inputs[0]->type) {            case AVMEDIA_TYPE_VIDEO:                //do_video_out函数将会做video编码                do_video_out(of->ctx, ost, filtered_frame, float_pts);                break;            case AVMEDIA_TYPE_AUDIO:                //do_audio_out函数将会做audioo编码                do_audio_out(of->ctx, ost, filtered_frame);                break;            default:                // TODO support subtitle filters                av_assert0(0);            }            av_frame_unref(filtered_frame);        }    }    return 0;}前一节说了,filter_frame(…)的最终结果是,把buffer存在了buffersink.c的FIFO里。 那么,这一节,说的其实就是一个从buffersink的FIFO读数据,并编码的过程。 从上面可知,av_buffersink_get_frame_flags函数,从buffersink读取一帧数据,放到filtered_frame。5.4.2. do_video_out //ffmpeg.cstatic void do_video_out(AVFormatContext *s,                         OutputStream *ost,                         AVFrame *next_picture,                         double sync_ipts){    int ret;    AVCodecContext *enc = ost->enc_ctx;    int nb_frames, nb0_frames, i;    //省略300字    for (i = 0; i < nb_frames; i++) {        AVFrame *in_picture;        if (i < nb0_frames && ost->last_frame) {            in_picture = ost->last_frame;        } else            in_picture = next_picture;        //省略300字        ost->frames_encoded++;        //开始编码        ret = avcodec_encode_video2(enc, &pkt, in_picture, &got_packet);    }    ///省略300字}该函数很长,做了很多杂事,但关键代码就是调用编码函数avcodec_encode_video2

5.5 函数流程图

下面给出ffmpeg使用filter时的函数流程图,主要把和filter相关的函数拉出来!

 

图7: ffmpeg带filter的流程图

 

ffmpeg常规的avcodec_register_all(…), avfilter_register_all(…), av_register_all(…)等函数用于注册各种资源

 

transcode_init() : 主要用于初始化前文提到的各种结构体。

transcode_step : 主要工作: 解码->送filter过滤->编码->继续解码….

 

choose_output() : 用于选择一个OutputStream。比如有一个audio,一个video,那要根据pts策略,比如谁的pts比较小,就挑哪个OutputStream先干活。

transcode_frome_filter() : 用于选个一个InputStream,用于下一步的process_input()。

process_input() : 主要是解码,并把解码的buffer送往filter处理。

reap_filters() : 主要是,从filter的FIFO拿出buffer,并编码。

 

 

6. 参考资料

FFmpeg官网: http://www.ffmpeg.org

FFmpeg doc : http://www.ffmpeg.org/documentation.html

FFmpeg wiki : https://trac.ffmpeg.org/wiki

CSDN大牛:

 

 

转载地址:http://knctd.baihongyu.com/

你可能感兴趣的文章
CGRectGet方法解释
查看>>
关于textField输入光标颜色及cleanButton大小和颜色的设置
查看>>
java连接Mysql时需要注意的问题
查看>>
struts2注解总结----@Action和@Result
查看>>
struts2 验证框架原理及实例
查看>>
Struts2 异常处理
查看>>
ecplise 使用 git
查看>>
Android开发之如何保证Service不被杀掉(broadcast+system/app)
查看>>
CANoe下载地址
查看>>
函数返回局部变量的几种情况
查看>>
(java)十进制转十六进制调用函数算法
查看>>
java蓝桥杯试题特殊文字
查看>>
java蓝桥杯杨辉三角
查看>>
centos 搭建基本环境 jdk+mysql
查看>>
docker 删除 image 踩到的坑
查看>>
idea提示参数设置
查看>>
Kafka Kafka-Connect Debezium 目标:同步Mysql
查看>>
AIX下安装proFTPD-支持虚拟用户和SFTP
查看>>
Jmeter性能测试内部分享
查看>>
Mysql日常使用维护命令总结
查看>>