【FFmpeg】ffplay 源代码结构图+播放模式分析

ffplay 源代码结构图

写在前面

这篇文章参考了雷神的文章：ffplay.c函数结构简单分析（画图）_雷霄骅(leixiaohua1020)

由于FFmpeg的不断更新，雷神画的结构图与现在版本FFmpeg的代码结构有所不同。为了深入理解FFmpeg，我参照雷神的代码结构图，重新画了一张图。

主要参考的FFmpeg官方代码：FFmpeg: fftools/ffplay.c File Reference

FFmpeg 版本：3.4.9 released on 2021-10-11

对应文档：FFmpeg: Main Page

源代码结构图

注：

1. 在event_loop函数中判断用户输入（键盘、鼠标等），用户输入有很多种，此结构图没有全部画出来。
2. 此结构图忽略了字幕的处理（打开、读入、解码、显示、关闭）

代码逻辑

播放模式分析

在音视频播放时有不同的模式：

enum ShowMode {
    SHOW_MODE_NONE = -1, 
    SHOW_MODE_VIDEO = 0,	// 播放视频画面
    SHOW_MODE_WAVES,		// 不显示视频画面，只显示波点
    SHOW_MODE_RDFT, 		// 不显示视频画面，只黑屏
    SHOW_MODE_NB
} show_mode;

SHOW_MODE_VIDEO

播放画面正常

对应播放的代码：

static void video_image_display(VideoState *is)
{
    Frame *vp;
    Frame *sp = NULL;
    SDL_Rect rect;

    vp = frame_queue_peek_last(&is->pictq);
    
    /*...process code for subtitile...*/

    calculate_display_rect(&rect, is->xleft, is->ytop, is->width, is->height, vp->width, vp->height, vp->sar);

    if (!vp->uploaded) {
        if (upload_texture(&is->vid_texture, vp->frame, &is->img_convert_ctx) < 0)
            return;
        vp->uploaded = 1;
        vp->flip_v = vp->frame->linesize[0] < 0;
    }

    SDL_RenderCopyEx(renderer, is->vid_texture, NULL, &rect, 0, NULL, vp->flip_v ? SDL_FLIP_VERTICAL : 0);

	/*...show subtitile code...*/
    
}

SHOW_MODE_WAVES

播放画面

此画面是由video_audio_display函数生成的。

对应播放的代码：

static void video_audio_display(VideoState *s)
{
    int i, i_start, x, y1, y, ys, delay, n, nb_display_channels;
    int ch, channels, h, h2;
    int64_t time_diff;
    int rdft_bits, nb_freq;

    for (rdft_bits = 1; (1 << rdft_bits) < 2 * s->height; rdft_bits++)
        ;
    nb_freq = 1 << (rdft_bits - 1);

    /* compute display index : center on currently output samples */
    channels = s->audio_tgt.channels;
    nb_display_channels = channels;
    if (!s->paused) {
        int data_used= s->show_mode == SHOW_MODE_WAVES ? s->width : (2*nb_freq);
        n = 2 * channels;
        delay = s->audio_write_buf_size;
        delay /= n;

        /* to be more precise, we take into account the time spent since
           the last buffer computation */
        if (audio_callback_time) {
            time_diff = av_gettime_relative() - audio_callback_time;
            delay -= (time_diff * s->audio_tgt.freq) / 1000000;
        }

        delay += 2 * data_used;
        if (delay < data_used)
            delay = data_used;

        i_start= x = compute_mod(s->sample_array_index - delay * channels, SAMPLE_ARRAY_SIZE);
        if (s->show_mode == SHOW_MODE_WAVES) {
            h = INT_MIN;
            for (i = 0; i < 1000; i += channels) {
                int idx = (SAMPLE_ARRAY_SIZE + x - i) % SAMPLE_ARRAY_SIZE;
                int a = s->sample_array[idx];
                int b = s->sample_array[(idx + 4 * channels) % SAMPLE_ARRAY_SIZE];
                int c = s->sample_array[(idx + 5 * channels) % SAMPLE_ARRAY_SIZE];
                int d = s->sample_array[(idx + 9 * channels) % SAMPLE_ARRAY_SIZE];
                int score = a - d;
                if (h < score && (b ^ c) < 0) {
                    h = score;
                    i_start = idx;
                }
            }
        }

        s->last_i_start = i_start;
    } else {
        i_start = s->last_i_start;
    }

    if (s->show_mode == SHOW_MODE_WAVES) {
        SDL_SetRenderDrawColor(renderer, 255, 255, 255, 255);

        /* total height for one channel */
        h = s->height / nb_display_channels;
        /* graph height / 2 */
        h2 = (h * 9) / 20;
        for (ch = 0; ch < nb_display_channels; ch++) {
            i = i_start + ch;
            y1 = s->ytop + ch * h + (h / 2); /* position of center line */
            for (x = 0; x < s->width; x++) {
                y = (s->sample_array[i] * h2) >> 15;
                if (y < 0) {
                    y = -y;
                    ys = y1 - y;
                } else {
                    ys = y1;
                }
                fill_rectangle(s->xleft + x, ys, 1, y);
                i += channels;
                if (i >= SAMPLE_ARRAY_SIZE)
                    i -= SAMPLE_ARRAY_SIZE;
            }
        }

        SDL_SetRenderDrawColor(renderer, 0, 0, 255, 255);

        for (ch = 1; ch < nb_display_channels; ch++) {
            y = s->ytop + ch * h;
            fill_rectangle(s->xleft, y, s->width, 1);
        }
    }
}

SHOW_MODE_RDFT

播放画面

此画面是由video_audio_display函数生成的。

对应播放的代码：

static void video_audio_display(VideoState *s)
{
    int i, i_start, x, y1, y, ys, delay, n, nb_display_channels;
    int ch, channels, h, h2;
    int64_t time_diff;
    int rdft_bits, nb_freq;

    for (rdft_bits = 1; (1 << rdft_bits) < 2 * s->height; rdft_bits++)
        ;
    nb_freq = 1 << (rdft_bits - 1);

    /* compute display index : center on currently output samples */
    channels = s->audio_tgt.channels;
    nb_display_channels = channels;
    if (!s->paused) {
        int data_used= s->show_mode == SHOW_MODE_WAVES ? s->width : (2*nb_freq);
        n = 2 * channels;
        delay = s->audio_write_buf_size;
        delay /= n;

        /* to be more precise, we take into account the time spent since
           the last buffer computation */
        if (audio_callback_time) {
            time_diff = av_gettime_relative() - audio_callback_time;
            delay -= (time_diff * s->audio_tgt.freq) / 1000000;
        }

        delay += 2 * data_used;
        if (delay < data_used)
            delay = data_used;

        i_start= x = compute_mod(s->sample_array_index - delay * channels, SAMPLE_ARRAY_SIZE);
        if (s->show_mode == SHOW_MODE_WAVES) {
            h = INT_MIN;
            for (i = 0; i < 1000; i += channels) {
                int idx = (SAMPLE_ARRAY_SIZE + x - i) % SAMPLE_ARRAY_SIZE;
                int a = s->sample_array[idx];
                int b = s->sample_array[(idx + 4 * channels) % SAMPLE_ARRAY_SIZE];
                int c = s->sample_array[(idx + 5 * channels) % SAMPLE_ARRAY_SIZE];
                int d = s->sample_array[(idx + 9 * channels) % SAMPLE_ARRAY_SIZE];
                int score = a - d;
                if (h < score && (b ^ c) < 0) {
                    h = score;
                    i_start = idx;
                }
            }
        }

        s->last_i_start = i_start;
    } else {
        i_start = s->last_i_start;
    }

    if (s->show_mode == SHOW_MODE_WAVES) {
        /* process for SHOW_MODE_WAVES*/
    } else {
        if (realloc_texture(&s->vis_texture, SDL_PIXELFORMAT_ARGB8888, s->width, s->height, SDL_BLENDMODE_NONE, 1) < 0)
            return;

        nb_display_channels= FFMIN(nb_display_channels, 2);
        if (rdft_bits != s->rdft_bits) {
            av_rdft_end(s->rdft);
            av_free(s->rdft_data);
            s->rdft = av_rdft_init(rdft_bits, DFT_R2C);
            s->rdft_bits = rdft_bits;
            s->rdft_data = av_malloc_array(nb_freq, 4 *sizeof(*s->rdft_data));
        }
        if (!s->rdft || !s->rdft_data){
            av_log(NULL, AV_LOG_ERROR, "Failed to allocate buffers for RDFT, switching to waves display\n");
            s->show_mode = SHOW_MODE_WAVES;
        } else {
            FFTSample *data[2];
            SDL_Rect rect = {.x = s->xpos, .y = 0, .w = 1, .h = s->height};
            uint32_t *pixels;
            int pitch;
            for (ch = 0; ch < nb_display_channels; ch++) {
                data[ch] = s->rdft_data + 2 * nb_freq * ch;
                i = i_start + ch;
                for (x = 0; x < 2 * nb_freq; x++) {
                    double w = (x-nb_freq) * (1.0 / nb_freq);
                    data[ch][x] = s->sample_array[i] * (1.0 - w * w);
                    i += channels;
                    if (i >= SAMPLE_ARRAY_SIZE)
                        i -= SAMPLE_ARRAY_SIZE;
                }
                av_rdft_calc(s->rdft, data[ch]);
            }
            /* Least efficient way to do this, we should of course
             * directly access it but it is more than fast enough. */
            if (!SDL_LockTexture(s->vis_texture, &rect, (void **)&pixels, &pitch)) {
                pitch >>= 2;
                pixels += pitch * s->height;
                for (y = 0; y < s->height; y++) {
                    double w = 1 / sqrt(nb_freq);
                    int a = sqrt(w * sqrt(data[0][2 * y + 0] * data[0][2 * y + 0] + data[0][2 * y + 1] * data[0][2 * y + 1]));
                    int b = (nb_display_channels == 2 ) ? sqrt(w * hypot(data[1][2 * y + 0], data[1][2 * y + 1]))
                                                        : a;
                    a = FFMIN(a, 255);
                    b = FFMIN(b, 255);
                    pixels -= pitch;
                    *pixels = (a << 16) + (b << 8) + ((a+b) >> 1);
                }
                SDL_UnlockTexture(s->vis_texture);
            }
            SDL_RenderCopy(renderer, s->vis_texture, NULL, NULL);
        }
        if (!s->paused)
            s->xpos++;
        if (s->xpos >= s->width)
            s->xpos= s->xleft;
    }
}

之后分析ffplay的音视频同步。