video-pc/libs/AVPlayer/av_player.cpp

#include "av_player.h"
#include <QDebug>
#include <QImage>
#include <QThread>
#include "threadpool.h"
#include "vframe.h"
#include "low_latency_config.h"

// 同步阈值定义已移至配置文件（low_latency_config.h）

//单帧视频时长阈值上限，用于适配低帧时同步，
//帧率过低视频帧超前不适合翻倍延迟，应特殊
//处理，这里设置上限一秒10帧

//同步操作摆烂阈值上限，此时同步已无意义


AVPlayer::AVPlayer()
    : m_decoder(new Decoder)
    , m_fmtCtx(nullptr)
    , m_audioFrame(av_frame_alloc())
    , m_imageWidth(300)
    , m_imageHeight(300)
    , m_swrCtx(nullptr)
    , m_swsCtx(nullptr)
    , m_buffer(nullptr)
    , m_audioBuf(nullptr)
    , m_duration(0)
    , m_volume(30)
    , m_exit(0)
    , m_pause(0)
    , m_playSpeed(1.0)
{
    m_sonicStream = nullptr;
}

AVPlayer::~AVPlayer()
{
    av_frame_free(&m_audioFrame);
    clearPlayer();
    delete m_decoder;
    if (m_swrCtx)
        swr_free(&m_swrCtx);
    if (m_swsCtx)
        sws_freeContext(m_swsCtx);
    if (m_audioBuf)
        av_free(m_audioBuf);
    if (m_buffer)
        av_free(m_buffer);
}

int AVPlayer::play(const QString& url)
{
    clearPlayer();
    if (!m_decoder->decode(url)) {
        qDebug() << "decode failed";
        return 0;
    }

    //解码成功可获取流时长
    m_duration = m_decoder->avDuration();
    emit AVDurationChanged(m_duration);

    m_pause = 0;
    m_clockInitFlag = -1;

    // 判断是否存在音/视频流
    m_audioIndex = m_decoder->audioIndex();
    m_videoIndex = m_decoder->videoIndex();
    m_hasAudio = (m_audioIndex >= 0);
    m_hasVideo = (m_videoIndex >= 0);

    bool ok = false;

    if (m_hasAudio) {
        if (initSDL()) {
            ok = true;
        } else {
            qDebug() << "init sdl failed!";
        }
    }

    if (m_hasVideo) {
        if (initVideo()) {
            ok = true;
        } else {
            qDebug() << "init video failed!";
        }
    }

    // 仅音频时，主动初始化时钟
    if (!m_hasVideo && m_hasAudio) {
        initAVClock();
    }

    return ok ? 1 : 0;
}

void fillAStreamCallback(void* userdata, uint8_t* stream, int len)
{
    memset(stream, 0, len);
    AVPlayer* is = (AVPlayer*) userdata;
    static double audioPts = 0.00;
    while (len > 0) {
        if (is->m_exit)
            return;
        if (is->m_audioBufIndex >= is->m_audioBufSize) { /*index到缓冲区末尾,重新填充数据*/
            int ret = is->m_decoder->getAFrame(is->m_audioFrame);
            if (ret) {
                is->m_audioBufIndex = 0;
                if ((is->m_targetSampleFmt != is->m_audioFrame->format
                     || is->m_targetChannelLayout != (int64_t)ffmpeg_get_frame_channel_layout(is->m_audioFrame)
                     || is->m_targetFreq != is->m_audioFrame->sample_rate
                     || is->m_targetNbSamples != is->m_audioFrame->nb_samples)
                    && !is->m_swrCtx) {
                    is->m_swrCtx = ffmpeg_swr_alloc_set_opts(nullptr,
                                                      is->m_targetChannelLayout,
                                                      is->m_targetSampleFmt,
                                                      is->m_targetFreq,
                                                      (int64_t)ffmpeg_get_frame_channel_layout(is->m_audioFrame),
                                                      (enum AVSampleFormat) is->m_audioFrame->format,
                                                      is->m_audioFrame->sample_rate,
                                                      0,
                                                      nullptr);
                    if (!is->m_swrCtx || swr_init(is->m_swrCtx) < 0) {
                        qDebug() << "swr_init failed";
                        return;
                    }
                }
                if (is->m_swrCtx) {
                    const uint8_t** in = (const uint8_t**) is->m_audioFrame->extended_data;
                    int out_count = (uint64_t) is->m_audioFrame->nb_samples * is->m_targetFreq
                                        / is->m_audioFrame->sample_rate
                                    + 256;
                    int out_size = ffmpeg_get_buffer_size(is->m_targetSampleFmt,
                                                              is->m_targetChannels,
                                                              out_count,
                                                              0);
                    if (out_size < 0) {
                        qDebug() << "av_samples_get_buffer_size failed";
                        return;
                    }
                    av_fast_malloc(&is->m_audioBuf, &is->m_audioBufSize, out_size);
                    if (!is->m_audioBuf) {
                        qDebug() << "av_fast_malloc failed";
                        return;
                    }
                    int len2 = swr_convert(is->m_swrCtx,
                                           &is->m_audioBuf,
                                           out_count,
                                           in,
                                           is->m_audioFrame->nb_samples);
                    if (len2 < 0) {
                        qDebug() << "swr_convert failed";
                        return;
                    }
                    if (is->m_playSpeed != 1.0) {
                        sonicSetSpeed(is->m_sonicStream, is->m_playSpeed);
                        int ret = sonicWriteShortToStream(is->m_sonicStream,
                                                          (short*) is->m_audioBuf,
                                                          len2);
                        int availSamples = sonicSamplesAvailable(is->m_sonicStream);
                        if (!availSamples) {
                            is->m_audioBufSize = is->m_audioBufIndex;
                            continue;
                        }
                        int numSamples = availSamples;
                        int bytes = numSamples * is->m_targetChannels
                                    * av_get_bytes_per_sample(is->m_targetSampleFmt);
                        if (bytes > out_size) {
                            av_fast_malloc(&is->m_audioBuf, &is->m_audioBufSize, bytes);
                        }
                        len2 = sonicReadShortFromStream(is->m_sonicStream,
                                                        (short*) is->m_audioBuf,
                                                        numSamples);
                    }
                    is->m_audioBufSize = len2 * is->m_targetChannels
                                         * av_get_bytes_per_sample(is->m_targetSampleFmt);
                } else {
                    is->m_audioBufSize = ffmpeg_get_buffer_size(is->m_targetSampleFmt,
                                                                    is->m_targetChannels,
                                                                    is->m_audioFrame->nb_samples,
                                                                    0);
                    av_fast_malloc(&is->m_audioBuf, &is->m_audioBufSize, is->m_audioBufSize + 256);
                    if (!is->m_audioBuf) {
                        qDebug() << "av_fast_malloc failed";
                        return;
                    }
                    memcpy(is->m_audioBuf, is->m_audioFrame->data[0], is->m_audioBufSize);
                }
                audioPts = is->m_audioFrame->pts
                           * av_q2d(is->m_fmtCtx->streams[is->m_audioIndex]->time_base);
                //qDebug()<<is->m_audioPts;
                av_frame_unref(is->m_audioFrame);
            } else {
                //判断是否真正播放到文件末尾
                if (is->m_decoder->isExit()) {
                    emit is->AVTerminate();
                }
                return;
            }
        }
        int len1 = is->m_audioBufSize - is->m_audioBufIndex;
        len1 = (len1 > len ? len : len1);
        SDL_MixAudio(stream, is->m_audioBuf + is->m_audioBufIndex, len1, is->m_volume);
        len -= len1;
        is->m_audioBufIndex += len1;
        stream += len1;
    }
    //记录音频时钟
    is->m_audioClock.setClock(audioPts);
    //发送时间戳变化信号,因为进度以整数秒单位变化展示，
    //所以大于一秒才发送，避免过于频繁的信号槽通信消耗性能
    uint32_t _pts = (uint32_t) audioPts;
    if (is->m_lastAudPts != _pts) {
        emit is->AVPtsChanged(_pts);
        is->m_lastAudPts = _pts;
    }
}

int AVPlayer::initSDL()
{
    // 无音频流直接返回失败，调用方根据 m_hasAudio 控制
    if (m_decoder->audioIndex() < 0)
        return 0;

    if (SDL_Init(SDL_INIT_AUDIO) != 0) {
        qDebug() << "SDL_Init failed";
        return 0;
    }

    m_exit = 0;

    m_audioBufSize = 0;
    m_audioBufIndex = 0;

    m_lastAudPts = -1;

    m_audioCodecPar = m_decoder->audioCodecPar();
    m_audioIndex = m_decoder->audioIndex();
    m_fmtCtx = m_decoder->formatContext();

    // 音频设备配置（平衡模式 - 类似VLC）
    SDL_AudioSpec wanted_spec;
    // wanted_spec.channels = m_audioCodecPar->channels;
    wanted_spec.channels = ffmpeg_get_codec_channels(m_fmtCtx->streams[m_audioIndex]);
    wanted_spec.freq = m_audioCodecPar->sample_rate;
    wanted_spec.format = AUDIO_S16SYS;
    wanted_spec.silence = 0;
    wanted_spec.callback = fillAStreamCallback;
    wanted_spec.userdata = this;
    // 平衡音频回调缓冲区，兼顾延迟与稳定性（需为2的幂）
    wanted_spec.samples = LowLatencyConfig::BALANCED_AUDIO_SAMPLES;  // 使用配置文件中的音频样本数

    if (SDL_OpenAudio(&wanted_spec, nullptr) < 0) {
        qDebug() << "SDL_OpenAudio failed";
        return 0;
    }
    m_targetSampleFmt = AV_SAMPLE_FMT_S16;
    // m_targetChannels = m_audioCodecPar->channels;
    m_targetChannels = ffmpeg_get_codec_channels(m_fmtCtx->streams[m_audioIndex]);
    m_targetFreq = m_audioCodecPar->sample_rate;
    m_targetChannelLayout = (int64_t)ffmpeg_get_default_channel_layout(m_targetChannels);
    m_targetNbSamples = m_audioCodecPar->frame_size;
    m_audioIndex = m_decoder->audioIndex();
    m_fmtCtx = m_decoder->formatContext();

    m_sonicStream = sonicCreateStream(m_targetFreq, m_targetChannels);
    sonicSetQuality(m_sonicStream, 1);

    SDL_PauseAudio(0);
    return 1;
}

int AVPlayer::initVideo()
{
    m_frameTimer = 0.00;

    m_videoCodecPar = m_decoder->videoCodecPar();
    m_videoIndex = m_decoder->videoIndex();
    m_fmtCtx = m_decoder->formatContext();

    m_imageWidth = m_videoCodecPar ? m_videoCodecPar->width : 0;
    m_imageHeight = m_videoCodecPar ? m_videoCodecPar->height : 0;

    m_dstPixFmt = AV_PIX_FMT_YUV422P;
    // 改用快速缩放算法以降低转换时延
    m_swsFlags = SWS_FAST_BILINEAR;

    // 仅当分辨率有效时分配缓存，否则延迟到首帧分配
    if (m_imageWidth > 0 && m_imageHeight > 0) {
        int bufSize = av_image_get_buffer_size(m_dstPixFmt, m_imageWidth, m_imageHeight, 1);
        m_buffer = (uint8_t*) av_realloc(m_buffer, bufSize * sizeof(uint8_t));
        av_image_fill_arrays(m_pixels, m_pitch, m_buffer, m_dstPixFmt, m_imageWidth, m_imageHeight, 1);
    }

    //视频帧播放回调递插入线程池任务队列
    if (!ThreadPool::addTask(std::bind(&AVPlayer::videoCallback, this, std::placeholders::_1),
                             std::make_shared<int>(0))) {
        qDebug() << "videoCallback add task failed!";
    }
    return 1;
}

void AVPlayer::pause(bool isPause)
{
    if (m_hasAudio) {
        if (SDL_GetAudioStatus() == SDL_AUDIO_STOPPED)
            return;
        if (isPause) {
            if (SDL_GetAudioStatus() == SDL_AUDIO_PLAYING) {
                SDL_PauseAudio(1);
                m_pauseTime = av_gettime_relative() / 1000000.0;
                m_pause = 1;
            }
        } else {
            if (SDL_GetAudioStatus() == SDL_AUDIO_PAUSED) {
                SDL_PauseAudio(0);
                m_frameTimer += av_gettime_relative() / 1000000.0 - m_pauseTime;
                m_pause = 0;
            }
        }
    } else if (m_hasVideo) {
        // 仅视频：通过标志控制回放线程
        double now = av_gettime_relative() / 1000000.0;
        if (isPause) {
            if (!m_pause) {
                m_pauseTime = now;
                m_pause = 1;
            }
        } else {
            if (m_pause) {
                m_frameTimer += now - m_pauseTime;
                m_pause = 0;
            }
        }
    }
}

void AVPlayer::clearPlayer()
{
    if (playState() != AV_STOPPED) {
        m_exit = 1;
        if (m_hasAudio && playState() == AV_PLAYING)
            SDL_PauseAudio(1);
        m_decoder->exit();
        if (m_hasAudio)
            SDL_CloseAudio();
        if (m_swrCtx)
            swr_free(&m_swrCtx);
        if (m_swsCtx)
            sws_freeContext(m_swsCtx);
        m_swrCtx = nullptr;
        m_swsCtx = nullptr;
        if (m_sonicStream)
            sonicDestroyStream(m_sonicStream);
        m_sonicStream = nullptr;
    }
}

AVTool::MediaInfo* AVPlayer::detectMediaInfo(const QString& url)
{
    return m_decoder->detectMediaInfo(url);
}

AVPlayer::PlayState AVPlayer::playState()
{
    if (m_hasAudio) {
        AVPlayer::PlayState state;
        switch (SDL_GetAudioStatus()) {
        case SDL_AUDIO_PLAYING:
            state = AVPlayer::AV_PLAYING;
            break;
        case SDL_AUDIO_PAUSED:
            state = AVPlayer::AV_PAUSED;
            break;
        case SDL_AUDIO_STOPPED:
            state = AVPlayer::AV_STOPPED;
            break;
        default:
            state = AVPlayer::AV_STOPPED;
            break;
        }
        return state;
    }

    if (m_hasVideo) {
        if (m_exit)
            return AV_STOPPED;
        if (m_pause)
            return AV_PAUSED;
        return AV_PLAYING;
    }

    return AV_STOPPED;
}

void AVPlayer::initAVClock()
{
    m_audioClock.setClock(0.00);
    m_videoClock.setClock(0.00);
    m_clockInitFlag = 1;
}

void AVPlayer::displayImage(AVFrame* frame)
{
    if (frame) {
        // 首帧兜底：探测不到分辨率时，用首帧尺寸初始化并分配缓存
        if (m_imageWidth <= 0 || m_imageHeight <= 0) {
            m_imageWidth = frame->width;
            m_imageHeight = frame->height;

            int bufSize = av_image_get_buffer_size(m_dstPixFmt, m_imageWidth, m_imageHeight, 1);
            m_buffer = (uint8_t*) av_realloc(m_buffer, bufSize * sizeof(uint8_t));
            av_image_fill_arrays(m_pixels, m_pitch, m_buffer, m_dstPixFmt, m_imageWidth, m_imageHeight, 1);
        }

        // 判断是否需要像素格式/分辨率转换
        bool needConvert =
            (frame->format != m_dstPixFmt) ||
            (frame->width != m_imageWidth) ||
            (frame->height != m_imageHeight);

        if (needConvert) {
            m_swsCtx = sws_getCachedContext(m_swsCtx,
                                            frame->width,
                                            frame->height,
                                            (enum AVPixelFormat) frame->format,
                                            m_imageWidth,
                                            m_imageHeight,
                                            m_dstPixFmt,
                                            m_swsFlags,
                                            nullptr,
                                            nullptr,
                                            nullptr);

            if (m_swsCtx) {
                // 确保输出缓存已按当前目标尺寸分配
                int bufSize = av_image_get_buffer_size(m_dstPixFmt, m_imageWidth, m_imageHeight, 1);
                m_buffer = (uint8_t*) av_realloc(m_buffer, bufSize * sizeof(uint8_t));
                av_image_fill_arrays(m_pixels, m_pitch, m_buffer, m_dstPixFmt, m_imageWidth, m_imageHeight, 1);

                sws_scale(m_swsCtx, frame->data, frame->linesize, 0, frame->height, m_pixels, m_pitch);

                uint8_t* planes[4] = { m_pixels[0], m_pixels[1], m_pixels[2], m_pixels[3] };
                int lines[4] = { m_pitch[0], m_pitch[1], m_pitch[2], m_pitch[3] };
                emit frameChanged(QSharedPointer<VideoFrame>::create(m_dstPixFmt,
                                                                     m_imageWidth,
                                                                     m_imageHeight,
                                                                     planes,
                                                                     lines));
            } else {
                // 回退：直接透传
                emit frameChanged(QSharedPointer<VideoFrame>::create((AVPixelFormat) frame->format,
                                                                     frame->width,
                                                                     frame->height,
                                                                     frame->data,
                                                                     frame->linesize));
            }
        } else {
            // 无需转换，直接透传
            emit frameChanged(QSharedPointer<VideoFrame>::create((AVPixelFormat) frame->format,
                                                                 frame->width,
                                                                 frame->height,
                                                                 frame->data,
                                                                 frame->linesize));
        }

        //记录视频时钟
        m_videoClock.setClock(frame->pts * av_q2d(m_fmtCtx->streams[m_videoIndex]->time_base));
    }
}

void AVPlayer::videoCallback(std::shared_ptr<void> par)
{
    double time = 0.00;
    double duration = 0.00;
    double delay = 0.00;
    if (m_clockInitFlag == -1) {
        initAVClock();
    }
    do {
        if (m_exit)
            break;
        if (m_pause) {
            std::this_thread::sleep_for(std::chrono::milliseconds(50));
            continue;
        }
        if (m_decoder->getRemainingVFrame()) {
            MyFrame* lastFrame = m_decoder->peekLastVFrame();
            MyFrame* frame = m_decoder->peekVFrame();

            //qDebug()<<"video pts:"<<frame->pts;

            if (frame->serial != m_decoder->vidPktSerial()) {
                m_decoder->setNextVFrame();
                continue;
            }

            if (frame->serial != lastFrame->serial)
                m_frameTimer = av_gettime_relative() / 1000000.0;

            duration = vpDuration(lastFrame, frame);
            delay = computeTargetDelay(duration);

            time = av_gettime_relative() / 1000000.0;

            //qDebug()<<"delay:"<<delay<<endl;

            //显示时长未到
            if (time < m_frameTimer + delay) {
                QThread::msleep(
                    (uint32_t) (FFMIN(LowLatencyConfig::BALANCED_SYNC_REJUDGE_THRESHOLD, m_frameTimer + delay - time) * 1000));
                continue;
            }

            m_frameTimer += delay;
            if (time - m_frameTimer > LowLatencyConfig::BALANCED_SYNC_THRESHOLD_MAX)
                m_frameTimer = time;

            //队列中未显示帧一帧以上执行逻辑丢帧判断,倍速播放和逐帧播放
            //都不跑进此逻辑，倍速易造成丢帧过多导致界面不流畅
            // 平衡：温和的丢帧策略，保证稳定性
            if (m_playSpeed == 1.0 && m_decoder->getRemainingVFrame() > 1) {  // 恢复为>1，更稳定
                MyFrame* nextFrame = m_decoder->peekNextVFrame();
                if (nextFrame) {
                    duration = nextFrame->pts - frame->pts;
                    //若主时钟超前到大于当前帧理论显示应持续的时间了，则当前帧立即丢弃
                    // 平衡：使用原始duration阈值，避免过度丢帧
                    if (time > m_frameTimer + duration) {
                        m_decoder->setNextVFrame();
                        qDebug() << "abandon vframe (balanced mode)" << Qt::endl;
                        continue;
                    }
                }
                
                // 温和：基于延迟的丢帧阈值使用配置文件参数
                if (delay > LowLatencyConfig::FRAME_DROP_THRESHOLD) {  // 使用配置文件中的丢帧阈值
                    m_decoder->setNextVFrame();
                    qDebug() << "drop frame due to high delay:" << delay << Qt::endl;
                    continue;
                }
            }

            displayImage(&frame->frame);

            // 无音频时，基于视频时钟更新对外进度
            if (!m_hasAudio) {
                uint32_t _pts = (uint32_t) m_videoClock.getClock();
                if (m_lastAudPts != _pts) {
                    emit AVPtsChanged(_pts);
                    m_lastAudPts = _pts;
                }
            }

            //读索引后移
            m_decoder->setNextVFrame();
        } else {
            QThread::msleep(10);
        }
    } while (true);
    //qDebug()<<"videoCallBack exit"<<endl;
    if (m_decoder->isExit()) {
        emit AVTerminate();
    }
}

double AVPlayer::computeTargetDelay(double delay)
{
    // 无音频流时，不进行音视频同步，直接按视频节奏播放
    if (!m_hasAudio)
        return delay;

    //视频当前显示帧与当前播放音频帧时间戳差值
    double diff = m_videoClock.getClock() - m_audioClock.getClock();

    //计算同步阈值
    double sync = FFMAX(LowLatencyConfig::BALANCED_SYNC_THRESHOLD_MIN, FFMIN(LowLatencyConfig::BALANCED_SYNC_THRESHOLD_MAX, delay));

    //不同步时间超过阈值直接放弃同步
    if (!isnan(diff) && fabs(diff) < LowLatencyConfig::BALANCED_NOSYNC_THRESHOLD) {
        if (diff
            <= -sync) { //视频已落后音频大于一帧的显示时长，delay值应为0，立马将当前帧显示追赶音频
            delay = FFMAX(0, diff + delay);
        } else if (diff >= sync
                   && delay
                          > LowLatencyConfig::BALANCED_SYNC_FRAMEDUP_THRESHOLD) { //视频超前音频大于一个视频帧的时间，延时一个视频帧时间+已超时时间，下次判定将至少被延时到下个将要显示的视频帧pts
            delay = diff + delay;
        } else if (diff >= sync) { //高帧率视频直接延时两个视频帧时间；；；；
            delay = 2 * delay;
        }
    }
    return delay;
}

double AVPlayer::vpDuration(MyFrame* lastFrame, MyFrame* curFrame)
{
    if (curFrame->serial == lastFrame->serial) {
        double duration = curFrame->pts - lastFrame->pts;
        if (isnan(duration) || duration > LowLatencyConfig::BALANCED_NOSYNC_THRESHOLD)
            return lastFrame->duration;
        else
            return duration;
    } else {
        return 0.00;
    }
}