基于FFmpeg的Android播放器

基于FFmpeg的Android播放器

1. 前言

FFmpeg是一个最有名的开源的编解码库,实现了通常的编解码逻辑。它还能够根据平台特性,与平台自有的硬件编解码器进行适配。经过一段时间的学习后,我使用FFmpeg在Android上实现了一个简单的支持硬件解码的视频播放器。在此写下这篇博客记录关键知识点。

代码在此:Android-VideoPlayer

整体上,该工程是基于FFmpeg的,不仅是基于FFmpeg的解码能力,而且一些标志位等直接使用FFmpeg的,不再另外包装。

2. 编译相关组件库

要使用FFmpeg,需要先将FFmpeg库本身以及它所需要的一些第三方库编译成Android平台库。具体编译脚本和库我已经完成并放到了github上,相关build下面都有readme,记录了编译的事项和相关步骤。

  • FFmpeg6.0:FFmpeg_build。基于最新的FFmpeg6.0编译,该版本的最大变化是支持了NDK的MediaCodec框架,在此之前,FFmpeg桥接Android硬件解码器的方法是通过反射调用到Java层的MediaCodec,需要将数据从Native拷贝到Java层,开销比较大,用起来也比较麻烦。另外一个重大更新是支持了av1-mediacodec,av1是一种比HEVC更高效的编码格式,体积小,但是编解码开销比较大,而且由于格式比较新,最近两三代的手机处理器才支持了av1的硬解,而编码器更是延后,今年苹果的A17 pro才支持了av1编码。但是这个格式将来一定会占有非常大的市场,目前从YouTube下载的HDR视频已经都是av1格式了。为了支持Android硬解,需要开启一些编译开关,最重要的是mediacodec后缀的那几个decoder,在FFmpeg源码目录运行./configure --list-decoders可以看到FFmpeg支持的所有的decoder。这个编译版本开启了非免费第三方库,不可用来商用,并且由于是练习,为了支持尽可能多的格式,没有进行过多裁切
  • libaom:aom_build, libaom是av1的编解码库,为了在老旧的cpu上支持av1解码,集成了该库。
  • libx265 & libx264:x264_buildx265_build, H264与HEVC的编解码库,兜底用。
  • fdk-aac & mp3lame:fdk-aac_buildlame_build,aac与mp3的音频编解码库,一般都可以硬解。这里是为了编码做准备。

3. 解码器

目前,在Android系统中,支持的解码器就是那明确的几个。如果要使用硬件解码器,必须以名字来查询而不是codec_id。

#define HW_DEC_COUNT 7

#define HW_DEC_H264 "h264_mediacodec"
#define HW_DEC_HEVC "hevc_mediacodec"
#define HW_DEC_VP8 "vp8_mediacodec"
#define HW_DEC_VP9 "vp9_mediacodec"
#define HW_DEC_AV1 "av1_mediacodec"
#define HW_DEC_MPEG2 "mpeg2_mediacodec"
#define HW_DEC_MPEG4 "mpeg4_mediacodec"

const static AVCodecID HW_DECODERS[HW_DEC_COUNT] = {
        AVCodecID::AV_CODEC_ID_H264,
        AVCodecID::AV_CODEC_ID_HEVC,
        AVCodecID::AV_CODEC_ID_VP8,
        AVCodecID::AV_CODEC_ID_VP9,
        AVCodecID::AV_CODEC_ID_AV1,
        AVCodecID::AV_CODEC_ID_MPEG2VIDEO,
        AVCodecID::AV_CODEC_ID_MPEG4
};

const static const char* HW_DECODER_NAMES[HW_DEC_COUNT] = {
        HW_DEC_H264,
        HW_DEC_HEVC,
        HW_DEC_VP8,
        HW_DEC_VP9,
        HW_DEC_AV1,
        HW_DEC_MPEG2,
        HW_DEC_MPEG4
};

static bool supportHWDec(AVCodecID codecId) {
    for (AVCodecID id : HW_DECODERS) {
        if (id == codecId) {
            return true;
        }
    }
    return false;
}

static const char* getHWDecName(AVCodecID codecId) {
    for (int i = 0; i < HW_DEC_COUNT; i++) {
        if (HW_DECODERS[i] == codecId) {
            return HW_DECODER_NAMES[i];
        }
    }
    return nullptr;
}
// 根据参数查找相应的decoder
bool FFmpegDecoder::init(AVCodecParameters *params, PreferCodecType preferType) {

    AVCodecID ffCodecID = AV_CODEC_ID_NONE;
    try {
        ffCodecID = AVCodecID(params->codec_id);
    } catch (...) {
        LOGE(TAG, "failed to convert %d to AVCodecID", params->codec_id);
        return false;
    }

	// 可以支持指定解码器类型。如果未指定,那就优先查找硬件解码器,找不到再去找软件解码器
    if (preferType == PreferCodecType::HW) {
        return findHWDecoder(params, ffCodecID);
    } else if (preferType == PreferCodecType::SW) {
        return findSWDecoder(params, ffCodecID);
    } else {
        if (findHWDecoder(params, ffCodecID)) {
            return true;
        }
        if (findSWDecoder(params, ffCodecID)) {
            return true;
        }
        return false;
    }
}

// 查找硬件解码器
bool FFmpegDecoder::findHWDecoder(AVCodecParameters *params, AVCodecID codecId) {
    release();
    int ret;
    const char *hwDecName = getHWDecName(codecId);
    if (hwDecName == nullptr) {
        return false;
    }

    const AVCodec * aCodec = avcodec_find_decoder_by_name(hwDecName);
    if (aCodec == nullptr) {
        LOGE(TAG, "Can't find hw decoder for codec: {id = %d, hw_name = %s}", codecId, hwDecName);
        return false;
    } else {
        codec = const_cast<AVCodec *>(aCodec);
    }

    codecCtx = avcodec_alloc_context3(codec);
    if (!codecCtx) {
        LOGE(TAG, "failed to alloc codec context");
        return false;
    }

    ret = avcodec_parameters_to_context(codecCtx, params);
    if (ret < 0) {
        LOGE(TAG, "copy decoder params failed, err = %d", ret);
        return false;
    }

    for (int i = 0;;i++) {
        const AVCodecHWConfig *config = avcodec_get_hw_config(codec, i);
        if (config == nullptr) {
            LOGE(TAG, "%s hw config is null", codec->name);
            break;
        }
        if ((config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX) &&
            config->device_type == AVHWDeviceType::AV_HWDEVICE_TYPE_MEDIACODEC) {
            // 该解码器支持硬件解码
            out_hw_pix_format = config->pix_fmt;
            hwPixFormat = config->pix_fmt;
            codecCtx->get_format = get_hw_format;
            if (initHWDecoder(codecCtx, AVHWDeviceType::AV_HWDEVICE_TYPE_MEDIACODEC) < 0) {
                LOGE(TAG, "initHWDecoder failed");
                return false;
            } else {
                break;
            }
        }
    }

    ret = avcodec_open2(codecCtx, codec, nullptr);

    if (ret < 0) {
        char buf[100];
        av_make_error_string(buf, 100, ret);
        LOGE(TAG, "open codec failed for %s, err = %s", codec->name, buf);
        return false;
    }
    codecType = CodecType::HW;
    return true;
}

// 查找软件解码器
bool FFmpegDecoder::findSWDecoder(AVCodecParameters *params, AVCodecID codecId) {
    release();
    int ret;
    const AVCodec * aCodec = avcodec_find_decoder(codecId);
    if (aCodec == nullptr) {
        LOGE(TAG, "Can't find decoder for codecID %d", codecId);
        return false;
    }
    codec = const_cast<AVCodec *>(aCodec);

    codecCtx = avcodec_alloc_context3(codec);
    if (!codecCtx) {
        LOGE(TAG, "failed to alloc codec context");
        return false;
    }

    ret = avcodec_parameters_to_context(codecCtx, params);
    if (ret < 0) {
        LOGE(TAG, "copy decoder params failed, err = %d", ret);
        return false;
    }

    ret = avcodec_open2(codecCtx, codec, nullptr);
    if (ret < 0) {
        char buf[100];
        av_make_error_string(buf, 100, ret);
        LOGE(TAG, "open codec failed for %s, err = %s", codec->name, buf);
        return false;
    }
    codecType = CodecType::SW;
    return true;
}

4. 解码流程

解码其实很简单。首先是先创建方便使用的音频和视频帧结构体,这两个结构体分别是AudioFrame和VideoFrame。其实也就是额外包含了一些属性方便访问和使用,内容物还是AVFrame。

对于解码,其实流程比较固定。

  • 打开文件,获取对应的FormatContext与Codec。
  • 通过AVFormatContext解复用,获取音频和视频的AVPacket,并将其存放到对应音频解码和视频解码的同步队列中,等待解码线程取用。
  • 音频和视频解码线程从各自的AVPacket同步队列中取出AVPacket,并进行解码。解码完成后获取AVFrame,然后将其设置到AudioFrame和VideoFrame中,并存放到对应的同步队列中,等待同步线程取用。
  • 同步线程从AudioFrame和VideoFrame的同步队列中取出已经解码好的数据,根据各自的pts来决定将AudioFrame和VideoFrame送到输出的时机。

音视频同步逻辑:在视频播放的过程中,音频是连续不断的,而视频却有不同帧率,所以同步是基于音频的时间戳。当然有些视频没有音频内容,此时就需要独立时钟来作为同步时间戳。

对于seek功能的支持:当用户进行seek时,置一个seekFlag为true。在解复用阶段,如果seekFlag为true,那解复用就对文件进行seek。同时对音视频AVPacket同步队列都清空,并向其中各存放一个seek标志的AVPacket。然后继续从seek点开始解复用。解码阶段在读取解复用存放的特殊AVPacket时,就对解码器进行reset,清空其内部缓存,然后存放两个seek标志分别到AVFrame的同步队列。在同步阶段,如果同步线程读取到任何一个具有seek标志的AVFrame,就停止输送到输出,并等待另一个同步队列读取到具有seek标志的AVFrame。在音视频都读取到seek标志之前,所有的AVFrame都弃用。读取到之后,重新进行同步。

解复用:

/*
 * Read packet data from source.
 * If the packet has some flags like STREAM_FLAG_SOUGHT, this packet won't
 * contain data.
 * */
void Player::readStreamLoop() {
    if (!formatCtx) {
        LOGE(TAG, "no format context");
        return;
    }
    int ret;
    bool pushSuccess = false;
    while (!stopReadFlag) {
        AVPacket *packet = av_packet_alloc();
        if (!packet) {
            LOGE(TAG, "av_packet_alloc failed");
            return;
        }

        if (seekFlag) {

            int64_t pts = (int64_t) (seekPtsMS / 1000.0f * AV_TIME_BASE);
            LOGD(TAG, "meet seek, time = %lld", pts);
            int streamIndex = -1;
//            if (audioStreamIndex >= 0) {
//                pts = (int64_t)(seekPtsMS / av_q2d(formatCtx->streams[audioStreamIndex]->time_base));
//                streamIndex = audioStreamIndex;
//            } else if (videoStreamIndex >= 0) {
//                pts = (int64_t)(seekPtsMS / av_q2d(formatCtx->streams[videoStreamIndex]->time_base));
//                streamIndex = videoStreamIndex;
//            }

            av_seek_frame(formatCtx, streamIndex, pts, AVSEEK_FLAG_BACKWARD);

            // put a empty packet width flag STREAM_FLAG_SOUGHT
            if (enableAudio) {
                audioPacketQueue.clear();
                PacketWrapper *p = playerContext.getEmptyPacketWrapper();
                p->flags = STREAM_FLAG_SOUGHT;
                audioPacketQueue.forcePush(p);
//                audioDecodeSeekFlag = true;
            }

            if (enableVideo) {
                videoPacketQueue.clear();
                PacketWrapper *p = playerContext.getEmptyPacketWrapper();
                p->flags = STREAM_FLAG_SOUGHT;
                videoPacketQueue.forcePush(p);
//                videoDecodeSeekFlag = true;
            }

//            syncSeekFlag = true;

            seekFlag = false;
        }

        ret = av_read_frame(formatCtx, packet);

        if (ret == 0) {
            if (packet->stream_index == audioStreamIndex && enableAudio) {
                PacketWrapper *pw = playerContext.getEmptyPacketWrapper();
                pw->setParams(packet);
                if (videoPacketQueue.getSize() == 0) {
                    audioPacketQueue.forcePush(pw);
                } else {
                    pushSuccess = audioPacketQueue.push(pw);
                    if (!pushSuccess) {
                        audioPacketQueue.forcePush(pw);
                    }
                }

            } else if (packet->stream_index == videoStreamIndex && enableVideo) {
                PacketWrapper *pw = playerContext.getEmptyPacketWrapper();
                pw->setParams(packet);
                if (audioPacketQueue.getSize() == 0) {
                    videoPacketQueue.forcePush(pw);
                } else {
                    pushSuccess = videoPacketQueue.push(pw);
                    if (!pushSuccess) {
                        videoPacketQueue.forcePush(pw);
                    }
                }
            } else {
                av_packet_unref(packet);
                av_packet_free(&packet);
            }
        } else if (ret == AVERROR_EOF) {
            av_packet_free(&packet);
            packet = nullptr;
            if (enableAudio) {
                PacketWrapper *pw = playerContext.getEmptyPacketWrapper();
                audioPacketQueue.forcePush(pw);
            }
            if (enableVideo) {
                PacketWrapper *pw = playerContext.getEmptyPacketWrapper();
                videoPacketQueue.forcePush(pw);
            }
        } else if (ret < 0) {
            LOGE(TAG, "av_read_frame failed");
            av_packet_free(&packet);
            packet = nullptr;
            return;
        }


    }
}

音频解码:

void Player::decodeAudioLoop() {
    if (!formatCtx) {
        return;
    }
    if (!audioDecoder) {
        LOGE(TAG, "audio decoder is null");
        return;
    }
    int ret;
    optional<PacketWrapper *> packetOpt;
    PacketWrapper *pw = nullptr;
    AVFrame *frame = nullptr;
    AudioFrame *audioFrame = nullptr;
    while (!stopDecodeAudioFlag && enableAudio) {

        packetOpt = audioPacketQueue.pop();
        if (!packetOpt.has_value()) {
            LOGE(TAG, "audio packetOpt has no value");
            break;
        }
        pw = packetOpt.value();

        if ((pw->flags & STREAM_FLAG_SOUGHT) == STREAM_FLAG_SOUGHT) {
            LOGD(TAG, "decode audio, meet a seek frame");
            playerContext.recyclePacketWrapper(pw);
            pw = nullptr;
            audioFrameQueue.clear();
            audioDecoder->flush();
            audioFrame = playerContext.getEmptyAudioFrame();
            audioFrame->flags |= STREAM_FLAG_SOUGHT;
            audioFrameQueue.forcePush(audioFrame);
            audioFrame = nullptr;
            continue;
        }

        ret = audioDecoder->sendPacket(pw->avPacket);
        if (ret < 0) {
            LOGE(TAG, "audio decoder send packet failed, err = %d", ret);
            break;
        }

        while (true) {
            frame = av_frame_alloc();
            ret = audioDecoder->receiveFrame(frame);
            if (ret < 0) {
                av_frame_unref(frame);
                av_frame_free(&frame);
                frame = nullptr;
                break;
            }
            audioFrame = playerContext.getEmptyAudioFrame();
            audioFrame->setParams(frame, audioStreamMap[audioStreamIndex].sampleFormat,
                                  formatCtx->streams[audioStreamIndex]->time_base);
            if (!audioFrameQueue.push(audioFrame)) {
                audioFrameQueue.push(audioFrame, false);
            }
            // DON'T delete AVFrame here, it will be carried to output by AudioFrame
            audioFrame = nullptr;
            frame = nullptr;
        }

        if (ret == AVERROR(EAGAIN)) {
//            LOGD(TAG, "audio stream again");
            continue;
        } else if (ret == AVERROR_EOF) {
//            LOGD(TAG, "audio stream meets eof");
            break;
        } else {
//            LOGE(TAG, "audio decoder error: %d", ret);
            break;
        }
    }

    if (pw) {
        playerContext.recyclePacketWrapper(pw);
    }

    if (frame) {
        av_frame_unref(frame);
        av_frame_free(&frame);
        frame = nullptr;
    }

    if (audioFrame) {
        audioFrameQueue.push(audioFrame, false);
        audioFrame = nullptr;
    }

    LOGD(TAG, "audio decode loop finish");

}

视频解码:

void Player::decodeVideoLoop() {
    if (!formatCtx) {
        return;
    }
    if (!videoDecoder) {
        LOGE(TAG, "video decoder is null");
        return;
    }
    int ret;
    optional<PacketWrapper *> packetOpt;
    PacketWrapper *pw = nullptr;
    AVFrame *frame = nullptr;
    VideoFrame *videoFrame = nullptr;
    while (!stopDecodeVideoFlag && enableVideo) {
        packetOpt = videoPacketQueue.pop();
        if (!packetOpt.has_value()) {
            LOGE(TAG, "video packetOpt has no value");
            break;
        }
        pw = packetOpt.value();

        if ((pw->flags & STREAM_FLAG_SOUGHT) == STREAM_FLAG_SOUGHT) {
            LOGD(TAG, "decode video, meet a seek frame");
            playerContext.recyclePacketWrapper(pw);
            pw = nullptr;
            videoFrameQueue.clear();
            videoDecoder->flush();
            videoFrame = playerContext.getEmptyVideoFrame();
            videoFrame->flags |= STREAM_FLAG_SOUGHT;
            videoFrameQueue.forcePush(videoFrame);
            videoFrame = nullptr;
            continue;
        }

        ret = videoDecoder->sendPacket(pw->avPacket);
        if (ret < 0) {
            LOGE(TAG, "video decoder send packet failed, err = %d", ret);
            break;
        }

        while (true) {
            frame = av_frame_alloc();
            ret = videoDecoder->receiveFrame(frame);
            if (ret < 0) {
                av_frame_unref(frame);
                av_frame_free(&frame);
                frame = nullptr;
                break;
            }
            videoFrame = playerContext.getEmptyVideoFrame();
            videoFrame->setParams(frame, AVPixelFormat(frame->format),
                                  formatCtx->streams[videoStreamIndex]->time_base);
            if (!videoFrameQueue.push(videoFrame)) {
                videoFrameQueue.push(videoFrame, false);
            }
            // DON'T delete AVFrame, it will be carried to output by VideoFrame.
            videoFrame = nullptr;
            frame = nullptr;

        }

        if (ret == AVERROR(EAGAIN)) {
//            LOGD(TAG, "video stream again");
            continue;
        } else if (ret == AVERROR_EOF) {
//            LOGD(TAG, "video stream meets eof");
            break;
        } else {
//            LOGE(TAG, "video decoder error: %d", ret);
            break;
        }
    }

    if (pw) {
        playerContext.recyclePacketWrapper(pw);
    }

    if (frame) {
        av_frame_unref(frame);
        av_frame_free(&frame);
        frame = nullptr;
    }

    if (videoFrame) {
        videoFrameQueue.push(videoFrame, false);
        videoFrame = nullptr;
    }

    LOGD(TAG, "video decode loop finish");

}

同步代码:

void Player::syncLoop() {
    chrono::system_clock::time_point lastAudioWriteTime;
    chrono::system_clock::time_point lastVideoWriteTime;

    int64_t lastAudioPts = -1;
    int64_t lastVideoPts = -1;

    if (stateListener != nullptr) {
        stateListener->playStateChanged(true);
    }

    AudioFrame *audioFrame = unPlayedAudioFrame;
    unPlayedAudioFrame = nullptr;
    VideoFrame *videoFrame = unPlayedVideoFrame;
    unPlayedVideoFrame = nullptr;

    while (!stopSyncFlag) {
        if (enableAudio && enableVideo) {
            if (audioFrame == nullptr) {
                optional<AudioFrame *> frameOpt = audioFrameQueue.pop();
                if (frameOpt.has_value()) {
                    audioFrame = frameOpt.value();
                } else {
                    break;
                }
            }

            if (videoFrame == nullptr) {
                optional<VideoFrame *> frameOpt = videoFrameQueue.pop();
                if (frameOpt.has_value()) {
                    videoFrame = frameOpt.value();
                } else {
                    break;
                }
            }

            if (lastAudioPts == -1) {
                lastAudioPts = audioFrame->pts;
            }

            if (lastVideoPts == -1) {
                lastVideoPts = videoFrame->pts;
            }

            if ((audioFrame->flags & STREAM_FLAG_SOUGHT) == STREAM_FLAG_SOUGHT
                    && (videoFrame->flags & STREAM_FLAG_SOUGHT) == STREAM_FLAG_SOUGHT) {
                LOGD(TAG, "sync, meet both audio and video seek frame");
                playerContext.recycleAudioFrame(audioFrame);
                audioFrame = nullptr;
                playerContext.recycleVideoFrame(videoFrame);
                videoFrame = nullptr;
                continue;
            } else if ((audioFrame->flags & STREAM_FLAG_SOUGHT) == STREAM_FLAG_SOUGHT) {
                LOGD(TAG, "sync, meet audio seek frame");
                playerContext.recycleVideoFrame(videoFrame);
                videoFrame = nullptr;
                continue;
            } else if ((videoFrame->flags & STREAM_FLAG_SOUGHT) == STREAM_FLAG_SOUGHT) {
                LOGD(TAG, "sync, meet video seek frame");
                playerContext.recycleAudioFrame(audioFrame);
                audioFrame = nullptr;
                continue;
            }
            int64_t audioOutputPts = audioFrame->getOutputPts();
            if (videoFrame->pts <= audioOutputPts) {
                lastVideoPts = videoFrame->pts;
                videoOutput->write(videoFrame);
                videoFrame = nullptr;
            } else {
                int64_t outputFrames = (videoFrame->pts + 3 - audioOutputPts) * 1.0f / 1000 * audioFrame->sampleRate;
                outputFrames = min((int64_t)(audioFrame->numFrames - audioFrame->outputStartIndex), outputFrames);
                audioFrame->outputFrameCount = outputFrames;
                lastAudioPts = audioOutputPts;
                audioOutput->write(audioFrame);
                audioFrame->outputStartIndex += audioFrame->outputFrameCount;
                if (audioFrame->outputStartIndex == audioFrame->numFrames) {
                    playerContext.recycleAudioFrame(audioFrame);
                    audioFrame = nullptr;
                }

            }

            if (stateListener != nullptr) {
                stateListener->progressChanged(lastAudioPts, false);
            }

        } else if (enableVideo) {
            if (videoFrame == nullptr) {
                optional<VideoFrame *> frameOpt = videoFrameQueue.pop();
                if (frameOpt.has_value()) {
                    videoFrame = frameOpt.value();
                } else {
                    break;
                }
            }
            lastVideoPts = videoFrame->pts;
            videoOutput->write(videoFrame);
            videoFrame = nullptr;

            this_thread::sleep_for(chrono::milliseconds(17));

            if (stateListener != nullptr) {
                stateListener->progressChanged(lastVideoPts, false);
            }
        } else if (enableAudio) {
            if (audioFrame == nullptr) {
                optional<AudioFrame *> frameOpt = audioFrameQueue.pop();
                if (frameOpt.has_value()) {
                    audioFrame = frameOpt.value();
                } else {
                    break;
                }
            }
            lastAudioPts = audioFrame->pts;
            audioOutput->write(audioFrame);
            playerContext.recycleAudioFrame(audioFrame);
            audioFrame = nullptr;
            if (stateListener != nullptr) {
                stateListener->progressChanged(lastAudioPts, false);
            }
        } else {
            LOGE(TAG, "both audio and video disabled, break");
            break;
        }

    }

    if (audioFrame) {
        unPlayedAudioFrame = audioFrame;
    }

    if (videoFrame) {
        unPlayedVideoFrame = videoFrame;
    }

    if (stateListener != nullptr) {
        stateListener->playStateChanged(false);
    }
}

5. 音频输出

不同的平台使用不同的音频框架进行音频输出。Android平台我选择使用oboe。对音频输出我定义了一下接口来统一各平台的调用。

//
// Created by 祖国瑞 on 2022/9/5.
//

#ifndef ANDROID_VIDEOPLAYER_IAUDIOOUTPUT_H
#define ANDROID_VIDEOPLAYER_IAUDIOOUTPUT_H

#include <stdlib.h>
#include "PlayerContext.h"
#include "AudioFrame.h"
extern "C" {
#include "FFmpeg/libavformat/avformat.h"
}

class IAudioOutput {
public:
    IAudioOutput(PlayerContext *playerContext) {
        this->playerCtx = playerContext;
    }

    virtual bool create(int sampleRate, int channels, AVSampleFormat sampleFormat) = 0;

    virtual void release() = 0;
    virtual void start() = 0;
    virtual void stop() = 0;
    virtual bool write(AudioFrame *audioFrame) = 0;
    virtual void write(uint8_t *buffer, int framesPerChannel) = 0;

protected:
    PlayerContext *playerCtx = nullptr;
};

#endif //ANDROID_VIDEOPLAYER_IAUDIOOUTPUT_H

由于oboe使用比较简单,这里就不单独列出来了。需要注意的是,由于源文件采样率及采样格式多种多样,音频框架不一定支持。所以在输出之前,使用FFmpeg的avresample来重采样。

6. 视频输出(需要优化)

视频输出使用OpenGL ES,这里还有很多需要优化,比如EGL的格式目前是写死的,对于超出8bit长度的像素格式,还是统一转换为了float32并建立纹理,YUV的非planner数据(例如NV21格式,U和V是混合在一个数据平面中交替出现的)需要切分成三个纹理使用。这样很多工作都是由cpu完成的,开销比较大,而且速度较慢。接下来会探索Android的ANativeBuffer支持的format,它与OpenGL ES的纹理格式是对应的,尽量减少cpu工作。

视频输出同样也对各平台规定了一个接口:

//
// Created by 祖国瑞 on 2022/9/5.
//

#ifndef ANDROID_VIDEOPLAYER_IVIDEOOUTPUT_H
#define ANDROID_VIDEOPLAYER_IVIDEOOUTPUT_H

#include <stdlib.h>
#include "VideoFrame.h"
#include "PlayerContext.h"
#include "SizeMode.h"



class IVideoOutput {
public:
    IVideoOutput(PlayerContext *playerContext) {
        this->playerCtx = playerContext;
    };
    virtual bool setFormat(AVPixelFormat pixelFormat, AVColorSpace colorSpace, bool isHDR) = 0;
    virtual bool create(void *surface) = 0;
    virtual void release() = 0;
    virtual void setScreenSize(int32_t width, int32_t height) = 0;
    virtual bool isReady() = 0;
    virtual void write(VideoFrame* frame) = 0;
    virtual void setSizeMode(SizeMode mode) = 0;

protected:
    PlayerContext *playerCtx;
    AVPixelFormat srcPixelFormat = AVPixelFormat::AV_PIX_FMT_NONE;
};


#endif //ANDROID_VIDEOPLAYER_IVIDEOOUTPUT_H

对于Android平台,我使用OpenGL ES来渲染。

首先,shader部分很简单,其主要作用就是yuv转rgb。但是由于GPU天然适合做大量简单计算,所以我们可以在shader里去最大化兼容各种像素格式,例如10bit int,16bit float等。以及各种大尾序小尾序等。

static const char *vertexShaderCode =
        "#version 300 es\n"
        "layout (location = 0) in vec3 aPos;\n"
        "layout (location = 1) in vec2 aTexCoord;\n"
        "out vec2 TexCoord;\n"
        "void main() {\n"
        "    gl_Position = vec4(aPos, 1.0f);\n"
        "    TexCoord = aTexCoord;\n"
        "}\n";


static const char *yuv2rgbShaderCode =
        "#version 300 es\n"
        "precision mediump float;\n"
        "uniform sampler2D tex_y;\n"
        "uniform sampler2D tex_u;\n"
        "uniform sampler2D tex_v;\n"
        "in vec2 TexCoord;\n"
        "out vec4 FragColor;\n"
        "void main() {\n"
        "    float y = texture(tex_y, TexCoord).r - 0.0625f;\n"
        "    float u = texture(tex_u, TexCoord).r - 0.5f;\n"
        "    float v = texture(tex_v, TexCoord).r - 0.5f;\n"
        "    float r = 1.164f * y + 1.793f * v;\n"
        "    float g = 1.164f * y - 0.213f * u - 0.533f * v;\n"
        "    float b = 1.164f * y + 2.112f * u;\n"
        "    //float a = texture(tex_y, TexCoord).r;\n"
        "    FragColor = vec4(r, g, b, 1.0f);\n"
        "}\n";

static const char *yuv16ui2rgbShaderCode =
        "#version 300 es\n"
        "precision mediump float;\n"
        "uniform usampler2D tex_y;\n"
        "uniform usampler2D tex_u;\n"
        "uniform usampler2D tex_v;\n"
        "in vec2 TexCoord;\n"
        "out vec4 FragColor;\n"
        "void main() {\n"
        "    float y = float(texture(tex_y, TexCoord).r) - 0.0625f;\n"
        "    float u = float(texture(tex_u, TexCoord).r) - 0.5f;\n"
        "    float v = float(texture(tex_v, TexCoord).r) - 0.5f;\n"
        "    float r = 1.164f * y + 1.793f * v;\n"
        "    float g = 1.164f * y - 0.213f * u - 0.533f * v;\n"
        "    float b = 1.164f * y + 2.112f * u;\n"
        "    //float a = texture(tex_y, TexCoord).r;\n"
        "    FragColor = vec4(r, g, b, 1.0f);\n"
        "}\n";

static const char *rgbShaderCode =
        "#version 300 se\n"
        "\n"
        "uniform sampler2D tex_rgb;\n"
        "\n"
        "in vec2 TexCoord;\n"
        "\n"
        "out vec4 FragColor;\n"
        "\n"
        "void main() {\n"
        "    FragColor = texture(tex_rgb, TexCoord);\n"
        "}\n";

然后就是像素处理了,我们需要把各种像素布局和像素格式的图片转换为OpenGL ES可以接受的texture,然后才能在shader里进行处理。

首先是根据像素格式来确定使用哪种shader。

bool GLESRender::setFormat(AVPixelFormat format, AVColorSpace colorSpace, bool isHDR) {

    LOGD(TAG, "setFormat");

    if (!eglWindow.isReady()) {
        LOGE(TAG, "eglWindow is not ready");
        return false;
    }

    pixelType = get_pixel_type(format);
    pixelLayout = get_pixel_layout(format);

    if (pixelType == PixelType::None || pixelLayout == PixelLayout::None) {
        LOGE(TAG, "unsupported pixel format: %d", format);
        return false;
    }
    // 根据像素格式来确定texture的格式等。注意这里GLES的format与GL的format并不完全一致,有很多是用不了的。
    // 你可以参考Android的硬件buffer format与GLES的format的对应关系。
    // 链接在此:https://developer.android.google.cn/ndk/reference/group/a-hardware-buffer
    if (pixelType == PixelType::RGB) {
        switch (format) {
            case AV_PIX_FMT_RGB24:
                glDataType = GL_UNSIGNED_BYTE;
                glInternalFormat = GL_RGB;
                glDataFormat = GL_UNSIGNED_BYTE;
                glSupportFormat = true;
                break;
            case AV_PIX_FMT_RGB565LE:
                glDataType = GL_UNSIGNED_SHORT_5_6_5;
                glInternalFormat = GL_RGB;
                glDataFormat = GL_UNSIGNED_SHORT_5_6_5;
                glSupportFormat = true;
                break;
            case AV_PIX_FMT_RGB444LE:
                glDataType = GL_UNSIGNED_SHORT_4_4_4_4;
                glInternalFormat = GL_RGB;
                glDataFormat = GL_UNSIGNED_SHORT_4_4_4_4;
                glSupportFormat = true;
                break;
            default:
                LOGE(TAG, "unsupported RGB format: %d", format);
                return false;
        }
        if (!shader.compileShader(vertexShaderCode, rgbShaderCode)) {
            LOGE(TAG, "format = RGB24, compile shader failed");
            return false;
        }

    } else if (pixelType == PixelType::YUV) {
        yuvCompDepth = get_yuv_comp_depth(format);
        if (yuvCompDepth < 0) {
            LOGE(TAG, "get_yuv_comp_depth failed, format = %d", format);
            return false;
        }
        const char *fragmentCode;
        if (yuvCompDepth <= 8) {
            glDataType = GL_UNSIGNED_BYTE;
            glInternalFormat = GL_LUMINANCE;
            glDataFormat = GL_LUMINANCE;
            fragmentCode = yuv2rgbShaderCode;
        } else if (yuvCompDepth <= 16) {
            glDataType = GL_FLOAT;
            glInternalFormat = GL_R32F;
            glDataFormat = GL_RED;
            fragmentCode = yuv2rgbShaderCode;
        } else {
            LOGE(TAG, "unsupported yuvCompDepth: %d", yuvCompDepth);
            return false;
        }
        if (!shader.compileShader(vertexShaderCode, fragmentCode)) {
            LOGE(TAG, "format = %d, compile shader failed", format);
            return false;
        }
        eglWindow.makeCurrent();
        LOGD(TAG, "yuvCompDepth = %d", yuvCompDepth);
    } else {
        LOGE(TAG, "unsupported pixel format: %d", format);
        return false;
    }

    this->format = format;
    this->colorSpace = colorSpace;
    this->isHDR = isHDR;

    LOGD(TAG, "setFormat: format = %d, pixType = %d, glDataType = 0x%x", format, pixelType, glDataType);

    if (!shader.isReady()) {
        LOGE(TAG, "shader is not ready");
        return false;
    }
    return true;
}

拿到视频数据后,某些yuv数据需要将其转换成三个独立的纹理,再给OpenGL ES去处理。最关键的信息就是像素类型、像素布局和像素深度。

enum class PixelType {
    None = -1,
    RGB,
    YUV
};

// 像素布局。
enum class PixelLayout {
    None = -1,
    // 多数yuv数据(例如YUV420P)都是planner,YUV三种像素是独立存储的,各占一个平面。
    Planner, 
    // 多数RGB数据(例如RGB565)都是packet,RGB三像素依次存储,混编在一个平面内。
    Packet,
    // 半平面,Android平台的NV21和NV12就是这种格式,Y像素单独存储到一个平面,UV数据依次存储,混编在一个平面
    Semi_Planner,
};

// 获取某种yuv格式的像素深度。一般都是8bit,现在有很多HDR视频是10bit。
// 目前还没有RGB相关的该方法,因为RGB图片本来就可以直接转为纹理,不需要单独处理。如果之后出现了与ANativeBuffer的格式不兼容的RGB格式,
// 也需要类似的处理。
int get_yuv_comp_depth(AVPixelFormat format);

然后就是将yuv数据分别读取到三个像素buffer中,方便之后转换为纹理。

bool read_yuv_pixel(AVFrame *frame, AVPixelFormat format, int64_t width, int64_t height,
                    uint8_t *yBuffer, int *yWidth, int *yHeight,
                    uint8_t *uBuffer, int *uWidth, int *uHeight,
                    uint8_t *vBuffer, int *vWidth, int *vHeight) {

    PixelLayout layout = get_pixel_layout(format);
    if (layout == PixelLayout::Planner) {
        return read_yuv_planner(frame, format, width, height, yBuffer, yWidth, yHeight, uBuffer, uWidth, uHeight, vBuffer, vWidth, vHeight);
    } else if (layout == PixelLayout::Packet) {
        return read_yuv_packet(frame, format, width, height, yBuffer, yWidth, yHeight, uBuffer, uWidth, uHeight, vBuffer, vWidth, vHeight);
    } else if (layout == PixelLayout::Semi_Planner) {
        return read_yuv_semi_planner(frame, format, width, height, yBuffer, yWidth, yHeight, uBuffer, uWidth, uHeight, vBuffer, vWidth, vHeight);
    } else {
        return false;
    }
}
评论 2
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