SkeyeRTSPLive高效转码之SkeyeVideoDecoder采用Nvidia独立显卡高效硬件解码解决方案（2)

Posted on by hackdl

在我之前写的一篇文章《SkeyeRTSPLive传统视频监控互联网+实现利器解决方案》中提到RTSP转RTMP的转流过程，简化流程就是通过SkeyeRTSPClient拉RTSP流，获取音视频编码数据，然后再通过SkeyeRTMPPusher推出去，流程非常简单；然后再实际开发过程中，我们发现其实这个过程并没有想象中那么简单；首先，RTSP协议支持多种音视频编码格式，如音频支持AAC,G711，G726,等，视频支持H264，H625，MJPEG, MPEG等等各种格式，而SkeyeRTMPPusher推流只支持H264（已扩展支持H265）格式，这时，音频我们可以通过SkeyeAACEncoder将音频转码成AAC格式，而视频我们可以通过SkeyeVideoDecoder解码成原始数据，然后再通过SkeyeVideoEncoder将原始数据转码成RTMP推送指定的格式，本文，我们将重点讲述SkeyeVideoDecoder基于Nvidia(英伟达)独立显卡的解码流程。

SkeyeVideoDecoder基Nvidia独立显卡的硬解码库SkeyeNvDecoder

SkeyeNvDecoder库是基于Nvidia独立显卡驱动的硬件解码程序，该解码程序效率非常高效且具有强大的并行解码效能力，其解码效率比ffmpeg软件解码效率提到至少5-6倍，最新的RTX系列显卡其解码效率甚至比软解码高10-12倍，轻松解码多路4K乃至8K高清视频无压力，本文采用的是截止目前（20190714）最新的显卡驱动，CUDA版本需要10.0或者以上版本支持。

1. 接口声明如下：
#ifndef SKEYENVDECODERAPI_H
#define SKEYENVDECODERAPI_H

#include 

//++ typedefine start
#ifndef SKEYENVDECODER_HANDLE
#define SKEYENVDECODER_HANDLE void*
#endif//SKEYENVDECODER_HANDLE

typedef enum _OutputFormat //native=默认解码器输出为NV12格式
{
    native = 0, bgrp, rgbp, bgra, rgba, bgra64, rgba64
}OutputFormat;


typedef enum _SKEYENvDecoder_CodecType {
    SKEYENvDecoder_Codec_MPEG1 = 0,                                         /**
2. SkeyeNvDecoder解码库调用流程

  • 第一步，初始化注册解码器
    注意，注册解码器函数全局只需调用一；

    int SKEYENvDecoder_Initsize(string &erroStr)
{
  try
  {
      if (!isInitsized) { //显卡只初始化一次
          ck(cuInit(0));
          int nGpu = 0;
          ck(cuDeviceGetCount(&nGpu));
          for (int i = 0; i
  • 第二步，创建解码器实例
SKEYENVDECODER_HANDLE SKEYENvDecoder_Create(NvDecoder_CodecType codec, int videoW, int videoH, bool bLowLatency, OutputFormat eOutputFormat, int& errCode, string &erroStr)
{
    //if (!isInitsized || !m_ctxV.size()) {
    //    return NULL;
    //}

    try {
        ck(cuInit(0));
        int nGpu = 0;
        ck(cuDeviceGetCount(&nGpu));
        CUcontext cuContext = NULL;
        m_curIndex++;
        m_curIndex = (m_curIndex) % nGpu;
        for (int i = 0; i  &v = m_ctxV.at(m_curIndex++ % m_ctxV.size());
        //std::cout = videoW && videoDecodeCaps.nMaxHeight >= videoH)
                {
                    NvDecoder* pDecoder = new NvDecoder(/*v.first*/cuContext, videoW, videoH, eOutputFormat== native?false:true,
                        (cudaVideoCodec)codec, NULL, bLowLatency, eOutputFormat);
                    pDecoder->Start();
                    return pDecoder;
                }
                else
                {
                    erroStr = "Width and height not supported on this GPU Decoder";
                    errCode = -2;
                }
            }
        }
    }
    catch (std::exception &e)
    {
        erroStr = e.what();
    }
    return NULL;
}
  • 第三步，调用解码函数解码
int SKEYENvDecoder_Decode(SKEYENVDECODER_HANDLE handle, const uint8_t *pData, int nSize, uint8_t ***pppFrame, int* pnFrameLen, int *pnFrameReturned)
{
    if (!handle)
        return -1;
    NvDecoder* pDecoder = (NvDecoder*)handle;

    int anSize[] = { 0, 3, 3, 4, 4, 8, 8 };
    //std::unique_ptr pImage(new uint8_t[nFrameSize]);
    std::vector* vecOutBuffer = pDecoder->GetFrameBufferVector();
    size_t nFrameSize = pDecoder->GetOutFrameSize();
    *pnFrameLen = nFrameSize;

    int nFrameReturned = 0, nFrame = 0;
    uint8_t **ppFrame = NULL;

    bool bLowLatency = pDecoder->IsSetLowLatency();
    bool bSuc = pDecoder->Decode(pData, nSize, &ppFrame, &nFrameReturned, CUVID_PKT_ENDOFPICTURE/*bLowLatency?CUVID_PKT_ENDOFPICTURE : 0*/);
    if (!bSuc)
        return -2;
    //if (!nFrame && nFrameReturned > 0)
    //LOG(INFO) GetVideoInfo();

    for (int i = 0; i GetSetOutputFormat())
        {
            if (i >= (*vecOutBuffer).size())
            {
                (*vecOutBuffer).push_back(new uint8_t[nFrameSize]);
            }
        }

        if (pDecoder->GetBitDepth() == 8)
        {
            switch (pDecoder->GetSetOutputFormat())
            {
            case native:
                //GetImage((CUdeviceptr)ppFrame[i], (*vecOutBuffer)[i], pDecoder->GetWidth(), pDecoder->GetHeight() + (pDecoder->GetChromaHeight() * pDecoder->GetNumChromaPlanes()));
                break;
            case bgrp:
                if (pDecoder->GetOutputFormat() == cudaVideoSurfaceFormat_YUV444)
                    YUV444ToColorPlanar((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                else
                    Nv12ToColorPlanar((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                GetImage(pDecoder->GetDeviceImagePtr(), (*vecOutBuffer)[i], pDecoder->GetWidth(), 3 * pDecoder->GetHeight());
                break;
            case rgbp:
                if (pDecoder->GetOutputFormat() == cudaVideoSurfaceFormat_YUV444)
                    YUV444ToColorPlanar((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                else
                    Nv12ToColorPlanar((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                GetImage(pDecoder->GetDeviceImagePtr(), (*vecOutBuffer)[i], pDecoder->GetWidth(), 3 * pDecoder->GetHeight());
                break;
            case bgra:
                if (pDecoder->GetOutputFormat() == cudaVideoSurfaceFormat_YUV444)
                    YUV444ToColor32((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 4 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                else
                    Nv12ToColor32((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 4 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                GetImage(pDecoder->GetDeviceImagePtr(), (*vecOutBuffer)[i], 4 * pDecoder->GetWidth(), pDecoder->GetHeight());
                break;
            case rgba:
                if (pDecoder->GetOutputFormat() == cudaVideoSurfaceFormat_YUV444)
                    YUV444ToColor32((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 4 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                else
                    Nv12ToColor32((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 4 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                GetImage(pDecoder->GetDeviceImagePtr(), (*vecOutBuffer)[i], 4 * pDecoder->GetWidth(), pDecoder->GetHeight());
                break;
            case bgra64:
                if (pDecoder->GetOutputFormat() == cudaVideoSurfaceFormat_YUV444)
                    YUV444ToColor64((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 8 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                else
                    Nv12ToColor64((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 8 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                GetImage(pDecoder->GetDeviceImagePtr(), (*vecOutBuffer)[i], 8 * pDecoder->GetWidth(), pDecoder->GetHeight());
                break;
            case rgba64:
                if (pDecoder->GetOutputFormat() == cudaVideoSurfaceFormat_YUV444)
                    YUV444ToColor64((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 8 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                else
                    Nv12ToColor64((uint8_t *)ppFrame[i], pDecoder->GetWidth(), (uint8_t*)pDecoder->GetDeviceImagePtr(), 8 * pDecoder->GetWidth(), pDecoder->GetWidth(), pDecoder->GetHeight());
                GetImage(pDecoder->GetDeviceImagePtr(), (*vecOutBuffer)[i], 8 * pDecoder->GetWidth(), pDecoder->GetHeight());
                break;
            }
        }
    }
    nFrame += nFrameReturned;

    if (nFrameReturned > 0)
    {
        if (pnFrameReturned)
            *pnFrameReturned = nFrameReturned;
        if (native != pDecoder->GetSetOutputFormat())
        {
            if (pppFrame && (*vecOutBuffer).size() > 0)
                *pppFrame = &(*vecOutBuffer)[0];
        }
        else
        {
            if (pppFrame && ppFrame)
                *pppFrame = ppFrame;
        }
    }
}
  • 第四步，停止解码，销毁解码器
void SKEYENvDecoder_Release(SKEYENVDECODER_HANDLE handle)
{
    if (!handle)
        return;
    NvDecoder* pDecoder = (NvDecoder*)handle;
    pDecoder->Stop();
    delete pDecoder;
    m_curIndex--;
    if (m_curIndex
  • 第五步，注销解码器，释放资源
int SKEYENvDecoder_Uninitsize()
{
    isInitsized = false;
    for (int nI = 0; nI

自此，SKEYENvDecoder的封装就完成了，我们可以通过其接口调用Nvidia的显卡进行硬件解码测试， 以下为真实应用效果，硬解12路效果图cpu I5占比11，730显卡点75-80，如下图所示：

有任何技术问题，欢迎大家和我技术交流：
295222688@qq.com

大家也可以加入SkeyePlayer流媒体播放器 QQ群进行讨论：
102644504

服务器托管，北京服务器托管，服务器租用 http://www.fwqtg.net

服务器托管，北京服务器托管，服务器租用，机房机柜带宽租用

服务器托管

咨询：董先生

电话13051898268 QQ/微信93663045！

上一篇:
下一篇: