EasyRTSPLive高效转码之EasyVideoDecoder采用Nvidia独立显卡高效硬件解码解决方案（附源码)
在我之前写的一篇文章《EasyRTSPLive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EasyRTSPClient拉RTSP流，获取音视频编码数据，然后再通过EasyRTMP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EasyRTMP推流只支持H264（已扩展支持H265）格式，这时，音频我们可以通过EasyAACEncoder将音频转码成AAC格式，而视频我们可以通过EasyVideoDecoder解码成原始数据，然后再通过EasyVideoEncoder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EasyVideoDecoder基于Nvidia(英伟达)独立显卡的解码流程。
EasyVideoDecoder基Nvidia独立显卡的硬解码库EasyNvDecoder
EasyNvDecoder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 EASYNVDECODERAPI_H
#define EASYNVDECODERAPI_H
#include
//++ typedefine start
#ifndef EASYNVDECODER_HANDLE
#define EASYNVDECODER_HANDLE void*
#endif//EASYNVDECODER_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 _EASYNvDecoder_CodecType {
EASYNvDecoder_Codec_MPEG1 = 0, /**< MPEG1 */
EASYNvDecoder_Codec_MPEG2, /**< MPEG2 */
EASYNvDecoder_Codec_MPEG4, /**< MPEG4 */
EASYNvDecoder_Codec_VC1, /**< VC1 */
EASYNvDecoder_Codec_H264, /**< H264 */
EASYNvDecoder_Codec_JPEG, /**< JPEG */
EASYNvDecoder_Codec_H264_SVC, /**< H264-SVC */
EASYNvDecoder_Codec_H264_MVC, /**< H264-MVC */
EASYNvDecoder_Codec_HEVC, /**< HEVC */
EASYNvDecoder_Codec_VP8, /**< VP8 */
EASYNvDecoder_Codec_VP9, /**< VP9 */
EASYNvDecoder_Codec_NumCodecs, /**< Max codecs */
} EASYNvDecoder_CodecType;
typedef enum _EASYNvDecoder_YUVType {
// Uncompressed YUV
EASYNvDecoder_YUV420 = (('I' << 24) | ('Y' << 16) | ('U' << 8) | ('V')), /**< Y,U,V (4:2:0) */
EASYNvDecoder_YV12 = (('Y' << 24) | ('V' << 16) | ('1' << 8) | ('2')), /**< Y,V,U (4:2:0) */
EASYNvDecoder_NV12 = (('N' << 24) | ('V' << 16) | ('1' << 8) | ('2')), /**< Y,UV (4:2:0) */
EASYNvDecoder_YUYV = (('Y' << 24) | ('U' << 16) | ('Y' << 8) | ('V')), /**< YUYV/YUY2 (4:2:2) */
EASYNvDecoder_UYVY = (('U' << 24) | ('Y' << 16) | ('V' << 8) | ('Y')) /**< UYVY (4:2:2) */
} EASYNvDecoder_YUVType;
#ifdef __cplusplus
extern "C"
{
#endif
int EASYNvDecoder_Initsize(std::string &erroStr);
//除非使用低延迟模式，否则请不要使用此标志bLowLatency，但是使用此标志很难获得硬件解码器100%的利用率。
EASYNVDECODER_HANDLE NvDecoder_Create(NvDecoder_CodecType codec, int videoW, int videoH, bool bLowLatency, OutputFormat eOutputFormat, int& errCode, std::string &erroStr);
int NvDecoder_Decode(NVDECODER_HANDLE handle, const uint8_t *pData, int nSize, uint8_t ***pppFrame, int* pnFrameLen, int *pnFrameReturned);
void EASYNvDecoder_Release(NVDECODER_HANDLE handle) ;
int NvDecoder_Uninitsize();
#ifdef __cplusplus
}
#endif
#endif // EASYNVDECODERAPI_H
2. EasyNvDecoder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 EASYNvDecoder_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 < nGpu; i++)
{
CUdevice cuDevice = 0;
ck(cuDeviceGet(&cuDevice, i));
char szDeviceName[128];
ck(cuDeviceGetName(szDeviceName, sizeof(szDeviceName), cuDevice));
LOG(INFO) << "Find Gpu: " << szDeviceName << std::endl;
CUcontext cuContext = NULL;
ck(cuCtxCreate(&cuContext, CU_CTX_SCHED_BLOCKING_SYNC, cuDevice));
m_ctxV.push_back({ cuContext,szDeviceName });
}
isInitsized = true;
m_curIndex = 0;
}
if (m_ctxV.empty()) {
return -1;
}
}
catch (const std::exception& ex)
{
erroStr = ex.what();
std::cout << ex.what();
return -2;
}
return 1;
}
- 第二步，创建解码器实例
EASYNVDECODER_HANDLE EASYNvDecoder_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 < nGpu; i++)
{
if (m_curIndex == i)
{
CUdevice cuDevice = 0;
ck(cuDeviceGet(&cuDevice, i));
char szDeviceName[128];
ck(cuDeviceGetName(szDeviceName, sizeof(szDeviceName), cuDevice));
LOG(INFO) << "Find Gpu: " << szDeviceName << std::endl;
ck(cuCtxCreate(&cuContext, CU_CTX_SCHED_BLOCKING_SYNC, cuDevice));
}
}
//std::pair &v = m_ctxV.at(m_curIndex++ % m_ctxV.size());
//std::cout << "Use Contex in " << v.second << std::endl;
const char *aszChromaFormat[] = { "4:0:0", "4:2:0", "4:2:2", "4:4:4" };
cudaVideoCodec aeCodec[] = { cudaVideoCodec_JPEG, cudaVideoCodec_MPEG1, cudaVideoCodec_MPEG2, cudaVideoCodec_MPEG4, cudaVideoCodec_H264, cudaVideoCodec_HEVC,
cudaVideoCodec_HEVC, cudaVideoCodec_HEVC, cudaVideoCodec_HEVC, cudaVideoCodec_HEVC, cudaVideoCodec_HEVC, cudaVideoCodec_VC1, cudaVideoCodec_VP8,
cudaVideoCodec_VP9, cudaVideoCodec_VP9, cudaVideoCodec_VP9 };
int anBitDepthMinus8[] = { 0, 0, 0, 0, 0, 0, 2, 4, 0, 2, 4, 0, 0, 0, 2, 4 };
cudaVideoChromaFormat aeChromaFormat[] = { cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420,
cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_444, cudaVideoChromaFormat_444,
cudaVideoChromaFormat_444, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420, cudaVideoChromaFormat_420 };
CUVIDDECODECAPS videoDecodeCaps = {};
videoDecodeCaps.eCodecType = (cudaVideoCodec)codec;
videoDecodeCaps.eChromaFormat = cudaVideoChromaFormat_420;
videoDecodeCaps.nBitDepthMinus8 = 0;
for (int i = 0; i < sizeof(aeCodec) / sizeof(aeCodec[0]); i++)
{
if (aeCodec[i] == codec)
{
videoDecodeCaps.eChromaFormat = aeChromaFormat[i];
videoDecodeCaps.nBitDepthMinus8 = anBitDepthMinus8[i];
break;
}
}
errCode = cuvidGetDecoderCaps(&videoDecodeCaps);
if (CUDA_SUCCESS == errCode) { //判断显卡是否支持1080p解码
LOG(INFO) << "cuvid Decoder Caps nMaxWidth " << videoDecodeCaps.nMaxWidth << " nMaxHeigth " << videoDecodeCaps.nMaxHeight << std::endl;
if (!videoDecodeCaps.bIsSupported) {
erroStr = "Codec not supported on this GPU Decoder";
errCode = -1;
}
else
{
//判断是否支持指定格式分辨率视频解码
if (videoDecodeCaps.nMaxWidth >= 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 EASYNvDecoder_Decode(EASYNVDECODER_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) << "nFrameReturned = " <GetVideoInfo();
for (int i = 0; i < nFrameReturned; i++)
{
if (native != pDecoder->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 EASYNvDecoder_Release(EASYNVDECODER_HANDLE handle)
{
if (!handle)
return;
NvDecoder* pDecoder = (NvDecoder*)handle;
pDecoder->Stop();
delete pDecoder;
m_curIndex--;
if (m_curIndex < 0)
m_curIndex = 0;
}
- 第五步，注销解码器，释放资源
int EASYNvDecoder_Uninitsize()
{
isInitsized = false;
for (int nI = 0; nI < m_ctxV.size(); nI++)
{
cuCtxDestroy(m_ctxV[nI].first);
}
m_ctxV.clear();
m_curIndex = 0;
return 1;
}
自此，EasyNvDecoder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：
[email protected]
大家也可以加入EasyPlayer流媒体播放器 QQ群进行讨论：
544917793