void
initCudaVideo()
{
// bind the context lock to the CUDA context
CUresult result = cuvidCtxLockCreate(&g_CtxLock, g_oContext);
if (result != CUDA_SUCCESS)
{
printf("cuvidCtxLockCreate failed: %d\n", result);
assert(0);
}
std::auto_ptr<VideoDecoder> apVideoDecoder(new VideoDecoder(g_pVideoSource->format(), g_oContext, g_eVideoCreateFlags, g_CtxLock));
std::auto_ptr<VideoParser> apVideoParser(new VideoParser(apVideoDecoder.get(), g_pFrameQueue));
g_pVideoSource->setParser(*apVideoParser.get());
g_pVideoParser = apVideoParser.release();
g_pVideoDecoder = apVideoDecoder.release();
// Create a Stream ID for handling Readback
if (g_bReadback)
{
checkCudaErrors(cuStreamCreate(&g_ReadbackSID, 0));
checkCudaErrors(cuStreamCreate(&g_KernelSID, 0));
printf("> initCudaVideo()\n");
printf(" CUDA Streams (%s) <g_ReadbackSID = %p>\n", ((g_ReadbackSID == 0) ? "Disabled" : "Enabled"), g_ReadbackSID);
printf(" CUDA Streams (%s) <g_KernelSID = %p>\n", ((g_KernelSID == 0) ? "Disabled" : "Enabled"), g_KernelSID);
}
}
void CudaVideoRender::initCudaVideo( )
{
// bind the context lock to the CUDA context
CUresult result = cuvidCtxLockCreate(&m_CtxLock, m_cuContext);
if (result != CUDA_SUCCESS) {
printf("cuvidCtxLockCreate failed: %d\n", result);
assert(0);
}
std::auto_ptr<VideoDecoder> apVideoDecoder(new VideoDecoder(m_pVideoSource->format(), m_cuContext, m_eVideoCreateFlags, m_CtxLock));
std::auto_ptr<VideoParser> apVideoParser(new VideoParser(apVideoDecoder.get(), m_pFrameQueue));
m_pVideoSource->setParser(*apVideoParser.get());
m_pVideoParser = apVideoParser.release();
m_pVideoDecoder = apVideoDecoder.release();
// Create a Stream ID for handling Readback
if (m_bReadback) {
cutilDrvSafeCallNoSync( cuStreamCreate(&m_ReadbackSID, 0) );
cutilDrvSafeCallNoSync( cuStreamCreate(&m_KernelSID, 0) );
printf("> initCudaVideo()\n");
printf(" CUDA Streams (%s) <m_ReadbackSID = %p>\n", ((m_ReadbackSID == 0) ? "Disabled" : "Enabled"), m_ReadbackSID );
printf(" CUDA Streams (%s) <m_KernelSID = %p>\n", ((m_KernelSID == 0) ? "Disabled" : "Enabled"), m_KernelSID );
}
}
bool VideoDecoderCUDAPrivate::initCuda()
{
CUresult result = cuInit(0);
if (result != CUDA_SUCCESS) {
available = false;
qWarning("cuInit(0) faile (%d)", result);
return false;
}
cudev = GetMaxGflopsGraphicsDeviceId();
int clockRate;
cuDeviceGetAttribute(&clockRate, CU_DEVICE_ATTRIBUTE_CLOCK_RATE, cudev);
int major, minor;
cuDeviceComputeCapability(&major, &minor, cudev);
char devname[256];
cuDeviceGetName(devname, 256, cudev);
description = QString("CUDA device: %1 %2.%3 %4 MHz").arg(devname).arg(major).arg(minor).arg(clockRate/1000);
//TODO: cuD3DCtxCreate > cuGLCtxCreate > cuCtxCreate
checkCudaErrors(cuCtxCreate(&cuctx, CU_CTX_SCHED_BLOCKING_SYNC, cudev)); //CU_CTX_SCHED_AUTO?
CUcontext cuCurrent = NULL;
result = cuCtxPopCurrent(&cuCurrent);
if (result != CUDA_SUCCESS) {
qWarning("cuCtxPopCurrent: %d\n", result);
return false;
}
checkCudaErrors(cuvidCtxLockCreate(&vid_ctx_lock, cuctx));
{
AutoCtxLock lock(this, vid_ctx_lock);
Q_UNUSED(lock);
//Flags- Parameters for stream creation (must be 0 (CU_STREAM_DEFAULT=0 in cuda5) in cuda 4.2, no CU_STREAM_NON_BLOCKING)
checkCudaErrors(cuStreamCreate(&stream, 0));//CU_STREAM_NON_BLOCKING)); //CU_STREAM_DEFAULT
//require compute capability >= 1.1
//flag: Reserved for future use, must be 0
//cuStreamAddCallback(stream, CUstreamCallback, this, 0);
}
return true;
}
cv::gpu::VideoReader_GPU::Impl::Impl(const cv::Ptr<VideoSource>& source) :
videoSource_(source),
lock_(0)
{
// init context
GpuMat temp(1, 1, CV_8UC1);
temp.release();
DeviceInfo devInfo;
CV_Assert( devInfo.supports(FEATURE_SET_COMPUTE_11) );
CUcontext ctx;
cuSafeCall( cuCtxGetCurrent(&ctx) );
cuSafeCall( cuvidCtxLockCreate(&lock_, ctx) );
frameQueue_.reset(new detail::FrameQueue);
videoDecoder_.reset(new detail::VideoDecoder(videoSource_->format(), lock_));
videoParser_.reset(new detail::VideoParser(videoDecoder_.get(), frameQueue_.get()));
videoSource_->setFrameQueue(frameQueue_.get());
videoSource_->setVideoParser(videoParser_.get());
videoSource_->start();
}
/*
* Main transcoding thread
* Initializes CUDA device, decodes frames with NVCUVID API and adds them to frame queue, which passes them to NVENC for encoding, then output
*/
void NVENCGUI::Transcode()
{
CUresult result;
// initialize CUDA
result = cuInit(0);
if (result != CUDA_SUCCESS)
{
emit Error(ERR_CUDA_INIT);
return;
}
NVENCSTATUS nvStatus = NV_ENC_SUCCESS;
// no input file
if (encodeConfig.inputFileName == NULL)
{
emit Error(ERR_INPUT);
return;
}
// no output file
if (encodeConfig.outputFileName == NULL)
{
emit Error(ERR_OUTPUT);
return;
}
// unable to open input file
if (!fopen(encodeConfig.inputFileName, "r"))
{
emit Error(ERR_INPUT);
return;
}
encodeConfig.fOutput = fopen(encodeConfig.outputFileName, "wb");
// unable to open output file
if (encodeConfig.fOutput == NULL)
{
emit Error(ERR_OUTPUT);
return;
}
// initialize CUDA on device and set CUDA context
CUcontext cudaCtx;
CUdevice device;
result = cuDeviceGet(&device, encodeConfig.deviceID);
if (result != CUDA_SUCCESS)
{
emit Error(ERR_CUDA_DEVICE);
return;
}
result = cuCtxCreate(&cudaCtx, CU_CTX_SCHED_AUTO, device);
if (result != CUDA_SUCCESS)
{
emit Error(ERR_CUDA_CTX);
return;
}
// initialize NVCUVID context
CUcontext curCtx;
CUvideoctxlock ctxLock;
result = cuCtxPopCurrent(&curCtx);
if (result != CUDA_SUCCESS)
{
emit Error(ERR_CUDA_CTX);
return;
}
result = cuvidCtxLockCreate(&ctxLock, curCtx);
if (result != CUDA_SUCCESS)
{
emit Error(ERR_CUDA_CTX);
return;
}
CudaDecoder* pDecoder = new CudaDecoder;
FrameQueue* pFrameQueue = new CUVIDFrameQueue(ctxLock);
pDecoder->InitVideoDecoder(encodeConfig.inputFileName, ctxLock, pFrameQueue, encodeConfig.width, encodeConfig.height);
int decodedW, decodedH, decodedFRN, decodedFRD;
pDecoder->GetCodecParam(&decodedW, &decodedH, &decodedFRN, &decodedFRD);
// If the width/height is not set, set to same as source
if (encodeConfig.width <= 0 || encodeConfig.height <= 0) {
encodeConfig.width = decodedW;
encodeConfig.height = decodedH;
}
// same, except for fps
if (encodeConfig.fps <= 0) {
if (decodedFRN <= 0 || decodedFRD <= 0)
encodeConfig.fps = 30;
else
encodeConfig.fps = decodedFRN / decodedFRD;
}
// initialize frame queue with width/height
pFrameQueue->init(encodeConfig.width, encodeConfig.height);
VideoEncoder* pEncoder = new VideoEncoder(ctxLock);
assert(pEncoder->GetHWEncoder());
// initialize NVENC HW Encoder
nvStatus = pEncoder->GetHWEncoder()->Initialize(cudaCtx, NV_ENC_DEVICE_TYPE_CUDA);
if (nvStatus != NV_ENC_SUCCESS)
{
emit Error(ERR_NVENC_ENC_INIT);
return;
}
// get preset GUID
encodeConfig.presetGUID = pEncoder->GetHWEncoder()->GetPresetGUID(encodeConfig.encoderPreset, encodeConfig.codec);
// create encoder
nvStatus = pEncoder->GetHWEncoder()->CreateEncoder(&encodeConfig);
if (nvStatus != NV_ENC_SUCCESS)
{
emit Error(ERR_NVENC_ENC_CREATE);
return;
}
// create buffer
nvStatus = pEncoder->AllocateIOBuffers(&encodeConfig);
if (nvStatus != NV_ENC_SUCCESS)
{
emit Error(ERR_NVENC_ENC_BUFFER);
return;
}
// print details to text window, start counter
emit PrintDetails();
NvQueryPerformanceCounter(&results.lStart);
//start decoding thread
#ifdef _WIN32
HANDLE decodeThread = CreateThread(NULL, 0, DecodeProc, (LPVOID)pDecoder, 0, NULL);
#else
pthread_t pid;
pthread_create(&pid, NULL, DecodeProc, (void*)pDecoder);
#endif
int encodedFrames = 0;
//start encoding thread
while (!(pFrameQueue->isEndOfDecode() && pFrameQueue->isEmpty()))
{
CUVIDPARSERDISPINFO pInfo;
if (pFrameQueue->dequeue(&pInfo))
{
CUdeviceptr dMappedFrame = 0;
unsigned int pitch;
CUVIDPROCPARAMS oVPP = { 0 };
oVPP.unpaired_field = 1;
oVPP.progressive_frame = 1;
cuvidMapVideoFrame(pDecoder->GetDecoder(), pInfo.picture_index, &dMappedFrame, &pitch, &oVPP);
EncodeFrameConfig stEncodeConfig = { 0 };
stEncodeConfig.dptr = dMappedFrame;
stEncodeConfig.pitch = pitch;
stEncodeConfig.width = encodeConfig.width;
stEncodeConfig.height = encodeConfig.height;
pEncoder->EncodeFrame(&stEncodeConfig);
cuvidUnmapVideoFrame(pDecoder->GetDecoder(), dMappedFrame);
pFrameQueue->releaseFrame(&pInfo);
//emit IncrementEncodedFrames();
}
}
// flush
pEncoder->EncodeFrame(NULL, true);
// end decoding thread
#ifdef _WIN32
WaitForSingleObject(decodeThread, INFINITE);
#else
pthread_join(pid, NULL);
#endif
// print transcoding details
if (pEncoder->GetEncodedFrames() > 0)
{
results.decodedFrames = pDecoder->m_decodedFrames;
results.encodedFrames = pEncoder->GetEncodedFrames();
NvQueryPerformanceCounter(&results.lEnd);
NvQueryPerformanceFrequency(&results.lFreq);
results.elapsedTime = (double)(results.lEnd - results.lStart) / (double)results.lFreq;
}
emit TranscodingEnd();
// clean up
cuvidCtxLockDestroy(ctxLock);
pEncoder->Deinitialize();
delete pDecoder;
delete pEncoder;
delete pFrameQueue;
result = cuCtxDestroy(cudaCtx);
if (result != CUDA_SUCCESS)
{
emit Error(ERR_CUDA_CTX_DESTROY);
return;
}
return;
}
