bool VideoDecoderCUDAPrivate::releaseCuda()
{
available = false;
if (!can_load)
return true;
if (dec) {
cuvidDestroyDecoder(dec);
dec = 0;
}
if (parser) {
cuvidDestroyVideoParser(parser);
parser = 0;
}
if (stream) {
cuStreamDestroy(stream);
stream = 0;
}
if (host_data) {
cuMemFreeHost(host_data);
host_data = 0;
host_data_size = 0;
}
if (vid_ctx_lock) {
cuvidCtxLockDestroy(vid_ctx_lock);
vid_ctx_lock = 0;
}
if (cuctx) {
checkCudaErrors(cuCtxDestroy(cuctx));
}
// TODO: dllapi unload
return true;
}
void CudaVideoRender::terminateCudaVideo(bool bDestroyContext)
{
if (m_pVideoParser) delete m_pVideoParser;
if (m_pVideoDecoder) delete m_pVideoDecoder;
if (m_pVideoSource) delete m_pVideoSource;
if (m_pFrameQueue) delete m_pFrameQueue;
if (m_CtxLock) {
cutilDrvSafeCallNoSync( cuvidCtxLockDestroy(m_CtxLock) );
}
if (m_cuContext && bDestroyContext) {
cutilDrvSafeCallNoSync( cuCtxDestroy(m_cuContext) );
m_cuContext = NULL;
}
if (m_ReadbackSID) cuStreamDestroy(m_ReadbackSID);
if (m_KernelSID) cuStreamDestroy(m_KernelSID);
}
void
freeCudaResources(bool bDestroyContext)
{
if (g_pVideoParser)
{
delete g_pVideoParser;
}
if (g_pVideoDecoder)
{
delete g_pVideoDecoder;
}
if (g_pVideoSource)
{
delete g_pVideoSource;
}
if (g_pFrameQueue)
{
delete g_pFrameQueue;
}
if (g_CtxLock)
{
checkCudaErrors(cuvidCtxLockDestroy(g_CtxLock));
}
if (g_oContext && bDestroyContext)
{
checkCudaErrors(cuCtxDestroy(g_oContext));
g_oContext = NULL;
}
if (g_ReadbackSID)
{
cuStreamDestroy(g_ReadbackSID);
}
if (g_KernelSID)
{
cuStreamDestroy(g_KernelSID);
}
}
bool VideoDecoderCUDAPrivate::releaseCuda()
{
if (!can_load)
return true;
if (dec) {
cuvidDestroyDecoder(dec);
dec = 0;
}
if (parser) {
cuvidDestroyVideoParser(parser);
parser = 0;
}
if (stream) {
cuStreamDestroy(stream);
stream = 0;
}
cuvidCtxLockDestroy(vid_ctx_lock);
if (cuctx) {
checkCudaErrors(cuCtxDestroy(cuctx));
}
return true;
}
/*
* 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;
}
