bool VideoDecoderCUDAPrivate::processDecodedData(CUVIDPARSERDISPINFO *cuviddisp, VideoFrame* outFrame) {
int num_fields = cuviddisp->progressive_frame ? 1 : 2+cuviddisp->repeat_first_field;
for (int active_field = 0; active_field < num_fields; ++active_field) {
CUVIDPROCPARAMS proc_params;
memset(&proc_params, 0, sizeof(CUVIDPROCPARAMS));
proc_params.progressive_frame = cuviddisp->progressive_frame; //check user config
proc_params.second_field = active_field == 1; //check user config
proc_params.top_field_first = cuviddisp->top_field_first;
proc_params.unpaired_field = cuviddisp->progressive_frame == 1;
CUdeviceptr devptr;
unsigned int pitch;
cuvidCtxLock(vid_ctx_lock, 0);
CUresult cuStatus = cuvidMapVideoFrame(dec, cuviddisp->picture_index, &devptr, &pitch, &proc_params);
if (cuStatus != CUDA_SUCCESS) {
qWarning("cuvidMapVideoFrame failed on index %d (%#x, %s)", cuviddisp->picture_index, cuStatus, _cudaGetErrorEnum(cuStatus));
cuvidUnmapVideoFrame(dec, devptr);
cuvidCtxUnlock(vid_ctx_lock, 0);
return false;
}
#define PAD_ALIGN(x,mask) ( (x + mask) & ~mask )
//uint w = dec_create_info.ulWidth;//PAD_ALIGN(dec_create_info.ulWidth, 0x3F);
uint h = dec_create_info.ulHeight;//PAD_ALIGN(dec_create_info.ulHeight, 0x0F); //?
#undef PAD_ALIGN
int size = pitch*h*3/2;
if (size > host_data_size && host_data) {
cuMemFreeHost(host_data);
host_data = 0;
host_data_size = 0;
}
if (!host_data) {
cuStatus = cuMemAllocHost((void**)&host_data, size);
if (cuStatus != CUDA_SUCCESS) {
qWarning("cuMemAllocHost failed (%#x, %s)", cuStatus, _cudaGetErrorEnum(cuStatus));
cuvidUnmapVideoFrame(dec, devptr);
cuvidCtxUnlock(vid_ctx_lock, 0);
return false;
}
host_data_size = size;
}
if (!host_data) {
qWarning("No valid staging memory!");
cuvidUnmapVideoFrame(dec, devptr);
cuvidCtxUnlock(vid_ctx_lock, 0);
return false;
}
cuStatus = cuMemcpyDtoHAsync(host_data, devptr, size, stream);
if (cuStatus != CUDA_SUCCESS) {
qWarning("cuMemcpyDtoHAsync failed (%#x, %s)", cuStatus, _cudaGetErrorEnum(cuStatus));
cuvidUnmapVideoFrame(dec, devptr);
cuvidCtxUnlock(vid_ctx_lock, 0);
return false;
}
cuStatus = cuCtxSynchronize();
if (cuStatus != CUDA_SUCCESS) {
qWarning("cuCtxSynchronize failed (%#x, %s)", cuStatus, _cudaGetErrorEnum(cuStatus));
}
cuvidUnmapVideoFrame(dec, devptr);
cuvidCtxUnlock(vid_ctx_lock, 0);
//qDebug("mark not in use pic_index: %d", cuviddisp->picture_index);
surface_in_use[cuviddisp->picture_index] = false;
uchar *planes[] = {
host_data,
host_data + pitch * h
};
int pitches[] = { (int)pitch, (int)pitch };
VideoFrame frame(codec_ctx->width, codec_ctx->height, VideoFormat::Format_NV12);
frame.setBits(planes);
frame.setBytesPerLine(pitches);
//TODO: is clone required? may crash on clone, I should review clone()
//frame = frame.clone();
if (outFrame) {
*outFrame = frame.clone();
}
#if COPY_ON_DECODE
frame_queue.put(frame.clone());
#endif
//qDebug("frame queue size: %d", frame_queue.size());
}
return true;
}
void
VideoDecoder::unmapFrame(CUdeviceptr pDevice)
{
CUresult oResult = cuvidUnmapVideoFrame(oDecoder_, pDevice);
assert(CUDA_SUCCESS == oResult);
}
void unmapFrame(cuda::GpuMat& frame)
{
cuSafeCall( cuvidUnmapVideoFrame(decoder_, (CUdeviceptr) frame.data) );
frame.release();
}
/*
* 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;
}
