#Encode video nvidia cuda toolkit how to
#Encode video nvidia cuda toolkit driver

cuD3D9UnmapResources – driver will release pointer back toĭ3D9.cudaPostProcessFrame – calls all subsequent CUDA post-processįunctions on that frame, and writes the result directly to the Mapped D3D.cuD3D9ResourceGetMappedPointer – gets a CUDA device pointer.cuvidMapVideoFrame – gets a CUDA device pointer from decoded.ThisĪllows the Video Processor and the Graphics hardware to run asynchronously. This CUDA Video Decoder library makes use two differentĮngines on the GPU, the Video Processor and the Graphics hardware (CUDA and 3D). It is by no meansĪ fully optimized application.

#Encode video nvidia cuda toolkit how to

Intended for simplicity and understanding of how to use this API case.

The CUDA Samples ( cudaDecodeGL and cudaDecodeD3D9) are VideoSource’s decoding thread and the application’s main thread, which is The FrameQueue is used to implement a producer-consumer pattern passingįrames (or better, references to decoded frames) between the Passed by the parser into the FrameQueue object. The sample application simply enqueues the pPicParams The implementation of HandlePictureDisplay() method in The parser calls this method for frames in the (as given to the decoder), and some information relevant for display likeįrame-time, field, etc. The HandlePictureDisplay() method is passed aĬUVIDPARSERDISPINFO struct which contains the necessaryĭata for displaying a frame i.e. Queue becomes available, it simply invokes theĬuvidDecodePicture() function, passing the Sample application waits if the output queue is full. The implementation of HandlePictureDecode() in the This picture index is later used to map the decoded frames to The decoded result gets associated with a picture-index value in theĬUVIDPICPARAMS struct, which is also provided by the Pointers to the video bitstream, information about frame size, flags ifįield or frame, bottom or top field, etc. Necessary for the decoder to decode a frame or field in particular The callback which can be passed straight on to theĬUVIDPICPARAMS struct contains all the information The CUDA VideoParser passes a CUVIDPICPARAMS struct to static int CUDAAPI HandlePictureDisplay( void *pUserData, Called by the video parser to display a video frame (in the case of field // pictures, there may be two decode calls per one display call, since two // fields make up one frame). static int CUDAAPI HandlePictureDecode( void *pUserData, Since the parser will // deliver data as fast as it can, we need to make sure that the picture index // we're attempting to use for decode is no longer used for display. The parser thread calls two callbacks to decode and display frames: // Called by the video parser to decode a single picture. Will need to write your own video source and parsing functions that Handle the decode and display of the frames. The stream, the VideoParser in turn calls back into two callbacks that The source calls aĬallback on the VideoParser class (Windows only) to parse

The decode sample application is driven by the VideoSourceĬlass (Windows only), which spawns its own thread.