How to use device memory by cuda runtime API when decode video?

I use the Xavier device for video decoding, and want to use cuda runtime API do somethings. How do I get device memory pointer by DMABUF FD of NvBuffer.

We have the demo code in jetson_multimedia_api samples. Please take a look at cuda_postprocess() in


I have the same question, and I want to avoid performance loss.
Does the device memory have to be accessed by EGLImage? Is there any other direct way?

For getting data pointer to the CUDA buffer, it has to be accessed by EGLImage, as demonstrated in cuda_postprocess().