CUDA runtime don't have api such as cudaMemcpyArrayToArrayAsync

Hi, NvTeam,
I’m a auto-driveing software developer on OrinX. I am developing a software that includes model inference, image processing, and NVENC. Since the NVENC needs to use cudaArray as input, I output an NV12 image in cudaArray format. At the same time, since I should not occupy the NvSCI buffer queue for a long time, in my pipeline, I designed to use cudaMemcpyArrayToArray to copy the processed image to NVENC’s input buffer. However, I found that there was no corresponding Async interface for this api, which forced me to use the default cuda stream, which prevented me from performing as quickly as I should in a competitive deployment scenario.

Hi,

Just want to confirm first. Do you use Jetson Orin or DRIVE Orin?

NVENC uses NvBufSurface rather than cudaArray.
Could you share a sample with us so we can known more about your usecase?

Thanks.