Importing NVMM Memory into a Vulkan Storage Image

Graphics / Linux Linux
,
1

I am currently trying to load memory from an NVMM buffer into a Vulkan Storage Image.

From what I have gathered so far, this is possible to do and I have gone through the CUDA reference manual (1. Introduction — CUDA C Programming Guide), where it is exemplified how it is done.

However I have some doubts.

  • Is importing the NVMM buffer pointer into a cudaImportExternalMemory and subsequently mapping the buffer using cudaExternalMemoryGetMappedBuffer enough to modify the resulting buffer pointer:
checkCudaErrors(cudaImportExternalMemory(&cudaExtMemImageBuffer, &cudaExtMemHandleDesc));
...
...
void* buff = nullptr;
checkCudaErrors(cudaExternalMemoryGetMappedBuffer(&buff, cudaExtMemImageBuffer, &desc));
checkCudaErrors(cudaMemcpy(buff, surface_params->dataPtr, desc.size, cudaMemcpyDeviceToDevice));

i.e Should the cudaMemCpy immediately updated the VkImage

  • Does the size I set in the cudaExternalMemoryBufferDesc reflect the actual size of the VkImage in bytes or something else.

  • Is any of the mipmapping section of the documentation necessary in order for a successful memory copy to occur.

I understand these questions may have obvious answers, but I would greatly appreciate a hint towards the right direction.

Thank you for your time,
Francisco