We have two way to get a page-lock memory with the type of portable and mapped, one is cudaHostAlloc() with the responding flags, another is cudaHostRegister(). For example , if i want to get a portable page-lock memory, i can call cudaHostAlloc(ptr, size, cudaHostAllocPortable); also i can malloc a memory, then call the cudaHostRegister(ptr, size, cudaHostRegisterPortbale). But for the type of write-combined, i only find one way that is cudaHostAlloc(ptr, size, cudaHostAllocWriteCombined), does the cudaHostRegister() api support write-combined? If support, how can i get a write-combined page-lock memory by this way?
There isn’t any way to request WC memory with
cudaHostRegister(). If you would like to see a change in CUDA, you can file a bug.