cudaHostAlloc with Huge TLB support?

any way to use cudaHostAlloc with HugeTLB support? currently cudaHostAlloc is a library call and programmer cannot control more detailed implementation of this API.
but for these applications & operations with large memory region usage on the host, may HUGETLB is a better choice for performance consideration.
anyone have any idea about cudaHostAlloc usage with HUGETLB support?
or NIVDIA will do that.

You don’t actually need to use the host memory allocation methods provided in the CUDA runtime/driver to allocate your host memory…they’re there for convenience (they offer a quick way to do some more advanced types of memory allocation).

I haven’t ever used HugeTLB, but if it’s just for managing host memory, I don’t see why you couldn’t use it with CUDA.

Thanks for your response…

Your are meaning that any CUDA memory allocation APIs can be replaced by the programmer manually? And following reference of these pointers are accepted by CUDA APIs?

Like, with CUDA stream, we need so-called page-locked memory.

Now CUDA API cudaHostAlloc can do that. From your opinion, programmer may use some C lib to do that also, and CUDA Stream accepted it, right?

I’m not trying to code this kind of code right now. But if it is, I get what I want.