[PATCH] D76365: [cuda][hip] Add CUDA builtin surface/texture reference support.

Thu Mar 19 17:00:18 PDT 2020

hliao added a comment.

In D76365#1932398 <https://reviews.llvm.org/D76365#1932398>, @tra wrote:

> I believe LLVM does have `nvvm.texsurf.handle` implemented: https://github.com/llvm/llvm-project/blob/d9972f848294b06807c8764615852ba2bc1e8a74/llvm/include/llvm/IR/IntrinsicsNVVM.td#L1150

This one only adds the definition but NVPTX backend doesn't handle it.

> We also appear to have some plumbing for it in clang: https://github.com/llvm/llvm-project/blob/31262d6722c7ae6a9966a76064af43e5b3a8df71/llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp#L724

Yeah, that's so far the `internal` version is used. The original one with one metadata as parameter that's only used to prevent CSE as the handle loading should be not optimized away or difficult for the backend to handle it. We should be able to add that intrinsic support easily. I could add that later. That should not be a big issue.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D76365/new/

https://reviews.llvm.org/D76365