[PATCH] D80464: [CUDA] Missing __syncthreads intrinsic in __clang_cuda_device_functions.h

[Artem Belevich via Phabricator via cfe-commits]

tra added a comment.

`__syncthreads` is clang's built-in and as such should not be in any header file:
https://github.com/llvm/llvm-project/blob/master/clang/include/clang/Basic/BuiltinsNVPTX.def#L406

My guess is that the root cause of this problem is that source parsing is done using C++, not CUDA and that leads to all sorts of issues.
Also, even if parsing is done as CUDA, in order to work, it will need to point clang to CUDA SDK installation, and that's another bit that's typically missing in various tools that use clang libraries.
Further source of problems will be the fact that CUDA compilation is actually a set of compilations and the tool will need to target appropriate GPU in order to have some GPU-specific builtins available.

In short -- I don't think this patch is needed. I'll chime in on YCM pull request.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D80464/new/

https://reviews.llvm.org/D80464