[flang-commits] [flang] [llvm] [mlir] [NFC][mlir] move NVVM dialect to mlir/lib/Dialect/NVVM/IR (PR #206570)

[Krzysztof Drewniak via flang-commits]

krzysz00 wrote:

> I'm not sure I'm totally onboard with the current direction: these dialects were intentionally positioned as sub-dialects of LLVM IR: they actually model LLVM intrinsics as dialect and as such are part of LLVM IR modeling.

On the other hand, LLVM itself lets you skip targets you're not going to use. Sure, that leaves the intrinsic definitions in, but I can very much agree that if you're not targetting NVVM/AMDGPU/... you shouldn't be getting an OptimizeForX.

https://github.com/llvm/llvm-project/pull/206570