[Mlir-commits] [mlir] [AMDGPU] fold `memref.subview` into `amdgpu.gather_to_lds` (PR #149851)

[Krzysztof Drewniak]

krzysz00 wrote:

Ok, given the general consensus from everyone else ... I can see why people are trying to avoid the build system coupling issues. Let's go with adding an AMDGPU-specific pass and populate function, especially since there's already precedent for that sort of split.

We can refactor this later and it'll provide good motivation for an interface

https://github.com/llvm/llvm-project/pull/149851