[Mlir-commits] [mlir] [mlir][gpu] Add address space modifier to Barrier (PR #110527)

[Fabian Mora]

https://github.com/fabianmcg requested changes to this pull request.

I'm preemptively blocking here because I think there  has not been enough attention on the RFC.

I think this PR is conflating memory fences and thread barriers and as @krzysz00  said `gpu.barrier` is more stronger than a memory fence.
I think the correct approach is adding a new op `gpu.memfence`.

https://github.com/llvm/llvm-project/pull/110527