[Mlir-commits] [mlir] [mlir][vector] Sink vector.extract/splat into load/store ops (PR #134389)

[Ivan Butygin]

Hardcode84 wrote:

Side comment from the offline discussion: `vector.load`/`store` currently supports optional OOB sematics while memref ops I will generate doesn't. So, technically, these patterns change semantics of the code, but I believe this semantics is not very useful in the current vague form, we should have an explicit flag in vector ops to enable/disable it. Explicit flag can also be useful during lowering (e.g. for AMDGPU buffer ops you need to explicitly decide if you want enable or disable OOB handling during lowering).

https://github.com/llvm/llvm-project/pull/134389