[Mlir-commits] [mlir] [MLIR] Implement emulation of static indexing subbyte type vector stores (PR #115922)

[llvmlistbot at llvm.org]

lialan wrote:

> What would happen if the start of the store is aligned? Would it still generate atomic stores?

It will be operation as usual, the atomic and the extra handling only happen when needed.

> I havent looked into the details yet, but we only need an atomic store if we cannot guarantee that there are no competing stores. This information we canno thave at this level, but a caller might have this information. I think it might be better to allow for a caller to indicate that the atomic stores are not needed. The default can be that you do the atomic stores.

Feel like in such a non-competing case we can do strength reduction and use masked store instead of atomic for the unaligned parts (the beginning and the end byte if it is unaligned). 

https://github.com/llvm/llvm-project/pull/115922