[Mlir-commits] [mlir] [mlir][vector] Refine vector.transfer_read hoisting/forwarding (PR #65770)

[Matthias Springer]

https://github.com/matthias-springer approved this pull request.

This fixes hoisting for `collapse_shape`, but it is still broken for other ops that introduce aliasing (e.g., `expand_shape`, `arith.select`, ...). We should query an alias analysis in `noAliasingUseInLoop` instead of hard-coding ops. Still, better than before so I'd say let's land it if it's blocking.


https://github.com/llvm/llvm-project/pull/65770