[Mlir-commits] [mlir] [MLIR][tensor] Improve `tensor.pack` verifier to catch more cases with unconditional runtime errors (PR #77217)

[Han-Chung Wang]

https://github.com/hanhanW approved this pull request.

LGTM, thanks for the patch. Now we should cover all the cases in the verifier.

> Ultimately, this smells like the "in_bounds" attribute that LLVM has on load/store and that we also added to vector.transfer_read/write.

IMO. we don't want to propagate "in_bounds" attribute to high level operations. This looks like a good stop point to me at tensor level, and we handle the case in vectorization. 

https://github.com/llvm/llvm-project/pull/77217