[Mlir-commits] [mlir] [mlir][vector] Add verification for incorrect vector.extract (PR #115824)

[Andrzej Warzyński]

banach-space wrote:

> This limitation is present in SPIR-V and causes a lot of pain when lowering to SRIR-V.

Could you elaborate? It’s not entirely clear whether you mean the challenges with handling 0-D vectors themselves or the fact that 0-D vectors are not fully supported. A bit more detail would help me understand your perspective better.

> I'd direct you to ArithToSPIRV and VectorToSPIRV specifically: you can see a lot of if else statements around the operand/result types to account for this limitation.

I’d really appreciate a specific pointer to the code or examples you find problematic. That way, I can look directly at the pain points you're highlighting instead of guessing. I’m not familiar with those specific parts of the codebase yet, so thanks for bearing with me. :)

No rush!

https://github.com/llvm/llvm-project/pull/115824