[Mlir-commits] [mlir] [mlir][linalg] Fix linalg.select crash with index type operands (PR #179056)

[Renato Golin]

rengolin wrote:

> AI tools (Claude) were used to assist with understanding the codebase and suggesting the implementation approach. The final code was reviewed and verified by me.

Thanks for the clarification. Please make sure you mention this on all your LLM-guided PRs (past, current and future) so that people know what they're dealing with.

However...

> I don't think this is a sufficient fix. The operation is conceptually a select, I don't see why it should be limited to integer+floats or integers+floats+index. It should support any element type that tensor supports.

I agree. This needs better thinking. Look across all the Linalg operations and let's fix the overall problem, not each particular one. Feel free to create an RFC after having done the analysis (hopefully without LLM, so that we know it was complete and accurate), and then we can discuss how we're going to solve it.

https://github.com/llvm/llvm-project/pull/179056