[Mlir-commits] [mlir] [mlir][gpu] Sink integer arithmetic into gpu.launch (PR #221462)

[Fabian Mora]

https://github.com/fabianmcg requested changes to this pull request.

Thank you for the PR. If you used AI, please remember to disclose its usage.

With respect to the patch, the above is not an issue in the pass. Sinking all addi, muli, subi, without restrictions is not a good policy and should be left to a dedicated pass to analyze the cost of sinking the ops. The reason we sink constants and some other ops, is that they can easily get hoisted (constants) or the cost of computing in the device is acceptable compared to wasting a kernel argument (memref.dim).

https://github.com/llvm/llvm-project/pull/221462