[Mlir-commits] [mlir] [mlir][arith] fix wrong floordivsi fold (#83079) (PR #83248)

[llvmlistbot at llvm.org]

lipracer wrote:

> This change almost certainly has performance implications. Do we have some way to measure or even guess at that? If this is fixing a correctness bug, I am also not the right person to verify that, but I trust the author in this

Yes, this may indeed affect performance. This implementation is based on tensorflow. For performance considerations, I think we can have a compilation option similar to ‘fast-expand’ for expansion. This change is only consistent with the behavior of llvm.

https://github.com/llvm/llvm-project/pull/83248