lhutton1 wrote: > not directly related to this PR but is there any way to propagate the static shapes to reciprocal in the example you shared Not that I'm aware of currently, but we're looking into ways to resolve this https://github.com/llvm/llvm-project/pull/171156