[llvm] [InstSimplify] Fold poison/nnan/ninf on @llvm.sqrt (PR #141821)

[Nikita Popov via llvm-commits]

nikic wrote:

> I think fixed length vectors would get their poison propagated if we extended the case in ConstantFoldScalarCall1.

Yes, fixed vectors are folded element-wise.

> Maybe we should look into constant scalable vector intrinsics with splatted operands instead then?

That would make sense to me. Surprised this isn't already the case...

https://github.com/llvm/llvm-project/pull/141821