[llvm] InstCombine: Infer nnan/ninf on rounding intrinsics (PR #177770)

[Yingwei Zheng via llvm-commits]

================
@@ -3070,6 +3070,14 @@ Value *InstCombinerImpl::SimplifyDemandedUseFPClass(Instruction *I,
         return Copysign;
       }
 
+      FastMathFlags InferredFMF =
----------------
dtcxzyw wrote:

Can we use `simplifyDemandedFPClassResult`?

https://github.com/llvm/llvm-project/pull/177770