[llvm-branch-commits] [llvm] ValueTracking: Improve handling for fma/fmuladd (PR #175614)

[Matt Arsenault via llvm-branch-commits]

================
@@ -332,6 +332,19 @@ KnownFPClass KnownFPClass::fmul(const KnownFPClass &KnownLHS,
   return Known;
 }
 
+KnownFPClass KnownFPClass::fma(const KnownFPClass &KnownLHS,
+                               const KnownFPClass &KnownRHS,
+                               const KnownFPClass &KnownAddend,
+                               DenormalMode Mode) {
+  return fadd(fmul(KnownLHS, KnownRHS, Mode), KnownAddend, Mode);
----------------
arsenm wrote:

The rounding can't change the possible class result. All we can get out of this is nan and sign bit handling which will be the same 

https://github.com/llvm/llvm-project/pull/175614