[llvm] [KnownFPClass] Fix fsub input denormal handling (PR #225605)

[Eric Ross via llvm-commits]

================
@@ -709,7 +721,7 @@ KnownFPClass KnownFPClass::fma(const KnownFPClass &KnownLHS,
   //
   // If the multiply is a -0 due to rounding, the final -0 + 0 will be -0,
   // unlike for a separate fadd.
-  KnownFPClass Known = fadd_impl(Mul, KnownAddend, Mode);
+  KnownFPClass Known = fadd_impl(Mul, KnownAddend, /*IsFMA=*/true);
----------------
ZERICO2005 wrote:

Since `DenormalMode` was unused in `fadd_impl`, the behavior of `KnownFPClass::fma` remains unchanged.


https://github.com/llvm/llvm-project/pull/225605