[llvm] APFloat: Fix signed zero handling in minnum/maxnum (PR #83376)

[Jay Foad via llvm-commits]

================
@@ -1397,6 +1397,8 @@ inline APFloat minnum(const APFloat &A, const APFloat &B) {
     return B;
   if (B.isNaN())
     return A;
+  if (A.isZero() && B.isZero() && (A.isNegative() != B.isNegative()))
----------------
jayfoad wrote:

IIUC this function now implements IEEE 2019 minimumNumber which is a refinement of IEEE 2008 minNum. Should update the doxygen comment accordingly.

https://github.com/llvm/llvm-project/pull/83376