[llvm] SelectionDAG: Improve expandFMINIMUM_FMAXIMUM (PR #137367)

[Matt Arsenault via llvm-commits]

================
@@ -8741,6 +8741,28 @@ SDValue TargetLowering::expandFMINNUM_FMAXNUM(SDNode *Node,
   return SDValue();
 }
 
+static SDValue determineFloatSign(SDValue N, SelectionDAG &DAG, bool Positive) {
+  SDLoc DL(N);
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  EVT VT = N->getValueType(0);
+  EVT IntVT = VT.changeTypeToInteger();
+  SDValue NTrunc = N;
+  if (!TLI.isTypeLegal(IntVT)) {
+    EVT FloatVT = VT.changeElementType(MVT::f32);
----------------
arsenm wrote:

Hardcoding this to i32/f32 doesn't feel right. Can you find the next smallest FP legal FP type?

https://github.com/llvm/llvm-project/pull/137367