[llvm] [AMDGPU] Adjust amdgpu fmax/fmin legalization for DAG (PR #203150)

[Matt Arsenault via llvm-commits]

================
@@ -8703,16 +8758,21 @@ SDValue SITargetLowering::lowerFMINNUM_FMAXNUM(SDValue Op,
   const MachineFunction &MF = DAG.getMachineFunction();
   const SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>();
   bool IsIEEEMode = Info->getMode().IEEE;
+  bool IsDivergent = Op.getNode()->isDivergent();
+  bool NeedExpand = (IsDivergent && !Subtarget->hasIEEEMinimumMaximumInsts()) ||
+                    (!IsDivergent && !Subtarget->hasSALUMinimumMaximumInsts());
----------------
arsenm wrote:

Shouldn't need to reconsider the features here. If it was legal it wouldn't reach here 

https://github.com/llvm/llvm-project/pull/203150