[llvm] [AMDGPU][GlobalISel] Lower G_FMINIMUM and G_FMAXIMUM (PR #151122)

[Matt Arsenault via llvm-commits]

================
@@ -8294,6 +8297,75 @@ LegalizerHelper::lowerFMinNumMaxNum(MachineInstr &MI) {
   return Legalized;
 }
 
+LegalizerHelper::LegalizeResult
+LegalizerHelper::lowerFMinimumMaximum(MachineInstr &MI) {
+  unsigned Opc = MI.getOpcode();
+  auto [Dst, Src0, Src1] = MI.getFirst3Regs();
+  LLT Ty = MRI.getType(Dst);
+  LLT CmpTy =
+      Ty.isScalar() ? LLT::scalar(1) : LLT::vector(Ty.getElementCount(), 1);
----------------
arsenm wrote:

```suggestion
  LLT CmpTy = Ty.changeElementSize(1);
```

https://github.com/llvm/llvm-project/pull/151122