[llvm] [AMDGPU][True16][MC] support more VOP3 inst in true16/fake16 format (PR #113603)

[Matt Arsenault via llvm-commits]

================
@@ -7412,6 +7414,28 @@ void SIInstrInfo::moveToVALUImpl(SIInstrWorklist &Worklist,
     Inst.eraseFromParent();
     return;
   }
+  case AMDGPU::S_MINIMUM_F16:
+  case AMDGPU::S_MAXIMUM_F16: {
+    const DebugLoc &DL = Inst.getDebugLoc();
+    Register NewDst;
+    if (ST.useRealTrue16Insts())
+      NewDst = MRI.createVirtualRegister(&AMDGPU::VGPR_16RegClass);
+    else
+      NewDst = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
----------------
arsenm wrote:

Unconditional MRI.createVirtualRegister call with conditional input class 

https://github.com/llvm/llvm-project/pull/113603