[llvm] [AMDGPU]:: Minor Unpacking Fixes. (PR #163992)

[Jeffrey Byrnes via llvm-commits]

================
@@ -612,10 +592,12 @@ void SIPreEmitPeephole::collectUnpackingCandidates(
 
   for (auto I = std::next(BeginMI.getIterator()); I != E; ++I) {
     MachineInstr &Instr = *I;
+    uint16_t UnpackedOpCode = mapToUnpackedOpcode(Instr);
     if (Instr.isMetaInstruction())
       continue;
     if ((Instr.isTerminator()) ||
-        (TII->isNeverCoissue(Instr) && !isUnpackingSupportedInstr(Instr)) ||
+        (TII->isNeverCoissue(Instr) &&
+         (UnpackedOpCode == std::numeric_limits<uint16_t>::max())) ||
----------------
jrbyrnes wrote:

Can you use a local variable -- `IsUnpackable = UnpackedOpCode == std::numeric_limits<uint16_t>::max()`

https://github.com/llvm/llvm-project/pull/163992