[llvm] [AMDGPU] Switch V_CNDMASK operands to shrink it into VOP2 (PR #135162)

[Christudasan Devadasan via llvm-commits]

================
@@ -842,6 +950,8 @@ bool SIShrinkInstructions::run(MachineFunction &MF) {
   unsigned VCCReg = ST->isWave32() ? AMDGPU::VCC_LO : AMDGPU::VCC;
 
   std::vector<unsigned> I1Defs;
+  Register OldVCC = AMDGPU::NoRegister;
----------------
cdevadas wrote:

Skip the initialization. The default value is NoRegister.

https://github.com/llvm/llvm-project/pull/135162