[llvm] [AMDGPU][True16] si-fold-operand selecting srcidx for v_mov_b16_t16_e64 (PR #162101)

[Brox Chen via llvm-commits]

================
@@ -932,7 +932,8 @@ static MachineOperand *lookUpCopyChain(const SIInstrInfo &TII,
   for (MachineInstr *SubDef = MRI.getVRegDef(SrcReg);
        SubDef && TII.isFoldableCopy(*SubDef);
        SubDef = MRI.getVRegDef(Sub->getReg())) {
-    unsigned SrcIdx = TII.getFoldableCopySrcIdx(*SubDef);
+    const int SrcIdx =
----------------
broxigarchen wrote:

I did a quick check, actually at this stage we always create a e64 inst, and actually cannot create a e32 inst since the dest reg of that t16_e32 is not allocatable. Thus the point 1 in the description is more cosmetic

https://github.com/llvm/llvm-project/pull/162101