RenameIndependentSubregs: Fix infinite loop

[arsenm]

Summary

Apparently this replacement can really be substituting the
same as the original register. Avoid restarting the loop
when there's been no change in the register uses.

[MatzeB]

LGTM

[arsenm]

r306441

lib/CodeGen/RenameIndependentSubregs.cpp

for (const SubRangeInfo &SRInfo : SubRangeInfos) {
       unsigned LocalID = SRInfo.ConEQ.getEqClass(VNI);
       // Global ID
       ID = Classes[LocalID + SRInfo.Index];
       break;
     }
 
     unsigned VReg = Intervals[ID]->reg;
     MO.setReg(VReg);
-    if (MO.isTied()) {
+
+    if (MO.isTied() && Reg != VReg) {
       /// Undef use operands are not tracked in the equivalence class but need
       /// to be update if they are tied.
       MO.getParent()->substituteRegister(Reg, VReg, 0, TRI);
 
       // substituteRegister breaks the iterator, so restart.
       I = MRI->reg_nodbg_begin(Reg);
     }
   }

test/CodeGen/AMDGPU/rename-independent-subregs-invalid-mac-operands.mir

bb.2:
     BUFFER_STORE_DWORD_OFFEN %6.sub2, %0, %sgpr0_sgpr1_sgpr2_sgpr3, %sgpr4, 8, 0, 0, 0, implicit %exec
     BUFFER_STORE_DWORD_OFFEN %6.sub1, %0, %sgpr0_sgpr1_sgpr2_sgpr3, %sgpr4, 4, 0, 0, 0, implicit %exec
     BUFFER_STORE_DWORD_OFFEN %6.sub0, %0, %sgpr0_sgpr1_sgpr2_sgpr3, %sgpr4, 0, 0, 0, 0, implicit %exec
     %sgpr30_sgpr31 = COPY %5
     %sgpr5 = COPY %sgpr5
     S_SETPC_B64_return %sgpr30_sgpr31, implicit %sgpr5
 
 ...
+
+# GCN-LABEL: name: inf_loop_tied_operand
+# GCN: bb.0:
+# GCN-NEXT: undef %2.sub0 = V_MAC_F32_e32 1073741824, undef %0, undef %2.sub0, implicit %exec
+# GCN-NEXT: dead undef %3.sub1 = COPY %2.sub0
+
+name:            inf_loop_tied_operand
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: vgpr_32, preferred-register: '' }
+  - { id: 1, class: vgpr_32, preferred-register: '' }
+  - { id: 2, class: vreg_128, preferred-register: '' }
+body:             |
+  bb.0:
+    %1 = V_MAC_F32_e32 1073741824, undef %0, undef %1, implicit %exec
+    undef %2.sub0 = COPY %1
+    %2.sub1 = COPY %1
+
+...