[AMDGPU] Only remove branches in SIInstrInfo::removeBranch

[critson]

Summary

Without this change _term instructions can be removed during
critical edge splitting.

[critson]

Rebase on top of pre-committed test.

[foad]

LGTM, given that it matches similar logic in analyzeBranch.

As a further cleanup maybe both analyzeBranch and removeBranch should stop using getFirstTerminator, and instead just look for one or two branch instructions right at the end of MBB?

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

 unsigned SIInstrInfo::removeBranch(MachineBasicBlock &MBB,
                                    int *BytesRemoved) const {
   MachineBasicBlock::iterator I = MBB.getFirstTerminator();
 
   unsigned Count = 0;
   unsigned RemovedSize = 0;
   while (I != MBB.end()) {
     MachineBasicBlock::iterator Next = std::next(I);
+    // Skip over artificial terminators when removing instructions.
+    if (I->isBranch() || I->isReturn()) {
       RemovedSize += getInstSizeInBytes(*I);
       I->eraseFromParent();
       ++Count;
+    }
     I = Next;
   }
 
   if (BytesRemoved)
     *BytesRemoved = RemovedSize;
 
   return Count;
 }

llvm/test/CodeGen/AMDGPU/artificial-terminators.mir

body: |
   ; CHECK-NEXT:   [[V_ADD_U32_e64_3:%[0-9]+]]:vgpr_32 = nsw V_ADD_U32_e64 killed [[V_SUB_U32_e64_]], killed [[S_MOV_B32_4]], 0, implicit $exec
   ; CHECK-NEXT:   S_BRANCH %bb.1
   ; CHECK-NEXT: {{  $}}
   ; CHECK-NEXT: bb.1:
   ; CHECK-NEXT:   successors: %bb.5(0x30000000), %bb.2(0x50000000)
   ; CHECK-NEXT: {{  $}}
   ; CHECK-NEXT:   [[V_CMP_LT_I32_e64_:%[0-9]+]]:sreg_32 = V_CMP_LT_I32_e64 [[V_ADD_U32_e64_3]], [[S_MOV_B32_1]], implicit $exec
   ; CHECK-NEXT:   [[S_XOR_B32_:%[0-9]+]]:sreg_32 = S_XOR_B32 $exec_lo, [[V_CMP_LT_I32_e64_]], implicit-def $scc
+  ; CHECK-NEXT:   $exec_lo = S_MOV_B32_term [[S_XOR_B32_]]
   ; CHECK-NEXT:   S_CBRANCH_EXECNZ %bb.2, implicit $exec
   ; CHECK-NEXT: {{  $}}
   ; CHECK-NEXT: bb.5:
   ; CHECK-NEXT:   successors: %bb.4(0x80000000)
   ; CHECK-NEXT: {{  $}}
   ; CHECK-NEXT:   [[COPY3:%[0-9]+]]:sreg_32 = COPY [[V_CMP_LT_I32_e64_]]
   ; CHECK-NEXT:   S_BRANCH %bb.4
   ; CHECK-NEXT: {{  $}}