diff --git a/llvm/lib/Target/AMDGPU/SIPreEmitPeephole.cpp b/llvm/lib/Target/AMDGPU/SIPreEmitPeephole.cpp
--- a/llvm/lib/Target/AMDGPU/SIPreEmitPeephole.cpp
+++ b/llvm/lib/Target/AMDGPU/SIPreEmitPeephole.cpp
@@ -30,6 +30,7 @@
   const SIRegisterInfo *TRI = nullptr;
 
   bool optimizeVccBranch(MachineInstr &MI) const;
+  bool optimizeSetGPR(MachineInstr &First, MachineInstr &MI) const;
 
 public:
   static char ID;
@@ -143,6 +144,54 @@
   return true;
 }
 
+bool SIPreEmitPeephole::optimizeSetGPR(MachineInstr &First,
+                                       MachineInstr &MI) const {
+  MachineBasicBlock &MBB = *MI.getParent();
+  const MachineFunction &MF = *MBB.getParent();
+  const MachineRegisterInfo &MRI = MF.getRegInfo();
+  MachineOperand *Idx = TII->getNamedOperand(MI, AMDGPU::OpName::src0);
+  Register IdxReg = Idx->isReg() ? Idx->getReg() : Register();
+  SmallVector<MachineInstr *, 4> ToRemove;
+  bool IdxOn = true;
+
+  if (!MI.isIdenticalTo(First))
+    return false;
+
+  // Scan back to find an identical S_SET_GPR_IDX_ON
+  for (MachineBasicBlock::iterator I = std::next(First.getIterator()),
+       E = MI.getIterator(); I != E; ++I) {
+    switch (I->getOpcode()) {
+    case AMDGPU::S_SET_GPR_IDX_MODE:
+      return false;
+    case AMDGPU::S_SET_GPR_IDX_OFF:
+      IdxOn = false;
+      ToRemove.push_back(&*I);
+      break;
+    default:
+      if (I->modifiesRegister(AMDGPU::M0, TRI))
+        return false;
+      if (IdxReg && I->modifiesRegister(IdxReg, TRI))
+        return false;
+      if (llvm::any_of(I->operands(),
+                       [&MRI, this](const MachineOperand &MO) {
+                         return MO.isReg() &&
+                                TRI->isVectorRegister(MRI, MO.getReg());
+                       })) {
+        // The only exception allowed here is another indirect V_MOV_B32_e32
+        // with the same mode.
+        if (!IdxOn || I->getOpcode() != AMDGPU::V_MOV_B32_e32 ||
+            !I->hasRegisterImplicitUseOperand(AMDGPU::M0))
+          return false;
+      }
+    }
+  }
+
+  MI.eraseFromParent();
+  for (MachineInstr *RI : ToRemove)
+    RI->eraseFromParent();
+  return true;
+}
+
 bool SIPreEmitPeephole::runOnMachineFunction(MachineFunction &MF) {
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   TII = ST.getInstrInfo();
@@ -150,18 +199,50 @@
   bool Changed = false;
 
   for (MachineBasicBlock &MBB : MF) {
-    MachineBasicBlock::iterator MBBI = MBB.getFirstTerminator();
-    if (MBBI == MBB.end())
+    MachineBasicBlock::iterator MBBE = MBB.getFirstTerminator();
+    if (MBBE != MBB.end()) {
+      MachineInstr &MI = *MBBE;
+      switch (MI.getOpcode()) {
+      case AMDGPU::S_CBRANCH_VCCZ:
+      case AMDGPU::S_CBRANCH_VCCNZ:
+        Changed |= optimizeVccBranch(MI);
+        continue;
+      default:
+        break;
+      }
+    }
+
+    if (!ST.hasVGPRIndexMode())
       continue;
 
-    MachineInstr &MI = *MBBI;
-    switch (MI.getOpcode()) {
-    case AMDGPU::S_CBRANCH_VCCZ:
-    case AMDGPU::S_CBRANCH_VCCNZ:
-      Changed |= optimizeVccBranch(MI);
-      break;
-    default:
-      break;
+    MachineInstr *SetGPRMI = nullptr;
+    const unsigned Threshold = 20;
+    unsigned Count = 0;
+    // Scan the block for two S_SET_GPR_IDX_ON instructions to see if a
+    // second is not needed. Do expensive checks in the optimizeSetGPR()
+    // and limit the distance to 20 instructions for compile time purposes.
+    for (MachineBasicBlock::iterator MBBI = MBB.begin(); MBBI != MBBE; ) {
+      MachineInstr &MI = *MBBI;
+      ++MBBI;
+
+      if (Count == Threshold)
+        SetGPRMI = nullptr;
+      else
+        ++Count;
+
+      if (MI.getOpcode() != AMDGPU::S_SET_GPR_IDX_ON)
+        continue;
+
+      Count = 0;
+      if (!SetGPRMI) {
+        SetGPRMI = &MI;
+        continue;
+      }
+
+      if (optimizeSetGPR(*SetGPRMI, MI))
+        Changed = true;
+      else
+        SetGPRMI = &MI;
     }
   }
 
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.ll
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.ll
@@ -428,8 +428,6 @@
 ; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
 ; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v17, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v18, v1
 ; GPRIDX-NEXT:    s_set_gpr_idx_off
 ; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
@@ -470,8 +468,6 @@
 ; GPRIDX-NEXT:    s_lshl_b32 s0, s2, 1
 ; GPRIDX-NEXT:    s_set_gpr_idx_on s0, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v16, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_set_gpr_idx_on s0, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v17, v1
 ; GPRIDX-NEXT:    s_set_gpr_idx_off
 ; GPRIDX-NEXT:    global_store_dwordx2 v[0:1], v[16:17], off
@@ -1036,8 +1032,6 @@
 ; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
 ; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v17, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v18, v1
 ; GPRIDX-NEXT:    s_set_gpr_idx_off
 ; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
@@ -1164,8 +1158,6 @@
 ; GPRIDX-NEXT:    v_cmp_eq_u32_e32 vcc, s6, v16
 ; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v17, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_set_gpr_idx_on s7, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v18, v1
 ; GPRIDX-NEXT:    s_set_gpr_idx_off
 ; GPRIDX-NEXT:    s_and_saveexec_b64 vcc, vcc
@@ -1298,8 +1290,6 @@
 ; GPRIDX-NEXT:    s_lshl_b32 s0, s2, 1
 ; GPRIDX-NEXT:    s_set_gpr_idx_on s0, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v32, v0
-; GPRIDX-NEXT:    s_set_gpr_idx_off
-; GPRIDX-NEXT:    s_set_gpr_idx_on s0, gpr_idx(SRC0)
 ; GPRIDX-NEXT:    v_mov_b32_e32 v0, v1
 ; GPRIDX-NEXT:    s_set_gpr_idx_off
 ; GPRIDX-NEXT:    v_readfirstlane_b32 s0, v32
diff --git a/llvm/test/CodeGen/AMDGPU/set-gpr-idx-peephole.mir b/llvm/test/CodeGen/AMDGPU/set-gpr-idx-peephole.mir
new file mode 100644
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/set-gpr-idx-peephole.mir
@@ -0,0 +1,338 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass si-pre-emit-peephole -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
+
+---
+name:            simple
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: simple
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            salu_in_between
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: salu_in_between
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $sgpr0 = S_MOV_B32 $sgpr2
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  $sgpr0 = S_MOV_B32 $sgpr2
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            valu_write_in_between
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: valu_write_in_between
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: $vgpr20 = V_MOV_B32_e32 1, implicit $exec
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  $vgpr20 = V_MOV_B32_e32 1, implicit $exec
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            valu_read_in_between
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: valu_read_in_between
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: V_NOP_e32 implicit $exec, implicit $vgpr0
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  V_NOP_e32 implicit $exec, implicit $vgpr0
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            changed_index
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: changed_index
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: $sgpr2 = S_MOV_B32 1
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  $sgpr2 = S_MOV_B32 1
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            implicitly_changed_index
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: implicitly_changed_index
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_NOP 0, implicit-def $sgpr2
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_NOP 0, implicit-def $sgpr2
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            changed_m0
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: changed_m0
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: $m0 = S_MOV_B32 1
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  $m0 = S_MOV_B32 1
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            implicitly_changed_m0
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: implicitly_changed_m0
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_NOP 0, implicit-def $m0
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_NOP 0, implicit-def $m0
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            same_imm_index
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: same_imm_index
+    ; GCN: S_SET_GPR_IDX_ON 1, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON 1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON 1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+name:            different_imm_index
+body:             |
+  bb.0:
+
+  S_SET_GPR_IDX_ON 1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON 2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            different_gpr_index
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: different_gpr_index
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr1, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            different_gpr_index_then_same_index
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: different_gpr_index_then_same_index
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_SET_GPR_IDX_ON $sgpr1, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            use_m0_with_idx_off
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: use_m0_with_idx_off
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  S_SET_GPR_IDX_OFF
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            three_in_a_row
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: three_in_a_row
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $vgpr17 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $vgpr18 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr17 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr18 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            different_gpr_index_then_two_same_indexes
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: different_gpr_index_then_two_same_indexes
+    ; GCN: S_SET_GPR_IDX_ON $sgpr1, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...
+
+---
+name:            two_same_indexes_then_different
+body:             |
+  bb.0:
+
+    ; GCN-LABEL: name: two_same_indexes_then_different
+    ; GCN: S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+    ; GCN: S_SET_GPR_IDX_ON killed $sgpr1, 1, implicit-def $m0, implicit undef $m0
+    ; GCN: $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+    ; GCN: S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON $sgpr2, 1, implicit-def $m0, implicit undef $m0
+  $vgpr16 = V_MOV_B32_e32 undef $vgpr1, implicit $exec, implicit $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+  S_SET_GPR_IDX_ON killed $sgpr1, 1, implicit-def $m0, implicit undef $m0
+  $vgpr15 = V_MOV_B32_e32 undef $vgpr0, implicit $exec, implicit killed $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15, implicit $m0
+  S_SET_GPR_IDX_OFF
+...