[AMDGPU] Enable carry out ADD/SUB operations divergence driven instruction selection.

[alex-t]

Summary

This change enables all kind of carry out ISD opcodes to be selected according to the node divergence.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

1064

Shouldn't need to scan all uses?

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3619–3622

This ignores other things it could be. You can just use .add()

3624–3626

You can just do COPY from SCC

[rampitec]

You need to add tests for selection and moveToVALU, including immediates and wave32.

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

1065

Should it look thru PHI and COPY? It may need a helper function though.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3806

It can me immediate, maybe even a FI.

3871

Same here, it can be an immediate, right?

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

5198

Formatting is off.

5232

Formatting is off.

5971

VCC_LO for wave32?

[alex-t]

Src1 immediate case handled. Formatting improved.

[alex-t]

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

1064

I assume that the one user selected to VALU is enough to make selecting carryout to SALU impractical. Also, how would you suggest to decide about VALU/SALU basing on the quantity of VALU users? Ratio VALU/SALU? Other heuristic?

1065

It works on the selection DAG on per block basis so no PHI nodes ever exist here. There are CopyToReg nodes representing cross block values. Those in order only have MVTs like i32, f32 etc...
Although, I could try to address their divergence using reg2value FunctionLoweringInfo map, similar to cross block input regs in isSDNodeSourceOfDivergence function. I'd consider this as a kind of further improvement.
Also, nobody guaranty the BBs selection order so the cross block scan may appear too complicated.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3624–3626

I did it this way because I noticed that the code in SIInstrInfo::CopyPhysReg restricted to 32bit destination.

if (RC == &AMDGPU::SReg_32_XM0RegClass ||
      RC == &AMDGPU::SReg_32RegClass) {
    if (SrcReg == AMDGPU::SCC) {
      BuildMI(MBB, MI, DL, get(AMDGPU::S_CSELECT_B32), DestReg)
          .addImm(1)
          .addImm(0);
      return;
    }

So, I expected this remark and postpone the decision for review :)
Would you mind me extending this to 64bit?

[rampitec]

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3805

Either operand can be immediate.

3808

VReg_64? Since it did not fail anywhere this case must be not covered by any tests.

3867

Again it can be an immediate.

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

5189

These are not necessarily registers too.

5216

Same here.

5963

use_nodbg_instructions()

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

1064

There's already a isVGPRImm function, which should be functionally the same

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

919–924

This is just broken. We already run it, and there shouldn't' be a reason to involve SIFixupVectorISel

[alex-t]

In D78091#1981388, @rampitec wrote:

You need to add tests for selection and moveToVALU, including immediates and wave32.

Carry outs - UADDO/USUBO are already covered by the existing uaddo.ll and usubo.ll. There exist examples with both divergent and uniform ISD::UADDO/ISD::USUBO nodes to select.
That test are already updated.
SIFixSGPRCopies::moveToVALU part is covered as well by the udiv64.ll, urem64.ll, sdiv64.ll, srem64.ll. All that tests contains identical tests one of which declared as kernel and therefore has uniform arguments and another one as function and has divergent arguments.
The former one contains uniform UADDO and ADDCARRY that are initially selected to S_ADD_CO_PSEUDO but later on converted in moveToVALU to V_ADD/V_ADDC.

So, I am only planning to add the pure selection MIR tests for ISD opcodes.
Wave64/32 tests are probably needed as wellas the immediate operands tests.

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

919–924

As I have seen debugging, FinalizaISel gets invoked from the TargetPassConfig base class just after a bundle of passes defined as InstrSelector by the Target. So in our case - after createSIAddIMGInitPass. That's why I had to add this.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3808

I maybe misunderstand the documentation, but it says that the size we only can have 32bit immediate aa operand.
I also did some experiments with different targets (gfx600,900,1010) and always have seen that 64bit size constant was split into 2 32bit parts for addition.
Please correct me if I understand it in a wrong way.

[rampitec]

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3808

Two lines below you are asking for sub0 of that RC. VGPR_32 does not have sub0.

[alex-t]

Both operands can be immediate - handled. Some other changes to follow up the discussion.

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

1064

Not really. isVGPRImm walks along the use list and check if each use can accept SGPR or not. It also tries to commute operands if possible to make SGPR acceptable. The main goal is to select SGPR instead of VGPR to keep immediate whenever possible. This works in pair with the SIFoldOperands.

Here we try to workaround the following sub-optimal selection pattern:

	s_sub_i32 s0, s0, s1
	s_cselect_b64 s[0:1], 1, 0
	v_cndmask_b32_e64 v0, 0, 1, s[0:1]

that appears instead of the:

	v_sub_co_u32_e32 v4, vcc, s0, v4
	v_cndmask_b32_e64 v5, 0, 1, vcc

We need s_cselect to copy SCC to SReg_64.
In this sense, any explicit use of the conditional register pair leads to one extra instruction in case of uniform UADDO/USUBO.
Same time it saves one VGPR.
The only case that does not produce extra instruction is true carry out operations where the only user of the UADDO/USUBO value 2 is the ADD/SUBCARRY node.
So, I decide to change this for explicit check. This is over-conservative but it is okay for now.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3808

Yes. And it is exactly what is expected :)
getSubRegClass returns VGPR_32RegClass itself in this case. In fact id does not matter what it is.
buildExtractSubRegOrImm does not use SubRC argument if operand is immediate.

if (Op.isImm()) {
   if (SubIdx == AMDGPU::sub0)
     return MachineOperand::CreateImm(static_cast<int32_t>(Op.getImm()));
   if (SubIdx == AMDGPU::sub1)
     return MachineOperand::CreateImm(static_cast<int32_t>(Op.getImm() >> 32));

   llvm_unreachable("Unhandled register index for immediate");
 }

Once again, I maybe don't understand what your objection is about.

For the simple i64 immediate addition like this:

%add = add i64 20015998343286, %a

we generate carry out:

	s_add_u32 s0, s2, 0x56789876
	s_addc_u32 s1, s3, 0x1234

for uniform

or

	v_add_co_u32_e32 v0, vcc, 0x56789876, v0
	v_mov_b32_e32 v1, 0x1234
	v_addc_co_u32_e32 v1, vcc, 0, v1, vcc

for divergent.

So, why do we need VReg_64?

[rampitec]

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3617

It can be immediate.

3798

You can probably use SIInstrInfo::getAddNoCarry() and extend it to produce sub as well or create e new helper. You are always using I32 version even if a no-carry U32 version is available.

3808

You are calling TRI->getSubRegClass(Src1RC, AMDGPU::sub1); on this RC. You want to have VGPR_32 as an answer. Even though getSubRegClass() may return RC itself if it does not have a requested subreg this sounds like a bug. It would be more natural for it to assert. To be on a safe side pass there VReg_64 to get the same VGPR_32 or just do not call it if it is an immediate.

[alex-t]

a few more corrections

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3798

Not sure which line this comment belong...
For LoOpc I indeed need carry out opcode.
getAddNoCarry returnc the addition that does not write the carry flag.

Anyway, the exact opcode is selected later on by the SIInstrInfo::pseudoToMCOpcode

renamable $vgpr0 = V_ADD_I32_e32 1450743926, killed $vgpr0, implicit-def $vcc, implicit $exec
renamable $vgpr1 = V_MOV_B32_e32 4660, implicit $exec
renamable $vgpr1 = V_ADDC_U32_e32 0, killed $vgpr1, implicit-def $vcc, implicit killed $vcc, implicit $exec

turns to the

	v_add_co_u32_e32 v0, vcc, 0x56789876, v0
	v_mov_b32_e32 v1, 0x1234
	v_addc_co_u32_e32 v1, vcc, 0, v1, vcc

for gfx9

but to the

	v_add_i32_e32 v0, vcc, 0x56789876, v0
	v_mov_b32_e32 v1, 0x1234
	v_addc_u32_e32 v1, vcc, 0, v1, vcc

for gfx6

3808

Fine, we have one weird piece - SIRegisterInfo::getSubRegClass that returns input argument back if does not succeed.
Then we have one more weird piece - SIInstrInfo::buildExtractSubRegOrImm that is in fact 2 separate functions. It does completely different things for register and immediate, and just ignores register class arguments for immediate. Unfortunately all arguments required. So, I have to pass to it register class and sub-register class despite of the fact they're not used.
Refactoring these 2 pieces should be separate change.
If you insist that using VGPR_32RegClass is misleading I have no choice but VReg_64RegClass, that is misleading either IMHO.
I'll have to add a FIXME comment to explain why we use VGPR_32RegClass for immediate that can be 32bit only.

[rampitec]

I think the code is fine now except couple formatting comments.
What is missing specific selection tests.

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

3798

Right, that is add which has these forms. If you need addc you have a dead carry out. So this is OK.

[alex-t]

Selection specific test (carryout-selection.ll) added

[rampitec]

Can you add tests with both immediates? You will probably need to -start-before=amdgpu-isel or -O0 if it will be constant folded.
Also fix formatting issues.

[alex-t]

In D78091#2011048, @rampitec wrote:

Can you add tests with both immediates? You will probably need to -start-before=amdgpu-isel or -O0 if it will be constant folded.
Also fix formatting issues.

There is a reason for not adding such a test.
SelectionDAG::getNode() performs trivial constant folding irrelative of the optLevel.

See SelectionDAG.cpp::5469 for details

// Perform trivial constant folding.
if (SDValue SV = FoldConstantArithmetic(Opcode, DL, VT, {N1, N2}))
  return SV;

The result of the visiting ISD::ADD with 2 immediates looks like:

So the add is selected to 2 moves.

%add = add i64 20015998343286, 46117495621    to the 

%5:sreg_32 = S_MOV_B32 4671
%6:sreg_32 = S_MOV_B32 323599291

[alex-t]

changed code passed through the clang-format

[rampitec]

LGTM

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

 }
 
 void AMDGPUDAGToDAGISel::SelectAddcSubb(SDNode *N) {
   SDLoc DL(N);
   SDValue LHS = N->getOperand(0);
   SDValue RHS = N->getOperand(1);
   SDValue CI = N->getOperand(2);
 
+  if (N->isDivergent()) {
     unsigned Opc = N->getOpcode() == ISD::ADDCARRY ? AMDGPU::V_ADDC_U32_e64
                                                    : AMDGPU::V_SUBB_U32_e64;
     CurDAG->SelectNodeTo(
         N, Opc, N->getVTList(),
-      {LHS, RHS, CI, CurDAG->getTargetConstant(0, {}, MVT::i1) /*clamp bit*/});
+        {LHS, RHS, CI,
+         CurDAG->getTargetConstant(0, {}, MVT::i1) /*clamp bit*/});
+  } else {
+    unsigned Opc = N->getOpcode() == ISD::ADDCARRY ? AMDGPU::S_ADD_CO_PSEUDO
+                                                   : AMDGPU::S_SUB_CO_PSEUDO;
+    CurDAG->SelectNodeTo(N, Opc, N->getVTList(), {LHS, RHS, CI});
+  }
 }
 
 void AMDGPUDAGToDAGISel::SelectUADDO_USUBO(SDNode *N) {
   // The name of the opcodes are misleading. v_add_i32/v_sub_i32 have unsigned
   // carry out despite the _i32 name. These were renamed in VI to _U32.
   // FIXME: We should probably rename the opcodes here.
-  unsigned Opc = N->getOpcode() == ISD::UADDO ?
-    AMDGPU::V_ADD_I32_e64 : AMDGPU::V_SUB_I32_e64;
+  bool IsAdd = N->getOpcode() == ISD::UADDO;
+  bool IsVALU = N->isDivergent();
+
+  for (SDNode::use_iterator UI = N->use_begin(), E = N->use_end(); UI != E;
+       ++UI)
+    if (UI.getUse().getResNo() == 1) {
+      if ((IsAdd && (UI->getOpcode() != ISD::ADDCARRY)) ||
+          (!IsAdd && (UI->getOpcode() != ISD::SUBCARRY))) {
+        IsVALU = true;
+        break;
+      }
+    }
+
+  if (IsVALU) {
+    unsigned Opc = IsAdd ? AMDGPU::V_ADD_I32_e64 : AMDGPU::V_SUB_I32_e64;
 
     CurDAG->SelectNodeTo(
         N, Opc, N->getVTList(),
         {N->getOperand(0), N->getOperand(1),
          CurDAG->getTargetConstant(0, {}, MVT::i1) /*clamp bit*/});
+  } else {
+    unsigned Opc = N->getOpcode() == ISD::UADDO ? AMDGPU::S_UADDO_PSEUDO
+                                                : AMDGPU::S_USUBO_PSEUDO;
+
+    CurDAG->SelectNodeTo(N, Opc, N->getVTList(),
+                         {N->getOperand(0), N->getOperand(1)});
+  }
 }
 
 void AMDGPUDAGToDAGISel::SelectFMA_W_CHAIN(SDNode *N) {
   SDLoc SL(N);
   //  src0_modifiers, src0,  src1_modifiers, src1, src2_modifiers, src2, clamp, omod
   SDValue Ops[10];
 
   SelectVOP3Mods0(N->getOperand(1), Ops[1], Ops[0], Ops[6], Ops[7]);

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

bool GCNPassConfig::addILPOpts() {
   TargetPassConfig::addILPOpts();
   return false;
 }
 
 bool GCNPassConfig::addInstSelector() {
   AMDGPUPassConfig::addInstSelector();
   addPass(&SIFixSGPRCopiesID);
   addPass(createSILowerI1CopiesPass());
+  // TODO: We have to add FinalizeISel
+  // to expand V_ADD/SUB_U64_PSEUDO before SIFixupVectorISel
+  // that expects V_ADD/SUB -> A_ADDC/SUBB pairs expanded.
+  // Will be removed as soon as SIFixupVectorISel is changed
+  // to work with V_ADD/SUB_U64_PSEUDO instead.
+  addPass(&FinalizeISelID);
   addPass(createSIFixupVectorISelPass());
   addPass(createSIAddIMGInitPass());
   return false;
 }
 
 bool GCNPassConfig::addIRTranslator() {
   addPass(new IRTranslator());
   return false;

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

if (TII->isMIMG(MI)) {
     }
     // Add a memoperand for mimg instructions so that they aren't assumed to
     // be ordered memory instuctions.
 
     return BB;
   }
 
   switch (MI.getOpcode()) {
+  case AMDGPU::S_UADDO_PSEUDO:
+  case AMDGPU::S_USUBO_PSEUDO: {
+    const DebugLoc &DL = MI.getDebugLoc();
+    MachineOperand &Dest0 = MI.getOperand(0);
+    MachineOperand &Dest1 = MI.getOperand(1);
+    MachineOperand &Src0 = MI.getOperand(2);
+    MachineOperand &Src1 = MI.getOperand(3);
+
+    unsigned Opc = (MI.getOpcode() == AMDGPU::S_UADDO_PSEUDO)
+                       ? AMDGPU::S_ADD_I32
+                       : AMDGPU::S_SUB_I32;
+    BuildMI(*BB, MI, DL, TII->get(Opc), Dest0.getReg()).add(Src0).add(Src1);
+
+    BuildMI(*BB, MI, DL, TII->get(AMDGPU::S_CSELECT_B64), Dest1.getReg())
+        .addImm(1)
+        .addImm(0);
+
+    MI.eraseFromParent();
+    return BB;
+  }
   case AMDGPU::S_ADD_U64_PSEUDO:
   case AMDGPU::S_SUB_U64_PSEUDO: {
     MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
     const GCNSubtarget &ST = MF->getSubtarget<GCNSubtarget>();
     const SIRegisterInfo *TRI = ST.getRegisterInfo();
     const TargetRegisterClass *BoolRC = TRI->getBoolRC();
     const DebugLoc &DL = MI.getDebugLoc();
 
     MachineOperand &Dest = MI.getOperand(0);
     MachineOperand &Src0 = MI.getOperand(1);
     MachineOperand &Src1 = MI.getOperand(2);
 
     Register DestSub0 = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
     Register DestSub1 = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
 
-    MachineOperand Src0Sub0 = TII->buildExtractSubRegOrImm(MI, MRI,
-     Src0, BoolRC, AMDGPU::sub0,
-     &AMDGPU::SReg_32RegClass);
-    MachineOperand Src0Sub1 = TII->buildExtractSubRegOrImm(MI, MRI,
-      Src0, BoolRC, AMDGPU::sub1,
-      &AMDGPU::SReg_32RegClass);
-
-    MachineOperand Src1Sub0 = TII->buildExtractSubRegOrImm(MI, MRI,
-      Src1, BoolRC, AMDGPU::sub0,
-      &AMDGPU::SReg_32RegClass);
-    MachineOperand Src1Sub1 = TII->buildExtractSubRegOrImm(MI, MRI,
-      Src1, BoolRC, AMDGPU::sub1,
-      &AMDGPU::SReg_32RegClass);
+    MachineOperand Src0Sub0 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src0, BoolRC, AMDGPU::sub0, &AMDGPU::SReg_32RegClass);
+    MachineOperand Src0Sub1 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src0, BoolRC, AMDGPU::sub1, &AMDGPU::SReg_32RegClass);
+
+    MachineOperand Src1Sub0 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src1, BoolRC, AMDGPU::sub0, &AMDGPU::SReg_32RegClass);
+    MachineOperand Src1Sub1 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src1, BoolRC, AMDGPU::sub1, &AMDGPU::SReg_32RegClass);
 
     bool IsAdd = (MI.getOpcode() == AMDGPU::S_ADD_U64_PSEUDO);
 
     unsigned LoOpc = IsAdd ? AMDGPU::S_ADD_U32 : AMDGPU::S_SUB_U32;
     unsigned HiOpc = IsAdd ? AMDGPU::S_ADDC_U32 : AMDGPU::S_SUBB_U32;
-    BuildMI(*BB, MI, DL, TII->get(LoOpc), DestSub0)
-      .add(Src0Sub0)
-      .add(Src1Sub0);
+    BuildMI(*BB, MI, DL, TII->get(LoOpc), DestSub0).add(Src0Sub0).add(Src1Sub0);
+    BuildMI(*BB, MI, DL, TII->get(HiOpc), DestSub1).add(Src0Sub1).add(Src1Sub1);
+    BuildMI(*BB, MI, DL, TII->get(TargetOpcode::REG_SEQUENCE), Dest.getReg())
+        .addReg(DestSub0)
+        .addImm(AMDGPU::sub0)
+        .addReg(DestSub1)
+        .addImm(AMDGPU::sub1);
+    MI.eraseFromParent();
+    return BB;
+  }
+  case AMDGPU::V_ADD_U64_PSEUDO:
+  case AMDGPU::V_SUB_U64_PSEUDO: {
+    MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+    const GCNSubtarget &ST = MF->getSubtarget<GCNSubtarget>();
+    const SIRegisterInfo *TRI = ST.getRegisterInfo();
+    const DebugLoc &DL = MI.getDebugLoc();
+
+    bool IsAdd = (MI.getOpcode() == AMDGPU::V_ADD_U64_PSEUDO);
+
+    const auto *CarryRC = TRI->getRegClass(AMDGPU::SReg_1_XEXECRegClassID);
+
+    Register DestSub0 = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+    Register DestSub1 = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+
+    Register CarryReg = MRI.createVirtualRegister(CarryRC);
+    Register DeadCarryReg = MRI.createVirtualRegister(CarryRC);
+
+    MachineOperand &Dest = MI.getOperand(0);
+    MachineOperand &Src0 = MI.getOperand(1);
+    MachineOperand &Src1 = MI.getOperand(2);
+
+    const TargetRegisterClass *Src0RC = Src0.isReg()
+                                            ? MRI.getRegClass(Src0.getReg())
+                                            : &AMDGPU::VReg_64RegClass;
+    const TargetRegisterClass *Src1RC = Src1.isReg()
+                                            ? MRI.getRegClass(Src1.getReg())
+                                            : &AMDGPU::VReg_64RegClass;
+
+    const TargetRegisterClass *Src0SubRC =
+        TRI->getSubRegClass(Src0RC, AMDGPU::sub0);
+    const TargetRegisterClass *Src1SubRC =
+        TRI->getSubRegClass(Src1RC, AMDGPU::sub1);
+
+    MachineOperand SrcReg0Sub0 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src0, Src0RC, AMDGPU::sub0, Src0SubRC);
+    MachineOperand SrcReg1Sub0 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src1, Src1RC, AMDGPU::sub0, Src1SubRC);
+
+    MachineOperand SrcReg0Sub1 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src0, Src0RC, AMDGPU::sub1, Src0SubRC);
+    MachineOperand SrcReg1Sub1 = TII->buildExtractSubRegOrImm(
+        MI, MRI, Src1, Src1RC, AMDGPU::sub1, Src1SubRC);
+
+    unsigned LoOpc = IsAdd ? AMDGPU::V_ADD_I32_e64 : AMDGPU::V_SUB_I32_e64;
+    MachineInstr *LoHalf = BuildMI(*BB, MI, DL, TII->get(LoOpc), DestSub0)
+                               .addReg(CarryReg, RegState::Define)
+                               .add(SrcReg0Sub0)
+                               .add(SrcReg1Sub0)
+                               .addImm(0); // clamp bit
+
+    unsigned HiOpc = IsAdd ? AMDGPU::V_ADDC_U32_e64 : AMDGPU::V_SUBB_U32_e64;
+    MachineInstr *HiHalf =
         BuildMI(*BB, MI, DL, TII->get(HiOpc), DestSub1)
-      .add(Src0Sub1)
-      .add(Src1Sub1);
+            .addReg(DeadCarryReg, RegState::Define | RegState::Dead)
+            .add(SrcReg0Sub1)
+            .add(SrcReg1Sub1)
+            .addReg(CarryReg, RegState::Kill)
+            .addImm(0); // clamp bit
+
     BuildMI(*BB, MI, DL, TII->get(TargetOpcode::REG_SEQUENCE), Dest.getReg())
         .addReg(DestSub0)
         .addImm(AMDGPU::sub0)
         .addReg(DestSub1)
         .addImm(AMDGPU::sub1);
+    TII->legalizeOperands(*LoHalf);
+    TII->legalizeOperands(*HiHalf);
+    MI.eraseFromParent();
+    return BB;
+  }
+  case AMDGPU::S_ADD_CO_PSEUDO:
+  case AMDGPU::S_SUB_CO_PSEUDO: {
+    // This pseudo has a chance to be selected
+    // only from uniform add/subcarry node. All the VGPR operands
+    // therefore assumed to be splat vectors.
+    MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+    const GCNSubtarget &ST = MF->getSubtarget<GCNSubtarget>();
+    const SIRegisterInfo *TRI = ST.getRegisterInfo();
+    MachineBasicBlock::iterator MII = MI;
+    const DebugLoc &DL = MI.getDebugLoc();
+    MachineOperand &Dest = MI.getOperand(0);
+    MachineOperand &Src0 = MI.getOperand(2);
+    MachineOperand &Src1 = MI.getOperand(3);
+    MachineOperand &Src2 = MI.getOperand(4);
+    unsigned Opc = (MI.getOpcode() == AMDGPU::S_ADD_CO_PSEUDO)
+                       ? AMDGPU::S_ADDC_U32
+                       : AMDGPU::S_SUBB_U32;
+    if (Src0.isReg() && TRI->isVectorRegister(MRI, Src0.getReg())) {
+      Register RegOp0 = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
+      BuildMI(*BB, MII, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32), RegOp0)
+          .addReg(Src0.getReg());
+      Src0.setReg(RegOp0);
+    }
+    if (Src1.isReg() && TRI->isVectorRegister(MRI, Src1.getReg())) {
+      Register RegOp1 = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
+      BuildMI(*BB, MII, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32), RegOp1)
+          .addReg(Src1.getReg());
+      Src1.setReg(RegOp1);
+    }
+    Register RegOp2 = MRI.createVirtualRegister(&AMDGPU::SReg_32RegClass);
+    if (TRI->isVectorRegister(MRI, Src2.getReg())) {
+      BuildMI(*BB, MII, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32), RegOp2)
+          .addReg(Src2.getReg());
+      Src2.setReg(RegOp2);
+    }
+
+    if (TRI->getRegSizeInBits(*MRI.getRegClass(Src2.getReg())) == 64) {
+      BuildMI(*BB, MII, DL, TII->get(AMDGPU::S_CMP_LG_U64))
+          .addReg(Src2.getReg())
+          .addImm(0);
+    } else {
+      BuildMI(*BB, MII, DL, TII->get(AMDGPU::S_CMPK_LG_U32))
+          .addReg(Src2.getReg())
+          .addImm(0);
+    }
+
+    BuildMI(*BB, MII, DL, TII->get(Opc), Dest.getReg()).add(Src0).add(Src1);
     MI.eraseFromParent();
     return BB;
   }
   case AMDGPU::SI_INIT_M0: {
     BuildMI(*BB, MI.getIterator(), MI.getDebugLoc(),
             TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
         .add(MI.getOperand(0));
     MI.eraseFromParent();
     return BB;
   }
   case AMDGPU::SI_INIT_EXEC:
     // This should be before all vector instructions.
     BuildMI(*BB, &*BB->begin(), MI.getDebugLoc(), TII->get(AMDGPU::S_MOV_B64),
             AMDGPU::EXEC)
         .addImm(MI.getOperand(0).getImm());
     MI.eraseFromParent();
     return BB;
 
   case AMDGPU::SI_INIT_EXEC_LO:
     // This should be before all vector instructions.
     BuildMI(*BB, &*BB->begin(), MI.getDebugLoc(), TII->get(AMDGPU::S_MOV_B32),
             AMDGPU::EXEC_LO)
         .addImm(MI.getOperand(0).getImm());
     MI.eraseFromParent();
     return BB;
 
   case AMDGPU::SI_INIT_EXEC_FROM_INPUT: {
     // Extract the thread count from an SGPR input and set EXEC accordingly.
     // Since BFM can't shift by 64, handle that case with CMP + CMOV.
     //
     // S_BFE_U32 count, input, {shift, 7}
     // S_BFM_B64 exec, count, 0
     // S_CMP_EQ_U32 count, 64
     // S_CMOV_B64 exec, -1
     MachineInstr *FirstMI = &*BB->begin();
     MachineRegisterInfo &MRI = MF->getRegInfo();
     Register InputReg = MI.getOperand(0).getReg();
MachineBasicBlock *SITargetLowering::EmitInstrWithCustomInserter(
   }
 
   case AMDGPU::GET_GROUPSTATICSIZE: {
     assert(getTargetMachine().getTargetTriple().getOS() == Triple::AMDHSA ||
            getTargetMachine().getTargetTriple().getOS() == Triple::AMDPAL);
     DebugLoc DL = MI.getDebugLoc();
     BuildMI(*BB, MI, DL, TII->get(AMDGPU::S_MOV_B32))
         .add(MI.getOperand(0))
         .addImm(MFI->getLDSSize());
     MI.eraseFromParent();
     return BB;
   }
   case AMDGPU::SI_INDIRECT_SRC_V1:
   case AMDGPU::SI_INDIRECT_SRC_V2:
   case AMDGPU::SI_INDIRECT_SRC_V4:
   case AMDGPU::SI_INDIRECT_SRC_V8:
   case AMDGPU::SI_INDIRECT_SRC_V16:
     return emitIndirectSrc(MI, *BB, *getSubtarget());
   case AMDGPU::SI_INDIRECT_DST_V1:
   case AMDGPU::SI_INDIRECT_DST_V2:
   case AMDGPU::SI_INDIRECT_DST_V4:

llvm/lib/Target/AMDGPU/SIInstrInfo.cpp

case AMDGPU::S_ANDN2_B32:
       splitScalarBinOpN2(Worklist, Inst, AMDGPU::S_AND_B32);
       Inst.eraseFromParent();
       continue;
 
     case AMDGPU::S_ORN2_B32:
       splitScalarBinOpN2(Worklist, Inst, AMDGPU::S_OR_B32);
       Inst.eraseFromParent();
       continue;
+
+    // TODO: remove as soon as everything is ready
+    // to replace VGPR to SGPR copy with V_READFIRSTLANEs.
+    // S_ADD/SUB_CO_PSEUDO as well as S_UADDO/USUBO_PSEUDO
+    // can only be selected from the uniform SDNode.
+    case AMDGPU::S_ADD_CO_PSEUDO:
+    case AMDGPU::S_SUB_CO_PSEUDO: {
+      unsigned Opc = (Inst.getOpcode() == AMDGPU::S_ADD_CO_PSEUDO)
+                         ? AMDGPU::V_ADDC_U32_e64
+                         : AMDGPU::V_SUBB_U32_e64;
+      const auto *CarryRC = RI.getRegClass(AMDGPU::SReg_1_XEXECRegClassID);
+      Register DummyCReg = MRI.createVirtualRegister(CarryRC);
+      Register CarryReg = MRI.createVirtualRegister(CarryRC);
+      Register DestReg = MRI.createVirtualRegister(RI.getEquivalentVGPRClass(
+          MRI.getRegClass(Inst.getOperand(0).getReg())));
+      BuildMI(*MBB, &Inst, Inst.getDebugLoc(), get(AMDGPU::COPY), CarryReg)
+          .addReg(Inst.getOperand(4).getReg());
+      MachineInstr *CarryOp =
+          BuildMI(*MBB, &Inst, Inst.getDebugLoc(), get(Opc), DestReg)
+              .addReg(DummyCReg, RegState::Define | RegState::Dead)
+              .add(Inst.getOperand(2))
+              .add(Inst.getOperand(3))
+              .addReg(CarryReg, RegState::Kill)
+              .addImm(0);
+      legalizeOperands(*CarryOp);
+      MRI.replaceRegWith(Inst.getOperand(0).getReg(), DestReg);
+      addUsersToMoveToVALUWorklist(DestReg, MRI, Worklist);
+      Inst.eraseFromParent();
+    }
+      continue;
+    case AMDGPU::S_UADDO_PSEUDO:
+    case AMDGPU::S_USUBO_PSEUDO: {
+      const DebugLoc &DL = Inst.getDebugLoc();
+      MachineOperand &Dest0 = Inst.getOperand(0);
+      MachineOperand &Dest1 = Inst.getOperand(1);
+      MachineOperand &Src0 = Inst.getOperand(2);
+      MachineOperand &Src1 = Inst.getOperand(3);
+
+      unsigned Opc = (Inst.getOpcode() == AMDGPU::S_UADDO_PSEUDO)
+                         ? AMDGPU::V_ADD_I32_e64
+                         : AMDGPU::V_SUB_I32_e64;
+      const TargetRegisterClass *NewRC =
+          RI.getEquivalentVGPRClass(MRI.getRegClass(Dest0.getReg()));
+      Register DestReg = MRI.createVirtualRegister(NewRC);
+      MachineInstr *NewInstr = BuildMI(*MBB, &Inst, DL, get(Opc), DestReg)
+                                   .addReg(Dest1.getReg(), RegState::Define)
+                                   .add(Src0)
+                                   .add(Src1)
+                                   .addImm(0); // clamp bit
+
+      legalizeOperands(*NewInstr, MDT);
+
+      MRI.replaceRegWith(Dest0.getReg(), DestReg);
+      addUsersToMoveToVALUWorklist(NewInstr->getOperand(0).getReg(), MRI,
+                                   Worklist);
+      Inst.eraseFromParent();
+    }
+      continue;
     }
 
     if (NewOpcode == AMDGPU::INSTRUCTION_LIST_END) {
       // We cannot move this instruction to the VALU, so we should try to
       // legalize its operands instead.
       legalizeOperands(Inst, MDT);
       continue;
     }
 
     // Use the new VALU Opcode.
     const MCInstrDesc &NewDesc = get(NewOpcode);
     Inst.setDesc(NewDesc);
 
     // Remove any references to SCC. Vector instructions can't read from it, and

 }
 
 void SIInstrInfo::addSCCDefUsersToVALUWorklist(MachineOperand &Op,
                                                MachineInstr &SCCDefInst,
                                                SetVectorType &Worklist) const {
   // Ensure that def inst defines SCC, which is still live.
   assert(Op.isReg() && Op.getReg() == AMDGPU::SCC && Op.isDef() &&
          !Op.isDead() && Op.getParent() == &SCCDefInst);
+  SmallVector<MachineInstr *, 4> CopyToDelete;
   // This assumes that all the users of SCC are in the same block
   // as the SCC def.
   for (MachineInstr &MI : // Skip the def inst itself.
        make_range(std::next(MachineBasicBlock::iterator(SCCDefInst)),
                   SCCDefInst.getParent()->end())) {
     // Check if SCC is used first.
-    if (MI.findRegisterUseOperandIdx(AMDGPU::SCC, false, &RI) != -1)
+    if (MI.findRegisterUseOperandIdx(AMDGPU::SCC, false, &RI) != -1) {
+      if (MI.isCopy()) {
+        MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
+        unsigned DestReg = MI.getOperand(0).getReg();
+        SmallVector<MachineInstr *, 4> Users;
+        for (auto &User : MRI.use_nodbg_instructions(DestReg)) {
+          if ((User.getOpcode() == AMDGPU::S_ADD_CO_PSEUDO) ||
+              (User.getOpcode() == AMDGPU::S_SUB_CO_PSEUDO)) {
+            Users.push_back(&User);
+            Worklist.insert(&User);
+          }
+        }
+        for (auto &U : Users)
+          U->getOperand(4).setReg(RI.getVCC());
+        CopyToDelete.push_back(&MI);
+      } else
         Worklist.insert(&MI);
+    }
     // Exit if we find another SCC def.
     if (MI.findRegisterDefOperandIdx(AMDGPU::SCC, false, false, &RI) != -1)
-      return;
+      break;
   }
+  for (auto &Copy : CopyToDelete)
+    Copy->eraseFromParent();
 }
 
 const TargetRegisterClass *SIInstrInfo::getDestEquivalentVGPRClass(
   const MachineInstr &Inst) const {
   const TargetRegisterClass *NewDstRC = getOpRegClass(Inst, 0);
 
   switch (Inst.getOpcode()) {
   // For target instructions, getOpRegClass just returns the virtual register

llvm/lib/Target/AMDGPU/SIInstructions.td

 }
 
 def V_SET_INACTIVE_B64 : VPseudoInstSI <(outs VReg_64:$vdst),
   (ins VReg_64: $src, VSrc_b64:$inactive),
   [(set i64:$vdst, (int_amdgcn_set_inactive i64:$src, i64:$inactive))]> {
   let Constraints = "$src = $vdst";
 }
 
+let usesCustomInserter = 1, Defs = [VCC, EXEC] in {
+def V_ADD_U64_PSEUDO : VPseudoInstSI <
+  (outs VReg_64:$vdst), (ins VSrc_b64:$src0, VSrc_b64:$src1),
+  [(set VReg_64:$vdst, (getDivergentFrag<add>.ret i64:$src0, i64:$src1))]
+>;
+
+def V_SUB_U64_PSEUDO : VPseudoInstSI <
+  (outs VReg_64:$vdst), (ins VSrc_b64:$src0, VSrc_b64:$src1),
+  [(set VReg_64:$vdst, (getDivergentFrag<sub>.ret i64:$src0, i64:$src1))]
+>;
+} // End usesCustomInserter = 1, Defs = [VCC, EXEC]
 
 let usesCustomInserter = 1, Defs = [SCC] in {
 def S_ADD_U64_PSEUDO : SPseudoInstSI <
-  (outs SReg_64:$vdst), (ins SSrc_b64:$src0, SSrc_b64:$src1),
-  [(set SReg_64:$vdst, (add i64:$src0, i64:$src1))]
+  (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1),
+  [(set SReg_64:$sdst, (UniformBinFrag<add> i64:$src0, i64:$src1))]
 >;
 
 def S_SUB_U64_PSEUDO : SPseudoInstSI <
-  (outs SReg_64:$vdst), (ins SSrc_b64:$src0, SSrc_b64:$src1),
-  [(set SReg_64:$vdst, (sub i64:$src0, i64:$src1))]
+  (outs SReg_64:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1),
+  [(set SReg_64:$sdst, (UniformBinFrag<sub> i64:$src0, i64:$src1))]
 >;
 
 def S_ADD_U64_CO_PSEUDO : SPseudoInstSI <
   (outs SReg_64:$vdst, VOPDstS64orS32:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1)
 >;
 
 def S_SUB_U64_CO_PSEUDO : SPseudoInstSI <
   (outs SReg_64:$vdst, VOPDstS64orS32:$sdst), (ins SSrc_b64:$src0, SSrc_b64:$src1)
 >;
+
+def S_ADD_CO_PSEUDO : SPseudoInstSI <
+  (outs SReg_32:$sdst, SSrc_i1:$scc_out), (ins SSrc_b32:$src0, SSrc_b32:$src1, SSrc_i1:$scc_in)
+>;
+
+def S_SUB_CO_PSEUDO : SPseudoInstSI <
+  (outs SReg_32:$sdst, SSrc_i1:$scc_out), (ins SSrc_b32:$src0, SSrc_b32:$src1, SSrc_i1:$scc_in)
+>;
+
+def S_UADDO_PSEUDO : SPseudoInstSI <
+  (outs SReg_32:$sdst, SSrc_i1:$scc_out), (ins SSrc_b32:$src0, SSrc_b32:$src1)
+>;
+
+def S_USUBO_PSEUDO : SPseudoInstSI <
+  (outs SReg_32:$sdst, SSrc_i1:$scc_out), (ins SSrc_b32:$src0, SSrc_b32:$src1)
+>;
+
 } // End usesCustomInserter = 1, Defs = [SCC]
 
 let usesCustomInserter = 1 in {
 def GET_GROUPSTATICSIZE : SPseudoInstSI <(outs SReg_32:$sdst), (ins),
   [(set SReg_32:$sdst, (int_amdgcn_groupstaticsize))]>;
 } // End let usesCustomInserter = 1, SALU = 1
 
 // Wrap an instruction by duplicating it, except for setting isTerminator.

llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll

 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_mul_lo_u32 v3, v0, s3
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v4, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v9, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GCN-NEXT:    s_mov_b32 s4, 0x976a7376
+; GCN-NEXT:    s_movk_i32 s4, 0x11e
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v6, v5
 ; GCN-NEXT:    v_mul_lo_u32 v6, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v3, v1, v3
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    s_mov_b32 s10, -1
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v6, v5
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v4, v3, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v9, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_add_i32_e64 v0, s[0:1], v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v0, s2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, s3
 ; GCN-NEXT:    v_addc_u32_e64 v2, vcc, v1, v3, s[0:1]
 ; GCN-NEXT:    v_mul_lo_u32 v6, v2, s3
-; GCN-NEXT:    s_movk_i32 s2, 0x11f
+; GCN-NEXT:    s_mov_b32 s2, 0x976a7377
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, s3
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v10, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v9, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v2, v4
-; GCN-NEXT:    s_mov_b32 s3, 0x976a7377
+; GCN-NEXT:    s_movk_i32 s3, 0x11f
 ; GCN-NEXT:    s_mov_b32 s9, s5
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v9, v6
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v8, v10, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v10, v2, v5
 ; GCN-NEXT:    v_mul_hi_u32 v5, v2, v5
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v10, v6
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v10
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v9, v5, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v11, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v4, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s6, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s6, v0
 ; GCN-NEXT:    v_mul_hi_u32 v4, s6, v1
 ; GCN-NEXT:    v_mul_hi_u32 v5, s7, v1
 ; GCN-NEXT:    v_mul_lo_u32 v1, s7, v1
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, s7, v0
 ; GCN-NEXT:    v_mul_hi_u32 v0, s7, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, v5, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v8, v2, vcc
-; GCN-NEXT:    v_mul_lo_u32 v2, v0, s2
-; GCN-NEXT:    v_mul_hi_u32 v3, v0, s3
-; GCN-NEXT:    v_mul_lo_u32 v4, v1, s3
-; GCN-NEXT:    v_mov_b32_e32 v5, s2
+; GCN-NEXT:    v_mul_lo_u32 v2, v0, s3
+; GCN-NEXT:    v_mul_hi_u32 v3, v0, s2
+; GCN-NEXT:    v_mul_lo_u32 v4, v1, s2
+; GCN-NEXT:    v_mov_b32_e32 v5, s3
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_mul_lo_u32 v3, v0, s3
+; GCN-NEXT:    v_mul_lo_u32 v3, v0, s2
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_sub_i32_e32 v4, vcc, s7, v2
 ; GCN-NEXT:    v_sub_i32_e32 v3, vcc, s6, v3
 ; GCN-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, vcc
-; GCN-NEXT:    v_subrev_i32_e64 v5, s[0:1], s3, v3
+; GCN-NEXT:    v_subrev_i32_e64 v5, s[0:1], s2, v3
 ; GCN-NEXT:    v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1]
-; GCN-NEXT:    s_movk_i32 s3, 0x11e
-; GCN-NEXT:    v_cmp_lt_u32_e64 s[0:1], s3, v4
+; GCN-NEXT:    v_cmp_lt_u32_e64 s[0:1], s4, v4
+; GCN-NEXT:    s_mov_b32 s2, 0x976a7376
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GCN-NEXT:    v_cmp_lt_u32_e64 s[0:1], s4, v5
+; GCN-NEXT:    v_cmp_lt_u32_e64 s[0:1], s2, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], s2, v4
+; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], s3, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v4, v6, v5, s[0:1]
 ; GCN-NEXT:    v_add_i32_e64 v5, s[0:1], 2, v0
 ; GCN-NEXT:    v_addc_u32_e64 v6, s[0:1], 0, v1, s[0:1]
 ; GCN-NEXT:    v_add_i32_e64 v7, s[0:1], 1, v0
 ; GCN-NEXT:    v_addc_u32_e64 v8, s[0:1], 0, v1, s[0:1]
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v4, v8, v6, s[0:1]
 ; GCN-NEXT:    v_mov_b32_e32 v6, s7
 ; GCN-NEXT:    v_subb_u32_e32 v2, vcc, v6, v2, vcc
-; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s3, v2
+; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v3
+; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s2, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s2, v2
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s3, v2
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, v6, v3, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, v7, v5, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GCN-NEXT:    s_endpgm
   %r = udiv i64 %x, 1235195949943

 ; GCN-NEXT:    v_mul_hi_u32 v9, v0, v8
 ; GCN-NEXT:    v_mul_hi_u32 v10, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v3, v5
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v9, v6
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v7, v10, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v10, v3, v8
 ; GCN-NEXT:    v_mul_hi_u32 v8, v3, v8
 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v5
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v10, v6
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v10
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v9, v8, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v11, v2, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v7, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v5, s[2:3]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v3
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc

 ; GCN-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GCN-NEXT:    v_mov_b32_e32 v7, 0
 ; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-NEXT:    v_mul_lo_u32 v2, v0, s2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, s3
 ; GCN-NEXT:    v_mul_lo_u32 v4, v1, s3
-; GCN-NEXT:    s_mov_b32 s12, 0x9761f7c9
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s8, s4
+; GCN-NEXT:    s_movk_i32 s12, 0x11f
+; GCN-NEXT:    s_mov_b32 s13, 0x9761f7c9
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_mul_lo_u32 v3, v0, s3
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v4, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v9, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GCN-NEXT:    s_movk_i32 s4, 0x11f
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_mov_b32 s9, s5
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v6, v5
 ; GCN-NEXT:    v_mul_lo_u32 v6, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v3, v1, v3
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
-; GCN-NEXT:    s_mov_b32 s9, s5
+; GCN-NEXT:    s_movk_i32 s5, 0x11e
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v6, v5
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v4, v3, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v9, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_add_i32_e64 v0, s[0:1], v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v0, s2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, s3
 ; GCN-NEXT:    v_addc_u32_e64 v2, vcc, v1, v3, s[0:1]
 ; GCN-NEXT:    v_mul_lo_u32 v6, v2, s3
-; GCN-NEXT:    s_movk_i32 s5, 0x11e
+; GCN-NEXT:    s_mov_b32 s8, s4
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, s3
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v10, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v9, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v2, v4
+; GCN-NEXT:    s_mov_b32 s4, 0x9761f7c8
 ; GCN-NEXT:    s_mov_b32 s11, 0xf000
-; GCN-NEXT:    s_mov_b32 s10, -1
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v9, v6
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v8, v10, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v10, v2, v5
 ; GCN-NEXT:    v_mul_hi_u32 v5, v2, v5
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v10, v6
+; GCN-NEXT:    s_mov_b32 s10, -1
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v10
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v9, v5, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v11, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v4, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s6, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s6, v0
 ; GCN-NEXT:    v_mul_hi_u32 v4, s6, v1
 ; GCN-NEXT:    v_mul_hi_u32 v5, s7, v1
 ; GCN-NEXT:    v_mul_lo_u32 v1, s7, v1
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, s7, v0
 ; GCN-NEXT:    v_mul_hi_u32 v0, s7, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, v5, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v8, v2, vcc
-; GCN-NEXT:    v_mul_lo_u32 v2, v0, s4
-; GCN-NEXT:    v_mul_hi_u32 v3, v0, s12
-; GCN-NEXT:    v_mul_lo_u32 v1, v1, s12
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s12
+; GCN-NEXT:    v_mul_lo_u32 v2, v0, s12
+; GCN-NEXT:    v_mul_hi_u32 v3, v0, s13
+; GCN-NEXT:    v_mul_lo_u32 v1, v1, s13
+; GCN-NEXT:    v_mul_lo_u32 v0, v0, s13
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, s7, v1
+; GCN-NEXT:    v_mov_b32_e32 v3, s12
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
-; GCN-NEXT:    v_mov_b32_e32 v3, s4
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, vcc
-; GCN-NEXT:    v_subrev_i32_e64 v4, s[0:1], s12, v0
+; GCN-NEXT:    v_subrev_i32_e64 v4, s[0:1], s13, v0
 ; GCN-NEXT:    v_subbrev_u32_e64 v5, s[2:3], 0, v2, s[0:1]
 ; GCN-NEXT:    v_cmp_lt_u32_e64 s[2:3], s5, v5
-; GCN-NEXT:    s_mov_b32 s6, 0x9761f7c8
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[2:3]
-; GCN-NEXT:    v_cmp_lt_u32_e64 s[2:3], s6, v4
-; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s12, v4
+; GCN-NEXT:    v_cmp_lt_u32_e64 s[2:3], s4, v4
+; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s13, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
-; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, v5
+; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s12, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, v6, v7, s[2:3]
 ; GCN-NEXT:    v_subbrev_u32_e64 v2, s[0:1], 0, v2, s[0:1]
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
 ; GCN-NEXT:    v_mov_b32_e32 v5, s7
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
 ; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s5, v1
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
-; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v0
+; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v0
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s4, v1
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s12, v1
 ; GCN-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, v4, v3, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GCN-NEXT:    s_endpgm
   %r = urem i64 %x, 1235195393993

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v5
 ; GCN-NEXT:    v_mul_lo_u32 v5, v2, v5
 ; GCN-NEXT:    v_mul_hi_u32 v6, v2, v4
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v8, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v10
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v10, v5
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v6, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    s_ashr_i32 s2, s11, 31
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v4, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s2

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v2, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, 0, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v5
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v7, v2
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v6, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v5, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s14
 ; GCN-NEXT:    s_addc_u32 s1, s11, s14

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v2, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, 0, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v5
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v7, v2
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v6, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v5, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s8
 ; GCN-NEXT:    s_addc_u32 s1, s11, s8

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v2, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, 0, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v5
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v7, v2
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v6, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v5, s[2:3]
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_ashr_i32 s2, s9, 31

 ; GCN-NEXT:    v_mul_hi_u32 v14, v2, v8
 ; GCN-NEXT:    v_mul_hi_u32 v13, v2, v9
 ; GCN-NEXT:    v_mul_hi_u32 v11, v3, v9
 ; GCN-NEXT:    v_mul_lo_u32 v9, v3, v9
 ; GCN-NEXT:    v_mul_hi_u32 v10, v3, v8
 ; GCN-NEXT:    v_add_i32_e32 v12, vcc, v13, v12
 ; GCN-NEXT:    v_addc_u32_e32 v13, vcc, 0, v14, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v8
-; GCN-NEXT:    v_add_i32_e32 v9, vcc, v9, v12
+; GCN-NEXT:    v_add_i32_e32 v9, vcc, v12, v9
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v13, v11, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v8, vcc, v10, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v9, v3
 ; GCN-NEXT:    v_addc_u32_e32 v8, vcc, v6, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v7
 ; GCN-NEXT:    s_ashr_i32 s14, s11, 31
 ; GCN-NEXT:    v_addc_u32_e64 v5, vcc, v5, v8, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s14

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v5
 ; GCN-NEXT:    v_mul_lo_u32 v5, v2, v5
 ; GCN-NEXT:    v_mul_hi_u32 v6, v2, v4
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v8, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v10
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v10, v5
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v6, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    s_ashr_i32 s2, s11, 31
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v4, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s2

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v2, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, 0, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v5
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v7, v2
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v6, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v5, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s14
 ; GCN-NEXT:    s_addc_u32 s1, s11, s14

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v2, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v2, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, 0, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v5
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v7, v2
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v6, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v5, s[2:3]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v14, v2, v8
 ; GCN-NEXT:    v_mul_hi_u32 v13, v2, v9
 ; GCN-NEXT:    v_mul_hi_u32 v11, v3, v9
 ; GCN-NEXT:    v_mul_lo_u32 v9, v3, v9
 ; GCN-NEXT:    v_mul_hi_u32 v10, v3, v8
 ; GCN-NEXT:    v_add_i32_e32 v12, vcc, v13, v12
 ; GCN-NEXT:    v_addc_u32_e32 v13, vcc, 0, v14, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v8
-; GCN-NEXT:    v_add_i32_e32 v9, vcc, v9, v12
+; GCN-NEXT:    v_add_i32_e32 v9, vcc, v12, v9
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v13, v11, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v8, vcc, v10, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v9, v3
 ; GCN-NEXT:    v_addc_u32_e32 v8, vcc, v6, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v7
 ; GCN-NEXT:    v_addc_u32_e64 v5, vcc, v5, v8, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s14
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3

llvm/test/CodeGen/AMDGPU/carryout-selection.ll

+; RUN: llc -march=amdgcn -stop-after=amdgpu-isel < %s | FileCheck -enable-var-scope -check-prefixes=GCN-ISEL                %s
+
+; RUN: llc -march=amdgcn -mcpu=verde   -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CISI    %s
+; RUN: llc -march=amdgcn -mcpu=fiji    -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI      %s
+; RUN: llc -march=amdgcn -mcpu=gfx900  -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9    %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX1010 %s
+
+; GCN-ISEL-LABEL: name:   sadd64rr 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: S_ADD_U64_PSEUDO
+
+; GCN-LABEL: @sadd64rr
+; GCN:       s_add_u32
+; GCN:       s_addc_u32
+define amdgpu_kernel void @sadd64rr(i64 addrspace(1)* %out, i64 %a, i64 %b) {
+entry:
+  %add = add i64 %a, %b
+  store i64 %add, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   sadd64ri 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: S_ADD_U64_PSEUDO
+
+; GCN-LABEL: @sadd64ri
+; GCN:       s_add_u32  s{{[0-9]+}}, s{{[0-9]+}}, 0x56789876
+; GCN:       s_addc_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x1234
+define amdgpu_kernel void @sadd64ri(i64 addrspace(1)* %out, i64 %a) {
+entry:
+  %add = add i64 20015998343286, %a
+  store i64 %add, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   vadd64rr 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: V_ADD_U64_PSEUDO
+
+; GCN-LABEL: @vadd64rr
+;
+; CISI: v_add_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; CISI: v_addc_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; VI: v_addc_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX9: v_addc_co_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX1010: v_add_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], s{{[0-9]+}}, v{{[0-9]+}}
+; GFX1010: v_add_co_ci_u32_e64 v{{[0-9]+}}, [[CARRY]], s{{[0-9]+}}, 0, [[CARRY]]
+define amdgpu_kernel void @vadd64rr(i64 addrspace(1)* %out, i64 %a) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.ext = sext i32 %tid to i64
+  %add = add i64 %a, %tid.ext
+  store i64 %add, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   vadd64ri 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: V_ADD_U64_PSEUDO
+
+; GCN-LABEL: @vadd64ri
+;
+; CISI: v_add_i32_e32 v0, vcc, 0x56789876, v0
+; CISI: v_mov_b32_e32 v1, 0x1234
+; CISI: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+;
+; VI: v_add_u32_e32 v0, vcc, 0x56789876, v0
+; VI: v_mov_b32_e32 v1, 0x1234
+; VI: v_addc_u32_e32 v1, vcc, 0, v1, vcc
+;
+; GFX9: v_add_co_u32_e32 v0, vcc, 0x56789876, v0 
+; GFX9: v_mov_b32_e32 v1, 0x1234                 
+; GFX9: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc    
+;
+; GFX1010: v_add_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], 0x56789876, v{{[0-9]+}}
+; GFX1010: v_add_co_ci_u32_e64 v{{[0-9]+}}, [[CARRY]], 0, 0x1234, [[CARRY]]
+define amdgpu_kernel void @vadd64ri(i64 addrspace(1)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.ext = sext i32 %tid to i64
+  %add = add i64 20015998343286, %tid.ext
+  store i64 %add, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   suaddo32
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: S_ADD_I32
+define amdgpu_kernel void @suaddo32(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 %a, i32 %b) #0 {
+  %uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
+  %val = extractvalue { i32, i1 } %uadd, 0
+  %carry = extractvalue { i32, i1 } %uadd, 1
+  store i32 %val, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+
+; GCN-ISEL-LABEL: name:   uaddo32_vcc_user
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: V_ADD_I32_e64
+
+; below we check selection to v_add/addc
+; because the only user of VCC produced by the UADDOis v_cndmask.
+; We select to VALU form to avoid unnecessary s_cselect to copy SCC to VCC
+
+; GCN-LABEL: @uaddo32_vcc_user
+;
+; CISI: v_add_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; CISI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
+;
+; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; VI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
+;
+; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX9: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
+;
+; GFX1010: v_add_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
+; GFX1010: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[CARRY]]
+define amdgpu_kernel void @uaddo32_vcc_user(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 %a, i32 %b) #0 {
+  %uadd = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
+  %val = extractvalue { i32, i1 } %uadd, 0
+  %carry = extractvalue { i32, i1 } %uadd, 1
+  store i32 %val, i32 addrspace(1)* %out, align 4
+  store i1 %carry, i1 addrspace(1)* %carryout
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   suaddo64
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: S_ADD_U64_PSEUDO
+
+; GCN-LABEL: @suaddo64
+;
+; GCN: s_add_u32
+; GCN: s_addc_u32
+define amdgpu_kernel void @suaddo64(i64 addrspace(1)* %out, i1 addrspace(1)* %carryout, i64 %a, i64 %b) #0 {
+  %uadd = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
+  %val = extractvalue { i64, i1 } %uadd, 0
+  %carry = extractvalue { i64, i1 } %uadd, 1
+  store i64 %val, i64 addrspace(1)* %out, align 8
+  store i1 %carry, i1 addrspace(1)* %carryout
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   vuaddo64
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: V_ADD_U64_PSEUDO
+
+; GCN-LABEL: @vuaddo64
+;
+; CISI: v_add_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v0
+; CISI: v_addc_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; VI: v_add_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v0
+; VI: v_addc_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX9: v_add_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v0
+; GFX9: v_addc_co_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX1010: v_add_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], s{{[0-9]+}}, v0
+; GFX1010: v_add_co_ci_u32_e64 v{{[0-9]+}}, [[CARRY]], s{{[0-9]+}}, 0, [[CARRY]]
+define amdgpu_kernel void @vuaddo64(i64 addrspace(1)* %out, i1 addrspace(1)* %carryout, i64 %a) #0 {
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.ext = sext i32 %tid to i64
+  %uadd = call { i64, i1 } @llvm.uadd.with.overflow.i64(i64 %a, i64 %tid.ext)
+  %val = extractvalue { i64, i1 } %uadd, 0
+  %carry = extractvalue { i64, i1 } %uadd, 1
+  store i64 %val, i64 addrspace(1)* %out, align 8
+  store i1 %carry, i1 addrspace(1)* %carryout
+  ret void
+}
+
+; RUN: llc -march=amdgcn -stop-after=amdgpu-isel < %s | FileCheck -enable-var-scope -check-prefixes=GCN-ISEL                %s
+
+; RUN: llc -march=amdgcn -mcpu=verde   -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,CISI    %s
+; RUN: llc -march=amdgcn -mcpu=fiji    -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,VI      %s
+; RUN: llc -march=amdgcn -mcpu=gfx900  -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX9    %s
+; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GCN,GFX1010 %s
+
+; GCN-ISEL-LABEL: name:   ssub64rr 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: S_SUB_U64_PSEUDO
+
+; GCN-LABEL: @ssub64rr
+; GCN:       s_sub_u32
+; GCN:       s_subb_u32
+define amdgpu_kernel void @ssub64rr(i64 addrspace(1)* %out, i64 %a, i64 %b) {
+entry:
+  %sub = sub i64 %a, %b
+  store i64 %sub, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   ssub64ri 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: S_SUB_U64_PSEUDO
+
+; GCN-LABEL: @ssub64ri
+; GCN:       s_sub_u32  s{{[0-9]+}}, 0x56789876, s{{[0-9]+}}
+; GCN:       s_subb_u32 s{{[0-9]+}}, 0x1234, s{{[0-9]+}}
+define amdgpu_kernel void @ssub64ri(i64 addrspace(1)* %out, i64 %a) {
+entry:
+  %sub = sub i64 20015998343286, %a
+  store i64 %sub, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   vsub64rr 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: V_SUB_U64_PSEUDO
+
+; GCN-LABEL: @vsub64rr
+;
+; CISI: v_sub_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; CISI: v_subbrev_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; VI: v_subbrev_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX9: v_subbrev_co_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX1010: v_sub_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], s{{[0-9]+}}, v{{[0-9]+}}
+; GFX1010: v_sub_co_ci_u32_e64 v{{[0-9]+}}, [[CARRY]], s{{[0-9]+}}, 0, [[CARRY]]
+define amdgpu_kernel void @vsub64rr(i64 addrspace(1)* %out, i64 %a) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.ext = sext i32 %tid to i64
+  %sub = sub i64 %a, %tid.ext
+  store i64 %sub, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   vsub64ri 
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0.entry:
+; GCN-ISEL: V_SUB_U64_PSEUDO
+
+; GCN-LABEL: @vsub64ri
+;
+; CISI: v_sub_i32_e32 v0, vcc, 0x56789876, v0
+; CISI: v_mov_b32_e32 v1, 0x1234
+; CISI: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
+;
+; VI: v_sub_u32_e32 v0, vcc, 0x56789876, v0
+; VI: v_mov_b32_e32 v1, 0x1234
+; VI: v_subbrev_u32_e32 v1, vcc, 0, v1, vcc
+;
+; GFX9: v_sub_co_u32_e32 v0, vcc, 0x56789876, v0 
+; GFX9: v_mov_b32_e32 v1, 0x1234                 
+; GFX9: v_subbrev_co_u32_e32 v1, vcc, 0, v1, vcc    
+;
+; GFX1010: v_sub_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], 0x56789876, v{{[0-9]+}}
+; GFX1010: v_sub_co_ci_u32_e64 v{{[0-9]+}}, [[CARRY]], 0x1234, 0, [[CARRY]]
+define amdgpu_kernel void @vsub64ri(i64 addrspace(1)* %out) {
+entry:
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.ext = sext i32 %tid to i64
+  %sub = sub i64 20015998343286, %tid.ext
+  store i64 %sub, i64 addrspace(1)* %out
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   susubo32
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: S_SUB_I32
+define amdgpu_kernel void @susubo32(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 %a, i32 %b) #0 {
+  %usub = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
+  %val = extractvalue { i32, i1 } %usub, 0
+  %carry = extractvalue { i32, i1 } %usub, 1
+  store i32 %val, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+
+; GCN-ISEL-LABEL: name:   usubo32_vcc_user
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: V_SUB_I32_e64
+
+; below we check selection to v_sub/subb
+; because the only user of VCC produced by the USUBOis v_cndmask.
+; We select to VALU form to avoid unnecessary s_cselect to copy SCC to VCC
+
+; GCN-LABEL: @usubo32_vcc_user
+;
+; CISI: v_sub_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; CISI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
+;
+; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; VI: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
+;
+; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GFX9: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, vcc
+;
+; GFX1010: v_sub_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
+; GFX1010: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1, [[CARRY]]
+define amdgpu_kernel void @usubo32_vcc_user(i32 addrspace(1)* %out, i1 addrspace(1)* %carryout, i32 %a, i32 %b) #0 {
+  %usub = call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
+  %val = extractvalue { i32, i1 } %usub, 0
+  %carry = extractvalue { i32, i1 } %usub, 1
+  store i32 %val, i32 addrspace(1)* %out, align 4
+  store i1 %carry, i1 addrspace(1)* %carryout
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   susubo64
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: S_SUB_U64_PSEUDO
+
+; GCN-LABEL: @susubo64
+;
+; GCN: s_sub_u32
+; GCN: s_subb_u32
+define amdgpu_kernel void @susubo64(i64 addrspace(1)* %out, i1 addrspace(1)* %carryout, i64 %a, i64 %b) #0 {
+  %usub = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
+  %val = extractvalue { i64, i1 } %usub, 0
+  %carry = extractvalue { i64, i1 } %usub, 1
+  store i64 %val, i64 addrspace(1)* %out, align 8
+  store i1 %carry, i1 addrspace(1)* %carryout
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   vusubo64
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.0
+; GCN-ISEL: V_SUB_U64_PSEUDO
+
+; GCN-LABEL: @vusubo64
+;
+; CISI: v_sub_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v0
+; CISI: v_subbrev_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; VI: v_sub_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v0
+; VI: v_subbrev_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX9: v_sub_co_u32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v0
+; GFX9: v_subbrev_co_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc
+;
+; GFX1010: v_sub_co_u32_e64 v{{[0-9]+}}, [[CARRY:s[0-9]+]], s{{[0-9]+}}, v0
+; GFX1010: v_sub_co_ci_u32_e64 v{{[0-9]+}}, [[CARRY]], s{{[0-9]+}}, 0, [[CARRY]]
+define amdgpu_kernel void @vusubo64(i64 addrspace(1)* %out, i1 addrspace(1)* %carryout, i64 %a) #0 {
+  %tid = call i32 @llvm.amdgcn.workitem.id.x()
+  %tid.ext = sext i32 %tid to i64
+  %usub = call { i64, i1 } @llvm.usub.with.overflow.i64(i64 %a, i64 %tid.ext)
+  %val = extractvalue { i64, i1 } %usub, 0
+  %carry = extractvalue { i64, i1 } %usub, 1
+  store i64 %val, i64 addrspace(1)* %out, align 8
+  store i1 %carry, i1 addrspace(1)* %carryout
+  ret void
+}
+
+; GCN-ISEL-LABEL: name:   sudiv64
+; GCN-ISEL-LABEL: body:
+; GCN-ISEL-LABEL: bb.3
+; GCN-ISEL: %[[CARRY:[0-9]+]]:sreg_64_xexec = V_ADD_I32_e64
+; GCN-ISEL: S_ADD_CO_PSEUDO %{{[0-9]+}}, %{{[0-9]+}}, %[[CARRY]]
+; GCN-ISEL: %[[CARRY:[0-9]+]]:sreg_64_xexec = V_SUB_I32_e64
+; GCN-ISEL: S_SUB_CO_PSEUDO %{{[0-9]+}}, %{{[0-9]+}}, %[[CARRY]]
+define amdgpu_kernel void @sudiv64(i64 addrspace(1)* %out, i64 %x, i64 %y) {
+  %result = udiv i64 %x, %y
+  store i64 %result, i64 addrspace(1)* %out
+  ret void
+}
+
+
+
+declare { i64, i1 } @llvm.uadd.with.overflow.i64(i64, i64) #1
+
+declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32) #1
+
+declare { i64, i1 } @llvm.usub.with.overflow.i64(i64, i64) #1
+
+declare { i32, i1 } @llvm.usub.with.overflow.i32(i32, i32) #1
+
+declare i32 @llvm.amdgcn.workitem.id.x() #1
+
+attributes #0 = { nounwind }
+attributes #1 = { nounwind readnone }
+

llvm/test/CodeGen/AMDGPU/cvt_f32_ubyte.ll

 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 2, v0
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v6, vcc, 3, v0
 ; VI-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_ubyte v8, v[2:3]
-; VI-NEXT:    flat_load_ubyte v2, v[4:5]
-; VI-NEXT:    flat_load_ubyte v3, v[6:7]
+; VI-NEXT:    flat_load_ubyte v4, v[4:5]
+; VI-NEXT:    flat_load_ubyte v5, v[6:7]
+; VI-NEXT:    flat_load_ubyte v6, v[2:3]
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
 ; VI-NEXT:    s_waitcnt vmcnt(3) lgkmcnt(3)
-; VI-NEXT:    v_cvt_f32_ubyte2_e32 v1, v8
+; VI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v4
 ; VI-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(2)
-; VI-NEXT:    v_cvt_f32_ubyte0_e32 v2, v2
+; VI-NEXT:    v_cvt_f32_ubyte0_e32 v3, v5
 ; VI-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
-; VI-NEXT:    v_cvt_f32_ubyte0_e32 v3, v3
+; VI-NEXT:    v_cvt_f32_ubyte2_e32 v1, v6
 ; VI-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
   %load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 1
   %cvt = uitofp <4 x i8> %load to <4 x float>

 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 3, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 2, v0
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; VI-NEXT:    v_add_u32_e32 v6, vcc, 6, v0
+; VI-NEXT:    flat_load_ubyte v12, v[4:5]
+; VI-NEXT:    v_add_u32_e32 v4, vcc, 6, v0
+; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
+; VI-NEXT:    v_add_u32_e32 v6, vcc, 4, v0
 ; VI-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
-; VI-NEXT:    v_add_u32_e32 v8, vcc, 4, v0
+; VI-NEXT:    v_add_u32_e32 v8, vcc, 5, v0
 ; VI-NEXT:    v_addc_u32_e32 v9, vcc, 0, v1, vcc
-; VI-NEXT:    v_add_u32_e32 v10, vcc, 5, v0
+; VI-NEXT:    v_add_u32_e32 v10, vcc, 1, v0
 ; VI-NEXT:    v_addc_u32_e32 v11, vcc, 0, v1, vcc
-; VI-NEXT:    v_add_u32_e32 v12, vcc, 1, v0
-; VI-NEXT:    v_addc_u32_e32 v13, vcc, 0, v1, vcc
+; VI-NEXT:    flat_load_ubyte v8, v[8:9]
+; VI-NEXT:    flat_load_ubyte v9, v[10:11]
 ; VI-NEXT:    flat_load_ubyte v6, v[6:7]
-; VI-NEXT:    flat_load_ubyte v7, v[8:9]
-; VI-NEXT:    flat_load_ubyte v8, v[10:11]
-; VI-NEXT:    flat_load_ubyte v9, v[12:13]
+; VI-NEXT:    flat_load_ubyte v7, v[4:5]
+; VI-NEXT:    flat_load_ubyte v2, v[2:3]
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    flat_load_ubyte v1, v[2:3]
-; VI-NEXT:    flat_load_ubyte v2, v[4:5]
-; VI-NEXT:    s_waitcnt vmcnt(6) lgkmcnt(6)
-; VI-NEXT:    v_cvt_f32_ubyte0_e32 v6, v6
 ; VI-NEXT:    s_waitcnt vmcnt(5) lgkmcnt(5)
-; VI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v7
-; VI-NEXT:    s_waitcnt vmcnt(4) lgkmcnt(4)
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v5, v8
+; VI-NEXT:    s_waitcnt vmcnt(4) lgkmcnt(4)
+; VI-NEXT:    v_cvt_f32_ubyte2_e32 v1, v9
+; VI-NEXT:    s_waitcnt vmcnt(3) lgkmcnt(3)
+; VI-NEXT:    v_cvt_f32_ubyte0_e32 v4, v6
 ; VI-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(2)
-; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
+; VI-NEXT:    v_cvt_f32_ubyte0_e32 v6, v7
 ; VI-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
-; VI-NEXT:    v_lshlrev_b32_e32 v3, 8, v1
-; VI-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_or_b32_sdwa v2, v3, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI-NEXT:    v_lshlrev_b32_e32 v2, 8, v2
+; VI-NEXT:    v_or_b32_sdwa v2, v2, v12 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; VI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v2
-; VI-NEXT:    v_cvt_f32_ubyte2_e32 v1, v9
+; VI-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
 ; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v2
 ; VI-NEXT:    buffer_store_dwordx3 v[4:6], off, s[4:7], 0 offset:16
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <7 x i8>, <7 x i8> addrspace(1)* %in, i32 %tid
   %load = load <7 x i8>, <7 x i8> addrspace(1)* %gep, align 1
   %cvt = uitofp <7 x i8> %load to <7 x float>

 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v0
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, 3, v0
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 2, v0
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
 ; VI-NEXT:    v_add_u32_e32 v6, vcc, 1, v0
 ; VI-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
+; VI-NEXT:    flat_load_ubyte v4, v[4:5]
+; VI-NEXT:    flat_load_ubyte v5, v[6:7]
 ; VI-NEXT:    flat_load_ubyte v2, v[2:3]
-; VI-NEXT:    flat_load_ubyte v3, v[4:5]
-; VI-NEXT:    flat_load_ubyte v4, v[6:7]
 ; VI-NEXT:    flat_load_ubyte v0, v[0:1]
-; VI-NEXT:    s_waitcnt vmcnt(3) lgkmcnt(3)
+; VI-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
 ; VI-NEXT:    v_lshlrev_b32_e32 v1, 8, v2
-; VI-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(2)
-; VI-NEXT:    v_or_b32_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; VI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v1
-; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v1
+; VI-NEXT:    v_or_b32_sdwa v1, v1, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; VI-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; VI-NEXT:    v_cvt_f32_ubyte0_e32 v0, v0
-; VI-NEXT:    v_cvt_f32_ubyte2_e32 v1, v4
+; VI-NEXT:    v_cvt_f32_ubyte3_e32 v3, v1
+; VI-NEXT:    v_cvt_f32_ubyte2_e32 v2, v1
+; VI-NEXT:    v_cvt_f32_ubyte2_e32 v1, v5
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep = getelementptr <4 x i8>, <4 x i8> addrspace(1)* %in, i32 %tid
   %load = load <4 x i8>, <4 x i8> addrspace(1)* %gep, align 1
   %ext = zext <4 x i8> %load to <4 x i32>
   %cvt = uitofp <4 x i32> %ext to <4 x float>
   store <4 x float> %cvt, <4 x float> addrspace(1)* %out, align 16

llvm/test/CodeGen/AMDGPU/fast-unaligned-load-store.global.ll

 define i32 @global_load_2xi16_align1(i16 addrspace(1)* %p) #0 {
 ; GFX7-ALIGNED-LABEL: global_load_2xi16_align1:
 ; GFX7-ALIGNED:       ; %bb.0:
 ; GFX7-ALIGNED-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-ALIGNED-NEXT:    v_add_i32_e32 v2, vcc, 2, v0
 ; GFX7-ALIGNED-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
 ; GFX7-ALIGNED-NEXT:    v_add_i32_e32 v4, vcc, 1, v0
 ; GFX7-ALIGNED-NEXT:    v_addc_u32_e32 v5, vcc, 0, v1, vcc
-; GFX7-ALIGNED-NEXT:    flat_load_ubyte v6, v[0:1]
-; GFX7-ALIGNED-NEXT:    v_add_i32_e32 v0, vcc, 3, v0
-; GFX7-ALIGNED-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; GFX7-ALIGNED-NEXT:    flat_load_ubyte v2, v[2:3]
-; GFX7-ALIGNED-NEXT:    flat_load_ubyte v3, v[4:5]
+; GFX7-ALIGNED-NEXT:    v_add_i32_e32 v6, vcc, 3, v0
+; GFX7-ALIGNED-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
 ; GFX7-ALIGNED-NEXT:    flat_load_ubyte v0, v[0:1]
+; GFX7-ALIGNED-NEXT:    flat_load_ubyte v1, v[6:7]
+; GFX7-ALIGNED-NEXT:    flat_load_ubyte v4, v[4:5]
+; GFX7-ALIGNED-NEXT:    flat_load_ubyte v2, v[2:3]
+; GFX7-ALIGNED-NEXT:    s_waitcnt vmcnt(2) lgkmcnt(2)
+; GFX7-ALIGNED-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX7-ALIGNED-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
-; GFX7-ALIGNED-NEXT:    v_lshlrev_b32_e32 v1, 8, v3
+; GFX7-ALIGNED-NEXT:    v_lshlrev_b32_e32 v3, 8, v4
 ; GFX7-ALIGNED-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; GFX7-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
-; GFX7-ALIGNED-NEXT:    v_or_b32_e32 v0, v0, v2
-; GFX7-ALIGNED-NEXT:    v_or_b32_e32 v1, v1, v6
-; GFX7-ALIGNED-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; GFX7-ALIGNED-NEXT:    v_or_b32_e32 v0, v1, v0
+; GFX7-ALIGNED-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX7-ALIGNED-NEXT:    v_or_b32_e32 v0, v3, v0
+; GFX7-ALIGNED-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX7-ALIGNED-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX7-ALIGNED-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX7-UNALIGNED-LABEL: global_load_2xi16_align1:
 ; GFX7-UNALIGNED:       ; %bb.0:
 ; GFX7-UNALIGNED-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-UNALIGNED-NEXT:    flat_load_dword v0, v[0:1]
 ; GFX7-UNALIGNED-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
 ; GFX7-UNALIGNED-NEXT:    s_setpc_b64 s[30:31]

llvm/test/CodeGen/AMDGPU/max.i16.ll

 ; VI-NEXT:    v_mov_b32_e32 v3, s1
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, s0, v4
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
 ; VI-NEXT:    v_mov_b32_e32 v5, s5
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, s4, v4
 ; VI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v5, vcc
 ; VI-NEXT:    v_add_u32_e32 v6, vcc, 4, v0
 ; VI-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
-; VI-NEXT:    flat_load_ushort v8, v[6:7]
-; VI-NEXT:    flat_load_dword v9, v[0:1]
+; VI-NEXT:    flat_load_ushort v6, v[6:7]
+; VI-NEXT:    flat_load_dword v7, v[0:1]
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, 4, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v3, vcc
 ; VI-NEXT:    flat_load_ushort v0, v[0:1]
-; VI-NEXT:    flat_load_dword v1, v[2:3]
-; VI-NEXT:    v_add_u32_e32 v6, vcc, 4, v4
-; VI-NEXT:    v_addc_u32_e32 v7, vcc, 0, v5, vcc
+; VI-NEXT:    flat_load_dword v8, v[2:3]
+; VI-NEXT:    v_add_u32_e32 v2, vcc, 4, v4
+; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v5, vcc
 ; VI-NEXT:    s_waitcnt vmcnt(1) lgkmcnt(1)
-; VI-NEXT:    v_max_i16_e32 v0, v8, v0
+; VI-NEXT:    v_max_i16_e32 v0, v6, v0
 ; VI-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_max_i16_e32 v2, v9, v1
-; VI-NEXT:    v_max_i16_sdwa v1, v9, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_e32 v1, v2, v1
-; VI-NEXT:    flat_store_short v[6:7], v0
+; VI-NEXT:    v_max_i16_e32 v1, v7, v8
+; VI-NEXT:    v_max_i16_sdwa v7, v7, v8 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI-NEXT:    v_or_b32_e32 v1, v1, v7
+; VI-NEXT:    flat_store_short v[2:3], v0
 ; VI-NEXT:    flat_store_dword v[4:5], v1
 ; VI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: v_test_imax_sge_v3i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x34
 ; GFX9-NEXT:    v_lshlrev_b32_e32 v4, 3, v0

llvm/test/CodeGen/AMDGPU/promote-constOffset-to-imm.ll

 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ;
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
-; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-4096
-; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
+; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
+; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-4096
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
 ;
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-2048
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-2048
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-2048

 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ;
+; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:2048
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:3072
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
-; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
-; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:2048
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:3072
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
+; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 ;
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:-2048
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024

 
 define amdgpu_kernel void @Offset64(i8 addrspace(1)*  %buffer) {
 ; GCN-LABEL: Offset64:
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ;
-; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-4096
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
+; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-4096
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ;
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:-2048
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 entry:

 ; TODO: Support load4 as anchor instruction.
 define amdgpu_kernel void @p32Offset64(i8 addrspace(1)*  %buffer) {
 ; GCN-LABEL: p32Offset64:
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
 ;
+; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:2048
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:3072
 ; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
-; GFX9:    global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ;
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:-2048
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dword {{v[0-9]+}}, v[{{[0-9]+:[0-9]+}}], off offset:1024
 entry:
   %call = tail call i64 @_Z13get_global_idj(i32 0)
   %conv = and i64 %call, 255

 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ; GFX8:    flat_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}]
 ;
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
-; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
+; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX9:    global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off offset:2048
 ;
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}
 ; GFX10:   global_load_dwordx2 v[{{[0-9]+:[0-9]+}}], v[{{[0-9]+:[0-9]+}}], off{{$}}

llvm/test/CodeGen/AMDGPU/sdiv64.ll

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v9, v3, v6
 ; GCN-NEXT:    v_mul_lo_u32 v6, v3, v6
 ; GCN-NEXT:    v_mul_hi_u32 v8, v3, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v7, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v5
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v10
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v10, v6
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v7, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_addc_u32_e64 v2, vcc, v2, v5, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s10, s14
 ; GCN-NEXT:    s_addc_u32 s1, s11, s14

 ; GCN-IR-NEXT:    v_cmp_ne_u64_e64 s[4:5], 63, v[11:12]
 ; GCN-IR-NEXT:    s_or_b64 s[6:7], s[6:7], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[8:9], s[6:7], -1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v18, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v6, v4
 ; GCN-IR-NEXT:    v_mov_b32_e32 v1, v5
 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v7, v10, 0, s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[8:9], s[4:5]
-; GCN-IR-NEXT:    v_mov_b32_e32 v17, v18
+; GCN-IR-NEXT:    v_mov_b32_e32 v15, v18
 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v9, 0, s[6:7]
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB1_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v15, vcc, 1, v11
-; GCN-IR-NEXT:    v_addc_u32_e32 v16, vcc, 0, v12, vcc
+; GCN-IR-NEXT:    v_add_i32_e32 v16, vcc, 1, v11
+; GCN-IR-NEXT:    v_addc_u32_e32 v17, vcc, 0, v12, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e64 v0, s[4:5], 63, v11
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[15:16], v[11:12]
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[16:17], v[11:12]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v11, 0
 ; GCN-IR-NEXT:    v_lshl_b64 v[7:8], v[9:10], v0
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB1_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, -1, v2
-; GCN-IR-NEXT:    v_lshr_b64 v[15:16], v[9:10], v15
+; GCN-IR-NEXT:    v_lshr_b64 v[16:17], v[9:10], v16
 ; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, -1, v3, vcc
 ; GCN-IR-NEXT:    v_not_b32_e32 v10, v13
 ; GCN-IR-NEXT:    v_not_b32_e32 v11, v18
 ; GCN-IR-NEXT:    v_add_i32_e32 v13, vcc, v10, v14
-; GCN-IR-NEXT:    v_addc_u32_e32 v14, vcc, v11, v17, vcc
-; GCN-IR-NEXT:    v_mov_b32_e32 v17, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v18, 0
+; GCN-IR-NEXT:    v_addc_u32_e32 v14, vcc, v11, v15, vcc
+; GCN-IR-NEXT:    v_mov_b32_e32 v19, 0
 ; GCN-IR-NEXT:  BB1_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GCN-IR-NEXT:    v_lshl_b64 v[15:16], v[15:16], 1
+; GCN-IR-NEXT:    v_lshl_b64 v[16:17], v[16:17], 1
 ; GCN-IR-NEXT:    v_lshrrev_b32_e32 v10, 31, v8
-; GCN-IR-NEXT:    v_or_b32_e32 v10, v15, v10
+; GCN-IR-NEXT:    v_or_b32_e32 v10, v16, v10
 ; GCN-IR-NEXT:    v_lshl_b64 v[7:8], v[7:8], 1
 ; GCN-IR-NEXT:    v_sub_i32_e32 v11, vcc, v0, v10
-; GCN-IR-NEXT:    v_subb_u32_e32 v11, vcc, v9, v16, vcc
-; GCN-IR-NEXT:    v_or_b32_e32 v7, v17, v7
-; GCN-IR-NEXT:    v_add_i32_e32 v17, vcc, 1, v13
+; GCN-IR-NEXT:    v_subb_u32_e32 v11, vcc, v9, v17, vcc
+; GCN-IR-NEXT:    v_or_b32_e32 v7, v18, v7
+; GCN-IR-NEXT:    v_add_i32_e32 v18, vcc, 1, v13
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v15, 31, v11
-; GCN-IR-NEXT:    v_or_b32_e32 v8, v18, v8
-; GCN-IR-NEXT:    v_addc_u32_e32 v18, vcc, 0, v14, vcc
-; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[17:18], v[13:14]
-; GCN-IR-NEXT:    v_mov_b32_e32 v13, v17
+; GCN-IR-NEXT:    v_or_b32_e32 v8, v19, v8
+; GCN-IR-NEXT:    v_addc_u32_e32 v19, vcc, 0, v14, vcc
+; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[18:19], v[13:14]
+; GCN-IR-NEXT:    v_mov_b32_e32 v13, v18
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-IR-NEXT:    v_and_b32_e32 v11, 1, v15
-; GCN-IR-NEXT:    v_and_b32_e32 v19, v15, v3
+; GCN-IR-NEXT:    v_and_b32_e32 v20, v15, v3
 ; GCN-IR-NEXT:    v_and_b32_e32 v15, v15, v2
-; GCN-IR-NEXT:    v_sub_i32_e64 v15, s[4:5], v10, v15
-; GCN-IR-NEXT:    v_mov_b32_e32 v14, v18
-; GCN-IR-NEXT:    v_mov_b32_e32 v18, v12
-; GCN-IR-NEXT:    v_subb_u32_e64 v16, s[4:5], v16, v19, s[4:5]
+; GCN-IR-NEXT:    v_sub_i32_e64 v16, s[4:5], v10, v15
+; GCN-IR-NEXT:    v_mov_b32_e32 v14, v19
+; GCN-IR-NEXT:    v_mov_b32_e32 v19, v12
+; GCN-IR-NEXT:    v_subb_u32_e64 v17, s[4:5], v17, v20, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v17, v11
+; GCN-IR-NEXT:    v_mov_b32_e32 v18, v11
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB1_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB1_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[7:8], 1
 ; GCN-IR-NEXT:    v_or_b32_e32 v7, v12, v3

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v6, 24, 0, s[4:5]
 ; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[4:5], -1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, v2
 ; GCN-IR-NEXT:    v_mov_b32_e32 v7, v11
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB11_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, 1, v4
-; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v5, vcc
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[6:7], v[4:5]
+; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, 1, v4
+; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, 0, v5, vcc
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[8:9], v[4:5]
 ; GCN-IR-NEXT:    v_sub_i32_e64 v4, s[4:5], 63, v4
-; GCN-IR-NEXT:    v_mov_b32_e32 v8, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v6, 0
 ; GCN-IR-NEXT:    v_lshl_b64 v[4:5], 24, v4
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB11_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    v_lshr_b64 v[12:13], 24, v6
-; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, -1, v0
-; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, -1, v1, vcc
+; GCN-IR-NEXT:    v_lshr_b64 v[12:13], 24, v8
+; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, -1, v0
+; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e32 v10, vcc, 58, v10
 ; GCN-IR-NEXT:    v_mov_b32_e32 v14, 0
 ; GCN-IR-NEXT:    v_subb_u32_e32 v11, vcc, 0, v11, vcc
 ; GCN-IR-NEXT:    v_mov_b32_e32 v15, 0
 ; GCN-IR-NEXT:  BB11_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    v_lshl_b64 v[12:13], v[12:13], 1
-; GCN-IR-NEXT:    v_lshrrev_b32_e32 v8, 31, v5
-; GCN-IR-NEXT:    v_or_b32_e32 v12, v12, v8
+; GCN-IR-NEXT:    v_lshrrev_b32_e32 v6, 31, v5
+; GCN-IR-NEXT:    v_or_b32_e32 v12, v12, v6
 ; GCN-IR-NEXT:    v_lshl_b64 v[4:5], v[4:5], 1
-; GCN-IR-NEXT:    v_sub_i32_e32 v8, vcc, v6, v12
-; GCN-IR-NEXT:    v_subb_u32_e32 v8, vcc, v7, v13, vcc
+; GCN-IR-NEXT:    v_sub_i32_e32 v6, vcc, v8, v12
+; GCN-IR-NEXT:    v_subb_u32_e32 v6, vcc, v9, v13, vcc
 ; GCN-IR-NEXT:    v_or_b32_e32 v4, v14, v4
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v14, 31, v8
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v14, 31, v6
 ; GCN-IR-NEXT:    v_and_b32_e32 v17, v14, v0
-; GCN-IR-NEXT:    v_and_b32_e32 v8, 1, v14
+; GCN-IR-NEXT:    v_and_b32_e32 v6, 1, v14
 ; GCN-IR-NEXT:    v_and_b32_e32 v16, v14, v1
 ; GCN-IR-NEXT:    v_add_i32_e32 v14, vcc, 1, v10
 ; GCN-IR-NEXT:    v_or_b32_e32 v5, v15, v5
 ; GCN-IR-NEXT:    v_addc_u32_e32 v15, vcc, 0, v11, vcc
 ; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[14:15], v[10:11]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v10, v14
-; GCN-IR-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
 ; GCN-IR-NEXT:    v_sub_i32_e64 v12, s[4:5], v12, v17
 ; GCN-IR-NEXT:    v_mov_b32_e32 v11, v15
-; GCN-IR-NEXT:    v_mov_b32_e32 v15, v9
+; GCN-IR-NEXT:    v_mov_b32_e32 v15, v7
 ; GCN-IR-NEXT:    v_subb_u32_e64 v13, s[4:5], v13, v16, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v14, v8
+; GCN-IR-NEXT:    v_mov_b32_e32 v14, v6
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB11_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB11_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[0:1], v[4:5], 1
-; GCN-IR-NEXT:    v_or_b32_e32 v7, v9, v1
-; GCN-IR-NEXT:    v_or_b32_e32 v6, v8, v0
+; GCN-IR-NEXT:    v_or_b32_e32 v7, v7, v1
+; GCN-IR-NEXT:    v_or_b32_e32 v6, v6, v0
 ; GCN-IR-NEXT:  BB11_6: ; %Flow4
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GCN-IR-NEXT:    v_xor_b32_e32 v0, v6, v2
 ; GCN-IR-NEXT:    v_xor_b32_e32 v1, v7, v3
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-IR-NEXT:    s_setpc_b64 s[30:31]
   %result = sdiv i64 24, %x

 ; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[4:5], -1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, v2
 ; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v7, v11
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB12_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, 1, v4
-; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v5, vcc
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[6:7], v[4:5]
+; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, 1, v4
+; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, 0, v5, vcc
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[8:9], v[4:5]
 ; GCN-IR-NEXT:    v_sub_i32_e64 v4, s[4:5], 63, v4
 ; GCN-IR-NEXT:    v_lshl_b64 v[4:5], s[8:9], v4
-; GCN-IR-NEXT:    v_mov_b32_e32 v8, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v6, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB12_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:    s_mov_b32 s4, 0x8000
-; GCN-IR-NEXT:    v_lshr_b64 v[12:13], s[4:5], v6
-; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, -1, v0
-; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, -1, v1, vcc
+; GCN-IR-NEXT:    v_lshr_b64 v[12:13], s[4:5], v8
+; GCN-IR-NEXT:    v_add_i32_e32 v8, vcc, -1, v0
+; GCN-IR-NEXT:    v_addc_u32_e32 v9, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e32 v10, vcc, 47, v10
 ; GCN-IR-NEXT:    v_mov_b32_e32 v14, 0
 ; GCN-IR-NEXT:    v_subb_u32_e32 v11, vcc, 0, v11, vcc
 ; GCN-IR-NEXT:    v_mov_b32_e32 v15, 0
 ; GCN-IR-NEXT:  BB12_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    v_lshl_b64 v[12:13], v[12:13], 1
-; GCN-IR-NEXT:    v_lshrrev_b32_e32 v8, 31, v5
-; GCN-IR-NEXT:    v_or_b32_e32 v12, v12, v8
+; GCN-IR-NEXT:    v_lshrrev_b32_e32 v6, 31, v5
+; GCN-IR-NEXT:    v_or_b32_e32 v12, v12, v6
 ; GCN-IR-NEXT:    v_lshl_b64 v[4:5], v[4:5], 1
-; GCN-IR-NEXT:    v_sub_i32_e32 v8, vcc, v6, v12
-; GCN-IR-NEXT:    v_subb_u32_e32 v8, vcc, v7, v13, vcc
+; GCN-IR-NEXT:    v_sub_i32_e32 v6, vcc, v8, v12
+; GCN-IR-NEXT:    v_subb_u32_e32 v6, vcc, v9, v13, vcc
 ; GCN-IR-NEXT:    v_or_b32_e32 v4, v14, v4
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v14, 31, v8
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v14, 31, v6
 ; GCN-IR-NEXT:    v_and_b32_e32 v17, v14, v0
-; GCN-IR-NEXT:    v_and_b32_e32 v8, 1, v14
+; GCN-IR-NEXT:    v_and_b32_e32 v6, 1, v14
 ; GCN-IR-NEXT:    v_and_b32_e32 v16, v14, v1
 ; GCN-IR-NEXT:    v_add_i32_e32 v14, vcc, 1, v10
 ; GCN-IR-NEXT:    v_or_b32_e32 v5, v15, v5
 ; GCN-IR-NEXT:    v_addc_u32_e32 v15, vcc, 0, v11, vcc
 ; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[14:15], v[10:11]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v10, v14
-; GCN-IR-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
 ; GCN-IR-NEXT:    v_sub_i32_e64 v12, s[4:5], v12, v17
 ; GCN-IR-NEXT:    v_mov_b32_e32 v11, v15
-; GCN-IR-NEXT:    v_mov_b32_e32 v15, v9
+; GCN-IR-NEXT:    v_mov_b32_e32 v15, v7
 ; GCN-IR-NEXT:    v_subb_u32_e64 v13, s[4:5], v13, v16, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v14, v8
+; GCN-IR-NEXT:    v_mov_b32_e32 v14, v6
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB12_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB12_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[0:1], v[4:5], 1
-; GCN-IR-NEXT:    v_or_b32_e32 v7, v9, v1
-; GCN-IR-NEXT:    v_or_b32_e32 v6, v8, v0
+; GCN-IR-NEXT:    v_or_b32_e32 v7, v7, v1
+; GCN-IR-NEXT:    v_or_b32_e32 v6, v6, v0
 ; GCN-IR-NEXT:  BB12_6: ; %Flow4
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GCN-IR-NEXT:    v_xor_b32_e32 v0, v6, v2
 ; GCN-IR-NEXT:    v_xor_b32_e32 v1, v7, v3
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, v0, v2
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-IR-NEXT:    s_setpc_b64 s[30:31]
   %result = sdiv i64 32768, %x

llvm/test/CodeGen/AMDGPU/srem64.ll

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v5
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v9, v3, v6
 ; GCN-NEXT:    v_mul_lo_u32 v6, v3, v6
 ; GCN-NEXT:    v_mul_hi_u32 v8, v3, v5
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v7, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v5
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v10
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v10, v6
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v7, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_addc_u32_e64 v2, vcc, v2, v5, s[0:1]
 ; GCN-NEXT:    s_add_u32 s0, s2, s10
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v3

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-IR-NEXT:    v_cmp_ne_u64_e32 vcc, 63, v[3:4]
 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v2, 24, 0, s[4:5]
 ; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[4:5], -1
 ; GCN-IR-NEXT:    v_mov_b32_e32 v5, v9
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB11_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v5, vcc, 1, v3
-; GCN-IR-NEXT:    v_addc_u32_e32 v6, vcc, 0, v4, vcc
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, 1, v3
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v4, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e64 v2, s[4:5], 63, v3
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[5:6], v[3:4]
-; GCN-IR-NEXT:    v_mov_b32_e32 v6, 0
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[6:7], v[3:4]
+; GCN-IR-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], 24, v2
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB11_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    v_add_i32_e32 v4, vcc, -1, v0
-; GCN-IR-NEXT:    v_lshr_b64 v[10:11], 24, v5
-; GCN-IR-NEXT:    v_addc_u32_e32 v5, vcc, -1, v1, vcc
+; GCN-IR-NEXT:    v_lshr_b64 v[10:11], 24, v6
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, -1, v0
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e32 v8, vcc, 58, v8
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-IR-NEXT:    v_subb_u32_e32 v9, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_mov_b32_e32 v13, 0
 ; GCN-IR-NEXT:  BB11_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    v_lshl_b64 v[10:11], v[10:11], 1
-; GCN-IR-NEXT:    v_lshrrev_b32_e32 v6, 31, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v6
+; GCN-IR-NEXT:    v_lshrrev_b32_e32 v4, 31, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v4
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_sub_i32_e32 v6, vcc, v4, v10
-; GCN-IR-NEXT:    v_subb_u32_e32 v6, vcc, v5, v11, vcc
+; GCN-IR-NEXT:    v_sub_i32_e32 v4, vcc, v6, v10
+; GCN-IR-NEXT:    v_subb_u32_e32 v4, vcc, v7, v11, vcc
 ; GCN-IR-NEXT:    v_or_b32_e32 v2, v12, v2
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v6
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v4
 ; GCN-IR-NEXT:    v_and_b32_e32 v15, v12, v0
-; GCN-IR-NEXT:    v_and_b32_e32 v6, 1, v12
+; GCN-IR-NEXT:    v_and_b32_e32 v4, 1, v12
 ; GCN-IR-NEXT:    v_and_b32_e32 v14, v12, v1
 ; GCN-IR-NEXT:    v_add_i32_e32 v12, vcc, 1, v8
 ; GCN-IR-NEXT:    v_or_b32_e32 v3, v13, v3
 ; GCN-IR-NEXT:    v_addc_u32_e32 v13, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[12:13], v[8:9]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v8, v12
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    v_sub_i32_e64 v10, s[4:5], v10, v15
 ; GCN-IR-NEXT:    v_mov_b32_e32 v9, v13
-; GCN-IR-NEXT:    v_mov_b32_e32 v13, v7
+; GCN-IR-NEXT:    v_mov_b32_e32 v13, v5
 ; GCN-IR-NEXT:    v_subb_u32_e64 v11, s[4:5], v11, v14, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v12, v6
+; GCN-IR-NEXT:    v_mov_b32_e32 v12, v4
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB11_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB11_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_or_b32_e32 v5, v7, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v2, v6, v2
+; GCN-IR-NEXT:    v_or_b32_e32 v5, v5, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v2, v4, v2
 ; GCN-IR-NEXT:  BB11_6: ; %Flow4
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GCN-IR-NEXT:    v_mul_lo_u32 v3, v0, v5
 ; GCN-IR-NEXT:    v_mul_hi_u32 v4, v0, v2
 ; GCN-IR-NEXT:    v_mul_lo_u32 v1, v1, v2
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, v2
 ; GCN-IR-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
 ; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1

 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v4, v4, 0, s[4:5]
 ; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[4:5], -1
 ; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v5, v9
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB12_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v4, vcc, 1, v2
-; GCN-IR-NEXT:    v_addc_u32_e32 v5, vcc, 0, v3, vcc
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[4:5], v[2:3]
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, 1, v2
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v3, vcc
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[6:7], v[2:3]
 ; GCN-IR-NEXT:    v_sub_i32_e64 v2, s[4:5], 63, v2
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], s[8:9], v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v6, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB12_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:    s_mov_b32 s4, 0x8000
-; GCN-IR-NEXT:    v_lshr_b64 v[10:11], s[4:5], v4
-; GCN-IR-NEXT:    v_add_i32_e32 v4, vcc, -1, v0
-; GCN-IR-NEXT:    v_addc_u32_e32 v5, vcc, -1, v1, vcc
+; GCN-IR-NEXT:    v_lshr_b64 v[10:11], s[4:5], v6
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, -1, v0
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e32 v8, vcc, 47, v8
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-IR-NEXT:    v_subb_u32_e32 v9, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_mov_b32_e32 v13, 0
 ; GCN-IR-NEXT:  BB12_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    v_lshl_b64 v[10:11], v[10:11], 1
-; GCN-IR-NEXT:    v_lshrrev_b32_e32 v6, 31, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v6
+; GCN-IR-NEXT:    v_lshrrev_b32_e32 v4, 31, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v4
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_sub_i32_e32 v6, vcc, v4, v10
-; GCN-IR-NEXT:    v_subb_u32_e32 v6, vcc, v5, v11, vcc
+; GCN-IR-NEXT:    v_sub_i32_e32 v4, vcc, v6, v10
+; GCN-IR-NEXT:    v_subb_u32_e32 v4, vcc, v7, v11, vcc
 ; GCN-IR-NEXT:    v_or_b32_e32 v2, v12, v2
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v6
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v4
 ; GCN-IR-NEXT:    v_and_b32_e32 v15, v12, v0
-; GCN-IR-NEXT:    v_and_b32_e32 v6, 1, v12
+; GCN-IR-NEXT:    v_and_b32_e32 v4, 1, v12
 ; GCN-IR-NEXT:    v_and_b32_e32 v14, v12, v1
 ; GCN-IR-NEXT:    v_add_i32_e32 v12, vcc, 1, v8
 ; GCN-IR-NEXT:    v_or_b32_e32 v3, v13, v3
 ; GCN-IR-NEXT:    v_addc_u32_e32 v13, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[12:13], v[8:9]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v8, v12
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    v_sub_i32_e64 v10, s[4:5], v10, v15
 ; GCN-IR-NEXT:    v_mov_b32_e32 v9, v13
-; GCN-IR-NEXT:    v_mov_b32_e32 v13, v7
+; GCN-IR-NEXT:    v_mov_b32_e32 v13, v5
 ; GCN-IR-NEXT:    v_subb_u32_e64 v11, s[4:5], v11, v14, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v12, v6
+; GCN-IR-NEXT:    v_mov_b32_e32 v12, v4
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB12_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB12_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_or_b32_e32 v5, v7, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v4, v6, v2
+; GCN-IR-NEXT:    v_or_b32_e32 v5, v5, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v4, v4, v2
 ; GCN-IR-NEXT:  BB12_6: ; %Flow4
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GCN-IR-NEXT:    v_mul_lo_u32 v2, v0, v5
 ; GCN-IR-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-IR-NEXT:    v_mul_lo_u32 v1, v1, v4
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, v4
 ; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1

llvm/test/CodeGen/AMDGPU/udiv64.ll

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v13, v1, v5
 ; GCN-NEXT:    v_mul_hi_u32 v12, v1, v6
 ; GCN-NEXT:    v_mul_hi_u32 v10, v3, v6
 ; GCN-NEXT:    v_mul_lo_u32 v6, v3, v6
 ; GCN-NEXT:    v_mul_hi_u32 v7, v3, v5
 ; GCN-NEXT:    v_add_i32_e32 v11, vcc, v12, v11
 ; GCN-NEXT:    v_addc_u32_e32 v12, vcc, v9, v13, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v5
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v11
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v11, v6
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v12, v10, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v7, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v9, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_addc_u32_e64 v2, vcc, v2, v5, s[2:3]
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_mov_b32_e32 v3, s6

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v2, v2, 0, s[4:5]
 ; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[4:5], -1
 ; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v3, v9
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB9_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v10, vcc, 1, v4
-; GCN-IR-NEXT:    v_addc_u32_e32 v11, vcc, 0, v5, vcc
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, 1, v4
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v5, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e64 v2, s[4:5], 63, v4
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[6:7], v[4:5]
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], s[8:9], v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v6, 0
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[10:11], v[4:5]
+; GCN-IR-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB9_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    v_add_i32_e32 v4, vcc, -1, v0
-; GCN-IR-NEXT:    v_addc_u32_e32 v5, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:    s_mov_b32 s4, 0x8000
+; GCN-IR-NEXT:    v_lshr_b64 v[10:11], s[4:5], v6
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, -1, v0
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e32 v8, vcc, 47, v8
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
-; GCN-IR-NEXT:    v_lshr_b64 v[10:11], s[4:5], v10
 ; GCN-IR-NEXT:    v_subb_u32_e32 v9, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_mov_b32_e32 v13, 0
 ; GCN-IR-NEXT:  BB9_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    v_lshl_b64 v[10:11], v[10:11], 1
-; GCN-IR-NEXT:    v_lshrrev_b32_e32 v6, 31, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v6
+; GCN-IR-NEXT:    v_lshrrev_b32_e32 v4, 31, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v4
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_sub_i32_e32 v6, vcc, v4, v10
-; GCN-IR-NEXT:    v_subb_u32_e32 v6, vcc, v5, v11, vcc
+; GCN-IR-NEXT:    v_sub_i32_e32 v4, vcc, v6, v10
+; GCN-IR-NEXT:    v_subb_u32_e32 v4, vcc, v7, v11, vcc
 ; GCN-IR-NEXT:    v_or_b32_e32 v2, v12, v2
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v6
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v4
 ; GCN-IR-NEXT:    v_and_b32_e32 v15, v12, v0
-; GCN-IR-NEXT:    v_and_b32_e32 v6, 1, v12
+; GCN-IR-NEXT:    v_and_b32_e32 v4, 1, v12
 ; GCN-IR-NEXT:    v_and_b32_e32 v14, v12, v1
 ; GCN-IR-NEXT:    v_add_i32_e32 v12, vcc, 1, v8
 ; GCN-IR-NEXT:    v_or_b32_e32 v3, v13, v3
 ; GCN-IR-NEXT:    v_addc_u32_e32 v13, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[12:13], v[8:9]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v8, v12
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    v_sub_i32_e64 v10, s[4:5], v10, v15
 ; GCN-IR-NEXT:    v_mov_b32_e32 v9, v13
-; GCN-IR-NEXT:    v_mov_b32_e32 v13, v7
+; GCN-IR-NEXT:    v_mov_b32_e32 v13, v5
 ; GCN-IR-NEXT:    v_subb_u32_e64 v11, s[4:5], v11, v14, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v12, v6
+; GCN-IR-NEXT:    v_mov_b32_e32 v12, v4
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB9_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB9_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[0:1], v[2:3], 1
-; GCN-IR-NEXT:    v_or_b32_e32 v3, v7, v1
-; GCN-IR-NEXT:    v_or_b32_e32 v2, v6, v0
+; GCN-IR-NEXT:    v_or_b32_e32 v3, v5, v1
+; GCN-IR-NEXT:    v_or_b32_e32 v2, v4, v0
 ; GCN-IR-NEXT:  BB9_6: ; %Flow4
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v0, v2
 ; GCN-IR-NEXT:    v_mov_b32_e32 v1, v3
 ; GCN-IR-NEXT:    s_setpc_b64 s[30:31]
   %result = udiv i64 32768, %x
   ret i64 %result
 }

 ; GCN-NEXT:    v_mul_hi_u32 v9, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v10, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v11, v2, v4
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v9, v5
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v8, v10, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v10, v2, v6
 ; GCN-NEXT:    v_mul_hi_u32 v6, v2, v6
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v5, vcc, v10, v5
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v10
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v9, v6, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v11, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v4, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v11, v2, v8
 ; GCN-NEXT:    v_mul_hi_u32 v12, v2, v6
 ; GCN-NEXT:    v_mul_hi_u32 v13, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v7, vcc, v11, v7
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v10, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v12, v4, v8
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v8
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v12, v7
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v12
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v8, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v13, v9, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v10, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[4:5]
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

llvm/test/CodeGen/AMDGPU/urem64.ll

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v12, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v11, v0, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v4, v7
 ; GCN-NEXT:    v_mul_lo_u32 v7, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v8, v4, v6
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, v11, v10
 ; GCN-NEXT:    v_addc_u32_e32 v11, vcc, v2, v12, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v6
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v10
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v10, v7
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v11, v9, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v8, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v6, vcc, v2, v6, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GCN-NEXT:    v_addc_u32_e64 v3, vcc, v3, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v9, v0, v6
 ; GCN-NEXT:    v_mul_hi_u32 v10, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v11, v2, v4
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v9, v5
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, v8, v10, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v10, v2, v6
 ; GCN-NEXT:    v_mul_hi_u32 v6, v2, v6
 ; GCN-NEXT:    v_mul_lo_u32 v2, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v5, vcc, v10, v5
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v10
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, v9, v6, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v11, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e64 v1, vcc, v1, v4, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc

 ; GCN-IR-NEXT:    v_cndmask_b32_e64 v4, v4, 0, s[4:5]
 ; GCN-IR-NEXT:    s_xor_b64 s[4:5], s[4:5], -1
 ; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v5, v9
 ; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], vcc
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[6:7], s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB8_6
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    v_add_i32_e32 v4, vcc, 1, v2
-; GCN-IR-NEXT:    v_addc_u32_e32 v5, vcc, 0, v3, vcc
-; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[4:5], v[2:3]
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, 1, v2
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, 0, v3, vcc
+; GCN-IR-NEXT:    v_cmp_ge_u64_e32 vcc, v[6:7], v[2:3]
 ; GCN-IR-NEXT:    v_sub_i32_e64 v2, s[4:5], 63, v2
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], s[8:9], v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v6, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    s_and_saveexec_b64 s[4:5], vcc
 ; GCN-IR-NEXT:    s_xor_b64 s[10:11], exec, s[4:5]
 ; GCN-IR-NEXT:    s_cbranch_execz BB8_5
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:    s_mov_b32 s4, 0x8000
-; GCN-IR-NEXT:    v_lshr_b64 v[10:11], s[4:5], v4
-; GCN-IR-NEXT:    v_add_i32_e32 v4, vcc, -1, v0
-; GCN-IR-NEXT:    v_addc_u32_e32 v5, vcc, -1, v1, vcc
+; GCN-IR-NEXT:    v_lshr_b64 v[10:11], s[4:5], v6
+; GCN-IR-NEXT:    v_add_i32_e32 v6, vcc, -1, v0
+; GCN-IR-NEXT:    v_addc_u32_e32 v7, vcc, -1, v1, vcc
 ; GCN-IR-NEXT:    v_sub_i32_e32 v8, vcc, 47, v8
 ; GCN-IR-NEXT:    v_mov_b32_e32 v12, 0
 ; GCN-IR-NEXT:    v_subb_u32_e32 v9, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_mov_b32_e32 v13, 0
 ; GCN-IR-NEXT:  BB8_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    v_lshl_b64 v[10:11], v[10:11], 1
-; GCN-IR-NEXT:    v_lshrrev_b32_e32 v6, 31, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v6
+; GCN-IR-NEXT:    v_lshrrev_b32_e32 v4, 31, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v10, v10, v4
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_sub_i32_e32 v6, vcc, v4, v10
-; GCN-IR-NEXT:    v_subb_u32_e32 v6, vcc, v5, v11, vcc
+; GCN-IR-NEXT:    v_sub_i32_e32 v4, vcc, v6, v10
+; GCN-IR-NEXT:    v_subb_u32_e32 v4, vcc, v7, v11, vcc
 ; GCN-IR-NEXT:    v_or_b32_e32 v2, v12, v2
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v6
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v12, 31, v4
 ; GCN-IR-NEXT:    v_and_b32_e32 v15, v12, v0
-; GCN-IR-NEXT:    v_and_b32_e32 v6, 1, v12
+; GCN-IR-NEXT:    v_and_b32_e32 v4, 1, v12
 ; GCN-IR-NEXT:    v_and_b32_e32 v14, v12, v1
 ; GCN-IR-NEXT:    v_add_i32_e32 v12, vcc, 1, v8
 ; GCN-IR-NEXT:    v_or_b32_e32 v3, v13, v3
 ; GCN-IR-NEXT:    v_addc_u32_e32 v13, vcc, 0, v9, vcc
 ; GCN-IR-NEXT:    v_cmp_lt_u64_e32 vcc, v[12:13], v[8:9]
 ; GCN-IR-NEXT:    v_mov_b32_e32 v8, v12
-; GCN-IR-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-IR-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-IR-NEXT:    v_sub_i32_e64 v10, s[4:5], v10, v15
 ; GCN-IR-NEXT:    v_mov_b32_e32 v9, v13
-; GCN-IR-NEXT:    v_mov_b32_e32 v13, v7
+; GCN-IR-NEXT:    v_mov_b32_e32 v13, v5
 ; GCN-IR-NEXT:    v_subb_u32_e64 v11, s[4:5], v11, v14, s[4:5]
 ; GCN-IR-NEXT:    s_or_b64 s[8:9], vcc, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v12, v6
+; GCN-IR-NEXT:    v_mov_b32_e32 v12, v4
 ; GCN-IR-NEXT:    s_andn2_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:    s_cbranch_execnz BB8_3
 ; GCN-IR-NEXT:  ; %bb.4: ; %Flow
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[8:9]
 ; GCN-IR-NEXT:  BB8_5: ; %Flow3
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[10:11]
 ; GCN-IR-NEXT:    v_lshl_b64 v[2:3], v[2:3], 1
-; GCN-IR-NEXT:    v_or_b32_e32 v5, v7, v3
-; GCN-IR-NEXT:    v_or_b32_e32 v4, v6, v2
+; GCN-IR-NEXT:    v_or_b32_e32 v5, v5, v3
+; GCN-IR-NEXT:    v_or_b32_e32 v4, v4, v2
 ; GCN-IR-NEXT:  BB8_6: ; %Flow4
 ; GCN-IR-NEXT:    s_or_b64 exec, exec, s[6:7]
 ; GCN-IR-NEXT:    v_mul_lo_u32 v2, v0, v5
 ; GCN-IR-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-IR-NEXT:    v_mul_lo_u32 v1, v1, v4
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, v4
 ; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1