[AMDGPU] Always select s_cselect_b32 for uniform 'select' SDNode

[alex-t]

Summary

This patch contains changes necessary to carry physical condition register (SCC) dependencies through the SDNode scheduler. It adds the edge in the SDNodeScheduler dependency graph instead of inserting the SCC copy between each definition and use. This approach lets the scheduler place instructions in an optimal way placing the copy only when the dependency cannot be resolved.

[rampitec]

llvm/include/llvm/CodeGen/TargetLowering.h

4002

Need a description of the function.

4003

Alignment is off.

llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp

119

Place it after the check for isVirtualRegister below?

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

12983

Alignment.

12998

We may actually later extend it beyond compares and to include VCC.

llvm/lib/Target/AMDGPU/SIISelLowering.h

483

Alignment.

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

935

getWaveMaskRegClass()? The difference that it returns SReg_32_XM0_XEXECRegClass and not SReg_32_XEXEC_HIRegClass.

[foad]

This patch contains changes necessary to carry physical condition register (SCC) dependencies through the SDNode scheduler. This allows avoiding inserting an SCC copy between each definition and use.

Please explain more. Why doesn't the generic support for physical registers work already? Why do we need to "avoid inserting an SCC copy"?

I was also working on this in D124450 but I have not looked at that patch for a few months.

[alex-t]

In D133593#3783316, @foad wrote:

This patch contains changes necessary to carry physical condition register (SCC) dependencies through the SDNode scheduler. This allows avoiding inserting an SCC copy between each definition and use.

Please explain more. Why doesn't the generic support for physical registers work already? Why do we need to "avoid inserting an SCC copy"?

I was also working on this in D124450 but I have not looked at that patch for a few months.

If we allow SCC to be copied (CopyCost() != -1) then any pair of the def-use will look like:

s_cmp_***
%1 = s_cselect_b64 -1, 0  <-- lowered CopyToReg  where srcreg == SCC
s_and_b64 %1, exec          <-- lowered CopyFromReg where dstreg == SCC
%res = s_cselect_b32 %trueval, %falseval  // not necessarily  s_cselect_b32 may be any SCC user

We could handle the trivial cases where the SCC copy is immediately copied back to SCC. Some more complex cases where these copies are interleaved by other instructions are more difficult to analyze.
At the same time, if we keep the SCC copy cost high, we let the SDNode scheduler place the instructions in an optimal way according to the register carried dependencies.
Literally, we will insert a copy only in case there is no other way to meet the data dependency.

llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp

119

It is deliberately here. In case we have a cross-BB SCC-carried dependency it is CopyToReg with the VReg as an operand 1.

CopyToReg chain, VregN, SCC def

and in successor BB:

CopyFromReg SCC, VregN

This is really SCC-carried dependency but it implemented via virtual register copy

[alex-t]

addressed changes requested by the reviewer

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

12998

What needs to be changed here right now for that?

[rampitec]

llvm/include/llvm/CodeGen/TargetLowering.h

4002

Could you please use the same comment style as the rest of the code?

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

12998

Nothing right now.

[alex-t]

Comments style changed.

[rampitec]

LGTM

[vpykhtin]

llvm/lib/CodeGen/MachineRegisterInfo.cpp

86

formatting: 2 spaces

[alex-t]

small formatting

llvm/include/llvm/CodeGen/TargetLowering.h

public:
   /// storing a legal type
   virtual bool splitValueIntoRegisterParts(SelectionDAG &DAG, const SDLoc &DL,
                                            SDValue Val, SDValue *Parts,
                                            unsigned NumParts, MVT PartVT,
                                            Optional<CallingConv::ID> CC) const {
     return false;
   }
 
+  virtual bool checkForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
+                                      const TargetRegisterInfo *TRI,
+                                      const TargetInstrInfo *TII,
+                                      unsigned &PhysReg, int &Cost) const {
+    return false;
+  }
+
   /// Target-specific combining of register parts into its original value
   virtual SDValue
   joinRegisterPartsIntoValue(SelectionDAG &DAG, const SDLoc &DL,
                              const SDValue *Parts, unsigned NumParts,
                              MVT PartVT, EVT ValueVT,
                              Optional<CallingConv::ID> CC) const {
     return SDValue();
   }

llvm/lib/CodeGen/MachineRegisterInfo.cpp

constrainRegClass(MachineRegisterInfo &MRI, Register Reg,
   MRI.setRegClass(Reg, NewRC);
   return NewRC;
 }
 
 const TargetRegisterClass *
 MachineRegisterInfo::constrainRegClass(Register Reg,
                                        const TargetRegisterClass *RC,
                                        unsigned MinNumRegs) {
+ if (Reg.isPhysical())
+    return nullptr;
   return ::constrainRegClass(*this, Reg, getRegClass(Reg), RC, MinNumRegs);
 }
 
 bool
 MachineRegisterInfo::constrainRegAttrs(Register Reg,
                                        Register ConstrainingReg,
                                        unsigned MinNumRegs) {
   const LLT RegTy = getType(Reg);

llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp

 }
 
 /// CheckForPhysRegDependency - Check if the dependency between def and use of
 /// a specified operand is a physical register dependency. If so, returns the
 /// register and the cost of copying the register.
 static void CheckForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
                                       const TargetRegisterInfo *TRI,
                                       const TargetInstrInfo *TII,
+                                      const TargetLowering &TLI,
                                       unsigned &PhysReg, int &Cost) {
   if (Op != 2 || User->getOpcode() != ISD::CopyToReg)
     return;
 
   unsigned Reg = cast<RegisterSDNode>(User->getOperand(1))->getReg();
+  if (TLI.checkForPhysRegDependency(Def, User, Op, TRI, TII, PhysReg, Cost))
+    return;
+
   if (Register::isVirtualRegister(Reg))
     return;
 
   unsigned ResNo = User->getOperand(2).getResNo();
   if (Def->getOpcode() == ISD::CopyFromReg &&
       cast<RegisterSDNode>(Def->getOperand(1))->getReg() == Reg) {
     PhysReg = Reg;
   } else if (Def->isMachineOpcode()) {
for (SDNode *N = SU.getNode(); N; N = N->getGluedNode()) {
 
         EVT OpVT = N->getOperand(i).getValueType();
         assert(OpVT != MVT::Glue && "Glued nodes should be in same sunit!");
         bool isChain = OpVT == MVT::Other;
 
         unsigned PhysReg = 0;
         int Cost = 1;
         // Determine if this is a physical register dependency.
-        CheckForPhysRegDependency(OpN, N, i, TRI, TII, PhysReg, Cost);
+        const TargetLowering &TLI = DAG->getTargetLoweringInfo();
+        CheckForPhysRegDependency(OpN, N, i, TRI, TII, TLI, PhysReg, Cost);
         assert((PhysReg == 0 || !isChain) &&
                "Chain dependence via physreg data?");
         // FIXME: See ScheduleDAGSDNodes::EmitCopyFromReg. For now, scheduler
         // emits a copy from the physical register to a virtual register unless
         // it requires a cross class copy (cost < 0). That means we are only
         // treating "expensive to copy" register dependency as physical register
         // dependency. This may change in the future though.
         if (Cost >= 0 && !StressSched)

llvm/lib/Target/AMDGPU/SIISelLowering.h

public:
 
   bool isCanonicalized(SelectionDAG &DAG, SDValue Op,
                        unsigned MaxDepth = 5) const;
   bool isCanonicalized(Register Reg, MachineFunction &MF,
                        unsigned MaxDepth = 5) const;
   bool denormalsEnabledForType(const SelectionDAG &DAG, EVT VT) const;
   bool denormalsEnabledForType(LLT Ty, MachineFunction &MF) const;
 
+  bool checkForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
+                                      const TargetRegisterInfo *TRI,
+                                      const TargetInstrInfo *TII,
+                                      unsigned &PhysReg, int &Cost) const override;
+
   bool isKnownNeverNaNForTargetNode(SDValue Op,
                                     const SelectionDAG &DAG,
                                     bool SNaN = false,
                                     unsigned Depth = 0) const override;
   AtomicExpansionKind shouldExpandAtomicRMWInIR(AtomicRMWInst *) const override;
   AtomicExpansionKind shouldExpandAtomicLoadInIR(LoadInst *LI) const override;
   AtomicExpansionKind shouldExpandAtomicStoreInIR(StoreInst *SI) const override;
   AtomicExpansionKind

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

 
 MachineMemOperand::Flags
 SITargetLowering::getTargetMMOFlags(const Instruction &I) const {
   // Propagate metadata set by AMDGPUAnnotateUniformValues to the MMO of a load.
   if (I.getMetadata("amdgpu.noclobber"))
     return MONoClobber;
   return MachineMemOperand::MONone;
 }
+
+bool SITargetLowering::checkForPhysRegDependency(SDNode *Def, SDNode *User, unsigned Op,
+                                      const TargetRegisterInfo *TRI,
+                                      const TargetInstrInfo *TII,
+                                      unsigned &PhysReg, int &Cost) const{
+  if (User->getOpcode() != ISD::CopyToReg)
+    return false;
+  if (!Def->isMachineOpcode())
+    return false;
+  MachineSDNode *MDef = dyn_cast<MachineSDNode>(Def);
+  if (!MDef)
+    return false;
+
+  unsigned ResNo = User->getOperand(Op).getResNo();
+  if(User->getOperand(Op)->getValueType(ResNo) != MVT::i1)
+    return false;
+  const MCInstrDesc &II = TII->get(MDef->getMachineOpcode());
+  if (II.isCompare() && II.hasImplicitDefOfPhysReg(AMDGPU::SCC)) {
+    PhysReg = AMDGPU::SCC;
+    const TargetRegisterClass *RC =
+        TRI->getMinimalPhysRegClass(PhysReg, Def->getSimpleValueType(ResNo));
+    Cost = RC->getCopyCost();
+    return true;
+  }
+  return false;
+}

llvm/lib/Target/AMDGPU/SIRegisterInfo.cpp

const TargetRegisterClass *SIRegisterInfo::getPointerRegClass(
   // private accesses, so this is correct in that case.
   return &AMDGPU::VGPR_32RegClass;
 }
 
 const TargetRegisterClass *
 SIRegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
   if (isAGPRClass(RC) && !ST.hasGFX90AInsts())
     return getEquivalentVGPRClass(RC);
+  if (RC == &AMDGPU::SCC_CLASSRegClass)
+    return ST.isWave32() ? &AMDGPU::SReg_32_XEXEC_HIRegClass
+                         : &AMDGPU::SReg_64_XEXECRegClass;
 
   return RC;
 }
 
 static unsigned getNumSubRegsForSpillOp(unsigned Op) {
 
   switch (Op) {
   case AMDGPU::SI_SPILL_S1024_SAVE:

llvm/lib/Target/AMDGPU/SOPInstructions.td

 
 // This pattern is restricted to certain subtargets (practically GFX8Plus)
 // because isel sometimes produces an sreg_64 copy to SCC as a by-product
 // of this pattern, and only for subtargets with hasScalarCompareEq64
 // is it possible to map such copy to a single instruction (S_CMP_LG_U64).
 class SelectPat<SDPatternOperator select> : PatFrag <
   (ops node:$src1, node:$src2),
   (select SCC, $src1, $src2),
-  [{ return Subtarget->hasScalarCompareEq64() &&
-            N->getOperand(0)->hasOneUse() && !N->isDivergent(); }]
+  [{ return !N->isDivergent(); }]
 >;
 
 let Uses = [SCC] in {
   let AddedComplexity = 20 in {
     def S_CSELECT_B32 : SOP2_32 <"s_cselect_b32",
       [(set i32:$sdst, (SelectPat<select> i32:$src0, i32:$src1))]
     >;
   }

llvm/test/CodeGen/AMDGPU/32-bit-local-address-space.ll

entry:
   %0 = getelementptr i32, i32 addrspace(3)* %in, i32 16385
   %1 = load i32, i32 addrspace(3)* %0
   store i32 %1, i32 addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}null_32bit_lds_ptr:
 ; GFX7 v_cmp_ne_u32
-; GFX7: v_cndmask_b32
+; GFX7: s_cselect_b32
 ; GFX8: s_cmp_lg_u32
 ; GFX8-NOT: v_cmp_ne_u32
 ; GFX8: s_cselect_b32
 define amdgpu_kernel void @null_32bit_lds_ptr(i32 addrspace(1)* %out, i32 addrspace(3)* %lds) nounwind {
   %cmp = icmp ne i32 addrspace(3)* %lds, null
   %x = select i1 %cmp, i32 123, i32 456
   store i32 %x, i32 addrspace(1)* %out
   ret void

llvm/test/CodeGen/AMDGPU/addrspacecast.ll

 ; RUN: llc -march=amdgcn -mtriple=amdgcn-amd-amdhsa -mcpu=kaveri --amdhsa-code-object-version=2 -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=HSA -check-prefix=CI %s
 ; RUN: llc -march=amdgcn -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 --amdhsa-code-object-version=2 -mattr=-promote-alloca -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=HSA -check-prefix=GFX9 %s
 
 ; HSA-LABEL: {{^}}use_group_to_flat_addrspacecast:
 ; HSA: enable_sgpr_private_segment_buffer = 1
 ; HSA: enable_sgpr_dispatch_ptr = 0
 ; CI: enable_sgpr_queue_ptr = 1
 ; GFX9: enable_sgpr_queue_ptr = 0
 
 ; CI-DAG: s_load_dword [[PTR:s[0-9]+]], s[6:7], 0x0{{$}}
 ; CI-DAG: s_load_dword [[APERTURE:s[0-9]+]], s[4:5], 0x10{{$}}
-; CI-DAG: v_mov_b32_e32 [[VAPERTURE:v[0-9]+]], [[APERTURE]]
 ; CI-DAG: s_cmp_lg_u32 [[PTR]], -1
-; CI-DAG: s_cselect_b64 vcc, -1, 0
-; CI-DAG: v_cndmask_b32_e32 v[[HI:[0-9]+]], 0, [[VAPERTURE]], vcc
-; CI-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
-; CI-DAG: v_cndmask_b32_e32 v[[LO:[0-9]+]], 0, [[VPTR]]
+; CI-DAG: s_cselect_b32 s[[HI:[0-9]+]], [[APERTURE]], 0
+; CI-DAG: s_cselect_b32 s[[LO:[0-9]+]], [[PTR]], 0
 
 ; HSA-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 7
 ; GFX9-DAG: s_load_dword [[PTR:s[0-9]+]], s[4:5], 0x0{{$}}
 ; GFX9-DAG: s_getreg_b32 [[SSRC_SHARED:s[0-9]+]], hwreg(HW_REG_SH_MEM_BASES, 16, 16)
 ; GFX9-DAG: s_lshl_b32 [[SSRC_SHARED_BASE:s[0-9]+]], [[SSRC_SHARED]], 16
-; GFX9-DAG: v_mov_b32_e32 [[VAPERTURE:v[0-9]+]], [[SSRC_SHARED_BASE]]
 
 ; GFX9-XXX: v_mov_b32_e32 [[VAPERTURE:v[0-9]+]], src_shared_base
 ; GFX9: s_cmp_lg_u32 [[PTR]], -1
-; GFX9: s_cselect_b64 vcc, -1, 0
-; GFX9: v_cndmask_b32_e32 v[[HI:[0-9]+]], 0, [[VAPERTURE]], vcc
-; GFX9-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
-; GFX9-DAG: v_cndmask_b32_e32 v[[LO:[0-9]+]], 0, [[VPTR]]
+; GFX9-DAG: s_cselect_b32 s[[HI:[0-9]+]], [[SSRC_SHARED_BASE]], 0
+; GFX9-DAG: s_cselect_b32 s[[LO:[0-9]+]], [[PTR]], 0
 
 ; HSA: flat_store_dword v[[[LO]]:[[HI]]], [[K]]
 
 ; At most 2 digits. Make sure src_shared_base is not counted as a high
 ; number SGPR.
 
-; CI: NumSgprs: {{[0-9][0-9]+}}
-; GFX9: NumSgprs: {{[0-9]+}}
+; HSA: NumSgprs: {{[0-9]+}}
 define amdgpu_kernel void @use_group_to_flat_addrspacecast(i32 addrspace(3)* %ptr) #0 {
   %stof = addrspacecast i32 addrspace(3)* %ptr to i32*
   store volatile i32 7, i32* %stof
   ret void
 }
 
 ; Test handling inside a non-kernel
 ; HSA-LABEL: {{^}}use_group_to_flat_addrspacecast_func:

 ; HSA-LABEL: {{^}}use_private_to_flat_addrspacecast:
 ; HSA: enable_sgpr_private_segment_buffer = 1
 ; HSA: enable_sgpr_dispatch_ptr = 0
 ; CI: enable_sgpr_queue_ptr = 1
 ; GFX9: enable_sgpr_queue_ptr = 0
 
 ; CI-DAG: s_load_dword [[PTR:s[0-9]+]], s[6:7], 0x0{{$}}
 ; CI-DAG: s_load_dword [[APERTURE:s[0-9]+]], s[4:5], 0x11{{$}}
-; CI-DAG: v_mov_b32_e32 [[VAPERTURE:v[0-9]+]], [[APERTURE]]
 
 ; CI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 7
 ; CI-DAG: s_cmp_lg_u32 [[PTR]], -1
-; CI-DAG: s_cselect_b64 vcc, -1, 0
-; CI-DAG: v_cndmask_b32_e32 v[[HI:[0-9]+]], 0, [[VAPERTURE]], vcc
-; CI-DAG: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
-; CI-DAG: v_cndmask_b32_e32 v[[LO:[0-9]+]], 0, [[VPTR]]
+; CI-DAG: s_cselect_b32 s[[HI:[0-9]+]], [[APERTURE]], 0
+; CI-DAG: s_cselect_b32 s[[LO:[0-9]+]], [[PTR]], 0
 
 ; GFX9-DAG: s_load_dword [[PTR:s[0-9]+]], s[4:5], 0x0{{$}}
 ; GFX9-DAG: s_getreg_b32 [[SSRC_PRIVATE:s[0-9]+]], hwreg(HW_REG_SH_MEM_BASES, 0, 16)
 ; GFX9-DAG: s_lshl_b32 [[SSRC_PRIVATE_BASE:s[0-9]+]], [[SSRC_PRIVATE]], 16
-; GFX9-DAG: v_mov_b32_e32 [[VAPERTURE:v[0-9]+]], [[SSRC_PRIVATE_BASE]]
 
 ; GFX9-XXX: v_mov_b32_e32 [[VAPERTURE:v[0-9]+]], src_private_base
 
 ; GFX9-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 7
 ; GFX9: s_cmp_lg_u32 [[PTR]], -1
-; GFX9: s_cselect_b64 vcc, -1, 0
-; GFX9: v_cndmask_b32_e32 v[[HI:[0-9]+]], 0, [[VAPERTURE]], vcc
-; GFX9: v_mov_b32_e32 [[VPTR:v[0-9]+]], [[PTR]]
-; GFX9-DAG: v_cndmask_b32_e32 v[[LO:[0-9]+]], 0, [[VPTR]]
+; GFX9: s_cselect_b32 s[[HI:[0-9]+]], [[SSRC_PRIVATE_BASE]], 0
+; GFX9: s_cselect_b32 s[[LO:[0-9]+]], [[PTR]], 0
 
 ; HSA: flat_store_dword v[[[LO]]:[[HI]]], [[K]]
 
-; CI: NumSgprs: {{[0-9][0-9]+}}
-; GFX9: NumSgprs: {{[0-9]+}}
+; HSA: NumSgprs: {{[0-9]+}}
 define amdgpu_kernel void @use_private_to_flat_addrspacecast(i32 addrspace(5)* %ptr) #0 {
   %stof = addrspacecast i32 addrspace(5)* %ptr to i32*
   store volatile i32 7, i32* %stof
   ret void
 }
 
 ; no-op
 ; HSA-LABEL: {{^}}use_global_to_flat_addrspacecast:

 }
 
 ; HSA-LABEL: {{^}}use_flat_to_group_addrspacecast:
 ; HSA: enable_sgpr_private_segment_buffer = 1
 ; HSA: enable_sgpr_dispatch_ptr = 0
 ; HSA: enable_sgpr_queue_ptr = 0
 
 ; HSA: s_load_dwordx2 s[[[PTR_LO:[0-9]+]]:[[PTR_HI:[0-9]+]]]
-; CI-DAG: v_cmp_ne_u64_e64 vcc, s[[[PTR_LO]]:[[PTR_HI]]], 0{{$}}
-; CI-DAG: v_mov_b32_e32 v[[VPTR_LO:[0-9]+]], s[[PTR_LO]]
-; CI-DAG: v_cndmask_b32_e32 [[CASTPTR:v[0-9]+]], -1, v[[VPTR_LO]]
+; CI-DAG: v_cmp_ne_u64_e64 s[[[CMP_LO:[0-9]+]]:[[CMP_HI:[0-9]+]]], s[[[PTR_LO]]:[[PTR_HI]]], 0{{$}}
+; CI-DAG: s_and_b64 s{{[[0-9]+:[0-9]+]}}, s[[[CMP_LO]]:[[CMP_HI]]], exec
+; CI-DAG: s_cselect_b32 [[CASTPTR:s[0-9]+]], s[[PTR_LO]], -1
+; CI-DAG: v_mov_b32_e32 [[VCASTPTR:v[0-9]+]], [[CASTPTR]]
 ; HSA-DAG: v_mov_b32_e32 v[[K:[0-9]+]], 0{{$}}
 ; GFX9-DAG: s_cmp_lg_u64 s[[[CMP_LO:[0-9]+]]:[[CMP_HI:[0-9]+]]], 0
 ; GFX9-DAG: s_cselect_b32 s[[PTR_LO]], s[[PTR_LO]], -1
 ; GFX9-DAG: v_mov_b32_e32 [[CASTPTR:v[0-9]+]], s[[PTR_LO]]
-; HSA: ds_write_b32 [[CASTPTR]], v[[K]]
+; CI-DAG: ds_write_b32 [[VCASTPTR]], v[[K]]
+; GFX9-DAG: ds_write_b32 [[CASTPTR]], v[[K]]
 define amdgpu_kernel void @use_flat_to_group_addrspacecast(i32* %ptr) #0 {
   %ftos = addrspacecast i32* %ptr to i32 addrspace(3)*
   store volatile i32 0, i32 addrspace(3)* %ftos
   ret void
 }
 
 ; HSA-LABEL: {{^}}use_flat_to_private_addrspacecast:
 ; HSA: enable_sgpr_private_segment_buffer = 1
 ; HSA: enable_sgpr_dispatch_ptr = 0
 ; HSA: enable_sgpr_queue_ptr = 0
 
 ; HSA: s_load_dwordx2 s[[[PTR_LO:[0-9]+]]:[[PTR_HI:[0-9]+]]]
-; CI-DAG: v_cmp_ne_u64_e64 vcc, s[[[PTR_LO]]:[[PTR_HI]]], 0{{$}}
-; CI-DAG: v_mov_b32_e32 v[[VPTR_LO:[0-9]+]], s[[PTR_LO]]
-; CI-DAG: v_cndmask_b32_e32 [[CASTPTR:v[0-9]+]], -1, v[[VPTR_LO]]
+; CI-DAG v_cmp_ne_u64_e64 vcc, s[[[PTR_LO]]:[[PTR_HI]]], 0{{$}}
+; CI-DAG v_mov_b32_e32 v[[VPTR_LO:[0-9]+]], s[[PTR_LO]]
+; CI-DAG v_cndmask_b32_e32 [[CASTPTR:v[0-9]+]], -1, v[[VPTR_LO]]
+; CI-DAG: v_cmp_ne_u64_e64 s[[[CMP_LO:[0-9]+]]:[[CMP_HI:[0-9]+]]], s[[[PTR_LO]]:[[PTR_HI]]], 0{{$}}
+; CI-DAG: s_and_b64 s{{[[0-9]+:[0-9]+]}}, s[[[CMP_LO]]:[[CMP_HI]]], exec
+; CI-DAG: s_cselect_b32 [[CASTPTR:s[0-9]+]], s[[PTR_LO]], -1
+; CI-DAG: v_mov_b32_e32 [[VCASTPTR:v[0-9]+]], [[CASTPTR]]
 ; HSA-DAG: v_mov_b32_e32 v[[K:[0-9]+]], 0{{$}}
 ; GFX9-DAG: s_cmp_lg_u64 s[[[CMP_LO:[0-9]+]]:[[CMP_HI:[0-9]+]]], 0
 ; GFX9-DAG: s_cselect_b32 s[[PTR_LO]], s[[PTR_LO]], -1
 ; GFX9-DAG: v_mov_b32_e32 [[CASTPTR:v[0-9]+]], s[[PTR_LO]]
-; HSA: buffer_store_dword v[[K]], [[CASTPTR]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen{{$}}
+; CI: buffer_store_dword v[[K]], [[VCASTPTR]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen{{$}}
+; GFX9: buffer_store_dword v[[K]], [[CASTPTR]], s{{\[[0-9]+:[0-9]+\]}}, 0 offen{{$}}
 define amdgpu_kernel void @use_flat_to_private_addrspacecast(i32* %ptr) #0 {
   %ftos = addrspacecast i32* %ptr to i32 addrspace(5)*
   store volatile i32 0, i32 addrspace(5)* %ftos
   ret void
 }
 
 ; HSA-LABEL: {{^}}use_flat_to_global_addrspacecast:
 ; HSA: enable_sgpr_queue_ptr = 0

llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll

; GFX90A-NEXT:    s_setpc_b64 s[30:31]
   call void asm sideeffect "; use $0 $1","{a3},{v[0:31]}"(i32 %agpr2, <32 x i32> %vgpr0)
   call void asm sideeffect "; use $0","${v32}"(i32 %v32)
   ret void
 }
 
 define amdgpu_kernel void @introduced_copy_to_sgpr(i64 %arg, i32 %arg1, i32 %arg2, i64 %arg3, <2 x half> %arg4, <2 x half> %arg5) #3 {
 ; GFX908-LABEL: introduced_copy_to_sgpr:
 ; GFX908:       ; %bb.0: ; %bb
-; GFX908-NEXT:    global_load_ushort v24, v[0:1], off glc
+; GFX908-NEXT:    global_load_ushort v16, v[0:1], off glc
 ; GFX908-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
-; GFX908-NEXT:    s_load_dwordx2 s[10:11], s[4:5], 0x10
-; GFX908-NEXT:    v_mov_b32_e32 v1, 0
-; GFX908-NEXT:    s_load_dword s5, s[4:5], 0x18
-; GFX908-NEXT:    s_mov_b32 s4, 0
+; GFX908-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10
+; GFX908-NEXT:    s_load_dword s9, s[4:5], 0x18
+; GFX908-NEXT:    s_mov_b32 s8, 0
+; GFX908-NEXT:    s_mov_b32 s5, s8
 ; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX908-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX908-NEXT:    s_sub_i32 s6, 0, s3
-; GFX908-NEXT:    s_lshl_b64 s[8:9], s[10:11], 5
-; GFX908-NEXT:    s_lshr_b32 s12, s5, 16
-; GFX908-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v25, s5
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v26, s12
-; GFX908-NEXT:    s_or_b32 s8, s8, 28
-; GFX908-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX908-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX908-NEXT:    v_mov_b32_e32 v6, s10
-; GFX908-NEXT:    v_mov_b32_e32 v7, s11
-; GFX908-NEXT:    v_mul_lo_u32 v2, s6, v0
-; GFX908-NEXT:    s_lshl_b64 s[6:7], s[0:1], 5
-; GFX908-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX908-NEXT:    v_add_u32_e32 v0, v0, v2
-; GFX908-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX908-NEXT:    v_mov_b32_e32 v2, s8
-; GFX908-NEXT:    v_mov_b32_e32 v3, s9
-; GFX908-NEXT:    v_mul_lo_u32 v4, v0, s3
-; GFX908-NEXT:    v_add_u32_e32 v5, 1, v0
-; GFX908-NEXT:    v_sub_u32_e32 v4, s2, v4
-; GFX908-NEXT:    v_cmp_le_u32_e32 vcc, s3, v4
-; GFX908-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
-; GFX908-NEXT:    v_subrev_u32_e32 v5, s3, v4
-; GFX908-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
-; GFX908-NEXT:    v_add_u32_e32 v5, 1, v0
-; GFX908-NEXT:    v_cmp_le_u32_e32 vcc, s3, v4
-; GFX908-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
-; GFX908-NEXT:    v_lshlrev_b64 v[4:5], 5, v[0:1]
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_readfirstlane_b32 s2, v24
-; GFX908-NEXT:    s_and_b32 s2, 0xffff, s2
-; GFX908-NEXT:    s_mul_i32 s1, s1, s2
-; GFX908-NEXT:    s_mul_hi_u32 s3, s0, s2
-; GFX908-NEXT:    s_mul_i32 s0, s0, s2
-; GFX908-NEXT:    s_add_i32 s1, s3, s1
-; GFX908-NEXT:    s_lshl_b64 s[8:9], s[0:1], 5
+; GFX908-NEXT:    s_sub_i32 s4, 0, s3
+; GFX908-NEXT:    v_cvt_f32_f16_e32 v17, s9
+; GFX908-NEXT:    v_mov_b32_e32 v19, 0
+; GFX908-NEXT:    v_rcp_iflag_f32_e32 v2, v0
+; GFX908-NEXT:    v_mov_b32_e32 v0, 0
+; GFX908-NEXT:    v_mov_b32_e32 v1, 0
+; GFX908-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
+; GFX908-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX908-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX908-NEXT:    s_mul_i32 s4, s4, s10
+; GFX908-NEXT:    s_mul_hi_u32 s4, s10, s4
+; GFX908-NEXT:    s_add_i32 s10, s10, s4
+; GFX908-NEXT:    s_mul_hi_u32 s4, s2, s10
+; GFX908-NEXT:    s_mul_i32 s10, s4, s3
+; GFX908-NEXT:    s_sub_i32 s2, s2, s10
+; GFX908-NEXT:    s_add_i32 s11, s4, 1
+; GFX908-NEXT:    s_sub_i32 s10, s2, s3
+; GFX908-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX908-NEXT:    s_cselect_b32 s4, s11, s4
+; GFX908-NEXT:    s_cselect_b32 s2, s10, s2
+; GFX908-NEXT:    s_add_i32 s10, s4, 1
+; GFX908-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX908-NEXT:    s_cselect_b32 s4, s10, s4
+; GFX908-NEXT:    s_lshr_b32 s9, s9, 16
+; GFX908-NEXT:    s_lshl_b64 s[12:13], s[4:5], 5
+; GFX908-NEXT:    v_cvt_f32_f16_e32 v18, s9
+; GFX908-NEXT:    s_lshl_b64 s[2:3], s[0:1], 5
+; GFX908-NEXT:    s_lshl_b64 s[10:11], s[6:7], 5
+; GFX908-NEXT:    s_or_b32 s10, s10, 28
+; GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GFX908-NEXT:    v_readfirstlane_b32 s5, v16
+; GFX908-NEXT:    s_and_b32 s5, 0xffff, s5
+; GFX908-NEXT:    s_mul_i32 s1, s1, s5
+; GFX908-NEXT:    s_mul_hi_u32 s9, s0, s5
+; GFX908-NEXT:    s_mul_i32 s0, s0, s5
+; GFX908-NEXT:    s_add_i32 s1, s9, s1
+; GFX908-NEXT:    s_lshl_b64 s[0:1], s[0:1], 5
 ; GFX908-NEXT:    s_branch .LBB3_2
 ; GFX908-NEXT:  .LBB3_1: ; %bb12
 ; GFX908-NEXT:    ; in Loop: Header=BB3_2 Depth=1
-; GFX908-NEXT:    v_add_co_u32_e32 v6, vcc, v6, v0
-; GFX908-NEXT:    v_addc_co_u32_e32 v7, vcc, 0, v7, vcc
-; GFX908-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX908-NEXT:    v_addc_co_u32_e32 v3, vcc, v3, v5, vcc
+; GFX908-NEXT:    s_add_u32 s6, s6, s4
+; GFX908-NEXT:    s_addc_u32 s7, s7, 0
+; GFX908-NEXT:    s_add_u32 s10, s10, s12
+; GFX908-NEXT:    s_addc_u32 s11, s11, s13
 ; GFX908-NEXT:  .LBB3_2: ; %bb9
 ; GFX908-NEXT:    ; =>This Loop Header: Depth=1
 ; GFX908-NEXT:    ; Child Loop BB3_5 Depth 2
 ; GFX908-NEXT:    s_cbranch_scc0 .LBB3_1
 ; GFX908-NEXT:  ; %bb.3: ; %bb14
 ; GFX908-NEXT:    ; in Loop: Header=BB3_2 Depth=1
-; GFX908-NEXT:    v_mov_b32_e32 v8, 0
-; GFX908-NEXT:    v_mov_b32_e32 v9, 0
-; GFX908-NEXT:    global_load_dwordx2 v[8:9], v[8:9], off
-; GFX908-NEXT:    s_mov_b32 s5, s4
-; GFX908-NEXT:    v_mov_b32_e32 v13, s5
-; GFX908-NEXT:    v_mov_b32_e32 v15, s5
-; GFX908-NEXT:    v_mov_b32_e32 v17, s5
-; GFX908-NEXT:    v_mov_b32_e32 v12, s4
-; GFX908-NEXT:    v_mov_b32_e32 v14, s4
-; GFX908-NEXT:    v_mov_b32_e32 v16, s4
-; GFX908-NEXT:    v_cmp_gt_i64_e64 s[0:1], 0, v[6:7]
-; GFX908-NEXT:    v_mov_b32_e32 v11, v3
-; GFX908-NEXT:    v_mov_b32_e32 v19, v13
-; GFX908-NEXT:    v_mov_b32_e32 v10, v2
-; GFX908-NEXT:    v_mov_b32_e32 v18, v12
-; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    v_readfirstlane_b32 s2, v8
-; GFX908-NEXT:    v_readfirstlane_b32 s3, v9
-; GFX908-NEXT:    s_add_u32 s2, s2, 1
-; GFX908-NEXT:    s_addc_u32 s3, s3, 0
-; GFX908-NEXT:    s_mul_hi_u32 s5, s6, s2
-; GFX908-NEXT:    s_mul_i32 s11, s7, s2
-; GFX908-NEXT:    s_mul_i32 s10, s6, s2
-; GFX908-NEXT:    s_mul_i32 s2, s6, s3
-; GFX908-NEXT:    s_add_i32 s2, s5, s2
-; GFX908-NEXT:    s_add_i32 s5, s2, s11
+; GFX908-NEXT:    global_load_dwordx2 v[2:3], v[0:1], off
+; GFX908-NEXT:    s_mov_b32 s9, s8
+; GFX908-NEXT:    v_mov_b32_e32 v4, s8
+; GFX908-NEXT:    v_mov_b32_e32 v6, s8
+; GFX908-NEXT:    v_mov_b32_e32 v8, s8
+; GFX908-NEXT:    v_mov_b32_e32 v5, s9
+; GFX908-NEXT:    v_mov_b32_e32 v7, s9
+; GFX908-NEXT:    v_mov_b32_e32 v9, s9
+; GFX908-NEXT:    v_cmp_lt_i64_e64 s[14:15], s[6:7], 0
+; GFX908-NEXT:    v_mov_b32_e32 v11, v5
+; GFX908-NEXT:    s_mov_b64 s[16:17], s[10:11]
+; GFX908-NEXT:    v_mov_b32_e32 v10, v4
+; GFX908-NEXT:    s_waitcnt vmcnt(0)
+; GFX908-NEXT:    v_readfirstlane_b32 s5, v2
+; GFX908-NEXT:    v_readfirstlane_b32 s9, v3
+; GFX908-NEXT:    s_add_u32 s5, s5, 1
+; GFX908-NEXT:    s_addc_u32 s9, s9, 0
+; GFX908-NEXT:    s_mul_hi_u32 s19, s2, s5
+; GFX908-NEXT:    s_mul_i32 s20, s3, s5
+; GFX908-NEXT:    s_mul_i32 s18, s2, s5
+; GFX908-NEXT:    s_mul_i32 s5, s2, s9
+; GFX908-NEXT:    s_add_i32 s5, s19, s5
+; GFX908-NEXT:    s_add_i32 s5, s5, s20
 ; GFX908-NEXT:    s_branch .LBB3_5
 ; GFX908-NEXT:  .LBB3_4: ; %bb58
 ; GFX908-NEXT:    ; in Loop: Header=BB3_5 Depth=2
-; GFX908-NEXT:    v_add_co_u32_sdwa v8, vcc, v8, v24 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
-; GFX908-NEXT:    v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
-; GFX908-NEXT:    v_cmp_gt_i64_e32 vcc, 0, v[8:9]
-; GFX908-NEXT:    v_mov_b32_e32 v20, s9
-; GFX908-NEXT:    v_add_co_u32_e64 v10, s[2:3], s8, v10
-; GFX908-NEXT:    v_addc_co_u32_e64 v11, s[2:3], v11, v20, s[2:3]
+; GFX908-NEXT:    v_add_co_u32_sdwa v2, vcc, v2, v16 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX908-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
+; GFX908-NEXT:    v_cmp_gt_i64_e32 vcc, 0, v[2:3]
+; GFX908-NEXT:    s_add_u32 s16, s16, s0
+; GFX908-NEXT:    s_addc_u32 s17, s17, s1
 ; GFX908-NEXT:    s_cbranch_vccz .LBB3_1
 ; GFX908-NEXT:  .LBB3_5: ; %bb16
 ; GFX908-NEXT:    ; Parent Loop BB3_2 Depth=1
 ; GFX908-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX908-NEXT:    v_mov_b32_e32 v21, s5
-; GFX908-NEXT:    v_add_co_u32_e32 v20, vcc, s10, v10
-; GFX908-NEXT:    v_addc_co_u32_e32 v21, vcc, v11, v21, vcc
-; GFX908-NEXT:    global_load_dword v28, v[20:21], off offset:-12 glc
+; GFX908-NEXT:    s_add_u32 s20, s16, s18
+; GFX908-NEXT:    s_addc_u32 s21, s17, s5
+; GFX908-NEXT:    global_load_dword v21, v19, s[20:21] offset:-12 glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    global_load_dword v27, v[20:21], off offset:-8 glc
+; GFX908-NEXT:    global_load_dword v20, v19, s[20:21] offset:-8 glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    global_load_dword v22, v[20:21], off offset:-4 glc
+; GFX908-NEXT:    global_load_dword v12, v19, s[20:21] offset:-4 glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    global_load_dword v20, v[20:21], off glc
+; GFX908-NEXT:    global_load_dword v12, v19, s[20:21] glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    ds_read_b64 v[20:21], v1
-; GFX908-NEXT:    ds_read_b64 v[22:23], v0
-; GFX908-NEXT:    s_and_b64 vcc, exec, s[0:1]
+; GFX908-NEXT:    ds_read_b64 v[12:13], v19
+; GFX908-NEXT:    ds_read_b64 v[14:15], v0
+; GFX908-NEXT:    s_and_b64 vcc, exec, s[14:15]
 ; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX908-NEXT:    s_cbranch_vccnz .LBB3_4
 ; GFX908-NEXT:  ; %bb.6: ; %bb51
 ; GFX908-NEXT:    ; in Loop: Header=BB3_5 Depth=2
-; GFX908-NEXT:    v_cvt_f32_f16_sdwa v29, v28 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v28, v28
-; GFX908-NEXT:    v_cvt_f32_f16_sdwa v30, v27 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v27, v27
-; GFX908-NEXT:    v_add_f32_e32 v31, v25, v20
-; GFX908-NEXT:    v_add_f32_e32 v32, v26, v21
-; GFX908-NEXT:    v_add_f32_e32 v33, 0, v20
-; GFX908-NEXT:    v_add_f32_e32 v34, 0, v21
-; GFX908-NEXT:    v_add_f32_e32 v23, v29, v23
-; GFX908-NEXT:    v_add_f32_e32 v22, v28, v22
-; GFX908-NEXT:    v_add_f32_e32 v21, v30, v21
-; GFX908-NEXT:    v_add_f32_e32 v20, v27, v20
-; GFX908-NEXT:    v_add_f32_e32 v13, v13, v32
-; GFX908-NEXT:    v_add_f32_e32 v12, v12, v31
-; GFX908-NEXT:    v_add_f32_e32 v15, v15, v34
-; GFX908-NEXT:    v_add_f32_e32 v14, v14, v33
-; GFX908-NEXT:    v_add_f32_e32 v16, v16, v22
-; GFX908-NEXT:    v_add_f32_e32 v17, v17, v23
-; GFX908-NEXT:    v_add_f32_e32 v18, v18, v20
-; GFX908-NEXT:    v_add_f32_e32 v19, v19, v21
+; GFX908-NEXT:    v_cvt_f32_f16_sdwa v22, v21 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX908-NEXT:    v_cvt_f32_f16_e32 v21, v21
+; GFX908-NEXT:    v_cvt_f32_f16_sdwa v23, v20 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX908-NEXT:    v_cvt_f32_f16_e32 v20, v20
+; GFX908-NEXT:    v_add_f32_e32 v24, v17, v12
+; GFX908-NEXT:    v_add_f32_e32 v25, v18, v13
+; GFX908-NEXT:    v_add_f32_e32 v26, 0, v12
+; GFX908-NEXT:    v_add_f32_e32 v27, 0, v13
+; GFX908-NEXT:    v_add_f32_e32 v15, v22, v15
+; GFX908-NEXT:    v_add_f32_e32 v14, v21, v14
+; GFX908-NEXT:    v_add_f32_e32 v13, v23, v13
+; GFX908-NEXT:    v_add_f32_e32 v12, v20, v12
+; GFX908-NEXT:    v_add_f32_e32 v5, v5, v25
+; GFX908-NEXT:    v_add_f32_e32 v4, v4, v24
+; GFX908-NEXT:    v_add_f32_e32 v7, v7, v27
+; GFX908-NEXT:    v_add_f32_e32 v6, v6, v26
+; GFX908-NEXT:    v_add_f32_e32 v8, v8, v14
+; GFX908-NEXT:    v_add_f32_e32 v9, v9, v15
+; GFX908-NEXT:    v_add_f32_e32 v10, v10, v12
+; GFX908-NEXT:    v_add_f32_e32 v11, v11, v13
 ; GFX908-NEXT:    s_branch .LBB3_4
 ;
 ; GFX90A-LABEL: introduced_copy_to_sgpr:
 ; GFX90A:       ; %bb.0: ; %bb
-; GFX90A-NEXT:    global_load_ushort v28, v[0:1], off glc
+; GFX90A-NEXT:    global_load_ushort v18, v[0:1], off glc
 ; GFX90A-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
-; GFX90A-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x10
-; GFX90A-NEXT:    s_load_dword s7, s[4:5], 0x18
-; GFX90A-NEXT:    v_mov_b32_e32 v1, 0
-; GFX90A-NEXT:    s_mov_b32 s6, 0
+; GFX90A-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x10
+; GFX90A-NEXT:    s_load_dword s9, s[4:5], 0x18
+; GFX90A-NEXT:    s_mov_b32 s8, 0
+; GFX90A-NEXT:    s_mov_b32 s5, s8
 ; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX90A-NEXT:    s_sub_i32 s12, 0, s3
-; GFX90A-NEXT:    s_lshr_b32 s13, s7, 16
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v2, s7
+; GFX90A-NEXT:    s_sub_i32 s4, 0, s3
+; GFX90A-NEXT:    v_mov_b32_e32 v19, 0
+; GFX90A-NEXT:    v_pk_mov_b32 v[2:3], 0, 0
 ; GFX90A-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v3, s13
-; GFX90A-NEXT:    s_lshl_b64 s[4:5], s[0:1], 5
-; GFX90A-NEXT:    s_lshl_b64 s[10:11], s[8:9], 5
 ; GFX90A-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX90A-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX90A-NEXT:    v_cvt_u32_f32_e32 v1, v0
+; GFX90A-NEXT:    v_cvt_f32_f16_e32 v0, s9
+; GFX90A-NEXT:    v_readfirstlane_b32 s10, v1
+; GFX90A-NEXT:    s_mul_i32 s4, s4, s10
+; GFX90A-NEXT:    s_mul_hi_u32 s4, s10, s4
+; GFX90A-NEXT:    s_add_i32 s10, s10, s4
+; GFX90A-NEXT:    s_mul_hi_u32 s4, s2, s10
+; GFX90A-NEXT:    s_mul_i32 s10, s4, s3
+; GFX90A-NEXT:    s_sub_i32 s2, s2, s10
+; GFX90A-NEXT:    s_add_i32 s11, s4, 1
+; GFX90A-NEXT:    s_sub_i32 s10, s2, s3
+; GFX90A-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX90A-NEXT:    s_cselect_b32 s4, s11, s4
+; GFX90A-NEXT:    s_cselect_b32 s2, s10, s2
+; GFX90A-NEXT:    s_add_i32 s10, s4, 1
+; GFX90A-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX90A-NEXT:    s_cselect_b32 s4, s10, s4
+; GFX90A-NEXT:    s_lshr_b32 s9, s9, 16
+; GFX90A-NEXT:    s_lshl_b64 s[12:13], s[4:5], 5
+; GFX90A-NEXT:    v_cvt_f32_f16_e32 v1, s9
+; GFX90A-NEXT:    s_lshl_b64 s[2:3], s[0:1], 5
+; GFX90A-NEXT:    s_lshl_b64 s[10:11], s[6:7], 5
 ; GFX90A-NEXT:    s_or_b32 s10, s10, 28
-; GFX90A-NEXT:    v_pk_mov_b32 v[4:5], s[8:9], s[8:9] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[6:7], s[10:11], s[10:11] op_sel:[0,1]
-; GFX90A-NEXT:    v_mul_lo_u32 v8, s12, v0
-; GFX90A-NEXT:    v_mul_hi_u32 v8, v0, v8
-; GFX90A-NEXT:    v_add_u32_e32 v0, v0, v8
-; GFX90A-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX90A-NEXT:    v_mul_lo_u32 v8, v0, s3
-; GFX90A-NEXT:    v_sub_u32_e32 v8, s2, v8
-; GFX90A-NEXT:    v_add_u32_e32 v9, 1, v0
-; GFX90A-NEXT:    v_cmp_le_u32_e32 vcc, s3, v8
-; GFX90A-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
-; GFX90A-NEXT:    v_subrev_u32_e32 v9, s3, v8
-; GFX90A-NEXT:    v_cndmask_b32_e32 v8, v8, v9, vcc
-; GFX90A-NEXT:    v_add_u32_e32 v9, 1, v0
-; GFX90A-NEXT:    v_cmp_le_u32_e32 vcc, s3, v8
-; GFX90A-NEXT:    v_cndmask_b32_e32 v0, v0, v9, vcc
-; GFX90A-NEXT:    v_lshlrev_b64 v[8:9], 5, v[0:1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[10:11], 0, 0
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_readfirstlane_b32 s2, v28
-; GFX90A-NEXT:    s_and_b32 s2, 0xffff, s2
-; GFX90A-NEXT:    s_mul_i32 s1, s1, s2
-; GFX90A-NEXT:    s_mul_hi_u32 s3, s0, s2
-; GFX90A-NEXT:    s_mul_i32 s0, s0, s2
-; GFX90A-NEXT:    s_add_i32 s1, s3, s1
-; GFX90A-NEXT:    s_lshl_b64 s[2:3], s[0:1], 5
+; GFX90A-NEXT:    v_readfirstlane_b32 s5, v18
+; GFX90A-NEXT:    s_and_b32 s5, 0xffff, s5
+; GFX90A-NEXT:    s_mul_i32 s1, s1, s5
+; GFX90A-NEXT:    s_mul_hi_u32 s9, s0, s5
+; GFX90A-NEXT:    s_mul_i32 s0, s0, s5
+; GFX90A-NEXT:    s_add_i32 s1, s9, s1
+; GFX90A-NEXT:    s_lshl_b64 s[0:1], s[0:1], 5
 ; GFX90A-NEXT:    s_branch .LBB3_2
 ; GFX90A-NEXT:  .LBB3_1: ; %bb12
 ; GFX90A-NEXT:    ; in Loop: Header=BB3_2 Depth=1
-; GFX90A-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v0
-; GFX90A-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
-; GFX90A-NEXT:    v_add_co_u32_e32 v6, vcc, v6, v8
-; GFX90A-NEXT:    v_addc_co_u32_e32 v7, vcc, v7, v9, vcc
+; GFX90A-NEXT:    s_add_u32 s6, s6, s4
+; GFX90A-NEXT:    s_addc_u32 s7, s7, 0
+; GFX90A-NEXT:    s_add_u32 s10, s10, s12
+; GFX90A-NEXT:    s_addc_u32 s11, s11, s13
 ; GFX90A-NEXT:  .LBB3_2: ; %bb9
 ; GFX90A-NEXT:    ; =>This Loop Header: Depth=1
 ; GFX90A-NEXT:    ; Child Loop BB3_5 Depth 2
 ; GFX90A-NEXT:    s_cbranch_scc0 .LBB3_1
 ; GFX90A-NEXT:  ; %bb.3: ; %bb14
 ; GFX90A-NEXT:    ; in Loop: Header=BB3_2 Depth=1
-; GFX90A-NEXT:    global_load_dwordx2 v[12:13], v[10:11], off
-; GFX90A-NEXT:    s_mov_b32 s7, s6
-; GFX90A-NEXT:    v_pk_mov_b32 v[16:17], s[6:7], s[6:7] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[18:19], s[6:7], s[6:7] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[20:21], s[6:7], s[6:7] op_sel:[0,1]
-; GFX90A-NEXT:    v_cmp_gt_i64_e64 s[0:1], 0, v[4:5]
-; GFX90A-NEXT:    v_pk_mov_b32 v[14:15], v[6:7], v[6:7] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[22:23], v[16:17], v[16:17] op_sel:[0,1]
+; GFX90A-NEXT:    global_load_dwordx2 v[4:5], v[2:3], off
+; GFX90A-NEXT:    s_mov_b32 s9, s8
+; GFX90A-NEXT:    v_pk_mov_b32 v[6:7], s[8:9], s[8:9] op_sel:[0,1]
+; GFX90A-NEXT:    v_pk_mov_b32 v[8:9], s[8:9], s[8:9] op_sel:[0,1]
+; GFX90A-NEXT:    v_pk_mov_b32 v[10:11], s[8:9], s[8:9] op_sel:[0,1]
+; GFX90A-NEXT:    v_cmp_lt_i64_e64 s[14:15], s[6:7], 0
+; GFX90A-NEXT:    s_mov_b64 s[16:17], s[10:11]
+; GFX90A-NEXT:    v_pk_mov_b32 v[12:13], v[6:7], v[6:7] op_sel:[0,1]
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_readfirstlane_b32 s7, v12
-; GFX90A-NEXT:    v_readfirstlane_b32 s8, v13
-; GFX90A-NEXT:    s_add_u32 s7, s7, 1
-; GFX90A-NEXT:    s_addc_u32 s9, s8, 0
-; GFX90A-NEXT:    s_mul_hi_u32 s10, s4, s7
-; GFX90A-NEXT:    s_mul_i32 s11, s5, s7
-; GFX90A-NEXT:    s_mul_i32 s8, s4, s7
-; GFX90A-NEXT:    s_mul_i32 s7, s4, s9
-; GFX90A-NEXT:    s_add_i32 s7, s10, s7
-; GFX90A-NEXT:    s_add_i32 s7, s7, s11
+; GFX90A-NEXT:    v_readfirstlane_b32 s5, v4
+; GFX90A-NEXT:    v_readfirstlane_b32 s9, v5
+; GFX90A-NEXT:    s_add_u32 s5, s5, 1
+; GFX90A-NEXT:    s_addc_u32 s9, s9, 0
+; GFX90A-NEXT:    s_mul_hi_u32 s19, s2, s5
+; GFX90A-NEXT:    s_mul_i32 s20, s3, s5
+; GFX90A-NEXT:    s_mul_i32 s18, s2, s5
+; GFX90A-NEXT:    s_mul_i32 s5, s2, s9
+; GFX90A-NEXT:    s_add_i32 s5, s19, s5
+; GFX90A-NEXT:    s_add_i32 s5, s5, s20
 ; GFX90A-NEXT:    s_branch .LBB3_5
 ; GFX90A-NEXT:  .LBB3_4: ; %bb58
 ; GFX90A-NEXT:    ; in Loop: Header=BB3_5 Depth=2
-; GFX90A-NEXT:    v_add_co_u32_sdwa v12, vcc, v12, v28 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
-; GFX90A-NEXT:    v_addc_co_u32_e32 v13, vcc, 0, v13, vcc
-; GFX90A-NEXT:    v_mov_b32_e32 v24, s3
-; GFX90A-NEXT:    v_add_co_u32_e32 v14, vcc, s2, v14
-; GFX90A-NEXT:    v_addc_co_u32_e32 v15, vcc, v15, v24, vcc
-; GFX90A-NEXT:    v_cmp_gt_i64_e32 vcc, 0, v[12:13]
+; GFX90A-NEXT:    v_add_co_u32_sdwa v4, vcc, v4, v18 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_0
+; GFX90A-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v5, vcc
+; GFX90A-NEXT:    s_add_u32 s16, s16, s0
+; GFX90A-NEXT:    v_cmp_gt_i64_e32 vcc, 0, v[4:5]
+; GFX90A-NEXT:    s_addc_u32 s17, s17, s1
 ; GFX90A-NEXT:    s_cbranch_vccz .LBB3_1
 ; GFX90A-NEXT:  .LBB3_5: ; %bb16
 ; GFX90A-NEXT:    ; Parent Loop BB3_2 Depth=1
 ; GFX90A-NEXT:    ; => This Inner Loop Header: Depth=2
-; GFX90A-NEXT:    v_mov_b32_e32 v25, s7
-; GFX90A-NEXT:    v_add_co_u32_e32 v24, vcc, s8, v14
-; GFX90A-NEXT:    v_addc_co_u32_e32 v25, vcc, v15, v25, vcc
-; GFX90A-NEXT:    global_load_dword v30, v[24:25], off offset:-12 glc
-; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    global_load_dword v29, v[24:25], off offset:-8 glc
+; GFX90A-NEXT:    s_add_u32 s20, s16, s18
+; GFX90A-NEXT:    s_addc_u32 s21, s17, s5
+; GFX90A-NEXT:    global_load_dword v21, v19, s[20:21] offset:-12 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    global_load_dword v26, v[24:25], off offset:-4 glc
+; GFX90A-NEXT:    global_load_dword v20, v19, s[20:21] offset:-8 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    global_load_dword v26, v[24:25], off glc
+; GFX90A-NEXT:    global_load_dword v14, v19, s[20:21] offset:-4 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    ; kill: killed $vgpr24 killed $vgpr25
-; GFX90A-NEXT:    ds_read_b64 v[24:25], v1
+; GFX90A-NEXT:    global_load_dword v14, v19, s[20:21] glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    ds_read_b64 v[26:27], v0
-; GFX90A-NEXT:    s_and_b64 vcc, exec, s[0:1]
+; GFX90A-NEXT:    ds_read_b64 v[14:15], v19
+; GFX90A-NEXT:    ds_read_b64 v[16:17], v0
+; GFX90A-NEXT:    s_and_b64 vcc, exec, s[14:15]
+; GFX90A-NEXT:    ; kill: killed $sgpr20 killed $sgpr21
 ; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NEXT:    s_cbranch_vccnz .LBB3_4
 ; GFX90A-NEXT:  ; %bb.6: ; %bb51
 ; GFX90A-NEXT:    ; in Loop: Header=BB3_5 Depth=2
-; GFX90A-NEXT:    v_cvt_f32_f16_sdwa v31, v30 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v30, v30
-; GFX90A-NEXT:    v_cvt_f32_f16_sdwa v33, v29 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v32, v29
-; GFX90A-NEXT:    v_pk_add_f32 v[34:35], v[2:3], v[24:25]
-; GFX90A-NEXT:    v_pk_add_f32 v[36:37], v[24:25], 0 op_sel_hi:[1,0]
-; GFX90A-NEXT:    v_pk_add_f32 v[26:27], v[30:31], v[26:27]
-; GFX90A-NEXT:    v_pk_add_f32 v[24:25], v[32:33], v[24:25]
-; GFX90A-NEXT:    v_pk_add_f32 v[16:17], v[16:17], v[34:35]
-; GFX90A-NEXT:    v_pk_add_f32 v[18:19], v[18:19], v[36:37]
-; GFX90A-NEXT:    v_pk_add_f32 v[20:21], v[20:21], v[26:27]
-; GFX90A-NEXT:    v_pk_add_f32 v[22:23], v[22:23], v[24:25]
+; GFX90A-NEXT:    v_cvt_f32_f16_sdwa v23, v21 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX90A-NEXT:    v_cvt_f32_f16_e32 v22, v21
+; GFX90A-NEXT:    v_cvt_f32_f16_sdwa v21, v20 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
+; GFX90A-NEXT:    v_cvt_f32_f16_e32 v20, v20
+; GFX90A-NEXT:    v_pk_add_f32 v[24:25], v[0:1], v[14:15]
+; GFX90A-NEXT:    v_pk_add_f32 v[26:27], v[14:15], 0 op_sel_hi:[1,0]
+; GFX90A-NEXT:    v_pk_add_f32 v[16:17], v[22:23], v[16:17]
+; GFX90A-NEXT:    v_pk_add_f32 v[14:15], v[20:21], v[14:15]
+; GFX90A-NEXT:    v_pk_add_f32 v[6:7], v[6:7], v[24:25]
+; GFX90A-NEXT:    v_pk_add_f32 v[8:9], v[8:9], v[26:27]
+; GFX90A-NEXT:    v_pk_add_f32 v[10:11], v[10:11], v[16:17]
+; GFX90A-NEXT:    v_pk_add_f32 v[12:13], v[12:13], v[14:15]
 ; GFX90A-NEXT:    s_branch .LBB3_4
 bb:
   %i = load volatile i16, i16 addrspace(4)* undef, align 2
   %i6 = zext i16 %i to i64
   %i7 = udiv i32 %arg1, %arg2
   %i8 = zext i32 %i7 to i64
   br label %bb9
 

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

 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; GFX6-NEXT:    s_mov_b32 s4, s0
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v1, v0, s3
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s2, v1
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_readfirstlane_b32 s0, v0
+; GFX6-NEXT:    s_mul_i32 s0, s0, s3
+; GFX6-NEXT:    s_sub_i32 s0, s2, s0
+; GFX6-NEXT:    s_sub_i32 s1, s0, s3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: udiv_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
 ; GFX9-NEXT:    s_sub_i32 s4, 0, s3
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    v_sub_u32_e32 v1, s2, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX9-NEXT:    s_mul_i32 s4, s4, s5
+; GFX9-NEXT:    s_mul_hi_u32 s4, s5, s4
+; GFX9-NEXT:    s_add_i32 s5, s5, s4
+; GFX9-NEXT:    s_mul_hi_u32 s4, s2, s5
+; GFX9-NEXT:    s_mul_i32 s5, s4, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s5
+; GFX9-NEXT:    s_add_i32 s6, s4, 1
+; GFX9-NEXT:    s_sub_i32 s5, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX9-NEXT:    s_cselect_b32 s2, s5, s2
+; GFX9-NEXT:    s_add_i32 s5, s4, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s5, s4
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = udiv i32 %x, %y
   store i32 %r, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @urem_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; CHECK-LABEL: @urem_i32(

 ; GFX6-NEXT:    s_sub_i32 s4, 0, s3
 ; GFX6-NEXT:    s_mov_b32 s5, s1
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; GFX6-NEXT:    s_mov_b32 s4, s0
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s3
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s3, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s3, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    v_readfirstlane_b32 s0, v0
+; GFX6-NEXT:    s_mul_i32 s0, s0, s3
+; GFX6-NEXT:    s_sub_i32 s0, s2, s0
+; GFX6-NEXT:    s_sub_i32 s1, s0, s3
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX6-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX6-NEXT:    s_sub_i32 s1, s0, s3
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX6-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX6-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: urem_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX6-NEXT:    s_xor_b32 s3, s3, s8
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s3
 ; GFX6-NEXT:    s_sub_i32 s4, 0, s3
 ; GFX6-NEXT:    s_ashr_i32 s9, s2, 31
 ; GFX6-NEXT:    s_add_i32 s2, s2, s9
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX6-NEXT:    s_xor_b32 s2, s2, s9
 ; GFX6-NEXT:    s_mov_b32 s5, s1
-; GFX6-NEXT:    s_xor_b32 s8, s9, s8
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; GFX6-NEXT:    s_mov_b32 s4, s0
+; GFX6-NEXT:    s_xor_b32 s0, s9, s8
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v1, v0, s3
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s2, v1
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s1, v0
+; GFX6-NEXT:    s_mul_i32 s1, s1, s3
+; GFX6-NEXT:    s_sub_i32 s1, s2, s1
+; GFX6-NEXT:    s_sub_i32 s2, s1, s3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s1, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    s_cselect_b32 s1, s2, s1
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s1, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_ashr_i32 s4, s3, 31
 ; GFX9-NEXT:    s_add_i32 s3, s3, s4
 ; GFX9-NEXT:    s_xor_b32 s3, s3, s4
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX9-NEXT:    s_sub_i32 s5, 0, s3
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, s5, v0
 ; GFX9-NEXT:    s_ashr_i32 s5, s2, 31
 ; GFX9-NEXT:    s_add_i32 s2, s2, s5
-; GFX9-NEXT:    s_xor_b32 s2, s2, s5
-; GFX9-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX9-NEXT:    s_xor_b32 s4, s5, s4
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    v_sub_u32_e32 v1, s2, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s4, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s4, v0
-; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s2, s2, s5
+; GFX9-NEXT:    s_sub_i32 s5, 0, s3
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX9-NEXT:    s_mul_i32 s5, s5, s6
+; GFX9-NEXT:    s_mul_hi_u32 s5, s6, s5
+; GFX9-NEXT:    s_add_i32 s6, s6, s5
+; GFX9-NEXT:    s_mul_hi_u32 s5, s2, s6
+; GFX9-NEXT:    s_mul_i32 s6, s5, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s6
+; GFX9-NEXT:    s_add_i32 s7, s5, 1
+; GFX9-NEXT:    s_sub_i32 s6, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX9-NEXT:    s_cselect_b32 s2, s6, s2
+; GFX9-NEXT:    s_add_i32 s6, s5, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s6, s5
+; GFX9-NEXT:    s_xor_b32 s2, s2, s4
+; GFX9-NEXT:    s_sub_i32 s2, s2, s4
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = sdiv i32 %x, %y
   store i32 %r, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @srem_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; CHECK-LABEL: @srem_i32(

 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s6, v0
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s4, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s4, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v0, s5, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX6-NEXT:    s_mul_i32 s7, s7, s4
+; GFX6-NEXT:    s_sub_i32 s6, s6, s7
+; GFX6-NEXT:    s_sub_i32 s7, s6, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX6-NEXT:    s_cselect_b32 s6, s7, s6
+; GFX6-NEXT:    s_sub_i32 s7, s6, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX6-NEXT:    s_cselect_b32 s4, s7, s6
+; GFX6-NEXT:    s_xor_b32 s4, s4, s5
+; GFX6-NEXT:    s_sub_i32 s4, s4, s5
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_ashr_i32 s5, s4, 16
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s5
 ; GFX6-NEXT:    s_sext_i32_i16 s4, s4
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GFX6-NEXT:    s_xor_b32 s4, s4, s5
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s4
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GFX6-NEXT:    buffer_store_short v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0

 ; CHECK-NEXT:    store i16 [[TMP23]], i16 addrspace(1)* [[OUT:%.*]], align 2
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: srem_i16:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_ashr_i32 s2, s4, 16
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s2
-; GFX6-NEXT:    s_sext_i32_i16 s3, s4
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s3
-; GFX6-NEXT:    s_xor_b32 s3, s3, s2
+; GFX6-NEXT:    s_ashr_i32 s5, s4, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s5
+; GFX6-NEXT:    s_sext_i32_i16 s2, s4
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s2
+; GFX6-NEXT:    s_xor_b32 s2, s2, s5
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GFX6-NEXT:    s_ashr_i32 s3, s3, 30
-; GFX6-NEXT:    s_or_b32 s3, s3, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s3
+; GFX6-NEXT:    s_ashr_i32 s2, s2, 30
+; GFX6-NEXT:    s_or_b32 s6, s2, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[2:3], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[2:3], s[2:3], exec
+; GFX6-NEXT:    s_cselect_b32 s2, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s2, v2
+; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s5
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    buffer_store_short v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c

 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s5, s4, 0x80008
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s5
 ; GFX6-NEXT:    s_sext_i32_i8 s4, s4
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GFX6-NEXT:    s_xor_b32 s4, s4, s5
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s4
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GFX6-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i8:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0

 ;
 ; GFX6-LABEL: srem_i8:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s2, s4, 0x80008
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s2
-; GFX6-NEXT:    s_sext_i32_i8 s5, s4
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s5
-; GFX6-NEXT:    s_xor_b32 s2, s5, s2
+; GFX6-NEXT:    s_sext_i32_i8 s3, s4
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s3
+; GFX6-NEXT:    s_xor_b32 s2, s3, s2
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s2, s2, 30
-; GFX6-NEXT:    s_or_b32 s2, s2, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s2
+; GFX6-NEXT:    s_lshr_b32 s5, s4, 8
+; GFX6-NEXT:    s_or_b32 s6, s2, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    s_lshr_b32 s3, s4, 8
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s3
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[2:3], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[2:3], s[2:3], exec
+; GFX6-NEXT:    s_cselect_b32 s2, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s2, v2
+; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s5
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_i8:
 ; GFX9:       ; %bb.0:

 ; CHECK-NEXT:    [[TMP126:%.*]] = add i32 [[TMP122]], 1
 ; CHECK-NEXT:    [[TMP127:%.*]] = select i1 [[TMP125]], i32 [[TMP126]], i32 [[TMP122]]
 ; CHECK-NEXT:    [[TMP128:%.*]] = insertelement <4 x i32> [[TMP96]], i32 [[TMP127]], i64 3
 ; CHECK-NEXT:    store <4 x i32> [[TMP128]], <4 x i32> addrspace(1)* [[OUT:%.*]], align 16
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: udiv_v4i32:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
-; GFX6-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s15, 0xf000
-; GFX6-NEXT:    s_mov_b32 s14, -1
+; GFX6-NEXT:    s_load_dwordx8 s[8:15], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dwordx2 s[16:17], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s19, 0xf000
+; GFX6-NEXT:    s_mov_b32 s18, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s8
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX6-NEXT:    s_sub_i32 s2, 0, s8
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v4, s10
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s12
+; GFX6-NEXT:    s_sub_i32 s2, 0, s12
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s13
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v4, s14
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v6, s11
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v6, s15
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v4
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v6, v6
+; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s8, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v0
-; GFX6-NEXT:    s_sub_i32 s2, 0, s9
-; GFX6-NEXT:    v_mul_lo_u32 v3, s2, v1
-; GFX6-NEXT:    s_sub_i32 s2, 0, s10
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s12
+; GFX6-NEXT:    s_sub_i32 s2, s8, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s12
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s12
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s12
+; GFX6-NEXT:    s_cselect_b64 s[2:3], -1, 0
+; GFX6-NEXT:    s_sub_i32 s4, 0, s13
+; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v1
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
-; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, v0, s8
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v0
-; GFX6-NEXT:    v_mul_lo_u32 v5, v1, s9
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s4, v2
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s8, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v2
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v4
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s5, v5
-; GFX6-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[2:3]
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
+; GFX6-NEXT:    v_mul_hi_u32 v1, s9, v1
+; GFX6-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v4
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
+; GFX6-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX6-NEXT:    s_mul_i32 s4, s4, s13
+; GFX6-NEXT:    s_sub_i32 s4, s9, s4
+; GFX6-NEXT:    s_sub_i32 s5, s4, s13
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s13
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v1
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s9, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
-; GFX6-NEXT:    v_mul_lo_u32 v4, s2, v2
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s9, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
-; GFX6-NEXT:    v_mul_hi_u32 v4, v2, v4
-; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v1
-; GFX6-NEXT:    s_sub_i32 s0, 0, s11
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v6
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; GFX6-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX6-NEXT:    v_mul_f32_e32 v4, 0x4f7ffffe, v4
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v4, v4
-; GFX6-NEXT:    v_mul_lo_u32 v3, v2, s10
-; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v2
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v4
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s6, v3
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s10, v3
-; GFX6-NEXT:    v_mul_hi_u32 v5, v4, v5
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v6, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v6, vcc, s10, v3
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
-; GFX6-NEXT:    v_mul_hi_u32 v4, s7, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v2
-; GFX6-NEXT:    v_mul_lo_u32 v6, v4, s11
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v4
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s7, v6
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v4, v5, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v4
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
-; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[12:15], 0
+; GFX6-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s13
+; GFX6-NEXT:    s_cselect_b64 s[4:5], -1, 0
+; GFX6-NEXT:    s_sub_i32 s6, 0, s14
+; GFX6-NEXT:    v_mul_lo_u32 v5, s6, v3
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v1
+; GFX6-NEXT:    v_mul_hi_u32 v5, v3, v5
+; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[4:5]
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
+; GFX6-NEXT:    v_mul_hi_u32 v3, s10, v3
+; GFX6-NEXT:    v_mul_f32_e32 v5, 0x4f7ffffe, v6
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v5, v5
+; GFX6-NEXT:    v_readfirstlane_b32 s6, v3
+; GFX6-NEXT:    s_mul_i32 s6, s6, s14
+; GFX6-NEXT:    s_sub_i32 s6, s10, s6
+; GFX6-NEXT:    s_sub_i32 s7, s6, s14
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s14
+; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v3
+; GFX6-NEXT:    s_cselect_b32 s6, s7, s6
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s14
+; GFX6-NEXT:    s_cselect_b64 s[6:7], -1, 0
+; GFX6-NEXT:    s_sub_i32 s8, 0, s15
+; GFX6-NEXT:    v_mul_lo_u32 v7, s8, v5
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
+; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v3
+; GFX6-NEXT:    v_mul_hi_u32 v7, v5, v7
+; GFX6-NEXT:    v_cndmask_b32_e64 v2, v3, v6, s[6:7]
+; GFX6-NEXT:    v_add_i32_e32 v5, vcc, v7, v5
+; GFX6-NEXT:    v_mul_hi_u32 v5, s11, v5
+; GFX6-NEXT:    v_readfirstlane_b32 s0, v5
+; GFX6-NEXT:    s_mul_i32 s0, s0, s15
+; GFX6-NEXT:    s_sub_i32 s0, s11, s0
+; GFX6-NEXT:    s_sub_i32 s1, s0, s15
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v5
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s15
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX6-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v3
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s15
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[16:19], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: udiv_v4i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s8
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s9
 ; GFX9-NEXT:    s_sub_i32 s2, 0, s8
-; GFX9-NEXT:    s_sub_i32 s3, 0, s9
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s10
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v5, s10
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v6, s11
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX9-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v5, v5
-; GFX9-NEXT:    v_mul_lo_u32 v2, s2, v0
-; GFX9-NEXT:    s_sub_i32 s2, 0, s10
-; GFX9-NEXT:    v_mul_lo_u32 v3, s3, v1
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v6, v6
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX9-NEXT:    v_mul_f32_e32 v6, 0x4f7ffffe, v6
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v3
-; GFX9-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v5
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, s8
-; GFX9-NEXT:    v_add_u32_e32 v7, 1, v0
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v6, v6
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_sub_u32_e32 v3, s4, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v7, s8, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v3
-; GFX9-NEXT:    v_mul_lo_u32 v3, s2, v2
-; GFX9-NEXT:    s_sub_i32 s2, 0, s11
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, s9
-; GFX9-NEXT:    v_add_u32_e32 v7, 1, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, v2, v3
-; GFX9-NEXT:    v_add_u32_e32 v8, 1, v1
-; GFX9-NEXT:    v_sub_u32_e32 v5, s5, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
-; GFX9-NEXT:    v_mul_lo_u32 v3, s2, v6
-; GFX9-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v8, vcc
-; GFX9-NEXT:    v_mul_hi_u32 v3, v6, v3
-; GFX9-NEXT:    v_mul_lo_u32 v8, v2, s10
-; GFX9-NEXT:    v_subrev_u32_e32 v7, s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
-; GFX9-NEXT:    v_add_u32_e32 v3, v6, v3
-; GFX9-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX9-NEXT:    v_add_u32_e32 v7, 1, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v5
-; GFX9-NEXT:    v_sub_u32_e32 v5, s6, v8
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v7, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s10, v5
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s10, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v6, v3, s11
-; GFX9-NEXT:    v_add_u32_e32 v7, 1, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v7, vcc
-; GFX9-NEXT:    v_add_u32_e32 v7, 1, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s10, v5
-; GFX9-NEXT:    v_sub_u32_e32 v5, s7, v6
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v7, vcc
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s11, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s11, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s11, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX9-NEXT:    s_mul_i32 s2, s2, s3
+; GFX9-NEXT:    s_mul_hi_u32 s2, s3, s2
+; GFX9-NEXT:    s_add_i32 s3, s3, s2
+; GFX9-NEXT:    s_mul_hi_u32 s2, s4, s3
+; GFX9-NEXT:    s_mul_i32 s3, s2, s8
+; GFX9-NEXT:    s_sub_i32 s3, s4, s3
+; GFX9-NEXT:    s_add_i32 s13, s2, 1
+; GFX9-NEXT:    s_sub_i32 s4, s3, s8
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s8
+; GFX9-NEXT:    s_cselect_b32 s2, s13, s2
+; GFX9-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX9-NEXT:    s_add_i32 s4, s2, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s8
+; GFX9-NEXT:    v_readfirstlane_b32 s12, v1
+; GFX9-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX9-NEXT:    s_sub_i32 s3, 0, s9
+; GFX9-NEXT:    s_mul_i32 s3, s3, s12
+; GFX9-NEXT:    s_mul_hi_u32 s3, s12, s3
+; GFX9-NEXT:    s_add_i32 s12, s12, s3
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v2
+; GFX9-NEXT:    s_mul_hi_u32 s3, s5, s12
+; GFX9-NEXT:    s_mul_i32 s4, s3, s9
+; GFX9-NEXT:    s_sub_i32 s4, s5, s4
+; GFX9-NEXT:    s_add_i32 s8, s3, 1
+; GFX9-NEXT:    s_sub_i32 s5, s4, s9
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s9
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_cselect_b32 s3, s8, s3
+; GFX9-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX9-NEXT:    s_add_i32 s5, s3, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s9
+; GFX9-NEXT:    s_cselect_b32 s3, s5, s3
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s11
+; GFX9-NEXT:    s_sub_i32 s4, 0, s10
+; GFX9-NEXT:    s_mul_i32 s4, s4, s5
+; GFX9-NEXT:    s_mul_hi_u32 s4, s5, s4
+; GFX9-NEXT:    s_add_i32 s5, s5, s4
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_mul_hi_u32 s4, s6, s5
+; GFX9-NEXT:    s_mul_i32 s5, s4, s10
+; GFX9-NEXT:    s_sub_i32 s5, s6, s5
+; GFX9-NEXT:    s_add_i32 s6, s4, 1
+; GFX9-NEXT:    s_sub_i32 s8, s5, s10
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s10
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX9-NEXT:    s_cselect_b32 s5, s8, s5
+; GFX9-NEXT:    s_add_i32 s6, s4, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s10
+; GFX9-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX9-NEXT:    s_sub_i32 s5, 0, s11
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX9-NEXT:    s_mul_i32 s5, s5, s6
+; GFX9-NEXT:    s_mul_hi_u32 s5, s6, s5
+; GFX9-NEXT:    s_add_i32 s6, s6, s5
+; GFX9-NEXT:    s_mul_hi_u32 s5, s7, s6
+; GFX9-NEXT:    s_mul_i32 s6, s5, s11
+; GFX9-NEXT:    s_sub_i32 s6, s7, s6
+; GFX9-NEXT:    s_add_i32 s7, s5, 1
+; GFX9-NEXT:    s_sub_i32 s8, s6, s11
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s11
+; GFX9-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX9-NEXT:    s_add_i32 s7, s5, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s11
+; GFX9-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = udiv <4 x i32> %x, %y
   store <4 x i32> %r, <4 x i32> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @urem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x, <4 x i32> %y) {

 ; CHECK-NEXT:    [[TMP120:%.*]] = insertelement <4 x i32> [[TMP90]], i32 [[TMP119]], i64 3
 ; CHECK-NEXT:    store <4 x i32> [[TMP120]], <4 x i32> addrspace(1)* [[OUT:%.*]], align 16
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: urem_v4i32:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s3, 0xf000
-; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s8
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX6-NEXT:    s_sub_i32 s12, 0, s8
-; GFX6-NEXT:    s_sub_i32 s13, 0, s9
+; GFX6-NEXT:    s_sub_i32 s2, 0, s8
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s9
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v3, s10
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v5, s11
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX6-NEXT:    v_mul_lo_u32 v2, s12, v0
-; GFX6-NEXT:    v_mul_lo_u32 v4, s13, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
-; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s8
-; GFX6-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v3
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s9
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    s_sub_i32 s4, 0, s10
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v2
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s5, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s9, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_mul_hi_u32 v3, v2, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v5
-; GFX6-NEXT:    s_sub_i32 s4, 0, s11
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v4
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s9, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX6-NEXT:    v_mul_lo_u32 v5, s4, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s10
-; GFX6-NEXT:    v_mul_hi_u32 v4, v3, v5
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s6, v2
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
-; GFX6-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s10, v2
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX6-NEXT:    v_mul_lo_u32 v3, v3, s11
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s10, v2
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s7, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s11, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s11, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s10
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s8
+; GFX6-NEXT:    s_sub_i32 s2, s4, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s8
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s8
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX6-NEXT:    s_cselect_b32 s4, s3, s2
+; GFX6-NEXT:    s_sub_i32 s2, 0, s9
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s11
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s9
+; GFX6-NEXT:    s_sub_i32 s2, s5, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s9
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s9
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s9
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s9
+; GFX6-NEXT:    s_cselect_b32 s5, s3, s2
+; GFX6-NEXT:    s_sub_i32 s2, 0, s10
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s6, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s10
+; GFX6-NEXT:    s_sub_i32 s2, s6, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s10
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s10
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s10
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s10
+; GFX6-NEXT:    s_cselect_b32 s6, s3, s2
+; GFX6-NEXT:    s_sub_i32 s2, 0, s11
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v2, s7, v0
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
+; GFX6-NEXT:    v_mov_b32_e32 v1, s5
+; GFX6-NEXT:    v_readfirstlane_b32 s4, v2
+; GFX6-NEXT:    s_mul_i32 s4, s4, s11
+; GFX6-NEXT:    s_sub_i32 s4, s7, s4
+; GFX6-NEXT:    s_sub_i32 s5, s4, s11
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s11
+; GFX6-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX6-NEXT:    s_sub_i32 s5, s4, s11
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s11
+; GFX6-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX6-NEXT:    v_mov_b32_e32 v2, s6
+; GFX6-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: urem_v4i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24

 ; CHECK-NEXT:    [[TMP162:%.*]] = xor i32 [[TMP161]], [[TMP128]]
 ; CHECK-NEXT:    [[TMP163:%.*]] = sub i32 [[TMP162]], [[TMP128]]
 ; CHECK-NEXT:    [[TMP164:%.*]] = insertelement <4 x i32> [[TMP123]], i32 [[TMP163]], i64 3
 ; CHECK-NEXT:    store <4 x i32> [[TMP164]], <4 x i32> addrspace(1)* [[OUT:%.*]], align 16
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: sdiv_v4i32:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
-; GFX6-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s15, 0xf000
-; GFX6-NEXT:    s_mov_b32 s14, -1
+; GFX6-NEXT:    s_load_dwordx8 s[8:15], s[0:1], 0xd
+; GFX6-NEXT:    s_load_dwordx2 s[16:17], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s19, 0xf000
+; GFX6-NEXT:    s_mov_b32 s18, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_ashr_i32 s2, s8, 31
-; GFX6-NEXT:    s_add_i32 s3, s8, s2
+; GFX6-NEXT:    s_ashr_i32 s2, s12, 31
+; GFX6-NEXT:    s_add_i32 s3, s12, s2
 ; GFX6-NEXT:    s_xor_b32 s3, s3, s2
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX6-NEXT:    s_ashr_i32 s8, s9, 31
-; GFX6-NEXT:    s_add_i32 s0, s9, s8
-; GFX6-NEXT:    s_xor_b32 s9, s0, s8
+; GFX6-NEXT:    s_sub_i32 s4, 0, s3
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX6-NEXT:    s_sub_i32 s1, 0, s3
-; GFX6-NEXT:    s_ashr_i32 s0, s4, 31
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    s_xor_b32 s2, s0, s2
-; GFX6-NEXT:    v_mul_lo_u32 v2, s1, v0
-; GFX6-NEXT:    s_add_i32 s1, s4, s0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX6-NEXT:    s_xor_b32 s1, s1, s0
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    s_sub_i32 s0, 0, s9
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GFX6-NEXT:    v_mul_hi_u32 v0, s1, v0
-; GFX6-NEXT:    v_mul_lo_u32 v2, s0, v1
-; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s3
-; GFX6-NEXT:    v_mul_hi_u32 v2, v1, v2
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v0
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s1, v3
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s3, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v3
-; GFX6-NEXT:    s_ashr_i32 s0, s5, 31
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX6-NEXT:    s_add_i32 s1, s5, s0
-; GFX6-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX6-NEXT:    s_ashr_i32 s3, s10, 31
-; GFX6-NEXT:    s_xor_b32 s1, s1, s0
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s2, v0
-; GFX6-NEXT:    s_xor_b32 s2, s0, s8
-; GFX6-NEXT:    s_add_i32 s0, s10, s3
-; GFX6-NEXT:    s_xor_b32 s4, s0, s3
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v3, s4
-; GFX6-NEXT:    v_mul_hi_u32 v1, s1, v1
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s9
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v1
-; GFX6-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s1, v2
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s9, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s9, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
-; GFX6-NEXT:    s_sub_i32 s0, 0, s4
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v3
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_mul_hi_u32 v2, v3, v5
-; GFX6-NEXT:    v_xor_b32_e32 v1, s2, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s2, v1
-; GFX6-NEXT:    s_ashr_i32 s2, s11, 31
-; GFX6-NEXT:    s_ashr_i32 s0, s6, 31
-; GFX6-NEXT:    s_add_i32 s5, s11, s2
-; GFX6-NEXT:    s_add_i32 s1, s6, s0
-; GFX6-NEXT:    s_xor_b32 s5, s5, s2
-; GFX6-NEXT:    s_xor_b32 s1, s1, s0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
+; GFX6-NEXT:    s_ashr_i32 s4, s8, 31
+; GFX6-NEXT:    s_add_i32 s5, s8, s4
+; GFX6-NEXT:    s_xor_b32 s5, s5, s4
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    s_xor_b32 s8, s4, s2
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s3
+; GFX6-NEXT:    s_sub_i32 s2, s5, s2
+; GFX6-NEXT:    s_sub_i32 s4, s2, s3
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX6-NEXT:    s_cselect_b64 s[2:3], -1, 0
+; GFX6-NEXT:    s_ashr_i32 s4, s13, 31
+; GFX6-NEXT:    s_add_i32 s5, s13, s4
+; GFX6-NEXT:    s_xor_b32 s5, s5, s4
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s5
+; GFX6-NEXT:    s_sub_i32 s6, 0, s5
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v1, s[2:3]
+; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
+; GFX6-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX6-NEXT:    v_mul_lo_u32 v3, s6, v2
+; GFX6-NEXT:    s_ashr_i32 s6, s9, 31
+; GFX6-NEXT:    s_add_i32 s7, s9, s6
+; GFX6-NEXT:    s_xor_b32 s7, s7, s6
+; GFX6-NEXT:    v_mul_hi_u32 v3, v2, v3
+; GFX6-NEXT:    s_xor_b32 s9, s6, s4
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v4, s5
-; GFX6-NEXT:    v_mul_hi_u32 v2, s1, v2
-; GFX6-NEXT:    s_xor_b32 s3, s0, s3
+; GFX6-NEXT:    v_mul_hi_u32 v2, s7, v2
+; GFX6-NEXT:    v_readfirstlane_b32 s4, v2
+; GFX6-NEXT:    s_mul_i32 s4, s4, s5
+; GFX6-NEXT:    s_sub_i32 s4, s7, s4
+; GFX6-NEXT:    s_sub_i32 s6, s4, s5
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s5
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v2
+; GFX6-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s5
+; GFX6-NEXT:    s_cselect_b64 s[4:5], -1, 0
+; GFX6-NEXT:    s_ashr_i32 s6, s14, 31
+; GFX6-NEXT:    s_add_i32 s7, s14, s6
+; GFX6-NEXT:    s_xor_b32 s7, s7, s6
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v4, s7
+; GFX6-NEXT:    s_sub_i32 s12, 0, s7
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v2
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v4
-; GFX6-NEXT:    v_mul_lo_u32 v3, v2, s4
-; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v2
+; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[4:5]
+; GFX6-NEXT:    v_xor_b32_e32 v2, s9, v2
 ; GFX6-NEXT:    v_mul_f32_e32 v4, 0x4f7ffffe, v4
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s1, v3
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v4, v4
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s4, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s4, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
-; GFX6-NEXT:    s_sub_i32 s0, 0, s5
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v4
-; GFX6-NEXT:    s_ashr_i32 s0, s7, 31
-; GFX6-NEXT:    s_add_i32 s1, s7, s0
-; GFX6-NEXT:    s_xor_b32 s1, s1, s0
+; GFX6-NEXT:    v_mul_lo_u32 v5, s12, v4
+; GFX6-NEXT:    s_ashr_i32 s12, s10, 31
+; GFX6-NEXT:    s_add_i32 s10, s10, s12
+; GFX6-NEXT:    s_xor_b32 s10, s10, s12
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v4, v5
-; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v2
-; GFX6-NEXT:    s_xor_b32 s2, s0, s2
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
-; GFX6-NEXT:    v_mul_hi_u32 v4, s1, v4
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v2, s3, v2
-; GFX6-NEXT:    v_mul_lo_u32 v3, v4, s5
+; GFX6-NEXT:    s_xor_b32 s12, s12, s6
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
+; GFX6-NEXT:    v_mul_hi_u32 v4, s10, v4
+; GFX6-NEXT:    v_readfirstlane_b32 s6, v4
+; GFX6-NEXT:    s_mul_i32 s6, s6, s7
+; GFX6-NEXT:    s_sub_i32 s6, s10, s6
+; GFX6-NEXT:    s_sub_i32 s10, s6, s7
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s7
 ; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v4
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s3, v2
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s1, v3
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s5, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v4, v5, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s5, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
+; GFX6-NEXT:    s_cselect_b32 s6, s10, s6
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s7
+; GFX6-NEXT:    s_cselect_b64 s[6:7], -1, 0
+; GFX6-NEXT:    s_ashr_i32 s10, s15, 31
+; GFX6-NEXT:    s_add_i32 s13, s15, s10
+; GFX6-NEXT:    s_xor_b32 s13, s13, s10
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v6, s13
+; GFX6-NEXT:    s_sub_i32 s0, 0, s13
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v5, vcc, 1, v4
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s5, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v4, v5, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v3, s2, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s2, v3
-; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[12:15], 0
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v6
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, v4, v5, s[6:7]
+; GFX6-NEXT:    v_xor_b32_e32 v4, s12, v4
+; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v1
+; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s9, v2
+; GFX6-NEXT:    v_mul_lo_u32 v2, s0, v3
+; GFX6-NEXT:    s_ashr_i32 s0, s11, 31
+; GFX6-NEXT:    s_add_i32 s1, s11, s0
+; GFX6-NEXT:    s_xor_b32 s1, s1, s0
+; GFX6-NEXT:    v_mul_hi_u32 v2, v3, v2
+; GFX6-NEXT:    s_xor_b32 s0, s0, s10
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_mul_hi_u32 v3, s1, v2
+; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s12, v4
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v3
+; GFX6-NEXT:    s_mul_i32 s2, s2, s13
+; GFX6-NEXT:    s_sub_i32 s1, s1, s2
+; GFX6-NEXT:    s_sub_i32 s2, s1, s13
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v3
+; GFX6-NEXT:    s_cmp_ge_u32 s1, s13
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX6-NEXT:    s_cselect_b32 s1, s2, s1
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v3
+; GFX6-NEXT:    s_cmp_ge_u32 s1, s13
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX6-NEXT:    v_xor_b32_e32 v3, s0, v3
+; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s0, v3
+; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[16:19], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_v4i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_ashr_i32 s2, s8, 31
 ; GFX9-NEXT:    s_add_i32 s3, s8, s2
 ; GFX9-NEXT:    s_xor_b32 s3, s3, s2
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX9-NEXT:    s_ashr_i32 s12, s9, 31
-; GFX9-NEXT:    s_add_i32 s9, s9, s12
-; GFX9-NEXT:    s_xor_b32 s9, s9, s12
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX9-NEXT:    s_sub_i32 s14, 0, s3
 ; GFX9-NEXT:    s_ashr_i32 s8, s4, 31
+; GFX9-NEXT:    s_add_i32 s4, s4, s8
+; GFX9-NEXT:    s_xor_b32 s2, s8, s2
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s4, s4, s8
+; GFX9-NEXT:    s_sub_i32 s8, 0, s3
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s12, v0
+; GFX9-NEXT:    s_mul_i32 s8, s8, s12
+; GFX9-NEXT:    s_mul_hi_u32 s8, s12, s8
+; GFX9-NEXT:    s_add_i32 s12, s12, s8
+; GFX9-NEXT:    s_mul_hi_u32 s8, s4, s12
+; GFX9-NEXT:    s_mul_i32 s12, s8, s3
+; GFX9-NEXT:    s_sub_i32 s4, s4, s12
+; GFX9-NEXT:    s_add_i32 s13, s8, 1
+; GFX9-NEXT:    s_sub_i32 s12, s4, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s3
+; GFX9-NEXT:    s_cselect_b32 s8, s13, s8
+; GFX9-NEXT:    s_cselect_b32 s4, s12, s4
+; GFX9-NEXT:    s_add_i32 s12, s8, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s3
+; GFX9-NEXT:    s_cselect_b32 s3, s12, s8
+; GFX9-NEXT:    s_ashr_i32 s4, s9, 31
+; GFX9-NEXT:    s_add_i32 s8, s9, s4
+; GFX9-NEXT:    s_xor_b32 s8, s8, s4
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s8
+; GFX9-NEXT:    s_ashr_i32 s9, s5, 31
+; GFX9-NEXT:    s_xor_b32 s3, s3, s2
+; GFX9-NEXT:    s_add_i32 s5, s5, s9
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s4, s9, s4
+; GFX9-NEXT:    s_sub_i32 s2, s3, s2
+; GFX9-NEXT:    s_xor_b32 s3, s5, s9
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_sub_i32 s5, 0, s8
+; GFX9-NEXT:    v_readfirstlane_b32 s9, v0
+; GFX9-NEXT:    s_mul_i32 s5, s5, s9
+; GFX9-NEXT:    s_mul_hi_u32 s5, s9, s5
+; GFX9-NEXT:    s_add_i32 s9, s9, s5
+; GFX9-NEXT:    s_mul_hi_u32 s5, s3, s9
+; GFX9-NEXT:    s_mul_i32 s9, s5, s8
+; GFX9-NEXT:    s_sub_i32 s3, s3, s9
+; GFX9-NEXT:    s_add_i32 s12, s5, 1
+; GFX9-NEXT:    s_sub_i32 s9, s3, s8
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s8
+; GFX9-NEXT:    s_cselect_b32 s5, s12, s5
+; GFX9-NEXT:    s_cselect_b32 s3, s9, s3
+; GFX9-NEXT:    s_add_i32 s9, s5, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s8
+; GFX9-NEXT:    s_cselect_b32 s3, s9, s5
+; GFX9-NEXT:    s_ashr_i32 s5, s10, 31
+; GFX9-NEXT:    s_add_i32 s8, s10, s5
+; GFX9-NEXT:    s_xor_b32 s8, s8, s5
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s8
+; GFX9-NEXT:    s_ashr_i32 s9, s6, 31
+; GFX9-NEXT:    s_xor_b32 s3, s3, s4
+; GFX9-NEXT:    s_add_i32 s6, s6, s9
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s5, s9, s5
+; GFX9-NEXT:    s_sub_i32 s3, s3, s4
+; GFX9-NEXT:    s_xor_b32 s4, s6, s9
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_sub_i32 s6, 0, s8
+; GFX9-NEXT:    v_readfirstlane_b32 s9, v0
+; GFX9-NEXT:    s_mul_i32 s6, s6, s9
+; GFX9-NEXT:    s_mul_hi_u32 s6, s9, s6
+; GFX9-NEXT:    s_add_i32 s9, s9, s6
+; GFX9-NEXT:    s_mul_hi_u32 s6, s4, s9
+; GFX9-NEXT:    s_mul_i32 s9, s6, s8
+; GFX9-NEXT:    s_sub_i32 s4, s4, s9
+; GFX9-NEXT:    s_add_i32 s10, s6, 1
+; GFX9-NEXT:    s_sub_i32 s9, s4, s8
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s8
+; GFX9-NEXT:    s_cselect_b32 s6, s10, s6
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s4
+; GFX9-NEXT:    s_add_i32 s9, s6, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s8
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s6
+; GFX9-NEXT:    s_ashr_i32 s6, s11, 31
+; GFX9-NEXT:    s_add_i32 s8, s11, s6
+; GFX9-NEXT:    s_xor_b32 s8, s8, s6
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s8
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    s_ashr_i32 s2, s7, 31
+; GFX9-NEXT:    s_xor_b32 s4, s4, s5
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX9-NEXT:    s_add_i32 s4, s4, s8
-; GFX9-NEXT:    s_xor_b32 s4, s4, s8
-; GFX9-NEXT:    v_mul_lo_u32 v2, s14, v0
+; GFX9-NEXT:    s_add_i32 s7, s7, s2
+; GFX9-NEXT:    s_xor_b32 s6, s2, s6
+; GFX9-NEXT:    s_sub_i32 s4, s4, s5
 ; GFX9-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    s_sub_i32 s14, 0, s9
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX9-NEXT:    s_ashr_i32 s13, s5, 31
-; GFX9-NEXT:    v_mul_lo_u32 v3, s14, v1
-; GFX9-NEXT:    s_add_i32 s5, s5, s13
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v2, v1, v3
-; GFX9-NEXT:    s_xor_b32 s5, s5, s13
-; GFX9-NEXT:    s_xor_b32 s2, s8, s2
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, 1, v0
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_sub_u32_e32 v3, s4, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v2
-; GFX9-NEXT:    s_ashr_i32 s3, s10, 31
-; GFX9-NEXT:    s_add_i32 s4, s10, s3
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    s_xor_b32 s4, s4, s3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v3, s4
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, s9
-; GFX9-NEXT:    v_add_u32_e32 v5, 1, v1
-; GFX9-NEXT:    s_ashr_i32 s8, s11, 31
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX9-NEXT:    v_sub_u32_e32 v2, s5, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; GFX9-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s9, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX9-NEXT:    s_sub_i32 s5, 0, s4
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v2
-; GFX9-NEXT:    v_mul_lo_u32 v2, s5, v3
-; GFX9-NEXT:    s_add_i32 s9, s11, s8
-; GFX9-NEXT:    v_add_u32_e32 v5, 1, v1
-; GFX9-NEXT:    s_xor_b32 s9, s9, s8
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
-; GFX9-NEXT:    v_mul_hi_u32 v2, v3, v2
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v5, s9
-; GFX9-NEXT:    s_ashr_i32 s5, s6, 31
-; GFX9-NEXT:    s_add_i32 s6, s6, s5
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v5
-; GFX9-NEXT:    s_xor_b32 s6, s6, s5
-; GFX9-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX9-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX9-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s2, v0
-; GFX9-NEXT:    s_xor_b32 s2, s13, s12
-; GFX9-NEXT:    v_mul_lo_u32 v5, v2, s4
-; GFX9-NEXT:    v_xor_b32_e32 v1, s2, v1
-; GFX9-NEXT:    v_subrev_u32_e32 v1, s2, v1
-; GFX9-NEXT:    s_xor_b32 s2, s5, s3
-; GFX9-NEXT:    s_sub_i32 s3, 0, s9
-; GFX9-NEXT:    v_mul_lo_u32 v7, s3, v3
-; GFX9-NEXT:    v_sub_u32_e32 v5, s6, v5
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s4, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX9-NEXT:    v_mul_hi_u32 v6, v3, v7
-; GFX9-NEXT:    s_ashr_i32 s3, s7, 31
-; GFX9-NEXT:    s_add_i32 s5, s7, s3
-; GFX9-NEXT:    s_xor_b32 s5, s5, s3
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v6
-; GFX9-NEXT:    v_mul_hi_u32 v3, s5, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v5
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, v3, s9
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v3
-; GFX9-NEXT:    v_xor_b32_e32 v2, s2, v2
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s2, v2
-; GFX9-NEXT:    v_sub_u32_e32 v5, s5, v5
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v5
-; GFX9-NEXT:    s_xor_b32 s2, s3, s8
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v3, s2, v3
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s2, v3
+; GFX9-NEXT:    s_xor_b32 s2, s7, s2
+; GFX9-NEXT:    s_sub_i32 s5, 0, s8
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v1
+; GFX9-NEXT:    s_mul_i32 s5, s5, s7
+; GFX9-NEXT:    s_mul_hi_u32 s5, s7, s5
+; GFX9-NEXT:    s_add_i32 s7, s7, s5
+; GFX9-NEXT:    s_mul_hi_u32 s5, s2, s7
+; GFX9-NEXT:    s_mul_i32 s7, s5, s8
+; GFX9-NEXT:    s_sub_i32 s2, s2, s7
+; GFX9-NEXT:    s_add_i32 s9, s5, 1
+; GFX9-NEXT:    s_sub_i32 s7, s2, s8
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX9-NEXT:    s_cselect_b32 s5, s9, s5
+; GFX9-NEXT:    s_cselect_b32 s2, s7, s2
+; GFX9-NEXT:    s_add_i32 s7, s5, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX9-NEXT:    s_cselect_b32 s2, s7, s5
+; GFX9-NEXT:    s_xor_b32 s2, s2, s6
+; GFX9-NEXT:    s_sub_i32 s2, s2, s6
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = sdiv <4 x i32> %x, %y
   store <4 x i32> %r, <4 x i32> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @srem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x, <4 x i32> %y) {

 ; CHECK-NEXT:    [[TMP152:%.*]] = insertelement <4 x i32> [[TMP114]], i32 [[TMP151]], i64 3
 ; CHECK-NEXT:    store <4 x i32> [[TMP152]], <4 x i32> addrspace(1)* [[OUT:%.*]], align 16
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: srem_v4i32:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_ashr_i32 s2, s8, 31
-; GFX6-NEXT:    s_add_i32 s8, s8, s2
-; GFX6-NEXT:    s_xor_b32 s8, s8, s2
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s8
-; GFX6-NEXT:    s_ashr_i32 s13, s9, 31
-; GFX6-NEXT:    s_add_i32 s9, s9, s13
-; GFX6-NEXT:    s_xor_b32 s9, s9, s13
+; GFX6-NEXT:    s_add_i32 s3, s8, s2
+; GFX6-NEXT:    s_xor_b32 s2, s3, s2
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s2
+; GFX6-NEXT:    s_sub_i32 s3, 0, s2
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    s_sub_i32 s14, 0, s8
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX6-NEXT:    s_ashr_i32 s12, s4, 31
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    s_add_i32 s4, s4, s12
-; GFX6-NEXT:    s_xor_b32 s4, s4, s12
-; GFX6-NEXT:    v_mul_lo_u32 v2, s14, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    s_sub_i32 s14, 0, s9
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    s_ashr_i32 s13, s5, 31
-; GFX6-NEXT:    s_add_i32 s5, s5, s13
-; GFX6-NEXT:    s_xor_b32 s5, s5, s13
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
+; GFX6-NEXT:    s_ashr_i32 s3, s4, 31
+; GFX6-NEXT:    s_add_i32 s4, s4, s3
+; GFX6-NEXT:    s_xor_b32 s4, s4, s3
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_mul_lo_u32 v2, s14, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX6-NEXT:    s_mul_i32 s8, s8, s2
+; GFX6-NEXT:    s_sub_i32 s4, s4, s8
+; GFX6-NEXT:    s_sub_i32 s8, s4, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX6-NEXT:    s_cselect_b32 s4, s8, s4
+; GFX6-NEXT:    s_sub_i32 s8, s4, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX6-NEXT:    s_cselect_b32 s2, s8, s4
+; GFX6-NEXT:    s_ashr_i32 s4, s9, 31
+; GFX6-NEXT:    s_add_i32 s8, s9, s4
+; GFX6-NEXT:    s_xor_b32 s4, s8, s4
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s4
+; GFX6-NEXT:    s_sub_i32 s8, 0, s4
+; GFX6-NEXT:    s_xor_b32 s2, s2, s3
+; GFX6-NEXT:    s_sub_i32 s9, s2, s3
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s8, v0
+; GFX6-NEXT:    s_ashr_i32 s8, s5, 31
+; GFX6-NEXT:    s_add_i32 s5, s5, s8
+; GFX6-NEXT:    s_xor_b32 s5, s5, s8
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s4
+; GFX6-NEXT:    s_sub_i32 s2, s5, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s4
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s4
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_ashr_i32 s3, s10, 31
+; GFX6-NEXT:    s_add_i32 s4, s10, s3
+; GFX6-NEXT:    s_xor_b32 s3, s4, s3
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s3
+; GFX6-NEXT:    s_sub_i32 s4, 0, s3
+; GFX6-NEXT:    s_xor_b32 s2, s2, s8
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
+; GFX6-NEXT:    s_ashr_i32 s4, s6, 31
+; GFX6-NEXT:    s_add_i32 s5, s6, s4
+; GFX6-NEXT:    s_xor_b32 s5, s5, s4
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    s_sub_i32 s6, s2, s8
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s3
+; GFX6-NEXT:    s_sub_i32 s2, s5, s2
+; GFX6-NEXT:    s_sub_i32 s5, s2, s3
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX6-NEXT:    s_cselect_b32 s2, s5, s2
+; GFX6-NEXT:    s_sub_i32 s5, s2, s3
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX6-NEXT:    s_cselect_b32 s5, s5, s2
+; GFX6-NEXT:    s_ashr_i32 s2, s11, 31
+; GFX6-NEXT:    s_add_i32 s3, s11, s2
+; GFX6-NEXT:    s_xor_b32 s8, s3, s2
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s8
+; GFX6-NEXT:    s_sub_i32 s10, 0, s8
+; GFX6-NEXT:    s_xor_b32 s5, s5, s4
+; GFX6-NEXT:    s_sub_i32 s4, s5, s4
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s8
+; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v0
+; GFX6-NEXT:    v_mov_b32_e32 v0, s9
+; GFX6-NEXT:    s_ashr_i32 s9, s7, 31
+; GFX6-NEXT:    s_add_i32 s7, s7, s9
+; GFX6-NEXT:    v_mul_lo_u32 v2, s10, v1
+; GFX6-NEXT:    s_xor_b32 s7, s7, s9
 ; GFX6-NEXT:    v_mul_hi_u32 v2, v1, v2
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GFX6-NEXT:    s_ashr_i32 s4, s10, 31
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    s_add_i32 s8, s10, s4
-; GFX6-NEXT:    s_xor_b32 s4, s8, s4
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s4
-; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v0, s12, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s9
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s12, v0
-; GFX6-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s5, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s9, v1
-; GFX6-NEXT:    s_sub_i32 s5, 0, s4
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_mul_lo_u32 v4, s5, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s9, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX6-NEXT:    v_mul_hi_u32 v3, v2, v4
-; GFX6-NEXT:    s_ashr_i32 s8, s11, 31
-; GFX6-NEXT:    s_add_i32 s9, s11, s8
-; GFX6-NEXT:    s_ashr_i32 s5, s6, 31
-; GFX6-NEXT:    s_xor_b32 s8, s9, s8
-; GFX6-NEXT:    s_add_i32 s6, s6, s5
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v3, s8
-; GFX6-NEXT:    s_xor_b32 s6, s6, s5
-; GFX6-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX6-NEXT:    v_xor_b32_e32 v1, s13, v1
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s13, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s4
-; GFX6-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s6, v2
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s4, v2
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v2
-; GFX6-NEXT:    s_sub_i32 s6, 0, s8
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, s6, v3
-; GFX6-NEXT:    s_ashr_i32 s6, s7, 31
-; GFX6-NEXT:    s_add_i32 s7, s7, s6
-; GFX6-NEXT:    s_xor_b32 s7, s7, s6
-; GFX6-NEXT:    v_mul_hi_u32 v4, v3, v4
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s4, v2
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
-; GFX6-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v2, s5, v2
-; GFX6-NEXT:    v_mul_lo_u32 v3, v3, s8
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s5, v2
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s7, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s8, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s8, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v3, s6, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s6, v3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GFX6-NEXT:    v_mul_hi_u32 v2, s7, v1
+; GFX6-NEXT:    v_mov_b32_e32 v1, s6
+; GFX6-NEXT:    v_readfirstlane_b32 s5, v2
+; GFX6-NEXT:    s_mul_i32 s5, s5, s8
+; GFX6-NEXT:    s_sub_i32 s5, s7, s5
+; GFX6-NEXT:    s_sub_i32 s6, s5, s8
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s8
+; GFX6-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX6-NEXT:    s_sub_i32 s6, s5, s8
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s8
+; GFX6-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX6-NEXT:    s_xor_b32 s5, s5, s9
+; GFX6-NEXT:    s_sub_i32 s5, s5, s9
+; GFX6-NEXT:    v_mov_b32_e32 v2, s4
+; GFX6-NEXT:    v_mov_b32_e32 v3, s5
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_v4i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24

 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_sext_i32_i16 s8, s6
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GFX6-NEXT:    s_sext_i32_i16 s9, s4
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s9
 ; GFX6-NEXT:    s_xor_b32 s8, s9, s8
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GFX6-NEXT:    s_ashr_i32 s6, s6, 16
 ; GFX6-NEXT:    s_ashr_i32 s8, s8, 30
-; GFX6-NEXT:    s_or_b32 s8, s8, 1
+; GFX6-NEXT:    s_or_b32 s10, s8, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
-; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s6
-; GFX6-NEXT:    v_mov_b32_e32 v3, s8
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s8, s10, 0
+; GFX6-NEXT:    s_ashr_i32 s6, s6, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 16
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v1
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s4
+; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v0
 ; GFX6-NEXT:    s_xor_b32 s4, s4, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
-; GFX6-NEXT:    v_mul_f32_e32 v3, v2, v3
+; GFX6-NEXT:    s_sext_i32_i16 s6, s7
+; GFX6-NEXT:    v_mul_f32_e32 v3, v1, v3
 ; GFX6-NEXT:    v_trunc_f32_e32 v3, v3
-; GFX6-NEXT:    v_mad_f32 v2, -v3, v1, v2
-; GFX6-NEXT:    v_mov_b32_e32 v4, s4
-; GFX6-NEXT:    s_sext_i32_i16 s4, s7
+; GFX6-NEXT:    v_mad_f32 v1, -v3, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v3, v3
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v1|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
-; GFX6-NEXT:    s_sext_i32_i16 s6, s5
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v1, v3
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s6
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v2
-; GFX6-NEXT:    s_xor_b32 s4, s6, s4
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, s8, v2
+; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s6
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, s4, v3
+; GFX6-NEXT:    s_sext_i32_i16 s4, s5
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s4
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v0
+; GFX6-NEXT:    s_xor_b32 s4, s4, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
 ; GFX6-NEXT:    s_or_b32 s4, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v4, v1, v4
 ; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
-; GFX6-NEXT:    v_mad_f32 v1, -v4, v2, v1
-; GFX6-NEXT:    v_mov_b32_e32 v5, s4
-; GFX6-NEXT:    s_ashr_i32 s4, s7, 16
+; GFX6-NEXT:    v_mad_f32 v1, -v4, v0, v1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v2|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, 0, v5, vcc
-; GFX6-NEXT:    s_ashr_i32 s5, s5, 16
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v4, v1
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s5
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v2
-; GFX6-NEXT:    s_xor_b32 s4, s5, s4
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    s_ashr_i32 s6, s7, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s6
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, s4, v4
+; GFX6-NEXT:    s_ashr_i32 s4, s5, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s4
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v0
+; GFX6-NEXT:    s_xor_b32 s4, s4, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v5, v4, v5
 ; GFX6-NEXT:    v_trunc_f32_e32 v5, v5
-; GFX6-NEXT:    v_mad_f32 v4, -v5, v2, v4
+; GFX6-NEXT:    v_mad_f32 v4, -v5, v0, v4
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v5, v5
-; GFX6-NEXT:    v_mov_b32_e32 v6, s4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v4|, |v2|
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v6, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
-; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v4|, |v0|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v5
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX6-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; GFX6-NEXT:    v_or_b32_e32 v1, v1, v2
-; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v3
-; GFX6-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX6-NEXT:    v_or_b32_e32 v0, v0, v2
+; GFX6-NEXT:    v_or_b32_e32 v1, v1, v0
+; GFX6-NEXT:    v_lshlrev_b32_e32 v0, 16, v3
+; GFX6-NEXT:    v_and_b32_e32 v2, 0xffff, v2
+; GFX6-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_v4i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0

 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_sext_i32_i16 s8, s6
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GFX6-NEXT:    s_sext_i32_i16 s9, s4
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s9
 ; GFX6-NEXT:    s_xor_b32 s8, s9, s8
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s8, s8, 30
-; GFX6-NEXT:    s_or_b32 s8, s8, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s8
+; GFX6-NEXT:    s_or_b32 s10, s8, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    s_ashr_i32 s9, s6, 16
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s8, s10, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s8, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s9
-; GFX6-NEXT:    s_lshr_b32 s8, s4, 16
-; GFX6-NEXT:    s_lshr_b32 s6, s6, 16
+; GFX6-NEXT:    s_ashr_i32 s8, s6, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s8
+; GFX6-NEXT:    s_lshr_b32 s10, s4, 16
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 16
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v1
-; GFX6-NEXT:    s_xor_b32 s4, s4, s9
+; GFX6-NEXT:    s_xor_b32 s4, s4, s8
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    s_lshr_b32 s6, s6, 16
 ; GFX6-NEXT:    v_mul_f32_e32 v3, v2, v3
 ; GFX6-NEXT:    v_trunc_f32_e32 v3, v3
 ; GFX6-NEXT:    v_mad_f32 v2, -v3, v1, v2
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v3, v3
-; GFX6-NEXT:    v_mov_b32_e32 v4, s4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v1|
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v2|, |v1|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, s4, v3
 ; GFX6-NEXT:    s_sext_i32_i16 s4, s7
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s6
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
 ; GFX6-NEXT:    s_sext_i32_i16 s6, s5
 ; GFX6-NEXT:    s_xor_b32 s4, s6, s4
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s8, v1
+; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s10, v1
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s6
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
 ; GFX6-NEXT:    s_or_b32 s4, s4, 1
-; GFX6-NEXT:    v_mov_b32_e32 v5, s4
+; GFX6-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v4, v1, v4
 ; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
 ; GFX6-NEXT:    v_mad_f32 v1, -v4, v2, v1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v2|
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX6-NEXT:    s_ashr_i32 s4, s7, 16
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v2|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, 0, v5, vcc
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    s_ashr_i32 s6, s7, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s6
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, s4, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s7
-; GFX6-NEXT:    s_lshr_b32 s6, s7, 16
+; GFX6-NEXT:    s_lshr_b32 s8, s7, 16
 ; GFX6-NEXT:    s_ashr_i32 s7, s5, 16
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s7
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v2
-; GFX6-NEXT:    s_xor_b32 s4, s7, s4
-; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    s_xor_b32 s6, s7, s6
+; GFX6-NEXT:    s_ashr_i32 s6, s6, 30
+; GFX6-NEXT:    s_lshr_b32 s4, s5, 16
 ; GFX6-NEXT:    v_mul_f32_e32 v5, v4, v5
 ; GFX6-NEXT:    v_trunc_f32_e32 v5, v5
 ; GFX6-NEXT:    v_mad_f32 v4, -v5, v2, v4
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v5, v5
-; GFX6-NEXT:    v_mov_b32_e32 v6, s4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v4|, |v2|
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v6, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
-; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s6
-; GFX6-NEXT:    s_lshr_b32 s4, s5, 16
+; GFX6-NEXT:    s_or_b32 s9, s6, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v4|, |v2|
+; GFX6-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GFX6-NEXT:    s_cselect_b32 s6, s9, 0
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, s6, v5
+; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s8
 ; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s5, v1
+; GFX6-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s4, v2
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
-; GFX6-NEXT:    v_and_b32_e32 v1, 0xffff, v1
 ; GFX6-NEXT:    v_or_b32_e32 v1, v1, v2
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v3
-; GFX6-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX6-NEXT:    v_or_b32_e32 v0, v0, v2
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_v4i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24

 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s5, s4, 0x30008
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s5
 ; GFX6-NEXT:    s_bfe_i32 s4, s4, 0x30000
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GFX6-NEXT:    s_xor_b32 s4, s4, s5
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s4
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GFX6-NEXT:    v_and_b32_e32 v0, 7, v0
 ; GFX6-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i3:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24

 ;
 ; GFX6-LABEL: srem_i3:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_bfe_i32 s2, s4, 0x30008
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s2
-; GFX6-NEXT:    s_bfe_i32 s5, s4, 0x30000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s5
-; GFX6-NEXT:    s_xor_b32 s2, s5, s2
+; GFX6-NEXT:    s_bfe_i32 s3, s4, 0x30000
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s3
+; GFX6-NEXT:    s_xor_b32 s2, s3, s2
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s2, s2, 30
-; GFX6-NEXT:    s_or_b32 s2, s2, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s2
+; GFX6-NEXT:    s_lshr_b32 s5, s4, 8
+; GFX6-NEXT:    s_or_b32 s6, s2, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    s_lshr_b32 s3, s4, 8
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s3
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[2:3], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[2:3], s[2:3], exec
+; GFX6-NEXT:    s_cselect_b32 s2, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s2, v2
+; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s5
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    v_and_b32_e32 v0, 7, v0
 ; GFX6-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_i3:

 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_sext_i32_i16 s8, s6
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GFX6-NEXT:    s_sext_i32_i16 s9, s4
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s9
 ; GFX6-NEXT:    s_xor_b32 s8, s9, s8
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GFX6-NEXT:    s_ashr_i32 s6, s6, 16
 ; GFX6-NEXT:    s_ashr_i32 s8, s8, 30
-; GFX6-NEXT:    s_or_b32 s8, s8, 1
+; GFX6-NEXT:    s_or_b32 s10, s8, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s8, s10, 0
+; GFX6-NEXT:    s_ashr_i32 s6, s6, 16
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s6
-; GFX6-NEXT:    v_mov_b32_e32 v3, s8
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 16
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, s8, v2
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v0
 ; GFX6-NEXT:    s_xor_b32 s4, s4, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    s_sext_i32_i16 s6, s7
 ; GFX6-NEXT:    v_mul_f32_e32 v3, v2, v3
 ; GFX6-NEXT:    v_trunc_f32_e32 v3, v3
-; GFX6-NEXT:    v_mad_f32 v2, -v3, v1, v2
-; GFX6-NEXT:    v_mov_b32_e32 v4, s4
-; GFX6-NEXT:    s_sext_i32_i16 s4, s7
+; GFX6-NEXT:    v_mad_f32 v2, -v3, v0, v2
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v3, v3
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v1|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
-; GFX6-NEXT:    s_sext_i32_i16 s5, s5
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s5
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v2
-; GFX6-NEXT:    s_xor_b32 s4, s5, s4
-; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
 ; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v2|, |v0|
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s6
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, s4, v3
+; GFX6-NEXT:    s_sext_i32_i16 s4, s5
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s4
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v0
+; GFX6-NEXT:    s_xor_b32 s4, s4, s6
+; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v4, v3, v4
 ; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
-; GFX6-NEXT:    v_mad_f32 v3, -v4, v2, v3
+; GFX6-NEXT:    v_mad_f32 v3, -v4, v0, v3
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX6-NEXT:    v_mov_b32_e32 v5, s4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
-; GFX6-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
-; GFX6-NEXT:    v_and_b32_e32 v0, 0xffff, v0
-; GFX6-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX6-NEXT:    buffer_store_short v2, off, s[0:3], 0 offset:4
-; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v3|, |v0|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v4
+; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 16, v2
+; GFX6-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; GFX6-NEXT:    v_or_b32_e32 v1, v1, v2
+; GFX6-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:4
+; GFX6-NEXT:    buffer_store_dword v1, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_v3i16:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_sext_i32_i16 s8, s6
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GFX6-NEXT:    s_sext_i32_i16 s9, s4
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s9
 ; GFX6-NEXT:    s_xor_b32 s8, s9, s8
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GFX6-NEXT:    s_ashr_i32 s8, s8, 30
-; GFX6-NEXT:    s_or_b32 s8, s8, 1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s8
+; GFX6-NEXT:    s_or_b32 s10, s8, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GFX6-NEXT:    s_ashr_i32 s9, s6, 16
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v1|, |v0|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s8, s10, 0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s8, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s9
-; GFX6-NEXT:    s_lshr_b32 s8, s4, 16
-; GFX6-NEXT:    s_lshr_b32 s6, s6, 16
+; GFX6-NEXT:    s_ashr_i32 s8, s6, 16
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v1, s8
+; GFX6-NEXT:    s_lshr_b32 s10, s4, 16
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 16
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v1
-; GFX6-NEXT:    s_xor_b32 s4, s4, s9
+; GFX6-NEXT:    s_xor_b32 s4, s4, s8
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    s_lshr_b32 s6, s6, 16
 ; GFX6-NEXT:    v_mul_f32_e32 v3, v2, v3
 ; GFX6-NEXT:    v_trunc_f32_e32 v3, v3
 ; GFX6-NEXT:    v_mad_f32 v2, -v3, v1, v2
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v3, v3
-; GFX6-NEXT:    v_mov_b32_e32 v4, s4
+; GFX6-NEXT:    s_or_b32 s4, s4, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v2|, |v1|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, s4, v3
 ; GFX6-NEXT:    s_sext_i32_i16 s4, s7
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v1|
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s6
 ; GFX6-NEXT:    s_sext_i32_i16 s6, s5
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s6
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; GFX6-NEXT:    s_xor_b32 s4, s6, s4
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
 ; GFX6-NEXT:    s_or_b32 s4, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v4, v3, v4
 ; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
 ; GFX6-NEXT:    v_mad_f32 v3, -v4, v2, v3
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX6-NEXT:    v_mov_b32_e32 v5, s4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v3|, |v2|
+; GFX6-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s4, 0
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, s4, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s7
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s8, v1
+; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s10, v1
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s5, v2
 ; GFX6-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX6-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX6-NEXT:    buffer_store_short v2, off, s[0:3], 0 offset:4
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;

 ; GFX6-NEXT:    s_lshr_b32 s5, s8, 15
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s5
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v3, v2
 ; GFX6-NEXT:    s_mov_b32 s0, s4
 ; GFX6-NEXT:    s_lshr_b32 s4, s6, 15
 ; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s4, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_and_b32_e32 v3, 0x7fff, v3
 ; GFX6-NEXT:    v_lshl_b64 v[0:1], v[0:1], 30
 ; GFX6-NEXT:    v_and_b32_e32 v2, 0x7fff, v6
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 15, v3
 ; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX6-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)

 ; GFX6-NEXT:    s_mov_b32 s0, s4
 ; GFX6-NEXT:    s_bfe_i32 s4, s8, 0xf0000
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s4
 ; GFX6-NEXT:    s_mov_b32 s1, s5
 ; GFX6-NEXT:    s_bfe_i32 s5, s6, 0xf0000
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s5
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; GFX6-NEXT:    s_xor_b32 s4, s5, s4
+; GFX6-NEXT:    v_mov_b32_e32 v0, s6
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
 ; GFX6-NEXT:    v_mul_f32_e32 v4, v3, v4
 ; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
 ; GFX6-NEXT:    v_mad_f32 v3, -v4, v2, v3
-; GFX6-NEXT:    v_mov_b32_e32 v5, s4
-; GFX6-NEXT:    s_bfe_i32 s4, s8, 0xf000f
+; GFX6-NEXT:    v_alignbit_b32 v0, s7, v0, 30
+; GFX6-NEXT:    s_or_b32 s7, s4, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v3|, |v2|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
-; GFX6-NEXT:    s_bfe_i32 s5, s6, 0xf000f
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s5
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v3
+; GFX6-NEXT:    s_cselect_b32 s4, s7, 0
+; GFX6-NEXT:    s_bfe_i32 s5, s8, 0xf000f
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, s5
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, s4, v4
+; GFX6-NEXT:    s_bfe_i32 s4, s6, 0xf000f
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s4
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v2
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s8
 ; GFX6-NEXT:    v_alignbit_b32 v1, s9, v1, 30
-; GFX6-NEXT:    s_xor_b32 s4, s5, s4
+; GFX6-NEXT:    s_xor_b32 s4, s4, s5
 ; GFX6-NEXT:    v_mul_f32_e32 v5, v4, v5
 ; GFX6-NEXT:    v_trunc_f32_e32 v5, v5
-; GFX6-NEXT:    v_mad_f32 v4, -v5, v3, v4
-; GFX6-NEXT:    v_bfe_i32 v1, v1, 0, 15
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
+; GFX6-NEXT:    v_mad_f32 v4, -v5, v2, v4
+; GFX6-NEXT:    v_bfe_i32 v1, v1, 0, 15
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v5, v5
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v4|, |v3|
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, v1
-; GFX6-NEXT:    v_mov_b32_e32 v0, s6
-; GFX6-NEXT:    s_or_b32 s4, s4, 1
-; GFX6-NEXT:    v_alignbit_b32 v0, s7, v0, 30
-; GFX6-NEXT:    v_mov_b32_e32 v6, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, 0, v6, vcc
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v4|, |v2|
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v2, v1
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
 ; GFX6-NEXT:    v_bfe_i32 v0, v0, 0, 15
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, s4, v5
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v6, v4
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v6, v2
 ; GFX6-NEXT:    v_xor_b32_e32 v0, v0, v1
 ; GFX6-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
 ; GFX6-NEXT:    v_or_b32_e32 v0, 1, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v1, v5, v6
 ; GFX6-NEXT:    v_trunc_f32_e32 v1, v1
-; GFX6-NEXT:    v_mad_f32 v5, -v1, v4, v5
+; GFX6-NEXT:    v_mad_f32 v5, -v1, v2, v5
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v5|, |v4|
+; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v5|, |v2|
 ; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX6-NEXT:    v_and_b32_e32 v3, 0x7fff, v3
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_and_b32_e32 v2, 0x7fff, v3
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_and_b32_e32 v3, 0x7fff, v4
 ; GFX6-NEXT:    v_lshl_b64 v[0:1], v[0:1], 30
-; GFX6-NEXT:    v_and_b32_e32 v2, 0x7fff, v2
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v3, 15, v3
 ; GFX6-NEXT:    v_or_b32_e32 v2, v2, v3
 ; GFX6-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
 ; GFX6-NEXT:    v_and_b32_e32 v0, 0x1fff, v1
 ; GFX6-NEXT:    buffer_store_short v0, off, s[0:3], 0 offset:4
 ; GFX6-NEXT:    s_endpgm

 ;
 ; GFX6-LABEL: srem_v3i15:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GFX6-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_mov_b32 s0, s4
+; GFX6-NEXT:    s_bfe_i32 s4, s8, 0xf0000
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s4
 ; GFX6-NEXT:    s_mov_b32 s1, s5
-; GFX6-NEXT:    s_bfe_i32 s5, s8, 0xf0000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s5
-; GFX6-NEXT:    v_mov_b32_e32 v2, s8
-; GFX6-NEXT:    v_alignbit_b32 v2, s9, v2, 30
-; GFX6-NEXT:    s_bfe_i32 s9, s6, 0xf0000
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s9
+; GFX6-NEXT:    s_bfe_i32 s5, s6, 0xf0000
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s5
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v6, v4
-; GFX6-NEXT:    s_xor_b32 s5, s9, s5
-; GFX6-NEXT:    s_ashr_i32 s5, s5, 30
-; GFX6-NEXT:    s_or_b32 s5, s5, 1
+; GFX6-NEXT:    s_xor_b32 s4, s5, s4
+; GFX6-NEXT:    v_mov_b32_e32 v0, s6
+; GFX6-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX6-NEXT:    v_mul_f32_e32 v6, v5, v6
 ; GFX6-NEXT:    v_trunc_f32_e32 v6, v6
+; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
 ; GFX6-NEXT:    v_mad_f32 v5, -v6, v4, v5
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v6, v6
-; GFX6-NEXT:    v_mov_b32_e32 v7, s5
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v5|, |v4|
-; GFX6-NEXT:    v_cndmask_b32_e32 v4, 0, v7, vcc
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
+; GFX6-NEXT:    v_alignbit_b32 v0, s7, v0, 30
+; GFX6-NEXT:    s_lshr_b32 s7, s6, 15
+; GFX6-NEXT:    v_alignbit_b32 v2, s9, v2, 30
+; GFX6-NEXT:    s_lshr_b32 s9, s8, 15
+; GFX6-NEXT:    s_or_b32 s10, s4, 1
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v5|, |v4|
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX6-NEXT:    s_cselect_b32 s4, s10, 0
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, s4, v6
+; GFX6-NEXT:    s_bfe_i32 s4, s8, 0xf000f
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s4
+; GFX6-NEXT:    s_bfe_i32 s5, s6, 0xf000f
+; GFX6-NEXT:    v_cvt_f32_i32_e32 v6, s5
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v4, s8
-; GFX6-NEXT:    s_bfe_i32 s5, s8, 0xf000f
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s5
-; GFX6-NEXT:    s_mov_b32 s0, s4
-; GFX6-NEXT:    v_mov_b32_e32 v0, s6
-; GFX6-NEXT:    s_lshr_b32 s4, s6, 15
-; GFX6-NEXT:    v_sub_i32_e32 v4, vcc, s6, v4
-; GFX6-NEXT:    s_bfe_i32 s6, s6, 0xf000f
-; GFX6-NEXT:    v_cvt_f32_i32_e32 v6, s6
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v7, v5
+; GFX6-NEXT:    s_xor_b32 s4, s5, s4
 ; GFX6-NEXT:    v_and_b32_e32 v3, 0x7fff, v2
-; GFX6-NEXT:    s_xor_b32 s5, s6, s5
-; GFX6-NEXT:    v_bfe_i32 v2, v2, 0, 15
+; GFX6-NEXT:    s_ashr_i32 s4, s4, 30
 ; GFX6-NEXT:    v_mul_f32_e32 v7, v6, v7
 ; GFX6-NEXT:    v_trunc_f32_e32 v7, v7
 ; GFX6-NEXT:    v_mad_f32 v6, -v7, v5, v6
-; GFX6-NEXT:    s_ashr_i32 s5, s5, 30
+; GFX6-NEXT:    v_bfe_i32 v2, v2, 0, 15
+; GFX6-NEXT:    v_sub_i32_e32 v4, vcc, s6, v4
+; GFX6-NEXT:    s_or_b32 s6, s4, 1
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v7, v7
-; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v6|, |v5|
+; GFX6-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v6|, |v5|
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v6, v2
-; GFX6-NEXT:    s_or_b32 s5, s5, 1
-; GFX6-NEXT:    v_alignbit_b32 v0, s7, v0, 30
-; GFX6-NEXT:    v_mov_b32_e32 v8, s5
+; GFX6-NEXT:    s_and_b64 s[4:5], s[4:5], exec
 ; GFX6-NEXT:    v_and_b32_e32 v1, 0x7fff, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v5, 0, v8, vcc
+; GFX6-NEXT:    s_cselect_b32 s4, s6, 0
 ; GFX6-NEXT:    v_bfe_i32 v0, v0, 0, 15
-; GFX6-NEXT:    v_add_i32_e32 v5, vcc, v5, v7
+; GFX6-NEXT:    v_add_i32_e32 v5, vcc, s4, v7
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v7, v0
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v8, v6
 ; GFX6-NEXT:    v_xor_b32_e32 v0, v0, v2
 ; GFX6-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
 ; GFX6-NEXT:    v_or_b32_e32 v0, 1, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v2, v7, v8
 ; GFX6-NEXT:    v_trunc_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mad_f32 v7, -v2, v6, v7
 ; GFX6-NEXT:    v_cvt_i32_f32_e32 v2, v2
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v7|, |v6|
-; GFX6-NEXT:    s_lshr_b32 s7, s8, 15
 ; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v5, v5, s7
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_mul_lo_u32 v5, v5, s9
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v0, v0, v3
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s4, v5
+; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s7, v5
 ; GFX6-NEXT:    v_and_b32_e32 v2, 0x7fff, v2
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, v1, v0
 ; GFX6-NEXT:    v_lshl_b64 v[0:1], v[0:1], 30
 ; GFX6-NEXT:    v_and_b32_e32 v3, 0x7fff, v4
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v2, 15, v2
 ; GFX6-NEXT:    v_or_b32_e32 v2, v3, v2
 ; GFX6-NEXT:    v_or_b32_e32 v0, v2, v0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0

 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x9
 ; GFX6-NEXT:    s_mov_b32 s11, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s10, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s2, 0x1000, s6
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s2
-; GFX6-NEXT:    s_lshl_b32 s3, 0x1000, s7
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s3
-; GFX6-NEXT:    s_sub_i32 s0, 0, s2
+; GFX6-NEXT:    s_sub_i32 s3, 0, s2
+; GFX6-NEXT:    s_lshl_b32 s6, 0x1000, s7
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s6
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, s0, v0
-; GFX6-NEXT:    s_sub_i32 s0, 0, s3
-; GFX6-NEXT:    v_mul_lo_u32 v3, s0, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
+; GFX6-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX6-NEXT:    s_mul_i32 s3, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s4, s3
+; GFX6-NEXT:    s_sub_i32 s4, s3, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
+; GFX6-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b64 s[2:3], -1, 0
+; GFX6-NEXT:    s_sub_i32 s4, 0, s6
+; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v1
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[2:3]
 ; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, v0, s2
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v0
-; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s3
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s4, v2
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s2, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s5, v4
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v1
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s3, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX6-NEXT:    v_readfirstlane_b32 s0, v1
+; GFX6-NEXT:    s_mul_i32 s0, s0, s6
+; GFX6-NEXT:    s_sub_i32 s0, s5, s0
+; GFX6-NEXT:    s_sub_i32 s1, s0, s6
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v1
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s6
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX6-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v1
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s6
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: udiv_v2i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_lshl_b32 s6, 0x1000, s6
-; GFX9-NEXT:    s_lshl_b32 s7, 0x1000, s7
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s6
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s7
-; GFX9-NEXT:    s_sub_i32 s2, 0, s6
-; GFX9-NEXT:    s_sub_i32 s3, 0, s7
+; GFX9-NEXT:    s_lshl_b32 s3, 0x1000, s6
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
+; GFX9-NEXT:    s_lshl_b32 s2, 0x1000, s7
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s2
+; GFX9-NEXT:    s_sub_i32 s6, 0, s3
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX9-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    v_mul_lo_u32 v2, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v3, s3, v1
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v2
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v3
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, s6
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, s7
-; GFX9-NEXT:    v_add_u32_e32 v5, 1, v0
-; GFX9-NEXT:    v_add_u32_e32 v6, 1, v1
-; GFX9-NEXT:    v_sub_u32_e32 v3, s4, v3
-; GFX9-NEXT:    v_sub_u32_e32 v4, s5, v4
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s6, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s6, v3
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s7, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v6, s[0:1]
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s7, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    v_add_u32_e32 v5, 1, v0
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, v4, v6, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s6, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s7, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX9-NEXT:    s_mul_i32 s6, s6, s7
+; GFX9-NEXT:    s_mul_hi_u32 s6, s7, s6
+; GFX9-NEXT:    s_add_i32 s7, s7, s6
+; GFX9-NEXT:    s_mul_hi_u32 s6, s4, s7
+; GFX9-NEXT:    s_mul_i32 s7, s6, s3
+; GFX9-NEXT:    s_sub_i32 s4, s4, s7
+; GFX9-NEXT:    s_add_i32 s9, s6, 1
+; GFX9-NEXT:    s_sub_i32 s7, s4, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s3
+; GFX9-NEXT:    s_cselect_b32 s6, s9, s6
+; GFX9-NEXT:    s_cselect_b32 s4, s7, s4
+; GFX9-NEXT:    s_add_i32 s7, s6, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s3
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
+; GFX9-NEXT:    s_cselect_b32 s3, s7, s6
+; GFX9-NEXT:    s_sub_i32 s4, 0, s2
+; GFX9-NEXT:    s_mul_i32 s4, s4, s8
+; GFX9-NEXT:    s_mul_hi_u32 s4, s8, s4
+; GFX9-NEXT:    s_add_i32 s8, s8, s4
+; GFX9-NEXT:    s_mul_hi_u32 s4, s5, s8
+; GFX9-NEXT:    s_mul_i32 s6, s4, s2
+; GFX9-NEXT:    s_sub_i32 s5, s5, s6
+; GFX9-NEXT:    s_add_i32 s7, s4, 1
+; GFX9-NEXT:    s_sub_i32 s6, s5, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX9-NEXT:    s_cselect_b32 s4, s7, s4
+; GFX9-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX9-NEXT:    s_add_i32 s6, s4, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX9-NEXT:    s_cselect_b32 s2, s6, s4
+; GFX9-NEXT:    v_mov_b32_e32 v0, s3
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl <2 x i32> <i32 4096, i32 4096>, %y
   %r = udiv <2 x i32> %x, %shl.y
   store <2 x i32> %r, <2 x i32> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @urem_i32_oddk_denom(i32 addrspace(1)* %out, i32 %x) {

 ; CHECK-NEXT:    [[TMP60:%.*]] = insertelement <2 x i32> [[TMP30]], i32 [[TMP59]], i64 1
 ; CHECK-NEXT:    store <2 x i32> [[TMP60]], <2 x i32> addrspace(1)* [[OUT:%.*]], align 8
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: urem_v2i32_pow2_shl_denom:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_lshl_b32 s6, 0x1000, s6
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s6
-; GFX6-NEXT:    s_lshl_b32 s7, 0x1000, s7
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s7
-; GFX6-NEXT:    s_sub_i32 s2, 0, s6
+; GFX6-NEXT:    s_lshl_b32 s2, 0x1000, s6
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s2
+; GFX6-NEXT:    s_sub_i32 s3, 0, s2
+; GFX6-NEXT:    s_lshl_b32 s6, 0x1000, s7
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s6
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v0
-; GFX6-NEXT:    s_sub_i32 s2, 0, s7
-; GFX6-NEXT:    v_mul_lo_u32 v3, s2, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX6-NEXT:    s_mul_i32 s3, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s4, s3
+; GFX6-NEXT:    s_sub_i32 s4, s3, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX6-NEXT:    s_sub_i32 s4, s3, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b32 s4, s4, s3
+; GFX6-NEXT:    s_sub_i32 s2, 0, s6
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
-; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s7
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s6, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s5, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s6, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s7, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s7, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX6-NEXT:    s_mul_i32 s7, s7, s6
+; GFX6-NEXT:    s_sub_i32 s5, s5, s7
+; GFX6-NEXT:    s_sub_i32 s7, s5, s6
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s6
+; GFX6-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX6-NEXT:    s_sub_i32 s7, s5, s6
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s6
+; GFX6-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
+; GFX6-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: urem_v2i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24

 ; GFX6-NEXT:    s_xor_b32 s3, s3, s8
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s3
 ; GFX6-NEXT:    s_sub_i32 s4, 0, s3
 ; GFX6-NEXT:    s_ashr_i32 s9, s2, 31
 ; GFX6-NEXT:    s_add_i32 s2, s2, s9
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX6-NEXT:    s_xor_b32 s2, s2, s9
 ; GFX6-NEXT:    s_mov_b32 s5, s1
-; GFX6-NEXT:    s_xor_b32 s8, s9, s8
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; GFX6-NEXT:    s_mov_b32 s4, s0
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v1, v0, s3
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s2, v1
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s3, v1
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v0
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s0, v0
+; GFX6-NEXT:    s_mul_i32 s0, s0, s3
+; GFX6-NEXT:    s_sub_i32 s0, s2, s0
+; GFX6-NEXT:    s_sub_i32 s1, s0, s3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    s_xor_b32 s0, s9, s8
+; GFX6-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s3, 0x1000, s3
 ; GFX9-NEXT:    s_ashr_i32 s4, s3, 31
 ; GFX9-NEXT:    s_add_i32 s3, s3, s4
 ; GFX9-NEXT:    s_xor_b32 s3, s3, s4
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX9-NEXT:    s_sub_i32 s5, 0, s3
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, s5, v0
+; GFX9-NEXT:    s_sub_i32 s6, 0, s3
 ; GFX9-NEXT:    s_ashr_i32 s5, s2, 31
 ; GFX9-NEXT:    s_add_i32 s2, s2, s5
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX9-NEXT:    s_xor_b32 s2, s2, s5
-; GFX9-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v0, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    v_sub_u32_e32 v1, s2, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX9-NEXT:    s_xor_b32 s2, s5, s4
-; GFX9-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s2, v0
-; GFX9-NEXT:    global_store_dword v2, v0, s[0:1]
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX9-NEXT:    s_mul_i32 s6, s6, s7
+; GFX9-NEXT:    s_mul_hi_u32 s6, s7, s6
+; GFX9-NEXT:    s_add_i32 s7, s7, s6
+; GFX9-NEXT:    s_mul_hi_u32 s6, s2, s7
+; GFX9-NEXT:    s_mul_i32 s8, s6, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s8
+; GFX9-NEXT:    s_add_i32 s7, s6, 1
+; GFX9-NEXT:    s_sub_i32 s8, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s6, s7, s6
+; GFX9-NEXT:    s_cselect_b32 s2, s8, s2
+; GFX9-NEXT:    s_add_i32 s7, s6, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s7, s6
+; GFX9-NEXT:    s_xor_b32 s3, s5, s4
+; GFX9-NEXT:    s_xor_b32 s2, s2, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl i32 4096, %y
   %r = sdiv i32 %x, %shl.y
   store i32 %r, i32 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @sdiv_v2i32_pow2k_denom(<2 x i32> addrspace(1)* %out, <2 x i32> %x) {

 ; CHECK-NEXT:    [[TMP80:%.*]] = xor i32 [[TMP79]], [[TMP46]]
 ; CHECK-NEXT:    [[TMP81:%.*]] = sub i32 [[TMP80]], [[TMP46]]
 ; CHECK-NEXT:    [[TMP82:%.*]] = insertelement <2 x i32> [[TMP41]], i32 [[TMP81]], i64 1
 ; CHECK-NEXT:    store <2 x i32> [[TMP82]], <2 x i32> addrspace(1)* [[OUT:%.*]], align 8
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: sdiv_v2i32_pow2_shl_denom:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xb
-; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s7, 0xf000
-; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_lshl_b32 s0, 0x1000, s10
-; GFX6-NEXT:    s_ashr_i32 s1, s0, 31
-; GFX6-NEXT:    s_add_i32 s0, s0, s1
-; GFX6-NEXT:    s_xor_b32 s2, s0, s1
+; GFX6-NEXT:    s_lshl_b32 s2, 0x1000, s6
+; GFX6-NEXT:    s_ashr_i32 s3, s2, 31
+; GFX6-NEXT:    s_add_i32 s2, s2, s3
+; GFX6-NEXT:    s_xor_b32 s2, s2, s3
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s2
-; GFX6-NEXT:    s_lshl_b32 s0, 0x1000, s11
-; GFX6-NEXT:    s_ashr_i32 s3, s0, 31
-; GFX6-NEXT:    s_add_i32 s0, s0, s3
+; GFX6-NEXT:    s_sub_i32 s6, 0, s2
+; GFX6-NEXT:    s_lshl_b32 s7, 0x1000, s7
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    s_sub_i32 s11, 0, s2
-; GFX6-NEXT:    s_xor_b32 s10, s0, s3
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s10
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    s_ashr_i32 s0, s8, 31
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    s_add_i32 s8, s8, s0
-; GFX6-NEXT:    v_mul_lo_u32 v2, s11, v0
-; GFX6-NEXT:    s_xor_b32 s8, s8, s0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    s_xor_b32 s11, s0, s1
-; GFX6-NEXT:    s_sub_i32 s0, 0, s10
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v0, s8, v0
-; GFX6-NEXT:    v_mul_lo_u32 v2, s0, v1
-; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s2
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v0
-; GFX6-NEXT:    v_mul_hi_u32 v2, v1, v2
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s8, v3
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s2, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
-; GFX6-NEXT:    s_ashr_i32 s0, s9, 31
-; GFX6-NEXT:    s_add_i32 s1, s9, s0
-; GFX6-NEXT:    s_xor_b32 s1, s1, s0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GFX6-NEXT:    v_mul_hi_u32 v1, s1, v1
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, 1, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s2, v3
-; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s10
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX6-NEXT:    s_xor_b32 s2, s0, s3
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v1
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s1, v2
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s10, v2
-; GFX6-NEXT:    v_xor_b32_e32 v0, s11, v0
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s10, v2
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s11, v0
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, 1, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v1, s2, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s2, v1
-; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s6, v0
+; GFX6-NEXT:    s_ashr_i32 s6, s4, 31
+; GFX6-NEXT:    s_add_i32 s4, s4, s6
+; GFX6-NEXT:    s_xor_b32 s4, s4, s6
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    s_xor_b32 s6, s6, s3
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX6-NEXT:    s_mul_i32 s3, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s4, s3
+; GFX6-NEXT:    s_sub_i32 s4, s3, s2
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_ashr_i32 s4, s7, 31
+; GFX6-NEXT:    s_add_i32 s7, s7, s4
+; GFX6-NEXT:    s_xor_b32 s7, s7, s4
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s7
+; GFX6-NEXT:    s_sub_i32 s8, 0, s7
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    v_xor_b32_e32 v0, s6, v0
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s6, v0
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    v_mul_lo_u32 v3, s8, v2
+; GFX6-NEXT:    s_ashr_i32 s8, s5, 31
+; GFX6-NEXT:    s_add_i32 s5, s5, s8
+; GFX6-NEXT:    s_xor_b32 s5, s5, s8
+; GFX6-NEXT:    v_mul_hi_u32 v1, v2, v3
+; GFX6-NEXT:    s_xor_b32 s4, s8, s4
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s6, v1
+; GFX6-NEXT:    s_mul_i32 s6, s6, s7
+; GFX6-NEXT:    s_sub_i32 s5, s5, s6
+; GFX6-NEXT:    s_sub_i32 s6, s5, s7
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v1
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s7
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX6-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v1
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s7
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX6-NEXT:    v_xor_b32_e32 v1, s4, v1
+; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s4, v1
+; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_v2i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_lshl_b32 s0, 0x1000, s6
-; GFX9-NEXT:    s_ashr_i32 s1, s0, 31
-; GFX9-NEXT:    s_add_i32 s0, s0, s1
-; GFX9-NEXT:    s_xor_b32 s0, s0, s1
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s0
+; GFX9-NEXT:    s_lshl_b32 s2, 0x1000, s6
+; GFX9-NEXT:    s_ashr_i32 s3, s2, 31
+; GFX9-NEXT:    s_add_i32 s2, s2, s3
+; GFX9-NEXT:    s_xor_b32 s2, s2, s3
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s2
 ; GFX9-NEXT:    s_lshl_b32 s6, 0x1000, s7
-; GFX9-NEXT:    s_ashr_i32 s8, s6, 31
-; GFX9-NEXT:    s_add_i32 s6, s6, s8
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX9-NEXT:    s_xor_b32 s6, s6, s8
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s6
-; GFX9-NEXT:    s_sub_i32 s10, 0, s0
-; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
 ; GFX9-NEXT:    s_ashr_i32 s7, s4, 31
 ; GFX9-NEXT:    s_add_i32 s4, s4, s7
-; GFX9-NEXT:    v_mul_lo_u32 v3, s10, v0
-; GFX9-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    s_sub_i32 s10, 0, s6
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v3
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s3, s7, s3
 ; GFX9-NEXT:    s_xor_b32 s4, s4, s7
-; GFX9-NEXT:    v_mul_lo_u32 v4, s10, v1
-; GFX9-NEXT:    s_ashr_i32 s9, s5, 31
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v3
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v4
-; GFX9-NEXT:    s_add_i32 s5, s5, s9
-; GFX9-NEXT:    s_xor_b32 s5, s5, s9
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s0
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v3
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_add_u32_e32 v3, 1, v0
-; GFX9-NEXT:    v_sub_u32_e32 v4, s4, v4
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s0, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s0, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v4, v3, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s0, v3
-; GFX9-NEXT:    v_mul_lo_u32 v3, v1, s6
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v1
-; GFX9-NEXT:    v_sub_u32_e32 v3, s5, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s6, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v4, s6, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s6, v3
-; GFX9-NEXT:    s_xor_b32 s1, s7, s1
-; GFX9-NEXT:    s_xor_b32 s0, s9, s8
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s1, v0
-; GFX9-NEXT:    v_xor_b32_e32 v1, s0, v1
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s1, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v1, s0, v1
-; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[2:3]
+; GFX9-NEXT:    s_sub_i32 s7, 0, s2
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_i32 s7, s7, s8
+; GFX9-NEXT:    s_mul_hi_u32 s7, s8, s7
+; GFX9-NEXT:    s_add_i32 s8, s8, s7
+; GFX9-NEXT:    s_mul_hi_u32 s7, s4, s8
+; GFX9-NEXT:    s_mul_i32 s8, s7, s2
+; GFX9-NEXT:    s_sub_i32 s4, s4, s8
+; GFX9-NEXT:    s_add_i32 s9, s7, 1
+; GFX9-NEXT:    s_sub_i32 s8, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s7, s9, s7
+; GFX9-NEXT:    s_cselect_b32 s4, s8, s4
+; GFX9-NEXT:    s_add_i32 s8, s7, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s2, s8, s7
+; GFX9-NEXT:    s_ashr_i32 s4, s6, 31
+; GFX9-NEXT:    s_add_i32 s6, s6, s4
+; GFX9-NEXT:    s_xor_b32 s6, s6, s4
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s6
+; GFX9-NEXT:    s_ashr_i32 s7, s5, 31
+; GFX9-NEXT:    s_xor_b32 s2, s2, s3
+; GFX9-NEXT:    s_add_i32 s5, s5, s7
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s4, s7, s4
+; GFX9-NEXT:    s_sub_i32 s2, s2, s3
+; GFX9-NEXT:    s_xor_b32 s3, s5, s7
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_sub_i32 s5, 0, s6
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX9-NEXT:    s_mul_i32 s5, s5, s7
+; GFX9-NEXT:    s_mul_hi_u32 s5, s7, s5
+; GFX9-NEXT:    s_add_i32 s7, s7, s5
+; GFX9-NEXT:    s_mul_hi_u32 s5, s3, s7
+; GFX9-NEXT:    s_mul_i32 s7, s5, s6
+; GFX9-NEXT:    s_sub_i32 s3, s3, s7
+; GFX9-NEXT:    s_add_i32 s8, s5, 1
+; GFX9-NEXT:    s_sub_i32 s7, s3, s6
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s6
+; GFX9-NEXT:    s_cselect_b32 s5, s8, s5
+; GFX9-NEXT:    s_cselect_b32 s3, s7, s3
+; GFX9-NEXT:    s_add_i32 s7, s5, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s6
+; GFX9-NEXT:    s_cselect_b32 s3, s7, s5
+; GFX9-NEXT:    s_xor_b32 s3, s3, s4
+; GFX9-NEXT:    s_sub_i32 s3, s3, s4
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl <2 x i32> <i32 4096, i32 4096>, %y
   %r = sdiv <2 x i32> %x, %shl.y
   store <2 x i32> %r, <2 x i32> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @srem_i32_oddk_denom(i32 addrspace(1)* %out, i32 %x) {

 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX6-NEXT:    s_xor_b32 s6, s2, s5
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s6, v0
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s4, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s4, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v0, s5, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX6-NEXT:    s_mul_i32 s7, s7, s4
+; GFX6-NEXT:    s_sub_i32 s6, s6, s7
+; GFX6-NEXT:    s_sub_i32 s7, s6, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX6-NEXT:    s_cselect_b32 s6, s7, s6
+; GFX6-NEXT:    s_sub_i32 s7, s6, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX6-NEXT:    s_cselect_b32 s4, s7, s6
+; GFX6-NEXT:    s_xor_b32 s4, s4, s5
+; GFX6-NEXT:    s_sub_i32 s4, s4, s5
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX6-LABEL: srem_v2i32_pow2_shl_denom:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_lshl_b32 s2, 0x1000, s6
 ; GFX6-NEXT:    s_ashr_i32 s3, s2, 31
 ; GFX6-NEXT:    s_add_i32 s2, s2, s3
-; GFX6-NEXT:    s_xor_b32 s6, s2, s3
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s6
-; GFX6-NEXT:    s_lshl_b32 s7, 0x1000, s7
-; GFX6-NEXT:    s_ashr_i32 s8, s7, 31
-; GFX6-NEXT:    s_add_i32 s7, s7, s8
+; GFX6-NEXT:    s_xor_b32 s2, s2, s3
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s2
+; GFX6-NEXT:    s_sub_i32 s3, 0, s2
+; GFX6-NEXT:    s_ashr_i32 s6, s4, 31
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    s_xor_b32 s7, s7, s8
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s7
-; GFX6-NEXT:    s_sub_i32 s9, 0, s6
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    s_ashr_i32 s8, s4, 31
-; GFX6-NEXT:    s_add_i32 s4, s4, s8
-; GFX6-NEXT:    v_mul_lo_u32 v2, s9, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX6-NEXT:    s_xor_b32 s4, s4, s8
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    s_sub_i32 s9, 0, s7
+; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
+; GFX6-NEXT:    s_add_i32 s3, s4, s6
+; GFX6-NEXT:    s_xor_b32 s3, s3, s6
+; GFX6-NEXT:    s_lshl_b32 s4, 0x1000, s7
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s3, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX6-NEXT:    s_mul_i32 s7, s7, s2
+; GFX6-NEXT:    s_sub_i32 s3, s3, s7
+; GFX6-NEXT:    s_sub_i32 s7, s3, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b32 s3, s7, s3
+; GFX6-NEXT:    s_sub_i32 s7, s3, s2
+; GFX6-NEXT:    s_cmp_ge_u32 s3, s2
+; GFX6-NEXT:    s_cselect_b32 s7, s7, s3
+; GFX6-NEXT:    s_ashr_i32 s2, s4, 31
+; GFX6-NEXT:    s_add_i32 s4, s4, s2
+; GFX6-NEXT:    s_xor_b32 s4, s4, s2
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s4
+; GFX6-NEXT:    s_sub_i32 s2, 0, s4
+; GFX6-NEXT:    s_ashr_i32 s8, s5, 31
+; GFX6-NEXT:    s_xor_b32 s7, s7, s6
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX6-NEXT:    s_sub_i32 s6, s7, s6
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX6-NEXT:    s_add_i32 s2, s5, s8
+; GFX6-NEXT:    s_xor_b32 s5, s2, s8
 ; GFX6-NEXT:    s_mov_b32 s2, -1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_mul_lo_u32 v2, s9, v1
-; GFX6-NEXT:    s_ashr_i32 s9, s5, 31
-; GFX6-NEXT:    s_add_i32 s5, s5, s9
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX6-NEXT:    v_mul_hi_u32 v2, v1, v2
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s6, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    s_xor_b32 s4, s5, s9
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
-; GFX6-NEXT:    v_mul_hi_u32 v1, s4, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s6, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s7
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s4, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s7, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s7, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT:    v_xor_b32_e32 v1, s9, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v1, vcc, s9, v1
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX6-NEXT:    s_mul_i32 s7, s7, s4
+; GFX6-NEXT:    s_sub_i32 s5, s5, s7
+; GFX6-NEXT:    s_sub_i32 s7, s5, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s4
+; GFX6-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX6-NEXT:    s_sub_i32 s7, s5, s4
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s4
+; GFX6-NEXT:    s_cselect_b32 s4, s7, s5
+; GFX6-NEXT:    s_xor_b32 s4, s4, s8
+; GFX6-NEXT:    s_sub_i32 s4, s4, s8
+; GFX6-NEXT:    v_mov_b32_e32 v0, s6
+; GFX6-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_v2i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24

 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v0, s4
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s5
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s5
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_mov_b32 s4, s0
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s5
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v1, v3

 ; GFX6-NEXT:    s_movk_i32 s2, 0x11e
 ; GFX6-NEXT:    v_cmp_lt_u32_e64 s[0:1], s2, v4
 ; GFX6-NEXT:    s_mov_b32 s9, 0x976a7376
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_lt_u32_e64 s[0:1], s9, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], s8, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v4, v6, v5, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v5, s[0:1], 2, v0
+; GFX6-NEXT:    v_add_i32_e64 v5, s[0:1], 1, v0
 ; GFX6-NEXT:    v_addc_u32_e64 v6, s[0:1], 0, v1, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 1, v0
+; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 2, v0
 ; GFX6-NEXT:    v_addc_u32_e64 v8, s[0:1], 0, v1, s[0:1]
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v8, v6, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, v5, v7, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, v6, v8, s[0:1]
 ; GFX6-NEXT:    v_mov_b32_e32 v6, s3
 ; GFX6-NEXT:    v_subb_u32_e32 v2, vcc, v6, v2, vcc
 ; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s2, v2
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s9, v3
 ; GFX6-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s8, v2
 ; GFX6-NEXT:    v_cndmask_b32_e32 v2, v6, v3, vcc
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v7, v5, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: udiv_i64_oddk_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4f176a73
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
 ; GFX9-NEXT:    v_madmk_f32 v0, v1, 0x438f8000, v0

 ; GFX9-NEXT:    s_cmpk_gt_u32 s6, 0x11e
 ; GFX9-NEXT:    s_mov_b32 s10, 0x976a7376
 ; GFX9-NEXT:    s_cselect_b32 s9, -1, 0
 ; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s10, v1
 ; GFX9-NEXT:    s_cmpk_eq_i32 s6, 0x11f
 ; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, s[0:1]
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s9
 ; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GFX9-NEXT:    s_add_u32 s6, s3, 2
 ; GFX9-NEXT:    v_cndmask_b32_e64 v1, v3, v1, s[0:1]
-; GFX9-NEXT:    s_addc_u32 s0, s2, 0
-; GFX9-NEXT:    s_add_u32 s9, s3, 1
-; GFX9-NEXT:    s_addc_u32 s1, s2, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s7, s7, s8
-; GFX9-NEXT:    s_cmpk_gt_u32 s7, 0x11e
-; GFX9-NEXT:    v_mov_b32_e32 v3, s1
-; GFX9-NEXT:    v_mov_b32_e32 v4, s0
+; GFX9-NEXT:    s_add_u32 s0, s3, 1
+; GFX9-NEXT:    s_addc_u32 s6, s2, 0
+; GFX9-NEXT:    s_add_u32 s1, s3, 2
+; GFX9-NEXT:    s_addc_u32 s9, s2, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v1
-; GFX9-NEXT:    s_cselect_b32 s8, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v1, s6
+; GFX9-NEXT:    v_mov_b32_e32 v4, s9
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s7, s8
+; GFX9-NEXT:    s_cmpk_gt_u32 s0, 0x11e
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
 ; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s10, v0
-; GFX9-NEXT:    s_cmpk_eq_i32 s7, 0x11f
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v3, v4, s[0:1]
+; GFX9-NEXT:    s_cmpk_eq_i32 s0, 0x11f
 ; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v3, s8
+; GFX9-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v3, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v4, v0, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v0, s9
-; GFX9-NEXT:    v_mov_b32_e32 v3, s6
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v3, s3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v0, s3
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v1, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
 ; GFX9-NEXT:    s_endpgm
   %r = udiv i64 %x, 1235195949943
   store i64 %r, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @udiv_i64_pow2k_denom(i64 addrspace(1)* %out, i64 %x) {

 ; GFX6-LABEL: udiv_v2i64_mixed_pow2k_denom:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    v_mov_b32_e32 v0, 0x4f800000
 ; GFX6-NEXT:    v_madak_f32 v0, 0, v0, 0x457ff000
 ; GFX6-NEXT:    v_rcp_f32_e32 v0, v0
 ; GFX6-NEXT:    s_movk_i32 s6, 0xf001
 ; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0xd
-; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_movk_i32 s8, 0xfff
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GFX6-NEXT:    v_trunc_f32_e32 v1, v1
 ; GFX6-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_lshr_b64 s[8:9], s[0:1], 12
-; GFX6-NEXT:    s_movk_i32 s0, 0xfff
+; GFX6-NEXT:    s_lshr_b64 s[0:1], s[0:1], 12
+; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    v_mul_hi_u32 v2, v0, s6
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s6
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s6
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v0, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2

 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s3, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s3, v0
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s0
-; GFX6-NEXT:    v_mul_hi_u32 v5, v0, s0
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v8, v0, s0
+; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s8
+; GFX6-NEXT:    v_mul_hi_u32 v5, v0, s8
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
+; GFX6-NEXT:    v_mul_lo_u32 v8, v0, s8
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v0
+; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 2, v0
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
 ; GFX6-NEXT:    v_mov_b32_e32 v5, s3
 ; GFX6-NEXT:    v_sub_i32_e32 v8, vcc, s2, v8
 ; GFX6-NEXT:    v_subb_u32_e32 v4, vcc, v5, v4, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s0, v8
+; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s8, v8
 ; GFX6-NEXT:    v_subbrev_u32_e32 v9, vcc, 0, v4, vcc
-; GFX6-NEXT:    s_movk_i32 s0, 0xffe
-; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v5
+; GFX6-NEXT:    s_movk_i32 s2, 0xffe
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s2, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
 ; GFX6-NEXT:    v_cndmask_b32_e32 v5, -1, v5, vcc
-; GFX6-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v8
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, -1, v5, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v3, v1, v3, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v6, v2, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v0, v1, s[0:1]
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    v_mov_b32_e32 v1, s9
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s2, v8
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, -1, v5, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v1, v3, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v0, v2, vcc
+; GFX6-NEXT:    v_mov_b32_e32 v0, s0
+; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: udiv_v2i64_mixed_pow2k_denom:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4f800000
-; GFX9-NEXT:    v_madak_f32 v0, 0, v0, 0x457ff000
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x457ff000
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
+; GFX9-NEXT:    v_mac_f32_e32 v0, 0, v1
 ; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_movk_i32 s2, 0xf001
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GFX9-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GFX9-NEXT:    v_trunc_f32_e32 v1, v1
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, s2
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, s2
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, s2
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_hi_u32 v5, v0, v3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_lo_u32 v6, v1, v3
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v4, vcc, v5, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v7, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v4, vcc, v4, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v3, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v4, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, s2
-; GFX9-NEXT:    v_mul_lo_u32 v3, v1, s2
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, s2
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v6, v0, v5
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v6, v3
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v7, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v7, v1, v5
-; GFX9-NEXT:    v_mul_hi_u32 v5, v1, v5
-; GFX9-NEXT:    s_movk_i32 s0, 0xfff
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_lshr_b64 s[4:5], s[4:5], 12
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v7
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v6, v5, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s6, v0
-; GFX9-NEXT:    v_mul_hi_u32 v5, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v6, s7, v1
-; GFX9-NEXT:    v_mul_lo_u32 v1, s7, v1
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, s7, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s7, v0
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v2, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, 2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, s0
-; GFX9-NEXT:    v_mul_hi_u32 v6, v0, s0
-; GFX9-NEXT:    v_mul_lo_u32 v9, v0, s0
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, 1, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v8, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_add_u32_e32 v5, v6, v5
-; GFX9-NEXT:    v_mov_b32_e32 v6, s7
-; GFX9-NEXT:    v_sub_co_u32_e32 v9, vcc, s6, v9
-; GFX9-NEXT:    v_subb_co_u32_e32 v5, vcc, v6, v5, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e32 v6, vcc, s0, v9
-; GFX9-NEXT:    v_subbrev_co_u32_e32 v10, vcc, 0, v5, vcc
-; GFX9-NEXT:    s_movk_i32 s0, 0xffe
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v10
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, -1, v6, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v9
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, -1, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v8, v3, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v1, v3, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v7, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    s_lshr_b64 s[0:1], s[4:5], 12
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX9-NEXT:    s_mul_hi_u32 s5, s4, 0xfffff001
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
+; GFX9-NEXT:    s_sub_i32 s5, s5, s4
+; GFX9-NEXT:    s_mul_i32 s9, s8, 0xfffff001
+; GFX9-NEXT:    s_add_i32 s5, s5, s9
+; GFX9-NEXT:    s_mul_i32 s11, s4, 0xfffff001
+; GFX9-NEXT:    s_mul_hi_u32 s9, s4, s5
+; GFX9-NEXT:    s_mul_i32 s10, s4, s5
+; GFX9-NEXT:    s_mul_hi_u32 s4, s4, s11
+; GFX9-NEXT:    s_add_u32 s4, s4, s10
+; GFX9-NEXT:    s_addc_u32 s9, 0, s9
+; GFX9-NEXT:    s_mul_hi_u32 s12, s8, s11
+; GFX9-NEXT:    s_mul_i32 s11, s8, s11
+; GFX9-NEXT:    s_add_u32 s4, s4, s11
+; GFX9-NEXT:    s_mul_hi_u32 s10, s8, s5
+; GFX9-NEXT:    s_addc_u32 s4, s9, s12
+; GFX9-NEXT:    s_addc_u32 s9, s10, 0
+; GFX9-NEXT:    s_mul_i32 s5, s8, s5
+; GFX9-NEXT:    s_add_u32 s4, s4, s5
+; GFX9-NEXT:    s_addc_u32 s5, 0, s9
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s4, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s4, s8, s5
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_hi_u32 s9, s8, 0xfffff001
+; GFX9-NEXT:    s_mul_i32 s5, s4, 0xfffff001
+; GFX9-NEXT:    s_sub_i32 s9, s9, s8
+; GFX9-NEXT:    s_add_i32 s9, s9, s5
+; GFX9-NEXT:    s_mul_i32 s11, s8, 0xfffff001
+; GFX9-NEXT:    s_mul_hi_u32 s5, s8, s9
+; GFX9-NEXT:    s_mul_i32 s10, s8, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s8, s11
+; GFX9-NEXT:    s_add_u32 s8, s8, s10
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
+; GFX9-NEXT:    s_mul_hi_u32 s12, s4, s11
+; GFX9-NEXT:    s_mul_i32 s11, s4, s11
+; GFX9-NEXT:    s_add_u32 s8, s8, s11
+; GFX9-NEXT:    s_mul_hi_u32 s10, s4, s9
+; GFX9-NEXT:    s_addc_u32 s5, s5, s12
+; GFX9-NEXT:    s_addc_u32 s8, s10, 0
+; GFX9-NEXT:    s_mul_i32 s9, s4, s9
+; GFX9-NEXT:    s_add_u32 s5, s5, s9
+; GFX9-NEXT:    s_addc_u32 s8, 0, s8
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s5, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s4, s4, s8
+; GFX9-NEXT:    v_readfirstlane_b32 s9, v0
+; GFX9-NEXT:    s_mul_i32 s8, s6, s4
+; GFX9-NEXT:    s_mul_hi_u32 s10, s6, s9
+; GFX9-NEXT:    s_mul_hi_u32 s5, s6, s4
+; GFX9-NEXT:    s_add_u32 s8, s10, s8
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
+; GFX9-NEXT:    s_mul_hi_u32 s11, s7, s9
+; GFX9-NEXT:    s_mul_i32 s9, s7, s9
+; GFX9-NEXT:    s_add_u32 s8, s8, s9
+; GFX9-NEXT:    s_mul_hi_u32 s10, s7, s4
+; GFX9-NEXT:    s_addc_u32 s5, s5, s11
+; GFX9-NEXT:    s_addc_u32 s8, s10, 0
+; GFX9-NEXT:    s_mul_i32 s4, s7, s4
+; GFX9-NEXT:    s_add_u32 s4, s5, s4
+; GFX9-NEXT:    s_addc_u32 s5, 0, s8
+; GFX9-NEXT:    s_add_u32 s8, s4, 1
+; GFX9-NEXT:    s_addc_u32 s9, s5, 0
+; GFX9-NEXT:    s_add_u32 s10, s4, 2
+; GFX9-NEXT:    s_mul_i32 s13, s5, 0xfff
+; GFX9-NEXT:    s_mul_hi_u32 s14, s4, 0xfff
+; GFX9-NEXT:    s_addc_u32 s11, s5, 0
+; GFX9-NEXT:    s_add_i32 s14, s14, s13
+; GFX9-NEXT:    s_mul_i32 s13, s4, 0xfff
+; GFX9-NEXT:    v_mov_b32_e32 v0, s13
+; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s6, v0
+; GFX9-NEXT:    s_movk_i32 s12, 0xfff
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s7, s14
+; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s12, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s7, s6, 0
+; GFX9-NEXT:    s_movk_i32 s12, 0xffe
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s12, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, -1, v1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, s8
+; GFX9-NEXT:    v_mov_b32_e32 v3, s10
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v2, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, s9
+; GFX9-NEXT:    v_mov_b32_e32 v3, s11
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s12, v0
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s4
-; GFX9-NEXT:    v_mov_b32_e32 v1, s5
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[2:3]
 ; GFX9-NEXT:    s_endpgm
   %r = udiv <2 x i64> %x, <i64 4096, i64 4095>
   store <2 x i64> %r, <2 x i64> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @udiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) {

 ; GFX6-NEXT:    v_cmp_lt_u32_e64 s[2:3], s6, v4
 ; GFX6-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[2:3], s4, v5
 ; GFX6-NEXT:    v_subrev_i32_e64 v3, s[0:1], s12, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, v6, v7, s[2:3]
 ; GFX6-NEXT:    v_subbrev_u32_e64 v2, s[0:1], 0, v2, s[0:1]
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
-; GFX6-NEXT:    v_mov_b32_e32 v5, s7
-; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GFX6-NEXT:    v_mov_b32_e32 v4, s7
+; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v4, v1, vcc
 ; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s5, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s4, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
 ; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v4, v3, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: urem_i64_oddk_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4f1761f8
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
 ; GFX9-NEXT:    v_madmk_f32 v0, v1, 0x438f8000, v0

 ; GFX9-NEXT:    s_mul_i32 s0, s0, 0x9761f7c9
 ; GFX9-NEXT:    s_add_i32 s9, s2, s1
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    s_sub_i32 s1, s7, s9
 ; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s6, v0
 ; GFX9-NEXT:    s_mov_b32 s8, 0x9761f7c9
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
 ; GFX9-NEXT:    s_subb_u32 s6, s1, 0x11f
-; GFX9-NEXT:    v_subrev_co_u32_e64 v3, s[0:1], s8, v0
+; GFX9-NEXT:    v_subrev_co_u32_e64 v1, s[0:1], s8, v0
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
 ; GFX9-NEXT:    s_subb_u32 s10, s6, 0
 ; GFX9-NEXT:    s_cmpk_gt_u32 s10, 0x11e
 ; GFX9-NEXT:    s_cselect_b32 s11, -1, 0
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[2:3], s12, v3
+; GFX9-NEXT:    v_cmp_lt_u32_e64 s[2:3], s12, v1
 ; GFX9-NEXT:    s_cmpk_eq_i32 s10, 0x11f
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, s[2:3]
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s11
 ; GFX9-NEXT:    s_cselect_b64 s[2:3], -1, 0
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v4, v1, s[2:3]
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[2:3]
 ; GFX9-NEXT:    s_subb_u32 s2, s6, 0x11f
-; GFX9-NEXT:    v_subrev_co_u32_e64 v4, s[0:1], s8, v3
+; GFX9-NEXT:    v_subrev_co_u32_e64 v4, s[0:1], s8, v1
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
-; GFX9-NEXT:    s_subb_u32 s0, s2, 0
+; GFX9-NEXT:    s_subb_u32 s2, s2, 0
+; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v3
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v1, v4, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v1, s10
+; GFX9-NEXT:    v_mov_b32_e32 v4, s2
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s2, s7, s9
-; GFX9-NEXT:    s_cmpk_gt_u32 s2, 0x11e
-; GFX9-NEXT:    v_mov_b32_e32 v5, s10
-; GFX9-NEXT:    v_mov_b32_e32 v6, s0
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v1
-; GFX9-NEXT:    s_cselect_b32 s3, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s7, s9
+; GFX9-NEXT:    s_cmpk_gt_u32 s0, 0x11e
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
 ; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s12, v0
-; GFX9-NEXT:    s_cmpk_eq_i32 s2, 0x11f
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v5, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v6, s3
+; GFX9-NEXT:    s_cmpk_eq_i32 s0, 0x11f
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v6, v5, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v6, s2
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v6, v1, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v4, v5, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s0
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v5, v1, vcc
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
 ; GFX9-NEXT:    s_endpgm
   %r = urem i64 %x, 1235195393993
   store i64 %r, i64 addrspace(1)* %out
   ret void
 }
 

 ; CHECK-NEXT:    store i64 [[R]], i64 addrspace(1)* [[OUT:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: sdiv_i64_oddk_denom:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    v_mov_b32_e32 v0, 0x4f800000
 ; GFX6-NEXT:    v_madak_f32 v0, 0, v0, 0x4996c7d8
 ; GFX6-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX6-NEXT:    s_mov_b32 s5, 0xffed2705
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s8, 0xffed2705
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GFX6-NEXT:    v_trunc_f32_e32 v1, v1
 ; GFX6-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_ashr_i32 s8, s3, 31
-; GFX6-NEXT:    s_add_u32 s2, s2, s8
-; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s5
-; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s5
-; GFX6-NEXT:    v_mul_lo_u32 v4, v0, s5
-; GFX6-NEXT:    s_mov_b32 s9, s8
-; GFX6-NEXT:    s_addc_u32 s3, s3, s8
+; GFX6-NEXT:    s_mov_b32 s0, s4
+; GFX6-NEXT:    s_mov_b32 s4, 0x12d8fb
+; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s8
+; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s8
+; GFX6-NEXT:    v_mul_lo_u32 v4, v0, s8
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    s_mov_b32 s1, s5
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
+; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v1, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX6-NEXT:    s_xor_b64 s[2:3], s[2:3], s[8:9]
-; GFX6-NEXT:    s_mov_b32 s4, s0
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v5, v4, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s5
-; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s5
-; GFX6-NEXT:    s_mov_b32 s0, 0x12d8fb
-; GFX6-NEXT:    s_mov_b32 s6, -1
+; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s8
+; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s8
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s5
+; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s8
 ; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v7, v5, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GFX6-NEXT:    s_ashr_i32 s8, s7, 31
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
+; GFX6-NEXT:    s_add_u32 s6, s6, s8
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    s_mov_b32 s9, s8
+; GFX6-NEXT:    s_addc_u32 s7, s7, s8
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v1
-; GFX6-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX6-NEXT:    v_mul_hi_u32 v4, s2, v1
-; GFX6-NEXT:    v_mul_hi_u32 v5, s3, v1
-; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v1
+; GFX6-NEXT:    s_xor_b64 s[6:7], s[6:7], s[8:9]
+; GFX6-NEXT:    v_mul_lo_u32 v2, s6, v1
+; GFX6-NEXT:    v_mul_hi_u32 v3, s6, v0
+; GFX6-NEXT:    v_mul_hi_u32 v4, s6, v1
+; GFX6-NEXT:    v_mul_hi_u32 v5, s7, v1
+; GFX6-NEXT:    v_mul_lo_u32 v1, s7, v1
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GFX6-NEXT:    v_mul_hi_u32 v0, s3, v0
-; GFX6-NEXT:    s_mov_b32 s5, s1
+; GFX6-NEXT:    v_mul_lo_u32 v4, s7, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s7, v0
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s0
-; GFX6-NEXT:    v_mul_hi_u32 v5, v0, s0
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v8, v0, s0
+; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s4
+; GFX6-NEXT:    v_mul_hi_u32 v5, v0, s4
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
+; GFX6-NEXT:    v_mul_lo_u32 v8, v0, s4
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v0
+; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 2, v0
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
-; GFX6-NEXT:    v_mov_b32_e32 v5, s3
-; GFX6-NEXT:    v_sub_i32_e32 v8, vcc, s2, v8
+; GFX6-NEXT:    v_mov_b32_e32 v5, s7
+; GFX6-NEXT:    v_sub_i32_e32 v8, vcc, s6, v8
 ; GFX6-NEXT:    v_subb_u32_e32 v4, vcc, v5, v4, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s0, v8
+; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s4, v8
 ; GFX6-NEXT:    v_subbrev_u32_e32 v9, vcc, 0, v4, vcc
-; GFX6-NEXT:    s_mov_b32 s0, 0x12d8fa
-; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v5
+; GFX6-NEXT:    s_mov_b32 s4, 0x12d8fa
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
 ; GFX6-NEXT:    v_cndmask_b32_e32 v5, -1, v5, vcc
-; GFX6-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v8
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, -1, v5, s[0:1]
-; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v8
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, -1, v5, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
 ; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
 ; GFX6-NEXT:    v_xor_b32_e32 v1, s8, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
-; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i64_oddk_denom:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4f800000
-; GFX9-NEXT:    v_madak_f32 v0, 0, v0, 0x4996c7d8
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4996c7d8
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
+; GFX9-NEXT:    v_mac_f32_e32 v0, 0, v1
 ; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_mov_b32 s2, 0xffed2705
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GFX9-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GFX9-NEXT:    v_trunc_f32_e32 v1, v1
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s2
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s2
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v7, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v6, v4, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v4, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s2
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s2
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX9-NEXT:    s_mul_hi_u32 s6, s5, 0xffed2705
+; GFX9-NEXT:    s_mul_i32 s7, s4, 0xffed2705
+; GFX9-NEXT:    s_add_i32 s6, s6, s7
+; GFX9-NEXT:    s_sub_i32 s6, s6, s5
+; GFX9-NEXT:    s_mul_i32 s9, s5, 0xffed2705
+; GFX9-NEXT:    s_mul_hi_u32 s7, s5, s6
+; GFX9-NEXT:    s_mul_i32 s8, s5, s6
+; GFX9-NEXT:    s_mul_hi_u32 s5, s5, s9
+; GFX9-NEXT:    s_add_u32 s5, s5, s8
+; GFX9-NEXT:    s_addc_u32 s7, 0, s7
+; GFX9-NEXT:    s_mul_hi_u32 s10, s4, s9
+; GFX9-NEXT:    s_mul_i32 s9, s4, s9
+; GFX9-NEXT:    s_add_u32 s5, s5, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s4, s6
+; GFX9-NEXT:    s_addc_u32 s5, s7, s10
+; GFX9-NEXT:    s_addc_u32 s7, s8, 0
+; GFX9-NEXT:    s_mul_i32 s6, s4, s6
+; GFX9-NEXT:    s_add_u32 s5, s5, s6
+; GFX9-NEXT:    s_addc_u32 s6, 0, s7
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s5, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s4, s4, s6
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX9-NEXT:    s_mul_i32 s5, s4, 0xffed2705
+; GFX9-NEXT:    s_mul_hi_u32 s7, s6, 0xffed2705
+; GFX9-NEXT:    s_add_i32 s7, s7, s5
+; GFX9-NEXT:    s_sub_i32 s5, s7, s6
+; GFX9-NEXT:    s_mul_i32 s8, s6, 0xffed2705
+; GFX9-NEXT:    s_mul_hi_u32 s11, s6, s5
+; GFX9-NEXT:    s_mul_i32 s12, s6, s5
+; GFX9-NEXT:    s_mul_hi_u32 s6, s6, s8
+; GFX9-NEXT:    s_add_u32 s6, s6, s12
+; GFX9-NEXT:    s_mul_hi_u32 s9, s4, s8
+; GFX9-NEXT:    s_mul_i32 s10, s4, s8
+; GFX9-NEXT:    s_addc_u32 s8, 0, s11
+; GFX9-NEXT:    s_add_u32 s6, s6, s10
+; GFX9-NEXT:    s_mul_hi_u32 s7, s4, s5
+; GFX9-NEXT:    s_addc_u32 s6, s8, s9
+; GFX9-NEXT:    s_addc_u32 s7, s7, 0
+; GFX9-NEXT:    s_mul_i32 s5, s4, s5
+; GFX9-NEXT:    s_add_u32 s5, s6, s5
+; GFX9-NEXT:    s_addc_u32 s6, 0, s7
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s5, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s6, s4, s6
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_ashr_i32 s2, s7, 31
-; GFX9-NEXT:    s_add_u32 s0, s6, s2
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v5, v1, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v6, vcc, v7, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v7, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v4, vcc, v6, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v7, v5, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v4, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
-; GFX9-NEXT:    s_mov_b32 s3, s2
-; GFX9-NEXT:    s_addc_u32 s1, s7, s2
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    s_xor_b64 s[0:1], s[0:1], s[2:3]
-; GFX9-NEXT:    v_mul_lo_u32 v2, s0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s0, v0
-; GFX9-NEXT:    v_mul_hi_u32 v5, s0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v6, s1, v1
-; GFX9-NEXT:    v_mul_lo_u32 v1, s1, v1
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, s1, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s1, v0
-; GFX9-NEXT:    s_mov_b32 s3, 0x12d8fb
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v2, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, 2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, s3
-; GFX9-NEXT:    v_mul_hi_u32 v6, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v9, v0, s3
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, 1, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v8, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_add_u32_e32 v5, v6, v5
-; GFX9-NEXT:    v_mov_b32_e32 v6, s1
-; GFX9-NEXT:    v_sub_co_u32_e32 v9, vcc, s0, v9
-; GFX9-NEXT:    v_subb_co_u32_e32 v5, vcc, v6, v5, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e32 v6, vcc, s3, v9
-; GFX9-NEXT:    v_subbrev_co_u32_e32 v10, vcc, 0, v5, vcc
-; GFX9-NEXT:    s_mov_b32 s0, 0x12d8fa
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v10
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, -1, v6, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v9
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, -1, v6, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v7, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v8, v3, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GFX9-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX9-NEXT:    v_xor_b32_e32 v1, s2, v1
-; GFX9-NEXT:    v_mov_b32_e32 v2, s2
-; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s2, v0
-; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[4:5]
+; GFX9-NEXT:    s_ashr_i32 s4, s3, 31
+; GFX9-NEXT:    s_add_u32 s2, s2, s4
+; GFX9-NEXT:    s_mov_b32 s5, s4
+; GFX9-NEXT:    s_addc_u32 s3, s3, s4
+; GFX9-NEXT:    s_xor_b64 s[2:3], s[2:3], s[4:5]
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_i32 s7, s2, s6
+; GFX9-NEXT:    s_mul_hi_u32 s9, s2, s8
+; GFX9-NEXT:    s_mul_hi_u32 s5, s2, s6
+; GFX9-NEXT:    s_add_u32 s7, s9, s7
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
+; GFX9-NEXT:    s_mul_hi_u32 s10, s3, s8
+; GFX9-NEXT:    s_mul_i32 s8, s3, s8
+; GFX9-NEXT:    s_add_u32 s7, s7, s8
+; GFX9-NEXT:    s_mul_hi_u32 s9, s3, s6
+; GFX9-NEXT:    s_addc_u32 s5, s5, s10
+; GFX9-NEXT:    s_addc_u32 s7, s9, 0
+; GFX9-NEXT:    s_mul_i32 s6, s3, s6
+; GFX9-NEXT:    s_add_u32 s5, s5, s6
+; GFX9-NEXT:    s_addc_u32 s6, 0, s7
+; GFX9-NEXT:    s_add_u32 s7, s5, 1
+; GFX9-NEXT:    s_addc_u32 s8, s6, 0
+; GFX9-NEXT:    s_add_u32 s9, s5, 2
+; GFX9-NEXT:    s_mul_i32 s12, s6, 0x12d8fb
+; GFX9-NEXT:    s_mul_hi_u32 s13, s5, 0x12d8fb
+; GFX9-NEXT:    s_addc_u32 s10, s6, 0
+; GFX9-NEXT:    s_add_i32 s13, s13, s12
+; GFX9-NEXT:    s_mul_i32 s12, s5, 0x12d8fb
+; GFX9-NEXT:    v_mov_b32_e32 v0, s12
+; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s2, v0
+; GFX9-NEXT:    s_mov_b32 s11, 0x12d8fb
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s2, s3, s13
+; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s11, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s3, s2, 0
+; GFX9-NEXT:    s_mov_b32 s11, 0x12d8fa
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s11, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s3, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, -1, v1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s7
+; GFX9-NEXT:    v_mov_b32_e32 v4, s9
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s8
+; GFX9-NEXT:    v_mov_b32_e32 v4, s10
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s11, v0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, s6
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v4, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
+; GFX9-NEXT:    v_xor_b32_e32 v1, s4, v1
+; GFX9-NEXT:    v_xor_b32_e32 v3, s4, v0
+; GFX9-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s4, v1
+; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v3, v4, vcc
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = sdiv i64 %x, 1235195
   store i64 %r, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @sdiv_i64_pow2k_denom(i64 addrspace(1)* %out, i64 %x) {
 ; CHECK-LABEL: @sdiv_i64_pow2k_denom(

 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, s4, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s5, v0
 ; GFX6-NEXT:    s_mov_b32 s5, s1
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GFX6-NEXT:    v_subrev_i32_e64 v5, s[0:1], s10, v3
 ; GFX6-NEXT:    v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s11, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s10, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], s11, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v4, v6, v5, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v5, s[0:1], 2, v0
+; GFX6-NEXT:    v_add_i32_e64 v5, s[0:1], 1, v0
 ; GFX6-NEXT:    v_addc_u32_e64 v6, s[0:1], 0, v1, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 1, v0
+; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 2, v0
 ; GFX6-NEXT:    v_addc_u32_e64 v8, s[0:1], 0, v1, s[0:1]
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v8, v6, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, v5, v7, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, v6, v8, s[0:1]
 ; GFX6-NEXT:    v_mov_b32_e32 v6, s3
 ; GFX6-NEXT:    v_subb_u32_e32 v2, vcc, v6, v2, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s11, v2
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v3
 ; GFX6-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s11, v2
 ; GFX6-NEXT:    v_cndmask_b32_e32 v2, v6, v3, vcc
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v7, v5, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GFX6-NEXT:    s_xor_b64 s[0:1], s[12:13], s[8:9]
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
 ; GFX6-NEXT:    v_xor_b32_e32 v0, s0, v0
 ; GFX6-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s1
 ; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;

 ; GFX9-NEXT:    s_subb_u32 s6, s6, 0
 ; GFX9-NEXT:    s_cmp_ge_u32 s6, s9
 ; GFX9-NEXT:    s_cselect_b32 s15, -1, 0
 ; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s8, v2
 ; GFX9-NEXT:    s_cmp_eq_u32 s6, s9
 ; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, -1, s[0:1]
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s15
 ; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GFX9-NEXT:    s_add_u32 s6, s12, 2
 ; GFX9-NEXT:    v_cndmask_b32_e64 v2, v3, v2, s[0:1]
-; GFX9-NEXT:    s_addc_u32 s0, s13, 0
-; GFX9-NEXT:    s_add_u32 s15, s12, 1
-; GFX9-NEXT:    s_addc_u32 s1, s13, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s7, s7, s14
-; GFX9-NEXT:    s_cmp_ge_u32 s7, s9
-; GFX9-NEXT:    v_mov_b32_e32 v3, s1
-; GFX9-NEXT:    v_mov_b32_e32 v4, s0
+; GFX9-NEXT:    s_add_u32 s0, s12, 1
+; GFX9-NEXT:    s_addc_u32 s6, s13, 0
+; GFX9-NEXT:    s_add_u32 s1, s12, 2
+; GFX9-NEXT:    s_addc_u32 s15, s13, 0
+; GFX9-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v2
-; GFX9-NEXT:    s_cselect_b32 s14, -1, 0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v1
-; GFX9-NEXT:    s_cmp_eq_u32 s7, s9
 ; GFX9-NEXT:    v_cndmask_b32_e64 v2, v3, v4, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v3, s6
+; GFX9-NEXT:    v_mov_b32_e32 v4, s15
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s7, s14
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s9
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
+; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s0, s9
 ; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v3, s14
+; GFX9-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v3, s13
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, s13
 ; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v2, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v2, s15
-; GFX9-NEXT:    v_mov_b32_e32 v3, s6
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v3, vcc
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s12
 ; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
 ; GFX9-NEXT:    s_xor_b64 s[0:1], s[10:11], s[2:3]
 ; GFX9-NEXT:    v_xor_b32_e32 v2, s0, v2
 ; GFX9-NEXT:    v_xor_b32_e32 v3, s1, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s0, v2
 ; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v3, v4, vcc

 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    s_ashr_i32 s8, s1, 31
 ; GFX6-NEXT:    s_lshr_b32 s8, s8, 20
 ; GFX6-NEXT:    v_mul_hi_u32 v2, v0, s6
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, s6
 ; GFX6-NEXT:    s_add_u32 s0, s0, s8
 ; GFX6-NEXT:    s_addc_u32 s1, s1, 0
-; GFX6-NEXT:    s_ashr_i64 s[8:9], s[0:1], 12
+; GFX6-NEXT:    s_ashr_i64 s[0:1], s[0:1], 12
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s6
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
+; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX6-NEXT:    s_ashr_i32 s10, s3, 31
-; GFX6-NEXT:    s_add_u32 s0, s2, s10
+; GFX6-NEXT:    s_ashr_i32 s8, s3, 31
+; GFX6-NEXT:    s_add_u32 s2, s2, s8
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v5, v3, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s6
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s6
-; GFX6-NEXT:    s_mov_b32 s11, s10
-; GFX6-NEXT:    s_addc_u32 s1, s3, s10
-; GFX6-NEXT:    s_xor_b64 s[0:1], s[0:1], s[10:11]
+; GFX6-NEXT:    s_mov_b32 s9, s8
+; GFX6-NEXT:    s_addc_u32 s3, s3, s8
+; GFX6-NEXT:    s_xor_b64 s[2:3], s[2:3], s[8:9]
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s6
 ; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v7, v5, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, s0, v1
-; GFX6-NEXT:    v_mul_hi_u32 v3, s0, v0
-; GFX6-NEXT:    v_mul_hi_u32 v4, s0, v1
-; GFX6-NEXT:    v_mul_hi_u32 v5, s1, v1
-; GFX6-NEXT:    v_mul_lo_u32 v1, s1, v1
+; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v1
+; GFX6-NEXT:    v_mul_hi_u32 v3, s2, v0
+; GFX6-NEXT:    v_mul_hi_u32 v4, s2, v1
+; GFX6-NEXT:    v_mul_hi_u32 v5, s3, v1
+; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v1
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, s1, v0
-; GFX6-NEXT:    v_mul_hi_u32 v0, s1, v0
-; GFX6-NEXT:    s_movk_i32 s2, 0xfff
+; GFX6-NEXT:    v_mul_lo_u32 v4, s3, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s3, v0
+; GFX6-NEXT:    s_movk_i32 s9, 0xfff
 ; GFX6-NEXT:    s_mov_b32 s6, -1
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s2
-; GFX6-NEXT:    v_mul_hi_u32 v5, v0, s2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 2, v0
-; GFX6-NEXT:    v_mul_lo_u32 v8, v0, s2
+; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s9
+; GFX6-NEXT:    v_mul_hi_u32 v5, v0, s9
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
+; GFX6-NEXT:    v_mul_lo_u32 v8, v0, s9
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 1, v0
+; GFX6-NEXT:    v_add_i32_e32 v6, vcc, 2, v0
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
-; GFX6-NEXT:    v_mov_b32_e32 v5, s1
-; GFX6-NEXT:    v_sub_i32_e32 v8, vcc, s0, v8
+; GFX6-NEXT:    v_mov_b32_e32 v5, s3
+; GFX6-NEXT:    v_sub_i32_e32 v8, vcc, s2, v8
 ; GFX6-NEXT:    v_subb_u32_e32 v4, vcc, v5, v4, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s2, v8
+; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s9, v8
 ; GFX6-NEXT:    v_subbrev_u32_e32 v9, vcc, 0, v4, vcc
-; GFX6-NEXT:    s_movk_i32 s0, 0xffe
-; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v5
+; GFX6-NEXT:    s_movk_i32 s2, 0xffe
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s2, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
 ; GFX6-NEXT:    v_cndmask_b32_e32 v5, -1, v5, vcc
-; GFX6-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v8
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, -1, v5, s[0:1]
-; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GFX6-NEXT:    v_xor_b32_e32 v0, s10, v0
-; GFX6-NEXT:    v_xor_b32_e32 v1, s10, v1
-; GFX6-NEXT:    v_mov_b32_e32 v3, s10
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s10, v0
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s2, v8
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, -1, v5, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
+; GFX6-NEXT:    v_xor_b32_e32 v1, s8, v1
+; GFX6-NEXT:    v_mov_b32_e32 v3, s8
+; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s8, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v3, vcc, v1, v3, vcc
-; GFX6-NEXT:    v_mov_b32_e32 v0, s8
-; GFX6-NEXT:    v_mov_b32_e32 v1, s9
+; GFX6-NEXT:    v_mov_b32_e32 v0, s0
+; GFX6-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: ssdiv_v2i64_mixed_pow2k_denom:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0x457ff000
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
-; GFX9-NEXT:    v_mac_f32_e32 v0, 0, v1
-; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_movk_i32 s8, 0xf001
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x457ff000
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0x4f800000
+; GFX9-NEXT:    v_mac_f32_e32 v1, 0, v2
+; GFX9-NEXT:    v_rcp_f32_e32 v1, v1
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
-; GFX9-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
-; GFX9-NEXT:    v_trunc_f32_e32 v1, v1
-; GFX9-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    v_mul_f32_e32 v1, 0x5f7ffffc, v1
+; GFX9-NEXT:    v_mul_f32_e32 v2, 0x2f800000, v1
+; GFX9-NEXT:    v_trunc_f32_e32 v2, v2
+; GFX9-NEXT:    v_mac_f32_e32 v1, 0xcf800000, v2
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_ashr_i32 s0, s5, 31
 ; GFX9-NEXT:    s_lshr_b32 s0, s0, 20
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, s8
-; GFX9-NEXT:    v_mul_lo_u32 v3, v1, s8
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s8
 ; GFX9-NEXT:    s_add_u32 s0, s4, s0
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v2
 ; GFX9-NEXT:    s_addc_u32 s1, s5, 0
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v5, v0, v4
-; GFX9-NEXT:    v_mul_hi_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v1, v2
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v5, v3
-; GFX9-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v6, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX9-NEXT:    s_ashr_i64 s[4:5], s[0:1], 12
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v4, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v7, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v4, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, s8
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s8
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s8
-; GFX9-NEXT:    s_ashr_i32 s8, s7, 31
-; GFX9-NEXT:    s_add_u32 s0, s6, s8
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v5, v1, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v6, vcc, v7, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v7, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v4, vcc, v6, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v7, v5, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v4, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
-; GFX9-NEXT:    s_mov_b32 s9, s8
-; GFX9-NEXT:    s_addc_u32 s1, s7, s8
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    s_xor_b64 s[0:1], s[0:1], s[8:9]
-; GFX9-NEXT:    v_mul_lo_u32 v2, s0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s0, v0
-; GFX9-NEXT:    v_mul_hi_u32 v5, s0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v6, s1, v1
-; GFX9-NEXT:    v_mul_lo_u32 v1, s1, v1
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v3, v2
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, s1, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s1, v0
-; GFX9-NEXT:    s_movk_i32 s6, 0xfff
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v1
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, 0, v2, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, 2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, s6
-; GFX9-NEXT:    v_mul_hi_u32 v6, v0, s6
-; GFX9-NEXT:    v_mul_lo_u32 v9, v0, s6
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, 1, v0
-; GFX9-NEXT:    v_addc_co_u32_e32 v8, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_add_u32_e32 v5, v6, v5
-; GFX9-NEXT:    v_mov_b32_e32 v6, s1
-; GFX9-NEXT:    v_sub_co_u32_e32 v9, vcc, s0, v9
-; GFX9-NEXT:    v_subb_co_u32_e32 v5, vcc, v6, v5, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e32 v6, vcc, s6, v9
-; GFX9-NEXT:    v_subbrev_co_u32_e32 v10, vcc, 0, v5, vcc
-; GFX9-NEXT:    s_movk_i32 s0, 0xffe
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v10
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, -1, v6, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v9
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, -1, v6, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v7, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v8, v3, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GFX9-NEXT:    v_xor_b32_e32 v0, s8, v0
-; GFX9-NEXT:    v_xor_b32_e32 v1, s8, v1
-; GFX9-NEXT:    v_mov_b32_e32 v3, s8
-; GFX9-NEXT:    v_subrev_co_u32_e32 v2, vcc, s8, v0
-; GFX9-NEXT:    v_subb_co_u32_e32 v3, vcc, v1, v3, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v0, s4
-; GFX9-NEXT:    v_mov_b32_e32 v1, s5
-; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[2:3]
+; GFX9-NEXT:    s_mul_hi_u32 s5, s4, 0xfffff001
+; GFX9-NEXT:    s_mul_i32 s9, s8, 0xfffff001
+; GFX9-NEXT:    s_add_i32 s5, s5, s9
+; GFX9-NEXT:    s_sub_i32 s5, s5, s4
+; GFX9-NEXT:    s_mul_i32 s11, s4, 0xfffff001
+; GFX9-NEXT:    s_ashr_i64 s[0:1], s[0:1], 12
+; GFX9-NEXT:    s_mul_hi_u32 s9, s4, s5
+; GFX9-NEXT:    s_mul_i32 s10, s4, s5
+; GFX9-NEXT:    s_mul_hi_u32 s4, s4, s11
+; GFX9-NEXT:    s_add_u32 s4, s4, s10
+; GFX9-NEXT:    s_addc_u32 s9, 0, s9
+; GFX9-NEXT:    s_mul_hi_u32 s12, s8, s11
+; GFX9-NEXT:    s_mul_i32 s11, s8, s11
+; GFX9-NEXT:    s_add_u32 s4, s4, s11
+; GFX9-NEXT:    s_mul_hi_u32 s10, s8, s5
+; GFX9-NEXT:    s_addc_u32 s4, s9, s12
+; GFX9-NEXT:    s_addc_u32 s9, s10, 0
+; GFX9-NEXT:    s_mul_i32 s5, s8, s5
+; GFX9-NEXT:    s_add_u32 s4, s4, s5
+; GFX9-NEXT:    s_addc_u32 s5, 0, s9
+; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, s4, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s4, s8, s5
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
+; GFX9-NEXT:    s_mul_i32 s5, s4, 0xfffff001
+; GFX9-NEXT:    s_mul_hi_u32 s9, s8, 0xfffff001
+; GFX9-NEXT:    s_add_i32 s9, s9, s5
+; GFX9-NEXT:    s_sub_i32 s5, s9, s8
+; GFX9-NEXT:    s_mul_i32 s10, s8, 0xfffff001
+; GFX9-NEXT:    s_mul_hi_u32 s13, s8, s5
+; GFX9-NEXT:    s_mul_i32 s14, s8, s5
+; GFX9-NEXT:    s_mul_hi_u32 s8, s8, s10
+; GFX9-NEXT:    s_add_u32 s8, s8, s14
+; GFX9-NEXT:    s_mul_hi_u32 s11, s4, s10
+; GFX9-NEXT:    s_mul_i32 s12, s4, s10
+; GFX9-NEXT:    s_addc_u32 s10, 0, s13
+; GFX9-NEXT:    s_add_u32 s8, s8, s12
+; GFX9-NEXT:    s_mul_hi_u32 s9, s4, s5
+; GFX9-NEXT:    s_addc_u32 s8, s10, s11
+; GFX9-NEXT:    s_addc_u32 s9, s9, 0
+; GFX9-NEXT:    s_mul_i32 s5, s4, s5
+; GFX9-NEXT:    s_add_u32 s5, s8, s5
+; GFX9-NEXT:    s_addc_u32 s8, 0, s9
+; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, s5, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s8, s4, s8
+; GFX9-NEXT:    s_ashr_i32 s4, s7, 31
+; GFX9-NEXT:    s_add_u32 s6, s6, s4
+; GFX9-NEXT:    s_mov_b32 s5, s4
+; GFX9-NEXT:    s_addc_u32 s7, s7, s4
+; GFX9-NEXT:    s_xor_b64 s[6:7], s[6:7], s[4:5]
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v1
+; GFX9-NEXT:    s_mul_i32 s9, s6, s8
+; GFX9-NEXT:    s_mul_hi_u32 s11, s6, s10
+; GFX9-NEXT:    s_mul_hi_u32 s5, s6, s8
+; GFX9-NEXT:    s_add_u32 s9, s11, s9
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
+; GFX9-NEXT:    s_mul_hi_u32 s12, s7, s10
+; GFX9-NEXT:    s_mul_i32 s10, s7, s10
+; GFX9-NEXT:    s_add_u32 s9, s9, s10
+; GFX9-NEXT:    s_mul_hi_u32 s11, s7, s8
+; GFX9-NEXT:    s_addc_u32 s5, s5, s12
+; GFX9-NEXT:    s_addc_u32 s9, s11, 0
+; GFX9-NEXT:    s_mul_i32 s8, s7, s8
+; GFX9-NEXT:    s_add_u32 s5, s5, s8
+; GFX9-NEXT:    s_addc_u32 s8, 0, s9
+; GFX9-NEXT:    s_add_u32 s9, s5, 1
+; GFX9-NEXT:    s_addc_u32 s10, s8, 0
+; GFX9-NEXT:    s_add_u32 s11, s5, 2
+; GFX9-NEXT:    s_mul_i32 s14, s8, 0xfff
+; GFX9-NEXT:    s_mul_hi_u32 s15, s5, 0xfff
+; GFX9-NEXT:    s_addc_u32 s12, s8, 0
+; GFX9-NEXT:    s_add_i32 s15, s15, s14
+; GFX9-NEXT:    s_mul_i32 s14, s5, 0xfff
+; GFX9-NEXT:    v_mov_b32_e32 v1, s14
+; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, s6, v1
+; GFX9-NEXT:    s_movk_i32 s13, 0xfff
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s7, s15
+; GFX9-NEXT:    v_subrev_co_u32_e32 v2, vcc, s13, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s7, s6, 0
+; GFX9-NEXT:    s_movk_i32 s13, 0xffe
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s13, v2
+; GFX9-NEXT:    s_cmp_eq_u32 s7, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, -1, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s9
+; GFX9-NEXT:    v_mov_b32_e32 v4, s11
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s10
+; GFX9-NEXT:    v_mov_b32_e32 v4, s12
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s13, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s6, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, -1, v1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v4, s8
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
+; GFX9-NEXT:    v_cndmask_b32_e32 v1, v4, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s5
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
+; GFX9-NEXT:    v_xor_b32_e32 v2, s4, v2
+; GFX9-NEXT:    v_xor_b32_e32 v1, s4, v1
+; GFX9-NEXT:    v_mov_b32_e32 v4, s4
+; GFX9-NEXT:    v_subrev_co_u32_e32 v3, vcc, s4, v2
+; GFX9-NEXT:    v_subb_co_u32_e32 v4, vcc, v1, v4, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    global_store_dwordx4 v0, v[1:4], s[2:3]
 ; GFX9-NEXT:    s_endpgm
   %r = sdiv <2 x i64> %x, <i64 4096, i64 4095>
   store <2 x i64> %r, <2 x i64> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %out, <2 x i64> %x, <2 x i64> %y) {
 ; CHECK-LABEL: @sdiv_v2i64_pow2_shl_denom(

 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX6-NEXT:    s_addc_u32 s1, s5, s16
 ; GFX6-NEXT:    s_xor_b64 s[4:5], s[0:1], s[16:17]
 ; GFX6-NEXT:    v_mul_lo_u32 v2, s10, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v3, s10, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s11, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s10, v0
-; GFX6-NEXT:    s_xor_b64 s[14:15], s[16:17], s[14:15]
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v1, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5

 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, s12, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v3, s12, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s13, v0
 ; GFX6-NEXT:    v_mov_b32_e32 v5, s13
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s12, v0
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GFX6-NEXT:    v_sub_i32_e32 v4, vcc, s5, v2
 ; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s4, v3
 ; GFX6-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, vcc
 ; GFX6-NEXT:    v_subrev_i32_e64 v5, s[0:1], s12, v3
 ; GFX6-NEXT:    v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s13, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s12, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], s13, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v4, v6, v5, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v5, s[0:1], 2, v0
+; GFX6-NEXT:    v_add_i32_e64 v5, s[0:1], 1, v0
 ; GFX6-NEXT:    v_addc_u32_e64 v6, s[0:1], 0, v1, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 1, v0
+; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 2, v0
 ; GFX6-NEXT:    v_addc_u32_e64 v8, s[0:1], 0, v1, s[0:1]
-; GFX6-NEXT:    s_ashr_i32 s4, s3, 31
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, v5, v7, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, v6, v8, s[0:1]
+; GFX6-NEXT:    s_xor_b64 s[0:1], s[16:17], s[14:15]
+; GFX6-NEXT:    s_ashr_i32 s4, s3, 31
 ; GFX6-NEXT:    s_add_u32 s2, s2, s4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v8, v6, s[0:1]
 ; GFX6-NEXT:    v_mov_b32_e32 v6, s5
 ; GFX6-NEXT:    s_mov_b32 s5, s4
 ; GFX6-NEXT:    s_addc_u32 s3, s3, s4
 ; GFX6-NEXT:    s_xor_b64 s[2:3], s[2:3], s[4:5]
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v8, s2
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v9, s3
 ; GFX6-NEXT:    v_subb_u32_e32 v2, vcc, v6, v2, vcc
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v6, s2
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v7, s3
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s13, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v8, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s12, v3
+; GFX6-NEXT:    v_mac_f32_e32 v6, 0x4f800000, v7
+; GFX6-NEXT:    v_rcp_f32_e32 v6, v6
 ; GFX6-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s13, v2
-; GFX6-NEXT:    v_mac_f32_e32 v8, 0x4f800000, v9
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v6, v3, vcc
-; GFX6-NEXT:    v_rcp_f32_e32 v3, v8
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v8, v3, vcc
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v7, v5, s[0:1]
-; GFX6-NEXT:    v_mul_f32_e32 v3, 0x5f7ffffc, v3
-; GFX6-NEXT:    v_mul_f32_e32 v4, 0x2f800000, v3
-; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
-; GFX6-NEXT:    v_mac_f32_e32 v3, 0xcf800000, v4
+; GFX6-NEXT:    v_mul_f32_e32 v2, 0x5f7ffffc, v6
+; GFX6-NEXT:    v_mul_f32_e32 v3, 0x2f800000, v2
+; GFX6-NEXT:    v_trunc_f32_e32 v3, v3
+; GFX6-NEXT:    v_mac_f32_e32 v2, 0xcf800000, v3
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v4, v4
-; GFX6-NEXT:    s_sub_u32 s0, 0, s2
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_mul_hi_u32 v2, s0, v3
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v4
-; GFX6-NEXT:    s_subb_u32 s1, 0, s3
-; GFX6-NEXT:    v_mul_lo_u32 v6, s1, v3
-; GFX6-NEXT:    s_ashr_i32 s12, s7, 31
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v3
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v6, v2
-; GFX6-NEXT:    v_mul_lo_u32 v6, v3, v2
-; GFX6-NEXT:    v_mul_hi_u32 v7, v3, v5
-; GFX6-NEXT:    v_mul_hi_u32 v8, v3, v2
-; GFX6-NEXT:    v_mul_hi_u32 v9, v4, v2
-; GFX6-NEXT:    v_mul_lo_u32 v2, v4, v2
+; GFX6-NEXT:    s_sub_u32 s12, 0, s2
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
+; GFX6-NEXT:    v_mul_hi_u32 v4, s12, v2
+; GFX6-NEXT:    v_mul_lo_u32 v5, s12, v3
+; GFX6-NEXT:    s_subb_u32 s13, 0, s3
+; GFX6-NEXT:    v_mul_lo_u32 v6, s13, v2
+; GFX6-NEXT:    v_xor_b32_e32 v0, s0, v0
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
+; GFX6-NEXT:    v_mul_lo_u32 v5, s12, v2
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
+; GFX6-NEXT:    v_mul_lo_u32 v6, v2, v4
+; GFX6-NEXT:    v_mul_hi_u32 v7, v2, v5
+; GFX6-NEXT:    v_mul_hi_u32 v8, v2, v4
+; GFX6-NEXT:    v_mul_hi_u32 v9, v3, v4
+; GFX6-NEXT:    v_mul_lo_u32 v4, v3, v4
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v8, v4, v5
-; GFX6-NEXT:    v_mul_hi_u32 v5, v4, v5
-; GFX6-NEXT:    s_mov_b32 s13, s12
-; GFX6-NEXT:    v_xor_b32_e32 v0, s14, v0
+; GFX6-NEXT:    v_mul_lo_u32 v8, v3, v5
+; GFX6-NEXT:    v_mul_hi_u32 v5, v3, v5
+; GFX6-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v6, v8
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, v7, v5, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v6, vcc, 0, v9, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v4, v5, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, s0, v3
-; GFX6-NEXT:    v_mul_hi_u32 v5, s0, v2
-; GFX6-NEXT:    v_mul_lo_u32 v6, s1, v2
-; GFX6-NEXT:    v_xor_b32_e32 v1, s15, v1
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v3, v5, vcc
+; GFX6-NEXT:    v_mul_lo_u32 v4, s12, v3
+; GFX6-NEXT:    v_mul_hi_u32 v5, s12, v2
+; GFX6-NEXT:    v_mul_lo_u32 v6, s13, v2
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v2
+; GFX6-NEXT:    v_mul_lo_u32 v5, s12, v2
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v8, v2, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v9, v2, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v10, v2, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v3, v5
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v3, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v3, v4
 ; GFX6-NEXT:    v_add_i32_e32 v8, vcc, v9, v8
 ; GFX6-NEXT:    v_addc_u32_e32 v9, vcc, 0, v10, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v3, v4
 ; GFX6-NEXT:    v_add_i32_e32 v5, vcc, v8, v5
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, v9, v7, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v6, vcc, 0, v6, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
+; GFX6-NEXT:    s_ashr_i32 s12, s7, 31
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
-; GFX6-NEXT:    s_add_u32 s0, s6, s12
+; GFX6-NEXT:    s_add_u32 s6, s6, s12
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
-; GFX6-NEXT:    s_addc_u32 s1, s7, s12
+; GFX6-NEXT:    s_mov_b32 s13, s12
+; GFX6-NEXT:    s_addc_u32 s7, s7, s12
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v3, v5, vcc
-; GFX6-NEXT:    s_xor_b64 s[6:7], s[0:1], s[12:13]
+; GFX6-NEXT:    s_xor_b64 s[6:7], s[6:7], s[12:13]
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s6, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v5, s6, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, s6, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, s7, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s7, v3
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v7, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v7, s7, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v2, s7, v2
-; GFX6-NEXT:    v_mov_b32_e32 v6, s15
+; GFX6-NEXT:    v_mov_b32_e32 v6, s1
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v7
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, v5, v2, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v8, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s2, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v5, s2, v2
-; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s14, v0
+; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v1, v6, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v6, s3, v2
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s2, v2
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
 ; GFX6-NEXT:    v_sub_i32_e32 v6, vcc, s7, v4
 ; GFX6-NEXT:    v_mov_b32_e32 v7, s3
 ; GFX6-NEXT:    v_sub_i32_e32 v5, vcc, s6, v5
 ; GFX6-NEXT:    v_subb_u32_e64 v6, s[0:1], v6, v7, vcc
 ; GFX6-NEXT:    v_subrev_i32_e64 v7, s[0:1], s2, v5
 ; GFX6-NEXT:    v_subbrev_u32_e64 v6, s[0:1], 0, v6, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v6
 ; GFX6-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v7
 ; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[0:1]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], s3, v6
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, v8, v7, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 2, v2
+; GFX6-NEXT:    v_add_i32_e64 v7, s[0:1], 1, v2
 ; GFX6-NEXT:    v_addc_u32_e64 v8, s[0:1], 0, v3, s[0:1]
-; GFX6-NEXT:    v_add_i32_e64 v9, s[0:1], 1, v2
+; GFX6-NEXT:    v_add_i32_e64 v9, s[0:1], 2, v2
 ; GFX6-NEXT:    v_addc_u32_e64 v10, s[0:1], 0, v3, s[0:1]
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
-; GFX6-NEXT:    v_cndmask_b32_e64 v6, v10, v8, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v6, v7, v9, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v7, v8, v10, s[0:1]
 ; GFX6-NEXT:    v_mov_b32_e32 v8, s7
 ; GFX6-NEXT:    v_subb_u32_e32 v4, vcc, v8, v4, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s3, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v8, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s2, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s3, v4
 ; GFX6-NEXT:    v_cndmask_b32_e32 v4, v8, v5, vcc
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v9, v7, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
 ; GFX6-NEXT:    s_xor_b64 s[0:1], s[12:13], s[4:5]
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v6, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
 ; GFX6-NEXT:    v_xor_b32_e32 v2, s0, v2
 ; GFX6-NEXT:    v_xor_b32_e32 v3, s1, v3
 ; GFX6-NEXT:    v_mov_b32_e32 v4, s1
 ; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s0, v2
 ; GFX6-NEXT:    v_subb_u32_e32 v3, vcc, v3, v4, vcc
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; GFX6-NEXT:    s_endpgm

 ; GFX9-NEXT:    s_subb_u32 s4, s4, 0
 ; GFX9-NEXT:    s_cmp_ge_u32 s4, s13
 ; GFX9-NEXT:    s_cselect_b32 s19, -1, 0
 ; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s12, v1
 ; GFX9-NEXT:    s_cmp_eq_u32 s4, s13
 ; GFX9-NEXT:    v_cndmask_b32_e64 v1, 0, -1, s[0:1]
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s19
 ; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GFX9-NEXT:    s_add_u32 s4, s16, 2
 ; GFX9-NEXT:    v_cndmask_b32_e64 v1, v2, v1, s[0:1]
-; GFX9-NEXT:    s_addc_u32 s0, s17, 0
-; GFX9-NEXT:    s_add_u32 s19, s16, 1
-; GFX9-NEXT:    s_addc_u32 s1, s17, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s5, s5, s18
-; GFX9-NEXT:    s_cmp_ge_u32 s5, s13
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_mov_b32_e32 v3, s0
+; GFX9-NEXT:    s_add_u32 s0, s16, 1
+; GFX9-NEXT:    s_addc_u32 s4, s17, 0
+; GFX9-NEXT:    s_add_u32 s1, s16, 2
+; GFX9-NEXT:    s_addc_u32 s19, s17, 0
+; GFX9-NEXT:    v_mov_b32_e32 v2, s0
+; GFX9-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v1
-; GFX9-NEXT:    s_cselect_b32 s18, -1, 0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s12, v0
-; GFX9-NEXT:    s_cmp_eq_u32 s5, s13
 ; GFX9-NEXT:    v_cndmask_b32_e64 v1, v2, v3, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v2, s4
+; GFX9-NEXT:    v_mov_b32_e32 v3, s19
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s5, s18
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s13
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
+; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s12, v0
+; GFX9-NEXT:    s_cmp_eq_u32 s0, s13
 ; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v2, s18
+; GFX9-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v0, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v2, s17
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v1, s19
-; GFX9-NEXT:    v_mov_b32_e32 v2, s4
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
 ; GFX9-NEXT:    s_xor_b64 s[0:1], s[14:15], s[8:9]
 ; GFX9-NEXT:    s_ashr_i32 s4, s11, 31
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v0, vcc
 ; GFX9-NEXT:    s_add_u32 s8, s10, s4
+; GFX9-NEXT:    v_mov_b32_e32 v3, s17
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; GFX9-NEXT:    s_mov_b32 s5, s4
 ; GFX9-NEXT:    s_addc_u32 s9, s11, s4
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v2, vcc
 ; GFX9-NEXT:    v_mov_b32_e32 v2, s16
 ; GFX9-NEXT:    s_xor_b64 s[8:9], s[8:9], s[4:5]
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v2, v1, vcc
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s8
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v3, s9
 ; GFX9-NEXT:    v_xor_b32_e32 v1, s0, v1
 ; GFX9-NEXT:    v_xor_b32_e32 v5, s1, v0
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s0, v1

 ; GFX9-NEXT:    s_subb_u32 s6, s6, 0
 ; GFX9-NEXT:    s_cmp_ge_u32 s6, s9
 ; GFX9-NEXT:    s_cselect_b32 s15, -1, 0
 ; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s8, v3
 ; GFX9-NEXT:    s_cmp_eq_u32 s6, s9
 ; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, s[0:1]
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s15
 ; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GFX9-NEXT:    s_add_u32 s6, s12, 2
 ; GFX9-NEXT:    v_cndmask_b32_e64 v3, v5, v3, s[0:1]
-; GFX9-NEXT:    s_addc_u32 s0, s13, 0
-; GFX9-NEXT:    s_add_u32 s15, s12, 1
-; GFX9-NEXT:    s_addc_u32 s1, s13, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s7, s7, s14
-; GFX9-NEXT:    s_cmp_ge_u32 s7, s9
-; GFX9-NEXT:    v_mov_b32_e32 v5, s1
-; GFX9-NEXT:    v_mov_b32_e32 v6, s0
+; GFX9-NEXT:    s_add_u32 s0, s12, 1
+; GFX9-NEXT:    s_addc_u32 s6, s13, 0
+; GFX9-NEXT:    s_add_u32 s1, s12, 2
+; GFX9-NEXT:    s_addc_u32 s15, s13, 0
+; GFX9-NEXT:    v_mov_b32_e32 v5, s0
+; GFX9-NEXT:    v_mov_b32_e32 v6, s1
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v3
-; GFX9-NEXT:    s_cselect_b32 s14, -1, 0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v2
-; GFX9-NEXT:    s_cmp_eq_u32 s7, s9
 ; GFX9-NEXT:    v_cndmask_b32_e64 v3, v5, v6, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v5, s6
+; GFX9-NEXT:    v_mov_b32_e32 v6, s15
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v5, v6, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s7, s14
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s9
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
+; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v2
+; GFX9-NEXT:    s_cmp_eq_u32 s0, s9
 ; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v5, s14
+; GFX9-NEXT:    v_mov_b32_e32 v6, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v5, v2, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v5, s13
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v6, s13
 ; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v5, v3, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v3, s15
-; GFX9-NEXT:    v_mov_b32_e32 v5, s6
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v6, v5, vcc
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s12
 ; GFX9-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
 ; GFX9-NEXT:    s_xor_b64 s[0:1], s[10:11], s[4:5]
 ; GFX9-NEXT:    v_xor_b32_e32 v3, s0, v3
 ; GFX9-NEXT:    v_xor_b32_e32 v5, s1, v2
 ; GFX9-NEXT:    v_mov_b32_e32 v6, s1
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v2, vcc, s0, v3
 ; GFX9-NEXT:    v_subb_co_u32_e32 v3, vcc, v5, v6, vcc

 ; CHECK-NEXT:    store i64 [[R]], i64 addrspace(1)* [[OUT:%.*]], align 4
 ; CHECK-NEXT:    ret void
 ;
 ; GFX6-LABEL: srem_i64_oddk_denom:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    v_mov_b32_e32 v0, 0x4f800000
 ; GFX6-NEXT:    v_madak_f32 v0, 0, v0, 0x4996c7d8
 ; GFX6-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX6-NEXT:    s_mov_b32 s4, 0xffed2705
-; GFX6-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s7, 0xf000
+; GFX6-NEXT:    s_mov_b32 s8, 0xffed2705
+; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GFX6-NEXT:    v_trunc_f32_e32 v1, v1
 ; GFX6-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    s_ashr_i32 s8, s3, 31
-; GFX6-NEXT:    s_add_u32 s2, s2, s8
-; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s4
-; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s4
-; GFX6-NEXT:    v_mul_lo_u32 v4, v0, s4
-; GFX6-NEXT:    s_mov_b32 s9, s8
-; GFX6-NEXT:    s_addc_u32 s3, s3, s8
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s8
+; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s8
+; GFX6-NEXT:    v_mul_lo_u32 v4, v0, s8
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v1, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX6-NEXT:    s_xor_b64 s[2:3], s[2:3], s[8:9]
-; GFX6-NEXT:    s_mov_b32 s5, s1
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v5, v4, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s4
-; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s4
-; GFX6-NEXT:    s_mov_b32 s6, -1
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s4
+; GFX6-NEXT:    v_mul_lo_u32 v2, v1, s8
+; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s8
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_mul_lo_u32 v3, v0, s8
 ; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v7, v5, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX6-NEXT:    s_ashr_i32 s8, s7, 31
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
+; GFX6-NEXT:    s_add_u32 s0, s6, s8
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    s_mov_b32 s9, s8
+; GFX6-NEXT:    s_addc_u32 s1, s7, s8
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v1
-; GFX6-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX6-NEXT:    v_mul_hi_u32 v4, s2, v1
-; GFX6-NEXT:    v_mul_hi_u32 v5, s3, v1
-; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v1
+; GFX6-NEXT:    s_xor_b64 s[6:7], s[0:1], s[8:9]
+; GFX6-NEXT:    v_mul_lo_u32 v2, s6, v1
+; GFX6-NEXT:    v_mul_hi_u32 v3, s6, v0
+; GFX6-NEXT:    v_mul_hi_u32 v4, s6, v1
+; GFX6-NEXT:    v_mul_hi_u32 v5, s7, v1
+; GFX6-NEXT:    v_mul_lo_u32 v1, s7, v1
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GFX6-NEXT:    v_mul_hi_u32 v0, s3, v0
-; GFX6-NEXT:    s_mov_b32 s4, s0
-; GFX6-NEXT:    s_mov_b32 s0, 0x12d8fb
+; GFX6-NEXT:    v_mul_lo_u32 v4, s7, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s7, v0
+; GFX6-NEXT:    s_mov_b32 s0, s4
+; GFX6-NEXT:    s_mov_b32 s4, 0x12d8fb
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s0
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, s0
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s0
+; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s4
+; GFX6-NEXT:    v_mul_hi_u32 v2, v0, s4
+; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s4
+; GFX6-NEXT:    s_mov_b32 s1, s5
 ; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GFX6-NEXT:    v_mov_b32_e32 v2, s3
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
+; GFX6-NEXT:    v_mov_b32_e32 v2, s7
+; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s0, v0
+; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s4, v0
 ; GFX6-NEXT:    v_subbrev_u32_e32 v3, vcc, 0, v1, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s0, v2
+; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s4, v2
 ; GFX6-NEXT:    v_subbrev_u32_e32 v5, vcc, 0, v3, vcc
-; GFX6-NEXT:    s_mov_b32 s0, 0x12d8fa
-; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v2
+; GFX6-NEXT:    s_mov_b32 s4, 0x12d8fa
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v2
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
 ; GFX6-NEXT:    v_cndmask_b32_e32 v6, -1, v6, vcc
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
-; GFX6-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v0
 ; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GFX6-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, -1, v5, s[0:1]
-; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
 ; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX6-NEXT:    v_cmp_lt_u32_e32 vcc, s4, v0
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
+; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, -1, v4, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
 ; GFX6-NEXT:    v_xor_b32_e32 v0, s8, v0
 ; GFX6-NEXT:    v_xor_b32_e32 v1, s8, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s8
 ; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
-; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: srem_i64_oddk_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    v_mov_b32_e32 v0, 0x4996c7d8
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0, v1
 ; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GFX9-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GFX9-NEXT:    v_trunc_f32_e32 v1, v1
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0xcf800000, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_readfirstlane_b32 s0, v1
-; GFX9-NEXT:    v_readfirstlane_b32 s1, v0
-; GFX9-NEXT:    s_mul_hi_u32 s2, s1, 0xffed2705
-; GFX9-NEXT:    s_mul_i32 s3, s0, 0xffed2705
-; GFX9-NEXT:    s_add_i32 s2, s2, s3
-; GFX9-NEXT:    s_sub_i32 s2, s2, s1
-; GFX9-NEXT:    s_mul_i32 s9, s1, 0xffed2705
-; GFX9-NEXT:    s_mul_hi_u32 s3, s1, s2
-; GFX9-NEXT:    s_mul_i32 s8, s1, s2
-; GFX9-NEXT:    s_mul_hi_u32 s1, s1, s9
-; GFX9-NEXT:    s_add_u32 s1, s1, s8
-; GFX9-NEXT:    s_addc_u32 s3, 0, s3
-; GFX9-NEXT:    s_mul_hi_u32 s10, s0, s9
-; GFX9-NEXT:    s_mul_i32 s9, s0, s9
-; GFX9-NEXT:    s_add_u32 s1, s1, s9
-; GFX9-NEXT:    s_mul_hi_u32 s8, s0, s2
-; GFX9-NEXT:    s_addc_u32 s1, s3, s10
-; GFX9-NEXT:    s_addc_u32 s3, s8, 0
-; GFX9-NEXT:    s_mul_i32 s2, s0, s2
-; GFX9-NEXT:    s_add_u32 s1, s1, s2
-; GFX9-NEXT:    s_addc_u32 s2, 0, s3
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s1, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX9-NEXT:    s_mul_hi_u32 s6, s5, 0xffed2705
+; GFX9-NEXT:    s_mul_i32 s7, s4, 0xffed2705
+; GFX9-NEXT:    s_add_i32 s6, s6, s7
+; GFX9-NEXT:    s_sub_i32 s6, s6, s5
+; GFX9-NEXT:    s_mul_i32 s9, s5, 0xffed2705
+; GFX9-NEXT:    s_mul_hi_u32 s7, s5, s6
+; GFX9-NEXT:    s_mul_i32 s8, s5, s6
+; GFX9-NEXT:    s_mul_hi_u32 s5, s5, s9
+; GFX9-NEXT:    s_add_u32 s5, s5, s8
+; GFX9-NEXT:    s_addc_u32 s7, 0, s7
+; GFX9-NEXT:    s_mul_hi_u32 s10, s4, s9
+; GFX9-NEXT:    s_mul_i32 s9, s4, s9
+; GFX9-NEXT:    s_add_u32 s5, s5, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s4, s6
+; GFX9-NEXT:    s_addc_u32 s5, s7, s10
+; GFX9-NEXT:    s_addc_u32 s7, s8, 0
+; GFX9-NEXT:    s_mul_i32 s6, s4, s6
+; GFX9-NEXT:    s_add_u32 s5, s5, s6
+; GFX9-NEXT:    s_addc_u32 s6, 0, s7
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s5, v0
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_addc_u32 s0, s0, s2
-; GFX9-NEXT:    v_readfirstlane_b32 s2, v0
-; GFX9-NEXT:    s_mul_i32 s1, s0, 0xffed2705
-; GFX9-NEXT:    s_mul_hi_u32 s3, s2, 0xffed2705
-; GFX9-NEXT:    s_add_i32 s3, s3, s1
-; GFX9-NEXT:    s_sub_i32 s1, s3, s2
-; GFX9-NEXT:    s_mul_i32 s8, s2, 0xffed2705
-; GFX9-NEXT:    s_mul_hi_u32 s11, s2, s1
-; GFX9-NEXT:    s_mul_i32 s12, s2, s1
-; GFX9-NEXT:    s_mul_hi_u32 s2, s2, s8
-; GFX9-NEXT:    s_add_u32 s2, s2, s12
-; GFX9-NEXT:    s_mul_hi_u32 s9, s0, s8
-; GFX9-NEXT:    s_mul_i32 s10, s0, s8
+; GFX9-NEXT:    s_addc_u32 s4, s4, s6
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX9-NEXT:    s_mul_i32 s5, s4, 0xffed2705
+; GFX9-NEXT:    s_mul_hi_u32 s7, s6, 0xffed2705
+; GFX9-NEXT:    s_add_i32 s7, s7, s5
+; GFX9-NEXT:    s_sub_i32 s5, s7, s6
+; GFX9-NEXT:    s_mul_i32 s8, s6, 0xffed2705
+; GFX9-NEXT:    s_mul_hi_u32 s11, s6, s5
+; GFX9-NEXT:    s_mul_i32 s12, s6, s5
+; GFX9-NEXT:    s_mul_hi_u32 s6, s6, s8
+; GFX9-NEXT:    s_add_u32 s6, s6, s12
+; GFX9-NEXT:    s_mul_hi_u32 s9, s4, s8
+; GFX9-NEXT:    s_mul_i32 s10, s4, s8
 ; GFX9-NEXT:    s_addc_u32 s8, 0, s11
-; GFX9-NEXT:    s_add_u32 s2, s2, s10
-; GFX9-NEXT:    s_mul_hi_u32 s3, s0, s1
-; GFX9-NEXT:    s_addc_u32 s2, s8, s9
-; GFX9-NEXT:    s_addc_u32 s3, s3, 0
-; GFX9-NEXT:    s_mul_i32 s1, s0, s1
-; GFX9-NEXT:    s_add_u32 s1, s2, s1
-; GFX9-NEXT:    s_addc_u32 s2, 0, s3
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s1, v0
+; GFX9-NEXT:    s_add_u32 s6, s6, s10
+; GFX9-NEXT:    s_mul_hi_u32 s7, s4, s5
+; GFX9-NEXT:    s_addc_u32 s6, s8, s9
+; GFX9-NEXT:    s_addc_u32 s7, s7, 0
+; GFX9-NEXT:    s_mul_i32 s5, s4, s5
+; GFX9-NEXT:    s_add_u32 s5, s6, s5
+; GFX9-NEXT:    s_addc_u32 s6, 0, s7
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s5, v0
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_addc_u32 s8, s0, s2
+; GFX9-NEXT:    s_addc_u32 s6, s4, s6
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_ashr_i32 s2, s7, 31
-; GFX9-NEXT:    s_add_u32 s0, s6, s2
-; GFX9-NEXT:    s_mov_b32 s3, s2
-; GFX9-NEXT:    s_addc_u32 s1, s7, s2
-; GFX9-NEXT:    s_xor_b64 s[0:1], s[0:1], s[2:3]
-; GFX9-NEXT:    v_readfirstlane_b32 s7, v0
-; GFX9-NEXT:    s_mul_i32 s6, s0, s8
-; GFX9-NEXT:    s_mul_hi_u32 s9, s0, s7
-; GFX9-NEXT:    s_mul_hi_u32 s3, s0, s8
-; GFX9-NEXT:    s_add_u32 s6, s9, s6
-; GFX9-NEXT:    s_addc_u32 s3, 0, s3
-; GFX9-NEXT:    s_mul_hi_u32 s10, s1, s7
-; GFX9-NEXT:    s_mul_i32 s7, s1, s7
-; GFX9-NEXT:    s_add_u32 s6, s6, s7
-; GFX9-NEXT:    s_mul_hi_u32 s9, s1, s8
-; GFX9-NEXT:    s_addc_u32 s3, s3, s10
-; GFX9-NEXT:    s_addc_u32 s6, s9, 0
-; GFX9-NEXT:    s_mul_i32 s7, s1, s8
-; GFX9-NEXT:    s_add_u32 s3, s3, s7
-; GFX9-NEXT:    s_addc_u32 s6, 0, s6
-; GFX9-NEXT:    s_mul_hi_u32 s8, s3, 0x12d8fb
-; GFX9-NEXT:    s_mul_i32 s3, s3, 0x12d8fb
+; GFX9-NEXT:    s_ashr_i32 s4, s3, 31
+; GFX9-NEXT:    s_add_u32 s2, s2, s4
+; GFX9-NEXT:    s_mov_b32 s5, s4
+; GFX9-NEXT:    s_addc_u32 s3, s3, s4
+; GFX9-NEXT:    s_xor_b64 s[2:3], s[2:3], s[4:5]
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_i32 s7, s2, s6
+; GFX9-NEXT:    s_mul_hi_u32 s9, s2, s8
+; GFX9-NEXT:    s_mul_hi_u32 s5, s2, s6
+; GFX9-NEXT:    s_add_u32 s7, s9, s7
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
+; GFX9-NEXT:    s_mul_hi_u32 s10, s3, s8
+; GFX9-NEXT:    s_mul_i32 s8, s3, s8
+; GFX9-NEXT:    s_add_u32 s7, s7, s8
+; GFX9-NEXT:    s_mul_hi_u32 s9, s3, s6
+; GFX9-NEXT:    s_addc_u32 s5, s5, s10
+; GFX9-NEXT:    s_addc_u32 s7, s9, 0
+; GFX9-NEXT:    s_mul_i32 s6, s3, s6
+; GFX9-NEXT:    s_add_u32 s5, s5, s6
+; GFX9-NEXT:    s_addc_u32 s6, 0, s7
+; GFX9-NEXT:    s_mul_hi_u32 s8, s5, 0x12d8fb
+; GFX9-NEXT:    s_mul_i32 s5, s5, 0x12d8fb
 ; GFX9-NEXT:    s_mul_i32 s6, s6, 0x12d8fb
-; GFX9-NEXT:    v_mov_b32_e32 v0, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s5
 ; GFX9-NEXT:    s_add_i32 s8, s8, s6
-; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s0, v0
+; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s2, v0
 ; GFX9-NEXT:    s_mov_b32 s7, 0x12d8fb
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s3, s1, s8
+; GFX9-NEXT:    s_subb_u32 s2, s3, s8
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s7, v0
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s0, s3, 0
+; GFX9-NEXT:    s_subb_u32 s3, s2, 0
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v3, vcc, s7, v1
 ; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s1, s0, 0
+; GFX9-NEXT:    s_subb_u32 s5, s3, 0
 ; GFX9-NEXT:    s_mov_b32 s6, 0x12d8fa
 ; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v1
-; GFX9-NEXT:    s_cmp_eq_u32 s0, 0
+; GFX9-NEXT:    s_cmp_eq_u32 s3, 0
 ; GFX9-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v4, -1, v4, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v5, s0
-; GFX9-NEXT:    v_mov_b32_e32 v6, s1
+; GFX9-NEXT:    v_mov_b32_e32 v5, s3
+; GFX9-NEXT:    v_mov_b32_e32 v6, s5
 ; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s6, v0
-; GFX9-NEXT:    s_cmp_eq_u32 s3, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v4, v5, v6, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, -1, v5, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v6, s3
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v1, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, v6, v4, s[0:1]
-; GFX9-NEXT:    v_xor_b32_e32 v0, s2, v0
-; GFX9-NEXT:    v_xor_b32_e32 v1, s2, v4
-; GFX9-NEXT:    v_mov_b32_e32 v3, s2
-; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s2, v0
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v0
+; GFX9-NEXT:    s_cmp_eq_u32 s2, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
+; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, -1, v3, vcc
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v3
+; GFX9-NEXT:    v_mov_b32_e32 v5, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v5, v4, vcc
+; GFX9-NEXT:    v_xor_b32_e32 v0, s4, v0
+; GFX9-NEXT:    v_xor_b32_e32 v1, s4, v3
+; GFX9-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s4, v0
 ; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[4:5]
+; GFX9-NEXT:    global_store_dwordx2 v2, v[0:1], s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = srem i64 %x, 1235195
   store i64 %r, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @srem_i64_pow2k_denom(i64 addrspace(1)* %out, i64 %x) {
 ; CHECK-LABEL: @srem_i64_pow2k_denom(

 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, s4, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s5, v0
 ; GFX6-NEXT:    s_mov_b32 s5, s1
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v0
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v2

 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s8, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v2, s8, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s9, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v0, s8, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s13, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s9
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s12, v0
 ; GFX6-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, vcc
 ; GFX6-NEXT:    v_subrev_i32_e64 v4, s[0:1], s8, v0
 ; GFX6-NEXT:    v_subbrev_u32_e64 v5, s[2:3], 0, v2, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[2:3], s9, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[2:3], s8, v4
 ; GFX6-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[2:3], s9, v5
 ; GFX6-NEXT:    v_subrev_i32_e64 v3, s[0:1], s8, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, v6, v7, s[2:3]
 ; GFX6-NEXT:    v_subbrev_u32_e64 v2, s[0:1], 0, v2, s[0:1]
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
-; GFX6-NEXT:    v_mov_b32_e32 v5, s13
-; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GFX6-NEXT:    v_mov_b32_e32 v4, s13
+; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v4, v1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
+; GFX6-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v4, v3, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
 ; GFX6-NEXT:    v_xor_b32_e32 v0, s10, v0
 ; GFX6-NEXT:    v_xor_b32_e32 v1, s10, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s10
 ; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s10, v0
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GFX6-NEXT:    s_endpgm
 ;

 ; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, s[2:3]
 ; GFX9-NEXT:    v_mov_b32_e32 v4, s13
 ; GFX9-NEXT:    s_cselect_b64 s[2:3], -1, 0
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
 ; GFX9-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[2:3]
 ; GFX9-NEXT:    s_subb_u32 s2, s6, s9
 ; GFX9-NEXT:    v_subrev_co_u32_e64 v4, s[0:1], s8, v2
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
-; GFX9-NEXT:    s_subb_u32 s0, s2, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s2, s7, s11
-; GFX9-NEXT:    s_cmp_ge_u32 s2, s9
-; GFX9-NEXT:    v_mov_b32_e32 v5, s12
-; GFX9-NEXT:    v_mov_b32_e32 v6, s0
+; GFX9-NEXT:    s_subb_u32 s2, s2, 0
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v3
-; GFX9-NEXT:    s_cselect_b32 s3, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v3, s12
+; GFX9-NEXT:    v_mov_b32_e32 v4, s2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s7, s11
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s9
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
 ; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v1
-; GFX9-NEXT:    s_cmp_eq_u32 s2, s9
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v5, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v6, s3
+; GFX9-NEXT:    s_cmp_eq_u32 s0, s9
+; GFX9-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v6, v5, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v6, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v4, v5, v4, vcc
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GFX9-NEXT:    v_mov_b32_e32 v5, s0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v6, v3, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
 ; GFX9-NEXT:    v_xor_b32_e32 v1, s10, v1
 ; GFX9-NEXT:    v_xor_b32_e32 v2, s10, v3
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s10
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s10, v1
 ; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v2, v3, vcc
 ; GFX9-NEXT:    global_store_dwordx2 v0, v[1:2], s[4:5]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl i64 4096, %y

 ; GFX6-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v3, s2, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s2, v0
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v2

 ; GFX6-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s16, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v2, s16, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s17, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v0, s16, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s5, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s17
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, vcc
 ; GFX6-NEXT:    v_subrev_i32_e64 v4, s[0:1], s16, v0
 ; GFX6-NEXT:    v_subbrev_u32_e64 v5, s[2:3], 0, v2, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[2:3], s17, v5
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[2:3], s16, v4
 ; GFX6-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[2:3], s17, v5
 ; GFX6-NEXT:    v_subrev_i32_e64 v3, s[0:1], s16, v4
 ; GFX6-NEXT:    v_cndmask_b32_e64 v6, v6, v7, s[2:3]
 ; GFX6-NEXT:    v_subbrev_u32_e64 v2, s[0:1], 0, v2, s[0:1]
-; GFX6-NEXT:    s_ashr_i32 s2, s15, 31
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
-; GFX6-NEXT:    s_add_u32 s4, s14, s2
+; GFX6-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
 ; GFX6-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
-; GFX6-NEXT:    v_mov_b32_e32 v5, s5
-; GFX6-NEXT:    s_mov_b32 s3, s2
-; GFX6-NEXT:    s_addc_u32 s5, s15, s2
-; GFX6-NEXT:    s_xor_b64 s[4:5], s[4:5], s[2:3]
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v6, s4
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v7, s5
-; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
+; GFX6-NEXT:    s_ashr_i32 s0, s15, 31
+; GFX6-NEXT:    s_add_u32 s2, s14, s0
+; GFX6-NEXT:    s_mov_b32 s1, s0
+; GFX6-NEXT:    s_addc_u32 s3, s15, s0
+; GFX6-NEXT:    v_mov_b32_e32 v4, s5
+; GFX6-NEXT:    s_xor_b64 s[4:5], s[2:3], s[0:1]
+; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v4, v1, vcc
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v4, s4
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v5, s5
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s17, v1
-; GFX6-NEXT:    v_mac_f32_e32 v6, 0x4f800000, v7
-; GFX6-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s16, v0
-; GFX6-NEXT:    v_rcp_f32_e32 v6, v6
-; GFX6-NEXT:    v_cndmask_b32_e64 v8, 0, -1, vcc
+; GFX6-NEXT:    v_mac_f32_e32 v4, 0x4f800000, v5
+; GFX6-NEXT:    v_rcp_f32_e32 v4, v4
+; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s17, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v5, v5, v8, vcc
+; GFX6-NEXT:    v_cndmask_b32_e32 v5, v6, v7, vcc
 ; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
 ; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v2, v4, v3, s[0:1]
-; GFX6-NEXT:    v_mul_f32_e32 v3, 0x5f7ffffc, v6
-; GFX6-NEXT:    v_mul_f32_e32 v4, 0x2f800000, v3
+; GFX6-NEXT:    v_mul_f32_e32 v2, 0x5f7ffffc, v4
+; GFX6-NEXT:    v_mul_f32_e32 v4, 0x2f800000, v2
 ; GFX6-NEXT:    v_trunc_f32_e32 v4, v4
-; GFX6-NEXT:    v_mac_f32_e32 v3, 0xcf800000, v4
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
+; GFX6-NEXT:    v_mac_f32_e32 v2, 0xcf800000, v4
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v4, v4
 ; GFX6-NEXT:    s_sub_u32 s0, 0, s4
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_mul_hi_u32 v2, s0, v3
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GFX6-NEXT:    v_mul_hi_u32 v3, s0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v4
 ; GFX6-NEXT:    s_subb_u32 s1, 0, s5
-; GFX6-NEXT:    v_mul_lo_u32 v6, s1, v3
+; GFX6-NEXT:    v_mul_lo_u32 v6, s1, v2
 ; GFX6-NEXT:    s_ashr_i32 s14, s7, 31
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
-; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v3
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v6, v2
-; GFX6-NEXT:    v_mul_lo_u32 v6, v3, v2
-; GFX6-NEXT:    v_mul_hi_u32 v7, v3, v5
-; GFX6-NEXT:    v_mul_hi_u32 v8, v3, v2
-; GFX6-NEXT:    v_mul_hi_u32 v9, v4, v2
-; GFX6-NEXT:    v_mul_lo_u32 v2, v4, v2
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
+; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v2
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v6, v3
+; GFX6-NEXT:    v_mul_lo_u32 v6, v2, v3
+; GFX6-NEXT:    v_mul_hi_u32 v7, v2, v5
+; GFX6-NEXT:    v_mul_hi_u32 v8, v2, v3
+; GFX6-NEXT:    v_mul_hi_u32 v9, v4, v3
+; GFX6-NEXT:    v_mul_lo_u32 v3, v4, v3
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GFX6-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v8, v4, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v4, v5
 ; GFX6-NEXT:    s_mov_b32 s15, s14
 ; GFX6-NEXT:    v_xor_b32_e32 v0, s12, v0
 ; GFX6-NEXT:    v_add_i32_e32 v6, vcc, v6, v8
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, v7, v5, vcc
 ; GFX6-NEXT:    v_addc_u32_e32 v6, vcc, 0, v9, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, v4, v5, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s0, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v5, s0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, s1, v2
 ; GFX6-NEXT:    v_xor_b32_e32 v1, s12, v1
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s0, v2
 ; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v8, v2, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v9, v2, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v10, v2, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v7, v3, v5
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v3, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v3, v4

 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GFX6-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, s4, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s5, v2
 ; GFX6-NEXT:    v_subrev_i32_e32 v0, vcc, s12, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v2, s4, v2
 ; GFX6-NEXT:    v_subb_u32_e32 v1, vcc, v1, v6, vcc
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GFX6-NEXT:    v_sub_i32_e32 v4, vcc, s7, v3
 ; GFX6-NEXT:    v_mov_b32_e32 v5, s5
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s6, v2
 ; GFX6-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, vcc
 ; GFX6-NEXT:    v_subrev_i32_e64 v6, s[0:1], s4, v2
 ; GFX6-NEXT:    v_subbrev_u32_e64 v7, s[2:3], 0, v4, s[0:1]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[2:3], s5, v7
 ; GFX6-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_le_u32_e64 s[2:3], s4, v6
 ; GFX6-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, s[0:1]
 ; GFX6-NEXT:    v_cndmask_b32_e64 v9, 0, -1, s[2:3]
 ; GFX6-NEXT:    v_cmp_eq_u32_e64 s[2:3], s5, v7
 ; GFX6-NEXT:    v_subrev_i32_e64 v5, s[0:1], s4, v6
 ; GFX6-NEXT:    v_cndmask_b32_e64 v8, v8, v9, s[2:3]
 ; GFX6-NEXT:    v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1]
 ; GFX6-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v8
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v7, v4, s[0:1]
-; GFX6-NEXT:    v_mov_b32_e32 v7, s7
-; GFX6-NEXT:    v_subb_u32_e32 v3, vcc, v7, v3, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v5, v6, v5, s[0:1]
+; GFX6-NEXT:    v_mov_b32_e32 v6, s7
+; GFX6-NEXT:    v_subb_u32_e32 v3, vcc, v6, v3, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s5, v3
-; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s4, v2
-; GFX6-NEXT:    v_cndmask_b32_e64 v8, 0, -1, vcc
+; GFX6-NEXT:    v_cndmask_b32_e64 v4, v7, v4, s[0:1]
+; GFX6-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
 ; GFX6-NEXT:    v_cmp_eq_u32_e32 vcc, s5, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v7, v7, v8, vcc
-; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v7
+; GFX6-NEXT:    v_cndmask_b32_e32 v6, v6, v7, vcc
+; GFX6-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
+; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
 ; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX6-NEXT:    v_cndmask_b32_e64 v4, v6, v5, s[0:1]
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
 ; GFX6-NEXT:    v_xor_b32_e32 v2, s14, v2
 ; GFX6-NEXT:    v_xor_b32_e32 v3, s14, v3
 ; GFX6-NEXT:    v_mov_b32_e32 v4, s14
 ; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s14, v2
 ; GFX6-NEXT:    v_subb_u32_e32 v3, vcc, v3, v4, vcc
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; GFX6-NEXT:    s_endpgm

 ; GFX9-NEXT:    v_cndmask_b32_e64 v2, 0, -1, s[2:3]
 ; GFX9-NEXT:    v_mov_b32_e32 v3, s17
 ; GFX9-NEXT:    s_cselect_b64 s[2:3], -1, 0
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
 ; GFX9-NEXT:    v_cndmask_b32_e64 v2, v3, v2, s[2:3]
 ; GFX9-NEXT:    s_subb_u32 s2, s4, s13
 ; GFX9-NEXT:    v_subrev_co_u32_e64 v3, s[0:1], s12, v1
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
-; GFX9-NEXT:    s_subb_u32 s0, s2, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s2, s5, s15
-; GFX9-NEXT:    s_cmp_ge_u32 s2, s13
-; GFX9-NEXT:    v_mov_b32_e32 v5, s16
-; GFX9-NEXT:    v_mov_b32_e32 v6, s0
+; GFX9-NEXT:    s_subb_u32 s2, s2, 0
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v2
-; GFX9-NEXT:    s_cselect_b32 s3, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v2, s16
+; GFX9-NEXT:    v_mov_b32_e32 v3, s2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s5, s15
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s13
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
 ; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s12, v0
-; GFX9-NEXT:    s_cmp_eq_u32 s2, s13
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v6, s3
+; GFX9-NEXT:    s_cmp_eq_u32 s0, s13
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v5, v3, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s0
 ; GFX9-NEXT:    s_ashr_i32 s0, s11, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v6, v5, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v6, s2
 ; GFX9-NEXT:    s_add_u32 s2, s10, s0
 ; GFX9-NEXT:    s_mov_b32 s1, s0
 ; GFX9-NEXT:    s_addc_u32 s3, s11, s0
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v3
 ; GFX9-NEXT:    s_xor_b64 s[4:5], s[2:3], s[0:1]
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s4
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v3, s5
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v2, v5, v2, vcc
 ; GFX9-NEXT:    v_xor_b32_e32 v0, s14, v0
 ; GFX9-NEXT:    v_xor_b32_e32 v2, s14, v2
 ; GFX9-NEXT:    v_mac_f32_e32 v1, 0x4f800000, v3
 ; GFX9-NEXT:    v_rcp_f32_e32 v3, v1
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s14
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s14, v0
 ; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v2, v5, vcc
 ; GFX9-NEXT:    v_mul_f32_e32 v2, 0x5f7ffffc, v3

 ; GFX9-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[2:3]
 ; GFX9-NEXT:    v_mov_b32_e32 v6, s13
 ; GFX9-NEXT:    s_cselect_b64 s[2:3], -1, 0
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
 ; GFX9-NEXT:    v_cndmask_b32_e64 v5, v6, v5, s[2:3]
 ; GFX9-NEXT:    s_subb_u32 s2, s6, s5
 ; GFX9-NEXT:    v_subrev_co_u32_e64 v6, s[0:1], s4, v3
 ; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
-; GFX9-NEXT:    s_subb_u32 s0, s2, 0
-; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
-; GFX9-NEXT:    s_subb_u32 s2, s7, s11
-; GFX9-NEXT:    s_cmp_ge_u32 s2, s5
-; GFX9-NEXT:    v_mov_b32_e32 v7, s12
-; GFX9-NEXT:    v_mov_b32_e32 v8, s0
+; GFX9-NEXT:    s_subb_u32 s2, s2, 0
 ; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    s_cselect_b32 s3, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v5, s12
+; GFX9-NEXT:    v_mov_b32_e32 v6, s2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v5, v5, v6, s[0:1]
+; GFX9-NEXT:    s_subb_u32 s0, s7, s11
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s5
+; GFX9-NEXT:    s_cselect_b32 s1, -1, 0
 ; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v2
-; GFX9-NEXT:    s_cmp_eq_u32 s2, s5
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, v7, v8, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v8, s3
+; GFX9-NEXT:    s_cmp_eq_u32 s0, s5
+; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v7, s1
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v7, v8, v7, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v8, s2
+; GFX9-NEXT:    v_cndmask_b32_e32 v6, v7, v6, vcc
+; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
+; GFX9-NEXT:    v_mov_b32_e32 v7, s0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v8, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v7, v5, vcc
 ; GFX9-NEXT:    v_xor_b32_e32 v2, s10, v2
 ; GFX9-NEXT:    v_xor_b32_e32 v3, s10, v5
 ; GFX9-NEXT:    v_mov_b32_e32 v5, s10
 ; GFX9-NEXT:    v_subrev_co_u32_e32 v2, vcc, s10, v2
 ; GFX9-NEXT:    v_subb_co_u32_e32 v3, vcc, v3, v5, vcc
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[8:9]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl <2 x i64> <i64 4096, i64 4096>, %y
   %r = srem <2 x i64> %x, %shl.y
   store <2 x i64> %r, <2 x i64> addrspace(1)* %out
   ret void
 }

llvm/test/CodeGen/AMDGPU/dagcombine-select.ll

define amdgpu_kernel void @select_and3(i32 addrspace(1)* %p, i32 %x, i32 %y) {
   %c = icmp slt i32 %x, 11
   %s = select i1 %c, i32 -1, i32 0
   %a = and i32 %y, %s
   store i32 %a, i32 addrspace(1)* %p, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_and_v4:
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
+; GCN:     s_cselect_b32 s[[SEL0:[0-9]+]], s{{[0-9]+}}, 0
+; GCN:     s_cselect_b32 s[[SEL1:[0-9]+]], s{{[0-9]+}}, 0
+; GCN:     s_cselect_b32 s[[SEL2:[0-9]+]], s{{[0-9]+}}, 0
+; GCN:     s_cselect_b32 s[[SEL3:[0-9]+]], s{{[0-9]+}}, 0
+; GCN:     v_mov_b32_e32 v[[V0:[0-9]+]], s[[SEL3]]
+; GCN:     v_mov_b32_e32 v[[V1:[0-9]+]], s[[SEL2]]
+; GCN:     v_mov_b32_e32 v[[V2:[0-9]+]], s[[SEL1]]
+; GCN:     v_mov_b32_e32 v[[V3:[0-9]+]], s[[SEL0]]
 ; GCN-NOT: v_and_b32
-; GCN:     store_dword
+; GCN:     global_store_dwordx4 v{{[0-9]+}}, v[[[V0]]:[[V3]]]
 define amdgpu_kernel void @select_and_v4(<4 x i32> addrspace(1)* %p, i32 %x, <4 x i32> %y) {
   %c = icmp slt i32 %x, 11
   %s = select i1 %c, <4 x i32> zeroinitializer, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
   %a = and <4 x i32> %s, %y
   store <4 x i32> %a, <4 x i32> addrspace(1)* %p, align 32
   ret void
 }
 
define amdgpu_kernel void @select_or3(i32 addrspace(1)* %p, i32 %x, i32 %y) {
   %c = icmp slt i32 %x, 11
   %s = select i1 %c, i32 -1, i32 0
   %a = or i32 %y, %s
   store i32 %a, i32 addrspace(1)* %p, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_or_v4:
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
-; GCN:     v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
+; GCN:     s_cselect_b32 s[[SEL0:[0-9]+]], s{{[0-9]+}}, -1
+; GCN:     s_cselect_b32 s[[SEL1:[0-9]+]], s{{[0-9]+}}, -1
+; GCN:     s_cselect_b32 s[[SEL2:[0-9]+]], s{{[0-9]+}}, -1
+; GCN:     s_cselect_b32 s[[SEL3:[0-9]+]], s{{[0-9]+}}, -1
 ; GCN-NOT: v_or_b32
-; GCN:     store_dword
+; GCN:     v_mov_b32_e32 v[[V0:[0-9]+]], s[[SEL3]]
+; GCN:     v_mov_b32_e32 v[[V1:[0-9]+]], s[[SEL2]]
+; GCN:     v_mov_b32_e32 v[[V2:[0-9]+]], s[[SEL1]]
+; GCN:     v_mov_b32_e32 v[[V3:[0-9]+]], s[[SEL0]]
+; GCN:     global_store_dwordx4 v{{[0-9]+}}, v[[[V0]]:[[V3]]]
 define amdgpu_kernel void @select_or_v4(<4 x i32> addrspace(1)* %p, i32 %x, <4 x i32> %y) {
   %c = icmp slt i32 %x, 11
   %s = select i1 %c, <4 x i32> zeroinitializer, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
   %a = or <4 x i32> %s, %y
   store <4 x i32> %a, <4 x i32> addrspace(1)* %p, align 32
   ret void
 }
 

 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants_v2i16(<2 x i16> addrspace(1)* %p, i1 %cond) {
   %sel = select i1 %cond, <2 x i16> <i16 -4, i16 2>, <2 x i16> <i16 3, i16 1>
   %bo = sub <2 x i16> <i16 5, i16 7>, %sel
   store <2 x i16> %bo, <2 x i16> addrspace(1)* %p, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}sel_constants_sub_constant_sel_constants_v4i32:
-; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 9,
-; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 6, 5,
-; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 10, 6,
-; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 14, 7,
+; GCN:     s_cselect_b32 s[[SEL0:[0-9]+]], 7, 14
+; GCN:     s_cselect_b32 s[[SEL1:[0-9]+]], 6, 10
+; GCN:     s_cselect_b32 s[[SEL2:[0-9]+]], 5, 6
+; GCN:     s_cselect_b32 s[[SEL3:[0-9]+]], 9, 2
+; GCN:     v_mov_b32_e32 v[[V0:[0-9]+]], s[[SEL3]]
+; GCN:     v_mov_b32_e32 v[[V1:[0-9]+]], s[[SEL2]]
+; GCN:     v_mov_b32_e32 v[[V2:[0-9]+]], s[[SEL1]]
+; GCN:     v_mov_b32_e32 v[[V3:[0-9]+]], s[[SEL0]]
+; GCN:     global_store_dwordx4 v{{[0-9]+}}, v[[[V0]]:[[V3]]]
 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants_v4i32(<4 x i32> addrspace(1)* %p, i1 %cond) {
   %sel = select i1 %cond, <4 x i32> <i32 -4, i32 2, i32 3, i32 4>, <4 x i32> <i32 3, i32 1, i32 -1, i32 -3>
   %bo = sub <4 x i32> <i32 5, i32 7, i32 9, i32 11>, %sel
   store <4 x i32> %bo, <4 x i32> addrspace(1)* %p, align 32
   ret void
 }
 
 ; GCN-LABEL: {{^}}sdiv_constant_sel_constants_i64:

 define amdgpu_kernel void @fsub_constant_sel_constants_v2f16(<2 x half> addrspace(1)* %p, i1 %cond) {
   %sel = select i1 %cond, <2 x half> <half -2.0, half -3.0>, <2 x half> <half -1.0, half 4.0>
   %bo = fsub <2 x half> <half -1.0, half 2.0>, %sel
   store <2 x half> %bo, <2 x half> addrspace(1)* %p, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}fsub_constant_sel_constants_v4f32:
-; GCN-DAG: v_mov_b32_e32 [[T2:v[0-9]+]], 0x40a00000
-; GCN-DAG: v_mov_b32_e32 [[T3:v[0-9]+]], 0x41100000
-; GCN-DAG: v_mov_b32_e32 [[T4:v[0-9]+]], 0x41500000
-; GCN-DAG: v_mov_b32_e32 [[F4:v[0-9]+]], 0x40c00000
-; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1.0,
-; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 2.0, [[T2]],
-; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 4.0, [[T3]],
-; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, [[F4]], [[T4]],
+; GCN:     s_mov_b32 [[T0:s[0-9]+]], 0x41500000
+; GCN:     s_cselect_b32 s[[SEL0:[0-9]+]], [[T0]], 0x40c00000
+; GCN:     s_cselect_b32 s[[SEL1:[0-9]+]], 0x41100000, 4.0
+; GCN:     s_cselect_b32 s[[SEL2:[0-9]+]], 0x40a00000, 2.0
+; GCN:     s_cselect_b32 s[[SEL3:[0-9]+]], 1.0, 0
+; GCN:     v_mov_b32_e32 v[[V0:[0-9]+]], s[[SEL3]]
+; GCN:     v_mov_b32_e32 v[[V1:[0-9]+]], s[[SEL2]]
+; GCN:     v_mov_b32_e32 v[[V2:[0-9]+]], s[[SEL1]]
+; GCN:     v_mov_b32_e32 v[[V3:[0-9]+]], s[[SEL0]]
+; GCN:     global_store_dwordx4 v{{[0-9]+}}, v[[[V0]]:[[V3]]]
 define amdgpu_kernel void @fsub_constant_sel_constants_v4f32(<4 x float> addrspace(1)* %p, i1 %cond) {
   %sel = select i1 %cond, <4 x float> <float -2.0, float -3.0, float -4.0, float -5.0>, <4 x float> <float -1.0, float 0.0, float 1.0, float 2.0>
   %bo = fsub <4 x float> <float -1.0, float 2.0, float 5.0, float 8.0>, %sel
   store <4 x float> %bo, <4 x float> addrspace(1)* %p, align 32
   ret void
 }
 
 ; GCN-LABEL: {{^}}fdiv_constant_sel_constants:

llvm/test/CodeGen/AMDGPU/expand-scalar-carry-out-select-user.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx704 < %s | FileCheck -check-prefix=GFX7 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefix=GFX9 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s
 
 define i32 @s_add_co_select_user() {
 ; GFX7-LABEL: s_add_co_select_user:
 ; GFX7:       ; %bb.0: ; %bb
 ; GFX7-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX7-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX7-NEXT:    s_load_dword s6, s[4:5], 0x0
 ; GFX7-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX7-NEXT:    v_add_i32_e64 v0, s[4:5], s6, s6
 ; GFX7-NEXT:    s_or_b32 s4, s4, s5
 ; GFX7-NEXT:    s_cmp_lg_u32 s4, 0
-; GFX7-NEXT:    s_addc_u32 s4, s6, 0
-; GFX7-NEXT:    v_mov_b32_e32 v1, s4
-; GFX7-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX7-NEXT:    s_addc_u32 s7, s6, 0
+; GFX7-NEXT:    s_cselect_b64 s[4:5], -1, 0
+; GFX7-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX7-NEXT:    s_cselect_b32 s4, s7, 0
 ; GFX7-NEXT:    s_cmp_gt_u32 s6, 31
-; GFX7-NEXT:    v_cndmask_b32_e32 v1, 0, v1, vcc
+; GFX7-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX7-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX7-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
 ; GFX7-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX9-LABEL: s_add_co_select_user:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX9-NEXT:    s_load_dword s6, s[4:5], 0x0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    v_add_co_u32_e64 v0, s[4:5], s6, s6
 ; GFX9-NEXT:    s_cmp_lg_u64 s[4:5], 0
-; GFX9-NEXT:    s_addc_u32 s4, s6, 0
-; GFX9-NEXT:    v_mov_b32_e32 v1, s4
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    s_addc_u32 s7, s6, 0
+; GFX9-NEXT:    s_cselect_b64 s[4:5], -1, 0
+; GFX9-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GFX9-NEXT:    s_cselect_b32 s4, s7, 0
 ; GFX9-NEXT:    s_cmp_gt_u32 s6, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, 0, v1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s4
 ; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
 ; GFX9-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: s_add_co_select_user:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX10-NEXT:    s_mov_b64 s[4:5], 0
 ; GFX10-NEXT:    s_load_dword s4, s[4:5], 0x0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_add_co_u32 v0, s5, s4, s4
 ; GFX10-NEXT:    s_cmpk_lg_u32 s5, 0x0
 ; GFX10-NEXT:    s_addc_u32 s5, s4, 0
 ; GFX10-NEXT:    s_cselect_b32 s6, -1, 0
+; GFX10-NEXT:    s_and_b32 s6, s6, exec_lo
+; GFX10-NEXT:    s_cselect_b32 s5, s5, 0
 ; GFX10-NEXT:    s_cmp_gt_u32 s4, 31
-; GFX10-NEXT:    v_cndmask_b32_e64 v1, 0, s5, s6
 ; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
-; GFX10-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc_lo
+; GFX10-NEXT:    v_cndmask_b32_e32 v0, s5, v0, vcc_lo
 ; GFX10-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX11-LABEL: s_add_co_select_user:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-NEXT:    s_mov_b64 s[0:1], 0
 ; GFX11-NEXT:    s_load_b32 s0, s[0:1], 0x0
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    v_add_co_u32 v0, s1, s0, s0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_cmpk_lg_u32 s1, 0x0
 ; GFX11-NEXT:    s_addc_u32 s1, s0, 0
 ; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
+; GFX11-NEXT:    s_and_b32 s2, s2, exec_lo
+; GFX11-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX11-NEXT:    s_cmp_gt_u32 s0, 31
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, 0, s1, s2
 ; GFX11-NEXT:    s_cselect_b32 vcc_lo, -1, 0
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_cndmask_b32_e32 v0, v1, v0, vcc_lo
+; GFX11-NEXT:    v_cndmask_b32_e32 v0, s1, v0, vcc_lo
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %i = load volatile i32, i32 addrspace(4)* null, align 8
   %i1 = add i32 %i, %i
   %i2 = icmp ult i32 %i1, %i
   %i3 = zext i1 %i2 to i32
   %i4 = add nuw nsw i32 %i3, 0
   %i5 = add i32 %i4, %i

llvm/test/CodeGen/AMDGPU/extract_vector_dynelt.ll

 entry:
   %ext = extractelement <4 x i32> <i32 0, i32 1, i32 2, i32 4>, i32 %sel
   store i32 %ext, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}double4_extelt:
 ; GCN-NOT: buffer_
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
-; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 3
-; GCN-DAG: s_cselect_b64 [[C3:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C3]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x3f847ae1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x47ae147b
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0xe147ae14, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x4000147a, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[[0-9]+}}, 3
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x40100a3d, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x70a3d70a, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @double4_extelt(double addrspace(1)* %out, i32 %sel) {
 entry:
   %ext = extractelement <4 x double> <double 0.01, double 1.01, double 2.01, double 4.01>, i32 %sel
   store double %ext, double addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}double5_extelt:
 ; GCN-NOT: buffer_
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
-; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 3
-; GCN-DAG: s_cselect_b64 [[C3:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 4
-; GCN-DAG: s_cselect_b64 [[C4:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C3]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C4]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x3f847ae1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x47ae147b
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0xe147ae14, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x4000147a, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[[0-9]+}}, 3
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x40100a3d, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x70a3d70a, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[[0-9]+}}, 4
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, 0x40140a3d, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @double5_extelt(double addrspace(1)* %out, i32 %sel) {
 entry:
   %ext = extractelement <5 x double> <double 0.01, double 1.01, double 2.01, double 4.01, double 5.01>, i32 %sel
   store double %ext, double addrspace(1)* %out
   ret void
 }
 

 entry:
   %ext = extractelement <2 x float> <float 0.0, float 1.0>, i32 %sel
   store float %ext, float addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}double2_extelt:
 ; GCN-NOT: buffer_
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, {{[^,]+}}, {{[^,]+}}, [[C1]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x3f847ae1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x47ae147b
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @double2_extelt(double addrspace(1)* %out, i32 %sel) {
 entry:
   %ext = extractelement <2 x double> <double 0.01, double 1.01>, i32 %sel
   store double %ext, double addrspace(1)* %out
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/extract_vector_elt-f64.ll

 ; RUN: llc -march=amdgcn -mtriple=amdgcn-- -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
 ; RUN: llc -march=amdgcn -mtriple=amdgcn-- -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
 
 ; GCN-LABEL: {{^}}extract_vector_elt_v3f64_2:
 ; GCN: buffer_load_dwordx4
 ; GCN: buffer_load_dwordx2
 ; GCN: buffer_store_dwordx2
 define amdgpu_kernel void @extract_vector_elt_v3f64_2(double addrspace(1)* %out, <3 x double> addrspace(1)* %in) #0 {
   %ld = load volatile <3 x double>, <3 x double> addrspace(1)* %in
   %elt = extractelement <3 x double> %ld, i32 2
   store volatile double %elt, double addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}dyn_extract_vector_elt_v3f64:
 ; GCN-NOT: buffer_load
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
-; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @dyn_extract_vector_elt_v3f64(double addrspace(1)* %out, <3 x double> %foo, i32 %elt) #0 {
   %dynelt = extractelement <3 x double> %foo, i32 %elt
   store volatile double %dynelt, double addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}dyn_extract_vector_elt_v4f64:
 ; GCN-NOT: buffer_load
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
-; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 3
-; GCN-DAG: s_cselect_b64 [[C3:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C3]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C3]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 3
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @dyn_extract_vector_elt_v4f64(double addrspace(1)* %out, <4 x double> %foo, i32 %elt) #0 {
   %dynelt = extractelement <4 x double> %foo, i32 %elt
   store volatile double %dynelt, double addrspace(1)* %out
   ret void
 }
 
 attributes #0 = { nounwind }

llvm/test/CodeGen/AMDGPU/extract_vector_elt-i64.ll

define amdgpu_kernel void @extract_vector_elt_v2i64(i64 addrspace(1)* %out, <2 x i64> %foo) #0 {
   %out1 = getelementptr i64, i64 addrspace(1)* %out, i32 1
   store volatile i64 %p1, i64 addrspace(1)* %out
   store volatile i64 %p0, i64 addrspace(1)* %out1
   ret void
 }
 
 ; GCN-LABEL: {{^}}dyn_extract_vector_elt_v2i64:
 ; GCN-NOT: buffer_load
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @dyn_extract_vector_elt_v2i64(i64 addrspace(1)* %out, <2 x i64> %foo, i32 %elt) #0 {
   %dynelt = extractelement <2 x i64> %foo, i32 %elt
   store volatile i64 %dynelt, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}dyn_extract_vector_elt_v2i64_2:
define amdgpu_kernel void @dyn_extract_vector_elt_v2i64_2(i64 addrspace(1)* %out, <2 x i64> addrspace(1)* %foo, i32 %elt, <2 x i64> %arst) #0 {
   %or = or <2 x i64> %load, %arst
   %dynelt = extractelement <2 x i64> %or, i32 %elt
   store volatile i64 %dynelt, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}dyn_extract_vector_elt_v3i64:
 ; GCN-NOT: buffer_load
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
-; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @dyn_extract_vector_elt_v3i64(i64 addrspace(1)* %out, <3 x i64> %foo, i32 %elt) #0 {
   %dynelt = extractelement <3 x i64> %foo, i32 %elt
   store volatile i64 %dynelt, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}dyn_extract_vector_elt_v4i64:
 ; GCN-NOT: buffer_load
-; GCN-DAG: s_cmp_eq_u32 [[IDX:s[0-9]+]], 1
-; GCN-DAG: s_cselect_b64 [[C1:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 2
-; GCN-DAG: s_cselect_b64 [[C2:[^,]+]], -1, 0
-; GCN-DAG: s_cmp_eq_u32 [[IDX]], 3
-; GCN-DAG: s_cselect_b64 [[C3:[^,]+]], -1, 0
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C1]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C2]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C3]]
-; GCN-DAG: v_cndmask_b32_e{{32|64}} v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[C3]]
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 1
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 2
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 3
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN-DAG: s_cselect_b32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN: store_dwordx2 v[{{[0-9:]+}}]
 define amdgpu_kernel void @dyn_extract_vector_elt_v4i64(i64 addrspace(1)* %out, <4 x i64> %foo, i32 %elt) #0 {
   %dynelt = extractelement <4 x i64> %foo, i32 %elt
   store volatile i64 %dynelt, i64 addrspace(1)* %out
   ret void
 }
 
 attributes #0 = { nounwind }

llvm/test/CodeGen/AMDGPU/fceil64.ll

 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=CI -check-prefix=FUNC %s
 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=CI -check-prefix=FUNC %s
 
 declare double @llvm.ceil.f64(double) nounwind readnone
 declare <2 x double> @llvm.ceil.v2f64(<2 x double>) nounwind readnone
 declare <3 x double> @llvm.ceil.v3f64(<3 x double>) nounwind readnone
 declare <4 x double> @llvm.ceil.v4f64(<4 x double>) nounwind readnone
 declare <8 x double> @llvm.ceil.v8f64(<8 x double>) nounwind readnone
 declare <16 x double> @llvm.ceil.v16f64(<16 x double>) nounwind readnone
 
 ; FUNC-LABEL: {{^}}fceil_f64:
 ; CI: v_ceil_f64_e32
 ; SI: s_bfe_u32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
 ; SI-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
-; FIXME: We should be using s_addk_i32 here, but the reg allocation hints
-;        are not always followed.
-; SI-DAG: s_add_i32 [[SEXP0:s[0-9]+]], [[SEXP]], 0xfffffc01
-; SI-DAG: s_lshr_b64 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], [[SEXP0]]
+; SI-DAG: s_addk_i32 [[SEXP]], 0xfc01
+; SI-DAG: s_lshr_b64 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], [[SEXP]]
 ; SI-DAG: s_andn2_b64
 ; SI-DAG: cmp_gt_i32
-; SI-DAG: cndmask_b32
-; SI-DAG: cndmask_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
 ; SI-DAG: cmp_lt_i32
-; SI-DAG: cndmask_b32
-; SI-DAG: cndmask_b32
-; SI-DAG: v_cmp_gt_f64
-; SI-DAG: v_cmp_lg_f64
-; SI-DAG: v_cndmask_b32
-; SI: v_cndmask_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: v_cmp_gt_f64_e64 [[FCMP:s[[0-9]+:[0-9]+]]]
+; SI-DAG: v_cmp_lg_f64_e32 vcc
+; SI-DAG: s_and_b64 [[AND1:s[[0-9]+:[0-9]+]]], [[FCMP]], vcc
+; SI-DAG: s_and_b64 [[AND1]], [[AND1]], exec
+; SI-DAG: s_cselect_b32 s{{[0-9]+}}, 0x3ff00000, 0
 ; SI: v_add_f64
 ; SI: s_endpgm
 define amdgpu_kernel void @fceil_f64(double addrspace(1)* %out, double %x) {
   %y = call double @llvm.ceil.f64(double %x) nounwind readnone
   store double %y, double addrspace(1)* %out
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/frem.ll

 ; SI-NEXT:    v_mul_f64 v[10:11], v[8:9], v[6:7]
 ; SI-NEXT:    v_fma_f64 v[12:13], -v[4:5], v[10:11], v[8:9]
 ; SI-NEXT:    v_cmp_eq_u32_e32 vcc, v3, v5
 ; SI-NEXT:    v_cmp_eq_u32_e64 s[0:1], v1, v9
 ; SI-NEXT:    s_xor_b64 vcc, s[0:1], vcc
 ; SI-NEXT:    s_nop 1
 ; SI-NEXT:    v_div_fmas_f64 v[4:5], v[12:13], v[6:7], v[10:11]
 ; SI-NEXT:    v_div_fixup_f64 v[4:5], v[4:5], v[2:3], v[0:1]
-; SI-NEXT:    v_bfe_u32 v6, v5, 20, 11
-; SI-NEXT:    v_add_i32_e32 v8, vcc, 0xfffffc01, v6
+; SI-NEXT:    v_readfirstlane_b32 s2, v5
+; SI-NEXT:    s_bfe_u32 s0, s2, 0xb0014
+; SI-NEXT:    s_add_i32 s3, s0, 0xfffffc01
 ; SI-NEXT:    s_mov_b32 s1, 0xfffff
 ; SI-NEXT:    s_mov_b32 s0, s6
-; SI-NEXT:    v_lshr_b64 v[6:7], s[0:1], v8
-; SI-NEXT:    v_not_b32_e32 v6, v6
+; SI-NEXT:    s_lshr_b64 s[0:1], s[0:1], s3
+; SI-NEXT:    v_not_b32_e32 v6, s0
 ; SI-NEXT:    v_and_b32_e32 v6, v4, v6
-; SI-NEXT:    v_not_b32_e32 v7, v7
-; SI-NEXT:    v_and_b32_e32 v7, v5, v7
-; SI-NEXT:    v_and_b32_e32 v9, 0x80000000, v5
-; SI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v8
-; SI-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
-; SI-NEXT:    v_cmp_lt_i32_e64 s[0:1], 51, v8
-; SI-NEXT:    v_cndmask_b32_e64 v5, v7, v5, s[0:1]
+; SI-NEXT:    v_not_b32_e32 v7, s1
+; SI-NEXT:    v_and_b32_e32 v5, v5, v7
+; SI-NEXT:    s_and_b32 s0, s2, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s3, 0
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
 ; SI-NEXT:    v_cndmask_b32_e64 v6, v6, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v4, v6, v4, s[0:1]
+; SI-NEXT:    v_mov_b32_e32 v7, s0
+; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
+; SI-NEXT:    s_cmp_gt_i32 s3, 51
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_mov_b32_e32 v7, s2
+; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
 ; SI-NEXT:    v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: frem_f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
; GFX11-NEXT:    s_endpgm
    %r2 = frem double %r0, %r1
    store double %r2, double addrspace(1)* %out, align 8
    ret void
 }
 
 define amdgpu_kernel void @fast_frem_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
 ; SI-LABEL: fast_frem_f64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
-; SI-NEXT:    s_mov_b32 s7, 0xf000
-; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
+; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_mov_b32 s4, s8
-; SI-NEXT:    s_mov_b32 s5, s9
-; SI-NEXT:    s_mov_b32 s8, s10
-; SI-NEXT:    s_mov_b32 s9, s11
-; SI-NEXT:    s_mov_b32 s10, s6
-; SI-NEXT:    s_mov_b32 s11, s7
-; SI-NEXT:    s_mov_b32 s2, s6
-; SI-NEXT:    s_mov_b32 s3, s7
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[8:11], 0
-; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[0:3], 0
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
+; SI-NEXT:    s_mov_b32 s4, s6
+; SI-NEXT:    s_mov_b32 s5, s7
+; SI-NEXT:    s_mov_b32 s6, s2
+; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, s2
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[4:7], 0
+; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[8:11], 0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_rcp_f64_e32 v[4:5], v[2:3]
 ; SI-NEXT:    v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
 ; SI-NEXT:    v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
 ; SI-NEXT:    v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
 ; SI-NEXT:    v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
 ; SI-NEXT:    v_mul_f64 v[6:7], v[0:1], v[4:5]
 ; SI-NEXT:    v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
 ; SI-NEXT:    v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
-; SI-NEXT:    v_bfe_u32 v6, v5, 20, 11
-; SI-NEXT:    v_add_i32_e32 v8, vcc, 0xfffffc01, v6
-; SI-NEXT:    s_mov_b32 s1, 0xfffff
-; SI-NEXT:    s_mov_b32 s0, s6
-; SI-NEXT:    v_lshr_b64 v[6:7], s[0:1], v8
-; SI-NEXT:    v_not_b32_e32 v6, v6
+; SI-NEXT:    v_readfirstlane_b32 s6, v5
+; SI-NEXT:    s_bfe_u32 s4, s6, 0xb0014
+; SI-NEXT:    s_add_i32 s7, s4, 0xfffffc01
+; SI-NEXT:    s_mov_b32 s5, 0xfffff
+; SI-NEXT:    s_mov_b32 s4, s2
+; SI-NEXT:    s_lshr_b64 s[4:5], s[4:5], s7
+; SI-NEXT:    v_not_b32_e32 v6, s4
 ; SI-NEXT:    v_and_b32_e32 v6, v4, v6
-; SI-NEXT:    v_not_b32_e32 v7, v7
-; SI-NEXT:    v_and_b32_e32 v7, v5, v7
-; SI-NEXT:    v_and_b32_e32 v9, 0x80000000, v5
-; SI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v8
-; SI-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
-; SI-NEXT:    v_cmp_lt_i32_e64 s[0:1], 51, v8
-; SI-NEXT:    v_cndmask_b32_e64 v5, v7, v5, s[0:1]
+; SI-NEXT:    v_not_b32_e32 v7, s5
+; SI-NEXT:    v_and_b32_e32 v5, v5, v7
+; SI-NEXT:    s_and_b32 s4, s6, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
 ; SI-NEXT:    v_cndmask_b32_e64 v6, v6, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v4, v6, v4, s[0:1]
+; SI-NEXT:    v_mov_b32_e32 v7, s4
+; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_mov_b32_e32 v7, s6
+; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
 ; SI-NEXT:    v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: fast_frem_f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
 ; CI-NEXT:    s_mov_b32 s11, 0xf000
 ; CI-NEXT:    s_mov_b32 s10, -1
; GFX11-NEXT:    s_endpgm
    %r2 = frem fast double %r0, %r1
    store double %r2, double addrspace(1)* %out, align 8
    ret void
 }
 
 define amdgpu_kernel void @unsafe_frem_f64(double addrspace(1)* %out, double addrspace(1)* %in1,
 ; SI-LABEL: unsafe_frem_f64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0x9
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
-; SI-NEXT:    s_mov_b32 s7, 0xf000
-; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; SI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
+; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_mov_b32 s4, s8
-; SI-NEXT:    s_mov_b32 s5, s9
-; SI-NEXT:    s_mov_b32 s8, s10
-; SI-NEXT:    s_mov_b32 s9, s11
-; SI-NEXT:    s_mov_b32 s10, s6
-; SI-NEXT:    s_mov_b32 s11, s7
-; SI-NEXT:    s_mov_b32 s2, s6
-; SI-NEXT:    s_mov_b32 s3, s7
-; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[8:11], 0
-; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[0:3], 0
+; SI-NEXT:    s_mov_b32 s0, s4
+; SI-NEXT:    s_mov_b32 s1, s5
+; SI-NEXT:    s_mov_b32 s4, s6
+; SI-NEXT:    s_mov_b32 s5, s7
+; SI-NEXT:    s_mov_b32 s6, s2
+; SI-NEXT:    s_mov_b32 s7, s3
+; SI-NEXT:    s_mov_b32 s10, s2
+; SI-NEXT:    s_mov_b32 s11, s3
+; SI-NEXT:    buffer_load_dwordx2 v[0:1], off, s[4:7], 0
+; SI-NEXT:    buffer_load_dwordx2 v[2:3], off, s[8:11], 0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_rcp_f64_e32 v[4:5], v[2:3]
 ; SI-NEXT:    v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
 ; SI-NEXT:    v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
 ; SI-NEXT:    v_fma_f64 v[6:7], -v[2:3], v[4:5], 1.0
 ; SI-NEXT:    v_fma_f64 v[4:5], v[6:7], v[4:5], v[4:5]
 ; SI-NEXT:    v_mul_f64 v[6:7], v[0:1], v[4:5]
 ; SI-NEXT:    v_fma_f64 v[8:9], -v[2:3], v[6:7], v[0:1]
 ; SI-NEXT:    v_fma_f64 v[4:5], v[8:9], v[4:5], v[6:7]
-; SI-NEXT:    v_bfe_u32 v6, v5, 20, 11
-; SI-NEXT:    v_add_i32_e32 v8, vcc, 0xfffffc01, v6
-; SI-NEXT:    s_mov_b32 s1, 0xfffff
-; SI-NEXT:    s_mov_b32 s0, s6
-; SI-NEXT:    v_lshr_b64 v[6:7], s[0:1], v8
-; SI-NEXT:    v_not_b32_e32 v6, v6
+; SI-NEXT:    v_readfirstlane_b32 s6, v5
+; SI-NEXT:    s_bfe_u32 s4, s6, 0xb0014
+; SI-NEXT:    s_add_i32 s7, s4, 0xfffffc01
+; SI-NEXT:    s_mov_b32 s5, 0xfffff
+; SI-NEXT:    s_mov_b32 s4, s2
+; SI-NEXT:    s_lshr_b64 s[4:5], s[4:5], s7
+; SI-NEXT:    v_not_b32_e32 v6, s4
 ; SI-NEXT:    v_and_b32_e32 v6, v4, v6
-; SI-NEXT:    v_not_b32_e32 v7, v7
-; SI-NEXT:    v_and_b32_e32 v7, v5, v7
-; SI-NEXT:    v_and_b32_e32 v9, 0x80000000, v5
-; SI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v8
-; SI-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
-; SI-NEXT:    v_cmp_lt_i32_e64 s[0:1], 51, v8
-; SI-NEXT:    v_cndmask_b32_e64 v5, v7, v5, s[0:1]
+; SI-NEXT:    v_not_b32_e32 v7, s5
+; SI-NEXT:    v_and_b32_e32 v5, v5, v7
+; SI-NEXT:    s_and_b32 s4, s6, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
 ; SI-NEXT:    v_cndmask_b32_e64 v6, v6, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v4, v6, v4, s[0:1]
+; SI-NEXT:    v_mov_b32_e32 v7, s4
+; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_mov_b32_e32 v7, s6
+; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v4, v6, v4, vcc
 ; SI-NEXT:    v_fma_f64 v[0:1], -v[4:5], v[2:3], v[0:1]
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: unsafe_frem_f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
 ; CI-NEXT:    s_mov_b32 s11, 0xf000
 ; CI-NEXT:    s_mov_b32 s10, -1

 ; SI-NEXT:    v_mul_f64 v[14:15], v[12:13], v[10:11]
 ; SI-NEXT:    v_fma_f64 v[16:17], -v[8:9], v[14:15], v[12:13]
 ; SI-NEXT:    v_cmp_eq_u32_e32 vcc, v7, v9
 ; SI-NEXT:    v_cmp_eq_u32_e64 s[0:1], v3, v13
 ; SI-NEXT:    s_xor_b64 vcc, s[0:1], vcc
 ; SI-NEXT:    s_nop 1
 ; SI-NEXT:    v_div_fmas_f64 v[8:9], v[16:17], v[10:11], v[14:15]
 ; SI-NEXT:    v_div_fixup_f64 v[8:9], v[8:9], v[6:7], v[2:3]
-; SI-NEXT:    v_bfe_u32 v10, v9, 20, 11
-; SI-NEXT:    v_add_i32_e32 v12, vcc, 0xfffffc01, v10
+; SI-NEXT:    v_readfirstlane_b32 s8, v9
+; SI-NEXT:    s_bfe_u32 s0, s8, 0xb0014
+; SI-NEXT:    s_add_i32 s9, s0, 0xfffffc01
 ; SI-NEXT:    s_mov_b32 s3, 0xfffff
-; SI-NEXT:    v_lshr_b64 v[10:11], s[2:3], v12
-; SI-NEXT:    v_not_b32_e32 v10, v10
+; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s9
+; SI-NEXT:    v_not_b32_e32 v10, s0
 ; SI-NEXT:    v_and_b32_e32 v10, v8, v10
-; SI-NEXT:    v_not_b32_e32 v11, v11
-; SI-NEXT:    v_and_b32_e32 v11, v9, v11
-; SI-NEXT:    v_and_b32_e32 v13, 0x80000000, v9
-; SI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v12
-; SI-NEXT:    v_cndmask_b32_e32 v11, v11, v13, vcc
-; SI-NEXT:    v_cmp_lt_i32_e64 s[0:1], 51, v12
-; SI-NEXT:    v_cndmask_b32_e64 v9, v11, v9, s[0:1]
+; SI-NEXT:    v_not_b32_e32 v11, s1
+; SI-NEXT:    v_and_b32_e32 v9, v9, v11
+; SI-NEXT:    s_and_b32 s0, s8, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s9, 0
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
 ; SI-NEXT:    v_cndmask_b32_e64 v10, v10, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v8, v10, v8, s[0:1]
+; SI-NEXT:    v_mov_b32_e32 v11, s0
+; SI-NEXT:    v_cndmask_b32_e32 v9, v9, v11, vcc
+; SI-NEXT:    s_cmp_gt_i32 s9, 51
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_mov_b32_e32 v11, s8
+; SI-NEXT:    v_cndmask_b32_e32 v9, v9, v11, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v8, v10, v8, vcc
 ; SI-NEXT:    v_fma_f64 v[2:3], -v[8:9], v[6:7], v[2:3]
 ; SI-NEXT:    v_div_scale_f64 v[6:7], s[0:1], v[4:5], v[4:5], v[0:1]
 ; SI-NEXT:    v_rcp_f64_e32 v[8:9], v[6:7]
 ; SI-NEXT:    v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
 ; SI-NEXT:    v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
 ; SI-NEXT:    v_fma_f64 v[10:11], -v[6:7], v[8:9], 1.0
 ; SI-NEXT:    v_fma_f64 v[8:9], v[8:9], v[10:11], v[8:9]
 ; SI-NEXT:    v_div_scale_f64 v[10:11], s[0:1], v[0:1], v[4:5], v[0:1]
 ; SI-NEXT:    v_mul_f64 v[12:13], v[10:11], v[8:9]
 ; SI-NEXT:    v_fma_f64 v[14:15], -v[6:7], v[12:13], v[10:11]
 ; SI-NEXT:    v_cmp_eq_u32_e32 vcc, v5, v7
 ; SI-NEXT:    v_cmp_eq_u32_e64 s[0:1], v1, v11
 ; SI-NEXT:    s_xor_b64 vcc, s[0:1], vcc
 ; SI-NEXT:    s_nop 1
 ; SI-NEXT:    v_div_fmas_f64 v[6:7], v[14:15], v[8:9], v[12:13]
 ; SI-NEXT:    v_div_fixup_f64 v[6:7], v[6:7], v[4:5], v[0:1]
-; SI-NEXT:    v_bfe_u32 v8, v7, 20, 11
-; SI-NEXT:    v_add_i32_e32 v10, vcc, 0xfffffc01, v8
-; SI-NEXT:    v_lshr_b64 v[8:9], s[2:3], v10
-; SI-NEXT:    v_not_b32_e32 v8, v8
+; SI-NEXT:    v_readfirstlane_b32 s8, v7
+; SI-NEXT:    s_bfe_u32 s0, s8, 0xb0014
+; SI-NEXT:    s_add_i32 s9, s0, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s9
+; SI-NEXT:    v_not_b32_e32 v8, s0
 ; SI-NEXT:    v_and_b32_e32 v8, v6, v8
-; SI-NEXT:    v_not_b32_e32 v9, v9
-; SI-NEXT:    v_and_b32_e32 v9, v7, v9
-; SI-NEXT:    v_and_b32_e32 v11, 0x80000000, v7
-; SI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v10
-; SI-NEXT:    v_cndmask_b32_e32 v9, v9, v11, vcc
-; SI-NEXT:    v_cmp_lt_i32_e64 s[0:1], 51, v10
-; SI-NEXT:    v_cndmask_b32_e64 v7, v9, v7, s[0:1]
+; SI-NEXT:    v_not_b32_e32 v9, s1
+; SI-NEXT:    v_and_b32_e32 v7, v7, v9
+; SI-NEXT:    s_and_b32 s0, s8, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s9, 0
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
 ; SI-NEXT:    v_cndmask_b32_e64 v8, v8, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v6, v8, v6, s[0:1]
+; SI-NEXT:    v_mov_b32_e32 v9, s0
+; SI-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
+; SI-NEXT:    s_cmp_gt_i32 s9, 51
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_mov_b32_e32 v9, s8
+; SI-NEXT:    v_cndmask_b32_e32 v7, v7, v9, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v6, v8, v6, vcc
 ; SI-NEXT:    v_fma_f64 v[0:1], -v[6:7], v[4:5], v[0:1]
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: frem_v2f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; CI-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd

llvm/test/CodeGen/AMDGPU/ftrunc.f64.ll

 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=bonaire < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
 
 declare double @llvm.trunc.f64(double) nounwind readnone
 declare <2 x double> @llvm.trunc.v2f64(<2 x double>) nounwind readnone
 declare <3 x double> @llvm.trunc.v3f64(<3 x double>) nounwind readnone
 declare <4 x double> @llvm.trunc.v4f64(<4 x double>) nounwind readnone
 declare <8 x double> @llvm.trunc.v8f64(<8 x double>) nounwind readnone
 declare <16 x double> @llvm.trunc.v16f64(<16 x double>) nounwind readnone
 
 ; FUNC-LABEL: {{^}}v_ftrunc_f64:
 ; CI: v_trunc_f64
-; SI: v_bfe_u32 {{v[0-9]+}}, {{v[0-9]+}}, 20, 11
+; SI: s_bfe_u32 {{s[0-9]+}}, {{s[0-9]+}}, 0xb0014
 ; SI: s_endpgm
 define amdgpu_kernel void @v_ftrunc_f64(double addrspace(1)* %out, double addrspace(1)* %in) {
   %x = load double, double addrspace(1)* %in, align 8
   %y = call double @llvm.trunc.f64(double %x) nounwind readnone
   store double %y, double addrspace(1)* %out, align 8
   ret void
 }
 
 ; FUNC-LABEL: {{^}}ftrunc_f64:
 ; CI: v_trunc_f64_e32
 
 ; SI: s_bfe_u32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
 ; SI-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
 ; SI-DAG: s_add_i32 [[SEXP1:s[0-9]+]], [[SEXP]], 0xfffffc01
 ; SI-DAG: s_lshr_b64 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], [[SEXP1]]
 ; SI-DAG: s_andn2_b64
 ; SI-DAG: cmp_gt_i32
-; SI-DAG: cndmask_b32
-; SI-DAG: cndmask_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
 ; SI-DAG: cmp_lt_i32
-; SI-DAG: cndmask_b32
-; SI-DAG: cndmask_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
 ; SI: s_endpgm
 define amdgpu_kernel void @ftrunc_f64(double addrspace(1)* %out, double %x) {
   %y = call double @llvm.trunc.f64(double %x) nounwind readnone
   store double %y, double addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}ftrunc_v2f64:

llvm/test/CodeGen/AMDGPU/idiv-licm.ll

 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX9-NEXT:    v_readfirstlane_b32 s6, v0
 ; GFX9-NEXT:    s_mul_i32 s7, s5, s6
 ; GFX9-NEXT:    s_mul_hi_u32 s7, s6, s7
 ; GFX9-NEXT:    s_add_i32 s6, s6, s7
 ; GFX9-NEXT:    s_mul_i32 s7, s3, s6
 ; GFX9-NEXT:    s_mul_hi_u32 s6, s2, s6
 ; GFX9-NEXT:    s_add_i32 s6, s6, s7
+; GFX9-NEXT:    s_not_b32 s9, s6
 ; GFX9-NEXT:    s_mul_i32 s7, s5, s6
+; GFX9-NEXT:    s_mul_i32 s9, s4, s9
+; GFX9-NEXT:    s_add_i32 s8, s6, 1
 ; GFX9-NEXT:    s_add_i32 s7, s2, s7
+; GFX9-NEXT:    s_add_i32 s9, s2, s9
 ; GFX9-NEXT:    s_cmp_ge_u32 s7, s4
-; GFX9-NEXT:    v_mov_b32_e32 v2, s6
-; GFX9-NEXT:    v_mov_b32_e32 v3, s7
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    s_add_i32 s7, s6, 1
-; GFX9-NEXT:    s_not_b32 s6, s6
-; GFX9-NEXT:    s_mul_i32 s6, s4, s6
-; GFX9-NEXT:    v_mov_b32_e32 v4, s7
-; GFX9-NEXT:    s_add_i32 s6, s2, s6
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v4, s6
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX9-NEXT:    v_add_u32_e32 v5, 1, v2
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX9-NEXT:    s_cselect_b32 s7, s9, s7
+; GFX9-NEXT:    s_add_i32 s8, s6, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s7, s4
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s6
 ; GFX9-NEXT:    s_add_u32 s2, s2, 1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v3
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    s_addc_u32 s3, s3, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
 ; GFX9-NEXT:    global_store_dword v1, v2, s[0:1]
 ; GFX9-NEXT:    s_add_u32 s0, s0, 4
 ; GFX9-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX9-NEXT:    s_cmpk_eq_i32 s2, 0x400
 ; GFX9-NEXT:    s_cbranch_scc0 .LBB0_1
 ; GFX9-NEXT:  ; %bb.2: ; %bb2
 ; GFX9-NEXT:    s_endpgm
 ;

 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX10-NEXT:    v_readfirstlane_b32 s6, v0
 ; GFX10-NEXT:    s_mul_i32 s7, s5, s6
 ; GFX10-NEXT:    s_mul_hi_u32 s7, s6, s7
 ; GFX10-NEXT:    s_add_i32 s6, s6, s7
 ; GFX10-NEXT:    s_mul_i32 s7, s3, s6
 ; GFX10-NEXT:    s_mul_hi_u32 s6, s2, s6
 ; GFX10-NEXT:    s_add_i32 s6, s6, s7
+; GFX10-NEXT:    s_not_b32 s8, s6
 ; GFX10-NEXT:    s_mul_i32 s7, s5, s6
+; GFX10-NEXT:    s_mul_i32 s8, s4, s8
 ; GFX10-NEXT:    s_add_i32 s7, s2, s7
+; GFX10-NEXT:    s_add_i32 s9, s6, 1
+; GFX10-NEXT:    s_add_i32 s8, s2, s8
 ; GFX10-NEXT:    s_cmp_ge_u32 s7, s4
-; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX10-NEXT:    s_cselect_b32 s6, s9, s6
+; GFX10-NEXT:    s_cselect_b32 s7, s8, s7
 ; GFX10-NEXT:    s_add_i32 s8, s6, 1
-; GFX10-NEXT:    s_not_b32 s9, s6
-; GFX10-NEXT:    v_mov_b32_e32 v2, s8
-; GFX10-NEXT:    s_mul_i32 s8, s4, s9
-; GFX10-NEXT:    s_add_i32 s8, s2, s8
+; GFX10-NEXT:    s_cmp_ge_u32 s7, s4
+; GFX10-NEXT:    s_cselect_b32 s6, s8, s6
 ; GFX10-NEXT:    s_add_u32 s2, s2, 1
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, s6, v2, vcc_lo
+; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    s_addc_u32 s3, s3, 0
-; GFX10-NEXT:    v_cndmask_b32_e32 v3, s7, v3, vcc_lo
-; GFX10-NEXT:    v_add_nc_u32_e32 v4, 1, v2
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s4, v3
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
 ; GFX10-NEXT:    global_store_dword v1, v2, s[0:1]
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_add_u32 s0, s0, 4
 ; GFX10-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-NEXT:    s_cmpk_eq_i32 s2, 0x400
 ; GFX10-NEXT:    s_cbranch_scc0 .LBB0_1
 ; GFX10-NEXT:  ; %bb.2: ; %bb2
 ; GFX10-NEXT:    s_endpgm

 ; GFX11-NEXT:    s_mul_i32 s7, s5, s6
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_mul_hi_u32 s7, s6, s7
 ; GFX11-NEXT:    s_add_i32 s6, s6, s7
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_mul_i32 s7, s3, s6
 ; GFX11-NEXT:    s_mul_hi_u32 s6, s2, s6
 ; GFX11-NEXT:    s_add_i32 s6, s6, s7
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_not_b32 s8, s6
 ; GFX11-NEXT:    s_mul_i32 s7, s5, s6
+; GFX11-NEXT:    s_mul_i32 s8, s4, s8
 ; GFX11-NEXT:    s_add_i32 s7, s2, s7
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_add_i32 s9, s6, 1
+; GFX11-NEXT:    s_add_i32 s8, s2, s8
 ; GFX11-NEXT:    s_cmp_ge_u32 s7, s4
-; GFX11-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX11-NEXT:    s_cselect_b32 s6, s9, s6
+; GFX11-NEXT:    s_cselect_b32 s7, s8, s7
 ; GFX11-NEXT:    s_add_i32 s8, s6, 1
-; GFX11-NEXT:    s_not_b32 s9, s6
-; GFX11-NEXT:    v_mov_b32_e32 v2, s8
-; GFX11-NEXT:    s_mul_i32 s8, s4, s9
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    s_add_i32 s8, s2, s8
+; GFX11-NEXT:    s_cmp_ge_u32 s7, s4
+; GFX11-NEXT:    s_cselect_b32 s6, s8, s6
 ; GFX11-NEXT:    s_add_u32 s2, s2, 1
-; GFX11-NEXT:    v_mov_b32_e32 v3, s8
-; GFX11-NEXT:    v_cndmask_b32_e32 v2, s6, v2, vcc_lo
+; GFX11-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX11-NEXT:    s_addc_u32 s3, s3, 0
-; GFX11-NEXT:    v_dual_cndmask_b32 v3, s7, v3 :: v_dual_add_nc_u32 v4, 1, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_cmp_le_u32_e32 vcc_lo, s4, v3
-; GFX11-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
 ; GFX11-NEXT:    global_store_b32 v1, v2, s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, 4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX11-NEXT:    s_cmpk_eq_i32 s2, 0x400
 ; GFX11-NEXT:    s_cbranch_scc0 .LBB0_1
 ; GFX11-NEXT:  ; %bb.2: ; %bb2
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
bb3:                                              ; preds = %bb3, %bb
   %tmp8 = icmp eq i32 %tmp7, 1024
   br i1 %tmp8, label %bb2, label %bb3
 }
 
 define amdgpu_kernel void @sdiv32_invariant_denom(i32 addrspace(1)* nocapture %arg, i32 %arg1) {
 ; GFX9-LABEL: sdiv32_invariant_denom:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s3, s[0:1], 0x2c
+; GFX9-NEXT:    s_mov_b32 s4, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_ashr_i32 s2, s3, 31
 ; GFX9-NEXT:    s_add_i32 s3, s3, s2
 ; GFX9-NEXT:    s_xor_b32 s3, s3, s2
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX9-NEXT:    s_sub_i32 s4, 0, s3
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_sub_i32 s5, 0, s3
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX9-NEXT:    v_mul_lo_u32 v1, s4, v0
-; GFX9-NEXT:    s_mov_b32 s4, 0
-; GFX9-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v1
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX9-NEXT:  .LBB2_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_mul_hi_u32 v2, s4, v0
-; GFX9-NEXT:    v_mul_lo_u32 v3, v2, s3
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v2
-; GFX9-NEXT:    v_sub_u32_e32 v3, s4, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v4, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v2, s2, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX9-NEXT:    s_mul_i32 s7, s5, s6
+; GFX9-NEXT:    s_mul_hi_u32 s7, s6, s7
+; GFX9-NEXT:    s_add_i32 s6, s6, s7
+; GFX9-NEXT:    s_mul_hi_u32 s6, s4, s6
+; GFX9-NEXT:    s_mul_i32 s7, s6, s3
+; GFX9-NEXT:    s_sub_i32 s7, s4, s7
+; GFX9-NEXT:    s_add_i32 s8, s6, 1
+; GFX9-NEXT:    s_sub_i32 s9, s7, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX9-NEXT:    s_cselect_b32 s7, s9, s7
+; GFX9-NEXT:    s_add_i32 s8, s6, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX9-NEXT:    s_xor_b32 s6, s6, s2
+; GFX9-NEXT:    s_sub_i32 s6, s6, s2
 ; GFX9-NEXT:    s_add_i32 s4, s4, 1
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s2, v2
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    global_store_dword v1, v2, s[0:1]
 ; GFX9-NEXT:    s_add_u32 s0, s0, 4
 ; GFX9-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX9-NEXT:    s_cmpk_eq_i32 s4, 0x400
 ; GFX9-NEXT:    s_cbranch_scc0 .LBB2_1
 ; GFX9-NEXT:  ; %bb.2: ; %bb2
 ; GFX9-NEXT:    s_endpgm
 ;

 ; GFX10-NEXT:  .LBB2_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX10-NEXT:    v_readfirstlane_b32 s6, v0
 ; GFX10-NEXT:    s_mul_i32 s7, s5, s6
 ; GFX10-NEXT:    s_mul_hi_u32 s7, s6, s7
 ; GFX10-NEXT:    s_add_i32 s6, s6, s7
 ; GFX10-NEXT:    s_mul_hi_u32 s6, s4, s6
 ; GFX10-NEXT:    s_mul_i32 s7, s6, s3
+; GFX10-NEXT:    s_add_i32 s8, s6, 1
 ; GFX10-NEXT:    s_sub_i32 s7, s4, s7
+; GFX10-NEXT:    s_sub_i32 s9, s7, s3
 ; GFX10-NEXT:    s_cmp_ge_u32 s7, s3
-; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX10-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX10-NEXT:    s_cselect_b32 s7, s9, s7
 ; GFX10-NEXT:    s_add_i32 s8, s6, 1
+; GFX10-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX10-NEXT:    s_cselect_b32 s6, s8, s6
 ; GFX10-NEXT:    s_add_i32 s4, s4, 1
-; GFX10-NEXT:    v_mov_b32_e32 v2, s8
-; GFX10-NEXT:    s_sub_i32 s8, s7, s3
-; GFX10-NEXT:    v_mov_b32_e32 v3, s8
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, s6, v2, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e32 v3, s7, v3, vcc_lo
-; GFX10-NEXT:    v_add_nc_u32_e32 v4, 1, v2
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s3, v3
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
-; GFX10-NEXT:    v_xor_b32_e32 v2, s2, v2
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v2, s2, v2
+; GFX10-NEXT:    s_xor_b32 s6, s6, s2
+; GFX10-NEXT:    s_sub_i32 s6, s6, s2
+; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    global_store_dword v1, v2, s[0:1]
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_add_u32 s0, s0, 4
 ; GFX10-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX10-NEXT:    s_cmpk_eq_i32 s4, 0x400
 ; GFX10-NEXT:    s_cbranch_scc0 .LBB2_1
 ; GFX10-NEXT:  ; %bb.2: ; %bb2
 ; GFX10-NEXT:    s_endpgm

 ; GFX11-NEXT:    v_readfirstlane_b32 s6, v0
 ; GFX11-NEXT:    s_mul_i32 s7, s5, s6
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_mul_hi_u32 s7, s6, s7
 ; GFX11-NEXT:    s_add_i32 s6, s6, s7
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
 ; GFX11-NEXT:    s_mul_hi_u32 s6, s4, s6
 ; GFX11-NEXT:    s_mul_i32 s7, s6, s3
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    s_add_i32 s8, s6, 1
 ; GFX11-NEXT:    s_sub_i32 s7, s4, s7
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT:    s_sub_i32 s9, s7, s3
 ; GFX11-NEXT:    s_cmp_ge_u32 s7, s3
-; GFX11-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX11-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX11-NEXT:    s_cselect_b32 s7, s9, s7
 ; GFX11-NEXT:    s_add_i32 s8, s6, 1
+; GFX11-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX11-NEXT:    s_cselect_b32 s6, s8, s6
 ; GFX11-NEXT:    s_add_i32 s4, s4, 1
-; GFX11-NEXT:    v_mov_b32_e32 v2, s8
-; GFX11-NEXT:    s_sub_i32 s8, s7, s3
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_mov_b32_e32 v3, s8
-; GFX11-NEXT:    v_cndmask_b32_e32 v2, s6, v2, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_dual_cndmask_b32 v3, s7, v3 :: v_dual_add_nc_u32 v4, 1, v2
-; GFX11-NEXT:    v_cmp_le_u32_e32 vcc_lo, s3, v3
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
-; GFX11-NEXT:    v_xor_b32_e32 v2, s2, v2
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX11-NEXT:    v_subrev_nc_u32_e32 v2, s2, v2
+; GFX11-NEXT:    s_xor_b32 s6, s6, s2
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    s_sub_i32 s6, s6, s2
+; GFX11-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX11-NEXT:    global_store_b32 v1, v2, s[0:1]
 ; GFX11-NEXT:    s_add_u32 s0, s0, 4
 ; GFX11-NEXT:    s_addc_u32 s1, s1, 0
 ; GFX11-NEXT:    s_cmpk_eq_i32 s4, 0x400
 ; GFX11-NEXT:    s_cbranch_scc0 .LBB2_1
 ; GFX11-NEXT:  ; %bb.2: ; %bb2
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm

llvm/test/CodeGen/AMDGPU/implicit-kernarg-backend-usage.ll

 define amdgpu_kernel void @addrspacecast(i32 addrspace(5)* %ptr.private, i32 addrspace(3)* %ptr.local) {
 ; GFX8V3-LABEL: addrspacecast:
 ; GFX8V3:       ; %bb.0:
 ; GFX8V3-NEXT:    s_load_dwordx2 s[0:1], s[6:7], 0x0
 ; GFX8V3-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x40
 ; GFX8V3-NEXT:    v_mov_b32_e32 v4, 1
 ; GFX8V3-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8V3-NEXT:    s_cmp_lg_u32 s0, -1
-; GFX8V3-NEXT:    v_mov_b32_e32 v0, s3
-; GFX8V3-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX8V3-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
-; GFX8V3-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8V3-NEXT:    s_cselect_b32 s3, s3, 0
+; GFX8V3-NEXT:    s_cselect_b32 s0, s0, 0
 ; GFX8V3-NEXT:    s_cmp_lg_u32 s1, -1
-; GFX8V3-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX8V3-NEXT:    v_mov_b32_e32 v2, s2
-; GFX8V3-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX8V3-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
+; GFX8V3-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8V3-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8V3-NEXT:    s_cselect_b32 s0, s2, 0
+; GFX8V3-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX8V3-NEXT:    v_mov_b32_e32 v2, s1
-; GFX8V3-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GFX8V3-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX8V3-NEXT:    flat_store_dword v[0:1], v4
 ; GFX8V3-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8V3-NEXT:    v_mov_b32_e32 v0, 2
 ; GFX8V3-NEXT:    flat_store_dword v[2:3], v0
 ; GFX8V3-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8V3-NEXT:    s_endpgm
 ;
 ; GFX8V4-LABEL: addrspacecast:
 ; GFX8V4:       ; %bb.0:
 ; GFX8V4-NEXT:    s_load_dwordx2 s[0:1], s[6:7], 0x0
 ; GFX8V4-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x40
 ; GFX8V4-NEXT:    v_mov_b32_e32 v4, 1
 ; GFX8V4-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8V4-NEXT:    s_cmp_lg_u32 s0, -1
-; GFX8V4-NEXT:    v_mov_b32_e32 v0, s3
-; GFX8V4-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX8V4-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
-; GFX8V4-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8V4-NEXT:    s_cselect_b32 s3, s3, 0
+; GFX8V4-NEXT:    s_cselect_b32 s0, s0, 0
 ; GFX8V4-NEXT:    s_cmp_lg_u32 s1, -1
-; GFX8V4-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX8V4-NEXT:    v_mov_b32_e32 v2, s2
-; GFX8V4-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX8V4-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
+; GFX8V4-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8V4-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8V4-NEXT:    s_cselect_b32 s0, s2, 0
+; GFX8V4-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX8V4-NEXT:    v_mov_b32_e32 v2, s1
-; GFX8V4-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GFX8V4-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX8V4-NEXT:    flat_store_dword v[0:1], v4
 ; GFX8V4-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8V4-NEXT:    v_mov_b32_e32 v0, 2
 ; GFX8V4-NEXT:    flat_store_dword v[2:3], v0
 ; GFX8V4-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8V4-NEXT:    s_endpgm
 ;
 ; GFX8V5-LABEL: addrspacecast:
 ; GFX8V5:       ; %bb.0:
 ; GFX8V5-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX8V5-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0xc8
 ; GFX8V5-NEXT:    v_mov_b32_e32 v4, 1
 ; GFX8V5-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8V5-NEXT:    s_cmp_lg_u32 s0, -1
-; GFX8V5-NEXT:    v_mov_b32_e32 v0, s2
-; GFX8V5-NEXT:    v_mov_b32_e32 v2, s0
-; GFX8V5-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX8V5-NEXT:    s_cselect_b32 s2, s2, 0
+; GFX8V5-NEXT:    s_cselect_b32 s0, s0, 0
 ; GFX8V5-NEXT:    s_cmp_lg_u32 s1, -1
-; GFX8V5-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
-; GFX8V5-NEXT:    v_cndmask_b32_e32 v0, 0, v2, vcc
-; GFX8V5-NEXT:    v_mov_b32_e32 v2, s3
-; GFX8V5-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX8V5-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
+; GFX8V5-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8V5-NEXT:    v_mov_b32_e32 v1, s2
+; GFX8V5-NEXT:    s_cselect_b32 s0, s3, 0
+; GFX8V5-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX8V5-NEXT:    v_mov_b32_e32 v2, s1
-; GFX8V5-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GFX8V5-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX8V5-NEXT:    flat_store_dword v[0:1], v4
 ; GFX8V5-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8V5-NEXT:    v_mov_b32_e32 v0, 2
 ; GFX8V5-NEXT:    flat_store_dword v[2:3], v0
 ; GFX8V5-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8V5-NEXT:    s_endpgm
 ;
 ; GFX9V3-LABEL: addrspacecast:
 ; GFX9V3:       ; %bb.0:
 ; GFX9V3-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9V3-NEXT:    s_getreg_b32 s2, hwreg(HW_REG_SH_MEM_BASES, 0, 16)
 ; GFX9V3-NEXT:    s_lshl_b32 s2, s2, 16
-; GFX9V3-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9V3-NEXT:    v_mov_b32_e32 v4, 1
 ; GFX9V3-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9V3-NEXT:    s_cmp_lg_u32 s0, -1
-; GFX9V3-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9V3-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
+; GFX9V3-NEXT:    s_cselect_b32 s0, s0, 0
 ; GFX9V3-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9V3-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_SH_MEM_BASES, 16, 16)
+; GFX9V3-NEXT:    s_cselect_b32 s2, s2, 0
 ; GFX9V3-NEXT:    s_lshl_b32 s0, s0, 16
 ; GFX9V3-NEXT:    s_cmp_lg_u32 s1, -1
-; GFX9V3-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX9V3-NEXT:    v_mov_b32_e32 v2, s0
-; GFX9V3-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9V3-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
+; GFX9V3-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9V3-NEXT:    s_cselect_b32 s0, s0, 0
+; GFX9V3-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX9V3-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9V3-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GFX9V3-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX9V3-NEXT:    flat_store_dword v[0:1], v4
 ; GFX9V3-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9V3-NEXT:    v_mov_b32_e32 v0, 2
 ; GFX9V3-NEXT:    flat_store_dword v[2:3], v0
 ; GFX9V3-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9V3-NEXT:    s_endpgm
 ;
 ; GFX9V4-LABEL: addrspacecast:
 ; GFX9V4:       ; %bb.0:
 ; GFX9V4-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9V4-NEXT:    s_getreg_b32 s2, hwreg(HW_REG_SH_MEM_BASES, 0, 16)
 ; GFX9V4-NEXT:    s_lshl_b32 s2, s2, 16
-; GFX9V4-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9V4-NEXT:    v_mov_b32_e32 v4, 1
 ; GFX9V4-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9V4-NEXT:    s_cmp_lg_u32 s0, -1
-; GFX9V4-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9V4-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
+; GFX9V4-NEXT:    s_cselect_b32 s0, s0, 0
 ; GFX9V4-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9V4-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_SH_MEM_BASES, 16, 16)
+; GFX9V4-NEXT:    s_cselect_b32 s2, s2, 0
 ; GFX9V4-NEXT:    s_lshl_b32 s0, s0, 16
 ; GFX9V4-NEXT:    s_cmp_lg_u32 s1, -1
-; GFX9V4-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX9V4-NEXT:    v_mov_b32_e32 v2, s0
-; GFX9V4-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9V4-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
+; GFX9V4-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9V4-NEXT:    s_cselect_b32 s0, s0, 0
+; GFX9V4-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX9V4-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9V4-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GFX9V4-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX9V4-NEXT:    flat_store_dword v[0:1], v4
 ; GFX9V4-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9V4-NEXT:    v_mov_b32_e32 v0, 2
 ; GFX9V4-NEXT:    flat_store_dword v[2:3], v0
 ; GFX9V4-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9V4-NEXT:    s_endpgm
 ;
 ; GFX9V5-LABEL: addrspacecast:
 ; GFX9V5:       ; %bb.0:
 ; GFX9V5-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX9V5-NEXT:    s_getreg_b32 s2, hwreg(HW_REG_SH_MEM_BASES, 0, 16)
 ; GFX9V5-NEXT:    s_lshl_b32 s2, s2, 16
-; GFX9V5-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9V5-NEXT:    v_mov_b32_e32 v4, 1
 ; GFX9V5-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9V5-NEXT:    s_cmp_lg_u32 s0, -1
-; GFX9V5-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9V5-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
+; GFX9V5-NEXT:    s_cselect_b32 s0, s0, 0
 ; GFX9V5-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9V5-NEXT:    s_getreg_b32 s0, hwreg(HW_REG_SH_MEM_BASES, 16, 16)
+; GFX9V5-NEXT:    s_cselect_b32 s2, s2, 0
 ; GFX9V5-NEXT:    s_lshl_b32 s0, s0, 16
 ; GFX9V5-NEXT:    s_cmp_lg_u32 s1, -1
-; GFX9V5-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GFX9V5-NEXT:    v_mov_b32_e32 v2, s0
-; GFX9V5-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9V5-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
+; GFX9V5-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9V5-NEXT:    s_cselect_b32 s0, s0, 0
+; GFX9V5-NEXT:    s_cselect_b32 s1, s1, 0
 ; GFX9V5-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9V5-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GFX9V5-NEXT:    v_mov_b32_e32 v3, s0
 ; GFX9V5-NEXT:    flat_store_dword v[0:1], v4
 ; GFX9V5-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9V5-NEXT:    v_mov_b32_e32 v0, 2
 ; GFX9V5-NEXT:    flat_store_dword v[2:3], v0
 ; GFX9V5-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9V5-NEXT:    s_endpgm
   %flat.private = addrspacecast i32 addrspace(5)* %ptr.private to i32*
   %flat.local = addrspacecast i32 addrspace(3)* %ptr.local to i32*

llvm/test/CodeGen/AMDGPU/indirect-call-known-callees.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s
 
 ; We have an indirect call with a known set of callees, which are
 ; known to not need any special inputs. The ABI still needs to use the
 ; register
 
 ; FIXME: Passing real values for workitem ID, and 0s that can be undef
 
 define amdgpu_kernel void @indirect_call_known_no_special_inputs() {
 ; CHECK-LABEL: indirect_call_known_no_special_inputs:
 ; CHECK:       ; %bb.0: ; %bb
 ; CHECK-NEXT:    s_add_u32 flat_scratch_lo, s4, s7
 ; CHECK-NEXT:    s_addc_u32 flat_scratch_hi, s5, 0
-; CHECK-NEXT:    s_mov_b64 s[4:5], 0
-; CHECK-NEXT:    s_load_dword s4, s[4:5], 0x0
 ; CHECK-NEXT:    s_add_u32 s0, s0, s7
 ; CHECK-NEXT:    s_addc_u32 s1, s1, 0
-; CHECK-NEXT:    s_mov_b32 s33, s6
-; CHECK-NEXT:    v_mov_b32_e32 v31, v0
-; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
-; CHECK-NEXT:    s_bitcmp1_b32 s4, 0
-; CHECK-NEXT:    s_cselect_b64 vcc, -1, 0
+; CHECK-NEXT:    s_mov_b64 s[4:5], 0
+; CHECK-NEXT:    s_load_dword s7, s[4:5], 0x0
 ; CHECK-NEXT:    s_getpc_b64 s[4:5]
 ; CHECK-NEXT:    s_add_u32 s4, s4, wobble@gotpcrel32@lo+4
 ; CHECK-NEXT:    s_addc_u32 s5, s5, wobble@gotpcrel32@hi+12
-; CHECK-NEXT:    s_getpc_b64 s[6:7]
-; CHECK-NEXT:    s_add_u32 s6, s6, snork@gotpcrel32@lo+4
-; CHECK-NEXT:    s_addc_u32 s7, s7, snork@gotpcrel32@hi+12
-; CHECK-NEXT:    s_load_dwordx2 s[8:9], s[6:7], 0x0
-; CHECK-NEXT:    s_load_dwordx2 s[10:11], s[4:5], 0x0
-; CHECK-NEXT:    s_mov_b32 s32, 0
-; CHECK-NEXT:    s_mov_b64 s[4:5], exec
+; CHECK-NEXT:    s_getpc_b64 s[8:9]
+; CHECK-NEXT:    s_add_u32 s8, s8, snork@gotpcrel32@lo+4
+; CHECK-NEXT:    s_addc_u32 s9, s9, snork@gotpcrel32@hi+12
+; CHECK-NEXT:    s_load_dwordx2 s[10:11], s[8:9], 0x0
+; CHECK-NEXT:    s_load_dwordx2 s[12:13], s[4:5], 0x0
+; CHECK-NEXT:    s_mov_b64 s[8:9], 0
 ; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
-; CHECK-NEXT:    v_mov_b32_e32 v0, s9
-; CHECK-NEXT:    v_mov_b32_e32 v1, s11
-; CHECK-NEXT:    v_mov_b32_e32 v2, s8
-; CHECK-NEXT:    v_mov_b32_e32 v4, s10
-; CHECK-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; CHECK-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; CHECK-NEXT:    s_and_b32 s4, 1, s7
+; CHECK-NEXT:    s_cmp_eq_u32 s4, 1
+; CHECK-NEXT:    v_mov_b32_e32 v31, v0
+; CHECK-NEXT:    s_cselect_b32 s5, s13, s11
+; CHECK-NEXT:    s_cselect_b32 s4, s12, s10
+; CHECK-NEXT:    s_mov_b32 s12, s6
 ; CHECK-NEXT:    v_mov_b32_e32 v1, 0
-; CHECK-NEXT:  .LBB0_1: ; =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    v_readfirstlane_b32 s4, v2
-; CHECK-NEXT:    v_readfirstlane_b32 s5, v3
-; CHECK-NEXT:    v_cmp_eq_u64_e32 vcc, s[4:5], v[2:3]
-; CHECK-NEXT:    s_and_saveexec_b64 s[34:35], vcc
-; CHECK-NEXT:    s_mov_b64 s[8:9], 0
-; CHECK-NEXT:    s_mov_b32 s12, s33
-; CHECK-NEXT:    v_mov_b32_e32 v4, v1
+; CHECK-NEXT:    v_mov_b32_e32 v4, 0
+; CHECK-NEXT:    s_mov_b32 s32, 0
 ; CHECK-NEXT:    s_swappc_b64 s[30:31], s[4:5]
-; CHECK-NEXT:    ; implicit-def: $vgpr2_vgpr3
-; CHECK-NEXT:    ; implicit-def: $vgpr31
-; CHECK-NEXT:    ; implicit-def: $vgpr1
-; CHECK-NEXT:    s_xor_b64 exec, exec, s[34:35]
-; CHECK-NEXT:    s_cbranch_execnz .LBB0_1
-; CHECK-NEXT:  ; %bb.2:
 ; CHECK-NEXT:    s_endpgm
 
-; CHECK: .amdhsa_kernarg_size 0
-; CHECK-NEXT: .amdhsa_user_sgpr_count 6
-; CHECK-NEXT: .amdhsa_user_sgpr_private_segment_buffer 1
-; CHECK-NEXT: .amdhsa_user_sgpr_dispatch_ptr 0
-; CHECK-NEXT: .amdhsa_user_sgpr_queue_ptr 0
-; CHECK-NEXT: .amdhsa_user_sgpr_kernarg_segment_ptr 0
-; CHECK-NEXT: .amdhsa_user_sgpr_dispatch_id 0
-; CHECK-NEXT: .amdhsa_user_sgpr_flat_scratch_init 1
-; CHECK-NEXT: .amdhsa_user_sgpr_private_segment_size 0
-; CHECK-NEXT: .amdhsa_uses_dynamic_stack 1
-; CHECK-NEXT: .amdhsa_system_sgpr_private_segment_wavefront_offset 1
-; CHECK-NEXT: .amdhsa_system_sgpr_workgroup_id_x 1
-; CHECK-NEXT: .amdhsa_system_sgpr_workgroup_id_y 0
-; CHECK-NEXT: .amdhsa_system_sgpr_workgroup_id_z 0
-; CHECK-NEXT: .amdhsa_system_sgpr_workgroup_info 0
-; CHECK-NEXT: .amdhsa_system_vgpr_workitem_id 0
 bb:
   %cond = load i1, i1 addrspace(4)* null
   %tmp = select i1 %cond, void (i8*, i32, i8*)* bitcast (void ()* @wobble to void (i8*, i32, i8*)*), void (i8*, i32, i8*)* bitcast (void ()* @snork to void (i8*, i32, i8*)*)
   call void %tmp(i8* undef, i32 undef, i8* undef)
   ret void
 }
 
 define void @wobble() {

llvm/test/CodeGen/AMDGPU/insert_vector_dynelt.ll

 }
 
 define amdgpu_kernel void @double2_inselt(<2 x double> addrspace(1)* %out, <2 x double> %vec, i32 %sel) {
 ; GCN-LABEL: double2_inselt:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_dword s2, s[0:1], 0x44
 ; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x34
 ; GCN-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GCN-NEXT:    v_mov_b32_e32 v0, 0x3ff00000
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_cmp_eq_u32 s2, 1
-; GCN-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v1, v0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s6
+; GCN-NEXT:    s_cselect_b32 s3, 0x3ff00000, s7
+; GCN-NEXT:    s_cselect_b32 s6, 0, s6
 ; GCN-NEXT:    s_cmp_eq_u32 s2, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v1, 0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    s_cselect_b32 s2, 0x3ff00000, s5
+; GCN-NEXT:    s_cselect_b32 s4, 0, s4
 ; GCN-NEXT:    v_mov_b32_e32 v5, s1
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s2
+; GCN-NEXT:    v_mov_b32_e32 v2, s6
+; GCN-NEXT:    v_mov_b32_e32 v3, s3
 ; GCN-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
 entry:
   %v = insertelement <2 x double> %vec, double 1.000000e+00, i32 %sel
   store <2 x double> %v, <2 x double> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @double5_inselt(<5 x double> addrspace(1)* %out, <5 x double> %vec, i32 %sel) {
 ; GCN-LABEL: double5_inselt:
 ; GCN:       ; %bb.0: ; %entry
 ; GCN-NEXT:    s_load_dword s12, s[0:1], 0xa4
 ; GCN-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x84
 ; GCN-NEXT:    s_load_dwordx2 s[10:11], s[0:1], 0x24
 ; GCN-NEXT:    s_load_dwordx8 s[0:7], s[0:1], 0x64
-; GCN-NEXT:    v_mov_b32_e32 v4, 0x3ff00000
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_cmp_eq_u32 s12, 4
-; GCN-NEXT:    v_mov_b32_e32 v0, s9
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v9, v0, v4, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s8
+; GCN-NEXT:    s_cselect_b32 s9, 0x3ff00000, s9
+; GCN-NEXT:    s_cselect_b32 s8, 0, s8
 ; GCN-NEXT:    s_cmp_eq_u32 s12, 1
-; GCN-NEXT:    v_cndmask_b32_e64 v8, v0, 0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v0, v4, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s2
+; GCN-NEXT:    s_cselect_b32 s3, 0x3ff00000, s3
+; GCN-NEXT:    s_cselect_b32 s2, 0, s2
 ; GCN-NEXT:    s_cmp_eq_u32 s12, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v0, 0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s1
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v0, v4, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v4, s8
+; GCN-NEXT:    v_mov_b32_e32 v5, s9
+; GCN-NEXT:    s_cselect_b32 s8, 0x3ff00000, s1
+; GCN-NEXT:    s_cselect_b32 s9, 0, s0
 ; GCN-NEXT:    s_cmp_eq_u32 s12, 3
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v5, s7
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v7, v5, v4, vcc
-; GCN-NEXT:    v_mov_b32_e32 v5, s6
+; GCN-NEXT:    s_cselect_b32 s0, 0x3ff00000, s7
+; GCN-NEXT:    s_cselect_b32 s1, 0, s6
 ; GCN-NEXT:    s_cmp_eq_u32 s12, 2
-; GCN-NEXT:    v_cndmask_b32_e64 v6, v5, 0, vcc
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
+; GCN-NEXT:    s_cselect_b32 s5, 0x3ff00000, s5
+; GCN-NEXT:    s_cselect_b32 s4, 0, s4
+; GCN-NEXT:    v_mov_b32_e32 v3, s0
 ; GCN-NEXT:    s_add_u32 s0, s10, 16
-; GCN-NEXT:    v_mov_b32_e32 v5, s5
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
 ; GCN-NEXT:    s_addc_u32 s1, s11, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v5, v5, v4, vcc
-; GCN-NEXT:    v_mov_b32_e32 v4, s4
-; GCN-NEXT:    v_mov_b32_e32 v11, s1
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v10, s0
-; GCN-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
+; GCN-NEXT:    v_mov_b32_e32 v7, s1
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    v_mov_b32_e32 v6, s0
+; GCN-NEXT:    flat_store_dwordx4 v[6:7], v[0:3]
+; GCN-NEXT:    v_mov_b32_e32 v6, s10
+; GCN-NEXT:    v_mov_b32_e32 v0, s9
+; GCN-NEXT:    v_mov_b32_e32 v1, s8
+; GCN-NEXT:    v_mov_b32_e32 v2, s2
+; GCN-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-NEXT:    v_mov_b32_e32 v7, s11
 ; GCN-NEXT:    s_add_u32 s0, s10, 32
-; GCN-NEXT:    v_mov_b32_e32 v4, s10
-; GCN-NEXT:    v_mov_b32_e32 v5, s11
-; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
+; GCN-NEXT:    flat_store_dwordx4 v[6:7], v[0:3]
 ; GCN-NEXT:    s_addc_u32 s1, s11, 0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-NEXT:    flat_store_dwordx2 v[0:1], v[8:9]
+; GCN-NEXT:    flat_store_dwordx2 v[0:1], v[4:5]
 ; GCN-NEXT:    s_endpgm
 entry:
   %v = insertelement <5 x double> %vec, double 1.000000e+00, i32 %sel
   store <5 x double> %v, <5 x double> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @double8_inselt(<8 x double> addrspace(1)* %out, <8 x double> %vec, i32 %sel) {

llvm/test/CodeGen/AMDGPU/insert_vector_elt.ll

 define amdgpu_kernel void @dynamic_insertelement_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, i32 %b) nounwind {
 ; SI-LABEL: dynamic_insertelement_v2i32:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; SI-NEXT:    s_load_dword s8, s[4:5], 0x4
 ; SI-NEXT:    s_mov_b32 s7, 0x100f000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    v_mov_b32_e32 v0, s3
+; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    s_cmp_lg_u32 s8, 1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s0, s3, 5
 ; SI-NEXT:    s_cmp_lg_u32 s8, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v0, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    s_mov_b32 s5, s1
-; SI-NEXT:    v_cndmask_b32_e32 v0, 5, v0, vcc
+; SI-NEXT:    s_cselect_b32 s1, s2, 5
+; SI-NEXT:    v_mov_b32_e32 v0, s1
+; SI-NEXT:    v_mov_b32_e32 v1, s0
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v2i32:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; VI-NEXT:    s_load_dword s8, s[4:5], 0x10
 ; VI-NEXT:    s_mov_b32 s7, 0x1100f000

 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dword s8, s[4:5], 0x8
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x4
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; SI-NEXT:    s_mov_b32 s7, 0x100f000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    s_cmp_lg_u32 s8, 2
-; SI-NEXT:    v_mov_b32_e32 v0, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s2, s2, 5
 ; SI-NEXT:    s_cmp_lg_u32 s8, 1
-; SI-NEXT:    v_cndmask_b32_e32 v2, 5, v0, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s1, s1, 5
 ; SI-NEXT:    s_cmp_lg_u32 s8, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v0, vcc
+; SI-NEXT:    s_cselect_b32 s0, s0, 5
 ; SI-NEXT:    v_mov_b32_e32 v0, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v0, 5, v0, vcc
+; SI-NEXT:    v_mov_b32_e32 v1, s1
+; SI-NEXT:    v_mov_b32_e32 v2, s2
 ; SI-NEXT:    buffer_store_dwordx3 v[0:2], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v3i32:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dword s8, s[4:5], 0x20
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x10
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
; VI-NEXT:    s_endpgm
   %vecins = insertelement <3 x i32> %a, i32 5, i32 %b
   store <3 x i32> %vecins, <3 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @dynamic_insertelement_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %a, i32 %b, [8 x i32], i32 %val) nounwind {
 ; SI-LABEL: dynamic_insertelement_v4i32:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dword s6, s[4:5], 0x8
-; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
-; SI-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x4
-; SI-NEXT:    s_load_dword s4, s[4:5], 0x11
-; SI-NEXT:    s_mov_b32 s3, 0x100f000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x4
+; SI-NEXT:    s_load_dword s8, s[4:5], 0x8
+; SI-NEXT:    s_load_dword s9, s[4:5], 0x11
+; SI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; SI-NEXT:    s_mov_b32 s7, 0x100f000
+; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_cmp_eq_u32 s6, 3
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s11
-; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    s_cmp_eq_u32 s6, 2
-; SI-NEXT:    v_cndmask_b32_e32 v3, v0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s10
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_eq_u32 s6, 1
-; SI-NEXT:    v_cndmask_b32_e32 v2, v0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s9
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_eq_u32 s6, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, v0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s8
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; SI-NEXT:    s_cmp_eq_u32 s8, 3
+; SI-NEXT:    s_cselect_b32 s3, s9, s3
+; SI-NEXT:    s_cmp_eq_u32 s8, 2
+; SI-NEXT:    s_cselect_b32 s2, s9, s2
+; SI-NEXT:    s_cmp_eq_u32 s8, 1
+; SI-NEXT:    s_cselect_b32 s1, s9, s1
+; SI-NEXT:    s_cmp_eq_u32 s8, 0
+; SI-NEXT:    s_cselect_b32 s0, s9, s0
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_mov_b32_e32 v1, s1
+; SI-NEXT:    v_mov_b32_e32 v2, s2
+; SI-NEXT:    v_mov_b32_e32 v3, s3
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v4i32:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x10
 ; VI-NEXT:    s_load_dword s8, s[4:5], 0x20
 ; VI-NEXT:    s_load_dword s9, s[4:5], 0x44
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0

 ; SI-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x4
 ; SI-NEXT:    s_load_dword s6, s[4:5], 0x8
 ; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; SI-NEXT:    s_mov_b32 s3, 0x100f000
 ; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    s_lshr_b32 s4, s11, 24
 ; SI-NEXT:    s_cmp_lg_u32 s6, 15
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_lshr_b32 s4, s11, 16
+; SI-NEXT:    s_cselect_b32 s4, s4, 5
+; SI-NEXT:    s_lshl_b32 s4, s4, 8
+; SI-NEXT:    s_lshr_b32 s5, s11, 16
 ; SI-NEXT:    s_cmp_lg_u32 s6, 14
-; SI-NEXT:    v_cndmask_b32_e32 v0, 5, v0, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v1, vcc
-; SI-NEXT:    s_lshr_b32 s4, s11, 8
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; SI-NEXT:    s_cselect_b32 s5, s5, 5
+; SI-NEXT:    s_and_b32 s5, s5, 0xff
+; SI-NEXT:    s_or_b32 s4, s5, s4
+; SI-NEXT:    s_lshl_b32 s4, s4, 16
+; SI-NEXT:    s_lshr_b32 s5, s11, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 13
-; SI-NEXT:    v_or_b32_e32 v0, v1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s5, s5, 5
+; SI-NEXT:    s_lshl_b32 s5, s5, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 12
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, s11
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v2, 5, v2, vcc
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
-; SI-NEXT:    v_and_b32_e32 v2, 0xff, v2
-; SI-NEXT:    v_or_b32_e32 v1, v2, v1
-; SI-NEXT:    s_lshr_b32 s4, s10, 24
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; SI-NEXT:    s_cselect_b32 s7, s11, 5
+; SI-NEXT:    s_and_b32 s7, s7, 0xff
+; SI-NEXT:    s_or_b32 s5, s7, s5
+; SI-NEXT:    s_and_b32 s5, s5, 0xffff
+; SI-NEXT:    s_or_b32 s4, s5, s4
+; SI-NEXT:    s_lshr_b32 s5, s10, 24
 ; SI-NEXT:    s_cmp_lg_u32 s6, 11
-; SI-NEXT:    v_or_b32_e32 v3, v1, v0
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_lshr_b32 s4, s10, 16
+; SI-NEXT:    s_cselect_b32 s5, s5, 5
+; SI-NEXT:    s_lshl_b32 s5, s5, 8
+; SI-NEXT:    s_lshr_b32 s7, s10, 16
 ; SI-NEXT:    s_cmp_lg_u32 s6, 10
-; SI-NEXT:    v_cndmask_b32_e32 v0, 5, v0, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v1, vcc
-; SI-NEXT:    s_lshr_b32 s4, s10, 8
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; SI-NEXT:    s_cselect_b32 s7, s7, 5
+; SI-NEXT:    s_and_b32 s7, s7, 0xff
+; SI-NEXT:    s_or_b32 s5, s7, s5
+; SI-NEXT:    s_lshl_b32 s5, s5, 16
+; SI-NEXT:    s_lshr_b32 s7, s10, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 9
-; SI-NEXT:    v_or_b32_e32 v0, v1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s7, s7, 5
+; SI-NEXT:    s_lshl_b32 s7, s7, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 8
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, s10
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v2, 5, v2, vcc
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
-; SI-NEXT:    v_and_b32_e32 v2, 0xff, v2
-; SI-NEXT:    v_or_b32_e32 v1, v2, v1
-; SI-NEXT:    s_lshr_b32 s4, s9, 24
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; SI-NEXT:    s_cselect_b32 s10, s10, 5
+; SI-NEXT:    s_and_b32 s10, s10, 0xff
+; SI-NEXT:    s_or_b32 s7, s10, s7
+; SI-NEXT:    s_and_b32 s7, s7, 0xffff
+; SI-NEXT:    s_or_b32 s5, s7, s5
+; SI-NEXT:    s_lshr_b32 s7, s9, 24
 ; SI-NEXT:    s_cmp_lg_u32 s6, 7
-; SI-NEXT:    v_or_b32_e32 v2, v1, v0
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_lshr_b32 s4, s9, 16
+; SI-NEXT:    s_cselect_b32 s7, s7, 5
+; SI-NEXT:    s_lshl_b32 s7, s7, 8
+; SI-NEXT:    s_lshr_b32 s10, s9, 16
 ; SI-NEXT:    s_cmp_lg_u32 s6, 6
-; SI-NEXT:    v_cndmask_b32_e32 v0, 5, v0, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v1, vcc
-; SI-NEXT:    s_lshr_b32 s4, s9, 8
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xff, v1
+; SI-NEXT:    s_cselect_b32 s10, s10, 5
+; SI-NEXT:    s_and_b32 s10, s10, 0xff
+; SI-NEXT:    s_or_b32 s7, s10, s7
+; SI-NEXT:    s_lshl_b32 s7, s7, 16
+; SI-NEXT:    s_lshr_b32 s10, s9, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 5
-; SI-NEXT:    v_or_b32_e32 v0, v1, v0
-; SI-NEXT:    v_mov_b32_e32 v1, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s10, s10, 5
+; SI-NEXT:    s_lshl_b32 s10, s10, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 4
-; SI-NEXT:    v_cndmask_b32_e32 v1, 5, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v4, s9
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v4, 5, v4, vcc
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
-; SI-NEXT:    v_and_b32_e32 v4, 0xff, v4
-; SI-NEXT:    v_or_b32_e32 v1, v4, v1
-; SI-NEXT:    s_lshr_b32 s4, s8, 24
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
+; SI-NEXT:    s_cselect_b32 s9, s9, 5
+; SI-NEXT:    s_and_b32 s9, s9, 0xff
+; SI-NEXT:    s_or_b32 s9, s9, s10
+; SI-NEXT:    s_and_b32 s9, s9, 0xffff
+; SI-NEXT:    s_or_b32 s7, s9, s7
+; SI-NEXT:    s_lshr_b32 s9, s8, 24
 ; SI-NEXT:    s_cmp_lg_u32 s6, 3
-; SI-NEXT:    v_or_b32_e32 v1, v1, v0
-; SI-NEXT:    v_mov_b32_e32 v0, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_lshr_b32 s4, s8, 16
+; SI-NEXT:    s_cselect_b32 s9, s9, 5
+; SI-NEXT:    s_lshl_b32 s9, s9, 8
+; SI-NEXT:    s_lshr_b32 s10, s8, 16
 ; SI-NEXT:    s_cmp_lg_u32 s6, 2
-; SI-NEXT:    v_cndmask_b32_e32 v0, 5, v0, vcc
-; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v4, 5, v4, vcc
-; SI-NEXT:    s_lshr_b32 s4, s8, 8
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 8, v0
-; SI-NEXT:    v_and_b32_e32 v4, 0xff, v4
+; SI-NEXT:    s_cselect_b32 s10, s10, 5
+; SI-NEXT:    s_and_b32 s10, s10, 0xff
+; SI-NEXT:    s_or_b32 s9, s10, s9
+; SI-NEXT:    s_lshl_b32 s9, s9, 16
+; SI-NEXT:    s_lshr_b32 s10, s8, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 1
-; SI-NEXT:    v_or_b32_e32 v0, v4, v0
-; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s10, s10, 5
+; SI-NEXT:    s_lshl_b32 s10, s10, 8
 ; SI-NEXT:    s_cmp_lg_u32 s6, 0
-; SI-NEXT:    v_cndmask_b32_e32 v4, 5, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s8
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v5, 5, v5, vcc
-; SI-NEXT:    v_lshlrev_b32_e32 v4, 8, v4
-; SI-NEXT:    v_and_b32_e32 v5, 0xff, v5
-; SI-NEXT:    v_or_b32_e32 v4, v5, v4
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v4, 0xffff, v4
-; SI-NEXT:    v_or_b32_e32 v0, v4, v0
+; SI-NEXT:    s_cselect_b32 s6, s8, 5
+; SI-NEXT:    s_and_b32 s6, s6, 0xff
+; SI-NEXT:    s_or_b32 s6, s6, s10
+; SI-NEXT:    s_and_b32 s6, s6, 0xffff
+; SI-NEXT:    s_or_b32 s6, s6, s9
+; SI-NEXT:    v_mov_b32_e32 v0, s6
+; SI-NEXT:    v_mov_b32_e32 v1, s7
+; SI-NEXT:    v_mov_b32_e32 v2, s5
+; SI-NEXT:    v_mov_b32_e32 v3, s4
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v16i8:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x10
 ; VI-NEXT:    s_load_dword s6, s[4:5], 0x20
 ; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0

 }
 
 define amdgpu_kernel void @dynamic_insertelement_v2f64(<2 x double> addrspace(1)* %out, [8 x i32], <2 x double> %a, [8 x i32], i32 %b) nounwind {
 ; SI-LABEL: dynamic_insertelement_v2f64:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dword s8, s[4:5], 0x18
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0xc
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
-; SI-NEXT:    v_mov_b32_e32 v1, 0x40200000
 ; SI-NEXT:    s_mov_b32 s7, 0x100f000
+; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    s_cmp_eq_u32 s8, 1
-; SI-NEXT:    v_mov_b32_e32 v0, s3
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s2
+; SI-NEXT:    s_cselect_b32 s3, 0x40200000, s3
+; SI-NEXT:    s_cselect_b32 s2, 0, s2
 ; SI-NEXT:    s_cmp_eq_u32 s8, 0
-; SI-NEXT:    v_cndmask_b32_e64 v2, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s1
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
+; SI-NEXT:    s_cselect_b32 s1, 0x40200000, s1
+; SI-NEXT:    s_cselect_b32 s0, 0, s0
 ; SI-NEXT:    v_mov_b32_e32 v0, s0
-; SI-NEXT:    s_mov_b32 s6, -1
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; SI-NEXT:    v_mov_b32_e32 v1, s1
+; SI-NEXT:    v_mov_b32_e32 v2, s2
+; SI-NEXT:    v_mov_b32_e32 v3, s3
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v2f64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dword s8, s[4:5], 0x60
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x30
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
-; VI-NEXT:    v_mov_b32_e32 v1, 0x40200000
 ; VI-NEXT:    s_mov_b32 s7, 0x1100f000
+; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    s_cmp_eq_u32 s8, 1
-; VI-NEXT:    v_mov_b32_e32 v0, s3
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s2
+; VI-NEXT:    s_cselect_b32 s3, 0x40200000, s3
+; VI-NEXT:    s_cselect_b32 s2, 0, s2
 ; VI-NEXT:    s_cmp_eq_u32 s8, 0
-; VI-NEXT:    v_cndmask_b32_e64 v2, v0, 0, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s1
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
+; VI-NEXT:    s_cselect_b32 s1, 0x40200000, s1
+; VI-NEXT:    s_cselect_b32 s0, 0, s0
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    s_mov_b32 s6, -1
-; VI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_mov_b32_e32 v2, s2
+; VI-NEXT:    v_mov_b32_e32 v3, s3
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
   %vecins = insertelement <2 x double> %a, double 8.0, i32 %b
   store <2 x double> %vecins, <2 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @dynamic_insertelement_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> %a, i32 %b) nounwind {
 ; SI-LABEL: dynamic_insertelement_v2i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dword s10, s[4:5], 0x8
+; SI-NEXT:    s_load_dword s8, s[4:5], 0x8
 ; SI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x4
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; SI-NEXT:    s_mov_b32 s7, 0x100f000
 ; SI-NEXT:    s_mov_b32 s6, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_cmp_eq_u32 s10, 1
-; SI-NEXT:    v_mov_b32_e32 v0, s3
-; SI-NEXT:    s_cselect_b64 s[8:9], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v3, v0, 0, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v0, s2
-; SI-NEXT:    s_cmp_eq_u32 s10, 0
-; SI-NEXT:    v_cndmask_b32_e64 v2, v0, 5, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v0, s1
-; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[2:3]
+; SI-NEXT:    s_cmp_eq_u32 s8, 1
+; SI-NEXT:    s_cselect_b32 s3, 0, s3
+; SI-NEXT:    s_cselect_b32 s2, 5, s2
+; SI-NEXT:    s_cmp_eq_u32 s8, 0
+; SI-NEXT:    s_cselect_b32 s1, 0, s1
+; SI-NEXT:    s_cselect_b32 s0, 5, s0
 ; SI-NEXT:    v_mov_b32_e32 v0, s0
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 5, s[2:3]
+; SI-NEXT:    v_mov_b32_e32 v1, s1
+; SI-NEXT:    v_mov_b32_e32 v2, s2
+; SI-NEXT:    v_mov_b32_e32 v3, s3
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v2i64:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dword s10, s[4:5], 0x20
+; VI-NEXT:    s_load_dword s8, s[4:5], 0x20
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x10
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
 ; VI-NEXT:    s_mov_b32 s7, 0x1100f000
 ; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_cmp_eq_u32 s10, 1
-; VI-NEXT:    v_mov_b32_e32 v0, s3
-; VI-NEXT:    s_cselect_b64 s[8:9], -1, 0
-; VI-NEXT:    v_cndmask_b32_e64 v3, v0, 0, s[8:9]
-; VI-NEXT:    v_mov_b32_e32 v0, s2
-; VI-NEXT:    s_cmp_eq_u32 s10, 0
-; VI-NEXT:    v_cndmask_b32_e64 v2, v0, 5, s[8:9]
-; VI-NEXT:    v_mov_b32_e32 v0, s1
-; VI-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; VI-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[2:3]
+; VI-NEXT:    s_cmp_eq_u32 s8, 1
+; VI-NEXT:    s_cselect_b32 s3, 0, s3
+; VI-NEXT:    s_cselect_b32 s2, 5, s2
+; VI-NEXT:    s_cmp_eq_u32 s8, 0
+; VI-NEXT:    s_cselect_b32 s1, 0, s1
+; VI-NEXT:    s_cselect_b32 s0, 5, s0
 ; VI-NEXT:    v_mov_b32_e32 v0, s0
-; VI-NEXT:    v_cndmask_b32_e64 v0, v0, 5, s[2:3]
+; VI-NEXT:    v_mov_b32_e32 v1, s1
+; VI-NEXT:    v_mov_b32_e32 v2, s2
+; VI-NEXT:    v_mov_b32_e32 v3, s3
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
   %vecins = insertelement <2 x i64> %a, i64 5, i32 %b
   store <2 x i64> %vecins, <2 x i64> addrspace(1)* %out, align 8
   ret void
 }
 
 define amdgpu_kernel void @dynamic_insertelement_v3i64(<3 x i64> addrspace(1)* %out, <3 x i64> %a, i32 %b) nounwind {
 ; SI-LABEL: dynamic_insertelement_v3i64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dword s12, s[4:5], 0x10
+; SI-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; SI-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x8
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0xc
 ; SI-NEXT:    s_mov_b32 s3, 0x100f000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_cmp_eq_u32 s12, 1
-; SI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s11
-; SI-NEXT:    v_cndmask_b32_e64 v3, v0, 0, s[6:7]
-; SI-NEXT:    v_mov_b32_e32 v0, s10
-; SI-NEXT:    s_cmp_eq_u32 s12, 0
-; SI-NEXT:    v_cndmask_b32_e64 v2, v0, 5, s[6:7]
-; SI-NEXT:    v_mov_b32_e32 v0, s9
-; SI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[6:7]
-; SI-NEXT:    v_mov_b32_e32 v0, s8
-; SI-NEXT:    s_cmp_eq_u32 s12, 2
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 5, s[6:7]
-; SI-NEXT:    v_mov_b32_e32 v4, s5
-; SI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v5, v4, 0, s[6:7]
-; SI-NEXT:    v_mov_b32_e32 v4, s4
+; SI-NEXT:    s_cmp_eq_u32 s6, 1
 ; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, 5, s[6:7]
-; SI-NEXT:    buffer_store_dwordx2 v[4:5], off, s[0:3], 0 offset:16
+; SI-NEXT:    s_cselect_b32 s7, 0, s11
+; SI-NEXT:    s_cselect_b32 s10, 5, s10
+; SI-NEXT:    s_cmp_eq_u32 s6, 0
+; SI-NEXT:    s_cselect_b32 s9, 0, s9
+; SI-NEXT:    s_cselect_b32 s8, 5, s8
+; SI-NEXT:    s_cmp_eq_u32 s6, 2
+; SI-NEXT:    s_cselect_b32 s5, 0, s5
+; SI-NEXT:    s_cselect_b32 s4, 5, s4
+; SI-NEXT:    v_mov_b32_e32 v0, s4
+; SI-NEXT:    v_mov_b32_e32 v1, s5
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0 offset:16
+; SI-NEXT:    v_mov_b32_e32 v0, s8
+; SI-NEXT:    v_mov_b32_e32 v1, s9
+; SI-NEXT:    v_mov_b32_e32 v2, s10
+; SI-NEXT:    v_mov_b32_e32 v3, s7
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v3i64:
 ; VI:       ; %bb.0:
-; VI-NEXT:    s_load_dword s12, s[4:5], 0x40
+; VI-NEXT:    s_load_dword s6, s[4:5], 0x40
 ; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; VI-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x20
 ; VI-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x30
 ; VI-NEXT:    s_mov_b32 s3, 0x1100f000
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
-; VI-NEXT:    s_cmp_eq_u32 s12, 1
-; VI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; VI-NEXT:    v_mov_b32_e32 v0, s11
-; VI-NEXT:    v_cndmask_b32_e64 v3, v0, 0, s[6:7]
-; VI-NEXT:    v_mov_b32_e32 v0, s10
-; VI-NEXT:    s_cmp_eq_u32 s12, 0
-; VI-NEXT:    v_cndmask_b32_e64 v2, v0, 5, s[6:7]
-; VI-NEXT:    v_mov_b32_e32 v0, s9
-; VI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; VI-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[6:7]
-; VI-NEXT:    v_mov_b32_e32 v0, s8
-; VI-NEXT:    s_cmp_eq_u32 s12, 2
-; VI-NEXT:    v_cndmask_b32_e64 v0, v0, 5, s[6:7]
-; VI-NEXT:    v_mov_b32_e32 v4, s5
-; VI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; VI-NEXT:    v_cndmask_b32_e64 v5, v4, 0, s[6:7]
-; VI-NEXT:    v_mov_b32_e32 v4, s4
+; VI-NEXT:    s_cmp_eq_u32 s6, 1
 ; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    v_cndmask_b32_e64 v4, v4, 5, s[6:7]
-; VI-NEXT:    buffer_store_dwordx2 v[4:5], off, s[0:3], 0 offset:16
+; VI-NEXT:    s_cselect_b32 s7, 0, s11
+; VI-NEXT:    s_cselect_b32 s10, 5, s10
+; VI-NEXT:    s_cmp_eq_u32 s6, 0
+; VI-NEXT:    s_cselect_b32 s9, 0, s9
+; VI-NEXT:    s_cselect_b32 s8, 5, s8
+; VI-NEXT:    s_cmp_eq_u32 s6, 2
+; VI-NEXT:    s_cselect_b32 s5, 0, s5
+; VI-NEXT:    s_cselect_b32 s4, 5, s4
+; VI-NEXT:    v_mov_b32_e32 v0, s4
+; VI-NEXT:    v_mov_b32_e32 v1, s5
+; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0 offset:16
+; VI-NEXT:    v_mov_b32_e32 v0, s8
+; VI-NEXT:    v_mov_b32_e32 v1, s9
+; VI-NEXT:    v_mov_b32_e32 v2, s10
+; VI-NEXT:    v_mov_b32_e32 v3, s7
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
   %vecins = insertelement <3 x i64> %a, i64 5, i32 %b
   store <3 x i64> %vecins, <3 x i64> addrspace(1)* %out, align 32
   ret void
 }
 
 define amdgpu_kernel void @dynamic_insertelement_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, i32 %b) nounwind {
 ; SI-LABEL: dynamic_insertelement_v4f64:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dword s6, s[4:5], 0x10
 ; SI-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x8
-; SI-NEXT:    v_mov_b32_e32 v4, 0x40200000
 ; SI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; SI-NEXT:    s_mov_b32 s3, 0x100f000
+; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
 ; SI-NEXT:    s_cmp_eq_u32 s6, 1
-; SI-NEXT:    v_mov_b32_e32 v0, s11
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v3, v0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s10
+; SI-NEXT:    s_cselect_b32 s4, 0x40200000, s11
+; SI-NEXT:    s_cselect_b32 s5, 0, s10
 ; SI-NEXT:    s_cmp_eq_u32 s6, 0
-; SI-NEXT:    v_cndmask_b32_e64 v2, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s9
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v1, v0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v0, s8
+; SI-NEXT:    s_cselect_b32 s7, 0x40200000, s9
+; SI-NEXT:    s_cselect_b32 s8, 0, s8
 ; SI-NEXT:    s_cmp_eq_u32 s6, 3
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s15
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v7, v5, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s14
+; SI-NEXT:    s_cselect_b32 s9, 0x40200000, s15
+; SI-NEXT:    s_cselect_b32 s10, 0, s14
 ; SI-NEXT:    s_cmp_eq_u32 s6, 2
-; SI-NEXT:    v_cndmask_b32_e64 v6, v5, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s13
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v5, v5, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v4, s12
-; SI-NEXT:    s_mov_b32 s2, -1
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; SI-NEXT:    s_cselect_b32 s6, 0x40200000, s13
+; SI-NEXT:    s_cselect_b32 s11, 0, s12
+; SI-NEXT:    v_mov_b32_e32 v0, s11
+; SI-NEXT:    v_mov_b32_e32 v1, s6
+; SI-NEXT:    v_mov_b32_e32 v2, s10
+; SI-NEXT:    v_mov_b32_e32 v3, s9
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:16
+; SI-NEXT:    s_nop 0
+; SI-NEXT:    v_mov_b32_e32 v0, s8
+; SI-NEXT:    v_mov_b32_e32 v1, s7
+; SI-NEXT:    v_mov_b32_e32 v2, s5
+; SI-NEXT:    v_mov_b32_e32 v3, s4
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: dynamic_insertelement_v4f64:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dword s6, s[4:5], 0x40
 ; VI-NEXT:    s_load_dwordx8 s[8:15], s[4:5], 0x20
-; VI-NEXT:    v_mov_b32_e32 v4, 0x40200000
 ; VI-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; VI-NEXT:    s_mov_b32 s3, 0x1100f000
+; VI-NEXT:    s_mov_b32 s2, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    s_cmp_eq_u32 s6, 1
-; VI-NEXT:    v_mov_b32_e32 v0, s11
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v3, v0, v4, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s10
+; VI-NEXT:    s_cselect_b32 s4, 0x40200000, s11
+; VI-NEXT:    s_cselect_b32 s5, 0, s10
 ; VI-NEXT:    s_cmp_eq_u32 s6, 0
-; VI-NEXT:    v_cndmask_b32_e64 v2, v0, 0, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s9
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v1, v0, v4, vcc
-; VI-NEXT:    v_mov_b32_e32 v0, s8
+; VI-NEXT:    s_cselect_b32 s7, 0x40200000, s9
+; VI-NEXT:    s_cselect_b32 s8, 0, s8
 ; VI-NEXT:    s_cmp_eq_u32 s6, 3
-; VI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; VI-NEXT:    v_mov_b32_e32 v5, s15
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v7, v5, v4, vcc
-; VI-NEXT:    v_mov_b32_e32 v5, s14
+; VI-NEXT:    s_cselect_b32 s9, 0x40200000, s15
+; VI-NEXT:    s_cselect_b32 s10, 0, s14
 ; VI-NEXT:    s_cmp_eq_u32 s6, 2
-; VI-NEXT:    v_cndmask_b32_e64 v6, v5, 0, vcc
-; VI-NEXT:    v_mov_b32_e32 v5, s13
-; VI-NEXT:    s_cselect_b64 vcc, -1, 0
-; VI-NEXT:    v_cndmask_b32_e32 v5, v5, v4, vcc
-; VI-NEXT:    v_mov_b32_e32 v4, s12
-; VI-NEXT:    s_mov_b32 s2, -1
-; VI-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; VI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; VI-NEXT:    s_cselect_b32 s6, 0x40200000, s13
+; VI-NEXT:    s_cselect_b32 s11, 0, s12
+; VI-NEXT:    v_mov_b32_e32 v0, s11
+; VI-NEXT:    v_mov_b32_e32 v1, s6
+; VI-NEXT:    v_mov_b32_e32 v2, s10
+; VI-NEXT:    v_mov_b32_e32 v3, s9
+; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0 offset:16
+; VI-NEXT:    s_nop 0
+; VI-NEXT:    v_mov_b32_e32 v0, s8
+; VI-NEXT:    v_mov_b32_e32 v1, s7
+; VI-NEXT:    v_mov_b32_e32 v2, s5
+; VI-NEXT:    v_mov_b32_e32 v3, s4
 ; VI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
   %vecins = insertelement <4 x double> %a, double 8.0, i32 %b
   store <4 x double> %vecins, <4 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 define amdgpu_kernel void @dynamic_insertelement_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, i32 %b) #0 {

llvm/test/CodeGen/AMDGPU/llvm.mulo.ll

 ; SI-NEXT:    v_mul_hi_u32 v1, s1, v0
 ; SI-NEXT:    s_mul_i32 s4, s1, s2
 ; SI-NEXT:    v_mov_b32_e32 v2, s3
 ; SI-NEXT:    v_mul_hi_u32 v3, s0, v2
 ; SI-NEXT:    s_mul_i32 s5, s0, s3
 ; SI-NEXT:    v_mul_hi_u32 v0, s0, v0
 ; SI-NEXT:    v_mul_hi_u32 v2, s1, v2
 ; SI-NEXT:    s_mul_i32 s1, s1, s3
-; SI-NEXT:    s_mul_i32 s0, s0, s2
+; SI-NEXT:    s_mul_i32 s2, s0, s2
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, s5, v0
 ; SI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s0
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, s4, v4
 ; SI-NEXT:    v_addc_u32_e32 v1, vcc, v3, v1, vcc
 ; SI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; SI-NEXT:    v_add_i32_e32 v3, vcc, s5, v0
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, s1, v1
 ; SI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, s4, v3
 ; SI-NEXT:    v_cmp_ne_u64_e32 vcc, 0, v[0:1]
 ; SI-NEXT:    v_cndmask_b32_e64 v1, v2, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v0, v5, 0, vcc
+; SI-NEXT:    s_and_b64 s[0:1], vcc, exec
+; SI-NEXT:    s_cselect_b32 s0, 0, s2
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    v_mov_b32_e32 v0, s0
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: umulo_i64_s:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_mul_i32 s7, s0, s3
 ; GFX9-NEXT:    s_mul_hi_u32 s8, s0, s2
 ; GFX9-NEXT:    s_mul_hi_u32 s5, s0, s3
 ; GFX9-NEXT:    s_add_u32 s9, s8, s7
 ; GFX9-NEXT:    s_mul_i32 s6, s1, s2
 ; GFX9-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX9-NEXT:    s_mul_hi_u32 s4, s1, s2
 ; GFX9-NEXT:    s_add_u32 s9, s9, s6
 ; GFX9-NEXT:    s_mul_hi_u32 s10, s1, s3
 ; GFX9-NEXT:    s_addc_u32 s4, s5, s4
 ; GFX9-NEXT:    s_addc_u32 s5, s10, 0
 ; GFX9-NEXT:    s_mul_i32 s1, s1, s3
 ; GFX9-NEXT:    s_add_u32 s4, s4, s1
 ; GFX9-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX9-NEXT:    s_add_i32 s1, s8, s7
 ; GFX9-NEXT:    s_add_i32 s1, s1, s6
+; GFX9-NEXT:    s_mul_i32 s0, s0, s2
 ; GFX9-NEXT:    s_cmp_lg_u64 s[4:5], 0
-; GFX9-NEXT:    s_mul_i32 s2, s0, s2
-; GFX9-NEXT:    v_mov_b32_e32 v0, s1
-; GFX9-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v0, s2
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[0:1]
+; GFX9-NEXT:    s_cselect_b32 s1, 0, s1
+; GFX9-NEXT:    s_cselect_b32 s0, 0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    global_store_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: umulo_i64_s:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_mul_i32 s7, s0, s3

 ; GFX10-NEXT:    s_addc_u32 s3, s5, s4
 ; GFX10-NEXT:    s_addc_u32 s5, s9, 0
 ; GFX10-NEXT:    s_add_u32 s4, s3, s1
 ; GFX10-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX10-NEXT:    s_add_i32 s1, s8, s7
 ; GFX10-NEXT:    s_mul_i32 s0, s0, s2
 ; GFX10-NEXT:    s_add_i32 s1, s1, s6
 ; GFX10-NEXT:    s_cmp_lg_u64 s[4:5], 0
-; GFX10-NEXT:    s_cselect_b32 s2, -1, 0
-; GFX10-NEXT:    v_cndmask_b32_e64 v1, s1, 0, s2
-; GFX10-NEXT:    v_cndmask_b32_e64 v0, s0, 0, s2
+; GFX10-NEXT:    s_cselect_b32 s0, 0, s0
+; GFX10-NEXT:    s_cselect_b32 s1, 0, s1
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    global_store_dwordx2 v[0:1], v[0:1], off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: umulo_i64_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_mul_i32 s7, s0, s3

 ; GFX11-NEXT:    s_addc_u32 s3, s5, s4
 ; GFX11-NEXT:    s_addc_u32 s5, s9, 0
 ; GFX11-NEXT:    s_add_u32 s4, s3, s1
 ; GFX11-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX11-NEXT:    s_add_i32 s1, s8, s7
 ; GFX11-NEXT:    s_mul_i32 s0, s0, s2
 ; GFX11-NEXT:    s_add_i32 s1, s1, s6
 ; GFX11-NEXT:    s_cmp_lg_u64 s[4:5], 0
-; GFX11-NEXT:    s_cselect_b32 s2, -1, 0
+; GFX11-NEXT:    s_cselect_b32 s0, 0, s0
+; GFX11-NEXT:    s_cselect_b32 s1, 0, s1
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, s1, 0, s2
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, s0, 0, s2
+; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    global_store_b64 v[0:1], v[0:1], off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %umulo = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %x, i64 %y)
   %mul = extractvalue { i64, i1 } %umulo, 0
   %overflow = extractvalue { i64, i1 } %umulo, 1
   %res = select i1 %overflow, i64 0, i64 %mul

 ; SI-NEXT:    v_mul_hi_u32 v1, s1, v0
 ; SI-NEXT:    s_mul_i32 s4, s1, s2
 ; SI-NEXT:    v_mov_b32_e32 v2, s3
 ; SI-NEXT:    v_mul_hi_u32 v3, s0, v2
 ; SI-NEXT:    s_mul_i32 s5, s0, s3
 ; SI-NEXT:    v_mul_hi_u32 v0, s0, v0
 ; SI-NEXT:    v_mul_hi_i32 v2, s1, v2
 ; SI-NEXT:    s_mul_i32 s6, s1, s3
-; SI-NEXT:    s_cmp_lt_i32 s1, 0
-; SI-NEXT:    s_mul_i32 s1, s0, s2
-; SI-NEXT:    v_add_i32_e32 v4, vcc, s5, v0
-; SI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s1
-; SI-NEXT:    v_add_i32_e32 v4, vcc, s4, v4
-; SI-NEXT:    v_addc_u32_e32 v1, vcc, v3, v1, vcc
-; SI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
+; SI-NEXT:    s_mul_i32 s8, s0, s2
+; SI-NEXT:    v_readfirstlane_b32 s9, v1
+; SI-NEXT:    v_readfirstlane_b32 s10, v3
+; SI-NEXT:    v_readfirstlane_b32 s11, v0
+; SI-NEXT:    v_readfirstlane_b32 s12, v2
 ; SI-NEXT:    v_add_i32_e32 v0, vcc, s5, v0
-; SI-NEXT:    v_add_i32_e32 v1, vcc, s6, v1
-; SI-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
-; SI-NEXT:    v_add_i32_e32 v4, vcc, s4, v0
-; SI-NEXT:    v_subrev_i32_e32 v3, vcc, s2, v1
-; SI-NEXT:    v_subbrev_u32_e32 v6, vcc, 0, v2, vcc
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_lt_i32 s3, 0
-; SI-NEXT:    v_ashrrev_i32_e32 v0, 31, v4
-; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; SI-NEXT:    v_cndmask_b32_e32 v6, v1, v3, vcc
+; SI-NEXT:    s_add_u32 s5, s11, s5
+; SI-NEXT:    v_add_i32_e32 v2, vcc, s4, v0
+; SI-NEXT:    s_addc_u32 s10, 0, s10
+; SI-NEXT:    v_ashrrev_i32_e32 v0, 31, v2
+; SI-NEXT:    s_add_u32 s4, s5, s4
 ; SI-NEXT:    v_mov_b32_e32 v1, v0
-; SI-NEXT:    v_subrev_i32_e32 v7, vcc, s0, v6
-; SI-NEXT:    v_subbrev_u32_e32 v3, vcc, 0, v2, vcc
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    v_cndmask_b32_e32 v3, v2, v3, vcc
-; SI-NEXT:    v_cndmask_b32_e32 v2, v6, v7, vcc
-; SI-NEXT:    v_cmp_ne_u64_e32 vcc, v[2:3], v[0:1]
-; SI-NEXT:    v_cndmask_b32_e64 v1, v4, 0, vcc
-; SI-NEXT:    v_cndmask_b32_e64 v0, v5, 0, vcc
+; SI-NEXT:    s_addc_u32 s4, s10, s9
+; SI-NEXT:    s_addc_u32 s5, s12, 0
+; SI-NEXT:    s_add_u32 s4, s4, s6
+; SI-NEXT:    s_addc_u32 s5, 0, s5
+; SI-NEXT:    s_sub_u32 s2, s4, s2
+; SI-NEXT:    s_subb_u32 s6, s5, 0
+; SI-NEXT:    s_cmp_lt_i32 s1, 0
+; SI-NEXT:    s_cselect_b32 s1, s6, s5
+; SI-NEXT:    s_cselect_b32 s2, s2, s4
+; SI-NEXT:    s_sub_u32 s0, s2, s0
+; SI-NEXT:    s_subb_u32 s4, s1, 0
+; SI-NEXT:    s_cmp_lt_i32 s3, 0
+; SI-NEXT:    s_cselect_b32 s1, s4, s1
+; SI-NEXT:    s_cselect_b32 s0, s0, s2
+; SI-NEXT:    v_cmp_ne_u64_e32 vcc, s[0:1], v[0:1]
+; SI-NEXT:    v_cndmask_b32_e64 v1, v2, 0, vcc
+; SI-NEXT:    s_and_b64 s[0:1], vcc, exec
+; SI-NEXT:    s_cselect_b32 s0, 0, s8
 ; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    v_mov_b32_e32 v0, s0
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: smulo_i64_s:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_mul_i32 s7, s0, s3
 ; GFX9-NEXT:    s_mul_hi_u32 s8, s0, s2
-; GFX9-NEXT:    s_mul_hi_u32 s6, s0, s3
+; GFX9-NEXT:    s_mul_hi_u32 s5, s0, s3
 ; GFX9-NEXT:    s_add_u32 s9, s8, s7
-; GFX9-NEXT:    s_mul_i32 s5, s1, s2
-; GFX9-NEXT:    s_addc_u32 s6, 0, s6
+; GFX9-NEXT:    s_mul_i32 s6, s1, s2
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX9-NEXT:    s_mul_hi_u32 s4, s1, s2
-; GFX9-NEXT:    s_add_u32 s9, s9, s5
+; GFX9-NEXT:    s_add_u32 s9, s9, s6
 ; GFX9-NEXT:    s_mul_hi_i32 s10, s1, s3
-; GFX9-NEXT:    s_addc_u32 s4, s6, s4
-; GFX9-NEXT:    s_addc_u32 s6, s10, 0
+; GFX9-NEXT:    s_addc_u32 s4, s5, s4
+; GFX9-NEXT:    s_addc_u32 s5, s10, 0
 ; GFX9-NEXT:    s_mul_i32 s9, s1, s3
 ; GFX9-NEXT:    s_add_u32 s4, s4, s9
-; GFX9-NEXT:    s_addc_u32 s6, 0, s6
+; GFX9-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX9-NEXT:    s_sub_u32 s9, s4, s2
-; GFX9-NEXT:    s_subb_u32 s10, s6, 0
+; GFX9-NEXT:    s_subb_u32 s10, s5, 0
 ; GFX9-NEXT:    s_cmp_lt_i32 s1, 0
-; GFX9-NEXT:    v_mov_b32_e32 v0, s6
-; GFX9-NEXT:    v_mov_b32_e32 v1, s10
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GFX9-NEXT:    v_mov_b32_e32 v1, s4
-; GFX9-NEXT:    v_mov_b32_e32 v2, s9
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v1, v2, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e32 v3, vcc, s0, v2
-; GFX9-NEXT:    v_subbrev_co_u32_e32 v1, vcc, 0, v0, vcc
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s4
+; GFX9-NEXT:    s_cselect_b32 s1, s10, s5
+; GFX9-NEXT:    s_sub_u32 s9, s4, s0
+; GFX9-NEXT:    s_subb_u32 s5, s1, 0
 ; GFX9-NEXT:    s_cmp_lt_i32 s3, 0
-; GFX9-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    s_cselect_b32 s5, s5, s1
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s4
 ; GFX9-NEXT:    s_add_i32 s1, s8, s7
-; GFX9-NEXT:    s_add_i32 s1, s1, s5
-; GFX9-NEXT:    s_ashr_i32 s4, s1, 31
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
-; GFX9-NEXT:    s_mov_b32 s5, s4
+; GFX9-NEXT:    s_add_i32 s1, s1, s6
+; GFX9-NEXT:    s_ashr_i32 s6, s1, 31
+; GFX9-NEXT:    s_mov_b32 s7, s6
 ; GFX9-NEXT:    s_mul_i32 s0, s0, s2
-; GFX9-NEXT:    v_cmp_ne_u64_e32 vcc, s[4:5], v[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    s_cmp_lg_u64 s[4:5], s[6:7]
+; GFX9-NEXT:    s_cselect_b32 s1, 0, s1
+; GFX9-NEXT:    s_cselect_b32 s0, 0, s0
 ; GFX9-NEXT:    v_mov_b32_e32 v0, s0
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v2, 0, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX9-NEXT:    global_store_dwordx2 v[0:1], v[0:1], off
 ; GFX9-NEXT:    s_endpgm
 ;
 ; GFX10-LABEL: smulo_i64_s:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_mul_i32 s7, s0, s3
 ; GFX10-NEXT:    s_mul_hi_u32 s8, s0, s2
-; GFX10-NEXT:    s_mul_hi_u32 s6, s0, s3
-; GFX10-NEXT:    s_mul_i32 s5, s1, s2
+; GFX10-NEXT:    s_mul_hi_u32 s5, s0, s3
+; GFX10-NEXT:    s_mul_i32 s6, s1, s2
 ; GFX10-NEXT:    s_add_u32 s11, s8, s7
 ; GFX10-NEXT:    s_mul_hi_u32 s4, s1, s2
-; GFX10-NEXT:    s_addc_u32 s6, 0, s6
+; GFX10-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX10-NEXT:    s_mul_hi_i32 s9, s1, s3
-; GFX10-NEXT:    s_add_u32 s11, s11, s5
+; GFX10-NEXT:    s_add_u32 s11, s11, s6
 ; GFX10-NEXT:    s_mul_i32 s10, s1, s3
-; GFX10-NEXT:    s_addc_u32 s4, s6, s4
-; GFX10-NEXT:    s_addc_u32 s6, s9, 0
+; GFX10-NEXT:    s_addc_u32 s4, s5, s4
+; GFX10-NEXT:    s_addc_u32 s5, s9, 0
 ; GFX10-NEXT:    s_add_u32 s4, s4, s10
-; GFX10-NEXT:    s_addc_u32 s6, 0, s6
+; GFX10-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX10-NEXT:    s_sub_u32 s9, s4, s2
-; GFX10-NEXT:    s_subb_u32 s10, s6, 0
-; GFX10-NEXT:    v_mov_b32_e32 v1, s9
+; GFX10-NEXT:    s_subb_u32 s10, s5, 0
 ; GFX10-NEXT:    s_cmp_lt_i32 s1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v0, s10
-; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX10-NEXT:    s_cselect_b32 s1, s9, s4
+; GFX10-NEXT:    s_cselect_b32 s4, s10, s5
+; GFX10-NEXT:    s_sub_u32 s9, s1, s0
+; GFX10-NEXT:    s_subb_u32 s5, s4, 0
 ; GFX10-NEXT:    s_cmp_lt_i32 s3, 0
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, s4, v1, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e32 v0, s6, v0, vcc_lo
-; GFX10-NEXT:    v_sub_co_u32 v3, vcc_lo, v2, s0
-; GFX10-NEXT:    v_subrev_co_ci_u32_e32 v1, vcc_lo, 0, v0, vcc_lo
-; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
-; GFX10-NEXT:    s_add_i32 s1, s8, s7
 ; GFX10-NEXT:    s_mul_i32 s0, s0, s2
-; GFX10-NEXT:    s_add_i32 s1, s1, s5
-; GFX10-NEXT:    v_cndmask_b32_e32 v1, v0, v1, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc_lo
-; GFX10-NEXT:    s_ashr_i32 s4, s1, 31
-; GFX10-NEXT:    s_mov_b32 s5, s4
-; GFX10-NEXT:    v_cmp_ne_u64_e32 vcc_lo, s[4:5], v[0:1]
-; GFX10-NEXT:    v_cndmask_b32_e64 v1, s1, 0, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e64 v0, s0, 0, vcc_lo
+; GFX10-NEXT:    s_cselect_b32 s5, s5, s4
+; GFX10-NEXT:    s_cselect_b32 s4, s9, s1
+; GFX10-NEXT:    s_add_i32 s1, s8, s7
+; GFX10-NEXT:    s_add_i32 s1, s1, s6
+; GFX10-NEXT:    s_ashr_i32 s6, s1, 31
+; GFX10-NEXT:    s_mov_b32 s7, s6
+; GFX10-NEXT:    s_cmp_lg_u64 s[4:5], s[6:7]
+; GFX10-NEXT:    s_cselect_b32 s0, 0, s0
+; GFX10-NEXT:    s_cselect_b32 s1, 0, s1
+; GFX10-NEXT:    v_mov_b32_e32 v0, s0
+; GFX10-NEXT:    v_mov_b32_e32 v1, s1
 ; GFX10-NEXT:    global_store_dwordx2 v[0:1], v[0:1], off
 ; GFX10-NEXT:    s_endpgm
 ;
 ; GFX11-LABEL: smulo_i64_s:
 ; GFX11:       ; %bb.0: ; %bb
 ; GFX11-NEXT:    s_load_b128 s[0:3], s[0:1], 0x24
 ; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-NEXT:    s_mul_i32 s7, s0, s3
 ; GFX11-NEXT:    s_mul_hi_u32 s8, s0, s2
-; GFX11-NEXT:    s_mul_hi_u32 s6, s0, s3
-; GFX11-NEXT:    s_mul_i32 s5, s1, s2
+; GFX11-NEXT:    s_mul_hi_u32 s5, s0, s3
+; GFX11-NEXT:    s_mul_i32 s6, s1, s2
 ; GFX11-NEXT:    s_add_u32 s11, s8, s7
 ; GFX11-NEXT:    s_mul_hi_u32 s4, s1, s2
-; GFX11-NEXT:    s_addc_u32 s6, 0, s6
+; GFX11-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX11-NEXT:    s_mul_hi_i32 s9, s1, s3
-; GFX11-NEXT:    s_add_u32 s11, s11, s5
+; GFX11-NEXT:    s_add_u32 s11, s11, s6
 ; GFX11-NEXT:    s_mul_i32 s10, s1, s3
-; GFX11-NEXT:    s_addc_u32 s4, s6, s4
-; GFX11-NEXT:    s_addc_u32 s6, s9, 0
+; GFX11-NEXT:    s_addc_u32 s4, s5, s4
+; GFX11-NEXT:    s_addc_u32 s5, s9, 0
 ; GFX11-NEXT:    s_add_u32 s4, s4, s10
-; GFX11-NEXT:    s_addc_u32 s6, 0, s6
+; GFX11-NEXT:    s_addc_u32 s5, 0, s5
 ; GFX11-NEXT:    s_sub_u32 s9, s4, s2
-; GFX11-NEXT:    s_subb_u32 s10, s6, 0
-; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
-; GFX11-NEXT:    v_dual_mov_b32 v1, s9 :: v_dual_mov_b32 v0, s10
+; GFX11-NEXT:    s_subb_u32 s10, s5, 0
 ; GFX11-NEXT:    s_cmp_lt_i32 s1, 0
-; GFX11-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX11-NEXT:    s_cselect_b32 s1, s9, s4
+; GFX11-NEXT:    s_cselect_b32 s4, s10, s5
+; GFX11-NEXT:    s_sub_u32 s9, s1, s0
+; GFX11-NEXT:    s_subb_u32 s5, s4, 0
 ; GFX11-NEXT:    s_cmp_lt_i32 s3, 0
-; GFX11-NEXT:    v_cndmask_b32_e32 v2, s4, v1, vcc_lo
-; GFX11-NEXT:    v_cndmask_b32_e32 v0, s6, v0, vcc_lo
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-NEXT:    v_sub_co_u32 v3, vcc_lo, v2, s0
-; GFX11-NEXT:    v_subrev_co_ci_u32_e32 v1, vcc_lo, 0, v0, vcc_lo
-; GFX11-NEXT:    s_cselect_b32 vcc_lo, -1, 0
-; GFX11-NEXT:    s_add_i32 s1, s8, s7
 ; GFX11-NEXT:    s_mul_i32 s0, s0, s2
-; GFX11-NEXT:    s_add_i32 s1, s1, s5
-; GFX11-NEXT:    v_dual_cndmask_b32 v1, v0, v1 :: v_dual_cndmask_b32 v0, v2, v3
-; GFX11-NEXT:    s_ashr_i32 s4, s1, 31
+; GFX11-NEXT:    s_cselect_b32 s5, s5, s4
+; GFX11-NEXT:    s_cselect_b32 s4, s9, s1
+; GFX11-NEXT:    s_add_i32 s1, s8, s7
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    s_add_i32 s1, s1, s6
+; GFX11-NEXT:    s_ashr_i32 s6, s1, 31
+; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-NEXT:    s_mov_b32 s7, s6
+; GFX11-NEXT:    s_cmp_lg_u64 s[4:5], s[6:7]
+; GFX11-NEXT:    s_cselect_b32 s0, 0, s0
+; GFX11-NEXT:    s_cselect_b32 s1, 0, s1
 ; GFX11-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
-; GFX11-NEXT:    s_mov_b32 s5, s4
-; GFX11-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
-; GFX11-NEXT:    v_cmp_ne_u64_e32 vcc_lo, s[4:5], v[0:1]
-; GFX11-NEXT:    v_cndmask_b32_e64 v1, s1, 0, vcc_lo
-; GFX11-NEXT:    v_cndmask_b32_e64 v0, s0, 0, vcc_lo
+; GFX11-NEXT:    v_dual_mov_b32 v0, s0 :: v_dual_mov_b32 v1, s1
 ; GFX11-NEXT:    global_store_b64 v[0:1], v[0:1], off
 ; GFX11-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
 ; GFX11-NEXT:    s_endpgm
 bb:
   %umulo = tail call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %x, i64 %y)
   %mul = extractvalue { i64, i1 } %umulo, 0
   %overflow = extractvalue { i64, i1 } %umulo, 1
   %res = select i1 %overflow, i64 0, i64 %mul

llvm/test/CodeGen/AMDGPU/llvm.round.f64.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI %s
 ; RUN: llc -march=amdgcn -mcpu=hawaii < %s | FileCheck -check-prefix=CI %s
 
 define amdgpu_kernel void @round_f64(double addrspace(1)* %out, double %x) #0 {
 ; SI-LABEL: round_f64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s10, -1
-; SI-NEXT:    s_mov_b32 s1, 0xfffff
-; SI-NEXT:    v_mov_b32_e32 v4, 0x3ff00000
-; SI-NEXT:    s_mov_b32 s11, 0xf000
+; SI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s6, -1
+; SI-NEXT:    s_mov_b32 s9, 0xfffff
+; SI-NEXT:    s_mov_b32 s8, s6
+; SI-NEXT:    v_mov_b32_e32 v2, 0x3ff00000
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_bfe_u32 s0, s7, 0xb0014
-; SI-NEXT:    s_mov_b32 s8, s4
-; SI-NEXT:    s_add_i32 s4, s0, 0xfffffc01
-; SI-NEXT:    s_mov_b32 s0, s10
-; SI-NEXT:    s_lshr_b64 s[0:1], s[0:1], s4
-; SI-NEXT:    s_andn2_b64 s[2:3], s[6:7], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s7, 0x80000000
-; SI-NEXT:    s_cmp_lt_i32 s4, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s3
-; SI-NEXT:    v_mov_b32_e32 v1, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s4, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s7
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; SI-NEXT:    v_mov_b32_e32 v0, s2
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, s6
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; SI-NEXT:    v_add_f64 v[2:3], s[6:7], -v[0:1]
+; SI-NEXT:    s_mov_b32 s4, s0
+; SI-NEXT:    s_bfe_u32 s0, s3, 0xb0014
+; SI-NEXT:    s_addk_i32 s0, 0xfc01
+; SI-NEXT:    s_lshr_b64 s[8:9], s[8:9], s0
+; SI-NEXT:    s_andn2_b64 s[8:9], s[2:3], s[8:9]
+; SI-NEXT:    s_and_b32 s5, s3, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s0, 0
+; SI-NEXT:    s_cselect_b32 s8, 0, s8
+; SI-NEXT:    s_cselect_b32 s5, s5, s9
+; SI-NEXT:    s_cmp_gt_i32 s0, 51
+; SI-NEXT:    s_cselect_b32 s8, s2, s8
+; SI-NEXT:    s_cselect_b32 s9, s3, s5
+; SI-NEXT:    v_mov_b32_e32 v0, s8
+; SI-NEXT:    v_mov_b32_e32 v1, s9
+; SI-NEXT:    v_add_f64 v[0:1], s[2:3], -v[0:1]
 ; SI-NEXT:    s_brev_b32 s0, -2
-; SI-NEXT:    v_mov_b32_e32 v5, s7
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[2:3]|, 0.5
-; SI-NEXT:    v_bfi_b32 v4, s0, v4, v5
-; SI-NEXT:    v_cndmask_b32_e32 v3, 0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, 0
-; SI-NEXT:    v_add_f64 v[0:1], v[0:1], v[2:3]
-; SI-NEXT:    s_mov_b32 s9, s5
-; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
+; SI-NEXT:    v_mov_b32_e32 v3, s3
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    v_bfi_b32 v2, s0, v2, v3
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v2, vcc
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    v_add_f64 v[0:1], s[8:9], v[0:1]
+; SI-NEXT:    s_mov_b32 s7, 0xf000
+; SI-NEXT:    s_mov_b32 s5, s1
+; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: round_f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; CI-NEXT:    s_brev_b32 s5, -2
 ; CI-NEXT:    v_mov_b32_e32 v4, 0x3ff00000
 ; CI-NEXT:    s_mov_b32 s7, 0xf000
; CI-NEXT:    s_endpgm
   %result = call double @llvm.round.f64(double %x) #1
   store double %result, double addrspace(1)* %out.gep
   ret void
 }
 
 define amdgpu_kernel void @round_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %in) #0 {
 ; SI-LABEL: round_v2f64:
 ; SI:       ; %bb.0:
-; SI-NEXT:    s_load_dwordx4 s[8:11], s[0:1], 0xd
-; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s6, -1
-; SI-NEXT:    s_mov_b32 s3, 0xfffff
-; SI-NEXT:    s_mov_b32 s2, s6
+; SI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
+; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
+; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s9, 0xfffff
+; SI-NEXT:    s_mov_b32 s8, s2
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_bfe_u32 s0, s11, 0xb0014
-; SI-NEXT:    s_add_i32 s7, s0, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s7
-; SI-NEXT:    s_andn2_b64 s[12:13], s[10:11], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s11, 0x80000000
-; SI-NEXT:    s_cmp_lt_i32 s7, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s13
-; SI-NEXT:    v_mov_b32_e32 v1, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s7, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; SI-NEXT:    s_bfe_u32 s3, s7, 0xb0014
+; SI-NEXT:    s_addk_i32 s3, 0xfc01
+; SI-NEXT:    s_lshr_b64 s[10:11], s[8:9], s3
+; SI-NEXT:    s_andn2_b64 s[10:11], s[6:7], s[10:11]
+; SI-NEXT:    s_and_b32 s12, s7, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s3, 0
+; SI-NEXT:    s_cselect_b32 s10, 0, s10
+; SI-NEXT:    s_cselect_b32 s11, s12, s11
+; SI-NEXT:    s_cmp_gt_i32 s3, 51
+; SI-NEXT:    s_cselect_b32 s10, s6, s10
+; SI-NEXT:    s_cselect_b32 s11, s7, s11
+; SI-NEXT:    v_mov_b32_e32 v0, s10
 ; SI-NEXT:    v_mov_b32_e32 v1, s11
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; SI-NEXT:    v_mov_b32_e32 v0, s12
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, s10
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; SI-NEXT:    v_add_f64 v[2:3], s[10:11], -v[0:1]
-; SI-NEXT:    s_bfe_u32 s0, s9, 0xb0014
-; SI-NEXT:    s_add_i32 s10, s0, 0xfffffc01
-; SI-NEXT:    s_brev_b32 s7, -2
-; SI-NEXT:    v_mov_b32_e32 v6, 0x3ff00000
-; SI-NEXT:    v_mov_b32_e32 v4, s11
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[2:3]|, 0.5
-; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s10
-; SI-NEXT:    v_bfi_b32 v4, s7, v6, v4
-; SI-NEXT:    s_andn2_b64 s[2:3], s[8:9], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s9, 0x80000000
-; SI-NEXT:    v_cndmask_b32_e32 v3, 0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, 0
+; SI-NEXT:    v_add_f64 v[0:1], s[6:7], -v[0:1]
+; SI-NEXT:    s_brev_b32 s3, -2
+; SI-NEXT:    v_mov_b32_e32 v4, 0x3ff00000
+; SI-NEXT:    v_mov_b32_e32 v2, s7
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    v_bfi_b32 v2, s3, v4, v2
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v2, vcc
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    s_bfe_u32 s6, s5, 0xb0014
+; SI-NEXT:    v_add_f64 v[2:3], s[10:11], v[0:1]
+; SI-NEXT:    s_add_i32 s10, s6, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[6:7], s[8:9], s10
+; SI-NEXT:    s_andn2_b64 s[6:7], s[4:5], s[6:7]
+; SI-NEXT:    s_and_b32 s8, s5, 0x80000000
 ; SI-NEXT:    s_cmp_lt_i32 s10, 0
-; SI-NEXT:    v_add_f64 v[2:3], v[0:1], v[2:3]
-; SI-NEXT:    v_mov_b32_e32 v0, s3
-; SI-NEXT:    v_mov_b32_e32 v1, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s6, 0, s6
+; SI-NEXT:    s_cselect_b32 s7, s8, s7
 ; SI-NEXT:    s_cmp_gt_i32 s10, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s9
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; SI-NEXT:    v_mov_b32_e32 v0, s2
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v4, s8
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
-; SI-NEXT:    v_add_f64 v[4:5], s[8:9], -v[0:1]
-; SI-NEXT:    v_mov_b32_e32 v7, s9
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[4:5]|, 0.5
-; SI-NEXT:    v_bfi_b32 v6, s7, v6, v7
-; SI-NEXT:    v_cndmask_b32_e32 v5, 0, v6, vcc
-; SI-NEXT:    v_mov_b32_e32 v4, 0
-; SI-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
-; SI-NEXT:    s_mov_b32 s7, 0xf000
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
+; SI-NEXT:    s_cselect_b32 s6, s4, s6
+; SI-NEXT:    s_cselect_b32 s7, s5, s7
+; SI-NEXT:    v_mov_b32_e32 v0, s6
+; SI-NEXT:    v_mov_b32_e32 v1, s7
+; SI-NEXT:    v_add_f64 v[0:1], s[4:5], -v[0:1]
+; SI-NEXT:    v_mov_b32_e32 v5, s5
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    v_bfi_b32 v4, s3, v4, v5
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    v_add_f64 v[0:1], s[6:7], v[0:1]
+; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: round_v2f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xd
 ; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; CI-NEXT:    s_brev_b32 s2, -2
 ; CI-NEXT:    v_mov_b32_e32 v6, 0x3ff00000
; CI-NEXT:    s_endpgm
   store <2 x double> %result, <2 x double> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @round_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %in) #0 {
 ; SI-LABEL: round_v4f64:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x11
-; SI-NEXT:    s_mov_b32 s14, -1
-; SI-NEXT:    s_load_dwordx2 s[12:13], s[0:1], 0x9
-; SI-NEXT:    s_mov_b32 s3, 0xfffff
-; SI-NEXT:    s_mov_b32 s2, s14
+; SI-NEXT:    s_mov_b32 s2, -1
+; SI-NEXT:    s_mov_b32 s13, 0xfffff
+; SI-NEXT:    v_mov_b32_e32 v8, 0x3ff00000
+; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_bfe_u32 s0, s7, 0xb0014
-; SI-NEXT:    s_add_i32 s18, s0, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s18
-; SI-NEXT:    s_andn2_b64 s[16:17], s[6:7], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s7, 0x80000000
-; SI-NEXT:    s_cmp_lt_i32 s18, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s17
-; SI-NEXT:    v_mov_b32_e32 v1, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s18, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s7
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; SI-NEXT:    v_mov_b32_e32 v0, s16
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, s6
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; SI-NEXT:    v_add_f64 v[2:3], s[6:7], -v[0:1]
-; SI-NEXT:    s_bfe_u32 s0, s5, 0xb0014
-; SI-NEXT:    s_add_i32 s17, s0, 0xfffffc01
+; SI-NEXT:    s_bfe_u32 s12, s7, 0xb0014
+; SI-NEXT:    s_add_i32 s16, s12, 0xfffffc01
+; SI-NEXT:    s_mov_b32 s12, s2
+; SI-NEXT:    s_lshr_b64 s[14:15], s[12:13], s16
+; SI-NEXT:    s_andn2_b64 s[14:15], s[6:7], s[14:15]
+; SI-NEXT:    s_and_b32 s17, s7, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s16, 0
+; SI-NEXT:    s_cselect_b32 s14, 0, s14
+; SI-NEXT:    s_cselect_b32 s15, s17, s15
+; SI-NEXT:    s_cmp_gt_i32 s16, 51
+; SI-NEXT:    s_cselect_b32 s14, s6, s14
+; SI-NEXT:    s_cselect_b32 s15, s7, s15
+; SI-NEXT:    v_mov_b32_e32 v0, s14
+; SI-NEXT:    v_mov_b32_e32 v1, s15
+; SI-NEXT:    v_add_f64 v[0:1], s[6:7], -v[0:1]
 ; SI-NEXT:    s_brev_b32 s16, -2
-; SI-NEXT:    v_mov_b32_e32 v12, 0x3ff00000
-; SI-NEXT:    v_mov_b32_e32 v4, s7
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[2:3]|, 0.5
-; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s17
-; SI-NEXT:    v_bfi_b32 v4, s16, v12, v4
-; SI-NEXT:    s_andn2_b64 s[6:7], s[4:5], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s5, 0x80000000
-; SI-NEXT:    v_cndmask_b32_e32 v3, 0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, 0
-; SI-NEXT:    s_cmp_lt_i32 s17, 0
-; SI-NEXT:    v_add_f64 v[2:3], v[0:1], v[2:3]
-; SI-NEXT:    v_mov_b32_e32 v0, s7
-; SI-NEXT:    v_mov_b32_e32 v1, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s17, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s5
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
+; SI-NEXT:    v_mov_b32_e32 v2, s7
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    v_bfi_b32 v2, s16, v8, v2
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v2, vcc
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    s_bfe_u32 s6, s5, 0xb0014
+; SI-NEXT:    v_add_f64 v[2:3], s[14:15], v[0:1]
+; SI-NEXT:    s_add_i32 s14, s6, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[6:7], s[12:13], s14
+; SI-NEXT:    s_andn2_b64 s[6:7], s[4:5], s[6:7]
+; SI-NEXT:    s_and_b32 s15, s5, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s14, 0
+; SI-NEXT:    s_cselect_b32 s6, 0, s6
+; SI-NEXT:    s_cselect_b32 s7, s15, s7
+; SI-NEXT:    s_cmp_gt_i32 s14, 51
+; SI-NEXT:    s_cselect_b32 s6, s4, s6
+; SI-NEXT:    s_cselect_b32 s7, s5, s7
 ; SI-NEXT:    v_mov_b32_e32 v0, s6
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; SI-NEXT:    v_mov_b32_e32 v1, s7
+; SI-NEXT:    v_add_f64 v[0:1], s[4:5], -v[0:1]
+; SI-NEXT:    s_bfe_u32 s4, s11, 0xb0014
+; SI-NEXT:    s_add_i32 s14, s4, 0xfffffc01
+; SI-NEXT:    v_mov_b32_e32 v4, s5
+; SI-NEXT:    s_lshr_b64 s[4:5], s[12:13], s14
+; SI-NEXT:    s_andn2_b64 s[4:5], s[10:11], s[4:5]
+; SI-NEXT:    s_and_b32 s15, s11, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s14, 0
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    s_cselect_b32 s4, 0, s4
+; SI-NEXT:    s_cselect_b32 s5, s15, s5
+; SI-NEXT:    s_cmp_gt_i32 s14, 51
+; SI-NEXT:    v_bfi_b32 v4, s16, v8, v4
+; SI-NEXT:    s_cselect_b32 s4, s10, s4
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
+; SI-NEXT:    s_cselect_b32 s5, s11, s5
 ; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
-; SI-NEXT:    s_bfe_u32 s0, s11, 0xb0014
-; SI-NEXT:    v_add_f64 v[4:5], s[4:5], -v[0:1]
-; SI-NEXT:    s_add_i32 s6, s0, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s6
-; SI-NEXT:    v_mov_b32_e32 v6, s5
+; SI-NEXT:    v_mov_b32_e32 v5, s5
+; SI-NEXT:    v_add_f64 v[4:5], s[10:11], -v[4:5]
+; SI-NEXT:    v_mov_b32_e32 v6, s11
 ; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[4:5]|, 0.5
-; SI-NEXT:    s_andn2_b64 s[4:5], s[10:11], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s11, 0x80000000
-; SI-NEXT:    v_bfi_b32 v6, s16, v12, v6
-; SI-NEXT:    s_cmp_lt_i32 s6, 0
-; SI-NEXT:    v_cndmask_b32_e32 v9, 0, v6, vcc
-; SI-NEXT:    v_mov_b32_e32 v4, s5
-; SI-NEXT:    v_mov_b32_e32 v5, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s6, 51
-; SI-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s11
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v5, v4, v5, s[0:1]
+; SI-NEXT:    v_bfi_b32 v6, s16, v8, v6
+; SI-NEXT:    v_cndmask_b32_e32 v5, 0, v6, vcc
+; SI-NEXT:    v_mov_b32_e32 v4, 0
+; SI-NEXT:    v_add_f64 v[6:7], s[4:5], v[4:5]
+; SI-NEXT:    s_bfe_u32 s4, s9, 0xb0014
+; SI-NEXT:    s_add_i32 s10, s4, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[4:5], s[12:13], s10
+; SI-NEXT:    s_andn2_b64 s[4:5], s[8:9], s[4:5]
+; SI-NEXT:    s_and_b32 s11, s9, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s10, 0
+; SI-NEXT:    s_cselect_b32 s4, 0, s4
+; SI-NEXT:    s_cselect_b32 s5, s11, s5
+; SI-NEXT:    s_cmp_gt_i32 s10, 51
+; SI-NEXT:    s_cselect_b32 s4, s8, s4
+; SI-NEXT:    s_cselect_b32 s5, s9, s5
 ; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v6, s10
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v6, s[0:1]
-; SI-NEXT:    v_add_f64 v[6:7], s[10:11], -v[4:5]
-; SI-NEXT:    s_bfe_u32 s0, s9, 0xb0014
-; SI-NEXT:    s_add_i32 s4, s0, 0xfffffc01
-; SI-NEXT:    v_mov_b32_e32 v10, s11
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[6:7]|, 0.5
-; SI-NEXT:    s_lshr_b64 s[0:1], s[2:3], s4
-; SI-NEXT:    v_bfi_b32 v10, s16, v12, v10
-; SI-NEXT:    s_andn2_b64 s[2:3], s[8:9], s[0:1]
-; SI-NEXT:    s_and_b32 s0, s9, 0x80000000
-; SI-NEXT:    v_cndmask_b32_e32 v7, 0, v10, vcc
-; SI-NEXT:    v_mov_b32_e32 v6, 0
-; SI-NEXT:    s_cmp_lt_i32 s4, 0
-; SI-NEXT:    v_add_f64 v[6:7], v[4:5], v[6:7]
-; SI-NEXT:    v_mov_b32_e32 v4, s3
-; SI-NEXT:    v_mov_b32_e32 v5, s0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s4, 51
-; SI-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s9
-; SI-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v5, v4, v5, s[0:1]
-; SI-NEXT:    v_mov_b32_e32 v4, s2
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v10, s8
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v10, s[0:1]
-; SI-NEXT:    v_add_f64 v[10:11], s[8:9], -v[4:5]
-; SI-NEXT:    v_mov_b32_e32 v13, s9
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[10:11]|, 0.5
-; SI-NEXT:    v_bfi_b32 v12, s16, v12, v13
-; SI-NEXT:    v_cndmask_b32_e32 v11, 0, v12, vcc
-; SI-NEXT:    v_mov_b32_e32 v10, 0
-; SI-NEXT:    v_mov_b32_e32 v8, 0
-; SI-NEXT:    v_add_f64 v[4:5], v[4:5], v[10:11]
-; SI-NEXT:    s_mov_b32 s15, 0xf000
-; SI-NEXT:    v_add_f64 v[0:1], v[0:1], v[8:9]
-; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[12:15], 0 offset:16
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[12:15], 0
+; SI-NEXT:    v_mov_b32_e32 v5, s5
+; SI-NEXT:    v_add_f64 v[4:5], s[8:9], -v[4:5]
+; SI-NEXT:    v_mov_b32_e32 v9, s9
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[4:5]|, 0.5
+; SI-NEXT:    v_bfi_b32 v8, s16, v8, v9
+; SI-NEXT:    v_cndmask_b32_e32 v5, 0, v8, vcc
+; SI-NEXT:    v_mov_b32_e32 v4, 0
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    v_add_f64 v[4:5], s[4:5], v[4:5]
+; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    v_add_f64 v[0:1], s[6:7], v[0:1]
+; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: round_v4f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x11
 ; CI-NEXT:    s_brev_b32 s12, -2
 ; CI-NEXT:    v_mov_b32_e32 v12, 0x3ff00000
 ; CI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
; CI-NEXT:    s_endpgm
   store <4 x double> %result, <4 x double> addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @round_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %in) #0 {
 ; SI-LABEL: round_v8f64:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x19
-; SI-NEXT:    s_mov_b32 s22, -1
+; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    s_mov_b32 s21, 0xfffff
-; SI-NEXT:    s_mov_b32 s20, s22
 ; SI-NEXT:    v_mov_b32_e32 v8, 0x3ff00000
+; SI-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    s_bfe_u32 s2, s7, 0xb0014
-; SI-NEXT:    s_add_i32 s26, s2, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[2:3], s[20:21], s26
-; SI-NEXT:    s_and_b32 s23, s7, 0x80000000
-; SI-NEXT:    s_andn2_b64 s[24:25], s[6:7], s[2:3]
-; SI-NEXT:    s_cmp_lt_i32 s26, 0
-; SI-NEXT:    v_mov_b32_e32 v0, s25
-; SI-NEXT:    v_mov_b32_e32 v1, s23
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s26, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s7
-; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[2:3]
-; SI-NEXT:    v_mov_b32_e32 v0, s24
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, s6
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[2:3]
-; SI-NEXT:    v_add_f64 v[2:3], s[6:7], -v[0:1]
-; SI-NEXT:    s_bfe_u32 s2, s5, 0xb0014
-; SI-NEXT:    s_add_i32 s24, s2, 0xfffffc01
-; SI-NEXT:    s_brev_b32 s23, -2
-; SI-NEXT:    v_mov_b32_e32 v4, s7
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[2:3]|, 0.5
-; SI-NEXT:    s_lshr_b64 s[2:3], s[20:21], s24
-; SI-NEXT:    v_bfi_b32 v4, s23, v8, v4
-; SI-NEXT:    s_andn2_b64 s[6:7], s[4:5], s[2:3]
-; SI-NEXT:    s_and_b32 s2, s5, 0x80000000
-; SI-NEXT:    v_cndmask_b32_e32 v3, 0, v4, vcc
-; SI-NEXT:    v_mov_b32_e32 v2, 0
+; SI-NEXT:    s_bfe_u32 s20, s7, 0xb0014
+; SI-NEXT:    s_add_i32 s24, s20, 0xfffffc01
+; SI-NEXT:    s_mov_b32 s20, s2
+; SI-NEXT:    s_lshr_b64 s[22:23], s[20:21], s24
+; SI-NEXT:    s_andn2_b64 s[22:23], s[6:7], s[22:23]
+; SI-NEXT:    s_and_b32 s25, s7, 0x80000000
 ; SI-NEXT:    s_cmp_lt_i32 s24, 0
-; SI-NEXT:    v_add_f64 v[2:3], v[0:1], v[2:3]
-; SI-NEXT:    v_mov_b32_e32 v0, s7
-; SI-NEXT:    v_mov_b32_e32 v1, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    s_cselect_b32 s22, 0, s22
+; SI-NEXT:    s_cselect_b32 s23, s25, s23
 ; SI-NEXT:    s_cmp_gt_i32 s24, 51
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; SI-NEXT:    v_mov_b32_e32 v1, s5
-; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[2:3]
-; SI-NEXT:    v_mov_b32_e32 v0, s6
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, 0, vcc
+; SI-NEXT:    s_cselect_b32 s22, s6, s22
+; SI-NEXT:    s_cselect_b32 s23, s7, s23
+; SI-NEXT:    v_mov_b32_e32 v0, s22
+; SI-NEXT:    v_mov_b32_e32 v1, s23
+; SI-NEXT:    v_add_f64 v[0:1], s[6:7], -v[0:1]
+; SI-NEXT:    s_brev_b32 s6, -2
+; SI-NEXT:    v_mov_b32_e32 v2, s7
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    v_bfi_b32 v2, s6, v8, v2
+; SI-NEXT:    s_bfe_u32 s7, s5, 0xb0014
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v2, vcc
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    s_addk_i32 s7, 0xfc01
+; SI-NEXT:    v_add_f64 v[2:3], s[22:23], v[0:1]
+; SI-NEXT:    s_lshr_b64 s[22:23], s[20:21], s7
+; SI-NEXT:    s_andn2_b64 s[22:23], s[4:5], s[22:23]
+; SI-NEXT:    s_and_b32 s24, s5, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b32 s22, 0, s22
+; SI-NEXT:    s_cselect_b32 s23, s24, s23
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b32 s22, s4, s22
+; SI-NEXT:    s_cselect_b32 s23, s5, s23
+; SI-NEXT:    v_mov_b32_e32 v0, s22
+; SI-NEXT:    v_mov_b32_e32 v1, s23
+; SI-NEXT:    v_add_f64 v[0:1], s[4:5], -v[0:1]
+; SI-NEXT:    v_mov_b32_e32 v4, s5
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[0:1]|, 0.5
+; SI-NEXT:    s_bfe_u32 s4, s11, 0xb0014
+; SI-NEXT:    v_bfi_b32 v4, s6, v8, v4
+; SI-NEXT:    s_add_i32 s7, s4, 0xfffffc01
+; SI-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
+; SI-NEXT:    v_mov_b32_e32 v0, 0
+; SI-NEXT:    s_lshr_b64 s[4:5], s[20:21], s7
+; SI-NEXT:    v_add_f64 v[0:1], s[22:23], v[0:1]
+; SI-NEXT:    s_andn2_b64 s[4:5], s[10:11], s[4:5]
+; SI-NEXT:    s_and_b32 s22, s11, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b32 s4, 0, s4
+; SI-NEXT:    s_cselect_b32 s5, s22, s5
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b32 s4, s10, s4
+; SI-NEXT:    s_cselect_b32 s5, s11, s5
 ; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[2:3]
-; SI-NEXT:    v_add_f64 v[4:5], s[4:5], -v[0:1]
-; SI-NEXT:    s_bfe_u32 s2, s11, 0xb0014
-; SI-NEXT:    s_add_i32 s6, s2, 0xfffffc01
-; SI-NEXT:    v_mov_b32_e32 v6, s5
+; SI-NEXT:    v_mov_b32_e32 v5, s5
+; SI-NEXT:    v_add_f64 v[4:5], s[10:11], -v[4:5]
+; SI-NEXT:    v_mov_b32_e32 v6, s11
 ; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[4:5]|, 0.5
-; SI-NEXT:    s_lshr_b64 s[2:3], s[20:21], s6
-; SI-NEXT:    v_bfi_b32 v6, s23, v8, v6
-; SI-NEXT:    s_andn2_b64 s[4:5], s[10:11], s[2:3]
-; SI-NEXT:    s_and_b32 s2, s11, 0x80000000
+; SI-NEXT:    v_bfi_b32 v6, s6, v8, v6
 ; SI-NEXT:    v_cndmask_b32_e32 v5, 0, v6, vcc
 ; SI-NEXT:    v_mov_b32_e32 v4, 0
-; SI-NEXT:    s_cmp_lt_i32 s6, 0
-; SI-NEXT:    v_add_f64 v[0:1], v[0:1], v[4:5]
-; SI-NEXT:    v_mov_b32_e32 v4, s5
-; SI-NEXT:    v_mov_b32_e32 v5, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s6, 51
-; SI-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s11
-; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v5, v4, v5, s[2:3]
-; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v6, s10
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v6, s[2:3]
-; SI-NEXT:    v_add_f64 v[6:7], s[10:11], -v[4:5]
-; SI-NEXT:    s_bfe_u32 s2, s9, 0xb0014
-; SI-NEXT:    s_add_i32 s6, s2, 0xfffffc01
-; SI-NEXT:    v_mov_b32_e32 v9, s11
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[6:7]|, 0.5
-; SI-NEXT:    s_lshr_b64 s[2:3], s[20:21], s6
-; SI-NEXT:    v_bfi_b32 v9, s23, v8, v9
-; SI-NEXT:    s_andn2_b64 s[4:5], s[8:9], s[2:3]
-; SI-NEXT:    s_and_b32 s2, s9, 0x80000000
-; SI-NEXT:    v_cndmask_b32_e32 v7, 0, v9, vcc
-; SI-NEXT:    v_mov_b32_e32 v6, 0
-; SI-NEXT:    s_cmp_lt_i32 s6, 0
-; SI-NEXT:    v_add_f64 v[6:7], v[4:5], v[6:7]
-; SI-NEXT:    v_mov_b32_e32 v4, s5
-; SI-NEXT:    v_mov_b32_e32 v5, s2
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s6, 51
-; SI-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
-; SI-NEXT:    v_mov_b32_e32 v5, s9
-; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v5, v4, v5, s[2:3]
+; SI-NEXT:    v_add_f64 v[6:7], s[4:5], v[4:5]
+; SI-NEXT:    s_bfe_u32 s4, s9, 0xb0014
+; SI-NEXT:    s_add_i32 s7, s4, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[4:5], s[20:21], s7
+; SI-NEXT:    s_andn2_b64 s[4:5], s[8:9], s[4:5]
+; SI-NEXT:    s_and_b32 s10, s9, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b32 s4, 0, s4
+; SI-NEXT:    s_cselect_b32 s5, s10, s5
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b32 s4, s8, s4
+; SI-NEXT:    s_cselect_b32 s5, s9, s5
 ; SI-NEXT:    v_mov_b32_e32 v4, s4
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v9, s8
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v9, s[2:3]
-; SI-NEXT:    s_bfe_u32 s2, s15, 0xb0014
-; SI-NEXT:    v_add_f64 v[9:10], s[8:9], -v[4:5]
-; SI-NEXT:    s_add_i32 s4, s2, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[2:3], s[20:21], s4
-; SI-NEXT:    v_mov_b32_e32 v11, s9
+; SI-NEXT:    v_mov_b32_e32 v5, s5
+; SI-NEXT:    v_add_f64 v[4:5], s[8:9], -v[4:5]
+; SI-NEXT:    v_mov_b32_e32 v9, s9
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[4:5]|, 0.5
+; SI-NEXT:    v_bfi_b32 v9, s6, v8, v9
+; SI-NEXT:    v_cndmask_b32_e32 v5, 0, v9, vcc
+; SI-NEXT:    v_mov_b32_e32 v4, 0
+; SI-NEXT:    v_add_f64 v[4:5], s[4:5], v[4:5]
+; SI-NEXT:    s_bfe_u32 s4, s15, 0xb0014
+; SI-NEXT:    s_add_i32 s7, s4, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[4:5], s[20:21], s7
+; SI-NEXT:    s_andn2_b64 s[4:5], s[14:15], s[4:5]
+; SI-NEXT:    s_and_b32 s8, s15, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b32 s4, 0, s4
+; SI-NEXT:    s_cselect_b32 s5, s8, s5
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b32 s5, s15, s5
+; SI-NEXT:    s_cselect_b32 s4, s14, s4
+; SI-NEXT:    v_mov_b32_e32 v10, s5
+; SI-NEXT:    v_mov_b32_e32 v9, s4
+; SI-NEXT:    v_add_f64 v[9:10], s[14:15], -v[9:10]
+; SI-NEXT:    v_mov_b32_e32 v11, s15
 ; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[9:10]|, 0.5
-; SI-NEXT:    s_andn2_b64 s[24:25], s[14:15], s[2:3]
-; SI-NEXT:    s_and_b32 s2, s15, 0x80000000
-; SI-NEXT:    v_bfi_b32 v11, s23, v8, v11
-; SI-NEXT:    s_cmp_lt_i32 s4, 0
+; SI-NEXT:    v_bfi_b32 v11, s6, v8, v11
 ; SI-NEXT:    v_cndmask_b32_e32 v10, 0, v11, vcc
 ; SI-NEXT:    v_mov_b32_e32 v9, 0
-; SI-NEXT:    s_cselect_b64 vcc, -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s4, 51
-; SI-NEXT:    v_add_f64 v[4:5], v[4:5], v[9:10]
-; SI-NEXT:    v_mov_b32_e32 v10, s2
-; SI-NEXT:    s_cselect_b64 s[2:3], -1, 0
+; SI-NEXT:    v_add_f64 v[10:11], s[4:5], v[9:10]
 ; SI-NEXT:    s_bfe_u32 s4, s13, 0xb0014
-; SI-NEXT:    s_add_i32 s6, s4, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[4:5], s[20:21], s6
-; SI-NEXT:    s_andn2_b64 s[26:27], s[12:13], s[4:5]
-; SI-NEXT:    s_and_b32 s4, s13, 0x80000000
-; SI-NEXT:    v_mov_b32_e32 v9, s25
-; SI-NEXT:    s_cmp_lt_i32 s6, 0
-; SI-NEXT:    v_cndmask_b32_e32 v15, v9, v10, vcc
-; SI-NEXT:    v_mov_b32_e32 v10, s4
-; SI-NEXT:    s_cselect_b64 s[4:5], -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s6, 51
-; SI-NEXT:    s_cselect_b64 s[6:7], -1, 0
-; SI-NEXT:    s_bfe_u32 s8, s19, 0xb0014
-; SI-NEXT:    s_add_i32 s10, s8, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[8:9], s[20:21], s10
-; SI-NEXT:    s_andn2_b64 s[28:29], s[18:19], s[8:9]
-; SI-NEXT:    s_and_b32 s8, s19, 0x80000000
-; SI-NEXT:    v_mov_b32_e32 v9, s27
-; SI-NEXT:    s_cmp_lt_i32 s10, 0
-; SI-NEXT:    v_cndmask_b32_e64 v17, v9, v10, s[4:5]
-; SI-NEXT:    v_mov_b32_e32 v9, s29
-; SI-NEXT:    v_mov_b32_e32 v10, s8
-; SI-NEXT:    s_cselect_b64 s[8:9], -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s10, 51
-; SI-NEXT:    v_cndmask_b32_e64 v9, v9, v10, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v10, s19
-; SI-NEXT:    s_cselect_b64 s[10:11], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v10, v9, v10, s[10:11]
-; SI-NEXT:    v_mov_b32_e32 v9, s28
-; SI-NEXT:    v_cndmask_b32_e64 v9, v9, 0, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v11, s18
-; SI-NEXT:    s_bfe_u32 s8, s17, 0xb0014
-; SI-NEXT:    v_cndmask_b32_e64 v9, v9, v11, s[10:11]
-; SI-NEXT:    s_add_i32 s10, s8, 0xfffffc01
-; SI-NEXT:    s_lshr_b64 s[8:9], s[20:21], s10
-; SI-NEXT:    s_andn2_b64 s[20:21], s[16:17], s[8:9]
-; SI-NEXT:    s_and_b32 s8, s17, 0x80000000
-; SI-NEXT:    s_cmp_lt_i32 s10, 0
-; SI-NEXT:    v_mov_b32_e32 v11, s21
+; SI-NEXT:    s_add_i32 s7, s4, 0xfffffc01
+; SI-NEXT:    s_lshr_b64 s[4:5], s[20:21], s7
+; SI-NEXT:    s_andn2_b64 s[4:5], s[12:13], s[4:5]
+; SI-NEXT:    s_and_b32 s8, s13, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b32 s4, 0, s4
+; SI-NEXT:    s_cselect_b32 s5, s8, s5
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b32 s5, s13, s5
+; SI-NEXT:    s_cselect_b32 s4, s12, s4
+; SI-NEXT:    s_bfe_u32 s7, s19, 0xb0014
+; SI-NEXT:    s_addk_i32 s7, 0xfc01
+; SI-NEXT:    s_lshr_b64 s[8:9], s[20:21], s7
+; SI-NEXT:    s_andn2_b64 s[8:9], s[18:19], s[8:9]
+; SI-NEXT:    s_and_b32 s10, s19, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    s_cselect_b32 s8, 0, s8
+; SI-NEXT:    s_cselect_b32 s9, s10, s9
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
+; SI-NEXT:    s_cselect_b32 s9, s19, s9
+; SI-NEXT:    s_cselect_b32 s8, s18, s8
+; SI-NEXT:    s_bfe_u32 s7, s17, 0xb0014
+; SI-NEXT:    v_mov_b32_e32 v13, s5
+; SI-NEXT:    s_addk_i32 s7, 0xfc01
+; SI-NEXT:    v_mov_b32_e32 v12, s4
+; SI-NEXT:    s_lshr_b64 s[10:11], s[20:21], s7
+; SI-NEXT:    v_add_f64 v[12:13], s[12:13], -v[12:13]
+; SI-NEXT:    s_andn2_b64 s[10:11], s[16:17], s[10:11]
+; SI-NEXT:    s_and_b32 s12, s17, 0x80000000
+; SI-NEXT:    s_cmp_lt_i32 s7, 0
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[12:13]|, 0.5
+; SI-NEXT:    v_mov_b32_e32 v13, s9
+; SI-NEXT:    s_cselect_b32 s10, 0, s10
+; SI-NEXT:    s_cselect_b32 s11, s12, s11
+; SI-NEXT:    s_cmp_gt_i32 s7, 51
 ; SI-NEXT:    v_mov_b32_e32 v12, s8
-; SI-NEXT:    s_cselect_b64 s[8:9], -1, 0
-; SI-NEXT:    s_cmp_gt_i32 s10, 51
-; SI-NEXT:    v_cndmask_b32_e64 v11, v11, v12, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v12, s17
-; SI-NEXT:    s_cselect_b64 s[10:11], -1, 0
-; SI-NEXT:    v_cndmask_b32_e64 v14, v11, v12, s[10:11]
-; SI-NEXT:    v_mov_b32_e32 v11, s20
-; SI-NEXT:    v_cndmask_b32_e64 v11, v11, 0, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v12, s16
-; SI-NEXT:    v_cndmask_b32_e64 v13, v11, v12, s[10:11]
-; SI-NEXT:    v_add_f64 v[11:12], s[16:17], -v[13:14]
-; SI-NEXT:    v_mov_b32_e32 v19, s17
-; SI-NEXT:    v_cmp_ge_f64_e64 s[8:9], |v[11:12]|, 0.5
-; SI-NEXT:    v_mov_b32_e32 v11, s19
-; SI-NEXT:    v_bfi_b32 v20, s23, v8, v11
-; SI-NEXT:    v_add_f64 v[11:12], s[18:19], -v[9:10]
-; SI-NEXT:    v_bfi_b32 v19, s23, v8, v19
-; SI-NEXT:    v_cmp_ge_f64_e64 s[10:11], |v[11:12]|, 0.5
-; SI-NEXT:    v_mov_b32_e32 v11, 0
-; SI-NEXT:    v_cndmask_b32_e64 v12, 0, v20, s[10:11]
-; SI-NEXT:    v_add_f64 v[11:12], v[9:10], v[11:12]
-; SI-NEXT:    v_cndmask_b32_e64 v10, 0, v19, s[8:9]
-; SI-NEXT:    v_mov_b32_e32 v9, 0
-; SI-NEXT:    v_mov_b32_e32 v16, s15
-; SI-NEXT:    v_add_f64 v[9:10], v[13:14], v[9:10]
-; SI-NEXT:    v_mov_b32_e32 v13, s24
-; SI-NEXT:    v_cndmask_b32_e64 v14, v15, v16, s[2:3]
-; SI-NEXT:    v_cndmask_b32_e64 v13, v13, 0, vcc
-; SI-NEXT:    v_mov_b32_e32 v15, s14
-; SI-NEXT:    v_cndmask_b32_e64 v13, v13, v15, s[2:3]
-; SI-NEXT:    v_mov_b32_e32 v15, s15
-; SI-NEXT:    v_bfi_b32 v19, s23, v8, v15
-; SI-NEXT:    v_mov_b32_e32 v15, s26
-; SI-NEXT:    v_mov_b32_e32 v18, s13
-; SI-NEXT:    v_cndmask_b32_e64 v15, v15, 0, s[4:5]
-; SI-NEXT:    v_mov_b32_e32 v16, s12
-; SI-NEXT:    v_cndmask_b32_e64 v18, v17, v18, s[6:7]
-; SI-NEXT:    v_cndmask_b32_e64 v17, v15, v16, s[6:7]
-; SI-NEXT:    v_mov_b32_e32 v15, s13
-; SI-NEXT:    v_bfi_b32 v8, s23, v8, v15
-; SI-NEXT:    v_add_f64 v[15:16], s[12:13], -v[17:18]
-; SI-NEXT:    s_load_dwordx2 s[20:21], s[0:1], 0x9
-; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[15:16]|, 0.5
-; SI-NEXT:    v_add_f64 v[15:16], s[14:15], -v[13:14]
-; SI-NEXT:    s_mov_b32 s23, 0xf000
-; SI-NEXT:    v_cmp_ge_f64_e64 s[0:1], |v[15:16]|, 0.5
-; SI-NEXT:    v_mov_b32_e32 v15, 0
-; SI-NEXT:    v_cndmask_b32_e64 v16, 0, v19, s[0:1]
-; SI-NEXT:    v_add_f64 v[15:16], v[13:14], v[15:16]
-; SI-NEXT:    v_cndmask_b32_e32 v14, 0, v8, vcc
-; SI-NEXT:    v_mov_b32_e32 v13, 0
-; SI-NEXT:    v_add_f64 v[13:14], v[17:18], v[13:14]
-; SI-NEXT:    s_waitcnt lgkmcnt(0)
-; SI-NEXT:    buffer_store_dwordx4 v[9:12], off, s[20:23], 0 offset:48
-; SI-NEXT:    buffer_store_dwordx4 v[13:16], off, s[20:23], 0 offset:32
-; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[20:23], 0 offset:16
-; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[20:23], 0
+; SI-NEXT:    s_cselect_b32 s11, s17, s11
+; SI-NEXT:    v_mov_b32_e32 v9, s13
+; SI-NEXT:    v_add_f64 v[12:13], s[18:19], -v[12:13]
+; SI-NEXT:    s_cselect_b32 s10, s16, s10
+; SI-NEXT:    v_mov_b32_e32 v15, s11
+; SI-NEXT:    v_bfi_b32 v9, s6, v8, v9
+; SI-NEXT:    v_mov_b32_e32 v14, s10
+; SI-NEXT:    v_cndmask_b32_e32 v17, 0, v9, vcc
+; SI-NEXT:    v_mov_b32_e32 v9, s19
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[12:13]|, 0.5
+; SI-NEXT:    v_add_f64 v[14:15], s[16:17], -v[14:15]
+; SI-NEXT:    v_bfi_b32 v9, s6, v8, v9
+; SI-NEXT:    v_cndmask_b32_e32 v13, 0, v9, vcc
+; SI-NEXT:    v_mov_b32_e32 v9, s17
+; SI-NEXT:    v_cmp_ge_f64_e64 vcc, |v[14:15]|, 0.5
+; SI-NEXT:    v_bfi_b32 v8, s6, v8, v9
+; SI-NEXT:    v_mov_b32_e32 v12, 0
+; SI-NEXT:    v_cndmask_b32_e32 v9, 0, v8, vcc
+; SI-NEXT:    v_mov_b32_e32 v8, 0
+; SI-NEXT:    v_mov_b32_e32 v16, 0
+; SI-NEXT:    v_add_f64 v[14:15], s[8:9], v[12:13]
+; SI-NEXT:    v_add_f64 v[12:13], s[10:11], v[8:9]
+; SI-NEXT:    s_mov_b32 s3, 0xf000
+; SI-NEXT:    v_add_f64 v[8:9], s[4:5], v[16:17]
+; SI-NEXT:    buffer_store_dwordx4 v[12:15], off, s[0:3], 0 offset:48
+; SI-NEXT:    buffer_store_dwordx4 v[8:11], off, s[0:3], 0 offset:32
+; SI-NEXT:    buffer_store_dwordx4 v[4:7], off, s[0:3], 0 offset:16
+; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; CI-LABEL: round_v8f64:
 ; CI:       ; %bb.0:
 ; CI-NEXT:    s_load_dwordx16 s[4:19], s[0:1], 0x19
 ; CI-NEXT:    s_brev_b32 s2, -2
 ; CI-NEXT:    v_mov_b32_e32 v16, 0x3ff00000
 ; CI-NEXT:    s_load_dwordx2 s[20:21], s[0:1], 0x9

llvm/test/CodeGen/AMDGPU/load-select-ptr.ll

 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
 
 ; Combine on select c, (load x), (load y) -> load (select c, x, y)
 ; drops MachinePointerInfo, so it can't be relied on for correctness.
 
 ; GCN-LABEL: {{^}}select_ptr_crash_i64_flat:
 ; GCN: s_load_dwordx2
 ; GCN: s_load_dwordx2
 ; GCN: s_load_dwordx2
 
 ; GCN: s_cmp_eq_u32
-; GCN: v_cndmask_b32
-; GCN: v_cndmask_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 
 ; GCN-NOT: load_dword
 ; GCN: flat_load_dwordx2
 ; GCN-NOT: load_dword
 
 ; GCN: flat_store_dwordx2
 define amdgpu_kernel void @select_ptr_crash_i64_flat(i32 %tmp, [8 x i32], i64* %ptr0, [8 x i32], i64* %ptr1, [8 x i32], i64 addrspace(1)* %ptr2) {
   %tmp2 = icmp eq i32 %tmp, 0
   %tmp3 = load i64, i64* %ptr0, align 8
   %tmp4 = load i64, i64* %ptr1, align 8
   %tmp5 = select i1 %tmp2, i64 %tmp3, i64 %tmp4
   store i64 %tmp5, i64 addrspace(1)* %ptr2, align 8
   ret void
 }
 
 ; The transform currently doesn't happen for non-addrspace 0, but it
 ; should.
 
 ; GCN-LABEL: {{^}}select_ptr_crash_i64_global:
 ; GCN: s_load_dwordx2
 ; GCN: s_load_dwordx2
 ; GCN: s_load_dwordx2
 ; GCN: s_load_dwordx2 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0x0{{$}}
 ; GCN: s_load_dwordx2 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 0x0{{$}}
-; GCN: v_cndmask_b32
-; GCN: v_cndmask_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 ; GCN: flat_store_dwordx2
 define amdgpu_kernel void @select_ptr_crash_i64_global(i32 %tmp, [8 x i32], i64 addrspace(1)* %ptr0, [8 x i32], i64 addrspace(1)* %ptr1, [8 x i32], i64 addrspace(1)* %ptr2) {
   %tmp2 = icmp eq i32 %tmp, 0
   %tmp3 = load i64, i64 addrspace(1)* %ptr0, align 8
   %tmp4 = load i64, i64 addrspace(1)* %ptr1, align 8
   %tmp5 = select i1 %tmp2, i64 %tmp3, i64 %tmp4
   store i64 %tmp5, i64 addrspace(1)* %ptr2, align 8
   ret void

llvm/test/CodeGen/AMDGPU/mad_uint24.ll

 ; SimplifyDemandedBits on the first operand of the mul instruction
 ; assuming that the pattern would be matched to a 24-bit mad.  This
 ; led to some instructions being incorrectly erased when the entire
 ; 24-bit mad pattern wasn't being matched.
 
 ; Check that the select instruction is not deleted.
 ; FUNC-LABEL: {{^}}i24_i32_i32_mad:
 ; EG: CNDE_INT
-; SI: v_cndmask
+; SI: s_cselect
 ; GCN2: s_cselect
 define amdgpu_kernel void @i24_i32_i32_mad(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %d) {
 entry:
   %0 = ashr i32 %a, 8
   %1 = icmp ne i32 %c, 0
   %2 = select i1 %1, i32 %0, i32 34
   %3 = mul i32 %2, %c
   %4 = add i32 %3, %d

llvm/test/CodeGen/AMDGPU/sad.ll

 
 ; GCN-LABEL: {{^}}s_sad_u32_i8_pat2:
 ; GCN: s_load_dword
 ; GCN-DAG: s_bfe_u32
 ; GCN-DAG: s_sub_i32
 ; GCN-DAG: s_and_b32
 ; GCN-DAG: s_sub_i32
 ; GCN-DAG: s_lshr_b32
-; GCN: v_add_i32_e32
+; GCN: s_add_i32
 define amdgpu_kernel void @s_sad_u32_i8_pat2(i8 addrspace(1)* %out, i8 zeroext %a, i8 zeroext %b, i8 zeroext %c) {
   %icmp0 = icmp ugt i8 %a, %b
   %sub0 = sub i8 %a, %b
   %sub1 = sub i8 %b, %a
   %ret0 = select i1 %icmp0, i8 %sub0, i8 %sub1
 
   %ret = add i8 %ret0, %c
 
   store i8 %ret, i8 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_sad_u32_mismatched_operands_pat1:
 ; GCN-DAG: s_cmp_le_u32 s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN-DAG: s_max_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
-; GCN: v_sub_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
-; GCN: v_add_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GCN: s_sub_i32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN: s_add_i32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 define amdgpu_kernel void @v_sad_u32_mismatched_operands_pat1(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %d) {
   %icmp0 = icmp ugt i32 %a, %b
   %t0 = select i1 %icmp0, i32 %a, i32 %b
 
   %icmp1 = icmp ule i32 %a, %b
   %t1 = select i1 %icmp1, i32 %a, i32 %d
 
   %ret0 = sub i32 %t0, %t1
   %ret = add i32 %ret0, %c
 
   store i32 %ret, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_sad_u32_mismatched_operands_pat2:
 ; GCN: s_sub_i32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 ; GCN: s_sub_i32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
-; GCN: v_add_i32_e32 v{{[0-9]+}}, vcc, s{{[0-9]+}}, v{{[0-9]+}}
+; GCN: s_add_i32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 define amdgpu_kernel void @v_sad_u32_mismatched_operands_pat2(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %d) {
   %icmp0 = icmp ugt i32 %a, %b
   %sub0 = sub i32 %a, %d
   %sub1 = sub i32 %b, %a
   %ret0 = select i1 %icmp0, i32 %sub0, i32 %sub1
 
   %ret = add i32 %ret0, %c
 
   store i32 %ret, i32 addrspace(1)* %out
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/sdiv.ll

 ; GCN-NEXT:    v_cvt_u32_f32_e32 v3, v3
 ; GCN-NEXT:    v_xor_b32_e32 v2, v5, v2
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v3, v4
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v3
 ; GCN-NEXT:    v_mul_lo_u32 v4, v3, v1
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, 1, v3
-; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v4, v0
-; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
+; GCN-NEXT:    v_sub_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_subrev_i32_e32 v4, vcc, v1, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
+; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
+; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, 1, v3
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v3, v4, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v0, v0, v2
 ; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v2, v0
 ; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;

 ; TONGA-NEXT:    v_cvt_u32_f32_e32 v3, v3
 ; TONGA-NEXT:    v_xor_b32_e32 v2, v5, v2
 ; TONGA-NEXT:    v_mul_lo_u32 v4, v4, v3
 ; TONGA-NEXT:    v_mul_hi_u32 v4, v3, v4
 ; TONGA-NEXT:    v_add_u32_e32 v3, vcc, v3, v4
 ; TONGA-NEXT:    v_mul_hi_u32 v3, v0, v3
 ; TONGA-NEXT:    v_mul_lo_u32 v4, v3, v1
 ; TONGA-NEXT:    v_add_u32_e32 v5, vcc, 1, v3
-; TONGA-NEXT:    v_subrev_u32_e32 v0, vcc, v4, v0
-; TONGA-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v1
-; TONGA-NEXT:    v_cndmask_b32_e64 v3, v3, v5, s[0:1]
+; TONGA-NEXT:    v_sub_u32_e32 v0, vcc, v0, v4
 ; TONGA-NEXT:    v_subrev_u32_e32 v4, vcc, v1, v0
-; TONGA-NEXT:    v_cndmask_b32_e64 v0, v0, v4, s[0:1]
+; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
+; TONGA-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
+; TONGA-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; TONGA-NEXT:    v_add_u32_e32 v4, vcc, 1, v3
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
 ; TONGA-NEXT:    v_cndmask_b32_e32 v0, v3, v4, vcc
 ; TONGA-NEXT:    v_xor_b32_e32 v0, v0, v2
 ; TONGA-NEXT:    v_subrev_u32_e32 v0, vcc, v2, v0
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s6, -1
 ; GFX9-NEXT:    s_mov_b32 s10, s6
 ; GFX9-NEXT:    s_mov_b32 s11, s7
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s8, s2
 ; GFX9-NEXT:    s_mov_b32 s9, s3
 ; GFX9-NEXT:    buffer_load_dwordx2 v[0:1], off, s[8:11], 0
 ; GFX9-NEXT:    s_mov_b32 s4, s0
 ; GFX9-NEXT:    s_mov_b32 s5, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_ashrrev_i32_e32 v2, 31, v1
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v2
-; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v2
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v3, v1
-; GFX9-NEXT:    v_sub_u32_e32 v4, 0, v1
-; GFX9-NEXT:    v_ashrrev_i32_e32 v5, 31, v0
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v5
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v5
-; GFX9-NEXT:    v_xor_b32_e32 v2, v5, v2
-; GFX9-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v4, v3
-; GFX9-NEXT:    v_mul_hi_u32 v4, v3, v4
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v4
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v3, v1
-; GFX9-NEXT:    v_add_u32_e32 v5, 1, v3
-; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v4
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
-; GFX9-NEXT:    v_sub_u32_e32 v4, v0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v3
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v3, v4, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v2
-; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX9-NEXT:    s_ashr_i32 s3, s2, 31
+; GFX9-NEXT:    s_add_i32 s2, s2, s3
+; GFX9-NEXT:    s_xor_b32 s2, s2, s3
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s2
+; GFX9-NEXT:    v_readfirstlane_b32 s0, v0
+; GFX9-NEXT:    s_ashr_i32 s1, s0, 31
+; GFX9-NEXT:    s_add_i32 s0, s0, s1
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX9-NEXT:    s_xor_b32 s3, s1, s3
+; GFX9-NEXT:    s_xor_b32 s0, s0, s1
+; GFX9-NEXT:    s_sub_i32 s1, 0, s2
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v1
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_i32 s1, s1, s8
+; GFX9-NEXT:    s_mul_hi_u32 s1, s8, s1
+; GFX9-NEXT:    s_add_i32 s8, s8, s1
+; GFX9-NEXT:    s_mul_hi_u32 s1, s0, s8
+; GFX9-NEXT:    s_mul_i32 s8, s1, s2
+; GFX9-NEXT:    s_sub_i32 s0, s0, s8
+; GFX9-NEXT:    s_add_i32 s9, s1, 1
+; GFX9-NEXT:    s_sub_i32 s8, s0, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s2
+; GFX9-NEXT:    s_cselect_b32 s1, s9, s1
+; GFX9-NEXT:    s_cselect_b32 s0, s8, s0
+; GFX9-NEXT:    s_add_i32 s8, s1, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s2
+; GFX9-NEXT:    s_cselect_b32 s0, s8, s1
+; GFX9-NEXT:    s_xor_b32 s0, s0, s3
+; GFX9-NEXT:    s_sub_i32 s0, s0, s3
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX9-NEXT:    s_endpgm
 ;
 ; EG-LABEL: sdiv_i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 26, @9, KC0[CB0:0-32], KC1[]

 ; GCN-NEXT:    s_mov_b32 s8, s2
 ; GCN-NEXT:    s_mov_b32 s9, s3
 ; GCN-NEXT:    buffer_load_dword v0, off, s[8:11], 0
 ; GCN-NEXT:    s_mov_b32 s2, 0x98a1930b
 ; GCN-NEXT:    s_mov_b32 s4, s0
 ; GCN-NEXT:    s_mov_b32 s5, s1
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_mul_hi_i32 v1, v0, s2
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GCN-NEXT:    v_lshrrev_b32_e32 v1, 31, v0
 ; GCN-NEXT:    v_ashrrev_i32_e32 v0, 11, v0
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; TONGA-LABEL: slow_sdiv_i32_3435:
 ; TONGA:       ; %bb.0:
 ; TONGA-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; TONGA-NEXT:    s_mov_b32 s7, 0xf000
 ; TONGA-NEXT:    s_mov_b32 s6, -1
 ; TONGA-NEXT:    s_mov_b32 s10, s6
 ; TONGA-NEXT:    s_mov_b32 s11, s7
 ; TONGA-NEXT:    s_waitcnt lgkmcnt(0)
 ; TONGA-NEXT:    s_mov_b32 s8, s2
 ; TONGA-NEXT:    s_mov_b32 s9, s3
 ; TONGA-NEXT:    buffer_load_dword v0, off, s[8:11], 0
 ; TONGA-NEXT:    s_mov_b32 s2, 0x98a1930b
 ; TONGA-NEXT:    s_mov_b32 s4, s0
 ; TONGA-NEXT:    s_mov_b32 s5, s1
 ; TONGA-NEXT:    s_waitcnt vmcnt(0)
 ; TONGA-NEXT:    v_mul_hi_i32 v1, v0, s2
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; TONGA-NEXT:    v_lshrrev_b32_e32 v1, 31, v0
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v0, 11, v0
 ; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: slow_sdiv_i32_3435:
 ; GFX9:       ; %bb.0:

 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v4, v0
 ; GCN-NEXT:    v_mul_lo_u32 v10, v10, v5
 ; GCN-NEXT:    v_mul_lo_u32 v11, v11, v7
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v6, v1
 ; GCN-NEXT:    v_xor_b32_e32 v0, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v4, v5, v10
 ; GCN-NEXT:    v_xor_b32_e32 v1, v1, v6
 ; GCN-NEXT:    v_mul_hi_u32 v6, v7, v11
-; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
-; GCN-NEXT:    v_add_i32_e32 v5, vcc, v7, v6
+; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v6, v7
 ; GCN-NEXT:    v_mul_hi_u32 v4, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v5
 ; GCN-NEXT:    v_mul_lo_u32 v6, v4, v2
 ; GCN-NEXT:    v_mul_lo_u32 v10, v5, v3
 ; GCN-NEXT:    v_add_i32_e32 v7, vcc, 1, v4
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, v0, v6
-; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v10, v1
+; GCN-NEXT:    v_sub_i32_e32 v1, vcc, v1, v10
 ; GCN-NEXT:    v_add_i32_e32 v11, vcc, 1, v5
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v2
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v3
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
 ; GCN-NEXT:    v_subrev_i32_e32 v6, vcc, v2, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v5, v5, v11, s[2:3]
+; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
 ; GCN-NEXT:    v_subrev_i32_e32 v7, vcc, v3, v1
+; GCN-NEXT:    v_cndmask_b32_e64 v5, v5, v11, s[2:3]
 ; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, 1, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[2:3]
 ; GCN-NEXT:    v_add_i32_e32 v7, vcc, 1, v5
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v4, v6, vcc
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v3
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v5, v7, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v0, v0, v8
 ; GCN-NEXT:    v_xor_b32_e32 v1, v1, v9
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, v0, v8
+; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v8, v0
 ; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v9, v1
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; TONGA-LABEL: sdiv_v2i32:
 ; TONGA:       ; %bb.0:
 ; TONGA-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; TONGA-NEXT:    s_mov_b32 s7, 0xf000

 ; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v4, v0
 ; TONGA-NEXT:    v_mul_lo_u32 v10, v10, v5
 ; TONGA-NEXT:    v_mul_lo_u32 v11, v11, v7
 ; TONGA-NEXT:    v_add_u32_e32 v1, vcc, v6, v1
 ; TONGA-NEXT:    v_xor_b32_e32 v0, v0, v4
 ; TONGA-NEXT:    v_mul_hi_u32 v4, v5, v10
 ; TONGA-NEXT:    v_xor_b32_e32 v1, v1, v6
 ; TONGA-NEXT:    v_mul_hi_u32 v6, v7, v11
-; TONGA-NEXT:    v_add_u32_e32 v4, vcc, v4, v5
-; TONGA-NEXT:    v_add_u32_e32 v5, vcc, v7, v6
+; TONGA-NEXT:    v_add_u32_e32 v4, vcc, v5, v4
+; TONGA-NEXT:    v_add_u32_e32 v5, vcc, v6, v7
 ; TONGA-NEXT:    v_mul_hi_u32 v4, v0, v4
 ; TONGA-NEXT:    v_mul_hi_u32 v5, v1, v5
 ; TONGA-NEXT:    v_mul_lo_u32 v6, v4, v2
 ; TONGA-NEXT:    v_mul_lo_u32 v10, v5, v3
 ; TONGA-NEXT:    v_add_u32_e32 v7, vcc, 1, v4
 ; TONGA-NEXT:    v_sub_u32_e32 v0, vcc, v0, v6
 ; TONGA-NEXT:    v_sub_u32_e32 v1, vcc, v1, v10
 ; TONGA-NEXT:    v_add_u32_e32 v11, vcc, 1, v5
 ; TONGA-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v2
 ; TONGA-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v3
-; TONGA-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
 ; TONGA-NEXT:    v_subrev_u32_e32 v6, vcc, v2, v0
-; TONGA-NEXT:    v_cndmask_b32_e64 v5, v5, v11, s[2:3]
+; TONGA-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
 ; TONGA-NEXT:    v_subrev_u32_e32 v7, vcc, v3, v1
+; TONGA-NEXT:    v_cndmask_b32_e64 v5, v5, v11, s[2:3]
 ; TONGA-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[0:1]
 ; TONGA-NEXT:    v_add_u32_e32 v6, vcc, 1, v4
 ; TONGA-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[2:3]
 ; TONGA-NEXT:    v_add_u32_e32 v7, vcc, 1, v5
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
 ; TONGA-NEXT:    v_cndmask_b32_e32 v0, v4, v6, vcc
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v3
 ; TONGA-NEXT:    v_cndmask_b32_e32 v1, v5, v7, vcc
 ; TONGA-NEXT:    v_xor_b32_e32 v0, v0, v8
 ; TONGA-NEXT:    v_xor_b32_e32 v1, v1, v9
 ; TONGA-NEXT:    v_subrev_u32_e32 v0, vcc, v8, v0
 ; TONGA-NEXT:    v_subrev_u32_e32 v1, vcc, v9, v1
 ; TONGA-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_v2i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX9-NEXT:    s_mov_b32 s7, 0xf000
-; GFX9-NEXT:    s_mov_b32 s6, -1
-; GFX9-NEXT:    s_mov_b32 s10, s6
-; GFX9-NEXT:    s_mov_b32 s11, s7
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    s_mov_b32 s10, s2
+; GFX9-NEXT:    s_mov_b32 s11, s3
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s8, s2
-; GFX9-NEXT:    s_mov_b32 s9, s3
+; GFX9-NEXT:    s_mov_b32 s8, s6
+; GFX9-NEXT:    s_mov_b32 s9, s7
 ; GFX9-NEXT:    buffer_load_dwordx4 v[0:3], off, s[8:11], 0
-; GFX9-NEXT:    s_mov_b32 s4, s0
-; GFX9-NEXT:    s_mov_b32 s5, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_ashrrev_i32_e32 v4, 31, v2
-; GFX9-NEXT:    v_ashrrev_i32_e32 v5, 31, v3
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v5
-; GFX9-NEXT:    v_xor_b32_e32 v2, v2, v4
-; GFX9-NEXT:    v_xor_b32_e32 v3, v3, v5
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v6, v2
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v7, v3
-; GFX9-NEXT:    v_sub_u32_e32 v10, 0, v2
-; GFX9-NEXT:    v_sub_u32_e32 v11, 0, v3
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v6, v6
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v7, v7
-; GFX9-NEXT:    v_ashrrev_i32_e32 v8, 31, v0
-; GFX9-NEXT:    v_ashrrev_i32_e32 v9, 31, v1
-; GFX9-NEXT:    v_mul_f32_e32 v6, 0x4f7ffffe, v6
-; GFX9-NEXT:    v_mul_f32_e32 v7, 0x4f7ffffe, v7
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v6, v6
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v7, v7
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v8
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v9
-; GFX9-NEXT:    v_mul_lo_u32 v10, v10, v6
-; GFX9-NEXT:    v_mul_lo_u32 v11, v11, v7
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v8
-; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v9
-; GFX9-NEXT:    v_mul_hi_u32 v10, v6, v10
-; GFX9-NEXT:    v_mul_hi_u32 v11, v7, v11
-; GFX9-NEXT:    v_xor_b32_e32 v4, v8, v4
-; GFX9-NEXT:    v_xor_b32_e32 v5, v9, v5
-; GFX9-NEXT:    v_add_u32_e32 v6, v6, v10
-; GFX9-NEXT:    v_add_u32_e32 v7, v7, v11
-; GFX9-NEXT:    v_mul_hi_u32 v6, v0, v6
-; GFX9-NEXT:    v_mul_hi_u32 v7, v1, v7
-; GFX9-NEXT:    v_mul_lo_u32 v8, v6, v2
-; GFX9-NEXT:    v_mul_lo_u32 v9, v7, v3
-; GFX9-NEXT:    v_add_u32_e32 v10, 1, v6
-; GFX9-NEXT:    v_add_u32_e32 v11, 1, v7
-; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v8
-; GFX9-NEXT:    v_sub_u32_e32 v1, v1, v9
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
-; GFX9-NEXT:    v_sub_u32_e32 v8, v0, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v10, vcc
-; GFX9-NEXT:    v_cmp_ge_u32_e64 s[0:1], v1, v3
-; GFX9-NEXT:    v_sub_u32_e32 v9, v1, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v8, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v11, s[0:1]
-; GFX9-NEXT:    v_add_u32_e32 v8, 1, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[0:1]
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
-; GFX9-NEXT:    v_add_u32_e32 v9, 1, v7
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v6, v8, vcc
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v7, v9, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v4
-; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v5
-; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v4
-; GFX9-NEXT:    v_sub_u32_e32 v1, v1, v5
-; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX9-NEXT:    v_readfirstlane_b32 s0, v2
+; GFX9-NEXT:    s_ashr_i32 s1, s0, 31
+; GFX9-NEXT:    s_add_i32 s0, s0, s1
+; GFX9-NEXT:    s_xor_b32 s6, s0, s1
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s6
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v0
+; GFX9-NEXT:    s_ashr_i32 s8, s7, 31
+; GFX9-NEXT:    s_add_i32 s7, s7, s8
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX9-NEXT:    s_xor_b32 s9, s8, s1
+; GFX9-NEXT:    s_xor_b32 s1, s7, s8
+; GFX9-NEXT:    s_sub_i32 s7, 0, s6
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v2
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_mov_b32 s0, s4
+; GFX9-NEXT:    v_readfirstlane_b32 s4, v3
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_i32 s7, s7, s8
+; GFX9-NEXT:    s_mul_hi_u32 s7, s8, s7
+; GFX9-NEXT:    s_add_i32 s8, s8, s7
+; GFX9-NEXT:    s_mul_hi_u32 s7, s1, s8
+; GFX9-NEXT:    s_mul_i32 s8, s7, s6
+; GFX9-NEXT:    s_sub_i32 s1, s1, s8
+; GFX9-NEXT:    s_add_i32 s10, s7, 1
+; GFX9-NEXT:    s_sub_i32 s8, s1, s6
+; GFX9-NEXT:    s_cmp_ge_u32 s1, s6
+; GFX9-NEXT:    s_cselect_b32 s7, s10, s7
+; GFX9-NEXT:    s_cselect_b32 s1, s8, s1
+; GFX9-NEXT:    s_add_i32 s8, s7, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s1, s6
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s7
+; GFX9-NEXT:    s_ashr_i32 s7, s4, 31
+; GFX9-NEXT:    s_add_i32 s4, s4, s7
+; GFX9-NEXT:    s_xor_b32 s4, s4, s7
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s4
+; GFX9-NEXT:    s_mov_b32 s1, s5
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v1
+; GFX9-NEXT:    s_ashr_i32 s8, s5, 31
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s6, s6, s9
+; GFX9-NEXT:    s_add_i32 s5, s5, s8
+; GFX9-NEXT:    s_xor_b32 s7, s8, s7
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_sub_i32 s6, s6, s9
+; GFX9-NEXT:    s_xor_b32 s5, s5, s8
+; GFX9-NEXT:    s_sub_i32 s8, 0, s4
+; GFX9-NEXT:    v_readfirstlane_b32 s9, v0
+; GFX9-NEXT:    s_mul_i32 s8, s8, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s9, s8
+; GFX9-NEXT:    s_add_i32 s9, s9, s8
+; GFX9-NEXT:    s_mul_hi_u32 s8, s5, s9
+; GFX9-NEXT:    s_mul_i32 s9, s8, s4
+; GFX9-NEXT:    s_sub_i32 s5, s5, s9
+; GFX9-NEXT:    s_add_i32 s10, s8, 1
+; GFX9-NEXT:    s_sub_i32 s9, s5, s4
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s4
+; GFX9-NEXT:    s_cselect_b32 s8, s10, s8
+; GFX9-NEXT:    s_cselect_b32 s5, s9, s5
+; GFX9-NEXT:    s_add_i32 s9, s8, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s4
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s8
+; GFX9-NEXT:    s_xor_b32 s4, s4, s7
+; GFX9-NEXT:    s_sub_i32 s4, s4, s7
+; GFX9-NEXT:    v_mov_b32_e32 v0, s6
+; GFX9-NEXT:    v_mov_b32_e32 v1, s4
+; GFX9-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX9-NEXT:    s_endpgm
 ;
 ; EG-LABEL: sdiv_v2i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 51, @9, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1

 ; GCN-NEXT:    v_mul_lo_u32 v9, v9, v8
 ; GCN-NEXT:    v_xor_b32_e32 v5, v5, v11
 ; GCN-NEXT:    v_xor_b32_e32 v6, v6, v13
 ; GCN-NEXT:    v_xor_b32_e32 v16, v10, v11
 ; GCN-NEXT:    v_mul_hi_u32 v9, v8, v9
 ; GCN-NEXT:    v_xor_b32_e32 v1, v1, v10
 ; GCN-NEXT:    v_cvt_f32_u32_e32 v10, v5
 ; GCN-NEXT:    v_cvt_f32_u32_e32 v11, v6
-; GCN-NEXT:    v_add_i32_e32 v8, vcc, v9, v8
+; GCN-NEXT:    v_add_i32_e32 v8, vcc, v8, v9
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v10, v10
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v11, v11
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v8
 ; GCN-NEXT:    v_ashrrev_i32_e32 v12, 31, v2
 ; GCN-NEXT:    v_mul_f32_e32 v9, 0x4f7ffffe, v10
 ; GCN-NEXT:    v_mul_f32_e32 v10, 0x4f7ffffe, v11
 ; GCN-NEXT:    v_mul_lo_u32 v11, v8, v4
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v9, v9

 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v4
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, 0, v6
 ; GCN-NEXT:    v_mul_lo_u32 v0, v0, v10
 ; GCN-NEXT:    v_xor_b32_e32 v4, v7, v14
 ; GCN-NEXT:    v_mul_hi_u32 v7, v9, v12
 ; GCN-NEXT:    v_cvt_f32_u32_e32 v12, v4
 ; GCN-NEXT:    v_mul_hi_u32 v0, v10, v0
 ; GCN-NEXT:    v_add_i32_e32 v11, vcc, 1, v8
-; GCN-NEXT:    v_add_i32_e32 v7, vcc, v9, v7
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v7, v9
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v7
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v10
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v10, v0
 ; GCN-NEXT:    v_mul_hi_u32 v0, v2, v0
 ; GCN-NEXT:    v_mul_lo_u32 v10, v7, v5
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v12, v12
 ; GCN-NEXT:    v_sub_i32_e32 v9, vcc, 0, v4
-; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v10, v1
+; GCN-NEXT:    v_sub_i32_e32 v1, vcc, v1, v10
 ; GCN-NEXT:    v_mul_lo_u32 v10, v0, v6
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v5
 ; GCN-NEXT:    v_mul_f32_e32 v12, 0x4f7ffffe, v12
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v12, v12
-; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v10, v2
+; GCN-NEXT:    v_sub_i32_e32 v2, vcc, v2, v10
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, 1, v7
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, v7, v10, s[2:3]
 ; GCN-NEXT:    v_add_i32_e32 v10, vcc, 1, v0
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[4:5], v2, v6
 ; GCN-NEXT:    v_cndmask_b32_e64 v10, v0, v10, s[4:5]
 ; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v5, v1
 ; GCN-NEXT:    v_cndmask_b32_e64 v0, v1, v0, s[2:3]
 ; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v6, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, v2, v1, s[4:5]
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, 1, v7
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v8, v8, v11, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v7, v1, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v1, v8, v15
 ; GCN-NEXT:    v_xor_b32_e32 v5, v0, v16
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, v1, v15
+; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v15, v1
 ; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v16, v5
 ; GCN-NEXT:    v_mul_lo_u32 v5, v9, v12
 ; GCN-NEXT:    v_ashrrev_i32_e32 v8, 31, v3
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v8, v3
 ; GCN-NEXT:    v_mul_hi_u32 v5, v12, v5
 ; GCN-NEXT:    v_xor_b32_e32 v3, v3, v8
 ; GCN-NEXT:    v_add_i32_e32 v7, vcc, 1, v10
 ; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v12
 ; GCN-NEXT:    v_mul_hi_u32 v5, v3, v5
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v6
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, v10, v7, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v2, v2, v17
 ; GCN-NEXT:    v_mul_lo_u32 v6, v5, v4
 ; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v17, v2
 ; GCN-NEXT:    v_xor_b32_e32 v7, v8, v14
-; GCN-NEXT:    v_subrev_i32_e32 v3, vcc, v6, v3
+; GCN-NEXT:    v_sub_i32_e32 v3, vcc, v3, v6
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, 1, v5
-; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v3, v4
-; GCN-NEXT:    v_cndmask_b32_e64 v5, v5, v6, s[0:1]
-; GCN-NEXT:    v_subrev_i32_e32 v6, vcc, v4, v3
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; GCN-NEXT:    v_subrev_i32_e32 v8, vcc, v4, v3
+; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v4
+; GCN-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, 1, v5
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v4
 ; GCN-NEXT:    v_cndmask_b32_e32 v3, v5, v6, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v3, v3, v7
 ; GCN-NEXT:    v_subrev_i32_e32 v3, vcc, v7, v3
 ; GCN-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; GCN-NEXT:    s_endpgm
 ;

 ; TONGA-NEXT:    v_mul_lo_u32 v9, v9, v8
 ; TONGA-NEXT:    v_xor_b32_e32 v5, v5, v11
 ; TONGA-NEXT:    v_xor_b32_e32 v6, v6, v13
 ; TONGA-NEXT:    v_xor_b32_e32 v16, v10, v11
 ; TONGA-NEXT:    v_mul_hi_u32 v9, v8, v9
 ; TONGA-NEXT:    v_xor_b32_e32 v1, v1, v10
 ; TONGA-NEXT:    v_cvt_f32_u32_e32 v10, v5
 ; TONGA-NEXT:    v_cvt_f32_u32_e32 v11, v6
-; TONGA-NEXT:    v_add_u32_e32 v8, vcc, v9, v8
+; TONGA-NEXT:    v_add_u32_e32 v8, vcc, v8, v9
 ; TONGA-NEXT:    v_rcp_iflag_f32_e32 v10, v10
 ; TONGA-NEXT:    v_rcp_iflag_f32_e32 v11, v11
 ; TONGA-NEXT:    v_mul_hi_u32 v8, v0, v8
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v12, 31, v2
 ; TONGA-NEXT:    v_mul_f32_e32 v9, 0x4f7ffffe, v10
 ; TONGA-NEXT:    v_mul_f32_e32 v10, 0x4f7ffffe, v11
 ; TONGA-NEXT:    v_mul_lo_u32 v11, v8, v4
 ; TONGA-NEXT:    v_cvt_u32_f32_e32 v9, v9

 ; TONGA-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v4
 ; TONGA-NEXT:    v_sub_u32_e32 v0, vcc, 0, v6
 ; TONGA-NEXT:    v_mul_lo_u32 v0, v0, v10
 ; TONGA-NEXT:    v_xor_b32_e32 v4, v7, v14
 ; TONGA-NEXT:    v_mul_hi_u32 v7, v9, v12
 ; TONGA-NEXT:    v_cvt_f32_u32_e32 v12, v4
 ; TONGA-NEXT:    v_mul_hi_u32 v0, v10, v0
 ; TONGA-NEXT:    v_add_u32_e32 v11, vcc, 1, v8
-; TONGA-NEXT:    v_add_u32_e32 v7, vcc, v9, v7
+; TONGA-NEXT:    v_add_u32_e32 v7, vcc, v7, v9
 ; TONGA-NEXT:    v_mul_hi_u32 v7, v1, v7
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v10
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v10, v0
 ; TONGA-NEXT:    v_mul_hi_u32 v0, v2, v0
 ; TONGA-NEXT:    v_mul_lo_u32 v10, v7, v5
 ; TONGA-NEXT:    v_rcp_iflag_f32_e32 v12, v12
 ; TONGA-NEXT:    v_sub_u32_e32 v9, vcc, 0, v4
 ; TONGA-NEXT:    v_sub_u32_e32 v1, vcc, v1, v10
 ; TONGA-NEXT:    v_mul_lo_u32 v10, v0, v6
 ; TONGA-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v5
 ; TONGA-NEXT:    v_mul_f32_e32 v12, 0x4f7ffffe, v12
 ; TONGA-NEXT:    v_cvt_u32_f32_e32 v12, v12
-; TONGA-NEXT:    v_subrev_u32_e32 v2, vcc, v10, v2
+; TONGA-NEXT:    v_sub_u32_e32 v2, vcc, v2, v10
 ; TONGA-NEXT:    v_add_u32_e32 v10, vcc, 1, v7
 ; TONGA-NEXT:    v_cndmask_b32_e64 v7, v7, v10, s[2:3]
 ; TONGA-NEXT:    v_add_u32_e32 v10, vcc, 1, v0
 ; TONGA-NEXT:    v_cmp_ge_u32_e64 s[4:5], v2, v6
 ; TONGA-NEXT:    v_cndmask_b32_e64 v10, v0, v10, s[4:5]
 ; TONGA-NEXT:    v_subrev_u32_e32 v0, vcc, v5, v1
 ; TONGA-NEXT:    v_cndmask_b32_e64 v0, v1, v0, s[2:3]
 ; TONGA-NEXT:    v_subrev_u32_e32 v1, vcc, v6, v2

 ; TONGA-NEXT:    v_add_u32_e32 v5, vcc, v5, v12
 ; TONGA-NEXT:    v_mul_hi_u32 v5, v3, v5
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v6
 ; TONGA-NEXT:    v_cndmask_b32_e32 v2, v10, v7, vcc
 ; TONGA-NEXT:    v_xor_b32_e32 v2, v2, v17
 ; TONGA-NEXT:    v_mul_lo_u32 v6, v5, v4
 ; TONGA-NEXT:    v_subrev_u32_e32 v2, vcc, v17, v2
 ; TONGA-NEXT:    v_xor_b32_e32 v7, v8, v14
-; TONGA-NEXT:    v_subrev_u32_e32 v3, vcc, v6, v3
+; TONGA-NEXT:    v_sub_u32_e32 v3, vcc, v3, v6
 ; TONGA-NEXT:    v_add_u32_e32 v6, vcc, 1, v5
-; TONGA-NEXT:    v_cmp_ge_u32_e64 s[0:1], v3, v4
-; TONGA-NEXT:    v_cndmask_b32_e64 v5, v5, v6, s[0:1]
-; TONGA-NEXT:    v_subrev_u32_e32 v6, vcc, v4, v3
-; TONGA-NEXT:    v_cndmask_b32_e64 v3, v3, v6, s[0:1]
+; TONGA-NEXT:    v_subrev_u32_e32 v8, vcc, v4, v3
+; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v4
+; TONGA-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
+; TONGA-NEXT:    v_cndmask_b32_e32 v3, v3, v8, vcc
 ; TONGA-NEXT:    v_add_u32_e32 v6, vcc, 1, v5
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v4
 ; TONGA-NEXT:    v_cndmask_b32_e32 v3, v5, v6, vcc
 ; TONGA-NEXT:    v_xor_b32_e32 v3, v3, v7
-; TONGA-NEXT:    v_sub_u32_e32 v3, vcc, v3, v7
+; TONGA-NEXT:    v_subrev_u32_e32 v3, vcc, v7, v3
 ; TONGA-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; TONGA-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: sdiv_v4i32:
 ; GFX9:       ; %bb.0:
-; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX9-NEXT:    s_mov_b32 s11, 0xf000
-; GFX9-NEXT:    s_mov_b32 s10, -1
-; GFX9-NEXT:    s_mov_b32 s6, s10
-; GFX9-NEXT:    s_mov_b32 s7, s11
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_mov_b32 s3, 0xf000
+; GFX9-NEXT:    s_mov_b32 s2, -1
+; GFX9-NEXT:    s_mov_b32 s10, s2
+; GFX9-NEXT:    s_mov_b32 s11, s3
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_mov_b32 s4, s2
-; GFX9-NEXT:    s_mov_b32 s5, s3
-; GFX9-NEXT:    buffer_load_dwordx4 v[0:3], off, s[4:7], 0
-; GFX9-NEXT:    buffer_load_dwordx4 v[4:7], off, s[4:7], 0 offset:16
-; GFX9-NEXT:    s_mov_b32 s8, s0
-; GFX9-NEXT:    s_mov_b32 s9, s1
+; GFX9-NEXT:    s_mov_b32 s8, s6
+; GFX9-NEXT:    s_mov_b32 s9, s7
+; GFX9-NEXT:    buffer_load_dwordx4 v[0:3], off, s[8:11], 0 offset:16
+; GFX9-NEXT:    buffer_load_dwordx4 v[4:7], off, s[8:11], 0
+; GFX9-NEXT:    s_mov_b32 s0, s4
 ; GFX9-NEXT:    s_waitcnt vmcnt(1)
-; GFX9-NEXT:    v_ashrrev_i32_e32 v8, 31, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s1, v0
+; GFX9-NEXT:    s_ashr_i32 s4, s1, 31
+; GFX9-NEXT:    s_add_i32 s1, s1, s4
+; GFX9-NEXT:    s_xor_b32 s6, s1, s4
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s6
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_ashrrev_i32_e32 v9, 31, v4
-; GFX9-NEXT:    v_add_u32_e32 v4, v4, v9
-; GFX9-NEXT:    v_ashrrev_i32_e32 v11, 31, v5
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v8
-; GFX9-NEXT:    v_xor_b32_e32 v4, v4, v9
-; GFX9-NEXT:    v_ashrrev_i32_e32 v10, 31, v1
-; GFX9-NEXT:    v_ashrrev_i32_e32 v13, 31, v6
-; GFX9-NEXT:    v_xor_b32_e32 v16, v8, v9
-; GFX9-NEXT:    v_add_u32_e32 v5, v5, v11
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v8
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v8, v4
-; GFX9-NEXT:    v_ashrrev_i32_e32 v12, 31, v2
-; GFX9-NEXT:    v_ashrrev_i32_e32 v15, 31, v7
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v10
-; GFX9-NEXT:    v_add_u32_e32 v6, v6, v13
-; GFX9-NEXT:    v_xor_b32_e32 v5, v5, v11
-; GFX9-NEXT:    v_ashrrev_i32_e32 v14, 31, v3
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v12
-; GFX9-NEXT:    v_add_u32_e32 v7, v7, v15
-; GFX9-NEXT:    v_xor_b32_e32 v17, v10, v11
-; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v10
-; GFX9-NEXT:    v_xor_b32_e32 v6, v6, v13
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v10, v5
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v14
-; GFX9-NEXT:    v_xor_b32_e32 v18, v12, v13
-; GFX9-NEXT:    v_xor_b32_e32 v2, v2, v12
-; GFX9-NEXT:    v_xor_b32_e32 v7, v7, v15
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v12, v6
-; GFX9-NEXT:    v_xor_b32_e32 v19, v14, v15
-; GFX9-NEXT:    v_xor_b32_e32 v3, v3, v14
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v14, v7
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v8, v8
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v10, v10
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v12, v12
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v14, v14
-; GFX9-NEXT:    v_mul_f32_e32 v8, 0x4f7ffffe, v8
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v8, v8
-; GFX9-NEXT:    v_mul_f32_e32 v10, 0x4f7ffffe, v10
-; GFX9-NEXT:    v_mul_f32_e32 v12, 0x4f7ffffe, v12
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v10, v10
-; GFX9-NEXT:    v_sub_u32_e32 v9, 0, v4
-; GFX9-NEXT:    v_mul_f32_e32 v14, 0x4f7ffffe, v14
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v12, v12
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v14, v14
-; GFX9-NEXT:    v_mul_lo_u32 v9, v9, v8
-; GFX9-NEXT:    v_sub_u32_e32 v11, 0, v5
-; GFX9-NEXT:    v_sub_u32_e32 v13, 0, v6
-; GFX9-NEXT:    v_mul_lo_u32 v11, v11, v10
-; GFX9-NEXT:    v_sub_u32_e32 v15, 0, v7
-; GFX9-NEXT:    v_mul_lo_u32 v13, v13, v12
-; GFX9-NEXT:    v_mul_lo_u32 v15, v15, v14
-; GFX9-NEXT:    v_mul_hi_u32 v9, v8, v9
-; GFX9-NEXT:    v_mul_hi_u32 v11, v10, v11
-; GFX9-NEXT:    v_mul_hi_u32 v13, v12, v13
-; GFX9-NEXT:    v_mul_hi_u32 v15, v14, v15
-; GFX9-NEXT:    v_add_u32_e32 v8, v8, v9
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v8
-; GFX9-NEXT:    v_add_u32_e32 v9, v10, v11
-; GFX9-NEXT:    v_add_u32_e32 v10, v12, v13
-; GFX9-NEXT:    v_mul_hi_u32 v9, v1, v9
-; GFX9-NEXT:    v_add_u32_e32 v11, v14, v15
-; GFX9-NEXT:    v_mul_hi_u32 v10, v2, v10
-; GFX9-NEXT:    v_mul_hi_u32 v11, v3, v11
-; GFX9-NEXT:    v_mul_lo_u32 v12, v8, v4
-; GFX9-NEXT:    v_mul_lo_u32 v14, v9, v5
-; GFX9-NEXT:    v_mul_lo_u32 v15, v10, v6
-; GFX9-NEXT:    v_add_u32_e32 v13, 1, v8
-; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v12
-; GFX9-NEXT:    v_mul_lo_u32 v12, v11, v7
-; GFX9-NEXT:    v_sub_u32_e32 v1, v1, v14
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v4
-; GFX9-NEXT:    v_add_u32_e32 v14, 1, v9
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v15
-; GFX9-NEXT:    v_cndmask_b32_e32 v8, v8, v13, vcc
-; GFX9-NEXT:    v_sub_u32_e32 v13, v0, v4
-; GFX9-NEXT:    v_cmp_ge_u32_e64 s[0:1], v1, v5
-; GFX9-NEXT:    v_add_u32_e32 v15, 1, v10
-; GFX9-NEXT:    v_sub_u32_e32 v3, v3, v12
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, v9, v14, s[0:1]
-; GFX9-NEXT:    v_sub_u32_e32 v14, v1, v5
-; GFX9-NEXT:    v_cmp_ge_u32_e64 s[2:3], v2, v6
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v13, vcc
-; GFX9-NEXT:    v_add_u32_e32 v12, 1, v11
-; GFX9-NEXT:    v_cndmask_b32_e64 v10, v10, v15, s[2:3]
-; GFX9-NEXT:    v_sub_u32_e32 v15, v2, v6
-; GFX9-NEXT:    v_cmp_ge_u32_e64 s[4:5], v3, v7
-; GFX9-NEXT:    v_add_u32_e32 v13, 1, v8
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v14, s[0:1]
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v11, v11, v12, s[4:5]
-; GFX9-NEXT:    v_sub_u32_e32 v12, v3, v7
-; GFX9-NEXT:    v_add_u32_e32 v14, 1, v9
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v2, v15, s[2:3]
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v8, v13, vcc
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v5
-; GFX9-NEXT:    v_add_u32_e32 v15, 1, v10
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v3, v12, s[4:5]
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v9, v14, vcc
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v6
-; GFX9-NEXT:    v_add_u32_e32 v12, 1, v11
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v10, v15, vcc
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v7
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v11, v12, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v16
-; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v17
-; GFX9-NEXT:    v_xor_b32_e32 v2, v2, v18
-; GFX9-NEXT:    v_xor_b32_e32 v3, v3, v19
-; GFX9-NEXT:    v_sub_u32_e32 v0, v0, v16
-; GFX9-NEXT:    v_sub_u32_e32 v1, v1, v17
-; GFX9-NEXT:    v_sub_u32_e32 v2, v2, v18
-; GFX9-NEXT:    v_sub_u32_e32 v3, v3, v19
-; GFX9-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v4
+; GFX9-NEXT:    s_ashr_i32 s9, s8, 31
+; GFX9-NEXT:    s_add_i32 s8, s8, s9
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s4, s9, s4
+; GFX9-NEXT:    s_xor_b32 s8, s8, s9
+; GFX9-NEXT:    s_sub_i32 s9, 0, s6
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v1
+; GFX9-NEXT:    s_mov_b32 s1, s5
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v3
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v0
+; GFX9-NEXT:    s_mul_i32 s9, s9, s10
+; GFX9-NEXT:    s_mul_hi_u32 s9, s10, s9
+; GFX9-NEXT:    s_add_i32 s10, s10, s9
+; GFX9-NEXT:    s_mul_hi_u32 s9, s8, s10
+; GFX9-NEXT:    s_mul_i32 s10, s9, s6
+; GFX9-NEXT:    s_sub_i32 s8, s8, s10
+; GFX9-NEXT:    s_add_i32 s11, s9, 1
+; GFX9-NEXT:    s_sub_i32 s10, s8, s6
+; GFX9-NEXT:    s_cmp_ge_u32 s8, s6
+; GFX9-NEXT:    s_cselect_b32 s9, s11, s9
+; GFX9-NEXT:    s_cselect_b32 s8, s10, s8
+; GFX9-NEXT:    s_add_i32 s10, s9, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s8, s6
+; GFX9-NEXT:    s_cselect_b32 s6, s10, s9
+; GFX9-NEXT:    s_ashr_i32 s8, s7, 31
+; GFX9-NEXT:    s_add_i32 s7, s7, s8
+; GFX9-NEXT:    s_xor_b32 s7, s7, s8
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s7
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v5
+; GFX9-NEXT:    s_ashr_i32 s11, s10, 31
+; GFX9-NEXT:    s_xor_b32 s6, s6, s4
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_add_i32 s10, s10, s11
+; GFX9-NEXT:    s_xor_b32 s8, s11, s8
+; GFX9-NEXT:    s_sub_i32 s4, s6, s4
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s6, s10, s11
+; GFX9-NEXT:    s_sub_i32 s10, 0, s7
+; GFX9-NEXT:    v_readfirstlane_b32 s9, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s11, v0
+; GFX9-NEXT:    s_mul_i32 s10, s10, s11
+; GFX9-NEXT:    s_mul_hi_u32 s10, s11, s10
+; GFX9-NEXT:    s_add_i32 s11, s11, s10
+; GFX9-NEXT:    s_mul_hi_u32 s10, s6, s11
+; GFX9-NEXT:    s_mul_i32 s11, s10, s7
+; GFX9-NEXT:    s_sub_i32 s6, s6, s11
+; GFX9-NEXT:    s_add_i32 s12, s10, 1
+; GFX9-NEXT:    s_sub_i32 s11, s6, s7
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s7
+; GFX9-NEXT:    s_cselect_b32 s10, s12, s10
+; GFX9-NEXT:    s_cselect_b32 s6, s11, s6
+; GFX9-NEXT:    s_add_i32 s11, s10, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s7
+; GFX9-NEXT:    s_cselect_b32 s6, s11, s10
+; GFX9-NEXT:    s_ashr_i32 s7, s9, 31
+; GFX9-NEXT:    s_add_i32 s9, s9, s7
+; GFX9-NEXT:    s_xor_b32 s9, s9, s7
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s9
+; GFX9-NEXT:    v_readfirstlane_b32 s11, v6
+; GFX9-NEXT:    s_ashr_i32 s12, s11, 31
+; GFX9-NEXT:    s_xor_b32 s6, s6, s8
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_add_i32 s11, s11, s12
+; GFX9-NEXT:    s_xor_b32 s7, s12, s7
+; GFX9-NEXT:    s_sub_i32 s6, s6, s8
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s8, s11, s12
+; GFX9-NEXT:    s_sub_i32 s11, 0, s9
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v7
+; GFX9-NEXT:    v_readfirstlane_b32 s12, v0
+; GFX9-NEXT:    s_mul_i32 s11, s11, s12
+; GFX9-NEXT:    s_mul_hi_u32 s11, s12, s11
+; GFX9-NEXT:    s_add_i32 s12, s12, s11
+; GFX9-NEXT:    s_mul_hi_u32 s11, s8, s12
+; GFX9-NEXT:    s_mul_i32 s12, s11, s9
+; GFX9-NEXT:    s_sub_i32 s8, s8, s12
+; GFX9-NEXT:    s_add_i32 s13, s11, 1
+; GFX9-NEXT:    s_sub_i32 s12, s8, s9
+; GFX9-NEXT:    s_cmp_ge_u32 s8, s9
+; GFX9-NEXT:    s_cselect_b32 s11, s13, s11
+; GFX9-NEXT:    s_cselect_b32 s8, s12, s8
+; GFX9-NEXT:    s_add_i32 s12, s11, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s8, s9
+; GFX9-NEXT:    s_cselect_b32 s8, s12, s11
+; GFX9-NEXT:    s_ashr_i32 s9, s5, 31
+; GFX9-NEXT:    s_add_i32 s5, s5, s9
+; GFX9-NEXT:    s_xor_b32 s5, s5, s9
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s5
+; GFX9-NEXT:    v_mov_b32_e32 v0, s4
+; GFX9-NEXT:    v_mov_b32_e32 v1, s6
+; GFX9-NEXT:    s_ashr_i32 s4, s10, 31
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX9-NEXT:    s_xor_b32 s6, s8, s7
+; GFX9-NEXT:    s_xor_b32 s8, s4, s9
+; GFX9-NEXT:    s_sub_i32 s6, s6, s7
+; GFX9-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX9-NEXT:    s_sub_i32 s7, 0, s5
+; GFX9-NEXT:    s_add_i32 s10, s10, s4
+; GFX9-NEXT:    s_xor_b32 s4, s10, s4
+; GFX9-NEXT:    v_readfirstlane_b32 s9, v2
+; GFX9-NEXT:    s_mul_i32 s7, s7, s9
+; GFX9-NEXT:    s_mul_hi_u32 s7, s9, s7
+; GFX9-NEXT:    s_add_i32 s9, s9, s7
+; GFX9-NEXT:    s_mul_hi_u32 s7, s4, s9
+; GFX9-NEXT:    s_mul_i32 s9, s7, s5
+; GFX9-NEXT:    s_sub_i32 s4, s4, s9
+; GFX9-NEXT:    s_add_i32 s10, s7, 1
+; GFX9-NEXT:    s_sub_i32 s9, s4, s5
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s5
+; GFX9-NEXT:    s_cselect_b32 s7, s10, s7
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s4
+; GFX9-NEXT:    s_add_i32 s9, s7, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s5
+; GFX9-NEXT:    s_cselect_b32 s4, s9, s7
+; GFX9-NEXT:    s_xor_b32 s4, s4, s8
+; GFX9-NEXT:    s_sub_i32 s4, s4, s8
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
+; GFX9-NEXT:    v_mov_b32_e32 v3, s4
+; GFX9-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX9-NEXT:    s_endpgm
 ;
 ; EG-LABEL: sdiv_v4i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 101, @11, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T2.XYZW, T0.X, 1

 ; GCN-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
 ; GCN-NEXT:    v_or_b32_e32 v0, 1, v0
 ; GCN-NEXT:    v_mul_f32_e32 v2, v3, v4
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v3, -v2, v1, v3
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
 ; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v1|
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; TONGA-LABEL: v_sdiv_i24:
 ; TONGA:       ; %bb.0:
 ; TONGA-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; TONGA-NEXT:    s_mov_b32 s7, 0xf000

 ; GCN-NEXT:    v_xor_b32_e32 v5, v5, v0
 ; GCN-NEXT:    v_xor_b32_e32 v0, v0, v1
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v3, v4
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
 ; GCN-NEXT:    v_mul_hi_u32 v3, v5, v3
 ; GCN-NEXT:    v_mul_lo_u32 v1, v3, v2
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, 1, v3
-; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v1, v5
-; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v1, v2
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
-; GCN-NEXT:    v_subrev_i32_e32 v4, vcc, v2, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; GCN-NEXT:    v_sub_i32_e32 v1, vcc, v5, v1
+; GCN-NEXT:    v_subrev_i32_e32 v5, vcc, v2, v1
+; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v2
+; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, 1, v3
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v2
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v1, v1, v0
 ; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v0, v1
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 25
 ; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm

 ; TONGA-NEXT:    v_xor_b32_e32 v5, v5, v0
 ; TONGA-NEXT:    v_xor_b32_e32 v0, v0, v1
 ; TONGA-NEXT:    v_mul_lo_u32 v4, v4, v3
 ; TONGA-NEXT:    v_mul_hi_u32 v4, v3, v4
 ; TONGA-NEXT:    v_add_u32_e32 v3, vcc, v3, v4
 ; TONGA-NEXT:    v_mul_hi_u32 v3, v5, v3
 ; TONGA-NEXT:    v_mul_lo_u32 v1, v3, v2
 ; TONGA-NEXT:    v_add_u32_e32 v4, vcc, 1, v3
-; TONGA-NEXT:    v_subrev_u32_e32 v1, vcc, v1, v5
-; TONGA-NEXT:    v_cmp_ge_u32_e64 s[0:1], v1, v2
-; TONGA-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
-; TONGA-NEXT:    v_subrev_u32_e32 v4, vcc, v2, v1
-; TONGA-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; TONGA-NEXT:    v_sub_u32_e32 v1, vcc, v5, v1
+; TONGA-NEXT:    v_subrev_u32_e32 v5, vcc, v2, v1
+; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v2
+; TONGA-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; TONGA-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
 ; TONGA-NEXT:    v_add_u32_e32 v4, vcc, 1, v3
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v2
 ; TONGA-NEXT:    v_cndmask_b32_e32 v1, v3, v4, vcc
 ; TONGA-NEXT:    v_xor_b32_e32 v1, v1, v0
 ; TONGA-NEXT:    v_subrev_u32_e32 v0, vcc, v0, v1
 ; TONGA-NEXT:    v_bfe_i32 v0, v0, 0, 25
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: v_sdiv_i25:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s6, -1
 ; GFX9-NEXT:    s_mov_b32 s10, s6
 ; GFX9-NEXT:    s_mov_b32 s11, s7
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_mov_b32 s8, s2
 ; GFX9-NEXT:    s_mov_b32 s9, s3
 ; GFX9-NEXT:    buffer_load_dwordx2 v[0:1], off, s[8:11], 0
 ; GFX9-NEXT:    s_mov_b32 s4, s0
 ; GFX9-NEXT:    s_mov_b32 s5, s1
 ; GFX9-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-NEXT:    v_bfe_i32 v2, v1, 0, 25
-; GFX9-NEXT:    v_bfe_i32 v1, v1, 24, 1
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v1
-; GFX9-NEXT:    v_xor_b32_e32 v2, v2, v1
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v3, v2
-; GFX9-NEXT:    v_sub_u32_e32 v4, 0, v2
-; GFX9-NEXT:    v_bfe_i32 v5, v0, 0, 25
-; GFX9-NEXT:    v_bfe_i32 v0, v0, 24, 1
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX9-NEXT:    v_add_u32_e32 v5, v5, v0
-; GFX9-NEXT:    v_xor_b32_e32 v5, v5, v0
-; GFX9-NEXT:    v_xor_b32_e32 v0, v0, v1
-; GFX9-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v4, v3
-; GFX9-NEXT:    v_mul_hi_u32 v4, v3, v4
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v4
-; GFX9-NEXT:    v_mul_hi_u32 v3, v5, v3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v1, 1, v3
-; GFX9-NEXT:    v_sub_u32_e32 v4, v5, v4
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v4, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v3, v1, vcc
-; GFX9-NEXT:    v_sub_u32_e32 v3, v4, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v4, v3, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v1
-; GFX9-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v1, v1, v0
-; GFX9-NEXT:    v_sub_u32_e32 v0, v1, v0
-; GFX9-NEXT:    v_bfe_i32 v0, v0, 0, 25
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX9-NEXT:    s_bfe_i32 s3, s2, 0x190000
+; GFX9-NEXT:    s_bfe_i32 s2, s2, 0x10018
+; GFX9-NEXT:    s_add_i32 s3, s3, s2
+; GFX9-NEXT:    s_xor_b32 s3, s3, s2
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s3
+; GFX9-NEXT:    v_readfirstlane_b32 s0, v0
+; GFX9-NEXT:    s_bfe_i32 s1, s0, 0x190000
+; GFX9-NEXT:    s_bfe_i32 s0, s0, 0x10018
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v1
+; GFX9-NEXT:    s_add_i32 s1, s1, s0
+; GFX9-NEXT:    s_xor_b32 s2, s0, s2
+; GFX9-NEXT:    s_xor_b32 s0, s1, s0
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_sub_i32 s1, 0, s3
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX9-NEXT:    s_mul_i32 s1, s1, s8
+; GFX9-NEXT:    s_mul_hi_u32 s1, s8, s1
+; GFX9-NEXT:    s_add_i32 s8, s8, s1
+; GFX9-NEXT:    s_mul_hi_u32 s1, s0, s8
+; GFX9-NEXT:    s_mul_i32 s8, s1, s3
+; GFX9-NEXT:    s_sub_i32 s0, s0, s8
+; GFX9-NEXT:    s_add_i32 s9, s1, 1
+; GFX9-NEXT:    s_sub_i32 s8, s0, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX9-NEXT:    s_cselect_b32 s1, s9, s1
+; GFX9-NEXT:    s_cselect_b32 s0, s8, s0
+; GFX9-NEXT:    s_add_i32 s8, s1, 1
+; GFX9-NEXT:    s_cmp_ge_u32 s0, s3
+; GFX9-NEXT:    s_cselect_b32 s0, s8, s1
+; GFX9-NEXT:    s_xor_b32 s0, s0, s2
+; GFX9-NEXT:    s_sub_i32 s0, s0, s2
+; GFX9-NEXT:    s_bfe_i32 s0, s0, 0x190000
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
 ; GFX9-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX9-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_sdiv_i25:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 1, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 37, @12, KC0[CB0:0-32], KC1[]

 ; TONGA-NEXT:    v_lshrrev_b32_e32 v4, 31, v0
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v0, 12, v0
 ; TONGA-NEXT:    v_lshrrev_b32_e32 v5, 31, v1
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v1, 12, v1
 ; TONGA-NEXT:    v_lshrrev_b32_e32 v6, 31, v2
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v2, 12, v2
 ; TONGA-NEXT:    v_lshrrev_b32_e32 v7, 31, v3
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v3, 12, v3
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v4, v0
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v4
 ; TONGA-NEXT:    v_add_u32_e32 v1, vcc, v1, v5
-; TONGA-NEXT:    v_add_u32_e32 v2, vcc, v2, v6
-; TONGA-NEXT:    v_add_u32_e32 v3, vcc, v7, v3
+; TONGA-NEXT:    v_add_u32_e32 v2, vcc, v6, v2
+; TONGA-NEXT:    v_add_u32_e32 v3, vcc, v3, v7
 ; TONGA-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
 ;
 ; GFX9-LABEL: scalarize_mulhs_4xi32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; GFX9-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX9-NEXT:    s_mov_b32 s6, -1

llvm/test/CodeGen/AMDGPU/sdiv64.ll

 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s4, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s5, v0
 ; GCN-NEXT:    s_xor_b64 s[2:3], s[2:3], s[12:13]
 ; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s4, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_subrev_i32_e64 v5, s[0:1], s10, v3
 ; GCN-NEXT:    v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s11, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s10, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], s11, 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_add_i32_e64 v5, s[0:1], 1, 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_add_i32_e64 v7, s[0:1], 2, 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_cndmask_b32_e64 v4, v5, v7, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v5, v6, v8, s[0:1]
 ; GCN-NEXT:    v_mov_b32_e32 v6, s3
 ; GCN-NEXT:    v_subb_u32_e32 v2, vcc, v6, v2, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s11, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s10, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s11, 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 v0, v0, v2, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GCN-NEXT:    s_xor_b64 s[0:1], s[12:13], s[8:9]
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v5, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v0, s0, v0
 ; GCN-NEXT:    v_xor_b32_e32 v1, s1, v1
 ; GCN-NEXT:    v_mov_b32_e32 v2, s1
 ; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0xd
+; GCN-IR-NEXT:    s_mov_b32 s15, 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i32 s0, s7, 31
 ; GCN-IR-NEXT:    s_mov_b32 s1, s0
 ; GCN-IR-NEXT:    s_ashr_i32 s2, s9, 31
 ; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[0:1], s[6:7]
 ; GCN-IR-NEXT:    s_mov_b32 s3, s2
 ; GCN-IR-NEXT:    s_sub_u32 s12, s6, s0
 ; GCN-IR-NEXT:    s_subb_u32 s13, s7, s0
 ; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[2:3], s[8:9]
 ; GCN-IR-NEXT:    s_sub_u32 s6, s6, s2
 ; GCN-IR-NEXT:    s_subb_u32 s7, s7, s2
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[12:13], 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[6:7], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[12:13], 0
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[8:9]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s6
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s7
+; GCN-IR-NEXT:    s_min_u32 s14, s8, s9
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s12
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s13
+; GCN-IR-NEXT:    s_min_u32 s18, s8, s9
+; GCN-IR-NEXT:    s_sub_u32 s16, s14, s18
+; GCN-IR-NEXT:    s_subb_u32 s17, 0, 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[20:21], s[16:17], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[16:17], 63
+; GCN-IR-NEXT:    s_or_b64 s[20:21], s[10:11], s[20:21]
+; GCN-IR-NEXT:    s_and_b64 s[10:11], s[20:21], exec
+; GCN-IR-NEXT:    s_cselect_b32 s11, 0, s13
+; GCN-IR-NEXT:    s_cselect_b32 s10, 0, s12
+; GCN-IR-NEXT:    s_or_b64 s[20:21], s[20:21], s[22:23]
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[10:11], s[14:15]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s6
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s7
-; GCN-IR-NEXT:    s_min_u32 s14, s10, s11
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s12
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s13
-; GCN-IR-NEXT:    s_min_u32 s18, s10, s11
-; GCN-IR-NEXT:    s_sub_u32 s10, s14, s18
-; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[10:11], 63
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[20:21]
-; GCN-IR-NEXT:    s_or_b64 s[20:21], s[16:17], s[22:23]
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[20:21]
-; GCN-IR-NEXT:    s_mov_b32 s15, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s16, s10, 1
-; GCN-IR-NEXT:    s_addc_u32 s17, s11, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[16:17], 0
-; GCN-IR-NEXT:    s_sub_i32 s10, 63, s10
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[20:21]
-; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[12:13], s10
+; GCN-IR-NEXT:    s_add_u32 s20, s16, 1
+; GCN-IR-NEXT:    s_addc_u32 s21, s17, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[20:21], 0
+; GCN-IR-NEXT:    s_sub_i32 s16, 63, s16
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[10:11]
+; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[12:13], s16
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[16:17], s[12:13], s16
+; GCN-IR-NEXT:    s_lshr_b64 s[16:17], s[12:13], s20
 ; GCN-IR-NEXT:    s_add_u32 s19, s6, -1
 ; GCN-IR-NEXT:    s_addc_u32 s20, s7, -1
 ; GCN-IR-NEXT:    s_not_b64 s[8:9], s[14:15]
 ; GCN-IR-NEXT:    s_add_u32 s12, s8, s18
 ; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], 0
 ; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:  .LBB0_3: ; %udiv-do-while

 ; GCN-IR-NEXT:    s_add_u32 s12, s12, 1
 ; GCN-IR-NEXT:    s_addc_u32 s13, s13, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], s[8:9]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[22:23]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_3
 ; GCN-IR-NEXT:  .LBB0_4: ; %Flow6
 ; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[10:11], 1
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[8:9], s[6:7]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-IR-NEXT:    s_branch .LBB0_6
-; GCN-IR-NEXT:  .LBB0_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s13
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[16:17]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s12
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[16:17]
-; GCN-IR-NEXT:  .LBB0_6: ; %udiv-end
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[8:9], s[6:7]
+; GCN-IR-NEXT:  .LBB0_5: ; %udiv-end
 ; GCN-IR-NEXT:    s_xor_b64 s[0:1], s[2:3], s[0:1]
-; GCN-IR-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GCN-IR-NEXT:    v_xor_b32_e32 v1, s1, v1
-; GCN-IR-NEXT:    v_mov_b32_e32 v2, s1
-; GCN-IR-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
+; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[10:11], s[0:1]
+; GCN-IR-NEXT:    s_sub_u32 s0, s2, s0
+; GCN-IR-NEXT:    s_subb_u32 s1, s3, s1
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = sdiv i64 %x, %y
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define i64 @v_test_sdiv(i64 %x, i64 %y) {

 ; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    s_mov_b32 s1, s5
 ; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-NEXT:    s_or_b32 s4, s4, 1
+; GCN-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv24_64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_mov_b32 s0, s4
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    s_mov_b32 s1, s5
 ; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-IR-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s4, s4, 1
+; GCN-IR-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 40
   %2 = ashr i64 %y, 40
   %result = sdiv i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out

 ; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s3
 ; GCN-NEXT:    s_mov_b32 s4, s0
 ; GCN-NEXT:    s_xor_b32 s0, s3, s8
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-NEXT:    s_ashr_i32 s0, s0, 30
-; GCN-NEXT:    s_or_b32 s0, s0, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s0
+; GCN-NEXT:    s_mov_b32 s5, s1
+; GCN-NEXT:    s_or_b32 s2, s0, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s2, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s0, v2
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv32_64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dword s8, s[0:1], 0xe
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s3
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    s_xor_b32 s0, s3, s8
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-IR-NEXT:    s_ashr_i32 s0, s0, 30
-; GCN-IR-NEXT:    s_or_b32 s0, s0, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s0
+; GCN-IR-NEXT:    s_mov_b32 s5, s1
+; GCN-IR-NEXT:    s_or_b32 s2, s0, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-IR-NEXT:    s_cselect_b32 s0, s2, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s0, v2
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 32
   %2 = ashr i64 %y, 32
   %result = sdiv i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void

 ; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 33
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    s_mov_b32 s1, s5
 ; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 33
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-NEXT:    s_or_b32 s4, s4, 1
+; GCN-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 31
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv31_64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_mov_b32 s0, s4
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 33
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    s_mov_b32 s1, s5
 ; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 33
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-IR-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s4, s4, 1
+; GCN-IR-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 31
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 33
   %2 = ashr i64 %y, 33
   %result = sdiv i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out

 ; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 41
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    s_mov_b32 s1, s5
 ; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 41
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-NEXT:    s_or_b32 s4, s4, 1
+; GCN-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 23
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv23_64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_mov_b32 s0, s4
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 41
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    s_mov_b32 s1, s5
 ; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 41
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-IR-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s4, s4, 1
+; GCN-IR-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 23
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 41
   %2 = ashr i64 %y, 41
   %result = sdiv i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out

 ; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 39
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    s_mov_b32 s1, s5
 ; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 39
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-NEXT:    s_or_b32 s4, s4, 1
+; GCN-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 25
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv25_64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_mov_b32 s0, s4
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 39
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    s_mov_b32 s1, s5
 ; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 39
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-IR-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s4, s4, 1
+; GCN-IR-NEXT:    s_or_b32 s6, s4, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 25
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 39
   %2 = ashr i64 %y, 39
   %result = sdiv i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out

 ; GCN-NEXT:    s_mov_b32 s2, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_ashr_i64 s[8:9], s[8:9], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    s_ashr_i64 s[4:5], s[4:5], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
+; GCN-NEXT:    s_ashr_i64 s[6:7], s[6:7], 40
 ; GCN-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-NEXT:    s_or_b32 s4, s4, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s4
+; GCN-NEXT:    s_ashr_i64 s[10:11], s[10:11], 40
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    s_ashr_i64 s[10:11], s[10:11], 40
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-NEXT:    s_or_b32 s7, s4, 1
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, s7, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v2, s10
-; GCN-NEXT:    s_ashr_i64 s[6:7], s[6:7], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v3, s6
 ; GCN-NEXT:    s_xor_b32 s4, s6, s10
-; GCN-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; GCN-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-NEXT:    s_or_b32 s4, s4, 1
-; GCN-NEXT:    v_mov_b32_e32 v5, s4
+; GCN-NEXT:    v_rcp_iflag_f32_e32 v4, v2
+; GCN-NEXT:    s_or_b32 s6, s4, 1
+; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
+; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    v_mul_f32_e32 v4, v3, v4
 ; GCN-NEXT:    v_trunc_f32_e32 v4, v4
 ; GCN-NEXT:    v_mad_f32 v3, -v4, v2, v3
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
-; GCN-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v3|, |v2|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, s4, v4
 ; GCN-NEXT:    v_bfe_i32 v2, v2, 0, 24
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    v_ashrrev_i32_e32 v3, 31, v2
 ; GCN-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv24_v2i64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
 ; GCN-IR-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[8:9], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[4:5], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    s_xor_b32 s4, s4, s8
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
+; GCN-IR-NEXT:    s_ashr_i64 s[6:7], s[6:7], 40
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s4, s4, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s4
+; GCN-IR-NEXT:    s_ashr_i64 s[10:11], s[10:11], 40
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    s_ashr_i64 s[10:11], s[10:11], 40
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GCN-IR-NEXT:    s_or_b32 s7, s4, 1
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s4, s7, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s4, v2
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v2, s10
-; GCN-IR-NEXT:    s_ashr_i64 s[6:7], s[6:7], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v3, s6
 ; GCN-IR-NEXT:    s_xor_b32 s4, s6, s10
-; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s4, s4, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v5, s4
+; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v4, v2
+; GCN-IR-NEXT:    s_or_b32 s6, s4, 1
+; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
+; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    v_mul_f32_e32 v4, v3, v4
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v4, v4
 ; GCN-IR-NEXT:    v_mad_f32 v3, -v4, v2, v3
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v4, v4
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v3|, |v2|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s4, s6, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, s4, v4
 ; GCN-IR-NEXT:    v_bfe_i32 v2, v2, 0, 24
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v3, 31, v2
 ; GCN-IR-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr <2 x i64> %x, <i64 40, i64 40>
   %2 = ashr <2 x i64> %y, <i64 40, i64 40>
   %result = sdiv <2 x i64> %1, %2
   store <2 x i64> %result, <2 x i64> addrspace(1)* %out
   ret void

 ; GCN-NEXT:    v_ashrrev_i32_e32 v0, 30, v0
 ; GCN-NEXT:    v_or_b32_e32 v0, 1, v0
 ; GCN-NEXT:    v_mul_f32_e32 v2, v3, v4
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v3, -v2, v1, v3
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
 ; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v1|
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GCN-NEXT:    buffer_store_short v1, off, s[0:3], 0 offset:4
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv24_48:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
+; GCN-IR-NEXT:    s_mov_b32 s15, 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_sext_i32_i16 s7, s7
 ; GCN-IR-NEXT:    s_ashr_i32 s0, s7, 31
 ; GCN-IR-NEXT:    s_sext_i32_i16 s3, s3
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[6:7], 24
 ; GCN-IR-NEXT:    s_mov_b32 s1, s0
 ; GCN-IR-NEXT:    s_ashr_i64 s[10:11], s[2:3], 24
 ; GCN-IR-NEXT:    s_ashr_i32 s2, s3, 31
 ; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[0:1], s[8:9]
 ; GCN-IR-NEXT:    s_mov_b32 s3, s2
 ; GCN-IR-NEXT:    s_sub_u32 s12, s6, s0
 ; GCN-IR-NEXT:    s_subb_u32 s13, s7, s0
 ; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[2:3], s[10:11]
 ; GCN-IR-NEXT:    s_sub_u32 s6, s6, s2
 ; GCN-IR-NEXT:    s_subb_u32 s7, s7, s2
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[6:7], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[12:13], 0
-; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[10:11], s[14:15]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s6
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s7
-; GCN-IR-NEXT:    s_min_u32 s14, s10, s11
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s12
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s13
-; GCN-IR-NEXT:    s_min_u32 s18, s10, s11
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[6:7], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[12:13], 0
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[8:9], s[10:11]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s6
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s7
+; GCN-IR-NEXT:    s_min_u32 s14, s8, s9
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s12
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s13
+; GCN-IR-NEXT:    s_min_u32 s18, s8, s9
 ; GCN-IR-NEXT:    s_sub_u32 s10, s14, s18
 ; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[20:21], s[10:11], 63
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[10:11], 63
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[20:21]
-; GCN-IR-NEXT:    s_or_b64 s[20:21], s[16:17], s[22:23]
+; GCN-IR-NEXT:    s_or_b64 s[20:21], s[16:17], s[20:21]
+; GCN-IR-NEXT:    s_and_b64 s[16:17], s[20:21], exec
+; GCN-IR-NEXT:    s_cselect_b32 s17, 0, s13
+; GCN-IR-NEXT:    s_cselect_b32 s16, 0, s12
+; GCN-IR-NEXT:    s_or_b64 s[20:21], s[20:21], s[22:23]
+; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[20:21]
-; GCN-IR-NEXT:    s_mov_b32 s15, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s16, s10, 1
 ; GCN-IR-NEXT:    s_addc_u32 s17, s11, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[16:17], 0
 ; GCN-IR-NEXT:    s_sub_i32 s10, 63, s10
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[20:21]
 ; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[12:13], s10

 ; GCN-IR-NEXT:    s_add_u32 s12, s12, 1
 ; GCN-IR-NEXT:    s_addc_u32 s13, s13, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[22:23], s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], s[8:9]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[22:23]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_3
 ; GCN-IR-NEXT:  .LBB9_4: ; %Flow3
 ; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[10:11], 1
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[8:9], s[6:7]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-IR-NEXT:    s_branch .LBB9_6
-; GCN-IR-NEXT:  .LBB9_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s13
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[16:17]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s12
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[16:17]
-; GCN-IR-NEXT:  .LBB9_6: ; %udiv-end
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[8:9], s[6:7]
+; GCN-IR-NEXT:  .LBB9_5: ; %udiv-end
 ; GCN-IR-NEXT:    s_xor_b64 s[0:1], s[2:3], s[0:1]
-; GCN-IR-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GCN-IR-NEXT:    v_xor_b32_e32 v1, s1, v1
-; GCN-IR-NEXT:    v_mov_b32_e32 v2, s1
-; GCN-IR-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
+; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[16:17], s[0:1]
+; GCN-IR-NEXT:    s_sub_u32 s0, s2, s0
+; GCN-IR-NEXT:    s_subb_u32 s1, s3, s1
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    buffer_store_short v1, off, s[4:7], 0 offset:4
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s1
+; GCN-IR-NEXT:    buffer_store_short v0, off, s[4:7], 0 offset:4
+; GCN-IR-NEXT:    s_waitcnt expcnt(0)
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-IR-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i48 %x, 24
   %2 = ashr i48 %y, 24
   %result = sdiv i48 %1, %2
   store i48 %result, i48 addrspace(1)* %out
   ret void
 }

 ; GCN-NEXT:    v_subrev_i32_e64 v4, s[0:1], s2, v2
 ; GCN-NEXT:    v_subbrev_u32_e64 v3, s[0:1], 0, v3, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], s3, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v3, v5, v4, s[0:1]
-; GCN-NEXT:    v_add_i32_e64 v4, s[0:1], 2, v0
+; GCN-NEXT:    v_add_i32_e64 v4, s[0:1], 1, v0
 ; GCN-NEXT:    v_addc_u32_e64 v5, s[0:1], 0, 0, s[0:1]
-; GCN-NEXT:    v_add_i32_e64 v6, s[0:1], 1, v0
+; GCN-NEXT:    v_add_i32_e64 v6, s[0:1], 2, v0
 ; GCN-NEXT:    v_addc_u32_e64 v7, s[0:1], 0, 0, s[0:1]
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v3
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v7, v5, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v6, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v4, v5, v7, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s3, v1
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v5, v2, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v6, v4, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
 ; GCN-NEXT:    v_xor_b32_e32 v0, s8, v0
 ; GCN-NEXT:    v_xor_b32_e32 v1, s8, v1
 ; GCN-NEXT:    v_mov_b32_e32 v2, s8
 ; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, s8, v0
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv_k_num_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i32 s4, s3, 31
 ; GCN-IR-NEXT:    s_mov_b32 s5, s4
 ; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[4:5], s[2:3]
 ; GCN-IR-NEXT:    s_sub_u32 s2, s2, s4
 ; GCN-IR-NEXT:    s_subb_u32 s3, s3, s4
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
-; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
-; GCN-IR-NEXT:    s_min_u32 s10, s8, s9
-; GCN-IR-NEXT:    s_add_u32 s8, s10, 0xffffffc5
-; GCN-IR-NEXT:    s_addc_u32 s9, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[8:9], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[8:9], 63
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[6:7], s[12:13]
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[12:13], s[14:15]
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[6:7]
-; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
+; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s2
+; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s3
+; GCN-IR-NEXT:    s_min_u32 s10, s10, s11
+; GCN-IR-NEXT:    s_add_u32 s12, s10, 0xffffffc5
+; GCN-IR-NEXT:    s_addc_u32 s13, 0, -1
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[14:15], s[12:13], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[12:13], 63
+; GCN-IR-NEXT:    s_or_b64 s[14:15], s[8:9], s[14:15]
+; GCN-IR-NEXT:    s_and_b64 s[8:9], s[14:15], exec
+; GCN-IR-NEXT:    s_cselect_b32 s8, 0, 24
+; GCN-IR-NEXT:    s_or_b64 s[14:15], s[14:15], s[16:17]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[14:15]
+; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
-; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[12:13], 0
-; GCN-IR-NEXT:    s_sub_i32 s8, 63, s8
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[14:15]
-; GCN-IR-NEXT:    s_lshl_b64 s[8:9], 24, s8
+; GCN-IR-NEXT:    s_add_u32 s14, s12, 1
+; GCN-IR-NEXT:    s_addc_u32 s15, s13, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[14:15], 0
+; GCN-IR-NEXT:    s_sub_i32 s11, 63, s12
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[8:9]
+; GCN-IR-NEXT:    s_lshl_b64 s[8:9], 24, s11
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[12:13], 24, s12
+; GCN-IR-NEXT:    s_lshr_b64 s[12:13], 24, s14
 ; GCN-IR-NEXT:    s_add_u32 s16, s2, -1
 ; GCN-IR-NEXT:    s_addc_u32 s17, s3, -1
 ; GCN-IR-NEXT:    s_sub_u32 s10, 58, s10
 ; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], 0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:  .LBB10_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1

 ; GCN-IR-NEXT:    s_add_u32 s10, s10, 1
 ; GCN-IR-NEXT:    s_addc_u32 s11, s11, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[10:11], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_3
 ; GCN-IR-NEXT:  .LBB10_4: ; %Flow5
 ; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[8:9], 1
-; GCN-IR-NEXT:    s_or_b64 s[2:3], s[6:7], s[2:3]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-IR-NEXT:    s_branch .LBB10_6
-; GCN-IR-NEXT:  .LBB10_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, 0
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, 24, 0, s[12:13]
-; GCN-IR-NEXT:  .LBB10_6: ; %udiv-end
-; GCN-IR-NEXT:    v_xor_b32_e32 v0, s4, v0
-; GCN-IR-NEXT:    v_xor_b32_e32 v1, s5, v1
-; GCN-IR-NEXT:    v_mov_b32_e32 v2, s5
-; GCN-IR-NEXT:    v_subrev_i32_e32 v0, vcc, s4, v0
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[2:3]
+; GCN-IR-NEXT:  .LBB10_5: ; %udiv-end
+; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[8:9], s[4:5]
+; GCN-IR-NEXT:    s_sub_u32 s4, s6, s4
+; GCN-IR-NEXT:    s_subb_u32 s5, s7, s5
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = sdiv i64 24, %x
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define i64 @v_test_sdiv_k_num_i64(i64 %x) {

 ; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s2
 ; GCN-NEXT:    s_mov_b32 s3, 0x41c00000
 ; GCN-NEXT:    s_mov_b32 s4, s0
 ; GCN-NEXT:    s_ashr_i32 s0, s2, 30
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v1, v0
-; GCN-NEXT:    s_or_b32 s0, s0, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s0
 ; GCN-NEXT:    s_mov_b32 s5, s1
+; GCN-NEXT:    s_or_b32 s2, s0, 1
 ; GCN-NEXT:    v_mul_f32_e32 v1, 0x41c00000, v1
 ; GCN-NEXT:    v_trunc_f32_e32 v1, v1
 ; GCN-NEXT:    v_mad_f32 v2, -v1, v0, s3
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v2|, |v0|
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s2, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s0, v1
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv24_k_num_i64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s2
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0x41c00000
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    s_ashr_i32 s0, s2, 30
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v1, v0
-; GCN-IR-NEXT:    s_or_b32 s0, s0, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s0
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
+; GCN-IR-NEXT:    s_or_b32 s2, s0, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v1, 0x41c00000, v1
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v1, v1
 ; GCN-IR-NEXT:    v_mad_f32 v2, -v1, v0, s3
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v2|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-IR-NEXT:    s_cselect_b32 s0, s2, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s0, v1
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %x.shr = ashr i64 %x, 40
   %result = sdiv i64 24, %x.shr
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_test_sdiv24_k_den_i64(i64 addrspace(1)* %out, i64 %x) {
 ; GCN-LABEL: s_test_sdiv24_k_den_i64:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-NEXT:    s_mov_b32 s8, 0x46b6fe00
 ; GCN-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s2
 ; GCN-NEXT:    s_mov_b32 s4, s0
 ; GCN-NEXT:    s_ashr_i32 s0, s2, 30
-; GCN-NEXT:    s_or_b32 s0, s0, 1
+; GCN-NEXT:    s_mov_b32 s5, s1
 ; GCN-NEXT:    v_mul_f32_e32 v1, 0x38331158, v0
 ; GCN-NEXT:    v_trunc_f32_e32 v1, v1
 ; GCN-NEXT:    v_mad_f32 v0, -v1, s8, v0
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GCN-NEXT:    v_mov_b32_e32 v2, s0
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, s8
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v2, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GCN-NEXT:    s_or_b32 s2, s0, 1
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v0|, s8
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s2, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s0, v1
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-NEXT:    s_mov_b32 s5, s1
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_sdiv24_k_den_i64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_mov_b32 s8, 0x46b6fe00
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s2
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    s_ashr_i32 s0, s2, 30
-; GCN-IR-NEXT:    s_or_b32 s0, s0, 1
+; GCN-IR-NEXT:    s_mov_b32 s5, s1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v1, 0x38331158, v0
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v1, v1
 ; GCN-IR-NEXT:    v_mad_f32 v0, -v1, s8, v0
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GCN-IR-NEXT:    v_mov_b32_e32 v2, s0
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, s8
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v2, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GCN-IR-NEXT:    s_or_b32 s2, s0, 1
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v0|, s8
+; GCN-IR-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-IR-NEXT:    s_cselect_b32 s0, s2, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s0, v1
 ; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %x.shr = ashr i64 %x, 40
   %result = sdiv i64 %x.shr, 23423
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }

llvm/test/CodeGen/AMDGPU/select-constant-cttz.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
 
 declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
 declare i32 @llvm.amdgcn.sffbh.i32(i32) nounwind readnone speculatable
 define amdgpu_kernel void @select_constant_cttz(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
 ; GCN-LABEL: select_constant_cttz:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_load_dword s2, s[2:3], 0x0
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-NEXT:    s_mov_b32 s6, -1
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    s_mov_b32 s5, s1
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_lshr_b32 s0, 1, s2
-; GCN-NEXT:    s_ff1_i32_b32 s0, s0
+; GCN-NEXT:    s_lshr_b32 s4, 1, s2
 ; GCN-NEXT:    s_cmp_lg_u32 s2, 0
-; GCN-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-NEXT:    s_cselect_b64 s[2:3], -1, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, -1, s[2:3]
-; GCN-NEXT:    v_ffbh_i32_e32 v1, v0
-; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v0
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, 31, v1
-; GCN-NEXT:    s_or_b64 s[0:1], s[2:3], s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, -1, s[0:1]
-; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GCN-NEXT:    s_ff1_i32_b32 s2, s4
+; GCN-NEXT:    s_cselect_b64 s[4:5], -1, 0
+; GCN-NEXT:    s_and_b64 s[6:7], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s2, -1, s2
+; GCN-NEXT:    s_flbit_i32 s6, s2
+; GCN-NEXT:    s_sub_i32 s8, 31, s6
+; GCN-NEXT:    s_cmp_eq_u32 s2, 0
+; GCN-NEXT:    s_cselect_b64 s[6:7], -1, 0
+; GCN-NEXT:    s_or_b64 s[4:5], s[4:5], s[6:7]
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s4, -1, s8
+; GCN-NEXT:    s_mov_b32 s2, -1
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
   %v    = load i32, i32 addrspace(1)* %arrayidx, align 4
   %sr   = lshr i32 1, %v
   %cmp  = icmp ne i32 %v, 0
   %cttz = call i32 @llvm.cttz.i32(i32 %sr, i1 true), !range !0
   %sel  = select i1 %cmp, i32 -1, i32 %cttz
   %ffbh = call i32 @llvm.amdgcn.sffbh.i32(i32 %sel)
   %sub  = sub i32 31, %ffbh
   %cmp2 = icmp eq i32 %sel, 0
   %or   = or i1 %cmp, %cmp2
   %sel2 = select i1 %or, i32 -1, i32 %sub
   store i32 %sel2, i32 addrspace(1)* %out
   ret void
 }
 
 !0 = !{i32 0, i32 33}

llvm/test/CodeGen/AMDGPU/select-opt.ll

 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
 
 ; Make sure to test with f32 and i32 compares. If we have to use float
 ; compares, we always have multiple condition registers. If we can do
 ; scalar compares, we don't want to use multiple condition registers.
 
 ; GCN-LABEL: {{^}}opt_select_i32_and_cmp_i32:
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP1:s\[[0-9]+:[0-9]+\]]], -1, 0
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP2:s\[[0-9]+:[0-9]+\]]], -1, 0
-; GCN: s_and_b64 vcc, [[CMP1]], [[CMP2]]
-; GCN: v_cndmask_b32_e32 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN-NOT: [[RESULT]]
-; GCN: buffer_store_dword [[RESULT]]
+; GCN: s_and_b64 [[AND1:s\[[0-9]+:[0-9]+\]]], [[CMP1]], [[CMP2]]
+; GCN: s_and_b64 [[AND2:s\[[0-9]+:[0-9]+\]]], [[AND1]], exec
+; GCN: s_cselect_b32 [[RESULT:s[0-9]+]]
+; GCN: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[RESULT]]
+; GCN: buffer_store_dword [[VRESULT]]
 define amdgpu_kernel void @opt_select_i32_and_cmp_i32(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %x, i32 %y) #0 {
   %icmp0 = icmp ne i32 %a, %b
   %icmp1 = icmp ne i32 %a, %c
   %and = and i1 %icmp0, %icmp1
   %select = select i1 %and, i32 %x, i32 %y
   store i32 %select, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i32_and_cmp_f32:
 ; GCN-DAG: v_cmp_lg_f32_e32 vcc
 ; GCN-DAG: v_cmp_lg_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
-; GCN: s_and_b64 vcc, vcc, [[CMP1]]
-; GCN: v_cndmask_b32_e32 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN-NOT: [[RESULT]]
-; GCN: buffer_store_dword [[RESULT]]
+; GCN: s_and_b64 [[CMP1]], vcc, [[CMP1]]
+; GCN: s_and_b64 [[AND:s\[[0-9]+:[0-9]+\]]], [[CMP1]], exec
+; GCN: s_cselect_b32 [[RESULT:s[0-9]+]]
+; GCN: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[RESULT]]
+; GCN: buffer_store_dword [[VRESULT]]
 define amdgpu_kernel void @opt_select_i32_and_cmp_f32(i32 addrspace(1)* %out, float %a, float %b, float %c, i32 %x, i32 %y) #0 {
   %fcmp0 = fcmp one float %a, %b
   %fcmp1 = fcmp one float %a, %c
   %and = and i1 %fcmp0, %fcmp1
   %select = select i1 %and, i32 %x, i32 %y
   store i32 %select, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i64_and_cmp_i32:
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP1:s\[[0-9]+:[0-9]+\]]], -1, 0
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP2:s\[[0-9]+:[0-9]+\]]], -1, 0
-; GCN: s_and_b64 vcc, [[CMP1]], [[CMP2]]
-; GCN: v_cndmask_b32_e32 v[[RESULT1:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: v_cndmask_b32_e32 v[[RESULT0:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: buffer_store_dwordx2 v[[[RESULT0]]:[[RESULT1]]]
+; GCN: s_and_b64 [[AND1:s\[[0-9]+:[0-9]+\]]], [[CMP1]], [[CMP2]]
+; GCN: s_and_b64 [[AND2:s\[[0-9]+:[0-9]+\]]], [[AND1]], exec
+; GCN-DAG: s_cselect_b32 [[RESULT0:s[0-9]+]]
+; GCN-DAG: s_cselect_b32 [[RESULT1:s[0-9]+]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT1:[0-9]+]], [[RESULT0]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT0:[0-9]+]], [[RESULT1]]
+; GCN: buffer_store_dwordx2 v[[[VRESULT0]]:[[VRESULT1]]]
 define amdgpu_kernel void @opt_select_i64_and_cmp_i32(i64 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i64 %x, i64 %y) #0 {
   %icmp0 = icmp ne i32 %a, %b
   %icmp1 = icmp ne i32 %a, %c
   %and = and i1 %icmp0, %icmp1
   %select = select i1 %and, i64 %x, i64 %y
   store i64 %select, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i64_and_cmp_f32:
 ; GCN-DAG: v_cmp_lg_f32_e32 vcc,
 ; GCN-DAG: v_cmp_lg_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
-; GCN: s_and_b64 vcc, vcc, [[CMP1]]
-; GCN: v_cndmask_b32_e32 v[[RESULT1:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: v_cndmask_b32_e32 v[[RESULT0:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: buffer_store_dwordx2 v[[[RESULT0]]:[[RESULT1]]]
+; GCN: s_and_b64 [[AND1:s\[[0-9]+:[0-9]+\]]], vcc, [[CMP1]]
+; GCN: s_and_b64 [[AND2:s\[[0-9]+:[0-9]+\]]], [[AND1]], exec
+; GCN-DAG: s_cselect_b32 [[RESULT0:s[0-9]+]]
+; GCN-DAG: s_cselect_b32 [[RESULT1:s[0-9]+]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT1:[0-9]+]], [[RESULT0]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT0:[0-9]+]], [[RESULT1]]
+; GCN: buffer_store_dwordx2 v[[[VRESULT0]]:[[VRESULT1]]]
 define amdgpu_kernel void @opt_select_i64_and_cmp_f32(i64 addrspace(1)* %out, float %a, float %b, float %c, i64 %x, i64 %y) #0 {
   %fcmp0 = fcmp one float %a, %b
   %fcmp1 = fcmp one float %a, %c
   %and = and i1 %fcmp0, %fcmp1
   %select = select i1 %and, i64 %x, i64 %y
   store i64 %select, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i32_or_cmp_i32:
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP1:s\[[0-9]+:[0-9]+\]]], -1, 0
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP2:s\[[0-9]+:[0-9]+\]]], -1, 0
-; GCN: s_or_b64 vcc, [[CMP1]], [[CMP2]]
-; GCN: v_cndmask_b32_e32 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN-NOT: [[RESULT]]
-; GCN: buffer_store_dword [[RESULT]]
+; GCN: s_or_b64 [[OR:s\[[0-9]+:[0-9]+\]]], [[CMP1]], [[CMP2]]
+; GCN: s_and_b64 [[AND:s\[[0-9]+:[0-9]+\]]], [[OR]], exec
+; GCN-DAG: s_cselect_b32 [[RESULT:s[0-9]+]]
+; GCN-DAG: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[RESULT]]
+; GCN: buffer_store_dword [[VRESULT]]
 ; GCN: s_endpgm
 define amdgpu_kernel void @opt_select_i32_or_cmp_i32(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %x, i32 %y) #0 {
   %icmp0 = icmp ne i32 %a, %b
   %icmp1 = icmp ne i32 %a, %c
   %or = or i1 %icmp0, %icmp1
   %select = select i1 %or, i32 %x, i32 %y
   store i32 %select, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i32_or_cmp_f32:
 ; GCN-DAG: v_cmp_lg_f32_e32 vcc
 ; GCN-DAG: v_cmp_lg_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
-; GCN: s_or_b64 vcc, vcc, [[CMP1]]
-; GCN: v_cndmask_b32_e32 [[RESULT:v[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN-NOT: [[RESULT]]
-; GCN: buffer_store_dword [[RESULT]]
+; GCN: s_or_b64 [[OR:s\[[0-9]+:[0-9]+\]]], vcc, [[CMP1]]
+; GCN: s_and_b64 [[AND:s\[[0-9]+:[0-9]+\]]], [[OR]], exec
+; GCN-DAG: s_cselect_b32 [[RESULT:s[0-9]+]]
+; GCN-DAG: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[RESULT]]
+; GCN: buffer_store_dword [[VRESULT]]
 define amdgpu_kernel void @opt_select_i32_or_cmp_f32(i32 addrspace(1)* %out, float %a, float %b, float %c, i32 %x, i32 %y) #0 {
   %fcmp0 = fcmp one float %a, %b
   %fcmp1 = fcmp one float %a, %c
   %or = or i1 %fcmp0, %fcmp1
   %select = select i1 %or, i32 %x, i32 %y
   store i32 %select, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i64_or_cmp_i32:
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP1:s\[[0-9]+:[0-9]+\]]], -1, 0
 ; GCN-DAG: s_cmp_lg_u32
 ; GCN: s_cselect_b64 [[CMP2:s\[[0-9]+:[0-9]+\]]], -1, 0
-; GCN: s_or_b64 vcc, [[CMP1]], [[CMP2]]
-; GCN: v_cndmask_b32_e32 v[[RESULT1:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: v_cndmask_b32_e32 v[[RESULT0:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: buffer_store_dwordx2 v[[[RESULT0]]:[[RESULT1]]]
+; GCN: s_or_b64 [[OR:s\[[0-9]+:[0-9]+\]]], [[CMP1]], [[CMP2]]
+; GCN: s_and_b64 [[AND:s\[[0-9]+:[0-9]+\]]], [[OR]], exec
+; GCN-DAG: s_cselect_b32 [[RESULT0:s[0-9]+]]
+; GCN-DAG: s_cselect_b32 [[RESULT1:s[0-9]+]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT1:[0-9]+]], [[RESULT0]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT0:[0-9]+]], [[RESULT1]]
+; GCN: buffer_store_dwordx2 v[[[VRESULT0]]:[[VRESULT1]]]
 define amdgpu_kernel void @opt_select_i64_or_cmp_i32(i64 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i64 %x, i64 %y) #0 {
   %icmp0 = icmp ne i32 %a, %b
   %icmp1 = icmp ne i32 %a, %c
   %or = or i1 %icmp0, %icmp1
   %select = select i1 %or, i64 %x, i64 %y
   store i64 %select, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}opt_select_i64_or_cmp_f32:
 ; GCN-DAG: v_cmp_lg_f32_e32 vcc,
 ; GCN-DAG: v_cmp_lg_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
-; GCN: s_or_b64 vcc, vcc, [[CMP1]]
-; GCN: v_cndmask_b32_e32 v[[RESULT1:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: v_cndmask_b32_e32 v[[RESULT0:[0-9]+]], {{v[0-9]+}}, {{v[0-9]+}}, vcc
-; GCN: buffer_store_dwordx2 v[[[RESULT0]]:[[RESULT1]]]
+; GCN: s_or_b64 [[OR:s\[[0-9]+:[0-9]+\]]], vcc, [[CMP1]]
+; GCN: s_and_b64 [[AND:s\[[0-9]+:[0-9]+\]]], [[OR]], exec
+; GCN-DAG: s_cselect_b32 [[RESULT0:s[0-9]+]]
+; GCN-DAG: s_cselect_b32 [[RESULT1:s[0-9]+]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT1:[0-9]+]], [[RESULT0]]
+; GCN-DAG: v_mov_b32_e32 v[[VRESULT0:[0-9]+]], [[RESULT1]]
+; GCN: buffer_store_dwordx2 v[[[VRESULT0]]:[[VRESULT1]]]
 define amdgpu_kernel void @opt_select_i64_or_cmp_f32(i64 addrspace(1)* %out, float %a, float %b, float %c, i64 %x, i64 %y) #0 {
   %fcmp0 = fcmp one float %a, %b
   %fcmp1 = fcmp one float %a, %c
   %or = or i1 %fcmp0, %fcmp1
   %select = select i1 %or, i64 %x, i64 %y
   store i64 %select, i64 addrspace(1)* %out
   ret void
 }

llvm/test/CodeGen/AMDGPU/select-vectors.ll

define amdgpu_kernel void @v_select_v16i8(<16 x i8> addrspace(1)* %out, <16 x i8> addrspace(1)* %a.ptr, <16 x i8> addrspace(1)* %b.ptr, i32 %c) #0 {
   store <16 x i8> %select, <16 x i8> addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_v4i8:
 ; GFX89: s_cselect_b32
 ; GFX89-NOT: s_cselect_b32
 
-; SI: v_cndmask_b32
+; SI: s_cselect_b32
 ; SI-NOT: cndmask
 define amdgpu_kernel void @select_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> %a, <4 x i8> %b, i8 %c) #0 {
   %cmp = icmp eq i8 %c, 0
   %select = select i1 %cmp, <4 x i8> %a, <4 x i8> %b
   store <4 x i8> %select, <4 x i8> addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_v2i16:
 ; GFX89: s_load_dwordx4
 ; GFX89: s_cselect_b32
 ; GFX89-NOT: s_cselect_b32
 
-; SI: v_cndmask_b32_e32
+; SI: s_cselect_b32
 ; SI-NOT: v_cndmask_b32e
 define amdgpu_kernel void @select_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> %a, <2 x i16> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x i16> %a, <2 x i16> %b
   store <2 x i16> %select, <2 x i16> addrspace(1)* %out, align 4
   ret void
 }
 
define amdgpu_kernel void @v_select_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %a.ptr, <2 x i16> addrspace(1)* %b.ptr, i32 %c) #0 {
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_select_v3i16:
 ; SI: v_cndmask_b32_e32
 ; SI: cndmask
 ; SI-NOT: cndmask
 
-; GFX89: v_cndmask_b32_e32
-; GFX89: cndmask
-; VI: cndmask
-; GFX89-NOT: cndmask
+; VI: s_cselect_b32
+; VI: s_cselect_b32
+; GFX9: cndmask
+; GFX9: cndmask
 define amdgpu_kernel void @v_select_v3i16(<3 x i16> addrspace(1)* %out, <3 x i16> addrspace(1)* %a.ptr, <3 x i16> addrspace(1)* %b.ptr, i32 %c) #0 {
   %a = load <3 x i16>, <3 x i16> addrspace(1)* %a.ptr
   %b = load <3 x i16>, <3 x i16> addrspace(1)* %b.ptr
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <3 x i16> %a, <3 x i16> %b
   store <3 x i16> %select, <3 x i16> addrspace(1)* %out, align 4
   ret void
 }
define amdgpu_kernel void @v_select_v8i16(<8 x i16> addrspace(1)* %out, <8 x i16> addrspace(1)* %a.ptr, <8 x i16> addrspace(1)* %b.ptr, i32 %c) #0 {
   store <8 x i16> %select, <8 x i16> addrspace(1)* %out, align 4
   ret void
 }
 
 ; FIXME: Expansion with bitwise operations may be better if doing a
 ; vector select with SGPR inputs.
 
 ; GCN-LABEL: {{^}}s_select_v2i32:
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 ; GCN: buffer_store_dwordx2
 define amdgpu_kernel void @s_select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x i32> %a, <2 x i32> %b
   store <2 x i32> %select, <2 x i32> addrspace(1)* %out, align 8
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_select_v4i32:
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 ; GCN: buffer_store_dwordx4
 define amdgpu_kernel void @s_select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %a, <4 x i32> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x i32> %a, <4 x i32> %b
   store <4 x i32> %select, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
bb:
   %tmp2 = icmp ult i32 %cond, 32
   %val = load <4 x i32>, <4 x i32> addrspace(1)* %in
   %tmp3 = select i1 %tmp2, <4 x i32> %val, <4 x i32> zeroinitializer
   store <4 x i32> %tmp3, <4 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_v8i32:
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 define amdgpu_kernel void @select_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> %a, <8 x i32> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x i32> %a, <8 x i32> %b
   store <8 x i32> %select, <8 x i32> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_select_v2f32:
-; GCN-DAG: s_load_dwordx4 s[[[ALO:[0-9]+]]:[[BHI:[0-9]+]]], s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
-
-; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s[[BHI]]
-; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s[[ALO]]
 ; GCN-DAG: s_cmp_eq_u32 s{{[0-9]+}}, 0{{$}}
-
-; GCN-DAG: v_cndmask_b32_e32
-; GCN-DAG: v_mov_b32_e32 v{{[0-9]+}}, s{{[0-9]+}}
-; GCN-DAG: v_cndmask_b32_e32
+; GCN-DAG: s_cselect_b32
+; GCN-DAG: s_cselect_b32
 ; GCN: buffer_store_dwordx2
 define amdgpu_kernel void @s_select_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x float> %a, <2 x float> %b
   store <2 x float> %select, <2 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_select_v3f32:
 ; GCN: s_cmp_eq_u32 s{{[0-9]+}}, 0{{$}}
 
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 
 ; GCN: buffer_store_dwordx
 define amdgpu_kernel void @s_select_v3f32(<3 x float> addrspace(1)* %out, <3 x float> %a, <3 x float> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <3 x float> %a, <3 x float> %b
   store <3 x float> %select, <3 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_select_v4f32:
 ; GCN: s_load_dwordx8
 ; GCN: s_cmp_eq_u32 s{{[0-9]+}}, 0{{$}}
 
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 
 ; GCN: buffer_store_dwordx4
 define amdgpu_kernel void @s_select_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x float> %a, <4 x float> %b
   store <4 x float> %select, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
bb:
   %tmp3 = select i1 %tmp2, <4 x float> %val, <4 x float> zeroinitializer
   store <4 x float> %tmp3, <4 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_select_v5f32:
 ; GCN: s_cmp_eq_u32 s{{[0-9]+}}, 0{{$}}
 
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 
 ; GCN: buffer_store_dwordx
 define amdgpu_kernel void @s_select_v5f32(<5 x float> addrspace(1)* %out, <5 x float> %a, <5 x float> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <5 x float> %a, <5 x float> %b
   store <5 x float> %select, <5 x float> addrspace(1)* %out, align 16
   ret void
 }

 define amdgpu_kernel void @select_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x float> %a, <8 x float> %b
   store <8 x float> %select, <8 x float> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_v2f64:
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 define amdgpu_kernel void @select_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %a, <2 x double> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <2 x double> %a, <2 x double> %b
   store <2 x double> %select, <2 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_v4f64:
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 define amdgpu_kernel void @select_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, <4 x double> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <4 x double> %a, <4 x double> %b
   store <4 x double> %select, <4 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_v8f64:
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 define amdgpu_kernel void @select_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, <8 x double> %b, i32 %c) #0 {
   %cmp = icmp eq i32 %c, 0
   %select = select i1 %cmp, <8 x double> %a, <8 x double> %b
   store <8 x double> %select, <8 x double> addrspace(1)* %out, align 16
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_select_v2f16:

llvm/test/CodeGen/AMDGPU/select64.ll

-; RUN: llc < %s -march=amdgcn -mcpu=tahiti -verify-machineinstrs | FileCheck -check-prefixes=SI,GCN %s
-; RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck -check-prefixes=VI,GCN %s
+; RUN: llc < %s -march=amdgcn -mcpu=tahiti -verify-machineinstrs | FileCheck -check-prefix=GCN %s
+; RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck -check-prefix=GCN %s
 
 ; GCN-LABEL: {{^}}select0:
 ; i64 select should be split into two i32 selects, and we shouldn't need
 ; to use a shfit to extract the hi dword of the input.
 ; GCN-NOT: s_lshr_b64
-; GCN: v_cndmask
-; GCN: v_cndmask
+; GCN: s_cselect_b32
+; GCN: s_cselect_b32
 define amdgpu_kernel void @select0(i64 addrspace(1)* %out, i32 %cond, i64 %in) {
 entry:
   %0 = icmp ugt i32 %cond, 5
   %1 = select i1 %0, i64 0, i64 %in
   store i64 %1, i64 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_trunc_i64:
-; VI: s_cselect_b32
-; VI-NOT: s_cselect_b32
-; SI: v_cndmask_b32
-; SI-NOT: v_cndmask_b32
+; GCN: s_cselect_b32
+; GCN-NOT: s_cselect_b32
 define amdgpu_kernel void @select_trunc_i64(i32 addrspace(1)* %out, i32 %cond, i64 %in) nounwind {
   %cmp = icmp ugt i32 %cond, 5
   %sel = select i1 %cmp, i64 0, i64 %in
   %trunc = trunc i64 %sel to i32
   store i32 %trunc, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}select_trunc_i64_2:
-; VI: s_cselect_b32
-; VI-NOT: s_cselect_b32
-; SI: v_cndmask_b32
-; SI-NOT: v_cndmask_b32
+; GCN: s_cselect_b32
+; GCN-NOT: s_cselect_b32
 define amdgpu_kernel void @select_trunc_i64_2(i32 addrspace(1)* %out, i32 %cond, i64 %a, i64 %b) nounwind {
   %cmp = icmp ugt i32 %cond, 5
   %sel = select i1 %cmp, i64 %a, i64 %b
   %trunc = trunc i64 %sel to i32
   store i32 %trunc, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_select_trunc_i64_2:
-; VI: s_cselect_b32
-; VI-NOT: s_cselect_b32
-; SI: v_cndmask_b32
-; SI-NOT: v_cndmask_b32
+; GCN: s_cselect_b32
+; GCN-NOT: s_cselect_b32
 define amdgpu_kernel void @v_select_trunc_i64_2(i32 addrspace(1)* %out, i32 %cond, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
   %cmp = icmp ugt i32 %cond, 5
   %a = load i64, i64 addrspace(1)* %aptr, align 8
   %b = load i64, i64 addrspace(1)* %bptr, align 8
   %sel = select i1 %cmp, i64 %a, i64 %b
   %trunc = trunc i64 %sel to i32
   store i32 %trunc, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_select_i64_split_imm:
-; GCN-DAG: v_cndmask_b32_e32 {{v[0-9]+}}, 0, {{v[0-9]+}}
-; GCN-DAG: v_cndmask_b32_e32 {{v[0-9]+}}, 63, {{v[0-9]+}}
+; GCN-DAG: s_cselect_b32
+; GCN-DAG: s_cselect_b32
 ; GCN: s_endpgm
 define amdgpu_kernel void @v_select_i64_split_imm(i64 addrspace(1)* %out, i32 %cond, i64 addrspace(1)* %aptr, i64 addrspace(1)* %bptr) nounwind {
   %cmp = icmp ugt i32 %cond, 5
   %a = load i64, i64 addrspace(1)* %aptr, align 8
   %b = load i64, i64 addrspace(1)* %bptr, align 8
   %sel = select i1 %cmp, i64 %a, i64 270582939648 ; 63 << 32
   store i64 %sel, i64 addrspace(1)* %out, align 8
   ret void
 }

llvm/test/CodeGen/AMDGPU/selectcc.ll

 ; RUN: llc -verify-machineinstrs -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
 ; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,SI -check-prefix=FUNC %s
 ; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,VI -check-prefix=FUNC %s
 
 ; FUNC-LABEL: {{^}}selectcc_i64:
 ; EG: XOR_INT
 ; EG: XOR_INT
 ; EG: OR_INT
 ; EG: CNDE_INT
 ; EG: CNDE_INT
 ; SI: v_cmp_eq_u64
 ; VI: s_cmp_eq_u64
-; GCN: v_cndmask
-; GCN: v_cndmask
+; GCN: s_cselect_b32
 define amdgpu_kernel void @selectcc_i64(i64 addrspace(1) * %out, i64 %lhs, i64 %rhs, i64 %true, i64 %false) {
 entry:
   %0 = icmp eq i64 %lhs, %rhs
   %1 = select i1 %0, i64 %true, i64 %false
   store i64 %1, i64 addrspace(1)* %out
   ret void
 }

llvm/test/CodeGen/AMDGPU/setcc64.ll

 entry:
   %tmp0 = icmp sle i64 %a, %b
   %tmp1 = sext i1 %tmp0 to i32
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}i128_sle:
-; GCN: v_cmp_le_u64
 ; GCN: v_cmp_le_i64
+; CGV: v_cndmask
 ; SI: v_cmp_eq_u64
 ; VI: s_cmp_eq_u64
 define amdgpu_kernel void @i128_sle(i32 addrspace(1)* %out, i128 %a, i128 %b) #0 {
 entry:
   %tmp0 = icmp sle i128 %a, %b
   %tmp1 = sext i1 %tmp0 to i32
   store i32 %tmp1, i32 addrspace(1)* %out
   ret void

llvm/test/CodeGen/AMDGPU/shift-i128.ll

 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   %shl = ashr i128 33, %rhs
   ret i128 %shl
 }
 
 define amdgpu_kernel void @s_shl_i128_ss(i128 %lhs, i128 %rhs) {
 ; GCN-LABEL: s_shl_i128_ss:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx8 s[4:11], s[4:5], 0x0
+; GCN-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x0
 ; GCN-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_sub_i32 s9, 64, s8
-; GCN-NEXT:    s_sub_i32 s2, s8, 64
-; GCN-NEXT:    s_lshl_b64 s[0:1], s[6:7], s8
-; GCN-NEXT:    s_lshr_b64 s[10:11], s[4:5], s9
-; GCN-NEXT:    s_lshl_b64 s[2:3], s[4:5], s2
-; GCN-NEXT:    s_or_b64 s[10:11], s[0:1], s[10:11]
-; GCN-NEXT:    s_cmp_lt_u32 s8, 64
-; GCN-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-NEXT:    v_mov_b32_e32 v1, s11
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    s_cmp_eq_u32 s8, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-NEXT:    v_mov_b32_e32 v1, s10
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s6
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v0, v1, s[0:1]
-; GCN-NEXT:    s_lshl_b64 s[0:1], s[4:5], s8
-; GCN-NEXT:    v_mov_b32_e32 v0, s1
-; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
+; GCN-NEXT:    s_sub_i32 s5, s4, 64
+; GCN-NEXT:    s_sub_i32 s12, 64, s4
+; GCN-NEXT:    s_lshl_b64 s[6:7], s[2:3], s4
+; GCN-NEXT:    s_lshl_b64 s[8:9], s[0:1], s4
+; GCN-NEXT:    s_lshl_b64 s[10:11], s[0:1], s5
+; GCN-NEXT:    s_lshr_b64 s[0:1], s[0:1], s12
+; GCN-NEXT:    s_or_b64 s[0:1], s[6:7], s[0:1]
+; GCN-NEXT:    s_cmp_lt_u32 s4, 64
+; GCN-NEXT:    s_cselect_b32 s0, s0, s10
+; GCN-NEXT:    s_cselect_b32 s1, s1, s11
+; GCN-NEXT:    s_cselect_b32 s5, s9, 0
+; GCN-NEXT:    s_cselect_b32 s6, s8, 0
+; GCN-NEXT:    s_cmp_eq_u32 s4, 0
+; GCN-NEXT:    s_cselect_b32 s1, s3, s1
+; GCN-NEXT:    s_cselect_b32 s0, s2, s0
+; GCN-NEXT:    v_mov_b32_e32 v0, s6
+; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    v_mov_b32_e32 v2, s0
+; GCN-NEXT:    v_mov_b32_e32 v3, s1
 ; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
   %shift = shl i128 %lhs, %rhs
   store i128 %shift, i128 addrspace(1)* null
   ret void
 }
 
 define amdgpu_kernel void @s_lshr_i128_ss(i128 %lhs, i128 %rhs) {
 ; GCN-LABEL: s_lshr_i128_ss:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx8 s[4:11], s[4:5], 0x0
+; GCN-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x0
 ; GCN-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_sub_i32 s9, 64, s8
-; GCN-NEXT:    s_sub_i32 s2, s8, 64
-; GCN-NEXT:    s_lshr_b64 s[0:1], s[4:5], s8
-; GCN-NEXT:    s_lshl_b64 s[10:11], s[6:7], s9
-; GCN-NEXT:    s_lshr_b64 s[2:3], s[6:7], s2
-; GCN-NEXT:    s_or_b64 s[10:11], s[0:1], s[10:11]
-; GCN-NEXT:    s_cmp_lt_u32 s8, 64
-; GCN-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-NEXT:    v_mov_b32_e32 v1, s11
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    s_cmp_eq_u32 s8, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-NEXT:    v_mov_b32_e32 v2, s10
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GCN-NEXT:    s_lshr_b64 s[0:1], s[6:7], s8
-; GCN-NEXT:    v_mov_b32_e32 v2, s1
-; GCN-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v2, s0
-; GCN-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
+; GCN-NEXT:    s_sub_i32 s5, s4, 64
+; GCN-NEXT:    s_sub_i32 s12, 64, s4
+; GCN-NEXT:    s_lshr_b64 s[6:7], s[0:1], s4
+; GCN-NEXT:    s_lshr_b64 s[8:9], s[2:3], s4
+; GCN-NEXT:    s_lshr_b64 s[10:11], s[2:3], s5
+; GCN-NEXT:    s_lshl_b64 s[2:3], s[2:3], s12
+; GCN-NEXT:    s_or_b64 s[2:3], s[6:7], s[2:3]
+; GCN-NEXT:    s_cmp_lt_u32 s4, 64
+; GCN-NEXT:    s_cselect_b32 s2, s2, s10
+; GCN-NEXT:    s_cselect_b32 s3, s3, s11
+; GCN-NEXT:    s_cselect_b32 s5, s9, 0
+; GCN-NEXT:    s_cselect_b32 s6, s8, 0
+; GCN-NEXT:    s_cmp_eq_u32 s4, 0
+; GCN-NEXT:    s_cselect_b32 s1, s1, s3
+; GCN-NEXT:    s_cselect_b32 s0, s0, s2
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
+; GCN-NEXT:    v_mov_b32_e32 v2, s6
+; GCN-NEXT:    v_mov_b32_e32 v3, s5
 ; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
   %shift = lshr i128 %lhs, %rhs
   store i128 %shift, i128 addrspace(1)* null
   ret void
 }
 
 define amdgpu_kernel void @s_ashr_i128_ss(i128 %lhs, i128 %rhs) {
 ; GCN-LABEL: s_ashr_i128_ss:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx8 s[4:11], s[4:5], 0x0
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_ashr_i32 s2, s7, 31
-; GCN-NEXT:    s_ashr_i64 s[0:1], s[6:7], s8
-; GCN-NEXT:    s_cmp_lt_u32 s8, 64
-; GCN-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s0
-; GCN-NEXT:    s_sub_i32 s0, s8, 64
-; GCN-NEXT:    s_ashr_i64 s[2:3], s[6:7], s0
-; GCN-NEXT:    s_sub_i32 s0, 64, s8
-; GCN-NEXT:    s_lshl_b64 s[0:1], s[6:7], s0
-; GCN-NEXT:    s_lshr_b64 s[6:7], s[4:5], s8
-; GCN-NEXT:    s_or_b64 s[6:7], s[6:7], s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-NEXT:    s_cmp_eq_u32 s8, 0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-NEXT:    s_cselect_b64 s[0:1], -1, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-NEXT:    v_mov_b32_e32 v4, s6
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
-; GCN-NEXT:    v_mov_b32_e32 v6, s4
+; GCN-NEXT:    s_load_dwordx8 s[0:7], s[4:5], 0x0
 ; GCN-NEXT:    v_mov_b32_e32 v4, 0
 ; GCN-NEXT:    v_mov_b32_e32 v5, 0
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[0:1]
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_sub_i32 s5, 64, s4
+; GCN-NEXT:    s_lshr_b64 s[6:7], s[0:1], s4
+; GCN-NEXT:    s_sub_i32 s10, s4, 64
+; GCN-NEXT:    s_lshl_b64 s[8:9], s[2:3], s5
+; GCN-NEXT:    s_ashr_i32 s12, s3, 31
+; GCN-NEXT:    s_ashr_i64 s[10:11], s[2:3], s10
+; GCN-NEXT:    s_or_b64 s[6:7], s[6:7], s[8:9]
+; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], s4
+; GCN-NEXT:    s_cmp_lt_u32 s4, 64
+; GCN-NEXT:    s_cselect_b32 s3, s3, s12
+; GCN-NEXT:    s_cselect_b32 s2, s2, s12
+; GCN-NEXT:    s_cselect_b32 s5, s6, s10
+; GCN-NEXT:    s_cselect_b32 s6, s7, s11
+; GCN-NEXT:    s_cmp_eq_u32 s4, 0
+; GCN-NEXT:    s_cselect_b32 s1, s1, s6
+; GCN-NEXT:    s_cselect_b32 s0, s0, s5
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
+; GCN-NEXT:    v_mov_b32_e32 v2, s2
+; GCN-NEXT:    v_mov_b32_e32 v3, s3
 ; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
   %shift = ashr i128 %lhs, %rhs
   store i128 %shift, i128 addrspace(1)* null
   ret void
 }
 
 define <2 x i128> @v_shl_v2i128_vv(<2 x i128> %lhs, <2 x i128> %rhs) {

 ; GCN-NEXT:    s_setpc_b64 s[30:31]
   %shl = ashr <2 x i128> %lhs, %rhs
   ret <2 x i128> %shl
 }
 
 define amdgpu_kernel void @s_shl_v2i128ss(<2 x i128> %lhs, <2 x i128> %rhs) {
 ; GCN-LABEL: s_shl_v2i128ss:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
-; GCN-NEXT:    v_mov_b32_e32 v10, 16
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
-; GCN-NEXT:    v_mov_b32_e32 v11, 0
-; GCN-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-NEXT:    s_load_dwordx16 s[0:15], s[4:5], 0x0
+; GCN-NEXT:    v_mov_b32_e32 v6, 16
+; GCN-NEXT:    v_mov_b32_e32 v4, 0
+; GCN-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_cmp_lt_u64_e64 s[0:1], s[12:13], 64
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[14:15], 0
-; GCN-NEXT:    s_sub_i32 s22, 64, s12
-; GCN-NEXT:    s_sub_i32 s20, s12, 64
-; GCN-NEXT:    s_lshr_b64 s[22:23], s[4:5], s22
-; GCN-NEXT:    s_lshl_b64 s[24:25], s[6:7], s12
-; GCN-NEXT:    s_lshl_b64 s[20:21], s[4:5], s20
-; GCN-NEXT:    s_or_b64 s[22:23], s[24:25], s[22:23]
-; GCN-NEXT:    s_and_b64 vcc, s[2:3], s[0:1]
-; GCN-NEXT:    s_or_b64 s[0:1], s[12:13], s[14:15]
-; GCN-NEXT:    v_mov_b32_e32 v0, s21
-; GCN-NEXT:    v_mov_b32_e32 v1, s23
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[0:1], 0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v0, s20
-; GCN-NEXT:    v_mov_b32_e32 v1, s22
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s6
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v0, v1, s[0:1]
-; GCN-NEXT:    v_cmp_lt_u64_e64 s[0:1], s[16:17], 64
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[18:19], 0
-; GCN-NEXT:    s_sub_i32 s13, 64, s16
-; GCN-NEXT:    s_sub_i32 s6, s16, 64
-; GCN-NEXT:    s_lshr_b64 s[14:15], s[8:9], s13
-; GCN-NEXT:    s_lshl_b64 s[20:21], s[10:11], s16
-; GCN-NEXT:    s_lshl_b64 s[6:7], s[8:9], s6
-; GCN-NEXT:    s_or_b64 s[14:15], s[20:21], s[14:15]
-; GCN-NEXT:    s_and_b64 s[0:1], s[2:3], s[0:1]
-; GCN-NEXT:    s_or_b64 s[2:3], s[16:17], s[18:19]
+; GCN-NEXT:    v_cmp_lt_u64_e64 s[16:17], s[8:9], 64
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[10:11], 0
+; GCN-NEXT:    s_sub_i32 s22, 64, s8
+; GCN-NEXT:    s_sub_i32 s20, s8, 64
+; GCN-NEXT:    s_lshr_b64 s[22:23], s[0:1], s22
+; GCN-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
+; GCN-NEXT:    s_lshl_b64 s[18:19], s[2:3], s8
+; GCN-NEXT:    s_lshl_b64 s[20:21], s[0:1], s20
+; GCN-NEXT:    s_or_b64 s[18:19], s[18:19], s[22:23]
+; GCN-NEXT:    s_and_b64 s[22:23], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s19, s19, s21
+; GCN-NEXT:    s_or_b64 s[10:11], s[8:9], s[10:11]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[10:11], 0
+; GCN-NEXT:    s_and_b64 s[22:23], s[10:11], exec
+; GCN-NEXT:    s_cselect_b32 s9, s3, s19
+; GCN-NEXT:    s_and_b64 s[22:23], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s3, s18, s20
+; GCN-NEXT:    s_and_b64 s[10:11], s[10:11], exec
+; GCN-NEXT:    v_cmp_lt_u64_e64 s[10:11], s[12:13], 64
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[14:15], 0
+; GCN-NEXT:    s_cselect_b32 s22, s2, s3
+; GCN-NEXT:    s_and_b64 s[2:3], s[18:19], s[10:11]
+; GCN-NEXT:    s_sub_i32 s18, 64, s12
+; GCN-NEXT:    s_sub_i32 s10, s12, 64
+; GCN-NEXT:    s_lshr_b64 s[18:19], s[4:5], s18
+; GCN-NEXT:    s_lshl_b64 s[20:21], s[6:7], s12
+; GCN-NEXT:    s_lshl_b64 s[10:11], s[4:5], s10
+; GCN-NEXT:    s_or_b64 s[18:19], s[20:21], s[18:19]
+; GCN-NEXT:    s_and_b64 s[20:21], s[2:3], exec
+; GCN-NEXT:    s_cselect_b32 s11, s19, s11
+; GCN-NEXT:    s_or_b64 s[14:15], s[12:13], s[14:15]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[14:15], 0
+; GCN-NEXT:    s_and_b64 s[20:21], s[14:15], exec
+; GCN-NEXT:    s_cselect_b32 s13, s7, s11
+; GCN-NEXT:    s_and_b64 s[20:21], s[2:3], exec
+; GCN-NEXT:    s_cselect_b32 s7, s18, s10
+; GCN-NEXT:    s_and_b64 s[10:11], s[14:15], exec
+; GCN-NEXT:    s_cselect_b32 s10, s6, s7
+; GCN-NEXT:    s_lshl_b64 s[0:1], s[0:1], s8
+; GCN-NEXT:    s_and_b64 s[6:7], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s6, s1, 0
+; GCN-NEXT:    s_cselect_b32 s7, s0, 0
+; GCN-NEXT:    s_lshl_b64 s[0:1], s[4:5], s12
+; GCN-NEXT:    s_and_b64 s[2:3], s[2:3], exec
+; GCN-NEXT:    s_cselect_b32 s1, s1, 0
+; GCN-NEXT:    s_cselect_b32 s0, s0, 0
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
+; GCN-NEXT:    v_mov_b32_e32 v2, s10
+; GCN-NEXT:    v_mov_b32_e32 v3, s13
+; GCN-NEXT:    flat_store_dwordx4 v[6:7], v[0:3]
+; GCN-NEXT:    s_nop 0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s7
-; GCN-NEXT:    v_mov_b32_e32 v1, s15
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[2:3], 0
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v1, s11
-; GCN-NEXT:    v_cndmask_b32_e64 v7, v0, v1, s[2:3]
-; GCN-NEXT:    v_mov_b32_e32 v0, s6
-; GCN-NEXT:    v_mov_b32_e32 v1, s14
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v1, s10
-; GCN-NEXT:    v_cndmask_b32_e64 v6, v0, v1, s[2:3]
-; GCN-NEXT:    s_lshl_b64 s[2:3], s[4:5], s12
-; GCN-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v0, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-NEXT:    s_lshl_b64 s[2:3], s[8:9], s16
-; GCN-NEXT:    v_mov_b32_e32 v4, s3
-; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, v4, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v4, s2
-; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, v4, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; GCN-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
-; GCN-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GCN-NEXT:    v_mov_b32_e32 v1, s6
+; GCN-NEXT:    v_mov_b32_e32 v2, s22
+; GCN-NEXT:    v_mov_b32_e32 v3, s9
+; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
   %shift = shl <2 x i128> %lhs, %rhs
   store <2 x i128> %shift, <2 x i128> addrspace(1)* null
   ret void
 }
 
 define amdgpu_kernel void @s_lshr_v2i128_ss(<2 x i128> %lhs, <2 x i128> %rhs) {
 ; GCN-LABEL: s_lshr_v2i128_ss:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
-; GCN-NEXT:    v_mov_b32_e32 v10, 16
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
-; GCN-NEXT:    v_mov_b32_e32 v11, 0
-; GCN-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-NEXT:    s_load_dwordx16 s[0:15], s[4:5], 0x0
+; GCN-NEXT:    v_mov_b32_e32 v6, 16
+; GCN-NEXT:    v_mov_b32_e32 v4, 0
+; GCN-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_cmp_lt_u64_e64 s[0:1], s[12:13], 64
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[14:15], 0
-; GCN-NEXT:    s_sub_i32 s22, 64, s12
-; GCN-NEXT:    s_sub_i32 s20, s12, 64
-; GCN-NEXT:    s_lshl_b64 s[22:23], s[6:7], s22
-; GCN-NEXT:    s_lshr_b64 s[24:25], s[4:5], s12
-; GCN-NEXT:    s_lshr_b64 s[20:21], s[6:7], s20
-; GCN-NEXT:    s_or_b64 s[22:23], s[24:25], s[22:23]
-; GCN-NEXT:    s_and_b64 vcc, s[2:3], s[0:1]
-; GCN-NEXT:    s_or_b64 s[0:1], s[12:13], s[14:15]
-; GCN-NEXT:    v_mov_b32_e32 v0, s21
-; GCN-NEXT:    v_mov_b32_e32 v1, s23
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[0:1], 0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v0, s20
-; GCN-NEXT:    v_mov_b32_e32 v2, s22
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GCN-NEXT:    v_cmp_lt_u64_e64 s[0:1], s[16:17], 64
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[18:19], 0
-; GCN-NEXT:    s_sub_i32 s13, 64, s16
-; GCN-NEXT:    s_sub_i32 s4, s16, 64
-; GCN-NEXT:    s_lshl_b64 s[14:15], s[10:11], s13
-; GCN-NEXT:    s_lshr_b64 s[20:21], s[8:9], s16
-; GCN-NEXT:    s_lshr_b64 s[4:5], s[10:11], s4
-; GCN-NEXT:    s_or_b64 s[14:15], s[20:21], s[14:15]
-; GCN-NEXT:    s_and_b64 s[0:1], s[2:3], s[0:1]
-; GCN-NEXT:    s_or_b64 s[2:3], s[16:17], s[18:19]
-; GCN-NEXT:    v_mov_b32_e32 v2, s5
-; GCN-NEXT:    v_mov_b32_e32 v3, s15
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[2:3], 0
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v3, s9
-; GCN-NEXT:    v_cndmask_b32_e64 v5, v2, v3, s[2:3]
-; GCN-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-NEXT:    v_mov_b32_e32 v3, s14
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v3, s8
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v2, v3, s[2:3]
+; GCN-NEXT:    v_cmp_lt_u64_e64 s[16:17], s[8:9], 64
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[10:11], 0
+; GCN-NEXT:    s_sub_i32 s22, 64, s8
+; GCN-NEXT:    s_sub_i32 s20, s8, 64
+; GCN-NEXT:    s_lshl_b64 s[22:23], s[2:3], s22
+; GCN-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
+; GCN-NEXT:    s_lshr_b64 s[18:19], s[0:1], s8
+; GCN-NEXT:    s_lshr_b64 s[20:21], s[2:3], s20
+; GCN-NEXT:    s_or_b64 s[18:19], s[18:19], s[22:23]
+; GCN-NEXT:    s_and_b64 s[22:23], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s19, s19, s21
+; GCN-NEXT:    s_or_b64 s[10:11], s[8:9], s[10:11]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[10:11], 0
+; GCN-NEXT:    s_and_b64 s[22:23], s[10:11], exec
+; GCN-NEXT:    s_cselect_b32 s9, s1, s19
+; GCN-NEXT:    s_and_b64 s[22:23], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s1, s18, s20
+; GCN-NEXT:    s_and_b64 s[10:11], s[10:11], exec
+; GCN-NEXT:    v_cmp_lt_u64_e64 s[10:11], s[12:13], 64
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[14:15], 0
+; GCN-NEXT:    s_cselect_b32 s22, s0, s1
+; GCN-NEXT:    s_and_b64 s[0:1], s[18:19], s[10:11]
+; GCN-NEXT:    s_sub_i32 s18, 64, s12
+; GCN-NEXT:    s_sub_i32 s10, s12, 64
+; GCN-NEXT:    s_lshl_b64 s[18:19], s[6:7], s18
+; GCN-NEXT:    s_lshr_b64 s[20:21], s[4:5], s12
+; GCN-NEXT:    s_lshr_b64 s[10:11], s[6:7], s10
+; GCN-NEXT:    s_or_b64 s[18:19], s[20:21], s[18:19]
+; GCN-NEXT:    s_and_b64 s[20:21], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s11, s19, s11
+; GCN-NEXT:    s_or_b64 s[14:15], s[12:13], s[14:15]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[14:15], 0
+; GCN-NEXT:    s_and_b64 s[20:21], s[14:15], exec
+; GCN-NEXT:    s_cselect_b32 s13, s5, s11
+; GCN-NEXT:    s_and_b64 s[20:21], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s5, s18, s10
+; GCN-NEXT:    s_and_b64 s[10:11], s[14:15], exec
+; GCN-NEXT:    s_cselect_b32 s10, s4, s5
+; GCN-NEXT:    s_lshr_b64 s[2:3], s[2:3], s8
+; GCN-NEXT:    s_and_b64 s[4:5], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s4, s3, 0
+; GCN-NEXT:    s_cselect_b32 s5, s2, 0
 ; GCN-NEXT:    s_lshr_b64 s[2:3], s[6:7], s12
-; GCN-NEXT:    v_mov_b32_e32 v2, s3
-; GCN-NEXT:    v_cndmask_b32_e32 v3, 0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v2, s2
-; GCN-NEXT:    s_lshr_b64 s[2:3], s[10:11], s16
-; GCN-NEXT:    v_mov_b32_e32 v6, s3
-; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, v6, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v6, s2
-; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, v6, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
-; GCN-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
-; GCN-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s3, 0
+; GCN-NEXT:    s_cselect_b32 s1, s2, 0
+; GCN-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-NEXT:    v_mov_b32_e32 v1, s13
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_mov_b32_e32 v3, s0
+; GCN-NEXT:    flat_store_dwordx4 v[6:7], v[0:3]
+; GCN-NEXT:    s_nop 0
+; GCN-NEXT:    v_mov_b32_e32 v0, s22
+; GCN-NEXT:    v_mov_b32_e32 v1, s9
+; GCN-NEXT:    v_mov_b32_e32 v2, s5
+; GCN-NEXT:    v_mov_b32_e32 v3, s4
+; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
   %shift = lshr <2 x i128> %lhs, %rhs
   store <2 x i128> %shift, <2 x i128> addrspace(1)* null
   ret void
 }
 
 define amdgpu_kernel void @s_ashr_v2i128_ss(<2 x i128> %lhs, <2 x i128> %rhs) {
 ; GCN-LABEL: s_ashr_v2i128_ss:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx16 s[4:19], s[4:5], 0x0
-; GCN-NEXT:    v_mov_b32_e32 v10, 16
-; GCN-NEXT:    v_mov_b32_e32 v8, 0
-; GCN-NEXT:    v_mov_b32_e32 v11, 0
-; GCN-NEXT:    v_mov_b32_e32 v9, 0
+; GCN-NEXT:    s_load_dwordx16 s[0:15], s[4:5], 0x0
+; GCN-NEXT:    v_mov_b32_e32 v6, 16
+; GCN-NEXT:    v_mov_b32_e32 v4, 0
+; GCN-NEXT:    v_mov_b32_e32 v7, 0
+; GCN-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_cmp_lt_u64_e64 s[0:1], s[12:13], 64
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[14:15], 0
-; GCN-NEXT:    s_sub_i32 s22, 64, s12
-; GCN-NEXT:    s_sub_i32 s20, s12, 64
-; GCN-NEXT:    s_lshl_b64 s[22:23], s[6:7], s22
-; GCN-NEXT:    s_lshr_b64 s[24:25], s[4:5], s12
-; GCN-NEXT:    s_ashr_i64 s[20:21], s[6:7], s20
-; GCN-NEXT:    s_or_b64 s[22:23], s[24:25], s[22:23]
-; GCN-NEXT:    s_and_b64 vcc, s[2:3], s[0:1]
-; GCN-NEXT:    s_or_b64 s[0:1], s[12:13], s[14:15]
-; GCN-NEXT:    v_mov_b32_e32 v0, s21
-; GCN-NEXT:    v_mov_b32_e32 v1, s23
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[0:1], s[0:1], 0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
-; GCN-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v0, v1, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v0, s20
-; GCN-NEXT:    v_mov_b32_e32 v2, s22
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GCN-NEXT:    v_cmp_lt_u64_e64 s[0:1], s[16:17], 64
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[18:19], 0
-; GCN-NEXT:    s_sub_i32 s13, 64, s16
-; GCN-NEXT:    s_sub_i32 s4, s16, 64
-; GCN-NEXT:    s_lshl_b64 s[14:15], s[10:11], s13
-; GCN-NEXT:    s_lshr_b64 s[20:21], s[8:9], s16
-; GCN-NEXT:    s_ashr_i64 s[4:5], s[10:11], s4
-; GCN-NEXT:    s_or_b64 s[14:15], s[20:21], s[14:15]
-; GCN-NEXT:    s_and_b64 s[0:1], s[2:3], s[0:1]
-; GCN-NEXT:    s_or_b64 s[2:3], s[16:17], s[18:19]
-; GCN-NEXT:    v_mov_b32_e32 v2, s5
-; GCN-NEXT:    v_mov_b32_e32 v3, s15
-; GCN-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[2:3], 0
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v3, s9
-; GCN-NEXT:    v_cndmask_b32_e64 v5, v2, v3, s[2:3]
-; GCN-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-NEXT:    v_mov_b32_e32 v3, s14
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
-; GCN-NEXT:    v_mov_b32_e32 v3, s8
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v2, v3, s[2:3]
-; GCN-NEXT:    s_ashr_i32 s4, s7, 31
+; GCN-NEXT:    v_cmp_lt_u64_e64 s[16:17], s[8:9], 64
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[10:11], 0
+; GCN-NEXT:    s_sub_i32 s22, 64, s8
+; GCN-NEXT:    s_sub_i32 s20, s8, 64
+; GCN-NEXT:    s_lshl_b64 s[22:23], s[2:3], s22
+; GCN-NEXT:    s_and_b64 s[16:17], s[18:19], s[16:17]
+; GCN-NEXT:    s_lshr_b64 s[18:19], s[0:1], s8
+; GCN-NEXT:    s_ashr_i64 s[20:21], s[2:3], s20
+; GCN-NEXT:    s_or_b64 s[18:19], s[18:19], s[22:23]
+; GCN-NEXT:    s_and_b64 s[22:23], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s19, s19, s21
+; GCN-NEXT:    s_or_b64 s[10:11], s[8:9], s[10:11]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[10:11], 0
+; GCN-NEXT:    s_and_b64 s[22:23], s[10:11], exec
+; GCN-NEXT:    s_cselect_b32 s9, s1, s19
+; GCN-NEXT:    s_and_b64 s[22:23], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s1, s18, s20
+; GCN-NEXT:    s_and_b64 s[10:11], s[10:11], exec
+; GCN-NEXT:    v_cmp_lt_u64_e64 s[10:11], s[12:13], 64
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[14:15], 0
+; GCN-NEXT:    s_cselect_b32 s22, s0, s1
+; GCN-NEXT:    s_and_b64 s[0:1], s[18:19], s[10:11]
+; GCN-NEXT:    s_sub_i32 s18, 64, s12
+; GCN-NEXT:    s_sub_i32 s10, s12, 64
+; GCN-NEXT:    s_lshl_b64 s[18:19], s[6:7], s18
+; GCN-NEXT:    s_lshr_b64 s[20:21], s[4:5], s12
+; GCN-NEXT:    s_ashr_i64 s[10:11], s[6:7], s10
+; GCN-NEXT:    s_or_b64 s[18:19], s[20:21], s[18:19]
+; GCN-NEXT:    s_and_b64 s[20:21], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s11, s19, s11
+; GCN-NEXT:    s_or_b64 s[14:15], s[12:13], s[14:15]
+; GCN-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[14:15], 0
+; GCN-NEXT:    s_and_b64 s[20:21], s[14:15], exec
+; GCN-NEXT:    s_cselect_b32 s13, s5, s11
+; GCN-NEXT:    s_and_b64 s[20:21], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s5, s18, s10
+; GCN-NEXT:    s_and_b64 s[10:11], s[14:15], exec
+; GCN-NEXT:    s_cselect_b32 s10, s4, s5
+; GCN-NEXT:    s_ashr_i32 s11, s3, 31
+; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], s8
+; GCN-NEXT:    s_and_b64 s[4:5], s[16:17], exec
+; GCN-NEXT:    s_cselect_b32 s4, s3, s11
+; GCN-NEXT:    s_cselect_b32 s5, s2, s11
+; GCN-NEXT:    s_ashr_i32 s8, s7, 31
 ; GCN-NEXT:    s_ashr_i64 s[2:3], s[6:7], s12
-; GCN-NEXT:    v_mov_b32_e32 v2, s4
-; GCN-NEXT:    v_mov_b32_e32 v3, s3
-; GCN-NEXT:    v_mov_b32_e32 v6, s2
-; GCN-NEXT:    s_ashr_i32 s4, s11, 31
-; GCN-NEXT:    s_ashr_i64 s[2:3], s[10:11], s16
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v2, v3, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GCN-NEXT:    v_mov_b32_e32 v6, s4
-; GCN-NEXT:    v_mov_b32_e32 v7, s3
-; GCN-NEXT:    v_mov_b32_e32 v12, s2
-; GCN-NEXT:    v_cndmask_b32_e64 v7, v6, v7, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e64 v6, v6, v12, s[0:1]
-; GCN-NEXT:    flat_store_dwordx4 v[10:11], v[4:7]
-; GCN-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s3, s8
+; GCN-NEXT:    s_cselect_b32 s1, s2, s8
+; GCN-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-NEXT:    v_mov_b32_e32 v1, s13
+; GCN-NEXT:    v_mov_b32_e32 v2, s1
+; GCN-NEXT:    v_mov_b32_e32 v3, s0
+; GCN-NEXT:    flat_store_dwordx4 v[6:7], v[0:3]
+; GCN-NEXT:    s_nop 0
+; GCN-NEXT:    v_mov_b32_e32 v0, s22
+; GCN-NEXT:    v_mov_b32_e32 v1, s9
+; GCN-NEXT:    v_mov_b32_e32 v2, s5
+; GCN-NEXT:    v_mov_b32_e32 v3, s4
+; GCN-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GCN-NEXT:    s_endpgm
   %shift = ashr <2 x i128> %lhs, %rhs
   store <2 x i128> %shift, <2 x i128> addrspace(1)* null
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/sint_to_fp.f64.ll

define amdgpu_kernel void @sint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
   store double %result, double addrspace(1)* %out
   ret void
 }
 
 ; We can't fold the SGPRs into v_cndmask_b32_e64, because it already
 ; uses an SGPR (implicit vcc).
 
 ; GCN-LABEL: {{^}}sint_to_fp_i1_f64:
-; VI-DAG: s_cmp_eq_u32
-; VI-DAG: s_cselect_b32 s[[SSEL:[0-9]+]], 0xbff00000, 0
-; VI-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
-; VI-DAG: v_mov_b32_e32 v[[SEL:[0-9]+]], s[[SSEL]]
-; VI: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v[[[ZERO]]:[[SEL]]]
-; VI: s_endpgm
-
-; CI-DAG: s_cmp_eq_u32
-; CI-DAG: s_cselect_b64 vcc, -1, 0
-; CI-DAG: v_cndmask_b32_e32 v[[SEL:[0-9]+]], 0, v{{[0-9]+}}, vcc
-; CI-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
-; CI: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v[[[ZERO]]:[[SEL]]]
-; CI: s_endpgm
+; GCN-DAG: s_cmp_eq_u32
+; GCN-DAG: s_cselect_b32 s[[SSEL:[0-9]+]], 0xbff00000, 0
+; GCN-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
+; GCN-DAG: v_mov_b32_e32 v[[SEL:[0-9]+]], s[[SSEL]]
+; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v[[[ZERO]]:[[SEL]]]
+; GCN: s_endpgm
 define amdgpu_kernel void @sint_to_fp_i1_f64(double addrspace(1)* %out, i32 %in) {
   %cmp = icmp eq i32 %in, 0
   %fp = sitofp i1 %cmp to double
   store double %fp, double addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}sint_to_fp_i1_f64_load:

llvm/test/CodeGen/AMDGPU/srem64.ll

 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s0, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s1, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_cmp_le_u32_e64 s[2:3], s12, v4
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s13, v5
 ; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s12, v4
 ; 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, s11
-; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GCN-NEXT:    v_mov_b32_e32 v4, s11
+; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v4, v1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s13, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s12, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s13, 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 v4, v4, v5, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
 ; 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:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[4:5], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
-; GCN-IR-NEXT:    s_flbit_i32_b32 s12, s4
-; GCN-IR-NEXT:    s_add_i32 s14, s12, 32
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[8:9], s[10:11]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s5
-; GCN-IR-NEXT:    s_min_u32 s10, s14, s8
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
-; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
-; GCN-IR-NEXT:    s_min_u32 s14, s8, s9
-; GCN-IR-NEXT:    s_sub_u32 s8, s10, s14
-; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 63
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[4:5], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[2:3], 0
+; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s4
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s5
+; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_min_u32 s10, s10, s11
+; GCN-IR-NEXT:    s_min_u32 s14, s6, s7
+; GCN-IR-NEXT:    s_sub_u32 s12, s10, s14
+; GCN-IR-NEXT:    s_subb_u32 s13, 0, 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[12:13], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[12:13], 63
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[8:9], s[16:17]
+; GCN-IR-NEXT:    s_and_b64 s[8:9], s[16:17], exec
+; GCN-IR-NEXT:    s_cselect_b32 s9, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s8, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[18:19]
+; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
-; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[12:13], 0
-; GCN-IR-NEXT:    s_sub_i32 s8, 63, s8
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
-; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[2:3], s8
+; GCN-IR-NEXT:    s_add_u32 s16, s12, 1
+; GCN-IR-NEXT:    s_addc_u32 s17, s13, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[16:17], 0
+; GCN-IR-NEXT:    s_sub_i32 s12, 63, s12
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[8:9]
+; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[2:3], s12
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[2:3], s12
+; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[2:3], s16
 ; GCN-IR-NEXT:    s_add_u32 s16, s4, -1
 ; GCN-IR-NEXT:    s_addc_u32 s17, s5, -1
 ; GCN-IR-NEXT:    s_not_b64 s[6:7], s[10:11]
 ; GCN-IR-NEXT:    s_add_u32 s10, s6, s14
 ; GCN-IR-NEXT:    s_addc_u32 s11, s7, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], 0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:  .LBB0_3: ; %udiv-do-while

 ; GCN-IR-NEXT:    s_add_u32 s10, s10, 1
 ; GCN-IR-NEXT:    s_addc_u32 s11, s11, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[10:11], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_3
 ; GCN-IR-NEXT:  .LBB0_4: ; %Flow6
 ; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[8:9], 1
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[6:7], s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-IR-NEXT:    s_branch .LBB0_6
-; GCN-IR-NEXT:  .LBB0_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[12:13]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[12:13]
-; GCN-IR-NEXT:  .LBB0_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, s4, v1
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, s4, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v3, s5, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, s4, v0
-; GCN-IR-NEXT:    s_mov_b32 s11, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GCN-IR-NEXT:  .LBB0_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s8
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GCN-IR-NEXT:    s_mov_b32 s12, s0
+; GCN-IR-NEXT:    s_mul_i32 s0, s4, s9
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s5, s8
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s4, s8
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    s_mov_b32 s10, -1
-; GCN-IR-NEXT:    s_mov_b32 s8, s0
-; GCN-IR-NEXT:    s_mov_b32 s9, s1
+; GCN-IR-NEXT:    s_mov_b32 s15, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s14, -1
+; GCN-IR-NEXT:    s_mov_b32 s13, s1
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[12:15], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = urem i64 %x, %y
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define i64 @v_test_srem(i64 %x, i64 %y) {
 ; GCN-LABEL: v_test_srem:

 ; GCN-IR-NEXT:    s_setpc_b64 s[30:31]
   %result = srem i64 %x, %y
   ret i64 %result
 }
 
 define amdgpu_kernel void @s_test_srem23_64(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-LABEL: s_test_srem23_64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-NEXT:    s_mov_b32 s6, -1
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-NEXT:    s_mov_b32 s2, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_ashr_i64 s[4:5], s[4:5], 41
-; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 41
-; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 41
+; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-NEXT:    s_mov_b32 s1, s5
+; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 41
+; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-NEXT:    s_or_b32 s3, s3, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 23
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-NEXT:    s_add_i32 s5, s6, s5
+; GCN-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-NEXT:    s_bfe_i32 s4, s4, 0x170000
+; GCN-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem23_64:
 ; GCN-IR:       ; %bb.0:
-; GCN-IR-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    s_mov_b32 s6, -1
+; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[4:5], 41
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 41
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-IR-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-IR-NEXT:    s_mov_b32 s0, s4
+; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 41
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-IR-NEXT:    s_mov_b32 s1, s5
+; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 41
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-IR-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-IR-NEXT:    s_or_b32 s3, s3, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-IR-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-IR-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-IR-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-IR-NEXT:    s_mov_b32 s4, s0
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 23
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-IR-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-IR-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-IR-NEXT:    s_add_i32 s5, s6, s5
+; GCN-IR-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-IR-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-IR-NEXT:    s_bfe_i32 s4, s4, 0x170000
+; GCN-IR-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 41
   %2 = ashr i64 %y, 41
   %result = srem i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_test_srem24_64(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-LABEL: s_test_srem24_64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-NEXT:    s_mov_b32 s6, -1
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-NEXT:    s_mov_b32 s2, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_ashr_i64 s[4:5], s[4:5], 40
-; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
-; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 40
+; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-NEXT:    s_mov_b32 s1, s5
+; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 40
+; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-NEXT:    s_or_b32 s3, s3, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-NEXT:    s_add_i32 s5, s6, s5
+; GCN-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-NEXT:    s_bfe_i32 s4, s4, 0x180000
+; GCN-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem24_64:
 ; GCN-IR:       ; %bb.0:
-; GCN-IR-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    s_mov_b32 s6, -1
+; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[4:5], 40
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-IR-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-IR-NEXT:    s_mov_b32 s0, s4
+; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 40
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-IR-NEXT:    s_mov_b32 s1, s5
+; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 40
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-IR-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-IR-NEXT:    s_or_b32 s3, s3, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-IR-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-IR-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-IR-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-IR-NEXT:    s_mov_b32 s4, s0
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-IR-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-IR-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-IR-NEXT:    s_add_i32 s5, s6, s5
+; GCN-IR-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-IR-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-IR-NEXT:    s_bfe_i32 s4, s4, 0x180000
+; GCN-IR-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 40
   %2 = ashr i64 %y, 40
   %result = srem i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
; GCN-IR-NEXT:    s_setpc_b64 s[30:31]
   %2 = ashr i64 %y, 40
   %result = srem i64 %1, %2
   ret i64 %result
 }
 
 define amdgpu_kernel void @s_test_srem25_64(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-LABEL: s_test_srem25_64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-NEXT:    s_mov_b32 s6, -1
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-NEXT:    s_mov_b32 s2, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_ashr_i64 s[4:5], s[4:5], 39
-; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 39
-; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 39
+; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-NEXT:    s_mov_b32 s1, s5
+; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 39
+; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-NEXT:    s_or_b32 s3, s3, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 25
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-NEXT:    s_add_i32 s5, s6, s5
+; GCN-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-NEXT:    s_bfe_i32 s4, s4, 0x190000
+; GCN-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem25_64:
 ; GCN-IR:       ; %bb.0:
-; GCN-IR-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    s_mov_b32 s6, -1
+; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[4:5], 39
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 39
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-IR-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-IR-NEXT:    s_mov_b32 s0, s4
+; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 39
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-IR-NEXT:    s_mov_b32 s1, s5
+; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 39
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-IR-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-IR-NEXT:    s_or_b32 s3, s3, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-IR-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-IR-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-IR-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-IR-NEXT:    s_mov_b32 s4, s0
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 25
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-IR-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-IR-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-IR-NEXT:    s_add_i32 s5, s6, s5
+; GCN-IR-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-IR-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-IR-NEXT:    s_bfe_i32 s4, s4, 0x190000
+; GCN-IR-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 39
   %2 = ashr i64 %y, 39
   %result = srem i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_test_srem31_64(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-LABEL: s_test_srem31_64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-NEXT:    s_mov_b32 s6, -1
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-NEXT:    s_mov_b32 s2, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_ashr_i64 s[4:5], s[4:5], 33
-; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 33
-; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    s_ashr_i64 s[8:9], s[0:1], 33
+; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-NEXT:    s_mov_b32 s1, s5
+; GCN-NEXT:    s_ashr_i64 s[4:5], s[6:7], 33
+; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-NEXT:    s_or_b32 s3, s3, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 31
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-NEXT:    s_add_i32 s5, s6, s5
+; GCN-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-NEXT:    s_bfe_i32 s4, s4, 0x1f0000
+; GCN-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem31_64:
 ; GCN-IR:       ; %bb.0:
-; GCN-IR-NEXT:    s_load_dword s5, s[0:1], 0xe
-; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    s_mov_b32 s6, -1
+; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-IR-NEXT:    s_load_dword s1, s[0:1], 0xe
+; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[4:5], 33
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 33
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s2
-; GCN-IR-NEXT:    s_xor_b32 s3, s2, s4
+; GCN-IR-NEXT:    s_mov_b32 s0, s4
+; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 33
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
+; GCN-IR-NEXT:    s_mov_b32 s1, s5
+; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[6:7], 33
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s4
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
-; GCN-IR-NEXT:    s_ashr_i32 s3, s3, 30
-; GCN-IR-NEXT:    s_or_b32 s3, s3, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s3
+; GCN-IR-NEXT:    s_xor_b32 s5, s4, s8
+; GCN-IR-NEXT:    s_ashr_i32 s5, s5, 30
+; GCN-IR-NEXT:    s_or_b32 s5, s5, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-IR-NEXT:    s_mov_b32 s4, s0
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 31
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[6:7], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[6:7], exec
+; GCN-IR-NEXT:    s_cselect_b32 s5, s5, 0
+; GCN-IR-NEXT:    v_readfirstlane_b32 s6, v2
+; GCN-IR-NEXT:    s_add_i32 s5, s6, s5
+; GCN-IR-NEXT:    s_mul_i32 s5, s5, s8
+; GCN-IR-NEXT:    s_sub_i32 s4, s4, s5
+; GCN-IR-NEXT:    s_bfe_i32 s4, s4, 0x1f0000
+; GCN-IR-NEXT:    s_ashr_i32 s5, s4, 31
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 33
   %2 = ashr i64 %y, 33
   %result = srem i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 ; 32 known sign bits
 define amdgpu_kernel void @s_test_srem32_64(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-LABEL: s_test_srem32_64:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dword s4, s[0:1], 0xe
+; GCN-NEXT:    s_load_dword s8, s[0:1], 0xe
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s4
+; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v1, s3
-; GCN-NEXT:    s_xor_b32 s2, s3, s4
+; GCN-NEXT:    s_xor_b32 s2, s3, s8
 ; GCN-NEXT:    s_ashr_i32 s2, s2, 30
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-NEXT:    s_or_b32 s2, s2, 1
-; GCN-NEXT:    v_mov_b32_e32 v3, s2
-; GCN-NEXT:    s_mov_b32 s5, s1
 ; GCN-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s4
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-NEXT:    s_cselect_b32 s2, s2, 0
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, s2, v2
+; GCN-NEXT:    v_mul_lo_u32 v0, v0, s8
 ; GCN-NEXT:    s_mov_b32 s4, s0
+; GCN-NEXT:    s_mov_b32 s5, s1
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s3, v0
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem32_64:
 ; GCN-IR:       ; %bb.0:
-; GCN-IR-NEXT:    s_load_dword s4, s[0:1], 0xe
+; GCN-IR-NEXT:    s_load_dword s8, s[0:1], 0xe
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s4
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s8
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v1, s3
-; GCN-IR-NEXT:    s_xor_b32 s2, s3, s4
+; GCN-IR-NEXT:    s_xor_b32 s2, s3, s8
 ; GCN-IR-NEXT:    s_ashr_i32 s2, s2, 30
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v2, v0
 ; GCN-IR-NEXT:    s_or_b32 s2, s2, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s2
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v2, v1, v2
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-IR-NEXT:    v_mad_f32 v1, -v2, v0, v1
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[4:5], |v1|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[4:5], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s2, s2, 0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s2, v2
+; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s8
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
+; GCN-IR-NEXT:    s_mov_b32 s5, s1
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s3, v0
 ; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 32
   %2 = ashr i64 %y, 32
   %result = srem i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out

 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s0, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s1, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_cmp_le_u32_e64 s[2:3], s12, v4
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s13, v5
 ; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s12, v4
 ; 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, s15
-; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GCN-NEXT:    v_mov_b32_e32 v4, s15
+; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v4, v1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s13, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s12, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s13, 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 v4, v4, v5, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; 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:    v_xor_b32_e32 v0, s6, v0
 ; GCN-NEXT:    v_xor_b32_e32 v1, s6, v1
 ; GCN-NEXT:    v_mov_b32_e32 v2, s6
 ; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, s6, v0
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem33_64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GCN-IR-NEXT:    s_mov_b32 s13, 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[6:7], 31
 ; GCN-IR-NEXT:    s_ashr_i64 s[8:9], s[0:1], 31
 ; GCN-IR-NEXT:    s_ashr_i32 s0, s7, 31
 ; GCN-IR-NEXT:    s_ashr_i32 s6, s1, 31
 ; GCN-IR-NEXT:    s_mov_b32 s1, s0
 ; GCN-IR-NEXT:    s_mov_b32 s7, s6
 ; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[2:3], s[0:1]
 ; GCN-IR-NEXT:    s_xor_b64 s[8:9], s[8:9], s[6:7]
 ; GCN-IR-NEXT:    s_sub_u32 s2, s2, s0
 ; GCN-IR-NEXT:    s_subb_u32 s3, s3, s0
 ; GCN-IR-NEXT:    s_sub_u32 s8, s8, s6
 ; GCN-IR-NEXT:    s_subb_u32 s9, s9, s6
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[8:9], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[8:9], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[6:7], s[10:11]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s8
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s9
+; GCN-IR-NEXT:    s_min_u32 s12, s6, s7
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_min_u32 s16, s6, s7
+; GCN-IR-NEXT:    s_sub_u32 s14, s12, s16
+; GCN-IR-NEXT:    s_subb_u32 s15, 0, 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[18:19], s[14:15], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[14:15], 63
+; GCN-IR-NEXT:    s_or_b64 s[18:19], s[10:11], s[18:19]
+; GCN-IR-NEXT:    s_and_b64 s[10:11], s[18:19], exec
+; GCN-IR-NEXT:    s_cselect_b32 s11, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s10, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[18:19], s[18:19], s[20:21]
 ; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
-; GCN-IR-NEXT:    s_or_b64 s[14:15], s[10:11], s[12:13]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s8
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s9
-; GCN-IR-NEXT:    s_min_u32 s12, s10, s11
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s2
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s3
-; GCN-IR-NEXT:    s_min_u32 s16, s10, s11
-; GCN-IR-NEXT:    s_sub_u32 s10, s12, s16
-; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[18:19], s[10:11], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    s_or_b64 s[14:15], s[14:15], s[18:19]
-; GCN-IR-NEXT:    s_or_b64 s[18:19], s[14:15], s[20:21]
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[18:19]
-; GCN-IR-NEXT:    s_mov_b32 s13, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s14, s10, 1
-; GCN-IR-NEXT:    s_addc_u32 s15, s11, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[14:15], 0
-; GCN-IR-NEXT:    s_sub_i32 s10, 63, s10
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[18:19]
-; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[2:3], s10
+; GCN-IR-NEXT:    s_add_u32 s18, s14, 1
+; GCN-IR-NEXT:    s_addc_u32 s19, s15, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[18:19], 0
+; GCN-IR-NEXT:    s_sub_i32 s14, 63, s14
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[10:11]
+; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[2:3], s14
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[14:15], s[2:3], s14
+; GCN-IR-NEXT:    s_lshr_b64 s[14:15], s[2:3], s18
 ; GCN-IR-NEXT:    s_add_u32 s18, s8, -1
 ; GCN-IR-NEXT:    s_addc_u32 s19, s9, -1
 ; GCN-IR-NEXT:    s_not_b64 s[6:7], s[12:13]
 ; GCN-IR-NEXT:    s_add_u32 s12, s6, s16
 ; GCN-IR-NEXT:    s_addc_u32 s13, s7, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[16:17], 0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:  .LBB8_3: ; %udiv-do-while

 ; GCN-IR-NEXT:    s_add_u32 s12, s12, 1
 ; GCN-IR-NEXT:    s_addc_u32 s13, s13, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[16:17], s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[20:21]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_3
 ; GCN-IR-NEXT:  .LBB8_4: ; %Flow6
 ; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[10:11], 1
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[6:7], s[10:11]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-IR-NEXT:    s_branch .LBB8_6
-; GCN-IR-NEXT:  .LBB8_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[14:15]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[14:15]
-; GCN-IR-NEXT:  .LBB8_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, s8, v1
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, s8, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v3, s9, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, s8, v0
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[6:7], s[10:11]
+; GCN-IR-NEXT:  .LBB8_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s8, v0
+; GCN-IR-NEXT:    s_mul_i32 s11, s8, s11
+; GCN-IR-NEXT:    s_mul_i32 s9, s9, s10
+; GCN-IR-NEXT:    s_mul_i32 s8, s8, s10
+; GCN-IR-NEXT:    v_readfirstlane_b32 s12, v0
+; GCN-IR-NEXT:    s_add_i32 s11, s12, s11
+; GCN-IR-NEXT:    s_add_i32 s11, s11, s9
+; GCN-IR-NEXT:    s_sub_u32 s2, s2, s8
+; GCN-IR-NEXT:    s_subb_u32 s3, s3, s11
+; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[2:3], s[0:1]
+; GCN-IR-NEXT:    s_sub_u32 s0, s2, s0
+; GCN-IR-NEXT:    s_subb_u32 s1, s3, s1
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
-; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
-; GCN-IR-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GCN-IR-NEXT:    v_xor_b32_e32 v1, s1, v1
-; GCN-IR-NEXT:    v_mov_b32_e32 v2, s1
-; GCN-IR-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i64 %x, 31
   %2 = ashr i64 %y, 31
   %result = srem i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }

 ; GCN-NEXT:    v_or_b32_e32 v5, 1, v5
 ; GCN-NEXT:    v_mul_f32_e32 v4, v3, v4
 ; GCN-NEXT:    v_trunc_f32_e32 v4, v4
 ; GCN-NEXT:    v_mad_f32 v3, -v4, v1, v3
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v4, v4
 ; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v1|
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v5, vcc
 ; GCN-NEXT:    s_mov_b32 s0, s4
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v4, v1
 ; GCN-NEXT:    v_mul_lo_u32 v0, v1, v0
 ; GCN-NEXT:    s_mov_b32 s1, s5
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, v2, v0
+; GCN-NEXT:    v_subrev_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GCN-NEXT:    buffer_store_short v1, off, s[0:3], 0 offset:4
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem24_48:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases

 ; GCN-IR-NEXT:    s_mov_b32 s1, s0
 ; GCN-IR-NEXT:    s_mov_b32 s13, s12
 ; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[8:9], s[0:1]
 ; GCN-IR-NEXT:    s_xor_b64 s[6:7], s[10:11], s[12:13]
 ; GCN-IR-NEXT:    s_sub_u32 s2, s2, s0
 ; GCN-IR-NEXT:    s_subb_u32 s3, s3, s0
 ; GCN-IR-NEXT:    s_sub_u32 s6, s6, s12
 ; GCN-IR-NEXT:    s_subb_u32 s7, s7, s12
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[6:7], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[6:7], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
+; GCN-IR-NEXT:    s_mov_b32 s13, 0
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[8:9], s[10:11]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s6
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s7
+; GCN-IR-NEXT:    s_min_u32 s12, s8, s9
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
+; GCN-IR-NEXT:    s_min_u32 s16, s8, s9
+; GCN-IR-NEXT:    s_sub_u32 s14, s12, s16
+; GCN-IR-NEXT:    s_subb_u32 s15, 0, 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[18:19], s[14:15], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[14:15], 63
+; GCN-IR-NEXT:    s_or_b64 s[18:19], s[10:11], s[18:19]
+; GCN-IR-NEXT:    s_and_b64 s[10:11], s[18:19], exec
+; GCN-IR-NEXT:    s_cselect_b32 s11, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s10, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[18:19], s[18:19], s[20:21]
 ; GCN-IR-NEXT:    s_mov_b64 s[8:9], 0
-; GCN-IR-NEXT:    s_or_b64 s[14:15], s[10:11], s[12:13]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s6
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s7
-; GCN-IR-NEXT:    s_min_u32 s12, s10, s11
-; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s2
-; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s3
-; GCN-IR-NEXT:    s_min_u32 s16, s10, s11
-; GCN-IR-NEXT:    s_sub_u32 s10, s12, s16
-; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[18:19], s[10:11], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[10:11], 63
-; GCN-IR-NEXT:    s_or_b64 s[14:15], s[14:15], s[18:19]
-; GCN-IR-NEXT:    s_or_b64 s[18:19], s[14:15], s[20:21]
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[18:19]
-; GCN-IR-NEXT:    s_mov_b32 s13, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s14, s10, 1
-; GCN-IR-NEXT:    s_addc_u32 s15, s11, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[14:15], 0
-; GCN-IR-NEXT:    s_sub_i32 s10, 63, s10
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[18:19]
-; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[2:3], s10
+; GCN-IR-NEXT:    s_add_u32 s18, s14, 1
+; GCN-IR-NEXT:    s_addc_u32 s19, s15, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[18:19], 0
+; GCN-IR-NEXT:    s_sub_i32 s14, 63, s14
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[10:11]
+; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[2:3], s14
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[14:15], s[2:3], s14
+; GCN-IR-NEXT:    s_lshr_b64 s[14:15], s[2:3], s18
 ; GCN-IR-NEXT:    s_add_u32 s18, s6, -1
 ; GCN-IR-NEXT:    s_addc_u32 s19, s7, -1
 ; GCN-IR-NEXT:    s_not_b64 s[8:9], s[12:13]
 ; GCN-IR-NEXT:    s_add_u32 s12, s8, s16
 ; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[16:17], 0
 ; GCN-IR-NEXT:    s_mov_b32 s9, 0
 ; GCN-IR-NEXT:  .LBB9_3: ; %udiv-do-while

 ; GCN-IR-NEXT:    s_add_u32 s12, s12, 1
 ; GCN-IR-NEXT:    s_addc_u32 s13, s13, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[20:21], s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[16:17], s[8:9]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[20:21]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB9_3
 ; GCN-IR-NEXT:  .LBB9_4: ; %Flow3
 ; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[10:11], 1
-; GCN-IR-NEXT:    s_or_b64 s[8:9], s[8:9], s[10:11]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s8
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s9
-; GCN-IR-NEXT:    s_branch .LBB9_6
-; GCN-IR-NEXT:  .LBB9_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[14:15]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[14:15]
-; GCN-IR-NEXT:  .LBB9_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, s6, v1
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, s6, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v3, s7, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, s6, v0
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[8:9], s[10:11]
+; GCN-IR-NEXT:  .LBB9_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s6, v0
+; GCN-IR-NEXT:    s_mul_i32 s8, s6, s11
+; GCN-IR-NEXT:    s_mul_i32 s7, s7, s10
+; GCN-IR-NEXT:    s_mul_i32 s6, s6, s10
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s8, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s7, v0
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s6
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
-; GCN-IR-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GCN-IR-NEXT:    v_xor_b32_e32 v1, s1, v1
+; GCN-IR-NEXT:    v_sub_i32_e32 v1, vcc, s2, v1
+; GCN-IR-NEXT:    v_subb_u32_e32 v0, vcc, v2, v0, vcc
+; GCN-IR-NEXT:    v_xor_b32_e32 v1, s0, v1
+; GCN-IR-NEXT:    v_xor_b32_e32 v0, s1, v0
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s1
-; GCN-IR-NEXT:    v_subrev_i32_e32 v0, vcc, s0, v0
-; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v1, v2, vcc
+; GCN-IR-NEXT:    v_subrev_i32_e32 v1, vcc, s0, v1
+; GCN-IR-NEXT:    v_subb_u32_e32 v0, vcc, v0, v2, vcc
+; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    buffer_store_short v1, off, s[4:7], 0 offset:4
-; GCN-IR-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GCN-IR-NEXT:    buffer_store_short v0, off, s[4:7], 0 offset:4
+; GCN-IR-NEXT:    buffer_store_dword v1, off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = ashr i48 %x, 24
   %2 = ashr i48 %y, 24
   %result = srem i48 %1, %2
   store i48 %result, i48 addrspace(1)* %out
   ret void
 }
 

 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s2, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s2, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s2, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s9, v5
 ; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s8, v4
 ; 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_subb_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
+; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
-; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s9, 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 v4, v4, v5, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
 ; 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:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem_k_num_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_ashr_i32 s6, s3, 31
-; GCN-IR-NEXT:    s_mov_b32 s7, s6
-; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[2:3], s[6:7]
-; GCN-IR-NEXT:    s_sub_u32 s4, s2, s6
-; GCN-IR-NEXT:    s_subb_u32 s5, s3, s6
-; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s4
-; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s5
-; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
-; GCN-IR-NEXT:    s_add_u32 s6, s8, 0xffffffc5
-; GCN-IR-NEXT:    s_addc_u32 s7, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[4:5], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[2:3], s[10:11]
-; GCN-IR-NEXT:    s_or_b64 s[2:3], s[10:11], s[12:13]
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[2:3]
-; GCN-IR-NEXT:    s_mov_b64 s[2:3], 0
+; GCN-IR-NEXT:    s_ashr_i32 s8, s3, 31
+; GCN-IR-NEXT:    s_mov_b32 s9, s8
+; GCN-IR-NEXT:    s_xor_b64 s[2:3], s[2:3], s[8:9]
+; GCN-IR-NEXT:    s_sub_u32 s4, s2, s8
+; GCN-IR-NEXT:    s_subb_u32 s5, s3, s8
+; GCN-IR-NEXT:    s_flbit_i32_b32 s2, s4
+; GCN-IR-NEXT:    s_add_i32 s2, s2, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s3, s5
+; GCN-IR-NEXT:    s_min_u32 s8, s2, s3
+; GCN-IR-NEXT:    s_add_u32 s2, s8, 0xffffffc5
+; GCN-IR-NEXT:    s_addc_u32 s3, 0, -1
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[4:5], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[2:3], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[2:3], 63
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[10:11], s[12:13]
+; GCN-IR-NEXT:    s_and_b64 s[10:11], s[12:13], exec
+; GCN-IR-NEXT:    s_cselect_b32 s10, 0, 24
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[14:15]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
+; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
-; GCN-IR-NEXT:    s_addc_u32 s11, s7, 0
+; GCN-IR-NEXT:    s_add_u32 s10, s2, 1
+; GCN-IR-NEXT:    s_addc_u32 s11, s3, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[10:11], 0
-; GCN-IR-NEXT:    s_sub_i32 s6, 63, s6
+; GCN-IR-NEXT:    s_sub_i32 s2, 63, s2
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], 24, s6
+; GCN-IR-NEXT:    s_lshl_b64 s[2:3], 24, s2
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    s_lshr_b64 s[10:11], 24, s10
 ; GCN-IR-NEXT:    s_add_u32 s14, s4, -1
 ; GCN-IR-NEXT:    s_addc_u32 s15, s5, -1
 ; GCN-IR-NEXT:    s_sub_u32 s8, 58, s8
 ; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], 0
-; GCN-IR-NEXT:    s_mov_b32 s3, 0
+; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:  .LBB10_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[10:11], 1
-; GCN-IR-NEXT:    s_lshr_b32 s2, s7, 31
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[2:3]
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[12:13], s[6:7]
-; GCN-IR-NEXT:    s_sub_u32 s2, s14, s10
-; GCN-IR-NEXT:    s_subb_u32 s2, s15, s11
-; GCN-IR-NEXT:    s_ashr_i32 s12, s2, 31
+; GCN-IR-NEXT:    s_lshr_b32 s6, s3, 31
+; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[2:3], 1
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[10:11], s[6:7]
+; GCN-IR-NEXT:    s_or_b64 s[2:3], s[12:13], s[2:3]
+; GCN-IR-NEXT:    s_sub_u32 s6, s14, s10
+; GCN-IR-NEXT:    s_subb_u32 s6, s15, s11
+; GCN-IR-NEXT:    s_ashr_i32 s12, s6, 31
 ; GCN-IR-NEXT:    s_mov_b32 s13, s12
-; GCN-IR-NEXT:    s_and_b32 s2, s12, 1
+; GCN-IR-NEXT:    s_and_b32 s6, s12, 1
 ; GCN-IR-NEXT:    s_and_b64 s[12:13], s[12:13], s[4:5]
 ; GCN-IR-NEXT:    s_sub_u32 s10, s10, s12
 ; GCN-IR-NEXT:    s_subb_u32 s11, s11, s13
 ; GCN-IR-NEXT:    s_add_u32 s8, s8, 1
 ; GCN-IR-NEXT:    s_addc_u32 s9, s9, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[8:9], 0
-; GCN-IR-NEXT:    s_mov_b64 s[12:13], s[2:3]
+; GCN-IR-NEXT:    s_mov_b64 s[12:13], s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB10_3
 ; GCN-IR-NEXT:  .LBB10_4: ; %Flow5
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[2:3], s[2:3], s[6:7]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-IR-NEXT:    s_branch .LBB10_6
-; GCN-IR-NEXT:  .LBB10_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, 0
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, 24, 0, s[10:11]
-; GCN-IR-NEXT:  .LBB10_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, s4, v1
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, s4, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v3, s5, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, s4, v0
+; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[2:3], 1
+; GCN-IR-NEXT:    s_or_b64 s[10:11], s[6:7], s[2:3]
+; GCN-IR-NEXT:  .LBB10_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s10
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GCN-IR-NEXT:    s_mul_i32 s6, s4, s11
+; GCN-IR-NEXT:    s_mul_i32 s5, s5, s10
+; GCN-IR-NEXT:    s_mul_i32 s4, s4, s10
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s6, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, s5, v0
+; GCN-IR-NEXT:    v_sub_i32_e64 v0, vcc, 24, s4
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, 24, v0
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = srem i64 24, %x
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
; GCN-IR-NEXT:    s_setpc_b64 s[30:31]
   %result = srem i64 %x, 32768
   ret i64 %result
 }
 
 define amdgpu_kernel void @s_test_srem24_k_num_i64(i64 addrspace(1)* %out, i64 %x) {
 ; GCN-LABEL: s_test_srem24_k_num_i64:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GCN-NEXT:    s_mov_b32 s7, 0xf000
+; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_ashr_i64 s[4:5], s[2:3], 40
-; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-NEXT:    s_mov_b32 s5, 0x41c00000
-; GCN-NEXT:    s_ashr_i32 s6, s4, 30
-; GCN-NEXT:    s_or_b32 s6, s6, 1
+; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
+; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s2
+; GCN-NEXT:    s_mov_b32 s3, 0x41c00000
+; GCN-NEXT:    s_mov_b32 s4, s0
+; GCN-NEXT:    s_ashr_i32 s0, s2, 30
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v1, v0
-; GCN-NEXT:    v_mov_b32_e32 v3, s6
-; GCN-NEXT:    s_mov_b32 s3, 0xf000
-; GCN-NEXT:    s_mov_b32 s2, -1
+; GCN-NEXT:    s_mov_b32 s5, s1
+; GCN-NEXT:    s_or_b32 s8, s0, 1
 ; GCN-NEXT:    v_mul_f32_e32 v1, 0x41c00000, v1
 ; GCN-NEXT:    v_trunc_f32_e32 v1, v1
-; GCN-NEXT:    v_mad_f32 v2, -v1, v0, s5
+; GCN-NEXT:    v_mad_f32 v2, -v1, v0, s3
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v0|
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, 24, v0
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v2|, |v0|
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s8, 0
+; GCN-NEXT:    v_readfirstlane_b32 s1, v1
+; GCN-NEXT:    s_add_i32 s0, s1, s0
+; GCN-NEXT:    s_mul_i32 s0, s0, s2
+; GCN-NEXT:    s_sub_i32 s0, 24, s0
+; GCN-NEXT:    s_bfe_i32 s0, s0, 0x180000
+; GCN-NEXT:    s_ashr_i32 s1, s0, 31
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem24_k_num_i64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s6, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_ashr_i64 s[4:5], s[2:3], 40
-; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s4
-; GCN-IR-NEXT:    s_mov_b32 s5, 0x41c00000
-; GCN-IR-NEXT:    s_ashr_i32 s6, s4, 30
-; GCN-IR-NEXT:    s_or_b32 s6, s6, 1
+; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
+; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s2
+; GCN-IR-NEXT:    s_mov_b32 s3, 0x41c00000
+; GCN-IR-NEXT:    s_mov_b32 s4, s0
+; GCN-IR-NEXT:    s_ashr_i32 s0, s2, 30
 ; GCN-IR-NEXT:    v_rcp_iflag_f32_e32 v1, v0
-; GCN-IR-NEXT:    v_mov_b32_e32 v3, s6
-; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
-; GCN-IR-NEXT:    s_mov_b32 s2, -1
+; GCN-IR-NEXT:    s_mov_b32 s5, s1
+; GCN-IR-NEXT:    s_or_b32 s8, s0, 1
 ; GCN-IR-NEXT:    v_mul_f32_e32 v1, 0x41c00000, v1
 ; GCN-IR-NEXT:    v_trunc_f32_e32 v1, v1
-; GCN-IR-NEXT:    v_mad_f32 v2, -v1, v0, s5
+; GCN-IR-NEXT:    v_mad_f32 v2, -v1, v0, s3
 ; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v1, v1
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v2|, |v0|
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s4
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, 24, v0
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v2|, |v0|
+; GCN-IR-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-IR-NEXT:    s_cselect_b32 s0, s8, 0
+; GCN-IR-NEXT:    v_readfirstlane_b32 s1, v1
+; GCN-IR-NEXT:    s_add_i32 s0, s1, s0
+; GCN-IR-NEXT:    s_mul_i32 s0, s0, s2
+; GCN-IR-NEXT:    s_sub_i32 s0, 24, s0
+; GCN-IR-NEXT:    s_bfe_i32 s0, s0, 0x180000
+; GCN-IR-NEXT:    s_ashr_i32 s1, s0, 31
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s1
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %x.shr = ashr i64 %x, 40
   %result = srem i64 24, %x.shr
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_test_srem24_k_den_i64(i64 addrspace(1)* %out, i64 %x) {
 ; GCN-LABEL: s_test_srem24_k_den_i64:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s4, 0x46b6fe00
+; GCN-NEXT:    s_mov_b32 s8, 0x46b6fe00
 ; GCN-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v0, s2
-; GCN-NEXT:    s_ashr_i32 s3, s2, 30
-; GCN-NEXT:    s_or_b32 s3, s3, 1
-; GCN-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-NEXT:    v_mul_f32_e32 v2, 0x38331158, v0
-; GCN-NEXT:    v_trunc_f32_e32 v2, v2
-; GCN-NEXT:    v_mad_f32 v0, -v2, s4, v0
-; GCN-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, s4
-; GCN-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
-; GCN-NEXT:    s_movk_i32 s3, 0x5b7f
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GCN-NEXT:    v_mul_lo_u32 v0, v0, s3
 ; GCN-NEXT:    s_mov_b32 s4, s0
+; GCN-NEXT:    s_ashr_i32 s0, s2, 30
 ; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
+; GCN-NEXT:    v_mul_f32_e32 v1, 0x38331158, v0
+; GCN-NEXT:    v_trunc_f32_e32 v1, v1
+; GCN-NEXT:    v_mad_f32 v0, -v1, s8, v0
+; GCN-NEXT:    v_cvt_i32_f32_e32 v1, v1
+; GCN-NEXT:    s_or_b32 s3, s0, 1
+; GCN-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v0|, s8
+; GCN-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-NEXT:    s_cselect_b32 s0, s3, 0
+; GCN-NEXT:    v_readfirstlane_b32 s1, v1
+; GCN-NEXT:    s_add_i32 s0, s1, s0
+; GCN-NEXT:    s_mulk_i32 s0, 0x5b7f
+; GCN-NEXT:    s_sub_i32 s0, s2, s0
+; GCN-NEXT:    s_bfe_i32 s0, s0, 0x180000
+; GCN-NEXT:    s_ashr_i32 s1, s0, 31
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_srem24_k_den_i64:
 ; GCN-IR:       ; %bb.0:
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b32 s4, 0x46b6fe00
+; GCN-IR-NEXT:    s_mov_b32 s8, 0x46b6fe00
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_ashr_i64 s[2:3], s[2:3], 40
 ; GCN-IR-NEXT:    v_cvt_f32_i32_e32 v0, s2
-; GCN-IR-NEXT:    s_ashr_i32 s3, s2, 30
-; GCN-IR-NEXT:    s_or_b32 s3, s3, 1
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-IR-NEXT:    v_mul_f32_e32 v2, 0x38331158, v0
-; GCN-IR-NEXT:    v_trunc_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_mad_f32 v0, -v2, s4, v0
-; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v2, v2
-; GCN-IR-NEXT:    v_cmp_ge_f32_e64 vcc, |v0|, s4
-; GCN-IR-NEXT:    v_cndmask_b32_e32 v0, 0, v1, vcc
-; GCN-IR-NEXT:    s_movk_i32 s3, 0x5b7f
-; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s3
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
+; GCN-IR-NEXT:    s_ashr_i32 s0, s2, 30
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    v_bfe_i32 v0, v0, 0, 24
-; GCN-IR-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
+; GCN-IR-NEXT:    v_mul_f32_e32 v1, 0x38331158, v0
+; GCN-IR-NEXT:    v_trunc_f32_e32 v1, v1
+; GCN-IR-NEXT:    v_mad_f32 v0, -v1, s8, v0
+; GCN-IR-NEXT:    v_cvt_i32_f32_e32 v1, v1
+; GCN-IR-NEXT:    s_or_b32 s3, s0, 1
+; GCN-IR-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v0|, s8
+; GCN-IR-NEXT:    s_and_b64 s[0:1], s[0:1], exec
+; GCN-IR-NEXT:    s_cselect_b32 s0, s3, 0
+; GCN-IR-NEXT:    v_readfirstlane_b32 s1, v1
+; GCN-IR-NEXT:    s_add_i32 s0, s1, s0
+; GCN-IR-NEXT:    s_mulk_i32 s0, 0x5b7f
+; GCN-IR-NEXT:    s_sub_i32 s0, s2, s0
+; GCN-IR-NEXT:    s_bfe_i32 s0, s0, 0x180000
+; GCN-IR-NEXT:    s_ashr_i32 s1, s0, 31
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %x.shr = ashr i64 %x, 40
   %result = srem i64 %x.shr, 23423
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 

llvm/test/CodeGen/AMDGPU/trunc.ll

define amdgpu_kernel void @sgpr_trunc_i32_to_i1(i32 addrspace(1)* %out, i32 %a) {
   store i32 %result, i32 addrspace(1)* %out, align 4
   ret void
 }
 
 ; GCN-LABEL: {{^}}s_trunc_i64_to_i1:
 ; SI: s_load_dwordx2 s[[[SLO:[0-9]+]]:{{[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0x13
 ; VI: s_load_dwordx2 s[[[SLO:[0-9]+]]:{{[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0x4c
 ; GCN: s_bitcmp1_b32 s[[SLO]], 0
-; SI: s_cselect_b64 s[[[VLO:[0-9]+]]:[[VHI:[0-9]+]]], -1, 0
-; SI: v_cndmask_b32_e64 {{v[0-9]+}}, -12, 63, s[[[VLO]]:[[VHI]]]
-; VI: s_cselect_b32 {{s[0-9]+}}, 63, -12
+; GCN: s_cselect_b32 {{s[0-9]+}}, 63, -12
 define amdgpu_kernel void @s_trunc_i64_to_i1(i32 addrspace(1)* %out, [8 x i32], i64 %x) {
   %trunc = trunc i64 %x to i1
   %sel = select i1 %trunc, i32 63, i32 -12
   store i32 %sel, i32 addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}v_trunc_i64_to_i1:

llvm/test/CodeGen/AMDGPU/udiv.ll

 ; SI-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
 ; SI-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; SI-NEXT:    v_mul_lo_u32 v3, v3, v2
 ; SI-NEXT:    v_mul_hi_u32 v3, v2, v3
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; SI-NEXT:    v_mul_hi_u32 v2, v0, v2
 ; SI-NEXT:    v_mul_lo_u32 v3, v2, v1
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, 1, v2
-; SI-NEXT:    v_subrev_i32_e32 v0, vcc, v3, v0
-; SI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v1
-; SI-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; SI-NEXT:    v_sub_i32_e32 v0, vcc, v0, v3
 ; SI-NEXT:    v_subrev_i32_e32 v3, vcc, v1, v0
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
+; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; SI-NEXT:    v_add_i32_e32 v3, vcc, 1, v2
 ; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
 ; SI-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
 ; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: udiv_i32:
 ; VI:       ; %bb.0:

 ; VI-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
 ; VI-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; VI-NEXT:    v_mul_lo_u32 v3, v3, v2
 ; VI-NEXT:    v_mul_hi_u32 v3, v2, v3
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
 ; VI-NEXT:    v_mul_hi_u32 v2, v0, v2
 ; VI-NEXT:    v_mul_lo_u32 v3, v2, v1
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 1, v2
-; VI-NEXT:    v_subrev_u32_e32 v0, vcc, v3, v0
-; VI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v1
-; VI-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
+; VI-NEXT:    v_sub_u32_e32 v0, vcc, v0, v3
 ; VI-NEXT:    v_subrev_u32_e32 v3, vcc, v1, v0
-; VI-NEXT:    v_cndmask_b32_e64 v0, v0, v3, s[0:1]
+; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
+; VI-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, 1, v2
 ; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
 ; VI-NEXT:    v_cndmask_b32_e32 v0, v2, v3, vcc
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GCN-LABEL: udiv_i32:
 ; GCN:       ; %bb.0:

 ; GCN-NEXT:    v_mul_lo_u32 v3, v3, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v2, v3
 ; GCN-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v0, v2
 ; GCN-NEXT:    v_mov_b32_e32 v2, s0
 ; GCN-NEXT:    v_mov_b32_e32 v3, s1
 ; GCN-NEXT:    v_mul_lo_u32 v5, v4, v1
 ; GCN-NEXT:    v_add_u32_e32 v6, vcc, 1, v4
-; GCN-NEXT:    v_subrev_u32_e32 v0, vcc, v5, v0
-; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v6, s[0:1]
+; GCN-NEXT:    v_sub_u32_e32 v0, vcc, v0, v5
 ; GCN-NEXT:    v_subrev_u32_e32 v5, vcc, v1, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v5, s[0:1]
+; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
+; GCN-NEXT:    v_cndmask_b32_e32 v4, v4, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
 ; GCN-NEXT:    v_add_u32_e32 v5, vcc, 1, v4
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v1
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v4, v5, vcc
 ; GCN-NEXT:    flat_store_dword v[2:3], v0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GFX1030-LABEL: udiv_i32:
 ; GFX1030:       ; %bb.0:
 ; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX1030-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-NEXT:    global_load_dwordx2 v[0:1], v2, s[2:3]
 ; GFX1030-NEXT:    s_waitcnt vmcnt(0)
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v3, v1
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v4, 0, v1
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX1030-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX1030-NEXT:    v_mul_lo_u32 v4, v4, v3
-; GFX1030-NEXT:    v_mul_hi_u32 v4, v3, v4
-; GFX1030-NEXT:    v_add_nc_u32_e32 v3, v3, v4
-; GFX1030-NEXT:    v_mul_hi_u32 v3, v0, v3
-; GFX1030-NEXT:    v_mul_lo_u32 v4, v3, v1
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v0, v0, v4
-; GFX1030-NEXT:    v_add_nc_u32_e32 v4, 1, v3
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v5, v0, v1
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v0, v1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc_lo
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc_lo
-; GFX1030-NEXT:    v_add_nc_u32_e32 v4, 1, v3
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v0, v1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v3, v4, vcc_lo
+; GFX1030-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v1, s2
+; GFX1030-NEXT:    s_sub_i32 s4, 0, s2
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX1030-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s3, v1
+; GFX1030-NEXT:    s_mul_i32 s4, s4, s3
+; GFX1030-NEXT:    s_mul_hi_u32 s4, s3, s4
+; GFX1030-NEXT:    s_add_i32 s3, s3, s4
+; GFX1030-NEXT:    s_mul_hi_u32 s3, s5, s3
+; GFX1030-NEXT:    s_mul_i32 s4, s3, s2
+; GFX1030-NEXT:    s_sub_i32 s4, s5, s4
+; GFX1030-NEXT:    s_add_i32 s5, s3, 1
+; GFX1030-NEXT:    s_sub_i32 s6, s4, s2
+; GFX1030-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX1030-NEXT:    s_cselect_b32 s3, s5, s3
+; GFX1030-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX1030-NEXT:    s_add_i32 s5, s3, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX1030-NEXT:    s_cselect_b32 s2, s5, s3
+; GFX1030-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX1030-NEXT:    global_store_dword v2, v0, s[0:1]
 ; GFX1030-NEXT:    s_endpgm
 ;
 ; EG-LABEL: udiv_i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 17, @9, KC0[CB0:0-32], KC1[]

 ; SI-NEXT:    s_sub_i32 s4, 0, s3
 ; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; SI-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; SI-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; SI-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    v_mul_hi_u32 v1, v0, v1
-; SI-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; SI-NEXT:    v_mul_hi_u32 v0, s2, v0
-; SI-NEXT:    v_mul_lo_u32 v1, v0, s3
-; SI-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; SI-NEXT:    v_sub_i32_e32 v1, vcc, s2, v1
-; SI-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v1
-; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; SI-NEXT:    v_subrev_i32_e32 v2, vcc, s3, v1
-; SI-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; SI-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; SI-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; SI-NEXT:    v_readfirstlane_b32 s0, v0
+; SI-NEXT:    s_mul_i32 s0, s0, s3
+; SI-NEXT:    s_sub_i32 s0, s2, s0
+; SI-NEXT:    s_sub_i32 s1, s0, s3
+; SI-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; SI-NEXT:    s_cmp_ge_u32 s0, s3
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; SI-NEXT:    s_cselect_b32 s0, s1, s0
+; SI-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; SI-NEXT:    s_cmp_ge_u32 s0, s3
+; SI-NEXT:    s_cselect_b64 vcc, -1, 0
+; SI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
 ; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: s_udiv_i32:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x24
 ; VI-NEXT:    s_mov_b32 s7, 0xf000
 ; VI-NEXT:    s_mov_b32 s6, -1
 ; VI-NEXT:    s_waitcnt lgkmcnt(0)
 ; VI-NEXT:    v_cvt_f32_u32_e32 v0, s3
 ; VI-NEXT:    s_sub_i32 s4, 0, s3
 ; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; VI-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; VI-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; VI-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; VI-NEXT:    s_mov_b32 s4, s0
 ; VI-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; VI-NEXT:    v_mul_hi_u32 v0, s2, v0
-; VI-NEXT:    v_mul_lo_u32 v1, v0, s3
-; VI-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
-; VI-NEXT:    v_sub_u32_e32 v1, vcc, s2, v1
-; VI-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v1
-; VI-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; VI-NEXT:    v_subrev_u32_e32 v2, vcc, s3, v1
-; VI-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; VI-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
-; VI-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; VI-NEXT:    v_readfirstlane_b32 s0, v0
+; VI-NEXT:    s_mul_i32 s0, s0, s3
+; VI-NEXT:    s_sub_i32 s0, s2, s0
+; VI-NEXT:    s_sub_i32 s1, s0, s3
+; VI-NEXT:    v_add_u32_e32 v1, vcc, 1, v0
+; VI-NEXT:    s_cmp_ge_u32 s0, s3
+; VI-NEXT:    s_cselect_b64 vcc, -1, 0
+; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; VI-NEXT:    s_cselect_b32 s0, s1, s0
+; VI-NEXT:    v_add_u32_e32 v1, vcc, 1, v0
+; VI-NEXT:    s_cmp_ge_u32 s0, s3
+; VI-NEXT:    s_cselect_b64 vcc, -1, 0
+; VI-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GCN-LABEL: s_udiv_i32:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GCN-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    v_cvt_f32_u32_e32 v0, s7
-; GCN-NEXT:    s_sub_i32 s0, 0, s7
+; GCN-NEXT:    v_cvt_f32_u32_e32 v0, s3
+; GCN-NEXT:    s_sub_i32 s4, 0, s3
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GCN-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GCN-NEXT:    v_mul_lo_u32 v1, s0, v0
+; GCN-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; GCN-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GCN-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
-; GCN-NEXT:    v_mul_hi_u32 v0, s6, v0
-; GCN-NEXT:    v_mul_lo_u32 v1, v0, s7
-; GCN-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
-; GCN-NEXT:    v_sub_u32_e32 v1, vcc, s6, v1
-; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s7, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GCN-NEXT:    v_subrev_u32_e32 v2, vcc, s7, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; GCN-NEXT:    v_add_u32_e32 v2, vcc, 1, v0
-; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v0, v2, vcc
-; GCN-NEXT:    v_mov_b32_e32 v0, s4
-; GCN-NEXT:    v_mov_b32_e32 v1, s5
+; GCN-NEXT:    v_mul_hi_u32 v0, s2, v0
+; GCN-NEXT:    v_readfirstlane_b32 s4, v0
+; GCN-NEXT:    s_mul_i32 s4, s4, s3
+; GCN-NEXT:    s_sub_i32 s2, s2, s4
+; GCN-NEXT:    s_sub_i32 s4, s2, s3
+; GCN-NEXT:    v_add_u32_e32 v1, vcc, 1, v0
+; GCN-NEXT:    s_cmp_ge_u32 s2, s3
+; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GCN-NEXT:    s_cselect_b32 s2, s4, s2
+; GCN-NEXT:    v_add_u32_e32 v1, vcc, 1, v0
+; GCN-NEXT:    s_cmp_ge_u32 s2, s3
+; GCN-NEXT:    s_cselect_b64 vcc, -1, 0
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v0, v1, vcc
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-NEXT:    flat_store_dword v[0:1], v2
 ; GCN-NEXT:    s_endpgm
 ;
 ; GFX1030-LABEL: s_udiv_i32:
 ; GFX1030:       ; %bb.0:
 ; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
-; GFX1030-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-NEXT:    v_cvt_f32_u32_e32 v0, s3
 ; GFX1030-NEXT:    s_sub_i32 s5, 0, s3
 ; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX1030-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX1030-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX1030-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX1030-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1030-NEXT:    s_mul_i32 s5, s5, s4
 ; GFX1030-NEXT:    s_mul_hi_u32 s5, s4, s5
 ; GFX1030-NEXT:    s_add_i32 s4, s4, s5
 ; GFX1030-NEXT:    s_mul_hi_u32 s4, s2, s4
 ; GFX1030-NEXT:    s_mul_i32 s5, s4, s3
 ; GFX1030-NEXT:    s_sub_i32 s2, s2, s5
+; GFX1030-NEXT:    s_add_i32 s5, s4, 1
+; GFX1030-NEXT:    s_sub_i32 s6, s2, s3
 ; GFX1030-NEXT:    s_cmp_ge_u32 s2, s3
-; GFX1030-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX1030-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX1030-NEXT:    s_cselect_b32 s2, s6, s2
 ; GFX1030-NEXT:    s_add_i32 s5, s4, 1
-; GFX1030-NEXT:    v_mov_b32_e32 v0, s5
-; GFX1030-NEXT:    s_sub_i32 s5, s2, s3
-; GFX1030-NEXT:    v_mov_b32_e32 v1, s5
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, s4, v0, vcc_lo
-; GFX1030-NEXT:    v_cndmask_b32_e32 v1, s2, v1, vcc_lo
-; GFX1030-NEXT:    v_add_nc_u32_e32 v2, 1, v0
-; GFX1030-NEXT:    v_cmp_le_u32_e32 vcc_lo, s3, v1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc_lo
-; GFX1030-NEXT:    global_store_dword v3, v0, s[0:1]
+; GFX1030-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX1030-NEXT:    s_cselect_b32 s2, s5, s4
+; GFX1030-NEXT:    v_mov_b32_e32 v1, s2
+; GFX1030-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX1030-NEXT:    s_endpgm
 ;
 ; EG-LABEL: s_udiv_i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 17, @4, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T0.X, T1.X, 1
 ; EG-NEXT:    CF_END
 ; EG-NEXT:    PAD

 ; SI-NEXT:    v_mul_f32_e32 v5, 0x4f7ffffe, v5
 ; SI-NEXT:    v_cvt_u32_f32_e32 v4, v4
 ; SI-NEXT:    v_cvt_u32_f32_e32 v5, v5
 ; SI-NEXT:    v_mul_lo_u32 v6, v6, v4
 ; SI-NEXT:    v_mul_lo_u32 v7, v7, v5
 ; SI-NEXT:    v_mul_hi_u32 v6, v4, v6
 ; SI-NEXT:    v_mul_hi_u32 v7, v5, v7
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
-; SI-NEXT:    v_add_i32_e32 v5, vcc, v7, v5
+; SI-NEXT:    v_add_i32_e32 v5, vcc, v5, v7
 ; SI-NEXT:    v_mul_hi_u32 v4, v0, v4
 ; SI-NEXT:    v_mul_hi_u32 v5, v1, v5
 ; SI-NEXT:    v_mul_lo_u32 v6, v4, v2
 ; SI-NEXT:    v_mul_lo_u32 v8, v5, v3
 ; SI-NEXT:    v_add_i32_e32 v7, vcc, 1, v4
-; SI-NEXT:    v_subrev_i32_e32 v0, vcc, v6, v0
+; SI-NEXT:    v_sub_i32_e32 v0, vcc, v0, v6
 ; SI-NEXT:    v_sub_i32_e32 v1, vcc, v1, v8
 ; SI-NEXT:    v_add_i32_e32 v9, vcc, 1, v5
 ; SI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v2
 ; SI-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v3
-; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
 ; SI-NEXT:    v_subrev_i32_e32 v6, vcc, v2, v0
-; SI-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[2:3]
+; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
 ; SI-NEXT:    v_subrev_i32_e32 v7, vcc, v3, v1
+; SI-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[2:3]
 ; SI-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[0:1]
 ; SI-NEXT:    v_add_i32_e32 v6, vcc, 1, v4
 ; SI-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[2:3]
 ; SI-NEXT:    v_add_i32_e32 v7, vcc, 1, v5
 ; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
 ; SI-NEXT:    v_cndmask_b32_e32 v0, v4, v6, vcc
 ; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v3
 ; SI-NEXT:    v_cndmask_b32_e32 v1, v5, v7, vcc

 ; VI-NEXT:    v_mul_f32_e32 v5, 0x4f7ffffe, v5
 ; VI-NEXT:    v_cvt_u32_f32_e32 v4, v4
 ; VI-NEXT:    v_cvt_u32_f32_e32 v5, v5
 ; VI-NEXT:    v_mul_lo_u32 v6, v6, v4
 ; VI-NEXT:    v_mul_lo_u32 v7, v7, v5
 ; VI-NEXT:    v_mul_hi_u32 v6, v4, v6
 ; VI-NEXT:    v_mul_hi_u32 v7, v5, v7
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, v6, v4
-; VI-NEXT:    v_add_u32_e32 v5, vcc, v7, v5
+; VI-NEXT:    v_add_u32_e32 v5, vcc, v5, v7
 ; VI-NEXT:    v_mul_hi_u32 v4, v0, v4
 ; VI-NEXT:    v_mul_hi_u32 v5, v1, v5
 ; VI-NEXT:    v_mul_lo_u32 v6, v4, v2
 ; VI-NEXT:    v_mul_lo_u32 v8, v5, v3
 ; VI-NEXT:    v_add_u32_e32 v7, vcc, 1, v4
 ; VI-NEXT:    v_sub_u32_e32 v0, vcc, v0, v6
-; VI-NEXT:    v_subrev_u32_e32 v1, vcc, v8, v1
+; VI-NEXT:    v_sub_u32_e32 v1, vcc, v1, v8
 ; VI-NEXT:    v_add_u32_e32 v9, vcc, 1, v5
 ; VI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v2
 ; VI-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v3
+; VI-NEXT:    v_subrev_u32_e32 v6, vcc, v2, v0
 ; VI-NEXT:    v_cndmask_b32_e64 v4, v4, v7, s[0:1]
-; VI-NEXT:    v_sub_u32_e32 v6, vcc, v0, v2
-; VI-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[2:3]
 ; VI-NEXT:    v_subrev_u32_e32 v7, vcc, v3, v1
+; VI-NEXT:    v_cndmask_b32_e64 v5, v5, v9, s[2:3]
 ; VI-NEXT:    v_cndmask_b32_e64 v0, v0, v6, s[0:1]
 ; VI-NEXT:    v_add_u32_e32 v6, vcc, 1, v4
 ; VI-NEXT:    v_cndmask_b32_e64 v1, v1, v7, s[2:3]
 ; VI-NEXT:    v_add_u32_e32 v7, vcc, 1, v5
 ; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
 ; VI-NEXT:    v_cndmask_b32_e32 v0, v4, v6, vcc
 ; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v3
 ; VI-NEXT:    v_cndmask_b32_e32 v1, v5, v7, vcc

 ; GCN-NEXT:    v_sub_u32_e32 v4, vcc, 0, v3
 ; GCN-NEXT:    v_mul_lo_u32 v8, v4, v7
 ; GCN-NEXT:    v_mul_hi_u32 v9, v6, v5
 ; GCN-NEXT:    v_mov_b32_e32 v4, s0
 ; GCN-NEXT:    v_mov_b32_e32 v5, s1
 ; GCN-NEXT:    v_mul_hi_u32 v8, v7, v8
 ; GCN-NEXT:    v_add_u32_e32 v6, vcc, v9, v6
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v6
-; GCN-NEXT:    v_add_u32_e32 v7, vcc, v8, v7
+; GCN-NEXT:    v_add_u32_e32 v7, vcc, v7, v8
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v7
 ; GCN-NEXT:    v_mul_lo_u32 v8, v6, v2
 ; GCN-NEXT:    v_add_u32_e32 v9, vcc, 1, v6
 ; GCN-NEXT:    v_mul_lo_u32 v10, v7, v3
 ; GCN-NEXT:    v_sub_u32_e32 v0, vcc, v0, v8
 ; GCN-NEXT:    v_add_u32_e32 v11, vcc, 1, v7
-; GCN-NEXT:    v_subrev_u32_e32 v1, vcc, v10, v1
+; GCN-NEXT:    v_sub_u32_e32 v1, vcc, v1, v10
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v2
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[2:3], v1, v3
+; GCN-NEXT:    v_subrev_u32_e32 v8, vcc, v2, v0
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, v6, v9, s[0:1]
-; GCN-NEXT:    v_sub_u32_e32 v8, vcc, v0, v2
-; GCN-NEXT:    v_cndmask_b32_e64 v7, v7, v11, s[2:3]
 ; GCN-NEXT:    v_subrev_u32_e32 v9, vcc, v3, v1
+; GCN-NEXT:    v_cndmask_b32_e64 v7, v7, v11, s[2:3]
 ; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v8, s[0:1]
 ; GCN-NEXT:    v_add_u32_e32 v8, vcc, 1, v6
 ; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s[2:3]
 ; GCN-NEXT:    v_add_u32_e32 v9, vcc, 1, v7
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v2
 ; GCN-NEXT:    v_cndmask_b32_e32 v0, v6, v8, vcc
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v1, v3
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v7, v9, vcc
 ; GCN-NEXT:    flat_store_dwordx2 v[4:5], v[0:1]
 ; GCN-NEXT:    s_endpgm
 ;
 ; GFX1030-LABEL: udiv_v2i32:
 ; GFX1030:       ; %bb.0:
-; GFX1030-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX1030-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1030-NEXT:    global_load_dwordx4 v[0:3], v4, s[6:7]
+; GFX1030-NEXT:    global_load_dwordx4 v[0:3], v4, s[2:3]
 ; GFX1030-NEXT:    s_waitcnt vmcnt(0)
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v5, v2
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v6, v3
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v7, 0, v2
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v8, 0, v3
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v5, v5
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v6, v6
-; GFX1030-NEXT:    v_mul_f32_e32 v5, 0x4f7ffffe, v5
-; GFX1030-NEXT:    v_mul_f32_e32 v6, 0x4f7ffffe, v6
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v5, v5
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v6, v6
-; GFX1030-NEXT:    v_mul_lo_u32 v7, v7, v5
-; GFX1030-NEXT:    v_mul_lo_u32 v8, v8, v6
-; GFX1030-NEXT:    v_mul_hi_u32 v7, v5, v7
-; GFX1030-NEXT:    v_mul_hi_u32 v8, v6, v8
-; GFX1030-NEXT:    v_add_nc_u32_e32 v5, v5, v7
-; GFX1030-NEXT:    v_add_nc_u32_e32 v6, v6, v8
-; GFX1030-NEXT:    v_mul_hi_u32 v5, v0, v5
-; GFX1030-NEXT:    v_mul_hi_u32 v6, v1, v6
-; GFX1030-NEXT:    v_mul_lo_u32 v7, v5, v2
-; GFX1030-NEXT:    v_mul_lo_u32 v8, v6, v3
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v0, v0, v7
-; GFX1030-NEXT:    v_add_nc_u32_e32 v7, 1, v5
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v1, v1, v8
-; GFX1030-NEXT:    v_add_nc_u32_e32 v8, 1, v6
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v0, v2
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v9, v1, v3
-; GFX1030-NEXT:    v_cmp_ge_u32_e64 s0, v1, v3
-; GFX1030-NEXT:    v_cndmask_b32_e32 v5, v5, v7, vcc_lo
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v7, v0, v2
-; GFX1030-NEXT:    v_cndmask_b32_e64 v6, v6, v8, s0
-; GFX1030-NEXT:    v_cndmask_b32_e64 v1, v1, v9, s0
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v0, v7, vcc_lo
-; GFX1030-NEXT:    v_add_nc_u32_e32 v7, 1, v5
-; GFX1030-NEXT:    v_add_nc_u32_e32 v8, 1, v6
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v0, v2
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v5, v7, vcc_lo
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v1, v3
-; GFX1030-NEXT:    v_cndmask_b32_e32 v1, v6, v8, vcc_lo
-; GFX1030-NEXT:    global_store_dwordx2 v4, v[0:1], s[4:5]
+; GFX1030-NEXT:    v_readfirstlane_b32 s2, v2
+; GFX1030-NEXT:    v_readfirstlane_b32 s3, v3
+; GFX1030-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v2, s2
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v3, s3
+; GFX1030-NEXT:    s_sub_i32 s5, 0, s2
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v3, v3
+; GFX1030-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
+; GFX1030-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v3
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX1030-NEXT:    v_readfirstlane_b32 s4, v2
+; GFX1030-NEXT:    v_readfirstlane_b32 s8, v0
+; GFX1030-NEXT:    s_mul_i32 s5, s5, s4
+; GFX1030-NEXT:    s_mul_hi_u32 s5, s4, s5
+; GFX1030-NEXT:    s_add_i32 s4, s4, s5
+; GFX1030-NEXT:    s_mul_hi_u32 s4, s6, s4
+; GFX1030-NEXT:    s_mul_i32 s5, s4, s2
+; GFX1030-NEXT:    s_sub_i32 s5, s6, s5
+; GFX1030-NEXT:    s_add_i32 s6, s4, 1
+; GFX1030-NEXT:    s_sub_i32 s7, s5, s2
+; GFX1030-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX1030-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX1030-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX1030-NEXT:    s_add_i32 s6, s4, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX1030-NEXT:    v_readfirstlane_b32 s5, v1
+; GFX1030-NEXT:    s_cselect_b32 s2, s6, s4
+; GFX1030-NEXT:    s_sub_i32 s4, 0, s3
+; GFX1030-NEXT:    v_mov_b32_e32 v0, s2
+; GFX1030-NEXT:    s_mul_i32 s4, s4, s8
+; GFX1030-NEXT:    s_mul_hi_u32 s4, s8, s4
+; GFX1030-NEXT:    s_add_i32 s8, s8, s4
+; GFX1030-NEXT:    s_mul_hi_u32 s4, s5, s8
+; GFX1030-NEXT:    s_mul_i32 s6, s4, s3
+; GFX1030-NEXT:    s_sub_i32 s5, s5, s6
+; GFX1030-NEXT:    s_add_i32 s6, s4, 1
+; GFX1030-NEXT:    s_sub_i32 s7, s5, s3
+; GFX1030-NEXT:    s_cmp_ge_u32 s5, s3
+; GFX1030-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX1030-NEXT:    s_cselect_b32 s5, s7, s5
+; GFX1030-NEXT:    s_add_i32 s6, s4, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s5, s3
+; GFX1030-NEXT:    s_cselect_b32 s3, s6, s4
+; GFX1030-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1030-NEXT:    global_store_dwordx2 v4, v[0:1], s[0:1]
 ; GFX1030-NEXT:    s_endpgm
 ;
 ; EG-LABEL: udiv_v2i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @8, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 0 @6
 ; EG-NEXT:    ALU 33, @9, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T1.XY, T0.X, 1

 ; SI-NEXT:    v_mul_lo_u32 v9, v9, v8
 ; SI-NEXT:    v_mul_lo_u32 v11, v11, v10
 ; SI-NEXT:    v_mul_lo_u32 v13, v13, v12
 ; SI-NEXT:    v_mul_lo_u32 v15, v15, v14
 ; SI-NEXT:    v_mul_hi_u32 v9, v8, v9
 ; SI-NEXT:    v_mul_hi_u32 v11, v10, v11
 ; SI-NEXT:    v_mul_hi_u32 v13, v12, v13
 ; SI-NEXT:    v_mul_hi_u32 v15, v14, v15
-; SI-NEXT:    v_add_i32_e32 v8, vcc, v9, v8
+; SI-NEXT:    v_add_i32_e32 v8, vcc, v8, v9
 ; SI-NEXT:    v_add_i32_e32 v9, vcc, v11, v10
-; SI-NEXT:    v_add_i32_e32 v10, vcc, v13, v12
+; SI-NEXT:    v_add_i32_e32 v10, vcc, v12, v13
 ; SI-NEXT:    v_add_i32_e32 v11, vcc, v15, v14
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_mul_hi_u32 v8, v4, v8
 ; SI-NEXT:    v_mul_hi_u32 v9, v5, v9
 ; SI-NEXT:    v_mul_hi_u32 v10, v6, v10
 ; SI-NEXT:    v_mul_hi_u32 v11, v7, v11
 ; SI-NEXT:    v_mul_lo_u32 v12, v8, v0
 ; SI-NEXT:    v_mul_lo_u32 v14, v9, v1
 ; SI-NEXT:    v_mul_lo_u32 v16, v10, v2
 ; SI-NEXT:    v_mul_lo_u32 v18, v11, v3
-; SI-NEXT:    v_subrev_i32_e32 v4, vcc, v12, v4
+; SI-NEXT:    v_sub_i32_e32 v4, vcc, v4, v12
 ; SI-NEXT:    v_sub_i32_e32 v5, vcc, v5, v14
 ; SI-NEXT:    v_sub_i32_e32 v6, vcc, v6, v16
 ; SI-NEXT:    v_sub_i32_e32 v7, vcc, v7, v18
 ; SI-NEXT:    v_add_i32_e32 v13, vcc, 1, v8
 ; SI-NEXT:    v_add_i32_e32 v15, vcc, 1, v9
 ; SI-NEXT:    v_add_i32_e32 v17, vcc, 1, v10
 ; SI-NEXT:    v_add_i32_e32 v19, vcc, 1, v11
 ; SI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v4, v0
 ; SI-NEXT:    v_cmp_ge_u32_e64 s[2:3], v5, v1
 ; SI-NEXT:    v_cmp_ge_u32_e64 s[4:5], v6, v2
 ; SI-NEXT:    v_cmp_ge_u32_e64 s[6:7], v7, v3
-; SI-NEXT:    v_cndmask_b32_e64 v8, v8, v13, s[0:1]
 ; SI-NEXT:    v_subrev_i32_e32 v12, vcc, v0, v4
-; SI-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s[2:3]
+; SI-NEXT:    v_cndmask_b32_e64 v8, v8, v13, s[0:1]
 ; SI-NEXT:    v_subrev_i32_e32 v13, vcc, v1, v5
-; SI-NEXT:    v_cndmask_b32_e64 v10, v10, v17, s[4:5]
+; SI-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s[2:3]
 ; SI-NEXT:    v_subrev_i32_e32 v14, vcc, v2, v6
+; SI-NEXT:    v_cndmask_b32_e64 v10, v10, v17, s[4:5]
+; SI-NEXT:    v_subrev_i32_e32 v15, vcc, v3, v7
 ; SI-NEXT:    v_cndmask_b32_e64 v11, v11, v19, s[6:7]
-; SI-NEXT:    v_sub_i32_e32 v15, vcc, v7, v3
 ; SI-NEXT:    v_cndmask_b32_e64 v4, v4, v12, s[0:1]
 ; SI-NEXT:    v_add_i32_e32 v12, vcc, 1, v8
 ; SI-NEXT:    v_cndmask_b32_e64 v5, v5, v13, s[2:3]
 ; SI-NEXT:    v_add_i32_e32 v13, vcc, 1, v9
 ; SI-NEXT:    v_cndmask_b32_e64 v6, v6, v14, s[4:5]
 ; SI-NEXT:    v_add_i32_e32 v14, vcc, 1, v10
 ; SI-NEXT:    v_cndmask_b32_e64 v7, v7, v15, s[6:7]
 ; SI-NEXT:    v_add_i32_e32 v15, vcc, 1, v11

 ; VI-NEXT:    v_mul_lo_u32 v9, v9, v8
 ; VI-NEXT:    v_mul_lo_u32 v11, v11, v10
 ; VI-NEXT:    v_mul_lo_u32 v13, v13, v12
 ; VI-NEXT:    v_mul_lo_u32 v15, v15, v14
 ; VI-NEXT:    v_mul_hi_u32 v9, v8, v9
 ; VI-NEXT:    v_mul_hi_u32 v11, v10, v11
 ; VI-NEXT:    v_mul_hi_u32 v13, v12, v13
 ; VI-NEXT:    v_mul_hi_u32 v15, v14, v15
-; VI-NEXT:    v_add_u32_e32 v8, vcc, v9, v8
-; VI-NEXT:    v_add_u32_e32 v9, vcc, v10, v11
+; VI-NEXT:    v_add_u32_e32 v8, vcc, v8, v9
+; VI-NEXT:    v_add_u32_e32 v9, vcc, v11, v10
 ; VI-NEXT:    v_add_u32_e32 v10, vcc, v12, v13
 ; VI-NEXT:    v_add_u32_e32 v11, vcc, v15, v14
 ; VI-NEXT:    s_waitcnt vmcnt(0)
 ; VI-NEXT:    v_mul_hi_u32 v8, v4, v8
 ; VI-NEXT:    v_mul_hi_u32 v9, v5, v9
 ; VI-NEXT:    v_mul_hi_u32 v10, v6, v10
 ; VI-NEXT:    v_mul_hi_u32 v11, v7, v11
 ; VI-NEXT:    v_mul_lo_u32 v12, v8, v0
 ; VI-NEXT:    v_mul_lo_u32 v14, v9, v1
 ; VI-NEXT:    v_mul_lo_u32 v16, v10, v2
 ; VI-NEXT:    v_mul_lo_u32 v18, v11, v3
 ; VI-NEXT:    v_sub_u32_e32 v4, vcc, v4, v12
-; VI-NEXT:    v_subrev_u32_e32 v5, vcc, v14, v5
-; VI-NEXT:    v_subrev_u32_e32 v6, vcc, v16, v6
+; VI-NEXT:    v_sub_u32_e32 v5, vcc, v5, v14
+; VI-NEXT:    v_sub_u32_e32 v6, vcc, v6, v16
 ; VI-NEXT:    v_sub_u32_e32 v7, vcc, v7, v18
 ; VI-NEXT:    v_add_u32_e32 v13, vcc, 1, v8
 ; VI-NEXT:    v_add_u32_e32 v15, vcc, 1, v9
 ; VI-NEXT:    v_add_u32_e32 v17, vcc, 1, v10
 ; VI-NEXT:    v_add_u32_e32 v19, vcc, 1, v11
 ; VI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v4, v0
 ; VI-NEXT:    v_cmp_ge_u32_e64 s[2:3], v5, v1
 ; VI-NEXT:    v_cmp_ge_u32_e64 s[4:5], v6, v2
 ; VI-NEXT:    v_cmp_ge_u32_e64 s[6:7], v7, v3
-; VI-NEXT:    v_cndmask_b32_e64 v8, v8, v13, s[0:1]
 ; VI-NEXT:    v_subrev_u32_e32 v12, vcc, v0, v4
-; VI-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s[2:3]
+; VI-NEXT:    v_cndmask_b32_e64 v8, v8, v13, s[0:1]
 ; VI-NEXT:    v_subrev_u32_e32 v13, vcc, v1, v5
-; VI-NEXT:    v_cndmask_b32_e64 v10, v10, v17, s[4:5]
+; VI-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s[2:3]
 ; VI-NEXT:    v_subrev_u32_e32 v14, vcc, v2, v6
-; VI-NEXT:    v_cndmask_b32_e64 v11, v11, v19, s[6:7]
+; VI-NEXT:    v_cndmask_b32_e64 v10, v10, v17, s[4:5]
 ; VI-NEXT:    v_subrev_u32_e32 v15, vcc, v3, v7
+; VI-NEXT:    v_cndmask_b32_e64 v11, v11, v19, s[6:7]
 ; VI-NEXT:    v_cndmask_b32_e64 v4, v4, v12, s[0:1]
 ; VI-NEXT:    v_add_u32_e32 v12, vcc, 1, v8
 ; VI-NEXT:    v_cndmask_b32_e64 v5, v5, v13, s[2:3]
 ; VI-NEXT:    v_add_u32_e32 v13, vcc, 1, v9
 ; VI-NEXT:    v_cndmask_b32_e64 v6, v6, v14, s[4:5]
 ; VI-NEXT:    v_add_u32_e32 v14, vcc, 1, v10
 ; VI-NEXT:    v_cndmask_b32_e64 v7, v7, v15, s[6:7]
 ; VI-NEXT:    v_add_u32_e32 v15, vcc, 1, v11

 ; GCN-NEXT:    v_mul_lo_u32 v11, v11, v10
 ; GCN-NEXT:    v_mul_lo_u32 v13, v13, v12
 ; GCN-NEXT:    v_mul_lo_u32 v15, v15, v14
 ; GCN-NEXT:    v_mul_lo_u32 v17, v17, v16
 ; GCN-NEXT:    v_mul_hi_u32 v11, v10, v11
 ; GCN-NEXT:    v_mul_hi_u32 v13, v12, v13
 ; GCN-NEXT:    v_mul_hi_u32 v15, v14, v15
 ; GCN-NEXT:    v_mul_hi_u32 v17, v16, v17
-; GCN-NEXT:    v_add_u32_e32 v10, vcc, v11, v10
-; GCN-NEXT:    v_add_u32_e32 v11, vcc, v12, v13
+; GCN-NEXT:    v_add_u32_e32 v10, vcc, v10, v11
+; GCN-NEXT:    v_add_u32_e32 v11, vcc, v13, v12
 ; GCN-NEXT:    v_add_u32_e32 v12, vcc, v14, v15
 ; GCN-NEXT:    v_add_u32_e32 v13, vcc, v17, v16
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_mul_hi_u32 v10, v4, v10
 ; GCN-NEXT:    v_mul_hi_u32 v11, v5, v11
 ; GCN-NEXT:    v_mul_hi_u32 v12, v6, v12
 ; GCN-NEXT:    v_mul_hi_u32 v13, v7, v13
 ; GCN-NEXT:    v_mul_lo_u32 v14, v10, v0
 ; GCN-NEXT:    v_mul_lo_u32 v16, v11, v1
 ; GCN-NEXT:    v_mul_lo_u32 v18, v12, v2
 ; GCN-NEXT:    v_mul_lo_u32 v19, v13, v3
 ; GCN-NEXT:    v_sub_u32_e32 v4, vcc, v4, v14
-; GCN-NEXT:    v_subrev_u32_e32 v5, vcc, v16, v5
-; GCN-NEXT:    v_subrev_u32_e32 v6, vcc, v18, v6
+; GCN-NEXT:    v_sub_u32_e32 v5, vcc, v5, v16
+; GCN-NEXT:    v_sub_u32_e32 v6, vcc, v6, v18
 ; GCN-NEXT:    v_sub_u32_e32 v7, vcc, v7, v19
 ; GCN-NEXT:    v_add_u32_e32 v15, vcc, 1, v10
 ; GCN-NEXT:    v_add_u32_e32 v17, vcc, 1, v11
 ; GCN-NEXT:    v_add_u32_e32 v14, vcc, 1, v12
 ; GCN-NEXT:    v_add_u32_e32 v16, vcc, 1, v13
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v4, v0
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[2:3], v5, v1
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[4:5], v6, v2
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[6:7], v7, v3
+; GCN-NEXT:    v_subrev_u32_e32 v18, vcc, v0, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v10, v10, v15, s[0:1]
-; GCN-NEXT:    v_subrev_u32_e32 v15, vcc, v0, v4
+; GCN-NEXT:    v_subrev_u32_e32 v15, vcc, v1, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v11, v11, v17, s[2:3]
-; GCN-NEXT:    v_subrev_u32_e32 v17, vcc, v1, v5
+; GCN-NEXT:    v_subrev_u32_e32 v17, vcc, v2, v6
 ; GCN-NEXT:    v_cndmask_b32_e64 v12, v12, v14, s[4:5]
-; GCN-NEXT:    v_subrev_u32_e32 v14, vcc, v2, v6
+; GCN-NEXT:    v_subrev_u32_e32 v14, vcc, v3, v7
 ; GCN-NEXT:    v_cndmask_b32_e64 v13, v13, v16, s[6:7]
-; GCN-NEXT:    v_subrev_u32_e32 v16, vcc, v3, v7
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v15, s[0:1]
-; GCN-NEXT:    v_add_u32_e32 v15, vcc, 1, v10
-; GCN-NEXT:    v_cndmask_b32_e64 v5, v5, v17, s[2:3]
-; GCN-NEXT:    v_add_u32_e32 v17, vcc, 1, v11
-; GCN-NEXT:    v_cndmask_b32_e64 v6, v6, v14, s[4:5]
-; GCN-NEXT:    v_add_u32_e32 v14, vcc, 1, v12
-; GCN-NEXT:    v_cndmask_b32_e64 v7, v7, v16, s[6:7]
-; GCN-NEXT:    v_add_u32_e32 v16, vcc, 1, v13
+; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v18, s[0:1]
+; GCN-NEXT:    v_add_u32_e32 v16, vcc, 1, v10
+; GCN-NEXT:    v_cndmask_b32_e64 v5, v5, v15, s[2:3]
+; GCN-NEXT:    v_add_u32_e32 v15, vcc, 1, v11
+; GCN-NEXT:    v_cndmask_b32_e64 v6, v6, v17, s[4:5]
+; GCN-NEXT:    v_add_u32_e32 v17, vcc, 1, v12
+; GCN-NEXT:    v_cndmask_b32_e64 v7, v7, v14, s[6:7]
+; GCN-NEXT:    v_add_u32_e32 v14, vcc, 1, v13
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v4, v0
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v10, v15, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v10, v16, vcc
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v5, v1
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v11, v17, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v11, v15, vcc
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v6, v2
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v12, v14, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v12, v17, vcc
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v7, v3
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v13, v16, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v3, v13, v14, vcc
 ; GCN-NEXT:    flat_store_dwordx4 v[8:9], v[0:3]
 ; GCN-NEXT:    s_endpgm
 ;
 ; GFX1030-LABEL: udiv_v4i32:
 ; GFX1030:       ; %bb.0:
-; GFX1030-NEXT:    s_load_dwordx4 s[4:7], s[4:5], 0x0
+; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX1030-NEXT:    v_mov_b32_e32 v8, 0
 ; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-NEXT:    s_clause 0x1
-; GFX1030-NEXT:    global_load_dwordx4 v[0:3], v8, s[6:7] offset:16
-; GFX1030-NEXT:    global_load_dwordx4 v[4:7], v8, s[6:7]
+; GFX1030-NEXT:    global_load_dwordx4 v[0:3], v8, s[2:3] offset:16
+; GFX1030-NEXT:    global_load_dwordx4 v[4:7], v8, s[2:3]
 ; GFX1030-NEXT:    s_waitcnt vmcnt(1)
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v9, v0
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v10, v1
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v11, v2
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v12, v3
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v13, 0, v0
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v9, v9
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v10, v10
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v11, v11
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v12, v12
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v14, 0, v1
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v15, 0, v2
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v16, 0, v3
-; GFX1030-NEXT:    v_mul_f32_e32 v9, 0x4f7ffffe, v9
-; GFX1030-NEXT:    v_mul_f32_e32 v10, 0x4f7ffffe, v10
-; GFX1030-NEXT:    v_mul_f32_e32 v11, 0x4f7ffffe, v11
-; GFX1030-NEXT:    v_mul_f32_e32 v12, 0x4f7ffffe, v12
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v9, v9
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v10, v10
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v11, v11
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v12, v12
-; GFX1030-NEXT:    v_mul_lo_u32 v13, v13, v9
-; GFX1030-NEXT:    v_mul_lo_u32 v14, v14, v10
-; GFX1030-NEXT:    v_mul_lo_u32 v15, v15, v11
-; GFX1030-NEXT:    v_mul_lo_u32 v16, v16, v12
-; GFX1030-NEXT:    v_mul_hi_u32 v13, v9, v13
-; GFX1030-NEXT:    v_mul_hi_u32 v14, v10, v14
-; GFX1030-NEXT:    v_mul_hi_u32 v15, v11, v15
-; GFX1030-NEXT:    v_mul_hi_u32 v16, v12, v16
-; GFX1030-NEXT:    v_add_nc_u32_e32 v9, v9, v13
-; GFX1030-NEXT:    v_add_nc_u32_e32 v10, v10, v14
-; GFX1030-NEXT:    v_add_nc_u32_e32 v11, v11, v15
-; GFX1030-NEXT:    v_add_nc_u32_e32 v12, v12, v16
+; GFX1030-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1030-NEXT:    v_readfirstlane_b32 s3, v1
 ; GFX1030-NEXT:    s_waitcnt vmcnt(0)
-; GFX1030-NEXT:    v_mul_hi_u32 v9, v4, v9
-; GFX1030-NEXT:    v_mul_hi_u32 v10, v5, v10
-; GFX1030-NEXT:    v_mul_hi_u32 v11, v6, v11
-; GFX1030-NEXT:    v_mul_hi_u32 v12, v7, v12
-; GFX1030-NEXT:    v_mul_lo_u32 v13, v9, v0
-; GFX1030-NEXT:    v_mul_lo_u32 v14, v10, v1
-; GFX1030-NEXT:    v_mul_lo_u32 v15, v11, v2
-; GFX1030-NEXT:    v_mul_lo_u32 v16, v12, v3
-; GFX1030-NEXT:    v_add_nc_u32_e32 v17, 1, v9
-; GFX1030-NEXT:    v_add_nc_u32_e32 v18, 1, v10
-; GFX1030-NEXT:    v_add_nc_u32_e32 v19, 1, v11
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v4, v4, v13
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v5, v5, v14
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v6, v6, v15
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v7, v7, v16
-; GFX1030-NEXT:    v_add_nc_u32_e32 v13, 1, v12
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v4, v0
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v14, v4, v0
-; GFX1030-NEXT:    v_cmp_ge_u32_e64 s0, v5, v1
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v15, v5, v1
-; GFX1030-NEXT:    v_cmp_ge_u32_e64 s1, v6, v2
-; GFX1030-NEXT:    v_cndmask_b32_e32 v9, v9, v17, vcc_lo
-; GFX1030-NEXT:    v_cndmask_b32_e32 v4, v4, v14, vcc_lo
-; GFX1030-NEXT:    v_cndmask_b32_e64 v10, v10, v18, s0
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v16, v6, v2
-; GFX1030-NEXT:    v_cmp_ge_u32_e64 s2, v7, v3
-; GFX1030-NEXT:    v_add_nc_u32_e32 v14, 1, v9
-; GFX1030-NEXT:    v_cndmask_b32_e64 v5, v5, v15, s0
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v4, v0
-; GFX1030-NEXT:    v_cndmask_b32_e64 v11, v11, v19, s1
-; GFX1030-NEXT:    v_cndmask_b32_e64 v12, v12, v13, s2
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v13, v7, v3
-; GFX1030-NEXT:    v_add_nc_u32_e32 v15, 1, v10
-; GFX1030-NEXT:    v_cndmask_b32_e64 v6, v6, v16, s1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v9, v14, vcc_lo
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v5, v1
-; GFX1030-NEXT:    v_add_nc_u32_e32 v16, 1, v11
-; GFX1030-NEXT:    v_cndmask_b32_e64 v7, v7, v13, s2
-; GFX1030-NEXT:    v_add_nc_u32_e32 v13, 1, v12
-; GFX1030-NEXT:    v_cndmask_b32_e32 v1, v10, v15, vcc_lo
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v6, v2
-; GFX1030-NEXT:    v_cndmask_b32_e32 v2, v11, v16, vcc_lo
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v7, v3
-; GFX1030-NEXT:    v_cndmask_b32_e32 v3, v12, v13, vcc_lo
-; GFX1030-NEXT:    global_store_dwordx4 v8, v[0:3], s[4:5]
+; GFX1030-NEXT:    v_readfirstlane_b32 s7, v4
+; GFX1030-NEXT:    v_readfirstlane_b32 s5, v2
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v0, s2
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v1, s3
+; GFX1030-NEXT:    s_sub_i32 s6, 0, s2
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX1030-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX1030-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s4, v0
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v0, s5
+; GFX1030-NEXT:    v_readfirstlane_b32 s9, v1
+; GFX1030-NEXT:    s_mul_i32 s6, s6, s4
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s4, s6
+; GFX1030-NEXT:    s_add_i32 s4, s4, s6
+; GFX1030-NEXT:    s_mul_hi_u32 s4, s7, s4
+; GFX1030-NEXT:    s_mul_i32 s6, s4, s2
+; GFX1030-NEXT:    s_sub_i32 s6, s7, s6
+; GFX1030-NEXT:    s_add_i32 s7, s4, 1
+; GFX1030-NEXT:    s_sub_i32 s8, s6, s2
+; GFX1030-NEXT:    s_cmp_ge_u32 s6, s2
+; GFX1030-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX1030-NEXT:    s_cselect_b32 s4, s7, s4
+; GFX1030-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX1030-NEXT:    s_add_i32 s7, s4, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s6, s2
+; GFX1030-NEXT:    v_readfirstlane_b32 s2, v3
+; GFX1030-NEXT:    s_cselect_b32 s4, s7, s4
+; GFX1030-NEXT:    s_sub_i32 s6, 0, s3
+; GFX1030-NEXT:    v_readfirstlane_b32 s7, v5
+; GFX1030-NEXT:    s_mul_i32 s6, s6, s9
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s9, s6
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v1, s2
+; GFX1030-NEXT:    s_add_i32 s9, s9, s6
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s7, s9
+; GFX1030-NEXT:    v_readfirstlane_b32 s10, v0
+; GFX1030-NEXT:    s_mul_i32 s8, s6, s3
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX1030-NEXT:    s_sub_i32 s7, s7, s8
+; GFX1030-NEXT:    s_add_i32 s8, s6, 1
+; GFX1030-NEXT:    s_sub_i32 s9, s7, s3
+; GFX1030-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX1030-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX1030-NEXT:    s_cselect_b32 s7, s9, s7
+; GFX1030-NEXT:    s_add_i32 s8, s6, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX1030-NEXT:    v_readfirstlane_b32 s7, v6
+; GFX1030-NEXT:    s_cselect_b32 s3, s8, s6
+; GFX1030-NEXT:    s_sub_i32 s6, 0, s5
+; GFX1030-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v1
+; GFX1030-NEXT:    s_mul_i32 s6, s6, s10
+; GFX1030-NEXT:    v_mov_b32_e32 v1, s3
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s10, s6
+; GFX1030-NEXT:    s_add_i32 s10, s10, s6
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s7, s10
+; GFX1030-NEXT:    s_mul_i32 s8, s6, s5
+; GFX1030-NEXT:    s_sub_i32 s7, s7, s8
+; GFX1030-NEXT:    s_add_i32 s8, s6, 1
+; GFX1030-NEXT:    s_sub_i32 s9, s7, s5
+; GFX1030-NEXT:    s_cmp_ge_u32 s7, s5
+; GFX1030-NEXT:    v_readfirstlane_b32 s10, v0
+; GFX1030-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX1030-NEXT:    s_cselect_b32 s7, s9, s7
+; GFX1030-NEXT:    s_add_i32 s8, s6, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s7, s5
+; GFX1030-NEXT:    v_readfirstlane_b32 s7, v7
+; GFX1030-NEXT:    s_cselect_b32 s5, s8, s6
+; GFX1030-NEXT:    s_sub_i32 s6, 0, s2
+; GFX1030-NEXT:    v_mov_b32_e32 v0, s4
+; GFX1030-NEXT:    s_mul_i32 s6, s6, s10
+; GFX1030-NEXT:    v_mov_b32_e32 v2, s5
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s10, s6
+; GFX1030-NEXT:    s_add_i32 s10, s10, s6
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s7, s10
+; GFX1030-NEXT:    s_mul_i32 s8, s6, s2
+; GFX1030-NEXT:    s_sub_i32 s7, s7, s8
+; GFX1030-NEXT:    s_add_i32 s8, s6, 1
+; GFX1030-NEXT:    s_sub_i32 s9, s7, s2
+; GFX1030-NEXT:    s_cmp_ge_u32 s7, s2
+; GFX1030-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX1030-NEXT:    s_cselect_b32 s7, s9, s7
+; GFX1030-NEXT:    s_add_i32 s8, s6, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s7, s2
+; GFX1030-NEXT:    s_cselect_b32 s2, s8, s6
+; GFX1030-NEXT:    v_mov_b32_e32 v3, s2
+; GFX1030-NEXT:    global_store_dwordx4 v8, v[0:3], s[0:1]
 ; GFX1030-NEXT:    s_endpgm
 ;
 ; EG-LABEL: udiv_v4i32:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @10, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 1 @6
 ; EG-NEXT:    ALU 65, @11, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    MEM_RAT_CACHELESS STORE_RAW T4.XYZW, T0.X, 1

 ; SI-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
 ; SI-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; SI-NEXT:    v_mul_lo_u32 v4, v4, v1
 ; SI-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; SI-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
 ; SI-NEXT:    v_mul_hi_u32 v1, v2, v1
 ; SI-NEXT:    v_mul_lo_u32 v3, v1, v0
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, 1, v1
-; SI-NEXT:    v_subrev_i32_e32 v2, vcc, v3, v2
-; SI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v2, v0
-; SI-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; SI-NEXT:    v_sub_i32_e32 v2, vcc, v2, v3
 ; SI-NEXT:    v_subrev_i32_e32 v3, vcc, v0, v2
-; SI-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
+; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v0
+; SI-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; SI-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
 ; SI-NEXT:    v_add_i32_e32 v3, vcc, 1, v1
 ; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v0
 ; SI-NEXT:    v_cndmask_b32_e32 v0, v1, v3, vcc
 ; SI-NEXT:    v_and_b32_e32 v0, 0xffffff, v0
 ; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: v_udiv_i24:

 ; VI-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; VI-NEXT:    v_mul_lo_u32 v4, v4, v1
 ; VI-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; VI-NEXT:    v_add_u32_e32 v1, vcc, v1, v4
 ; VI-NEXT:    v_mul_hi_u32 v1, v2, v1
 ; VI-NEXT:    v_mul_lo_u32 v3, v1, v0
 ; VI-NEXT:    v_add_u32_e32 v4, vcc, 1, v1
 ; VI-NEXT:    v_sub_u32_e32 v2, vcc, v2, v3
-; VI-NEXT:    v_cmp_ge_u32_e64 s[0:1], v2, v0
-; VI-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
-; VI-NEXT:    v_sub_u32_e32 v3, vcc, v2, v0
-; VI-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
+; VI-NEXT:    v_subrev_u32_e32 v3, vcc, v0, v2
+; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v0
+; VI-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
+; VI-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
 ; VI-NEXT:    v_add_u32_e32 v3, vcc, 1, v1
 ; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v0
 ; VI-NEXT:    v_cndmask_b32_e32 v0, v1, v3, vcc
 ; VI-NEXT:    v_and_b32_e32 v0, 0xffffff, v0
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GCN-LABEL: v_udiv_i24:

 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; GCN-NEXT:    v_add_u32_e32 v0, vcc, v1, v4
 ; GCN-NEXT:    v_mul_hi_u32 v4, v2, v0
 ; GCN-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-NEXT:    v_mov_b32_e32 v1, s1
 ; GCN-NEXT:    v_mul_lo_u32 v5, v4, v3
 ; GCN-NEXT:    v_add_u32_e32 v6, vcc, 1, v4
 ; GCN-NEXT:    v_sub_u32_e32 v2, vcc, v2, v5
-; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v2, v3
-; GCN-NEXT:    v_cndmask_b32_e64 v4, v4, v6, s[0:1]
-; GCN-NEXT:    v_sub_u32_e32 v5, vcc, v2, v3
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v2, v5, s[0:1]
+; GCN-NEXT:    v_subrev_u32_e32 v5, vcc, v3, v2
+; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v3
+; GCN-NEXT:    v_cndmask_b32_e32 v4, v4, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
 ; GCN-NEXT:    v_add_u32_e32 v5, vcc, 1, v4
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v3
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, v4, v5, vcc
 ; GCN-NEXT:    v_and_b32_e32 v2, 0xffffff, v2
 ; GCN-NEXT:    flat_store_dword v[0:1], v2
 ; GCN-NEXT:    s_endpgm
 ;
 ; GFX1030-LABEL: v_udiv_i24:
 ; GFX1030:       ; %bb.0:
 ; GFX1030-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x0
 ; GFX1030-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-NEXT:    s_clause 0x3
 ; GFX1030-NEXT:    global_load_ubyte v1, v0, s[2:3] offset:6
 ; GFX1030-NEXT:    global_load_ushort v2, v0, s[2:3] offset:4
 ; GFX1030-NEXT:    global_load_ubyte v3, v0, s[2:3] offset:2
 ; GFX1030-NEXT:    global_load_ushort v4, v0, s[2:3]
 ; GFX1030-NEXT:    s_waitcnt vmcnt(3)
-; GFX1030-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX1030-NEXT:    s_waitcnt vmcnt(2)
+; GFX1030-NEXT:    v_readfirstlane_b32 s3, v2
 ; GFX1030-NEXT:    s_waitcnt vmcnt(1)
-; GFX1030-NEXT:    v_lshlrev_b32_e32 v3, 16, v3
-; GFX1030-NEXT:    v_or_b32_e32 v1, v2, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s4, v3
 ; GFX1030-NEXT:    s_waitcnt vmcnt(0)
-; GFX1030-NEXT:    v_or_b32_e32 v3, v4, v3
-; GFX1030-NEXT:    v_cvt_f32_u32_e32 v2, v1
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v5, 0, v1
-; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v2, v2
-; GFX1030-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
-; GFX1030-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX1030-NEXT:    v_mul_lo_u32 v5, v5, v2
-; GFX1030-NEXT:    v_mul_hi_u32 v5, v2, v5
-; GFX1030-NEXT:    v_add_nc_u32_e32 v2, v2, v5
-; GFX1030-NEXT:    v_mul_hi_u32 v2, v3, v2
-; GFX1030-NEXT:    v_mul_lo_u32 v4, v2, v1
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v3, v3, v4
-; GFX1030-NEXT:    v_add_nc_u32_e32 v4, 1, v2
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v5, v3, v1
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v3, v1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
-; GFX1030-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc_lo
-; GFX1030-NEXT:    v_add_nc_u32_e32 v4, 1, v2
-; GFX1030-NEXT:    v_cmp_ge_u32_e32 vcc_lo, v3, v1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v1, v2, v4, vcc_lo
-; GFX1030-NEXT:    v_and_b32_e32 v1, 0xffffff, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s5, v4
+; GFX1030-NEXT:    s_lshl_b32 s2, s2, 16
+; GFX1030-NEXT:    s_or_b32 s2, s3, s2
+; GFX1030-NEXT:    s_lshl_b32 s4, s4, 16
+; GFX1030-NEXT:    v_cvt_f32_u32_e32 v1, s2
+; GFX1030-NEXT:    s_sub_i32 s6, 0, s2
+; GFX1030-NEXT:    s_or_b32 s4, s5, s4
+; GFX1030-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX1030-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX1030-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX1030-NEXT:    v_readfirstlane_b32 s3, v1
+; GFX1030-NEXT:    s_mul_i32 s6, s6, s3
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s3, s6
+; GFX1030-NEXT:    s_add_i32 s3, s3, s6
+; GFX1030-NEXT:    s_mul_hi_u32 s3, s4, s3
+; GFX1030-NEXT:    s_mul_i32 s5, s3, s2
+; GFX1030-NEXT:    s_sub_i32 s4, s4, s5
+; GFX1030-NEXT:    s_add_i32 s5, s3, 1
+; GFX1030-NEXT:    s_sub_i32 s6, s4, s2
+; GFX1030-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX1030-NEXT:    s_cselect_b32 s3, s5, s3
+; GFX1030-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX1030-NEXT:    s_add_i32 s5, s3, 1
+; GFX1030-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX1030-NEXT:    s_cselect_b32 s2, s5, s3
+; GFX1030-NEXT:    s_and_b32 s2, s2, 0xffffff
+; GFX1030-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX1030-NEXT:    global_store_dword v0, v1, s[0:1]
 ; GFX1030-NEXT:    s_endpgm
 ;
 ; EG-LABEL: v_udiv_i24:
 ; EG:       ; %bb.0:
 ; EG-NEXT:    ALU 0, @14, KC0[CB0:0-32], KC1[]
 ; EG-NEXT:    TEX 3 @6
 ; EG-NEXT:    ALU 23, @15, KC0[CB0:0-32], KC1[]

 ; SI-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; SI-NEXT:    v_or_b32_e32 v0, 1, v0
 ; SI-NEXT:    v_mul_f32_e32 v1, v3, v4
 ; SI-NEXT:    v_trunc_f32_e32 v1, v1
 ; SI-NEXT:    v_mad_f32 v3, -v1, v2, v3
 ; SI-NEXT:    v_cvt_i32_f32_e32 v1, v1
 ; SI-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
 ; SI-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; SI-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; SI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
 ; VI-LABEL: fdiv_test_denormals:
 ; VI:       ; %bb.0: ; %bb
 ; VI-NEXT:    s_mov_b32 s0, 0
 ; VI-NEXT:    s_mov_b32 s3, 0xf000
 ; VI-NEXT:    s_mov_b32 s2, -1

 ; VI-NEXT:    v_rcp_iflag_f32_e32 v4, v2
 ; VI-NEXT:    v_or_b32_e32 v0, 1, v0
 ; VI-NEXT:    v_mul_f32_e32 v1, v3, v4
 ; VI-NEXT:    v_trunc_f32_e32 v1, v1
 ; VI-NEXT:    v_mad_f32 v3, -v1, v2, v3
 ; VI-NEXT:    v_cvt_i32_f32_e32 v1, v1
 ; VI-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
 ; VI-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
+; VI-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; VI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
 ; GCN-LABEL: fdiv_test_denormals:
 ; GCN:       ; %bb.0: ; %bb
 ; GCN-NEXT:    flat_load_sbyte v2, v[0:1]
 ; GCN-NEXT:    v_mov_b32_e32 v0, 0
 ; GCN-NEXT:    v_mov_b32_e32 v1, 0
 ; GCN-NEXT:    flat_load_sbyte v3, v[0:1]
 ; GCN-NEXT:    s_waitcnt vmcnt(1)
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v4, v2
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_cvt_f32_i32_e32 v5, v3
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v6, v4
 ; GCN-NEXT:    v_xor_b32_e32 v2, v3, v2
 ; GCN-NEXT:    v_ashrrev_i32_e32 v2, 30, v2
 ; GCN-NEXT:    v_or_b32_e32 v2, 1, v2
 ; GCN-NEXT:    v_mul_f32_e32 v3, v5, v6
 ; GCN-NEXT:    v_trunc_f32_e32 v3, v3
 ; GCN-NEXT:    v_mad_f32 v5, -v3, v4, v5
 ; GCN-NEXT:    v_cvt_i32_f32_e32 v3, v3
 ; GCN-NEXT:    v_cmp_ge_f32_e64 vcc, |v5|, |v4|
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, 0, v2, vcc
-; GCN-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_u32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    flat_store_byte v[0:1], v2
 ; GCN-NEXT:    s_endpgm
 ;
 ; GFX1030-LABEL: fdiv_test_denormals:
 ; GFX1030:       ; %bb.0: ; %bb
 ; GFX1030-NEXT:    global_load_sbyte v2, v[0:1], off
 ; GFX1030-NEXT:    v_mov_b32_e32 v0, 0
 ; GFX1030-NEXT:    v_mov_b32_e32 v1, 0

 ; SI-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; SI-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; SI-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; SI-NEXT:    v_addc_u32_e32 v3, vcc, v3, v5, vcc
 ; SI-NEXT:    v_mul_hi_u32 v4, v2, s4
 ; SI-NEXT:    v_mul_lo_u32 v5, v3, s4
 ; SI-NEXT:    v_mul_lo_u32 v6, v2, s4
 ; SI-NEXT:    s_mov_b32 s4, 0x186a0
-; SI-NEXT:    v_subrev_i32_e32 v4, vcc, v2, v4
+; SI-NEXT:    v_sub_i32_e32 v4, vcc, v4, v2
 ; SI-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
 ; SI-NEXT:    v_mul_lo_u32 v5, v2, v4
 ; SI-NEXT:    v_mul_hi_u32 v7, v2, v6
 ; SI-NEXT:    v_mul_hi_u32 v8, v2, v4
 ; SI-NEXT:    v_mul_hi_u32 v9, v3, v4
 ; SI-NEXT:    v_mul_lo_u32 v4, v3, v4
 ; SI-NEXT:    v_add_i32_e32 v5, vcc, v7, v5
 ; SI-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; VI-NEXT:    v_mul_f32_e32 v2, 0x5f7ffffc, v2
 ; VI-NEXT:    v_mul_f32_e32 v3, 0x2f800000, v2
 ; VI-NEXT:    v_trunc_f32_e32 v3, v3
 ; VI-NEXT:    v_mac_f32_e32 v2, 0xcf800000, v3
 ; VI-NEXT:    v_cvt_u32_f32_e32 v6, v2
 ; VI-NEXT:    v_cvt_u32_f32_e32 v7, v3
 ; VI-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v6, s6, 0
 ; VI-NEXT:    v_mul_lo_u32 v4, v7, s6
-; VI-NEXT:    v_subrev_u32_e32 v3, vcc, v6, v3
-; VI-NEXT:    v_add_u32_e32 v8, vcc, v3, v4
+; VI-NEXT:    v_sub_u32_e32 v3, vcc, v3, v6
+; VI-NEXT:    v_add_u32_e32 v8, vcc, v4, v3
 ; VI-NEXT:    v_mul_hi_u32 v5, v6, v2
 ; VI-NEXT:    v_mad_u64_u32 v[3:4], s[4:5], v6, v8, 0
 ; VI-NEXT:    v_add_u32_e32 v9, vcc, v5, v3
 ; VI-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v7, v2, 0
 ; VI-NEXT:    v_addc_u32_e32 v10, vcc, 0, v4, vcc
 ; VI-NEXT:    v_mad_u64_u32 v[4:5], s[4:5], v7, v8, 0
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, v9, v2
 ; VI-NEXT:    v_addc_u32_e32 v2, vcc, v10, v3, vcc
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v5, vcc
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
 ; VI-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
 ; VI-NEXT:    v_add_u32_e32 v6, vcc, v6, v2
 ; VI-NEXT:    v_addc_u32_e32 v7, vcc, v7, v3, vcc
 ; VI-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v6, s6, 0
 ; VI-NEXT:    v_mul_lo_u32 v4, v7, s6
-; VI-NEXT:    v_subrev_u32_e32 v3, vcc, v6, v3
+; VI-NEXT:    v_sub_u32_e32 v3, vcc, v3, v6
 ; VI-NEXT:    v_add_u32_e32 v5, vcc, v3, v4
 ; VI-NEXT:    v_mad_u64_u32 v[3:4], s[4:5], v6, v5, 0
 ; VI-NEXT:    v_mul_hi_u32 v8, v6, v2
 ; VI-NEXT:    v_add_u32_e32 v8, vcc, v8, v3
 ; VI-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v7, v2, 0
 ; VI-NEXT:    v_addc_u32_e32 v9, vcc, 0, v4, vcc
 ; VI-NEXT:    v_mad_u64_u32 v[4:5], s[4:5], v7, v5, 0
 ; VI-NEXT:    v_add_u32_e32 v2, vcc, v8, v2

 ; GCN-NEXT:    v_mul_f32_e32 v2, 0x5f7ffffc, v2
 ; GCN-NEXT:    v_mul_f32_e32 v3, 0x2f800000, v2
 ; GCN-NEXT:    v_trunc_f32_e32 v3, v3
 ; GCN-NEXT:    v_mac_f32_e32 v2, 0xcf800000, v3
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v6, v2
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v7, v3
 ; GCN-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v6, s6, 0
 ; GCN-NEXT:    v_mul_lo_u32 v4, v7, s6
-; GCN-NEXT:    v_subrev_u32_e32 v3, vcc, v6, v3
-; GCN-NEXT:    v_add_u32_e32 v8, vcc, v3, v4
+; GCN-NEXT:    v_sub_u32_e32 v3, vcc, v3, v6
+; GCN-NEXT:    v_add_u32_e32 v8, vcc, v4, v3
 ; GCN-NEXT:    v_mul_hi_u32 v5, v6, v2
 ; GCN-NEXT:    v_mad_u64_u32 v[3:4], s[4:5], v6, v8, 0
 ; GCN-NEXT:    v_add_u32_e32 v9, vcc, v5, v3
 ; GCN-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v7, v2, 0
 ; GCN-NEXT:    v_addc_u32_e32 v10, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_mad_u64_u32 v[4:5], s[4:5], v7, v8, 0
 ; GCN-NEXT:    v_add_u32_e32 v2, vcc, v9, v2
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, v10, v3, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_u32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v3, vcc
 ; GCN-NEXT:    v_add_u32_e32 v6, vcc, v6, v2
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, v7, v3, vcc
 ; GCN-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v6, s6, 0
 ; GCN-NEXT:    v_mul_lo_u32 v4, v7, s6
-; GCN-NEXT:    v_subrev_u32_e32 v3, vcc, v6, v3
+; GCN-NEXT:    v_sub_u32_e32 v3, vcc, v3, v6
 ; GCN-NEXT:    v_add_u32_e32 v5, vcc, v3, v4
 ; GCN-NEXT:    v_mad_u64_u32 v[3:4], s[4:5], v6, v5, 0
 ; GCN-NEXT:    v_mul_hi_u32 v8, v6, v2
 ; GCN-NEXT:    v_add_u32_e32 v8, vcc, v8, v3
 ; GCN-NEXT:    v_mad_u64_u32 v[2:3], s[4:5], v7, v2, 0
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_mad_u64_u32 v[4:5], s[4:5], v7, v5, 0
 ; GCN-NEXT:    v_add_u32_e32 v2, vcc, v8, v2

llvm/test/CodeGen/AMDGPU/udiv64.ll

 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s4, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s5, v0
 ; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s4, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_subrev_i32_e64 v5, s[0:1], s8, v3
 ; GCN-NEXT:    v_subbrev_u32_e64 v4, s[0:1], 0, v4, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s9, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s8, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], s9, 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_add_i32_e64 v5, s[0:1], 1, 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_add_i32_e64 v7, s[0:1], 2, 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_cndmask_b32_e64 v4, v5, v7, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v5, v6, v8, s[0:1]
 ; GCN-NEXT:    v_mov_b32_e32 v6, s3
 ; GCN-NEXT:    v_subb_u32_e32 v2, vcc, v6, v2, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s9, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s8, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s9, 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:    v_cndmask_b32_e32 v1, v1, v5, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v4, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_udiv_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[4:5], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
-; GCN-IR-NEXT:    s_flbit_i32_b32 s12, s4
-; GCN-IR-NEXT:    s_add_i32 s14, s12, 32
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[8:9], s[10:11]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s5
-; GCN-IR-NEXT:    s_min_u32 s10, s14, s8
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
-; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
-; GCN-IR-NEXT:    s_min_u32 s14, s8, s9
-; GCN-IR-NEXT:    s_sub_u32 s8, s10, s14
-; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 63
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[4:5], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[2:3], 0
+; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s4
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s5
+; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_min_u32 s10, s10, s11
+; GCN-IR-NEXT:    s_min_u32 s14, s6, s7
+; GCN-IR-NEXT:    s_sub_u32 s12, s10, s14
+; GCN-IR-NEXT:    s_subb_u32 s13, 0, 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[12:13], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[12:13], 63
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[8:9], s[16:17]
+; GCN-IR-NEXT:    s_and_b64 s[8:9], s[16:17], exec
+; GCN-IR-NEXT:    s_cselect_b32 s9, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s8, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[18:19]
+; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
-; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[12:13], 0
-; GCN-IR-NEXT:    s_sub_i32 s8, 63, s8
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
-; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[2:3], s8
+; GCN-IR-NEXT:    s_add_u32 s16, s12, 1
+; GCN-IR-NEXT:    s_addc_u32 s17, s13, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[16:17], 0
+; GCN-IR-NEXT:    s_sub_i32 s12, 63, s12
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[8:9]
+; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[2:3], s12
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[2:3], s12
+; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[2:3], s16
 ; GCN-IR-NEXT:    s_add_u32 s15, s4, -1
 ; GCN-IR-NEXT:    s_addc_u32 s16, s5, -1
 ; GCN-IR-NEXT:    s_not_b64 s[2:3], s[10:11]
 ; GCN-IR-NEXT:    s_add_u32 s2, s2, s14
 ; GCN-IR-NEXT:    s_addc_u32 s3, s3, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:  .LBB0_3: ; %udiv-do-while

 ; GCN-IR-NEXT:    s_add_u32 s2, s2, 1
 ; GCN-IR-NEXT:    s_addc_u32 s3, s3, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[2:3], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_3
 ; GCN-IR-NEXT:  .LBB0_4: ; %Flow6
 ; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[8:9], 1
-; GCN-IR-NEXT:    s_or_b64 s[2:3], s[6:7], s[2:3]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-IR-NEXT:    s_branch .LBB0_6
-; GCN-IR-NEXT:  .LBB0_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[12:13]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[12:13]
-; GCN-IR-NEXT:  .LBB0_6: ; %udiv-end
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[2:3]
+; GCN-IR-NEXT:  .LBB0_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s8
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s9
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = udiv i64 %x, %y
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define i64 @v_test_udiv_i64(i64 %x, i64 %y) {
; GCN-IR-NEXT:    s_endpgm
   %result = udiv i64 %1, %2
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_test_udiv24_i48(i48 addrspace(1)* %out, i48 %x, i48 %y) {
 ; GCN-LABEL: s_test_udiv24_i48:
 ; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
+; GCN-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0xd
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-NEXT:    v_mov_b32_e32 v2, 0x4f800000
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-NEXT:    s_mov_b32 s6, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_and_b32 s4, s4, 0xff000000
-; GCN-NEXT:    s_and_b32 s5, s5, 0xffff
-; GCN-NEXT:    v_mov_b32_e32 v0, s4
-; GCN-NEXT:    v_alignbit_b32 v0, s5, v0, 24
+; GCN-NEXT:    s_and_b32 s0, s2, 0xff000000
+; GCN-NEXT:    s_and_b32 s1, s3, 0xffff
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    v_alignbit_b32 v0, s1, v0, 24
 ; GCN-NEXT:    v_cvt_f32_u32_e32 v1, v0
-; GCN-NEXT:    s_and_b32 s8, s3, 0xffff
-; GCN-NEXT:    s_and_b32 s9, s2, 0xff000000
-; GCN-NEXT:    s_lshr_b64 s[2:3], s[4:5], 24
+; GCN-NEXT:    s_and_b32 s7, s7, 0xffff
+; GCN-NEXT:    s_and_b32 s6, s6, 0xff000000
+; GCN-NEXT:    s_lshr_b64 s[0:1], s[0:1], 24
 ; GCN-NEXT:    v_mac_f32_e32 v1, 0, v2
 ; GCN-NEXT:    v_rcp_f32_e32 v1, v1
-; GCN-NEXT:    s_sub_u32 s2, 0, s2
-; GCN-NEXT:    s_subb_u32 s3, 0, s3
-; GCN-NEXT:    s_mov_b32 s4, s0
+; GCN-NEXT:    s_sub_u32 s8, 0, s0
+; GCN-NEXT:    s_subb_u32 s9, 0, s1
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-NEXT:    v_mul_f32_e32 v1, 0x5f7ffffc, v1
 ; GCN-NEXT:    v_mul_f32_e32 v2, 0x2f800000, v1
 ; GCN-NEXT:    v_trunc_f32_e32 v2, v2
 ; GCN-NEXT:    v_mac_f32_e32 v1, 0xcf800000, v2
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    v_mul_lo_u32 v3, s2, v2
-; GCN-NEXT:    v_mul_hi_u32 v4, s2, v1
-; GCN-NEXT:    v_mul_lo_u32 v5, s3, v1
-; GCN-NEXT:    v_mul_lo_u32 v6, s2, v1
+; GCN-NEXT:    s_mov_b32 s2, -1
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    v_mul_lo_u32 v3, s8, v2
+; GCN-NEXT:    v_mul_hi_u32 v4, s8, v1
+; GCN-NEXT:    v_mul_lo_u32 v5, s9, v1
+; GCN-NEXT:    v_mul_lo_u32 v6, s8, v1
+; GCN-NEXT:    s_mov_b32 s1, s5
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GCN-NEXT:    v_mul_lo_u32 v4, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v6
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v2, v3
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v7, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v7, v2, v6
 ; GCN-NEXT:    v_mul_hi_u32 v6, v2, v6
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v7
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v5, v6, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
-; GCN-NEXT:    v_mul_lo_u32 v3, s2, v2
-; GCN-NEXT:    v_mul_hi_u32 v4, s2, v1
-; GCN-NEXT:    v_mul_lo_u32 v5, s3, v1
+; GCN-NEXT:    v_mul_lo_u32 v3, s8, v2
+; GCN-NEXT:    v_mul_hi_u32 v4, s8, v1
+; GCN-NEXT:    v_mul_lo_u32 v5, s9, v1
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
-; GCN-NEXT:    v_mul_lo_u32 v4, s2, v1
+; GCN-NEXT:    v_mul_lo_u32 v4, s8, v1
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GCN-NEXT:    v_mul_lo_u32 v7, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v1, v4
 ; GCN-NEXT:    v_mul_hi_u32 v9, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v6, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v4, v2, v4
 ; GCN-NEXT:    v_mul_hi_u32 v5, v2, v3
 ; GCN-NEXT:    v_add_i32_e32 v7, vcc, v8, v7
 ; GCN-NEXT:    v_addc_u32_e32 v8, vcc, 0, v9, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v3, v2, v3
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v7, v4
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, v8, v6, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
-; GCN-NEXT:    v_mov_b32_e32 v3, s9
+; GCN-NEXT:    v_mov_b32_e32 v3, s6
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, v2, v4, vcc
-; GCN-NEXT:    v_alignbit_b32 v3, s8, v3, 24
+; GCN-NEXT:    v_alignbit_b32 v3, s7, v3, 24
 ; GCN-NEXT:    v_mul_lo_u32 v4, v3, v2
 ; GCN-NEXT:    v_mul_hi_u32 v1, v3, v1
 ; GCN-NEXT:    v_mul_hi_u32 v2, v3, v2
 ; GCN-NEXT:    v_mov_b32_e32 v5, 0
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, 0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, 0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, 0, v2, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v1
-; GCN-NEXT:    v_add_i32_e32 v4, vcc, 2, v1
+; GCN-NEXT:    v_add_i32_e32 v4, vcc, 1, v1
 ; GCN-NEXT:    v_mul_lo_u32 v10, v0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v2, vcc
-; GCN-NEXT:    v_add_i32_e32 v8, vcc, 1, v1
+; GCN-NEXT:    v_add_i32_e32 v8, vcc, 2, v1
 ; GCN-NEXT:    v_addc_u32_e32 v9, vcc, 0, v2, vcc
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, v6, v7
 ; GCN-NEXT:    v_sub_i32_e32 v3, vcc, v3, v10
 ; GCN-NEXT:    v_subb_u32_e32 v6, vcc, 0, v6, vcc
 ; GCN-NEXT:    v_sub_i32_e32 v7, vcc, v3, v0
 ; GCN-NEXT:    v_subbrev_u32_e32 v10, vcc, 0, v6, vcc
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v7, v0
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v10
-; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v3, v0
 ; GCN-NEXT:    v_cndmask_b32_e32 v7, -1, v7, vcc
-; GCN-NEXT:    v_cndmask_b32_e64 v0, 0, -1, s[0:1]
-; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v6
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v7
-; GCN-NEXT:    v_cndmask_b32_e64 v0, -1, v0, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v4, v8, v4, vcc
-; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v1, v4, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v1, v9, v5, vcc
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v2, v1, s[0:1]
-; GCN-NEXT:    buffer_store_short v1, off, s[4:7], 0 offset:4
-; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
+; GCN-NEXT:    v_cndmask_b32_e32 v5, v5, v9, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v4, v4, v8, vcc
+; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v3, v0
+; GCN-NEXT:    v_cndmask_b32_e64 v0, 0, -1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v6
+; GCN-NEXT:    v_cndmask_b32_e32 v0, -1, v0, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v1, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v2, v5, vcc
+; GCN-NEXT:    buffer_store_short v1, off, s[0:3], 0 offset:4
+; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_udiv24_i48:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0xd
+; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_and_b32 s3, s7, 0xffff
 ; GCN-IR-NEXT:    s_and_b32 s2, s6, 0xff000000
 ; GCN-IR-NEXT:    s_and_b32 s1, s1, 0xffff
 ; GCN-IR-NEXT:    s_and_b32 s0, s0, 0xff000000
 ; GCN-IR-NEXT:    s_lshr_b64 s[8:9], s[2:3], 24
-; GCN-IR-NEXT:    s_lshr_b64 s[2:3], s[0:1], 24
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[8:9], 0
-; GCN-IR-NEXT:    s_mov_b64 s[0:1], 0
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[6:7], s[10:11]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
-; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
-; GCN-IR-NEXT:    s_min_u32 s10, s6, s7
-; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s8
-; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s9
-; GCN-IR-NEXT:    s_min_u32 s14, s6, s7
+; GCN-IR-NEXT:    s_lshr_b64 s[0:1], s[0:1], 24
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[2:3], s[0:1], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[8:9], 0
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[2:3], s[6:7]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s2, s0
+; GCN-IR-NEXT:    s_add_i32 s2, s2, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s3, s1
+; GCN-IR-NEXT:    s_min_u32 s10, s2, s3
+; GCN-IR-NEXT:    s_flbit_i32_b32 s2, s8
+; GCN-IR-NEXT:    s_add_i32 s2, s2, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s3, s9
+; GCN-IR-NEXT:    s_min_u32 s14, s2, s3
 ; GCN-IR-NEXT:    s_sub_u32 s6, s10, s14
 ; GCN-IR-NEXT:    s_subb_u32 s7, 0, 0
 ; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[6:7], 63
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[6:7], 63
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[16:17]
+; GCN-IR-NEXT:    s_and_b64 s[12:13], s[16:17], exec
+; GCN-IR-NEXT:    s_cselect_b32 s13, 0, s9
+; GCN-IR-NEXT:    s_cselect_b32 s12, 0, s8
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[18:19]
+; GCN-IR-NEXT:    s_mov_b64 s[2:3], 0
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
-; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
 ; GCN-IR-NEXT:    s_add_u32 s12, s6, 1
 ; GCN-IR-NEXT:    s_addc_u32 s13, s7, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[12:13], 0
 ; GCN-IR-NEXT:    s_sub_i32 s6, 63, s6
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[8:9], s6
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
 ; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[8:9], s12
-; GCN-IR-NEXT:    s_add_u32 s15, s2, -1
-; GCN-IR-NEXT:    s_addc_u32 s16, s3, -1
-; GCN-IR-NEXT:    s_not_b64 s[0:1], s[10:11]
-; GCN-IR-NEXT:    s_add_u32 s8, s0, s14
-; GCN-IR-NEXT:    s_addc_u32 s9, s1, 0
+; GCN-IR-NEXT:    s_add_u32 s15, s0, -1
+; GCN-IR-NEXT:    s_addc_u32 s16, s1, -1
+; GCN-IR-NEXT:    s_not_b64 s[2:3], s[10:11]
+; GCN-IR-NEXT:    s_add_u32 s8, s2, s14
+; GCN-IR-NEXT:    s_addc_u32 s9, s3, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
-; GCN-IR-NEXT:    s_mov_b32 s1, 0
+; GCN-IR-NEXT:    s_mov_b32 s3, 0
 ; GCN-IR-NEXT:  .LBB7_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    s_lshl_b64 s[12:13], s[12:13], 1
-; GCN-IR-NEXT:    s_lshr_b32 s0, s7, 31
+; GCN-IR-NEXT:    s_lshr_b32 s2, s7, 31
 ; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[0:1]
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[2:3]
 ; GCN-IR-NEXT:    s_or_b64 s[6:7], s[10:11], s[6:7]
-; GCN-IR-NEXT:    s_sub_u32 s0, s15, s12
-; GCN-IR-NEXT:    s_subb_u32 s0, s16, s13
-; GCN-IR-NEXT:    s_ashr_i32 s10, s0, 31
+; GCN-IR-NEXT:    s_sub_u32 s2, s15, s12
+; GCN-IR-NEXT:    s_subb_u32 s2, s16, s13
+; GCN-IR-NEXT:    s_ashr_i32 s10, s2, 31
 ; GCN-IR-NEXT:    s_mov_b32 s11, s10
-; GCN-IR-NEXT:    s_and_b32 s0, s10, 1
-; GCN-IR-NEXT:    s_and_b64 s[10:11], s[10:11], s[2:3]
+; GCN-IR-NEXT:    s_and_b32 s2, s10, 1
+; GCN-IR-NEXT:    s_and_b64 s[10:11], s[10:11], s[0:1]
 ; GCN-IR-NEXT:    s_sub_u32 s12, s12, s10
 ; GCN-IR-NEXT:    s_subb_u32 s13, s13, s11
 ; GCN-IR-NEXT:    s_add_u32 s8, s8, 1
 ; GCN-IR-NEXT:    s_addc_u32 s9, s9, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 0
-; GCN-IR-NEXT:    s_mov_b64 s[10:11], s[0:1]
+; GCN-IR-NEXT:    s_mov_b64 s[10:11], s[2:3]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_3
 ; GCN-IR-NEXT:  .LBB7_4: ; %Flow3
-; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[0:1], s[0:1], s[2:3]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s1
-; GCN-IR-NEXT:    s_branch .LBB7_6
-; GCN-IR-NEXT:  .LBB7_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s9
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[12:13]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s8
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[12:13]
-; GCN-IR-NEXT:  .LBB7_6: ; %udiv-end
+; GCN-IR-NEXT:    s_lshl_b64 s[0:1], s[6:7], 1
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[2:3], s[0:1]
+; GCN-IR-NEXT:  .LBB7_5: ; %udiv-end
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    buffer_store_short v1, off, s[4:7], 0 offset:4
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s13
+; GCN-IR-NEXT:    buffer_store_short v0, off, s[4:7], 0 offset:4
+; GCN-IR-NEXT:    s_waitcnt expcnt(0)
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s12
 ; GCN-IR-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
   %1 = lshr i48 %x, 24
   %2 = lshr i48 %y, 24
   %result = udiv i48 %1, %2
   store i48 %result, i48 addrspace(1)* %out
   ret void
 }

 ; GCN-NEXT:    v_subrev_i32_e64 v4, s[0:1], s2, v2
 ; GCN-NEXT:    v_subbrev_u32_e64 v3, s[0:1], 0, v3, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s3, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, s[0:1]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], s3, v3
 ; GCN-NEXT:    v_cndmask_b32_e64 v3, v5, v4, s[0:1]
-; GCN-NEXT:    v_add_i32_e64 v4, s[0:1], 2, v0
+; GCN-NEXT:    v_add_i32_e64 v4, s[0:1], 1, v0
 ; GCN-NEXT:    v_addc_u32_e64 v5, s[0:1], 0, 0, s[0:1]
-; GCN-NEXT:    v_add_i32_e64 v6, s[0:1], 1, v0
+; GCN-NEXT:    v_add_i32_e64 v6, s[0:1], 2, v0
 ; GCN-NEXT:    v_addc_u32_e64 v7, s[0:1], 0, 0, s[0:1]
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v3
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v3, v7, v5, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v6, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v4, v5, v7, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s3, v1
 ; GCN-NEXT:    v_cndmask_b32_e32 v1, v5, v2, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v2, v6, v4, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v3, vcc
-; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v1, 0, v4, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_udiv_k_num_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
-; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
-; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
-; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
-; GCN-IR-NEXT:    s_add_u32 s6, s8, 0xffffffc5
-; GCN-IR-NEXT:    s_addc_u32 s7, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[4:5], s[10:11]
-; GCN-IR-NEXT:    s_or_b64 s[4:5], s[10:11], s[12:13]
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
+; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_min_u32 s8, s8, s9
+; GCN-IR-NEXT:    s_add_u32 s10, s8, 0xffffffc5
+; GCN-IR-NEXT:    s_addc_u32 s11, 0, -1
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[10:11], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[10:11], 63
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[6:7], s[12:13]
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[12:13], exec
+; GCN-IR-NEXT:    s_cselect_b32 s6, 0, 24
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[14:15]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
+; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
-; GCN-IR-NEXT:    s_addc_u32 s11, s7, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[10:11], 0
-; GCN-IR-NEXT:    s_sub_i32 s6, 63, s6
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], 24, s6
+; GCN-IR-NEXT:    s_add_u32 s12, s10, 1
+; GCN-IR-NEXT:    s_addc_u32 s13, s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[12:13], 0
+; GCN-IR-NEXT:    s_sub_i32 s9, 63, s10
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[6:7]
+; GCN-IR-NEXT:    s_lshl_b64 s[6:7], 24, s9
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[10:11], 24, s10
+; GCN-IR-NEXT:    s_lshr_b64 s[10:11], 24, s12
 ; GCN-IR-NEXT:    s_add_u32 s14, s2, -1
 ; GCN-IR-NEXT:    s_addc_u32 s15, s3, -1
 ; GCN-IR-NEXT:    s_sub_u32 s8, 58, s8
 ; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:  .LBB8_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1

 ; GCN-IR-NEXT:    s_add_u32 s8, s8, 1
 ; GCN-IR-NEXT:    s_addc_u32 s9, s9, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[8:9], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], s[4:5]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB8_3
 ; GCN-IR-NEXT:  .LBB8_4: ; %Flow5
 ; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[2:3], s[4:5], s[2:3]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-IR-NEXT:    s_branch .LBB8_6
-; GCN-IR-NEXT:  .LBB8_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, 0
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, 24, 0, s[10:11]
-; GCN-IR-NEXT:  .LBB8_6: ; %udiv-end
+; GCN-IR-NEXT:    s_or_b64 s[6:7], s[4:5], s[2:3]
+; GCN-IR-NEXT:  .LBB8_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = udiv i64 24, %x
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 ; define i64 @v_test_udiv_k_num_i64(i64 %x) {

 }
 
 define amdgpu_kernel void @s_test_udiv_k_den_i64(i64 addrspace(1)* %out, i64 %x) {
 ; GCN-LABEL: s_test_udiv_k_den_i64:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    v_mov_b32_e32 v0, 0x4f800000
 ; GCN-NEXT:    v_madak_f32 v0, 0, v0, 0x41c00000
 ; GCN-NEXT:    v_rcp_f32_e32 v0, v0
-; GCN-NEXT:    s_movk_i32 s4, 0xffe8
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
+; GCN-NEXT:    s_movk_i32 s8, 0xffe8
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GCN-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GCN-NEXT:    v_trunc_f32_e32 v1, v1
 ; 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:    s_mov_b32 s2, -1
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    s_mov_b32 s6, -1
-; GCN-NEXT:    v_mul_hi_u32 v2, v0, s4
-; GCN-NEXT:    v_mul_lo_u32 v4, v1, s4
-; GCN-NEXT:    v_mul_lo_u32 v3, v0, s4
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    v_mul_hi_u32 v2, v0, s8
+; GCN-NEXT:    v_mul_lo_u32 v4, v1, s8
+; GCN-NEXT:    v_mul_lo_u32 v3, v0, s8
+; GCN-NEXT:    s_mov_b32 s1, s5
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, v3
 ; GCN-NEXT:    v_mul_lo_u32 v4, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v6, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v3, v1, v3
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v5, v3, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GCN-NEXT:    v_mul_hi_u32 v2, v0, s4
-; GCN-NEXT:    v_mul_lo_u32 v3, v1, s4
-; GCN-NEXT:    v_mul_lo_u32 v4, v0, s4
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_mul_hi_u32 v2, v0, s8
+; GCN-NEXT:    v_mul_lo_u32 v3, v1, s8
+; GCN-NEXT:    v_mul_lo_u32 v4, v0, s8
+; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_mul_lo_u32 v3, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v6, v1, v4
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v6
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v5, v4, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GCN-NEXT:    v_mul_lo_u32 v2, s2, v1
-; GCN-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GCN-NEXT:    v_mul_hi_u32 v4, s2, v1
-; GCN-NEXT:    v_mul_hi_u32 v5, s3, v1
-; GCN-NEXT:    v_mul_lo_u32 v1, s3, v1
+; 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, 0, v4, vcc
-; GCN-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GCN-NEXT:    v_mul_hi_u32 v0, s3, v0
+; 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, v2, v4
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v4, v1, 24
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, 24
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, 2, v0
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
 ; GCN-NEXT:    v_mul_lo_u32 v8, v0, 24
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v1, vcc
-; GCN-NEXT:    v_add_i32_e32 v6, vcc, 1, v0
+; GCN-NEXT:    v_add_i32_e32 v6, vcc, 2, v0
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
-; GCN-NEXT:    v_mov_b32_e32 v5, s3
-; GCN-NEXT:    v_sub_i32_e32 v8, vcc, s2, v8
+; GCN-NEXT:    v_mov_b32_e32 v5, s7
+; GCN-NEXT:    v_sub_i32_e32 v8, vcc, s6, v8
 ; GCN-NEXT:    v_subb_u32_e32 v4, vcc, v5, v4, vcc
 ; GCN-NEXT:    v_subrev_i32_e32 v5, vcc, 24, v8
 ; GCN-NEXT:    v_subbrev_u32_e32 v9, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, 23, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v9
 ; GCN-NEXT:    v_cndmask_b32_e32 v5, -1, v5, vcc
-; GCN-NEXT:    v_cmp_lt_u32_e64 s[0:1], 23, v8
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v4
-; GCN-NEXT:    v_cndmask_b32_e64 v4, -1, v5, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e32 v3, v7, v3, vcc
-; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GCN-NEXT:    v_cndmask_b32_e32 v2, v6, v2, vcc
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v7, vcc
+; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, 23, v8
+; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
+; GCN-NEXT:    v_cndmask_b32_e32 v4, -1, v5, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_udiv_k_den_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
 ; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
 ; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
 ; GCN-IR-NEXT:    s_min_u32 s10, s6, s7
-; GCN-IR-NEXT:    s_sub_u32 s6, 59, s10
-; GCN-IR-NEXT:    s_subb_u32 s7, 0, 0
+; GCN-IR-NEXT:    s_sub_u32 s8, 59, s10
+; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[8:9], s[6:7], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_or_b64 s[8:9], s[4:5], s[8:9]
-; GCN-IR-NEXT:    s_or_b64 s[4:5], s[8:9], s[12:13]
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[6:7], s[8:9], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[8:9], 63
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[4:5], s[6:7]
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s7, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s6, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[4:5], s[12:13]
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB11_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s8, s6, 1
-; GCN-IR-NEXT:    s_addc_u32 s9, s7, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[8:9], 0
-; GCN-IR-NEXT:    s_sub_i32 s6, 63, s6
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[2:3], s6
+; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
+; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[12:13], 0
+; GCN-IR-NEXT:    s_sub_i32 s8, 63, s8
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[6:7]
+; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[2:3], s8
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB11_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[8:9], s[2:3], s8
+; GCN-IR-NEXT:    s_lshr_b64 s[8:9], s[2:3], s12
 ; GCN-IR-NEXT:    s_add_u32 s2, s10, 0xffffffc4
 ; GCN-IR-NEXT:    s_addc_u32 s3, 0, -1
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], 0
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:  .LBB11_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[8:9], 1
 ; GCN-IR-NEXT:    s_lshr_b32 s4, s7, 31

 ; GCN-IR-NEXT:    s_add_u32 s2, s2, 1
 ; GCN-IR-NEXT:    s_addc_u32 s3, s3, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[2:3], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[10:11], s[4:5]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[12:13]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB11_3
 ; GCN-IR-NEXT:  .LBB11_4: ; %Flow5
 ; GCN-IR-NEXT:    s_lshl_b64 s[2:3], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[2:3], s[4:5], s[2:3]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s3
-; GCN-IR-NEXT:    s_branch .LBB11_6
-; GCN-IR-NEXT:  .LBB11_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[8:9]
-; GCN-IR-NEXT:  .LBB11_6: ; %udiv-end
+; GCN-IR-NEXT:    s_or_b64 s[6:7], s[4:5], s[2:3]
+; GCN-IR-NEXT:  .LBB11_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
 ; GCN-IR-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-IR-NEXT:    s_mov_b32 s2, -1
+; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
 ; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = udiv i64 %x, 24
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define i64 @v_test_udiv_k_den_i64(i64 %x) {

llvm/test/CodeGen/AMDGPU/udivrem.ll

 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ; R600-NEXT:     ADD_INT * T1.W, PV.W, 1,
 ; R600-NEXT:     CNDE_INT T2.X, T3.W, T0.W, PV.W,
 ; R600-NEXT:     LSHR * T3.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX6-LABEL: test_udivrem:
 ; GFX6:       ; %bb.0:
-; GFX6-NEXT:    s_load_dword s2, s[0:1], 0x26
+; GFX6-NEXT:    s_load_dword s8, s[0:1], 0x26
+; GFX6-NEXT:    s_load_dword s9, s[0:1], 0x1d
 ; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
-; GFX6-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x13
+; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x13
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s6, -1
-; GFX6-NEXT:    s_mov_b32 s10, s6
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s2
-; GFX6-NEXT:    s_sub_i32 s3, 0, s2
-; GFX6-NEXT:    s_mov_b32 s11, s7
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s8
+; GFX6-NEXT:    s_sub_i32 s2, 0, s8
+; GFX6-NEXT:    s_mov_b32 s3, s7
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
-; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x1d
+; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX6-NEXT:    s_mov_b32 s2, s6
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
-; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX6-NEXT:    v_mul_hi_u32 v0, s3, v0
-; GFX6-NEXT:    v_mul_lo_u32 v1, v0, s2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s3, v1
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s2, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v1, v1, v2, s[0:1]
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
-; GFX6-NEXT:    v_cmp_le_u32_e64 s[0:1], s2, v1
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s2, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s9, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s10, v0
+; GFX6-NEXT:    s_mul_i32 s10, s10, s8
+; GFX6-NEXT:    s_sub_i32 s9, s9, s10
+; GFX6-NEXT:    s_sub_i32 s10, s9, s8
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s9, s8
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    s_cselect_b32 s9, s10, s9
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    s_sub_i32 s10, s9, s8
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, 1, v0
+; GFX6-NEXT:    s_cmp_ge_u32 s9, s8
+; GFX6-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX6-NEXT:    s_cselect_b32 s8, s10, s9
 ; GFX6-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GFX6-NEXT:    s_waitcnt expcnt(0)
-; GFX6-NEXT:    v_cndmask_b32_e64 v0, v1, v2, s[0:1]
-; GFX6-NEXT:    buffer_store_dword v0, off, s[8:11], 0
+; GFX6-NEXT:    v_mov_b32_e32 v0, s8
+; GFX6-NEXT:    buffer_store_dword v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_udivrem:
 ; GFX8:       ; %bb.0:
-; GFX8-NEXT:    s_load_dword s6, s[0:1], 0x98
-; GFX8-NEXT:    s_load_dword s7, s[0:1], 0x74
-; GFX8-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x4c
+; GFX8-NEXT:    s_load_dword s4, s[0:1], 0x98
+; GFX8-NEXT:    s_load_dword s5, s[0:1], 0x74
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX8-NEXT:    v_cvt_f32_u32_e32 v0, s6
-; GFX8-NEXT:    s_sub_i32 s2, 0, s6
+; GFX8-NEXT:    v_cvt_f32_u32_e32 v0, s4
+; GFX8-NEXT:    s_sub_i32 s2, 0, s4
 ; GFX8-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX8-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX8-NEXT:    v_cvt_u32_f32_e32 v0, v0
 ; GFX8-NEXT:    v_mul_lo_u32 v1, s2, v0
 ; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x4c
 ; GFX8-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
-; GFX8-NEXT:    v_mul_hi_u32 v2, s7, v0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_mov_b32_e32 v2, s0
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
+; GFX8-NEXT:    v_mul_hi_u32 v4, s5, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_mov_b32_e32 v1, s3
-; GFX8-NEXT:    v_mul_lo_u32 v3, v2, s6
-; GFX8-NEXT:    v_add_u32_e32 v4, vcc, 1, v2
-; GFX8-NEXT:    v_sub_u32_e32 v3, vcc, s7, v3
-; GFX8-NEXT:    v_cmp_le_u32_e64 s[0:1], s6, v3
-; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s6, v3
-; GFX8-NEXT:    v_cndmask_b32_e64 v3, v3, v4, s[0:1]
-; GFX8-NEXT:    v_add_u32_e32 v4, vcc, 1, v2
-; GFX8-NEXT:    v_cmp_le_u32_e64 s[0:1], s6, v3
-; GFX8-NEXT:    v_cndmask_b32_e64 v2, v2, v4, s[0:1]
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s6, v3
-; GFX8-NEXT:    flat_store_dword v[0:1], v2
-; GFX8-NEXT:    v_mov_b32_e32 v0, s4
-; GFX8-NEXT:    v_cndmask_b32_e64 v2, v3, v4, s[0:1]
-; GFX8-NEXT:    v_mov_b32_e32 v1, s5
-; GFX8-NEXT:    flat_store_dword v[0:1], v2
+; GFX8-NEXT:    v_readfirstlane_b32 s0, v4
+; GFX8-NEXT:    s_mul_i32 s0, s0, s4
+; GFX8-NEXT:    s_sub_i32 s0, s5, s0
+; GFX8-NEXT:    s_sub_i32 s1, s0, s4
+; GFX8-NEXT:    v_add_u32_e32 v5, vcc, 1, v4
+; GFX8-NEXT:    s_cmp_ge_u32 s0, s4
+; GFX8-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX8-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX8-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
+; GFX8-NEXT:    s_sub_i32 s1, s0, s4
+; GFX8-NEXT:    v_add_u32_e32 v5, vcc, 1, v4
+; GFX8-NEXT:    s_cmp_ge_u32 s0, s4
+; GFX8-NEXT:    s_cselect_b64 vcc, -1, 0
+; GFX8-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
+; GFX8-NEXT:    s_cselect_b32 s0, s1, s0
+; GFX8-NEXT:    flat_store_dword v[0:1], v4
+; GFX8-NEXT:    v_mov_b32_e32 v0, s0
+; GFX8-NEXT:    flat_store_dword v[2:3], v0
 ; GFX8-NEXT:    s_endpgm
   %result0 = udiv i32 %x, %y
   store i32 %result0, i32 addrspace(1)* %out0
   %result1 = urem i32 %x, %y
   store i32 %result1, i32 addrspace(1)* %out1
   ret void
 }
 

 ; R600-NEXT:     CNDE_INT T0.X, PV.W, T0.Z, PS,
 ; R600-NEXT:     LSHR * T1.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX6-LABEL: test_udivrem_v2:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0xb
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s6
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s7
 ; GFX6-NEXT:    s_sub_i32 s2, 0, s6
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s7
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
+; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_mul_lo_u32 v2, s2, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s6
+; GFX6-NEXT:    s_sub_i32 s2, s4, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s6
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s6
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s6
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s6
+; GFX6-NEXT:    s_cselect_b32 s4, s3, s2
 ; GFX6-NEXT:    s_sub_i32 s2, 0, s7
-; GFX6-NEXT:    v_mul_lo_u32 v3, s2, v1
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
-; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s7
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s5, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s6, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s7, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s6, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, s7, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_readfirstlane_b32 s6, v0
+; GFX6-NEXT:    s_mul_i32 s6, s6, s7
+; GFX6-NEXT:    s_sub_i32 s5, s5, s6
+; GFX6-NEXT:    s_sub_i32 s6, s5, s7
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s7
+; GFX6-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX6-NEXT:    s_sub_i32 s6, s5, s7
+; GFX6-NEXT:    s_cmp_ge_u32 s5, s7
+; GFX6-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
+; GFX6-NEXT:    v_mov_b32_e32 v1, s5
 ; GFX6-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_udivrem_v2:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x2c
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
 ; GFX8-NEXT:    s_sub_i32 s3, s2, s6
 ; GFX8-NEXT:    s_cmp_ge_u32 s2, s6
 ; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
 ; GFX8-NEXT:    s_sub_i32 s3, 0, s7
 ; GFX8-NEXT:    v_mul_lo_u32 v0, s3, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s0
 ; GFX8-NEXT:    v_mul_hi_u32 v0, v1, v0
-; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
 ; GFX8-NEXT:    v_mul_hi_u32 v1, s5, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX8-NEXT:    v_readfirstlane_b32 s2, v1
 ; GFX8-NEXT:    s_mul_i32 s2, s2, s7
 ; GFX8-NEXT:    s_sub_i32 s2, s5, s2
 ; GFX8-NEXT:    s_sub_i32 s3, s2, s7
 ; GFX8-NEXT:    s_cmp_ge_u32 s2, s7
 ; GFX8-NEXT:    s_cselect_b32 s2, s3, s2

 ; R600-NEXT:     CNDE_INT T3.X, PV.W, T0.Y, PS,
 ; R600-NEXT:     LSHR * T0.X, KC0[2].Y, literal.x,
 ; R600-NEXT:    2(2.802597e-45), 0(0.000000e+00)
 ;
 ; GFX6-LABEL: test_udivrem_v4:
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0xd
 ; GFX6-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x9
-; GFX6-NEXT:    s_mov_b32 s3, 0xf000
-; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v0, s8
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX6-NEXT:    s_sub_i32 s12, 0, s8
-; GFX6-NEXT:    s_sub_i32 s13, 0, s9
+; GFX6-NEXT:    s_sub_i32 s2, 0, s8
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s9
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v3, s10
-; GFX6-NEXT:    v_cvt_f32_u32_e32 v5, s11
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX6-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX6-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX6-NEXT:    v_mul_lo_u32 v2, s12, v0
-; GFX6-NEXT:    v_mul_lo_u32 v4, s13, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v4
-; GFX6-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s8
-; GFX6-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v3
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s9
-; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v3, vcc, s8, v0
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX6-NEXT:    s_sub_i32 s4, 0, s10
-; GFX6-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v3, s4, v2
-; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s5, v1
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s9, v1
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_mul_hi_u32 v3, v2, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_rcp_iflag_f32_e32 v4, v5
-; GFX6-NEXT:    s_sub_i32 s4, 0, s11
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v4
-; GFX6-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s9, v1
-; GFX6-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX6-NEXT:    v_mul_lo_u32 v5, s4, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX6-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX6-NEXT:    v_mul_lo_u32 v2, v2, s10
-; GFX6-NEXT:    v_mul_hi_u32 v4, v3, v5
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s6, v2
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
-; GFX6-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v5, vcc, s10, v2
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX6-NEXT:    v_mul_lo_u32 v3, v3, s11
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s10, v2
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX6-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s7, v3
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s11, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX6-NEXT:    v_subrev_i32_e32 v4, vcc, s11, v3
-; GFX6-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX6-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s10
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s8
+; GFX6-NEXT:    s_sub_i32 s2, s4, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s8
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s8
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX6-NEXT:    s_cselect_b32 s4, s3, s2
+; GFX6-NEXT:    s_sub_i32 s2, 0, s9
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX6-NEXT:    v_cvt_f32_u32_e32 v2, s11
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s9
+; GFX6-NEXT:    s_sub_i32 s2, s5, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s9
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s9
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s9
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s9
+; GFX6-NEXT:    s_cselect_b32 s5, s3, s2
+; GFX6-NEXT:    s_sub_i32 s2, 0, s10
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    v_rcp_iflag_f32_e32 v2, v2
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_mul_hi_u32 v0, s6, v0
+; GFX6-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX6-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX6-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX6-NEXT:    s_mul_i32 s2, s2, s10
+; GFX6-NEXT:    s_sub_i32 s2, s6, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s10
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s10
+; GFX6-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX6-NEXT:    s_sub_i32 s3, s2, s10
+; GFX6-NEXT:    s_cmp_ge_u32 s2, s10
+; GFX6-NEXT:    s_cselect_b32 s6, s3, s2
+; GFX6-NEXT:    s_sub_i32 s2, 0, s11
+; GFX6-NEXT:    v_mul_lo_u32 v0, s2, v1
+; GFX6-NEXT:    s_mov_b32 s3, 0xf000
+; GFX6-NEXT:    s_mov_b32 s2, -1
+; GFX6-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_mul_hi_u32 v2, s7, v0
+; GFX6-NEXT:    v_mov_b32_e32 v0, s4
+; GFX6-NEXT:    v_mov_b32_e32 v1, s5
+; GFX6-NEXT:    v_readfirstlane_b32 s4, v2
+; GFX6-NEXT:    s_mul_i32 s4, s4, s11
+; GFX6-NEXT:    s_sub_i32 s4, s7, s4
+; GFX6-NEXT:    s_sub_i32 s5, s4, s11
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s11
+; GFX6-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX6-NEXT:    s_sub_i32 s5, s4, s11
+; GFX6-NEXT:    s_cmp_ge_u32 s4, s11
+; GFX6-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX6-NEXT:    v_mov_b32_e32 v2, s6
+; GFX6-NEXT:    v_mov_b32_e32 v3, s4
 ; GFX6-NEXT:    buffer_store_dwordx4 v[0:3], off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
 ; GFX8-LABEL: test_udivrem_v4:
 ; GFX8:       ; %bb.0:
 ; GFX8-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)

 ; GFX8-NEXT:    s_cselect_b32 s3, s4, s3
 ; GFX8-NEXT:    s_sub_i32 s4, s3, s9
 ; GFX8-NEXT:    s_cmp_ge_u32 s3, s9
 ; GFX8-NEXT:    s_cselect_b32 s3, s4, s3
 ; GFX8-NEXT:    s_sub_i32 s4, 0, s10
 ; GFX8-NEXT:    v_mul_lo_u32 v0, s4, v1
 ; GFX8-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX8-NEXT:    v_mul_hi_u32 v0, v1, v0
-; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
 ; GFX8-NEXT:    v_mul_hi_u32 v0, s6, v0
 ; GFX8-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
 ; GFX8-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX8-NEXT:    v_readfirstlane_b32 s4, v0
 ; GFX8-NEXT:    s_mul_i32 s4, s4, s10
 ; GFX8-NEXT:    s_sub_i32 s4, s6, s4
 ; GFX8-NEXT:    s_sub_i32 s5, s4, s10
 ; GFX8-NEXT:    s_cmp_ge_u32 s4, s10

llvm/test/CodeGen/AMDGPU/uint_to_fp.f64.ll

define amdgpu_kernel void @s_uint_to_fp_v4i32_to_v4f64(<4 x double> addrspace(1)* %out, <4 x i32> %in) {
   ret void
 }
 
 ; We can't fold the SGPRs into v_cndmask_b32_e32, because it already
 ; uses an SGPR (implicit vcc).
 
 ; GCN-LABEL: {{^}}uint_to_fp_i1_to_f64:
 ; VI-DAG: s_cmp_eq_u32
-; VI-DAG: s_cselect_b32 s[[SSEL:[0-9]+]], 0x3ff00000, 0
-; VI-DAG: v_mov_b32_e32 v[[SEL:[0-9]+]], s[[SSEL]]
-; SI-DAG: s_cmp_eq_u32
-; SI-DAG: s_cselect_b64 vcc, -1, 0
-; SI-DAG: v_cndmask_b32_e32 v[[SEL:[0-9]+]], 0, {{v[0-9]+}}, vcc
+; GCN-DAG: s_cselect_b32 s[[SSEL:[0-9]+]], 0x3ff00000, 0
 ; GCN-DAG: v_mov_b32_e32 v[[ZERO:[0-9]+]], 0{{$}}
+; GCN-DAG: v_mov_b32_e32 v[[SEL:[0-9]+]], s[[SSEL]]
 ; GCN: flat_store_dwordx2 v{{\[[0-9]+:[0-9]+\]}}, v[[[ZERO]]:[[SEL]]]
 ; GCN: s_endpgm
 define amdgpu_kernel void @uint_to_fp_i1_to_f64(double addrspace(1)* %out, i32 %in) {
   %cmp = icmp eq i32 %in, 0
   %fp = uitofp i1 %cmp to double
   store double %fp, double addrspace(1)* %out, align 4
   ret void
 }

llvm/test/CodeGen/AMDGPU/urem64.ll

 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s0, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s1, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_cmp_le_u32_e64 s[2:3], s12, v4
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s13, v5
 ; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s12, v4
 ; 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, s11
-; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v5, v1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GCN-NEXT:    v_mov_b32_e32 v4, s11
+; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v4, v1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s13, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s12, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
+; GCN-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s13, 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 v4, v4, v5, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
 ; 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:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_urem_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0xd
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[4:5], 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[10:11], s[2:3], 0
-; GCN-IR-NEXT:    s_flbit_i32_b32 s12, s4
-; GCN-IR-NEXT:    s_add_i32 s14, s12, 32
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[8:9], s[10:11]
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s5
-; GCN-IR-NEXT:    s_min_u32 s10, s14, s8
-; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
-; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
-; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
-; GCN-IR-NEXT:    s_min_u32 s14, s8, s9
-; GCN-IR-NEXT:    s_sub_u32 s8, s10, s14
-; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[8:9], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[8:9], 63
-; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[16:17]
-; GCN-IR-NEXT:    s_or_b64 s[16:17], s[12:13], s[18:19]
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[4:5], 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[2:3], 0
+; GCN-IR-NEXT:    s_flbit_i32_b32 s10, s4
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s11, s5
+; GCN-IR-NEXT:    s_add_i32 s10, s10, 32
+; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
+; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
+; GCN-IR-NEXT:    s_min_u32 s10, s10, s11
+; GCN-IR-NEXT:    s_min_u32 s14, s6, s7
+; GCN-IR-NEXT:    s_sub_u32 s12, s10, s14
+; GCN-IR-NEXT:    s_subb_u32 s13, 0, 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[16:17], s[12:13], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[12:13], 63
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[8:9], s[16:17]
+; GCN-IR-NEXT:    s_and_b64 s[8:9], s[16:17], exec
+; GCN-IR-NEXT:    s_cselect_b32 s9, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s8, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[16:17], s[16:17], s[18:19]
+; GCN-IR-NEXT:    s_mov_b64 s[6:7], 0
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_mov_b32 s11, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s12, s8, 1
-; GCN-IR-NEXT:    s_addc_u32 s13, s9, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[12:13], 0
-; GCN-IR-NEXT:    s_sub_i32 s8, 63, s8
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[16:17]
-; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[2:3], s8
+; GCN-IR-NEXT:    s_add_u32 s16, s12, 1
+; GCN-IR-NEXT:    s_addc_u32 s17, s13, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[8:9], s[16:17], 0
+; GCN-IR-NEXT:    s_sub_i32 s12, 63, s12
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[8:9]
+; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[2:3], s12
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[2:3], s12
+; GCN-IR-NEXT:    s_lshr_b64 s[12:13], s[2:3], s16
 ; GCN-IR-NEXT:    s_add_u32 s16, s4, -1
 ; GCN-IR-NEXT:    s_addc_u32 s17, s5, -1
 ; GCN-IR-NEXT:    s_not_b64 s[6:7], s[10:11]
 ; GCN-IR-NEXT:    s_add_u32 s10, s6, s14
 ; GCN-IR-NEXT:    s_addc_u32 s11, s7, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], 0
 ; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:  .LBB0_3: ; %udiv-do-while

 ; GCN-IR-NEXT:    s_add_u32 s10, s10, 1
 ; GCN-IR-NEXT:    s_addc_u32 s11, s11, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[18:19], s[10:11], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[14:15], s[6:7]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[18:19]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB0_3
 ; GCN-IR-NEXT:  .LBB0_4: ; %Flow6
 ; GCN-IR-NEXT:    s_lshl_b64 s[8:9], s[8:9], 1
-; GCN-IR-NEXT:    s_or_b64 s[6:7], s[6:7], s[8:9]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s7
-; GCN-IR-NEXT:    s_branch .LBB0_6
-; GCN-IR-NEXT:  .LBB0_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[12:13]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[12:13]
-; GCN-IR-NEXT:  .LBB0_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, s4, v1
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, s4, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v3, s5, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, s4, v0
-; GCN-IR-NEXT:    s_mov_b32 s11, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
+; GCN-IR-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GCN-IR-NEXT:  .LBB0_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s8
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s4, v0
+; GCN-IR-NEXT:    s_mov_b32 s12, s0
+; GCN-IR-NEXT:    s_mul_i32 s0, s4, s9
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s5, s8
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s4, s8
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    s_mov_b32 s10, -1
-; GCN-IR-NEXT:    s_mov_b32 s8, s0
-; GCN-IR-NEXT:    s_mov_b32 s9, s1
+; GCN-IR-NEXT:    s_mov_b32 s15, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s14, -1
+; GCN-IR-NEXT:    s_mov_b32 s13, s1
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[12:15], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = urem i64 %x, %y
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define i64 @v_test_urem_i64(i64 %x, i64 %y) {
 ; GCN-LABEL: v_test_urem_i64:

 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v8, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v2, s0, v1
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s1, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v3
 ; GCN-NEXT:    v_mul_hi_u32 v8, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v1, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v6, vcc, v7, v6
 ; GCN-NEXT:    v_addc_u32_e32 v7, vcc, 0, v8, vcc

 ; GCN-NEXT:    v_mul_hi_u32 v0, v0, 24
 ; GCN-NEXT:    v_mul_hi_u32 v1, v1, 24
 ; GCN-NEXT:    v_mov_b32_e32 v3, s7
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v1, s7, v0
 ; GCN-NEXT:    v_mul_hi_u32 v2, s6, v0
 ; GCN-NEXT:    v_mul_lo_u32 v0, s6, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, 0, v1
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, 24, v0
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, vcc
 ; GCN-NEXT:    v_subrev_i32_e64 v4, s[0:1], s6, v0
 ; GCN-NEXT:    v_subbrev_u32_e64 v5, s[2:3], 0, v2, s[0:1]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[2:3], s7, v5
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[2:3]
 ; GCN-NEXT:    v_cmp_le_u32_e64 s[2:3], s6, v4
 ; GCN-NEXT:    v_subb_u32_e64 v2, s[0:1], v2, v3, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
 ; GCN-NEXT:    v_cmp_eq_u32_e64 s[2:3], s7, v5
 ; GCN-NEXT:    v_subrev_i32_e64 v3, s[0:1], s6, v4
 ; 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_subb_u32_e32 v1, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
 ; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
+; GCN-NEXT:    v_cndmask_b32_e64 v3, v4, v3, s[0:1]
+; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
+; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
 ; GCN-NEXT:    v_cndmask_b32_e64 v2, v5, v2, s[0:1]
 ; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, vcc
-; GCN-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, s7, 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 v4, v4, v5, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
 ; 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:    v_cndmask_b32_e32 v0, v0, v3, vcc
 ; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_urem_k_num_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
-; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
-; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
-; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
-; GCN-IR-NEXT:    s_add_u32 s6, s8, 0xffffffc5
-; GCN-IR-NEXT:    s_addc_u32 s7, 0, -1
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[4:5], s[10:11]
-; GCN-IR-NEXT:    s_or_b64 s[4:5], s[10:11], s[12:13]
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
+; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-IR-NEXT:    s_flbit_i32_b32 s8, s2
+; GCN-IR-NEXT:    s_flbit_i32_b32 s9, s3
+; GCN-IR-NEXT:    s_add_i32 s8, s8, 32
+; GCN-IR-NEXT:    s_min_u32 s8, s8, s9
+; GCN-IR-NEXT:    s_add_u32 s10, s8, 0xffffffc5
+; GCN-IR-NEXT:    s_addc_u32 s11, 0, -1
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[2:3], 0
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[12:13], s[10:11], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[10:11], 63
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[6:7], s[12:13]
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[12:13], exec
+; GCN-IR-NEXT:    s_cselect_b32 s6, 0, 24
+; GCN-IR-NEXT:    s_or_b64 s[12:13], s[12:13], s[14:15]
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
+; GCN-IR-NEXT:    s_mov_b32 s7, 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB6_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
-; GCN-IR-NEXT:    s_addc_u32 s11, s7, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[10:11], 0
-; GCN-IR-NEXT:    s_sub_i32 s6, 63, s6
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], 24, s6
+; GCN-IR-NEXT:    s_add_u32 s12, s10, 1
+; GCN-IR-NEXT:    s_addc_u32 s13, s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[12:13], 0
+; GCN-IR-NEXT:    s_sub_i32 s9, 63, s10
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[6:7]
+; GCN-IR-NEXT:    s_lshl_b64 s[6:7], 24, s9
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB6_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[10:11], 24, s10
+; GCN-IR-NEXT:    s_lshr_b64 s[10:11], 24, s12
 ; GCN-IR-NEXT:    s_add_u32 s14, s2, -1
 ; GCN-IR-NEXT:    s_addc_u32 s15, s3, -1
 ; GCN-IR-NEXT:    s_sub_u32 s8, 58, s8
 ; GCN-IR-NEXT:    s_subb_u32 s9, 0, 0
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:  .LBB6_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1

 ; GCN-IR-NEXT:    s_add_u32 s8, s8, 1
 ; GCN-IR-NEXT:    s_addc_u32 s9, s9, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[16:17], s[8:9], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], s[4:5]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[16:17]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB6_3
 ; GCN-IR-NEXT:  .LBB6_4: ; %Flow5
 ; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[4:5], s[4:5], s[6:7]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-IR-NEXT:    s_branch .LBB6_6
-; GCN-IR-NEXT:  .LBB6_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, 0
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, 24, 0, s[10:11]
-; GCN-IR-NEXT:  .LBB6_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, s2, v1
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, s2, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v3, s3, v0
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, s2, v0
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
-; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, 24, v0
-; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    s_mov_b32 s4, s0
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
+; GCN-IR-NEXT:    s_or_b64 s[6:7], s[4:5], s[6:7]
+; GCN-IR-NEXT:  .LBB6_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s6
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s2, v0
+; GCN-IR-NEXT:    s_mov_b32 s8, s0
+; GCN-IR-NEXT:    s_mul_i32 s0, s2, s7
+; GCN-IR-NEXT:    s_mov_b32 s11, 0xf000
+; GCN-IR-NEXT:    v_add_i32_e32 v0, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s3, s6
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s2, s6
+; GCN-IR-NEXT:    v_sub_i32_e64 v0, vcc, 24, s0
+; GCN-IR-NEXT:    s_mov_b32 s10, -1
+; GCN-IR-NEXT:    s_mov_b32 s9, s1
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, 0, v1, vcc
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = urem i64 24, %x
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 define amdgpu_kernel void @s_test_urem_k_den_i64(i64 addrspace(1)* %out, i64 %x) {
 ; GCN-LABEL: s_test_urem_k_den_i64:
 ; GCN:       ; %bb.0:
 ; GCN-NEXT:    v_mov_b32_e32 v0, 0x4f800000
 ; GCN-NEXT:    v_madak_f32 v0, 0, v0, 0x41c00000
 ; GCN-NEXT:    v_rcp_f32_e32 v0, v0
-; GCN-NEXT:    s_movk_i32 s4, 0xffe8
-; GCN-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
-; GCN-NEXT:    s_mov_b32 s7, 0xf000
+; GCN-NEXT:    s_movk_i32 s2, 0xffe8
+; GCN-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x9
+; GCN-NEXT:    s_mov_b32 s3, 0xf000
 ; GCN-NEXT:    v_mul_f32_e32 v0, 0x5f7ffffc, v0
 ; GCN-NEXT:    v_mul_f32_e32 v1, 0x2f800000, v0
 ; GCN-NEXT:    v_trunc_f32_e32 v1, v1
 ; 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:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_mov_b32 s5, s1
-; GCN-NEXT:    s_mov_b32 s6, -1
-; GCN-NEXT:    v_mul_hi_u32 v2, v0, s4
-; GCN-NEXT:    v_mul_lo_u32 v4, v1, s4
-; GCN-NEXT:    v_mul_lo_u32 v3, v0, s4
-; GCN-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GCN-NEXT:    s_mov_b32 s0, s4
+; GCN-NEXT:    s_mov_b32 s1, s5
+; GCN-NEXT:    v_mul_hi_u32 v2, v0, s2
+; GCN-NEXT:    v_mul_lo_u32 v4, v1, s2
+; GCN-NEXT:    v_mul_lo_u32 v3, v0, s2
+; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, v3
 ; GCN-NEXT:    v_mul_lo_u32 v4, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v5, v4
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v6, v1, v3
 ; GCN-NEXT:    v_mul_hi_u32 v3, v1, v3
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v5, v3, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GCN-NEXT:    v_mul_hi_u32 v2, v0, s4
-; GCN-NEXT:    v_mul_lo_u32 v3, v1, s4
-; GCN-NEXT:    v_mul_lo_u32 v4, v0, s4
-; GCN-NEXT:    s_mov_b32 s4, s0
-; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_mul_hi_u32 v2, v0, s2
+; GCN-NEXT:    v_mul_lo_u32 v3, v1, s2
+; GCN-NEXT:    v_mul_lo_u32 v4, v0, s2
+; GCN-NEXT:    s_mov_b32 s2, -1
+; GCN-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_mul_lo_u32 v3, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, v4
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, v1, v2
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GCN-NEXT:    v_addc_u32_e32 v5, vcc, 0, v6, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v6, v1, v4
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v6
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, v5, v4, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v4, vcc, 0, v7, vcc
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GCN-NEXT:    v_addc_u32_e32 v3, vcc, 0, v4, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, v1, v3, vcc
-; GCN-NEXT:    v_mul_lo_u32 v2, s2, v1
-; GCN-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GCN-NEXT:    v_mul_hi_u32 v4, s2, v1
-; GCN-NEXT:    v_mul_hi_u32 v5, s3, v1
-; GCN-NEXT:    v_mul_lo_u32 v1, s3, v1
+; 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, 0, v4, vcc
-; GCN-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GCN-NEXT:    v_mul_hi_u32 v0, s3, v0
+; 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, v2, v4
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, v3, v0, vcc
 ; GCN-NEXT:    v_addc_u32_e32 v2, vcc, 0, v5, vcc
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GCN-NEXT:    v_addc_u32_e32 v1, vcc, 0, v2, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v1, v1, 24
 ; GCN-NEXT:    v_mul_hi_u32 v2, v0, 24
 ; GCN-NEXT:    v_mul_lo_u32 v0, v0, 24
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-NEXT:    v_mov_b32_e32 v2, s3
-; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
+; GCN-NEXT:    v_mov_b32_e32 v2, s7
+; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
 ; GCN-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
 ; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, 24, v0
 ; GCN-NEXT:    v_subbrev_u32_e32 v3, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_subrev_i32_e32 v4, vcc, 24, v2
 ; GCN-NEXT:    v_subbrev_u32_e32 v5, vcc, 0, v3, vcc
 ; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, 23, v2
 ; GCN-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
 ; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v3
 ; GCN-NEXT:    v_cndmask_b32_e32 v6, -1, v6, vcc
 ; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
-; GCN-NEXT:    v_cmp_lt_u32_e64 s[0:1], 23, v0
 ; GCN-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GCN-NEXT:    v_cndmask_b32_e64 v5, 0, -1, s[0:1]
-; GCN-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v1
-; GCN-NEXT:    v_cndmask_b32_e64 v5, -1, v5, s[0:1]
-; GCN-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v5
 ; GCN-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GCN-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
-; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-NEXT:    v_cmp_lt_u32_e32 vcc, 23, v0
+; GCN-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
+; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
+; GCN-NEXT:    v_cndmask_b32_e32 v4, -1, v4, vcc
+; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v4
+; GCN-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
+; GCN-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GCN-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; GCN-NEXT:    s_endpgm
 ;
 ; GCN-IR-LABEL: s_test_urem_k_den_i64:
 ; GCN-IR:       ; %bb.0: ; %_udiv-special-cases
 ; GCN-IR-NEXT:    s_load_dwordx4 s[0:3], s[0:1], 0x9
 ; GCN-IR-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-IR-NEXT:    s_flbit_i32_b32 s6, s2
 ; GCN-IR-NEXT:    s_flbit_i32_b32 s7, s3
 ; GCN-IR-NEXT:    s_add_i32 s6, s6, 32
 ; GCN-IR-NEXT:    s_min_u32 s8, s6, s7
-; GCN-IR-NEXT:    s_sub_u32 s6, 59, s8
-; GCN-IR-NEXT:    s_subb_u32 s7, 0, 0
+; GCN-IR-NEXT:    s_sub_u32 s10, 59, s8
+; GCN-IR-NEXT:    s_subb_u32 s11, 0, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[4:5], s[2:3], 0
-; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[10:11], s[6:7], 63
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[6:7], 63
-; GCN-IR-NEXT:    s_or_b64 s[10:11], s[4:5], s[10:11]
-; GCN-IR-NEXT:    s_or_b64 s[4:5], s[10:11], s[12:13]
+; GCN-IR-NEXT:    v_cmp_gt_u64_e64 s[6:7], s[10:11], 63
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[10:11], 63
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[4:5], s[6:7]
+; GCN-IR-NEXT:    s_and_b64 s[6:7], s[4:5], exec
+; GCN-IR-NEXT:    s_cselect_b32 s7, 0, s3
+; GCN-IR-NEXT:    s_cselect_b32 s6, 0, s2
+; GCN-IR-NEXT:    s_or_b64 s[4:5], s[4:5], s[12:13]
 ; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[4:5]
 ; GCN-IR-NEXT:    s_mov_b64 s[4:5], 0
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_5
 ; GCN-IR-NEXT:  ; %bb.1: ; %udiv-bb1
-; GCN-IR-NEXT:    s_add_u32 s10, s6, 1
-; GCN-IR-NEXT:    s_addc_u32 s11, s7, 0
-; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[12:13], s[10:11], 0
-; GCN-IR-NEXT:    s_sub_i32 s6, 63, s6
-; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[12:13]
-; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[2:3], s6
+; GCN-IR-NEXT:    s_add_u32 s12, s10, 1
+; GCN-IR-NEXT:    s_addc_u32 s13, s11, 0
+; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[6:7], s[12:13], 0
+; GCN-IR-NEXT:    s_sub_i32 s9, 63, s10
+; GCN-IR-NEXT:    s_andn2_b64 vcc, exec, s[6:7]
+; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[2:3], s9
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_4
 ; GCN-IR-NEXT:  ; %bb.2: ; %udiv-preheader
-; GCN-IR-NEXT:    s_lshr_b64 s[10:11], s[2:3], s10
+; GCN-IR-NEXT:    s_lshr_b64 s[10:11], s[2:3], s12
 ; GCN-IR-NEXT:    s_add_u32 s8, s8, 0xffffffc4
 ; GCN-IR-NEXT:    s_addc_u32 s9, 0, -1
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], 0
 ; GCN-IR-NEXT:    s_mov_b32 s5, 0
 ; GCN-IR-NEXT:  .LBB7_3: ; %udiv-do-while
 ; GCN-IR-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-IR-NEXT:    s_lshl_b64 s[10:11], s[10:11], 1
 ; GCN-IR-NEXT:    s_lshr_b32 s4, s7, 31

 ; GCN-IR-NEXT:    s_add_u32 s8, s8, 1
 ; GCN-IR-NEXT:    s_addc_u32 s9, s9, 0
 ; GCN-IR-NEXT:    v_cmp_eq_u64_e64 s[14:15], s[8:9], 0
 ; GCN-IR-NEXT:    s_mov_b64 s[12:13], s[4:5]
 ; GCN-IR-NEXT:    s_and_b64 vcc, exec, s[14:15]
 ; GCN-IR-NEXT:    s_cbranch_vccz .LBB7_3
 ; GCN-IR-NEXT:  .LBB7_4: ; %Flow5
 ; GCN-IR-NEXT:    s_lshl_b64 s[6:7], s[6:7], 1
-; GCN-IR-NEXT:    s_or_b64 s[4:5], s[4:5], s[6:7]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s4
-; GCN-IR-NEXT:    v_mov_b32_e32 v1, s5
-; GCN-IR-NEXT:    s_branch .LBB7_6
-; GCN-IR-NEXT:  .LBB7_5:
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s3
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v1, v0, 0, s[10:11]
-; GCN-IR-NEXT:    v_mov_b32_e32 v0, s2
-; GCN-IR-NEXT:    v_cndmask_b32_e64 v0, v0, 0, s[10:11]
-; GCN-IR-NEXT:  .LBB7_6: ; %udiv-end
-; GCN-IR-NEXT:    v_mul_lo_u32 v1, v1, 24
-; GCN-IR-NEXT:    v_mul_hi_u32 v2, v0, 24
-; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, 24
-; GCN-IR-NEXT:    s_mov_b32 s7, 0xf000
-; GCN-IR-NEXT:    s_mov_b32 s6, -1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
+; GCN-IR-NEXT:    s_or_b64 s[6:7], s[4:5], s[6:7]
+; GCN-IR-NEXT:  .LBB7_5: ; %udiv-end
+; GCN-IR-NEXT:    v_mul_hi_u32 v0, s6, 24
+; GCN-IR-NEXT:    s_mov_b32 s8, s0
+; GCN-IR-NEXT:    s_mul_i32 s0, s7, 24
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, s0, v0
+; GCN-IR-NEXT:    s_mul_i32 s0, s6, 24
+; GCN-IR-NEXT:    v_mov_b32_e32 v0, s0
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
-; GCN-IR-NEXT:    s_mov_b32 s4, s0
-; GCN-IR-NEXT:    s_mov_b32 s5, s1
+; GCN-IR-NEXT:    s_mov_b32 s11, 0xf000
+; GCN-IR-NEXT:    s_mov_b32 s10, -1
+; GCN-IR-NEXT:    s_mov_b32 s9, s1
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
-; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GCN-IR-NEXT:    buffer_store_dwordx2 v[0:1], off, s[8:11], 0
 ; GCN-IR-NEXT:    s_endpgm
   %result = urem i64 %x, 24
   store i64 %result, i64 addrspace(1)* %out
   ret void
 }
 
 ; FIXME: Constant bus violation
 ; define i64 @v_test_urem_k_num_i64(i64 %x) {

llvm/test/CodeGen/AMDGPU/vselect.ll

 ;RUN: llc < %s -march=amdgcn -verify-machineinstrs | FileCheck --check-prefix=SI --check-prefix=FUNC %s
 ;RUN: llc < %s -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs | FileCheck --check-prefix=VI --check-prefix=FUNC %s
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG --check-prefix=FUNC %s
 
 ; FUNC-LABEL: {{^}}test_select_v2i32:
 
 ; EG-DAG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW]}}, KC0[3].Z
 ; EG-DAG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW]}}, KC0[3].Y
 
 ; VI: s_cmp_gt_i32
 ; VI: s_cselect_b32
 ; VI: s_cmp_gt_i32
 ; VI: s_cselect_b32
 
 ; SI-DAG: s_cmp_gt_i32
-; SI-DAG: v_cndmask_b32_e64
+; SI-DAG: s_cselect_b32
 ; SI-DAG: s_cmp_gt_i32
-; SI-DAG: v_cndmask_b32_e32
+; SI-DAG: s_cselect_b32
 
 define amdgpu_kernel void @test_select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in0, <2 x i32> addrspace(1)* %in1, <2 x i32> %val) {
 entry:
   %load0 = load <2 x i32>, <2 x i32> addrspace(1)* %in0
   %load1 = load <2 x i32>, <2 x i32> addrspace(1)* %in1
   %cmp = icmp sgt <2 x i32> %load0, %load1
   %result = select <2 x i1> %cmp, <2 x i32> %val, <2 x i32> %load0
   store <2 x i32> %result, <2 x i32> addrspace(1)* %out

 ; EG-DAG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW]}}, KC0[3].Z
 ; EG-DAG: CNDE_INT {{\** *}}T{{[0-9]+\.[XYZW], PV\.[XYZW], T[0-9]+\.[XYZW]}}, KC0[3].Y
 
 ; VI: s_cselect_b32
 ; VI: s_cselect_b32
 ; VI: s_cselect_b32
 ; VI: s_cselect_b32
 
-; SI-DAG: v_cndmask_b32_e64
-; SI-DAG: v_cndmask_b32_e64
-; SI-DAG: v_cndmask_b32_e64
-; SI-DAG: v_cndmask_b32_e32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
+; SI-DAG: s_cselect_b32
 
 define amdgpu_kernel void @test_select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in0, <4 x i32> addrspace(1)* %in1, <4 x i32> %val) {
 entry:
   %load0 = load <4 x i32>, <4 x i32> addrspace(1)* %in0
   %load1 = load <4 x i32>, <4 x i32> addrspace(1)* %in1
   %cmp = icmp sgt <4 x i32> %load0, %load1
   %result = select <4 x i1> %cmp, <4 x i32> %val, <4 x i32> %load0
   store <4 x i32> %result, <4 x i32> addrspace(1)* %out