[AMDGPU] Lowering VGPR to SGPR copies to v_readfirstlane_b32 if profitable.

Since the divergence-driven instruction selection has been enabled for AMDGPU,
 all the uniform instructions are expected to be selected to SALU form, except those not having one.
 VGPR to SGPR copies appear in MIR to connect values producers and consumers. This change implements an algorithm
 that evolves a reasonable tradeoff between the profit achieved from keeping the uniform instructions in SALU form
 and overhead introduced by the data transfer between the VGPRs and SGPRs.

Reviewed By: rampitec

Differential Revision: https://reviews.llvm.org/D128252

diff --git a/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp b/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
index f54778535b7cbb008f98fc65246e3cb0674df436..3e95c55df57e8ad17b5b31eec6e0ba7bb23c57cd 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFixSGPRCopies.cpp
@@ -67,6 +67,7 @@
 #include "AMDGPU.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "llvm/ADT/SetOperations.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Target/TargetMachine.h"
@@ -81,9 +82,9 @@ static cl::opt<bool> EnableM0Merge(
   cl::init(true));
 
 namespace {
-
 class SIFixSGPRCopies : public MachineFunctionPass {
   MachineDominatorTree *MDT;
+  unsigned NextVGPRToSGPRCopyID;
 
 public:
   static char ID;
@@ -92,9 +93,16 @@ public:
   const SIRegisterInfo *TRI;
   const SIInstrInfo *TII;
 
-  SIFixSGPRCopies() : MachineFunctionPass(ID) {}
+  SIFixSGPRCopies() : MachineFunctionPass(ID), NextVGPRToSGPRCopyID(0) {}
 
   bool runOnMachineFunction(MachineFunction &MF) override;
+  unsigned getNextVGPRToSGPRCopyId() { return ++NextVGPRToSGPRCopyID; }
+  void lowerVGPR2SGPRCopies(MachineFunction &MF);
+  // Handles copies which source register is:
+  // 1. Physical register
+  // 2. AGPR
+  // 3. Defined by the instruction the merely moves the immediate
+  bool lowerSpecialCase(MachineInstr &MI);
 
   MachineBasicBlock *processPHINode(MachineInstr &MI);
 
@@ -569,6 +577,14 @@ bool SIFixSGPRCopies::runOnMachineFunction(MachineFunction &MF) {
   TII = ST.getInstrInfo();
   MDT = &getAnalysis<MachineDominatorTree>();
 
+  // We have to lower VGPR to SGPR copies before the main loop
+  // because the REG_SEQUENCE and PHI lowering in main loop
+  // convert the def-use chains to VALU and close the opportunities
+  // for keeping them scalar.
+  // TODO: REG_SEQENCE and PHIs are semantically copies. The next patch
+  // addresses their lowering and unify the processing in one main loop.
+  lowerVGPR2SGPRCopies(MF);
+
   for (MachineFunction::iterator BI = MF.begin(), BE = MF.end();
                                                   BI != BE; ++BI) {
     MachineBasicBlock *MBB = &*BI;
@@ -640,42 +656,7 @@ bool SIFixSGPRCopies::runOnMachineFunction(MachineFunction &MF) {
           continue;
         }
 
-        if (isVGPRToSGPRCopy(SrcRC, DstRC, *TRI)) {
-          Register SrcReg = MI.getOperand(1).getReg();
-          if (!SrcReg.isVirtual()) {
-            MachineBasicBlock *NewBB = TII->moveToVALU(MI, MDT);
-            if (NewBB && NewBB != MBB) {
-              MBB = NewBB;
-              E = MBB->end();
-              BI = MachineFunction::iterator(MBB);
-              BE = MF.end();
-            }
-            assert((!NewBB || NewBB == I->getParent()) &&
-                   "moveToVALU did not return the right basic block");
-            break;
-          }
-
-          MachineInstr *DefMI = MRI->getVRegDef(SrcReg);
-          unsigned SMovOp;
-          int64_t Imm;
-          // If we are just copying an immediate, we can replace the copy with
-          // s_mov_b32.
-          if (isSafeToFoldImmIntoCopy(&MI, DefMI, TII, SMovOp, Imm)) {
-            MI.getOperand(1).ChangeToImmediate(Imm);
-            MI.addImplicitDefUseOperands(MF);
-            MI.setDesc(TII->get(SMovOp));
-            break;
-          }
-          MachineBasicBlock *NewBB = TII->moveToVALU(MI, MDT);
-          if (NewBB && NewBB != MBB) {
-            MBB = NewBB;
-            E = MBB->end();
-            BI = MachineFunction::iterator(MBB);
-            BE = MF.end();
-          }
-          assert((!NewBB || NewBB == I->getParent()) &&
-                 "moveToVALU did not return the right basic block");
-        } else if (isSGPRToVGPRCopy(SrcRC, DstRC, *TRI)) {
+        if (isSGPRToVGPRCopy(SrcRC, DstRC, *TRI)) {
           tryChangeVGPRtoSGPRinCopy(MI, TRI, TII);
         }
 
@@ -916,3 +897,269 @@ MachineBasicBlock *SIFixSGPRCopies::processPHINode(MachineInstr &MI) {
   }
   return CreatedBB;
 }
+
+bool SIFixSGPRCopies::lowerSpecialCase(MachineInstr &MI) {
+  MachineBasicBlock *MBB = MI.getParent();
+  const TargetRegisterClass *SrcRC, *DstRC;
+  std::tie(SrcRC, DstRC) = getCopyRegClasses(MI, *TRI, *MRI);
+
+  // We return true to indicate that no further processing needed
+  if (!isVGPRToSGPRCopy(SrcRC, DstRC, *TRI))
+    return true;
+
+  Register SrcReg = MI.getOperand(1).getReg();
+  if (!SrcReg.isVirtual() || TRI->isAGPR(*MRI, SrcReg)) {
+    TII->moveToVALU(MI, MDT);
+    return true;
+  }
+
+  unsigned SMovOp;
+  int64_t Imm;
+  // If we are just copying an immediate, we can replace the copy with
+  // s_mov_b32.
+  if (isSafeToFoldImmIntoCopy(&MI, MRI->getVRegDef(SrcReg), TII, SMovOp, Imm)) {
+    MI.getOperand(1).ChangeToImmediate(Imm);
+    MI.addImplicitDefUseOperands(*MBB->getParent());
+    MI.setDesc(TII->get(SMovOp));
+    return true;
+  }
+  return false;
+}
+
+class V2SCopyInfo {
+public:
+  // VGPR to SGPR copy being processed
+  MachineInstr *Copy;
+  // All SALU instructions reachable from this copy in SSA graph
+  DenseSet<MachineInstr *> SChain;
+  // Number of SGPR to VGPR copies that are used to put the SALU computation
+  // results back to VALU.
+  unsigned NumSVCopies;
+
+  unsigned Score;
+  // Actual count of v_readfirstlane_b32
+  // which need to be inserted to keep SChain SALU
+  unsigned NumReadfirstlanes;
+  // Current score state. To speedup selection V2SCopyInfos for processing
+  bool NeedToBeConvertedToVALU = false;
+  // Unique ID. Used as a key for mapping to keep permanent order.
+  unsigned ID;
+
+  // Count of another VGPR to SGPR copies that contribute to the
+  // current copy SChain
+  unsigned SiblingPenalty = 0;
+  SetVector<unsigned> Siblings;
+  V2SCopyInfo() : Copy(nullptr), ID(0){};
+  V2SCopyInfo(unsigned Id, MachineInstr *C, unsigned Width)
+      : Copy(C), NumSVCopies(0), NumReadfirstlanes(Width / 32), ID(Id){};
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+  void dump() {
+    dbgs() << ID << " : " << *Copy << "\n\tS:" << SChain.size()
+           << "\n\tSV:" << NumSVCopies << "\n\tSP: " << SiblingPenalty
+           << "\nScore: " << Score << "\n";
+  }
+#endif
+};
+
+void SIFixSGPRCopies::lowerVGPR2SGPRCopies(MachineFunction &MF) {
+
+  DenseMap<unsigned, V2SCopyInfo> Copies;
+  DenseMap<MachineInstr *, SetVector<unsigned>> SiblingPenalty;
+
+  // The main function that computes the VGPR to SGPR copy score
+  // and determines copy further lowering way: v_readfirstlane_b32 or moveToVALU
+  auto needToBeConvertedToVALU = [&](V2SCopyInfo *I) -> bool {
+    if (I->SChain.empty())
+      return true;
+    I->Siblings = SiblingPenalty[*std::max_element(
+        I->SChain.begin(), I->SChain.end(),
+        [&](MachineInstr *A, MachineInstr *B) -> bool {
+          return SiblingPenalty[A].size() < SiblingPenalty[B].size();
+        })];
+    I->Siblings.remove_if([&](unsigned ID) { return ID == I->ID; });
+    // The loop below computes the number of another VGPR to SGPR copies
+    // which contribute to the current copy SALU chain. We assume that all the
+    // copies with the same source virtual register will be squashed to one by
+    // regalloc. Also we take careof the copies of the differnt subregs of the
+    // same register.
+    SmallSet<std::pair<Register, unsigned>, 4> SrcRegs;
+    for (auto J : I->Siblings) {
+      auto InfoIt = Copies.find(J);
+      if (InfoIt != Copies.end()) {
+        MachineInstr *SiblingCopy = InfoIt->getSecond().Copy;
+        if (SiblingCopy->isImplicitDef())
+          // the COPY has already been MoveToVALUed
+          continue;
+
+        SrcRegs.insert(std::make_pair(SiblingCopy->getOperand(1).getReg(),
+                                      SiblingCopy->getOperand(1).getSubReg()));
+      }
+    }
+    I->SiblingPenalty = SrcRegs.size();
+
+    unsigned Penalty =
+        I->NumSVCopies + I->SiblingPenalty + I->NumReadfirstlanes;
+    unsigned Profit = I->SChain.size();
+    I->Score = Penalty > Profit ? 0 : Profit - Penalty;
+    I->NeedToBeConvertedToVALU = I->Score < 3;
+    return I->NeedToBeConvertedToVALU;
+  };
+
+  auto needProcessing = [](MachineInstr &MI) -> bool {
+    switch (MI.getOpcode()) {
+    case AMDGPU::COPY:
+    case AMDGPU::WQM:
+    case AMDGPU::STRICT_WQM:
+    case AMDGPU::SOFT_WQM:
+    case AMDGPU::STRICT_WWM:
+      return true;
+    default:
+      return false;
+    }
+  };
+
+  for (MachineFunction::iterator BI = MF.begin(), BE = MF.end(); BI != BE;
+       ++BI) {
+    MachineBasicBlock *MBB = &*BI;
+    for (MachineBasicBlock::iterator I = MBB->begin(), E = MBB->end(); I != E;
+         ++I) {
+      MachineInstr &MI = *I;
+      if (!needProcessing(MI))
+        continue;
+      if (lowerSpecialCase(MI))
+        continue;
+
+      // Compute the COPY width to pass it to V2SCopyInfo Ctor
+      Register DstReg = MI.getOperand(0).getReg();
+
+      const TargetRegisterClass *DstRC = TRI->getRegClassForReg(*MRI, DstReg);
+
+      V2SCopyInfo In(getNextVGPRToSGPRCopyId(), &MI,
+                     TRI->getRegSizeInBits(*DstRC));
+
+      SmallVector<MachineInstr *, 8> AnalysisWorklist;
+      // Needed because the SSA is not a tree but a graph and may have
+      // forks and joins. We should not then go same way twice.
+      DenseSet<MachineInstr *> Visited;
+      AnalysisWorklist.push_back(&MI);
+      while (!AnalysisWorklist.empty()) {
+
+        MachineInstr *Inst = AnalysisWorklist.pop_back_val();
+
+        if (!Visited.insert(Inst).second)
+          continue;
+
+        // Copies and REG_SEQUENCE do not contribute to the final assembly
+        // So, skip them but take care of the SGPR to VGPR copies bookkeeping.
+        if (Inst->isCopy() || Inst->isRegSequence()) {
+          if (TRI->isVGPR(*MRI, Inst->getOperand(0).getReg())) {
+            if (!Inst->isCopy() ||
+                !tryChangeVGPRtoSGPRinCopy(*Inst, TRI, TII)) {
+              In.NumSVCopies++;
+              continue;
+            }
+          }
+        }
+
+        SiblingPenalty[Inst].insert(In.ID);
+
+        SmallVector<MachineInstr *, 4> Users;
+        if ((TII->isSALU(*Inst) && Inst->isCompare()) ||
+            (Inst->isCopy() && Inst->getOperand(0).getReg() == AMDGPU::SCC)) {
+          auto I = Inst->getIterator();
+          auto E = Inst->getParent()->end();
+          while (++I != E && !I->findRegisterDefOperand(AMDGPU::SCC)) {
+            if (I->readsRegister(AMDGPU::SCC))
+              Users.push_back(&*I);
+          }
+        } else if (Inst->getNumExplicitDefs() != 0) {
+          Register Reg = Inst->getOperand(0).getReg();
+          if (TRI->isSGPRReg(*MRI, Reg))
+            for (auto &U : MRI->use_instructions(Reg))
+              Users.push_back(&U);
+        }
+        for (auto U : Users) {
+          if (TII->isSALU(*U))
+            In.SChain.insert(U);
+          AnalysisWorklist.push_back(U);
+        }
+      }
+      Copies[In.ID] = In;
+    }
+  }
+
+  SmallVector<unsigned, 8> LoweringWorklist;
+  for (auto &C : Copies) {
+    if (needToBeConvertedToVALU(&C.second))
+      LoweringWorklist.push_back(C.second.ID);
+  }
+
+  while (!LoweringWorklist.empty()) {
+    unsigned CurID = LoweringWorklist.pop_back_val();
+    auto CurInfoIt = Copies.find(CurID);
+    if (CurInfoIt != Copies.end()) {
+      V2SCopyInfo C = CurInfoIt->getSecond();
+      LLVM_DEBUG(dbgs() << "Processing ...\n"; C.dump());
+      for (auto S : C.Siblings) {
+        auto SibInfoIt = Copies.find(S);
+        if (SibInfoIt != Copies.end()) {
+          V2SCopyInfo &SI = SibInfoIt->getSecond();
+          LLVM_DEBUG(dbgs() << "Sibling:\n"; SI.dump());
+          if (!SI.NeedToBeConvertedToVALU) {
+            set_subtract(SI.SChain, C.SChain);
+            if (needToBeConvertedToVALU(&SI))
+              LoweringWorklist.push_back(SI.ID);
+          }
+          SI.Siblings.remove_if([&](unsigned ID) { return ID == C.ID; });
+        }
+      }
+      LLVM_DEBUG(dbgs() << "V2S copy " << *C.Copy
+                        << " is being turned to VALU\n");
+      Copies.erase(C.ID);
+      TII->moveToVALU(*C.Copy, MDT);
+    }
+  }
+
+  // Now do actual lowering
+  for (auto C : Copies) {
+    MachineInstr *MI = C.second.Copy;
+    MachineBasicBlock *MBB = MI->getParent();
+    // We decide to turn V2S copy to v_readfirstlane_b32
+    // remove it from the V2SCopies and remove it from all its siblings
+    LLVM_DEBUG(dbgs() << "V2S copy " << *MI
+                      << " is being turned to v_readfirstlane_b32"
+                      << " Score: " << C.second.Score << "\n");
+    Register DstReg = MI->getOperand(0).getReg();
+    Register SrcReg = MI->getOperand(1).getReg();
+    unsigned SubReg = MI->getOperand(1).getSubReg();
+    const TargetRegisterClass *SrcRC = TRI->getRegClassForReg(*MRI, SrcReg);
+    SrcRC = TRI->getSubRegClass(SrcRC, SubReg);
+    size_t SrcSize = TRI->getRegSizeInBits(*SrcRC);
+    if (SrcSize == 16) {
+      // HACK to handle possible 16bit VGPR source
+      auto MIB = BuildMI(*MBB, MI, MI->getDebugLoc(),
+                         TII->get(AMDGPU::V_READFIRSTLANE_B32), DstReg);
+      MIB.addReg(SrcReg, 0, AMDGPU::NoSubRegister);
+    } else if (SrcSize == 32) {
+      auto MIB = BuildMI(*MBB, MI, MI->getDebugLoc(),
+                         TII->get(AMDGPU::V_READFIRSTLANE_B32), DstReg);
+      MIB.addReg(SrcReg, 0, SubReg);
+    } else {
+      auto Result = BuildMI(*MBB, MI, MI->getDebugLoc(),
+                            TII->get(AMDGPU::REG_SEQUENCE), DstReg);
+      int N = TRI->getRegSizeInBits(*SrcRC) / 32;
+      for (int i = 0; i < N; i++) {
+        Register PartialSrc = TII->buildExtractSubReg(
+            Result, *MRI, MI->getOperand(1), SrcRC,
+            TRI->getSubRegFromChannel(i), &AMDGPU::VGPR_32RegClass);
+        Register PartialDst =
+            MRI->createVirtualRegister(&AMDGPU::SReg_32RegClass);
+        BuildMI(*MBB, *Result, Result->getDebugLoc(),
+                TII->get(AMDGPU::V_READFIRSTLANE_B32), PartialDst)
+            .addReg(PartialSrc);
+        Result.addReg(PartialDst).addImm(TRI->getSubRegFromChannel(i));
+      }
+    }
+    MI->eraseFromParent();
+  }
+}

diff --git a/llvm/test/CodeGen/AMDGPU/add3.ll b/llvm/test/CodeGen/AMDGPU/add3.ll
index 4144bb797b627f39b2a32ab7a2924f14667d73c2..3c9f2328f1185e0a1259a5dbbd605fc1671b3700 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/add3.ll
+++ b/llvm/test/CodeGen/AMDGPU/add3.ll
@@ -223,7 +223,7 @@ define amdgpu_ps float @add3_uniform_vgpr(float inreg %a, float inreg %b, float
 ; VI-NEXT:    v_mov_b32_e32 v2, 0x40400000
 ; VI-NEXT:    v_add_f32_e32 v2, s4, v2
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
-; VI-NEXT:    v_add_u32_e32 v0, vcc, v2, v0
+; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; VI-NEXT:    ; return to shader part epilog
 ;
 ; GFX9-LABEL: add3_uniform_vgpr:

diff --git a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
index 6a2d2539b34ec8133acf7462b96e63994b2eaadf..3c729182d1b09a4ff2704a541f875150f084c264 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
+++ b/llvm/test/CodeGen/AMDGPU/agpr-copy-no-free-registers.ll
@@ -521,175 +521,168 @@ define void @v32_asm_def_use(float %v0, float %v1) #0 {
 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 v0, v[0:1], off glc
+; GFX908-NEXT:    global_load_ushort v24, v[0:1], off glc
 ; GFX908-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
 ; GFX908-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x10
-; GFX908-NEXT:    s_load_dword s7, s[4:5], 0x18
-; GFX908-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x0
-; GFX908-NEXT:    s_mov_b32 s6, 0
+; GFX908-NEXT:    s_load_dword s8, s[4:5], 0x18
+; GFX908-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x0
+; GFX908-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX908-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX908-NEXT:    v_cvt_f32_u32_e32 v1, s1
+; GFX908-NEXT:    v_cvt_f32_u32_e32 v0, s1
 ; GFX908-NEXT:    s_sub_i32 s4, 0, s1
-; GFX908-NEXT:    s_lshl_b64 s[10:11], s[2:3], 5
-; GFX908-NEXT:    s_or_b32 s10, s10, 28
-; GFX908-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX908-NEXT:    v_mov_b32_e32 v35, s10
-; GFX908-NEXT:    s_lshr_b32 s12, s7, 16
-; GFX908-NEXT:    v_mov_b32_e32 v10, s11
-; GFX908-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX908-NEXT:    v_cvt_u32_f32_e32 v2, v1
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v26, s7
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v27, s12
-; GFX908-NEXT:    v_accvgpr_write_b32 a0, v35
-; GFX908-NEXT:    v_mul_lo_u32 v1, s4, v2
-; GFX908-NEXT:    v_accvgpr_write_b32 a1, v10
-; GFX908-NEXT:    s_lshl_b64 s[4:5], s[8:9], 5
-; GFX908-NEXT:    v_mul_hi_u32 v3, v2, v1
-; GFX908-NEXT:    v_mov_b32_e32 v1, 0
-; GFX908-NEXT:    v_add_u32_e32 v2, v2, v3
-; GFX908-NEXT:    v_mul_hi_u32 v4, s0, v2
-; GFX908-NEXT:    v_mul_lo_u32 v5, v4, s1
-; GFX908-NEXT:    v_add_u32_e32 v6, 1, v4
-; GFX908-NEXT:    v_sub_u32_e32 v5, s0, v5
-; GFX908-NEXT:    v_cmp_le_u32_e32 vcc, s1, v5
-; GFX908-NEXT:    v_cndmask_b32_e32 v4, v4, v6, vcc
-; GFX908-NEXT:    v_subrev_u32_e32 v6, s1, v5
-; GFX908-NEXT:    v_cndmask_b32_e32 v5, v5, v6, vcc
-; GFX908-NEXT:    v_add_u32_e32 v7, 1, v4
-; GFX908-NEXT:    v_cmp_le_u32_e32 vcc, s1, v5
+; GFX908-NEXT:    s_lshr_b32 s11, s8, 16
+; GFX908-NEXT:    v_cvt_f32_f16_e32 v25, s8
+; GFX908-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX908-NEXT:    s_lshl_b64 s[8:9], s[2:3], 5
+; GFX908-NEXT:    v_cvt_f32_f16_e32 v26, s11
+; 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 v7, s3
+; GFX908-NEXT:    s_mov_b32 s10, 0
+; GFX908-NEXT:    v_mov_b32_e32 v6, s2
+; GFX908-NEXT:    v_mul_lo_u32 v2, s4, v0
+; GFX908-NEXT:    s_lshl_b64 s[4:5], s[6:7], 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, s0, 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, s1
+; GFX908-NEXT:    v_add_u32_e32 v5, 1, v0
+; GFX908-NEXT:    v_sub_u32_e32 v4, s0, v4
+; GFX908-NEXT:    v_cmp_le_u32_e32 vcc, s1, v4
+; GFX908-NEXT:    v_cndmask_b32_e32 v0, v0, v5, vcc
+; GFX908-NEXT:    v_subrev_u32_e32 v5, s1, 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, s1, 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_and_b32_e32 v28, 0xffff, v0
-; GFX908-NEXT:    v_cndmask_b32_e32 v0, v4, v7, vcc
-; GFX908-NEXT:    v_mul_lo_u32 v8, s9, v28
-; GFX908-NEXT:    v_mul_hi_u32 v9, s8, v28
-; GFX908-NEXT:    v_lshlrev_b64 v[2:3], 5, v[0:1]
-; GFX908-NEXT:    v_mul_lo_u32 v6, s8, v28
-; GFX908-NEXT:    v_add_u32_e32 v7, v9, v8
-; GFX908-NEXT:    v_accvgpr_write_b32 a2, v2
-; GFX908-NEXT:    v_accvgpr_write_b32 a3, v3
-; GFX908-NEXT:    v_lshlrev_b64 v[6:7], 5, v[6:7]
-; GFX908-NEXT:    v_mov_b32_e32 v9, s3
-; GFX908-NEXT:    v_mov_b32_e32 v8, s2
+; GFX908-NEXT:    v_readfirstlane_b32 s0, v24
+; GFX908-NEXT:    s_and_b32 s0, 0xffff, s0
+; GFX908-NEXT:    s_mul_i32 s1, s7, s0
+; GFX908-NEXT:    s_mul_hi_u32 s7, s6, s0
+; GFX908-NEXT:    s_mul_i32 s0, s6, s0
+; GFX908-NEXT:    s_add_i32 s1, s7, s1
+; GFX908-NEXT:    s_lshl_b64 s[6:7], 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 v8, vcc, v8, v0
-; GFX908-NEXT:    s_nop 0
-; GFX908-NEXT:    v_accvgpr_read_b32 v3, a1
-; GFX908-NEXT:    v_accvgpr_read_b32 v5, a3
-; GFX908-NEXT:    v_addc_co_u32_e32 v9, vcc, 0, v9, vcc
-; GFX908-NEXT:    v_accvgpr_read_b32 v2, a0
-; GFX908-NEXT:    v_accvgpr_read_b32 v4, a2
+; 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_nop 0
-; GFX908-NEXT:    v_accvgpr_write_b32 a0, v2
-; GFX908-NEXT:    v_accvgpr_write_b32 a1, v3
 ; 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 v10, 0
-; GFX908-NEXT:    v_mov_b32_e32 v11, 0
-; GFX908-NEXT:    global_load_dwordx2 v[10:11], v[10:11], off
-; GFX908-NEXT:    s_mov_b32 s7, s6
-; GFX908-NEXT:    v_cmp_gt_i64_e64 s[0:1], 0, v[8:9]
-; GFX908-NEXT:    v_accvgpr_read_b32 v13, a1
-; GFX908-NEXT:    v_mov_b32_e32 v15, s7
-; GFX908-NEXT:    v_mov_b32_e32 v17, s7
-; GFX908-NEXT:    v_accvgpr_read_b32 v12, a0
-; GFX908-NEXT:    v_mov_b32_e32 v14, s6
-; GFX908-NEXT:    v_mov_b32_e32 v16, s6
+; 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 s11, s10
+; GFX908-NEXT:    v_mov_b32_e32 v13, s11
+; GFX908-NEXT:    v_mov_b32_e32 v15, s11
+; GFX908-NEXT:    v_mov_b32_e32 v17, s11
+; GFX908-NEXT:    v_mov_b32_e32 v12, s10
+; GFX908-NEXT:    v_mov_b32_e32 v14, s10
+; GFX908-NEXT:    v_mov_b32_e32 v16, s10
+; 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_add_co_u32_e32 v20, vcc, 1, v10
-; GFX908-NEXT:    v_addc_co_u32_e32 v18, vcc, 0, v11, vcc
-; GFX908-NEXT:    v_mul_lo_u32 v21, s4, v18
-; GFX908-NEXT:    v_mul_hi_u32 v22, s4, v20
-; GFX908-NEXT:    v_mul_lo_u32 v23, s5, v20
-; GFX908-NEXT:    v_mul_lo_u32 v29, s4, v20
-; GFX908-NEXT:    v_mov_b32_e32 v19, s7
-; GFX908-NEXT:    v_add_u32_e32 v20, v22, v21
-; GFX908-NEXT:    v_add_u32_e32 v30, v20, v23
-; GFX908-NEXT:    v_mov_b32_e32 v21, s7
-; GFX908-NEXT:    v_mov_b32_e32 v18, s6
-; GFX908-NEXT:    v_mov_b32_e32 v20, s6
+; 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 s9, s4, s2
+; GFX908-NEXT:    s_mul_i32 s11, s5, s2
+; GFX908-NEXT:    s_mul_i32 s8, s4, s2
+; GFX908-NEXT:    s_mul_i32 s2, s4, s3
+; GFX908-NEXT:    s_add_i32 s2, s9, s2
+; GFX908-NEXT:    s_add_i32 s9, s2, s11
 ; 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_e32 v10, vcc, v10, v28
-; GFX908-NEXT:    v_addc_co_u32_e32 v11, vcc, 0, v11, vcc
-; GFX908-NEXT:    v_cmp_gt_i64_e32 vcc, 0, v[10:11]
-; GFX908-NEXT:    v_add_co_u32_e64 v12, s[2:3], v12, v6
-; GFX908-NEXT:    v_addc_co_u32_e64 v13, s[2:3], v13, v7, s[2:3]
+; 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, s7
+; GFX908-NEXT:    v_add_co_u32_e64 v10, s[2:3], s6, v10
+; GFX908-NEXT:    v_addc_co_u32_e64 v11, s[2:3], v11, v20, s[2:3]
 ; 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_add_co_u32_e32 v22, vcc, v12, v29
-; GFX908-NEXT:    v_addc_co_u32_e32 v23, vcc, v13, v30, vcc
-; GFX908-NEXT:    global_load_dword v32, v[22:23], off offset:-12 glc
+; GFX908-NEXT:    v_mov_b32_e32 v21, s9
+; GFX908-NEXT:    v_add_co_u32_e32 v20, vcc, s8, 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_waitcnt vmcnt(0)
-; GFX908-NEXT:    global_load_dword v31, v[22:23], off offset:-8 glc
+; GFX908-NEXT:    global_load_dword v27, v[20:21], off offset:-8 glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    global_load_dword v24, v[22:23], off offset:-4 glc
+; GFX908-NEXT:    global_load_dword v22, v[20:21], off offset:-4 glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    global_load_dword v22, v[22:23], off glc
+; GFX908-NEXT:    global_load_dword v20, v[20:21], off glc
 ; GFX908-NEXT:    s_waitcnt vmcnt(0)
-; GFX908-NEXT:    ds_read_b64 v[22:23], v1
-; GFX908-NEXT:    ds_read_b64 v[24:25], v0
+; 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:    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 v33, v32 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v32, v32
-; GFX908-NEXT:    v_cvt_f32_f16_sdwa v34, v31 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX908-NEXT:    v_cvt_f32_f16_e32 v31, v31
-; GFX908-NEXT:    v_add_f32_e32 v4, v26, v22
-; GFX908-NEXT:    v_add_f32_e32 v5, v27, v23
-; GFX908-NEXT:    v_add_f32_e32 v2, 0, v22
-; GFX908-NEXT:    v_add_f32_e32 v3, 0, v23
-; GFX908-NEXT:    v_add_f32_e32 v25, v33, v25
-; GFX908-NEXT:    v_add_f32_e32 v24, v32, v24
-; GFX908-NEXT:    v_add_f32_e32 v23, v34, v23
-; GFX908-NEXT:    v_add_f32_e32 v22, v31, v22
-; GFX908-NEXT:    v_add_f32_e32 v15, v15, v5
-; GFX908-NEXT:    v_add_f32_e32 v14, v14, v4
-; GFX908-NEXT:    v_add_f32_e32 v17, v17, v3
-; GFX908-NEXT:    v_add_f32_e32 v16, v16, v2
-; GFX908-NEXT:    v_add_f32_e32 v18, v18, v24
-; GFX908-NEXT:    v_add_f32_e32 v19, v19, v25
-; GFX908-NEXT:    v_add_f32_e32 v20, v20, v22
-; GFX908-NEXT:    v_add_f32_e32 v21, v21, v23
+; 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:    s_branch .LBB3_4
 ;
 ; GFX90A-LABEL: introduced_copy_to_sgpr:
 ; GFX90A:       ; %bb.0: ; %bb
-; GFX90A-NEXT:    global_load_ushort v10, v[0:1], off glc
+; GFX90A-NEXT:    global_load_ushort v28, v[0:1], off glc
 ; GFX90A-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
 ; GFX90A-NEXT:    s_load_dwordx2 s[6:7], s[4:5], 0x8
-; GFX90A-NEXT:    s_load_dwordx2 s[8:9], s[4:5], 0x10
+; GFX90A-NEXT:    s_load_dwordx2 s[10:11], s[4:5], 0x10
 ; GFX90A-NEXT:    s_load_dword s2, s[4:5], 0x18
 ; GFX90A-NEXT:    v_mov_b32_e32 v1, 0
-; GFX90A-NEXT:    s_mov_b32 s4, 0
+; GFX90A-NEXT:    s_mov_b32 s8, 0
 ; GFX90A-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX90A-NEXT:    v_cvt_f32_u32_e32 v0, s7
-; GFX90A-NEXT:    s_sub_i32 s5, 0, s7
+; GFX90A-NEXT:    s_sub_i32 s9, 0, s7
+; GFX90A-NEXT:    s_lshl_b64 s[4:5], s[10:11], 5
+; GFX90A-NEXT:    s_or_b32 s4, s4, 28
+; GFX90A-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX90A-NEXT:    s_lshr_b32 s12, s2, 16
+; GFX90A-NEXT:    v_pk_mov_b32 v[6:7], s[4:5], s[4:5] op_sel:[0,1]
 ; GFX90A-NEXT:    v_cvt_f32_f16_e32 v2, s2
-; GFX90A-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v3, s12
-; GFX90A-NEXT:    s_lshl_b64 s[10:11], s[8:9], 5
-; GFX90A-NEXT:    s_or_b32 s10, s10, 28
 ; GFX90A-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX90A-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX90A-NEXT:    v_cvt_f32_f16_e32 v3, s12
 ; GFX90A-NEXT:    s_lshl_b64 s[2:3], s[0:1], 5
-; 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, s5, v0
+; GFX90A-NEXT:    v_pk_mov_b32 v[4:5], s[10:11], s[10:11] op_sel:[0,1]
+; GFX90A-NEXT:    v_mul_lo_u32 v8, s9, 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, s6, v0
@@ -704,14 +697,15 @@ define amdgpu_kernel void @introduced_copy_to_sgpr(i64 %arg, i32 %arg1, i32 %arg
 ; GFX90A-NEXT:    v_cmp_le_u32_e32 vcc, s7, 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_and_b32_e32 v30, 0xffff, v10
-; GFX90A-NEXT:    v_mul_lo_u32 v11, s1, v30
-; GFX90A-NEXT:    v_mul_hi_u32 v12, s0, v30
-; GFX90A-NEXT:    v_mul_lo_u32 v10, s0, v30
-; GFX90A-NEXT:    v_add_u32_e32 v11, v12, v11
-; GFX90A-NEXT:    v_lshlrev_b64 v[10:11], 5, v[10:11]
-; GFX90A-NEXT:    v_pk_mov_b32 v[12:13], 0, 0
+; GFX90A-NEXT:    v_readfirstlane_b32 s4, v28
+; GFX90A-NEXT:    s_and_b32 s4, 0xffff, s4
+; GFX90A-NEXT:    s_mul_i32 s1, s1, s4
+; GFX90A-NEXT:    s_mul_hi_u32 s5, s0, s4
+; GFX90A-NEXT:    s_mul_i32 s0, s0, s4
+; GFX90A-NEXT:    s_add_i32 s1, s5, s1
+; GFX90A-NEXT:    s_lshl_b64 s[4:5], s[0:1], 5
 ; GFX90A-NEXT:    s_branch .LBB3_2
 ; GFX90A-NEXT:  .LBB3_1: ; %bb12
 ; GFX90A-NEXT:    ; in Loop: Header=BB3_2 Depth=1
@@ -725,66 +719,70 @@ define amdgpu_kernel void @introduced_copy_to_sgpr(i64 %arg, i32 %arg1, i32 %arg
 ; 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[14:15], v[12:13], off
-; GFX90A-NEXT:    s_mov_b32 s5, s4
+; GFX90A-NEXT:    global_load_dwordx2 v[12:13], v[10:11], off
+; GFX90A-NEXT:    s_mov_b32 s9, s8
+; GFX90A-NEXT:    v_pk_mov_b32 v[16:17], s[8:9], s[8:9] op_sel:[0,1]
+; GFX90A-NEXT:    v_pk_mov_b32 v[18:19], s[8:9], s[8:9] op_sel:[0,1]
+; GFX90A-NEXT:    v_pk_mov_b32 v[20:21], s[8:9], s[8:9] 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[16:17], v[6:7], v[6:7] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[18:19], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[20:21], s[4:5], s[4:5] op_sel:[0,1]
-; GFX90A-NEXT:    v_pk_mov_b32 v[22:23], s[4:5], s[4:5] op_sel:[0,1]
+; 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:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    v_add_co_u32_e32 v24, vcc, 1, v14
-; GFX90A-NEXT:    v_addc_co_u32_e32 v25, vcc, 0, v15, vcc
-; GFX90A-NEXT:    v_mul_lo_u32 v25, s2, v25
-; GFX90A-NEXT:    v_mul_hi_u32 v26, s2, v24
-; GFX90A-NEXT:    v_mul_lo_u32 v27, s3, v24
-; GFX90A-NEXT:    v_mul_lo_u32 v31, s2, v24
-; GFX90A-NEXT:    v_add_u32_e32 v24, v26, v25
-; GFX90A-NEXT:    v_add_u32_e32 v32, v24, v27
-; GFX90A-NEXT:    v_pk_mov_b32 v[24:25], s[4:5], s[4:5] op_sel:[0,1]
+; GFX90A-NEXT:    v_readfirstlane_b32 s6, v12
+; GFX90A-NEXT:    v_readfirstlane_b32 s7, v13
+; GFX90A-NEXT:    s_add_u32 s6, s6, 1
+; GFX90A-NEXT:    s_addc_u32 s7, s7, 0
+; GFX90A-NEXT:    s_mul_hi_u32 s9, s2, s6
+; GFX90A-NEXT:    s_mul_i32 s7, s2, s7
+; GFX90A-NEXT:    s_mul_i32 s10, s3, s6
+; GFX90A-NEXT:    s_add_i32 s7, s9, s7
+; GFX90A-NEXT:    s_mul_i32 s6, s2, s6
+; GFX90A-NEXT:    s_add_i32 s7, s7, s10
 ; 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_e32 v14, vcc, v14, v30
-; GFX90A-NEXT:    v_addc_co_u32_e32 v15, vcc, 0, v15, vcc
-; GFX90A-NEXT:    v_add_co_u32_e32 v16, vcc, v16, v10
-; GFX90A-NEXT:    v_addc_co_u32_e32 v17, vcc, v17, v11, vcc
-; GFX90A-NEXT:    v_cmp_gt_i64_e32 vcc, 0, v[14:15]
+; 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, s5
+; GFX90A-NEXT:    v_add_co_u32_e32 v14, vcc, s4, 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:    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_add_co_u32_e32 v26, vcc, v16, v31
-; GFX90A-NEXT:    v_addc_co_u32_e32 v27, vcc, v17, v32, vcc
-; GFX90A-NEXT:    global_load_dword v34, v[26:27], off offset:-12 glc
+; GFX90A-NEXT:    v_mov_b32_e32 v25, s7
+; GFX90A-NEXT:    v_add_co_u32_e32 v24, vcc, s6, 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 v33, v[26:27], off offset:-8 glc
+; GFX90A-NEXT:    global_load_dword v29, v[24:25], off offset:-8 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    global_load_dword v28, v[26:27], off offset:-4 glc
+; GFX90A-NEXT:    global_load_dword v26, v[24:25], off offset:-4 glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    global_load_dword v28, v[26:27], off glc
+; GFX90A-NEXT:    global_load_dword v26, v[24:25], off glc
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    ; kill: killed $vgpr26 killed $vgpr27
-; GFX90A-NEXT:    ds_read_b64 v[26:27], v1
+; GFX90A-NEXT:    ; kill: killed $vgpr24 killed $vgpr25
+; GFX90A-NEXT:    ds_read_b64 v[24:25], v1
 ; GFX90A-NEXT:    s_waitcnt vmcnt(0)
-; GFX90A-NEXT:    ds_read_b64 v[28:29], v0
+; GFX90A-NEXT:    ds_read_b64 v[26:27], v0
 ; GFX90A-NEXT:    s_and_b64 vcc, exec, s[0:1]
 ; 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 v35, v34 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v34, v34
-; GFX90A-NEXT:    v_cvt_f32_f16_sdwa v37, v33 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1
-; GFX90A-NEXT:    v_cvt_f32_f16_e32 v36, v33
-; GFX90A-NEXT:    v_pk_add_f32 v[38:39], v[2:3], v[26:27]
-; GFX90A-NEXT:    v_pk_add_f32 v[40:41], v[26:27], 0 op_sel_hi:[1,0]
-; GFX90A-NEXT:    v_pk_add_f32 v[28:29], v[34:35], v[28:29]
-; GFX90A-NEXT:    v_pk_add_f32 v[26:27], v[36:37], v[26:27]
-; GFX90A-NEXT:    v_pk_add_f32 v[18:19], v[18:19], v[38:39]
-; GFX90A-NEXT:    v_pk_add_f32 v[20:21], v[20:21], v[40:41]
-; GFX90A-NEXT:    v_pk_add_f32 v[22:23], v[22:23], v[28:29]
-; GFX90A-NEXT:    v_pk_add_f32 v[24:25], v[24:25], v[26:27]
+; 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:    s_branch .LBB3_4
 bb:
   %i = load volatile i16, i16 addrspace(4)* undef, align 2

diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
index 089ee88e78fcacdb5b24901e92106a84443e2b72..de75f84110a5206bce18b80b1c11445b82bb1dc6 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-idiv.ll
@@ -51,7 +51,7 @@ define amdgpu_kernel void @udiv_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; GFX6-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; 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
@@ -145,7 +145,7 @@ define amdgpu_kernel void @urem_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s2, v0
 ; GFX6-NEXT:    s_mov_b32 s2, -1
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; 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 v0, v0, s5
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
@@ -161,27 +161,28 @@ define amdgpu_kernel void @urem_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; GFX9-LABEL: urem_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-NEXT:    v_mov_b32_e32 v1, 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, s3
 ; GFX9-NEXT:    s_sub_i32 s4, 0, s3
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; 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_mov_b32_e32 v1, 0
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s3
-; GFX9-NEXT:    v_sub_u32_e32 v0, s2, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; 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 s4, s4, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s4
+; GFX9-NEXT:    s_sub_i32 s4, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX9-NEXT:    s_sub_i32 s4, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s4, s2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
 ; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
 ; GFX9-NEXT:    s_endpgm
   %r = urem i32 %x, %y
@@ -372,7 +373,7 @@ define amdgpu_kernel void @srem_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; 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, 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_lo_u32 v0, v0, s4
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
@@ -390,35 +391,36 @@ define amdgpu_kernel void @srem_i32(i32 addrspace(1)* %out, i32 %x, i32 %y) {
 ; GFX9-LABEL: srem_i32:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; 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 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:    s_ashr_i32 s4, s2, 31
+; GFX9-NEXT:    s_add_i32 s2, s2, s4
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s2, s2, s4
 ; 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_ashr_i32 s4, s2, 31
-; GFX9-NEXT:    s_add_i32 s2, s2, s4
+; 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 s5, s5, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s5
+; GFX9-NEXT:    s_sub_i32 s5, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s5, s2
+; GFX9-NEXT:    s_sub_i32 s5, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s5, s2
 ; GFX9-NEXT:    s_xor_b32 s2, s2, s4
-; 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_mov_b32_e32 v1, 0
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s3
-; GFX9-NEXT:    v_sub_u32_e32 v0, s2, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s4, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s4, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; 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 = srem i32 %x, %y
@@ -697,7 +699,7 @@ define amdgpu_kernel void @srem_i16(i16 addrspace(1)* %out, i16 %x, i16 %y) {
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; 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
@@ -919,7 +921,7 @@ define amdgpu_kernel void @sdiv_i8(i8 addrspace(1)* %out, i8 %x, i8 %y) {
 ; 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_add_i32_e32 v0, vcc, v2, v0
 ; GFX6-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; GFX6-NEXT:    s_endpgm
 ;
@@ -1198,9 +1200,9 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; GFX6-NEXT:    s_sub_i32 s2, 0, s10
 ; 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 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_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
@@ -1226,7 +1228,7 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; 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, v4, v2
+; 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
@@ -1522,7 +1524,7 @@ define amdgpu_kernel void @urem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; 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_add_i32_e32 v2, vcc, v2, v3
 ; 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
@@ -1564,68 +1566,76 @@ define amdgpu_kernel void @urem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; 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:    s_sub_i32 s3, 0, s9
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; 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_mul_f32_e32 v2, 0x4f7ffffe, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, s2, v0
+; 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 s2, s2, s8
+; GFX9-NEXT:    s_sub_i32 s2, s4, s2
+; GFX9-NEXT:    s_sub_i32 s3, s2, s8
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX9-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX9-NEXT:    s_sub_i32 s3, s2, s8
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX9-NEXT:    v_readfirstlane_b32 s12, v1
+; GFX9-NEXT:    s_cselect_b32 s2, s3, 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:    s_mul_hi_u32 s3, s5, s12
+; GFX9-NEXT:    s_mul_i32 s3, s3, s9
+; GFX9-NEXT:    s_sub_i32 s3, s5, s3
+; GFX9-NEXT:    s_sub_i32 s4, s3, s9
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX9-NEXT:    v_mul_lo_u32 v5, s3, v1
-; GFX9-NEXT:    s_sub_i32 s2, 0, s10
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v3
-; GFX9-NEXT:    v_mul_hi_u32 v5, v1, v5
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v3
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v3, s11
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v5
-; GFX9-NEXT:    v_mul_lo_u32 v5, s2, v2
-; GFX9-NEXT:    s_sub_i32 s2, 0, s11
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v5, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s8
-; GFX9-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX9-NEXT:    v_mul_lo_u32 v5, s2, v3
-; GFX9-NEXT:    v_mul_lo_u32 v1, v1, s9
-; GFX9-NEXT:    v_sub_u32_e32 v0, s4, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s8, v0
-; GFX9-NEXT:    v_mul_hi_u32 v5, v3, v5
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, v2, s10
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX9-NEXT:    v_sub_u32_e32 v1, s5, v1
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s8, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v6, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s9, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v3, v3, s11
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s9, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX9-NEXT:    v_sub_u32_e32 v2, s6, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v6, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s10, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s10, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX9-NEXT:    v_sub_u32_e32 v3, s7, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s11, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s11, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s9
+; GFX9-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX9-NEXT:    s_sub_i32 s4, s3, s9
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s9
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s11
+; GFX9-NEXT:    s_cselect_b32 s3, s4, s3
+; GFX9-NEXT:    s_sub_i32 s4, 0, s10
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v2
+; 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 s4, s4, s10
+; GFX9-NEXT:    s_sub_i32 s4, s6, s4
+; GFX9-NEXT:    s_sub_i32 s5, s4, s10
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s10
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX9-NEXT:    s_sub_i32 s5, s4, s10
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s10
+; GFX9-NEXT:    s_cselect_b32 s4, s5, 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 s5, s5, s11
+; GFX9-NEXT:    s_sub_i32 s5, s7, s5
+; GFX9-NEXT:    s_sub_i32 s6, s5, s11
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s11
+; GFX9-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX9-NEXT:    s_sub_i32 s6, s5, s11
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s11
+; GFX9-NEXT:    s_cselect_b32 s5, s6, 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 = urem <4 x i32> %x, %y
@@ -1831,7 +1841,7 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; 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_add_i32_e32 v0, vcc, v0, v2
 ; 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
@@ -1903,7 +1913,7 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; 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_add_i32_e32 v4, vcc, v4, v5
 ; 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
@@ -2323,105 +2333,113 @@ define amdgpu_kernel void @srem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %x
 ; 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 s8, s8, s2
-; GFX9-NEXT:    s_xor_b32 s2, s8, s2
+; GFX9-NEXT:    s_add_i32 s3, s8, s2
+; GFX9-NEXT:    s_xor_b32 s2, s3, s2
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s2
-; GFX9-NEXT:    s_ashr_i32 s3, s9, 31
-; GFX9-NEXT:    s_add_i32 s8, s9, s3
-; GFX9-NEXT:    s_sub_i32 s12, 0, s2
+; GFX9-NEXT:    s_sub_i32 s8, 0, s2
+; GFX9-NEXT:    s_ashr_i32 s3, s4, 31
+; GFX9-NEXT:    s_add_i32 s4, s4, s3
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX9-NEXT:    s_xor_b32 s3, s8, s3
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s3
-; GFX9-NEXT:    s_ashr_i32 s8, s4, 31
+; GFX9-NEXT:    s_xor_b32 s4, s4, s3
 ; 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_add_i32 s4, s4, s8
-; GFX9-NEXT:    s_xor_b32 s4, s4, s8
-; GFX9-NEXT:    v_mul_lo_u32 v2, s12, v0
-; GFX9-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    s_sub_i32 s12, 0, s3
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX9-NEXT:    s_ashr_i32 s9, s5, 31
-; GFX9-NEXT:    v_mul_lo_u32 v3, s12, v1
-; GFX9-NEXT:    s_ashr_i32 s12, s10, 31
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX9-NEXT:    s_add_i32 s5, s5, s9
-; GFX9-NEXT:    s_xor_b32 s5, s5, s9
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s2
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v3
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_sub_u32_e32 v0, s4, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s2, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s2, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v0
-; GFX9-NEXT:    s_add_i32 s2, s10, s12
-; GFX9-NEXT:    s_xor_b32 s2, s2, s12
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s2
-; GFX9-NEXT:    v_mul_lo_u32 v1, v1, s3
-; GFX9-NEXT:    v_xor_b32_e32 v0, s8, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s8, 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 s8, s8, s2
+; GFX9-NEXT:    s_sub_i32 s4, s4, s8
+; GFX9-NEXT:    s_sub_i32 s8, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s4, s8, s4
+; GFX9-NEXT:    s_sub_i32 s8, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s2, s8, s4
+; GFX9-NEXT:    s_ashr_i32 s4, s9, 31
+; GFX9-NEXT:    s_add_i32 s8, s9, s4
+; GFX9-NEXT:    s_xor_b32 s4, s8, s4
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s4
+; GFX9-NEXT:    s_ashr_i32 s8, s5, 31
+; GFX9-NEXT:    s_xor_b32 s2, s2, s3
+; GFX9-NEXT:    s_add_i32 s5, s5, s8
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_sub_i32 s2, s2, s3
+; GFX9-NEXT:    s_xor_b32 s3, s5, s8
+; GFX9-NEXT:    s_sub_i32 s5, 0, s4
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; 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 s5, s5, s4
+; GFX9-NEXT:    s_sub_i32 s3, s3, s5
+; GFX9-NEXT:    s_sub_i32 s5, s3, s4
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s4
+; GFX9-NEXT:    s_cselect_b32 s3, s5, s3
+; GFX9-NEXT:    s_sub_i32 s5, s3, s4
+; GFX9-NEXT:    s_cmp_ge_u32 s3, s4
+; GFX9-NEXT:    s_cselect_b32 s3, s5, s3
+; GFX9-NEXT:    s_ashr_i32 s4, s10, 31
+; GFX9-NEXT:    s_add_i32 s5, s10, s4
+; GFX9-NEXT:    s_xor_b32 s4, s5, s4
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s4
+; GFX9-NEXT:    s_xor_b32 s3, s3, s8
+; GFX9-NEXT:    s_sub_i32 s3, s3, s8
+; GFX9-NEXT:    s_sub_i32 s8, 0, s4
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_ashr_i32 s5, s6, 31
+; GFX9-NEXT:    s_add_i32 s6, s6, s5
+; GFX9-NEXT:    s_xor_b32 s6, s6, s5
+; GFX9-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v0, v0
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; 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, s6, s9
+; GFX9-NEXT:    s_mul_i32 s8, s8, s4
+; GFX9-NEXT:    s_sub_i32 s6, s6, s8
+; GFX9-NEXT:    s_sub_i32 s8, s6, s4
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX9-NEXT:    s_cselect_b32 s6, s8, s6
+; GFX9-NEXT:    s_sub_i32 s8, s6, s4
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX9-NEXT:    s_cselect_b32 s4, s8, s6
+; GFX9-NEXT:    s_ashr_i32 s6, s11, 31
+; GFX9-NEXT:    s_add_i32 s8, s11, s6
+; GFX9-NEXT:    s_xor_b32 s6, s8, s6
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s6
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    s_ashr_i32 s2, s7, 31
+; GFX9-NEXT:    s_xor_b32 s3, s4, s5
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v2, v2
-; GFX9-NEXT:    v_sub_u32_e32 v1, s5, v1
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
+; GFX9-NEXT:    s_add_i32 s4, s7, s2
+; GFX9-NEXT:    s_sub_i32 s3, s3, s5
+; GFX9-NEXT:    s_sub_i32 s5, 0, s6
 ; GFX9-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v2
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v1
-; GFX9-NEXT:    s_sub_i32 s3, 0, s2
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v3, s3, v2
-; GFX9-NEXT:    s_ashr_i32 s3, s11, 31
-; GFX9-NEXT:    s_add_i32 s4, s11, s3
-; GFX9-NEXT:    s_xor_b32 s3, s4, s3
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v5, s3
-; GFX9-NEXT:    v_mul_hi_u32 v3, v2, v3
-; GFX9-NEXT:    s_ashr_i32 s4, s6, 31
-; GFX9-NEXT:    s_add_i32 s5, s6, s4
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v5, v5
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v3
-; GFX9-NEXT:    s_xor_b32 s5, s5, s4
-; GFX9-NEXT:    v_mul_hi_u32 v2, s5, v2
-; GFX9-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v5
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX9-NEXT:    s_sub_i32 s6, 0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v2, v2, s2
-; GFX9-NEXT:    v_xor_b32_e32 v1, s9, v1
-; GFX9-NEXT:    v_mul_lo_u32 v5, s6, v3
-; GFX9-NEXT:    v_subrev_u32_e32 v1, s9, v1
-; GFX9-NEXT:    v_sub_u32_e32 v2, s5, v2
-; GFX9-NEXT:    s_ashr_i32 s5, s7, 31
-; GFX9-NEXT:    v_mul_hi_u32 v5, v3, v5
-; GFX9-NEXT:    s_add_i32 s6, s7, s5
-; GFX9-NEXT:    s_xor_b32 s6, s6, s5
-; GFX9-NEXT:    v_subrev_u32_e32 v6, s2, v2
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, s6, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v6, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s2, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, v3, s3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v2, s4, v2
-; GFX9-NEXT:    v_sub_u32_e32 v3, s6, v3
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s3, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v5, s3, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v3, s5, v3
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s4, v2
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s5, v3
+; GFX9-NEXT:    s_xor_b32 s4, s4, s2
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v2
+; 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, s4, s7
+; GFX9-NEXT:    s_mul_i32 s5, s5, s6
+; GFX9-NEXT:    s_sub_i32 s4, s4, s5
+; GFX9-NEXT:    s_sub_i32 s5, s4, s6
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s6
+; GFX9-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX9-NEXT:    s_sub_i32 s5, s4, s6
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s6
+; GFX9-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX9-NEXT:    s_xor_b32 s4, s4, s2
+; GFX9-NEXT:    s_sub_i32 s2, s4, s2
+; GFX9-NEXT:    v_mov_b32_e32 v2, 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 = srem <4 x i32> %x, %y
@@ -2770,7 +2788,7 @@ define amdgpu_kernel void @urem_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %x
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v3
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GFX6-NEXT:    s_lshr_b32 s4, s7, 16
-; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v1, v2
+; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v2, v1
 ; GFX6-NEXT:    v_mul_f32_e32 v1, v4, v5
 ; GFX6-NEXT:    v_cvt_f32_u32_e32 v5, s4
 ; GFX6-NEXT:    s_lshr_b32 s6, s5, 16
@@ -2995,7 +3013,7 @@ define amdgpu_kernel void @sdiv_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %x
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s8
 ; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 16
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; 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, s6
@@ -3263,7 +3281,7 @@ define amdgpu_kernel void @srem_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %x
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v1|, |v0|
 ; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GFX6-NEXT:    v_mov_b32_e32 v1, s4
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GFX6-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX6-NEXT:    v_alignbit_b32 v2, s7, v2, 16
 ; GFX6-NEXT:    v_bfe_i32 v3, v2, 0, 16
@@ -3283,7 +3301,7 @@ define amdgpu_kernel void @srem_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %x
 ; GFX6-NEXT:    v_or_b32_e32 v3, 1, v3
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v6|, |v4|
 ; GFX6-NEXT:    v_cndmask_b32_e32 v3, 0, v3, vcc
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GFX6-NEXT:    s_sext_i32_i16 s4, s7
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v3, v2
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v3, s4
@@ -3617,7 +3635,7 @@ define amdgpu_kernel void @sdiv_i3(i3 addrspace(1)* %out, i3 %x, i3 %y) {
 ; 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_add_i32_e32 v0, vcc, v2, 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
@@ -3701,7 +3719,7 @@ define amdgpu_kernel void @srem_i3(i3 addrspace(1)* %out, i3 %x, i3 %y) {
 ; 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_add_i32_e32 v0, vcc, v2, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s3
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
@@ -4191,7 +4209,7 @@ define amdgpu_kernel void @sdiv_v3i16(<3 x i16> addrspace(1)* %out, <3 x i16> %x
 ; GFX6-NEXT:    v_mov_b32_e32 v3, s8
 ; GFX6-NEXT:    v_cndmask_b32_e32 v0, 0, v3, vcc
 ; GFX6-NEXT:    s_ashr_i32 s4, s4, 16
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v2
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; 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, s6
@@ -4207,7 +4225,7 @@ define amdgpu_kernel void @sdiv_v3i16(<3 x i16> addrspace(1)* %out, <3 x i16> %x
 ; 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, v1, v3
+; 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
@@ -4435,7 +4453,7 @@ define amdgpu_kernel void @srem_v3i16(<3 x i16> addrspace(1)* %out, <3 x i16> %x
 ; GFX6-NEXT:    v_mov_b32_e32 v6, s4
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v4|
 ; GFX6-NEXT:    v_cndmask_b32_e32 v3, 0, v6, vcc
-; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
+; GFX6-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v3, s7
 ; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, v1, v2
 ; GFX6-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
@@ -4821,7 +4839,7 @@ define amdgpu_kernel void @urem_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX6-NEXT:    v_mul_lo_u32 v2, v3, v2
 ; GFX6-NEXT:    s_lshr_b32 s3, s2, 15
 ; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s3, 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
@@ -5008,7 +5026,7 @@ define amdgpu_kernel void @sdiv_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX6-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX6-NEXT:    v_cndmask_b32_e32 v2, 0, v5, vcc
 ; GFX6-NEXT:    s_bfe_i32 s1, s2, 0xf000f
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v4, s1
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v5, v3
 ; GFX6-NEXT:    s_xor_b32 s0, s1, s0
@@ -5024,7 +5042,7 @@ define amdgpu_kernel void @sdiv_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX6-NEXT:    v_mov_b32_e32 v6, s0
 ; GFX6-NEXT:    v_cndmask_b32_e32 v3, 0, v6, vcc
 ; 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 v3, vcc, v5, v3
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, v0
 ; GFX6-NEXT:    v_rcp_iflag_f32_e32 v6, v4
 ; GFX6-NEXT:    v_xor_b32_e32 v0, v0, v1
@@ -5233,7 +5251,7 @@ define amdgpu_kernel void @srem_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX6-NEXT:    v_cmp_ge_f32_e64 vcc, |v5|, |v4|
 ; GFX6-NEXT:    v_cndmask_b32_e32 v4, 0, v7, vcc
 ; GFX6-NEXT:    v_and_b32_e32 v3, 0x7fff, v2
-; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
+; GFX6-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v4, s0
 ; GFX6-NEXT:    s_bfe_i32 s0, s0, 0xf000f
 ; GFX6-NEXT:    v_cvt_f32_i32_e32 v5, s0
@@ -5256,7 +5274,7 @@ define amdgpu_kernel void @srem_v3i15(<3 x i15> addrspace(1)* %out, <3 x i15> %x
 ; GFX6-NEXT:    v_and_b32_e32 v1, 0x7fff, v0
 ; GFX6-NEXT:    v_cndmask_b32_e32 v5, 0, v8, vcc
 ; 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, v7, v5
 ; 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
@@ -5764,7 +5782,7 @@ define amdgpu_kernel void @urem_i32_oddk_denom(i32 addrspace(1)* %out, i32 %x) {
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
 ; GFX6-NEXT:    v_sub_i32_e32 v1, vcc, s4, v0
 ; GFX6-NEXT:    v_lshrrev_b32_e32 v1, 1, v1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GFX6-NEXT:    v_lshrrev_b32_e32 v0, 20, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s2
 ; GFX6-NEXT:    s_mov_b32 s2, -1
@@ -5998,7 +6016,7 @@ define amdgpu_kernel void @urem_v2i32_pow2_shl_denom(<2 x i32> addrspace(1)* %ou
 ; 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_add_i32_e32 v1, vcc, v3, v1
+; 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 v0, v0, s6
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s7
@@ -6022,46 +6040,49 @@ define amdgpu_kernel void @urem_v2i32_pow2_shl_denom(<2 x i32> addrspace(1)* %ou
 ; 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
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b32 s3, 0x1000, s6
-; GFX9-NEXT:    s_lshl_b32 s2, 0x1000, s7
 ; 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:    s_sub_i32 s7, 0, s2
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; 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, s6, v0
-; GFX9-NEXT:    v_mul_lo_u32 v3, s7, v1
-; 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 v0, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v1, v1, s2
-; GFX9-NEXT:    v_sub_u32_e32 v0, s4, v0
-; GFX9-NEXT:    v_sub_u32_e32 v1, s5, v1
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v4, s2, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v4, s2, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; 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 s6, s6, s3
+; GFX9-NEXT:    s_sub_i32 s4, s4, s6
+; GFX9-NEXT:    s_sub_i32 s6, s4, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s3
+; GFX9-NEXT:    s_cselect_b32 s4, s6, s4
+; GFX9-NEXT:    s_sub_i32 s6, s4, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s3
+; GFX9-NEXT:    v_readfirstlane_b32 s8, v1
+; GFX9-NEXT:    s_cselect_b32 s3, s6, s4
+; 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 s4, s4, s2
+; GFX9-NEXT:    s_sub_i32 s4, s5, s4
+; GFX9-NEXT:    s_sub_i32 s5, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX9-NEXT:    s_sub_i32 s5, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s2, s5, 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
@@ -6179,7 +6200,7 @@ define amdgpu_kernel void @sdiv_i32_pow2_shl_denom(i32 addrspace(1)* %out, i32 %
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
 ; GFX6-NEXT:    s_xor_b32 s1, s1, s0
 ; GFX6-NEXT:    s_xor_b32 s2, s0, s8
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s1, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v0, s3
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, 1, v0
@@ -6483,7 +6504,7 @@ define amdgpu_kernel void @sdiv_v2i32_pow2_shl_denom(<2 x i32> addrspace(1)* %ou
 ; 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_add_i32_e32 v1, vcc, v1, v2
 ; 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
@@ -6588,18 +6609,19 @@ define amdgpu_kernel void @srem_i32_oddk_denom(i32 addrspace(1)* %out, i32 %x) {
 ; GFX6:       ; %bb.0:
 ; GFX6-NEXT:    s_load_dword s4, s[0:1], 0xb
 ; GFX6-NEXT:    v_mov_b32_e32 v0, 0xd9528441
-; GFX6-NEXT:    s_mov_b32 s2, 0x12d8fb
 ; 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_mul_hi_i32 v0, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, s4, v0
-; GFX6-NEXT:    v_lshrrev_b32_e32 v1, 31, v0
-; GFX6-NEXT:    v_ashrrev_i32_e32 v0, 20, v0
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, 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:    v_readfirstlane_b32 s5, v0
+; GFX6-NEXT:    s_add_i32 s5, s5, s4
+; GFX6-NEXT:    s_lshr_b32 s6, s5, 31
+; GFX6-NEXT:    s_ashr_i32 s5, s5, 20
+; GFX6-NEXT:    s_add_i32 s5, s5, s6
+; GFX6-NEXT:    s_mul_i32 s5, s5, 0x12d8fb
+; 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
 ;
@@ -6693,7 +6715,7 @@ define amdgpu_kernel void @srem_i32_pow2_shl_denom(i32 addrspace(1)* %out, i32 %
 ; 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, 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_lo_u32 v0, v0, s4
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
@@ -6711,36 +6733,37 @@ define amdgpu_kernel void @srem_i32_pow2_shl_denom(i32 addrspace(1)* %out, i32 %
 ; GFX9-LABEL: srem_i32_pow2_shl_denom:
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x2c
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; 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 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:    s_ashr_i32 s4, s2, 31
+; GFX9-NEXT:    s_add_i32 s2, s2, s4
 ; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_xor_b32 s2, s2, s4
 ; 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_ashr_i32 s4, s2, 31
-; GFX9-NEXT:    s_add_i32 s2, s2, s4
+; 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 s5, s5, s3
+; GFX9-NEXT:    s_sub_i32 s2, s2, s5
+; GFX9-NEXT:    s_sub_i32 s5, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s5, s2
+; GFX9-NEXT:    s_sub_i32 s5, s2, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s2, s3
+; GFX9-NEXT:    s_cselect_b32 s2, s5, s2
 ; GFX9-NEXT:    s_xor_b32 s2, s2, s4
-; 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_mov_b32_e32 v1, 0
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s3
-; GFX9-NEXT:    v_sub_u32_e32 v0, s2, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v2, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s4, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s4, v0
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; 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
   %shl.y = shl i32 4096, %y
@@ -6919,7 +6942,7 @@ define amdgpu_kernel void @srem_v2i32_pow2_shl_denom(<2 x i32> addrspace(1)* %ou
 ; GFX6-NEXT:    s_sub_i32 s9, 0, s7
 ; GFX6-NEXT:    s_mov_b32 s3, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s2, -1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
+; 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
@@ -6954,62 +6977,65 @@ define amdgpu_kernel void @srem_v2i32_pow2_shl_denom(<2 x i32> addrspace(1)* %ou
 ; 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
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_lshl_b32 s3, 0x1000, s6
-; GFX9-NEXT:    s_ashr_i32 s6, s3, 31
-; GFX9-NEXT:    s_add_i32 s3, s3, s6
-; GFX9-NEXT:    s_lshl_b32 s2, 0x1000, s7
-; GFX9-NEXT:    s_xor_b32 s3, s3, s6
-; GFX9-NEXT:    s_ashr_i32 s7, s2, 31
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX9-NEXT:    s_add_i32 s2, s2, s7
-; GFX9-NEXT:    s_xor_b32 s2, s2, s7
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s2
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX9-NEXT:    s_sub_i32 s8, 0, s3
+; 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 s3, 0x1000, s7
+; GFX9-NEXT:    s_sub_i32 s7, 0, s2
 ; GFX9-NEXT:    s_ashr_i32 s6, s4, 31
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; 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 v0, v0
 ; GFX9-NEXT:    s_add_i32 s4, s4, s6
-; 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, s8, v0
-; GFX9-NEXT:    s_sub_i32 s8, 0, s2
 ; GFX9-NEXT:    s_xor_b32 s4, s4, s6
-; GFX9-NEXT:    v_mul_lo_u32 v3, s8, v1
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX9-NEXT:    s_ashr_i32 s7, s5, 31
-; GFX9-NEXT:    s_add_i32 s5, s5, s7
-; GFX9-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX9-NEXT:    v_add_u32_e32 v0, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX9-NEXT:    s_xor_b32 s5, s5, s7
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v3
-; GFX9-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s3
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v2, 0
-; GFX9-NEXT:    v_mul_lo_u32 v1, v1, s2
-; GFX9-NEXT:    v_sub_u32_e32 v0, s4, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s3, v0
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s3, v0
-; GFX9-NEXT:    v_sub_u32_e32 v1, s5, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s2, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s2, v1
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s6, v0
-; GFX9-NEXT:    v_xor_b32_e32 v1, s7, v1
-; GFX9-NEXT:    v_subrev_u32_e32 v0, s6, v0
-; GFX9-NEXT:    v_subrev_u32_e32 v1, s7, v1
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; 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 s7, s7, s2
+; GFX9-NEXT:    s_sub_i32 s4, s4, s7
+; GFX9-NEXT:    s_sub_i32 s7, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s4, s7, s4
+; GFX9-NEXT:    s_sub_i32 s7, s4, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s4, s2
+; GFX9-NEXT:    s_cselect_b32 s2, s7, s4
+; 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_xor_b32 s2, s2, s6
+; GFX9-NEXT:    s_sub_i32 s2, s2, s6
+; GFX9-NEXT:    s_sub_i32 s6, 0, s3
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v0, v0
+; GFX9-NEXT:    s_ashr_i32 s4, s5, 31
+; GFX9-NEXT:    s_add_i32 s5, s5, s4
+; GFX9-NEXT:    s_xor_b32 s5, s5, s4
+; 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, s5, s7
+; GFX9-NEXT:    s_mul_i32 s6, s6, s3
+; GFX9-NEXT:    s_sub_i32 s5, s5, s6
+; GFX9-NEXT:    s_sub_i32 s6, s5, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s3
+; GFX9-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX9-NEXT:    s_sub_i32 s6, s5, s3
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s3
+; GFX9-NEXT:    s_cselect_b32 s3, s6, 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
@@ -7046,8 +7072,8 @@ define amdgpu_kernel void @udiv_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s5
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v0, s5
 ; GFX6-NEXT:    s_mov_b32 s7, 0xf000
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v0, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
@@ -7070,7 +7096,7 @@ define amdgpu_kernel void @udiv_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; 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_mul_lo_u32 v4, v0, v2
@@ -7151,116 +7177,130 @@ define amdgpu_kernel void @udiv_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
 ; GFX9-NEXT:    v_madmk_f32 v0, v1, 0x438f8000, v0
 ; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_movk_i32 s2, 0xfee0
-; GFX9-NEXT:    s_mov_b32 s3, 0x68958c89
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], 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 v0, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    v_mul_lo_u32 v2, v0, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, s3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, v2
-; 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, v3, v5
-; 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:    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, v0, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, s3
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, s3
-; GFX9-NEXT:    s_movk_i32 s2, 0x11f
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, v0, v5
-; 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, v4, v3
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v6, v1, v5
-; GFX9-NEXT:    v_mul_hi_u32 v5, v1, v5
-; GFX9-NEXT:    s_mov_b32 s3, 0x976a7377
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v4, v5, 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_readfirstlane_b32 s0, v0
+; GFX9-NEXT:    s_mul_i32 s1, s0, 0xfffffee0
+; GFX9-NEXT:    s_mul_hi_u32 s2, s0, 0x68958c89
+; GFX9-NEXT:    s_add_i32 s1, s2, s1
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX9-NEXT:    s_mul_i32 s3, s2, 0x68958c89
+; GFX9-NEXT:    s_add_i32 s1, s1, s3
+; GFX9-NEXT:    s_mul_i32 s9, s0, 0x68958c89
+; GFX9-NEXT:    s_mul_hi_u32 s3, s0, s1
+; GFX9-NEXT:    s_mul_i32 s8, s0, s1
+; GFX9-NEXT:    s_mul_hi_u32 s0, s0, s9
+; GFX9-NEXT:    s_add_u32 s0, s0, s8
+; GFX9-NEXT:    s_addc_u32 s3, 0, s3
+; GFX9-NEXT:    s_mul_hi_u32 s10, s2, s9
+; GFX9-NEXT:    s_mul_i32 s9, s2, s9
+; GFX9-NEXT:    s_add_u32 s0, s0, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s2, s1
+; GFX9-NEXT:    s_addc_u32 s0, s3, s10
+; GFX9-NEXT:    s_addc_u32 s3, s8, 0
+; GFX9-NEXT:    s_mul_i32 s1, s2, s1
+; GFX9-NEXT:    s_add_u32 s0, s0, s1
+; GFX9-NEXT:    s_addc_u32 s1, 0, s3
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s0, s2, s1
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX9-NEXT:    s_mul_i32 s3, s2, 0xfffffee0
+; GFX9-NEXT:    s_mul_hi_u32 s8, s2, 0x68958c89
+; GFX9-NEXT:    s_mul_i32 s1, s0, 0x68958c89
+; GFX9-NEXT:    s_add_i32 s3, s8, s3
+; GFX9-NEXT:    s_add_i32 s3, s3, s1
+; GFX9-NEXT:    s_mul_i32 s9, s2, 0x68958c89
+; GFX9-NEXT:    s_mul_hi_u32 s1, s2, s3
+; GFX9-NEXT:    s_mul_i32 s8, s2, s3
+; GFX9-NEXT:    s_mul_hi_u32 s2, s2, s9
+; GFX9-NEXT:    s_add_u32 s2, s2, s8
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_hi_u32 s10, s0, s9
+; GFX9-NEXT:    s_mul_i32 s9, s0, s9
+; GFX9-NEXT:    s_add_u32 s2, s2, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s0, s3
+; GFX9-NEXT:    s_addc_u32 s1, s1, s10
+; GFX9-NEXT:    s_addc_u32 s2, s8, 0
+; GFX9-NEXT:    s_mul_i32 s3, s0, s3
+; GFX9-NEXT:    s_add_u32 s1, s1, s3
+; GFX9-NEXT:    s_addc_u32 s2, 0, s2
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s1, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s0, s0, s2
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mul_lo_u32 v2, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s6, v0
-; GFX9-NEXT:    v_mul_hi_u32 v4, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, 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, v4, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v4, s7, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s7, v0
-; GFX9-NEXT:    v_mov_b32_e32 v6, 0x11f
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, 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_mul_lo_u32 v2, v0, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, s3
-; GFX9-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_sub_u32_e32 v4, s7, v2
-; GFX9-NEXT:    v_sub_co_u32_e32 v3, vcc, s6, v3
-; GFX9-NEXT:    v_subb_co_u32_e64 v4, s[0:1], v4, v6, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v6, s[0:1], s3, v3
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v4, s[0:1], 0, v4, s[0:1]
-; GFX9-NEXT:    s_movk_i32 s3, 0x11e
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s3, v4
-; GFX9-NEXT:    s_mov_b32 s6, 0x976a7376
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s6, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s2, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, v7, v6, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v6, s[0:1], 2, v0
-; GFX9-NEXT:    v_addc_co_u32_e64 v7, s[0:1], 0, v1, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v8, s[0:1], 1, v0
-; GFX9-NEXT:    v_addc_co_u32_e64 v9, s[0:1], 0, v1, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, v9, v7, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v7, s7
-; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v7, v2, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s3, v2
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v3
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v7, v3, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v8, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    global_store_dwordx2 v5, v[0:1], s[4:5]
+; GFX9-NEXT:    s_mul_i32 s2, s6, s0
+; GFX9-NEXT:    s_mul_hi_u32 s8, s6, s3
+; GFX9-NEXT:    s_mul_hi_u32 s1, s6, s0
+; GFX9-NEXT:    s_add_u32 s2, s8, s2
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_hi_u32 s9, s7, s3
+; GFX9-NEXT:    s_mul_i32 s3, s7, s3
+; GFX9-NEXT:    s_add_u32 s2, s2, s3
+; GFX9-NEXT:    s_mul_hi_u32 s8, s7, s0
+; GFX9-NEXT:    s_addc_u32 s1, s1, s9
+; GFX9-NEXT:    s_addc_u32 s2, s8, 0
+; GFX9-NEXT:    s_mul_i32 s0, s7, s0
+; GFX9-NEXT:    s_add_u32 s3, s1, s0
+; GFX9-NEXT:    s_addc_u32 s2, 0, s2
+; GFX9-NEXT:    s_mul_i32 s0, s3, 0x11f
+; GFX9-NEXT:    s_mul_hi_u32 s8, s3, 0x976a7377
+; GFX9-NEXT:    s_add_i32 s0, s8, s0
+; GFX9-NEXT:    s_mul_i32 s8, s2, 0x976a7377
+; GFX9-NEXT:    s_mul_i32 s9, s3, 0x976a7377
+; GFX9-NEXT:    s_add_i32 s8, s0, s8
+; GFX9-NEXT:    v_mov_b32_e32 v0, s9
+; GFX9-NEXT:    s_sub_i32 s0, s7, s8
+; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s6, v0
+; GFX9-NEXT:    s_mov_b32 s1, 0x976a7377
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s0, 0x11f
+; GFX9-NEXT:    v_subrev_co_u32_e64 v1, s[0:1], s1, v0
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; GFX9-NEXT:    s_subb_u32 s6, s6, 0
+; 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:    v_cmp_ne_u32_e64 s[0:1], 0, v1
+; GFX9-NEXT:    s_cselect_b32 s8, -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:    v_cndmask_b32_e64 v0, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s8
+; 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_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:    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
@@ -7421,8 +7461,8 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)*
 ; 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_subrev_i32_e32 v2, vcc, v0, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; 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
@@ -7443,8 +7483,8 @@ define amdgpu_kernel void @udiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)*
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v1, s6
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v0, s6
 ; GFX6-NEXT:    s_mov_b32 s6, -1
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v5, v0, v4
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
@@ -7702,8 +7742,8 @@ define amdgpu_kernel void @urem_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s3
 ; GFX6-NEXT:    v_mul_lo_u32 v5, v0, s3
 ; GFX6-NEXT:    s_mov_b32 s12, 0x9761f7c9
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v3, v0, v2
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v0, v5
 ; GFX6-NEXT:    v_mul_hi_u32 v6, v0, v2
@@ -7727,7 +7767,7 @@ define amdgpu_kernel void @urem_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v1, s3
 ; GFX6-NEXT:    s_mov_b32 s11, 0xf000
 ; GFX6-NEXT:    s_mov_b32 s10, -1
-; 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, s3
 ; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
 ; GFX6-NEXT:    v_mul_lo_u32 v4, v0, v2
@@ -7764,8 +7804,8 @@ define amdgpu_kernel void @urem_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; GFX6-NEXT:    v_mul_hi_u32 v3, v0, s12
 ; GFX6-NEXT:    v_mul_lo_u32 v1, v1, s12
 ; GFX6-NEXT:    v_mul_lo_u32 v0, v0, s12
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
+; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
 ; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, s7, v1
 ; GFX6-NEXT:    v_mov_b32_e32 v3, 0x11f
 ; GFX6-NEXT:    v_sub_i32_e32 v0, vcc, s6, v0
@@ -7805,115 +7845,128 @@ define amdgpu_kernel void @urem_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0x4f800000
 ; GFX9-NEXT:    v_madmk_f32 v0, v1, 0x438f8000, v0
 ; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_movk_i32 s2, 0xfee0
-; GFX9-NEXT:    s_mov_b32 s3, 0x689e0837
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_mov_b32 s12, 0x9761f7c8
+; 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 v0, v0
 ; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX9-NEXT:    s_movk_i32 s8, 0x11f
-; GFX9-NEXT:    s_mov_b32 s9, 0x9761f7c9
-; GFX9-NEXT:    v_mul_lo_u32 v2, v0, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, s3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v0, s3
-; GFX9-NEXT:    s_mov_b32 s10, 0x9761f7c8
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, v2
-; 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, v3, v5
-; 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:    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, v0, s2
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, s3
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, s3
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_lo_u32 v3, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, v0, v5
-; 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, v4, v3
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v6, v1, v5
-; GFX9-NEXT:    v_mul_hi_u32 v5, v1, v5
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v6
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v4, v5, 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_readfirstlane_b32 s0, v0
+; GFX9-NEXT:    s_mul_i32 s1, s0, 0xfffffee0
+; GFX9-NEXT:    s_mul_hi_u32 s2, s0, 0x689e0837
+; GFX9-NEXT:    s_add_i32 s1, s2, s1
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX9-NEXT:    s_mul_i32 s3, s2, 0x689e0837
+; GFX9-NEXT:    s_add_i32 s1, s1, s3
+; GFX9-NEXT:    s_mul_i32 s9, s0, 0x689e0837
+; GFX9-NEXT:    s_mul_hi_u32 s3, s0, s1
+; GFX9-NEXT:    s_mul_i32 s8, s0, s1
+; GFX9-NEXT:    s_mul_hi_u32 s0, s0, s9
+; GFX9-NEXT:    s_add_u32 s0, s0, s8
+; GFX9-NEXT:    s_addc_u32 s3, 0, s3
+; GFX9-NEXT:    s_mul_hi_u32 s10, s2, s9
+; GFX9-NEXT:    s_mul_i32 s9, s2, s9
+; GFX9-NEXT:    s_add_u32 s0, s0, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s2, s1
+; GFX9-NEXT:    s_addc_u32 s0, s3, s10
+; GFX9-NEXT:    s_addc_u32 s3, s8, 0
+; GFX9-NEXT:    s_mul_i32 s1, s2, s1
+; GFX9-NEXT:    s_add_u32 s0, s0, s1
+; GFX9-NEXT:    s_addc_u32 s1, 0, s3
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s0, s2, s1
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX9-NEXT:    s_mul_i32 s3, s2, 0xfffffee0
+; GFX9-NEXT:    s_mul_hi_u32 s8, s2, 0x689e0837
+; GFX9-NEXT:    s_mul_i32 s1, s0, 0x689e0837
+; GFX9-NEXT:    s_add_i32 s3, s8, s3
+; GFX9-NEXT:    s_add_i32 s3, s3, s1
+; GFX9-NEXT:    s_mul_i32 s9, s2, 0x689e0837
+; GFX9-NEXT:    s_mul_hi_u32 s1, s2, s3
+; GFX9-NEXT:    s_mul_i32 s8, s2, s3
+; GFX9-NEXT:    s_mul_hi_u32 s2, s2, s9
+; GFX9-NEXT:    s_add_u32 s2, s2, s8
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_hi_u32 s10, s0, s9
+; GFX9-NEXT:    s_mul_i32 s9, s0, s9
+; GFX9-NEXT:    s_add_u32 s2, s2, s9
+; GFX9-NEXT:    s_mul_hi_u32 s8, s0, s3
+; GFX9-NEXT:    s_addc_u32 s1, s1, s10
+; GFX9-NEXT:    s_addc_u32 s2, s8, 0
+; GFX9-NEXT:    s_mul_i32 s3, s0, s3
+; GFX9-NEXT:    s_add_u32 s1, s1, s3
+; GFX9-NEXT:    s_addc_u32 s2, 0, s2
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s1, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s0, s0, s2
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mul_lo_u32 v2, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s6, v0
-; GFX9-NEXT:    v_mul_hi_u32 v4, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, 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, v4, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v4, s7, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s7, v0
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, 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_mul_lo_u32 v2, v0, s8
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, s9
-; GFX9-NEXT:    v_mul_lo_u32 v1, v1, s9
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s9
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v1, v2, v1
-; GFX9-NEXT:    v_sub_u32_e32 v2, s7, v1
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0x11f
+; GFX9-NEXT:    s_mul_i32 s2, s6, s0
+; GFX9-NEXT:    s_mul_hi_u32 s8, s6, s3
+; GFX9-NEXT:    s_mul_hi_u32 s1, s6, s0
+; GFX9-NEXT:    s_add_u32 s2, s8, s2
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_hi_u32 s9, s7, s3
+; GFX9-NEXT:    s_mul_i32 s3, s7, s3
+; GFX9-NEXT:    s_add_u32 s2, s2, s3
+; GFX9-NEXT:    s_mul_hi_u32 s8, s7, s0
+; GFX9-NEXT:    s_addc_u32 s1, s1, s9
+; GFX9-NEXT:    s_addc_u32 s2, s8, 0
+; GFX9-NEXT:    s_mul_i32 s0, s7, s0
+; GFX9-NEXT:    s_add_u32 s0, s1, s0
+; GFX9-NEXT:    s_addc_u32 s1, 0, s2
+; GFX9-NEXT:    s_mul_i32 s2, s0, 0x11f
+; GFX9-NEXT:    s_mul_hi_u32 s3, s0, 0x9761f7c9
+; GFX9-NEXT:    s_add_i32 s2, s3, s2
+; GFX9-NEXT:    s_mul_i32 s1, s1, 0x9761f7c9
+; 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:    v_subb_co_u32_e64 v2, s[0:1], v2, v3, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v5, s[0:1], s9, v0
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v6, s[2:3], 0, v2, s[0:1]
-; GFX9-NEXT:    s_movk_i32 s6, 0x11e
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[2:3], s6, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[2:3], s10, v5
-; GFX9-NEXT:    v_subb_co_u32_e64 v2, s[0:1], v2, v3, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s8, v6
-; GFX9-NEXT:    v_subrev_co_u32_e64 v3, s[0:1], s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v8, s[2:3]
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v2, s[0:1], 0, v2, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v6, v2, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v6, s7
-; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v6, v1, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v1
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s10, v0
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s8, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v7, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v3, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    global_store_dwordx2 v4, v[0:1], s[4:5]
+; 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:    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:    s_cmpk_eq_i32 s10, 0x11f
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, 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:    s_subb_u32 s2, s6, 0x11f
+; GFX9-NEXT:    v_subrev_co_u32_e64 v4, s[0:1], s8, 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, 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_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_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 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
@@ -8138,8 +8191,8 @@ define amdgpu_kernel void @sdiv_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; 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:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
+; 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
@@ -8162,7 +8215,7 @@ define amdgpu_kernel void @sdiv_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; 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_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_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
@@ -8434,8 +8487,8 @@ define amdgpu_kernel void @sdiv_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 %
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s5, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s4, v0
 ; GFX6-NEXT:    s_addc_u32 s3, s3, s12
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; 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 v6, v0, v2
@@ -8495,9 +8548,9 @@ define amdgpu_kernel void @sdiv_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 %
 ; GFX6-NEXT:    v_mul_hi_u32 v3, s10, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s11, v0
 ; GFX6-NEXT:    v_mov_b32_e32 v5, s11
-; 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, s10, 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, s3, v2
 ; GFX6-NEXT:    v_sub_i32_e32 v3, vcc, s2, v3
 ; GFX6-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, vcc
@@ -8549,125 +8602,143 @@ define amdgpu_kernel void @sdiv_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 %
 ; GFX9-NEXT:    s_xor_b64 s[8:9], s[4:5], s[2:3]
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s8
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX9-NEXT:    s_sub_u32 s10, 0, s8
-; GFX9-NEXT:    s_subb_u32 s4, 0, s9
+; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_sub_u32 s0, 0, s8
+; GFX9-NEXT:    s_subb_u32 s1, 0, s9
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0x4f800000, v1
-; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; 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_rcp_f32_e32 v1, v0
+; 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 v2, v2
 ; 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, s10, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s10, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s4, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s10, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v6, v1, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; 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 v3, vcc, v3, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v6, 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, s10, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s10, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s4, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s10, v0
-; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v5
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, v1, v5
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, v5
-; 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 v5, vcc, v6, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v7, v4, 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_readfirstlane_b32 s10, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s11, v1
+; GFX9-NEXT:    s_mul_i32 s12, s0, s10
+; GFX9-NEXT:    s_mul_hi_u32 s14, s0, s11
+; GFX9-NEXT:    s_mul_i32 s13, s1, s11
+; GFX9-NEXT:    s_add_i32 s12, s14, s12
+; GFX9-NEXT:    s_add_i32 s12, s12, s13
+; GFX9-NEXT:    s_mul_i32 s15, s0, s11
+; GFX9-NEXT:    s_mul_hi_u32 s13, s11, s12
+; GFX9-NEXT:    s_mul_i32 s14, s11, s12
+; GFX9-NEXT:    s_mul_hi_u32 s11, s11, s15
+; GFX9-NEXT:    s_add_u32 s11, s11, s14
+; GFX9-NEXT:    s_addc_u32 s13, 0, s13
+; GFX9-NEXT:    s_mul_hi_u32 s16, s10, s15
+; GFX9-NEXT:    s_mul_i32 s15, s10, s15
+; GFX9-NEXT:    s_add_u32 s11, s11, s15
+; GFX9-NEXT:    s_mul_hi_u32 s14, s10, s12
+; GFX9-NEXT:    s_addc_u32 s11, s13, s16
+; GFX9-NEXT:    s_addc_u32 s13, s14, 0
+; GFX9-NEXT:    s_mul_i32 s12, s10, s12
+; GFX9-NEXT:    s_add_u32 s11, s11, s12
+; GFX9-NEXT:    s_addc_u32 s12, 0, s13
+; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, s11, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s10, s10, s12
+; GFX9-NEXT:    v_readfirstlane_b32 s12, v1
+; GFX9-NEXT:    s_mul_i32 s11, s0, s10
+; GFX9-NEXT:    s_mul_hi_u32 s13, s0, s12
+; GFX9-NEXT:    s_add_i32 s11, s13, s11
+; GFX9-NEXT:    s_mul_i32 s1, s1, s12
+; GFX9-NEXT:    s_add_i32 s11, s11, s1
+; GFX9-NEXT:    s_mul_i32 s0, s0, s12
+; GFX9-NEXT:    s_mul_hi_u32 s13, s10, s0
+; GFX9-NEXT:    s_mul_i32 s14, s10, s0
+; GFX9-NEXT:    s_mul_i32 s16, s12, s11
+; GFX9-NEXT:    s_mul_hi_u32 s0, s12, s0
+; GFX9-NEXT:    s_mul_hi_u32 s15, s12, s11
+; GFX9-NEXT:    s_add_u32 s0, s0, s16
+; GFX9-NEXT:    s_addc_u32 s12, 0, s15
+; GFX9-NEXT:    s_add_u32 s0, s0, s14
+; GFX9-NEXT:    s_mul_hi_u32 s1, s10, s11
+; GFX9-NEXT:    s_addc_u32 s0, s12, s13
+; GFX9-NEXT:    s_addc_u32 s1, s1, 0
+; GFX9-NEXT:    s_mul_i32 s11, s10, s11
+; GFX9-NEXT:    s_add_u32 s0, s0, s11
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, s0, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s12, s10, s1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_ashr_i32 s10, s7, 31
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
 ; GFX9-NEXT:    s_add_u32 s0, s6, s10
-; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, v0, v2
 ; GFX9-NEXT:    s_mov_b32 s11, s10
 ; GFX9-NEXT:    s_addc_u32 s1, s7, s10
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
 ; GFX9-NEXT:    s_xor_b64 s[6:7], s[0:1], s[10:11]
-; GFX9-NEXT:    v_mul_lo_u32 v2, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s6, v0
-; GFX9-NEXT:    v_mul_hi_u32 v4, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, 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, v4, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v4, s7, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s7, v0
-; GFX9-NEXT:    v_mov_b32_e32 v6, s9
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, 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_mul_lo_u32 v2, s8, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s8, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s9, v0
-; GFX9-NEXT:    v_mov_b32_e32 v5, 0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, s8, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_sub_u32_e32 v4, s7, v2
-; GFX9-NEXT:    v_sub_co_u32_e32 v3, vcc, s6, v3
-; GFX9-NEXT:    v_subb_co_u32_e64 v4, s[0:1], v4, v6, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v6, s[0:1], s8, v3
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v4, s[0:1], 0, v4, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s9, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s8, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s9, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, v7, v6, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v6, s[0:1], 2, v0
-; GFX9-NEXT:    v_addc_co_u32_e64 v7, s[0:1], 0, v1, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v8, s[0:1], 1, v0
-; GFX9-NEXT:    v_addc_co_u32_e64 v9, s[0:1], 0, v1, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, v9, v7, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v7, s7
-; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v7, v2, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v2
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v3
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s9, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v7, v3, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v2
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v8, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
+; GFX9-NEXT:    v_readfirstlane_b32 s13, v1
+; GFX9-NEXT:    s_mul_i32 s1, s6, s12
+; GFX9-NEXT:    s_mul_hi_u32 s14, s6, s13
+; GFX9-NEXT:    s_mul_hi_u32 s0, s6, s12
+; GFX9-NEXT:    s_add_u32 s1, s14, s1
+; GFX9-NEXT:    s_addc_u32 s0, 0, s0
+; GFX9-NEXT:    s_mul_hi_u32 s15, s7, s13
+; GFX9-NEXT:    s_mul_i32 s13, s7, s13
+; GFX9-NEXT:    s_add_u32 s1, s1, s13
+; GFX9-NEXT:    s_mul_hi_u32 s14, s7, s12
+; GFX9-NEXT:    s_addc_u32 s0, s0, s15
+; GFX9-NEXT:    s_addc_u32 s1, s14, 0
+; GFX9-NEXT:    s_mul_i32 s12, s7, s12
+; GFX9-NEXT:    s_add_u32 s12, s0, s12
+; GFX9-NEXT:    s_addc_u32 s13, 0, s1
+; GFX9-NEXT:    s_mul_i32 s0, s8, s13
+; GFX9-NEXT:    s_mul_hi_u32 s1, s8, s12
+; GFX9-NEXT:    s_add_i32 s0, s1, s0
+; GFX9-NEXT:    s_mul_i32 s1, s9, s12
+; GFX9-NEXT:    s_add_i32 s14, s0, s1
+; GFX9-NEXT:    s_mul_i32 s1, s8, s12
+; GFX9-NEXT:    v_mov_b32_e32 v1, s1
+; GFX9-NEXT:    s_sub_i32 s0, s7, s14
+; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, s6, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s0, s9
+; GFX9-NEXT:    v_subrev_co_u32_e64 v2, s[0:1], s8, v1
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; 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:    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_cndmask_b32_e64 v1, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v3, s14
+; 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_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_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_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s0, v0
-; GFX9-NEXT:    v_xor_b32_e32 v1, s1, v1
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
-; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_store_dwordx2 v5, v[0:1], s[4:5]
+; 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
+; GFX9-NEXT:    global_store_dwordx2 v0, v[1:2], s[4:5]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl i64 4096, %y
   %r = sdiv i64 %x, %shl.y
@@ -8772,9 +8843,9 @@ define amdgpu_kernel void @ssdiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)*
 ; 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:    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, v0, s6
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
+; GFX6-NEXT:    v_sub_i32_e32 v2, vcc, v2, v0
 ; 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
@@ -8798,7 +8869,7 @@ define amdgpu_kernel void @ssdiv_v2i64_mixed_pow2k_denom(<2 x i64> addrspace(1)*
 ; 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:    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, s6
 ; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
 ; GFX6-NEXT:    v_mul_lo_u32 v6, v0, v2
@@ -9046,8 +9117,8 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; 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, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; 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
@@ -9106,9 +9177,9 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; 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
@@ -9159,9 +9230,9 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; 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, v2, v5
+; 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, v2, v6
+; 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
@@ -9273,257 +9344,293 @@ define amdgpu_kernel void @sdiv_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX9-NEXT:    s_mov_b64 s[2:3], 0x1000
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b64 s[10:11], s[2:3], s10
 ; GFX9-NEXT:    s_lshl_b64 s[2:3], s[2:3], s8
-; GFX9-NEXT:    s_ashr_i32 s12, s3, 31
-; GFX9-NEXT:    s_add_u32 s2, s2, s12
-; GFX9-NEXT:    s_mov_b32 s13, s12
-; GFX9-NEXT:    s_addc_u32 s3, s3, s12
-; GFX9-NEXT:    s_xor_b64 s[8:9], s[2:3], s[12:13]
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s8
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX9-NEXT:    s_sub_u32 s2, 0, s8
-; GFX9-NEXT:    s_subb_u32 s3, 0, s9
-; GFX9-NEXT:    s_ashr_i32 s14, s5, 31
-; GFX9-NEXT:    v_mac_f32_e32 v0, 0x4f800000, v1
-; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_mov_b32 s15, s14
+; GFX9-NEXT:    s_ashr_i32 s8, s3, 31
+; GFX9-NEXT:    s_add_u32 s2, s2, s8
+; GFX9-NEXT:    s_mov_b32 s9, s8
+; GFX9-NEXT:    s_addc_u32 s3, s3, s8
+; GFX9-NEXT:    s_xor_b64 s[12:13], s[2:3], s[8:9]
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s12
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s13
+; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX9-NEXT:    s_sub_u32 s0, 0, s12
+; GFX9-NEXT:    s_subb_u32 s1, 0, s13
+; GFX9-NEXT:    v_mac_f32_e32 v0, 0x4f800000, v1
+; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
 ; 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, s2, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s2, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v6, v1, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; 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 v3, vcc, v3, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v6, 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, s2, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s2, v0
-; GFX9-NEXT:    s_add_u32 s2, s4, s14
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v5
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, v1, v5
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, v5
-; 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 v5, vcc, v6, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v7, v4, 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_addc_u32 s3, s5, s14
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    s_xor_b64 s[4:5], s[2:3], s[14:15]
-; GFX9-NEXT:    v_mul_lo_u32 v2, s4, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s4, v0
-; GFX9-NEXT:    v_mul_hi_u32 v4, s4, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, s5, v1
-; GFX9-NEXT:    v_mul_lo_u32 v1, s5, v1
-; 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_mul_lo_u32 v4, s5, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s5, v0
-; GFX9-NEXT:    v_mov_b32_e32 v6, s9
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, v0, v1
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v2, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v3, s8, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, s8, v1
-; GFX9-NEXT:    v_mul_lo_u32 v5, s9, v1
-; GFX9-NEXT:    s_xor_b64 s[12:13], s[14:15], s[12:13]
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-NEXT:    v_add_u32_e32 v3, v4, v3
-; GFX9-NEXT:    v_mul_lo_u32 v4, s8, v1
-; GFX9-NEXT:    v_add_u32_e32 v3, v3, v5
-; GFX9-NEXT:    v_sub_u32_e32 v5, s5, v3
-; GFX9-NEXT:    v_sub_co_u32_e32 v4, vcc, s4, v4
-; GFX9-NEXT:    v_subb_co_u32_e64 v5, s[0:1], v5, v6, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v6, s[0:1], s8, v4
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v5, s[0:1], 0, v5, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s8, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s9, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, v7, v6, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v6, s[0:1], 2, v1
-; GFX9-NEXT:    v_addc_co_u32_e64 v7, s[0:1], 0, v2, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v8, s[0:1], 1, v1
-; GFX9-NEXT:    v_addc_co_u32_e64 v9, s[0:1], 0, v2, s[0:1]
+; GFX9-NEXT:    v_readfirstlane_b32 s14, v1
+; GFX9-NEXT:    v_readfirstlane_b32 s15, v0
+; GFX9-NEXT:    s_mul_i32 s16, s0, s14
+; GFX9-NEXT:    s_mul_hi_u32 s18, s0, s15
+; GFX9-NEXT:    s_mul_i32 s17, s1, s15
+; GFX9-NEXT:    s_add_i32 s16, s18, s16
+; GFX9-NEXT:    s_add_i32 s16, s16, s17
+; GFX9-NEXT:    s_mul_i32 s19, s0, s15
+; GFX9-NEXT:    s_mul_hi_u32 s17, s15, s16
+; GFX9-NEXT:    s_mul_i32 s18, s15, s16
+; GFX9-NEXT:    s_mul_hi_u32 s15, s15, s19
+; GFX9-NEXT:    s_add_u32 s15, s15, s18
+; GFX9-NEXT:    s_addc_u32 s17, 0, s17
+; GFX9-NEXT:    s_mul_hi_u32 s20, s14, s19
+; GFX9-NEXT:    s_mul_i32 s19, s14, s19
+; GFX9-NEXT:    s_add_u32 s15, s15, s19
+; GFX9-NEXT:    s_mul_hi_u32 s18, s14, s16
+; GFX9-NEXT:    s_addc_u32 s15, s17, s20
+; GFX9-NEXT:    s_addc_u32 s17, s18, 0
+; GFX9-NEXT:    s_mul_i32 s16, s14, s16
+; GFX9-NEXT:    s_add_u32 s15, s15, s16
+; GFX9-NEXT:    s_addc_u32 s16, 0, s17
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s15, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s14, s14, s16
+; GFX9-NEXT:    v_readfirstlane_b32 s16, v0
+; GFX9-NEXT:    s_mul_i32 s15, s0, s14
+; GFX9-NEXT:    s_mul_hi_u32 s17, s0, s16
+; GFX9-NEXT:    s_add_i32 s15, s17, s15
+; GFX9-NEXT:    s_mul_i32 s1, s1, s16
+; GFX9-NEXT:    s_add_i32 s15, s15, s1
+; GFX9-NEXT:    s_mul_i32 s0, s0, s16
+; GFX9-NEXT:    s_mul_hi_u32 s17, s14, s0
+; GFX9-NEXT:    s_mul_i32 s18, s14, s0
+; GFX9-NEXT:    s_mul_i32 s20, s16, s15
+; GFX9-NEXT:    s_mul_hi_u32 s0, s16, s0
+; GFX9-NEXT:    s_mul_hi_u32 s19, s16, s15
+; GFX9-NEXT:    s_add_u32 s0, s0, s20
+; GFX9-NEXT:    s_addc_u32 s16, 0, s19
+; GFX9-NEXT:    s_add_u32 s0, s0, s18
+; GFX9-NEXT:    s_mul_hi_u32 s1, s14, s15
+; GFX9-NEXT:    s_addc_u32 s0, s16, s17
+; GFX9-NEXT:    s_addc_u32 s1, s1, 0
+; GFX9-NEXT:    s_mul_i32 s15, s14, s15
+; GFX9-NEXT:    s_add_u32 s0, s0, s15
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s16, s14, s1
+; GFX9-NEXT:    s_ashr_i32 s14, s5, 31
+; GFX9-NEXT:    s_add_u32 s0, s4, s14
+; GFX9-NEXT:    s_mov_b32 s15, s14
+; GFX9-NEXT:    s_addc_u32 s1, s5, s14
+; GFX9-NEXT:    s_xor_b64 s[4:5], s[0:1], s[14:15]
+; GFX9-NEXT:    v_readfirstlane_b32 s17, v0
+; GFX9-NEXT:    s_mul_i32 s1, s4, s16
+; GFX9-NEXT:    s_mul_hi_u32 s18, s4, s17
+; GFX9-NEXT:    s_mul_hi_u32 s0, s4, s16
+; GFX9-NEXT:    s_add_u32 s1, s18, s1
+; GFX9-NEXT:    s_addc_u32 s0, 0, s0
+; GFX9-NEXT:    s_mul_hi_u32 s19, s5, s17
+; GFX9-NEXT:    s_mul_i32 s17, s5, s17
+; GFX9-NEXT:    s_add_u32 s1, s1, s17
+; GFX9-NEXT:    s_mul_hi_u32 s18, s5, s16
+; GFX9-NEXT:    s_addc_u32 s0, s0, s19
+; GFX9-NEXT:    s_addc_u32 s1, s18, 0
+; GFX9-NEXT:    s_mul_i32 s16, s5, s16
+; GFX9-NEXT:    s_add_u32 s16, s0, s16
+; GFX9-NEXT:    s_addc_u32 s17, 0, s1
+; GFX9-NEXT:    s_mul_i32 s0, s12, s17
+; GFX9-NEXT:    s_mul_hi_u32 s1, s12, s16
+; GFX9-NEXT:    s_add_i32 s0, s1, s0
+; GFX9-NEXT:    s_mul_i32 s1, s13, s16
+; GFX9-NEXT:    s_add_i32 s18, s0, s1
+; GFX9-NEXT:    s_mul_i32 s1, s12, s16
+; GFX9-NEXT:    v_mov_b32_e32 v0, s1
+; GFX9-NEXT:    s_sub_i32 s0, s5, s18
+; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s4, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s4, s0, s13
+; GFX9-NEXT:    v_subrev_co_u32_e64 v1, s[0:1], s12, v0
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; 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:    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_cndmask_b32_e64 v0, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v2, s18
+; 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_cmp_ne_u32_e64 s[0:1], 0, v5
-; GFX9-NEXT:    s_add_u32 s10, s10, s4
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, v9, v7, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v7, s5
+; GFX9-NEXT:    s_add_u32 s8, s10, s4
 ; GFX9-NEXT:    s_mov_b32 s5, s4
-; GFX9-NEXT:    s_addc_u32 s11, s11, s4
-; GFX9-NEXT:    s_xor_b64 s[10:11], s[10:11], s[4:5]
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v9, s10
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v10, s11
-; GFX9-NEXT:    v_subb_co_u32_e32 v3, vcc, v7, v3, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v3
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v4, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s9, v3
-; GFX9-NEXT:    v_mac_f32_e32 v9, 0x4f800000, v10
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v7, v4, vcc
-; GFX9-NEXT:    v_rcp_f32_e32 v4, v9
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v3
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v8, v6, s[0:1]
-; GFX9-NEXT:    v_mul_f32_e32 v4, 0x5f7ffffc, v4
-; GFX9-NEXT:    v_mul_f32_e32 v5, 0x2f800000, v4
-; GFX9-NEXT:    v_trunc_f32_e32 v5, v5
-; GFX9-NEXT:    v_mac_f32_e32 v4, 0xcf800000, v5
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v4, v4
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v5, v5
-; GFX9-NEXT:    s_sub_u32 s0, 0, s10
-; GFX9-NEXT:    s_subb_u32 s1, 0, s11
-; GFX9-NEXT:    v_mul_hi_u32 v6, s0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v7, s0, v5
-; GFX9-NEXT:    v_mul_lo_u32 v8, s1, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v3, s0, v4
-; GFX9-NEXT:    v_add_u32_e32 v6, v6, v7
-; GFX9-NEXT:    v_add_u32_e32 v6, v6, v8
-; GFX9-NEXT:    v_mul_lo_u32 v7, v4, v6
-; GFX9-NEXT:    v_mul_hi_u32 v8, v4, v3
-; GFX9-NEXT:    v_mul_hi_u32 v9, v4, v6
-; GFX9-NEXT:    v_mul_hi_u32 v10, v5, v6
-; GFX9-NEXT:    v_mul_lo_u32 v6, v5, v6
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, v8, v7
-; GFX9-NEXT:    v_addc_co_u32_e32 v8, vcc, 0, v9, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v9, v5, v3
-; GFX9-NEXT:    v_mul_hi_u32 v3, v5, v3
-; GFX9-NEXT:    s_ashr_i32 s8, s7, 31
-; GFX9-NEXT:    s_mov_b32 s9, s8
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, v7, v9
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v8, v3, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v7, vcc, 0, v10, vcc
-; 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_add_co_u32_e32 v3, vcc, v4, v3
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v5, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, s0, v4
-; GFX9-NEXT:    v_mul_hi_u32 v6, s0, v3
-; GFX9-NEXT:    v_mul_lo_u32 v7, s1, v3
-; GFX9-NEXT:    v_mul_lo_u32 v8, s0, v3
-; GFX9-NEXT:    s_add_u32 s0, s6, s8
-; GFX9-NEXT:    v_add_u32_e32 v5, v6, v5
-; GFX9-NEXT:    v_add_u32_e32 v5, v5, v7
-; GFX9-NEXT:    v_mul_lo_u32 v9, v3, v5
-; GFX9-NEXT:    v_mul_hi_u32 v10, v3, v8
-; GFX9-NEXT:    v_mul_hi_u32 v11, v3, v5
-; GFX9-NEXT:    v_mul_hi_u32 v7, v4, v8
-; GFX9-NEXT:    v_mul_lo_u32 v8, v4, v8
-; GFX9-NEXT:    v_mul_hi_u32 v6, v4, v5
-; GFX9-NEXT:    v_add_co_u32_e32 v9, vcc, v10, v9
-; GFX9-NEXT:    v_addc_co_u32_e32 v10, vcc, 0, v11, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, v4, v5
-; GFX9-NEXT:    v_add_co_u32_e32 v8, vcc, v9, v8
-; GFX9-NEXT:    v_addc_co_u32_e32 v7, vcc, v10, v7, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v5, vcc, v7, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; GFX9-NEXT:    s_addc_u32 s1, s7, s8
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v4, v6, vcc
-; GFX9-NEXT:    s_xor_b64 s[6:7], s[0:1], s[8:9]
-; GFX9-NEXT:    v_mul_lo_u32 v5, s6, v4
-; GFX9-NEXT:    v_mul_hi_u32 v6, s6, v3
-; GFX9-NEXT:    v_mul_hi_u32 v8, s6, v4
-; GFX9-NEXT:    v_mul_hi_u32 v9, s7, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, s7, v4
-; GFX9-NEXT:    v_add_co_u32_e32 v5, vcc, v6, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v8, s7, v3
-; GFX9-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX9-NEXT:    v_xor_b32_e32 v1, s12, v1
-; GFX9-NEXT:    v_xor_b32_e32 v2, s13, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v5, vcc, v5, v8
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v6, v3, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v9, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, s10, v4
-; GFX9-NEXT:    v_mul_hi_u32 v6, s10, v3
-; GFX9-NEXT:    v_mul_lo_u32 v8, s11, v3
-; GFX9-NEXT:    v_mov_b32_e32 v7, s13
-; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s12, v1
-; GFX9-NEXT:    v_add_u32_e32 v5, v6, v5
-; GFX9-NEXT:    v_mul_lo_u32 v6, s10, v3
-; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v2, v7, vcc
-; GFX9-NEXT:    v_add_u32_e32 v5, v5, v8
-; GFX9-NEXT:    v_sub_u32_e32 v7, s7, v5
-; GFX9-NEXT:    v_mov_b32_e32 v8, s11
-; GFX9-NEXT:    v_sub_co_u32_e32 v6, vcc, s6, v6
-; GFX9-NEXT:    v_subb_co_u32_e64 v7, s[0:1], v7, v8, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v8, s[0:1], s10, v6
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v7, s[0:1], 0, v7, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s11, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[0:1], s10, v8
-; GFX9-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], s11, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, v9, v8, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v8, s[0:1], 2, v3
-; GFX9-NEXT:    v_addc_co_u32_e64 v9, s[0:1], 0, v4, s[0:1]
-; GFX9-NEXT:    v_add_co_u32_e64 v10, s[0:1], 1, v3
-; GFX9-NEXT:    v_addc_co_u32_e64 v11, s[0:1], 0, v4, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, v11, v9, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v9, s7
-; GFX9-NEXT:    v_subb_co_u32_e32 v5, vcc, v9, v5, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s11, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s10, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s11, v5
-; GFX9-NEXT:    v_cndmask_b32_e32 v5, v9, v6, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, v10, v8, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    s_xor_b64 s[0:1], s[8:9], s[4:5]
-; GFX9-NEXT:    v_cndmask_b32_e32 v4, v4, v7, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v3, s0, v3
-; GFX9-NEXT:    v_xor_b32_e32 v4, s1, v4
+; GFX9-NEXT:    s_addc_u32 s9, s11, s4
+; 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:    v_mac_f32_e32 v2, 0x4f800000, v3
+; GFX9-NEXT:    v_rcp_f32_e32 v2, v2
+; GFX9-NEXT:    s_sub_u32 s0, 0, s8
+; GFX9-NEXT:    v_mov_b32_e32 v6, s1
+; GFX9-NEXT:    s_subb_u32 s1, 0, s9
+; GFX9-NEXT:    v_mul_f32_e32 v2, 0x5f7ffffc, v2
+; GFX9-NEXT:    v_mul_f32_e32 v3, 0x2f800000, v2
+; GFX9-NEXT:    v_trunc_f32_e32 v3, v3
+; GFX9-NEXT:    v_mac_f32_e32 v2, 0xcf800000, v3
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
+; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v5, v6, vcc
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s13, v3
+; GFX9-NEXT:    s_mul_hi_u32 s12, s0, s10
+; GFX9-NEXT:    s_mul_i32 s14, s0, s13
+; GFX9-NEXT:    s_mul_i32 s11, s1, s10
+; GFX9-NEXT:    s_add_i32 s12, s12, s14
+; GFX9-NEXT:    s_add_i32 s12, s12, s11
+; GFX9-NEXT:    s_mul_i32 s15, s0, s10
+; GFX9-NEXT:    s_mul_hi_u32 s11, s10, s12
+; GFX9-NEXT:    s_mul_i32 s14, s10, s12
+; GFX9-NEXT:    s_mul_hi_u32 s10, s10, s15
+; GFX9-NEXT:    s_add_u32 s10, s10, s14
+; GFX9-NEXT:    s_addc_u32 s11, 0, s11
+; GFX9-NEXT:    s_mul_hi_u32 s16, s13, s15
+; GFX9-NEXT:    s_mul_i32 s15, s13, s15
+; GFX9-NEXT:    s_add_u32 s10, s10, s15
+; GFX9-NEXT:    s_mul_hi_u32 s14, s13, s12
+; GFX9-NEXT:    s_addc_u32 s10, s11, s16
+; GFX9-NEXT:    s_addc_u32 s11, s14, 0
+; GFX9-NEXT:    s_mul_i32 s12, s13, s12
+; GFX9-NEXT:    s_add_u32 s10, s10, s12
+; GFX9-NEXT:    s_addc_u32 s11, 0, s11
+; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, s10, v2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s10, s13, s11
+; GFX9-NEXT:    v_readfirstlane_b32 s12, v2
+; GFX9-NEXT:    s_mul_i32 s11, s0, s10
+; GFX9-NEXT:    s_mul_hi_u32 s13, s0, s12
+; GFX9-NEXT:    s_add_i32 s11, s13, s11
+; GFX9-NEXT:    s_mul_i32 s1, s1, s12
+; GFX9-NEXT:    s_add_i32 s11, s11, s1
+; GFX9-NEXT:    s_mul_i32 s0, s0, s12
+; GFX9-NEXT:    s_mul_hi_u32 s13, s10, s0
+; GFX9-NEXT:    s_mul_i32 s14, s10, s0
+; GFX9-NEXT:    s_mul_i32 s16, s12, s11
+; GFX9-NEXT:    s_mul_hi_u32 s0, s12, s0
+; GFX9-NEXT:    s_mul_hi_u32 s15, s12, s11
+; GFX9-NEXT:    s_add_u32 s0, s0, s16
+; GFX9-NEXT:    s_addc_u32 s12, 0, s15
+; GFX9-NEXT:    s_add_u32 s0, s0, s14
+; GFX9-NEXT:    s_mul_hi_u32 s1, s10, s11
+; GFX9-NEXT:    s_addc_u32 s0, s12, s13
+; GFX9-NEXT:    s_addc_u32 s1, s1, 0
+; GFX9-NEXT:    s_mul_i32 s11, s10, s11
+; GFX9-NEXT:    s_add_u32 s0, s0, s11
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, s0, v2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s12, s10, s1
+; GFX9-NEXT:    s_ashr_i32 s10, s7, 31
+; GFX9-NEXT:    s_add_u32 s0, s6, s10
+; GFX9-NEXT:    s_mov_b32 s11, s10
+; GFX9-NEXT:    s_addc_u32 s1, s7, s10
+; GFX9-NEXT:    s_xor_b64 s[6:7], s[0:1], s[10:11]
+; GFX9-NEXT:    v_readfirstlane_b32 s13, v2
+; GFX9-NEXT:    s_mul_i32 s1, s6, s12
+; GFX9-NEXT:    s_mul_hi_u32 s14, s6, s13
+; GFX9-NEXT:    s_mul_hi_u32 s0, s6, s12
+; GFX9-NEXT:    s_add_u32 s1, s14, s1
+; GFX9-NEXT:    s_addc_u32 s0, 0, s0
+; GFX9-NEXT:    s_mul_hi_u32 s15, s7, s13
+; GFX9-NEXT:    s_mul_i32 s13, s7, s13
+; GFX9-NEXT:    s_add_u32 s1, s1, s13
+; GFX9-NEXT:    s_mul_hi_u32 s14, s7, s12
+; GFX9-NEXT:    s_addc_u32 s0, s0, s15
+; GFX9-NEXT:    s_addc_u32 s1, s14, 0
+; GFX9-NEXT:    s_mul_i32 s12, s7, s12
+; GFX9-NEXT:    s_add_u32 s12, s0, s12
+; GFX9-NEXT:    s_addc_u32 s13, 0, s1
+; GFX9-NEXT:    s_mul_i32 s0, s8, s13
+; GFX9-NEXT:    s_mul_hi_u32 s1, s8, s12
+; GFX9-NEXT:    s_add_i32 s0, s1, s0
+; GFX9-NEXT:    s_mul_i32 s1, s9, s12
+; GFX9-NEXT:    s_add_i32 s14, s0, s1
+; GFX9-NEXT:    s_mul_i32 s1, s8, s12
+; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; GFX9-NEXT:    s_sub_i32 s0, s7, s14
+; GFX9-NEXT:    v_sub_co_u32_e32 v2, vcc, s6, v2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s0, s9
+; GFX9-NEXT:    v_subrev_co_u32_e64 v3, s[0:1], s8, v2
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; 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_subrev_co_u32_e32 v3, vcc, s0, v3
-; GFX9-NEXT:    v_subb_co_u32_e32 v4, vcc, v4, v5, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v6, s0
+; 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_cndmask_b32_e64 v2, 0, -1, vcc
+; GFX9-NEXT:    v_mov_b32_e32 v5, s14
+; 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_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_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
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    global_store_dwordx4 v0, v[1:4], s[2:3]
+; GFX9-NEXT:    global_store_dwordx4 v4, v[0:3], s[2:3]
 ; GFX9-NEXT:    s_endpgm
   %shl.y = shl <2 x i64> <i64 4096, i64 4096>, %y
   %r = sdiv <2 x i64> %x, %shl.y
@@ -9559,8 +9666,8 @@ define amdgpu_kernel void @srem_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ; 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:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_subrev_i32_e32 v2, vcc, v0, v2
+; 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
@@ -9653,113 +9760,127 @@ define amdgpu_kernel void @srem_i64_oddk_denom(i64 addrspace(1)* %out, i64 %x) {
 ;
 ; GFX9-LABEL: srem_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:    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 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:    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 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_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s8, s0, s2
 ; 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 v4, s0, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, 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, v4, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v4, s1, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s1, v0
-; GFX9-NEXT:    s_mov_b32 s3, 0x12d8fb
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, 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_mul_lo_u32 v1, v1, s3
-; GFX9-NEXT:    v_mul_hi_u32 v2, v0, s3
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s3
-; GFX9-NEXT:    v_mov_b32_e32 v3, 0
-; GFX9-NEXT:    v_add_u32_e32 v1, v2, v1
-; GFX9-NEXT:    v_mov_b32_e32 v2, s1
+; 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_mul_i32 s6, s6, 0x12d8fb
+; GFX9-NEXT:    v_mov_b32_e32 v0, s3
+; GFX9-NEXT:    s_add_i32 s8, s8, s6
 ; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s0, v0
-; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v2, v1, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e32 v2, vcc, s3, v0
-; GFX9-NEXT:    v_subbrev_co_u32_e32 v4, vcc, 0, v1, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e32 v5, vcc, s3, v2
-; GFX9-NEXT:    v_subbrev_co_u32_e32 v6, vcc, 0, v4, vcc
-; GFX9-NEXT:    s_mov_b32 s0, 0x12d8fa
-; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s0, v2
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v7, -1, v7, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v7
-; GFX9-NEXT:    v_cmp_lt_u32_e64 s[0:1], s0, v0
-; GFX9-NEXT:    v_cndmask_b32_e32 v4, v4, v6, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, s[0:1]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[0:1], 0, v1
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, -1, v6, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v6
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, v0, v2, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v4, s[0:1]
+; 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:    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:    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_mov_b32 s6, 0x12d8fa
+; GFX9-NEXT:    v_cmp_lt_u32_e32 vcc, s6, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s0, 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_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, v1
-; GFX9-NEXT:    v_mov_b32_e32 v2, s2
+; 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_subb_co_u32_e32 v1, vcc, v1, v2, vcc
-; GFX9-NEXT:    global_store_dwordx2 v3, v[0:1], s[4:5]
+; 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:    s_endpgm
   %r = srem i64 %x, 1235195
   store i64 %r, i64 addrspace(1)* %out
@@ -9855,8 +9976,8 @@ define amdgpu_kernel void @srem_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 %
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s5, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s4, v0
 ; GFX6-NEXT:    s_addc_u32 s3, s3, s10
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; 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 v6, v0, v2
@@ -9968,123 +10089,140 @@ define amdgpu_kernel void @srem_i64_pow2_shl_denom(i64 addrspace(1)* %out, i64 %
 ; GFX9-NEXT:    s_xor_b64 s[8:9], s[2:3], s[4:5]
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s8
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s9
-; GFX9-NEXT:    s_sub_u32 s2, 0, s8
-; GFX9-NEXT:    s_subb_u32 s3, 0, s9
 ; GFX9-NEXT:    s_load_dwordx4 s[4:7], s[0:1], 0x24
+; GFX9-NEXT:    s_sub_u32 s0, 0, s8
+; GFX9-NEXT:    s_subb_u32 s1, 0, s9
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0x4f800000, v1
-; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
+; GFX9-NEXT:    v_rcp_f32_e32 v1, v0
+; 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 v2, v2
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v1
+; GFX9-NEXT:    s_mul_i32 s10, s0, s2
+; GFX9-NEXT:    s_mul_hi_u32 s12, s0, s3
+; GFX9-NEXT:    s_mul_i32 s11, s1, s3
+; GFX9-NEXT:    s_add_i32 s10, s12, s10
+; GFX9-NEXT:    s_add_i32 s10, s10, s11
+; GFX9-NEXT:    s_mul_i32 s13, s0, s3
+; GFX9-NEXT:    s_mul_hi_u32 s11, s3, s10
+; GFX9-NEXT:    s_mul_i32 s12, s3, s10
+; GFX9-NEXT:    s_mul_hi_u32 s3, s3, s13
+; GFX9-NEXT:    s_add_u32 s3, s3, s12
+; GFX9-NEXT:    s_addc_u32 s11, 0, s11
+; GFX9-NEXT:    s_mul_hi_u32 s14, s2, s13
+; GFX9-NEXT:    s_mul_i32 s13, s2, s13
+; GFX9-NEXT:    s_add_u32 s3, s3, s13
+; GFX9-NEXT:    s_mul_hi_u32 s12, s2, s10
+; GFX9-NEXT:    s_addc_u32 s3, s11, s14
+; GFX9-NEXT:    s_addc_u32 s11, s12, 0
+; GFX9-NEXT:    s_mul_i32 s10, s2, s10
+; GFX9-NEXT:    s_add_u32 s3, s3, s10
+; GFX9-NEXT:    s_addc_u32 s10, 0, s11
+; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, s3, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s2, s2, s10
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v1
+; GFX9-NEXT:    s_mul_i32 s3, s0, s2
+; GFX9-NEXT:    s_mul_hi_u32 s11, s0, s10
+; GFX9-NEXT:    s_add_i32 s3, s11, s3
+; GFX9-NEXT:    s_mul_i32 s1, s1, s10
+; GFX9-NEXT:    s_add_i32 s3, s3, s1
+; GFX9-NEXT:    s_mul_i32 s0, s0, s10
+; GFX9-NEXT:    s_mul_hi_u32 s11, s2, s0
+; GFX9-NEXT:    s_mul_i32 s12, s2, s0
+; GFX9-NEXT:    s_mul_i32 s14, s10, s3
+; GFX9-NEXT:    s_mul_hi_u32 s0, s10, s0
+; GFX9-NEXT:    s_mul_hi_u32 s13, s10, s3
+; GFX9-NEXT:    s_add_u32 s0, s0, s14
+; GFX9-NEXT:    s_addc_u32 s10, 0, s13
+; GFX9-NEXT:    s_add_u32 s0, s0, s12
+; GFX9-NEXT:    s_mul_hi_u32 s1, s2, s3
+; GFX9-NEXT:    s_addc_u32 s0, s10, s11
+; GFX9-NEXT:    s_addc_u32 s1, s1, 0
+; GFX9-NEXT:    s_mul_i32 s3, s2, s3
+; GFX9-NEXT:    s_add_u32 s0, s0, s3
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, s0, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s2, s2, s1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_ashr_i32 s10, s7, 31
-; 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:    s_add_u32 s0, s6, s10
 ; GFX9-NEXT:    s_mov_b32 s11, s10
-; GFX9-NEXT:    v_mul_lo_u32 v2, s2, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s2, v0
 ; GFX9-NEXT:    s_addc_u32 s1, s7, s10
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v6, v1, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; 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 v3, vcc, v3, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v6, 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, s2, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s2, v0
 ; GFX9-NEXT:    s_xor_b64 s[6:7], s[0:1], s[10:11]
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v5
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, v1, v5
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, v5
-; 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 v5, vcc, v6, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v7, v4, 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:    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 v4, s6, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, 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, v4, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v4, s7, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s7, v0
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, 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_mul_lo_u32 v1, s8, v1
-; GFX9-NEXT:    v_mul_hi_u32 v2, s8, v0
-; GFX9-NEXT:    v_mul_lo_u32 v3, s9, v0
-; GFX9-NEXT:    v_mul_lo_u32 v0, s8, v0
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
-; GFX9-NEXT:    v_add_u32_e32 v1, v2, v1
-; GFX9-NEXT:    v_add_u32_e32 v1, v1, v3
-; GFX9-NEXT:    v_sub_u32_e32 v2, s7, v1
-; GFX9-NEXT:    v_mov_b32_e32 v3, s9
-; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s6, v0
-; GFX9-NEXT:    v_subb_co_u32_e64 v2, s[0:1], v2, v3, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v5, s[0:1], s8, v0
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v6, s[2:3], 0, v2, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s9, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s8, v5
-; GFX9-NEXT:    v_subb_co_u32_e64 v2, s[0:1], v2, v3, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s9, v6
-; GFX9-NEXT:    v_subrev_co_u32_e64 v3, s[0:1], s8, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v8, s[2:3]
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v2, s[0:1], 0, v2, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v6, v2, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v6, s7
-; GFX9-NEXT:    v_subb_co_u32_e32 v1, vcc, v6, v1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s9, v1
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v7, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v1
+; GFX9-NEXT:    s_mul_i32 s1, s6, s2
+; GFX9-NEXT:    s_mul_hi_u32 s11, s6, s3
+; GFX9-NEXT:    s_mul_hi_u32 s0, s6, s2
+; GFX9-NEXT:    s_add_u32 s1, s11, s1
+; GFX9-NEXT:    s_addc_u32 s0, 0, s0
+; GFX9-NEXT:    s_mul_hi_u32 s12, s7, s3
+; GFX9-NEXT:    s_mul_i32 s3, s7, s3
+; GFX9-NEXT:    s_add_u32 s1, s1, s3
+; GFX9-NEXT:    s_mul_hi_u32 s11, s7, s2
+; GFX9-NEXT:    s_addc_u32 s0, s0, s12
+; GFX9-NEXT:    s_addc_u32 s1, s11, 0
+; GFX9-NEXT:    s_mul_i32 s2, s7, s2
+; GFX9-NEXT:    s_add_u32 s0, s0, s2
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_i32 s1, s8, s1
+; GFX9-NEXT:    s_mul_hi_u32 s2, s8, s0
+; GFX9-NEXT:    s_add_i32 s1, s2, s1
+; GFX9-NEXT:    s_mul_i32 s2, s9, s0
+; GFX9-NEXT:    s_mul_i32 s0, s8, s0
+; GFX9-NEXT:    s_add_i32 s11, s1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    s_sub_i32 s1, s7, s11
+; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, s6, v1
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s1, s9
+; GFX9-NEXT:    v_subrev_co_u32_e64 v2, s[0:1], s8, v1
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; GFX9-NEXT:    s_subb_u32 s12, s6, 0
+; GFX9-NEXT:    s_cmp_ge_u32 s12, s9
+; GFX9-NEXT:    s_cselect_b32 s13, -1, 0
+; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s8, v2
+; GFX9-NEXT:    s_cmp_eq_u32 s12, s9
+; 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:    v_cmp_ne_u32_e64 s[0:1], 0, v3
+; GFX9-NEXT:    s_cselect_b32 s3, -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_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 v1, v1, v2, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v2, v5, v3, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v0, s10, v0
+; GFX9-NEXT:    v_cndmask_b32_e32 v3, v6, v3, vcc
 ; GFX9-NEXT:    v_xor_b32_e32 v1, s10, v1
-; GFX9-NEXT:    v_mov_b32_e32 v2, s10
-; GFX9-NEXT:    v_subrev_co_u32_e32 v0, vcc, s10, 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:    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
   %r = srem i64 %x, %shl.y
@@ -10212,8 +10350,8 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; GFX6-NEXT:    v_mul_lo_u32 v5, s3, v0
 ; GFX6-NEXT:    v_mul_lo_u32 v4, s2, v0
 ; GFX6-NEXT:    s_mov_b32 s10, -1
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; 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
@@ -10322,9 +10460,9 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; GFX6-NEXT:    s_subb_u32 s1, 0, s5
 ; GFX6-NEXT:    v_mul_lo_u32 v6, s1, v3
 ; GFX6-NEXT:    s_ashr_i32 s14, s7, 31
-; GFX6-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; 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, v2, v6
+; 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
@@ -10390,8 +10528,8 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; 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
@@ -10434,6 +10572,7 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_load_dwordx8 s[4:11], s[0:1], 0x34
 ; GFX9-NEXT:    s_mov_b64 s[2:3], 0x1000
+; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_lshl_b64 s[10:11], s[2:3], s10
 ; GFX9-NEXT:    s_lshl_b64 s[2:3], s[2:3], s8
@@ -10444,243 +10583,276 @@ define amdgpu_kernel void @srem_v2i64_pow2_shl_denom(<2 x i64> addrspace(1)* %ou
 ; GFX9-NEXT:    s_xor_b64 s[12:13], s[2:3], s[8:9]
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s12
 ; GFX9-NEXT:    v_cvt_f32_u32_e32 v1, s13
-; GFX9-NEXT:    s_sub_u32 s2, 0, s12
-; GFX9-NEXT:    s_subb_u32 s3, 0, s13
-; GFX9-NEXT:    s_ashr_i32 s8, s5, 31
+; GFX9-NEXT:    s_load_dwordx2 s[8:9], s[0:1], 0x24
+; GFX9-NEXT:    s_sub_u32 s0, 0, s12
+; GFX9-NEXT:    s_subb_u32 s1, 0, s13
 ; GFX9-NEXT:    v_mac_f32_e32 v0, 0x4f800000, v1
 ; GFX9-NEXT:    v_rcp_f32_e32 v0, v0
-; GFX9-NEXT:    s_mov_b32 s9, s8
 ; 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, s2, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s2, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v5
-; GFX9-NEXT:    v_mul_hi_u32 v3, v0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v5, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v6, v1, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, v1, v4
-; GFX9-NEXT:    v_mul_hi_u32 v8, v1, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; 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 v3, vcc, v3, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v5, v6, 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, s2, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_mul_lo_u32 v4, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v5, s2, v0
-; GFX9-NEXT:    s_add_u32 s2, s4, s8
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_mul_lo_u32 v6, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v7, v0, v5
-; GFX9-NEXT:    v_mul_hi_u32 v8, v0, v2
-; GFX9-NEXT:    v_mul_hi_u32 v4, v1, v5
-; GFX9-NEXT:    v_mul_lo_u32 v5, v1, v5
-; 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 v5, vcc, v6, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v7, v4, 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_addc_u32 s3, s5, s8
-; GFX9-NEXT:    v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
-; GFX9-NEXT:    s_xor_b64 s[14:15], s[2:3], s[8:9]
-; GFX9-NEXT:    v_mul_lo_u32 v2, s14, v1
-; GFX9-NEXT:    v_mul_hi_u32 v3, s14, v0
-; GFX9-NEXT:    v_mul_hi_u32 v4, s14, v1
-; GFX9-NEXT:    v_mul_hi_u32 v5, s15, v1
-; GFX9-NEXT:    v_mul_lo_u32 v1, s15, v1
-; 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_mul_lo_u32 v4, s15, v0
-; GFX9-NEXT:    v_mul_hi_u32 v0, s15, v0
-; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
-; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, v2, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, v3, v0, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v2, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v1, vcc, v0, v1
-; GFX9-NEXT:    v_addc_co_u32_e32 v0, vcc, 0, v2, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v2, s12, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, s12, v1
-; GFX9-NEXT:    v_mul_lo_u32 v4, s13, v1
-; GFX9-NEXT:    v_mul_lo_u32 v1, s12, v1
-; GFX9-NEXT:    v_mov_b32_e32 v0, 0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_add_u32_e32 v2, v2, v4
-; GFX9-NEXT:    v_sub_u32_e32 v3, s15, v2
-; GFX9-NEXT:    v_mov_b32_e32 v4, s13
-; GFX9-NEXT:    v_sub_co_u32_e32 v1, vcc, s14, v1
-; GFX9-NEXT:    v_subb_co_u32_e64 v3, s[0:1], v3, v4, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v5, s[0:1], s12, v1
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v6, s[2:3], 0, v3, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s13, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s12, v5
-; GFX9-NEXT:    v_cndmask_b32_e64 v8, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s13, v6
-; GFX9-NEXT:    v_cndmask_b32_e64 v7, v7, v8, s[2:3]
-; GFX9-NEXT:    s_ashr_i32 s2, s11, 31
-; GFX9-NEXT:    v_subb_co_u32_e64 v3, s[0:1], v3, v4, s[0:1]
-; GFX9-NEXT:    s_add_u32 s10, s10, s2
-; GFX9-NEXT:    v_subrev_co_u32_e64 v4, s[0:1], s12, v5
-; GFX9-NEXT:    s_mov_b32 s3, s2
-; GFX9-NEXT:    s_addc_u32 s11, s11, s2
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v3, s[0:1], 0, v3, s[0:1]
-; GFX9-NEXT:    s_xor_b64 s[10:11], s[10:11], s[2:3]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v7
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v7, s10
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v8, s11
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v6, v3, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v6, s15
-; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v6, v2, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s13, v2
-; GFX9-NEXT:    v_mac_f32_e32 v7, 0x4f800000, v8
-; GFX9-NEXT:    v_cndmask_b32_e64 v6, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s12, v1
-; GFX9-NEXT:    v_rcp_f32_e32 v7, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s13, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v6, v6, v9, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v6
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX9-NEXT:    s_mul_i32 s14, s0, s2
+; GFX9-NEXT:    s_mul_hi_u32 s16, s0, s3
+; GFX9-NEXT:    s_mul_i32 s15, s1, s3
+; GFX9-NEXT:    s_add_i32 s14, s16, s14
+; GFX9-NEXT:    s_add_i32 s14, s14, s15
+; GFX9-NEXT:    s_mul_i32 s17, s0, s3
+; GFX9-NEXT:    s_mul_hi_u32 s15, s3, s14
+; GFX9-NEXT:    s_mul_i32 s16, s3, s14
+; GFX9-NEXT:    s_mul_hi_u32 s3, s3, s17
+; GFX9-NEXT:    s_add_u32 s3, s3, s16
+; GFX9-NEXT:    s_addc_u32 s15, 0, s15
+; GFX9-NEXT:    s_mul_hi_u32 s18, s2, s17
+; GFX9-NEXT:    s_mul_i32 s17, s2, s17
+; GFX9-NEXT:    s_add_u32 s3, s3, s17
+; GFX9-NEXT:    s_mul_hi_u32 s16, s2, s14
+; GFX9-NEXT:    s_addc_u32 s3, s15, s18
+; GFX9-NEXT:    s_addc_u32 s15, s16, 0
+; GFX9-NEXT:    s_mul_i32 s14, s2, s14
+; GFX9-NEXT:    s_add_u32 s3, s3, s14
+; GFX9-NEXT:    s_addc_u32 s14, 0, s15
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s3, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s2, s2, s14
+; GFX9-NEXT:    v_readfirstlane_b32 s14, v0
+; GFX9-NEXT:    s_mul_i32 s3, s0, s2
+; GFX9-NEXT:    s_mul_hi_u32 s15, s0, s14
+; GFX9-NEXT:    s_add_i32 s3, s15, s3
+; GFX9-NEXT:    s_mul_i32 s1, s1, s14
+; GFX9-NEXT:    s_add_i32 s3, s3, s1
+; GFX9-NEXT:    s_mul_i32 s0, s0, s14
+; GFX9-NEXT:    s_mul_hi_u32 s15, s2, s0
+; GFX9-NEXT:    s_mul_i32 s16, s2, s0
+; GFX9-NEXT:    s_mul_i32 s18, s14, s3
+; GFX9-NEXT:    s_mul_hi_u32 s0, s14, s0
+; GFX9-NEXT:    s_mul_hi_u32 s17, s14, s3
+; GFX9-NEXT:    s_add_u32 s0, s0, s18
+; GFX9-NEXT:    s_addc_u32 s14, 0, s17
+; GFX9-NEXT:    s_add_u32 s0, s0, s16
+; GFX9-NEXT:    s_mul_hi_u32 s1, s2, s3
+; GFX9-NEXT:    s_addc_u32 s0, s14, s15
+; GFX9-NEXT:    s_addc_u32 s1, s1, 0
+; GFX9-NEXT:    s_mul_i32 s3, s2, s3
+; GFX9-NEXT:    s_add_u32 s0, s0, s3
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v0, vcc, s0, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s2, s2, s1
+; GFX9-NEXT:    s_ashr_i32 s14, s5, 31
+; GFX9-NEXT:    s_add_u32 s0, s4, s14
+; GFX9-NEXT:    s_mov_b32 s15, s14
+; GFX9-NEXT:    s_addc_u32 s1, s5, s14
+; GFX9-NEXT:    s_xor_b64 s[4:5], s[0:1], s[14:15]
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX9-NEXT:    s_mul_i32 s1, s4, s2
+; GFX9-NEXT:    s_mul_hi_u32 s15, s4, s3
+; GFX9-NEXT:    s_mul_hi_u32 s0, s4, s2
+; GFX9-NEXT:    s_add_u32 s1, s15, s1
+; GFX9-NEXT:    s_addc_u32 s0, 0, s0
+; GFX9-NEXT:    s_mul_hi_u32 s16, s5, s3
+; GFX9-NEXT:    s_mul_i32 s3, s5, s3
+; GFX9-NEXT:    s_add_u32 s1, s1, s3
+; GFX9-NEXT:    s_mul_hi_u32 s15, s5, s2
+; GFX9-NEXT:    s_addc_u32 s0, s0, s16
+; GFX9-NEXT:    s_addc_u32 s1, s15, 0
+; GFX9-NEXT:    s_mul_i32 s2, s5, s2
+; GFX9-NEXT:    s_add_u32 s0, s0, s2
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_i32 s1, s12, s1
+; GFX9-NEXT:    s_mul_hi_u32 s2, s12, s0
+; GFX9-NEXT:    s_add_i32 s1, s2, s1
+; GFX9-NEXT:    s_mul_i32 s2, s13, s0
+; GFX9-NEXT:    s_mul_i32 s0, s12, s0
+; GFX9-NEXT:    s_add_i32 s15, s1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v0, s0
+; GFX9-NEXT:    s_sub_i32 s1, s5, s15
+; GFX9-NEXT:    v_sub_co_u32_e32 v0, vcc, s4, v0
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s4, s1, s13
+; GFX9-NEXT:    v_subrev_co_u32_e64 v1, s[0:1], s12, v0
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; GFX9-NEXT:    s_subb_u32 s16, s4, 0
+; GFX9-NEXT:    s_cmp_ge_u32 s16, s13
+; GFX9-NEXT:    s_cselect_b32 s17, -1, 0
+; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s12, v1
+; GFX9-NEXT:    s_cmp_eq_u32 s16, s13
+; 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:    v_cmp_ne_u32_e64 s[0:1], 0, v2
+; GFX9-NEXT:    s_cselect_b32 s3, -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_cselect_b64 vcc, -1, 0
+; GFX9-NEXT:    v_cndmask_b32_e64 v1, v1, v3, s[0:1]
+; 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:    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_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_mul_f32_e32 v3, 0x2f800000, v2
+; GFX9-NEXT:    v_trunc_f32_e32 v3, v3
+; GFX9-NEXT:    v_mac_f32_e32 v2, 0xcf800000, v3
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v2, v2
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v3, v3
+; GFX9-NEXT:    s_sub_u32 s0, 0, s4
+; GFX9-NEXT:    s_subb_u32 s1, 0, s5
+; GFX9-NEXT:    v_readfirstlane_b32 s2, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s11, v3
+; GFX9-NEXT:    s_mul_hi_u32 s10, s0, s2
+; GFX9-NEXT:    s_mul_i32 s12, s0, s11
+; GFX9-NEXT:    s_mul_i32 s3, s1, s2
+; GFX9-NEXT:    s_add_i32 s10, s10, s12
+; GFX9-NEXT:    s_add_i32 s10, s10, s3
+; GFX9-NEXT:    s_mul_i32 s13, s0, s2
+; GFX9-NEXT:    s_mul_hi_u32 s3, s2, s10
+; GFX9-NEXT:    s_mul_i32 s12, s2, s10
+; GFX9-NEXT:    s_mul_hi_u32 s2, s2, s13
+; GFX9-NEXT:    s_add_u32 s2, s2, s12
+; GFX9-NEXT:    s_addc_u32 s3, 0, s3
+; GFX9-NEXT:    s_mul_hi_u32 s14, s11, s13
+; GFX9-NEXT:    s_mul_i32 s13, s11, s13
+; GFX9-NEXT:    s_add_u32 s2, s2, s13
+; GFX9-NEXT:    s_mul_hi_u32 s12, s11, s10
+; GFX9-NEXT:    s_addc_u32 s2, s3, s14
+; GFX9-NEXT:    s_addc_u32 s3, s12, 0
+; GFX9-NEXT:    s_mul_i32 s10, s11, s10
+; GFX9-NEXT:    s_add_u32 s2, s2, s10
+; GFX9-NEXT:    s_addc_u32 s3, 0, s3
+; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, s2, v2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s2, s11, s3
+; GFX9-NEXT:    v_readfirstlane_b32 s10, v2
+; GFX9-NEXT:    s_mul_i32 s3, s0, s2
+; GFX9-NEXT:    s_mul_hi_u32 s11, s0, s10
+; GFX9-NEXT:    s_add_i32 s3, s11, s3
+; GFX9-NEXT:    s_mul_i32 s1, s1, s10
+; GFX9-NEXT:    s_add_i32 s3, s3, s1
+; GFX9-NEXT:    s_mul_i32 s0, s0, s10
+; GFX9-NEXT:    s_mul_hi_u32 s11, s2, s0
+; GFX9-NEXT:    s_mul_i32 s12, s2, s0
+; GFX9-NEXT:    s_mul_i32 s14, s10, s3
+; GFX9-NEXT:    s_mul_hi_u32 s0, s10, s0
+; GFX9-NEXT:    s_mul_hi_u32 s13, s10, s3
+; GFX9-NEXT:    s_add_u32 s0, s0, s14
+; GFX9-NEXT:    s_addc_u32 s10, 0, s13
+; GFX9-NEXT:    s_add_u32 s0, s0, s12
+; GFX9-NEXT:    s_mul_hi_u32 s1, s2, s3
+; GFX9-NEXT:    s_addc_u32 s0, s10, s11
+; GFX9-NEXT:    s_addc_u32 s1, s1, 0
+; GFX9-NEXT:    s_mul_i32 s3, s2, s3
+; GFX9-NEXT:    s_add_u32 s0, s0, s3
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    v_add_co_u32_e32 v2, vcc, s0, v2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_addc_u32 s2, s2, s1
+; GFX9-NEXT:    s_ashr_i32 s10, s7, 31
+; GFX9-NEXT:    s_add_u32 s0, s6, s10
+; GFX9-NEXT:    s_mov_b32 s11, s10
+; GFX9-NEXT:    s_addc_u32 s1, s7, s10
+; GFX9-NEXT:    s_xor_b64 s[6:7], s[0:1], s[10:11]
+; GFX9-NEXT:    v_readfirstlane_b32 s3, v2
+; GFX9-NEXT:    s_mul_i32 s1, s6, s2
+; GFX9-NEXT:    s_mul_hi_u32 s11, s6, s3
+; GFX9-NEXT:    s_mul_hi_u32 s0, s6, s2
+; GFX9-NEXT:    s_add_u32 s1, s11, s1
+; GFX9-NEXT:    s_addc_u32 s0, 0, s0
+; GFX9-NEXT:    s_mul_hi_u32 s12, s7, s3
+; GFX9-NEXT:    s_mul_i32 s3, s7, s3
+; GFX9-NEXT:    s_add_u32 s1, s1, s3
+; GFX9-NEXT:    s_mul_hi_u32 s11, s7, s2
+; GFX9-NEXT:    s_addc_u32 s0, s0, s12
+; GFX9-NEXT:    s_addc_u32 s1, s11, 0
+; GFX9-NEXT:    s_mul_i32 s2, s7, s2
+; GFX9-NEXT:    s_add_u32 s0, s0, s2
+; GFX9-NEXT:    s_addc_u32 s1, 0, s1
+; GFX9-NEXT:    s_mul_i32 s1, s4, s1
+; GFX9-NEXT:    s_mul_hi_u32 s2, s4, s0
+; GFX9-NEXT:    s_add_i32 s1, s2, s1
+; GFX9-NEXT:    s_mul_i32 s2, s5, s0
+; GFX9-NEXT:    s_mul_i32 s0, s4, s0
+; GFX9-NEXT:    s_add_i32 s11, s1, s2
+; GFX9-NEXT:    v_mov_b32_e32 v2, s0
+; GFX9-NEXT:    s_sub_i32 s1, s7, s11
+; GFX9-NEXT:    v_sub_co_u32_e32 v2, vcc, s6, v2
+; GFX9-NEXT:    s_cmp_lg_u64 vcc, 0
+; GFX9-NEXT:    s_subb_u32 s6, s1, s5
+; GFX9-NEXT:    v_subrev_co_u32_e64 v3, s[0:1], s4, v2
+; GFX9-NEXT:    s_cmp_lg_u64 s[0:1], 0
+; GFX9-NEXT:    s_subb_u32 s12, s6, 0
+; GFX9-NEXT:    s_cmp_ge_u32 s12, s5
+; GFX9-NEXT:    s_cselect_b32 s13, -1, 0
+; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s4, v3
+; GFX9-NEXT:    s_cmp_eq_u32 s12, s5
+; 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:    v_cmp_ne_u32_e64 s[0:1], 0, v5
+; GFX9-NEXT:    s_cselect_b32 s3, -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_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 v2, v2, v3, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v3, v5, v4, s[0:1]
-; GFX9-NEXT:    v_mul_f32_e32 v4, 0x5f7ffffc, v7
-; GFX9-NEXT:    v_mul_f32_e32 v5, 0x2f800000, v4
-; GFX9-NEXT:    v_trunc_f32_e32 v5, v5
-; GFX9-NEXT:    v_mac_f32_e32 v4, 0xcf800000, v5
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v4, v4
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v5, v5
-; GFX9-NEXT:    s_sub_u32 s0, 0, s10
-; GFX9-NEXT:    s_subb_u32 s1, 0, s11
-; GFX9-NEXT:    v_mul_hi_u32 v6, s0, v4
-; GFX9-NEXT:    v_mul_lo_u32 v7, s0, v5
-; GFX9-NEXT:    v_mul_lo_u32 v8, s1, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v3, s0, v4
-; GFX9-NEXT:    v_add_u32_e32 v6, v6, v7
-; GFX9-NEXT:    v_add_u32_e32 v6, v6, v8
-; GFX9-NEXT:    v_mul_lo_u32 v7, v4, v6
-; GFX9-NEXT:    v_mul_hi_u32 v8, v4, v3
-; GFX9-NEXT:    v_mul_hi_u32 v9, v4, v6
-; GFX9-NEXT:    v_mul_hi_u32 v10, v5, v6
-; GFX9-NEXT:    v_mul_lo_u32 v6, v5, v6
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, v8, v7
-; GFX9-NEXT:    v_addc_co_u32_e32 v8, vcc, 0, v9, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v9, v5, v3
-; GFX9-NEXT:    v_mul_hi_u32 v3, v5, v3
-; GFX9-NEXT:    s_ashr_i32 s12, s7, 31
-; GFX9-NEXT:    s_mov_b32 s13, s12
-; GFX9-NEXT:    v_add_co_u32_e32 v7, vcc, v7, v9
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v8, v3, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v7, vcc, 0, v10, vcc
-; 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_add_co_u32_e32 v3, vcc, v4, v3
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v5, v6, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, s0, v4
-; GFX9-NEXT:    v_mul_hi_u32 v6, s0, v3
-; GFX9-NEXT:    v_mul_lo_u32 v7, s1, v3
-; GFX9-NEXT:    v_mul_lo_u32 v8, s0, v3
-; GFX9-NEXT:    s_add_u32 s0, s6, s12
-; GFX9-NEXT:    v_add_u32_e32 v5, v6, v5
-; GFX9-NEXT:    v_add_u32_e32 v5, v5, v7
-; GFX9-NEXT:    v_mul_lo_u32 v9, v3, v5
-; GFX9-NEXT:    v_mul_hi_u32 v10, v3, v8
-; GFX9-NEXT:    v_mul_hi_u32 v11, v3, v5
-; GFX9-NEXT:    v_mul_hi_u32 v7, v4, v8
-; GFX9-NEXT:    v_mul_lo_u32 v8, v4, v8
-; GFX9-NEXT:    v_mul_hi_u32 v6, v4, v5
-; GFX9-NEXT:    v_add_co_u32_e32 v9, vcc, v10, v9
-; GFX9-NEXT:    v_addc_co_u32_e32 v10, vcc, 0, v11, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v5, v4, v5
-; GFX9-NEXT:    v_add_co_u32_e32 v8, vcc, v9, v8
-; GFX9-NEXT:    v_addc_co_u32_e32 v7, vcc, v10, v7, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v5, vcc, v7, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v6, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v5
-; GFX9-NEXT:    s_addc_u32 s1, s7, s12
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, v4, v6, vcc
-; GFX9-NEXT:    s_xor_b64 s[6:7], s[0:1], s[12:13]
-; GFX9-NEXT:    v_mul_lo_u32 v5, s6, v4
-; GFX9-NEXT:    v_mul_hi_u32 v6, s6, v3
-; GFX9-NEXT:    v_mul_hi_u32 v8, s6, v4
-; GFX9-NEXT:    v_mul_hi_u32 v9, s7, v4
-; GFX9-NEXT:    v_mul_lo_u32 v4, s7, v4
-; GFX9-NEXT:    v_add_co_u32_e32 v5, vcc, v6, v5
-; GFX9-NEXT:    v_addc_co_u32_e32 v6, vcc, 0, v8, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v8, s7, v3
-; GFX9-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX9-NEXT:    v_xor_b32_e32 v1, s8, v1
-; GFX9-NEXT:    v_xor_b32_e32 v2, s8, v2
-; GFX9-NEXT:    v_add_co_u32_e32 v5, vcc, v5, v8
-; GFX9-NEXT:    v_addc_co_u32_e32 v3, vcc, v6, v3, vcc
-; GFX9-NEXT:    v_addc_co_u32_e32 v5, vcc, 0, v9, vcc
-; GFX9-NEXT:    v_add_co_u32_e32 v3, vcc, v3, v4
-; GFX9-NEXT:    v_addc_co_u32_e32 v4, vcc, 0, v5, vcc
-; GFX9-NEXT:    v_mul_lo_u32 v4, s10, v4
-; GFX9-NEXT:    v_mul_hi_u32 v5, s10, v3
-; GFX9-NEXT:    v_mul_lo_u32 v6, s11, v3
-; GFX9-NEXT:    v_mul_lo_u32 v3, s10, v3
-; GFX9-NEXT:    v_mov_b32_e32 v7, s8
-; GFX9-NEXT:    v_subrev_co_u32_e32 v1, vcc, s8, v1
-; GFX9-NEXT:    v_add_u32_e32 v4, v5, v4
-; GFX9-NEXT:    v_subb_co_u32_e32 v2, vcc, v2, v7, vcc
-; GFX9-NEXT:    v_add_u32_e32 v4, v4, v6
-; GFX9-NEXT:    v_sub_u32_e32 v5, s7, v4
-; GFX9-NEXT:    v_mov_b32_e32 v6, s11
-; GFX9-NEXT:    v_sub_co_u32_e32 v3, vcc, s6, v3
-; GFX9-NEXT:    v_subb_co_u32_e64 v5, s[0:1], v5, v6, vcc
-; GFX9-NEXT:    v_subrev_co_u32_e64 v7, s[0:1], s10, v3
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v8, s[2:3], 0, v5, s[0:1]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s11, v8
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_le_u32_e64 s[2:3], s10, v7
-; GFX9-NEXT:    v_subb_co_u32_e64 v5, s[0:1], v5, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e64 v10, 0, -1, s[2:3]
-; GFX9-NEXT:    v_cmp_eq_u32_e64 s[2:3], s11, v8
-; GFX9-NEXT:    v_subrev_co_u32_e64 v6, s[0:1], s10, v7
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, v9, v10, s[2:3]
-; GFX9-NEXT:    v_subbrev_co_u32_e64 v5, s[0:1], 0, v5, s[0:1]
-; GFX9-NEXT:    v_cmp_ne_u32_e64 s[0:1], 0, v9
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, v8, v5, s[0:1]
-; GFX9-NEXT:    v_mov_b32_e32 v8, s7
-; GFX9-NEXT:    v_subb_co_u32_e32 v4, vcc, v8, v4, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s11, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v8, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s10, v3
-; GFX9-NEXT:    v_cndmask_b32_e64 v9, 0, -1, vcc
-; GFX9-NEXT:    v_cmp_eq_u32_e32 vcc, s11, v4
-; GFX9-NEXT:    v_cndmask_b32_e32 v8, v8, v9, vcc
-; GFX9-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v8
-; GFX9-NEXT:    v_cndmask_b32_e32 v4, v4, v5, vcc
-; GFX9-NEXT:    v_cndmask_b32_e64 v5, v7, v6, s[0:1]
-; GFX9-NEXT:    v_cndmask_b32_e32 v3, v3, v5, vcc
-; GFX9-NEXT:    v_xor_b32_e32 v3, s12, v3
-; GFX9-NEXT:    v_xor_b32_e32 v4, s12, v4
-; GFX9-NEXT:    v_mov_b32_e32 v5, s12
-; GFX9-NEXT:    v_subrev_co_u32_e32 v3, vcc, s12, v3
-; GFX9-NEXT:    v_subb_co_u32_e32 v4, vcc, v4, v5, vcc
+; GFX9-NEXT:    v_cndmask_b32_e32 v5, v8, 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 v0, v[1:4], s[4:5]
+; 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

diff --git a/llvm/test/CodeGen/AMDGPU/cf-loop-on-constant.ll b/llvm/test/CodeGen/AMDGPU/cf-loop-on-constant.ll
index e5926f3fc937d0a48d08d51234c424fcf785505f..535173bb0f9ed5e8e4ba396711ca67654128e37b 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/cf-loop-on-constant.ll
+++ b/llvm/test/CodeGen/AMDGPU/cf-loop-on-constant.ll
@@ -1,14 +1,73 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
-; RUN: llc -march=amdgcn -verify-machineinstrs -O0 < %s
+; RUN: llc -march=amdgcn -verify-machineinstrs -O0 < %s | FileCheck -check-prefix=GCN_DBG %s
 
-; GCN-LABEL: {{^}}test_loop:
-; GCN: s_and_b64 s[0:1], exec, -1
-; GCN: [[LABEL:.LBB[0-9]+_[0-9]+]]: ; %for.body{{$}}
-; GCN: ds_read_b32
-; GCN: ds_write_b32
-; GCN: s_cbranch_vccnz [[LABEL]]
-; GCN: s_endpgm
 define amdgpu_kernel void @test_loop(float addrspace(3)* %ptr, i32 %n) nounwind {
+; GCN-LABEL: test_loop:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_load_dword s2, s[0:1], 0xa
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_cmp_eq_u32 s2, -1
+; GCN-NEXT:    s_cbranch_scc1 .LBB0_3
+; GCN-NEXT:  ; %bb.1: ; %for.body.preheader
+; GCN-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_addk_i32 s0, 0x80
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    s_and_b64 s[0:1], exec, -1
+; GCN-NEXT:    s_mov_b32 m0, -1
+; GCN-NEXT:  .LBB0_2: ; %for.body
+; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN-NEXT:    ds_read_b32 v1, v0
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GCN-NEXT:    ds_write_b32 v0, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, 4, v0
+; GCN-NEXT:    s_mov_b64 vcc, s[0:1]
+; GCN-NEXT:    s_cbranch_vccnz .LBB0_2
+; GCN-NEXT:  .LBB0_3: ; %for.exit
+; GCN-NEXT:    s_endpgm
+;
+; GCN_DBG-LABEL: test_loop:
+; GCN_DBG:       ; %bb.0: ; %entry
+; GCN_DBG-NEXT:    s_load_dword s2, s[0:1], 0x9
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s2, 0
+; GCN_DBG-NEXT:    s_load_dword s1, s[0:1], 0xa
+; GCN_DBG-NEXT:    s_mov_b32 s0, 0
+; GCN_DBG-NEXT:    s_mov_b32 s2, -1
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    s_cmp_lg_u32 s1, s2
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_cbranch_scc1 .LBB0_2
+; GCN_DBG-NEXT:  ; %bb.1: ; %for.exit
+; GCN_DBG-NEXT:    s_endpgm
+; GCN_DBG-NEXT:  .LBB0_2: ; %for.body
+; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
+; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
+; GCN_DBG-NEXT:    s_mov_b32 s1, 2
+; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_read_b32 v0, v0
+; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_write_b32 v0, v1
+; GCN_DBG-NEXT:    s_mov_b32 s1, 1
+; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
+; GCN_DBG-NEXT:    s_mov_b64 s[2:3], -1
+; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB0_2
+; GCN_DBG-NEXT:  ; %bb.3: ; %DummyReturnBlock
+; GCN_DBG-NEXT:    s_endpgm
 entry:
   %cmp = icmp eq i32 %n, -1
   br i1 %cmp, label %for.exit, label %for.body
@@ -27,12 +86,58 @@ for.body:
   br label %for.body
 }
 
-; GCN-LABEL: @loop_const_true
-; GCN: [[LABEL:.LBB[0-9]+_[0-9]+]]:
-; GCN: ds_read_b32
-; GCN: ds_write_b32
-; GCN: s_branch [[LABEL]]
 define amdgpu_kernel void @loop_const_true(float addrspace(3)* %ptr, i32 %n) nounwind {
+; GCN-LABEL: loop_const_true:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    s_addk_i32 s0, 0x80
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    s_mov_b32 m0, -1
+; GCN-NEXT:  .LBB1_1: ; %for.body
+; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN-NEXT:    ds_read_b32 v1, v0
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GCN-NEXT:    ds_write_b32 v0, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, 4, v0
+; GCN-NEXT:    s_branch .LBB1_1
+;
+; GCN_DBG-LABEL: loop_const_true:
+; GCN_DBG:       ; %bb.0: ; %entry
+; GCN_DBG-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
+; GCN_DBG-NEXT:    s_mov_b32 s0, 0
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_branch .LBB1_2
+; GCN_DBG-NEXT:  .LBB1_1: ; %for.exit
+; GCN_DBG-NEXT:    s_endpgm
+; GCN_DBG-NEXT:  .LBB1_2: ; %for.body
+; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
+; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
+; GCN_DBG-NEXT:    s_mov_b32 s1, 2
+; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_read_b32 v0, v0
+; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_write_b32 v0, v1
+; GCN_DBG-NEXT:    s_mov_b32 s1, 1
+; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
+; GCN_DBG-NEXT:    s_mov_b64 s[2:3], 0
+; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB1_1
+; GCN_DBG-NEXT:    s_branch .LBB1_2
 entry:
   br label %for.body
 
@@ -50,10 +155,54 @@ for.body:
   br i1 true, label %for.body, label %for.exit
 }
 
-; GCN-LABEL: {{^}}loop_const_false:
-; GCN-NOT: s_branch
-; GCN: s_endpgm
 define amdgpu_kernel void @loop_const_false(float addrspace(3)* %ptr, i32 %n) nounwind {
+; GCN-LABEL: loop_const_false:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    s_mov_b32 m0, -1
+; GCN-NEXT:    ds_read_b32 v1, v0 offset:128
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GCN-NEXT:    ds_write_b32 v0, v1 offset:128
+; GCN-NEXT:    s_endpgm
+;
+; GCN_DBG-LABEL: loop_const_false:
+; GCN_DBG:       ; %bb.0: ; %entry
+; GCN_DBG-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
+; GCN_DBG-NEXT:    s_mov_b32 s0, 0
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_branch .LBB2_2
+; GCN_DBG-NEXT:  .LBB2_1: ; %for.exit
+; GCN_DBG-NEXT:    s_endpgm
+; GCN_DBG-NEXT:  .LBB2_2: ; %for.body
+; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
+; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
+; GCN_DBG-NEXT:    s_mov_b32 s1, 2
+; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_read_b32 v0, v0
+; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_write_b32 v0, v1
+; GCN_DBG-NEXT:    s_mov_b32 s1, 1
+; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
+; GCN_DBG-NEXT:    s_mov_b64 s[2:3], -1
+; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB2_1
+; GCN_DBG-NEXT:    s_branch .LBB2_2
 entry:
   br label %for.body
 
@@ -72,10 +221,52 @@ for.body:
   br i1 false, label %for.body, label %for.exit
 }
 
-; GCN-LABEL: {{^}}loop_const_undef:
-; GCN-NOT: s_branch
-; GCN: s_endpgm
 define amdgpu_kernel void @loop_const_undef(float addrspace(3)* %ptr, i32 %n) nounwind {
+; GCN-LABEL: loop_const_undef:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_mov_b32_e32 v0, s0
+; GCN-NEXT:    s_mov_b32 m0, -1
+; GCN-NEXT:    ds_read_b32 v1, v0 offset:128
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GCN-NEXT:    ds_write_b32 v0, v1 offset:128
+; GCN-NEXT:    s_endpgm
+;
+; GCN_DBG-LABEL: loop_const_undef:
+; GCN_DBG:       ; %bb.0: ; %entry
+; GCN_DBG-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
+; GCN_DBG-NEXT:    s_mov_b32 s0, 0
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_branch .LBB3_2
+; GCN_DBG-NEXT:  .LBB3_1: ; %for.exit
+; GCN_DBG-NEXT:    s_endpgm
+; GCN_DBG-NEXT:  .LBB3_2: ; %for.body
+; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 1
+; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 0
+; GCN_DBG-NEXT:    s_mov_b32 s1, 2
+; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 s2, 0x80
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_read_b32 v0, v0
+; GCN_DBG-NEXT:    s_mov_b32 s2, 1.0
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s2
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_write_b32 v0, v1
+; GCN_DBG-NEXT:    s_mov_b32 s1, 1
+; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    s_cbranch_scc1 .LBB3_1
+; GCN_DBG-NEXT:    s_branch .LBB3_2
 entry:
   br label %for.body
 
@@ -94,18 +285,81 @@ for.body:
   br i1 undef, label %for.body, label %for.exit
 }
 
-; GCN-LABEL: {{^}}loop_arg_0:
-; GCN: v_and_b32_e32 v{{[0-9]+}}, 1, v{{[0-9]+}}
-; GCN: v_cmp_eq_u32{{[^,]*}}, 1,
-; GCN: s_add_i32 s2, s0, 0x80
-
-; GCN: [[LOOPBB:.LBB[0-9]+_[0-9]+]]
-; GCN: _add_i32_e32 v0, vcc, 4, v0
-
-; GCN: s_cbranch_{{vccz|vccnz}} [[LOOPBB]]
-; GCN-NEXT: ; %bb.2
-; GCN-NEXT: s_endpgm
 define amdgpu_kernel void @loop_arg_0(float addrspace(3)* %ptr, i32 %n) nounwind {
+; GCN-LABEL: loop_arg_0:
+; GCN:       ; %bb.0: ; %entry
+; GCN-NEXT:    v_mov_b32_e32 v0, 0
+; GCN-NEXT:    s_mov_b32 m0, -1
+; GCN-NEXT:    ds_read_u8 v0, v0
+; GCN-NEXT:    s_load_dword s2, s[0:1], 0x9
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_readfirstlane_b32 s0, v0
+; GCN-NEXT:    s_bitcmp1_b32 s0, 0
+; GCN-NEXT:    s_cselect_b64 s[0:1], -1, 0
+; GCN-NEXT:    s_addk_i32 s2, 0x80
+; GCN-NEXT:    s_xor_b64 s[0:1], s[0:1], -1
+; GCN-NEXT:    v_mov_b32_e32 v0, s2
+; GCN-NEXT:    s_and_b64 s[0:1], exec, s[0:1]
+; GCN-NEXT:  .LBB4_1: ; %for.body
+; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN-NEXT:    ds_read_b32 v1, v0
+; GCN-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN-NEXT:    v_add_f32_e32 v1, 1.0, v1
+; GCN-NEXT:    ds_write_b32 v0, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, 4, v0
+; GCN-NEXT:    s_mov_b64 vcc, s[0:1]
+; GCN-NEXT:    s_cbranch_vccz .LBB4_1
+; GCN-NEXT:  ; %bb.2: ; %for.exit
+; GCN-NEXT:    s_endpgm
+;
+; GCN_DBG-LABEL: loop_arg_0:
+; GCN_DBG:       ; %bb.0: ; %entry
+; GCN_DBG-NEXT:    s_load_dword s0, s[0:1], 0x9
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 0
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, 0
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    ds_read_u8 v0, v0
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_readfirstlane_b32 s0, v0
+; GCN_DBG-NEXT:    s_and_b32 s0, 1, s0
+; GCN_DBG-NEXT:    s_cmp_eq_u32 s0, 1
+; GCN_DBG-NEXT:    s_cselect_b64 s[0:1], -1, 0
+; GCN_DBG-NEXT:    s_mov_b64 s[2:3], -1
+; GCN_DBG-NEXT:    s_xor_b64 s[0:1], s[0:1], s[2:3]
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 1
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s1, 2
+; GCN_DBG-NEXT:    s_mov_b32 s0, 0
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 3
+; GCN_DBG-NEXT:    s_branch .LBB4_2
+; GCN_DBG-NEXT:  .LBB4_1: ; %for.exit
+; GCN_DBG-NEXT:    s_endpgm
+; GCN_DBG-NEXT:  .LBB4_2: ; %for.body
+; GCN_DBG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GCN_DBG-NEXT:    v_readlane_b32 s0, v2, 3
+; GCN_DBG-NEXT:    v_readlane_b32 s2, v2, 1
+; GCN_DBG-NEXT:    v_readlane_b32 s3, v2, 2
+; GCN_DBG-NEXT:    v_readlane_b32 s4, v2, 0
+; GCN_DBG-NEXT:    s_mov_b32 s1, 2
+; GCN_DBG-NEXT:    s_lshl_b32 s1, s0, s1
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s4
+; GCN_DBG-NEXT:    s_mov_b32 s4, 0x80
+; GCN_DBG-NEXT:    s_add_i32 s1, s1, s4
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_read_b32 v0, v0
+; GCN_DBG-NEXT:    s_mov_b32 s4, 1.0
+; GCN_DBG-NEXT:    s_waitcnt lgkmcnt(0)
+; GCN_DBG-NEXT:    v_add_f32_e64 v1, v0, s4
+; GCN_DBG-NEXT:    s_mov_b32 m0, -1
+; GCN_DBG-NEXT:    v_mov_b32_e32 v0, s1
+; GCN_DBG-NEXT:    ds_write_b32 v0, v1
+; GCN_DBG-NEXT:    s_mov_b32 s1, 1
+; GCN_DBG-NEXT:    s_add_i32 s0, s0, s1
+; GCN_DBG-NEXT:    s_and_b64 vcc, exec, s[2:3]
+; GCN_DBG-NEXT:    v_writelane_b32 v2, s0, 3
+; GCN_DBG-NEXT:    s_cbranch_vccnz .LBB4_1
+; GCN_DBG-NEXT:    s_branch .LBB4_2
 entry:
   %cond = load volatile i1, i1 addrspace(3)* null
   br label %for.body

diff --git a/llvm/test/CodeGen/AMDGPU/idiv-licm.ll b/llvm/test/CodeGen/AMDGPU/idiv-licm.ll
index 54bc6ddcca45fd994757eb59f61da51c548cf7b2..a93492e14bf41a348bf4541abb37fed49cb5f9aa 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/idiv-licm.ll
+++ b/llvm/test/CodeGen/AMDGPU/idiv-licm.ll
@@ -9,35 +9,41 @@ define amdgpu_kernel void @udiv32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX9-NEXT:    s_mov_b64 s[2:3], 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:    v_cvt_f32_u32_e32 v0, s4
 ; GFX9-NEXT:    s_sub_i32 s5, 0, s4
 ; 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:    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:  .LBB0_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_mul_lo_u32 v2, s3, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, s5, v2
-; GFX9-NEXT:    v_not_b32_e32 v5, v2
-; GFX9-NEXT:    v_mul_lo_u32 v5, s4, v5
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v2
-; GFX9-NEXT:    v_add_u32_e32 v3, s2, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v3
+; 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_mul_i32 s7, s5, s6
+; GFX9-NEXT:    s_add_i32 s7, s2, s7
+; 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_add_u32_e32 v4, s2, v5
+; 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_add_u32 s2, s2, 1
-; GFX9-NEXT:    v_add_u32_e32 v4, 1, v2
 ; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v3
 ; GFX9-NEXT:    s_addc_u32 s3, s3, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
+; 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
@@ -49,34 +55,38 @@ define amdgpu_kernel void @udiv32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX10-LABEL: udiv32_invariant_denom:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_load_dword s4, s[0:1], 0x2c
-; GFX10-NEXT:    s_mov_b64 s[2:3], 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX10-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_u32_e32 v0, s4
 ; GFX10-NEXT:    s_sub_i32 s5, 0, s4
 ; GFX10-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX10-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX10-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX10-NEXT:    v_mul_lo_u32 v1, s5, v0
-; GFX10-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, v0, v1
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:  .LBB0_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_mul_lo_u32 v2, s3, v0
-; GFX10-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX10-NEXT:    v_add_nc_u32_e32 v2, v3, v2
-; GFX10-NEXT:    v_not_b32_e32 v3, v2
-; GFX10-NEXT:    v_mul_lo_u32 v4, s5, v2
-; GFX10-NEXT:    v_add_nc_u32_e32 v5, 1, v2
-; GFX10-NEXT:    v_mul_lo_u32 v3, s4, v3
-; GFX10-NEXT:    v_add_nc_u32_e32 v4, s2, v4
-; GFX10-NEXT:    v_add_nc_u32_e32 v3, s2, v3
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s4, v4
+; 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_mul_i32 s7, s5, s6
+; GFX10-NEXT:    s_add_i32 s7, s2, s7
+; GFX10-NEXT:    s_cmp_ge_u32 s7, s4
+; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; 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_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:    s_addc_u32 s3, s3, 0
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc_lo
-; GFX10-NEXT:    v_cndmask_b32_e32 v3, v4, v3, 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, s4, v3
 ; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc_lo
@@ -165,33 +175,35 @@ define amdgpu_kernel void @urem32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX9-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX9-NEXT:    s_mov_b64 s[2:3], 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:    v_cvt_f32_u32_e32 v0, s4
 ; GFX9-NEXT:    s_sub_i32 s5, 0, s4
 ; 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:    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:  .LBB1_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_mul_lo_u32 v2, s3, v0
-; GFX9-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX9-NEXT:    v_add_u32_e32 v2, v3, v2
-; GFX9-NEXT:    v_mul_lo_u32 v3, s5, v2
-; GFX9-NEXT:    v_not_b32_e32 v2, v2
-; GFX9-NEXT:    v_mul_lo_u32 v2, s4, v2
-; GFX9-NEXT:    v_add_u32_e32 v3, s2, v3
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v3
-; GFX9-NEXT:    v_add_u32_e32 v2, s2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v3, v2, vcc
+; 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_mul_i32 s7, s5, s6
+; GFX9-NEXT:    s_not_b32 s6, s6
+; GFX9-NEXT:    s_mul_i32 s6, s4, s6
+; GFX9-NEXT:    s_add_i32 s7, s2, s7
+; GFX9-NEXT:    s_add_i32 s6, s2, s6
+; GFX9-NEXT:    s_cmp_ge_u32 s7, s4
+; GFX9-NEXT:    s_cselect_b32 s6, s6, s7
+; GFX9-NEXT:    s_sub_i32 s7, s6, s4
+; GFX9-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX9-NEXT:    s_cselect_b32 s6, s7, s6
 ; GFX9-NEXT:    s_add_u32 s2, s2, 1
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s4, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s4, v2
+; GFX9-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX9-NEXT:    s_addc_u32 s3, s3, 0
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, 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
@@ -203,35 +215,37 @@ define amdgpu_kernel void @urem32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX10-LABEL: urem32_invariant_denom:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_load_dword s4, s[0:1], 0x2c
-; GFX10-NEXT:    s_mov_b64 s[2:3], 0
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX10-NEXT:    s_mov_b64 s[2:3], 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    v_cvt_f32_u32_e32 v0, s4
 ; GFX10-NEXT:    s_sub_i32 s5, 0, s4
 ; GFX10-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX10-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX10-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX10-NEXT:    v_mul_lo_u32 v1, s5, v0
-; GFX10-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, v0, v1
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:  .LBB1_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_mul_lo_u32 v2, s3, v0
-; GFX10-NEXT:    v_mul_hi_u32 v3, s2, v0
-; GFX10-NEXT:    v_add_nc_u32_e32 v2, v3, v2
-; GFX10-NEXT:    v_not_b32_e32 v3, v2
-; GFX10-NEXT:    v_mul_lo_u32 v2, s5, v2
-; GFX10-NEXT:    v_mul_lo_u32 v3, s4, v3
-; GFX10-NEXT:    v_add_nc_u32_e32 v2, s2, v2
-; GFX10-NEXT:    v_add_nc_u32_e32 v3, s2, v3
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s4, v2
+; 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 s7, s6
+; GFX10-NEXT:    s_mul_i32 s6, s5, s6
+; GFX10-NEXT:    s_mul_i32 s7, s4, s7
+; GFX10-NEXT:    s_add_i32 s6, s2, s6
+; GFX10-NEXT:    s_add_i32 s7, s2, s7
+; GFX10-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX10-NEXT:    s_cselect_b32 s6, s7, s6
+; GFX10-NEXT:    s_sub_i32 s7, s6, s4
+; GFX10-NEXT:    s_cmp_ge_u32 s6, s4
+; GFX10-NEXT:    s_cselect_b32 s6, s7, s6
 ; GFX10-NEXT:    s_add_u32 s2, s2, 1
+; GFX10-NEXT:    v_mov_b32_e32 v2, s6
 ; GFX10-NEXT:    s_addc_u32 s3, s3, 0
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc_lo
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v3, s4, v2
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s4, v2
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v3, 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
@@ -355,38 +369,41 @@ define amdgpu_kernel void @sdiv32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX10-LABEL: sdiv32_invariant_denom:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_load_dword s3, s[0:1], 0x2c
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX10-NEXT:    s_mov_b32 s4, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_ashr_i32 s2, s3, 31
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_add_i32 s3, s3, s2
 ; GFX10-NEXT:    s_xor_b32 s3, s3, s2
 ; GFX10-NEXT:    v_cvt_f32_u32_e32 v0, s3
-; GFX10-NEXT:    s_sub_i32 s4, 0, s3
+; GFX10-NEXT:    s_sub_i32 s5, 0, s3
 ; GFX10-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX10-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX10-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX10-NEXT:    v_mul_lo_u32 v1, s4, v0
-; GFX10-NEXT:    s_mov_b32 s4, 0
-; GFX10-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, v0, v1
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:  .LBB2_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_mul_hi_u32 v2, s4, v0
-; GFX10-NEXT:    v_mul_lo_u32 v3, v2, s3
-; GFX10-NEXT:    v_add_nc_u32_e32 v4, 1, v2
-; GFX10-NEXT:    v_sub_nc_u32_e32 v3, s4, v3
+; 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_sub_i32 s7, s4, s7
+; GFX10-NEXT:    s_cmp_ge_u32 s7, s3
+; GFX10-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX10-NEXT:    s_add_i32 s8, s6, 1
 ; GFX10-NEXT:    s_add_i32 s4, s4, 1
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v5, s3, v3
-; 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_cndmask_b32_e32 v3, v3, v5, vcc_lo
+; 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_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    global_store_dword v1, v2, s[0:1]
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_add_u32 s0, s0, 4
@@ -470,34 +487,35 @@ define amdgpu_kernel void @srem32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX9-LABEL: srem32_invariant_denom:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; 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_sub_i32 s3, 0, s2
-; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    s_mov_b32 s3, 0
+; GFX9-NEXT:    s_sub_i32 s4, 0, s2
 ; 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, s3, v0
-; GFX9-NEXT:    s_mov_b32 s3, 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:  .LBB3_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_mul_hi_u32 v2, s3, v0
-; GFX9-NEXT:    v_mul_lo_u32 v2, v2, s2
-; GFX9-NEXT:    v_sub_u32_e32 v2, s3, v2
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s2, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
-; GFX9-NEXT:    v_subrev_u32_e32 v3, s2, v2
-; GFX9-NEXT:    v_cmp_le_u32_e32 vcc, s2, v2
+; GFX9-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX9-NEXT:    s_mul_i32 s6, s4, s5
+; GFX9-NEXT:    s_mul_hi_u32 s6, s5, s6
+; GFX9-NEXT:    s_add_i32 s5, s5, s6
+; GFX9-NEXT:    s_mul_hi_u32 s5, s3, s5
+; GFX9-NEXT:    s_mul_i32 s5, s5, s2
+; GFX9-NEXT:    s_sub_i32 s5, s3, s5
+; GFX9-NEXT:    s_sub_i32 s6, s5, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX9-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX9-NEXT:    s_sub_i32 s6, s5, s2
+; GFX9-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX9-NEXT:    s_cselect_b32 s5, s6, s5
 ; GFX9-NEXT:    s_add_i32 s3, s3, 1
-; GFX9-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v2, s5
 ; 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
@@ -509,34 +527,35 @@ define amdgpu_kernel void @srem32_invariant_denom(i32 addrspace(1)* nocapture %a
 ; GFX10-LABEL: srem32_invariant_denom:
 ; GFX10:       ; %bb.0: ; %bb
 ; GFX10-NEXT:    s_load_dword s2, s[0:1], 0x2c
+; GFX10-NEXT:    v_mov_b32_e32 v1, 0
+; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_ashr_i32 s3, s2, 31
-; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX10-NEXT:    s_add_i32 s2, s2, s3
 ; GFX10-NEXT:    s_xor_b32 s2, s2, s3
+; GFX10-NEXT:    s_mov_b32 s3, 0
 ; GFX10-NEXT:    v_cvt_f32_u32_e32 v0, s2
-; GFX10-NEXT:    s_sub_i32 s3, 0, s2
+; GFX10-NEXT:    s_sub_i32 s4, 0, s2
 ; GFX10-NEXT:    v_rcp_iflag_f32_e32 v0, v0
 ; GFX10-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX10-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX10-NEXT:    v_mul_lo_u32 v1, s3, v0
-; GFX10-NEXT:    s_mov_b32 s3, 0
-; GFX10-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, v0, v1
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX10-NEXT:  .LBB3_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_mul_hi_u32 v2, s3, v0
-; GFX10-NEXT:    v_mul_lo_u32 v2, v2, s2
-; GFX10-NEXT:    v_sub_nc_u32_e32 v2, s3, v2
+; GFX10-NEXT:    v_readfirstlane_b32 s5, v0
+; GFX10-NEXT:    s_mul_i32 s6, s4, s5
+; GFX10-NEXT:    s_mul_hi_u32 s6, s5, s6
+; GFX10-NEXT:    s_add_i32 s5, s5, s6
+; GFX10-NEXT:    s_mul_hi_u32 s5, s3, s5
+; GFX10-NEXT:    s_mul_i32 s5, s5, s2
+; GFX10-NEXT:    s_sub_i32 s5, s3, s5
+; GFX10-NEXT:    s_sub_i32 s6, s5, s2
+; GFX10-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX10-NEXT:    s_cselect_b32 s5, s6, s5
+; GFX10-NEXT:    s_sub_i32 s6, s5, s2
+; GFX10-NEXT:    s_cmp_ge_u32 s5, s2
+; GFX10-NEXT:    s_cselect_b32 s5, s6, s5
 ; GFX10-NEXT:    s_add_i32 s3, s3, 1
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v3, s2, v2
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s2, v2
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc_lo
-; GFX10-NEXT:    v_subrev_nc_u32_e32 v3, s2, v2
-; GFX10-NEXT:    v_cmp_le_u32_e32 vcc_lo, s2, v2
-; GFX10-NEXT:    v_cndmask_b32_e32 v2, v2, v3, vcc_lo
-; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX10-NEXT:    v_mov_b32_e32 v2, s5
 ; GFX10-NEXT:    global_store_dword v1, v2, s[0:1]
 ; GFX10-NEXT:    s_waitcnt_depctr 0xffe3
 ; GFX10-NEXT:    s_add_u32 s0, s0, 4
@@ -615,32 +634,33 @@ define amdgpu_kernel void @udiv16_invariant_denom(i16 addrspace(1)* nocapture %a
 ; GFX9-LABEL: udiv16_invariant_denom:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x2c
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_movk_i32 s4, 0x400
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    s_mov_b32 s5, 0
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    s_movk_i32 s6, 0x400
+; GFX9-NEXT:    s_mov_b32 s7, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_and_b32 s2, 0xffff, s2
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v2, s2
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v0, s2
 ; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v2
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v0
 ; GFX9-NEXT:  .LBB4_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX9-NEXT:    v_cvt_f32_u32_e32 v8, v0
-; GFX9-NEXT:    v_lshlrev_b64 v[5:6], 1, v[0:1]
+; GFX9-NEXT:    s_and_b32 s4, 0xffff, s7
+; GFX9-NEXT:    v_cvt_f32_u32_e32 v4, s4
+; GFX9-NEXT:    v_add_u16_e64 v3, s7, 1
+; GFX9-NEXT:    v_readfirstlane_b32 s7, v3
+; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, s6, v3
+; GFX9-NEXT:    v_mul_f32_e32 v3, v4, v1
+; GFX9-NEXT:    v_trunc_f32_e32 v3, v3
+; GFX9-NEXT:    v_cvt_u32_f32_e32 v5, v3
+; GFX9-NEXT:    s_lshl_b64 s[0:1], s[4:5], 1
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v7, s3
-; GFX9-NEXT:    v_add_co_u32_e64 v5, s[0:1], s2, v5
-; GFX9-NEXT:    v_mul_f32_e32 v0, v8, v3
-; GFX9-NEXT:    v_trunc_f32_e32 v0, v0
-; GFX9-NEXT:    v_addc_co_u32_e64 v6, s[0:1], v7, v6, s[0:1]
-; GFX9-NEXT:    v_cvt_u32_f32_e32 v7, v0
-; GFX9-NEXT:    v_add_u16_e32 v4, 1, v4
-; GFX9-NEXT:    v_mad_f32 v0, -v0, v2, v8
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, s4, v4
-; GFX9-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v0|, v2
-; GFX9-NEXT:    v_addc_co_u32_e64 v0, s[0:1], 0, v7, s[0:1]
-; GFX9-NEXT:    global_store_short v[5:6], v0, off
+; GFX9-NEXT:    s_add_u32 s8, s2, s0
+; GFX9-NEXT:    v_mad_f32 v3, -v3, v0, v4
+; GFX9-NEXT:    s_addc_u32 s9, s3, s1
+; GFX9-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v3|, v0
+; GFX9-NEXT:    v_addc_co_u32_e64 v3, s[0:1], 0, v5, s[0:1]
+; GFX9-NEXT:    global_store_short v2, v3, s[8:9]
 ; GFX9-NEXT:    s_cbranch_vccz .LBB4_1
 ; GFX9-NEXT:  ; %bb.2: ; %bb2
 ; GFX9-NEXT:    s_endpgm
@@ -650,29 +670,31 @@ define amdgpu_kernel void @udiv16_invariant_denom(i16 addrspace(1)* nocapture %a
 ; GFX10-NEXT:    s_clause 0x1
 ; GFX10-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v4, 0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-NEXT:    s_mov_b32 s1, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_and_b32 s0, 0xffff, s4
-; GFX10-NEXT:    v_cvt_f32_u32_e32 v2, s0
-; GFX10-NEXT:    v_rcp_iflag_f32_e32 v3, v2
+; GFX10-NEXT:    s_mov_b32 s4, 0
+; GFX10-NEXT:    v_cvt_f32_u32_e32 v0, s0
+; GFX10-NEXT:    v_rcp_iflag_f32_e32 v1, v0
 ; GFX10-NEXT:  .LBB4_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_add_nc_u16 v4, v4, 1
-; GFX10-NEXT:    v_cvt_f32_u32_e32 v7, v0
-; GFX10-NEXT:    v_lshlrev_b64 v[5:6], 1, v[0:1]
-; GFX10-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0x400, v4
-; GFX10-NEXT:    v_mul_f32_e32 v0, v7, v3
-; GFX10-NEXT:    v_add_co_u32 v5, s0, s2, v5
-; GFX10-NEXT:    v_add_co_ci_u32_e64 v6, s0, s3, v6, s0
-; GFX10-NEXT:    v_trunc_f32_e32 v0, v0
+; GFX10-NEXT:    s_and_b32 s0, 0xffff, s4
+; GFX10-NEXT:    v_add_nc_u16 v3, s4, 1
+; GFX10-NEXT:    v_cvt_f32_u32_e32 v4, s0
+; GFX10-NEXT:    s_lshl_b64 s[4:5], s[0:1], 1
+; GFX10-NEXT:    s_add_u32 s6, s2, s4
+; GFX10-NEXT:    v_readfirstlane_b32 s4, v3
+; GFX10-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0x400, v3
+; GFX10-NEXT:    v_mul_f32_e32 v3, v4, v1
+; GFX10-NEXT:    s_addc_u32 s7, s3, s5
 ; GFX10-NEXT:    s_and_b32 vcc_lo, exec_lo, vcc_lo
-; GFX10-NEXT:    v_mad_f32 v7, -v0, v2, v7
-; GFX10-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX10-NEXT:    v_cmp_ge_f32_e64 s0, |v7|, v2
-; GFX10-NEXT:    v_add_co_ci_u32_e64 v0, s0, 0, v0, s0
-; GFX10-NEXT:    global_store_short v[5:6], v0, off
+; GFX10-NEXT:    v_trunc_f32_e32 v3, v3
+; GFX10-NEXT:    v_mad_f32 v4, -v3, v0, v4
+; GFX10-NEXT:    v_cvt_u32_f32_e32 v3, v3
+; GFX10-NEXT:    v_cmp_ge_f32_e64 s0, |v4|, v0
+; GFX10-NEXT:    v_add_co_ci_u32_e64 v3, s0, 0, v3, s0
+; GFX10-NEXT:    global_store_short v2, v3, s[6:7]
 ; GFX10-NEXT:    s_cbranch_vccz .LBB4_1
 ; GFX10-NEXT:  ; %bb.2: ; %bb2
 ; GFX10-NEXT:    s_endpgm
@@ -862,37 +884,38 @@ define amdgpu_kernel void @sdiv16_invariant_denom(i16 addrspace(1)* nocapture %a
 ; GFX9-LABEL: sdiv16_invariant_denom:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x2c
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_mov_b32 s3, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
 ; GFX9-NEXT:    s_movk_i32 s5, 0x400
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX9-NEXT:    s_sext_i32_i16 s4, s2
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX9-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v2
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v0, s4
+; GFX9-NEXT:    s_mov_b32 s6, 0
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v0
 ; GFX9-NEXT:  .LBB6_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_bfe_i32 v5, v4, 0, 16
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v9, v5
-; GFX9-NEXT:    v_xor_b32_e32 v8, s4, v5
-; GFX9-NEXT:    v_lshlrev_b64 v[5:6], 1, v[0:1]
-; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    v_mov_b32_e32 v7, s3
-; GFX9-NEXT:    v_add_co_u32_e64 v5, s[0:1], s2, v5
-; GFX9-NEXT:    v_addc_co_u32_e64 v6, s[0:1], v7, v6, s[0:1]
-; GFX9-NEXT:    v_mul_f32_e32 v7, v9, v3
-; GFX9-NEXT:    v_trunc_f32_e32 v7, v7
-; GFX9-NEXT:    v_ashrrev_i32_e32 v0, 30, v8
-; GFX9-NEXT:    v_cvt_i32_f32_e32 v8, v7
-; GFX9-NEXT:    v_mad_f32 v7, -v7, v2, v9
-; GFX9-NEXT:    v_add_u16_e32 v4, 1, v4
-; GFX9-NEXT:    v_or_b32_e32 v0, 1, v0
-; GFX9-NEXT:    v_cmp_ge_f32_e64 s[0:1], |v7|, |v2|
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, s5, v4
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, v0, s[0:1]
-; GFX9-NEXT:    v_add_u32_e32 v0, v8, v0
-; GFX9-NEXT:    global_store_short v[5:6], v0, off
+; GFX9-NEXT:    s_sext_i32_i16 s2, s6
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v4, s2
+; GFX9-NEXT:    s_xor_b32 s7, s2, s4
+; GFX9-NEXT:    s_ashr_i32 s2, s7, 30
+; GFX9-NEXT:    s_or_b32 s2, s2, 1
+; GFX9-NEXT:    v_mul_f32_e32 v5, v4, v1
+; GFX9-NEXT:    v_trunc_f32_e32 v5, v5
+; GFX9-NEXT:    v_mad_f32 v4, -v5, v0, v4
+; GFX9-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v4|, |v0|
+; GFX9-NEXT:    v_cvt_i32_f32_e32 v5, v5
+; GFX9-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX9-NEXT:    s_cselect_b32 s7, s2, 0
+; GFX9-NEXT:    s_and_b32 s2, s6, 0xffff
+; GFX9-NEXT:    v_add_u16_e64 v3, s6, 1
+; GFX9-NEXT:    s_lshl_b64 s[8:9], s[2:3], 1
+; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, s5, v3
+; GFX9-NEXT:    s_add_u32 s8, s0, s8
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v3
+; GFX9-NEXT:    v_add_u32_e32 v3, s7, v5
+; GFX9-NEXT:    s_addc_u32 s9, s1, s9
+; GFX9-NEXT:    global_store_short v2, v3, s[8:9]
 ; GFX9-NEXT:    s_cbranch_vccz .LBB6_1
 ; GFX9-NEXT:  ; %bb.2: ; %bb2
 ; GFX9-NEXT:    s_endpgm
@@ -902,34 +925,36 @@ define amdgpu_kernel void @sdiv16_invariant_denom(i16 addrspace(1)* nocapture %a
 ; GFX10-NEXT:    s_clause 0x1
 ; GFX10-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v4, 0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-NEXT:    s_mov_b32 s1, 0
+; GFX10-NEXT:    s_mov_b32 s5, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX10-NEXT:    s_sext_i32_i16 s4, s4
-; GFX10-NEXT:    v_cvt_f32_i32_e32 v2, s4
-; GFX10-NEXT:    v_rcp_iflag_f32_e32 v3, v2
+; GFX10-NEXT:    v_cvt_f32_i32_e32 v0, s4
+; GFX10-NEXT:    v_rcp_iflag_f32_e32 v1, v0
 ; GFX10-NEXT:  .LBB6_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_bfe_i32 v5, v4, 0, 16
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_add_nc_u16 v4, v4, 1
-; GFX10-NEXT:    v_cvt_f32_i32_e32 v7, v5
-; GFX10-NEXT:    v_xor_b32_e32 v8, s4, v5
-; GFX10-NEXT:    v_lshlrev_b64 v[5:6], 1, v[0:1]
-; GFX10-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0x400, v4
-; GFX10-NEXT:    v_mul_f32_e32 v0, v7, v3
-; GFX10-NEXT:    v_ashrrev_i32_e32 v8, 30, v8
-; GFX10-NEXT:    v_add_co_u32 v5, s0, s2, v5
-; GFX10-NEXT:    v_trunc_f32_e32 v0, v0
-; GFX10-NEXT:    v_or_b32_e32 v8, 1, v8
-; GFX10-NEXT:    v_add_co_ci_u32_e64 v6, s0, s3, v6, s0
-; GFX10-NEXT:    s_and_b32 vcc_lo, exec_lo, vcc_lo
-; GFX10-NEXT:    v_mad_f32 v7, -v0, v2, v7
-; GFX10-NEXT:    v_cvt_i32_f32_e32 v0, v0
-; GFX10-NEXT:    v_cmp_ge_f32_e64 s1, |v7|, |v2|
-; GFX10-NEXT:    v_cndmask_b32_e64 v7, 0, v8, s1
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, v0, v7
-; GFX10-NEXT:    global_store_short v[5:6], v0, off
+; GFX10-NEXT:    s_sext_i32_i16 s0, s5
+; GFX10-NEXT:    v_add_nc_u16 v3, s5, 1
+; GFX10-NEXT:    v_cvt_f32_i32_e32 v4, s0
+; GFX10-NEXT:    s_xor_b32 s0, s0, s4
+; GFX10-NEXT:    s_ashr_i32 s0, s0, 30
+; GFX10-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0x400, v3
+; GFX10-NEXT:    v_mul_f32_e32 v5, v4, v1
+; GFX10-NEXT:    s_or_b32 s0, s0, 1
+; GFX10-NEXT:    v_trunc_f32_e32 v5, v5
+; GFX10-NEXT:    v_mad_f32 v4, -v5, v0, v4
+; GFX10-NEXT:    v_cmp_ge_f32_e64 s6, |v4|, |v0|
+; GFX10-NEXT:    v_cvt_i32_f32_e32 v4, v5
+; GFX10-NEXT:    s_and_b32 s6, s6, exec_lo
+; GFX10-NEXT:    s_cselect_b32 s6, s0, 0
+; GFX10-NEXT:    s_and_b32 s0, s5, 0xffff
+; GFX10-NEXT:    v_readfirstlane_b32 s5, v3
+; GFX10-NEXT:    v_add_nc_u32_e32 v3, s6, v4
+; GFX10-NEXT:    s_lshl_b64 s[6:7], s[0:1], 1
+; GFX10-NEXT:    s_add_u32 s6, s2, s6
+; GFX10-NEXT:    s_addc_u32 s7, s3, s7
+; GFX10-NEXT:    global_store_short v2, v3, s[6:7]
 ; GFX10-NEXT:    s_cbranch_vccz .LBB6_1
 ; GFX10-NEXT:  ; %bb.2: ; %bb2
 ; GFX10-NEXT:    s_endpgm
@@ -1002,38 +1027,40 @@ define amdgpu_kernel void @srem16_invariant_denom(i16 addrspace(1)* nocapture %a
 ; GFX9-LABEL: srem16_invariant_denom:
 ; GFX9:       ; %bb.0: ; %bb
 ; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x2c
-; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x24
-; GFX9-NEXT:    v_mov_b32_e32 v1, 0
-; GFX9-NEXT:    s_movk_i32 s7, 0x400
-; GFX9-NEXT:    v_mov_b32_e32 v4, 0
+; GFX9-NEXT:    s_mov_b32 s3, 0
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX9-NEXT:    v_mov_b32_e32 v2, 0
+; GFX9-NEXT:    s_movk_i32 s5, 0x400
 ; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX9-NEXT:    s_sext_i32_i16 s6, s2
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v2, s6
-; GFX9-NEXT:    v_rcp_iflag_f32_e32 v3, v2
+; GFX9-NEXT:    s_sext_i32_i16 s4, s2
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v0, s4
+; GFX9-NEXT:    s_mov_b32 s6, 0
+; GFX9-NEXT:    v_rcp_iflag_f32_e32 v1, v0
 ; GFX9-NEXT:  .LBB7_1: ; %bb3
 ; GFX9-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX9-NEXT:    v_bfe_i32 v7, v4, 0, 16
-; GFX9-NEXT:    v_cvt_f32_i32_e32 v10, v7
-; GFX9-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX9-NEXT:    v_xor_b32_e32 v9, s6, v7
-; GFX9-NEXT:    v_lshlrev_b64 v[5:6], 1, v[0:1]
-; GFX9-NEXT:    v_ashrrev_i32_e32 v0, 30, v9
-; GFX9-NEXT:    v_mul_f32_e32 v9, v10, v3
-; GFX9-NEXT:    v_trunc_f32_e32 v9, v9
-; GFX9-NEXT:    v_cvt_i32_f32_e32 v11, v9
-; GFX9-NEXT:    v_mad_f32 v9, -v9, v2, v10
-; GFX9-NEXT:    v_or_b32_e32 v0, 1, v0
-; GFX9-NEXT:    v_cmp_ge_f32_e64 s[2:3], |v9|, |v2|
-; GFX9-NEXT:    v_cndmask_b32_e64 v0, 0, v0, s[2:3]
-; GFX9-NEXT:    v_add_u32_e32 v0, v11, v0
-; GFX9-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX9-NEXT:    v_add_u16_e32 v4, 1, v4
-; GFX9-NEXT:    v_mov_b32_e32 v8, s5
-; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, s7, v4
-; GFX9-NEXT:    v_add_co_u32_e64 v5, s[0:1], s4, v5
-; GFX9-NEXT:    v_addc_co_u32_e64 v6, s[0:1], v8, v6, s[0:1]
-; GFX9-NEXT:    v_sub_u32_e32 v0, v7, v0
-; GFX9-NEXT:    global_store_short v[5:6], v0, off
+; GFX9-NEXT:    s_sext_i32_i16 s7, s6
+; GFX9-NEXT:    v_cvt_f32_i32_e32 v4, s7
+; GFX9-NEXT:    s_xor_b32 s2, s7, s4
+; GFX9-NEXT:    s_ashr_i32 s2, s2, 30
+; GFX9-NEXT:    s_or_b32 s2, s2, 1
+; GFX9-NEXT:    v_mul_f32_e32 v5, v4, v1
+; GFX9-NEXT:    v_trunc_f32_e32 v5, v5
+; GFX9-NEXT:    v_mad_f32 v4, -v5, v0, v4
+; GFX9-NEXT:    v_cvt_i32_f32_e32 v5, v5
+; GFX9-NEXT:    v_cmp_ge_f32_e64 s[8:9], |v4|, |v0|
+; GFX9-NEXT:    s_and_b64 s[8:9], s[8:9], exec
+; GFX9-NEXT:    v_add_u16_e64 v3, s6, 1
+; GFX9-NEXT:    s_cselect_b32 s8, s2, 0
+; GFX9-NEXT:    v_cmp_eq_u16_e32 vcc, s5, v3
+; GFX9-NEXT:    s_and_b32 s2, s6, 0xffff
+; GFX9-NEXT:    v_readfirstlane_b32 s6, v3
+; GFX9-NEXT:    v_add_u32_e32 v3, s8, v5
+; GFX9-NEXT:    v_mul_lo_u32 v3, v3, s4
+; GFX9-NEXT:    s_lshl_b64 s[8:9], s[2:3], 1
+; GFX9-NEXT:    s_add_u32 s8, s0, s8
+; GFX9-NEXT:    s_addc_u32 s9, s1, s9
+; GFX9-NEXT:    v_sub_u32_e32 v3, s7, v3
+; GFX9-NEXT:    global_store_short v2, v3, s[8:9]
 ; GFX9-NEXT:    s_cbranch_vccz .LBB7_1
 ; GFX9-NEXT:  ; %bb.2: ; %bb2
 ; GFX9-NEXT:    s_endpgm
@@ -1043,35 +1070,38 @@ define amdgpu_kernel void @srem16_invariant_denom(i16 addrspace(1)* nocapture %a
 ; GFX10-NEXT:    s_clause 0x1
 ; GFX10-NEXT:    s_load_dword s4, s[0:1], 0x2c
 ; GFX10-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
-; GFX10-NEXT:    v_mov_b32_e32 v1, 0
-; GFX10-NEXT:    v_mov_b32_e32 v4, 0
+; GFX10-NEXT:    v_mov_b32_e32 v2, 0
+; GFX10-NEXT:    s_mov_b32 s1, 0
+; GFX10-NEXT:    s_mov_b32 s5, 0
 ; GFX10-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX10-NEXT:    s_sext_i32_i16 s1, s4
-; GFX10-NEXT:    v_cvt_f32_i32_e32 v2, s1
-; GFX10-NEXT:    v_rcp_iflag_f32_e32 v3, v2
+; GFX10-NEXT:    s_sext_i32_i16 s4, s4
+; GFX10-NEXT:    v_cvt_f32_i32_e32 v0, s4
+; GFX10-NEXT:    v_rcp_iflag_f32_e32 v1, v0
 ; GFX10-NEXT:  .LBB7_1: ; %bb3
 ; GFX10-NEXT:    ; =>This Inner Loop Header: Depth=1
-; GFX10-NEXT:    v_bfe_i32 v7, v4, 0, 16
-; GFX10-NEXT:    v_and_b32_e32 v0, 0xffff, v4
-; GFX10-NEXT:    v_add_nc_u16 v4, v4, 1
-; GFX10-NEXT:    v_cvt_f32_i32_e32 v5, v7
-; GFX10-NEXT:    v_xor_b32_e32 v6, s1, v7
-; GFX10-NEXT:    v_mul_f32_e32 v8, v5, v3
-; GFX10-NEXT:    v_ashrrev_i32_e32 v6, 30, v6
-; GFX10-NEXT:    v_trunc_f32_e32 v8, v8
-; GFX10-NEXT:    v_or_b32_e32 v6, 1, v6
-; GFX10-NEXT:    v_mad_f32 v5, -v8, v2, v5
-; GFX10-NEXT:    v_cvt_i32_f32_e32 v8, v8
-; GFX10-NEXT:    v_cmp_ge_f32_e64 vcc_lo, |v5|, |v2|
-; GFX10-NEXT:    v_cndmask_b32_e32 v9, 0, v6, vcc_lo
-; GFX10-NEXT:    v_lshlrev_b64 v[5:6], 1, v[0:1]
-; GFX10-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0x400, v4
-; GFX10-NEXT:    v_add_nc_u32_e32 v0, v8, v9
-; GFX10-NEXT:    v_add_co_u32 v5, s0, s2, v5
-; GFX10-NEXT:    v_add_co_ci_u32_e64 v6, s0, s3, v6, s0
-; GFX10-NEXT:    v_mul_lo_u32 v0, v0, s1
-; GFX10-NEXT:    v_sub_nc_u32_e32 v0, v7, v0
-; GFX10-NEXT:    global_store_short v[5:6], v0, off
+; GFX10-NEXT:    s_sext_i32_i16 s8, s5
+; GFX10-NEXT:    v_add_nc_u16 v3, s5, 1
+; GFX10-NEXT:    v_cvt_f32_i32_e32 v4, s8
+; GFX10-NEXT:    s_xor_b32 s0, s8, s4
+; GFX10-NEXT:    s_ashr_i32 s0, s0, 30
+; GFX10-NEXT:    v_cmp_eq_u16_e32 vcc_lo, 0x400, v3
+; GFX10-NEXT:    v_mul_f32_e32 v5, v4, v1
+; GFX10-NEXT:    s_or_b32 s0, s0, 1
+; GFX10-NEXT:    v_trunc_f32_e32 v5, v5
+; GFX10-NEXT:    v_mad_f32 v4, -v5, v0, v4
+; GFX10-NEXT:    v_cmp_ge_f32_e64 s6, |v4|, |v0|
+; GFX10-NEXT:    v_cvt_i32_f32_e32 v4, v5
+; GFX10-NEXT:    s_and_b32 s6, s6, exec_lo
+; GFX10-NEXT:    s_cselect_b32 s6, s0, 0
+; GFX10-NEXT:    s_and_b32 s0, s5, 0xffff
+; GFX10-NEXT:    v_add_nc_u32_e32 v4, s6, v4
+; GFX10-NEXT:    v_readfirstlane_b32 s5, v3
+; GFX10-NEXT:    s_lshl_b64 s[6:7], s[0:1], 1
+; GFX10-NEXT:    s_add_u32 s6, s2, s6
+; GFX10-NEXT:    v_mul_lo_u32 v3, v4, s4
+; GFX10-NEXT:    s_addc_u32 s7, s3, s7
+; GFX10-NEXT:    v_sub_nc_u32_e32 v3, s8, v3
+; GFX10-NEXT:    global_store_short v2, v3, s[6:7]
 ; GFX10-NEXT:    s_cbranch_vccz .LBB7_1
 ; GFX10-NEXT:  ; %bb.2: ; %bb2
 ; GFX10-NEXT:    s_endpgm

diff --git a/llvm/test/CodeGen/AMDGPU/multilevel-break.ll b/llvm/test/CodeGen/AMDGPU/multilevel-break.ll
index bf757bd0b26da1a98936fc96d55281a553be0e47..6f8ebbcc73d1e90fbc8d90cfa54f807670c7dfc5 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/multilevel-break.ll
+++ b/llvm/test/CodeGen/AMDGPU/multilevel-break.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S -mtriple=amdgcn-- -structurizecfg -si-annotate-control-flow < %s | FileCheck -check-prefix=OPT %s
 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
@@ -177,20 +178,21 @@ define amdgpu_kernel void @multi_if_break_loop(i32 %arg) #0 {
 ; GCN-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GCN-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
+; GCN-NEXT:    v_readfirstlane_b32 s8, v1
 ; GCN-NEXT:    s_mov_b64 s[4:5], -1
-; GCN-NEXT:    v_cmp_gt_i32_e32 vcc, 1, v1
+; GCN-NEXT:    s_cmp_lt_i32 s8, 1
 ; GCN-NEXT:    s_mov_b64 s[6:7], -1
-; GCN-NEXT:    s_cbranch_vccnz .LBB1_6
+; GCN-NEXT:    s_cbranch_scc1 .LBB1_6
 ; GCN-NEXT:  ; %bb.3: ; %LeafBlock1
 ; GCN-NEXT:    ; in Loop: Header=BB1_2 Depth=1
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 1, v1
+; GCN-NEXT:    s_cmp_eq_u32 s8, 1
 ; GCN-NEXT:    s_mov_b64 s[4:5], -1
-; GCN-NEXT:    s_cbranch_vccz .LBB1_5
+; GCN-NEXT:    s_cbranch_scc0 .LBB1_5
 ; GCN-NEXT:  ; %bb.4: ; %case1
 ; GCN-NEXT:    ; in Loop: Header=BB1_2 Depth=1
-; GCN-NEXT:    buffer_load_dword v2, off, s[0:3], 0 glc
+; GCN-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_cmp_ge_i32_e32 vcc, v0, v2
+; GCN-NEXT:    v_cmp_ge_i32_e32 vcc, v0, v1
 ; GCN-NEXT:    s_orn2_b64 s[4:5], vcc, exec
 ; GCN-NEXT:  .LBB1_5: ; %Flow3
 ; GCN-NEXT:    ; in Loop: Header=BB1_2 Depth=1
@@ -201,8 +203,8 @@ define amdgpu_kernel void @multi_if_break_loop(i32 %arg) #0 {
 ; GCN-NEXT:    s_cbranch_vccz .LBB1_1
 ; GCN-NEXT:  ; %bb.7: ; %LeafBlock
 ; GCN-NEXT:    ; in Loop: Header=BB1_2 Depth=1
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 0, v1
-; GCN-NEXT:    s_cbranch_vccz .LBB1_1
+; GCN-NEXT:    s_cmp_eq_u32 s8, 0
+; GCN-NEXT:    s_cbranch_scc0 .LBB1_1
 ; GCN-NEXT:  ; %bb.8: ; %case0
 ; GCN-NEXT:    ; in Loop: Header=BB1_2 Depth=1
 ; GCN-NEXT:    buffer_load_dword v1, off, s[0:3], 0 glc

diff --git a/llvm/test/CodeGen/AMDGPU/s_add_co_pseudo_lowering.mir b/llvm/test/CodeGen/AMDGPU/s_add_co_pseudo_lowering.mir
index 5e8d66d68e6c3d96181fd13411bac061327c28fa..f2d9aefa6673140be15a09c94e56613157f4d0db 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/s_add_co_pseudo_lowering.mir
+++ b/llvm/test/CodeGen/AMDGPU/s_add_co_pseudo_lowering.mir
@@ -14,27 +14,27 @@ body:             |
     ; GCN-NEXT: [[DEF:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
     ; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
     ; GCN-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
-    ; GCN-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr0
-    ; GCN-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr1
-    ; GCN-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr2
-    ; GCN-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY [[COPY3]]
-    ; GCN-NEXT: [[V_MUL_LO_U32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32_e64 [[COPY]], [[COPY4]], implicit $exec
-    ; GCN-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 killed [[V_MUL_LO_U32_e64_]], [[COPY6]], 0, implicit $exec
-    ; GCN-NEXT: [[S_MUL_HI_U32_:%[0-9]+]]:sreg_32 = S_MUL_HI_U32 [[COPY4]], [[COPY5]]
+    ; GCN-NEXT: [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+    ; GCN-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr1
+    ; GCN-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2
+    ; GCN-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY2]]
+    ; GCN-NEXT: [[V_MUL_LO_U32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32_e64 [[COPY]], [[COPY3]], implicit $exec
+    ; GCN-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 killed [[V_MUL_LO_U32_e64_]], [[COPY5]], 0, implicit $exec
+    ; GCN-NEXT: [[S_MUL_HI_U32_:%[0-9]+]]:sreg_32 = S_MUL_HI_U32 [[COPY3]], [[COPY4]]
     ; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -614296167
-    ; GCN-NEXT: [[V_MUL_LO_U32_e64_1:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32_e64 [[COPY]], [[COPY3]], implicit $exec
-    ; GCN-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[S_MOV_B32_]]
-    ; GCN-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 killed [[V_MUL_LO_U32_e64_1]], [[COPY7]], [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
-    ; GCN-NEXT: [[V_MUL_HI_U32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_HI_U32_e64 [[COPY4]], [[V_ADDC_U32_e64_]], implicit $exec
+    ; GCN-NEXT: [[V_MUL_LO_U32_e64_1:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32_e64 [[COPY]], [[COPY2]], implicit $exec
+    ; GCN-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY killed [[S_MOV_B32_]]
+    ; GCN-NEXT: [[V_ADDC_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADDC_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 killed [[V_MUL_LO_U32_e64_1]], [[COPY6]], [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
+    ; GCN-NEXT: [[V_MUL_HI_U32_e64_:%[0-9]+]]:vgpr_32 = V_MUL_HI_U32_e64 [[COPY3]], [[V_ADDC_U32_e64_]], implicit $exec
     ; GCN-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -181084736
     ; GCN-NEXT: [[V_MUL_LO_U32_e64_2:%[0-9]+]]:vgpr_32 = V_MUL_LO_U32_e64 [[V_MUL_HI_U32_e64_]], [[S_MOV_B32_1]], implicit $exec
-    ; GCN-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY killed [[S_MOV_B32_1]]
-    ; GCN-NEXT: [[V_ADDC_U32_e64_2:%[0-9]+]]:vgpr_32, [[V_ADDC_U32_e64_3:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 [[COPY8]], killed [[V_MUL_LO_U32_e64_2]], [[V_ADDC_U32_e64_1]], 0, implicit $exec
+    ; GCN-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[S_MOV_B32_1]]
+    ; GCN-NEXT: [[V_ADDC_U32_e64_2:%[0-9]+]]:vgpr_32, [[V_ADDC_U32_e64_3:%[0-9]+]]:sreg_64_xexec = V_ADDC_U32_e64 [[COPY7]], killed [[V_MUL_LO_U32_e64_2]], [[V_ADDC_U32_e64_1]], 0, implicit $exec
     %0:vgpr_32 = COPY $vgpr0
     %6:sreg_32 = COPY %0
     %1:vgpr_32 = COPY $vgpr1
     %2:vgpr_32 = COPY $vgpr2
-    %3:sreg_32 = COPY $sgpr0
+    %3:sreg_32 = COPY %2
     %4:sreg_32 = COPY $sgpr1
     %5:sreg_32 = COPY $sgpr2
     %20:vgpr_32 = COPY %3

diff --git a/llvm/test/CodeGen/AMDGPU/saddo.ll b/llvm/test/CodeGen/AMDGPU/saddo.ll
index b212a25e4473ac6bd87fb81b55ba0ad8c6b301c1..a9a3945e2f94d83eb84f21e494ba306aa4d61735 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/saddo.ll
+++ b/llvm/test/CodeGen/AMDGPU/saddo.ll
@@ -251,7 +251,7 @@ define amdgpu_kernel void @v_saddo_i32(i32 addrspace(1)* %out, i1 addrspace(1)*
 ; SI-NEXT:    s_mov_b32 s4, s2
 ; SI-NEXT:    s_mov_b32 s5, s3
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_add_i32_e32 v2, vcc, v1, v0
+; SI-NEXT:    v_add_i32_e32 v2, vcc, v0, v1
 ; SI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v1
 ; SI-NEXT:    v_cmp_lt_i32_e64 s[0:1], v2, v0
 ; SI-NEXT:    s_xor_b64 s[0:1], vcc, s[0:1]
@@ -275,7 +275,7 @@ define amdgpu_kernel void @v_saddo_i32(i32 addrspace(1)* %out, i1 addrspace(1)*
 ; VI-NEXT:    v_mov_b32_e32 v2, s2
 ; VI-NEXT:    v_mov_b32_e32 v3, s3
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_add_u32_e32 v6, vcc, v5, v4
+; VI-NEXT:    v_add_u32_e32 v6, vcc, v4, v5
 ; VI-NEXT:    v_cmp_gt_i32_e32 vcc, 0, v5
 ; VI-NEXT:    v_cmp_lt_i32_e64 s[0:1], v6, v4
 ; VI-NEXT:    s_xor_b64 s[0:1], vcc, s[0:1]

diff --git a/llvm/test/CodeGen/AMDGPU/scalar_to_vector.ll b/llvm/test/CodeGen/AMDGPU/scalar_to_vector.ll
index dccee2d1dcef8c7b9e60a48c5b38b79455ca8ea5..21c8260cbfefd1f559391636103e2fcde007b38f 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/scalar_to_vector.ll
+++ b/llvm/test/CodeGen/AMDGPU/scalar_to_vector.ll
@@ -104,14 +104,17 @@ define amdgpu_kernel void @scalar_to_vector_v4i16() {
 ; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v0
-; SI-NEXT:    v_or_b32_e32 v0, v1, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xff00, v0
-; SI-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
-; SI-NEXT:    v_or_b32_e32 v1, v2, v1
-; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v1
-; SI-NEXT:    v_or_b32_e32 v1, v1, v2
-; SI-NEXT:    v_or_b32_e32 v0, v0, v2
+; SI-NEXT:    v_readfirstlane_b32 s0, v0
+; SI-NEXT:    s_lshl_b32 s1, s0, 8
+; SI-NEXT:    s_or_b32 s0, s1, s0
+; SI-NEXT:    s_and_b32 s1, s0, 0xff00
+; SI-NEXT:    s_lshr_b32 s4, s0, 8
+; SI-NEXT:    s_or_b32 s1, s4, s1
+; SI-NEXT:    s_lshl_b32 s4, s1, 16
+; SI-NEXT:    s_or_b32 s1, s1, s4
+; SI-NEXT:    s_or_b32 s0, s0, s4
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_mov_b32_e32 v1, s1
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;
@@ -145,14 +148,17 @@ define amdgpu_kernel void @scalar_to_vector_v4f16() {
 ; SI-NEXT:    s_mov_b32 s2, -1
 ; SI-NEXT:    buffer_load_ubyte v0, off, s[0:3], 0
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 8, v0
-; SI-NEXT:    v_or_b32_e32 v0, v1, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xff00, v0
-; SI-NEXT:    v_lshrrev_b32_e32 v2, 8, v0
-; SI-NEXT:    v_or_b32_e32 v1, v2, v1
-; SI-NEXT:    v_lshlrev_b32_e32 v2, 16, v1
-; SI-NEXT:    v_or_b32_e32 v1, v1, v2
-; SI-NEXT:    v_or_b32_e32 v0, v0, v2
+; SI-NEXT:    v_readfirstlane_b32 s0, v0
+; SI-NEXT:    s_lshl_b32 s1, s0, 8
+; SI-NEXT:    s_or_b32 s0, s1, s0
+; SI-NEXT:    s_and_b32 s1, s0, 0xff00
+; SI-NEXT:    s_lshr_b32 s4, s0, 8
+; SI-NEXT:    s_or_b32 s1, s4, s1
+; SI-NEXT:    s_lshl_b32 s4, s1, 16
+; SI-NEXT:    s_or_b32 s1, s1, s4
+; SI-NEXT:    s_or_b32 s0, s0, s4
+; SI-NEXT:    v_mov_b32_e32 v0, s0
+; SI-NEXT:    v_mov_b32_e32 v1, s1
 ; SI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[0:3], 0
 ; SI-NEXT:    s_endpgm
 ;

diff --git a/llvm/test/CodeGen/AMDGPU/sdiv.ll b/llvm/test/CodeGen/AMDGPU/sdiv.ll
index c0c609a24f8449aac921eb1f536a416805d42cb4..5c83eeaadd0a3726268016c1b5206a27178f76c3 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/sdiv.ll
+++ b/llvm/test/CodeGen/AMDGPU/sdiv.ll
@@ -42,7 +42,7 @@ define amdgpu_kernel void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; 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, v4, v3
+; 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
@@ -55,7 +55,7 @@ define amdgpu_kernel void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; 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_sub_i32_e32 v0, vcc, 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
 ;
@@ -87,7 +87,7 @@ define amdgpu_kernel void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; 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, v4, v3
+; 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
@@ -100,7 +100,7 @@ define amdgpu_kernel void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; 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_sub_u32_e32 v0, vcc, 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
 ;
@@ -214,7 +214,7 @@ define amdgpu_kernel void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)*
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; GCN-NEXT:    v_lshrrev_b32_e32 v1, 30, v1
-; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
 ; GCN-NEXT:    v_ashrrev_i32_e32 v0, 2, v0
 ; GCN-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
@@ -235,7 +235,7 @@ define amdgpu_kernel void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)*
 ; TONGA-NEXT:    s_waitcnt vmcnt(0)
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v1, 31, v0
 ; TONGA-NEXT:    v_lshrrev_b32_e32 v1, 30, v1
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v0, 2, v0
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
@@ -433,21 +433,21 @@ define amdgpu_kernel void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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, v6, v7
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v7, v6
 ; 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_sub_i32_e32 v1, vcc, v1, v10
+; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v10, v1
 ; 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_sub_i32_e32 v6, vcc, v0, v2
+; 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_sub_i32_e32 v7, vcc, v1, v3
+; GCN-NEXT:    v_subrev_i32_e32 v7, vcc, v3, v1
 ; 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]
@@ -459,7 +459,7 @@ define amdgpu_kernel void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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_sub_i32_e32 v1, vcc, v1, v9
+; 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
 ;
@@ -506,7 +506,7 @@ define amdgpu_kernel void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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, v6, v7
+; TONGA-NEXT:    v_add_u32_e32 v5, vcc, v7, v6
 ; 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
@@ -518,9 +518,9 @@ define amdgpu_kernel void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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_sub_u32_e32 v6, vcc, v0, v2
+; 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_sub_u32_e32 v7, vcc, v1, v3
+; TONGA-NEXT:    v_subrev_u32_e32 v7, vcc, v3, v1
 ; 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]
@@ -853,7 +853,7 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; GCN-NEXT:    v_add_i32_e32 v11, vcc, 1, v8
 ; GCN-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v4
 ; GCN-NEXT:    v_cndmask_b32_e64 v8, v8, v11, s[0:1]
-; GCN-NEXT:    v_sub_i32_e32 v11, vcc, v0, v4
+; GCN-NEXT:    v_subrev_i32_e32 v11, vcc, v4, v0
 ; GCN-NEXT:    v_ashrrev_i32_e32 v14, 31, v7
 ; GCN-NEXT:    v_cndmask_b32_e64 v0, v0, v11, s[0:1]
 ; GCN-NEXT:    v_add_i32_e32 v7, vcc, v14, v7
@@ -865,27 +865,27 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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, v7, v9
+; GCN-NEXT:    v_add_i32_e32 v7, vcc, v9, v7
 ; GCN-NEXT:    v_mul_hi_u32 v7, v1, v7
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v10
 ; 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_sub_i32_e32 v1, vcc, v1, v10
+; GCN-NEXT:    v_subrev_i32_e32 v1, vcc, v10, v1
 ; 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_sub_i32_e32 v2, vcc, v2, v10
+; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v10, v2
 ; 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_sub_i32_e32 v0, vcc, v1, v5
+; 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_sub_i32_e32 v1, vcc, v2, v6
+; 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
@@ -894,7 +894,7 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_sub_i32_e32 v1, vcc, v5, v16
+; 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
@@ -907,19 +907,19 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_sub_i32_e32 v2, vcc, v2, v17
+; GCN-NEXT:    v_subrev_i32_e32 v2, vcc, v17, v2
 ; GCN-NEXT:    v_xor_b32_e32 v7, v8, v14
-; GCN-NEXT:    v_sub_i32_e32 v3, vcc, v3, v6
+; GCN-NEXT:    v_subrev_i32_e32 v3, vcc, v6, v3
 ; 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_sub_i32_e32 v6, vcc, v3, v4
+; 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_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_sub_i32_e32 v3, vcc, 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
 ;
@@ -984,7 +984,7 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; TONGA-NEXT:    v_add_u32_e32 v11, vcc, 1, v8
 ; TONGA-NEXT:    v_cmp_ge_u32_e64 s[0:1], v0, v4
 ; TONGA-NEXT:    v_cndmask_b32_e64 v8, v8, v11, s[0:1]
-; TONGA-NEXT:    v_sub_u32_e32 v11, vcc, v0, v4
+; TONGA-NEXT:    v_subrev_u32_e32 v11, vcc, v4, v0
 ; TONGA-NEXT:    v_ashrrev_i32_e32 v14, 31, v7
 ; TONGA-NEXT:    v_cndmask_b32_e64 v0, v0, v11, s[0:1]
 ; TONGA-NEXT:    v_add_u32_e32 v7, vcc, v14, v7
@@ -996,7 +996,7 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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, v7, v9
+; TONGA-NEXT:    v_add_u32_e32 v7, vcc, v9, v7
 ; TONGA-NEXT:    v_mul_hi_u32 v7, v1, v7
 ; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v10
 ; TONGA-NEXT:    v_mul_hi_u32 v0, v2, v0
@@ -1008,15 +1008,15 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_sub_u32_e32 v2, vcc, v2, v10
+; TONGA-NEXT:    v_subrev_u32_e32 v2, vcc, v10, v2
 ; 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_sub_u32_e32 v0, vcc, v1, v5
+; 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_sub_u32_e32 v1, vcc, v2, v6
+; TONGA-NEXT:    v_subrev_u32_e32 v1, vcc, v6, v2
 ; TONGA-NEXT:    v_cndmask_b32_e64 v2, v2, v1, s[4:5]
 ; TONGA-NEXT:    v_add_u32_e32 v1, vcc, 1, v7
 ; TONGA-NEXT:    v_cmp_ge_u32_e32 vcc, v0, v5
@@ -1040,17 +1040,17 @@ define amdgpu_kernel void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_sub_u32_e32 v3, vcc, v3, v6
+; TONGA-NEXT:    v_subrev_u32_e32 v3, vcc, v6, v3
 ; 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_sub_u32_e32 v6, vcc, v3, v4
+; 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_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_subrev_u32_e32 v3, vcc, v7, v3
+; TONGA-NEXT:    v_sub_u32_e32 v3, vcc, v3, v7
 ; TONGA-NEXT:    buffer_store_dwordx4 v[0:3], off, s[8:11], 0
 ; TONGA-NEXT:    s_endpgm
 ;
@@ -1515,7 +1515,7 @@ define amdgpu_kernel void @v_sdiv_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %i
 ; TONGA-NEXT:    v_cvt_i32_f32_e32 v1, v1
 ; TONGA-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
 ; TONGA-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
 ; TONGA-NEXT:    v_bfe_i32 v0, v0, 0, 8
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
@@ -1680,7 +1680,7 @@ define amdgpu_kernel void @v_sdiv_i23(i32 addrspace(1)* %out, i23 addrspace(1)*
 ; TONGA-NEXT:    v_cvt_i32_f32_e32 v1, v1
 ; TONGA-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v2|
 ; TONGA-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
 ; TONGA-NEXT:    v_bfe_i32 v0, v0, 0, 23
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
@@ -1824,7 +1824,7 @@ define amdgpu_kernel void @v_sdiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)*
 ; 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:    buffer_store_dword v0, off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
@@ -1865,7 +1865,7 @@ define amdgpu_kernel void @v_sdiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)*
 ; TONGA-NEXT:    v_cvt_i32_f32_e32 v2, v2
 ; TONGA-NEXT:    v_cmp_ge_f32_e64 vcc, |v3|, |v1|
 ; TONGA-NEXT:    v_cndmask_b32_e32 v0, 0, v0, vcc
-; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v2, v0
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v0, v2
 ; TONGA-NEXT:    v_bfe_i32 v0, v0, 0, 24
 ; TONGA-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
@@ -1993,21 +1993,21 @@ define amdgpu_kernel void @v_sdiv_i25(i32 addrspace(1)* %out, i25 addrspace(1)*
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_bfe_i32 v2, v1, 0, 25
 ; GCN-NEXT:    v_bfe_i32 v1, v1, 24, 1
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v1, v2
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v1
 ; GCN-NEXT:    v_xor_b32_e32 v2, v2, v1
 ; GCN-NEXT:    v_cvt_f32_u32_e32 v3, v2
 ; GCN-NEXT:    v_sub_i32_e32 v4, vcc, 0, v2
 ; GCN-NEXT:    v_bfe_i32 v5, v0, 0, 25
 ; GCN-NEXT:    v_rcp_iflag_f32_e32 v3, v3
 ; GCN-NEXT:    v_bfe_i32 v0, v0, 24, 1
-; GCN-NEXT:    v_add_i32_e32 v5, vcc, v0, v5
+; GCN-NEXT:    v_add_i32_e32 v5, vcc, v5, v0
 ; GCN-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v3, v3
 ; 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, v4, v3
+; 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
@@ -2020,7 +2020,7 @@ define amdgpu_kernel void @v_sdiv_i25(i32 addrspace(1)* %out, i25 addrspace(1)*
 ; 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_sub_i32_e32 v0, vcc, 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
@@ -2041,21 +2041,21 @@ define amdgpu_kernel void @v_sdiv_i25(i32 addrspace(1)* %out, i25 addrspace(1)*
 ; TONGA-NEXT:    s_waitcnt vmcnt(0)
 ; TONGA-NEXT:    v_bfe_i32 v2, v1, 0, 25
 ; TONGA-NEXT:    v_bfe_i32 v1, v1, 24, 1
-; TONGA-NEXT:    v_add_u32_e32 v2, vcc, v1, v2
+; TONGA-NEXT:    v_add_u32_e32 v2, vcc, v2, v1
 ; TONGA-NEXT:    v_xor_b32_e32 v2, v2, v1
 ; TONGA-NEXT:    v_cvt_f32_u32_e32 v3, v2
 ; TONGA-NEXT:    v_sub_u32_e32 v4, vcc, 0, v2
 ; TONGA-NEXT:    v_bfe_i32 v5, v0, 0, 25
 ; TONGA-NEXT:    v_rcp_iflag_f32_e32 v3, v3
 ; TONGA-NEXT:    v_bfe_i32 v0, v0, 24, 1
-; TONGA-NEXT:    v_add_u32_e32 v5, vcc, v0, v5
+; TONGA-NEXT:    v_add_u32_e32 v5, vcc, v5, v0
 ; TONGA-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v3
 ; TONGA-NEXT:    v_cvt_u32_f32_e32 v3, v3
 ; 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, v4, v3
+; 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
@@ -2232,7 +2232,7 @@ define amdgpu_kernel void @scalarize_mulhs_4xi32(<4 x i32> addrspace(1)* nocaptu
 ; GCN-NEXT:    v_ashrrev_i32_e32 v3, 12, v3
 ; GCN-NEXT:    v_add_i32_e32 v0, vcc, v0, v4
 ; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v5
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v6
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v6, v2
 ; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v7
 ; GCN-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; GCN-NEXT:    s_endpgm
@@ -2262,10 +2262,10 @@ define amdgpu_kernel void @scalarize_mulhs_4xi32(<4 x i32> addrspace(1)* nocaptu
 ; 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, v0, v4
+; TONGA-NEXT:    v_add_u32_e32 v0, vcc, v4, v0
 ; 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, v3, v7
+; TONGA-NEXT:    v_add_u32_e32 v3, vcc, v7, v3
 ; TONGA-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; TONGA-NEXT:    s_endpgm
 ;

diff --git a/llvm/test/CodeGen/AMDGPU/sdiv64.ll b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
index 56719aee1acbd1eed9ac01e099b6b2f218340d06..ac15a6a14a5b14d1f1a5f8f4f17512e3c0091473 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/sdiv64.ll
+++ b/llvm/test/CodeGen/AMDGPU/sdiv64.ll
@@ -36,8 +36,8 @@ define amdgpu_kernel void @s_test_sdiv(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-NEXT:    v_mul_lo_u32 v5, s5, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s4, v0
 ; GCN-NEXT:    s_addc_u32 s3, s3, s12
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -97,9 +97,9 @@ define amdgpu_kernel void @s_test_sdiv(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-NEXT:    v_mul_hi_u32 v3, s10, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s11, v0
 ; GCN-NEXT:    v_mov_b32_e32 v5, s11
-; 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, s10, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_sub_i32_e32 v4, vcc, s3, v2
 ; GCN-NEXT:    v_sub_i32_e32 v3, vcc, s2, v3
 ; GCN-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, vcc
@@ -524,7 +524,7 @@ define amdgpu_kernel void @s_test_sdiv24_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_add_i32_e32 v0, vcc, v2, 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
@@ -628,7 +628,7 @@ define amdgpu_kernel void @s_test_sdiv32_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_add_i32_e32 v0, vcc, v2, 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
@@ -664,7 +664,7 @@ define amdgpu_kernel void @s_test_sdiv31_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_add_i32_e32 v0, vcc, v2, 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[0:3], 0
@@ -731,7 +731,7 @@ define amdgpu_kernel void @s_test_sdiv23_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_add_i32_e32 v0, vcc, v2, 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[0:3], 0
@@ -798,7 +798,7 @@ define amdgpu_kernel void @s_test_sdiv25_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_add_i32_e32 v0, vcc, v2, 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[0:3], 0
@@ -926,7 +926,7 @@ define amdgpu_kernel void @s_test_sdiv24_v2i64(<2 x i64> addrspace(1)* %out, <2
 ; 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, v2, v4
+; GCN-IR-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; 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
@@ -1115,8 +1115,8 @@ define amdgpu_kernel void @s_test_sdiv_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s5, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s4, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -1165,7 +1165,7 @@ define amdgpu_kernel void @s_test_sdiv_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v1, s3, v0
 ; GCN-NEXT:    v_mul_hi_u32 v2, s2, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, s2, v0
 ; GCN-NEXT:    v_sub_i32_e32 v3, vcc, 0, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, 24, v2
@@ -1800,7 +1800,7 @@ define amdgpu_kernel void @s_test_sdiv24_k_num_i64(i64 addrspace(1)* %out, i64 %
 ; 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, v0, v1
+; GCN-NEXT:    v_add_i32_e32 v0, vcc, v1, 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
@@ -1827,7 +1827,7 @@ define amdgpu_kernel void @s_test_sdiv24_k_num_i64(i64 addrspace(1)* %out, i64 %
 ; 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_add_i32_e32 v0, vcc, v1, 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
@@ -1858,7 +1858,7 @@ define amdgpu_kernel void @s_test_sdiv24_k_den_i64(i64 addrspace(1)* %out, i64 %
 ; 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:    v_add_i32_e32 v0, vcc, v1, v0
 ; 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

diff --git a/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll b/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll
index f7092e525adc55c62015664c4f570b98960ccfe6..b2d13e5d988d107f10916a8631534b466d96614b 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll
+++ b/llvm/test/CodeGen/AMDGPU/sdwa-peephole.ll
@@ -487,7 +487,10 @@ entry:
 ; GCN-LABEL: {{^}}add_bb_v2i16:
 ; NOSDWA-NOT: v_add_{{(_co)?}}_u32_sdwa
 
-; VI: v_add_u32_sdwa v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; VI: v_readfirstlane_b32 [[LO:s[0-9]+]]
+; VI: v_readfirstlane_b32 [[HI:s[0-9]+]]
+; VI: s_lshr_b32 [[LOSH:s[0-9]+]], [[LO]], 16
+; VI: s_lshr_b32 [[HISH:s[0-9]+]], [[HI]], 16
 
 ; GFX9_10: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 

diff --git a/llvm/test/CodeGen/AMDGPU/shl.ll b/llvm/test/CodeGen/AMDGPU/shl.ll
index 70dd26200712e1e0e130e3950852a7f5db0f2d11..a2f169a83b3eeb3f69f7ed4966070781d7a881b3 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/shl.ll
+++ b/llvm/test/CodeGen/AMDGPU/shl.ll
@@ -83,9 +83,9 @@ define amdgpu_kernel void @shl_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> add
 ; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_lshl_b32_e32 v3, v3, v7
-; SI-NEXT:    v_lshl_b32_e32 v2, v2, v6
-; SI-NEXT:    v_lshl_b32_e32 v1, v1, v5
+; SI-NEXT:    v_lshlrev_b32_e32 v3, v7, v3
+; SI-NEXT:    v_lshlrev_b32_e32 v2, v6, v2
+; SI-NEXT:    v_lshlrev_b32_e32 v1, v5, v1
 ; SI-NEXT:    v_lshl_b32_e32 v0, v0, v4
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm

diff --git a/llvm/test/CodeGen/AMDGPU/sra.ll b/llvm/test/CodeGen/AMDGPU/sra.ll
index 3474235c3f5f9136cc1128f77c52906507cffc3e..2d43c58aa38a21768bce9101aeda57aedd002a86 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/sra.ll
+++ b/llvm/test/CodeGen/AMDGPU/sra.ll
@@ -85,9 +85,9 @@ define amdgpu_kernel void @ashr_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_ashr_i32_e32 v3, v3, v7
-; SI-NEXT:    v_ashr_i32_e32 v2, v2, v6
-; SI-NEXT:    v_ashr_i32_e32 v1, v1, v5
+; SI-NEXT:    v_ashrrev_i32_e32 v3, v7, v3
+; SI-NEXT:    v_ashrrev_i32_e32 v2, v6, v2
+; SI-NEXT:    v_ashrrev_i32_e32 v1, v5, v1
 ; SI-NEXT:    v_ashr_i32_e32 v0, v0, v4
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
@@ -159,14 +159,17 @@ define amdgpu_kernel void @ashr_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> ad
 ; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_bfe_i32 v2, v0, 0, 16
-; SI-NEXT:    v_ashrrev_i32_e32 v0, 16, v0
-; SI-NEXT:    v_lshrrev_b32_e32 v3, 16, v1
-; SI-NEXT:    v_ashrrev_i32_e32 v0, v3, v0
-; SI-NEXT:    v_ashrrev_i32_e32 v1, v1, v2
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v1, 0xffff, v1
-; SI-NEXT:    v_or_b32_e32 v0, v1, v0
+; SI-NEXT:    v_readfirstlane_b32 s0, v0
+; SI-NEXT:    v_readfirstlane_b32 s1, v1
+; SI-NEXT:    s_sext_i32_i16 s2, s0
+; SI-NEXT:    s_ashr_i32 s0, s0, 16
+; SI-NEXT:    s_lshr_b32 s3, s1, 16
+; SI-NEXT:    s_ashr_i32 s0, s0, s3
+; SI-NEXT:    s_ashr_i32 s1, s2, s1
+; SI-NEXT:    s_lshl_b32 s0, s0, 16
+; SI-NEXT:    s_and_b32 s1, s1, 0xffff
+; SI-NEXT:    s_or_b32 s0, s1, s0
+; SI-NEXT:    v_mov_b32_e32 v0, s0
 ; SI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;
@@ -184,9 +187,18 @@ define amdgpu_kernel void @ashr_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> ad
 ; VI-NEXT:    s_mov_b32 s4, s0
 ; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ashrrev_i32_sdwa v2, sext(v1), sext(v0) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
-; VI-NEXT:    v_ashrrev_i32_sdwa v0, sext(v1), sext(v0) dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_sdwa v0, v2, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_readfirstlane_b32 s0, v0
+; VI-NEXT:    v_readfirstlane_b32 s1, v1
+; VI-NEXT:    s_ashr_i32 s2, s0, 16
+; VI-NEXT:    s_sext_i32_i16 s0, s0
+; VI-NEXT:    s_ashr_i32 s3, s1, 16
+; VI-NEXT:    s_sext_i32_i16 s1, s1
+; VI-NEXT:    s_ashr_i32 s0, s0, s1
+; VI-NEXT:    s_ashr_i32 s1, s2, s3
+; VI-NEXT:    s_lshl_b32 s1, s1, 16
+; VI-NEXT:    s_and_b32 s0, s0, 0xffff
+; VI-NEXT:    s_or_b32 s0, s0, s1
+; VI-NEXT:    v_mov_b32_e32 v0, s0
 ; VI-NEXT:    buffer_store_dword v0, off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;
@@ -243,22 +255,28 @@ define amdgpu_kernel void @ashr_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> ad
 ; SI-NEXT:    s_mov_b32 s4, s0
 ; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
-; SI-NEXT:    v_bfe_i32 v4, v0, 0, 16
-; SI-NEXT:    v_ashrrev_i32_e32 v0, 16, v0
-; SI-NEXT:    v_bfe_i32 v5, v1, 0, 16
-; SI-NEXT:    v_ashrrev_i32_e32 v1, 16, v1
-; SI-NEXT:    v_lshrrev_b32_e32 v6, 16, v2
-; SI-NEXT:    v_lshrrev_b32_e32 v7, 16, v3
-; SI-NEXT:    v_ashr_i32_e32 v1, v1, v7
-; SI-NEXT:    v_ashr_i32_e32 v3, v5, v3
-; SI-NEXT:    v_ashr_i32_e32 v0, v0, v6
-; SI-NEXT:    v_ashr_i32_e32 v2, v4, v2
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
-; SI-NEXT:    v_and_b32_e32 v3, 0xffff, v3
-; SI-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
-; SI-NEXT:    v_and_b32_e32 v2, 0xffff, v2
-; SI-NEXT:    v_or_b32_e32 v1, v3, v1
-; SI-NEXT:    v_or_b32_e32 v0, v2, v0
+; SI-NEXT:    v_readfirstlane_b32 s0, v3
+; SI-NEXT:    v_readfirstlane_b32 s1, v2
+; SI-NEXT:    v_readfirstlane_b32 s2, v1
+; SI-NEXT:    v_readfirstlane_b32 s3, v0
+; SI-NEXT:    s_sext_i32_i16 s8, s3
+; SI-NEXT:    s_ashr_i32 s3, s3, 16
+; SI-NEXT:    s_sext_i32_i16 s9, s2
+; SI-NEXT:    s_ashr_i32 s2, s2, 16
+; SI-NEXT:    s_lshr_b32 s10, s1, 16
+; SI-NEXT:    s_lshr_b32 s11, s0, 16
+; SI-NEXT:    s_ashr_i32 s2, s2, s11
+; SI-NEXT:    s_ashr_i32 s0, s9, s0
+; SI-NEXT:    s_ashr_i32 s3, s3, s10
+; SI-NEXT:    s_ashr_i32 s1, s8, s1
+; SI-NEXT:    s_lshl_b32 s2, s2, 16
+; SI-NEXT:    s_and_b32 s0, s0, 0xffff
+; SI-NEXT:    s_lshl_b32 s3, s3, 16
+; SI-NEXT:    s_and_b32 s1, s1, 0xffff
+; SI-NEXT:    s_or_b32 s0, s0, s2
+; SI-NEXT:    s_or_b32 s1, s1, s3
+; 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
 ;
@@ -276,12 +294,30 @@ define amdgpu_kernel void @ashr_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> ad
 ; VI-NEXT:    s_mov_b32 s4, s0
 ; VI-NEXT:    s_mov_b32 s5, s1
 ; VI-NEXT:    s_waitcnt vmcnt(0)
-; VI-NEXT:    v_ashrrev_i32_sdwa v4, sext(v2), sext(v0) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
-; VI-NEXT:    v_ashrrev_i32_sdwa v0, sext(v2), sext(v0) dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; VI-NEXT:    v_ashrrev_i32_sdwa v2, sext(v3), sext(v1) dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:WORD_0
-; VI-NEXT:    v_ashrrev_i32_sdwa v1, sext(v3), sext(v1) dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; VI-NEXT:    v_or_b32_sdwa v1, v2, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
-; VI-NEXT:    v_or_b32_sdwa v0, v4, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD
+; VI-NEXT:    v_readfirstlane_b32 s0, v2
+; VI-NEXT:    v_readfirstlane_b32 s1, v3
+; VI-NEXT:    v_readfirstlane_b32 s2, v0
+; VI-NEXT:    v_readfirstlane_b32 s3, v1
+; VI-NEXT:    s_ashr_i32 s8, s3, 16
+; VI-NEXT:    s_sext_i32_i16 s3, s3
+; VI-NEXT:    s_ashr_i32 s9, s2, 16
+; VI-NEXT:    s_sext_i32_i16 s2, s2
+; VI-NEXT:    s_ashr_i32 s10, s1, 16
+; VI-NEXT:    s_sext_i32_i16 s1, s1
+; VI-NEXT:    s_ashr_i32 s11, s0, 16
+; VI-NEXT:    s_sext_i32_i16 s0, s0
+; VI-NEXT:    s_ashr_i32 s0, s2, s0
+; VI-NEXT:    s_ashr_i32 s2, s9, s11
+; VI-NEXT:    s_ashr_i32 s1, s3, s1
+; VI-NEXT:    s_ashr_i32 s3, s8, s10
+; VI-NEXT:    s_lshl_b32 s3, s3, 16
+; VI-NEXT:    s_and_b32 s1, s1, 0xffff
+; VI-NEXT:    s_lshl_b32 s2, s2, 16
+; VI-NEXT:    s_and_b32 s0, s0, 0xffff
+; VI-NEXT:    s_or_b32 s1, s1, s3
+; VI-NEXT:    s_or_b32 s0, s0, s2
+; VI-NEXT:    v_mov_b32_e32 v0, s0
+; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; VI-NEXT:    s_endpgm
 ;

diff --git a/llvm/test/CodeGen/AMDGPU/srem64.ll b/llvm/test/CodeGen/AMDGPU/srem64.ll
index c7cadbf49cb117112f81a0d4caede4b756c82f5d..32d54d90b6f1d38b835e2bf25791078c7986f7ba 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/srem64.ll
+++ b/llvm/test/CodeGen/AMDGPU/srem64.ll
@@ -28,8 +28,8 @@ define amdgpu_kernel void @s_test_srem(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s1, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -86,8 +86,8 @@ define amdgpu_kernel void @s_test_srem(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; GCN-NEXT:    v_mul_hi_u32 v2, s12, v0
 ; GCN-NEXT:    v_mul_lo_u32 v3, s13, v0
 ; GCN-NEXT:    v_mul_lo_u32 v0, s12, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, s11, v1
 ; GCN-NEXT:    v_mov_b32_e32 v3, s13
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s10, v0
@@ -202,8 +202,8 @@ define amdgpu_kernel void @s_test_srem(i64 addrspace(1)* %out, i64 %x, i64 %y) {
 ; 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, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; 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:    s_mov_b32 s10, -1
@@ -669,7 +669,7 @@ define amdgpu_kernel void @s_test_srem25_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_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
@@ -701,7 +701,7 @@ define amdgpu_kernel void @s_test_srem25_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_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
@@ -740,7 +740,7 @@ define amdgpu_kernel void @s_test_srem31_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_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
@@ -772,7 +772,7 @@ define amdgpu_kernel void @s_test_srem31_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_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
@@ -810,7 +810,7 @@ define amdgpu_kernel void @s_test_srem32_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_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, s3, v0
@@ -839,7 +839,7 @@ define amdgpu_kernel void @s_test_srem32_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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_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, s3, v0
@@ -889,8 +889,8 @@ define amdgpu_kernel void @s_test_srem33_64(i64 addrspace(1)* %out, i64 %x, i64
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s1, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v6, v0, v2
@@ -950,8 +950,8 @@ define amdgpu_kernel void @s_test_srem33_64(i64 addrspace(1)* %out, i64 %x, i64
 ; GCN-NEXT:    v_mul_hi_u32 v2, s12, v0
 ; GCN-NEXT:    v_mul_lo_u32 v3, s13, v0
 ; GCN-NEXT:    v_mul_lo_u32 v0, s12, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, s15, v1
 ; GCN-NEXT:    v_mov_b32_e32 v3, s13
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s14, v0
@@ -1078,13 +1078,15 @@ define amdgpu_kernel void @s_test_srem33_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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:    v_mul_lo_u32 v1, s8, v1
 ; 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_mov_b32 s6, -1
+; GCN-IR-NEXT:    v_readfirstlane_b32 s10, v2
+; GCN-IR-NEXT:    v_mul_lo_u32 v2, s9, v0
+; GCN-IR-NEXT:    v_mul_lo_u32 v0, s8, v0
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, s10, v1
+; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
 ; 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
@@ -1092,7 +1094,6 @@ define amdgpu_kernel void @s_test_srem33_64(i64 addrspace(1)* %out, i64 %x, i64
 ; 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:    buffer_store_dwordx2 v[0:1], off, s[4:7], 0
 ; GCN-IR-NEXT:    s_endpgm
@@ -1239,8 +1240,8 @@ define amdgpu_kernel void @s_test_srem24_48(i48 addrspace(1)* %out, i48 %x, i48
 ; 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:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; 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, s5
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s4, v0
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, v2, v1, vcc
@@ -1291,8 +1292,8 @@ define amdgpu_kernel void @s_test_srem_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_mul_hi_u32 v3, s2, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s3, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s2, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -1340,7 +1341,7 @@ define amdgpu_kernel void @s_test_srem_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_mul_lo_u32 v1, s9, v0
 ; GCN-NEXT:    v_mul_hi_u32 v2, s8, v0
 ; GCN-NEXT:    v_mul_lo_u32 v0, s8, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
 ; 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
@@ -1447,8 +1448,8 @@ define amdgpu_kernel void @s_test_srem_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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 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_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 s2, -1
 ; GCN-IR-NEXT:    v_subb_u32_e32 v1, vcc, 0, v1, vcc
@@ -1970,7 +1971,7 @@ define amdgpu_kernel void @s_test_srem24_k_num_i64(i64 addrspace(1)* %out, i64 %
 ; 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, v0, v1
+; 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
@@ -2030,7 +2031,7 @@ define amdgpu_kernel void @s_test_srem24_k_den_i64(i64 addrspace(1)* %out, i64 %
 ; 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, v0, v2
+; 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_mov_b32 s5, s1
@@ -2059,7 +2060,7 @@ define amdgpu_kernel void @s_test_srem24_k_den_i64(i64 addrspace(1)* %out, i64 %
 ; 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_add_i32_e32 v0, vcc, v2, v0
 ; GCN-IR-NEXT:    v_mul_lo_u32 v0, v0, s3
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0
 ; GCN-IR-NEXT:    s_mov_b32 s5, s1

diff --git a/llvm/test/CodeGen/AMDGPU/srl.ll b/llvm/test/CodeGen/AMDGPU/srl.ll
index f2d392144344cffa631072dd35b325a01df93c88..9693c44fac30a206672cd5285d0e530548763c9d 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/srl.ll
+++ b/llvm/test/CodeGen/AMDGPU/srl.ll
@@ -138,9 +138,9 @@ define amdgpu_kernel void @lshr_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; SI-NEXT:    s_mov_b32 s5, s1
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    v_lshr_b32_e32 v3, v3, v7
-; SI-NEXT:    v_lshr_b32_e32 v2, v2, v6
-; SI-NEXT:    v_lshr_b32_e32 v1, v1, v5
-; SI-NEXT:    v_lshr_b32_e32 v0, v0, v4
+; SI-NEXT:    v_lshrrev_b32_e32 v2, v6, v2
+; SI-NEXT:    v_lshrrev_b32_e32 v1, v5, v1
+; SI-NEXT:    v_lshrrev_b32_e32 v0, v4, v0
 ; SI-NEXT:    buffer_store_dwordx4 v[0:3], off, s[4:7], 0
 ; SI-NEXT:    s_endpgm
 ;

diff --git a/llvm/test/CodeGen/AMDGPU/sub.ll b/llvm/test/CodeGen/AMDGPU/sub.ll
index e38f8fa00f8acc5277c432a7fff094ea0cd9af09..290ba016f40645ea2346774a0db329e1b9cde6ed 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/sub.ll
+++ b/llvm/test/CodeGen/AMDGPU/sub.ll
@@ -47,7 +47,7 @@ define amdgpu_kernel void @test_sub_imm_i32(i32 addrspace(1)* %out, i32 addrspac
 
 ; GCN-LABEL: {{^}}test_sub_v2i32:
 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
-; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
+; SI: v_subrev_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
 
 ; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
 ; GFX9: v_sub_u32_e32 v{{[0-9]+, v[0-9]+, v[0-9]+}}
@@ -61,7 +61,7 @@ define amdgpu_kernel void @test_sub_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32
 }
 
 ; GCN-LABEL: {{^}}test_sub_v4i32:
-; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
+; SI: v_subrev_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}
 ; SI: v_sub_i32_e32 v{{[0-9]+, vcc, v[0-9]+, v[0-9]+}}

diff --git a/llvm/test/CodeGen/AMDGPU/udiv.ll b/llvm/test/CodeGen/AMDGPU/udiv.ll
index ab2322d03cc886cba4dc13a96492c47c9d0cf1e4..b3196c33f57f142a96d30a5edf173a0b9401eccc 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/udiv.ll
+++ b/llvm/test/CodeGen/AMDGPU/udiv.ll
@@ -27,7 +27,7 @@ define amdgpu_kernel void @udiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; 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, v3, v2
+; 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
@@ -63,7 +63,7 @@ define amdgpu_kernel void @udiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; 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, v3, v2
+; 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
@@ -93,7 +93,7 @@ define amdgpu_kernel void @udiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %i
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v2, v2
 ; 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, v3, v2
+; 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
@@ -193,7 +193,7 @@ define amdgpu_kernel void @s_udiv_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) {
 ; SI-NEXT:    v_mul_lo_u32 v1, s4, v0
 ; SI-NEXT:    s_load_dwordx2 s[4:5], s[0:1], 0x9
 ; SI-NEXT:    v_mul_hi_u32 v1, v0, v1
-; SI-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; SI-NEXT:    v_add_i32_e32 v0, vcc, v0, v1
 ; 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
@@ -271,25 +271,29 @@ define amdgpu_kernel void @s_udiv_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) {
 ; GFX1030-LABEL: s_udiv_i32:
 ; GFX1030:       ; %bb.0:
 ; GFX1030-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x8
+; GFX1030-NEXT:    v_mov_b32_e32 v3, 0
 ; GFX1030-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1030-NEXT:    v_cvt_f32_u32_e32 v0, s1
-; GFX1030-NEXT:    s_sub_i32 s2, 0, s1
+; GFX1030-NEXT:    s_sub_i32 s3, 0, s1
 ; 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_mul_lo_u32 v1, s2, v0
+; GFX1030-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX1030-NEXT:    s_mul_i32 s3, s3, s2
+; GFX1030-NEXT:    s_mul_hi_u32 s3, s2, s3
+; GFX1030-NEXT:    s_add_i32 s2, s2, s3
+; GFX1030-NEXT:    s_mul_hi_u32 s6, s0, s2
+; GFX1030-NEXT:    s_mul_i32 s2, s6, s1
+; GFX1030-NEXT:    s_sub_i32 s0, s0, s2
 ; GFX1030-NEXT:    s_load_dwordx2 s[2:3], s[4:5], 0x0
-; GFX1030-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX1030-NEXT:    v_add_nc_u32_e32 v0, v0, v1
-; GFX1030-NEXT:    v_mul_hi_u32 v0, s0, v0
-; GFX1030-NEXT:    v_mul_lo_u32 v1, v0, s1
-; GFX1030-NEXT:    v_add_nc_u32_e32 v2, 1, v0
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v1, s0, v1
-; GFX1030-NEXT:    v_subrev_nc_u32_e32 v3, s1, v1
-; GFX1030-NEXT:    v_cmp_le_u32_e32 vcc_lo, s1, v1
-; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc_lo
-; GFX1030-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc_lo
-; GFX1030-NEXT:    v_mov_b32_e32 v3, 0
+; GFX1030-NEXT:    s_cmp_ge_u32 s0, s1
+; GFX1030-NEXT:    s_cselect_b32 vcc_lo, -1, 0
+; GFX1030-NEXT:    s_add_i32 s7, s6, 1
+; GFX1030-NEXT:    s_sub_i32 s4, s0, s1
+; GFX1030-NEXT:    v_mov_b32_e32 v0, s7
+; GFX1030-NEXT:    v_mov_b32_e32 v1, s4
+; GFX1030-NEXT:    v_cndmask_b32_e32 v0, s6, v0, vcc_lo
+; GFX1030-NEXT:    v_cndmask_b32_e32 v1, s0, v1, vcc_lo
 ; GFX1030-NEXT:    v_add_nc_u32_e32 v2, 1, v0
 ; GFX1030-NEXT:    v_cmp_le_u32_e32 vcc_lo, s1, v1
 ; GFX1030-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc_lo
@@ -368,7 +372,7 @@ define amdgpu_kernel void @udiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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_subrev_i32_e32 v1, vcc, v8, v1
+; 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
@@ -422,13 +426,13 @@ define amdgpu_kernel void @udiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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_subrev_u32_e32 v0, vcc, v6, v0
+; VI-NEXT:    v_sub_u32_e32 v0, vcc, v0, v6
 ; VI-NEXT:    v_subrev_u32_e32 v1, vcc, v8, v1
 ; 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_cndmask_b32_e64 v4, v4, v7, s[0:1]
-; VI-NEXT:    v_subrev_u32_e32 v6, vcc, v2, v0
+; 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 v0, v0, v6, s[0:1]
@@ -473,13 +477,13 @@ define amdgpu_kernel void @udiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> ad
 ; 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_subrev_u32_e32 v0, vcc, v8, v0
+; 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_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 v6, v6, v9, s[0:1]
-; GCN-NEXT:    v_subrev_u32_e32 v8, vcc, v2, v0
+; 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 v0, v0, v8, s[0:1]
@@ -654,9 +658,9 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_subrev_i32_e32 v5, vcc, v14, v5
-; SI-NEXT:    v_subrev_i32_e32 v6, vcc, v16, v6
-; SI-NEXT:    v_subrev_i32_e32 v7, vcc, v18, v7
+; 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
@@ -670,9 +674,9 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; SI-NEXT:    v_cndmask_b32_e64 v9, v9, v15, s[2:3]
 ; 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_subrev_i32_e32 v14, vcc, v2, v6
+; SI-NEXT:    v_sub_i32_e32 v14, vcc, v6, v2
 ; SI-NEXT:    v_cndmask_b32_e64 v11, v11, v19, s[6:7]
-; SI-NEXT:    v_subrev_i32_e32 v15, vcc, v3, v7
+; 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]
@@ -736,8 +740,8 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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, v11, v10
-; VI-NEXT:    v_add_u32_e32 v10, vcc, v13, v12
+; VI-NEXT:    v_add_u32_e32 v9, vcc, v10, v11
+; 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
@@ -748,10 +752,10 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_subrev_u32_e32 v4, vcc, v12, v4
+; 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_subrev_u32_e32 v7, vcc, v18, v7
+; 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
@@ -831,8 +835,8 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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, v13, v12
-; GCN-NEXT:    v_add_u32_e32 v12, vcc, v15, v14
+; GCN-NEXT:    v_add_u32_e32 v11, vcc, v12, v13
+; 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
@@ -843,10 +847,10 @@ define amdgpu_kernel void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> ad
 ; 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_subrev_u32_e32 v4, vcc, v14, v4
+; 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_subrev_u32_e32 v7, vcc, v19, v7
+; 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
@@ -1854,10 +1858,10 @@ define amdgpu_kernel void @v_udiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)*
 ; 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_sub_i32_e32 v2, vcc, v2, v3
 ; 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_subrev_i32_e32 v3, vcc, v0, v2
+; SI-NEXT:    v_sub_i32_e32 v3, vcc, v2, v0
 ; SI-NEXT:    v_cndmask_b32_e64 v2, v2, v3, s[0:1]
 ; SI-NEXT:    v_add_i32_e32 v3, vcc, 1, v1
 ; SI-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v0
@@ -1897,14 +1901,14 @@ define amdgpu_kernel void @v_udiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)*
 ; 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, v4, v1
+; 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_subrev_u32_e32 v2, vcc, v3, v2
+; 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_subrev_u32_e32 v3, vcc, v0, v2
+; 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_add_u32_e32 v3, vcc, 1, v1
 ; VI-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v0
@@ -1950,16 +1954,16 @@ define amdgpu_kernel void @v_udiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)*
 ; GCN-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GCN-NEXT:    v_mul_lo_u32 v4, v4, v1
 ; GCN-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GCN-NEXT:    v_add_u32_e32 v0, vcc, v4, v1
+; 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_subrev_u32_e32 v2, vcc, v5, v2
+; 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_subrev_u32_e32 v5, vcc, v3, v2
+; 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_add_u32_e32 v5, vcc, 1, v4
 ; GCN-NEXT:    v_cmp_ge_u32_e32 vcc, v2, v3
@@ -2351,7 +2355,7 @@ define amdgpu_kernel void @fdiv_test_denormals(i8 addrspace(1)* nocapture readon
 ; 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
 ;
@@ -2377,7 +2381,7 @@ define amdgpu_kernel void @fdiv_test_denormals(i8 addrspace(1)* nocapture readon
 ; 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, v1, v0
+; VI-NEXT:    v_add_u32_e32 v0, vcc, v0, v1
 ; VI-NEXT:    buffer_store_byte v0, off, s[0:3], 0
 ; VI-NEXT:    s_endpgm
 ;
@@ -2401,7 +2405,7 @@ define amdgpu_kernel void @fdiv_test_denormals(i8 addrspace(1)* nocapture readon
 ; 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, v3, v2
+; GCN-NEXT:    v_add_u32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    flat_store_byte v[0:1], v2
 ; GCN-NEXT:    s_endpgm
 ;
@@ -2597,7 +2601,7 @@ define i64 @v_test_udiv64_mulhi_fold(i64 %arg) {
 ; 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_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
@@ -2684,7 +2688,7 @@ define i64 @v_test_udiv64_mulhi_fold(i64 %arg) {
 ; 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_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
@@ -2765,32 +2769,34 @@ define i64 @v_test_udiv64_mulhi_fold(i64 %arg) {
 ; GFX1030-NEXT:    s_addc_u32 s5, 0, 0
 ; GFX1030-NEXT:    v_add_co_u32 v2, s4, 0xa9000000, s4
 ; GFX1030-NEXT:    s_cmpk_lg_u32 s4, 0x0
-; GFX1030-NEXT:    s_addc_u32 s4, s5, 0xa7c5
-; GFX1030-NEXT:    v_mul_hi_u32 v3, 0xfffe7960, v2
-; GFX1030-NEXT:    v_mul_lo_u32 v4, 0xfffe7960, v2
-; GFX1030-NEXT:    s_mul_i32 s5, s4, 0xfffe7960
-; GFX1030-NEXT:    v_sub_nc_u32_e32 v3, v3, v2
-; GFX1030-NEXT:    v_mul_hi_u32 v5, v2, v4
-; GFX1030-NEXT:    v_mul_hi_u32 v8, s4, v4
-; GFX1030-NEXT:    v_mul_lo_u32 v4, s4, v4
-; GFX1030-NEXT:    v_add_nc_u32_e32 v3, s5, v3
-; GFX1030-NEXT:    v_mul_lo_u32 v6, v2, v3
-; GFX1030-NEXT:    v_mul_hi_u32 v7, v2, v3
-; GFX1030-NEXT:    v_mul_hi_u32 v9, s4, v3
-; GFX1030-NEXT:    v_mul_lo_u32 v3, s4, v3
-; GFX1030-NEXT:    v_add_co_u32 v5, vcc_lo, v5, v6
-; GFX1030-NEXT:    v_add_co_ci_u32_e32 v6, vcc_lo, 0, v7, vcc_lo
-; GFX1030-NEXT:    v_add_co_u32 v4, vcc_lo, v5, v4
-; GFX1030-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v6, v8, vcc_lo
-; GFX1030-NEXT:    v_add_co_ci_u32_e32 v5, vcc_lo, 0, v9, vcc_lo
-; GFX1030-NEXT:    v_add_co_u32 v3, vcc_lo, v4, v3
-; GFX1030-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, 0, v5, vcc_lo
-; GFX1030-NEXT:    v_add_co_u32 v5, vcc_lo, v2, v3
-; GFX1030-NEXT:    v_add_co_ci_u32_e32 v6, vcc_lo, s4, v4, vcc_lo
-; GFX1030-NEXT:    v_mul_hi_u32 v8, v0, v5
-; GFX1030-NEXT:    v_mad_u64_u32 v[4:5], null, v1, v5, 0
-; GFX1030-NEXT:    v_mad_u64_u32 v[2:3], null, v0, v6, 0
-; GFX1030-NEXT:    v_mad_u64_u32 v[6:7], null, v1, v6, 0
+; GFX1030-NEXT:    s_addc_u32 s5, s5, 0xa7c5
+; GFX1030-NEXT:    v_readfirstlane_b32 s4, v2
+; GFX1030-NEXT:    s_mul_i32 s6, s5, 0xfffe7960
+; GFX1030-NEXT:    s_mul_hi_u32 s7, s4, 0xfffe7960
+; GFX1030-NEXT:    s_mul_i32 s8, s4, 0xfffe7960
+; GFX1030-NEXT:    s_sub_i32 s7, s7, s4
+; GFX1030-NEXT:    s_mul_hi_u32 s9, s4, s8
+; GFX1030-NEXT:    s_add_i32 s7, s7, s6
+; GFX1030-NEXT:    s_mul_hi_u32 s10, s5, s8
+; GFX1030-NEXT:    s_mul_i32 s6, s5, s8
+; GFX1030-NEXT:    s_mul_hi_u32 s8, s4, s7
+; GFX1030-NEXT:    s_mul_i32 s4, s4, s7
+; GFX1030-NEXT:    s_mul_hi_u32 s11, s5, s7
+; GFX1030-NEXT:    s_add_u32 s4, s9, s4
+; GFX1030-NEXT:    s_addc_u32 s8, 0, s8
+; GFX1030-NEXT:    s_add_u32 s4, s4, s6
+; GFX1030-NEXT:    s_mul_i32 s7, s5, s7
+; GFX1030-NEXT:    s_addc_u32 s4, s8, s10
+; GFX1030-NEXT:    s_addc_u32 s6, s11, 0
+; GFX1030-NEXT:    s_add_u32 s4, s4, s7
+; GFX1030-NEXT:    s_addc_u32 s6, 0, s6
+; GFX1030-NEXT:    v_add_co_u32 v4, s4, v2, s4
+; GFX1030-NEXT:    s_cmpk_lg_u32 s4, 0x0
+; GFX1030-NEXT:    s_addc_u32 s4, s5, s6
+; GFX1030-NEXT:    v_mul_hi_u32 v8, v0, v4
+; GFX1030-NEXT:    v_mad_u64_u32 v[2:3], null, v0, s4, 0
+; GFX1030-NEXT:    v_mad_u64_u32 v[4:5], null, v1, v4, 0
+; GFX1030-NEXT:    v_mad_u64_u32 v[6:7], null, v1, s4, 0
 ; GFX1030-NEXT:    v_add_co_u32 v2, vcc_lo, v8, v2
 ; GFX1030-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
 ; GFX1030-NEXT:    v_add_co_u32 v2, vcc_lo, v2, v4

diff --git a/llvm/test/CodeGen/AMDGPU/udiv64.ll b/llvm/test/CodeGen/AMDGPU/udiv64.ll
index 3186ad61d7c48eea8be03739345a8cbc9ed13c00..e1dc9904b812352cbb85a0e28497a4cf3d1f8a29 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/udiv64.ll
+++ b/llvm/test/CodeGen/AMDGPU/udiv64.ll
@@ -26,8 +26,8 @@ define amdgpu_kernel void @s_test_udiv_i64(i64 addrspace(1)* %out, i64 %x, i64 %
 ; GCN-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s5, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s4, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -86,9 +86,9 @@ define amdgpu_kernel void @s_test_udiv_i64(i64 addrspace(1)* %out, i64 %x, i64 %
 ; GCN-NEXT:    v_mul_hi_u32 v3, s8, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s9, v0
 ; GCN-NEXT:    v_mov_b32_e32 v5, s9
-; 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, s8, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v4
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
 ; GCN-NEXT:    v_sub_i32_e32 v4, vcc, s3, v2
 ; GCN-NEXT:    v_sub_i32_e32 v3, vcc, s2, v3
 ; GCN-NEXT:    v_subb_u32_e64 v4, s[0:1], v4, v5, vcc
@@ -704,8 +704,8 @@ define amdgpu_kernel void @s_test_udiv24_i48(i48 addrspace(1)* %out, i48 %x, i48
 ; GCN-NEXT:    v_mul_hi_u32 v4, s0, v1
 ; GCN-NEXT:    v_mul_lo_u32 v5, s1, v1
 ; GCN-NEXT:    v_mul_lo_u32 v6, s0, v1
-; GCN-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
-; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
+; 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
@@ -725,9 +725,9 @@ define amdgpu_kernel void @s_test_udiv24_i48(i48 addrspace(1)* %out, i48 %x, i48
 ; GCN-NEXT:    v_mul_lo_u32 v3, s0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v4, s0, v1
 ; GCN-NEXT:    v_mul_lo_u32 v5, s1, v1
-; GCN-NEXT:    v_add_i32_e32 v3, vcc, v4, v3
+; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v4
 ; GCN-NEXT:    v_mul_lo_u32 v4, s0, v1
-; GCN-NEXT:    v_add_i32_e32 v3, vcc, v5, v3
+; GCN-NEXT:    v_add_i32_e32 v3, vcc, v3, v5
 ; 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
@@ -908,8 +908,8 @@ define amdgpu_kernel void @s_test_udiv_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_mul_hi_u32 v3, s4, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s5, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s4, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -958,7 +958,7 @@ define amdgpu_kernel void @s_test_udiv_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_addc_u32_e32 v0, vcc, 0, v1, vcc
 ; GCN-NEXT:    v_mul_lo_u32 v1, s3, v0
 ; GCN-NEXT:    v_mul_hi_u32 v2, s2, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
 ; GCN-NEXT:    v_mul_lo_u32 v2, s2, v0
 ; GCN-NEXT:    v_sub_i32_e32 v3, vcc, 0, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, 24, v2
@@ -1356,8 +1356,8 @@ define amdgpu_kernel void @s_test_udiv_k_den_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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_subrev_i32_e32 v2, vcc, v0, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; 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
@@ -1378,8 +1378,8 @@ define amdgpu_kernel void @s_test_udiv_k_den_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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_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
@@ -1531,8 +1531,8 @@ define i64 @v_test_udiv_k_den_i64(i64 %x) {
 ; GCN-NEXT:    v_mul_hi_u32 v4, v2, s4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v3, s4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v2, s4
-; GCN-NEXT:    v_subrev_i32_e32 v4, vcc, v2, v4
-; GCN-NEXT:    v_add_i32_e32 v4, vcc, v6, v4
+; GCN-NEXT:    v_sub_i32_e32 v4, vcc, v4, v2
+; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v6
 ; GCN-NEXT:    v_mul_hi_u32 v7, v2, v5
 ; GCN-NEXT:    v_mul_lo_u32 v6, v2, v4
 ; GCN-NEXT:    v_mul_hi_u32 v8, v2, v4
@@ -1552,7 +1552,7 @@ define i64 @v_test_udiv_k_den_i64(i64 %x) {
 ; GCN-NEXT:    v_mul_hi_u32 v4, v2, s4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v3, s4
 ; GCN-NEXT:    v_mul_lo_u32 v6, v2, s4
-; GCN-NEXT:    v_subrev_i32_e32 v4, vcc, v2, v4
+; GCN-NEXT:    v_sub_i32_e32 v4, vcc, v4, v2
 ; GCN-NEXT:    v_add_i32_e32 v4, vcc, v4, v5
 ; GCN-NEXT:    v_mul_lo_u32 v5, v2, v4
 ; GCN-NEXT:    v_mul_hi_u32 v7, v2, v6

diff --git a/llvm/test/CodeGen/AMDGPU/udivrem.ll b/llvm/test/CodeGen/AMDGPU/udivrem.ll
index 86ad4598131fdbb73b3d9c1efbde7f4564f827e9..58c1445b9f303b1eb0b7ac718b280e51e58da535 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/udivrem.ll
+++ b/llvm/test/CodeGen/AMDGPU/udivrem.ll
@@ -52,7 +52,7 @@ define amdgpu_kernel void @test_udivrem(i32 addrspace(1)* %out0, [8 x i32], i32
 ; GFX6-NEXT:    v_mul_lo_u32 v1, s3, v0
 ; GFX6-NEXT:    s_load_dword s3, s[0:1], 0x1d
 ; GFX6-NEXT:    v_mul_hi_u32 v1, v0, v1
-; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v1, v0
+; 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
@@ -86,7 +86,7 @@ define amdgpu_kernel void @test_udivrem(i32 addrspace(1)* %out0, [8 x i32], i32
 ; GFX8-NEXT:    v_mul_lo_u32 v1, s2, v0
 ; GFX8-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x24
 ; GFX8-NEXT:    v_mul_hi_u32 v1, v0, v1
-; 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 v2, s7, v0
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v0, s2
@@ -176,7 +176,7 @@ define amdgpu_kernel void @test_udivrem_v2(<2 x i32> addrspace(1)* %out, <2 x i3
 ; 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, v3, v1
+; 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
@@ -204,41 +204,45 @@ define amdgpu_kernel void @test_udivrem_v2(<2 x i32> addrspace(1)* %out, <2 x i3
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_cvt_f32_u32_e32 v0, s6
-; GFX8-NEXT:    v_cvt_f32_u32_e32 v1, s7
 ; GFX8-NEXT:    s_sub_i32 s2, 0, s6
+; GFX8-NEXT:    v_cvt_f32_u32_e32 v2, s7
 ; GFX8-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX8-NEXT:    v_rcp_iflag_f32_e32 v1, v1
+; GFX8-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX8-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX8-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX8-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX8-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX8-NEXT:    v_mul_lo_u32 v2, s2, v0
-; GFX8-NEXT:    s_sub_i32 s2, 0, s7
-; GFX8-NEXT:    v_mul_lo_u32 v3, s2, v1
-; GFX8-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX8-NEXT:    v_mul_hi_u32 v3, v1, v3
-; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v2, v0
-; GFX8-NEXT:    v_add_u32_e32 v1, vcc, v3, v1
+; GFX8-NEXT:    v_mul_lo_u32 v1, s2, v0
+; GFX8-NEXT:    v_mul_hi_u32 v1, v0, v1
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
 ; GFX8-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX8-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX8-NEXT:    v_mul_lo_u32 v0, v0, s6
-; GFX8-NEXT:    v_mul_lo_u32 v1, v1, s7
-; GFX8-NEXT:    v_sub_u32_e32 v0, vcc, s4, v0
-; GFX8-NEXT:    v_sub_u32_e32 v1, vcc, s5, v1
-; GFX8-NEXT:    v_subrev_u32_e32 v2, vcc, s6, v0
-; GFX8-NEXT:    v_subrev_u32_e32 v3, vcc, s7, v1
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v3, vcc
-; GFX8-NEXT:    v_subrev_u32_e32 v2, vcc, s6, v0
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s6, v0
-; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v2, vcc
-; GFX8-NEXT:    v_subrev_u32_e32 v2, vcc, s7, v1
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s7, v1
-; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v2, vcc
+; GFX8-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX8-NEXT:    v_cvt_u32_f32_e32 v1, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX8-NEXT:    s_mul_i32 s2, s2, s6
+; GFX8-NEXT:    s_sub_i32 s2, s4, 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, 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_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
+; GFX8-NEXT:    s_sub_i32 s3, s2, s7
+; GFX8-NEXT:    s_cmp_ge_u32 s2, s7
+; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s2
 ; GFX8-NEXT:    flat_store_dwordx2 v[2:3], v[0:1]
 ; GFX8-NEXT:    s_endpgm
   %result0 = udiv <2 x i32> %x, %y
@@ -341,7 +345,7 @@ define amdgpu_kernel void @test_udivrem_v4(<4 x i32> addrspace(1)* %out, <4 x i3
 ; GFX6-NEXT:    v_mul_hi_u32 v4, v1, v4
 ; GFX6-NEXT:    v_add_i32_e32 v0, vcc, v2, v0
 ; GFX6-NEXT:    v_mul_hi_u32 v0, s4, v0
-; GFX6-NEXT:    v_add_i32_e32 v1, vcc, v4, v1
+; 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
@@ -374,7 +378,7 @@ define amdgpu_kernel void @test_udivrem_v4(<4 x i32> addrspace(1)* %out, <4 x i3
 ; 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, v4, v3
+; 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
@@ -399,75 +403,83 @@ define amdgpu_kernel void @test_udivrem_v4(<4 x i32> addrspace(1)* %out, <4 x i3
 ; GFX8-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x24
 ; GFX8-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX8-NEXT:    v_cvt_f32_u32_e32 v0, s8
-; GFX8-NEXT:    v_cvt_f32_u32_e32 v1, s9
 ; GFX8-NEXT:    s_sub_i32 s2, 0, s8
-; GFX8-NEXT:    s_sub_i32 s3, 0, s9
+; GFX8-NEXT:    v_cvt_f32_u32_e32 v2, s9
+; GFX8-NEXT:    v_mov_b32_e32 v5, s1
 ; GFX8-NEXT:    v_rcp_iflag_f32_e32 v0, v0
-; GFX8-NEXT:    v_rcp_iflag_f32_e32 v1, v1
-; GFX8-NEXT:    v_cvt_f32_u32_e32 v3, s10
-; GFX8-NEXT:    v_cvt_f32_u32_e32 v5, s11
+; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    v_rcp_iflag_f32_e32 v2, v2
 ; GFX8-NEXT:    v_mul_f32_e32 v0, 0x4f7ffffe, v0
 ; GFX8-NEXT:    v_cvt_u32_f32_e32 v0, v0
-; GFX8-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v1
-; GFX8-NEXT:    v_cvt_u32_f32_e32 v1, v1
-; GFX8-NEXT:    v_rcp_iflag_f32_e32 v3, v3
-; GFX8-NEXT:    v_mul_lo_u32 v2, s2, v0
-; GFX8-NEXT:    s_sub_i32 s2, 0, s10
-; GFX8-NEXT:    v_mul_lo_u32 v4, s3, v1
-; GFX8-NEXT:    v_mul_hi_u32 v2, v0, v2
-; GFX8-NEXT:    v_mul_hi_u32 v4, v1, v4
-; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v2, v0
+; GFX8-NEXT:    v_mul_lo_u32 v1, s2, v0
+; 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 v0, s4, v0
-; GFX8-NEXT:    v_add_u32_e32 v1, vcc, v4, v1
-; GFX8-NEXT:    v_mul_hi_u32 v1, s5, v1
-; GFX8-NEXT:    v_mul_lo_u32 v0, v0, s8
-; GFX8-NEXT:    v_mul_f32_e32 v2, 0x4f7ffffe, v3
-; GFX8-NEXT:    v_cvt_u32_f32_e32 v2, v2
-; GFX8-NEXT:    v_mul_lo_u32 v1, v1, s9
-; GFX8-NEXT:    v_sub_u32_e32 v0, vcc, s4, v0
-; GFX8-NEXT:    v_subrev_u32_e32 v3, vcc, s8, v0
-; GFX8-NEXT:    v_sub_u32_e32 v1, vcc, s5, v1
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX8-NEXT:    v_subrev_u32_e32 v3, vcc, s8, v0
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s8, v0
-; GFX8-NEXT:    v_cndmask_b32_e32 v0, v0, v3, vcc
-; GFX8-NEXT:    v_mul_lo_u32 v3, s2, v2
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s9, v1
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX8-NEXT:    v_mul_hi_u32 v3, v2, v3
-; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX8-NEXT:    v_rcp_iflag_f32_e32 v4, v5
-; GFX8-NEXT:    s_sub_i32 s2, 0, s11
-; GFX8-NEXT:    v_add_u32_e32 v2, vcc, v3, v2
-; GFX8-NEXT:    v_mul_f32_e32 v3, 0x4f7ffffe, v4
-; GFX8-NEXT:    v_cvt_u32_f32_e32 v3, v3
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s9, v1
-; GFX8-NEXT:    v_mul_hi_u32 v2, s6, v2
-; GFX8-NEXT:    v_mul_lo_u32 v5, s2, v3
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s9, v1
-; GFX8-NEXT:    v_cndmask_b32_e32 v1, v1, v4, vcc
-; GFX8-NEXT:    v_mul_lo_u32 v2, v2, s10
-; GFX8-NEXT:    v_mul_hi_u32 v4, v3, v5
-; GFX8-NEXT:    v_sub_u32_e32 v2, vcc, s6, v2
-; GFX8-NEXT:    v_add_u32_e32 v3, vcc, v4, v3
-; GFX8-NEXT:    v_mul_hi_u32 v3, s7, v3
-; GFX8-NEXT:    v_subrev_u32_e32 v5, vcc, s10, v2
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX8-NEXT:    v_mul_lo_u32 v3, v3, s11
-; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v5, vcc
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s10, v2
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s10, v2
-; GFX8-NEXT:    v_cndmask_b32_e32 v2, v2, v4, vcc
-; GFX8-NEXT:    v_sub_u32_e32 v3, vcc, s7, v3
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s11, v3
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX8-NEXT:    v_subrev_u32_e32 v4, vcc, s11, v3
-; GFX8-NEXT:    v_cmp_le_u32_e32 vcc, s11, v3
-; GFX8-NEXT:    v_cndmask_b32_e32 v3, v3, v4, vcc
-; GFX8-NEXT:    v_mov_b32_e32 v5, s1
-; GFX8-NEXT:    v_mov_b32_e32 v4, s0
+; GFX8-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX8-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX8-NEXT:    v_cvt_f32_u32_e32 v2, s10
+; GFX8-NEXT:    v_readfirstlane_b32 s2, v0
+; GFX8-NEXT:    s_mul_i32 s2, s2, s8
+; GFX8-NEXT:    s_sub_i32 s2, s4, s2
+; GFX8-NEXT:    s_sub_i32 s3, s2, s8
+; GFX8-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX8-NEXT:    s_sub_i32 s3, s2, s8
+; GFX8-NEXT:    s_cmp_ge_u32 s2, s8
+; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX8-NEXT:    s_sub_i32 s3, 0, s9
+; GFX8-NEXT:    v_mul_lo_u32 v0, s3, 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, v0, v1
+; GFX8-NEXT:    v_mul_hi_u32 v0, s5, v0
+; GFX8-NEXT:    v_mul_f32_e32 v1, 0x4f7ffffe, v2
+; GFX8-NEXT:    v_cvt_u32_f32_e32 v1, v1
+; GFX8-NEXT:    v_cvt_f32_u32_e32 v2, s11
+; GFX8-NEXT:    v_readfirstlane_b32 s3, v0
+; GFX8-NEXT:    s_mul_i32 s3, s3, s9
+; GFX8-NEXT:    s_sub_i32 s3, s5, 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, 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_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
+; GFX8-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX8-NEXT:    s_sub_i32 s5, s4, s10
+; GFX8-NEXT:    s_cmp_ge_u32 s4, s10
+; GFX8-NEXT:    s_cselect_b32 s4, s5, s4
+; GFX8-NEXT:    s_sub_i32 s5, 0, s11
+; GFX8-NEXT:    v_mul_lo_u32 v0, s5, v1
+; GFX8-NEXT:    v_mov_b32_e32 v2, s4
+; GFX8-NEXT:    v_mul_hi_u32 v0, v1, v0
+; GFX8-NEXT:    v_add_u32_e32 v0, vcc, v1, v0
+; GFX8-NEXT:    v_mul_hi_u32 v3, s7, v0
+; GFX8-NEXT:    v_mov_b32_e32 v0, s2
+; GFX8-NEXT:    v_mov_b32_e32 v1, s3
+; GFX8-NEXT:    v_readfirstlane_b32 s2, v3
+; GFX8-NEXT:    s_mul_i32 s2, s2, s11
+; GFX8-NEXT:    s_sub_i32 s2, s7, s2
+; GFX8-NEXT:    s_sub_i32 s3, s2, s11
+; GFX8-NEXT:    s_cmp_ge_u32 s2, s11
+; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX8-NEXT:    s_sub_i32 s3, s2, s11
+; GFX8-NEXT:    s_cmp_ge_u32 s2, s11
+; GFX8-NEXT:    s_cselect_b32 s2, s3, s2
+; GFX8-NEXT:    v_mov_b32_e32 v3, s2
 ; GFX8-NEXT:    flat_store_dwordx4 v[4:5], v[0:3]
 ; GFX8-NEXT:    s_endpgm
   %result0 = udiv <4 x i32> %x, %y

diff --git a/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll b/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll
index f494922139d9af8195e312bb4cbfb713d16ea443..006f7cbef149770e1fd8f95f967410c892f05a58 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll
+++ b/llvm/test/CodeGen/AMDGPU/unstructured-cfg-def-use-issue.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: llc -mtriple=amdgcn-amdhsa -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefix=GCN %s
 ; RUN: opt -S -si-annotate-control-flow -mtriple=amdgcn-amdhsa -verify-machineinstrs -simplifycfg-require-and-preserve-domtree=1 < %s | FileCheck -check-prefix=SI-OPT %s
@@ -19,10 +20,11 @@ define hidden void @widget() {
 ; GCN-NEXT:    flat_load_dword v0, v[0:1]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_cmp_gt_i32_e32 vcc, 21, v0
+; GCN-NEXT:    v_readfirstlane_b32 s15, v0
 ; GCN-NEXT:    s_cbranch_vccz .LBB0_3
 ; GCN-NEXT:  ; %bb.1: ; %bb4
-; GCN-NEXT:    v_cmp_ne_u32_e32 vcc, 9, v0
-; GCN-NEXT:    s_cbranch_vccnz .LBB0_4
+; GCN-NEXT:    s_cmp_lg_u32 s15, 9
+; GCN-NEXT:    s_cbranch_scc1 .LBB0_4
 ; GCN-NEXT:  ; %bb.2: ; %bb7
 ; GCN-NEXT:    s_getpc_b64 s[16:17]
 ; GCN-NEXT:    s_add_u32 s16, s16, wibble@rel32@lo+4
@@ -30,8 +32,8 @@ define hidden void @widget() {
 ; GCN-NEXT:    s_swappc_b64 s[30:31], s[16:17]
 ; GCN-NEXT:    s_branch .LBB0_7
 ; GCN-NEXT:  .LBB0_3: ; %bb2
-; GCN-NEXT:    v_cmp_eq_u32_e32 vcc, 21, v0
-; GCN-NEXT:    s_cbranch_vccnz .LBB0_6
+; GCN-NEXT:    s_cmp_eq_u32 s15, 21
+; GCN-NEXT:    s_cbranch_scc1 .LBB0_6
 ; GCN-NEXT:  .LBB0_4: ; %bb9
 ; GCN-NEXT:    s_getpc_b64 s[16:17]
 ; GCN-NEXT:    s_add_u32 s16, s16, wibble@rel32@lo+4

diff --git a/llvm/test/CodeGen/AMDGPU/urem.ll b/llvm/test/CodeGen/AMDGPU/urem.ll
index c6e0d8383eae2323546ed1492e7d64b240ac665d..40731ce14e83a7375f04d6129ea167cfe8a367ec 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/urem.ll
+++ b/llvm/test/CodeGen/AMDGPU/urem.ll
@@ -21,9 +21,9 @@ define amdgpu_kernel void @test_urem_i32(i32 addrspace(1)* %out, i32 addrspace(1
 ; FUNC-LABEL: {{^}}test_urem_i32_7:
 ; SI: s_mov_b32 [[MAGIC:s[0-9]+]], 0x24924925
 ; SI: v_mul_hi_u32 {{v[0-9]+}}, {{v[0-9]+}}, [[MAGIC]]
-; SI: v_subrev_{{[iu]}}32
-; SI: v_mul_lo_u32
 ; SI: v_sub_{{[iu]}}32
+; SI: v_mul_lo_u32
+; SI: v_subrev_{{[iu]}}32
 ; SI: buffer_store_dword
 ; SI: s_endpgm
 define amdgpu_kernel void @test_urem_i32_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {

diff --git a/llvm/test/CodeGen/AMDGPU/urem64.ll b/llvm/test/CodeGen/AMDGPU/urem64.ll
index 1821045bf6963db2707f9b0accfc07f34315ae16..399652e6697ddb1aa232e77fc8b71b7fefb4709b 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/urem64.ll
+++ b/llvm/test/CodeGen/AMDGPU/urem64.ll
@@ -28,8 +28,8 @@ define amdgpu_kernel void @s_test_urem_i64(i64 addrspace(1)* %out, i64 %x, i64 %
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s1, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -86,8 +86,8 @@ define amdgpu_kernel void @s_test_urem_i64(i64 addrspace(1)* %out, i64 %x, i64 %
 ; GCN-NEXT:    v_mul_hi_u32 v2, s12, v0
 ; GCN-NEXT:    v_mul_lo_u32 v3, s13, v0
 ; GCN-NEXT:    v_mul_lo_u32 v0, s12, v0
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v2, v1
-; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v3, v1
 ; GCN-NEXT:    v_sub_i32_e32 v2, vcc, s11, v1
 ; GCN-NEXT:    v_mov_b32_e32 v3, s13
 ; GCN-NEXT:    v_sub_i32_e32 v0, vcc, s10, v0
@@ -202,8 +202,8 @@ define amdgpu_kernel void @s_test_urem_i64(i64 addrspace(1)* %out, i64 %x, i64 %
 ; 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, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; 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:    s_mov_b32 s10, -1
@@ -733,8 +733,8 @@ define amdgpu_kernel void @s_test_urem_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; GCN-NEXT:    v_mul_hi_u32 v3, s0, v0
 ; GCN-NEXT:    v_mul_lo_u32 v5, s1, v0
 ; GCN-NEXT:    v_mul_lo_u32 v4, s0, v0
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v3, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v5
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
+; GCN-NEXT:    v_add_i32_e32 v2, vcc, v5, v2
 ; GCN-NEXT:    v_mul_hi_u32 v3, v0, v4
 ; GCN-NEXT:    v_mul_lo_u32 v5, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v7, v0, v2
@@ -782,7 +782,7 @@ define amdgpu_kernel void @s_test_urem_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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, v2, v1
+; GCN-NEXT:    v_add_i32_e32 v1, vcc, v1, v2
 ; 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
@@ -884,8 +884,8 @@ define amdgpu_kernel void @s_test_urem_k_num_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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, v2, v1
-; GCN-IR-NEXT:    v_add_i32_e32 v1, vcc, v1, v3
+; 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
@@ -919,8 +919,8 @@ define amdgpu_kernel void @s_test_urem_k_den_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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_subrev_i32_e32 v2, vcc, v0, v2
-; GCN-NEXT:    v_add_i32_e32 v2, vcc, v4, v2
+; 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
@@ -941,7 +941,7 @@ define amdgpu_kernel void @s_test_urem_k_den_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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_sub_i32_e32 v2, vcc, v2, v0
 ; GCN-NEXT:    v_add_i32_e32 v2, vcc, v2, v3
 ; GCN-NEXT:    v_mul_lo_u32 v3, v0, v2
 ; GCN-NEXT:    v_mul_hi_u32 v5, v0, v4
@@ -1071,7 +1071,7 @@ define amdgpu_kernel void @s_test_urem_k_den_i64(i64 addrspace(1)* %out, i64 %x)
 ; 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:    v_add_i32_e32 v1, vcc, v1, v2
 ; GCN-IR-NEXT:    v_mov_b32_e32 v2, s3
 ; GCN-IR-NEXT:    v_sub_i32_e32 v0, vcc, s2, v0
 ; GCN-IR-NEXT:    s_mov_b32 s4, s0

diff --git a/llvm/test/CodeGen/AMDGPU/usubo.ll b/llvm/test/CodeGen/AMDGPU/usubo.ll
index f7eb8570090f4aae39407e42184b57ba871c236f..bc631d8ca7626bdca2a8c1685f0a8e4e1e25b709 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/usubo.ll
+++ b/llvm/test/CodeGen/AMDGPU/usubo.ll
@@ -135,7 +135,7 @@ define amdgpu_kernel void @v_usubo_i64(i64 addrspace(1)* %out, i1 addrspace(1)*
 }
 
 ; FUNC-LABEL: {{^}}v_usubo_i16:
-; SI: v_subrev_i32_e32
+; SI: v_sub_i32_e32
 ; SI: v_and_b32
 ; SI: v_cmp_ne_u32_e32
 

diff --git a/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll b/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
index f7285051247cba20ff16cf9d543430e01fdcdd23..e8c847223947bdfb75fed54b5c032bd0b95d0a4f 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
+++ b/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
@@ -234,7 +234,7 @@ define amdgpu_ps float @loop(i32 %z, float %v, i32 inreg %bound, float(float)* %
   ; SI-NEXT: bb.1.Flow:
   ; SI-NEXT:   successors: %bb.2(0x40000000), %bb.10(0x40000000)
   ; SI-NEXT: {{  $}}
-  ; SI-NEXT:   [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %29:vgpr_32, %bb.0, %4, %bb.9
+  ; SI-NEXT:   [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %43:vgpr_32, %bb.0, %4, %bb.9
   ; SI-NEXT:   [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[COPY4]], %bb.0, undef %45:vgpr_32, %bb.9
   ; SI-NEXT:   [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[COPY3]], %bb.0, undef %47:vgpr_32, %bb.9
   ; SI-NEXT:   [[PHI3:%[0-9]+]]:vgpr_32 = PHI [[COPY2]], %bb.0, undef %49:vgpr_32, %bb.9
@@ -357,7 +357,7 @@ define amdgpu_ps float @loop_with_use(i32 %z, float %v, i32 inreg %bound, float(
   ; SI-NEXT: bb.1.Flow:
   ; SI-NEXT:   successors: %bb.2(0x40000000), %bb.10(0x40000000)
   ; SI-NEXT: {{  $}}
-  ; SI-NEXT:   [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %30:vgpr_32, %bb.0, %4, %bb.9
+  ; SI-NEXT:   [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %44:vgpr_32, %bb.0, %4, %bb.9
   ; SI-NEXT:   [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[COPY3]], %bb.0, undef %46:vgpr_32, %bb.9
   ; SI-NEXT:   [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[COPY2]], %bb.0, undef %48:vgpr_32, %bb.9
   ; SI-NEXT:   [[SI_ELSE:%[0-9]+]]:sreg_32 = SI_ELSE killed [[SI_IF]], %bb.10, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
@@ -510,7 +510,7 @@ define amdgpu_kernel void @livevariables_update_missed_block(i8 addrspace(1)* %s
   ; SI-NEXT: bb.6.sw.bb18:
   ; SI-NEXT:   successors: %bb.5(0x80000000)
   ; SI-NEXT: {{  $}}
-  ; SI-NEXT:   [[PHI1:%[0-9]+]]:vgpr_32 = PHI undef %39:vgpr_32, %bb.3, [[GLOBAL_LOAD_UBYTE1]], %bb.4
+  ; SI-NEXT:   [[PHI1:%[0-9]+]]:vgpr_32 = PHI undef %37:vgpr_32, %bb.3, [[GLOBAL_LOAD_UBYTE1]], %bb.4
   ; SI-NEXT:   [[V_MOV_B2:%[0-9]+]]:vreg_64 = V_MOV_B64_PSEUDO 0, implicit $exec
   ; SI-NEXT:   GLOBAL_STORE_BYTE killed [[V_MOV_B2]], killed [[PHI1]], 0, 0, implicit $exec :: (store (s8) into `i8 addrspace(1)* null`, addrspace 1)
   ; SI-NEXT:   S_BRANCH %bb.5
@@ -596,27 +596,17 @@ define protected amdgpu_kernel void @nested_waterfalls(%tex* addrspace(1)* %tex.
   ; SI-NEXT: bb.2:
   ; SI-NEXT:   successors: %bb.3(0x80000000)
   ; SI-NEXT: {{  $}}
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub0, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub1, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub0, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub1, implicit $exec
   ; SI-NEXT:   [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1
-  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE3]], [[REG_SEQUENCE2]].sub0_sub1, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub2, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub3, implicit $exec
+  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE3]], [[GLOBAL_LOAD_DWORDX4_2]].sub0_sub1, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub2, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub3, implicit $exec
   ; SI-NEXT:   [[REG_SEQUENCE4:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1
-  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE4]], [[REG_SEQUENCE2]].sub2_sub3, implicit $exec
+  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE4]], [[GLOBAL_LOAD_DWORDX4_2]].sub2_sub3, implicit $exec
   ; SI-NEXT:   [[S_AND_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[V_CMP_EQ_U64_e64_]], killed [[V_CMP_EQ_U64_e64_1]], implicit-def dead $scc
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub4, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub5, implicit $exec
-  ; SI-NEXT:   [[REG_SEQUENCE5:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_4]], %subreg.sub0, [[V_READFIRSTLANE_B32_5]], %subreg.sub1
-  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE5]], [[REG_SEQUENCE2]].sub4_sub5, implicit $exec
-  ; SI-NEXT:   [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[S_AND_B32_]], killed [[V_CMP_EQ_U64_e64_2]], implicit-def dead $scc
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub6, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub7, implicit $exec
-  ; SI-NEXT:   [[REG_SEQUENCE6:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_6]], %subreg.sub0, [[V_READFIRSTLANE_B32_7]], %subreg.sub1
-  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE6]], [[REG_SEQUENCE2]].sub6_sub7, implicit $exec
-  ; SI-NEXT:   [[S_AND_B32_2:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[S_AND_B32_1]], killed [[V_CMP_EQ_U64_e64_3]], implicit-def dead $scc
-  ; SI-NEXT:   [[REG_SEQUENCE7:%[0-9]+]]:sgpr_256 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_1]], %subreg.sub1, killed [[V_READFIRSTLANE_B32_2]], %subreg.sub2, killed [[V_READFIRSTLANE_B32_3]], %subreg.sub3, killed [[V_READFIRSTLANE_B32_4]], %subreg.sub4, killed [[V_READFIRSTLANE_B32_5]], %subreg.sub5, killed [[V_READFIRSTLANE_B32_6]], %subreg.sub6, killed [[V_READFIRSTLANE_B32_7]], %subreg.sub7
-  ; SI-NEXT:   [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_2]], implicit-def $exec, implicit-def dead $scc, implicit $exec
+  ; SI-NEXT:   [[REG_SEQUENCE5:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_1]], %subreg.sub1, killed [[V_READFIRSTLANE_B32_2]], %subreg.sub2, killed [[V_READFIRSTLANE_B32_3]], %subreg.sub3
+  ; SI-NEXT:   [[S_AND_SAVEEXEC_B32_:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_]], implicit-def $exec, implicit-def dead $scc, implicit $exec
   ; SI-NEXT: {{  $}}
   ; SI-NEXT: bb.3:
   ; SI-NEXT:   successors: %bb.4(0x80000000)
@@ -626,22 +616,32 @@ define protected amdgpu_kernel void @nested_waterfalls(%tex* addrspace(1)* %tex.
   ; SI-NEXT: bb.4:
   ; SI-NEXT:   successors: %bb.5(0x80000000)
   ; SI-NEXT: {{  $}}
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_8:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub0, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_9:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub1, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub0, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_5:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub1, implicit $exec
+  ; SI-NEXT:   [[REG_SEQUENCE6:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_4]], %subreg.sub0, [[V_READFIRSTLANE_B32_5]], %subreg.sub1
+  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_2:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE6]], [[REG_SEQUENCE2]].sub0_sub1, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_6:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub2, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_7:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub3, implicit $exec
+  ; SI-NEXT:   [[REG_SEQUENCE7:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_6]], %subreg.sub0, [[V_READFIRSTLANE_B32_7]], %subreg.sub1
+  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE7]], [[REG_SEQUENCE2]].sub2_sub3, implicit $exec
+  ; SI-NEXT:   [[S_AND_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[V_CMP_EQ_U64_e64_2]], killed [[V_CMP_EQ_U64_e64_3]], implicit-def dead $scc
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_8:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub4, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_9:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub5, implicit $exec
   ; SI-NEXT:   [[REG_SEQUENCE8:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_8]], %subreg.sub0, [[V_READFIRSTLANE_B32_9]], %subreg.sub1
-  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_4:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE8]], [[GLOBAL_LOAD_DWORDX4_2]].sub0_sub1, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_10:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub2, implicit $exec
-  ; SI-NEXT:   [[V_READFIRSTLANE_B32_11:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[GLOBAL_LOAD_DWORDX4_2]].sub3, implicit $exec
+  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_4:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE8]], [[REG_SEQUENCE2]].sub4_sub5, implicit $exec
+  ; SI-NEXT:   [[S_AND_B32_2:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[S_AND_B32_1]], killed [[V_CMP_EQ_U64_e64_4]], implicit-def dead $scc
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_10:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub6, implicit $exec
+  ; SI-NEXT:   [[V_READFIRSTLANE_B32_11:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[REG_SEQUENCE2]].sub7, implicit $exec
   ; SI-NEXT:   [[REG_SEQUENCE9:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[V_READFIRSTLANE_B32_10]], %subreg.sub0, [[V_READFIRSTLANE_B32_11]], %subreg.sub1
-  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_5:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE9]], [[GLOBAL_LOAD_DWORDX4_2]].sub2_sub3, implicit $exec
-  ; SI-NEXT:   [[S_AND_B32_3:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[V_CMP_EQ_U64_e64_4]], killed [[V_CMP_EQ_U64_e64_5]], implicit-def dead $scc
-  ; SI-NEXT:   [[REG_SEQUENCE10:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_8]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_9]], %subreg.sub1, killed [[V_READFIRSTLANE_B32_10]], %subreg.sub2, killed [[V_READFIRSTLANE_B32_11]], %subreg.sub3
+  ; SI-NEXT:   [[V_CMP_EQ_U64_e64_5:%[0-9]+]]:sreg_32_xm0_xexec = V_CMP_EQ_U64_e64 killed [[REG_SEQUENCE9]], [[REG_SEQUENCE2]].sub6_sub7, implicit $exec
+  ; SI-NEXT:   [[S_AND_B32_3:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_B32 killed [[S_AND_B32_2]], killed [[V_CMP_EQ_U64_e64_5]], implicit-def dead $scc
+  ; SI-NEXT:   [[REG_SEQUENCE10:%[0-9]+]]:sgpr_256 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_4]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_5]], %subreg.sub1, killed [[V_READFIRSTLANE_B32_6]], %subreg.sub2, killed [[V_READFIRSTLANE_B32_7]], %subreg.sub3, killed [[V_READFIRSTLANE_B32_8]], %subreg.sub4, killed [[V_READFIRSTLANE_B32_9]], %subreg.sub5, killed [[V_READFIRSTLANE_B32_10]], %subreg.sub6, killed [[V_READFIRSTLANE_B32_11]], %subreg.sub7
   ; SI-NEXT:   [[S_AND_SAVEEXEC_B32_1:%[0-9]+]]:sreg_32_xm0_xexec = S_AND_SAVEEXEC_B32 killed [[S_AND_B32_3]], implicit-def $exec, implicit-def dead $scc, implicit $exec
   ; SI-NEXT: {{  $}}
   ; SI-NEXT: bb.5:
   ; SI-NEXT:   successors: %bb.4(0x40000000), %bb.6(0x40000000)
   ; SI-NEXT: {{  $}}
-  ; SI-NEXT:   [[IMAGE_SAMPLE_V1_V2_gfx10_:%[0-9]+]]:vgpr_32 = IMAGE_SAMPLE_V1_V2_gfx10 undef %27:vreg_64, [[REG_SEQUENCE7]], killed [[REG_SEQUENCE10]], 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32) from custom "ImageResource")
+  ; SI-NEXT:   [[IMAGE_SAMPLE_V1_V2_gfx10_:%[0-9]+]]:vgpr_32 = IMAGE_SAMPLE_V1_V2_gfx10 undef %27:vreg_64, killed [[REG_SEQUENCE10]], [[REG_SEQUENCE5]], 1, 1, 0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load (s32) from custom "ImageResource")
   ; SI-NEXT:   $exec_lo = S_XOR_B32_term $exec_lo, killed [[S_AND_SAVEEXEC_B32_1]], implicit-def dead $scc
   ; SI-NEXT:   SI_WATERFALL_LOOP %bb.4, implicit $exec
   ; SI-NEXT: {{  $}}

diff --git a/llvm/test/CodeGen/AMDGPU/wave32.ll b/llvm/test/CodeGen/AMDGPU/wave32.ll
index ec91016353a64b33b026d48f0b06f2556b2b2ff9..9ad95bb0ab6f240ea7a9486c479f4c3a15eac6c0 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/wave32.ll
+++ b/llvm/test/CodeGen/AMDGPU/wave32.ll
@@ -330,24 +330,15 @@ bb:
 }
 
 ; GCN-LABEL: {{^}}test_udiv64:
-; GFX1032: v_add_co_u32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
-; GFX1032: v_add_co_u32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}, vcc_lo
-; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
-; GFX1032: v_add_co_u32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0, v{{[0-9]+}}, vcc_lo
-; GFX1032: v_add_co_u32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1032: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, v{{[0-9]+}}, v{{[0-9]+}}, vcc_lo
-; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
-; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}, vcc{{$}}
-; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
-; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, 0, v{{[0-9]+}}, vcc{{$}}
-; GFX1064: v_add_co_u32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}
-; GFX1064: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, v{{[0-9]+}}, vcc{{$}}
+; GCN: s_add_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN: s_addc_u32 s{{[0-9]+}}, 0, s{{[0-9]+}}
+; GCN: s_add_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN: s_addc_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN: s_addc_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0
+; GCN: s_add_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN: s_add_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
+; GCN: s_addc_u32 s{{[0-9]+}}, 0, s{{[0-9]+}}
+; GCN: s_addc_u32 s{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
 define amdgpu_kernel void @test_udiv64(i64 addrspace(1)* %arg) #0 {
 bb:
   %tmp = getelementptr inbounds i64, i64 addrspace(1)* %arg, i64 1