[llvm] [SCEV] Introduce UDiv::mayTriggerUB (PR #217064)

Ramkumar Ramachandra via llvm-commits llvm-commits at lists.llvm.org
Tue Aug 18 08:56:28 PDT 2026


https://github.com/artagnon created https://github.com/llvm/llvm-project/pull/217064

Use it to clean up the logic in SCEVExpander, leading to one improvement in LoopVectorize. While at it, make a related non-functional change in ScalarEvolution.

>From c4f28e0f0f75cc51b783a25f695fa35c387cd985 Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <artagnon at tenstorrent.com>
Date: Tue, 18 Aug 2026 16:43:58 +0100
Subject: [PATCH] [SCEV] Introduce UDiv::mayTriggerUB

Use it to clean up the logic in SCEVExpander, leading to one improvement
in LoopVectorize. While at it, make a related non-functional change in
ScalarEvolution.
---
 .../Analysis/ScalarEvolutionExpressions.h     |   5 +
 llvm/lib/Analysis/ScalarEvolution.cpp         | 265 ++++----
 .../Utils/ScalarEvolutionExpander.cpp         |  26 +-
 llvm/test/Transforms/LoopVectorize/pr38697.ll | 615 ++++++++++++++++--
 4 files changed, 720 insertions(+), 191 deletions(-)

diff --git a/llvm/include/llvm/Analysis/ScalarEvolutionExpressions.h b/llvm/include/llvm/Analysis/ScalarEvolutionExpressions.h
index ebee63963c701..6aea576c181cb 100644
--- a/llvm/include/llvm/Analysis/ScalarEvolutionExpressions.h
+++ b/llvm/include/llvm/Analysis/ScalarEvolutionExpressions.h
@@ -319,6 +319,11 @@ class SCEVUDivExpr : public SCEV {
 
   /// Methods for support type inquiry through isa, cast, and dyn_cast:
   static bool classof(const SCEV *S) { return S->getSCEVType() == scUDivExpr; }
+
+  /// Returns true if the expression may trigger undefined-behavior.
+  bool mayTriggerUB(ScalarEvolution &SE) const {
+    return !SE.isKnownNonZero(getRHS());
+  }
 };
 
 /// This node represents a polynomial recurrence on the trip count
diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp
index 1707c6c17fe23..cd3c3e6bba777 100644
--- a/llvm/lib/Analysis/ScalarEvolution.cpp
+++ b/llvm/lib/Analysis/ScalarEvolution.cpp
@@ -3514,161 +3514,160 @@ const SCEV *ScalarEvolution::getUDivExpr(SCEVUse LHS, SCEVUse RHS) {
     return LHS;
 
   if (const SCEVConstant *RHSC = dyn_cast<SCEVConstant>(RHS)) {
-    if (RHSC->getValue()->isOne())
-      return LHS;                               // X udiv 1 --> x
     // If the denominator is zero, the result of the udiv is undefined. Don't
     // try to analyze it, because the resolution chosen here may differ from
     // the resolution chosen in other parts of the compiler.
-    if (!RHSC->getValue()->isZero()) {
-      // Determine if the division can be folded into the operands of
-      // its operands.
-      // TODO: Generalize this to non-constants by using known-bits information.
-      Type *Ty = LHS->getType();
-      unsigned LZ = RHSC->getAPInt().countl_zero();
-      unsigned MaxShiftAmt = getTypeSizeInBits(Ty) - LZ - 1;
-      // For non-power-of-two values, effectively round the value up to the
-      // nearest power of two.
-      if (!RHSC->getAPInt().isPowerOf2())
-        ++MaxShiftAmt;
-      IntegerType *ExtTy =
+    if (RHSC->getValue()->isZero())
+      return getOrCreateUDivExpr(LHS, RHS);
+
+    if (RHSC->getValue()->isOne())
+      return LHS; // X udiv 1 --> x
+
+    // Determine if the division can be folded into the operands of
+    // its operands.
+    // TODO: Generalize this to non-constants by using known-bits information.
+    Type *Ty = LHS->getType();
+    unsigned LZ = RHSC->getAPInt().countl_zero();
+    unsigned MaxShiftAmt = getTypeSizeInBits(Ty) - LZ - 1;
+    // For non-power-of-two values, effectively round the value up to the
+    // nearest power of two.
+    if (!RHSC->getAPInt().isPowerOf2())
+      ++MaxShiftAmt;
+    IntegerType *ExtTy =
         IntegerType::get(getContext(), getTypeSizeInBits(Ty) + MaxShiftAmt);
-      if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(LHS))
-        if (const SCEVConstant *Step =
-            dyn_cast<SCEVConstant>(AR->getStepRecurrence(*this))) {
-          // {X,+,N}/C --> {X/C,+,N/C} if safe and N/C can be folded.
-          const APInt &StepInt = Step->getAPInt();
-          const APInt &DivInt = RHSC->getAPInt();
-          if (!StepInt.urem(DivInt) &&
-              getZeroExtendExpr(AR, ExtTy) ==
-              getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
-                            getZeroExtendExpr(Step, ExtTy),
-                            AR->getLoop(), SCEV::FlagAnyWrap)) {
-            SmallVector<SCEVUse, 4> Operands;
-            for (const SCEV *Op : AR->operands())
-              Operands.push_back(getUDivExpr(Op, RHS));
-            return getAddRecExpr(Operands, AR->getLoop(), SCEV::FlagNW);
-          }
-          /// Get a canonical UDivExpr for a recurrence.
-          /// {X,+,N}/C => {Y,+,N}/C where Y=X-(X%N). Safe when C%N=0.
-          const APInt *StartRem;
-          if (!DivInt.urem(StepInt) && match(getURemExpr(AR->getStart(), Step),
-                                             m_scev_APInt(StartRem))) {
-            bool NoWrap =
-                getZeroExtendExpr(AR, ExtTy) ==
+    if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(LHS))
+      if (const SCEVConstant *Step =
+              dyn_cast<SCEVConstant>(AR->getStepRecurrence(*this))) {
+        // {X,+,N}/C --> {X/C,+,N/C} if safe and N/C can be folded.
+        const APInt &StepInt = Step->getAPInt();
+        const APInt &DivInt = RHSC->getAPInt();
+        if (!StepInt.urem(DivInt) &&
+            getZeroExtendExpr(AR, ExtTy) ==
                 getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
                               getZeroExtendExpr(Step, ExtTy), AR->getLoop(),
-                              SCEV::FlagAnyWrap);
-
-            // With N <= C and both N, C as powers-of-2, the transformation
-            // {X,+,N}/C => {(X - X%N),+,N}/C preserves division results even
-            // if wrapping occurs, as the division results remain equivalent for
-            // all offsets in [[(X - X%N), X).
-            bool CanFoldWithWrap = StepInt.ule(DivInt) && // N <= C
-                                   StepInt.isPowerOf2() && DivInt.isPowerOf2();
-            // Only fold if the subtraction can be folded in the start
-            // expression.
-            const SCEV *NewStart =
-                getMinusSCEV(AR->getStart(), getConstant(*StartRem));
-            if (*StartRem != 0 && (NoWrap || CanFoldWithWrap) &&
-                !isa<SCEVAddExpr>(NewStart)) {
-              const SCEV *NewLHS =
-                  getAddRecExpr(NewStart, Step, AR->getLoop(),
-                                NoWrap ? SCEV::FlagNW : SCEV::FlagAnyWrap);
-              if (LHS != NewLHS)
-                return getUDivExpr(NewLHS, RHS);
-            }
-          }
+                              SCEV::FlagAnyWrap)) {
+          SmallVector<SCEVUse, 4> Operands;
+          for (const SCEV *Op : AR->operands())
+            Operands.push_back(getUDivExpr(Op, RHS));
+          return getAddRecExpr(Operands, AR->getLoop(), SCEV::FlagNW);
         }
-      // (A*B)/C --> A*(B/C) if safe and B/C can be folded.
-      if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(LHS)) {
-        SmallVector<SCEVUse, 4> Operands;
-        for (const SCEV *Op : M->operands())
-          Operands.push_back(getZeroExtendExpr(Op, ExtTy));
-        if (getZeroExtendExpr(M, ExtTy) == getMulExpr(Operands)) {
-          // Find an operand that's safely divisible.
-          for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
-            const SCEV *Op = M->getOperand(i);
-            const SCEV *Div = getUDivExpr(Op, RHSC);
-            if (!isa<SCEVUDivExpr>(Div) && getMulExpr(Div, RHSC) == Op) {
-              Operands = SmallVector<SCEVUse, 4>(M->operands());
-              Operands[i] = Div;
-              return getMulExpr(Operands);
-            }
-          }
-
-          // Even if it's not divisible, try to remove a common factor.
-          if (const auto *LHSC = dyn_cast<SCEVConstant>(M->getOperand(0))) {
-            APInt Factor = APIntOps::GreatestCommonDivisor(LHSC->getAPInt(),
-                                                           RHSC->getAPInt());
-            if (!Factor.isIntN(1)) {
-              SmallVector<SCEVUse, 2> NewOperands;
-              NewOperands.push_back(getConstant(LHSC->getAPInt().udiv(Factor)));
-              append_range(NewOperands, M->operands().drop_front());
-              const SCEV *NewMul = getMulExpr(NewOperands);
-              return getUDivExpr(NewMul,
-                                 getConstant(RHSC->getAPInt().udiv(Factor)));
-            }
+        /// Get a canonical UDivExpr for a recurrence.
+        /// {X,+,N}/C => {Y,+,N}/C where Y=X-(X%N). Safe when C%N=0.
+        const APInt *StartRem;
+        if (!DivInt.urem(StepInt) &&
+            match(getURemExpr(AR->getStart(), Step), m_scev_APInt(StartRem))) {
+          bool NoWrap = getZeroExtendExpr(AR, ExtTy) ==
+                        getAddRecExpr(getZeroExtendExpr(AR->getStart(), ExtTy),
+                                      getZeroExtendExpr(Step, ExtTy),
+                                      AR->getLoop(), SCEV::FlagAnyWrap);
+
+          // With N <= C and both N, C as powers-of-2, the transformation
+          // {X,+,N}/C => {(X - X%N),+,N}/C preserves division results even
+          // if wrapping occurs, as the division results remain equivalent for
+          // all offsets in [[(X - X%N), X).
+          bool CanFoldWithWrap = StepInt.ule(DivInt) && // N <= C
+                                 StepInt.isPowerOf2() && DivInt.isPowerOf2();
+          // Only fold if the subtraction can be folded in the start
+          // expression.
+          const SCEV *NewStart =
+              getMinusSCEV(AR->getStart(), getConstant(*StartRem));
+          if (*StartRem != 0 && (NoWrap || CanFoldWithWrap) &&
+              !isa<SCEVAddExpr>(NewStart)) {
+            const SCEV *NewLHS =
+                getAddRecExpr(NewStart, Step, AR->getLoop(),
+                              NoWrap ? SCEV::FlagNW : SCEV::FlagAnyWrap);
+            if (LHS != NewLHS)
+              return getUDivExpr(NewLHS, RHS);
           }
         }
       }
+    // (A*B)/C --> A*(B/C) if safe and B/C can be folded.
+    if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(LHS)) {
+      SmallVector<SCEVUse, 4> Operands;
+      for (const SCEV *Op : M->operands())
+        Operands.push_back(getZeroExtendExpr(Op, ExtTy));
+      if (getZeroExtendExpr(M, ExtTy) == getMulExpr(Operands)) {
+        // Find an operand that's safely divisible.
+        for (unsigned i = 0, e = M->getNumOperands(); i != e; ++i) {
+          const SCEV *Op = M->getOperand(i);
+          const SCEV *Div = getUDivExpr(Op, RHSC);
+          if (!isa<SCEVUDivExpr>(Div) && getMulExpr(Div, RHSC) == Op) {
+            Operands = SmallVector<SCEVUse, 4>(M->operands());
+            Operands[i] = Div;
+            return getMulExpr(Operands);
+          }
+        }
 
-      // (A/B)/C --> A/(B*C) if safe and B*C can be folded.
-      if (const SCEVUDivExpr *OtherDiv = dyn_cast<SCEVUDivExpr>(LHS)) {
-        if (auto *DivisorConstant =
-                dyn_cast<SCEVConstant>(OtherDiv->getRHS())) {
-          bool Overflow = false;
-          APInt NewRHS =
-              DivisorConstant->getAPInt().umul_ov(RHSC->getAPInt(), Overflow);
-          if (Overflow) {
-            return getConstant(RHSC->getType(), 0, false);
+        // Even if it's not divisible, try to remove a common factor.
+        if (const auto *LHSC = dyn_cast<SCEVConstant>(M->getOperand(0))) {
+          APInt Factor = APIntOps::GreatestCommonDivisor(LHSC->getAPInt(),
+                                                         RHSC->getAPInt());
+          if (!Factor.isIntN(1)) {
+            SmallVector<SCEVUse, 2> NewOperands;
+            NewOperands.push_back(getConstant(LHSC->getAPInt().udiv(Factor)));
+            append_range(NewOperands, M->operands().drop_front());
+            const SCEV *NewMul = getMulExpr(NewOperands);
+            return getUDivExpr(NewMul,
+                               getConstant(RHSC->getAPInt().udiv(Factor)));
           }
-          return getUDivExpr(OtherDiv->getLHS(), getConstant(NewRHS));
         }
       }
+    }
 
-      // (A+B)/C --> (A/C + B/C) if safe and A/C and B/C can be folded.
-      if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(LHS)) {
-        SmallVector<SCEVUse, 4> Operands;
-        for (const SCEV *Op : A->operands())
-          Operands.push_back(getZeroExtendExpr(Op, ExtTy));
-        if (getZeroExtendExpr(A, ExtTy) == getAddExpr(Operands)) {
-          Operands.clear();
-          for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i) {
-            const SCEV *Op = getUDivExpr(A->getOperand(i), RHS);
-            if (isa<SCEVUDivExpr>(Op) ||
-                getMulExpr(Op, RHS) != A->getOperand(i))
-              break;
-            Operands.push_back(Op);
-          }
-          if (Operands.size() == A->getNumOperands())
-            return getAddExpr(Operands);
+    // (A/B)/C --> A/(B*C) if safe and B*C can be folded.
+    if (const SCEVUDivExpr *OtherDiv = dyn_cast<SCEVUDivExpr>(LHS)) {
+      if (auto *DivisorConstant = dyn_cast<SCEVConstant>(OtherDiv->getRHS())) {
+        bool Overflow = false;
+        APInt NewRHS =
+            DivisorConstant->getAPInt().umul_ov(RHSC->getAPInt(), Overflow);
+        if (Overflow) {
+          return getConstant(RHSC->getType(), 0, false);
         }
+        return getUDivExpr(OtherDiv->getLHS(), getConstant(NewRHS));
       }
+    }
 
-      // ((N - M) + (M * A)) / N --> ((N - 1) + (M * A)) / N
-      // This is an idiom for rounding A up to the next multiple of N, where A
-      // is aready known to be a multiple of M. In this case, instcombine can
-      // see that some low bits of the added constant are unused, so can clear
-      // them, but we want to canonicalise to set the low bits. This makes the
-      // pattern easier to match, without needing to check for known bits in
-      // A*M.
-      const APInt &N = RHSC->getAPInt();
-      const APInt *NMinusM, *M;
-      const SCEV *A;
-      if (match(LHS, m_scev_Add(m_scev_APInt(NMinusM),
-                                m_scev_Mul(m_scev_APInt(M), m_SCEV(A))))) {
-        if (N.isPowerOf2() && M->isPowerOf2() && M->ult(N) &&
-            *NMinusM == N - *M) {
-          return getUDivExpr(
-              getAddExpr(getConstant(N - 1), getMulExpr(getConstant(*M), A)),
-              RHS);
+    // (A+B)/C --> (A/C + B/C) if safe and A/C and B/C can be folded.
+    if (const SCEVAddExpr *A = dyn_cast<SCEVAddExpr>(LHS)) {
+      SmallVector<SCEVUse, 4> Operands;
+      for (const SCEV *Op : A->operands())
+        Operands.push_back(getZeroExtendExpr(Op, ExtTy));
+      if (getZeroExtendExpr(A, ExtTy) == getAddExpr(Operands)) {
+        Operands.clear();
+        for (unsigned i = 0, e = A->getNumOperands(); i != e; ++i) {
+          const SCEV *Op = getUDivExpr(A->getOperand(i), RHS);
+          if (isa<SCEVUDivExpr>(Op) || getMulExpr(Op, RHS) != A->getOperand(i))
+            break;
+          Operands.push_back(Op);
         }
+        if (Operands.size() == A->getNumOperands())
+          return getAddExpr(Operands);
       }
+    }
 
-      // Fold if both operands are constant.
-      if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(LHS))
-        return getConstant(LHSC->getAPInt().udiv(RHSC->getAPInt()));
+    // ((N - M) + (M * A)) / N --> ((N - 1) + (M * A)) / N
+    // This is an idiom for rounding A up to the next multiple of N, where A
+    // is aready known to be a multiple of M. In this case, instcombine can
+    // see that some low bits of the added constant are unused, so can clear
+    // them, but we want to canonicalise to set the low bits. This makes the
+    // pattern easier to match, without needing to check for known bits in
+    // A*M.
+    const APInt &N = RHSC->getAPInt();
+    const APInt *NMinusM, *M;
+    const SCEV *A;
+    if (match(LHS, m_scev_Add(m_scev_APInt(NMinusM),
+                              m_scev_Mul(m_scev_APInt(M), m_SCEV(A))))) {
+      if (N.isPowerOf2() && M->isPowerOf2() && M->ult(N) &&
+          *NMinusM == N - *M) {
+        return getUDivExpr(
+            getAddExpr(getConstant(N - 1), getMulExpr(getConstant(*M), A)),
+            RHS);
+      }
     }
+
+    // Fold if both operands are constant.
+    if (const SCEVConstant *LHSC = dyn_cast<SCEVConstant>(LHS))
+      return getConstant(LHSC->getAPInt().udiv(RHSC->getAPInt()));
   }
 
   // ((-C + (C smax %x)) /u %x) evaluates to zero, for any positive constant C.
diff --git a/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp b/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
index a7ca2104e59c3..e7879d8b09a55 100644
--- a/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
+++ b/llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp
@@ -725,12 +725,12 @@ Value *SCEVExpander::visitUDivExpr(SCEVUseT<const SCEVUDivExpr *> S) {
     // We need an umax if either RHSExpr is not known to be zero, or if it is
     // not guaranteed to be non-poison. In the later case, the frozen poison may
     // be 0.
-    if (!SE.isKnownNonZero(RHSExpr) || !GuaranteedNotPoison)
+    if (S->mayTriggerUB(SE) || !GuaranteedNotPoison)
       RHS = Builder.CreateIntrinsic(RHS->getType(), Intrinsic::umax,
                                     {RHS, ConstantInt::get(RHS->getType(), 1)});
   }
   return InsertBinop(Instruction::UDiv, LHS, RHS, SCEV::FlagAnyWrap,
-                     /*IsSafeToHoist*/ SE.isKnownNonZero(S->getRHS()));
+                     /*IsSafeToHoist=*/!S->mayTriggerUB(SE));
 }
 
 /// Determine if this is a well-behaved chain of instructions leading back to
@@ -1667,20 +1667,11 @@ Value *SCEVExpander::expand(SCEVUse S) {
 
   // We can move insertion point only if there is no div or rem operations
   // otherwise we are risky to move it over the check for zero denominator.
-  auto SafeToHoist = [](const SCEV *S) {
-    return !SCEVExprContains(S, [](const SCEV *S) {
-              if (const auto *D = dyn_cast<SCEVUDivExpr>(S)) {
-                if (const auto *SC = dyn_cast<SCEVConstant>(D->getRHS()))
-                  // Division by non-zero constants can be hoisted.
-                  return SC->getValue()->isZero();
-                // All other divisions should not be moved as they may be
-                // divisions by zero and should be kept within the
-                // conditions of the surrounding loops that guard their
-                // execution (see PR35406).
-                return true;
-              }
-              return false;
-            });
+  auto SafeToHoist = [this](const SCEV *S) {
+    return !SCEVExprContains(S, [this](const SCEV *S) {
+      const auto *D = dyn_cast<SCEVUDivExpr>(S);
+      return D && D->mayTriggerUB(SE);
+    });
   };
   if (SafeToHoist(S)) {
     for (Loop *L = SE.LI.getLoopFor(Builder.GetInsertBlock());;
@@ -2507,8 +2498,7 @@ struct SCEVFindUnsafe {
 
   bool follow(const SCEV *S) {
     if (const SCEVUDivExpr *D = dyn_cast<SCEVUDivExpr>(S)) {
-      if (!SE.isKnownNonZero(D->getRHS()) ||
-          !SE.isGuaranteedNotToBePoison(D->getRHS())) {
+      if (D->mayTriggerUB(SE) || !SE.isGuaranteedNotToBePoison(D->getRHS())) {
         IsUnsafe = true;
         return false;
       }
diff --git a/llvm/test/Transforms/LoopVectorize/pr38697.ll b/llvm/test/Transforms/LoopVectorize/pr38697.ll
index 5570a1d8f7bd0..7caa7a99b509b 100644
--- a/llvm/test/Transforms/LoopVectorize/pr38697.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr38697.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph" --version 6
 ; RUN: opt -passes=loop-vectorize -force-vector-width=2 -S < %s 2>&1 | FileCheck %s
 ; RUN: opt -passes=indvars -S < %s 2>&1 | FileCheck %s -check-prefix=INDVARCHECK
 
@@ -31,17 +32,114 @@ target triple = "x86_64-unknown-linux-gnu"
 ; Verify that a 'udiv' does not appear in the 'loop1.preheader' block, and that
 ; a 'udiv' has been inserted at the top of the 'while.body.preheader' block.
 define void @testCountIncrLoop(ptr %ptr, i32 %lim, i32 %count, i32 %val) mustprogress {
-; CHECK-LABEL: @testCountIncrLoop(
-; CHECK-NEXT:  entry:
-; CHECK:       loop1.preheader:
-; CHECK-NOT:     udiv
-; CHECK:       loop1.body:
-; CHECK:       while.cond.preheader:
-; CHECK:       while.body.preheader:
-; CHECK:         [[TMP1:%.*]] = udiv i32 [[TMP0:%.*]], [[COUNT:%.*]]
-; CHECK:       vector.ph:
-; CHECK:       exit:
-; CHECK:         ret void
+; CHECK-LABEL: define void @testCountIncrLoop(
+; CHECK-SAME: ptr [[PTR:%.*]], i32 [[LIM:%.*]], i32 [[COUNT:%.*]], i32 [[VAL:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[LIM]], 0
+; CHECK-NEXT:    br i1 [[CMP1]], label %[[LOOP1_PREHEADER:.*]], [[EXIT:label %.*]]
+; CHECK:       [[LOOP1_PREHEADER]]:
+; CHECK-NEXT:    [[CMP2:%.*]] = icmp sgt i32 [[COUNT]], 0
+; CHECK-NEXT:    [[CMP4:%.*]] = icmp slt i32 [[COUNT]], 8
+; CHECK-NEXT:    br label %[[LOOP1_BODY:.*]]
+; CHECK:       [[LOOP1_BODY]]:
+; CHECK-NEXT:    [[OUTER_I:%.*]] = phi i32 [ 0, %[[LOOP1_PREHEADER]] ], [ [[OUTER_I_1:%.*]], %[[LOOP1_INC:.*]] ]
+; CHECK-NEXT:    [[INX_1:%.*]] = phi i32 [ 0, %[[LOOP1_PREHEADER]] ], [ [[INX_2:%.*]], %[[LOOP1_INC]] ]
+; CHECK-NEXT:    br i1 [[CMP2]], label %[[WHILE_COND_PREHEADER:.*]], label %[[LOOP1_INC]]
+; CHECK:       [[WHILE_COND_PREHEADER]]:
+; CHECK-NEXT:    br i1 [[CMP4]], label %[[WHILE_BODY_PREHEADER:.*]], [[WHILE_END:label %.*]]
+; CHECK:       [[WHILE_BODY_PREHEADER]]:
+; CHECK-NEXT:    [[TMP6:%.*]] = shl i32 [[COUNT]], 1
+; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @llvm.smax.i32(i32 [[TMP6]], i32 8)
+; CHECK-NEXT:    [[TMP2:%.*]] = mul i32 [[COUNT]], -2
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[TMP7]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP3]], i32 1)
+; CHECK-NEXT:    [[TMP5:%.*]] = sub i32 [[TMP7]], [[TMP4]]
+; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[TMP5]], [[TMP2]]
+; CHECK-NEXT:    [[TMP1:%.*]] = udiv i32 [[TMP0]], [[COUNT]]
+; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[TMP4]], [[TMP1]]
+; CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP8]], 1
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP9]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[TMP10:%.*]] = and i32 [[TMP9]], 3
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[TMP9]], [[TMP10]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[COUNT]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = mul i32 [[N_VEC]], [[COUNT]]
+; CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[COUNT]], [[TMP11]]
+; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[VAL]], i64 0
+; CHECK-NEXT:    [[TMP14:%.*]] = mul <2 x i32> splat (i32 2), [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i32> poison, i32 [[VAL]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT1]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = mul nsw <2 x i32> <i32 0, i32 1>, [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[INDUCTION:%.*]] = add nsw <2 x i32> [[BROADCAST_SPLAT]], [[TMP15]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP13]], %[[VECTOR_PH]] ], [ [[TMP18:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI3:%.*]] = phi <2 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP19:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[STEP_ADD:%.*]] = add nsw <2 x i32> [[VEC_IND]], [[TMP14]]
+; CHECK-NEXT:    [[TMP16:%.*]] = ashr <2 x i32> [[BROADCAST_SPLAT2]], [[VEC_IND]]
+; CHECK-NEXT:    [[TMP17:%.*]] = ashr <2 x i32> [[BROADCAST_SPLAT2]], [[STEP_ADD]]
+; CHECK-NEXT:    [[TMP18]] = add <2 x i32> [[TMP16]], [[VEC_PHI]]
+; CHECK-NEXT:    [[TMP19]] = add <2 x i32> [[TMP17]], [[VEC_PHI3]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nsw <2 x i32> [[STEP_ADD]], [[TMP14]]
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[BIN_RDX:%.*]] = add <2 x i32> [[TMP19]], [[TMP18]]
+; CHECK-NEXT:    [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[BIN_RDX]])
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[TMP9]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], [[WHILE_END_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+;
+; INDVARCHECK-LABEL: define void @testCountIncrLoop(
+; INDVARCHECK-SAME: ptr [[PTR:%.*]], i32 [[LIM:%.*]], i32 [[COUNT:%.*]], i32 [[VAL:%.*]]) #[[ATTR0:[0-9]+]] {
+; INDVARCHECK-NEXT:  [[ENTRY:.*:]]
+; INDVARCHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[LIM]], 0
+; INDVARCHECK-NEXT:    br i1 [[CMP1]], label %[[LOOP1_PREHEADER:.*]], label %[[EXIT:.*]]
+; INDVARCHECK:       [[LOOP1_PREHEADER]]:
+; INDVARCHECK-NEXT:    [[CMP2:%.*]] = icmp sgt i32 [[COUNT]], 0
+; INDVARCHECK-NEXT:    [[CMP4:%.*]] = icmp slt i32 [[COUNT]], 8
+; INDVARCHECK-NEXT:    br label %[[LOOP1_BODY:.*]]
+; INDVARCHECK:       [[LOOP1_BODY]]:
+; INDVARCHECK-NEXT:    [[OUTER_I:%.*]] = phi i32 [ 0, %[[LOOP1_PREHEADER]] ], [ [[OUTER_I_1:%.*]], %[[LOOP1_INC:.*]] ]
+; INDVARCHECK-NEXT:    [[INX_1:%.*]] = phi i32 [ 0, %[[LOOP1_PREHEADER]] ], [ [[INX_2:%.*]], %[[LOOP1_INC]] ]
+; INDVARCHECK-NEXT:    br i1 [[CMP2]], label %[[WHILE_COND_PREHEADER:.*]], label %[[LOOP1_INC]]
+; INDVARCHECK:       [[WHILE_COND_PREHEADER]]:
+; INDVARCHECK-NEXT:    br i1 [[CMP4]], label %[[WHILE_BODY_PREHEADER:.*]], label %[[WHILE_END:.*]]
+; INDVARCHECK:       [[WHILE_BODY_PREHEADER]]:
+; INDVARCHECK-NEXT:    br label %[[WHILE_BODY:.*]]
+; INDVARCHECK:       [[WHILE_BODY]]:
+; INDVARCHECK-NEXT:    [[TMP:%.*]] = phi i32 [ [[ADD3:%.*]], %[[WHILE_BODY]] ], [ [[COUNT]], %[[WHILE_BODY_PREHEADER]] ]
+; INDVARCHECK-NEXT:    [[RESULT_1:%.*]] = phi i32 [ [[ADD:%.*]], %[[WHILE_BODY]] ], [ [[VAL]], %[[WHILE_BODY_PREHEADER]] ]
+; INDVARCHECK-NEXT:    [[SHR:%.*]] = ashr i32 [[VAL]], [[TMP]]
+; INDVARCHECK-NEXT:    [[ADD]] = add nsw i32 [[SHR]], [[RESULT_1]]
+; INDVARCHECK-NEXT:    [[ADD3]] = add nsw i32 [[TMP]], [[COUNT]]
+; INDVARCHECK-NEXT:    [[CMP3:%.*]] = icmp slt i32 [[ADD3]], 8
+; INDVARCHECK-NEXT:    br i1 [[CMP3]], label %[[WHILE_BODY]], label %[[WHILE_END_LOOPEXIT:.*]]
+; INDVARCHECK:       [[WHILE_END_LOOPEXIT]]:
+; INDVARCHECK-NEXT:    [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD]], %[[WHILE_BODY]] ]
+; INDVARCHECK-NEXT:    br label %[[WHILE_END]]
+; INDVARCHECK:       [[WHILE_END]]:
+; INDVARCHECK-NEXT:    [[RESULT_0_LCSSA:%.*]] = phi i32 [ [[VAL]], %[[WHILE_COND_PREHEADER]] ], [ [[ADD_LCSSA]], %[[WHILE_END_LOOPEXIT]] ]
+; INDVARCHECK-NEXT:    [[CONV:%.*]] = trunc i32 [[RESULT_0_LCSSA]] to i8
+; INDVARCHECK-NEXT:    [[INC:%.*]] = add nsw i32 [[INX_1]], 1
+; INDVARCHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[INX_1]] to i64
+; INDVARCHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i64 [[IDXPROM]]
+; INDVARCHECK-NEXT:    store i8 [[CONV]], ptr [[ARRAYIDX]], align 1
+; INDVARCHECK-NEXT:    br label %[[LOOP1_INC]]
+; INDVARCHECK:       [[LOOP1_INC]]:
+; INDVARCHECK-NEXT:    [[INX_2]] = phi i32 [ [[INC]], %[[WHILE_END]] ], [ [[INX_1]], %[[LOOP1_BODY]] ]
+; INDVARCHECK-NEXT:    [[OUTER_I_1]] = add nuw nsw i32 [[OUTER_I]], 1
+; INDVARCHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[OUTER_I_1]], [[LIM]]
+; INDVARCHECK-NEXT:    br i1 [[EXITCOND]], label %[[EXIT_LOOPEXIT:.*]], label %[[LOOP1_BODY]]
+; INDVARCHECK:       [[EXIT_LOOPEXIT]]:
+; INDVARCHECK-NEXT:    br label %[[EXIT]]
+; INDVARCHECK:       [[EXIT]]:
+; INDVARCHECK-NEXT:    ret void
 ;
 entry:
   %cmp1 = icmp sgt i32 %lim, 0
@@ -115,14 +213,121 @@ exit:
 ;
 ; Verify that the 'udiv' is hoisted to the preheader, and is not in the loop body.
 define i32 @NonZeroDivHoist(ptr nocapture readonly %ptr, i32 %start1, i32 %start2) {
-; INDVARCHECK-LABEL: @NonZeroDivHoist(
-; INDVARCHECK-NEXT:  entry:
-; INDVARCHECK:       for.body3.lr.ph:
-; INDVARCHECK-NEXT:    [[TMP0:%.*]] = udiv i64 16, [[INDVARS_IV:%.*]]
-; INDVARCHECK-NEXT:    br label [[FOR_BODY3:%.*]]
-; INDVARCHECK:       for.body3:
-; INDVARCHECK-NOT:     udiv
-; INDVARCHECK:       for.end10:
+; CHECK-LABEL: define i32 @NonZeroDivHoist(
+; CHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br label %[[FOR_COND:.*]]
+; CHECK:       [[FOR_COND]]:
+; CHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END:.*]] ]
+; CHECK-NEXT:    [[COUNTER1_0:%.*]] = phi i32 [ 1, %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_END]] ]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[COUNTER1_0]], 100
+; CHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; CHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; CHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; CHECK:       [[FOR_BODY3_LR_PH]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = udiv i32 16, [[COUNTER1_0]]
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 10, [[START2]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[N_RND_UP]], 1
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[TMP2]]
+; CHECK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[ADD]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = udiv i32 16, [[COUNTER1_0]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2:.*]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ule <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP6:%.*]] = mul i32 [[INDEX]], [[TMP0]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i1> [[TMP5]], i64 0
+; CHECK-NEXT:    br i1 [[TMP7]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
+; CHECK:       [[PRED_LOAD_IF]]:
+; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[TMP4]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
+; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i32> poison, i32 [[TMP11]], i64 0
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE]]
+; CHECK:       [[PRED_LOAD_CONTINUE]]:
+; CHECK-NEXT:    [[TMP13:%.*]] = phi <2 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP12]], %[[PRED_LOAD_IF]] ]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP5]], i64 1
+; CHECK-NEXT:    br i1 [[TMP14]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_IF1]]:
+; CHECK-NEXT:    [[TMP15:%.*]] = mul i32 1, [[TMP0]]
+; CHECK-NEXT:    [[TMP16:%.*]] = add i32 [[TMP6]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP4]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[TMP19]], align 4
+; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP20]], i64 1
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_CONTINUE2]]:
+; CHECK-NEXT:    [[TMP22:%.*]] = phi <2 x i32> [ [[TMP13]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP21]], %[[PRED_LOAD_IF1]] ]
+; CHECK-NEXT:    [[TMP23]] = add <2 x i32> [[TMP22]], [[VEC_PHI]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nuw <2 x i32> [[VEC_IND]], splat (i32 2)
+; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP25:%.*]] = select <2 x i1> [[TMP5]], <2 x i32> [[TMP23]], <2 x i32> [[VEC_PHI]]
+; CHECK-NEXT:    [[TMP26:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP25]])
+; CHECK-NEXT:    br label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; CHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; CHECK-NEXT:    br label %[[FOR_END]]
+; CHECK:       [[FOR_END]]:
+; CHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[TMP26]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; CHECK-NEXT:    [[INC9]] = add i32 [[COUNTER1_0]], 1
+; CHECK-NEXT:    br label %[[FOR_COND]]
+; CHECK:       [[FOR_END10]]:
+; CHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; CHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
+;
+; INDVARCHECK-LABEL: define i32 @NonZeroDivHoist(
+; INDVARCHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; INDVARCHECK-NEXT:  [[ENTRY:.*]]:
+; INDVARCHECK-NEXT:    br label %[[FOR_COND:.*]]
+; INDVARCHECK:       [[FOR_COND]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_END:.*]] ], [ 1, %[[ENTRY]] ]
+; INDVARCHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END]] ]
+; INDVARCHECK-NEXT:    [[TMP0:%.*]] = udiv i64 16, [[INDVARS_IV]]
+; INDVARCHECK-NEXT:    [[EXITCOND4:%.*]] = icmp ne i64 [[INDVARS_IV]], 100
+; INDVARCHECK-NEXT:    br i1 [[EXITCOND4]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; INDVARCHECK:       [[FOR_BODY]]:
+; INDVARCHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; INDVARCHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; INDVARCHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; INDVARCHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_BODY3_LR_PH]]:
+; INDVARCHECK-NEXT:    br label %[[FOR_BODY3:.*]]
+; INDVARCHECK:       [[FOR_BODY3]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[FOR_BODY3]] ], [ 0, %[[FOR_BODY3_LR_PH]] ]
+; INDVARCHECK-NEXT:    [[VAL_126:%.*]] = phi i32 [ [[ADD]], %[[FOR_BODY3_LR_PH]] ], [ [[ADD7:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[COUNTER2_025:%.*]] = phi i32 [ [[START2]], %[[FOR_BODY3_LR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT2]] = add nuw nsw i64 [[INDVARS_IV1]], [[TMP0]]
+; INDVARCHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
+; INDVARCHECK-NEXT:    [[ADD7]] = add i32 [[TMP1]], [[VAL_126]]
+; INDVARCHECK-NEXT:    [[INC]] = add nuw nsw i32 [[COUNTER2_025]], 1
+; INDVARCHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INC]], 10
+; INDVARCHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_BODY3]], label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; INDVARCHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; INDVARCHECK-NEXT:    [[SPLIT:%.*]] = phi i32 [ [[ADD7]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    br label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_END]]:
+; INDVARCHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[SPLIT]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; INDVARCHECK-NEXT:    br label %[[FOR_COND]]
+; INDVARCHECK:       [[FOR_END10]]:
+; INDVARCHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; INDVARCHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
 ;
 entry:
   br label %for.cond
@@ -178,12 +383,123 @@ for.end10:
 ; Verify that the 'udiv' is not hoisted to the preheader, and it remains in the
 ; loop body.
 define i32 @ZeroDivNoHoist(ptr nocapture readonly %ptr, i32 %start1, i32 %start2) {
-; INDVARCHECK-LABEL: @ZeroDivNoHoist(
-; INDVARCHECK-NEXT:  entry:
-; INDVARCHECK-NOT:     udiv
-; INDVARCHECK:       for.body3:
-; INDVARCHECK:         [[TMP1:%.*]] = udiv i64 16, [[INDVARS_IV:%.*]]
-; INDVARCHECK:       for.cond1.for.end_crit_edge:
+; CHECK-LABEL: define i32 @ZeroDivNoHoist(
+; CHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br label %[[FOR_COND:.*]]
+; CHECK:       [[FOR_COND]]:
+; CHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END:.*]] ]
+; CHECK-NEXT:    [[COUNTER1_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_END]] ]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[COUNTER1_0]], 100
+; CHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; CHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; CHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; CHECK:       [[FOR_BODY3_LR_PH]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = udiv i32 16, [[COUNTER1_0]]
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 10, [[START2]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[N_RND_UP]], 1
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[TMP2]]
+; CHECK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[ADD]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = udiv i32 16, [[COUNTER1_0]]
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2:.*]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ule <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP6:%.*]] = mul i32 [[INDEX]], [[TMP0]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i1> [[TMP5]], i64 0
+; CHECK-NEXT:    br i1 [[TMP7]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
+; CHECK:       [[PRED_LOAD_IF]]:
+; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[TMP4]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
+; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i32> poison, i32 [[TMP11]], i64 0
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE]]
+; CHECK:       [[PRED_LOAD_CONTINUE]]:
+; CHECK-NEXT:    [[TMP13:%.*]] = phi <2 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP12]], %[[PRED_LOAD_IF]] ]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP5]], i64 1
+; CHECK-NEXT:    br i1 [[TMP14]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_IF1]]:
+; CHECK-NEXT:    [[TMP15:%.*]] = mul i32 1, [[TMP0]]
+; CHECK-NEXT:    [[TMP16:%.*]] = add i32 [[TMP6]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP4]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[TMP19]], align 4
+; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP20]], i64 1
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_CONTINUE2]]:
+; CHECK-NEXT:    [[TMP22:%.*]] = phi <2 x i32> [ [[TMP13]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP21]], %[[PRED_LOAD_IF1]] ]
+; CHECK-NEXT:    [[TMP23]] = add <2 x i32> [[TMP22]], [[VEC_PHI]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nuw <2 x i32> [[VEC_IND]], splat (i32 2)
+; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP25:%.*]] = select <2 x i1> [[TMP5]], <2 x i32> [[TMP23]], <2 x i32> [[VEC_PHI]]
+; CHECK-NEXT:    [[TMP26:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP25]])
+; CHECK-NEXT:    br label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; CHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; CHECK-NEXT:    br label %[[FOR_END]]
+; CHECK:       [[FOR_END]]:
+; CHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[TMP26]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; CHECK-NEXT:    [[INC9]] = add i32 [[COUNTER1_0]], 1
+; CHECK-NEXT:    br label %[[FOR_COND]]
+; CHECK:       [[FOR_END10]]:
+; CHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; CHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
+;
+; INDVARCHECK-LABEL: define i32 @ZeroDivNoHoist(
+; INDVARCHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; INDVARCHECK-NEXT:  [[ENTRY:.*]]:
+; INDVARCHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[START1]] to i64
+; INDVARCHECK-NEXT:    br label %[[FOR_COND:.*]]
+; INDVARCHECK:       [[FOR_COND]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_END:.*]] ], [ [[TMP0]], %[[ENTRY]] ]
+; INDVARCHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END]] ]
+; INDVARCHECK-NEXT:    [[INDVARS3:%.*]] = trunc i64 [[INDVARS_IV]] to i32
+; INDVARCHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[INDVARS3]], 100
+; INDVARCHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; INDVARCHECK:       [[FOR_BODY]]:
+; INDVARCHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; INDVARCHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; INDVARCHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; INDVARCHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_BODY3_LR_PH]]:
+; INDVARCHECK-NEXT:    br label %[[FOR_BODY3:.*]]
+; INDVARCHECK:       [[FOR_BODY3]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[FOR_BODY3]] ], [ 0, %[[FOR_BODY3_LR_PH]] ]
+; INDVARCHECK-NEXT:    [[VAL_126:%.*]] = phi i32 [ [[ADD]], %[[FOR_BODY3_LR_PH]] ], [ [[ADD7:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[COUNTER2_025:%.*]] = phi i32 [ [[START2]], %[[FOR_BODY3_LR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = udiv i64 16, [[INDVARS_IV]]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT2]] = add nuw nsw i64 [[INDVARS_IV1]], [[TMP1]]
+; INDVARCHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
+; INDVARCHECK-NEXT:    [[ADD7]] = add i32 [[TMP1]], [[VAL_126]]
+; INDVARCHECK-NEXT:    [[INC]] = add nuw nsw i32 [[COUNTER2_025]], 1
+; INDVARCHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INC]], 10
+; INDVARCHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_BODY3]], label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; INDVARCHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; INDVARCHECK-NEXT:    [[SPLIT:%.*]] = phi i32 [ [[ADD7]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    br label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_END]]:
+; INDVARCHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[SPLIT]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; INDVARCHECK-NEXT:    br label %[[FOR_COND]]
+; INDVARCHECK:       [[FOR_END10]]:
+; INDVARCHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; INDVARCHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
 ;
 entry:
   br label %for.cond
@@ -238,14 +554,123 @@ for.end10:
 ; Verify that the division-operation is hoisted, and that it appears as a
 ; right-shift ('lshr') rather than an explicit division.
 define i32 @DivBy16Hoist(ptr nocapture readonly %ptr, i32 %start1, i32 %start2) {
-; INDVARCHECK-LABEL: @DivBy16Hoist(
-; INDVARCHECK-NEXT:  entry:
-; INDVARCHECK:       for.cond:
-; INDVARCHECK:         [[TMP1:%.*]] = lshr i64 [[INDVARS_IV:%.*]], 4
-; INDVARCHECK:       for.body:
-; INDVARCHECK-NOT:     lshr
-; INDVARCHECK-NOT:     udiv
-; INDVARCHECK:       for.end10:
+; CHECK-LABEL: define i32 @DivBy16Hoist(
+; CHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br label %[[FOR_COND:.*]]
+; CHECK:       [[FOR_COND]]:
+; CHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END:.*]] ]
+; CHECK-NEXT:    [[COUNTER1_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_END]] ]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[COUNTER1_0]], 100
+; CHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; CHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; CHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; CHECK:       [[FOR_BODY3_LR_PH]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = lshr i32 [[COUNTER1_0]], 4
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 10, [[START2]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[N_RND_UP]], 1
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[TMP2]]
+; CHECK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[ADD]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = lshr i32 [[COUNTER1_0]], 4
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2:.*]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ule <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP6:%.*]] = mul i32 [[INDEX]], [[TMP0]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i1> [[TMP5]], i64 0
+; CHECK-NEXT:    br i1 [[TMP7]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
+; CHECK:       [[PRED_LOAD_IF]]:
+; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[TMP4]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
+; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i32> poison, i32 [[TMP11]], i64 0
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE]]
+; CHECK:       [[PRED_LOAD_CONTINUE]]:
+; CHECK-NEXT:    [[TMP13:%.*]] = phi <2 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP12]], %[[PRED_LOAD_IF]] ]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP5]], i64 1
+; CHECK-NEXT:    br i1 [[TMP14]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_IF1]]:
+; CHECK-NEXT:    [[TMP15:%.*]] = mul i32 1, [[TMP0]]
+; CHECK-NEXT:    [[TMP16:%.*]] = add i32 [[TMP6]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP4]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[TMP19]], align 4
+; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP20]], i64 1
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_CONTINUE2]]:
+; CHECK-NEXT:    [[TMP22:%.*]] = phi <2 x i32> [ [[TMP13]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP21]], %[[PRED_LOAD_IF1]] ]
+; CHECK-NEXT:    [[TMP23]] = add <2 x i32> [[TMP22]], [[VEC_PHI]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nuw <2 x i32> [[VEC_IND]], splat (i32 2)
+; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP25:%.*]] = select <2 x i1> [[TMP5]], <2 x i32> [[TMP23]], <2 x i32> [[VEC_PHI]]
+; CHECK-NEXT:    [[TMP26:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP25]])
+; CHECK-NEXT:    br label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; CHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; CHECK-NEXT:    br label %[[FOR_END]]
+; CHECK:       [[FOR_END]]:
+; CHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[TMP26]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; CHECK-NEXT:    [[INC9]] = add i32 [[COUNTER1_0]], 1
+; CHECK-NEXT:    br label %[[FOR_COND]]
+; CHECK:       [[FOR_END10]]:
+; CHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; CHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
+;
+; INDVARCHECK-LABEL: define i32 @DivBy16Hoist(
+; INDVARCHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; INDVARCHECK-NEXT:  [[ENTRY:.*]]:
+; INDVARCHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[START1]] to i64
+; INDVARCHECK-NEXT:    br label %[[FOR_COND:.*]]
+; INDVARCHECK:       [[FOR_COND]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_END:.*]] ], [ [[TMP0]], %[[ENTRY]] ]
+; INDVARCHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END]] ]
+; INDVARCHECK-NEXT:    [[INDVARS3:%.*]] = trunc i64 [[INDVARS_IV]] to i32
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = lshr i64 [[INDVARS_IV]], 4
+; INDVARCHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[INDVARS3]], 100
+; INDVARCHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; INDVARCHECK:       [[FOR_BODY]]:
+; INDVARCHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; INDVARCHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; INDVARCHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; INDVARCHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_BODY3_LR_PH]]:
+; INDVARCHECK-NEXT:    br label %[[FOR_BODY3:.*]]
+; INDVARCHECK:       [[FOR_BODY3]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[FOR_BODY3]] ], [ 0, %[[FOR_BODY3_LR_PH]] ]
+; INDVARCHECK-NEXT:    [[VAL_126:%.*]] = phi i32 [ [[ADD]], %[[FOR_BODY3_LR_PH]] ], [ [[ADD7:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[COUNTER2_025:%.*]] = phi i32 [ [[START2]], %[[FOR_BODY3_LR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT2]] = add nuw nsw i64 [[INDVARS_IV1]], [[TMP1]]
+; INDVARCHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
+; INDVARCHECK-NEXT:    [[ADD7]] = add i32 [[TMP1]], [[VAL_126]]
+; INDVARCHECK-NEXT:    [[INC]] = add nuw nsw i32 [[COUNTER2_025]], 1
+; INDVARCHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INC]], 10
+; INDVARCHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_BODY3]], label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; INDVARCHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; INDVARCHECK-NEXT:    [[SPLIT:%.*]] = phi i32 [ [[ADD7]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    br label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_END]]:
+; INDVARCHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[SPLIT]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; INDVARCHECK-NEXT:    br label %[[FOR_COND]]
+; INDVARCHECK:       [[FOR_END10]]:
+; INDVARCHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; INDVARCHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
 ;
 entry:
   br label %for.cond
@@ -299,13 +724,123 @@ for.end10:
 ;
 ; Verify that the division-operation is hoisted.
 define i32 @DivBy17Hoist(ptr nocapture readonly %ptr, i32 %start1, i32 %start2) {
-; INDVARCHECK-LABEL: @DivBy17Hoist(
-; INDVARCHECK-NEXT:  entry:
-; INDVARCHECK:       for.cond:
-; INDVARCHECK:         [[TMP1:%.*]] = udiv i64 [[INDVARS_IV:%.*]], 17
-; INDVARCHECK:       for.body:
-; INDVARCHECK-NOT:     udiv
-; INDVARCHECK:       for.end10:
+; CHECK-LABEL: define i32 @DivBy17Hoist(
+; CHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    br label %[[FOR_COND:.*]]
+; CHECK:       [[FOR_COND]]:
+; CHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END:.*]] ]
+; CHECK-NEXT:    [[COUNTER1_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[INC9:%.*]], %[[FOR_END]] ]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[COUNTER1_0]], 100
+; CHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; CHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; CHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; CHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; CHECK:       [[FOR_BODY3_LR_PH]]:
+; CHECK-NEXT:    [[TMP0:%.*]] = udiv i32 [[COUNTER1_0]], 17
+; CHECK-NEXT:    [[TMP1:%.*]] = sub i32 10, [[START2]]
+; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_RND_UP:%.*]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = and i32 [[N_RND_UP]], 1
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[N_RND_UP]], [[TMP2]]
+; CHECK-NEXT:    [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x i32> zeroinitializer, i32 [[ADD]], i64 0
+; CHECK-NEXT:    [[TMP4:%.*]] = udiv i32 [[COUNTER1_0]], 17
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[TRIP_COUNT_MINUS_1]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2:.*]] ]
+; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ [[TMP3]], %[[VECTOR_PH]] ], [ [[TMP23:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[PRED_LOAD_CONTINUE2]] ]
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ule <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[TMP6:%.*]] = mul i32 [[INDEX]], [[TMP0]]
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i1> [[TMP5]], i64 0
+; CHECK-NEXT:    br i1 [[TMP7]], label %[[PRED_LOAD_IF:.*]], label %[[PRED_LOAD_CONTINUE:.*]]
+; CHECK:       [[PRED_LOAD_IF]]:
+; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[TMP4]], [[TMP6]]
+; CHECK-NEXT:    [[TMP9:%.*]] = zext i32 [[TMP8]] to i64
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = load i32, ptr [[TMP10]], align 4
+; CHECK-NEXT:    [[TMP12:%.*]] = insertelement <2 x i32> poison, i32 [[TMP11]], i64 0
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE]]
+; CHECK:       [[PRED_LOAD_CONTINUE]]:
+; CHECK-NEXT:    [[TMP13:%.*]] = phi <2 x i32> [ poison, %[[VECTOR_BODY]] ], [ [[TMP12]], %[[PRED_LOAD_IF]] ]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <2 x i1> [[TMP5]], i64 1
+; CHECK-NEXT:    br i1 [[TMP14]], label %[[PRED_LOAD_IF1:.*]], label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_IF1]]:
+; CHECK-NEXT:    [[TMP15:%.*]] = mul i32 1, [[TMP0]]
+; CHECK-NEXT:    [[TMP16:%.*]] = add i32 [[TMP6]], [[TMP15]]
+; CHECK-NEXT:    [[TMP17:%.*]] = add i32 [[TMP4]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = zext i32 [[TMP17]] to i64
+; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[TMP19]], align 4
+; CHECK-NEXT:    [[TMP21:%.*]] = insertelement <2 x i32> [[TMP13]], i32 [[TMP20]], i64 1
+; CHECK-NEXT:    br label %[[PRED_LOAD_CONTINUE2]]
+; CHECK:       [[PRED_LOAD_CONTINUE2]]:
+; CHECK-NEXT:    [[TMP22:%.*]] = phi <2 x i32> [ [[TMP13]], %[[PRED_LOAD_CONTINUE]] ], [ [[TMP21]], %[[PRED_LOAD_IF1]] ]
+; CHECK-NEXT:    [[TMP23]] = add <2 x i32> [[TMP22]], [[VEC_PHI]]
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nuw <2 x i32> [[VEC_IND]], splat (i32 2)
+; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP24]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[TMP25:%.*]] = select <2 x i1> [[TMP5]], <2 x i32> [[TMP23]], <2 x i32> [[VEC_PHI]]
+; CHECK-NEXT:    [[TMP26:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP25]])
+; CHECK-NEXT:    br label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; CHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; CHECK-NEXT:    br label %[[FOR_END]]
+; CHECK:       [[FOR_END]]:
+; CHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[TMP26]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; CHECK-NEXT:    [[INC9]] = add i32 [[COUNTER1_0]], 1
+; CHECK-NEXT:    br label %[[FOR_COND]]
+; CHECK:       [[FOR_END10]]:
+; CHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; CHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
+;
+; INDVARCHECK-LABEL: define i32 @DivBy17Hoist(
+; INDVARCHECK-SAME: ptr readonly captures(none) [[PTR:%.*]], i32 [[START1:%.*]], i32 [[START2:%.*]]) {
+; INDVARCHECK-NEXT:  [[ENTRY:.*]]:
+; INDVARCHECK-NEXT:    [[TMP0:%.*]] = zext i32 [[START1]] to i64
+; INDVARCHECK-NEXT:    br label %[[FOR_COND:.*]]
+; INDVARCHECK:       [[FOR_COND]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_END:.*]] ], [ [[TMP0]], %[[ENTRY]] ]
+; INDVARCHECK-NEXT:    [[VAL_0:%.*]] = phi i32 [ [[START1]], %[[ENTRY]] ], [ [[VAL_1_LCSSA:%.*]], %[[FOR_END]] ]
+; INDVARCHECK-NEXT:    [[INDVARS3:%.*]] = trunc i64 [[INDVARS_IV]] to i32
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = udiv i64 [[INDVARS_IV]], 17
+; INDVARCHECK-NEXT:    [[CMP:%.*]] = icmp ult i32 [[INDVARS3]], 100
+; INDVARCHECK-NEXT:    br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END10:.*]]
+; INDVARCHECK:       [[FOR_BODY]]:
+; INDVARCHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[PTR]], align 4
+; INDVARCHECK-NEXT:    [[ADD:%.*]] = add i32 [[TMP]], [[VAL_0]]
+; INDVARCHECK-NEXT:    [[CMP224:%.*]] = icmp ult i32 [[START2]], 10
+; INDVARCHECK-NEXT:    br i1 [[CMP224]], label %[[FOR_BODY3_LR_PH:.*]], label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_BODY3_LR_PH]]:
+; INDVARCHECK-NEXT:    br label %[[FOR_BODY3:.*]]
+; INDVARCHECK:       [[FOR_BODY3]]:
+; INDVARCHECK-NEXT:    [[INDVARS_IV1:%.*]] = phi i64 [ [[INDVARS_IV_NEXT2:%.*]], %[[FOR_BODY3]] ], [ 0, %[[FOR_BODY3_LR_PH]] ]
+; INDVARCHECK-NEXT:    [[VAL_126:%.*]] = phi i32 [ [[ADD]], %[[FOR_BODY3_LR_PH]] ], [ [[ADD7:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[COUNTER2_025:%.*]] = phi i32 [ [[START2]], %[[FOR_BODY3_LR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT2]] = add nuw nsw i64 [[INDVARS_IV1]], [[TMP1]]
+; INDVARCHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, ptr [[PTR]], i64 [[INDVARS_IV_NEXT2]]
+; INDVARCHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX6]], align 4
+; INDVARCHECK-NEXT:    [[ADD7]] = add i32 [[TMP1]], [[VAL_126]]
+; INDVARCHECK-NEXT:    [[INC]] = add nuw nsw i32 [[COUNTER2_025]], 1
+; INDVARCHECK-NEXT:    [[EXITCOND:%.*]] = icmp ne i32 [[INC]], 10
+; INDVARCHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_BODY3]], label %[[FOR_COND1_FOR_END_CRIT_EDGE:.*]]
+; INDVARCHECK:       [[FOR_COND1_FOR_END_CRIT_EDGE]]:
+; INDVARCHECK-NEXT:    [[SPLIT:%.*]] = phi i32 [ [[ADD7]], %[[FOR_BODY3]] ]
+; INDVARCHECK-NEXT:    br label %[[FOR_END]]
+; INDVARCHECK:       [[FOR_END]]:
+; INDVARCHECK-NEXT:    [[VAL_1_LCSSA]] = phi i32 [ [[SPLIT]], %[[FOR_COND1_FOR_END_CRIT_EDGE]] ], [ [[ADD]], %[[FOR_BODY]] ]
+; INDVARCHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; INDVARCHECK-NEXT:    br label %[[FOR_COND]]
+; INDVARCHECK:       [[FOR_END10]]:
+; INDVARCHECK-NEXT:    [[VAL_0_LCSSA:%.*]] = phi i32 [ [[VAL_0]], %[[FOR_COND]] ]
+; INDVARCHECK-NEXT:    ret i32 [[VAL_0_LCSSA]]
 ;
 entry:
   br label %for.cond



More information about the llvm-commits mailing list