[SCEV] Improve the run-time checking of the NoWrap predicate

[sbaranga]

Summary

This implements a new method of run-time checking the NoWrap
SCEV predicates, which should be easier to optimize and nicer
for targets that don't correctly handle multiplication/addition
of large integer types (like i128).

If the AddRec is {a,+,b} and the backedge taken count is c,
the idea is to check that |b| * c doesn't have unsigned overflow,
and depending on the sign of b, that:

a + |b| * c >= a (b >= 0) or
a - |b| * c <= a (b <= 0)

where the comparisons above are signed or unsigned, depending on
the flag that we're checking.

The advantage of doing this is that we avoid extending to a larger
type and we avoid the multiplication of large types (multiplying
i128 can be expensive).

[sanjoy]

lib/Analysis/ScalarEvolutionExpander.cpp

2020 ↗

(On Diff #54202)

One of these has to include Step == 0.

2021 ↗

(On Diff #54202)

Spelling: "Step > 0"

2022 ↗

(On Diff #54202)

Nit: I'd be explicit here about "... doesn't unsigned overflow"

2038 ↗

(On Diff #54202)

Why not just check if Step < 0? (Assuming later passes don't optimize this) That way we won't have compute -Step if Step is positive.

2054 ↗

(On Diff #54202)

I'd keep the operands in the same order in both the cases.

2084 ↗

(On Diff #54202)

Why do you need this extra check? If Step is 0 then won't both the less than and the greater than check return false?

[sbaranga]

lib/Analysis/ScalarEvolutionExpander.cpp

2038 ↗

(On Diff #54202)

Sure, that makes sense.

2084 ↗

(On Diff #54202)

The problem is that we have an edge case where we discard bits by truncating the backedge taken count and the step is 0. In this case we don't have overflow. It would be better to create it only when we need to truncate the backedge taken count and create the and with the 'BackedgeCount < MAX_UINT(Type(Start))' comparison.

[sbaranga]

Only compare the step with 0 when we need to truncate the backedge taken count.
Otherwise, the case where the step is 0 is correctly handled.

Address the other comments received in the review.

llvm/trunk/lib/Analysis/ScalarEvolutionExpander.cpp

 Value *SCEVExpander::generateOverflowCheck(const SCEVAddRecExpr *AR,
                                            Instruction *Loc, bool Signed) {
   assert(AR->isAffine() && "Cannot generate RT check for "
                            "non-affine expression");
 
   SCEVUnionPredicate Pred;
   const SCEV *ExitCount =
       SE.getPredicatedBackedgeTakenCount(AR->getLoop(), Pred);
+
+  assert(ExitCount != SE.getCouldNotCompute() && "Invalid loop count");
+
   const SCEV *Step = AR->getStepRecurrence(SE);
   const SCEV *Start = AR->getStart();
 
-  unsigned DstBits = SE.getTypeSizeInBits(AR->getType());
   unsigned SrcBits = SE.getTypeSizeInBits(ExitCount->getType());
-  unsigned MaxBits = 2 * std::max(DstBits, SrcBits);
+  unsigned DstBits = SE.getTypeSizeInBits(AR->getType());
 
-  auto *TripCount = SE.getTruncateOrZeroExtend(ExitCount, AR->getType());
-  IntegerType *MaxTy = IntegerType::get(Loc->getContext(), MaxBits);
+  // The expression {Start,+,Step} has nusw/nssw if
+  //   Step < 0, Start - |Step| * Backedge <= Start
+  //   Step >= 0, Start + |Step| * Backedge > Start
+  // and |Step| * Backedge doesn't unsigned overflow.
 
-  assert(ExitCount != SE.getCouldNotCompute() && "Invalid loop count");
+  IntegerType *CountTy = IntegerType::get(Loc->getContext(), SrcBits);
+  Builder.SetInsertPoint(Loc);
+  Value *TripCountVal = expandCodeFor(ExitCount, CountTy, Loc);
+
+  IntegerType *Ty =
+      IntegerType::get(Loc->getContext(), SE.getTypeSizeInBits(AR->getType()));
+
+  Value *StepValue = expandCodeFor(Step, Ty, Loc);
+  Value *NegStepValue = expandCodeFor(SE.getNegativeSCEV(Step), Ty, Loc);
+  Value *StartValue = expandCodeFor(Start, Ty, Loc);
 
-  const auto *ExtendedTripCount = SE.getZeroExtendExpr(ExitCount, MaxTy);
-  const auto *ExtendedStep = SE.getSignExtendExpr(Step, MaxTy);
-  const auto *ExtendedStart = Signed ? SE.getSignExtendExpr(Start, MaxTy)
-                                     : SE.getZeroExtendExpr(Start, MaxTy);
-
-  const SCEV *End = SE.getAddExpr(Start, SE.getMulExpr(TripCount, Step));
-  const SCEV *RHS = Signed ? SE.getSignExtendExpr(End, MaxTy)
-                           : SE.getZeroExtendExpr(End, MaxTy);
-
-  const SCEV *LHS = SE.getAddExpr(
-      ExtendedStart, SE.getMulExpr(ExtendedTripCount, ExtendedStep));
-
-  // Do all SCEV expansions now.
-  Value *LHSVal = expandCodeFor(LHS, MaxTy, Loc);
-  Value *RHSVal = expandCodeFor(RHS, MaxTy, Loc);
+  ConstantInt *Zero =
+      ConstantInt::get(Loc->getContext(), APInt::getNullValue(DstBits));
 
   Builder.SetInsertPoint(Loc);
+  // Compute |Step|
+  Value *StepCompare = Builder.CreateICmp(ICmpInst::ICMP_SLT, StepValue, Zero);
+  Value *AbsStep = Builder.CreateSelect(StepCompare, NegStepValue, StepValue);
+
+  // Get the backedge taken count and truncate or extended to the AR type.
+  Value *TruncTripCount = Builder.CreateZExtOrTrunc(TripCountVal, Ty);
+  auto *MulF = Intrinsic::getDeclaration(Loc->getModule(),
+                                         Intrinsic::umul_with_overflow, Ty);
+
+  // Compute |Step| * Backedge
+  CallInst *Mul = Builder.CreateCall(MulF, {AbsStep, TruncTripCount}, "mul");
+  Value *MulV = Builder.CreateExtractValue(Mul, 0, "mul.result");
+  Value *OfMul = Builder.CreateExtractValue(Mul, 1, "mul.overflow");
+
+  // Compute:
+  //   Start + |Step| * Backedge < Start
+  //   Start - |Step| * Backedge > Start
+  Value *Add = Builder.CreateAdd(StartValue, MulV);
+  Value *Sub = Builder.CreateSub(StartValue, MulV);
+
+  Value *EndCompareGT = Builder.CreateICmp(
+      Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT, Sub, StartValue);
+
+  Value *EndCompareLT = Builder.CreateICmp(
+      Signed ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT, Add, StartValue);
+
+  // Select the answer based on the sign of Step.
+  Value *EndCheck =
+      Builder.CreateSelect(StepCompare, EndCompareGT, EndCompareLT);
+
+  // If the backedge taken count type is larger than the AR type,
+  // check that we don't drop any bits by truncating it. If we are
+  // droping bits, then we have overflow (unless the step is zero).
+  if (SE.getTypeSizeInBits(CountTy) > SE.getTypeSizeInBits(Ty)) {
+    auto MaxVal = APInt::getMaxValue(DstBits).zext(SrcBits);
+    auto *BackedgeCheck =
+        Builder.CreateICmp(ICmpInst::ICMP_UGT, TripCountVal,
+                           ConstantInt::get(Loc->getContext(), MaxVal));
+    BackedgeCheck = Builder.CreateAnd(
+        BackedgeCheck, Builder.CreateICmp(ICmpInst::ICMP_NE, StepValue, Zero));
+
+    EndCheck = Builder.CreateOr(EndCheck, BackedgeCheck);
+  }
 
-  return Builder.CreateICmp(ICmpInst::ICMP_NE, RHSVal, LHSVal);
+  EndCheck = Builder.CreateOr(EndCheck, OfMul);
+  return EndCheck;
 }
 
 Value *SCEVExpander::expandWrapPredicate(const SCEVWrapPredicate *Pred,
                                          Instruction *IP) {
   const auto *A = cast<SCEVAddRecExpr>(Pred->getExpr());
   Value *NSSWCheck = nullptr, *NUSWCheck = nullptr;
 
   // Add a check for NUSW

llvm/trunk/test/Analysis/LoopAccessAnalysis/wrapping-pointer-versioning.ll

 
 ; LAA: [PSE]  %arrayidxA = getelementptr i16, i16* %a, i64 %mul_ext:
 ; LAA-NEXT: ((2 * (zext i32 {0,+,2}<%for.body> to i64)) + %a)
 ; LAA-NEXT: --> {%a,+,4}<%for.body>
 
 
 ; LV-LABEL: f1
 ; LV-LABEL: for.body.lver.check
-; LV: [[PredCheck0:%[^ ]*]] = icmp ne i128
-; LV: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
-; LV: [[PredCheck1:%[^ ]*]] = icmp ne i128
+
+; LV:      [[BETrunc:%[^ ]*]] = trunc i64 [[BE:%[^ ]*]] to i32
+; LV-NEXT: [[OFMul:%[^ ]*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[BETrunc]])
+; LV-NEXT: [[OFMulResult:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 0
+; LV-NEXT: [[OFMulOverflow:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 1
+; LV-NEXT: [[AddEnd:%[^ ]*]] = add i32 0, [[OFMulResult]]
+; LV-NEXT: [[SubEnd:%[^ ]*]] = sub i32 0, [[OFMulResult]]
+; LV-NEXT: [[CmpNeg:%[^ ]*]] = icmp ugt i32 [[SubEnd]], 0
+; LV-NEXT: [[CmpPos:%[^ ]*]] = icmp ult i32 [[AddEnd]], 0
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 false, i1 [[CmpNeg]], i1 [[CmpPos]]
+; LV-NEXT: [[BECheck:%[^ ]*]] = icmp ugt i64 [[BE]], 4294967295
+; LV-NEXT: [[CheckOr0:%[^ ]*]] = or i1 [[Cmp]], [[BECheck]]
+; LV-NEXT: [[PredCheck0:%[^ ]*]] = or i1 [[CheckOr0]], [[OFMulOverflow]]
+
+; LV-NEXT: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
+
+; LV-NEXT: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE]])
+; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0
+; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1
+; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 %a2, [[OFMulResult1]]
+; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 %a2, [[OFMulResult1]]
+; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], %a2
+; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], %a2
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 false, i1 [[CmpNeg1]], i1 [[CmpPos1]]
+; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]]
+
 ; LV: [[FinalCheck:%[^ ]*]] = or i1 [[Or0]], [[PredCheck1]]
 ; LV: br i1 [[FinalCheck]], label %for.body.ph.lver.orig, label %for.body.ph
 define void @f1(i16* noalias %a,
                 i16* noalias %b, i64 %N) {
 entry:
   br label %for.body
 
 for.body:                                         ; preds = %for.body, %entry

 ;     i64 {zext i32 (2 * (trunc i64 %N to i32)) to i64,+,-2}<%for.body>
 
 ; LAA: [PSE]  %arrayidxA = getelementptr i16, i16* %a, i64 %mul_ext:
 ; LAA-NEXT: ((2 * (zext i32 {(2 * (trunc i64 %N to i32)),+,-2}<%for.body> to i64)) + %a)
 ; LAA-NEXT: --> {((2 * (zext i32 (2 * (trunc i64 %N to i32)) to i64)) + %a),+,-4}<%for.body>
 
 ; LV-LABEL: f2
 ; LV-LABEL: for.body.lver.check
-; LV: [[PredCheck0:%[^ ]*]] = icmp ne i128
-; LV: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
-; LV: [[PredCheck1:%[^ ]*]] = icmp ne i128
+
+; LV: [[OFMul:%[^ ]*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[BETrunc:%[^ ]*]])
+; LV-NEXT: [[OFMulResult:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 0
+; LV-NEXT: [[OFMulOverflow:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 1
+; LV-NEXT: [[AddEnd:%[^ ]*]] = add i32 [[Start:%[^ ]*]], [[OFMulResult]]
+; LV-NEXT: [[SubEnd:%[^ ]*]] = sub i32 [[Start]], [[OFMulResult]]
+; LV-NEXT: [[CmpNeg:%[^ ]*]] = icmp ugt i32 [[SubEnd]], [[Start]]
+; LV-NEXT: [[CmpPos:%[^ ]*]] = icmp ult i32 [[AddEnd]], [[Start]]
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 true, i1 [[CmpNeg]], i1 [[CmpPos]]
+; LV-NEXT: [[BECheck:%[^ ]*]] = icmp ugt i64 [[BE]], 4294967295
+; LV-NEXT: [[CheckOr0:%[^ ]*]] = or i1 [[Cmp]], [[BECheck]]
+; LV-NEXT: [[PredCheck0:%[^ ]*]] = or i1 [[CheckOr0]], [[OFMulOverflow]]
+
+; LV-NEXT: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
+
+; LV: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE]])
+; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0
+; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1
+; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 [[Start:%[^ ]*]], [[OFMulResult1]]
+; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 [[Start]], [[OFMulResult1]]
+; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], [[Start]]
+; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], [[Start]]
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 true, i1 [[CmpNeg1]], i1 [[CmpPos1]]
+; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]]
+
 ; LV: [[FinalCheck:%[^ ]*]] = or i1 [[Or0]], [[PredCheck1]]
 ; LV: br i1 [[FinalCheck]], label %for.body.ph.lver.orig, label %for.body.ph
 define void @f2(i16* noalias %a,
                 i16* noalias %b, i64 %N) {
 entry:
   %TruncN = trunc i64 %N to i32
   br label %for.body
 

 ;     i64 {0,+,2}<%for.body>
 
 ; LAA: [PSE]  %arrayidxA = getelementptr i16, i16* %a, i64 %mul_ext:
 ; LAA-NEXT: ((2 * (sext i32 {0,+,2}<%for.body> to i64)) + %a)
 ; LAA-NEXT: --> {%a,+,4}<%for.body>
 
 ; LV-LABEL: f3
 ; LV-LABEL: for.body.lver.check
-; LV: [[PredCheck0:%[^ ]*]] = icmp ne i128
-; LV: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
-; LV: [[PredCheck1:%[^ ]*]] = icmp ne i128
+
+; LV: [[OFMul:%[^ ]*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[BETrunc:%[^ ]*]])
+; LV-NEXT: [[OFMulResult:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 0
+; LV-NEXT: [[OFMulOverflow:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 1
+; LV-NEXT: [[AddEnd:%[^ ]*]] = add i32 0, [[OFMulResult]]
+; LV-NEXT: [[SubEnd:%[^ ]*]] = sub i32 0, [[OFMulResult]]
+; LV-NEXT: [[CmpNeg:%[^ ]*]] = icmp sgt i32 [[SubEnd]], 0
+; LV-NEXT: [[CmpPos:%[^ ]*]] = icmp slt i32 [[AddEnd]], 0
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 false, i1 [[CmpNeg]], i1 [[CmpPos]]
+; LV-NEXT: [[BECheck:%[^ ]*]] = icmp ugt i64 [[BE]], 4294967295
+; LV-NEXT: [[CheckOr0:%[^ ]*]] = or i1 [[Cmp]], [[BECheck]]
+; LV-NEXT: [[PredCheck0:%[^ ]*]] = or i1 [[CheckOr0]], [[OFMulOverflow]]
+
+; LV-NEXT: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
+
+; LV: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE:%[^ ]*]])
+; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0
+; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1
+; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 %a2, [[OFMulResult1]]
+; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 %a2, [[OFMulResult1]]
+; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], %a2
+; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], %a2
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 false, i1 [[CmpNeg1]], i1 [[CmpPos1]]
+; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]]
+
 ; LV: [[FinalCheck:%[^ ]*]] = or i1 [[Or0]], [[PredCheck1]]
 ; LV: br i1 [[FinalCheck]], label %for.body.ph.lver.orig, label %for.body.ph
 define void @f3(i16* noalias %a,
                 i16* noalias %b, i64 %N) {
 entry:
   br label %for.body
 
 for.body:                                         ; preds = %for.body, %entry

 ;     i64 {sext i32 (2 * (trunc i64 %N to i32)) to i64,+,-2}<%for.body>
 
 ; LAA: [PSE]  %arrayidxA = getelementptr i16, i16* %a, i64 %mul_ext:
 ; LAA-NEXT: ((2 * (sext i32 {(2 * (trunc i64 %N to i32)),+,-2}<%for.body> to i64)) + %a)
 ; LAA-NEXT: --> {((2 * (sext i32 (2 * (trunc i64 %N to i32)) to i64)) + %a),+,-4}<%for.body>
 
 ; LV-LABEL: f4
 ; LV-LABEL: for.body.lver.check
-; LV: [[PredCheck0:%[^ ]*]] = icmp ne i128
-; LV: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
-; LV: [[PredCheck1:%[^ ]*]] = icmp ne i128
+
+; LV: [[OFMul:%[^ ]*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[BETrunc:%[^ ]*]])
+; LV-NEXT: [[OFMulResult:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 0
+; LV-NEXT: [[OFMulOverflow:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 1
+; LV-NEXT: [[AddEnd:%[^ ]*]] = add i32 [[Start:%[^ ]*]], [[OFMulResult]]
+; LV-NEXT: [[SubEnd:%[^ ]*]] = sub i32 [[Start]], [[OFMulResult]]
+; LV-NEXT: [[CmpNeg:%[^ ]*]] = icmp sgt i32 [[SubEnd]], [[Start]]
+; LV-NEXT: [[CmpPos:%[^ ]*]] = icmp slt i32 [[AddEnd]], [[Start]]
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 true, i1 [[CmpNeg]], i1 [[CmpPos]]
+; LV-NEXT: [[BECheck:%[^ ]*]] = icmp ugt i64 [[BE]], 4294967295
+; LV-NEXT: [[CheckOr0:%[^ ]*]] = or i1 [[Cmp]], [[BECheck]]
+; LV-NEXT: [[PredCheck0:%[^ ]*]] = or i1 [[CheckOr0]], [[OFMulOverflow]]
+
+; LV-NEXT: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
+
+; LV: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE:%[^ ]*]])
+; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0
+; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1
+; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 [[Start:%[^ ]*]], [[OFMulResult1]]
+; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 [[Start]], [[OFMulResult1]]
+; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], [[Start]]
+; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], [[Start]]
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 true, i1 [[CmpNeg1]], i1 [[CmpPos1]]
+; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]]
+
 ; LV: [[FinalCheck:%[^ ]*]] = or i1 [[Or0]], [[PredCheck1]]
 ; LV: br i1 [[FinalCheck]], label %for.body.ph.lver.orig, label %for.body.ph
 define void @f4(i16* noalias %a,
                 i16* noalias %b, i64 %N) {
 entry:
   %TruncN = trunc i64 %N to i32
   br label %for.body
 

 ; LAA-NEXT: {((2 * (sext i32 (2 * (trunc i64 %N to i32)) to i64)) + %a),+,-4}<%for.body> Added Flags: <nusw>
 
 ; LAA: [PSE]  %arrayidxA = getelementptr inbounds i16, i16* %a, i32 %mul:
 ; LAA-NEXT: ((2 * (sext i32 {(2 * (trunc i64 %N to i32)),+,-2}<%for.body> to i64))<nsw> + %a)<nsw>
 ; LAA-NEXT: --> {((2 * (sext i32 (2 * (trunc i64 %N to i32)) to i64)) + %a),+,-4}<%for.body>
 
 ; LV-LABEL: f5
 ; LV-LABEL: for.body.lver.check
+; LV: [[OFMul:%[^ ]*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 2, i32 [[BETrunc:%[^ ]*]])
+; LV-NEXT: [[OFMulResult:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 0
+; LV-NEXT: [[OFMulOverflow:%[^ ]*]] = extractvalue { i32, i1 } [[OFMul]], 1
+; LV-NEXT: [[AddEnd:%[^ ]*]] = add i32 [[Start:%[^ ]*]], [[OFMulResult]]
+; LV-NEXT: [[SubEnd:%[^ ]*]] = sub i32 [[Start]], [[OFMulResult]]
+; LV-NEXT: [[CmpNeg:%[^ ]*]] = icmp sgt i32 [[SubEnd]], [[Start]]
+; LV-NEXT: [[CmpPos:%[^ ]*]] = icmp slt i32 [[AddEnd]], [[Start]]
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 true, i1 [[CmpNeg]], i1 [[CmpPos]]
+; LV-NEXT: [[BECheck:%[^ ]*]] = icmp ugt i64 [[BE]], 4294967295
+; LV-NEXT: [[CheckOr0:%[^ ]*]] = or i1 [[Cmp]], [[BECheck]]
+; LV-NEXT: [[PredCheck0:%[^ ]*]] = or i1 [[CheckOr0]], [[OFMulOverflow]]
+
+; LV-NEXT: [[Or0:%[^ ]*]] = or i1 false, [[PredCheck0]]
+
+; LV: [[OFMul1:%[^ ]*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 4, i64 [[BE:%[^ ]*]])
+; LV-NEXT: [[OFMulResult1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 0
+; LV-NEXT: [[OFMulOverflow1:%[^ ]*]] = extractvalue { i64, i1 } [[OFMul1]], 1
+; LV-NEXT: [[AddEnd1:%[^ ]*]] = add i64 [[Start:%[^ ]*]], [[OFMulResult1]]
+; LV-NEXT: [[SubEnd1:%[^ ]*]] = sub i64 [[Start]], [[OFMulResult1]]
+; LV-NEXT: [[CmpNeg1:%[^ ]*]] = icmp ugt i64 [[SubEnd1]], [[Start]]
+; LV-NEXT: [[CmpPos1:%[^ ]*]] = icmp ult i64 [[AddEnd1]], [[Start]]
+; LV-NEXT: [[Cmp:%[^ ]*]] = select i1 true, i1 [[CmpNeg1]], i1 [[CmpPos1]]
+; LV-NEXT: [[PredCheck1:%[^ ]*]] = or i1 [[Cmp]], [[OFMulOverflow1]]
+
+; LV: [[FinalCheck:%[^ ]*]] = or i1 [[Or0]], [[PredCheck1]]
+; LV: br i1 [[FinalCheck]], label %for.body.ph.lver.orig, label %for.body.ph
 define void @f5(i16* noalias %a,
                 i16* noalias %b, i64 %N) {
 entry:
   %TruncN = trunc i64 %N to i32
   br label %for.body
 
 for.body:                                         ; preds = %for.body, %entry
   %ind = phi i64 [ 0, %entry ], [ %inc, %for.body ]