[LAA] Use PSE::getSymbolicMaxBackedgeTakenCount. (#93499)

Update LAA to use PSE::getSymbolicMaxBackedgeTakenCount which returns
the minimum of the countable exits.

When analyzing dependences and computing runtime checks, we need the
smallest upper bound on the number of iterations. In terms of memory
safety, it shouldn't matter if any uncomputable exits leave the loop,
as long as we prove that there are no dependences given the minimum of
the countable exits. The same should apply also for generating runtime
checks.

Note that this shifts the responsiblity of checking whether all exit
counts are computable or handling early-exits to the users of LAA.

Depends on #93498

PR: #93499

Author: fhahn

llvm/lib/Analysis/LoopAccessAnalysis.cpp

@@ -214,7 +214,7 @@ getStartAndEndForAccess(const Loop *Lp, const SCEV *PtrExpr, Type *AccessTy,
   if (SE->isLoopInvariant(PtrExpr, Lp)) {
     ScStart = ScEnd = PtrExpr;
   } else if (auto *AR = dyn_cast<SCEVAddRecExpr>(PtrExpr)) {
-    const SCEV *Ex = PSE.getBackedgeTakenCount();
+    const SCEV *Ex = PSE.getSymbolicMaxBackedgeTakenCount();
 
     ScStart = AR->getStart();
     ScEnd = AR->evaluateAtIteration(Ex, *SE);
@@ -1796,28 +1796,28 @@ void MemoryDepChecker::mergeInStatus(VectorizationSafetyStatus S) {
 /// Given a dependence-distance \p Dist between two
 /// memory accesses, that have strides in the same direction whose absolute
 /// value of the maximum stride is given in \p MaxStride, and that have the same
-/// type size \p TypeByteSize, in a loop whose takenCount is \p
-/// BackedgeTakenCount, check if it is possible to prove statically that the
-/// dependence distance is larger than the range that the accesses will travel
-/// through the execution of the loop. If so, return true; false otherwise. This
-/// is useful for example in loops such as the following (PR31098):
+/// type size \p TypeByteSize, in a loop whose maximum backedge taken count is
+/// \p MaxBTC, check if it is possible to prove statically that the dependence
+/// distance is larger than the range that the accesses will travel through the
+/// execution of the loop. If so, return true; false otherwise. This is useful
+/// for example in loops such as the following (PR31098):
 ///     for (i = 0; i < D; ++i) {
 ///                = out[i];
 ///       out[i+D] =
 ///     }
 static bool isSafeDependenceDistance(const DataLayout &DL, ScalarEvolution &SE,
-                                     const SCEV &BackedgeTakenCount,
-                                     const SCEV &Dist, uint64_t MaxStride,
+                                     const SCEV &MaxBTC, const SCEV &Dist,
+                                     uint64_t MaxStride,
                                      uint64_t TypeByteSize) {
 
   // If we can prove that
-  //      (**) |Dist| > BackedgeTakenCount * Step
+  //      (**) |Dist| > MaxBTC * Step
   // where Step is the absolute stride of the memory accesses in bytes,
   // then there is no dependence.
   //
   // Rationale:
   // We basically want to check if the absolute distance (|Dist/Step|)
-  // is >= the loop iteration count (or > BackedgeTakenCount).
+  // is >= the loop iteration count (or > MaxBTC).
   // This is equivalent to the Strong SIV Test (Practical Dependence Testing,
   // Section 4.2.1); Note, that for vectorization it is sufficient to prove
   // that the dependence distance is >= VF; This is checked elsewhere.
@@ -1828,8 +1828,8 @@ static bool isSafeDependenceDistance(const DataLayout &DL, ScalarEvolution &SE,
   // also guarantees that distance >= VF.
   //
   const uint64_t ByteStride = MaxStride * TypeByteSize;
-  const SCEV *Step = SE.getConstant(BackedgeTakenCount.getType(), ByteStride);
-  const SCEV *Product = SE.getMulExpr(&BackedgeTakenCount, Step);
+  const SCEV *Step = SE.getConstant(MaxBTC.getType(), ByteStride);
+  const SCEV *Product = SE.getMulExpr(&MaxBTC, Step);
 
   const SCEV *CastedDist = &Dist;
   const SCEV *CastedProduct = Product;
@@ -1844,13 +1844,13 @@ static bool isSafeDependenceDistance(const DataLayout &DL, ScalarEvolution &SE,
   else
     CastedDist = SE.getNoopOrSignExtend(&Dist, Product->getType());
 
-  // Is  Dist - (BackedgeTakenCount * Step) > 0 ?
+  // Is  Dist - (MaxBTC * Step) > 0 ?
   // (If so, then we have proven (**) because |Dist| >= Dist)
   const SCEV *Minus = SE.getMinusSCEV(CastedDist, CastedProduct);
   if (SE.isKnownPositive(Minus))
     return true;
 
-  // Second try: Is  -Dist - (BackedgeTakenCount * Step) > 0 ?
+  // Second try: Is  -Dist - (MaxBTC * Step) > 0 ?
   // (If so, then we have proven (**) because |Dist| >= -1*Dist)
   const SCEV *NegDist = SE.getNegativeSCEV(CastedDist);
   Minus = SE.getMinusSCEV(NegDist, CastedProduct);
@@ -2034,12 +2034,13 @@ MemoryDepChecker::Dependence::DepType MemoryDepChecker::isDependent(
   uint64_t MaxStride = std::max(StrideA, StrideB);
 
   // If the distance between the acecsses is larger than their maximum absolute
-  // stride multiplied by the backedge taken count, the accesses are independet,
-  // i.e. they are far enough appart that accesses won't access the same
-  // location across all loop ierations.
-  if (HasSameSize &&
-      isSafeDependenceDistance(DL, SE, *(PSE.getBackedgeTakenCount()), *Dist,
-                               MaxStride, TypeByteSize))
+  // stride multiplied by the symbolic maximum backedge taken count (which is an
+  // upper bound of the number of iterations), the accesses are independet, i.e.
+  // they are far enough appart that accesses won't access the same location
+  // across all loop ierations.
+  if (HasSameSize && isSafeDependenceDistance(
+                         DL, SE, *(PSE.getSymbolicMaxBackedgeTakenCount()),
+                         *Dist, MaxStride, TypeByteSize))
     return Dependence::NoDep;
 
   const SCEVConstant *C = dyn_cast<SCEVConstant>(Dist);
@@ -2374,8 +2375,10 @@ bool LoopAccessInfo::canAnalyzeLoop() {
     return false;
   }
 
-  // ScalarEvolution needs to be able to find the exit count.
-  const SCEV *ExitCount = PSE->getBackedgeTakenCount();
+  // ScalarEvolution needs to be able to find the symbolic max backedge taken
+  // count, which is an upper bound on the number of loop iterations. The loop
+  // may execute fewer iterations, if it exits via an uncountable exit.
+  const SCEV *ExitCount = PSE->getSymbolicMaxBackedgeTakenCount();
   if (isa<SCEVCouldNotCompute>(ExitCount)) {
     recordAnalysis("CantComputeNumberOfIterations")
         << "could not determine number of loop iterations";
@@ -2984,25 +2987,25 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
   // of various possible stride specializations, considering the alternatives
   // of using gather/scatters (if available).
 
-  const SCEV *BETakenCount = PSE->getBackedgeTakenCount();
+  const SCEV *MaxBTC = PSE->getSymbolicMaxBackedgeTakenCount();
 
-  // Match the types so we can compare the stride and the BETakenCount.
+  // Match the types so we can compare the stride and the MaxBTC.
   // The Stride can be positive/negative, so we sign extend Stride;
-  // The backedgeTakenCount is non-negative, so we zero extend BETakenCount.
+  // The backedgeTakenCount is non-negative, so we zero extend MaxBTC.
   const DataLayout &DL = TheLoop->getHeader()->getModule()->getDataLayout();
   uint64_t StrideTypeSizeBits = DL.getTypeSizeInBits(StrideExpr->getType());
-  uint64_t BETypeSizeBits = DL.getTypeSizeInBits(BETakenCount->getType());
+  uint64_t BETypeSizeBits = DL.getTypeSizeInBits(MaxBTC->getType());
   const SCEV *CastedStride = StrideExpr;
-  const SCEV *CastedBECount = BETakenCount;
+  const SCEV *CastedBECount = MaxBTC;
   ScalarEvolution *SE = PSE->getSE();
   if (BETypeSizeBits >= StrideTypeSizeBits)
-    CastedStride = SE->getNoopOrSignExtend(StrideExpr, BETakenCount->getType());
+    CastedStride = SE->getNoopOrSignExtend(StrideExpr, MaxBTC->getType());
   else
-    CastedBECount = SE->getZeroExtendExpr(BETakenCount, StrideExpr->getType());
+    CastedBECount = SE->getZeroExtendExpr(MaxBTC, StrideExpr->getType());
   const SCEV *StrideMinusBETaken = SE->getMinusSCEV(CastedStride, CastedBECount);
   // Since TripCount == BackEdgeTakenCount + 1, checking:
   // "Stride >= TripCount" is equivalent to checking:
-  // Stride - BETakenCount > 0
+  // Stride - MaxBTC> 0
   if (SE->isKnownPositive(StrideMinusBETaken)) {
     LLVM_DEBUG(
         dbgs() << "LAA: Stride>=TripCount; No point in versioning as the "

llvm/lib/Transforms/Utils/ScalarEvolutionExpander.cpp

@@ -2079,7 +2079,7 @@ Value *SCEVExpander::generateOverflowCheck(const SCEVAddRecExpr *AR,
   // FIXME: It is highly suspicious that we're ignoring the predicates here.
   SmallVector<const SCEVPredicate *, 4> Pred;
   const SCEV *ExitCount =
-      SE.getPredicatedBackedgeTakenCount(AR->getLoop(), Pred);
+      SE.getPredicatedSymbolicMaxBackedgeTakenCount(AR->getLoop(), Pred);
 
   assert(!isa<SCEVCouldNotCompute>(ExitCount) && "Invalid loop count");
 

llvm/lib/Transforms/Vectorize/LoopVectorizationLegality.cpp

@@ -1506,6 +1506,16 @@ bool LoopVectorizationLegality::canVectorize(bool UseVPlanNativePath) {
       return false;
   }
 
+  if (isa<SCEVCouldNotCompute>(PSE.getBackedgeTakenCount())) {
+    reportVectorizationFailure("could not determine number of loop iterations",
+                               "could not determine number of loop iterations",
+                               "CantComputeNumberOfIterations", ORE, TheLoop);
+    if (DoExtraAnalysis)
+      Result = false;
+    else
+      return false;
+  }
+
   LLVM_DEBUG(dbgs() << "LV: We can vectorize this loop"
                     << (LAI->getRuntimePointerChecking()->Need
                             ? " (with a runtime bound check)"

llvm/test/Analysis/LoopAccessAnalysis/early-exit-runtime-checks.ll

@@ -4,10 +4,21 @@
 define void @all_exits_dominate_latch_countable_exits_at_most_500_iterations(ptr %A, ptr %B) {
 ; CHECK-LABEL: 'all_exits_dominate_latch_countable_exits_at_most_500_iterations'
 ; CHECK-NEXT:    loop.header:
-; CHECK-NEXT:      Report: could not determine number of loop iterations
+; CHECK-NEXT:      Memory dependences are safe with run-time checks
 ; CHECK-NEXT:      Dependences:
 ; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Check 0:
+; CHECK-NEXT:        Comparing group ([[GRP1:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.B = getelementptr inbounds i32, ptr %B, i64 %iv
+; CHECK-NEXT:        Against group ([[GRP2:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.A = getelementptr inbounds i32, ptr %A, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
+; CHECK-NEXT:        Group [[GRP1]]:
+; CHECK-NEXT:          (Low: %B High: (2000 + %B))
+; CHECK-NEXT:            Member: {%B,+,4}<nuw><%loop.header>
+; CHECK-NEXT:        Group [[GRP2]]:
+; CHECK-NEXT:          (Low: %A High: (2000 + %A))
+; CHECK-NEXT:            Member: {%A,+,4}<nuw><%loop.header>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:
@@ -53,10 +64,21 @@ e.2:
 define i32 @all_exits_dominate_latch_countable_exits_at_most_1000_iterations(ptr %A, ptr %B) {
 ; CHECK-LABEL: 'all_exits_dominate_latch_countable_exits_at_most_1000_iterations'
 ; CHECK-NEXT:    loop.header:
-; CHECK-NEXT:      Report: could not determine number of loop iterations
+; CHECK-NEXT:      Memory dependences are safe with run-time checks
 ; CHECK-NEXT:      Dependences:
 ; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Check 0:
+; CHECK-NEXT:        Comparing group ([[GRP3:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.B = getelementptr inbounds i32, ptr %B, i64 %iv
+; CHECK-NEXT:        Against group ([[GRP4:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.A = getelementptr inbounds i32, ptr %A, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
+; CHECK-NEXT:        Group [[GRP3]]:
+; CHECK-NEXT:          (Low: %B High: (4004 + %B))
+; CHECK-NEXT:            Member: {%B,+,4}<nuw><%loop.header>
+; CHECK-NEXT:        Group [[GRP4]]:
+; CHECK-NEXT:          (Low: %A High: (4004 + %A))
+; CHECK-NEXT:            Member: {%A,+,4}<nuw><%loop.header>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:
@@ -145,10 +167,21 @@ e.2:
 define i32 @b3_does_not_dominate_latch(ptr %A, ptr %B) {
 ; CHECK-LABEL: 'b3_does_not_dominate_latch'
 ; CHECK-NEXT:    loop.header:
-; CHECK-NEXT:      Report: could not determine number of loop iterations
+; CHECK-NEXT:      Memory dependences are safe with run-time checks
 ; CHECK-NEXT:      Dependences:
 ; CHECK-NEXT:      Run-time memory checks:
+; CHECK-NEXT:      Check 0:
+; CHECK-NEXT:        Comparing group ([[GRP5:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.B = getelementptr inbounds i32, ptr %B, i64 %iv
+; CHECK-NEXT:        Against group ([[GRP6:0x[0-9a-f]+]]):
+; CHECK-NEXT:          %gep.A = getelementptr inbounds i32, ptr %A, i64 %iv
 ; CHECK-NEXT:      Grouped accesses:
+; CHECK-NEXT:        Group [[GRP5]]:
+; CHECK-NEXT:          (Low: %B High: (4004 + %B))
+; CHECK-NEXT:            Member: {%B,+,4}<nuw><%loop.header>
+; CHECK-NEXT:        Group [[GRP6]]:
+; CHECK-NEXT:          (Low: %A High: (4004 + %A))
+; CHECK-NEXT:            Member: {%A,+,4}<nuw><%loop.header>
 ; CHECK-EMPTY:
 ; CHECK-NEXT:      Non vectorizable stores to invariant address were not found in loop.
 ; CHECK-NEXT:      SCEV assumptions:

llvm/test/Transforms/LoopDistribute/early-exit.ll

@@ -0,0 +1,96 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; REQUIRES: x86-registered-target
+; RUN: opt -aa-pipeline=basic-aa -passes=loop-distribute -enable-loop-distribute -verify-loop-info -verify-dom-info -S %s | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.10.0"
+
+@B = common global ptr null, align 8
+@A = common global ptr null, align 8
+@C = common global ptr null, align 8
+@D = common global ptr null, align 8
+@E = common global ptr null, align 8
+
+define void @f() {
+; CHECK-LABEL: define void @f() {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[A:%.*]] = load ptr, ptr @A, align 8
+; CHECK-NEXT:    [[B:%.*]] = load ptr, ptr @B, align 8
+; CHECK-NEXT:    [[C:%.*]] = load ptr, ptr @C, align 8
+; CHECK-NEXT:    [[D:%.*]] = load ptr, ptr @D, align 8
+; CHECK-NEXT:    [[E:%.*]] = load ptr, ptr @E, align 8
+; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[IND:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[ADD:%.*]], %[[LATCH:.*]] ]
+; CHECK-NEXT:    [[ARRAYIDXA:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[IND]]
+; CHECK-NEXT:    [[LOADA:%.*]] = load i32, ptr [[ARRAYIDXA]], align 4
+; CHECK-NEXT:    [[ARRAYIDXB:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[IND]]
+; CHECK-NEXT:    [[LOADB:%.*]] = load i32, ptr [[ARRAYIDXB]], align 4
+; CHECK-NEXT:    [[UNCOUNTABLE_C:%.*]] = icmp eq i32 [[LOADB]], 10
+; CHECK-NEXT:    br i1 [[UNCOUNTABLE_C]], label %[[FOR_END:.*]], label %[[LATCH]]
+; CHECK:       [[LATCH]]:
+; CHECK-NEXT:    [[MULA:%.*]] = mul i32 [[LOADB]], [[LOADA]]
+; CHECK-NEXT:    [[ADD]] = add nuw nsw i64 [[IND]], 1
+; CHECK-NEXT:    [[ARRAYIDXA_PLUS_4:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[ADD]]
+; CHECK-NEXT:    store i32 [[MULA]], ptr [[ARRAYIDXA_PLUS_4]], align 4
+; CHECK-NEXT:    [[ARRAYIDXD:%.*]] = getelementptr inbounds i32, ptr [[D]], i64 [[IND]]
+; CHECK-NEXT:    [[LOADD:%.*]] = load i32, ptr [[ARRAYIDXD]], align 4
+; CHECK-NEXT:    [[ARRAYIDXE:%.*]] = getelementptr inbounds i32, ptr [[E]], i64 [[IND]]
+; CHECK-NEXT:    [[LOADE:%.*]] = load i32, ptr [[ARRAYIDXE]], align 4
+; CHECK-NEXT:    [[MULC:%.*]] = mul i32 [[LOADD]], [[LOADE]]
+; CHECK-NEXT:    [[ARRAYIDXC:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[IND]]
+; CHECK-NEXT:    store i32 [[MULC]], ptr [[ARRAYIDXC]], align 4
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[ADD]], 20
+; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]]
+; CHECK:       [[FOR_END]]:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = load ptr, ptr @A, align 8
+  %b = load ptr, ptr @B, align 8
+  %c = load ptr, ptr @C, align 8
+  %d = load ptr, ptr @D, align 8
+  %e = load ptr, ptr @E, align 8
+  br label %for.body
+
+for.body:
+  %ind = phi i64 [ 0, %entry ], [ %add, %latch ]
+
+  %arrayidxA = getelementptr inbounds i32, ptr %a, i64 %ind
+  %loadA = load i32, ptr %arrayidxA, align 4
+
+  %arrayidxB = getelementptr inbounds i32, ptr %b, i64 %ind
+  %loadB = load i32, ptr %arrayidxB, align 4
+  %uncountable.c = icmp eq i32 %loadB, 10
+  br i1 %uncountable.c, label %for.end, label %latch
+
+latch:
+  %mulA = mul i32 %loadB, %loadA
+
+  %add = add nuw nsw i64 %ind, 1
+  %arrayidxA_plus_4 = getelementptr inbounds i32, ptr %a, i64 %add
+  store i32 %mulA, ptr %arrayidxA_plus_4, align 4
+
+  %arrayidxD = getelementptr inbounds i32, ptr %d, i64 %ind
+  %loadD = load i32, ptr %arrayidxD, align 4
+
+  %arrayidxE = getelementptr inbounds i32, ptr %e, i64 %ind
+  %loadE = load i32, ptr %arrayidxE, align 4
+
+  %mulC = mul i32 %loadD, %loadE
+
+  %arrayidxC = getelementptr inbounds i32, ptr %c, i64 %ind
+  store i32 %mulC, ptr %arrayidxC, align 4
+
+  %exitcond = icmp eq i64 %add, 20
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret void
+}
+
+attributes #0 = { nounwind readnone convergent }
+attributes #1 = { nounwind convergent }
+
+!0 = distinct !{!0, !1}
+!1 = !{!"llvm.loop.distribute.enable", i1 true}

llvm/test/Transforms/LoopLoadElim/early-exit.ll

@@ -0,0 +1,61 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=loop-load-elim -S %s | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+
+define void @f(ptr %A, ptr %B, ptr %C, i64 %N) {
+; CHECK-LABEL: define void @f(
+; CHECK-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[C:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT:  [[FOR_BODY_LVER_CHECK:.*]]:
+; CHECK-NEXT:    br label %[[FOR_BODY:.*]]
+; CHECK:       [[FOR_BODY]]:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 0, %[[FOR_BODY_LVER_CHECK]] ], [ [[INDVARS_IV_NEXT:%.*]], %[[LATCH:.*]] ]
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[AIDX_NEXT:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
+; CHECK-NEXT:    [[BIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[CIDX:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[AIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[B:%.*]] = load i32, ptr [[BIDX]], align 4
+; CHECK-NEXT:    [[UNCOUNTABLE_C:%.*]] = icmp eq i32 [[B]], 10
+; CHECK-NEXT:    br i1 [[UNCOUNTABLE_C]], label %[[LATCH]], label %[[FOR_END:.*]]
+; CHECK:       [[LATCH]]:
+; CHECK-NEXT:    [[A_P1:%.*]] = add i32 [[B]], 2
+; CHECK-NEXT:    store i32 [[A_P1]], ptr [[AIDX_NEXT]], align 4
+; CHECK-NEXT:    [[A:%.*]] = load i32, ptr [[AIDX]], align 1
+; CHECK-NEXT:    [[C:%.*]] = mul i32 [[A]], 2
+; CHECK-NEXT:    store i32 [[C]], ptr [[CIDX]], align 4
+; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[EXITCOND]], label %[[FOR_END]], label %[[FOR_BODY]]
+; CHECK:       [[FOR_END]]:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %for.body, %entry
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %latch ]
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+
+  %Aidx_next = getelementptr inbounds i32, ptr %A, i64 %indvars.iv.next
+  %Bidx = getelementptr inbounds i32, ptr %B, i64 %indvars.iv
+  %Cidx = getelementptr inbounds i32, ptr %C, i64 %indvars.iv
+  %Aidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
+
+  %b = load i32, ptr %Bidx, align 4
+  %uncountable.c = icmp eq i32 %b, 10
+  br i1 %uncountable.c, label  %latch, label %for.end
+
+latch:
+  %a_p1 = add i32 %b, 2
+  store i32 %a_p1, ptr %Aidx_next, align 4
+
+  %a = load i32, ptr %Aidx, align 1
+  %c = mul i32 %a, 2
+  store i32 %c, ptr %Cidx, align 4
+
+  %exitcond = icmp eq i64 %indvars.iv.next, %N
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body
+  ret void
+}