[OpenMP][OpenMPIRBuilder] Implement loop unrolling.

[Meinersbur]

Summary

Add methods for loop unrolling to the OpenMPIRBuilder class and use them in Clang if -fopenmp-enable-irbuilder is enabled. The unrolling methods are:

• unrollLoopFull
• unrollLoopPartial
• unrollLoopHeuristic

unrollLoopPartial and unrollLoopHeuristic can use compiler heuristics to automatically determine the unroll factor. If possible, that is if no CanonicalLoopInfo is required to pass to another method, metadata for LLVM's LoopUnrollPass is added. Otherwise the unroll factor is determined using the same heurstics as user by LoopUnrollPass. Not requiring a CanonicalLoopInfo, especially with unrollLoopHeuristic allows greater flexibility.

With full unrolling and partial unrolling with known unroll factor, instead of duplicating instructions by the OpenMPIRBuilder, the full unroll is still delegated to the LoopUnrollPass. In case of partial unrolling the loop is first tiled using the existing tileLoops methods, then the inner loop fully unrolled using the same mechanism.

[Meinersbur]

• Use TargetMachine for unroll factor heuristic

[jdoerfert]

LG, minor comments below.

clang/lib/CodeGen/CGStmtOpenMP.cpp

2636–2640

make it an unreachable or sink the assert and combine the conditions. Later allows to remove all braces.

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

2096

Nit: Add an assert for latch to exist. Just for my sanity ;)

2138

Should we cache it in the OpenMPIRBuilder?

2260

2294

This code matches the pattern we have in the 2 members above. Can we make it a member as well and make it clear in the name of all 3 that we use metadata and the unroller pass? No strong feelings, more of an idea.

[kiranchandramohan]

I have a few minor questions.

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

2074

Nit: In the body, it seems we are prepending.

2187–2188

Should this be settable for experimentation or additional control? If not can you provide an explanation for these values?

llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp

1667

Are we not calling ompbuilder finalize or verify because it is only adding metadata?

1747

Should we add tests for workshare loops with unroll?

[Meinersbur]

• Address reviews

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

2074

In the body, first we append Existing->operands(), then Properties. I.e. we append the Properties passes as an argument to after the existing properties to form a new list NewLoopProperties.

2187–2188

Made them configurable using the new -openmp-ir-builder-unroll-threshold-factor switch.

I selected 1.5 as default to make it match approximately the unroll factor the LoopUnrollPass would derive in an -O3 pass pass pipeline. No large-scale experiments, just a loop that LoopUnrollPass would unroll by a factor of 4 (out of possible: 1 (no unrolling), 2, 4, 8) and make #pragma omp unroll partial also unroll by a factor of 4.

2294

I think the API should abstract over how the unrolling is performed, frontends should not need to be concerned about how unrolling is actually applied. The metadata version is preferred unless we cannot use it because we need the unrolled loop's CFG wrapped by CanonicalLoopInfo.

I changed the parameters to make it more idiomatic that the caller only receives a CanonicalLoopInfo if requested, and that's the only purpose of the 4th parameter.

llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp

1667

Added

OMPBuilder.finalize();
EXPECT_FALSE(verifyModule(*M, &errs()));

to the unroll tests. Thanks for noticing.

1747

I think this is beyond the scope of these tests here. To test thoroughly, we'd have to test all combinations of loop transformations and loop-associated directives. The CanonicalLoopInfo::assertOK() should already check the invariants that ensure that the CLI can be used as input for other loop-associated directives. Additionally, the composition is tested with clang's tests.

[kiranchandramohan]

LGTM. Thanks for the responses.

[lkail]

This looks have broken build if -DBUILD_SHARED_LIBS=On is specified.

CMake Error: The inter-target dependency graph contains the following strongly connected component (cycle):
  "LLVMFrontendOpenMP" of type SHARED_LIBRARY
    depends on "LLVMPasses" (weak)
  "LLVMipo" of type SHARED_LIBRARY
    depends on "LLVMFrontendOpenMP" (weak)
  "LLVMCoroutines" of type SHARED_LIBRARY
    depends on "LLVMipo" (weak)
  "LLVMPasses" of type SHARED_LIBRARY
    depends on "LLVMCoroutines" (weak)
    depends on "LLVMipo" (weak)
At least one of these targets is not a STATIC_LIBRARY.  Cyclic dependencies are allowed only among static libraries.
CMake Generate step failed.  Build files cannot be regenerated correctly.

[david-arm]

Hi, this has broken the build for me too.

[Meinersbur]

Register required passes individually instead of asking the PassBuilder to register them. The PassBuilder sits in the LLVMPasses component. This creates a circular dependency since LLVMPasses requires LLVMipo containing the OpenMPOpt pass that uses the OpenMPIRBuilder.

[Meinersbur]

• Also register indirect dependencies

clang/include/clang/Sema/Sema.h

public:
 
   /// Initialization of captured region for OpenMP region.
   void ActOnOpenMPRegionStart(OpenMPDirectiveKind DKind, Scope *CurScope);
 
   /// Called for syntactical loops (ForStmt or CXXForRangeStmt) associated to
   /// an OpenMP loop directive.
   StmtResult ActOnOpenMPCanonicalLoop(Stmt *AStmt);
 
+  /// Process a canonical OpenMP loop nest that can either be a canonical
+  /// literal loop (ForStmt or CXXForRangeStmt), or the generated loop of an
+  /// OpenMP loop transformation construct.
+  StmtResult ActOnOpenMPLoopnest(Stmt *AStmt);
+
   /// End of OpenMP region.
   ///
   /// \param S Statement associated with the current OpenMP region.
   /// \param Clauses List of clauses for the current OpenMP region.
   ///
   /// \returns Statement for finished OpenMP region.
   StmtResult ActOnOpenMPRegionEnd(StmtResult S, ArrayRef<OMPClause *> Clauses);
   StmtResult ActOnOpenMPExecutableDirective(

clang/lib/CodeGen/CGStmtOpenMP.cpp

emitCapturedStmtCall(CodeGenFunction &ParentCGF, EmittedClosureTy Cap,
 
   return ParentCGF.Builder.CreateCall(Cap.first, EffectiveArgs);
 }
 
 llvm::CanonicalLoopInfo *
 CodeGenFunction::EmitOMPCollapsedCanonicalLoopNest(const Stmt *S, int Depth) {
   assert(Depth == 1 && "Nested loops with OpenMPIRBuilder not yet implemented");
 
+  // The caller is processing the loop-associated directive processing the \p
+  // Depth loops nested in \p S. Put the previous pending loop-associated
+  // directive to the stack. If the current loop-associated directive is a loop
+  // transformation directive, it will push its generated loops onto the stack
+  // such that together with the loops left here they form the combined loop
+  // nest for the parent loop-associated directive.
+  int ParentExpectedOMPLoopDepth = ExpectedOMPLoopDepth;
+  ExpectedOMPLoopDepth = Depth;
+
   EmitStmt(S);
   assert(OMPLoopNestStack.size() >= (size_t)Depth && "Found too few loops");
 
   // The last added loop is the outermost one.
-  return OMPLoopNestStack.back();
+  llvm::CanonicalLoopInfo *Result = OMPLoopNestStack.back();
+
+  // Pop the \p Depth loops requested by the call from that stack and restore
+  // the previous context.
+  OMPLoopNestStack.set_size(OMPLoopNestStack.size() - Depth);
+  ExpectedOMPLoopDepth = ParentExpectedOMPLoopDepth;
+
+  return Result;
 }
 
 void CodeGenFunction::EmitOMPCanonicalLoop(const OMPCanonicalLoop *S) {
   const Stmt *SyntacticalLoop = S->getLoopStmt();
   if (!getLangOpts().OpenMPIRBuilder) {
     // Ignore if OpenMPIRBuilder is not enabled.
     EmitStmt(SyntacticalLoop);
     return;

 
 void CodeGenFunction::EmitOMPTileDirective(const OMPTileDirective &S) {
   // Emit the de-sugared statement.
   OMPTransformDirectiveScopeRAII TileScope(*this, &S);
   EmitStmt(S.getTransformedStmt());
 }
 
 void CodeGenFunction::EmitOMPUnrollDirective(const OMPUnrollDirective &S) {
+  bool UseOMPIRBuilder = CGM.getLangOpts().OpenMPIRBuilder;
+
+  if (UseOMPIRBuilder) {
+    auto DL = SourceLocToDebugLoc(S.getBeginLoc());
+    const Stmt *Inner = S.getRawStmt();
+
+    // Consume nested loop. Clear the entire remaining loop stack because a
+    // fully unrolled loop is non-transformable. For partial unrolling the
+    // generated outer loop is pushed back to the stack.
+    llvm::CanonicalLoopInfo *CLI = EmitOMPCollapsedCanonicalLoopNest(Inner, 1);
+    OMPLoopNestStack.clear();
+
+    llvm::OpenMPIRBuilder &OMPBuilder = CGM.getOpenMPRuntime().getOMPBuilder();
+
+    bool NeedsUnrolledCLI = ExpectedOMPLoopDepth >= 1;
+    llvm::CanonicalLoopInfo *UnrolledCLI = nullptr;
+
+    if (S.hasClausesOfKind<OMPFullClause>()) {
+      assert(ExpectedOMPLoopDepth == 0);
+      OMPBuilder.unrollLoopFull(DL, CLI);
+    } else if (auto *PartialClause = S.getSingleClause<OMPPartialClause>()) {
+      uint64_t Factor = 0;
+      if (Expr *FactorExpr = PartialClause->getFactor()) {
+        Factor = FactorExpr->EvaluateKnownConstInt(getContext()).getZExtValue();
+        assert(Factor >= 1 && "Only positive factors are valid");
+      }
+      OMPBuilder.unrollLoopPartial(DL, CLI, Factor,
+                                   NeedsUnrolledCLI ? &UnrolledCLI : nullptr);
+    } else {
+      OMPBuilder.unrollLoopHeuristic(DL, CLI);
+    }
+
+    assert((!NeedsUnrolledCLI || UnrolledCLI) &&
+           "NeedsUnrolledCLI implies UnrolledCLI to be set");
+    if (UnrolledCLI)
+      OMPLoopNestStack.push_back(UnrolledCLI);
+
+    return;
+  }
+
   // This function is only called if the unrolled loop is not consumed by any
   // other loop-associated construct. Such a loop-associated construct will have
   // used the transformed AST.
 
   // Set the unroll metadata for the next emitted loop.
   LoopStack.setUnrollState(LoopAttributes::Enable);
 
   if (S.hasClausesOfKind<OMPFullClause>()) {

clang/lib/CodeGen/CodeGenFunction.h

public:
   /// OMPEmitOMPCanonicalLoop called by it will add its CanonicalLoopInfo to
   /// this stack when done. Entering a new loop requires clearing this list; it
   /// either means we start parsing a new loop nest (in which case the previous
   /// loop nest goes out of scope) or a second loop in the same level in which
   /// case it would be ambiguous into which of the two (or more) loops the loop
   /// nest would extend.
   SmallVector<llvm::CanonicalLoopInfo *, 4> OMPLoopNestStack;
 
+  /// Number of nested loop to be consumed by the last surrounding
+  /// loop-associated directive.
+  int ExpectedOMPLoopDepth = 0;
+
   // CodeGen lambda for loops and support for ordered clause
   typedef llvm::function_ref<void(CodeGenFunction &, const OMPLoopDirective &,
                                   JumpDest)>
       CodeGenLoopTy;
   typedef llvm::function_ref<void(CodeGenFunction &, SourceLocation,
                                   const unsigned, const bool)>
       CodeGenOrderedTy;
 

clang/lib/Parse/ParseOpenMP.cpp

if (HasAssociatedStatement) {
       // should have at least one compound statement scope within it.
       ParsingOpenMPDirectiveRAII NormalScope(*this, /*Value=*/false);
       {
         Sema::CompoundScopeRAII Scope(Actions);
         AssociatedStmt = ParseStatement();
 
         if (AssociatedStmt.isUsable() && isOpenMPLoopDirective(DKind) &&
             getLangOpts().OpenMPIRBuilder)
-          AssociatedStmt =
-              Actions.ActOnOpenMPCanonicalLoop(AssociatedStmt.get());
+          AssociatedStmt = Actions.ActOnOpenMPLoopnest(AssociatedStmt.get());
       }
       AssociatedStmt = Actions.ActOnOpenMPRegionEnd(AssociatedStmt, Clauses);
     } else if (DKind == OMPD_target_update || DKind == OMPD_target_enter_data ||
                DKind == OMPD_target_exit_data) {
       Actions.ActOnOpenMPRegionStart(DKind, getCurScope());
       AssociatedStmt = (Sema::CompoundScopeRAII(Actions),
                         Actions.ActOnCompoundStmt(Loc, Loc, llvm::None,
                                                   /*isStmtExpr=*/false));

clang/lib/Sema/SemaOpenMP.cpp

StmtResult Sema::ActOnOpenMPCanonicalLoop(Stmt *AStmt) {
   CapturedStmt *LoopVarFunc = buildLoopVarFunc(
       *this, LVTy, LogicalTy, CounterRef, Step, isa<CXXForRangeStmt>(AStmt));
   DeclRefExpr *LVRef = BuildDeclRefExpr(LUVDecl, LUVDecl->getType(), VK_LValue,
                                         {}, nullptr, nullptr, {}, nullptr);
   return OMPCanonicalLoop::create(getASTContext(), AStmt, DistanceFunc,
                                   LoopVarFunc, LVRef);
 }
 
+StmtResult Sema::ActOnOpenMPLoopnest(Stmt *AStmt) {
+  // Handle a literal loop.
+  if (isa<ForStmt>(AStmt) || isa<CXXForRangeStmt>(AStmt))
+    return ActOnOpenMPCanonicalLoop(AStmt);
+
+  // If not a literal loop, it must be the result of a loop transformation.
+  OMPExecutableDirective *LoopTransform = cast<OMPExecutableDirective>(AStmt);
+  assert(
+      isOpenMPLoopTransformationDirective(LoopTransform->getDirectiveKind()) &&
+      "Loop transformation directive expected");
+  return LoopTransform;
+}
+
 static ExprResult buildUserDefinedMapperRef(Sema &SemaRef, Scope *S,
                                             CXXScopeSpec &MapperIdScopeSpec,
                                             const DeclarationNameInfo &MapperId,
                                             QualType Type,
                                             Expr *UnresolvedMapper);
 
 /// Perform DFS through the structure/class data members trying to find
 /// member(s) with user-defined 'default' mapper and generate implicit map

clang/test/OpenMP/irbuilder_unroll_full.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_full(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_HEADER]]:
+// CHECK-NEXT:    %[[OMP_LOOP_IV:.+]] = phi i32 [ 0, %[[OMP_LOOP_PREHEADER]] ], [ %[[OMP_LOOP_NEXT:.+]], %[[OMP_LOOP_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_LOOP_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_COND]]:
+// CHECK-NEXT:    %[[OMP_LOOP_CMP:.+]] = icmp ult i32 %[[OMP_LOOP_IV]], %[[DOTCOUNT]]
+// CHECK-NEXT:    br i1 %[[OMP_LOOP_CMP]], label %[[OMP_LOOP_BODY:.+]], label %[[OMP_LOOP_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[OMP_LOOP_IV]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP3:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP4]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP3]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP5:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP6:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP7]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP6]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP5]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP9:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP10]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP9]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP11:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP11]]
+// CHECK-NEXT:    %[[TMP12:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP13:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP13]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP12]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_INC]]:
+// CHECK-NEXT:    %[[OMP_LOOP_NEXT]] = add nuw i32 %[[OMP_LOOP_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+void unroll_full(float *a, float *b, float *c, float *d) {
+#pragma omp unroll full
+  for (int i = 0; i < 2; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 2, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.full"}

clang/test/OpenMP/irbuilder_unroll_heuristic.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_heuristic(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_HEADER]]:
+// CHECK-NEXT:    %[[OMP_LOOP_IV:.+]] = phi i32 [ 0, %[[OMP_LOOP_PREHEADER]] ], [ %[[OMP_LOOP_NEXT:.+]], %[[OMP_LOOP_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_LOOP_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_COND]]:
+// CHECK-NEXT:    %[[OMP_LOOP_CMP:.+]] = icmp ult i32 %[[OMP_LOOP_IV]], %[[DOTCOUNT]]
+// CHECK-NEXT:    br i1 %[[OMP_LOOP_CMP]], label %[[OMP_LOOP_BODY:.+]], label %[[OMP_LOOP_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[OMP_LOOP_IV]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP3:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP4]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP3]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP5:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP6:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP7]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP6]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP5]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP9:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP10]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP9]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP11:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP11]]
+// CHECK-NEXT:    %[[TMP12:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP13:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP13]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP12]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_INC]]:
+// CHECK-NEXT:    %[[OMP_LOOP_NEXT]] = add nuw i32 %[[OMP_LOOP_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+void unroll_heuristic(float *a, float *b, float *c, float *d) {
+#pragma omp unroll
+  for (int i = 0; i < 128; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 128, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}

clang/test/OpenMP/irbuilder_unroll_partial_factor.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_factor(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_HEADER]]:
+// CHECK-NEXT:    %[[OMP_LOOP_IV:.+]] = phi i32 [ 0, %[[OMP_LOOP_PREHEADER]] ], [ %[[OMP_LOOP_NEXT:.+]], %[[OMP_LOOP_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_LOOP_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_COND]]:
+// CHECK-NEXT:    %[[OMP_LOOP_CMP:.+]] = icmp ult i32 %[[OMP_LOOP_IV]], %[[DOTCOUNT]]
+// CHECK-NEXT:    br i1 %[[OMP_LOOP_CMP]], label %[[OMP_LOOP_BODY:.+]], label %[[OMP_LOOP_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[OMP_LOOP_IV]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP3:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP4]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP3]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP5:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP6:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP7]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP6]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP5]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP9:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP10]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP9]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP11:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP11]]
+// CHECK-NEXT:    %[[TMP12:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP13:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP13]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP12]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_INC]]:
+// CHECK-NEXT:    %[[OMP_LOOP_NEXT]] = add nuw i32 %[[OMP_LOOP_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+void unroll_partial_factor(float *a, float *b, float *c, float *d) {
+#pragma omp unroll partial(3)
+  for (int i = 0; i < 2; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 2, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 3}

clang/test/OpenMP/irbuilder_unroll_partial_factor_for.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_heuristic_for(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[N_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LASTITER:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LOWERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_UPPERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_STRIDE:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32 %[[N:.+]], i32* %[[N_ADDR]], align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 1
+// CHECK-NEXT:    store i32* %[[N_ADDR]], i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP2:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[TMP2]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    %[[TMP4:.+]] = udiv i32 %[[DOTCOUNT]], 13
+// CHECK-NEXT:    %[[TMP5:.+]] = urem i32 %[[DOTCOUNT]], 13
+// CHECK-NEXT:    %[[TMP6:.+]] = icmp ne i32 %[[TMP5]], 0
+// CHECK-NEXT:    %[[TMP7:.+]] = zext i1 %[[TMP6]] to i32
+// CHECK-NEXT:    %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP4]], %[[TMP7]]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_PREHEADER]]:
+// CHECK-NEXT:    store i32 0, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP8:.+]] = sub i32 %[[OMP_FLOOR0_TRIPCOUNT]], 1
+// CHECK-NEXT:    store i32 %[[TMP8]], i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[P_STRIDE]], align 4
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[P_LASTITER]], i32* %[[P_LOWERBOUND]], i32* %[[P_UPPERBOUND]], i32* %[[P_STRIDE]], i32 1, i32 1)
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP11:.+]] = sub i32 %[[TMP10]], %[[TMP9]]
+// CHECK-NEXT:    %[[TMP12:.+]] = add i32 %[[TMP11]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_COND]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV]], %[[TMP12]]
+// CHECK-NEXT:    br i1 %[[OMP_FLOOR0_CMP]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_BODY]]:
+// CHECK-NEXT:    %[[TMP13:.+]] = add i32 %[[OMP_FLOOR0_IV]], %[[TMP9]]
+// CHECK-NEXT:    %[[TMP14:.+]] = icmp eq i32 %[[TMP13]], %[[OMP_FLOOR0_TRIPCOUNT]]
+// CHECK-NEXT:    %[[TMP15:.+]] = select i1 %[[TMP14]], i32 %[[TMP5]], i32 13
+// CHECK-NEXT:    br label %[[OMP_TILE0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_TILE0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_COND]]:
+// CHECK-NEXT:    %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV]], %[[TMP15]]
+// CHECK-NEXT:    br i1 %[[OMP_TILE0_CMP]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_BODY]]:
+// CHECK-NEXT:    %[[TMP16:.+]] = mul nuw i32 13, %[[TMP13]]
+// CHECK-NEXT:    %[[TMP17:.+]] = add nuw i32 %[[TMP16]], %[[OMP_TILE0_IV]]
+// CHECK-NEXT:    br label %[[OMP_LOOP_BODY:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[TMP17]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP18:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP19:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP19]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP18]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP20:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP21:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP22:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP22]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP21]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP23:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP20]], %[[TMP23]]
+// CHECK-NEXT:    %[[TMP24:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP25:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP25]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP24]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP26:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP26]]
+// CHECK-NEXT:    %[[TMP27:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP28:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP28]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP27]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_INC]]:
+// CHECK-NEXT:    %[[OMP_TILE0_NEXT]] = add nuw i32 %[[OMP_TILE0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_INC]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_NEXT]] = add nuw i32 %[[OMP_FLOOR0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_EXIT]]:
+// CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]])
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM9:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @2, i32 %[[OMP_GLOBAL_THREAD_NUM9]])
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+void unroll_partial_heuristic_for(int n, float *a, float *b, float *c, float *d) {
+#pragma omp for
+#pragma omp unroll partial(13)
+  for (int i = 0; i < n; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 1
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32*, i32** %[[TMP4]], align 8
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[TMP5]], align 4
+// CHECK-NEXT:    store i32 %[[TMP6]], i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP7]], %[[TMP8]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP9]], %[[TMP10]]
+// CHECK-NEXT:    %[[TMP11:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP11]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP12:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP12]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 13}

clang/test/OpenMP/irbuilder_unroll_partial_factor_for_collapse.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_factor_for_collapse(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[M_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[DOTOMP_IV:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[TMP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[TMP1:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[J:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTUNROLLED_IV_J:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTOMP_LB:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[DOTOMP_UB:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[I6:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTUNROLLED_IV_J7:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTUNROLL_INNER_IV_J:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32 %[[M:.+]], i32* %[[M_ADDR]], align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load i32, i32* %[[M_ADDR]], align 4
+// CHECK-NEXT:    store i32 %[[TMP0]], i32* %[[DOTCAPTURE_EXPR_]], align 4
+// CHECK-NEXT:    store i32 0, i32* %[[J]], align 4
+// CHECK-NEXT:    %[[TMP1_1:.+]] = load i32, i32* %[[DOTCAPTURE_EXPR_]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP1_1]], 0
+// CHECK-NEXT:    %[[DIV:.+]] = sdiv i32 %[[SUB]], 1
+// CHECK-NEXT:    %[[CONV:.+]] = sext i32 %[[DIV]] to i64
+// CHECK-NEXT:    %[[MUL:.+]] = mul nsw i64 %[[CONV]], 2
+// CHECK-NEXT:    %[[SUB3:.+]] = sub nsw i64 %[[MUL]], 1
+// CHECK-NEXT:    store i64 %[[SUB3]], i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 0, i32* %[[DOTUNROLLED_IV_J]], align 4
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[DOTCAPTURE_EXPR_]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 0, %[[TMP2]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_PRECOND_THEN]]:
+// CHECK-NEXT:    store i64 0, i64* %[[DOTOMP_LB]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i64, i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    store i64 %[[TMP3]], i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    store i64 1, i64* %[[DOTOMP_STRIDE]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[DOTOMP_IS_LAST]], align 4
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @3)
+// CHECK-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[DOTOMP_IS_LAST]], i64* %[[DOTOMP_LB]], i64* %[[DOTOMP_UB]], i64* %[[DOTOMP_STRIDE]], i64 1, i64 1)
+// CHECK-NEXT:    %[[TMP4:.+]] = load i64, i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    %[[TMP5:.+]] = load i64, i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    %[[CMP8:.+]] = icmp sgt i64 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP8]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i64, i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    %[[TMP7:.+]] = load i64, i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i64 [ %[[TMP6]], %[[COND_TRUE]] ], [ %[[TMP7]], %[[COND_FALSE]] ]
+// CHECK-NEXT:    store i64 %[[COND]], i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    %[[TMP8:.+]] = load i64, i64* %[[DOTOMP_LB]], align 8
+// CHECK-NEXT:    store i64 %[[TMP8]], i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    br label %[[OMP_INNER_FOR_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_COND]]:
+// CHECK-NEXT:    %[[TMP9:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[TMP10:.+]] = load i64, i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    %[[CMP10:.+]] = icmp sle i64 %[[TMP9]], %[[TMP10]]
+// CHECK-NEXT:    br i1 %[[CMP10]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_BODY]]:
+// CHECK-NEXT:    %[[TMP11:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[DIV12:.+]] = sdiv i64 %[[TMP11]], 2
+// CHECK-NEXT:    %[[MUL13:.+]] = mul nsw i64 %[[DIV12]], 1
+// CHECK-NEXT:    %[[ADD:.+]] = add nsw i64 0, %[[MUL13]]
+// CHECK-NEXT:    %[[CONV14:.+]] = trunc i64 %[[ADD]] to i32
+// CHECK-NEXT:    store i32 %[[CONV14]], i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[TMP12:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[TMP13:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[DIV15:.+]] = sdiv i64 %[[TMP13]], 2
+// CHECK-NEXT:    %[[MUL16:.+]] = mul nsw i64 %[[DIV15]], 2
+// CHECK-NEXT:    %[[SUB17:.+]] = sub nsw i64 %[[TMP12]], %[[MUL16]]
+// CHECK-NEXT:    %[[MUL18:.+]] = mul nsw i64 %[[SUB17]], 4
+// CHECK-NEXT:    %[[ADD19:.+]] = add nsw i64 0, %[[MUL18]]
+// CHECK-NEXT:    %[[CONV20:.+]] = trunc i64 %[[ADD19]] to i32
+// CHECK-NEXT:    store i32 %[[CONV20]], i32* %[[DOTUNROLLED_IV_J7]], align 4
+// CHECK-NEXT:    %[[TMP14:.+]] = load i32, i32* %[[DOTUNROLLED_IV_J7]], align 4
+// CHECK-NEXT:    store i32 %[[TMP14]], i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    br label %[[FOR_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_COND]]:
+// CHECK-NEXT:    %[[TMP15:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[TMP16:.+]] = load i32, i32* %[[DOTUNROLLED_IV_J7]], align 4
+// CHECK-NEXT:    %[[ADD21:.+]] = add nsw i32 %[[TMP16]], 4
+// CHECK-NEXT:    %[[CMP22:.+]] = icmp sle i32 %[[TMP15]], %[[ADD21]]
+// CHECK-NEXT:    br i1 %[[CMP22]], label %[[LAND_RHS:.+]], label %[[LAND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[LAND_RHS]]:
+// CHECK-NEXT:    %[[TMP17:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[CMP24:.+]] = icmp sle i32 %[[TMP17]], 8
+// CHECK-NEXT:    br label %[[LAND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[LAND_END]]:
+// CHECK-NEXT:    %[[TMP18:.+]] = phi i1 [ false, %[[FOR_COND]] ], [ %[[CMP24]], %[[LAND_RHS]] ]
+// CHECK-NEXT:    br i1 %[[TMP18]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_BODY]]:
+// CHECK-NEXT:    %[[TMP19:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[MUL26:.+]] = mul nsw i32 %[[TMP19]], 1
+// CHECK-NEXT:    %[[ADD27:.+]] = add nsw i32 0, %[[MUL26]]
+// CHECK-NEXT:    store i32 %[[ADD27]], i32* %[[J]], align 4
+// CHECK-NEXT:    %[[TMP20:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP21:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP21]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP20]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP22:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP23:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP24:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM28:.+]] = sext i32 %[[TMP24]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX29:.+]] = getelementptr inbounds float, float* %[[TMP23]], i64 %[[IDXPROM28]]
+// CHECK-NEXT:    %[[TMP25:.+]] = load float, float* %[[ARRAYIDX29]], align 4
+// CHECK-NEXT:    %[[TMP26:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP27:.+]] = load i32, i32* %[[J]], align 4
+// CHECK-NEXT:    %[[IDXPROM30:.+]] = sext i32 %[[TMP27]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX31:.+]] = getelementptr inbounds float, float* %[[TMP26]], i64 %[[IDXPROM30]]
+// CHECK-NEXT:    %[[TMP28:.+]] = load float, float* %[[ARRAYIDX31]], align 4
+// CHECK-NEXT:    %[[MUL32:.+]] = fmul float %[[TMP25]], %[[TMP28]]
+// CHECK-NEXT:    %[[ADD33:.+]] = fadd float %[[TMP22]], %[[MUL32]]
+// CHECK-NEXT:    %[[TMP29:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP30:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM34:.+]] = sext i32 %[[TMP30]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX35:.+]] = getelementptr inbounds float, float* %[[TMP29]], i64 %[[IDXPROM34]]
+// CHECK-NEXT:    %[[TMP31:.+]] = load float, float* %[[ARRAYIDX35]], align 4
+// CHECK-NEXT:    %[[ADD36:.+]] = fadd float %[[TMP31]], %[[ADD33]]
+// CHECK-NEXT:    store float %[[ADD36]], float* %[[ARRAYIDX35]], align 4
+// CHECK-NEXT:    br label %[[FOR_INC:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_INC]]:
+// CHECK-NEXT:    %[[TMP32:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[INC:.+]] = add nsw i32 %[[TMP32]], 1
+// CHECK-NEXT:    store i32 %[[INC]], i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_END]]:
+// CHECK-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_BODY_CONTINUE]]:
+// CHECK-NEXT:    br label %[[OMP_INNER_FOR_INC:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_INC]]:
+// CHECK-NEXT:    %[[TMP33:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[ADD37:.+]] = add nsw i64 %[[TMP33]], 1
+// CHECK-NEXT:    store i64 %[[ADD37]], i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    br label %[[OMP_INNER_FOR_COND]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_END]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_EXIT]]:
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM38:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @5)
+// CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM38]])
+// CHECK-NEXT:    br label %[[OMP_PRECOND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_PRECOND_END]]:
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM39:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @7)
+// CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @6, i32 %[[OMP_GLOBAL_THREAD_NUM39]])
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+void unroll_partial_factor_for_collapse(int m, float *a, float *b, float *c, float *d) {
+#pragma omp for collapse(2)
+  for (int i = 0; i < m; i++) {
+#pragma omp unroll partial(4)
+    for (int j = 0; j < 8; j++) {
+      a[i] += b[i] + c[i] * d[j];
+    }
+  }
+}
+
+#endif // HEADER
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.mustprogress"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 4}

clang/test/OpenMP/irbuilder_unroll_partial_heuristic.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_heuristic(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_HEADER]]:
+// CHECK-NEXT:    %[[OMP_LOOP_IV:.+]] = phi i32 [ 0, %[[OMP_LOOP_PREHEADER]] ], [ %[[OMP_LOOP_NEXT:.+]], %[[OMP_LOOP_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_LOOP_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_COND]]:
+// CHECK-NEXT:    %[[OMP_LOOP_CMP:.+]] = icmp ult i32 %[[OMP_LOOP_IV]], %[[DOTCOUNT]]
+// CHECK-NEXT:    br i1 %[[OMP_LOOP_CMP]], label %[[OMP_LOOP_BODY:.+]], label %[[OMP_LOOP_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[OMP_LOOP_IV]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP3:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP4]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP3]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP5:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP6:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP7]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP6]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP5]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP9:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP10]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP9]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP11:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP11]]
+// CHECK-NEXT:    %[[TMP12:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP13:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP13]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP12]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_INC]]:
+// CHECK-NEXT:    %[[OMP_LOOP_NEXT]] = add nuw i32 %[[OMP_LOOP_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_LOOP_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+void unroll_partial_heuristic(float *a, float *b, float *c, float *d) {
+#pragma omp unroll partial
+  for (int i = 0; i < 2; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 2, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}

clang/test/OpenMP/irbuilder_unroll_partial_heuristic_constant_for.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+// REQUIRES: x86-registered-target
+
+// TODO: The unroll-factor heuristic might be able to use the information that the trip count is constant, but currently is not able to determine that.
+
+#ifndef HEADER
+#define HEADER
+
+double sind(double);
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_heuristic_constant_for(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[E_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[OFFSET_ADDR:.+]] = alloca float, align 4
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LASTITER:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LOWERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_UPPERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_STRIDE:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[E:.+]], float** %[[E_ADDR]], align 8
+// CHECK-NEXT:    store float %[[OFFSET:.+]], float* %[[OFFSET_ADDR]], align 4
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    %[[TMP3:.+]] = udiv i32 %[[DOTCOUNT]], 4
+// CHECK-NEXT:    %[[TMP4:.+]] = urem i32 %[[DOTCOUNT]], 4
+// CHECK-NEXT:    %[[TMP5:.+]] = icmp ne i32 %[[TMP4]], 0
+// CHECK-NEXT:    %[[TMP6:.+]] = zext i1 %[[TMP5]] to i32
+// CHECK-NEXT:    %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP3]], %[[TMP6]]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_PREHEADER]]:
+// CHECK-NEXT:    store i32 0, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = sub i32 %[[OMP_FLOOR0_TRIPCOUNT]], 1
+// CHECK-NEXT:    store i32 %[[TMP7]], i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[P_STRIDE]], align 4
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[P_LASTITER]], i32* %[[P_LOWERBOUND]], i32* %[[P_UPPERBOUND]], i32* %[[P_STRIDE]], i32 1, i32 1)
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP10:.+]] = sub i32 %[[TMP9]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP11:.+]] = add i32 %[[TMP10]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_COND]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV]], %[[TMP11]]
+// CHECK-NEXT:    br i1 %[[OMP_FLOOR0_CMP]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_BODY]]:
+// CHECK-NEXT:    %[[TMP12:.+]] = add i32 %[[OMP_FLOOR0_IV]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP13:.+]] = icmp eq i32 %[[TMP12]], %[[OMP_FLOOR0_TRIPCOUNT]]
+// CHECK-NEXT:    %[[TMP14:.+]] = select i1 %[[TMP13]], i32 %[[TMP4]], i32 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_TILE0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_COND]]:
+// CHECK-NEXT:    %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV]], %[[TMP14]]
+// CHECK-NEXT:    br i1 %[[OMP_TILE0_CMP]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_BODY]]:
+// CHECK-NEXT:    %[[TMP15:.+]] = mul nuw i32 4, %[[TMP12]]
+// CHECK-NEXT:    %[[TMP16:.+]] = add nuw i32 %[[TMP15]], %[[OMP_TILE0_IV]]
+// CHECK-NEXT:    br label %[[OMP_LOOP_BODY:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[TMP16]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP17:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP18:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP18]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP17]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP19:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[CONV:.+]] = fpext float %[[TMP19]] to double
+// CHECK-NEXT:    %[[CALL:.+]] = call double @sind(double %[[CONV]])
+// CHECK-NEXT:    %[[TMP20:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP21:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP21]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP20]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP22:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[CONV4:.+]] = fpext float %[[TMP22]] to double
+// CHECK-NEXT:    %[[MUL:.+]] = fmul double %[[CALL]], %[[CONV4]]
+// CHECK-NEXT:    %[[TMP23:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP24:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM5:.+]] = sext i32 %[[TMP24]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX6:.+]] = getelementptr inbounds float, float* %[[TMP23]], i64 %[[IDXPROM5]]
+// CHECK-NEXT:    %[[TMP25:.+]] = load float, float* %[[ARRAYIDX6]], align 4
+// CHECK-NEXT:    %[[CONV7:.+]] = fpext float %[[TMP25]] to double
+// CHECK-NEXT:    %[[MUL8:.+]] = fmul double %[[MUL]], %[[CONV7]]
+// CHECK-NEXT:    %[[TMP26:.+]] = load float*, float** %[[E_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP27:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM9:.+]] = sext i32 %[[TMP27]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX10:.+]] = getelementptr inbounds float, float* %[[TMP26]], i64 %[[IDXPROM9]]
+// CHECK-NEXT:    %[[TMP28:.+]] = load float, float* %[[ARRAYIDX10]], align 4
+// CHECK-NEXT:    %[[CONV11:.+]] = fpext float %[[TMP28]] to double
+// CHECK-NEXT:    %[[MUL12:.+]] = fmul double %[[MUL8]], %[[CONV11]]
+// CHECK-NEXT:    %[[TMP29:.+]] = load float, float* %[[OFFSET_ADDR]], align 4
+// CHECK-NEXT:    %[[CONV13:.+]] = fpext float %[[TMP29]] to double
+// CHECK-NEXT:    %[[ADD:.+]] = fadd double %[[MUL12]], %[[CONV13]]
+// CHECK-NEXT:    %[[TMP30:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP31:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM14:.+]] = sext i32 %[[TMP31]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX15:.+]] = getelementptr inbounds float, float* %[[TMP30]], i64 %[[IDXPROM14]]
+// CHECK-NEXT:    %[[TMP32:.+]] = load float, float* %[[ARRAYIDX15]], align 4
+// CHECK-NEXT:    %[[CONV16:.+]] = fpext float %[[TMP32]] to double
+// CHECK-NEXT:    %[[ADD17:.+]] = fadd double %[[CONV16]], %[[ADD]]
+// CHECK-NEXT:    %[[CONV18:.+]] = fptrunc double %[[ADD17]] to float
+// CHECK-NEXT:    store float %[[CONV18]], float* %[[ARRAYIDX15]], align 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_INC]]:
+// CHECK-NEXT:    %[[OMP_TILE0_NEXT]] = add nuw i32 %[[OMP_TILE0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_INC]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_NEXT]] = add nuw i32 %[[OMP_FLOOR0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_EXIT]]:
+// CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]])
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM19:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @2, i32 %[[OMP_GLOBAL_THREAD_NUM19]])
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+void unroll_partial_heuristic_constant_for(float *a, float *b, float *c, float *d, float *e, float offset) {
+#pragma omp for
+#pragma omp unroll partial
+  for (int i = 0; i < 128; i++) {
+    a[i] += sind(b[i]) * c[i] * d[i] * e[i] + offset;
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 128, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 4}

clang/test/OpenMP/irbuilder_unroll_partial_heuristic_for_collapse.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+// REQUIRES: x86-registered-target
+
+#ifndef HEADER
+#define HEADER
+
+double sind(double);
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_heuristic_for(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[M_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[E_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[OFFSET_ADDR:.+]] = alloca float, align 4
+// CHECK-NEXT:    %[[DOTOMP_IV:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[TMP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[TMP1:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTCAPTURE_EXPR_:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[J:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTCAPTURE_EXPR_2:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTUNROLLED_IV_J:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTOMP_LB:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[DOTOMP_UB:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[DOTOMP_STRIDE:.+]] = alloca i64, align 8
+// CHECK-NEXT:    %[[DOTOMP_IS_LAST:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[I6:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTUNROLLED_IV_J7:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTUNROLL_INNER_IV_J:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32 %[[M:.+]], i32* %[[M_ADDR]], align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[E:.+]], float** %[[E_ADDR]], align 8
+// CHECK-NEXT:    store float %[[OFFSET:.+]], float* %[[OFFSET_ADDR]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = load i32, i32* %[[M_ADDR]], align 4
+// CHECK-NEXT:    store i32 %[[TMP0]], i32* %[[DOTCAPTURE_EXPR_]], align 4
+// CHECK-NEXT:    store i32 0, i32* %[[J]], align 4
+// CHECK-NEXT:    %[[TMP1_1:.+]] = load i32, i32* %[[DOTCAPTURE_EXPR_]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP1_1]], 0
+// CHECK-NEXT:    %[[DIV:.+]] = sdiv i32 %[[SUB]], 1
+// CHECK-NEXT:    %[[CONV:.+]] = sext i32 %[[DIV]] to i64
+// CHECK-NEXT:    %[[MUL:.+]] = mul nsw i64 %[[CONV]], 4
+// CHECK-NEXT:    %[[SUB3:.+]] = sub nsw i64 %[[MUL]], 1
+// CHECK-NEXT:    store i64 %[[SUB3]], i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 0, i32* %[[DOTUNROLLED_IV_J]], align 4
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[DOTCAPTURE_EXPR_]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 0, %[[TMP2]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[OMP_PRECOND_THEN:.+]], label %[[OMP_PRECOND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_PRECOND_THEN]]:
+// CHECK-NEXT:    store i64 0, i64* %[[DOTOMP_LB]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i64, i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    store i64 %[[TMP3]], i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    store i64 1, i64* %[[DOTOMP_STRIDE]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[DOTOMP_IS_LAST]], align 4
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @3)
+// CHECK-NEXT:    call void @__kmpc_for_static_init_8(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[DOTOMP_IS_LAST]], i64* %[[DOTOMP_LB]], i64* %[[DOTOMP_UB]], i64* %[[DOTOMP_STRIDE]], i64 1, i64 1)
+// CHECK-NEXT:    %[[TMP4:.+]] = load i64, i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    %[[TMP5:.+]] = load i64, i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    %[[CMP8:.+]] = icmp sgt i64 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP8]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i64, i64* %[[DOTCAPTURE_EXPR_2]], align 8
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    %[[TMP7:.+]] = load i64, i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i64 [ %[[TMP6]], %[[COND_TRUE]] ], [ %[[TMP7]], %[[COND_FALSE]] ]
+// CHECK-NEXT:    store i64 %[[COND]], i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    %[[TMP8:.+]] = load i64, i64* %[[DOTOMP_LB]], align 8
+// CHECK-NEXT:    store i64 %[[TMP8]], i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    br label %[[OMP_INNER_FOR_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_COND]]:
+// CHECK-NEXT:    %[[TMP9:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[TMP10:.+]] = load i64, i64* %[[DOTOMP_UB]], align 8
+// CHECK-NEXT:    %[[CMP10:.+]] = icmp sle i64 %[[TMP9]], %[[TMP10]]
+// CHECK-NEXT:    br i1 %[[CMP10]], label %[[OMP_INNER_FOR_BODY:.+]], label %[[OMP_INNER_FOR_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_BODY]]:
+// CHECK-NEXT:    %[[TMP11:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[DIV12:.+]] = sdiv i64 %[[TMP11]], 4
+// CHECK-NEXT:    %[[MUL13:.+]] = mul nsw i64 %[[DIV12]], 1
+// CHECK-NEXT:    %[[ADD:.+]] = add nsw i64 0, %[[MUL13]]
+// CHECK-NEXT:    %[[CONV14:.+]] = trunc i64 %[[ADD]] to i32
+// CHECK-NEXT:    store i32 %[[CONV14]], i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[TMP12:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[TMP13:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[DIV15:.+]] = sdiv i64 %[[TMP13]], 4
+// CHECK-NEXT:    %[[MUL16:.+]] = mul nsw i64 %[[DIV15]], 4
+// CHECK-NEXT:    %[[SUB17:.+]] = sub nsw i64 %[[TMP12]], %[[MUL16]]
+// CHECK-NEXT:    %[[MUL18:.+]] = mul nsw i64 %[[SUB17]], 2
+// CHECK-NEXT:    %[[ADD19:.+]] = add nsw i64 0, %[[MUL18]]
+// CHECK-NEXT:    %[[CONV20:.+]] = trunc i64 %[[ADD19]] to i32
+// CHECK-NEXT:    store i32 %[[CONV20]], i32* %[[DOTUNROLLED_IV_J7]], align 4
+// CHECK-NEXT:    %[[TMP14:.+]] = load i32, i32* %[[DOTUNROLLED_IV_J7]], align 4
+// CHECK-NEXT:    store i32 %[[TMP14]], i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    br label %[[FOR_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_COND]]:
+// CHECK-NEXT:    %[[TMP15:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[TMP16:.+]] = load i32, i32* %[[DOTUNROLLED_IV_J7]], align 4
+// CHECK-NEXT:    %[[ADD21:.+]] = add nsw i32 %[[TMP16]], 2
+// CHECK-NEXT:    %[[CMP22:.+]] = icmp sle i32 %[[TMP15]], %[[ADD21]]
+// CHECK-NEXT:    br i1 %[[CMP22]], label %[[LAND_RHS:.+]], label %[[LAND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[LAND_RHS]]:
+// CHECK-NEXT:    %[[TMP17:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[CMP24:.+]] = icmp sle i32 %[[TMP17]], 8
+// CHECK-NEXT:    br label %[[LAND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[LAND_END]]:
+// CHECK-NEXT:    %[[TMP18:.+]] = phi i1 [ false, %[[FOR_COND]] ], [ %[[CMP24]], %[[LAND_RHS]] ]
+// CHECK-NEXT:    br i1 %[[TMP18]], label %[[FOR_BODY:.+]], label %[[FOR_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_BODY]]:
+// CHECK-NEXT:    %[[TMP19:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[MUL26:.+]] = mul nsw i32 %[[TMP19]], 1
+// CHECK-NEXT:    %[[ADD27:.+]] = add nsw i32 0, %[[MUL26]]
+// CHECK-NEXT:    store i32 %[[ADD27]], i32* %[[J]], align 4
+// CHECK-NEXT:    %[[TMP20:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP21:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP21]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP20]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP22:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[CONV28:.+]] = fpext float %[[TMP22]] to double
+// CHECK-NEXT:    %[[CALL:.+]] = call double @sind(double %[[CONV28]])
+// CHECK-NEXT:    %[[TMP23:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP24:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM29:.+]] = sext i32 %[[TMP24]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX30:.+]] = getelementptr inbounds float, float* %[[TMP23]], i64 %[[IDXPROM29]]
+// CHECK-NEXT:    %[[TMP25:.+]] = load float, float* %[[ARRAYIDX30]], align 4
+// CHECK-NEXT:    %[[CONV31:.+]] = fpext float %[[TMP25]] to double
+// CHECK-NEXT:    %[[MUL32:.+]] = fmul double %[[CALL]], %[[CONV31]]
+// CHECK-NEXT:    %[[TMP26:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP27:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM33:.+]] = sext i32 %[[TMP27]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX34:.+]] = getelementptr inbounds float, float* %[[TMP26]], i64 %[[IDXPROM33]]
+// CHECK-NEXT:    %[[TMP28:.+]] = load float, float* %[[ARRAYIDX34]], align 4
+// CHECK-NEXT:    %[[CONV35:.+]] = fpext float %[[TMP28]] to double
+// CHECK-NEXT:    %[[MUL36:.+]] = fmul double %[[MUL32]], %[[CONV35]]
+// CHECK-NEXT:    %[[TMP29:.+]] = load float*, float** %[[E_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP30:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM37:.+]] = sext i32 %[[TMP30]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX38:.+]] = getelementptr inbounds float, float* %[[TMP29]], i64 %[[IDXPROM37]]
+// CHECK-NEXT:    %[[TMP31:.+]] = load float, float* %[[ARRAYIDX38]], align 4
+// CHECK-NEXT:    %[[CONV39:.+]] = fpext float %[[TMP31]] to double
+// CHECK-NEXT:    %[[MUL40:.+]] = fmul double %[[MUL36]], %[[CONV39]]
+// CHECK-NEXT:    %[[TMP32:.+]] = load float, float* %[[OFFSET_ADDR]], align 4
+// CHECK-NEXT:    %[[CONV41:.+]] = fpext float %[[TMP32]] to double
+// CHECK-NEXT:    %[[ADD42:.+]] = fadd double %[[MUL40]], %[[CONV41]]
+// CHECK-NEXT:    %[[TMP33:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP34:.+]] = load i32, i32* %[[I6]], align 4
+// CHECK-NEXT:    %[[IDXPROM43:.+]] = sext i32 %[[TMP34]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX44:.+]] = getelementptr inbounds float, float* %[[TMP33]], i64 %[[IDXPROM43]]
+// CHECK-NEXT:    %[[TMP35:.+]] = load float, float* %[[ARRAYIDX44]], align 4
+// CHECK-NEXT:    %[[CONV45:.+]] = fpext float %[[TMP35]] to double
+// CHECK-NEXT:    %[[ADD46:.+]] = fadd double %[[CONV45]], %[[ADD42]]
+// CHECK-NEXT:    %[[CONV47:.+]] = fptrunc double %[[ADD46]] to float
+// CHECK-NEXT:    store float %[[CONV47]], float* %[[ARRAYIDX44]], align 4
+// CHECK-NEXT:    br label %[[FOR_INC:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_INC]]:
+// CHECK-NEXT:    %[[TMP36:.+]] = load i32, i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    %[[INC:.+]] = add nsw i32 %[[TMP36]], 1
+// CHECK-NEXT:    store i32 %[[INC]], i32* %[[DOTUNROLL_INNER_IV_J]], align 4
+// CHECK-NEXT:    br label %[[FOR_COND]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[FOR_END]]:
+// CHECK-NEXT:    br label %[[OMP_BODY_CONTINUE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_BODY_CONTINUE]]:
+// CHECK-NEXT:    br label %[[OMP_INNER_FOR_INC:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_INC]]:
+// CHECK-NEXT:    %[[TMP37:.+]] = load i64, i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    %[[ADD48:.+]] = add nsw i64 %[[TMP37]], 1
+// CHECK-NEXT:    store i64 %[[ADD48]], i64* %[[DOTOMP_IV]], align 8
+// CHECK-NEXT:    br label %[[OMP_INNER_FOR_COND]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_INNER_FOR_END]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_EXIT]]:
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM49:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @5)
+// CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM49]])
+// CHECK-NEXT:    br label %[[OMP_PRECOND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_PRECOND_END]]:
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM50:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @7)
+// CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @6, i32 %[[OMP_GLOBAL_THREAD_NUM50]])
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+void unroll_partial_heuristic_for(int m, float *a, float *b, float *c, float *d, float *e, float offset) {
+#pragma omp for collapse(2)
+  for (int i = 0; i < m; i++) {
+#pragma omp unroll partial
+    for (int j = 0; j < 8; j++) {
+      a[i] += sind(b[i]) * c[i] * d[i] * e[i] + offset;
+    }
+  }
+}
+
+#endif // HEADER
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.mustprogress"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 2}

clang/test/OpenMP/irbuilder_unroll_partial_heuristic_runtime_for.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+// REQUIRES: x86-registered-target
+
+#ifndef HEADER
+#define HEADER
+
+double sind(double);
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_heuristic_runtime_for(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[N_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[E_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[OFFSET_ADDR:.+]] = alloca float, align 4
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LASTITER:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LOWERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_UPPERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_STRIDE:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32 %[[N:.+]], i32* %[[N_ADDR]], align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[E:.+]], float** %[[E_ADDR]], align 8
+// CHECK-NEXT:    store float %[[OFFSET:.+]], float* %[[OFFSET_ADDR]], align 4
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 1
+// CHECK-NEXT:    store i32* %[[N_ADDR]], i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP2:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[TMP2]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    %[[TMP4:.+]] = udiv i32 %[[DOTCOUNT]], 4
+// CHECK-NEXT:    %[[TMP5:.+]] = urem i32 %[[DOTCOUNT]], 4
+// CHECK-NEXT:    %[[TMP6:.+]] = icmp ne i32 %[[TMP5]], 0
+// CHECK-NEXT:    %[[TMP7:.+]] = zext i1 %[[TMP6]] to i32
+// CHECK-NEXT:    %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP4]], %[[TMP7]]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_PREHEADER]]:
+// CHECK-NEXT:    store i32 0, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP8:.+]] = sub i32 %[[OMP_FLOOR0_TRIPCOUNT]], 1
+// CHECK-NEXT:    store i32 %[[TMP8]], i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[P_STRIDE]], align 4
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[P_LASTITER]], i32* %[[P_LOWERBOUND]], i32* %[[P_UPPERBOUND]], i32* %[[P_STRIDE]], i32 1, i32 1)
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP11:.+]] = sub i32 %[[TMP10]], %[[TMP9]]
+// CHECK-NEXT:    %[[TMP12:.+]] = add i32 %[[TMP11]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_COND]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV]], %[[TMP12]]
+// CHECK-NEXT:    br i1 %[[OMP_FLOOR0_CMP]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_BODY]]:
+// CHECK-NEXT:    %[[TMP13:.+]] = add i32 %[[OMP_FLOOR0_IV]], %[[TMP9]]
+// CHECK-NEXT:    %[[TMP14:.+]] = icmp eq i32 %[[TMP13]], %[[OMP_FLOOR0_TRIPCOUNT]]
+// CHECK-NEXT:    %[[TMP15:.+]] = select i1 %[[TMP14]], i32 %[[TMP5]], i32 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_TILE0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_COND]]:
+// CHECK-NEXT:    %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV]], %[[TMP15]]
+// CHECK-NEXT:    br i1 %[[OMP_TILE0_CMP]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_BODY]]:
+// CHECK-NEXT:    %[[TMP16:.+]] = mul nuw i32 4, %[[TMP13]]
+// CHECK-NEXT:    %[[TMP17:.+]] = add nuw i32 %[[TMP16]], %[[OMP_TILE0_IV]]
+// CHECK-NEXT:    br label %[[OMP_LOOP_BODY:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[TMP17]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP18:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP19:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP19]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP18]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP20:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[CONV:.+]] = fpext float %[[TMP20]] to double
+// CHECK-NEXT:    %[[CALL:.+]] = call double @sind(double %[[CONV]])
+// CHECK-NEXT:    %[[TMP21:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP22:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP22]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP21]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP23:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[CONV4:.+]] = fpext float %[[TMP23]] to double
+// CHECK-NEXT:    %[[MUL:.+]] = fmul double %[[CALL]], %[[CONV4]]
+// CHECK-NEXT:    %[[TMP24:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP25:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM5:.+]] = sext i32 %[[TMP25]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX6:.+]] = getelementptr inbounds float, float* %[[TMP24]], i64 %[[IDXPROM5]]
+// CHECK-NEXT:    %[[TMP26:.+]] = load float, float* %[[ARRAYIDX6]], align 4
+// CHECK-NEXT:    %[[CONV7:.+]] = fpext float %[[TMP26]] to double
+// CHECK-NEXT:    %[[MUL8:.+]] = fmul double %[[MUL]], %[[CONV7]]
+// CHECK-NEXT:    %[[TMP27:.+]] = load float*, float** %[[E_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP28:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM9:.+]] = sext i32 %[[TMP28]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX10:.+]] = getelementptr inbounds float, float* %[[TMP27]], i64 %[[IDXPROM9]]
+// CHECK-NEXT:    %[[TMP29:.+]] = load float, float* %[[ARRAYIDX10]], align 4
+// CHECK-NEXT:    %[[CONV11:.+]] = fpext float %[[TMP29]] to double
+// CHECK-NEXT:    %[[MUL12:.+]] = fmul double %[[MUL8]], %[[CONV11]]
+// CHECK-NEXT:    %[[TMP30:.+]] = load float, float* %[[OFFSET_ADDR]], align 4
+// CHECK-NEXT:    %[[CONV13:.+]] = fpext float %[[TMP30]] to double
+// CHECK-NEXT:    %[[ADD:.+]] = fadd double %[[MUL12]], %[[CONV13]]
+// CHECK-NEXT:    %[[TMP31:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP32:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM14:.+]] = sext i32 %[[TMP32]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX15:.+]] = getelementptr inbounds float, float* %[[TMP31]], i64 %[[IDXPROM14]]
+// CHECK-NEXT:    %[[TMP33:.+]] = load float, float* %[[ARRAYIDX15]], align 4
+// CHECK-NEXT:    %[[CONV16:.+]] = fpext float %[[TMP33]] to double
+// CHECK-NEXT:    %[[ADD17:.+]] = fadd double %[[CONV16]], %[[ADD]]
+// CHECK-NEXT:    %[[CONV18:.+]] = fptrunc double %[[ADD17]] to float
+// CHECK-NEXT:    store float %[[CONV18]], float* %[[ARRAYIDX15]], align 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_INC]]:
+// CHECK-NEXT:    %[[OMP_TILE0_NEXT]] = add nuw i32 %[[OMP_TILE0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_INC]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_NEXT]] = add nuw i32 %[[OMP_FLOOR0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_EXIT]]:
+// CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]])
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM19:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @2, i32 %[[OMP_GLOBAL_THREAD_NUM19]])
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+void unroll_partial_heuristic_runtime_for(int n, float *a, float *b, float *c, float *d, float *e, float offset) {
+#pragma omp for
+#pragma omp unroll partial
+  for (int i = 0; i < n; i++) {
+    a[i] += sind(b[i]) * c[i] * d[i] * e[i] + offset;
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 1
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32*, i32** %[[TMP4]], align 8
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[TMP5]], align 4
+// CHECK-NEXT:    store i32 %[[TMP6]], i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP7]], %[[TMP8]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP10:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP9]], %[[TMP10]]
+// CHECK-NEXT:    %[[TMP11:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP11]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP12:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP12]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 4}

clang/test/OpenMP/irbuilder_unroll_unroll_partial_factor.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_partial_factor_for(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LASTITER:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_LOWERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_UPPERBOUND:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[P_STRIDE:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    %[[TMP3:.+]] = udiv i32 %[[DOTCOUNT]], 2
+// CHECK-NEXT:    %[[TMP4:.+]] = urem i32 %[[DOTCOUNT]], 2
+// CHECK-NEXT:    %[[TMP5:.+]] = icmp ne i32 %[[TMP4]], 0
+// CHECK-NEXT:    %[[TMP6:.+]] = zext i1 %[[TMP5]] to i32
+// CHECK-NEXT:    %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP3]], %[[TMP6]]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_PREHEADER]]:
+// CHECK-NEXT:    store i32 0, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = sub i32 %[[OMP_FLOOR0_TRIPCOUNT]], 1
+// CHECK-NEXT:    store i32 %[[TMP7]], i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[P_STRIDE]], align 4
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_for_static_init_4u(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]], i32 34, i32* %[[P_LASTITER]], i32* %[[P_LOWERBOUND]], i32* %[[P_UPPERBOUND]], i32* %[[P_STRIDE]], i32 1, i32 1)
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[P_LOWERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32, i32* %[[P_UPPERBOUND]], align 4
+// CHECK-NEXT:    %[[TMP10:.+]] = sub i32 %[[TMP9]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP11:.+]] = add i32 %[[TMP10]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_COND]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV]], %[[TMP11]]
+// CHECK-NEXT:    br i1 %[[OMP_FLOOR0_CMP]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_BODY]]:
+// CHECK-NEXT:    %[[TMP12:.+]] = add i32 %[[OMP_FLOOR0_IV]], %[[TMP8]]
+// CHECK-NEXT:    %[[TMP13:.+]] = icmp eq i32 %[[TMP12]], %[[OMP_FLOOR0_TRIPCOUNT]]
+// CHECK-NEXT:    %[[TMP14:.+]] = select i1 %[[TMP13]], i32 %[[TMP4]], i32 2
+// CHECK-NEXT:    br label %[[OMP_TILE0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_TILE0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_COND]]:
+// CHECK-NEXT:    %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV]], %[[TMP14]]
+// CHECK-NEXT:    br i1 %[[OMP_TILE0_CMP]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_BODY]]:
+// CHECK-NEXT:    %[[TMP15:.+]] = mul nuw i32 2, %[[TMP12]]
+// CHECK-NEXT:    %[[TMP16:.+]] = add nuw i32 %[[TMP15]], %[[OMP_TILE0_IV]]
+// CHECK-NEXT:    br label %[[OMP_LOOP_BODY:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[TMP16]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP17:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP18:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP18]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP17]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP19:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP20:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP21:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP21]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP20]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP22:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP19]], %[[TMP22]]
+// CHECK-NEXT:    %[[TMP23:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP24:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP24]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP23]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP25:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP25]]
+// CHECK-NEXT:    %[[TMP26:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP27:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP27]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP26]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_INC]]:
+// CHECK-NEXT:    %[[OMP_TILE0_NEXT]] = add nuw i32 %[[OMP_TILE0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_INC]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_NEXT]] = add nuw i32 %[[OMP_FLOOR0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_EXIT]]:
+// CHECK-NEXT:    call void @__kmpc_for_static_fini(%struct.ident_t* @1, i32 %[[OMP_GLOBAL_THREAD_NUM]])
+// CHECK-NEXT:    %[[OMP_GLOBAL_THREAD_NUM9:.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @1)
+// CHECK-NEXT:    call void @__kmpc_barrier(%struct.ident_t* @2, i32 %[[OMP_GLOBAL_THREAD_NUM9]])
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+void unroll_partial_factor_for(float *a, float *b, float *c, float *d) {
+#pragma omp for
+#pragma omp unroll partial(2)
+  for (int i = 0; i < 2; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 2, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 2}

clang/test/OpenMP/irbuilder_unroll_unroll_partial_heuristic.c

+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs
+// RUN: %clang_cc1 -fopenmp-enable-irbuilder -verify -fopenmp -fopenmp-version=51 -x c -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+
+// CHECK-LABEL: define {{.*}}@unroll_unroll_partial_heuristic(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[A_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[B_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[C_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[D_ADDR:.+]] = alloca float*, align 8
+// CHECK-NEXT:    %[[I:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[AGG_CAPTURED:.+]] = alloca %struct.anon, align 8
+// CHECK-NEXT:    %[[AGG_CAPTURED1:.+]] = alloca %struct.anon.0, align 4
+// CHECK-NEXT:    %[[DOTCOUNT_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store float* %[[A:.+]], float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[B:.+]], float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[C:.+]], float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    store float* %[[D:.+]], float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    store i32 0, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[TMP0:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[AGG_CAPTURED]], i32 0, i32 0
+// CHECK-NEXT:    store i32* %[[I]], i32** %[[TMP0]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[AGG_CAPTURED1]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    store i32 %[[TMP2]], i32* %[[TMP1]], align 4
+// CHECK-NEXT:    call void @__captured_stmt(i32* %[[DOTCOUNT_ADDR]], %struct.anon* %[[AGG_CAPTURED]])
+// CHECK-NEXT:    %[[DOTCOUNT:.+]] = load i32, i32* %[[DOTCOUNT_ADDR]], align 4
+// CHECK-NEXT:    br label %[[OMP_LOOP_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_PREHEADER]]:
+// CHECK-NEXT:    %[[TMP3:.+]] = udiv i32 %[[DOTCOUNT]], 8
+// CHECK-NEXT:    %[[TMP4:.+]] = urem i32 %[[DOTCOUNT]], 8
+// CHECK-NEXT:    %[[TMP5:.+]] = icmp ne i32 %[[TMP4]], 0
+// CHECK-NEXT:    %[[TMP6:.+]] = zext i1 %[[TMP5]] to i32
+// CHECK-NEXT:    %[[OMP_FLOOR0_TRIPCOUNT:.+]] = add nuw i32 %[[TMP3]], %[[TMP6]]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_IV:.+]] = phi i32 [ 0, %[[OMP_FLOOR0_PREHEADER]] ], [ %[[OMP_FLOOR0_NEXT:.+]], %[[OMP_FLOOR0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_COND]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_CMP:.+]] = icmp ult i32 %[[OMP_FLOOR0_IV]], %[[OMP_FLOOR0_TRIPCOUNT]]
+// CHECK-NEXT:    br i1 %[[OMP_FLOOR0_CMP]], label %[[OMP_FLOOR0_BODY:.+]], label %[[OMP_FLOOR0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_BODY]]:
+// CHECK-NEXT:    %[[TMP7:.+]] = icmp eq i32 %[[OMP_FLOOR0_IV]], %[[OMP_FLOOR0_TRIPCOUNT]]
+// CHECK-NEXT:    %[[TMP8:.+]] = select i1 %[[TMP7]], i32 %[[TMP4]], i32 8
+// CHECK-NEXT:    br label %[[OMP_TILE0_PREHEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_PREHEADER]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_HEADER]]:
+// CHECK-NEXT:    %[[OMP_TILE0_IV:.+]] = phi i32 [ 0, %[[OMP_TILE0_PREHEADER]] ], [ %[[OMP_TILE0_NEXT:.+]], %[[OMP_TILE0_INC:.+]] ]
+// CHECK-NEXT:    br label %[[OMP_TILE0_COND:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_COND]]:
+// CHECK-NEXT:    %[[OMP_TILE0_CMP:.+]] = icmp ult i32 %[[OMP_TILE0_IV]], %[[TMP8]]
+// CHECK-NEXT:    br i1 %[[OMP_TILE0_CMP]], label %[[OMP_TILE0_BODY:.+]], label %[[OMP_TILE0_EXIT:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_BODY]]:
+// CHECK-NEXT:    %[[TMP9:.+]] = mul nuw i32 8, %[[OMP_FLOOR0_IV]]
+// CHECK-NEXT:    %[[TMP10:.+]] = add nuw i32 %[[TMP9]], %[[OMP_TILE0_IV]]
+// CHECK-NEXT:    br label %[[OMP_LOOP_BODY:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_BODY]]:
+// CHECK-NEXT:    call void @__captured_stmt.1(i32* %[[I]], i32 %[[TMP10]], %struct.anon.0* %[[AGG_CAPTURED1]])
+// CHECK-NEXT:    %[[TMP11:.+]] = load float*, float** %[[B_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP12:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM:.+]] = sext i32 %[[TMP12]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX:.+]] = getelementptr inbounds float, float* %[[TMP11]], i64 %[[IDXPROM]]
+// CHECK-NEXT:    %[[TMP13:.+]] = load float, float* %[[ARRAYIDX]], align 4
+// CHECK-NEXT:    %[[TMP14:.+]] = load float*, float** %[[C_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP15:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM2:.+]] = sext i32 %[[TMP15]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX3:.+]] = getelementptr inbounds float, float* %[[TMP14]], i64 %[[IDXPROM2]]
+// CHECK-NEXT:    %[[TMP16:.+]] = load float, float* %[[ARRAYIDX3]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = fmul float %[[TMP13]], %[[TMP16]]
+// CHECK-NEXT:    %[[TMP17:.+]] = load float*, float** %[[D_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP18:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM4:.+]] = sext i32 %[[TMP18]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX5:.+]] = getelementptr inbounds float, float* %[[TMP17]], i64 %[[IDXPROM4]]
+// CHECK-NEXT:    %[[TMP19:.+]] = load float, float* %[[ARRAYIDX5]], align 4
+// CHECK-NEXT:    %[[MUL6:.+]] = fmul float %[[MUL]], %[[TMP19]]
+// CHECK-NEXT:    %[[TMP20:.+]] = load float*, float** %[[A_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP21:.+]] = load i32, i32* %[[I]], align 4
+// CHECK-NEXT:    %[[IDXPROM7:.+]] = sext i32 %[[TMP21]] to i64
+// CHECK-NEXT:    %[[ARRAYIDX8:.+]] = getelementptr inbounds float, float* %[[TMP20]], i64 %[[IDXPROM7]]
+// CHECK-NEXT:    store float %[[MUL6]], float* %[[ARRAYIDX8]], align 4
+// CHECK-NEXT:    br label %[[OMP_TILE0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_INC]]:
+// CHECK-NEXT:    %[[OMP_TILE0_NEXT]] = add nuw i32 %[[OMP_TILE0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_TILE0_HEADER]], !llvm.loop ![[LOOP3:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_TILE0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_TILE0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_INC]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_INC]]:
+// CHECK-NEXT:    %[[OMP_FLOOR0_NEXT]] = add nuw i32 %[[OMP_FLOOR0_IV]], 1
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_HEADER]], !llvm.loop ![[LOOP6:[0-9]+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_EXIT]]:
+// CHECK-NEXT:    br label %[[OMP_FLOOR0_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_FLOOR0_AFTER]]:
+// CHECK-NEXT:    br label %[[OMP_LOOP_AFTER:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[OMP_LOOP_AFTER]]:
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+void unroll_unroll_partial_heuristic(float *a, float *b, float *c, float *d) {
+#pragma omp unroll partial
+#pragma omp unroll partial
+  for (int i = 0; i < 2; i++) {
+    a[i] = b[i] * c[i] * d[i];
+  }
+}
+
+#endif // HEADER
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[DISTANCE_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon*, align 8
+// CHECK-NEXT:    %[[DOTSTART:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTOP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[DOTSTEP:.+]] = alloca i32, align 4
+// CHECK-NEXT:    store i32* %[[DISTANCE:.+]], i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store %struct.anon* %[[__CONTEXT:.+]], %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon*, %struct.anon** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon, %struct.anon* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32*, i32** %[[TMP1]], align 8
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[TMP2]], align 4
+// CHECK-NEXT:    store i32 %[[TMP3]], i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    store i32 2, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    store i32 1, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[TMP5:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[CMP:.+]] = icmp slt i32 %[[TMP4]], %[[TMP5]]
+// CHECK-NEXT:    br i1 %[[CMP]], label %[[COND_TRUE:.+]], label %[[COND_FALSE:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_TRUE]]:
+// CHECK-NEXT:    %[[TMP6:.+]] = load i32, i32* %[[DOTSTOP]], align 4
+// CHECK-NEXT:    %[[TMP7:.+]] = load i32, i32* %[[DOTSTART]], align 4
+// CHECK-NEXT:    %[[SUB:.+]] = sub nsw i32 %[[TMP6]], %[[TMP7]]
+// CHECK-NEXT:    %[[TMP8:.+]] = load i32, i32* %[[DOTSTEP]], align 4
+// CHECK-NEXT:    %[[DIV:.+]] = udiv i32 %[[SUB]], %[[TMP8]]
+// CHECK-NEXT:    br label %[[COND_END:.+]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_FALSE]]:
+// CHECK-NEXT:    br label %[[COND_END]]
+// CHECK-EMPTY:
+// CHECK-NEXT:  [[COND_END]]:
+// CHECK-NEXT:    %[[COND:.+]] = phi i32 [ %[[DIV]], %[[COND_TRUE]] ], [ 0, %[[COND_FALSE]] ]
+// CHECK-NEXT:    %[[TMP9:.+]] = load i32*, i32** %[[DISTANCE_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[COND]], i32* %[[TMP9]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK-LABEL: define {{.*}}@__captured_stmt.1(
+// CHECK-NEXT:  [[ENTRY:.*]]:
+// CHECK-NEXT:    %[[LOOPVAR_ADDR:.+]] = alloca i32*, align 8
+// CHECK-NEXT:    %[[LOGICAL_ADDR:.+]] = alloca i32, align 4
+// CHECK-NEXT:    %[[__CONTEXT_ADDR:.+]] = alloca %struct.anon.0*, align 8
+// CHECK-NEXT:    store i32* %[[LOOPVAR:.+]], i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[LOGICAL:.+]], i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    store %struct.anon.0* %[[__CONTEXT:.+]], %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP0:.+]] = load %struct.anon.0*, %struct.anon.0** %[[__CONTEXT_ADDR]], align 8
+// CHECK-NEXT:    %[[TMP1:.+]] = getelementptr inbounds %struct.anon.0, %struct.anon.0* %[[TMP0]], i32 0, i32 0
+// CHECK-NEXT:    %[[TMP2:.+]] = load i32, i32* %[[TMP1]], align 4
+// CHECK-NEXT:    %[[TMP3:.+]] = load i32, i32* %[[LOGICAL_ADDR]], align 4
+// CHECK-NEXT:    %[[MUL:.+]] = mul i32 1, %[[TMP3]]
+// CHECK-NEXT:    %[[ADD:.+]] = add i32 %[[TMP2]], %[[MUL]]
+// CHECK-NEXT:    %[[TMP4:.+]] = load i32*, i32** %[[LOOPVAR_ADDR]], align 8
+// CHECK-NEXT:    store i32 %[[ADD]], i32* %[[TMP4]], align 4
+// CHECK-NEXT:    ret void
+// CHECK-NEXT:  }
+
+
+// CHECK: ![[META0:[0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+// CHECK: ![[META1:[0-9]+]] = !{i32 7, !"openmp", i32 51}
+// CHECK: ![[META2:[0-9]+]] =
+// CHECK: ![[LOOP3]] = distinct !{![[LOOP3]], ![[LOOPPROP4:[0-9]+]], ![[LOOPPROP5:[0-9]+]]}
+// CHECK: ![[LOOPPROP4]] = !{!"llvm.loop.unroll.enable"}
+// CHECK: ![[LOOPPROP5]] = !{!"llvm.loop.unroll.count", i32 8}
+// CHECK: ![[LOOP6]] = distinct !{![[LOOP6]], ![[LOOPPROP4]]}

llvm/include/llvm/Analysis/LoopInfo.h

   
 /// Returns true if Name is applied to TheLoop and enabled.
 bool getBooleanLoopAttribute(const Loop *TheLoop, StringRef Name);
 
 /// Find named metadata for a loop with an integer value.
 llvm::Optional<int>
 getOptionalIntLoopAttribute(const Loop *TheLoop, StringRef Name);
 
+/// Find named metadata for a loop with an integer value. Return \p Default if
+/// not set.
+int getIntLoopAttribute(const Loop *TheLoop, StringRef Name, int Default = 0);
+
 /// Find string metadata for loop
 ///
 /// If it has a value (e.g. {"llvm.distribute", 1} return the value as an
 /// operand or null otherwise.  If the string metadata is not found return
 /// Optional's not-a-value.
 Optional<const MDOperand *> findStringMetadataForLoop(const Loop *TheLoop,
                                                       StringRef Name);
 

llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h

public:
   ///
   /// \returns A list of generated loops. Contains twice as many loops as the
   ///          input loop nest; the first half are the floor loops and the
   ///          second half are the tile loops.
   std::vector<CanonicalLoopInfo *>
   tileLoops(DebugLoc DL, ArrayRef<CanonicalLoopInfo *> Loops,
             ArrayRef<Value *> TileSizes);
 
+  /// Fully unroll a loop.
+  ///
+  /// Instead of unrolling the loop immediately (and duplicating its body
+  /// instructions), it is deferred to LLVM's LoopUnrollPass by adding loop
+  /// metadata.
+  ///
+  /// \param DL   Debug location for instructions added by unrolling.
+  /// \param Loop The loop to unroll. The loop will be invalidated.
+  void unrollLoopFull(DebugLoc DL, CanonicalLoopInfo *Loop);
+
+  /// Fully or partially unroll a loop. How the loop is unrolled is determined
+  /// using LLVM's LoopUnrollPass.
+  ///
+  /// \param DL   Debug location for instructions added by unrolling.
+  /// \param Loop The loop to unroll. The loop will be invalidated.
+  void unrollLoopHeuristic(DebugLoc DL, CanonicalLoopInfo *Loop);
+
+  /// Partially unroll a loop.
+  ///
+  /// The CanonicalLoopInfo of the unrolled loop for use with chained
+  /// loop-associated directive can be requested using \p UnrolledCLI. Not
+  /// needing the CanonicalLoopInfo allows more efficient code generation by
+  /// deferring the actual unrolling to the LoopUnrollPass using loop metadata.
+  /// A loop-associated directive applied to the unrolled loop needs to know the
+  /// new trip count which means that if using a heuristically determined unroll
+  /// factor (\p Factor == 0), that factor must be computed immediately. We are
+  /// using the same logic as the LoopUnrollPass to derived the unroll factor,
+  /// but which assumes that some canonicalization has taken place (e.g.
+  /// Mem2Reg, LICM, GVN, Inlining, etc.). That is, the heuristic will perform
+  /// better when the unrolled loop's CanonicalLoopInfo is not needed.
+  ///
+  /// \param DL          Debug location for instructions added by unrolling.
+  /// \param Loop        The loop to unroll. The loop will be invalidated.
+  /// \param Factor      The factor to unroll the loop by. A factor of 0
+  ///                    indicates that a heuristic should be used to determine
+  ///                    the unroll-factor.
+  /// \param UnrolledCLI If non-null, receives the CanonicalLoopInfo of the
+  ///                    partially unrolled loop. Otherwise, uses loop metadata
+  ///                    to defer unrolling to the LoopUnrollPass.
+  void unrollLoopPartial(DebugLoc DL, CanonicalLoopInfo *Loop, int32_t Factor,
+                         CanonicalLoopInfo **UnrolledCLI);
+
   /// Generator for '#omp flush'
   ///
   /// \param Loc The location where the flush directive was encountered
   void createFlush(const LocationDescription &Loc);
 
   /// Generator for '#omp taskwait'
   ///
   /// \param Loc The location where the taskwait directive was encountered.

llvm/lib/Analysis/LoopInfo.cpp

llvm::Optional<int> llvm::getOptionalIntLoopAttribute(const Loop *TheLoop,
 
   ConstantInt *IntMD = mdconst::extract_or_null<ConstantInt>(AttrMD->get());
   if (!IntMD)
     return None;
 
   return IntMD->getSExtValue();
 }
 
+int llvm::getIntLoopAttribute(const Loop *TheLoop, StringRef Name,
+                              int Default) {
+  return getOptionalIntLoopAttribute(TheLoop, Name).getValueOr(Default);
+}
+
 static const char *LLVMLoopMustProgress = "llvm.loop.mustprogress";
 
 bool llvm::hasMustProgress(const Loop *L) {
   return getBooleanLoopAttribute(L, LLVMLoopMustProgress);
 }
 
 bool llvm::isMustProgress(const Loop *L) {
   return L->getHeader()->getParent()->mustProgress() || hasMustProgress(L);

llvm/lib/Frontend/OpenMP/CMakeLists.txt

 add_llvm_component_library(LLVMFrontendOpenMP
   OMP.cpp
   OMPContext.cpp
   OMPIRBuilder.cpp
 
   ADDITIONAL_HEADER_DIRS
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/Frontend
   ${LLVM_MAIN_INCLUDE_DIR}/llvm/Frontend/OpenMP
 
   DEPENDS
   intrinsics_gen
   omp_gen
 
   LINK_COMPONENTS
   Core
   Support
+  Passes
   TransformUtils
   )

llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

 /// convenient way to create LLVM instructions for OpenMP directives.
 ///
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Frontend/OpenMP/OMPIRBuilder.h"
 
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/Triple.h"
+#include "llvm/Analysis/CodeMetrics.h"
+#include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/IR/CFG.h"
 #include "llvm/IR/DebugInfo.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/IR/Value.h"
+#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Error.h"
+#include "llvm/Support/TargetRegistry.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/Target/TargetOptions.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/CodeExtractor.h"
+#include "llvm/Transforms/Utils/LoopPeel.h"
+#include "llvm/Transforms/Utils/UnrollLoop.h"
 
 #include <sstream>
 
 #define DEBUG_TYPE "openmp-ir-builder"
 
 using namespace llvm;
 using namespace omp;
 
 static cl::opt<bool>
     OptimisticAttributes("openmp-ir-builder-optimistic-attributes", cl::Hidden,
                          cl::desc("Use optimistic attributes describing "
                                   "'as-if' properties of runtime calls."),
                          cl::init(false));
 
+static cl::opt<double> UnrollThresholdFactor(
+    "openmp-ir-builder-unroll-threshold-factor", cl::Hidden,
+    cl::desc("Factor for the unroll threshold to account for code "
+             "simplifications still taking place"),
+    cl::init(1.5));
+
 void OpenMPIRBuilder::addAttributes(omp::RuntimeFunction FnID, Function &Fn) {
   LLVMContext &Ctx = Fn.getContext();
 
   // Get the function's current attributes.
   auto Attrs = Fn.getAttributes();
   auto FnAttrs = Attrs.getFnAttrs();
   auto RetAttrs = Attrs.getRetAttrs();
   SmallVector<AttributeSet, 4> ArgAttrs;

 
 #ifndef NDEBUG
   for (CanonicalLoopInfo *GenL : Result)
     GenL->assertOK();
 #endif
   return Result;
 }
 
+/// Attach loop metadata \p Properties to the loop described by \p Loop. If the
+/// loop already has metadata, the loop properties are appended.
+static void addLoopMetadata(CanonicalLoopInfo *Loop,
+                            ArrayRef<Metadata *> Properties) {
+  assert(Loop->isValid() && "Expecting a valid CanonicalLoopInfo");
+
+  // Nothing to do if no property to attach.
+  if (Properties.empty())
+    return;
+
+  LLVMContext &Ctx = Loop->getFunction()->getContext();
+  SmallVector<Metadata *> NewLoopProperties;
+  NewLoopProperties.push_back(nullptr);
+
+  // If the loop already has metadata, prepend it to the new metadata.
+  BasicBlock *Latch = Loop->getLatch();
+  assert(Latch && "A valid CanonicalLoopInfo must have a unique latch");
+  MDNode *Existing = Latch->getTerminator()->getMetadata(LLVMContext::MD_loop);
+  if (Existing)
+    append_range(NewLoopProperties, drop_begin(Existing->operands(), 1));
+
+  append_range(NewLoopProperties, Properties);
+  MDNode *LoopID = MDNode::getDistinct(Ctx, NewLoopProperties);
+  LoopID->replaceOperandWith(0, LoopID);
+
+  Latch->getTerminator()->setMetadata(LLVMContext::MD_loop, LoopID);
+}
+
+void OpenMPIRBuilder::unrollLoopFull(DebugLoc, CanonicalLoopInfo *Loop) {
+  LLVMContext &Ctx = Builder.getContext();
+  addLoopMetadata(
+      Loop, {MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.enable")),
+             MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.full"))});
+}
+
+void OpenMPIRBuilder::unrollLoopHeuristic(DebugLoc, CanonicalLoopInfo *Loop) {
+  LLVMContext &Ctx = Builder.getContext();
+  addLoopMetadata(
+      Loop, {
+                MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.enable")),
+            });
+}
+
+/// Create the TargetMachine object to query the backend for optimization
+/// preferences.
+///
+/// Ideally, this would be passed from the front-end to the OpenMPBuilder, but
+/// e.g. Clang does not pass it to its CodeGen layer and creates it only when
+/// needed for the LLVM pass pipline. We use some default options to avoid
+/// having to pass too many settings from the frontend that probably do not
+/// matter.
+///
+/// Currently, TargetMachine is only used sometimes by the unrollLoopPartial
+/// method. If we are going to use TargetMachine for more purposes, especially
+/// those that are sensitive to TargetOptions, RelocModel and CodeModel, it
+/// might become be worth requiring front-ends to pass on their TargetMachine,
+/// or at least cache it between methods. Note that while fontends such as Clang
+/// have just a single main TargetMachine per translation unit, "target-cpu" and
+/// "target-features" that determine the TargetMachine are per-function and can
+/// be overrided using __attribute__((target("OPTIONS"))).
+static std::unique_ptr<TargetMachine>
+createTargetMachine(Function *F, CodeGenOpt::Level OptLevel) {
+  Module *M = F->getParent();
+
+  StringRef CPU = F->getFnAttribute("target-cpu").getValueAsString();
+  StringRef Features = F->getFnAttribute("target-features").getValueAsString();
+  const std::string &Triple = M->getTargetTriple();
+
+  std::string Error;
+  const llvm::Target *TheTarget = TargetRegistry::lookupTarget(Triple, Error);
+  if (!TheTarget)
+    return {};
+
+  llvm::TargetOptions Options;
+  return std::unique_ptr<TargetMachine>(TheTarget->createTargetMachine(
+      Triple, CPU, Features, Options, /*RelocModel=*/None, /*CodeModel=*/None,
+      OptLevel));
+}
+
+/// Heuristically determine the best-performant unroll factor for \p CLI. This
+/// depends on the target processor. We are re-using the same heuristics as the
+/// LoopUnrollPass.
+static int32_t computeHeuristicUnrollFactor(CanonicalLoopInfo *CLI) {
+  Function *F = CLI->getFunction();
+
+  // Assume the user requests the most aggressive unrolling, even if the rest of
+  // the code is optimized using a lower setting.
+  CodeGenOpt::Level OptLevel = CodeGenOpt::Aggressive;
+  std::unique_ptr<TargetMachine> TM = createTargetMachine(F, OptLevel);
+
+  llvm::PassBuilder PB;
+  FunctionAnalysisManager FAM;
+  PB.registerFunctionAnalyses(FAM);
+  TargetIRAnalysis TIRA;
+  if (TM)
+    TIRA = TargetIRAnalysis(
+        [&](const Function &F) { return TM->getTargetTransformInfo(F); });
+  TargetIRAnalysis::Result &&TTI = TIRA.run(*F, FAM);
+  ScalarEvolutionAnalysis SEA;
+  ScalarEvolution &&SE = SEA.run(*F, FAM);
+  DominatorTreeAnalysis DTA;
+  DominatorTree &&DT = DTA.run(*F, FAM);
+  LoopAnalysis LIA;
+  LoopInfo &&LI = LIA.run(*F, FAM);
+  AssumptionAnalysis ACT;
+  AssumptionCache &&AC = ACT.run(*F, FAM);
+  OptimizationRemarkEmitter ORE{F};
+
+  Loop *L = LI.getLoopFor(CLI->getHeader());
+  assert(L && "Expecting CanonicalLoopInfo to be recognized as a loop");
+
+  TargetTransformInfo::UnrollingPreferences UP =
+      gatherUnrollingPreferences(L, SE, TTI,
+                                 /*BlockFrequencyInfo=*/nullptr,
+                                 /*ProfileSummaryInfo=*/nullptr, ORE, OptLevel,
+                                 /*UserThreshold=*/None,
+                                 /*UserCount=*/None,
+                                 /*UserAllowPartial=*/true,
+                                 /*UserAllowRuntime=*/true,
+                                 /*UserUpperBound=*/None,
+                                 /*UserFullUnrollMaxCount=*/None);
+
+  UP.Force = true;
+
+  // Account for additional optimizations taking place before the LoopUnrollPass
+  // would unroll the loop.
+  UP.Threshold *= UnrollThresholdFactor;
+  UP.PartialThreshold *= UnrollThresholdFactor;
+
+  // Use normal unroll factors even if the rest of the code is optimized for
+  // size.
+  UP.OptSizeThreshold = UP.Threshold;
+  UP.PartialOptSizeThreshold = UP.PartialThreshold;
+
+  LLVM_DEBUG(dbgs() << "Unroll heuristic thresholds:\n"
+                    << "  Threshold=" << UP.Threshold << "\n"
+                    << "  PartialThreshold=" << UP.PartialThreshold << "\n"
+                    << "  OptSizeThreshold=" << UP.OptSizeThreshold << "\n"
+                    << "  PartialOptSizeThreshold="
+                    << UP.PartialOptSizeThreshold << "\n");
+
+  // Disable peeling.
+  TargetTransformInfo::PeelingPreferences PP =
+      gatherPeelingPreferences(L, SE, TTI,
+                               /*UserAllowPeeling=*/false,
+                               /*UserAllowProfileBasedPeeling=*/false,
+                               /*UserUnrollingSpecficValues=*/false);
+
+  SmallPtrSet<const Value *, 32> EphValues;
+  CodeMetrics::collectEphemeralValues(L, &AC, EphValues);
+
+  // Assume that reads and writes to stack variables can be eliminated by
+  // Mem2Reg, SROA or LICM. That is, don't count them towards the loop body's
+  // size.
+  for (BasicBlock *BB : L->blocks()) {
+    for (Instruction &I : *BB) {
+      Value *Ptr;
+      if (auto *Load = dyn_cast<LoadInst>(&I)) {
+        Ptr = Load->getPointerOperand();
+      } else if (auto *Store = dyn_cast<StoreInst>(&I)) {
+        Ptr = Store->getPointerOperand();
+      } else
+        continue;
+
+      Ptr = Ptr->stripPointerCasts();
+
+      if (auto *Alloca = dyn_cast<AllocaInst>(Ptr)) {
+        if (Alloca->getParent() == &F->getEntryBlock())
+          EphValues.insert(&I);
+      }
+    }
+  }
+
+  unsigned NumInlineCandidates;
+  bool NotDuplicatable;
+  bool Convergent;
+  unsigned LoopSize =
+      ApproximateLoopSize(L, NumInlineCandidates, NotDuplicatable, Convergent,
+                          TTI, EphValues, UP.BEInsns);
+  LLVM_DEBUG(dbgs() << "Estimated loop size is " << LoopSize << "\n");
+
+  // Loop is not unrollable if the loop contains certain instructions.
+  if (NotDuplicatable || Convergent) {
+    LLVM_DEBUG(dbgs() << "Loop not considered unrollable\n");
+    return 1;
+  }
+
+  // TODO: Determine trip count of \p CLI if constant, computeUnrollCount might
+  // be able to use it.
+  int TripCount = 0;
+  int MaxTripCount = 0;
+  bool MaxOrZero = false;
+  unsigned TripMultiple = 0;
+
+  bool UseUpperBound = false;
+  computeUnrollCount(L, TTI, DT, &LI, SE, EphValues, &ORE, TripCount,
+                     MaxTripCount, MaxOrZero, TripMultiple, LoopSize, UP, PP,
+                     UseUpperBound);
+  unsigned Factor = UP.Count;
+  LLVM_DEBUG(dbgs() << "Suggesting unroll factor of " << Factor << "\n");
+
+  // This function returns 1 to signal to not unroll a loop.
+  if (Factor == 0)
+    return 1;
+  return Factor;
+}
+
+void OpenMPIRBuilder::unrollLoopPartial(DebugLoc DL, CanonicalLoopInfo *Loop,
+                                        int32_t Factor,
+                                        CanonicalLoopInfo **UnrolledCLI) {
+  assert(Factor >= 0 && "Unroll factor must not be negative");
+
+  Function *F = Loop->getFunction();
+  LLVMContext &Ctx = F->getContext();
+
+  // If the unrolled loop is not used for another loop-associated directive, it
+  // is sufficient to add metadata for the LoopUnrollPass.
+  if (!UnrolledCLI) {
+    SmallVector<Metadata *, 2> LoopMetadata;
+    LoopMetadata.push_back(
+        MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.enable")));
+
+    if (Factor >= 1) {
+      ConstantAsMetadata *FactorConst = ConstantAsMetadata::get(
+          ConstantInt::get(Type::getInt32Ty(Ctx), APInt(32, Factor)));
+      LoopMetadata.push_back(MDNode::get(
+          Ctx, {MDString::get(Ctx, "llvm.loop.unroll.count"), FactorConst}));
+    }
+
+    addLoopMetadata(Loop, LoopMetadata);
+    return;
+  }
+
+  // Heuristically determine the unroll factor.
+  if (Factor == 0)
+    Factor = computeHeuristicUnrollFactor(Loop);
+
+  // No change required with unroll factor 1.
+  if (Factor == 1) {
+    *UnrolledCLI = Loop;
+    return;
+  }
+
+  assert(Factor >= 2 &&
+         "unrolling only makes sense with a factor of 2 or larger");
+
+  Type *IndVarTy = Loop->getIndVarType();
+
+  // Apply partial unrolling by tiling the loop by the unroll-factor, then fully
+  // unroll the inner loop.
+  Value *FactorVal =
+      ConstantInt::get(IndVarTy, APInt(IndVarTy->getIntegerBitWidth(), Factor,
+                                       /*isSigned=*/false));
+  std::vector<CanonicalLoopInfo *> LoopNest =
+      tileLoops(DL, {Loop}, {FactorVal});
+  assert(LoopNest.size() == 2 && "Expect 2 loops after tiling");
+  *UnrolledCLI = LoopNest[0];
+  CanonicalLoopInfo *InnerLoop = LoopNest[1];
+
+  // LoopUnrollPass can only fully unroll loops with constant trip count.
+  // Unroll by the unroll factor with a fallback epilog for the remainder
+  // iterations if necessary.
+  ConstantAsMetadata *FactorConst = ConstantAsMetadata::get(
+      ConstantInt::get(Type::getInt32Ty(Ctx), APInt(32, Factor)));
+  addLoopMetadata(
+      InnerLoop,
+      {MDNode::get(Ctx, MDString::get(Ctx, "llvm.loop.unroll.enable")),
+       MDNode::get(
+           Ctx, {MDString::get(Ctx, "llvm.loop.unroll.count"), FactorConst})});
+
+#ifndef NDEBUG
+  (*UnrolledCLI)->assertOK();
+#endif
+}
+
 OpenMPIRBuilder::InsertPointTy
 OpenMPIRBuilder::createCopyPrivate(const LocationDescription &Loc,
                                    llvm::Value *BufSize, llvm::Value *CpyBuf,
                                    llvm::Value *CpyFn, llvm::Value *DidIt) {
   if (!updateToLocation(Loc))
     return Loc.IP;
 
   Constant *SrcLocStr = getOrCreateSrcLocStr(Loc);

llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp

 #include "llvm/Frontend/OpenMP/OMPIRBuilder.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/DIBuilder.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/InstIterator.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Verifier.h"
+#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "gtest/gtest.h"
 
 using namespace llvm;
 using namespace omp;
 
 namespace {
 
void SetUp() override {
     DL = DILocation::get(Ctx, 3, 7, Scope);
   }
 
   void TearDown() override {
     BB = nullptr;
     M.reset();
   }
 
+  /// Create a function with a simple loop that calls printf using the logical
+  /// loop counter for use with tests that need a CanonicalLoopInfo object.
+  CanonicalLoopInfo *buildSingleLoopFunction(DebugLoc DL,
+                                             OpenMPIRBuilder &OMPBuilder,
+                                             Instruction **Call = nullptr,
+                                             BasicBlock **BodyCode = nullptr) {
+    OMPBuilder.initialize();
+    F->setName("func");
+
+    IRBuilder<> Builder(BB);
+    OpenMPIRBuilder::LocationDescription Loc({Builder.saveIP(), DL});
+    Value *TripCount = F->getArg(0);
+
+    auto LoopBodyGenCB = [&](OpenMPIRBuilder::InsertPointTy CodeGenIP,
+                             llvm::Value *LC) {
+      Builder.restoreIP(CodeGenIP);
+      if (BodyCode)
+        *BodyCode = Builder.GetInsertBlock();
+
+      // Add something that consumes the induction variable to the body.
+      CallInst *CallInst = createPrintfCall(Builder, "%d\\n", {LC});
+      if (Call)
+        *Call = CallInst;
+    };
+    CanonicalLoopInfo *Loop =
+        OMPBuilder.createCanonicalLoop(Loc, LoopBodyGenCB, TripCount);
+
+    // Finalize the function.
+    Builder.restoreIP(Loop->getAfterIP());
+    Builder.CreateRetVoid();
+
+    return Loop;
+  }
+
   LLVMContext Ctx;
   std::unique_ptr<Module> M;
   Function *F;
   BasicBlock *BB;
   DebugLoc DL;
 };
 
 class OpenMPIRBuilderTestWithParams
TEST_F(OpenMPIRBuilderTest, CollapseNestedLoops) {
   EXPECT_EQ(InnerIV->getOperand(0), Collapsed->getIndVar());
   EXPECT_EQ(InnerIV->getOperand(1), InnerTripCount);
 
   EXPECT_EQ(InbetweenLead->getOperand(1), OuterIV);
   EXPECT_EQ(InbetweenTrail->getOperand(1), OuterIV);
 }
 
 TEST_F(OpenMPIRBuilderTest, TileSingleLoop) {
-  using InsertPointTy = OpenMPIRBuilder::InsertPointTy;
   OpenMPIRBuilder OMPBuilder(*M);
-  OMPBuilder.initialize();
-  F->setName("func");
-
-  IRBuilder<> Builder(BB);
-  OpenMPIRBuilder::LocationDescription Loc({Builder.saveIP(), DL});
-  Value *TripCount = F->getArg(0);
-
-  BasicBlock *BodyCode = nullptr;
-  Instruction *Call = nullptr;
-  auto LoopBodyGenCB = [&](InsertPointTy CodeGenIP, llvm::Value *LC) {
-    Builder.restoreIP(CodeGenIP);
-    BodyCode = Builder.GetInsertBlock();
-
-    // Add something that consumes the induction variable to the body.
-    Call = createPrintfCall(Builder, "%d\\n", {LC});
-  };
+  Instruction *Call;
+  BasicBlock *BodyCode;
   CanonicalLoopInfo *Loop =
-      OMPBuilder.createCanonicalLoop(Loc, LoopBodyGenCB, TripCount);
-
-  // Finalize the function.
-  Builder.restoreIP(Loop->getAfterIP());
-  Builder.CreateRetVoid();
+      buildSingleLoopFunction(DL, OMPBuilder, &Call, &BodyCode);
 
   Instruction *OrigIndVar = Loop->getIndVar();
   EXPECT_EQ(Call->getOperand(1), OrigIndVar);
 
   // Tile the loop.
   Constant *TileSize = ConstantInt::get(Loop->getIndVarType(), APInt(32, 7));
   std::vector<CanonicalLoopInfo *> GenLoops =
       OMPBuilder.tileLoops(DL, {Loop}, {TileSize});
TEST_F(OpenMPIRBuilderTest, TileSingleLoopCounts) {
 
   // Finalize the function.
   Builder.CreateRetVoid();
   OMPBuilder.finalize();
 
   EXPECT_FALSE(verifyModule(*M, &errs()));
 }
 
+TEST_F(OpenMPIRBuilderTest, UnrollLoopFull) {
+  OpenMPIRBuilder OMPBuilder(*M);
+
+  CanonicalLoopInfo *CLI = buildSingleLoopFunction(DL, OMPBuilder);
+
+  // Unroll the loop.
+  OMPBuilder.unrollLoopFull(DL, CLI);
+
+  OMPBuilder.finalize();
+  EXPECT_FALSE(verifyModule(*M, &errs()));
+
+  PassBuilder PB;
+  FunctionAnalysisManager FAM;
+  PB.registerFunctionAnalyses(FAM);
+  LoopInfo &LI = FAM.getResult<LoopAnalysis>(*F);
+
+  const std::vector<Loop *> &TopLvl = LI.getTopLevelLoops();
+  EXPECT_EQ(TopLvl.size(), 1u);
+
+  Loop *L = TopLvl.front();
+  EXPECT_TRUE(getBooleanLoopAttribute(L, "llvm.loop.unroll.enable"));
+  EXPECT_TRUE(getBooleanLoopAttribute(L, "llvm.loop.unroll.full"));
+}
+
+TEST_F(OpenMPIRBuilderTest, UnrollLoopPartial) {
+  OpenMPIRBuilder OMPBuilder(*M);
+  CanonicalLoopInfo *CLI = buildSingleLoopFunction(DL, OMPBuilder);
+
+  // Unroll the loop.
+  CanonicalLoopInfo *UnrolledLoop = nullptr;
+  OMPBuilder.unrollLoopPartial(DL, CLI, 5, &UnrolledLoop);
+  ASSERT_NE(UnrolledLoop, nullptr);
+
+  OMPBuilder.finalize();
+  EXPECT_FALSE(verifyModule(*M, &errs()));
+  UnrolledLoop->assertOK();
+
+  PassBuilder PB;
+  FunctionAnalysisManager FAM;
+  PB.registerFunctionAnalyses(FAM);
+  LoopInfo &LI = FAM.getResult<LoopAnalysis>(*F);
+
+  const std::vector<Loop *> &TopLvl = LI.getTopLevelLoops();
+  EXPECT_EQ(TopLvl.size(), 1u);
+  Loop *Outer = TopLvl.front();
+  EXPECT_EQ(Outer->getHeader(), UnrolledLoop->getHeader());
+  EXPECT_EQ(Outer->getLoopLatch(), UnrolledLoop->getLatch());
+  EXPECT_EQ(Outer->getExitingBlock(), UnrolledLoop->getCond());
+  EXPECT_EQ(Outer->getExitBlock(), UnrolledLoop->getExit());
+
+  EXPECT_EQ(Outer->getSubLoops().size(), 1u);
+  Loop *Inner = Outer->getSubLoops().front();
+
+  EXPECT_TRUE(getBooleanLoopAttribute(Inner, "llvm.loop.unroll.enable"));
+  EXPECT_EQ(getIntLoopAttribute(Inner, "llvm.loop.unroll.count"), 5);
+}
+
+TEST_F(OpenMPIRBuilderTest, UnrollLoopHeuristic) {
+  OpenMPIRBuilder OMPBuilder(*M);
+
+  CanonicalLoopInfo *CLI = buildSingleLoopFunction(DL, OMPBuilder);
+
+  // Unroll the loop.
+  OMPBuilder.unrollLoopHeuristic(DL, CLI);
+
+  OMPBuilder.finalize();
+  EXPECT_FALSE(verifyModule(*M, &errs()));
+
+  PassBuilder PB;
+  FunctionAnalysisManager FAM;
+  PB.registerFunctionAnalyses(FAM);
+  LoopInfo &LI = FAM.getResult<LoopAnalysis>(*F);
+
+  const std::vector<Loop *> &TopLvl = LI.getTopLevelLoops();
+  EXPECT_EQ(TopLvl.size(), 1u);
+
+  Loop *L = TopLvl.front();
+  EXPECT_TRUE(getBooleanLoopAttribute(L, "llvm.loop.unroll.enable"));
+}
+
 TEST_F(OpenMPIRBuilderTest, StaticWorkShareLoop) {
   using InsertPointTy = OpenMPIRBuilder::InsertPointTy;
   OpenMPIRBuilder OMPBuilder(*M);
   OMPBuilder.initialize();
   IRBuilder<> Builder(BB);
   OpenMPIRBuilder::LocationDescription Loc({Builder.saveIP(), DL});
 
   Type *LCTy = Type::getInt32Ty(Ctx);
   Value *StartVal = ConstantInt::get(LCTy, 10);