[OPENMP] Codegen for distribute directive

This patch provide basic implementation of codegen for teams directive, excluding all clauses except dist_schedule. It also fixes parts of AST reader/writer to enable correct pre-compiled header handling.

http://reviews.llvm.org/D17170

llvm-svn: 262741

6 files changed, 360 insertions, 106 deletions

diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index a38d6fdd3af..f45afaf2c89 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -425,6 +425,9 @@ enum OpenMPSchedType {
   OMP_ord_runtime = 69,
   OMP_ord_auto = 70,
   OMP_sch_default = OMP_sch_static,
+  /// \brief dist_schedule types
+  OMP_dist_sch_static_chunked = 91,
+  OMP_dist_sch_static = 92,
 };
 
 enum OpenMPRTLFunction {
@@ -2148,12 +2151,26 @@ static OpenMPSchedType getRuntimeSchedule(OpenMPScheduleClauseKind ScheduleKind,
   llvm_unreachable("Unexpected runtime schedule");
 }
 
+/// \brief Map the OpenMP distribute schedule to the runtime enumeration.
+static OpenMPSchedType
+getRuntimeSchedule(OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) {
+  // only static is allowed for dist_schedule
+  return Chunked ? OMP_dist_sch_static_chunked : OMP_dist_sch_static;
+}
+
 bool CGOpenMPRuntime::isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind,
                                          bool Chunked) const {
   auto Schedule = getRuntimeSchedule(ScheduleKind, Chunked, /*Ordered=*/false);
   return Schedule == OMP_sch_static;
 }
 
+bool CGOpenMPRuntime::isStaticNonchunked(
+    OpenMPDistScheduleClauseKind ScheduleKind, bool Chunked) const {
+  auto Schedule = getRuntimeSchedule(ScheduleKind, Chunked);
+  return Schedule == OMP_dist_sch_static;
+}
+
+
 bool CGOpenMPRuntime::isDynamic(OpenMPScheduleClauseKind ScheduleKind) const {
   auto Schedule =
       getRuntimeSchedule(ScheduleKind, /*Chunked=*/false, /*Ordered=*/false);
@@ -2194,6 +2211,55 @@ void CGOpenMPRuntime::emitForDispatchInit(CodeGenFunction &CGF,
   CGF.EmitRuntimeCall(createDispatchInitFunction(IVSize, IVSigned), Args);
 }
 
+static void emitForStaticInitCall(CodeGenFunction &CGF,
+                                  SourceLocation Loc,
+                                  llvm::Value * UpdateLocation,
+                                  llvm::Value * ThreadId,
+                                  llvm::Constant * ForStaticInitFunction,
+                                  OpenMPSchedType Schedule,
+                                  unsigned IVSize, bool IVSigned, bool Ordered,
+                                  Address IL, Address LB, Address UB,
+                                  Address ST, llvm::Value *Chunk) {
+  if (!CGF.HaveInsertPoint())
+     return;
+
+   assert(!Ordered);
+   assert(Schedule == OMP_sch_static || Schedule == OMP_sch_static_chunked ||
+          Schedule == OMP_ord_static || Schedule == OMP_ord_static_chunked ||
+          Schedule == OMP_dist_sch_static ||
+          Schedule == OMP_dist_sch_static_chunked);
+
+   // Call __kmpc_for_static_init(
+   //          ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
+   //          kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
+   //          kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
+   //          kmp_int[32|64] incr, kmp_int[32|64] chunk);
+   if (Chunk == nullptr) {
+     assert((Schedule == OMP_sch_static || Schedule == OMP_ord_static ||
+             Schedule == OMP_dist_sch_static) &&
+            "expected static non-chunked schedule");
+     // If the Chunk was not specified in the clause - use default value 1.
+       Chunk = CGF.Builder.getIntN(IVSize, 1);
+   } else {
+     assert((Schedule == OMP_sch_static_chunked ||
+             Schedule == OMP_ord_static_chunked ||
+             Schedule == OMP_dist_sch_static_chunked) &&
+            "expected static chunked schedule");
+   }
+   llvm::Value *Args[] = {
+     UpdateLocation,
+     ThreadId,
+     CGF.Builder.getInt32(Schedule), // Schedule type
+     IL.getPointer(),                // &isLastIter
+     LB.getPointer(),                // &LB
+     UB.getPointer(),                // &UB
+     ST.getPointer(),                // &Stride
+     CGF.Builder.getIntN(IVSize, 1), // Incr
+     Chunk                           // Chunk
+   };
+   CGF.EmitRuntimeCall(ForStaticInitFunction, Args);
+}
+
 void CGOpenMPRuntime::emitForStaticInit(CodeGenFunction &CGF,
                                         SourceLocation Loc,
                                         OpenMPScheduleClauseKind ScheduleKind,
@@ -2201,41 +2267,27 @@ void CGOpenMPRuntime::emitForStaticInit(CodeGenFunction &CGF,
                                         bool Ordered, Address IL, Address LB,
                                         Address UB, Address ST,
                                         llvm::Value *Chunk) {
-  if (!CGF.HaveInsertPoint())
-    return;
-  OpenMPSchedType Schedule =
-    getRuntimeSchedule(ScheduleKind, Chunk != nullptr, Ordered);
-  assert(!Ordered);
-  assert(Schedule == OMP_sch_static || Schedule == OMP_sch_static_chunked ||
-         Schedule == OMP_ord_static || Schedule == OMP_ord_static_chunked);
-
-  // Call __kmpc_for_static_init(
-  //          ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
-  //          kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
-  //          kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
-  //          kmp_int[32|64] incr, kmp_int[32|64] chunk);
-  if (Chunk == nullptr) {
-    assert((Schedule == OMP_sch_static || Schedule == OMP_ord_static) &&
-           "expected static non-chunked schedule");
-    // If the Chunk was not specified in the clause - use default value 1.
-      Chunk = CGF.Builder.getIntN(IVSize, 1);
-  } else {
-    assert((Schedule == OMP_sch_static_chunked ||
-            Schedule == OMP_ord_static_chunked) &&
-           "expected static chunked schedule");
-  }
-  llvm::Value *Args[] = {
-      emitUpdateLocation(CGF, Loc),
-      getThreadID(CGF, Loc),
-      CGF.Builder.getInt32(Schedule), // Schedule type
-      IL.getPointer(),                // &isLastIter
-      LB.getPointer(),                // &LB
-      UB.getPointer(),                // &UB
-      ST.getPointer(),                // &Stride
-      CGF.Builder.getIntN(IVSize, 1), // Incr
-      Chunk                           // Chunk
-  };
-  CGF.EmitRuntimeCall(createForStaticInitFunction(IVSize, IVSigned), Args);
+  OpenMPSchedType ScheduleNum = getRuntimeSchedule(ScheduleKind, Chunk != nullptr,
+                                                   Ordered);
+  auto *UpdatedLocation = emitUpdateLocation(CGF, Loc);
+  auto *ThreadId = getThreadID(CGF, Loc);
+  auto *StaticInitFunction = createForStaticInitFunction(IVSize, IVSigned);
+  emitForStaticInitCall(CGF, Loc, UpdatedLocation, ThreadId, StaticInitFunction,
+      ScheduleNum, IVSize, IVSigned, Ordered, IL, LB, UB, ST, Chunk);
+}
+
+void CGOpenMPRuntime::emitDistributeStaticInit(CodeGenFunction &CGF,
+    SourceLocation Loc, OpenMPDistScheduleClauseKind SchedKind,
+    unsigned IVSize, bool IVSigned,
+    bool Ordered, Address IL, Address LB,
+    Address UB, Address ST,
+    llvm::Value *Chunk) {
+  OpenMPSchedType ScheduleNum = getRuntimeSchedule(SchedKind, Chunk != nullptr);
+  auto *UpdatedLocation = emitUpdateLocation(CGF, Loc);
+  auto *ThreadId = getThreadID(CGF, Loc);
+  auto *StaticInitFunction = createForStaticInitFunction(IVSize, IVSigned);
+  emitForStaticInitCall(CGF, Loc, UpdatedLocation, ThreadId, StaticInitFunction,
+      ScheduleNum, IVSize, IVSigned, Ordered, IL, LB, UB, ST, Chunk);
 }
 
 void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.h b/clang/lib/CodeGen/CGOpenMPRuntime.h
index 0044bb706f7..975e10ad292 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.h
@@ -493,6 +493,14 @@ public:
   virtual bool isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind,
                                   bool Chunked) const;
 
+  /// \brief Check if the specified \a ScheduleKind is static non-chunked.
+  /// This kind of distribute directive is emitted without outer loop.
+  /// \param ScheduleKind Schedule kind specified in the 'dist_schedule' clause.
+  /// \param Chunked True if chunk is specified in the clause.
+  ///
+  virtual bool isStaticNonchunked(OpenMPDistScheduleClauseKind ScheduleKind,
+                                  bool Chunked) const;
+
   /// \brief Check if the specified \a ScheduleKind is dynamic.
   /// This kind of worksharing directive is emitted without outer loop.
   /// \param ScheduleKind Schedule Kind specified in the 'schedule' clause.
@@ -536,6 +544,31 @@ public:
                                  Address UB, Address ST,
                                  llvm::Value *Chunk = nullptr);
 
+  ///
+  /// \param CGF Reference to current CodeGenFunction.
+  /// \param Loc Clang source location.
+  /// \param SchedKind Schedule kind, specified by the 'dist_schedule' clause.
+  /// \param IVSize Size of the iteration variable in bits.
+  /// \param IVSigned Sign of the interation variable.
+  /// \param Ordered true if loop is ordered, false otherwise.
+  /// \param IL Address of the output variable in which the flag of the
+  /// last iteration is returned.
+  /// \param LB Address of the output variable in which the lower iteration
+  /// number is returned.
+  /// \param UB Address of the output variable in which the upper iteration
+  /// number is returned.
+  /// \param ST Address of the output variable in which the stride value is
+  /// returned nesessary to generated the static_chunked scheduled loop.
+  /// \param Chunk Value of the chunk for the static_chunked scheduled loop.
+  /// For the default (nullptr) value, the chunk 1 will be used.
+  ///
+  virtual void emitDistributeStaticInit(CodeGenFunction &CGF, SourceLocation Loc,
+                                        OpenMPDistScheduleClauseKind SchedKind,
+                                        unsigned IVSize, bool IVSigned,
+                                        bool Ordered, Address IL, Address LB,
+                                        Address UB, Address ST,
+                                        llvm::Value *Chunk = nullptr);
+
   /// \brief Call the appropriate runtime routine to notify that we finished
   /// iteration of the ordered loop with the dynamic scheduling.
   ///
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index 8434cdf2da2..8984aca7b70 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -1410,82 +1410,15 @@ void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
   CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_simd, CodeGen);
 }
 
-void CodeGenFunction::EmitOMPForOuterLoop(
-    OpenMPScheduleClauseKind ScheduleKind, bool IsMonotonic,
+void CodeGenFunction::EmitOMPOuterLoop(bool DynamicOrOrdered, bool IsMonotonic,
     const OMPLoopDirective &S, OMPPrivateScope &LoopScope, bool Ordered,
     Address LB, Address UB, Address ST, Address IL, llvm::Value *Chunk) {
   auto &RT = CGM.getOpenMPRuntime();
 
-  // Dynamic scheduling of the outer loop (dynamic, guided, auto, runtime).
-  const bool DynamicOrOrdered = Ordered || RT.isDynamic(ScheduleKind);
-
-  assert((Ordered ||
-          !RT.isStaticNonchunked(ScheduleKind, /*Chunked=*/Chunk != nullptr)) &&
-         "static non-chunked schedule does not need outer loop");
-
-  // Emit outer loop.
-  //
-  // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
-  // When schedule(dynamic,chunk_size) is specified, the iterations are
-  // distributed to threads in the team in chunks as the threads request them.
-  // Each thread executes a chunk of iterations, then requests another chunk,
-  // until no chunks remain to be distributed. Each chunk contains chunk_size
-  // iterations, except for the last chunk to be distributed, which may have
-  // fewer iterations. When no chunk_size is specified, it defaults to 1.
-  //
-  // When schedule(guided,chunk_size) is specified, the iterations are assigned
-  // to threads in the team in chunks as the executing threads request them.
-  // Each thread executes a chunk of iterations, then requests another chunk,
-  // until no chunks remain to be assigned. For a chunk_size of 1, the size of
-  // each chunk is proportional to the number of unassigned iterations divided
-  // by the number of threads in the team, decreasing to 1. For a chunk_size
-  // with value k (greater than 1), the size of each chunk is determined in the
-  // same way, with the restriction that the chunks do not contain fewer than k
-  // iterations (except for the last chunk to be assigned, which may have fewer
-  // than k iterations).
-  //
-  // When schedule(auto) is specified, the decision regarding scheduling is
-  // delegated to the compiler and/or runtime system. The programmer gives the
-  // implementation the freedom to choose any possible mapping of iterations to
-  // threads in the team.
-  //
-  // When schedule(runtime) is specified, the decision regarding scheduling is
-  // deferred until run time, and the schedule and chunk size are taken from the
-  // run-sched-var ICV. If the ICV is set to auto, the schedule is
-  // implementation defined
-  //
-  // while(__kmpc_dispatch_next(&LB, &UB)) {
-  //   idx = LB;
-  //   while (idx <= UB) { BODY; ++idx;
-  //   __kmpc_dispatch_fini_(4|8)[u](); // For ordered loops only.
-  //   } // inner loop
-  // }
-  //
-  // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
-  // When schedule(static, chunk_size) is specified, iterations are divided into
-  // chunks of size chunk_size, and the chunks are assigned to the threads in
-  // the team in a round-robin fashion in the order of the thread number.
-  //
-  // while(UB = min(UB, GlobalUB), idx = LB, idx < UB) {
-  //   while (idx <= UB) { BODY; ++idx; } // inner loop
-  //   LB = LB + ST;
-  //   UB = UB + ST;
-  // }
-  //
-
   const Expr *IVExpr = S.getIterationVariable();
   const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
   const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
 
-  if (DynamicOrOrdered) {
-    llvm::Value *UBVal = EmitScalarExpr(S.getLastIteration());
-    RT.emitForDispatchInit(*this, S.getLocStart(), ScheduleKind,
-                           IVSize, IVSigned, Ordered, UBVal, Chunk);
-  } else {
-    RT.emitForStaticInit(*this, S.getLocStart(), ScheduleKind,
-                         IVSize, IVSigned, Ordered, IL, LB, UB, ST, Chunk);
-  }
-
   auto LoopExit = getJumpDestInCurrentScope("omp.dispatch.end");
 
   // Start the loop with a block that tests the condition.
@@ -1565,6 +1498,111 @@ void CodeGenFunction::EmitOMPForOuterLoop(
   // Tell the runtime we are done.
   if (!DynamicOrOrdered)
     RT.emitForStaticFinish(*this, S.getLocEnd());
+
+}
+
+void CodeGenFunction::EmitOMPForOuterLoop(
+    OpenMPScheduleClauseKind ScheduleKind, bool IsMonotonic,
+    const OMPLoopDirective &S, OMPPrivateScope &LoopScope, bool Ordered,
+    Address LB, Address UB, Address ST, Address IL, llvm::Value *Chunk) {
+  auto &RT = CGM.getOpenMPRuntime();
+
+  // Dynamic scheduling of the outer loop (dynamic, guided, auto, runtime).
+  const bool DynamicOrOrdered = Ordered || RT.isDynamic(ScheduleKind);
+
+  assert((Ordered ||
+          !RT.isStaticNonchunked(ScheduleKind, /*Chunked=*/Chunk != nullptr)) &&
+         "static non-chunked schedule does not need outer loop");
+
+  // Emit outer loop.
+  //
+  // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
+  // When schedule(dynamic,chunk_size) is specified, the iterations are
+  // distributed to threads in the team in chunks as the threads request them.
+  // Each thread executes a chunk of iterations, then requests another chunk,
+  // until no chunks remain to be distributed. Each chunk contains chunk_size
+  // iterations, except for the last chunk to be distributed, which may have
+  // fewer iterations. When no chunk_size is specified, it defaults to 1.
+  //
+  // When schedule(guided,chunk_size) is specified, the iterations are assigned
+  // to threads in the team in chunks as the executing threads request them.
+  // Each thread executes a chunk of iterations, then requests another chunk,
+  // until no chunks remain to be assigned. For a chunk_size of 1, the size of
+  // each chunk is proportional to the number of unassigned iterations divided
+  // by the number of threads in the team, decreasing to 1. For a chunk_size
+  // with value k (greater than 1), the size of each chunk is determined in the
+  // same way, with the restriction that the chunks do not contain fewer than k
+  // iterations (except for the last chunk to be assigned, which may have fewer
+  // than k iterations).
+  //
+  // When schedule(auto) is specified, the decision regarding scheduling is
+  // delegated to the compiler and/or runtime system. The programmer gives the
+  // implementation the freedom to choose any possible mapping of iterations to
+  // threads in the team.
+  //
+  // When schedule(runtime) is specified, the decision regarding scheduling is
+  // deferred until run time, and the schedule and chunk size are taken from the
+  // run-sched-var ICV. If the ICV is set to auto, the schedule is
+  // implementation defined
+  //
+  // while(__kmpc_dispatch_next(&LB, &UB)) {
+  //   idx = LB;
+  //   while (idx <= UB) { BODY; ++idx;
+  //   __kmpc_dispatch_fini_(4|8)[u](); // For ordered loops only.
+  //   } // inner loop
+  // }
+  //
+  // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
+  // When schedule(static, chunk_size) is specified, iterations are divided into
+  // chunks of size chunk_size, and the chunks are assigned to the threads in
+  // the team in a round-robin fashion in the order of the thread number.
+  //
+  // while(UB = min(UB, GlobalUB), idx = LB, idx < UB) {
+  //   while (idx <= UB) { BODY; ++idx; } // inner loop
+  //   LB = LB + ST;
+  //   UB = UB + ST;
+  // }
+  //
+
+  const Expr *IVExpr = S.getIterationVariable();
+  const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
+  const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
+
+  if (DynamicOrOrdered) {
+    llvm::Value *UBVal = EmitScalarExpr(S.getLastIteration());
+    RT.emitForDispatchInit(*this, S.getLocStart(), ScheduleKind,
+                           IVSize, IVSigned, Ordered, UBVal, Chunk);
+  } else {
+    RT.emitForStaticInit(*this, S.getLocStart(), ScheduleKind, IVSize, IVSigned,
+                         Ordered, IL, LB, UB, ST, Chunk);
+  }
+
+  EmitOMPOuterLoop(IsMonotonic, DynamicOrOrdered, S, LoopScope, Ordered, LB, UB,
+                   ST, IL, Chunk);
+}
+
+void CodeGenFunction::EmitOMPDistributeOuterLoop(
+    OpenMPDistScheduleClauseKind ScheduleKind,
+    const OMPDistributeDirective &S, OMPPrivateScope &LoopScope,
+    Address LB, Address UB, Address ST, Address IL, llvm::Value *Chunk) {
+
+  auto &RT = CGM.getOpenMPRuntime();
+
+  // Emit outer loop.
+  // Same behavior as a OMPForOuterLoop, except that schedule cannot be
+  // dynamic
+  //
+
+  const Expr *IVExpr = S.getIterationVariable();
+  const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
+  const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
+
+  RT.emitDistributeStaticInit(*this, S.getLocStart(), ScheduleKind,
+                              IVSize, IVSigned, /* Ordered = */ false,
+                              IL, LB, UB, ST, Chunk);
+
+  EmitOMPOuterLoop(/* DynamicOrOrdered = */ false, /* IsMonotonic = */ false,
+                   S, LoopScope, /* Ordered = */ false, LB, UB, ST, IL, Chunk);
 }
 
 /// \brief Emit a helper variable and return corresponding lvalue.
@@ -2191,9 +2229,130 @@ void CodeGenFunction::EmitOMPFlushDirective(const OMPFlushDirective &S) {
   }(), S.getLocStart());
 }
 
+void CodeGenFunction::EmitOMPDistributeLoop(const OMPDistributeDirective &S) {
+  // Emit the loop iteration variable.
+  auto IVExpr = cast<DeclRefExpr>(S.getIterationVariable());
+  auto IVDecl = cast<VarDecl>(IVExpr->getDecl());
+  EmitVarDecl(*IVDecl);
+
+  // Emit the iterations count variable.
+  // If it is not a variable, Sema decided to calculate iterations count on each
+  // iteration (e.g., it is foldable into a constant).
+  if (auto LIExpr = dyn_cast<DeclRefExpr>(S.getLastIteration())) {
+    EmitVarDecl(*cast<VarDecl>(LIExpr->getDecl()));
+    // Emit calculation of the iterations count.
+    EmitIgnoredExpr(S.getCalcLastIteration());
+  }
+
+  auto &RT = CGM.getOpenMPRuntime();
+
+  // Check pre-condition.
+  {
+    // Skip the entire loop if we don't meet the precondition.
+    // If the condition constant folds and can be elided, avoid emitting the
+    // whole loop.
+    bool CondConstant;
+    llvm::BasicBlock *ContBlock = nullptr;
+    if (ConstantFoldsToSimpleInteger(S.getPreCond(), CondConstant)) {
+      if (!CondConstant)
+        return;
+    } else {
+      auto *ThenBlock = createBasicBlock("omp.precond.then");
+      ContBlock = createBasicBlock("omp.precond.end");
+      emitPreCond(*this, S, S.getPreCond(), ThenBlock, ContBlock,
+                  getProfileCount(&S));
+      EmitBlock(ThenBlock);
+      incrementProfileCounter(&S);
+    }
+
+    // Emit 'then' code.
+    {
+      // Emit helper vars inits.
+      LValue LB =
+          EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getLowerBoundVariable()));
+      LValue UB =
+          EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getUpperBoundVariable()));
+      LValue ST =
+          EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getStrideVariable()));
+      LValue IL =
+          EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getIsLastIterVariable()));
+
+      OMPPrivateScope LoopScope(*this);
+      emitPrivateLoopCounters(*this, LoopScope, S.counters(),
+                              S.private_counters());
+      (void)LoopScope.Privatize();
+
+      // Detect the distribute schedule kind and chunk.
+      llvm::Value *Chunk = nullptr;
+      OpenMPDistScheduleClauseKind ScheduleKind = OMPC_DIST_SCHEDULE_unknown;
+      if (auto *C = S.getSingleClause<OMPDistScheduleClause>()) {
+        ScheduleKind = C->getDistScheduleKind();
+        if (const auto *Ch = C->getChunkSize()) {
+          Chunk = EmitScalarExpr(Ch);
+          Chunk = EmitScalarConversion(Chunk, Ch->getType(),
+          S.getIterationVariable()->getType(),
+          S.getLocStart());
+        }
+      }
+      const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
+      const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
+
+      // OpenMP [2.10.8, distribute Construct, Description]
+      // If dist_schedule is specified, kind must be static. If specified,
+      // iterations are divided into chunks of size chunk_size, chunks are
+      // assigned to the teams of the league in a round-robin fashion in the
+      // order of the team number. When no chunk_size is specified, the
+      // iteration space is divided into chunks that are approximately equal
+      // in size, and at most one chunk is distributed to each team of the
+      // league. The size of the chunks is unspecified in this case.
+      if (RT.isStaticNonchunked(ScheduleKind,
+                                /* Chunked */ Chunk != nullptr)) {
+        RT.emitDistributeStaticInit(*this, S.getLocStart(), ScheduleKind,
+                             IVSize, IVSigned, /* Ordered = */ false,
+                             IL.getAddress(), LB.getAddress(),
+                             UB.getAddress(), ST.getAddress());
+        auto LoopExit =
+            getJumpDestInCurrentScope(createBasicBlock("omp.loop.exit"));
+        // UB = min(UB, GlobalUB);
+        EmitIgnoredExpr(S.getEnsureUpperBound());
+        // IV = LB;
+        EmitIgnoredExpr(S.getInit());
+        // while (idx <= UB) { BODY; ++idx; }
+        EmitOMPInnerLoop(S, LoopScope.requiresCleanups(), S.getCond(),
+                         S.getInc(),
+                         [&S, LoopExit](CodeGenFunction &CGF) {
+                           CGF.EmitOMPLoopBody(S, LoopExit);
+                           CGF.EmitStopPoint(&S);
+                         },
+                         [](CodeGenFunction &) {});
+        EmitBlock(LoopExit.getBlock());
+        // Tell the runtime we are done.
+        RT.emitForStaticFinish(*this, S.getLocStart());
+      } else {
+        // Emit the outer loop, which requests its work chunk [LB..UB] from
+        // runtime and runs the inner loop to process it.
+        EmitOMPDistributeOuterLoop(ScheduleKind, S, LoopScope,
+                            LB.getAddress(), UB.getAddress(), ST.getAddress(),
+                            IL.getAddress(), Chunk);
+      }
+    }
+
+    // We're now done with the loop, so jump to the continuation block.
+    if (ContBlock) {
+      EmitBranch(ContBlock);
+      EmitBlock(ContBlock, true);
+    }
+  }
+}
+
 void CodeGenFunction::EmitOMPDistributeDirective(
     const OMPDistributeDirective &S) {
-  llvm_unreachable("CodeGen for 'omp distribute' is not supported yet.");
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    CGF.EmitOMPDistributeLoop(S);
+  };
+  CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_distribute, CodeGen,
+                                              false);
 }
 
 static llvm::Function *emitOutlinedOrderedFunction(CodeGenModule &CGM,
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index 5339d51d523..a84a0f0e73c 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -2364,6 +2364,7 @@ public:
   void EmitOMPTaskLoopDirective(const OMPTaskLoopDirective &S);
   void EmitOMPTaskLoopSimdDirective(const OMPTaskLoopSimdDirective &S);
   void EmitOMPDistributeDirective(const OMPDistributeDirective &S);
+  void EmitOMPDistributeLoop(const OMPDistributeDirective &S);
 
   /// \brief Emit inner loop of the worksharing/simd construct.
   ///
@@ -2393,11 +2394,18 @@ private:
   /// \return true, if this construct has any lastprivate clause, false -
   /// otherwise.
   bool EmitOMPWorksharingLoop(const OMPLoopDirective &S);
+  void EmitOMPOuterLoop(bool IsMonotonic, bool DynamicOrOrdered,
+      const OMPLoopDirective &S, OMPPrivateScope &LoopScope, bool Ordered,
+      Address LB, Address UB, Address ST, Address IL, llvm::Value *Chunk);
   void EmitOMPForOuterLoop(OpenMPScheduleClauseKind ScheduleKind,
                            bool IsMonotonic, const OMPLoopDirective &S,
                            OMPPrivateScope &LoopScope, bool Ordered, Address LB,
                            Address UB, Address ST, Address IL,
                            llvm::Value *Chunk);
+  void EmitOMPDistributeOuterLoop(
+      OpenMPDistScheduleClauseKind ScheduleKind,
+      const OMPDistributeDirective &S, OMPPrivateScope &LoopScope,
+      Address LB, Address UB, Address ST, Address IL, llvm::Value *Chunk);
   /// \brief Emit code for sections directive.
   void EmitSections(const OMPExecutableDirective &S);
 
diff --git a/clang/lib/Serialization/ASTReaderStmt.cpp b/clang/lib/Serialization/ASTReaderStmt.cpp
index dd4997b708a..e5791d81f20 100644
--- a/clang/lib/Serialization/ASTReaderStmt.cpp
+++ b/clang/lib/Serialization/ASTReaderStmt.cpp
@@ -2307,7 +2307,8 @@ void ASTStmtReader::VisitOMPLoopDirective(OMPLoopDirective *D) {
   D->setInit(Reader.ReadSubExpr());
   D->setInc(Reader.ReadSubExpr());
   if (isOpenMPWorksharingDirective(D->getDirectiveKind()) ||
-      isOpenMPTaskLoopDirective(D->getDirectiveKind())) {
+      isOpenMPTaskLoopDirective(D->getDirectiveKind()) ||
+      isOpenMPDistributeDirective(D->getDirectiveKind())) {
     D->setIsLastIterVariable(Reader.ReadSubExpr());
     D->setLowerBoundVariable(Reader.ReadSubExpr());
     D->setUpperBoundVariable(Reader.ReadSubExpr());
diff --git a/clang/lib/Serialization/ASTWriterStmt.cpp b/clang/lib/Serialization/ASTWriterStmt.cpp
index ba2fd48563e..02d7a60eb30 100644
--- a/clang/lib/Serialization/ASTWriterStmt.cpp
+++ b/clang/lib/Serialization/ASTWriterStmt.cpp
@@ -2095,7 +2095,8 @@ void ASTStmtWriter::VisitOMPLoopDirective(OMPLoopDirective *D) {
   Writer.AddStmt(D->getInit());
   Writer.AddStmt(D->getInc());
   if (isOpenMPWorksharingDirective(D->getDirectiveKind()) ||
-      isOpenMPTaskLoopDirective(D->getDirectiveKind())) {
+      isOpenMPTaskLoopDirective(D->getDirectiveKind()) ||
+      isOpenMPDistributeDirective(D->getDirectiveKind())) {
     Writer.AddStmt(D->getIsLastIterVariable());
     Writer.AddStmt(D->getLowerBoundVariable());
     Writer.AddStmt(D->getUpperBoundVariable());