[OPNEMP]Allow num_tasks clause in combined task-based directives.

The expression of the num_tasks clause must be captured in the combined
task-based directives, like 'parallel master taskloop' directive.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@374819 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 49 insertions, 22 deletions

diff --git a/include/clang/AST/OpenMPClause.h b/include/clang/AST/OpenMPClause.h
index db780f7ed3..911c1cfb77 100644
--- a/include/clang/AST/OpenMPClause.h
+++ b/include/clang/AST/OpenMPClause.h
@@ -5375,7 +5375,7 @@ public:
 /// \endcode
 /// In this example directive '#pragma omp taskloop' has clause 'num_tasks'
 /// with single expression '4'.
-class OMPNumTasksClause : public OMPClause {
+class OMPNumTasksClause : public OMPClause, public OMPClauseWithPreInit {
   friend class OMPClauseReader;
 
   /// Location of '('.
@@ -5391,16 +5391,23 @@ public:
   /// Build 'num_tasks' clause.
   ///
   /// \param Size Expression associated with this clause.
+  /// \param HelperSize Helper grainsize for the construct.
+  /// \param CaptureRegion Innermost OpenMP region where expressions in this
+  /// clause must be captured.
   /// \param StartLoc Starting location of the clause.
   /// \param EndLoc Ending location of the clause.
-  OMPNumTasksClause(Expr *Size, SourceLocation StartLoc,
+  OMPNumTasksClause(Expr *Size, Stmt *HelperSize,
+                    OpenMPDirectiveKind CaptureRegion, SourceLocation StartLoc,
                     SourceLocation LParenLoc, SourceLocation EndLoc)
-      : OMPClause(OMPC_num_tasks, StartLoc, EndLoc), LParenLoc(LParenLoc),
-        NumTasks(Size) {}
+      : OMPClause(OMPC_num_tasks, StartLoc, EndLoc), OMPClauseWithPreInit(this),
+        LParenLoc(LParenLoc), NumTasks(Size) {
+    setPreInitStmt(HelperSize, CaptureRegion);
+  }
 
   /// Build an empty clause.
   explicit OMPNumTasksClause()
-      : OMPClause(OMPC_num_tasks, SourceLocation(), SourceLocation()) {}
+      : OMPClause(OMPC_num_tasks, SourceLocation(), SourceLocation()),
+        OMPClauseWithPreInit(this) {}
 
   /// Sets the location of '('.
   void setLParenLoc(SourceLocation Loc) { LParenLoc = Loc; }
@@ -5417,11 +5424,10 @@ public:
     return const_child_range(&NumTasks, &NumTasks + 1);
   }
 
-  child_range used_children() {
-    return child_range(child_iterator(), child_iterator());
-  }
+  child_range used_children();
   const_child_range used_children() const {
-    return const_child_range(const_child_iterator(), const_child_iterator());
+    auto Children = const_cast<OMPNumTasksClause *>(this)->used_children();
+    return const_child_range(Children.begin(), Children.end());
   }
 
   static bool classof(const OMPClause *T) {
diff --git a/include/clang/AST/RecursiveASTVisitor.h b/include/clang/AST/RecursiveASTVisitor.h
index ba5232a2f4..09a6f7840d 100644
--- a/include/clang/AST/RecursiveASTVisitor.h
+++ b/include/clang/AST/RecursiveASTVisitor.h
@@ -3283,6 +3283,7 @@ bool RecursiveASTVisitor<Derived>::VisitOMPGrainsizeClause(
 template <typename Derived>
 bool RecursiveASTVisitor<Derived>::VisitOMPNumTasksClause(
     OMPNumTasksClause *C) {
+  TRY_TO(VisitOMPClauseWithPreInit(C));
   TRY_TO(TraverseStmt(C->getNumTasks()));
   return true;
 }
diff --git a/lib/AST/OpenMPClause.cpp b/lib/AST/OpenMPClause.cpp
index b97607f8c6..c1aeaf6a6b 100644
--- a/lib/AST/OpenMPClause.cpp
+++ b/lib/AST/OpenMPClause.cpp
@@ -86,6 +86,8 @@ const OMPClauseWithPreInit *OMPClauseWithPreInit::get(const OMPClause *C) {
     return static_cast<const OMPDeviceClause *>(C);
   case OMPC_grainsize:
     return static_cast<const OMPGrainsizeClause *>(C);
+  case OMPC_num_tasks:
+    return static_cast<const OMPNumTasksClause *>(C);
   case OMPC_default:
   case OMPC_proc_bind:
   case OMPC_final:
@@ -116,7 +118,6 @@ const OMPClauseWithPreInit *OMPClauseWithPreInit::get(const OMPClause *C) {
   case OMPC_map:
   case OMPC_priority:
   case OMPC_nogroup:
-  case OMPC_num_tasks:
   case OMPC_hint:
   case OMPC_defaultmap:
   case OMPC_unknown:
@@ -241,6 +242,12 @@ OMPClause::child_range OMPGrainsizeClause::used_children() {
   return child_range(&Grainsize, &Grainsize + 1);
 }
 
+OMPClause::child_range OMPNumTasksClause::used_children() {
+  if (Stmt **C = getAddrOfExprAsWritten(getPreInitStmt()))
+    return child_range(C, C + 1);
+  return child_range(&NumTasks, &NumTasks + 1);
+}
+
 OMPOrderedClause *OMPOrderedClause::Create(const ASTContext &C, Expr *Num,
                                            unsigned NumLoops,
                                            SourceLocation StartLoc,
diff --git a/lib/AST/StmtProfile.cpp b/lib/AST/StmtProfile.cpp
index 6c65f8a1d0..9c19305a7d 100644
--- a/lib/AST/StmtProfile.cpp
+++ b/lib/AST/StmtProfile.cpp
@@ -745,6 +745,7 @@ void OMPClauseProfiler::VisitOMPGrainsizeClause(const OMPGrainsizeClause *C) {
     Profiler->VisitStmt(C->getGrainsize());
 }
 void OMPClauseProfiler::VisitOMPNumTasksClause(const OMPNumTasksClause *C) {
+  VistOMPClauseWithPreInit(C);
   if (C->getNumTasks())
     Profiler->VisitStmt(C->getNumTasks());
 }
diff --git a/lib/Sema/SemaOpenMP.cpp b/lib/Sema/SemaOpenMP.cpp
index 94cd80a4b0..4d6ff009c5 100644
--- a/lib/Sema/SemaOpenMP.cpp
+++ b/lib/Sema/SemaOpenMP.cpp
@@ -4595,12 +4595,16 @@ StmtResult Sema::ActOnOpenMPExecutableDirective(
         if (isOpenMPParallelDirective(DSAStack->getCurrentDirective()))
           break;
         continue;
+      case OMPC_num_tasks:
+        // Do not analyze if no parent parallel directive.
+        if (isOpenMPParallelDirective(DSAStack->getCurrentDirective()))
+          break;
+        continue;
       case OMPC_ordered:
       case OMPC_device:
       case OMPC_num_teams:
       case OMPC_thread_limit:
       case OMPC_priority:
-      case OMPC_num_tasks:
       case OMPC_hint:
       case OMPC_collapse:
       case OMPC_safelen:
@@ -10778,6 +10782,7 @@ static OpenMPDirectiveKind getOpenMPCaptureRegionForClause(
     }
     break;
   case OMPC_grainsize:
+  case OMPC_num_tasks:
     switch (DKind) {
     case OMPD_task:
     case OMPD_taskloop:
@@ -10881,7 +10886,6 @@ static OpenMPDirectiveKind getOpenMPCaptureRegionForClause(
   case OMPC_map:
   case OMPC_priority:
   case OMPC_nogroup:
-  case OMPC_num_tasks:
   case OMPC_hint:
   case OMPC_defaultmap:
   case OMPC_unknown:
@@ -15952,15 +15956,20 @@ OMPClause *Sema::ActOnOpenMPNumTasksClause(Expr *NumTasks,
                                            SourceLocation LParenLoc,
                                            SourceLocation EndLoc) {
   Expr *ValExpr = NumTasks;
+  Stmt *HelperValStmt = nullptr;
+  OpenMPDirectiveKind CaptureRegion = OMPD_unknown;
 
   // OpenMP [2.9.2, taskloop Constrcut]
   // The parameter of the num_tasks clause must be a positive integer
   // expression.
-  if (!isNonNegativeIntegerValue(ValExpr, *this, OMPC_num_tasks,
-                                 /*StrictlyPositive=*/true))
+  if (!isNonNegativeIntegerValue(
+          ValExpr, *this, OMPC_num_tasks,
+          /*StrictlyPositive=*/true, /*BuildCapture=*/true,
+          DSAStack->getCurrentDirective(), &CaptureRegion, &HelperValStmt))
     return nullptr;
 
-  return new (Context) OMPNumTasksClause(ValExpr, StartLoc, LParenLoc, EndLoc);
+  return new (Context) OMPNumTasksClause(ValExpr, HelperValStmt, CaptureRegion,
+                                         StartLoc, LParenLoc, EndLoc);
 }
 
 OMPClause *Sema::ActOnOpenMPHintClause(Expr *Hint, SourceLocation StartLoc,
diff --git a/lib/Serialization/ASTReader.cpp b/lib/Serialization/ASTReader.cpp
index 0a7958f2c2..60be1570bd 100644
--- a/lib/Serialization/ASTReader.cpp
+++ b/lib/Serialization/ASTReader.cpp
@@ -12940,6 +12940,7 @@ void OMPClauseReader::VisitOMPGrainsizeClause(OMPGrainsizeClause *C) {
 }
 
 void OMPClauseReader::VisitOMPNumTasksClause(OMPNumTasksClause *C) {
+  VisitOMPClauseWithPreInit(C);
   C->setNumTasks(Record.readSubExpr());
   C->setLParenLoc(Record.readSourceLocation());
 }
diff --git a/lib/Serialization/ASTWriter.cpp b/lib/Serialization/ASTWriter.cpp
index 57c9242504..3badff45fb 100644
--- a/lib/Serialization/ASTWriter.cpp
+++ b/lib/Serialization/ASTWriter.cpp
@@ -6944,6 +6944,7 @@ void OMPClauseWriter::VisitOMPGrainsizeClause(OMPGrainsizeClause *C) {
 }
 
 void OMPClauseWriter::VisitOMPNumTasksClause(OMPNumTasksClause *C) {
+  VisitOMPClauseWithPreInit(C);
   Record.AddStmt(C->getNumTasks());
   Record.AddSourceLocation(C->getLParenLoc());
 }
diff --git a/test/OpenMP/parallel_master_taskloop_ast_print.cpp b/test/OpenMP/parallel_master_taskloop_ast_print.cpp
index 8095652de4..23cf67c0a5 100644
--- a/test/OpenMP/parallel_master_taskloop_ast_print.cpp
+++ b/test/OpenMP/parallel_master_taskloop_ast_print.cpp
@@ -60,9 +60,9 @@ int main(int argc, char **argv) {
   static int a;
 // CHECK: static int a;
 #pragma omp taskgroup task_reduction(+: d)
-#pragma omp parallel master taskloop if(parallel: a) default(none) shared(a) final(b) priority(5) num_tasks(argc) reduction(*: g)
+#pragma omp parallel master taskloop if(parallel: a) default(none) shared(a, argc) final(b) priority(5) num_tasks(argc) reduction(*: g)
   // CHECK-NEXT: #pragma omp taskgroup task_reduction(+: d)
-  // CHECK-NEXT: #pragma omp parallel master taskloop if(parallel: a) default(none) shared(a) final(b) priority(5) num_tasks(argc) reduction(*: g)
+  // CHECK-NEXT: #pragma omp parallel master taskloop if(parallel: a) default(none) shared(a,argc) final(b) priority(5) num_tasks(argc) reduction(*: g)
   for (int i = 0; i < 2; ++i)
     a = 2;
 // CHECK-NEXT: for (int i = 0; i < 2; ++i)
diff --git a/test/OpenMP/parallel_master_taskloop_codegen.cpp b/test/OpenMP/parallel_master_taskloop_codegen.cpp
index 2a2f4eb598..289687cff2 100644
--- a/test/OpenMP/parallel_master_taskloop_codegen.cpp
+++ b/test/OpenMP/parallel_master_taskloop_codegen.cpp
@@ -15,9 +15,9 @@ int main(int argc, char **argv) {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* [[DEFLOC:@.+]])
 // CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEFLOC]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* [[OMP_OUTLINED1:@.+]] to void (i32*, i32*, ...)*))
 // CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEFLOC]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i64)* [[OMP_OUTLINED2:@.+]] to void (i32*, i32*, ...)*), i64 [[GRAINSIZE:%.+]])
-// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEFLOC]], i32 3, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***, i64)* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i32* [[ARGC:%.+]], i8*** [[ARGV:%.+]], i64 [[COND:%.+]])
+// CHECK: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* [[DEFLOC]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, i8***, i64, i64)* [[OMP_OUTLINED3:@.+]] to void (i32*, i32*, ...)*), i32* [[ARGC:%.+]], i8*** [[ARGV:%.+]], i64 [[COND:%.+]], i64 [[NUM_TASKS:%.+]])
 // CHECK: call void @__kmpc_serialized_parallel(%struct.ident_t* [[DEFLOC]], i32 [[GTID]])
-// CHECK: call void [[OMP_OUTLINED3]](i32* %{{.+}}, i32* %{{.+}}, i32* [[ARGC]], i8*** [[ARGV]], i64 [[COND]])
+// CHECK: call void [[OMP_OUTLINED3]](i32* %{{.+}}, i32* %{{.+}}, i32* [[ARGC]], i8*** [[ARGV]], i64 [[COND]], i64 [[NUM_TASKS]])
 // CHECK: call void @__kmpc_end_serialized_parallel(%struct.ident_t* [[DEFLOC]], i32 [[GTID]])
 
 
@@ -132,7 +132,7 @@ int main(int argc, char **argv) {
 #pragma omp parallel master taskloop nogroup grainsize(argc)
   for (int i = 0; i < 10; ++i)
     ;
-// CHECK: define internal void [[OMP_OUTLINED3]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, i8*** dereferenceable(8) %{{.+}}, i64 %{{.+}})
+// CHECK: define internal void [[OMP_OUTLINED3]](i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable(4) %{{.+}}, i8*** dereferenceable(8) %{{.+}}, i64 %{{.+}}, i64 %{{.+}})
 // CHECK:       [[RES:%.+]] = call {{.*}}i32 @__kmpc_master(%struct.ident_t* [[DEFLOC]], i32 [[GTID:%.+]])
 // CHECK-NEXT:  [[IS_MASTER:%.+]] = icmp ne i32 [[RES]], 0
 // CHECK-NEXT:  br i1 [[IS_MASTER]], label {{%?}}[[THEN:.+]], label {{%?}}[[EXIT:.+]]
@@ -151,7 +151,8 @@ int main(int argc, char **argv) {
 // CHECK: [[ST:%.+]] = getelementptr inbounds [[TD_TY]], [[TD_TY]]* [[TASK_DATA]], i32 0, i32 7
 // CHECK: store i64 1, i64* [[ST]],
 // CHECK: [[ST_VAL:%.+]] = load i64, i64* [[ST]],
-// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 1, i32 2, i64 4, i8* null)
+// CHECK: [[NUM_TASKS:%.+]] = zext i32 %{{.+}} to i64
+// CHECK: call void @__kmpc_taskloop(%struct.ident_t* [[DEFLOC]], i32 [[GTID]], i8* [[TASKV]], i32 [[IF_INT]], i64* [[DOWN]], i64* [[UP]], i64 [[ST_VAL]], i32 1, i32 2, i64 [[NUM_TASKS]], i8* null)
 // CHECK: call void @__kmpc_end_taskgroup(%struct.ident_t* [[DEFLOC]], i32 [[GTID]])
 // CHECK-NEXT:  call {{.*}}void @__kmpc_end_master(%struct.ident_t* [[DEFLOC]], i32 [[GTID]])
 // CHECK-NEXT:  br label {{%?}}[[EXIT]]
@@ -176,7 +177,7 @@ int main(int argc, char **argv) {
 // CHECK: ret i32 0
 
   int i;
-#pragma omp parallel master taskloop if(argc) shared(argc, argv) collapse(2) num_tasks(4)
+#pragma omp parallel master taskloop if(argc) shared(argc, argv) collapse(2) num_tasks(argc)
   for (i = 0; i < argc; ++i)
   for (int j = argc; j < argv[argc][argc]; ++j)
     ;