[OPENMP] Initial codegen for 'parallel sections' directive.

[ABataev]

Summary

Emits code for outlined 'parallel' directive with the implicitly inlined 'sections' directive:

...
call __kmpc_fork_call(..., outlined_function, ...);
...

define internal void outlined_function(...) {
  <code for implicit sections directive>;
}

[rjmccall]

Looks good.

lib/CodeGen/CGStmtOpenMP.cpp

888 ↗

(On Diff #23666)

You use this twice; it doesn't seem worthwhile.

[ABataev]

John, thanks for the review.

lib/CodeGen/CGStmtOpenMP.cpp

888 ↗

(On Diff #23666)

Removed, thanks!

cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp

static LValue createSectionLVal(CodeGenFunction &CGF, QualType Ty,
                                 const Twine &Name,
                                 llvm::Value *Init = nullptr) {
   auto LVal = CGF.MakeNaturalAlignAddrLValue(CGF.CreateMemTemp(Ty, Name), Ty);
   if (Init)
     CGF.EmitScalarInit(Init, LVal);
   return LVal;
 }
 
-void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
-  LexicalScope Scope(*this, S.getSourceRange());
+static OpenMPDirectiveKind emitSections(CodeGenFunction &CGF,
+                                        const OMPExecutableDirective &S) {
   auto *Stmt = cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt();
   auto *CS = dyn_cast<CompoundStmt>(Stmt);
   if (CS && CS->size() > 1) {
     auto &&CodeGen = [&S, CS](CodeGenFunction &CGF) {
       auto &C = CGF.CGM.getContext();
       auto KmpInt32Ty = C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
       // Emit helper vars inits.
       LValue LB = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.lb.",
                                     CGF.Builder.getInt32(0));
       auto *GlobalUBVal = CGF.Builder.getInt32(CS->size() - 1);
       LValue UB =
           createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.ub.", GlobalUBVal);
       LValue ST = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.st.",
                                     CGF.Builder.getInt32(1));
       LValue IL = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.il.",
                                     CGF.Builder.getInt32(0));
       // Loop counter.
       LValue IV = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.iv.");
       OpaqueValueExpr IVRefExpr(S.getLocStart(), KmpInt32Ty, VK_LValue);
-      OpaqueValueMapping OpaqueIV(CGF, &IVRefExpr, IV);
+      CodeGenFunction::OpaqueValueMapping OpaqueIV(CGF, &IVRefExpr, IV);
       OpaqueValueExpr UBRefExpr(S.getLocStart(), KmpInt32Ty, VK_LValue);
-      OpaqueValueMapping OpaqueUB(CGF, &UBRefExpr, UB);
+      CodeGenFunction::OpaqueValueMapping OpaqueUB(CGF, &UBRefExpr, UB);
       // Generate condition for loop.
       BinaryOperator Cond(&IVRefExpr, &UBRefExpr, BO_LE, C.BoolTy, VK_RValue,
                           OK_Ordinary, S.getLocStart(),
                           /*fpContractable=*/false);
       // Increment for loop counter.
       UnaryOperator Inc(&IVRefExpr, UO_PreInc, KmpInt32Ty, VK_RValue,
                         OK_Ordinary, S.getLocStart());
       auto BodyGen = [CS, &S, &IV](CodeGenFunction &CGF) {
auto &&CodeGen = [&S, CS](CodeGenFunction &CGF) {
       CGF.EmitStoreOfScalar(CGF.EmitLoadOfScalar(LB, S.getLocStart()), IV);
       // while (idx <= UB) { BODY; ++idx; }
       CGF.EmitOMPInnerLoop(S, /*RequiresCleanup=*/false, &Cond, &Inc, BodyGen);
       // Tell the runtime we are done.
       CGF.CGM.getOpenMPRuntime().emitForFinish(CGF, S.getLocStart(),
                                                OMPC_SCHEDULE_static);
     };
 
-    CGM.getOpenMPRuntime().emitInlinedDirective(*this, CodeGen);
-  } else {
-    // If only one section is found - no need to generate loop, emit as a
-    // single
-    // region.
-    auto &&CodeGen = [&S](CodeGenFunction &CGF) {
-      CGF.EmitStmt(
-          cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
-      CGF.EnsureInsertPoint();
-    };
-    CGM.getOpenMPRuntime().emitSingleRegion(*this, CodeGen, S.getLocStart(),
-                                            llvm::None, llvm::None, llvm::None,
-                                            llvm::None);
-  }
+    CGF.CGM.getOpenMPRuntime().emitInlinedDirective(CGF, CodeGen);
+    return OMPD_sections;
+  }
+  // If only one section is found - no need to generate loop, emit as a single
+  auto &&CodeGen = [Stmt](CodeGenFunction &CGF) {
+    CGF.EmitStmt(Stmt);
+  CGF.CGM.getOpenMPRuntime().emitSingleRegion(CGF, CodeGen, S.getLocStart(),
+                                              llvm::None, llvm::None,
+                                              llvm::None, llvm::None);
+  return OMPD_single;
 
+void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
+  LexicalScope Scope(*this, S.getSourceRange());
+  OpenMPDirectiveKind EmittedAs = emitSections(*this, S);
   // Emit an implicit barrier at the end.
   if (!S.getSingleClause(OMPC_nowait)) {
-    CGM.getOpenMPRuntime().emitBarrierCall(
-        *this, S.getLocStart(),
-        (CS && CS->size() > 1) ? OMPD_sections : OMPD_single);
+    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), EmittedAs);
   }
 }
 
 void CodeGenFunction::EmitOMPSectionDirective(const OMPSectionDirective &S) {
   LexicalScope Scope(*this, S.getSourceRange());
   auto &&CodeGen = [&S](CodeGenFunction &CGF) {
     CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
     CGF.EnsureInsertPoint();

 }
 
 void CodeGenFunction::EmitOMPParallelForSimdDirective(
     const OMPParallelForSimdDirective &) {
   llvm_unreachable("CodeGen for 'omp parallel for simd' is not supported yet.");
 }
 
 void CodeGenFunction::EmitOMPParallelSectionsDirective(
-    const OMPParallelSectionsDirective &) {
-  llvm_unreachable("CodeGen for 'omp parallel sections' is not supported yet.");
+    const OMPParallelSectionsDirective &S) {
+  // Emit directive as a combined directive that consists of two implicit
+  // directives: 'parallel' with 'sections' directive.
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    (void)emitSections(CGF, S);
+    // Emit implicit barrier at the end of parallel region.
+    CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                               OMPD_parallel);
+  };
+  emitCommonOMPParallelDirective(*this, S, CodeGen);
 }
 
 void CodeGenFunction::EmitOMPTaskDirective(const OMPTaskDirective &S) {
   // Emit outlined function for task construct.
   LexicalScope Scope(*this, S.getSourceRange());
   auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
   auto CapturedStruct = GenerateCapturedStmtArgument(*CS);
   auto *I = CS->getCapturedDecl()->param_begin();

cfe/trunk/lib/CodeGen/CodeGenFunction.h

#endif
   void EmitOMPBarrierDirective(const OMPBarrierDirective &S);
   void EmitOMPTaskwaitDirective(const OMPTaskwaitDirective &S);
   void EmitOMPFlushDirective(const OMPFlushDirective &S);
   void EmitOMPOrderedDirective(const OMPOrderedDirective &S);
   void EmitOMPAtomicDirective(const OMPAtomicDirective &S);
   void EmitOMPTargetDirective(const OMPTargetDirective &S);
   void EmitOMPTeamsDirective(const OMPTeamsDirective &S);
 
+  void
+  EmitOMPInnerLoop(const Stmt &S, bool RequiresCleanup, const Expr *LoopCond,
+                   const Expr *IncExpr,
+                   const llvm::function_ref<void(CodeGenFunction &)> &BodyGen);
+
 private:
 
   /// Helpers for the OpenMP loop directives.
   void EmitOMPLoopBody(const OMPLoopDirective &Directive,
                        bool SeparateIter = false);
-  void
-  EmitOMPInnerLoop(const Stmt &S, bool RequiresCleanup, const Expr *LoopCond,
-                   const Expr *IncExpr,
-                   const llvm::function_ref<void(CodeGenFunction &)> &BodyGen);
   void EmitOMPSimdFinal(const OMPLoopDirective &S);
   void EmitOMPWorksharingLoop(const OMPLoopDirective &S);
   void EmitOMPForOuterLoop(OpenMPScheduleClauseKind ScheduleKind,
                            const OMPLoopDirective &S,
                            OMPPrivateScope &LoopScope, llvm::Value *LB,
                            llvm::Value *UB, llvm::Value *ST, llvm::Value *IL,
                            llvm::Value *Chunk);
 

cfe/trunk/lib/Sema/SemaOpenMP.cpp

Sema::CapturedParamNameType Params[] = {
         std::make_pair(".bound_tid.", KmpInt32PtrTy),
         std::make_pair(StringRef(), QualType()) // __context with shared vars
     };
     ActOnCapturedRegionStart(DSAStack->getConstructLoc(), CurScope, CR_OpenMP,
                              Params);
     break;
   }
   case OMPD_parallel_sections: {
+    QualType KmpInt32Ty = Context.getIntTypeForBitwidth(32, 1);
+    QualType KmpInt32PtrTy = Context.getPointerType(KmpInt32Ty);
     Sema::CapturedParamNameType Params[] = {
+        std::make_pair(".global_tid.", KmpInt32PtrTy),
+        std::make_pair(".bound_tid.", KmpInt32PtrTy),
         std::make_pair(StringRef(), QualType()) // __context with shared vars
     };
     ActOnCapturedRegionStart(DSAStack->getConstructLoc(), CurScope, CR_OpenMP,
                              Params);
     break;
   }
   case OMPD_task: {
     QualType KmpInt32Ty = Context.getIntTypeForBitwidth(32, 1);

cfe/trunk/test/OpenMP/parallel_sections_codegen.cpp

+// RUN: %clang_cc1 -verify -fopenmp=libiomp5 -x c++ -emit-llvm -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -o - %s | FileCheck %s
+// RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -fexceptions -fcxx-exceptions -triple x86_64-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp=libiomp5 -x c++ -std=c++11 -include-pch %t -fsyntax-only -verify %s -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-llvm -o - | FileCheck %s
+// expected-no-diagnostics
+
+#ifndef HEADER
+#define HEADER
+// CHECK: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8*
+// CHECK-LABEL: foo
+void foo() {};
+// CHECK-LABEL: bar
+void bar() {};
+
+template <class T>
+T tmain() {
+#pragma omp parallel sections
+  {
+    foo();
+  }
+  return T();
+}
+
+// CHECK-LABEL: @main
+int main() {
+// CHECK: call void (%{{.+}}*, i32, void (i32*, i32*, ...)*, ...)* @__kmpc_fork_call(%{{.+}}* @{{.+}}, i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %{{.+}}*)* [[OMP_PARALLEL_FUNC:@.+]] to void (i32*, i32*, ...)*), i8* %{{.+}})
+// CHECK-LABEL: }
+// CHECK: define internal void [[OMP_PARALLEL_FUNC]](i32* [[GTID_PARAM_ADDR:%.+]], i32* %{{.+}}, %{{.+}}* %{{.+}})
+// CHECK: store i32* [[GTID_PARAM_ADDR]], i32** [[GTID_REF_ADDR:%.+]],
+#pragma omp parallel sections
+  {
+// CHECK:      store i32 0, i32* [[LB_PTR:%.+]],
+// CHECK:      store i32 1, i32* [[UB_PTR:%.+]],
+// CHECK:      [[GTID_REF:%.+]] = load i32*, i32** [[GTID_REF_ADDR]],
+// CHECK:      [[GTID:%.+]] = load i32, i32* [[GTID_REF]],
+// CHECK:      call void @__kmpc_for_static_init_4(%{{.+}}* @{{.+}}, i32 [[GTID]], i32 34, i32* [[IS_LAST_PTR:%.+]], i32* [[LB_PTR]], i32* [[UB_PTR]], i32* [[STRIDE_PTR:%.+]], i32 1, i32 1)
+// <<UB = min(UB, GlobalUB);>>
+// CHECK:      [[UB:%.+]] = load i32, i32* [[UB_PTR]]
+// CHECK:      [[CMP:%.+]] = icmp slt i32 [[UB]], 1
+// CHECK:      [[MIN_UB_GLOBALUB:%.+]] = select i1 [[CMP]], i32 [[UB]], i32 1
+// CHECK:      store i32 [[MIN_UB_GLOBALUB]], i32* [[UB_PTR]]
+// <<IV = LB;>>
+// CHECK:      [[LB:%.+]] = load i32, i32* [[LB_PTR]]
+// CHECK:      store i32 [[LB]], i32* [[IV_PTR:%.+]]
+// CHECK:      br label %[[INNER_FOR_COND:.+]]
+// CHECK:      [[INNER_FOR_COND]]
+// <<IV <= UB?>>
+// CHECK:      [[IV:%.+]] = load i32, i32* [[IV_PTR]]
+// CHECK:      [[UB:%.+]] = load i32, i32* [[UB_PTR]]
+// CHECK:      [[CMP:%.+]] = icmp sle i32 [[IV]], [[UB]]
+// CHECK:      br i1 [[CMP]], label %[[INNER_LOOP_BODY:.+]], label %[[INNER_LOOP_END:.+]]
+// CHECK:      [[INNER_LOOP_BODY]]
+// <<TRUE>> - > <BODY>
+// CHECK:      [[IV:%.+]] = load i32, i32* [[IV_PTR]]
+// CHECK:      switch i32 [[IV]], label %[[SECTIONS_EXIT:.+]] [
+// CHECK-NEXT: i32 0, label %[[SECTIONS_CASE0:.+]]
+// CHECK-NEXT: i32 1, label %[[SECTIONS_CASE1:.+]]
+#pragma omp section
+// CHECK:      [[SECTIONS_CASE0]]
+// CHECK-NEXT: invoke void @{{.*}}foo{{.*}}()
+// CHECK:      br label %[[SECTIONS_EXIT]]
+    foo();
+#pragma omp section
+// CHECK:      [[SECTIONS_CASE1]]
+// CHECK-NEXT: invoke void @{{.*}}bar{{.*}}()
+// CHECK:      br label %[[SECTIONS_EXIT]]
+    bar();
+// CHECK:      [[SECTIONS_EXIT]]
+// <<++IV;>>
+// CHECK:      [[IV:%.+]] = load i32, i32* [[IV_PTR]]
+// CHECK-NEXT: [[INC:%.+]] = add nsw i32 [[IV]], 1
+// CHECK-NEXT: store i32 [[INC]], i32* [[IV_PTR]]
+// CHECK-NEXT: br label %[[INNER_FOR_COND]]
+// CHECK:      [[INNER_LOOP_END]]
+  }
+// CHECK:      call void @__kmpc_for_static_fini(%{{.+}}* @{{.+}}, i32 [[GTID]])
+// CHECK:      call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]],
+  return tmain<int>();
+}
+
+// CHECK-LABEL: tmain
+// CHECK:       call void {{.*}} @__kmpc_fork_call(
+// CHECK-NOT:   __kmpc_global_thread_num
+// CHECK:       [[RES:%.+]] = call i32 @__kmpc_single(
+// CHECK-NEXT:  [[BOOLRES:%.+]] = icmp ne i32 [[RES]], 0
+// CHECK-NEXT:  br i1 [[BOOLRES]], label %[[THEN:.+]], label %[[END:.+]]
+// CHECK:       [[THEN]]
+// CHECK-NEXT:  invoke void @{{.*}}foo{{.*}}()
+// CHECK-NEXT:  unwind label %[[TERM_LPAD:.+]]
+// CHECK:       call void @__kmpc_end_single(
+// CHECK-NEXT:  br label %[[END]]
+// CHECK:       [[END]]
+// CHECK-NEXT:  call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_LOC]],
+// CHECK-NEXT:  ret
+// CHECK:       [[TERM_LPAD]]
+// CHECK:       call void @__clang_call_terminate(i8*
+// CHECK-NEXT:  unreachable
+
+#endif