[OPENMP] Do not emit implicit barrier for single directive with 'copyprivate' clause(s).

[ABataev]

Summary

Runtime function for 'copyprivate' directive generates implicit barriers, so no need to emit it.

[rjmccall]

LGTM.

lib/CodeGen/CGStmtOpenMP.cpp

void CodeGenFunction::EmitOMPSingleDirective(const OMPSingleDirective &S) {
   // Emit code for 'single' region along with 'copyprivate' clauses
   auto &&CodeGen = [&S](CodeGenFunction &CGF) {
     CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
     CGF.EnsureInsertPoint();
   };
   CGM.getOpenMPRuntime().emitSingleRegion(*this, CodeGen, S.getLocStart(),
                                           CopyprivateVars, DestExprs, SrcExprs,
                                           AssignmentOps);
-  // Emit an implicit barrier at the end.
-  if (!S.getSingleClause(OMPC_nowait)) {
+  // Emit an implicit barrier at the end (if no 'nowait' clause and no
+  // 'copyprivate' clause).
+  if (!S.getSingleClause(OMPC_nowait) && CopyprivateVars.empty()) {
     CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), OMPD_single);
   }
 }
 
 void CodeGenFunction::EmitOMPMasterDirective(const OMPMasterDirective &S) {
   LexicalScope Scope(*this, S.getSourceRange());
   auto &&CodeGen = [&S](CodeGenFunction &CGF) {
     CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());

test/OpenMP/single_codegen.cpp

 // CHECK:       [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
 // CHECK-NEXT:  [[IS_SINGLE:%.+]] = icmp ne i32 [[RES]], 0
 // CHECK-NEXT:  br i1 [[IS_SINGLE]], label {{%?}}[[THEN:.+]], label {{%?}}[[EXIT:.+]]
 // CHECK:       [[THEN]]
 // CHECK-NEXT:  store i8 2, i8* [[A_ADDR]]
 // CHECK-NEXT:  call void @__kmpc_end_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
 // CHECK-NEXT:  br label {{%?}}[[EXIT]]
 // CHECK:       [[EXIT]]
-// CHECK-NOT:   __kmpc_cancel_barrier
+// CHECK-NOT:   call {{.+}} @__kmpc_cancel_barrier
 #pragma omp single nowait
   a = 2;
 // CHECK:       [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
 // CHECK-NEXT:  [[IS_SINGLE:%.+]] = icmp ne i32 [[RES]], 0
 // CHECK-NEXT:  br i1 [[IS_SINGLE]], label {{%?}}[[THEN:.+]], label {{%?}}[[EXIT:.+]]
 // CHECK:       [[THEN]]
+// CHECK-NEXT:  store i8 2, i8* [[A_ADDR]]
+// CHECK-NEXT:  call void @__kmpc_end_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
+// CHECK-NEXT:  br label {{%?}}[[EXIT]]
+// CHECK:       [[EXIT]]
+// CHECK:       call{{.*}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_SINGLE_LOC]], i32 [[GTID]])
+#pragma omp single
+  a = 2;
+// CHECK:       [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
+// CHECK-NEXT:  [[IS_SINGLE:%.+]] = icmp ne i32 [[RES]], 0
+// CHECK-NEXT:  br i1 [[IS_SINGLE]], label {{%?}}[[THEN:.+]], label {{%?}}[[EXIT:.+]]
+// CHECK:       [[THEN]]
 // CHECK-NEXT:  invoke void [[FOO]]()
 // CHECK:       to label {{%?}}[[CONT:.+]] unwind
 // CHECK:       [[CONT]]
 // CHECK:       store i32 1, i32* [[DID_IT]]
 // CHECK:       call void @__kmpc_end_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
 // CHECK-NEXT:  br label {{%?}}[[EXIT]]
 // CHECK:       [[EXIT]]
 // CHECK:       [[A_PTR_REF:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[COPY_LIST]], i{{[0-9]+}} 0, i{{[0-9]+}} 0

 // CHECK:       [[TC2_PTR_REF:%.+]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[COPY_LIST]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
 // CHECK:       [[TC2_THREADPRIVATE_ADDR_VOID_PTR:%.+]] = call{{.*}} i8* @__kmpc_threadprivate_cached
 // CHECK:       [[TC2_THREADPRIVATE_ADDR:%.+]] = bitcast i8* [[TC2_THREADPRIVATE_ADDR_VOID_PTR]] to [2 x [[TEST_CLASS_TY]]]*
 // CHECK:       [[TC2_PTR_REF_VOID_PTR:%.+]] = bitcast [2 x [[TEST_CLASS_TY]]]* [[TC2_THREADPRIVATE_ADDR]] to i8*
 // CHECK:       store i8* [[TC2_PTR_REF_VOID_PTR]], i8** [[TC2_PTR_REF]],
 // CHECK:       [[COPY_LIST_VOID_PTR:%.+]] = bitcast [5 x i8*]* [[COPY_LIST]] to i8*
 // CHECK:       [[DID_IT_VAL:%.+]] = load i32, i32* [[DID_IT]],
 // CHECK:       call void @__kmpc_copyprivate([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32 40, i8* [[COPY_LIST_VOID_PTR]], void (i8*, i8*)* [[COPY_FUNC:@.+]], i32 [[DID_IT_VAL]])
-// CHECK:       call{{.*}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_SINGLE_LOC]], i32 [[GTID]])
+// CHECK-NOT:   call {{.+}} @__kmpc_cancel_barrier
 #pragma omp single copyprivate(a, c, tc, a2, tc2)
   foo();
 // CHECK-NOT:   call i32 @__kmpc_single
 // CHECK-NOT:   call void @__kmpc_end_single
   return a;
 }
 
 // CHECK: void [[COPY_FUNC]](i8*, i8*)