[OPENMP] Improved codegen for implicit/explicit 'barrier' constructs.

Replace boolean IsExplicit parameter of OpenMPRuntime::emitBarrierCall() method by OpenMPDirectiveKind Kind for better compatibility with the runtime library. Also add processing of 'nowait' clause on worksharing directives.
Differential Revision: http://reviews.llvm.org/D8659

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@233511 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGOpenMPRuntime.cpp b/lib/CodeGen/CGOpenMPRuntime.cpp
index ef2d2147aca50a913f38a8eefc1c6784e206a758..2dbaae5624c771b2f48ad80e933fe13d896c6064 100644 (file)
--- a/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -179,7 +179,7 @@ void CGOpenMPRegionInfo::EmitBody(CodeGenFunction &CGF, const Stmt *S) {
   if (PrivateScope.Privatize())
     // Emit implicit barrier to synchronize threads and avoid data races.
     CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, Directive.getLocStart(),
-                                               /*IsExplicit=*/false);
+                                               OMPD_unknown);
   CGCapturedStmtInfo::EmitBody(CGF, S);
 }
 
@@ -1138,11 +1138,23 @@ void CGOpenMPRuntime::emitSingleRegion(CodeGenFunction &CGF,
 }
 
 void CGOpenMPRuntime::emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
-                                      bool IsExplicit) {
+                                      OpenMPDirectiveKind Kind) {
   // Build call __kmpc_cancel_barrier(loc, thread_id);
-  auto Flags = static_cast<OpenMPLocationFlags>(
-      OMP_IDENT_KMPC |
-      (IsExplicit ? OMP_IDENT_BARRIER_EXPL : OMP_IDENT_BARRIER_IMPL));
+  OpenMPLocationFlags Flags = OMP_IDENT_KMPC;
+  if (Kind == OMPD_for) {
+    Flags =
+        static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_IMPL_FOR);
+  } else if (Kind == OMPD_sections) {
+    Flags = static_cast<OpenMPLocationFlags>(Flags |
+                                             OMP_IDENT_BARRIER_IMPL_SECTIONS);
+  } else if (Kind == OMPD_single) {
+    Flags =
+        static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_IMPL_SINGLE);
+  } else if (Kind == OMPD_barrier) {
+    Flags = static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_EXPL);
+  } else {
+    Flags = static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_IMPL);
+  }
   // Build call __kmpc_cancel_barrier(loc, thread_id);
   // Replace __kmpc_barrier() function by __kmpc_cancel_barrier() because this
   // one provides the same functionality and adds initial support for

diff --git a/lib/CodeGen/CGOpenMPRuntime.h b/lib/CodeGen/CGOpenMPRuntime.h
index f8849e627c8e9b1e0bd943ac604bd325a3355a77..d32e2ab17329ad84ca4d62d4eefbb7999f8c944c 100644 (file)
--- a/lib/CodeGen/CGOpenMPRuntime.h
+++ b/lib/CodeGen/CGOpenMPRuntime.h
@@ -358,11 +358,12 @@ public:
                                 ArrayRef<const Expr *> DstExprs,
                                 ArrayRef<const Expr *> AssignmentOps);
 
-  /// \brief Emits explicit barrier for OpenMP threads.
-  /// \param IsExplicit true, if it is explicitly specified barrier.
+  /// \brief Emit an implicit/explicit barrier for OpenMP threads.
+  /// \param Kind Directive for which this implicit barrier call must be
+  /// generated. Must be OMPD_barrier for explicit barrier generation.
   ///
   virtual void emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
-                               bool IsExplicit = true);
+                               OpenMPDirectiveKind Kind);
 
   /// \brief Check if the specified \a ScheduleKind is static non-chunked.
   /// This kind of worksharing directive is emitted without outer loop.

diff --git a/lib/CodeGen/CGStmtOpenMP.cpp b/lib/CodeGen/CGStmtOpenMP.cpp
index 9af74749f17c4532fc01bece5faaa95c3282bfeb..f74e47a4ca5a137ac09dac1aeacdada7df55224e 100644 (file)
--- a/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/lib/CodeGen/CGStmtOpenMP.cpp
@@ -785,8 +785,9 @@ void CodeGenFunction::EmitOMPForDirective(const OMPForDirective &S) {
   EmitOMPWorksharingLoop(S);
 
   // Emit an implicit barrier at the end.
-  CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(),
-                                         /*IsExplicit*/ false);
+  if (!S.getSingleClause(OMPC_nowait)) {
+    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), OMPD_for);
+  }
 }
 
 void CodeGenFunction::EmitOMPForSimdDirective(const OMPForSimdDirective &) {
@@ -886,9 +887,11 @@ void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
   }
 
   // Emit an implicit barrier at the end.
-  if (!S.getSingleClause(OMPC_nowait))
-    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(),
-                                           /*IsExplicit=*/false);
+  if (!S.getSingleClause(OMPC_nowait)) {
+    CGM.getOpenMPRuntime().emitBarrierCall(
+        *this, S.getLocStart(),
+        (CS && CS->size() > 1) ? OMPD_sections : OMPD_single);
+  }
 }
 
 void CodeGenFunction::EmitOMPSectionDirective(const OMPSectionDirective &S) {
@@ -927,9 +930,9 @@ void CodeGenFunction::EmitOMPSingleDirective(const OMPSingleDirective &S) {
     EnsureInsertPoint();
   }, S.getLocStart(), CopyprivateVars, SrcExprs, DstExprs, AssignmentOps);
   // Emit an implicit barrier at the end.
-  if (!S.getSingleClause(OMPC_nowait))
-    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(),
-                                           /*IsExplicit=*/false);
+  if (!S.getSingleClause(OMPC_nowait)) {
+    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), OMPD_single);
+  }
 }
 
 void CodeGenFunction::EmitOMPMasterDirective(const OMPMasterDirective &S) {
@@ -1001,7 +1004,7 @@ void CodeGenFunction::EmitOMPTaskyieldDirective(
 }
 
 void CodeGenFunction::EmitOMPBarrierDirective(const OMPBarrierDirective &S) {
-  CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart());
+  CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), OMPD_barrier);
 }
 
 void CodeGenFunction::EmitOMPTaskwaitDirective(const OMPTaskwaitDirective &) {

diff --git a/test/OpenMP/for_codegen.cpp b/test/OpenMP/for_codegen.cpp
index f24f97ce231f54639cb910e3a4257e233eed271f..a53af80ac645c2aab3912b1f9e4b52bdec43ac4f 100644 (file)
--- a/test/OpenMP/for_codegen.cpp
+++ b/test/OpenMP/for_codegen.cpp
@@ -8,10 +8,11 @@
 #define HEADER
 
 // CHECK: [[IDENT_T_TY:%.+]] = type { i32, i32, i32, i32, i8* }
+// CHECK: [[IMPLICIT_BARRIER_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 66, i32 0, i32 0, i8*
 // CHECK-LABEL: define {{.*void}} @{{.*}}without_schedule_clause{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}})
 void without_schedule_clause(float *a, float *b, float *c, float *d) {
 // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num([[IDENT_T_TY]]* [[DEFAULT_LOC:[@%].+]])
-  #pragma omp for
+  #pragma omp for nowait
 // CHECK: call void @__kmpc_for_static_init_4([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]], i32 34, i32* [[IS_LAST:%[^,]+]], i32* [[OMP_LB:%[^,]+]], i32* [[OMP_UB:%[^,]+]], i32* [[OMP_ST:%[^,]+]], i32 1, i32 1)
 // UB = min(UB, GlobalUB)
 // CHECK-NEXT: [[UB:%.+]] = load i32, i32* [[OMP_UB]]
@@ -44,7 +45,7 @@ void without_schedule_clause(float *a, float *b, float *c, float *d) {
   }
 // CHECK: [[LOOP1_END]]
 // CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK-NOT: __kmpc_cancel_barrier
 // CHECK: ret void
 }
 
@@ -84,7 +85,7 @@ void static_not_chunked(float *a, float *b, float *c, float *d) {
   }
 // CHECK: [[LOOP1_END]]
 // CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
 // CHECK: ret void
 }
 
@@ -143,7 +144,7 @@ void static_chunked(float *a, float *b, float *c, float *d) {
 
 // CHECK: [[O_LOOP1_END]]
 // CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
 // CHECK: ret void
 }
 
@@ -184,7 +185,7 @@ void dynamic1(float *a, float *b, float *c, float *d) {
   }
 // CHECK: [[LOOP1_END]]
 // CHECK: [[O_LOOP1_END]]
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
 // CHECK: ret void
 }
 
@@ -225,7 +226,7 @@ void guided7(float *a, float *b, float *c, float *d) {
   }
 // CHECK: [[LOOP1_END]]
 // CHECK: [[O_LOOP1_END]]
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
 // CHECK: ret void
 }
 
@@ -269,7 +270,7 @@ void test_auto(float *a, float *b, float *c, float *d) {
   }
 // CHECK: [[LOOP1_END]]
 // CHECK: [[O_LOOP1_END]]
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
 // CHECK: ret void
 }
 
@@ -310,7 +311,7 @@ void runtime(float *a, float *b, float *c, float *d) {
   }
 // CHECK: [[LOOP1_END]]
 // CHECK: [[O_LOOP1_END]]
-// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[DEFAULT_LOC_BARRIER:[@%].+]], i32 [[GTID]])
+// CHECK: call {{.+}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_LOC]], i32 [[GTID]])
 // CHECK: ret void
 }
 

diff --git a/test/OpenMP/sections_codegen.cpp b/test/OpenMP/sections_codegen.cpp
index 9485d9ffe209dd69199b47832f9b091fa76f9a2b..e385550e883e3ddc1baad177a1366eb9b9c4b2f5 100644 (file)
--- a/test/OpenMP/sections_codegen.cpp
+++ b/test/OpenMP/sections_codegen.cpp
@@ -5,7 +5,8 @@
 
 #ifndef HEADER
 #define HEADER
-
+// CHECK: [[IMPLICIT_BARRIER_SECTIONS_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 194, i32 0, i32 0, i8*
+// CHECK: [[IMPLICIT_BARRIER_SINGLE_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 322, i32 0, i32 0, i8*
 // CHECK-LABEL: foo
 void foo() {};
 // CHECK-LABEL: bar
@@ -26,7 +27,7 @@ int main() {
   float l = 0.0; // Used as a base point in checks.
 // CHECK: [[GTID:%.+]] = call{{.*}} i32 @__kmpc_global_thread_num({{.*}})
 // CHECK: store float
-#pragma omp sections nowait
+#pragma omp sections
   {
 // CHECK:      store i32 0, i32* [[LB_PTR:%.+]],
 // CHECK:      store i32 1, i32* [[UB_PTR:%.+]],
@@ -71,6 +72,13 @@ int main() {
 // CHECK:      [[INNER_LOOP_END]]
   }
 // CHECK:      call void @__kmpc_for_static_fini(%{{.+}}* @{{.+}}, i32 [[GTID]])
+// CHECK:      call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_SECTIONS_LOC]],
+#pragma omp sections nowait
+  {
+    foo();
+#pragma omp section
+    bar();
+  }
 // CHECK-NOT:  __kmpc_cancel_barrier
   return tmain<int>();
 }
@@ -87,7 +95,7 @@ int main() {
 // CHECK:       call void @__kmpc_end_single(
 // CHECK-NEXT:  br label %[[END]]
 // CHECK:       [[END]]
-// CHECK-NEXT:  call i32 @__kmpc_cancel_barrier(
+// CHECK-NEXT:  call i32 @__kmpc_cancel_barrier(%{{.+}}* [[IMPLICIT_BARRIER_SINGLE_LOC]],
 // CHECK-NEXT:  ret
 // CHECK:       [[TERM_LPAD]]
 // CHECK:       call void @__clang_call_terminate(i8*

diff --git a/test/OpenMP/single_codegen.cpp b/test/OpenMP/single_codegen.cpp
index 2b8a9cd5886f2b7f970da3e45accf0f0af6bbf31..5114a8e2b2a15e56aaebb53289e5c4ed52ea3345 100644 (file)
--- a/test/OpenMP/single_codegen.cpp
+++ b/test/OpenMP/single_codegen.cpp
@@ -18,6 +18,7 @@ public:
 
 // CHECK-DAG:   [[TEST_CLASS_TY:%.+]] = type { i{{[0-9]+}} }
 // CHECK:       [[IDENT_T_TY:%.+]] = type { i32, i32, i32, i32, i8* }
+// CHECK:       [[IMPLICIT_BARRIER_SINGLE_LOC:@.+]] = private unnamed_addr constant %{{.+}} { i32 0, i32 322, i32 0, i32 0, i8*
 
 // CHECK:       define void [[FOO:@.+]]()
 
@@ -47,6 +48,7 @@ int main() {
 // 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
 #pragma omp single nowait
   a = 2;
 // CHECK:       [[RES:%.+]] = call i32 @__kmpc_single([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
@@ -73,7 +75,7 @@ int main() {
 // CHECK:       [[COPY_LIST_VOID_PTR:%.+]] = bitcast [3 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 24, i8* [[COPY_LIST_VOID_PTR]], void (i8*, i8*)* [[COPY_FUNC:@.+]], i32 [[DID_IT_VAL]])
-// CHECK:       call{{.*}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* {{@.+}}, i32 [[GTID]])
+// CHECK:       call{{.*}} @__kmpc_cancel_barrier([[IDENT_T_TY]]* [[IMPLICIT_BARRIER_SINGLE_LOC]], i32 [[GTID]])
 #pragma omp single copyprivate(a, c, tc)
   foo();
 // CHECK-NOT:   call i32 @__kmpc_single