Hi rjmccall, hfinkel,
Inner bodies of OpenMP worksharing loop-based constructs with dynamic or guided
scheduling are allowed to be marked with !llvm.mem.parallel_loop_access
metadata for better optimization. Worksharing constructs with static scheduling
cannot be marked this way (according to OpenMP standard "A data dependence
between the same logical iterations in two such loops is guaranteed").
Constructs with auto and runtime scheduling are also not marked because
automatically chosen scheduling may be static also.
http://reviews.llvm.org/D9518
Files:
lib/CodeGen/CGStmtOpenMP.cpp
test/OpenMP/for_codegen.cpp
test/OpenMP/ordered_codegen.cpp
EMAIL PREFERENCES
http://reviews.llvm.org/settings/panel/emailpreferences/
Index: test/OpenMP/ordered_codegen.cpp
===================================================================
--- test/OpenMP/ordered_codegen.cpp
+++ test/OpenMP/ordered_codegen.cpp
@@ -39,6 +39,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
// CHECK-NEXT: call void @__kmpc_end_ordered([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
// ... end of ordered region ...
#pragma omp ordered
@@ -86,6 +87,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
// CHECK-NEXT: call void @__kmpc_end_ordered([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
// ... end of ordered region ...
#pragma omp ordered
@@ -139,6 +141,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
// CHECK-NEXT: call void @__kmpc_end_ordered([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
// ... end of ordered region ...
#pragma omp ordered
@@ -189,6 +192,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
// CHECK-NEXT: call void @__kmpc_end_ordered([[IDENT_T_TY]]* [[DEFAULT_LOC]], i32 [[GTID]])
// ... end of ordered region ...
#pragma omp ordered
Index: test/OpenMP/for_codegen.cpp
===================================================================
--- test/OpenMP/for_codegen.cpp
+++ test/OpenMP/for_codegen.cpp
@@ -41,6 +41,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
@@ -81,6 +82,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
@@ -129,6 +131,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add i32 [[IV1_2]], 1
@@ -180,7 +183,7 @@
// CHECK-NEXT: store i64 [[CALC_I_2]], i64* [[LC_I:.+]]
// ... loop body ...
// End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK: store float [[RESULT:%.+]], float* {{%.+}}!llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add i64 [[IV1_2]], 1
@@ -221,7 +224,7 @@
// CHECK-NEXT: store i64 [[CALC_I_2]], i64* [[LC_I:.+]]
// ... loop body ...
// End of body: store into a[i]:
-// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK: store float [[RESULT:%.+]], float* {{%.+}}!llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add i64 [[IV1_2]], 1
@@ -266,6 +269,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i64 [[IV1_2]], 1
@@ -307,6 +311,7 @@
// ... loop body ...
// End of body: store into a[i]:
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}
+// CHECK-NOT: !llvm.mem.parallel_loop_access
a[i] = b[i] * c[i] * d[i];
// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}
// CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
Index: lib/CodeGen/CGStmtOpenMP.cpp
===================================================================
--- lib/CodeGen/CGStmtOpenMP.cpp
+++ lib/CodeGen/CGStmtOpenMP.cpp
@@ -886,6 +886,11 @@
bool DynamicWithOrderedClause =
Dynamic && S.getSingleClause(OMPC_ordered) != nullptr;
SourceLocation Loc = S.getLocStart();
+ // Generate !llvm.loop.parallel metadata for loads and stores for loops with
+ // dynamic/guided scheduling and without ordered clause.
+ LoopStack.setParallel((ScheduleKind == OMPC_SCHEDULE_dynamic ||
+ ScheduleKind == OMPC_SCHEDULE_guided) &&
+ !DynamicWithOrderedClause);
EmitOMPInnerLoop(
S, LoopScope.requiresCleanups(), S.getCond(/*SeparateIter=*/false),
S.getInc(),
_______________________________________________
cfe-commits mailing list
cfe-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
