r236693 - [OPENMP] Generate !llvm.mem.loop_parallel_access metadata for loops with dynamic/guided scheduling.
Alexey Bataev
a.bataev at hotmail.com
Wed May 6 21:25:18 PDT 2015
Author: abataev
Date: Wed May 6 23:25:17 2015
New Revision: 236693
URL: http://llvm.org/viewvc/llvm-project?rev=236693&view=rev
Log:
[OPENMP] Generate !llvm.mem.loop_parallel_access metadata for loops with dynamic/guided scheduling.
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.
Differential Revision: http://reviews.llvm.org/D9518
Modified:
cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp
cfe/trunk/test/OpenMP/for_codegen.cpp
cfe/trunk/test/OpenMP/ordered_codegen.cpp
Modified: cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp?rev=236693&r1=236692&r2=236693&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGStmtOpenMP.cpp Wed May 6 23:25:17 2015
@@ -886,6 +886,11 @@ void CodeGenFunction::EmitOMPForOuterLoo
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(),
Modified: cfe/trunk/test/OpenMP/for_codegen.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/OpenMP/for_codegen.cpp?rev=236693&r1=236692&r2=236693&view=diff
==============================================================================
--- cfe/trunk/test/OpenMP/for_codegen.cpp (original)
+++ cfe/trunk/test/OpenMP/for_codegen.cpp Wed May 6 23:25:17 2015
@@ -41,6 +41,7 @@ void without_schedule_clause(float *a, f
// ... 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 @@ void static_not_chunked(float *a, float
// ... 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 @@ void static_chunked(float *a, float *b,
// ... 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 @@ void dynamic1(float *a, float *b, float
// 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 @@ void guided7(float *a, float *b, float *
// 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 @@ void test_auto(float *a, float *b, float
// ... 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 @@ void runtime(float *a, float *b, float *
// ... 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
Modified: cfe/trunk/test/OpenMP/ordered_codegen.cpp
URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/OpenMP/ordered_codegen.cpp?rev=236693&r1=236692&r2=236693&view=diff
==============================================================================
--- cfe/trunk/test/OpenMP/ordered_codegen.cpp (original)
+++ cfe/trunk/test/OpenMP/ordered_codegen.cpp Wed May 6 23:25:17 2015
@@ -39,6 +39,7 @@ void static_not_chunked(float *a, float
// ... 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 @@ void dynamic1(float *a, float *b, float
// ... 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 @@ void test_auto(float *a, float *b, float
// ... 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 @@ void runtime(float *a, float *b, float *
// ... 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
More information about the cfe-commits
mailing list