Author: abataev
Date: Thu Oct 17 10:12:03 2019
New Revision: 375134
URL: http://llvm.org/viewvc/llvm-project?rev=375134&view=rev
Log:
[OPENMP]Improve use of the global tid parameter.
If we can determined, that the global tid parameter can be used in the
function, better to use it rather than calling __kmpc_global_thread_num
function.
Modified:
cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp
cfe/trunk/test/OpenMP/openmp_win_codegen.cpp
cfe/trunk/test/OpenMP/parallel_for_codegen.cpp
Modified: cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp
URL:
http://llvm.org/viewvc/llvm-project/cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp?rev=375134&r1=375133&r2=375134&view=diff
==============================================================================
--- cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp (original)
+++ cfe/trunk/lib/CodeGen/CGOpenMPRuntime.cpp Thu Oct 17 10:12:03 2019
@@ -1697,18 +1697,23 @@ llvm::Value *CGOpenMPRuntime::getThreadI
return ThreadID;
}
// If exceptions are enabled, do not use parameter to avoid possible crash.
- if (!CGF.EHStack.requiresLandingPad() || !CGF.getLangOpts().Exceptions ||
- !CGF.getLangOpts().CXXExceptions ||
- CGF.Builder.GetInsertBlock() == CGF.AllocaInsertPt->getParent()) {
- if (auto *OMPRegionInfo =
- dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo)) {
- if (OMPRegionInfo->getThreadIDVariable()) {
- // Check if this an outlined function with thread id passed as
argument.
- LValue LVal = OMPRegionInfo->getThreadIDVariableLValue(CGF);
+ if (auto *OMPRegionInfo =
+ dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo)) {
+ if (OMPRegionInfo->getThreadIDVariable()) {
+ // Check if this an outlined function with thread id passed as argument.
+ LValue LVal = OMPRegionInfo->getThreadIDVariableLValue(CGF);
+ llvm::BasicBlock *TopBlock = CGF.AllocaInsertPt->getParent();
+ if (!CGF.EHStack.requiresLandingPad() || !CGF.getLangOpts().Exceptions ||
+ !CGF.getLangOpts().CXXExceptions ||
+ CGF.Builder.GetInsertBlock() == TopBlock ||
+ !isa<llvm::Instruction>(LVal.getPointer()) ||
+ cast<llvm::Instruction>(LVal.getPointer())->getParent() == TopBlock
||
+ cast<llvm::Instruction>(LVal.getPointer())->getParent() ==
+ CGF.Builder.GetInsertBlock()) {
ThreadID = CGF.EmitLoadOfScalar(LVal, Loc);
// If value loaded in entry block, cache it and use it everywhere in
// function.
- if (CGF.Builder.GetInsertBlock() == CGF.AllocaInsertPt->getParent()) {
+ if (CGF.Builder.GetInsertBlock() == TopBlock) {
auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
Elem.second.ThreadID = ThreadID;
}
Modified: cfe/trunk/test/OpenMP/openmp_win_codegen.cpp
URL:
http://llvm.org/viewvc/llvm-project/cfe/trunk/test/OpenMP/openmp_win_codegen.cpp?rev=375134&r1=375133&r2=375134&view=diff
==============================================================================
--- cfe/trunk/test/OpenMP/openmp_win_codegen.cpp (original)
+++ cfe/trunk/test/OpenMP/openmp_win_codegen.cpp Thu Oct 17 10:12:03 2019
@@ -50,11 +50,10 @@ int main() {
}
// CHECK: define internal void [[OUTLINED]](
-// CHECK: [[GID:%.+]] = {{.*}}call i32
@__kmpc_global_thread_num(%struct.ident_t* {{.*}}@0)
// CHECK: invoke void @{{.+}}foo
// CHECK: [[CATCHSWITCH:%.+]] = catchswitch within none
// CHECK: [[CATCHPAD:%.+]] = catchpad within [[CATCHSWITCH]]
-// CHECK: call void @__kmpc_critical(%struct.ident_t* {{.*}}@0, i32 [[GID]],
+// CHECK: call void @__kmpc_critical(%struct.ident_t* {{.*}}@0, i32
[[GID:%.+]],
// CHECK: invoke void @{{.+}}bar
// CHECK: call void @__kmpc_end_critical(%struct.ident_t* {{.*}}@0, i32
[[GID]],
// CHECK: catchret from [[CATCHPAD]] to
Modified: cfe/trunk/test/OpenMP/parallel_for_codegen.cpp
URL:
http://llvm.org/viewvc/llvm-project/cfe/trunk/test/OpenMP/parallel_for_codegen.cpp?rev=375134&r1=375133&r2=375134&view=diff
==============================================================================
--- cfe/trunk/test/OpenMP/parallel_for_codegen.cpp (original)
+++ cfe/trunk/test/OpenMP/parallel_for_codegen.cpp Thu Oct 17 10:12:03 2019
@@ -40,7 +40,7 @@ void with_var_schedule() {
// CHECK: [[CHUNK_VAL:%.+]] = load i8, i8* %
// CHECK: [[CHUNK_SIZE:%.+]] = sext i8 [[CHUNK_VAL]] to i64
// CHECK: call void @__kmpc_for_static_init_8u([[IDENT_T_TY]]* [[LOOP_LOC]],
i32 [[GTID:%[^,]+]], i32 33, i32* [[IS_LAST:%[^,]+]], i64* [[OMP_LB:%[^,]+]],
i64* [[OMP_UB:%[^,]+]], i64* [[OMP_ST:%[^,]+]], i64 1, i64 [[CHUNK_SIZE]])
-// CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[LOOP_LOC]], i32
[[GTID]])
+// CHECK: call void @__kmpc_for_static_fini([[IDENT_T_TY]]* [[LOOP_LOC]], i32
[[GTID:%.+]])
#pragma omp parallel for schedule(static, char(a)) private(a)
for (unsigned long long i = 1; i < 2 + a; ++i) {
}
@@ -284,7 +284,7 @@ void test_auto(float *a, float *b, float
// CHECK: store i32* [[GTID_PARAM_ADDR]], i32** [[GTID_REF_ADDR:%.+]],
// CHECK: call void @__kmpc_dispatch_init_8([[IDENT_T_TY]]* [[DEFAULT_LOC]],
i32 [[GTID:%.+]], i32 38, i64 0, i64 [[LAST_ITER:%[^,]+]], i64 1, i64 1)
//
-// CHECK: [[HASWORK:%.+]] = call i32 @__kmpc_dispatch_next_8([[IDENT_T_TY]]*
[[DEFAULT_LOC]], i32 [[GTID]], i32* [[OMP_ISLAST:%[^,]+]], i64*
[[OMP_LB:%[^,]+]], i64* [[OMP_UB:%[^,]+]], i64* [[OMP_ST:%[^,]+]])
+// CHECK: [[HASWORK:%.+]] = call i32 @__kmpc_dispatch_next_8([[IDENT_T_TY]]*
[[DEFAULT_LOC]], i32 [[GTID:%.+]], i32* [[OMP_ISLAST:%[^,]+]], i64*
[[OMP_LB:%[^,]+]], i64* [[OMP_UB:%[^,]+]], i64* [[OMP_ST:%[^,]+]])
// CHECK-NEXT: [[O_CMP:%.+]] = icmp ne i32 [[HASWORK]], 0
// CHECK-NEXT: br i1 [[O_CMP]], label %[[O_LOOP1_BODY:[^,]+]], label
%[[O_LOOP1_END:[^,]+]]
@@ -494,7 +494,6 @@ void range_for_single() {
// OMP5: [[IS_LAST:%.+]] = alloca i32,
// OMP5: [[BEGIN:%.+]] = alloca i32*,
// OMP5: [[A_PTR:%.+]] = alloca i32*,
-// OMP5: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(
// __range = arr;
// OMP5: [[ARR:%.+]] = load [10 x i32]*, [10 x i32]** [[ARR_ADDR]],
@@ -551,7 +550,7 @@ void range_for_single() {
// OMP5: store i32 0, i32* [[IS_LAST]],
// loop.
-// OMP5: call void @__kmpc_for_static_init_8(%struct.ident_t* {{.+}}, i32
[[GTID]], i32 34, i32* [[IS_LAST]], i64* [[LB]], i64* [[UB]], i64* [[STRIDE]],
i64 1, i64 1)
+// OMP5: call void @__kmpc_for_static_init_8(%struct.ident_t* {{.+}}, i32
[[GTID:%.+]], i32 34, i32* [[IS_LAST]], i64* [[LB]], i64* [[UB]], i64*
[[STRIDE]], i64 1, i64 1)
// ub = (ub > number_of_elems ? number_of_elems : ub);
// OMP5: [[UB_VAL:%.+]] = load i64, i64* [[UB]],
@@ -613,7 +612,7 @@ void range_for_single() {
// end:
// OMP5: [[END]]:
-// OMP5: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32
[[GTID]])
+// OMP5: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32
[[GTID:%.+]])
// exit:
// OMP5: [[EXIT]]:
// OMP5: ret void
@@ -621,12 +620,12 @@ void range_for_single() {
// OMP5-LABEL: range_for_collapsed
void range_for_collapsed() {
int arr[10] = {0};
-// OMP5: call void @__kmpc_for_static_init_8(%struct.ident_t* {{.+}}, i32
[[GTID%.+]], i32 34, i32* %{{.+}}, i64* %{{.+}}, i64* %{{.+}}, i64* %{{.+}},
i64 1, i64 1)
+// OMP5: call void @__kmpc_for_static_init_8(%struct.ident_t* {{.+}}, i32
[[GTID:%.+]], i32 34, i32* %{{.+}}, i64* %{{.+}}, i64* %{{.+}}, i64* %{{.+}},
i64 1, i64 1)
#pragma omp parallel for collapse(2)
for (auto &a : arr)
for (auto b : arr)
a = b;
-// OMP5: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32
[[GTID]])
+// OMP5: call void @__kmpc_for_static_fini(%struct.ident_t* {{.+}}, i32
[[GTID:%.+]])
}
#endif // OMP5
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
