Skip to content

Commit 0b49f9e

Browse files
alexey-bataevalexey-bataev
committed
[OPENMP] Codegen for distribute parallel for simd directive.
Initial codegen for `#pragma omp distribute parallel for simd` directive and its clauses. llvm-svn: 319079
1 parent 974acd6 commit 0b49f9e

8 files changed

+4245
-7
lines changed

‎clang/lib/CodeGen/CGStmtOpenMP.cpp

Lines changed: 5 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -2044,14 +2044,12 @@ void CodeGenFunction::EmitOMPDistributeParallelForDirective(
20442044

20452045
void CodeGenFunction::EmitOMPDistributeParallelForSimdDirective(
20462046
const OMPDistributeParallelForSimdDirective &S) {
2047+
auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &) {
2048+
CGF.EmitOMPDistributeLoop(S, emitInnerParallelForWhenCombined,
2049+
S.getDistInc());
2050+
};
20472051
OMPLexicalScope Scope(*this, S, /*AsInlined=*/true);
2048-
CGM.getOpenMPRuntime().emitInlinedDirective(
2049-
*this, OMPD_distribute_parallel_for_simd,
2050-
[&S](CodeGenFunction &CGF, PrePostActionTy &) {
2051-
OMPLoopScope PreInitScope(CGF, S);
2052-
CGF.EmitStmt(
2053-
cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
2054-
});
2052+
CGM.getOpenMPRuntime().emitInlinedDirective(*this, OMPD_distribute, CodeGen);
20552053
}
20562054

20572055
void CodeGenFunction::EmitOMPDistributeSimdDirective(

‎clang/test/OpenMP/distribute_parallel_for_simd_codegen.cpp

Lines changed: 2262 additions & 0 deletions
Large diffs are not rendered by default.

‎clang/test/OpenMP/distribute_parallel_for_simd_firstprivate_codegen.cpp

Lines changed: 619 additions & 0 deletions
Large diffs are not rendered by default.

‎clang/test/OpenMP/distribute_parallel_for_simd_if_codegen.cpp

Lines changed: 192 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,192 @@
1+
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -emit-llvm %s -o - | FileCheck %s
2+
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -std=c++11 -triple %itanium_abi_triple -emit-pch -o %t %s
3+
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fopenmp-targets=powerpc64le-ibm-linux-gnu -x c++ -triple %itanium_abi_triple -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s
4+
// expected-no-diagnostics
5+
#ifndef HEADER
6+
#define HEADER
7+
8+
void fn1();
9+
void fn2();
10+
void fn3();
11+
void fn4();
12+
void fn5();
13+
void fn6();
14+
15+
int Arg;
16+
17+
// CHECK-LABEL: define {{.*}}void @{{.+}}gtid_test
18+
void gtid_test() {
19+
#pragma omp target
20+
#pragma omp teams
21+
// CHECK: call i{{[0-9]+}} @__tgt_target_teams(
22+
// CHECK: call void [[OFFLOADING_FUN_0:@.+]](
23+
// CHECK: call i{{[0-9]+}} @__tgt_target_teams(
24+
// CHECK: call void [[OFFLOADING_FUN_1:@.+]](
25+
#pragma omp distribute parallel for simd
26+
for(int i = 0 ; i < 100; i++) {}
27+
// CHECK: define internal void [[OFFLOADING_FUN_0]](
28+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_0:@.+]] to {{.+}})
29+
// CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_0]](
30+
// CHECK: call void @__kmpc_for_static_init_4(
31+
// CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_0:@.+]] to void
32+
// CHECK: call void @__kmpc_for_static_fini(
33+
34+
// CHECK: define{{.+}} void [[OMP_OUTLINED_0]](
35+
// CHECK: call void @__kmpc_for_static_init_4(
36+
// CHECK: call void @__kmpc_for_static_fini(
37+
// CHECK: ret
38+
#pragma omp target
39+
#pragma omp teams
40+
#pragma omp distribute parallel for simd if (parallel: false)
41+
for(int i = 0 ; i < 100; i++) {
42+
// CHECK: define internal void [[OFFLOADING_FUN_1]](
43+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_1:@.+]] to {{.+}})
44+
// CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_1]](
45+
// CHECK: call void @__kmpc_for_static_init_4(
46+
// CHECK: call void @__kmpc_serialized_parallel(
47+
// CHECK: call void [[OMP_OUTLINED_1:@.+]](
48+
// CHECK: call void @__kmpc_end_serialized_parallel(
49+
// CHECK: call void @__kmpc_for_static_fini(
50+
// CHECK: define{{.+}} void [[OMP_OUTLINED_1]](
51+
// CHECK: call void @__kmpc_for_static_init_4(
52+
// CHECK: call void @{{.+}}gtid_test
53+
// CHECK: call void @__kmpc_for_static_fini(
54+
// CHECK: ret
55+
gtid_test();
56+
}
57+
}
58+
59+
60+
template <typename T>
61+
int tmain(T Arg) {
62+
#pragma omp target
63+
#pragma omp teams
64+
#pragma omp distribute parallel for simd if (true)
65+
for(int i = 0 ; i < 100; i++) {
66+
fn1();
67+
}
68+
#pragma omp target
69+
#pragma omp teams
70+
#pragma omp distribute parallel for simd if (false)
71+
for(int i = 0 ; i < 100; i++) {
72+
fn2();
73+
}
74+
#pragma omp target
75+
#pragma omp teams
76+
#pragma omp distribute parallel for simd if (parallel: Arg)
77+
for(int i = 0 ; i < 100; i++) {
78+
fn3();
79+
}
80+
return 0;
81+
}
82+
83+
// CHECK-LABEL: define {{.*}}i{{[0-9]+}} @main()
84+
int main() {
85+
// CHECK: call i{{[0-9]+}} @__tgt_target_teams(
86+
// CHECK: call void [[OFFLOADING_FUN_0:@.+]](
87+
// CHECK: call i{{[0-9]+}} @__tgt_target_teams(
88+
// CHECK: call void [[OFFLOADING_FUN_1:@.+]](
89+
// CHECK: call i{{[0-9]+}} @__tgt_target_teams(
90+
// CHECK: call void [[OFFLOADING_FUN_2:@.+]](
91+
// CHECK: = call {{.*}}i{{.+}} @{{.+}}tmain
92+
#pragma omp target
93+
#pragma omp teams
94+
#pragma omp distribute parallel for simd if (true)
95+
for(int i = 0 ; i < 100; i++) {
96+
// CHECK: define internal void [[OFFLOADING_FUN_0]](
97+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_0:@.+]] to {{.+}})
98+
// CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_0]](
99+
100+
// CHECK: call void @__kmpc_for_static_init_4(
101+
// CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_2:@.+]] to void
102+
// CHECK: call void @__kmpc_for_static_fini(
103+
// CHECK: define{{.+}} void [[OMP_OUTLINED_2]](
104+
// CHECK: call void @__kmpc_for_static_init_4(
105+
// CHECK: call {{.*}}void @{{.+}}fn4
106+
// CHECK: call void @__kmpc_for_static_fini(
107+
108+
fn4();
109+
}
110+
111+
#pragma omp target
112+
#pragma omp teams
113+
#pragma omp distribute parallel for simd if (false)
114+
for(int i = 0 ; i < 100; i++) {
115+
// CHECK: define internal void [[OFFLOADING_FUN_1]](
116+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 0, {{.+}}* [[OMP_TEAMS_OUTLINED_1:@.+]] to {{.+}})
117+
// CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_1]](
118+
119+
// CHECK: call void @__kmpc_for_static_init_4(
120+
// CHECK: call void @__kmpc_serialized_parallel(
121+
// CHECK: call void [[OMP_OUTLINED_3:@.+]](
122+
// CHECK: call void @__kmpc_end_serialized_parallel(
123+
// CHECK: call void @__kmpc_for_static_fini(
124+
125+
// CHECK: define{{.+}} void [[OMP_OUTLINED_3]](
126+
// CHECK: call void @__kmpc_for_static_init_4(
127+
// CHECK: call {{.*}}void @{{.+}}fn5
128+
// CHECK: call void @__kmpc_for_static_fini(
129+
fn5();
130+
}
131+
132+
#pragma omp target
133+
#pragma omp teams
134+
#pragma omp distribute parallel for simd if (Arg)
135+
for(int i = 0 ; i < 100; i++) {
136+
// CHECK: define internal void [[OFFLOADING_FUN_2]](
137+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}}* [[OMP_TEAMS_OUTLINED_2:@.+]] to {{.+}})
138+
// CHECK: define{{.+}} void [[OMP_TEAMS_OUTLINED_2]](
139+
140+
// CHECK: call void @__kmpc_for_static_init_4(
141+
// CHECK: call void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, {{.+}}* [[OMP_OUTLINED_4:@.+]] to void
142+
// CHECK: call void @__kmpc_serialized_parallel(
143+
// CHECK: call void [[OMP_OUTLINED_4:@.+]](
144+
// CHECK: call void @__kmpc_end_serialized_parallel(
145+
// CHECK: call void @__kmpc_for_static_fini(
146+
147+
// CHECK: define{{.+}} void [[OMP_OUTLINED_4]](
148+
// CHECK: call void @__kmpc_for_static_init_4(
149+
// CHECK: call {{.*}}void @{{.+}}fn6
150+
// CHECK: call void @__kmpc_for_static_fini(
151+
fn6();
152+
}
153+
154+
return tmain(Arg);
155+
}
156+
157+
// CHECK-LABEL: define {{.+}} @{{.+}}tmain
158+
159+
// CHECK: call void @__kmpc_for_static_init_4(
160+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, void {{.+}}* [[T_OUTLINE_FUN_1:@.+]] to void
161+
// CHECK: call void @__kmpc_for_static_fini(
162+
163+
// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_1]]
164+
// CHECK: call void @__kmpc_for_static_init_4(
165+
// CHECK: call {{.*}}void @{{.+}}fn1
166+
// CHECK: call void @__kmpc_for_static_fini(
167+
// CHECK: ret void
168+
169+
// CHECK: call void @__kmpc_for_static_init_4(
170+
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(
171+
// CHECK: call void [[T_OUTLINE_FUN_2:@.+]](
172+
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(
173+
// CHECK: call void @__kmpc_for_static_fini(
174+
175+
// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_2]]
176+
// CHECK: call void @__kmpc_for_static_init_4(
177+
// CHECK: call {{.*}}void @{{.+}}fn2
178+
// CHECK: call void @__kmpc_for_static_fini(
179+
// CHECK: ret void
180+
181+
// CHECK: call void @__kmpc_for_static_init_4(
182+
// CHECK: call {{.*}}void {{.+}} @__kmpc_fork_call(%{{.+}}* @{{.+}}, i{{.+}} 2, void {{.+}}* [[T_OUTLINE_FUN_3:@.+]] to void
183+
// CHECK: call {{.*}}void @__kmpc_serialized_parallel(
184+
// call void [[T_OUTLINE_FUN_3:@.+]](
185+
// CHECK: call {{.*}}void @__kmpc_end_serialized_parallel(
186+
187+
// CHECK: define internal {{.*}}void [[T_OUTLINE_FUN_3]]
188+
// CHECK: call void @__kmpc_for_static_init_4(
189+
// CHECK: call {{.*}}void @{{.+}}fn3
190+
// CHECK: call void @__kmpc_for_static_fini(
191+
// CHECK: ret void
192+
#endif

0 commit comments

Comments
 (0)