• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // expected-no-diagnostics
2 #ifndef HEADER
3 #define HEADER
4 // Test host codegen.
5 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
6 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
7 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-64
8 // RUN: %clang_cc1 -DCK1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32
9 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
10 // RUN: %clang_cc1 -DCK1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK1 --check-prefix CK1-32
11 
12 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
13 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
14 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
15 // RUN: %clang_cc1 -DCK1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
16 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
17 // RUN: %clang_cc1 -DCK1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
18 // SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
19 #ifdef CK1
20 
21 int a[100];
22 
23 // CK1: define {{.*}}i32 @{{.+}}teams_argument_globali(
teams_argument_global(int n)24 int teams_argument_global(int n) {
25   int i;
26   int te = n / 128;
27   int th = 128;
28   // discard n_addr and i
29   // CK1: alloca i32,
30   // CK1: alloca i32,
31   // CK1: [[TE:%.+]] = alloca i32,
32   // CK1: [[TH:%.+]] = alloca i32,
33   // CK1: [[TE_CAST:%.+]] = alloca i{{32|64}},
34   // CK1: [[TH_CAST:%.+]] = alloca i{{32|64}},
35   // CK1: [[TE_PAR:%.+]] = load{{.+}}, {{.+}} [[TE_CAST]],
36   // CK1: [[TH_PAR:%.+]] = load{{.+}}, {{.+}} [[TH_CAST]],
37 
38   // CK1: call void @__kmpc_push_target_tripcount(%struct.ident_t* @{{.+}}, i64 -1, i64 %{{.+}})
39   // CK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 5, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 {{.+}}, i32 1)
40 
41   // CK1: call void @[[OFFL1:.+]](i{{32|64}} [[TE_PAR]], i{{32|64}} [[TH_PAR]],
42   #pragma omp target
43   #pragma omp teams distribute simd num_teams(te), thread_limit(th) aligned(a) simdlen(16) linear(i)
44   for(i = 0; i < n; i++) {
45     a[i] = 0;
46   }
47 
48   // CK1: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 2, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 0, i32 1)
49   // CK1: call void @[[OFFL2:.+]](i{{64|32}} %{{.+}})
50   #pragma omp target
51   {{{
52   #pragma omp teams distribute simd safelen(32)
53   for(int i = 0; i < n; i++) {
54     a[i] = 0;
55   }
56   }}}
57 
58   // outlined target regions
59   // CK1: define internal void @[[OFFL1]](i{{32|64}} [[TE_ARG:%.+]], i{{32|64}} [[TH_ARG:%.+]], [100 x i{{32|64}}]* {{.+}}, i{{32|64}} {{.+}}, {{.+}})
60   // CK1: [[TE_ADDR:%.+]] = alloca i{{32|64}},
61   // CK1: [[TH_ADDR:%.+]] = alloca i{{32|64}},
62   // CK1: store{{.+}} [[TE_ARG]], {{.+}} [[TE_ADDR]],
63   // CK1: store{{.+}} [[TH_ARG]], {{.+}} [[TH_ADDR]],
64   // CK1-64: [[TE_CONV:%.+]] = bitcast{{.+}} [[TE_ADDR]] to
65   // CK1-64: [[TH_CONV:%.+]] = bitcast{{.+}} [[TH_ADDR]] to
66   // CK1-64: [[TE_VAL:%.+]] = load i32, i32* [[TE_CONV]],
67   // CK1-64: [[TH_VAL:%.+]] = load i32, i32* [[TH_CONV]],
68   // CK1-32: [[TE_VAL:%.+]] = load i32, i32* [[TE_ADDR]],
69   // CK1-32: [[TH_VAL:%.+]] = load i32, i32* [[TH_ADDR]],
70   // CK1: call void @__kmpc_push_num_teams({{.+}}, {{.+}}, i32 [[TE_VAL]], i32 [[TH_VAL]])
71   // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}})
72   // CK1: ret void
73 
74   // CK1: define internal void @[[OUTL1]]({{.+}})
75   // CK1: call void @__kmpc_for_static_init_4(
76   // CK1: call void @__kmpc_for_static_fini(
77   // CK1: ret void
78 
79   // CK1: define internal void @[[OFFL2]]({{.+}}, {{.+}})
80   // CK1: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 2, {{.+}} @[[OUTL2:.+]] to {{.+}}, {{.+}}, {{.+}})
81   // CK1: ret void
82 
83   // CK1: define internal void @[[OUTL2]]({{.+}})
84   // CK1: call void @__kmpc_for_static_init_4(
85   // CK1: call void @__kmpc_for_static_fini(
86   // CK1: ret void
87 
88   return a[0];
89 }
90 
91 // CK1-DAG: !{!"llvm.loop.vectorize.width", i32 16}
92 // CK1-DAG: !{!"llvm.loop.vectorize.enable", i1 true}
93 // CK1-DAG: !{!"llvm.loop.vectorize.width", i32 32}
94 
95 #endif // CK1
96 
97 // Test host codegen.
98 // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
99 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
100 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-64
101 // RUN: %clang_cc1 -DCK2 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32
102 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
103 // RUN: %clang_cc1 -DCK2 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK2 --check-prefix CK2-32
104 
105 // RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
106 // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
107 // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
108 // RUN: %clang_cc1 -DCK2 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY1 %s
109 // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
110 // RUN: %clang_cc1 -DCK2 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
111 // SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
112 #ifdef CK2
113 
114 // CK2: define {{.*}}i32 @{{.+}}teams_local_argv(
teams_local_arg(void)115 int teams_local_arg(void) {
116   int n = 100;
117   int a[n];
118 
119   // CK2: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 0, i32 1)
120   // CK2: call void @[[OFFL1:.+]](i{{64|32}} %{{.+}})
121   #pragma omp target
122   #pragma omp teams distribute simd
123   for(int i = 0; i < n; i++) {
124     a[i] = 0;
125   }
126 
127   // outlined target region
128   // CK2: define internal void @[[OFFL1]]({{.+}}, {{.+}})
129   // CK2: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}})
130   // CK2: ret void
131 
132   // CK2: define internal void @[[OUTL1]]({{.+}})
133   // CK2: call void @__kmpc_for_static_init_4(
134   // CK2: call void @__kmpc_for_static_fini(
135   // CK2: ret void
136 
137   return a[0];
138 }
139 // CK2: !{!"llvm.loop.vectorize.enable", i1 true}
140 #endif // CK2
141 
142 // Test host codegen.
143 // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64 --check-prefix OMP3_45
144 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
145 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64 --check-prefix OMP3_45
146 // RUN: %clang_cc1 -DCK3 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 --check-prefix OMP3_45
147 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
148 // RUN: %clang_cc1 -DCK3 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 --check-prefix OMP3_45
149 // RUN: %clang_cc1 -DCK3 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64 --check-prefix OMP3_50
150 // RUN: %clang_cc1 -DCK3 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
151 // RUN: %clang_cc1 -DCK3 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-64 -check-prefix OMP3_50
152 // RUN: %clang_cc1 -DCK3 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 -check-prefix OMP3_50
153 // RUN: %clang_cc1 -DCK3 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
154 // RUN: %clang_cc1 -DCK3 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK3 --check-prefix CK3-32 -check-prefix OMP3_50
155 
156 // RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
157 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
158 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
159 // RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
160 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
161 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
162 // RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
163 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
164 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
165 // RUN: %clang_cc1 -DCK3 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY2 %s
166 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
167 // RUN: %clang_cc1 -DCK3 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY2 %s
168 // SIMD-ONLY2-NOT: {{__kmpc|__tgt}}
169 #ifdef CK3
170 
171 // CK3: [[SSI:%.+]] = type { [{{.+}} x i32], float }
172 
173 template <typename T, int X, long long Y>
174 struct SS{
175   T a[X];
176   float b;
177   // CK3: define {{.*}}i32 @{{.+}}foo{{.+}}(
fooSS178   int foo(void) {
179 
180   // CK3: call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* %{{.+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 0, i32 1)
181   // CK3: call void @[[OFFL1:.+]]([[SSI]]* %{{.+}})
182     #pragma omp target
183 #ifdef OMP5
184     #pragma omp teams distribute simd if(b) nontemporal(a, b)
185 #else
186     #pragma omp teams distribute simd
187 #endif // OMP5
188     for(int i = 0; i < X; i++) {
189       a[i] = (T)b;
190     }
191 
192       // outlined target region
193   // CK3: define internal void @[[OFFL1]]([[SSI]]* {{.+}})
194   // OMP3_45: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}})
195   // OMP3_50: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 2, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}}, {{.+}})
196   // CK3: ret void
197 
198   // CK3: define internal void @[[OUTL1]]({{.+}})
199   // CK3: call void @__kmpc_for_static_init_4(
200   // OMP3_45-NOT: !nontemporal
201   // OMP3_50: load float,{{.*}}!nontemporal
202   // CK3: call void @__kmpc_for_static_fini(
203   // CK3: ret void
204 
205     return a[0];
206   }
207 };
208 
teams_template_struct(void)209 int teams_template_struct(void) {
210   SS<int, 123, 456> V;
211   return V.foo();
212 
213 }
214 // CK3-DAG: !{!"llvm.loop.vectorize.enable", i1 true}
215 // OMP3_45-NOT: !{!"llvm.loop.vectorize.enable", i1 false}
216 // OMP3_50-DAG: !{!"llvm.loop.vectorize.enable", i1 false}
217 #endif // CK3
218 
219 // Test host codegen.
220 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64 --check-prefix OMP4_45
221 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
222 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64 --check-prefix OMP4_45
223 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 --check-prefix OMP4_45
224 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
225 // RUN: %clang_cc1 -DCK4 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 --check-prefix OMP4_45
226 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64 --check-prefix OMP4_50
227 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
228 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-64 --check-prefix OMP4_50
229 // RUN: %clang_cc1 -DCK4 -verify -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 --check-prefix OMP4_50
230 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
231 // RUN: %clang_cc1 -DCK4 -fopenmp -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp -fopenmp-version=50 -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CK4 --check-prefix CK4-32 --check-prefix OMP4_50
232 
233 // RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
234 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
235 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
236 // RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
237 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
238 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
239 // RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
240 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
241 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
242 // RUN: %clang_cc1 -DCK4 -verify -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY3 %s
243 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
244 // RUN: %clang_cc1 -DCK4 -fopenmp-simd -fopenmp-version=50 -DOMP5 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY3 %s
245 // SIMD-ONLY3-NOT: {{__kmpc|__tgt}}
246 
247 #ifdef CK4
248 
249 template <typename T, int n>
tmain(T argc)250 int tmain(T argc) {
251   T a[n];
252   int te = n/128;
253   int th = 128;
254 #pragma omp target
255 #pragma omp teams distribute simd num_teams(te) thread_limit(th)
256   for(int i = 0; i < n; i++) {
257     a[i] = (T)0;
258   }
259   return 0;
260 }
261 
main(int argc,char ** argv)262 int main (int argc, char **argv) {
263   int n = 100;
264   int a[n];
265 #pragma omp target
266 #ifdef OMP5
267 #pragma omp teams distribute simd if(simd:argc)
268 #else
269 #pragma omp teams distribute simd
270 #endif // OMP5
271   for(int i = 0; i < n; i++) {
272     a[i] = 0;
273   }
274   return tmain<int, 10>(argc);
275 }
276 
277 // CK4:  define {{.*}}i32 @{{[^,]+}}(i{{.+}}{{.+}} %[[ARGC:.+]], {{.+}})
278 // OMP4_45:   call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 0, i32 1)
279 // OMP4_50:   call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 4, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}, i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 0, i32 1)
280 // CK4: call void @[[OFFL1:.+]]({{.+}})
281 // CK4: {{%.+}} = call{{.*}} i32 @[[TMAIN:.+]]({{.+}})
282 // CK4:  ret
283 
284 // CK4:  define {{.*}}void @[[OFFL1]]({{.+}})
285 // OMP4_45: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 3, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}})
286 // OMP4_50: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 4, {{.+}} @[[OUTL1:.+]] to {{.+}}, {{.+}}, {{.+}})
287 // CK4: ret void
288 
289 // CK4: define internal void @[[OUTL1]]({{.+}})
290 // CK4: call void @__kmpc_for_static_init_4(
291 // CK4: call void @__kmpc_for_static_fini(
292 // CK4: ret void
293 
294 // CK4:  define {{.*}}i32 @[[TMAIN]]({{.+}})
295 // CK4:   call i32 @__tgt_target_teams_mapper(%struct.ident_t* @{{.+}}, i64 -1, i8* @{{[^,]+}}, i32 3, i8** %{{[^,]+}}, i8** %{{[^,]+}}, i{{64|32}}* {{.+}}@{{[^,]+}}, i32 0, i32 0), i64* {{.+}}@{{[^,]+}}, i32 0, i32 0), i8** null, i8** null, i32 {{.+}}, i32 1)
296 // CK4: call void @[[OFFLT:.+]]({{.+}})
297 // CK4:  ret
298 // CK4-NEXT: }
299 
300 // CK4: define {{.*}}void @[[OFFLT]](i{{32|64}} [[TE_ARG:%.+]], i{{32|64}} [[TH_ARG:%.+]], {{.+}})
301 // CK4: [[TE_ADDR:%.+]] = alloca i{{32|64}},
302 // CK4: [[TH_ADDR:%.+]] = alloca i{{32|64}},
303 // CK4: store{{.+}} [[TE_ARG]], {{.+}} [[TE_ADDR]],
304 // CK4: store{{.+}} [[TH_ARG]], {{.+}} [[TH_ADDR]],
305 // CK4-64: [[TE_CONV:%.+]] = bitcast{{.+}} [[TE_ADDR]] to
306 // CK4-64: [[TH_CONV:%.+]] = bitcast{{.+}} [[TH_ADDR]] to
307 // CK4-64: [[TE_VAL:%.+]] = load i32, i32* [[TE_CONV]],
308 // CK4-64: [[TH_VAL:%.+]] = load i32, i32* [[TH_CONV]],
309 // CK4-32: [[TE_VAL:%.+]] = load i32, i32* [[TE_ADDR]],
310 // CK4-32: [[TH_VAL:%.+]] = load i32, i32* [[TH_ADDR]],
311 // CK4: call void @__kmpc_push_num_teams({{.+}}, {{.+}}, i32 [[TE_VAL]], i32 [[TH_VAL]])
312 // CK4: call void {{.+}} @__kmpc_fork_teams({{.+}}, i32 1, {{.+}} @[[OUTLT:.+]] to {{.+}}, {{.+}}, {{.+}})
313 // CK4: ret void
314 
315 // CK4: define internal void @[[OUTLT]]({{.+}})
316 // CK4: call void @__kmpc_for_static_init_4(
317 // CK4: call void @__kmpc_for_static_fini(
318 // CK4: ret void
319 
320 // CK4-DAG: !{!"llvm.loop.vectorize.enable", i1 true}
321 // OMP4_45-NOT: !{!"llvm.loop.vectorize.enable", i1 false}
322 // OMP5_50-DAG: !{!"llvm.loop.vectorize.enable", i1 false}
323 #endif // CK4
324 #endif
325 
326