1 // RUN: %libomp-compile && env OMP_NUM_THREADS=2,2,2,2,2 %libomp-run
2 #include <stdio.h>
3 #include "omp_testsuite.h"
4
5 // When compiler supports num_threads clause list format, remove the following
6 // and use num_threads clause directly
7 #if defined(__cplusplus)
8 extern "C" {
9 #endif
10
11 int __kmpc_global_thread_num(void *loc);
12 void __kmpc_push_num_threads_list(void *loc, int gtid, unsigned length,
13 int *list);
14
15 #if defined(__cplusplus)
16 }
17 #endif
18
test_omp_parallel_num_threads_list()19 int test_omp_parallel_num_threads_list() {
20 int num_failed = 0;
21
22 // Initially, 5 levels specified via OMP_NUM_THREADS with 2 threads per level
23 // Check top 3 levels
24 #pragma omp parallel reduction(+ : num_failed) // 1st level
25 {
26 #pragma omp single
27 num_failed = num_failed + !(omp_get_num_threads() == 2);
28 #pragma omp parallel reduction(+ : num_failed) // 2nd level
29 {
30 #pragma omp single
31 num_failed = num_failed + !(omp_get_num_threads() == 2);
32 #pragma omp parallel reduction(+ : num_failed) // 3rd level
33 {
34 #pragma omp single
35 num_failed = num_failed + !(omp_get_num_threads() == 2);
36 } // end 3rd level parallel
37 } // end 2nd level parallel
38 } // end 1st level parallel
39
40 // Make sure that basic single element num_threads clause works
41 #pragma omp parallel reduction(+ : num_failed) num_threads(4) // 1st level
42 {
43 #pragma omp single
44 num_failed = num_failed + !(omp_get_num_threads() == 4);
45 #pragma omp parallel reduction(+ : num_failed) // 2nd level
46 {
47 #pragma omp single
48 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
49 #pragma omp parallel reduction(+ : num_failed) // 3rd level
50 {
51 #pragma omp single
52 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
53 } // end 3rd level parallel
54 } // end 2nd level parallel
55 } // end 1st level parallel
56
57 // Check that basic single element num_threads clause works on second level
58 #pragma omp parallel reduction(+ : num_failed) // 1st level
59 {
60 #pragma omp single
61 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
62 #pragma omp parallel reduction(+ : num_failed) num_threads(4) // 2nd level
63 {
64 #pragma omp single
65 num_failed = num_failed + !(omp_get_num_threads() == 4);
66 #pragma omp parallel reduction(+ : num_failed) // 3rd level
67 {
68 #pragma omp single
69 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
70 } // end 3rd level parallel
71 } // end 2nd level parallel
72 } // end 1st level parallel
73
74 // Try a short list. It should completely overwrite the old settings.
75 // We need to use the compiler interface for now.
76 int threads[2] = {3, 3};
77 __kmpc_push_num_threads_list(NULL, __kmpc_global_thread_num(NULL), 2,
78 threads);
79 #pragma omp parallel reduction(+ : num_failed) // num_threads(3,3) // 1st level
80 {
81 #pragma omp single
82 num_failed = num_failed + !(omp_get_num_threads() == 3);
83 #pragma omp parallel reduction(+ : num_failed) // 2nd level
84 {
85 #pragma omp single
86 num_failed = num_failed + !(omp_get_num_threads() == 3);
87 #pragma omp parallel reduction(+ : num_failed) // 3rd level
88 {
89 // NOTE: should just keep using last element in list, to nesting depth
90 #pragma omp single
91 num_failed = num_failed + !(omp_get_num_threads() == 3);
92 } // end 3rd level parallel
93 } // end 2nd level parallel
94 } // end 1st level parallel
95
96 // Similar, but at a lower level.
97 #pragma omp parallel reduction(+ : num_failed) // 1st level
98 {
99 #pragma omp single
100 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
101 int threads[2] = {3, 3};
102 __kmpc_push_num_threads_list(NULL, __kmpc_global_thread_num(NULL), 2,
103 threads);
104 #pragma omp parallel reduction(+ : num_failed) // num_threads(3,3) // 2nd level
105 {
106 #pragma omp single
107 num_failed = num_failed + !(omp_get_num_threads() == 3);
108 #pragma omp parallel reduction(+ : num_failed) // 3rd level
109 {
110 // NOTE: just keep using last element in list, to nesting depth
111 #pragma omp single
112 num_failed = num_failed + !(omp_get_num_threads() == 3);
113 } // end 3rd level parallel
114 } // end 2nd level parallel
115 // Make sure a second inner parallel is NOT affected by the clause
116 #pragma omp parallel reduction(+ : num_failed) // 2nd level
117 {
118 #pragma omp single
119 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
120 #pragma omp parallel reduction(+ : num_failed) // 3rd level
121 {
122 #pragma omp single
123 // NOTE: just keep using last element in list, to nesting depth
124 num_failed = num_failed + !(omp_get_num_threads() == 2); // Unaffected
125 } // end 3rd level parallel
126 } // end 2nd level parallel
127 } // end 1st level parallel
128
129 // Test lists at multiple levels
130 int threads2[2] = {3,2};
131 __kmpc_push_num_threads_list(NULL, __kmpc_global_thread_num(NULL), 2,
132 threads2);
133 #pragma omp parallel reduction(+ : num_failed) // num_threads(3,2) // 1st level
134 {
135 #pragma omp single
136 num_failed = num_failed + !(omp_get_num_threads() == 3);
137 #pragma omp parallel reduction(+ : num_failed) // 2nd level
138 {
139 #pragma omp single
140 num_failed = num_failed + !(omp_get_num_threads() == 2);
141 #pragma omp parallel reduction(+ : num_failed) // 3rd level
142 {
143 #pragma omp single
144 num_failed = num_failed + !(omp_get_num_threads() == 2);
145 int threads3[2] = {3,1};
146 __kmpc_push_num_threads_list(NULL, __kmpc_global_thread_num(NULL), 2,
147 threads3);
148 #pragma omp parallel reduction(+ : num_failed) // num_threads(3,1) // 4th level
149 {
150 #pragma omp single
151 num_failed = num_failed + !(omp_get_num_threads() == 3);
152 #pragma omp parallel reduction(+ : num_failed) // 5th level
153 {
154 #pragma omp single
155 num_failed = num_failed + !(omp_get_num_threads() == 1);
156 #pragma omp parallel reduction(+ : num_failed) // 6th level
157 {
158 #pragma omp single
159 num_failed = num_failed + !(omp_get_num_threads() == 1);
160 } // end 6th level parallel
161 } // end 5th level parallel
162 } // end 4th level parallel
163 #pragma omp parallel reduction(+ : num_failed) // 4th level
164 {
165 #pragma omp single
166 num_failed = num_failed + !(omp_get_num_threads() == 2);
167 } // end 4th level parallel
168 } // end 3rd level parallel
169 } // end 2nd level parallel
170 #pragma omp parallel reduction(+ : num_failed) // 2nd level
171 {
172 #pragma omp single
173 num_failed = num_failed + !(omp_get_num_threads() == 2);
174 #pragma omp parallel reduction(+ : num_failed) // 3rd level
175 {
176 #pragma omp single
177 num_failed = num_failed + !(omp_get_num_threads() == 2);
178 } // end 3rd level parallel
179 } // end 2nd level parallel
180 } // end 1st level parallel
181
182 // Now we should be back to the way we started.
183 #pragma omp parallel reduction(+ : num_failed) // 1st level
184 {
185 #pragma omp single
186 num_failed = num_failed + !(omp_get_num_threads() == 2);
187 #pragma omp parallel reduction(+ : num_failed) // 2nd level
188 {
189 #pragma omp single
190 num_failed = num_failed + !(omp_get_num_threads() == 2);
191 #pragma omp parallel reduction(+ : num_failed) // 3rd level
192 {
193 #pragma omp single
194 num_failed = num_failed + !(omp_get_num_threads() == 2);
195 } // end 3rd level parallel
196 } // end 2nd level parallel
197 } // end 1st level parallel
198
199 return (!num_failed);
200 }
201
main()202 int main() {
203 int i;
204 int num_failed = 0;
205
206 for (i = 0; i < REPETITIONS; i++) {
207 if (!test_omp_parallel_num_threads_list()) {
208 num_failed++;
209 }
210 }
211 return num_failed;
212 }
213