1 | // RUN: %libomp-compile-and-run | FileCheck %s |
2 | // REQUIRES: ompt |
3 | |
4 | // The runtime currently does not get dependency information from GCC. |
5 | // UNSUPPORTED: gcc |
6 | |
7 | // Tests OMP 5.x task dependence "omp_all_memory", |
8 | // emulates compiler codegen versions for new dep kind |
9 | // |
10 | // Task tree created: |
11 | // task0 - task1 (in: i1, i2) |
12 | // \ |
13 | // task2 (inoutset: i2), (in: i1) |
14 | // / |
15 | // task3 (omp_all_memory) via flag=0x80 |
16 | // / |
17 | // task4 - task5 (in: i1, i2) |
18 | // / |
19 | // task6 (omp_all_memory) via addr=-1 |
20 | // / |
21 | // task7 (omp_all_memory) via flag=0x80 |
22 | // / |
23 | // task8 (in: i3) |
24 | // |
25 | |
26 | // CHECK: ompt_event_dependences: |
27 | // CHECK-SAME: ompt_dependence_type_in |
28 | // CHECK-SAME: ompt_dependence_type_in |
29 | // CHECK-SAME: ndeps=2 |
30 | |
31 | // CHECK: ompt_event_dependences: |
32 | // CHECK-SAME: ompt_dependence_type_in |
33 | // CHECK-SAME: ompt_dependence_type_in |
34 | // CHECK-SAME: ndeps=2 |
35 | |
36 | // CHECK: ompt_event_dependences: |
37 | // CHECK-SAME: ompt_dependence_type_in |
38 | // CHECK-SAME: ompt_dependence_type_inoutset |
39 | // CHECK-SAME: ndeps=2 |
40 | |
41 | // CHECK: ompt_event_dependences: |
42 | // CHECK-SAME: ompt_dependence_type_in |
43 | // CHECK-SAME: ompt_dependence_type_out_all_memory |
44 | // CHECK-SAME: ndeps=2 |
45 | |
46 | // CHECK: ompt_event_dependences: |
47 | // CHECK-SAME: ompt_dependence_type_in |
48 | // CHECK-SAME: ompt_dependence_type_in |
49 | // CHECK-SAME: ndeps=2 |
50 | |
51 | // CHECK: ompt_event_dependences: |
52 | // CHECK-SAME: ompt_dependence_type_in |
53 | // CHECK-SAME: ompt_dependence_type_in |
54 | // CHECK-SAME: ndeps=2 |
55 | |
56 | // CHECK: ompt_event_dependences: |
57 | // CHECK-SAME: ompt_dependence_type_out_all_memory |
58 | // CHECK-SAME: ndeps=1 |
59 | |
60 | // CHECK: ompt_event_dependences: |
61 | // CHECK-SAME: ompt_dependence_type_out_all_memory |
62 | // CHECK-SAME: ompt_dependence_type_mutexinoutset |
63 | // CHECK-SAME: ndeps=2 |
64 | |
65 | // CHECK: ompt_event_dependences: |
66 | // CHECK-SAME: ompt_dependence_type_in |
67 | // CHECK-SAME: ndeps=1 |
68 | |
69 | #include "callback.h" |
70 | #include <stdio.h> |
71 | #include <omp.h> |
72 | |
73 | #ifdef _WIN32 |
74 | #include <windows.h> |
75 | #define mysleep(n) Sleep(n) |
76 | #else |
77 | #include <unistd.h> |
78 | #define mysleep(n) usleep((n)*1000) |
79 | #endif |
80 | |
81 | // to check the # of concurrent tasks (must be 1 for MTX, <3 for other kinds) |
82 | static int checker = 0; |
83 | static int err = 0; |
84 | #ifndef DELAY |
85 | #define DELAY 100 |
86 | #endif |
87 | |
88 | // --------------------------------------------------------------------------- |
89 | // internal data to emulate compiler codegen |
90 | typedef struct DEP { |
91 | size_t addr; |
92 | size_t len; |
93 | unsigned char flags; |
94 | } dep; |
95 | #define DEP_ALL_MEM 0x80 |
96 | typedef struct task { |
97 | void **shareds; |
98 | void *entry; |
99 | int part_id; |
100 | void *destr_thunk; |
101 | int priority; |
102 | long long device_id; |
103 | int f_priv; |
104 | } task_t; |
105 | #define TIED 1 |
106 | typedef int (*entry_t)(int, task_t *); |
107 | typedef struct ID { |
108 | int reserved_1; |
109 | int flags; |
110 | int reserved_2; |
111 | int reserved_3; |
112 | char *psource; |
113 | } id; |
114 | // thunk routine for tasks with ALL dependency |
115 | int thunk_m(int gtid, task_t *ptask) { |
116 | int lcheck, th; |
117 | #pragma omp atomic capture |
118 | lcheck = ++checker; |
119 | th = omp_get_thread_num(); |
120 | printf(format: "task m_%d, th %d, checker %d\n" , ptask->f_priv, th, lcheck); |
121 | if (lcheck != 1) { // no more than 1 task at a time |
122 | err++; |
123 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
124 | } |
125 | mysleep(DELAY); |
126 | #pragma omp atomic read |
127 | lcheck = checker; // must still be equal to 1 |
128 | if (lcheck != 1) { |
129 | err++; |
130 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
131 | } |
132 | #pragma omp atomic |
133 | --checker; |
134 | return 0; |
135 | } |
136 | // thunk routine for tasks with inoutset dependency |
137 | int thunk_s(int gtid, task_t *ptask) { |
138 | int lcheck, th; |
139 | #pragma omp atomic capture |
140 | lcheck = ++checker; // 1 |
141 | th = omp_get_thread_num(); |
142 | printf(format: "task 2_%d, th %d, checker %d\n" , ptask->f_priv, th, lcheck); |
143 | if (lcheck != 1) { // no more than 1 task at a time |
144 | err++; |
145 | printf(format: "Error s1, checker %d != 1\n" , lcheck); |
146 | } |
147 | mysleep(DELAY); |
148 | #pragma omp atomic read |
149 | lcheck = checker; // must still be equal to 1 |
150 | if (lcheck != 1) { |
151 | err++; |
152 | printf(format: "Error s2, checker %d != 1\n" , lcheck); |
153 | } |
154 | #pragma omp atomic |
155 | --checker; |
156 | return 0; |
157 | } |
158 | |
159 | #ifdef __cplusplus |
160 | extern "C" { |
161 | #endif |
162 | int __kmpc_global_thread_num(id *); |
163 | task_t *__kmpc_omp_task_alloc(id *loc, int gtid, int flags, size_t sz, |
164 | size_t shar, entry_t rtn); |
165 | int __kmpc_omp_task_with_deps(id *loc, int gtid, task_t *task, int ndeps, |
166 | dep *dep_lst, int nd_noalias, dep *noalias_lst); |
167 | static id loc = {.reserved_1: 0, .flags: 2, .reserved_2: 0, .reserved_3: 0, .psource: ";file;func;0;0;;" }; |
168 | #ifdef __cplusplus |
169 | } // extern "C" |
170 | #endif |
171 | // End of internal data |
172 | // --------------------------------------------------------------------------- |
173 | |
174 | int main() { |
175 | char *ompx_all_memory = (void *)0xffffffffffffffff; |
176 | int i1, i2, i3; |
177 | omp_set_num_threads(8); |
178 | omp_set_dynamic(0); |
179 | #pragma omp parallel |
180 | { |
181 | #pragma omp single nowait |
182 | { |
183 | dep sdep[2]; |
184 | task_t *ptr; |
185 | int gtid = __kmpc_global_thread_num(&loc); |
186 | int t = omp_get_thread_num(); |
187 | #pragma omp task depend(in : i1, i2) |
188 | { // task 0 |
189 | int lcheck, th; |
190 | #pragma omp atomic capture |
191 | lcheck = ++checker; // 1 or 2 |
192 | th = omp_get_thread_num(); |
193 | printf(format: "task 0_%d, th %d, checker %d\n" , t, th, lcheck); |
194 | if (lcheck > 2 || lcheck < 1) { |
195 | err++; // no more than 2 tasks concurrently |
196 | printf(format: "Error1, checker %d, not 1 or 2\n" , lcheck); |
197 | } |
198 | mysleep(DELAY); |
199 | #pragma omp atomic read |
200 | lcheck = checker; // 1 or 2 |
201 | if (lcheck > 2 || lcheck < 1) { |
202 | #pragma omp atomic |
203 | err++; |
204 | printf(format: "Error2, checker %d, not 1 or 2\n" , lcheck); |
205 | } |
206 | #pragma omp atomic |
207 | --checker; |
208 | } |
209 | #pragma omp task depend(in : i1, i2) |
210 | { // task 1 |
211 | int lcheck, th; |
212 | #pragma omp atomic capture |
213 | lcheck = ++checker; // 1 or 2 |
214 | th = omp_get_thread_num(); |
215 | printf(format: "task 1_%d, th %d, checker %d\n" , t, th, lcheck); |
216 | if (lcheck > 2 || lcheck < 1) { |
217 | err++; // no more than 2 tasks concurrently |
218 | printf(format: "Error3, checker %d, not 1 or 2\n" , lcheck); |
219 | } |
220 | mysleep(DELAY); |
221 | #pragma omp atomic read |
222 | lcheck = checker; // 1 or 2 |
223 | if (lcheck > 2 || lcheck < 1) { |
224 | err++; |
225 | printf(format: "Error4, checker %d, not 1 or 2\n" , lcheck); |
226 | } |
227 | #pragma omp atomic |
228 | --checker; |
229 | } |
230 | // compiler codegen start |
231 | // task2 |
232 | ptr = __kmpc_omp_task_alloc(loc: &loc, gtid, TIED, sz: sizeof(task_t), shar: 0, rtn: thunk_s); |
233 | sdep[0].addr = (size_t)&i1; |
234 | sdep[0].len = 0; // not used |
235 | sdep[0].flags = 1; // IN |
236 | sdep[1].addr = (size_t)&i2; |
237 | sdep[1].len = 0; // not used |
238 | sdep[1].flags = 8; // INOUTSET |
239 | ptr->f_priv = t + 10; // init single first-private variable |
240 | __kmpc_omp_task_with_deps(loc: &loc, gtid, task: ptr, ndeps: 2, dep_lst: sdep, nd_noalias: 0, noalias_lst: 0); |
241 | |
242 | // task3 |
243 | #pragma omp task depend(in : i1) depend(inout : ompx_all_memory[0]) |
244 | { |
245 | int lcheck, th; |
246 | #pragma omp atomic capture |
247 | lcheck = ++checker; |
248 | th = omp_get_thread_num(); |
249 | printf(format: "task m_%d, th %d, checker %d\n" , t, th, lcheck); |
250 | if (lcheck != 1) { // no more than 1 task at a time |
251 | err++; |
252 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
253 | } |
254 | mysleep(DELAY); |
255 | #pragma omp atomic read |
256 | lcheck = checker; // must still be equal to 1 |
257 | if (lcheck != 1) { |
258 | err++; |
259 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
260 | } |
261 | #pragma omp atomic |
262 | --checker; |
263 | } |
264 | // compiler codegen end |
265 | #pragma omp task depend(in : i1, i2) |
266 | { // task 4 |
267 | int lcheck, th; |
268 | #pragma omp atomic capture |
269 | lcheck = ++checker; // 1 or 2 |
270 | th = omp_get_thread_num(); |
271 | printf(format: "task 4_%d, th %d, checker %d\n" , t, th, lcheck); |
272 | if (lcheck > 2 || lcheck < 1) { |
273 | err++; // no more than 2 tasks concurrently |
274 | printf(format: "Error5, checker %d, not 1 or 2\n" , lcheck); |
275 | } |
276 | mysleep(DELAY); |
277 | #pragma omp atomic read |
278 | lcheck = checker; // 1 or 2 |
279 | if (lcheck > 2 || lcheck < 1) { |
280 | err++; |
281 | printf(format: "Error6, checker %d, not 1 or 2\n" , lcheck); |
282 | } |
283 | #pragma omp atomic |
284 | --checker; |
285 | } |
286 | #pragma omp task depend(in : i1, i2) |
287 | { // task 5 |
288 | int lcheck, th; |
289 | #pragma omp atomic capture |
290 | lcheck = ++checker; // 1 or 2 |
291 | th = omp_get_thread_num(); |
292 | printf(format: "task 5_%d, th %d, checker %d\n" , t, th, lcheck); |
293 | if (lcheck > 2 || lcheck < 1) { |
294 | err++; // no more than 2 tasks concurrently |
295 | printf(format: "Error7, checker %d, not 1 or 2\n" , lcheck); |
296 | } |
297 | mysleep(DELAY); |
298 | #pragma omp atomic read |
299 | lcheck = checker; // 1 or 2 |
300 | if (lcheck > 2 || lcheck < 1) { |
301 | err++; |
302 | printf(format: "Error8, checker %d, not 1 or 2\n" , lcheck); |
303 | } |
304 | #pragma omp atomic |
305 | --checker; |
306 | } |
307 | // task6 |
308 | #pragma omp task depend(inout : ompx_all_memory[0]) |
309 | { |
310 | int lcheck, th; |
311 | #pragma omp atomic capture |
312 | lcheck = ++checker; |
313 | th = omp_get_thread_num(); |
314 | printf(format: "task m_%d, th %d, checker %d\n" , t, th, lcheck); |
315 | if (lcheck != 1) { // no more than 1 task at a time |
316 | err++; |
317 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
318 | } |
319 | mysleep(DELAY); |
320 | #pragma omp atomic read |
321 | lcheck = checker; // must still be equal to 1 |
322 | if (lcheck != 1) { |
323 | err++; |
324 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
325 | } |
326 | #pragma omp atomic |
327 | --checker; |
328 | } |
329 | // task7 |
330 | #pragma omp task depend(inout : ompx_all_memory[0]) depend(mutexinoutset : i3) |
331 | { |
332 | int lcheck, th; |
333 | #pragma omp atomic capture |
334 | lcheck = ++checker; |
335 | th = omp_get_thread_num(); |
336 | printf(format: "task m_%d, th %d, checker %d\n" , t, th, lcheck); |
337 | if (lcheck != 1) { // no more than 1 task at a time |
338 | err++; |
339 | printf(format: "Error m1, checker %d != 1\n" , lcheck); |
340 | } |
341 | mysleep(DELAY); |
342 | #pragma omp atomic read |
343 | lcheck = checker; // must still be equal to 1 |
344 | if (lcheck != 1) { |
345 | err++; |
346 | printf(format: "Error m2, checker %d != 1\n" , lcheck); |
347 | } |
348 | #pragma omp atomic |
349 | --checker; |
350 | } |
351 | #pragma omp task depend(in : i3) |
352 | { // task 8 |
353 | int lcheck, th; |
354 | #pragma omp atomic capture |
355 | lcheck = ++checker; // 1 |
356 | th = omp_get_thread_num(); |
357 | printf(format: "task 8_%d, th %d, checker %d\n" , t, th, lcheck); |
358 | if (lcheck != 1) { |
359 | err++; |
360 | printf(format: "Error9, checker %d, != 1\n" , lcheck); |
361 | } |
362 | mysleep(DELAY); |
363 | #pragma omp atomic read |
364 | lcheck = checker; |
365 | if (lcheck != 1) { |
366 | err++; |
367 | printf(format: "Error10, checker %d, != 1\n" , lcheck); |
368 | } |
369 | #pragma omp atomic |
370 | --checker; |
371 | } |
372 | } // single |
373 | } // parallel |
374 | if (err == 0 && checker == 0) { |
375 | printf(format: "passed\n" ); |
376 | return 0; |
377 | } else { |
378 | printf(format: "failed, err = %d, checker = %d\n" , err, checker); |
379 | return 1; |
380 | } |
381 | } |
382 | |