1 | /* |
2 | Copyright (c) 2005-2019 Intel Corporation |
3 | |
4 | Licensed under the Apache License, Version 2.0 (the "License"); |
5 | you may not use this file except in compliance with the License. |
6 | You may obtain a copy of the License at |
7 | |
8 | http://www.apache.org/licenses/LICENSE-2.0 |
9 | |
10 | Unless required by applicable law or agreed to in writing, software |
11 | distributed under the License is distributed on an "AS IS" BASIS, |
12 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
13 | See the License for the specific language governing permissions and |
14 | limitations under the License. |
15 | */ |
16 | |
17 | #include "harness.h" |
18 | |
19 | #if __TBB_CPF_BUILD |
20 | #define TBB_DEPRECATED_FLOW_NODE_EXTRACTION 1 |
21 | #endif |
22 | |
23 | #include "harness_graph.h" |
24 | |
25 | #include "tbb/task_scheduler_init.h" |
26 | #include "tbb/tick_count.h" |
27 | |
28 | #define N 1000 |
29 | #define C 10 |
30 | |
31 | template< typename T > |
32 | void spin_try_get( tbb::flow::buffer_node<T> &b, T &value ) { |
33 | while ( b.try_get(value) != true ) {} |
34 | } |
35 | |
36 | template< typename T > |
37 | void check_item( T* count_value, T &value ) { |
38 | count_value[value / N] += value % N; |
39 | } |
40 | |
41 | template< typename T > |
42 | struct parallel_puts : NoAssign { |
43 | |
44 | tbb::flow::buffer_node<T> &my_b; |
45 | |
46 | parallel_puts( tbb::flow::buffer_node<T> &b ) : my_b(b) {} |
47 | |
48 | void operator()(int i) const { |
49 | for (int j = 0; j < N; ++j) { |
50 | bool msg = my_b.try_put( T(N*i + j) ); |
51 | ASSERT( msg == true, NULL ); |
52 | } |
53 | } |
54 | }; |
55 | |
56 | template< typename T > |
57 | struct touches { |
58 | |
59 | bool **my_touches; |
60 | int my_num_threads; |
61 | |
62 | touches( int num_threads ) : my_num_threads(num_threads) { |
63 | my_touches = new bool* [my_num_threads]; |
64 | for ( int p = 0; p < my_num_threads; ++p) { |
65 | my_touches[p] = new bool[N]; |
66 | for ( int n = 0; n < N; ++n) |
67 | my_touches[p][n] = false; |
68 | } |
69 | } |
70 | |
71 | ~touches() { |
72 | for ( int p = 0; p < my_num_threads; ++p) { |
73 | delete [] my_touches[p]; |
74 | } |
75 | delete [] my_touches; |
76 | } |
77 | |
78 | bool check( T v ) { |
79 | ASSERT ( my_touches[v/N][v%N] == false, NULL); |
80 | my_touches[v/N][v%N] = true; |
81 | return true; |
82 | } |
83 | |
84 | bool validate_touches() { |
85 | for ( int p = 0; p < my_num_threads; ++p) { |
86 | for ( int n = 0; n < N; ++n) { |
87 | ASSERT ( my_touches[p][n] == true, NULL); |
88 | } |
89 | } |
90 | return true; |
91 | } |
92 | }; |
93 | |
94 | template< typename T > |
95 | struct parallel_gets : NoAssign { |
96 | |
97 | tbb::flow::buffer_node<T> &my_b; |
98 | touches<T> &my_touches; |
99 | |
100 | parallel_gets( tbb::flow::buffer_node<T> &b, touches<T> &t) : my_b(b), my_touches(t) {} |
101 | |
102 | void operator()(int) const { |
103 | for (int j = 0; j < N; ++j) { |
104 | T v; |
105 | spin_try_get( my_b, v ); |
106 | my_touches.check( v ); |
107 | } |
108 | } |
109 | |
110 | }; |
111 | |
112 | template< typename T > |
113 | struct parallel_put_get : NoAssign { |
114 | |
115 | tbb::flow::buffer_node<T> &my_b; |
116 | touches<T> &my_touches; |
117 | |
118 | parallel_put_get( tbb::flow::buffer_node<T> &b, touches<T> &t ) : my_b(b), my_touches(t) {} |
119 | |
120 | void operator()(int tid) const { |
121 | |
122 | for ( int i = 0; i < N; i+=C ) { |
123 | int j_end = ( N < i + C ) ? N : i + C; |
124 | // dump about C values into the buffer |
125 | for ( int j = i; j < j_end; ++j ) { |
126 | ASSERT( my_b.try_put( T (N*tid + j ) ) == true, NULL ); |
127 | } |
128 | // receiver about C values from the buffer |
129 | for ( int j = i; j < j_end; ++j ) { |
130 | T v; |
131 | spin_try_get( my_b, v ); |
132 | my_touches.check( v ); |
133 | } |
134 | } |
135 | } |
136 | |
137 | }; |
138 | |
139 | // |
140 | // Tests |
141 | // |
142 | // Item can be reserved, released, consumed ( single serial receiver ) |
143 | // |
144 | template< typename T > |
145 | int test_reservation() { |
146 | tbb::flow::graph g; |
147 | T bogus_value(-1); |
148 | |
149 | // Simple tests |
150 | tbb::flow::buffer_node<T> b(g); |
151 | |
152 | b.try_put(T(1)); |
153 | b.try_put(T(2)); |
154 | b.try_put(T(3)); |
155 | |
156 | T v, vsum; |
157 | ASSERT( b.try_reserve(v) == true, NULL ); |
158 | ASSERT( b.try_release() == true, NULL ); |
159 | v = bogus_value; |
160 | g.wait_for_all(); |
161 | ASSERT( b.try_reserve(v) == true, NULL ); |
162 | ASSERT( b.try_consume() == true, NULL ); |
163 | vsum += v; |
164 | v = bogus_value; |
165 | g.wait_for_all(); |
166 | |
167 | ASSERT( b.try_get(v) == true, NULL ); |
168 | vsum += v; |
169 | v = bogus_value; |
170 | g.wait_for_all(); |
171 | |
172 | ASSERT( b.try_reserve(v) == true, NULL ); |
173 | ASSERT( b.try_release() == true, NULL ); |
174 | v = bogus_value; |
175 | g.wait_for_all(); |
176 | ASSERT( b.try_reserve(v) == true, NULL ); |
177 | ASSERT( b.try_consume() == true, NULL ); |
178 | vsum += v; |
179 | ASSERT( vsum == T(6), NULL); |
180 | v = bogus_value; |
181 | g.wait_for_all(); |
182 | |
183 | return 0; |
184 | } |
185 | |
186 | // |
187 | // Tests |
188 | // |
189 | // multiple parallel senders, items in arbitrary order |
190 | // multiple parallel senders, multiple parallel receivers, items in arbitrary order and all items received |
191 | // * overlapped puts / gets |
192 | // * all puts finished before any getS |
193 | // |
194 | template< typename T > |
195 | int test_parallel(int num_threads) { |
196 | tbb::flow::graph g; |
197 | tbb::flow::buffer_node<T> b(g); |
198 | tbb::flow::buffer_node<T> b2(g); |
199 | tbb::flow::buffer_node<T> b3(g); |
200 | T bogus_value(-1); |
201 | T j = bogus_value; |
202 | |
203 | NativeParallelFor( num_threads, parallel_puts<T>(b) ); |
204 | |
205 | T *next_value = new T[num_threads]; |
206 | for (int tid = 0; tid < num_threads; ++tid) next_value[tid] = T(0); |
207 | |
208 | for (int i = 0; i < num_threads * N; ++i ) { |
209 | spin_try_get( b, j ); |
210 | check_item( next_value, j ); |
211 | j = bogus_value; |
212 | } |
213 | for (int tid = 0; tid < num_threads; ++tid) { |
214 | ASSERT( next_value[tid] == T((N*(N-1))/2), NULL ); |
215 | } |
216 | |
217 | j = bogus_value; |
218 | g.wait_for_all(); |
219 | ASSERT( b.try_get( j ) == false, NULL ); |
220 | ASSERT( j == bogus_value, NULL ); |
221 | |
222 | NativeParallelFor( num_threads, parallel_puts<T>(b) ); |
223 | |
224 | { |
225 | touches< T > t( num_threads ); |
226 | NativeParallelFor( num_threads, parallel_gets<T>(b, t) ); |
227 | g.wait_for_all(); |
228 | ASSERT( t.validate_touches(), NULL ); |
229 | } |
230 | j = bogus_value; |
231 | ASSERT( b.try_get( j ) == false, NULL ); |
232 | ASSERT( j == bogus_value, NULL ); |
233 | |
234 | g.wait_for_all(); |
235 | { |
236 | touches< T > t( num_threads ); |
237 | NativeParallelFor( num_threads, parallel_put_get<T>(b, t) ); |
238 | g.wait_for_all(); |
239 | ASSERT( t.validate_touches(), NULL ); |
240 | } |
241 | j = bogus_value; |
242 | ASSERT( b.try_get( j ) == false, NULL ); |
243 | ASSERT( j == bogus_value, NULL ); |
244 | |
245 | tbb::flow::make_edge( b, b2 ); |
246 | tbb::flow::make_edge( b2, b3 ); |
247 | |
248 | NativeParallelFor( num_threads, parallel_puts<T>(b) ); |
249 | { |
250 | touches< T > t( num_threads ); |
251 | NativeParallelFor( num_threads, parallel_gets<T>(b3, t) ); |
252 | g.wait_for_all(); |
253 | ASSERT( t.validate_touches(), NULL ); |
254 | } |
255 | j = bogus_value; |
256 | g.wait_for_all(); |
257 | ASSERT( b.try_get( j ) == false, NULL ); |
258 | g.wait_for_all(); |
259 | ASSERT( b2.try_get( j ) == false, NULL ); |
260 | g.wait_for_all(); |
261 | ASSERT( b3.try_get( j ) == false, NULL ); |
262 | ASSERT( j == bogus_value, NULL ); |
263 | |
264 | // test copy constructor |
265 | ASSERT( b.remove_successor( b2 ), NULL ); |
266 | // fill up b: |
267 | NativeParallelFor( num_threads, parallel_puts<T>(b) ); |
268 | // copy b: |
269 | tbb::flow::buffer_node<T> b_copy(b); |
270 | |
271 | // b_copy should be empty |
272 | j = bogus_value; |
273 | g.wait_for_all(); |
274 | ASSERT( b_copy.try_get( j ) == false, NULL ); |
275 | |
276 | // hook them together: |
277 | ASSERT( b.register_successor(b_copy) == true, NULL ); |
278 | // try to get content from b_copy |
279 | { |
280 | touches< T > t( num_threads ); |
281 | NativeParallelFor( num_threads, parallel_gets<T>(b_copy, t) ); |
282 | g.wait_for_all(); |
283 | ASSERT( t.validate_touches(), NULL ); |
284 | } |
285 | // now both should be empty |
286 | j = bogus_value; |
287 | g.wait_for_all(); |
288 | ASSERT( b.try_get( j ) == false, NULL ); |
289 | g.wait_for_all(); |
290 | ASSERT( b_copy.try_get( j ) == false, NULL ); |
291 | ASSERT( j == bogus_value, NULL ); |
292 | |
293 | delete [] next_value; |
294 | return 0; |
295 | } |
296 | |
297 | // |
298 | // Tests |
299 | // |
300 | // Predecessors cannot be registered |
301 | // Empty buffer rejects item requests |
302 | // Single serial sender, items in arbitrary order |
303 | // Chained buffers ( 2 & 3 ), single sender, items at last buffer in arbitrary order |
304 | // |
305 | |
306 | template< typename T > |
307 | int test_serial() { |
308 | tbb::flow::graph g; |
309 | T bogus_value(-1); |
310 | |
311 | tbb::flow::buffer_node<T> b(g); |
312 | tbb::flow::buffer_node<T> b2(g); |
313 | T j = bogus_value; |
314 | |
315 | // |
316 | // Rejects attempts to add / remove predecessor |
317 | // Rejects request from empty buffer |
318 | // |
319 | ASSERT( b.register_predecessor( b2 ) == false, NULL ); |
320 | ASSERT( b.remove_predecessor( b2 ) == false, NULL ); |
321 | ASSERT( b.try_get( j ) == false, NULL ); |
322 | ASSERT( j == bogus_value, NULL ); |
323 | |
324 | // |
325 | // Simple puts and gets |
326 | // |
327 | |
328 | for (int i = 0; i < N; ++i) { |
329 | bool msg = b.try_put( T(i) ); |
330 | ASSERT( msg == true, NULL ); |
331 | } |
332 | |
333 | T vsum = T(0); |
334 | for (int i = 0; i < N; ++i) { |
335 | j = bogus_value; |
336 | spin_try_get( b, j ); |
337 | vsum += j; |
338 | } |
339 | ASSERT( vsum == (N*(N-1))/2, NULL); |
340 | j = bogus_value; |
341 | g.wait_for_all(); |
342 | ASSERT( b.try_get( j ) == false, NULL ); |
343 | ASSERT( j == bogus_value, NULL ); |
344 | |
345 | tbb::flow::make_edge(b, b2); |
346 | #if TBB_DEPRECATED_FLOW_NODE_EXTRACTION |
347 | ASSERT( b.successor_count() == 1, NULL); |
348 | ASSERT( b.predecessor_count() == 0, NULL); |
349 | ASSERT( b2.successor_count() == 0, NULL); |
350 | ASSERT( b2.predecessor_count() == 1, NULL); |
351 | typename tbb::flow::buffer_node<T>::successor_list_type my_succs; |
352 | b.copy_successors(my_succs); |
353 | ASSERT(my_succs.size() == 1, NULL); |
354 | typename tbb::flow::buffer_node<T>::predecessor_list_type my_preds; |
355 | b.copy_predecessors(my_preds); |
356 | ASSERT(my_preds.size() == 0, NULL); |
357 | #endif |
358 | |
359 | vsum = T(0); |
360 | for (int i = 0; i < N; ++i) { |
361 | bool msg = b.try_put( T(i) ); |
362 | ASSERT( msg == true, NULL ); |
363 | } |
364 | |
365 | for (int i = 0; i < N; ++i) { |
366 | j = bogus_value; |
367 | spin_try_get( b2, j ); |
368 | vsum += j; |
369 | } |
370 | ASSERT( vsum == (N*(N-1))/2, NULL); |
371 | j = bogus_value; |
372 | g.wait_for_all(); |
373 | ASSERT( b.try_get( j ) == false, NULL ); |
374 | g.wait_for_all(); |
375 | ASSERT( b2.try_get( j ) == false, NULL ); |
376 | ASSERT( j == bogus_value, NULL ); |
377 | |
378 | tbb::flow::remove_edge(b, b2); |
379 | ASSERT( b.try_put( 1 ) == true, NULL ); |
380 | g.wait_for_all(); |
381 | ASSERT( b2.try_get( j ) == false, NULL ); |
382 | ASSERT( j == bogus_value, NULL ); |
383 | g.wait_for_all(); |
384 | ASSERT( b.try_get( j ) == true, NULL ); |
385 | ASSERT( j == 1, NULL ); |
386 | |
387 | tbb::flow::buffer_node<T> b3(g); |
388 | tbb::flow::make_edge( b, b2 ); |
389 | tbb::flow::make_edge( b2, b3 ); |
390 | |
391 | vsum = T(0); |
392 | for (int i = 0; i < N; ++i) { |
393 | bool msg = b.try_put( T(i) ); |
394 | ASSERT( msg == true, NULL ); |
395 | } |
396 | |
397 | for (int i = 0; i < N; ++i) { |
398 | j = bogus_value; |
399 | spin_try_get( b3, j ); |
400 | vsum += j; |
401 | } |
402 | ASSERT( vsum == (N*(N-1))/2, NULL); |
403 | j = bogus_value; |
404 | g.wait_for_all(); |
405 | ASSERT( b.try_get( j ) == false, NULL ); |
406 | g.wait_for_all(); |
407 | ASSERT( b2.try_get( j ) == false, NULL ); |
408 | g.wait_for_all(); |
409 | ASSERT( b3.try_get( j ) == false, NULL ); |
410 | ASSERT( j == bogus_value, NULL ); |
411 | |
412 | tbb::flow::remove_edge(b, b2); |
413 | ASSERT( b.try_put( 1 ) == true, NULL ); |
414 | g.wait_for_all(); |
415 | ASSERT( b2.try_get( j ) == false, NULL ); |
416 | ASSERT( j == bogus_value, NULL ); |
417 | g.wait_for_all(); |
418 | ASSERT( b3.try_get( j ) == false, NULL ); |
419 | ASSERT( j == bogus_value, NULL ); |
420 | g.wait_for_all(); |
421 | ASSERT( b.try_get( j ) == true, NULL ); |
422 | ASSERT( j == 1, NULL ); |
423 | |
424 | return 0; |
425 | } |
426 | |
427 | int TestMain() { |
428 | tbb::tick_count start = tbb::tick_count::now(), stop; |
429 | for (int p = 2; p <= 4; ++p) { |
430 | tbb::task_scheduler_init init(p); |
431 | test_serial<int>(); |
432 | test_parallel<int>(p); |
433 | } |
434 | stop = tbb::tick_count::now(); |
435 | REMARK("Buffer_Node Time=%6.6f\n" , (stop-start).seconds()); |
436 | test_resets<int,tbb::flow::buffer_node<int> >(); |
437 | test_resets<float,tbb::flow::buffer_node<float> >(); |
438 | #if TBB_DEPRECATED_FLOW_NODE_EXTRACTION |
439 | test_buffer_extract<tbb::flow::buffer_node<int> >().run_tests(); |
440 | #endif |
441 | return Harness::Done; |
442 | } |
443 | |