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 | #define HARNESS_DEFAULT_MIN_THREADS 0 |
18 | #define HARNESS_DEFAULT_MAX_THREADS 4 |
19 | |
20 | #include "tbb/enumerable_thread_specific.h" |
21 | #include "tbb/task_scheduler_init.h" |
22 | #include "tbb/parallel_for.h" |
23 | #include "tbb/parallel_reduce.h" |
24 | #include "tbb/blocked_range.h" |
25 | #include "tbb/tick_count.h" |
26 | #include "tbb/tbb_allocator.h" |
27 | #include "tbb/tbb_thread.h" |
28 | #include "tbb/atomic.h" |
29 | |
30 | #include <cstring> |
31 | #include <vector> |
32 | #include <deque> |
33 | #include <list> |
34 | #include <map> |
35 | #include <utility> |
36 | |
37 | #include "harness_assert.h" |
38 | #include "harness.h" |
39 | #include "harness_checktype.h" |
40 | |
41 | #include "../tbbmalloc/shared_utils.h" |
42 | using rml::internal::estimatedCacheLineSize; |
43 | |
44 | #if __TBB_GCC_WARNING_SUPPRESSION_PRESENT |
45 | #pragma GCC diagnostic ignored "-Wuninitialized" |
46 | #endif |
47 | |
48 | static tbb::atomic<int> construction_counter; |
49 | static tbb::atomic<int> destruction_counter; |
50 | |
51 | #if TBB_USE_DEBUG |
52 | const int REPETITIONS = 4; |
53 | const int N = 10000; |
54 | const int RANGE_MIN=1000; |
55 | #else |
56 | const int REPETITIONS = 10; |
57 | const int N = 100000; |
58 | const int RANGE_MIN=10000; |
59 | #endif |
60 | const int VALID_NUMBER_OF_KEYS = 100; |
61 | const double EXPECTED_SUM = (REPETITIONS + 1) * N; |
62 | |
63 | //! A minimal class that occupies N bytes. |
64 | /** Defines default and copy constructor, and allows implicit operator&. |
65 | Hides operator=. */ |
66 | template<size_t N=tbb::internal::NFS_MaxLineSize> |
67 | class minimal: NoAssign { |
68 | private: |
69 | int my_value; |
70 | bool is_constructed; |
71 | char pad[N-sizeof(int) - sizeof(bool)]; |
72 | public: |
73 | minimal() : NoAssign(), my_value(0) { ++construction_counter; is_constructed = true; } |
74 | minimal( const minimal &m ) : NoAssign(), my_value(m.my_value) { ++construction_counter; is_constructed = true; } |
75 | ~minimal() { ++destruction_counter; ASSERT(is_constructed, NULL); is_constructed = false; } |
76 | void set_value( const int i ) { ASSERT(is_constructed, NULL); my_value = i; } |
77 | int value( ) const { ASSERT(is_constructed, NULL); return my_value; } |
78 | }; |
79 | |
80 | static size_t AlignMask = 0; // set to cache-line-size - 1 |
81 | |
82 | template<typename T> |
83 | T& check_alignment(T& t, const char *aname) { |
84 | if( !tbb::internal::is_aligned(&t, AlignMask)) { |
85 | REPORT_ONCE("alignment error with %s allocator (%x)\n" , aname, (int)size_t(&t) & (AlignMask-1)); |
86 | } |
87 | return t; |
88 | } |
89 | |
90 | template<typename T> |
91 | const T& check_alignment(const T& t, const char *aname) { |
92 | if( !tbb::internal::is_aligned(&t, AlignMask)) { |
93 | REPORT_ONCE("alignment error with %s allocator (%x)\n" , aname, (int)size_t(&t) & (AlignMask-1)); |
94 | } |
95 | return t; |
96 | } |
97 | |
98 | // Test constructors which throw. If an ETS constructor throws before completion, |
99 | // the already-built objects are un-constructed. Do not call the destructor if |
100 | // this occurs. |
101 | |
102 | static tbb::atomic<int> gThrowValue; |
103 | static int targetThrowValue = 3; |
104 | |
105 | class Thrower { |
106 | public: |
107 | Thrower() { |
108 | #if TBB_USE_EXCEPTIONS |
109 | if(++gThrowValue == targetThrowValue) { |
110 | throw std::bad_alloc(); |
111 | } |
112 | #endif |
113 | } |
114 | }; |
115 | |
116 | // MyThrower field of ThrowingConstructor will throw after a certain number of |
117 | // construction calls. The constructor unwinder wshould unconstruct the instance |
118 | // of check_type<int> that was constructed just before. |
119 | class ThrowingConstructor { |
120 | check_type<int> m_checktype; |
121 | Thrower m_throwing_field; |
122 | public: |
123 | int m_cnt; |
124 | ThrowingConstructor() : m_checktype(), m_throwing_field() { m_cnt = 0;} |
125 | private: |
126 | }; |
127 | |
128 | // |
129 | // A helper class that simplifies writing the tests since minimal does not |
130 | // define = or + operators. |
131 | // |
132 | |
133 | template< typename T > |
134 | struct test_helper { |
135 | static inline void init(T &e) { e = static_cast<T>(0); } |
136 | static inline void sum(T &e, const int addend ) { e += static_cast<T>(addend); } |
137 | static inline void sum(T &e, const double addend ) { e += static_cast<T>(addend); } |
138 | static inline void set(T &e, const int value ) { e = static_cast<T>(value); } |
139 | static inline double get(const T &e ) { return static_cast<double>(e); } |
140 | }; |
141 | |
142 | template<size_t N> |
143 | struct test_helper<minimal<N> > { |
144 | static inline void init(minimal<N> &sum) { sum.set_value( 0 ); } |
145 | static inline void sum(minimal<N> &sum, const int addend ) { sum.set_value( sum.value() + addend); } |
146 | static inline void sum(minimal<N> &sum, const double addend ) { sum.set_value( sum.value() + static_cast<int>(addend)); } |
147 | static inline void sum(minimal<N> &sum, const minimal<N> &addend ) { sum.set_value( sum.value() + addend.value()); } |
148 | static inline void set(minimal<N> &v, const int value ) { v.set_value( static_cast<int>(value) ); } |
149 | static inline double get(const minimal<N> &sum ) { return static_cast<double>(sum.value()); } |
150 | }; |
151 | |
152 | template<> |
153 | struct test_helper<ThrowingConstructor> { |
154 | static inline void init(ThrowingConstructor &sum) { sum.m_cnt = 0; } |
155 | static inline void sum(ThrowingConstructor &sum, const int addend ) { sum.m_cnt += addend; } |
156 | static inline void sum(ThrowingConstructor &sum, const double addend ) { sum.m_cnt += static_cast<int>(addend); } |
157 | static inline void sum(ThrowingConstructor &sum, const ThrowingConstructor &addend ) { sum.m_cnt += addend.m_cnt; } |
158 | static inline void set(ThrowingConstructor &v, const int value ) { v.m_cnt = static_cast<int>(value); } |
159 | static inline double get(const ThrowingConstructor &sum ) { return static_cast<double>(sum.m_cnt); } |
160 | }; |
161 | |
162 | //! Tag class used to make certain constructors hard to invoke accidentally. |
163 | struct SecretTagType {} SecretTag; |
164 | |
165 | //// functors and routines for initialization and combine |
166 | |
167 | //! Counts instances of FunctorFinit |
168 | static tbb::atomic<int> FinitCounter; |
169 | |
170 | template <typename T, int Value> |
171 | struct FunctorFinit { |
172 | FunctorFinit( const FunctorFinit& ) {++FinitCounter;} |
173 | FunctorFinit( SecretTagType ) {++FinitCounter;} |
174 | ~FunctorFinit() {--FinitCounter;} |
175 | T operator()() { return Value; } |
176 | }; |
177 | |
178 | template <int Value> |
179 | struct FunctorFinit<ThrowingConstructor,Value> { |
180 | FunctorFinit( const FunctorFinit& ) {++FinitCounter;} |
181 | FunctorFinit( SecretTagType ) {++FinitCounter;} |
182 | ~FunctorFinit() {--FinitCounter;} |
183 | ThrowingConstructor operator()() { ThrowingConstructor temp; temp.m_cnt = Value; return temp; } |
184 | }; |
185 | |
186 | template <size_t N, int Value> |
187 | struct FunctorFinit<minimal<N>,Value> { |
188 | FunctorFinit( const FunctorFinit& ) {++FinitCounter;} |
189 | FunctorFinit( SecretTagType ) {++FinitCounter;} |
190 | ~FunctorFinit() {--FinitCounter;} |
191 | minimal<N> operator()() { |
192 | minimal<N> result; |
193 | result.set_value( Value ); |
194 | return result; |
195 | } |
196 | }; |
197 | |
198 | // Addition |
199 | |
200 | template <typename T> |
201 | struct FunctorAddCombineRef { |
202 | T operator()(const T& left, const T& right) const { |
203 | return left+right; |
204 | } |
205 | }; |
206 | |
207 | template <size_t N> |
208 | struct FunctorAddCombineRef<minimal<N> > { |
209 | minimal<N> operator()(const minimal<N>& left, const minimal<N>& right) const { |
210 | minimal<N> result; |
211 | result.set_value( left.value() + right.value() ); |
212 | return result; |
213 | } |
214 | }; |
215 | |
216 | template <> |
217 | struct FunctorAddCombineRef<ThrowingConstructor> { |
218 | ThrowingConstructor operator()(const ThrowingConstructor& left, const ThrowingConstructor& right) const { |
219 | ThrowingConstructor result; |
220 | result.m_cnt = ( left.m_cnt + right.m_cnt ); |
221 | return result; |
222 | } |
223 | }; |
224 | |
225 | template <typename T> |
226 | struct FunctorAddCombine { |
227 | T operator()(T left, T right ) const { |
228 | return FunctorAddCombineRef<T>()( left, right ); |
229 | } |
230 | }; |
231 | |
232 | template <typename T> |
233 | T FunctionAddByRef( const T &left, const T &right) { |
234 | return FunctorAddCombineRef<T>()( left, right ); |
235 | } |
236 | |
237 | template <typename T> |
238 | T FunctionAdd( T left, T right) { return FunctionAddByRef(left,right); } |
239 | |
240 | template <typename T> |
241 | class Accumulator { |
242 | public: |
243 | Accumulator(T& _result) : my_result(_result) {} |
244 | Accumulator& operator=(const Accumulator& other) { |
245 | test_helper<T>::set(my_result, test_helper<T>::get(other)); |
246 | return *this; |
247 | } |
248 | void operator()(const T& new_bit) { test_helper<T>::sum(my_result, new_bit); } |
249 | private: |
250 | T& my_result; |
251 | }; |
252 | |
253 | template <typename T> |
254 | class ClearingAccumulator { |
255 | public: |
256 | ClearingAccumulator(T& _result) : my_result(_result) {} |
257 | ClearingAccumulator& operator=(const ClearingAccumulator& other) { |
258 | test_helper<T>::set(my_result, test_helper<T>::get(other)); |
259 | return *this; |
260 | } |
261 | void operator()(T& new_bit) { |
262 | test_helper<T>::sum(my_result, new_bit); |
263 | test_helper<T>::init(new_bit); |
264 | } |
265 | static void AssertClean(const T& thread_local_value) { |
266 | T zero; |
267 | test_helper<T>::init(zero); |
268 | ASSERT(test_helper<T>::get(thread_local_value)==test_helper<T>::get(zero), |
269 | "combine_each does not allow to modify thread local values?" ); |
270 | } |
271 | private: |
272 | T& my_result; |
273 | }; |
274 | |
275 | //// end functors and routines |
276 | |
277 | template< typename T > |
278 | void run_serial_scalar_tests(const char *test_name) { |
279 | tbb::tick_count t0; |
280 | T sum; |
281 | test_helper<T>::init(sum); |
282 | |
283 | REMARK("Testing serial %s... " , test_name); |
284 | for (int t = -1; t < REPETITIONS; ++t) { |
285 | if (Verbose && t == 0) t0 = tbb::tick_count::now(); |
286 | for (int i = 0; i < N; ++i) { |
287 | test_helper<T>::sum(sum,1); |
288 | } |
289 | } |
290 | |
291 | double result_value = test_helper<T>::get(sum); |
292 | ASSERT( EXPECTED_SUM == result_value, NULL); |
293 | REMARK("done\nserial %s, 0, %g, %g\n" , test_name, result_value, ( tbb::tick_count::now() - t0).seconds()); |
294 | } |
295 | |
296 | |
297 | template <typename T, template<class> class Allocator> |
298 | class parallel_scalar_body: NoAssign { |
299 | typedef tbb::enumerable_thread_specific<T, Allocator<T> > ets_type; |
300 | ets_type &sums; |
301 | const char* allocator_name; |
302 | |
303 | public: |
304 | |
305 | parallel_scalar_body ( ets_type &_sums, const char *alloc_name ) : sums(_sums), allocator_name(alloc_name) { } |
306 | |
307 | void operator()( const tbb::blocked_range<int> &r ) const { |
308 | for (int i = r.begin(); i != r.end(); ++i) |
309 | test_helper<T>::sum( check_alignment(sums.local(),allocator_name), 1 ); |
310 | } |
311 | |
312 | }; |
313 | |
314 | template< typename T, template<class> class Allocator> |
315 | void run_parallel_scalar_tests_nocombine(const char *test_name, const char *allocator_name) { |
316 | |
317 | typedef tbb::enumerable_thread_specific<T, Allocator<T> > ets_type; |
318 | |
319 | Check<T> my_check; |
320 | gThrowValue = 0; |
321 | { |
322 | // We assume that static_sums zero-initialized or has a default constructor that zeros it. |
323 | static ets_type static_sums = ets_type( T() ); |
324 | |
325 | T exemplar; |
326 | test_helper<T>::init(exemplar); |
327 | |
328 | for (int p = MinThread; p <= MaxThread; ++p) { |
329 | REMARK("Testing parallel %s with allocator %s on %d thread(s)... " , test_name, allocator_name, p); |
330 | tbb::task_scheduler_init init(p); |
331 | tbb::tick_count t0; |
332 | |
333 | T iterator_sum; |
334 | test_helper<T>::init(iterator_sum); |
335 | |
336 | T finit_ets_sum; |
337 | test_helper<T>::init(finit_ets_sum); |
338 | |
339 | T const_iterator_sum; |
340 | test_helper<T>::init(const_iterator_sum); |
341 | |
342 | T range_sum; |
343 | test_helper<T>::init(range_sum); |
344 | |
345 | T const_range_sum; |
346 | test_helper<T>::init(const_range_sum); |
347 | |
348 | T cconst_sum; |
349 | test_helper<T>::init(cconst_sum); |
350 | |
351 | T assign_sum; |
352 | test_helper<T>::init(assign_sum); |
353 | |
354 | T cassgn_sum; |
355 | test_helper<T>::init(cassgn_sum); |
356 | T non_cassgn_sum; |
357 | test_helper<T>::init(non_cassgn_sum); |
358 | |
359 | T static_sum; |
360 | test_helper<T>::init(static_sum); |
361 | |
362 | for (int t = -1; t < REPETITIONS; ++t) { |
363 | if (Verbose && t == 0) t0 = tbb::tick_count::now(); |
364 | |
365 | static_sums.clear(); |
366 | |
367 | ets_type sums(exemplar); |
368 | FunctorFinit<T,0> my_finit(SecretTag); |
369 | ets_type finit_ets(my_finit); |
370 | |
371 | ASSERT( sums.empty(), NULL); |
372 | tbb::parallel_for( tbb::blocked_range<int>( 0, N, RANGE_MIN ), parallel_scalar_body<T,Allocator>( sums, allocator_name ) ); |
373 | ASSERT( !sums.empty(), NULL); |
374 | |
375 | ASSERT( finit_ets.empty(), NULL); |
376 | tbb::parallel_for( tbb::blocked_range<int>( 0, N, RANGE_MIN ), parallel_scalar_body<T,Allocator>( finit_ets, allocator_name ) ); |
377 | ASSERT( !finit_ets.empty(), NULL); |
378 | |
379 | ASSERT(static_sums.empty(), NULL); |
380 | tbb::parallel_for( tbb::blocked_range<int>( 0, N, RANGE_MIN ), parallel_scalar_body<T,Allocator>( static_sums, allocator_name ) ); |
381 | ASSERT( !static_sums.empty(), NULL); |
382 | |
383 | // use iterator |
384 | typename ets_type::size_type size = 0; |
385 | for ( typename ets_type::iterator i = sums.begin(); i != sums.end(); ++i ) { |
386 | ++size; |
387 | test_helper<T>::sum(iterator_sum, *i); |
388 | } |
389 | ASSERT( sums.size() == size, NULL); |
390 | |
391 | // use const_iterator |
392 | for ( typename ets_type::const_iterator i = sums.begin(); i != sums.end(); ++i ) { |
393 | test_helper<T>::sum(const_iterator_sum, *i); |
394 | } |
395 | |
396 | // use range_type |
397 | typename ets_type::range_type r = sums.range(); |
398 | for ( typename ets_type::range_type::const_iterator i = r.begin(); i != r.end(); ++i ) { |
399 | test_helper<T>::sum(range_sum, *i); |
400 | } |
401 | |
402 | // use const_range_type |
403 | typename ets_type::const_range_type cr = sums.range(); |
404 | for ( typename ets_type::const_range_type::iterator i = cr.begin(); i != cr.end(); ++i ) { |
405 | test_helper<T>::sum(const_range_sum, *i); |
406 | } |
407 | |
408 | // test copy constructor, with TLS-cached locals |
409 | typedef typename tbb::enumerable_thread_specific<T, Allocator<T>, tbb::ets_key_per_instance> cached_ets_type; |
410 | |
411 | cached_ets_type cconst(sums); |
412 | |
413 | for ( typename cached_ets_type::const_iterator i = cconst.begin(); i != cconst.end(); ++i ) { |
414 | test_helper<T>::sum(cconst_sum, *i); |
415 | } |
416 | |
417 | // test assignment |
418 | ets_type assigned; |
419 | assigned = sums; |
420 | |
421 | for ( typename ets_type::const_iterator i = assigned.begin(); i != assigned.end(); ++i ) { |
422 | test_helper<T>::sum(assign_sum, *i); |
423 | } |
424 | |
425 | // test assign to and from cached locals |
426 | cached_ets_type cassgn; |
427 | cassgn = sums; |
428 | for ( typename cached_ets_type::const_iterator i = cassgn.begin(); i != cassgn.end(); ++i ) { |
429 | test_helper<T>::sum(cassgn_sum, *i); |
430 | } |
431 | |
432 | ets_type non_cassgn; |
433 | non_cassgn = cassgn; |
434 | for ( typename ets_type::const_iterator i = non_cassgn.begin(); i != non_cassgn.end(); ++i ) { |
435 | test_helper<T>::sum(non_cassgn_sum, *i); |
436 | } |
437 | |
438 | // test finit-initialized ets |
439 | for(typename ets_type::const_iterator i = finit_ets.begin(); i != finit_ets.end(); ++i) { |
440 | test_helper<T>::sum(finit_ets_sum, *i); |
441 | } |
442 | |
443 | // test static ets |
444 | for(typename ets_type::const_iterator i = static_sums.begin(); i != static_sums.end(); ++i) { |
445 | test_helper<T>::sum(static_sum, *i); |
446 | } |
447 | |
448 | } |
449 | |
450 | ASSERT( EXPECTED_SUM == test_helper<T>::get(iterator_sum), NULL); |
451 | ASSERT( EXPECTED_SUM == test_helper<T>::get(const_iterator_sum), NULL); |
452 | ASSERT( EXPECTED_SUM == test_helper<T>::get(range_sum), NULL); |
453 | ASSERT( EXPECTED_SUM == test_helper<T>::get(const_range_sum), NULL); |
454 | |
455 | ASSERT( EXPECTED_SUM == test_helper<T>::get(cconst_sum), NULL); |
456 | ASSERT( EXPECTED_SUM == test_helper<T>::get(assign_sum), NULL); |
457 | ASSERT( EXPECTED_SUM == test_helper<T>::get(cassgn_sum), NULL); |
458 | ASSERT( EXPECTED_SUM == test_helper<T>::get(non_cassgn_sum), NULL); |
459 | ASSERT( EXPECTED_SUM == test_helper<T>::get(finit_ets_sum), NULL); |
460 | ASSERT( EXPECTED_SUM == test_helper<T>::get(static_sum), NULL); |
461 | |
462 | REMARK("done\nparallel %s, %d, %g, %g\n" , test_name, p, test_helper<T>::get(iterator_sum), |
463 | ( tbb::tick_count::now() - t0).seconds()); |
464 | } |
465 | } // Check block |
466 | } |
467 | |
468 | template< typename T, template<class> class Allocator> |
469 | void run_parallel_scalar_tests(const char *test_name, const char *allocator_name) { |
470 | |
471 | typedef tbb::enumerable_thread_specific<T, Allocator<T> > ets_type; |
472 | bool exception_caught = false; |
473 | |
474 | // We assume that static_sums zero-initialized or has a default constructor that zeros it. |
475 | static ets_type static_sums = ets_type( T() ); |
476 | |
477 | T exemplar; |
478 | test_helper<T>::init(exemplar); |
479 | |
480 | int test_throw_count = 10; |
481 | // the test will be performed repeatedly until it does not throw. For non-throwing types |
482 | // this means once; for the throwing type test it may loop two or three times. The |
483 | // value of targetThrowValue will determine when and if the test will throw. |
484 | do { |
485 | targetThrowValue = test_throw_count; // keep testing until we get no exception |
486 | exception_caught = false; |
487 | #if TBB_USE_EXCEPTIONS |
488 | try { |
489 | #endif |
490 | run_parallel_scalar_tests_nocombine<T,Allocator>(test_name, allocator_name); |
491 | #if TBB_USE_EXCEPTIONS |
492 | } |
493 | catch(...) { |
494 | REMARK("Exception caught %d\n" , targetThrowValue); |
495 | } |
496 | #endif |
497 | for (int p = MinThread; p <= MaxThread; ++p) { |
498 | REMARK("Testing parallel %s with allocator %s on %d thread(s)... " , test_name, allocator_name, p); |
499 | tbb::task_scheduler_init init(p); |
500 | tbb::tick_count t0; |
501 | |
502 | gThrowValue = 0; |
503 | |
504 | T combine_sum; |
505 | test_helper<T>::init(combine_sum); |
506 | |
507 | T combine_ref_sum; |
508 | test_helper<T>::init(combine_ref_sum); |
509 | |
510 | T accumulator_sum; |
511 | test_helper<T>::init(accumulator_sum); |
512 | |
513 | T static_sum; |
514 | test_helper<T>::init(static_sum); |
515 | |
516 | T clearing_accumulator_sum; |
517 | test_helper<T>::init(clearing_accumulator_sum); |
518 | |
519 | { |
520 | Check<T> my_check; |
521 | #if TBB_USE_EXCEPTIONS |
522 | try |
523 | #endif |
524 | { |
525 | for (int t = -1; t < REPETITIONS; ++t) { |
526 | if (Verbose && t == 0) t0 = tbb::tick_count::now(); |
527 | |
528 | static_sums.clear(); |
529 | |
530 | ets_type sums(exemplar); |
531 | |
532 | ASSERT( sums.empty(), NULL); |
533 | tbb::parallel_for( tbb::blocked_range<int>( 0, N, RANGE_MIN ), |
534 | parallel_scalar_body<T,Allocator>( sums, allocator_name ) ); |
535 | ASSERT( !sums.empty(), NULL); |
536 | |
537 | ASSERT(static_sums.empty(), NULL); |
538 | tbb::parallel_for( tbb::blocked_range<int>( 0, N, RANGE_MIN ), |
539 | parallel_scalar_body<T,Allocator>( static_sums, allocator_name ) ); |
540 | ASSERT( !static_sums.empty(), NULL); |
541 | |
542 | // Use combine |
543 | test_helper<T>::sum(combine_sum, sums.combine(FunctionAdd<T>)); |
544 | test_helper<T>::sum(combine_ref_sum, sums.combine(FunctionAddByRef<T>)); |
545 | test_helper<T>::sum(static_sum, static_sums.combine(FunctionAdd<T>)); |
546 | |
547 | // Accumulate with combine_each |
548 | sums.combine_each(Accumulator<T>(accumulator_sum)); |
549 | // Accumulate and clear thread-local values |
550 | sums.combine_each(ClearingAccumulator<T>(clearing_accumulator_sum)); |
551 | // Check that the values were cleared |
552 | sums.combine_each(ClearingAccumulator<T>::AssertClean); |
553 | } |
554 | } |
555 | #if TBB_USE_EXCEPTIONS |
556 | catch(...) { |
557 | REMARK("Exception caught %d\n" , targetThrowValue); |
558 | exception_caught = true; |
559 | } |
560 | #endif |
561 | } |
562 | |
563 | ASSERT( EXPECTED_SUM == test_helper<T>::get(combine_sum) || exception_caught, NULL); |
564 | ASSERT( EXPECTED_SUM == test_helper<T>::get(combine_ref_sum) || exception_caught, NULL); |
565 | ASSERT( EXPECTED_SUM == test_helper<T>::get(static_sum) || exception_caught, NULL); |
566 | ASSERT( EXPECTED_SUM == test_helper<T>::get(accumulator_sum) || exception_caught, NULL); |
567 | ASSERT( EXPECTED_SUM == test_helper<T>::get(clearing_accumulator_sum) || exception_caught, NULL); |
568 | |
569 | REMARK("done\nparallel combine %s, %d, %g, %g\n" , test_name, p, test_helper<T>::get(combine_sum), |
570 | ( tbb::tick_count::now() - t0).seconds()); |
571 | } // MinThread .. MaxThread |
572 | test_throw_count += 10; // keep testing until we don't get an exception |
573 | } while (exception_caught && test_throw_count < 200); |
574 | ASSERT(!exception_caught, "No non-exception test completed" ); |
575 | } |
576 | |
577 | template <typename T, template<class> class Allocator> |
578 | class parallel_vector_for_body: NoAssign { |
579 | typedef std::vector<T, tbb::tbb_allocator<T> > container_type; |
580 | typedef tbb::enumerable_thread_specific< container_type, Allocator<container_type> > ets_type; |
581 | ets_type &locals; |
582 | const char *allocator_name; |
583 | |
584 | public: |
585 | |
586 | parallel_vector_for_body ( ets_type &_locals, const char *aname ) : locals(_locals), allocator_name(aname) { } |
587 | |
588 | void operator()( const tbb::blocked_range<int> &r ) const { |
589 | T one; |
590 | test_helper<T>::set(one, 1); |
591 | |
592 | for (int i = r.begin(); i < r.end(); ++i) { |
593 | check_alignment(locals.local(),allocator_name).push_back( one ); |
594 | } |
595 | } |
596 | |
597 | }; |
598 | |
599 | template <typename R, typename T> |
600 | struct parallel_vector_reduce_body { |
601 | |
602 | T sum; |
603 | size_t count; |
604 | typedef std::vector<T, tbb::tbb_allocator<T> > container_type; |
605 | |
606 | parallel_vector_reduce_body ( ) : count(0) { test_helper<T>::init(sum); } |
607 | parallel_vector_reduce_body ( parallel_vector_reduce_body<R, T> &, tbb::split ) : count(0) { test_helper<T>::init(sum); } |
608 | |
609 | void operator()( const R &r ) { |
610 | for (typename R::iterator ri = r.begin(); ri != r.end(); ++ri) { |
611 | const container_type &v = *ri; |
612 | ++count; |
613 | for (typename container_type::const_iterator vi = v.begin(); vi != v.end(); ++vi) { |
614 | test_helper<T>::sum(sum, *vi); |
615 | } |
616 | } |
617 | } |
618 | |
619 | void join( const parallel_vector_reduce_body &b ) { |
620 | test_helper<T>::sum(sum,b.sum); |
621 | count += b.count; |
622 | } |
623 | |
624 | }; |
625 | |
626 | template< typename T, template<class> class Allocator> |
627 | void run_parallel_vector_tests(const char *test_name, const char *allocator_name) { |
628 | tbb::tick_count t0; |
629 | typedef std::vector<T, tbb::tbb_allocator<T> > container_type; |
630 | typedef tbb::enumerable_thread_specific< container_type, Allocator<container_type> > ets_type; |
631 | |
632 | for (int p = MinThread; p <= MaxThread; ++p) { |
633 | REMARK("Testing parallel %s with allocator %s on %d thread(s)... " , test_name, allocator_name, p); |
634 | tbb::task_scheduler_init init(p); |
635 | |
636 | T sum; |
637 | test_helper<T>::init(sum); |
638 | |
639 | for (int t = -1; t < REPETITIONS; ++t) { |
640 | if (Verbose && t == 0) t0 = tbb::tick_count::now(); |
641 | ets_type vs; |
642 | |
643 | ASSERT( vs.empty(), NULL ); |
644 | tbb::parallel_for( tbb::blocked_range<int> (0, N, RANGE_MIN), |
645 | parallel_vector_for_body<T,Allocator>( vs, allocator_name ) ); |
646 | ASSERT( !vs.empty(), NULL ); |
647 | |
648 | // copy construct |
649 | ets_type vs2(vs); // this causes an assertion failure, related to allocators... |
650 | |
651 | // assign |
652 | ets_type vs3; |
653 | vs3 = vs; |
654 | |
655 | parallel_vector_reduce_body< typename ets_type::const_range_type, T > pvrb; |
656 | tbb::parallel_reduce ( vs.range(1), pvrb ); |
657 | |
658 | test_helper<T>::sum(sum, pvrb.sum); |
659 | |
660 | ASSERT( vs.size() == pvrb.count, NULL ); |
661 | ASSERT( vs2.size() == pvrb.count, NULL ); |
662 | ASSERT( vs3.size() == pvrb.count, NULL ); |
663 | |
664 | tbb::flattened2d<ets_type> fvs = flatten2d(vs); |
665 | size_t ccount = fvs.size(); |
666 | ASSERT( ccount == size_t(N), NULL ); |
667 | size_t elem_cnt = 0; |
668 | for(typename tbb::flattened2d<ets_type>::const_iterator i = fvs.begin(); i != fvs.end(); ++i) { |
669 | ++elem_cnt; |
670 | }; |
671 | ASSERT( ccount == elem_cnt, NULL ); |
672 | |
673 | elem_cnt = 0; |
674 | for(typename tbb::flattened2d<ets_type>::iterator i = fvs.begin(); i != fvs.end(); ++i) { |
675 | ++elem_cnt; |
676 | }; |
677 | ASSERT( ccount == elem_cnt, NULL ); |
678 | |
679 | #if __TBB_ETS_USE_CPP11 |
680 | // Test the ETS constructor with multiple args |
681 | T minus_one; |
682 | test_helper<T>::set(minus_one, -1); |
683 | // Set ETS to construct "local" vectors pre-occupied with 25 "minus_one"s |
684 | // Cast 25 to size_type to prevent Intel Compiler SFINAE compilation issues with gcc 5. |
685 | ets_type vvs( typename container_type::size_type(25), minus_one, tbb::tbb_allocator<T>() ); |
686 | ASSERT( vvs.empty(), NULL ); |
687 | tbb::parallel_for ( tbb::blocked_range<int> (0, N, RANGE_MIN), parallel_vector_for_body<T,Allocator>( vvs, allocator_name ) ); |
688 | ASSERT( !vvs.empty(), NULL ); |
689 | |
690 | parallel_vector_reduce_body< typename ets_type::const_range_type, T > pvrb2; |
691 | tbb::parallel_reduce ( vvs.range(1), pvrb2 ); |
692 | ASSERT( pvrb2.count == vvs.size(), NULL ); |
693 | ASSERT( test_helper<T>::get(pvrb2.sum) == N-pvrb2.count*25, NULL ); |
694 | |
695 | tbb::flattened2d<ets_type> fvvs = flatten2d(vvs); |
696 | ccount = fvvs.size(); |
697 | ASSERT( ccount == N+pvrb2.count*25, NULL ); |
698 | #endif |
699 | } |
700 | |
701 | double result_value = test_helper<T>::get(sum); |
702 | ASSERT( EXPECTED_SUM == result_value, NULL); |
703 | REMARK("done\nparallel %s, %d, %g, %g\n" , test_name, p, result_value, ( tbb::tick_count::now() - t0).seconds()); |
704 | } |
705 | } |
706 | |
707 | template<typename T, template<class> class Allocator> |
708 | void run_cross_type_vector_tests(const char *test_name) { |
709 | tbb::tick_count t0; |
710 | const char* allocator_name = "default" ; |
711 | typedef std::vector<T, tbb::tbb_allocator<T> > container_type; |
712 | |
713 | for (int p = MinThread; p <= MaxThread; ++p) { |
714 | REMARK("Testing parallel %s on %d thread(s)... " , test_name, p); |
715 | tbb::task_scheduler_init init(p); |
716 | |
717 | T sum; |
718 | test_helper<T>::init(sum); |
719 | |
720 | for (int t = -1; t < REPETITIONS; ++t) { |
721 | if (Verbose && t == 0) t0 = tbb::tick_count::now(); |
722 | typedef typename tbb::enumerable_thread_specific< container_type, Allocator<container_type>, tbb::ets_no_key > ets_nokey_type; |
723 | typedef typename tbb::enumerable_thread_specific< container_type, Allocator<container_type>, tbb::ets_key_per_instance > ets_tlskey_type; |
724 | ets_nokey_type vs; |
725 | |
726 | ASSERT( vs.empty(), NULL); |
727 | tbb::parallel_for ( tbb::blocked_range<int> (0, N, RANGE_MIN), parallel_vector_for_body<T, Allocator>( vs, allocator_name ) ); |
728 | ASSERT( !vs.empty(), NULL); |
729 | |
730 | // copy construct |
731 | ets_tlskey_type vs2(vs); |
732 | |
733 | // assign |
734 | ets_nokey_type vs3; |
735 | vs3 = vs2; |
736 | |
737 | parallel_vector_reduce_body< typename ets_nokey_type::const_range_type, T > pvrb; |
738 | tbb::parallel_reduce ( vs3.range(1), pvrb ); |
739 | |
740 | test_helper<T>::sum(sum, pvrb.sum); |
741 | |
742 | ASSERT( vs3.size() == pvrb.count, NULL); |
743 | |
744 | tbb::flattened2d<ets_nokey_type> fvs = flatten2d(vs3); |
745 | size_t ccount = fvs.size(); |
746 | size_t elem_cnt = 0; |
747 | for(typename tbb::flattened2d<ets_nokey_type>::const_iterator i = fvs.begin(); i != fvs.end(); ++i) { |
748 | ++elem_cnt; |
749 | }; |
750 | ASSERT(ccount == elem_cnt, NULL); |
751 | |
752 | elem_cnt = 0; |
753 | for(typename tbb::flattened2d<ets_nokey_type>::iterator i = fvs.begin(); i != fvs.end(); ++i) { |
754 | ++elem_cnt; |
755 | }; |
756 | ASSERT(ccount == elem_cnt, NULL); |
757 | } |
758 | |
759 | double result_value = test_helper<T>::get(sum); |
760 | ASSERT( EXPECTED_SUM == result_value, NULL); |
761 | REMARK("done\nparallel %s, %d, %g, %g\n" , test_name, p, result_value, ( tbb::tick_count::now() - t0).seconds()); |
762 | } |
763 | } |
764 | |
765 | template< typename T > |
766 | void run_serial_vector_tests(const char *test_name) { |
767 | tbb::tick_count t0; |
768 | T sum; |
769 | test_helper<T>::init(sum); |
770 | T one; |
771 | test_helper<T>::set(one, 1); |
772 | |
773 | REMARK("Testing serial %s... " , test_name); |
774 | for (int t = -1; t < REPETITIONS; ++t) { |
775 | if (Verbose && t == 0) t0 = tbb::tick_count::now(); |
776 | std::vector<T, tbb::tbb_allocator<T> > v; |
777 | for (int i = 0; i < N; ++i) { |
778 | v.push_back( one ); |
779 | } |
780 | for (typename std::vector<T, tbb::tbb_allocator<T> >::const_iterator i = v.begin(); i != v.end(); ++i) |
781 | test_helper<T>::sum(sum, *i); |
782 | } |
783 | |
784 | double result_value = test_helper<T>::get(sum); |
785 | ASSERT( EXPECTED_SUM == result_value, NULL); |
786 | REMARK("done\nserial %s, 0, %g, %g\n" , test_name, result_value, ( tbb::tick_count::now() - t0).seconds()); |
787 | } |
788 | |
789 | const size_t line_size = tbb::internal::NFS_MaxLineSize; |
790 | |
791 | void run_serial_tests() { |
792 | run_serial_scalar_tests<int>("int" ); |
793 | run_serial_scalar_tests<double>("double" ); |
794 | run_serial_scalar_tests<minimal<> >("minimal<>" ); |
795 | run_serial_vector_tests<int>("std::vector<int, tbb::tbb_allocator<int> >" ); |
796 | run_serial_vector_tests<double>("std::vector<double, tbb::tbb_allocator<double> >" ); |
797 | } |
798 | |
799 | template<template<class>class Allocator> |
800 | void run_parallel_tests(const char *allocator_name) { |
801 | run_parallel_scalar_tests<int, Allocator>("int" ,allocator_name); |
802 | run_parallel_scalar_tests<double, Allocator>("double" ,allocator_name); |
803 | run_parallel_scalar_tests_nocombine<minimal<>,Allocator>("minimal<>" ,allocator_name); |
804 | run_parallel_scalar_tests<ThrowingConstructor, Allocator>("ThrowingConstructor" , allocator_name); |
805 | run_parallel_vector_tests<int, Allocator>("std::vector<int, tbb::tbb_allocator<int> >" ,allocator_name); |
806 | run_parallel_vector_tests<double, Allocator>("std::vector<double, tbb::tbb_allocator<double> >" ,allocator_name); |
807 | } |
808 | |
809 | void run_cross_type_tests() { |
810 | // cross-type scalar tests are part of run_parallel_scalar_tests_nocombine |
811 | run_cross_type_vector_tests<int, tbb::tbb_allocator>("std::vector<int, tbb::tbb_allocator<int> >" ); |
812 | run_cross_type_vector_tests<double, tbb::tbb_allocator>("std::vector<double, tbb::tbb_allocator<double> >" ); |
813 | } |
814 | |
815 | typedef tbb::enumerable_thread_specific<minimal<line_size> > flogged_ets; |
816 | |
817 | class set_body { |
818 | flogged_ets *a; |
819 | |
820 | public: |
821 | set_body( flogged_ets*_a ) : a(_a) { } |
822 | |
823 | void operator() ( ) const { |
824 | for (int i = 0; i < VALID_NUMBER_OF_KEYS; ++i) { |
825 | check_alignment(a[i].local(), "default" ).set_value(i + 1); |
826 | } |
827 | } |
828 | |
829 | }; |
830 | |
831 | void do_tbb_threads( int max_threads, flogged_ets a[] ) { |
832 | std::vector< tbb::tbb_thread * > threads; |
833 | |
834 | for (int p = 0; p < max_threads; ++p) { |
835 | threads.push_back( new tbb::tbb_thread ( set_body( a ) ) ); |
836 | } |
837 | |
838 | for (int p = 0; p < max_threads; ++p) { |
839 | threads[p]->join(); |
840 | } |
841 | |
842 | for(int p = 0; p < max_threads; ++p) { |
843 | delete threads[p]; |
844 | } |
845 | } |
846 | |
847 | void flog_key_creation_and_deletion() { |
848 | const int FLOG_REPETITIONS = 100; |
849 | |
850 | for (int p = MinThread; p <= MaxThread; ++p) { |
851 | REMARK("Testing repeated deletes on %d threads... " , p); |
852 | |
853 | for (int j = 0; j < FLOG_REPETITIONS; ++j) { |
854 | construction_counter = 0; |
855 | destruction_counter = 0; |
856 | |
857 | // causes VALID_NUMBER_OF_KEYS exemplar instances to be constructed |
858 | flogged_ets* a = new flogged_ets[VALID_NUMBER_OF_KEYS]; |
859 | ASSERT(int(construction_counter) == 0, NULL); // no exemplars or actual locals have been constructed |
860 | ASSERT(int(destruction_counter) == 0, NULL); // and none have been destroyed |
861 | |
862 | // causes p * VALID_NUMBER_OF_KEYS minimals to be created |
863 | do_tbb_threads(p, a); |
864 | |
865 | for (int i = 0; i < VALID_NUMBER_OF_KEYS; ++i) { |
866 | int pcnt = 0; |
867 | for ( flogged_ets::iterator tli = a[i].begin(); tli != a[i].end(); ++tli ) { |
868 | ASSERT( (*tli).value() == i+1, NULL ); |
869 | ++pcnt; |
870 | } |
871 | ASSERT( pcnt == p, NULL); // should be one local per thread. |
872 | } |
873 | delete[] a; |
874 | } |
875 | |
876 | ASSERT( int(construction_counter) == (p)*VALID_NUMBER_OF_KEYS, NULL ); |
877 | ASSERT( int(destruction_counter) == (p)*VALID_NUMBER_OF_KEYS, NULL ); |
878 | |
879 | REMARK("done\nTesting repeated clears on %d threads... " , p); |
880 | |
881 | construction_counter = 0; |
882 | destruction_counter = 0; |
883 | |
884 | // causes VALID_NUMBER_OF_KEYS exemplar instances to be constructed |
885 | flogged_ets* a = new flogged_ets[VALID_NUMBER_OF_KEYS]; |
886 | |
887 | for (int j = 0; j < FLOG_REPETITIONS; ++j) { |
888 | |
889 | // causes p * VALID_NUMBER_OF_KEYS minimals to be created |
890 | do_tbb_threads(p, a); |
891 | |
892 | for (int i = 0; i < VALID_NUMBER_OF_KEYS; ++i) { |
893 | for ( flogged_ets::iterator tli = a[i].begin(); tli != a[i].end(); ++tli ) { |
894 | ASSERT( (*tli).value() == i+1, NULL ); |
895 | } |
896 | a[i].clear(); |
897 | ASSERT( static_cast<int>(a[i].end() - a[i].begin()) == 0, NULL ); |
898 | } |
899 | |
900 | } |
901 | |
902 | delete[] a; |
903 | |
904 | ASSERT( int(construction_counter) == (FLOG_REPETITIONS*p)*VALID_NUMBER_OF_KEYS, NULL ); |
905 | ASSERT( int(destruction_counter) == (FLOG_REPETITIONS*p)*VALID_NUMBER_OF_KEYS, NULL ); |
906 | |
907 | REMARK("done\n" ); |
908 | } |
909 | |
910 | } |
911 | |
912 | template <typename inner_container> |
913 | void flog_segmented_interator() { |
914 | |
915 | bool found_error = false; |
916 | typedef typename inner_container::value_type T; |
917 | typedef std::vector< inner_container > nested_vec; |
918 | inner_container my_inner_container; |
919 | my_inner_container.clear(); |
920 | nested_vec my_vec; |
921 | |
922 | // simple nested vector (neither level empty) |
923 | const int maxval = 10; |
924 | for(int i=0; i < maxval; i++) { |
925 | my_vec.push_back(my_inner_container); |
926 | for(int j = 0; j < maxval; j++) { |
927 | my_vec.at(i).push_back((T)(maxval * i + j)); |
928 | } |
929 | } |
930 | |
931 | tbb::internal::segmented_iterator<nested_vec, T> my_si(my_vec); |
932 | |
933 | T ii; |
934 | for(my_si=my_vec.begin(), ii=0; my_si != my_vec.end(); ++my_si, ++ii) { |
935 | if((*my_si) != ii) { |
936 | found_error = true; |
937 | REMARK( "*my_si=%d\n" , int(*my_si)); |
938 | } |
939 | } |
940 | |
941 | // outer level empty |
942 | my_vec.clear(); |
943 | for(my_si=my_vec.begin(); my_si != my_vec.end(); ++my_si) { |
944 | found_error = true; |
945 | } |
946 | |
947 | // inner levels empty |
948 | my_vec.clear(); |
949 | for(int i =0; i < maxval; ++i) { |
950 | my_vec.push_back(my_inner_container); |
951 | } |
952 | for(my_si = my_vec.begin(); my_si != my_vec.end(); ++my_si) { |
953 | found_error = true; |
954 | } |
955 | |
956 | // every other inner container is empty |
957 | my_vec.clear(); |
958 | for(int i=0; i < maxval; ++i) { |
959 | my_vec.push_back(my_inner_container); |
960 | if(i%2) { |
961 | for(int j = 0; j < maxval; ++j) { |
962 | my_vec.at(i).push_back((T)(maxval * (i/2) + j)); |
963 | } |
964 | } |
965 | } |
966 | for(my_si = my_vec.begin(), ii=0; my_si != my_vec.end(); ++my_si, ++ii) { |
967 | if((*my_si) != ii) { |
968 | found_error = true; |
969 | REMARK("*my_si=%d, ii=%d\n" , (int)(*my_si), (int)ii); |
970 | } |
971 | } |
972 | |
973 | tbb::internal::segmented_iterator<nested_vec, const T> my_csi(my_vec); |
974 | for(my_csi=my_vec.begin(), ii=0; my_csi != my_vec.end(); ++my_csi, ++ii) { |
975 | if((*my_csi) != ii) { |
976 | found_error = true; |
977 | REMARK( "*my_csi=%d\n" , int(*my_csi)); |
978 | } |
979 | } |
980 | |
981 | // outer level empty |
982 | my_vec.clear(); |
983 | for(my_csi=my_vec.begin(); my_csi != my_vec.end(); ++my_csi) { |
984 | found_error = true; |
985 | } |
986 | |
987 | // inner levels empty |
988 | my_vec.clear(); |
989 | for(int i =0; i < maxval; ++i) { |
990 | my_vec.push_back(my_inner_container); |
991 | } |
992 | for(my_csi = my_vec.begin(); my_csi != my_vec.end(); ++my_csi) { |
993 | found_error = true; |
994 | } |
995 | |
996 | // every other inner container is empty |
997 | my_vec.clear(); |
998 | for(int i=0; i < maxval; ++i) { |
999 | my_vec.push_back(my_inner_container); |
1000 | if(i%2) { |
1001 | for(int j = 0; j < maxval; ++j) { |
1002 | my_vec.at(i).push_back((T)(maxval * (i/2) + j)); |
1003 | } |
1004 | } |
1005 | } |
1006 | for(my_csi = my_vec.begin(), ii=0; my_csi != my_vec.end(); ++my_csi, ++ii) { |
1007 | if((*my_csi) != ii) { |
1008 | found_error = true; |
1009 | REMARK("*my_csi=%d, ii=%d\n" , (int)(*my_csi), (int)ii); |
1010 | } |
1011 | } |
1012 | |
1013 | |
1014 | if(found_error) REPORT("segmented_iterator failed\n" ); |
1015 | } |
1016 | |
1017 | template <typename Key, typename Val> |
1018 | void flog_segmented_iterator_map() { |
1019 | typedef typename std::map<Key, Val> my_map; |
1020 | typedef std::vector< my_map > nested_vec; |
1021 | my_map my_inner_container; |
1022 | my_inner_container.clear(); |
1023 | nested_vec my_vec; |
1024 | my_vec.clear(); |
1025 | bool found_error = false; |
1026 | |
1027 | // simple nested vector (neither level empty) |
1028 | const int maxval = 4; |
1029 | for(int i=0; i < maxval; i++) { |
1030 | my_vec.push_back(my_inner_container); |
1031 | for(int j = 0; j < maxval; j++) { |
1032 | my_vec.at(i).insert(std::make_pair<Key,Val>(maxval * i + j, 2*(maxval*i + j))); |
1033 | } |
1034 | } |
1035 | |
1036 | tbb::internal::segmented_iterator<nested_vec, std::pair<const Key, Val> > my_si(my_vec); |
1037 | Key ii; |
1038 | for(my_si=my_vec.begin(), ii=0; my_si != my_vec.end(); ++my_si, ++ii) { |
1039 | if(((*my_si).first != ii) || ((*my_si).second != 2*ii)) { |
1040 | found_error = true; |
1041 | REMARK( "ii=%d, (*my_si).first=%d, second=%d\n" ,ii, int((*my_si).first), int((*my_si).second)); |
1042 | } |
1043 | } |
1044 | |
1045 | tbb::internal::segmented_iterator<nested_vec, const std::pair<const Key, Val> > my_csi(my_vec); |
1046 | for(my_csi=my_vec.begin(), ii=0; my_csi != my_vec.end(); ++my_csi, ++ii) { |
1047 | if(((*my_csi).first != ii) || ((*my_csi).second != 2*ii)) { |
1048 | found_error = true; |
1049 | REMARK( "ii=%d, (*my_csi).first=%d, second=%d\n" ,ii, int((*my_csi).first), int((*my_csi).second)); |
1050 | } |
1051 | } |
1052 | if(found_error) REPORT("segmented_iterator_map failed\n" ); |
1053 | } |
1054 | |
1055 | void run_segmented_iterator_tests() { |
1056 | // only the following containers can be used with the segmented iterator. |
1057 | REMARK("Running Segmented Iterator Tests\n" ); |
1058 | flog_segmented_interator<std::vector< int > >(); |
1059 | flog_segmented_interator<std::vector< double > >(); |
1060 | flog_segmented_interator<std::deque< int > >(); |
1061 | flog_segmented_interator<std::deque< double > >(); |
1062 | flog_segmented_interator<std::list< int > >(); |
1063 | flog_segmented_interator<std::list< double > >(); |
1064 | |
1065 | flog_segmented_iterator_map<int, int>(); |
1066 | flog_segmented_iterator_map<int, double>(); |
1067 | } |
1068 | |
1069 | template<typename T, template<class> class Allocator, typename Init> |
1070 | tbb::enumerable_thread_specific<T,Allocator<T> > MakeETS( Init init ) { |
1071 | return tbb::enumerable_thread_specific<T,Allocator<T> >(init); |
1072 | } |
1073 | #if __TBB_ETS_USE_CPP11 |
1074 | // In some GCC versions, parameter packs in lambdas might cause compile errors |
1075 | template<typename ETS, typename... P> |
1076 | struct MakeETS_Functor { |
1077 | ETS operator()( typename tbb::internal::strip<P>::type&&... params ) { |
1078 | return ETS(std::move(params)...); |
1079 | } |
1080 | }; |
1081 | template<typename T, template<class> class Allocator, typename... P> |
1082 | tbb::enumerable_thread_specific<T,Allocator<T> > MakeETS( tbb::internal::stored_pack<P...> pack ) { |
1083 | typedef tbb::enumerable_thread_specific<T,Allocator<T> > result_type; |
1084 | return tbb::internal::call_and_return< result_type >( |
1085 | MakeETS_Functor<result_type,P...>(), std::move(pack) |
1086 | ); |
1087 | } |
1088 | #endif |
1089 | |
1090 | template<typename T, template<class> class Allocator, typename InitSrc, typename InitDst, typename Validator> |
1091 | void ets_copy_assign_test( InitSrc init1, InitDst init2, Validator check, const char *allocator_name ) { |
1092 | typedef tbb::enumerable_thread_specific<T, Allocator<T> > ets_type; |
1093 | |
1094 | // Create the source instance |
1095 | const ets_type& cref_binder = MakeETS<T, Allocator>(init1); |
1096 | ets_type& source = const_cast<ets_type&>(cref_binder); |
1097 | check(check_alignment(source.local(),allocator_name)); |
1098 | |
1099 | // Test copy construction |
1100 | bool existed = false; |
1101 | ets_type copy(source); |
1102 | check(check_alignment(copy.local(existed),allocator_name)); |
1103 | ASSERT(existed, "Local data not created by ETS copy constructor" ); |
1104 | copy.clear(); |
1105 | check(check_alignment(copy.local(),allocator_name)); |
1106 | |
1107 | // Test assignment |
1108 | existed = false; |
1109 | ets_type assign(init2); |
1110 | assign = source; |
1111 | check(check_alignment(assign.local(existed),allocator_name)); |
1112 | ASSERT(existed, "Local data not created by ETS assignment" ); |
1113 | assign.clear(); |
1114 | check(check_alignment(assign.local(),allocator_name)); |
1115 | |
1116 | #if __TBB_ETS_USE_CPP11 |
1117 | // Create the source instance |
1118 | ets_type&& rvref_binder = MakeETS<T, Allocator>(init1); |
1119 | check(check_alignment(rvref_binder.local(),allocator_name)); |
1120 | |
1121 | // Test move construction |
1122 | existed = false; |
1123 | ets_type moved(rvref_binder); |
1124 | check(check_alignment(moved.local(existed),allocator_name)); |
1125 | ASSERT(existed, "Local data not created by ETS move constructor" ); |
1126 | moved.clear(); |
1127 | check(check_alignment(moved.local(),allocator_name)); |
1128 | |
1129 | // Test assignment |
1130 | existed = false; |
1131 | ets_type move_assign(init2); |
1132 | move_assign = std::move(moved); |
1133 | check(check_alignment(move_assign.local(existed),allocator_name)); |
1134 | ASSERT(existed, "Local data not created by ETS move assignment" ); |
1135 | move_assign.clear(); |
1136 | check(check_alignment(move_assign.local(),allocator_name)); |
1137 | #endif |
1138 | } |
1139 | |
1140 | template<typename T, int Expected> |
1141 | struct Validator { |
1142 | void operator()( const T& value ) { |
1143 | ASSERT(test_helper<T>::get(value) == Expected, NULL); |
1144 | } |
1145 | void operator()( const std::pair<int,T>& value ) { |
1146 | ASSERT(value.first > 0, NULL); |
1147 | ASSERT(test_helper<T>::get(value.second) == Expected*value.first, NULL); |
1148 | } |
1149 | }; |
1150 | |
1151 | template <typename T, template<class> class Allocator> |
1152 | void run_assign_and_copy_constructor_test(const char *test_name, const char *allocator_name) { |
1153 | REMARK("Testing assignment and copy construction for %s with allocator %s\n" , test_name, allocator_name); |
1154 | #define EXPECTED 3142 |
1155 | |
1156 | // test with exemplar initializer |
1157 | T src_init; |
1158 | test_helper<T>::set(src_init,EXPECTED); |
1159 | T other_init; |
1160 | test_helper<T>::init(other_init); |
1161 | ets_copy_assign_test<T, Allocator>(src_init, other_init, Validator<T,EXPECTED>(), allocator_name); |
1162 | |
1163 | // test with function initializer |
1164 | FunctorFinit<T,EXPECTED> src_finit(SecretTag); |
1165 | FunctorFinit<T,0> other_finit(SecretTag); |
1166 | ets_copy_assign_test<T, Allocator>(src_finit, other_finit, Validator<T,EXPECTED>(), allocator_name); |
1167 | |
1168 | #if __TBB_ETS_USE_CPP11 |
1169 | // test with multi-argument "emplace" initializer |
1170 | // The arguments are wrapped into tbb::internal::stored_pack to avoid variadic templates in ets_copy_assign_test. |
1171 | test_helper<T>::set(src_init,EXPECTED*17); |
1172 | ets_copy_assign_test< std::pair<int,T>, Allocator>(tbb::internal::save_pack(17,src_init), std::make_pair(-1,T()), Validator<T,EXPECTED>(), allocator_name); |
1173 | #endif |
1174 | #undef EXPECTED |
1175 | } |
1176 | |
1177 | template< template<class> class Allocator> |
1178 | void run_assignment_and_copy_constructor_tests(const char* allocator_name) { |
1179 | REMARK("Running assignment and copy constructor tests\n" ); |
1180 | run_assign_and_copy_constructor_test<int, Allocator>("int" , allocator_name); |
1181 | run_assign_and_copy_constructor_test<double, Allocator>("double" , allocator_name); |
1182 | // Try class sizes that are close to a cache line in size, in order to check padding calculations. |
1183 | run_assign_and_copy_constructor_test<minimal<line_size-1>, Allocator >("minimal<line_size-1>" , allocator_name); |
1184 | run_assign_and_copy_constructor_test<minimal<line_size>, Allocator >("minimal<line_size>" , allocator_name); |
1185 | run_assign_and_copy_constructor_test<minimal<line_size+1>, Allocator >("minimal<line_size+1>" , allocator_name); |
1186 | ASSERT(FinitCounter==0, NULL); |
1187 | } |
1188 | |
1189 | // Class with no default constructor |
1190 | class HasNoDefaultConstructor { |
1191 | HasNoDefaultConstructor(); |
1192 | public: |
1193 | HasNoDefaultConstructor( SecretTagType ) {} |
1194 | }; |
1195 | // Initialization functor for HasNoDefaultConstructor |
1196 | struct HasNoDefaultConstructorFinit { |
1197 | HasNoDefaultConstructor operator()() { |
1198 | return HasNoDefaultConstructor(SecretTag); |
1199 | } |
1200 | }; |
1201 | // Combine functor for HasNoDefaultConstructor |
1202 | struct HasNoDefaultConstructorCombine { |
1203 | HasNoDefaultConstructor operator()( HasNoDefaultConstructor, HasNoDefaultConstructor ) { |
1204 | return HasNoDefaultConstructor(SecretTag); |
1205 | } |
1206 | }; |
1207 | |
1208 | #if __TBB_ETS_USE_CPP11 |
1209 | // Class that only has a constructor with multiple parameters and a move constructor |
1210 | class HasSpecialAndMoveCtor : NoCopy { |
1211 | HasSpecialAndMoveCtor(); |
1212 | public: |
1213 | HasSpecialAndMoveCtor( SecretTagType, size_t = size_t(0), const char* = "" ) {} |
1214 | HasSpecialAndMoveCtor( HasSpecialAndMoveCtor&& ) {} |
1215 | }; |
1216 | #endif |
1217 | |
1218 | // No-op combine-each functor |
1219 | template<typename V> |
1220 | struct EmptyCombineEach { |
1221 | void operator()( const V& ) { } |
1222 | }; |
1223 | |
1224 | int |
1225 | align_val(void * const p) { |
1226 | size_t tmp = (size_t)p; |
1227 | int a = 1; |
1228 | while((tmp&0x1) == 0) { a <<=1; tmp >>= 1; } |
1229 | return a; |
1230 | } |
1231 | |
1232 | bool is_between(void* lowp, void *highp, void *testp) { |
1233 | if((size_t)lowp < (size_t)testp && (size_t)testp < (size_t)highp) return true; |
1234 | return (size_t)lowp > (size_t)testp && (size_t)testp > (size_t)highp; |
1235 | } |
1236 | |
1237 | template<class U> struct alignment_of { |
1238 | typedef struct { char t; U padded; } test_alignment; |
1239 | static const size_t value = sizeof(test_alignment) - sizeof(U); |
1240 | }; |
1241 | using tbb::interface6::internal::ets_element; |
1242 | template<typename T, typename OtherType> |
1243 | void allocate_ets_element_on_stack(const char *name) { |
1244 | typedef T aligning_element_type; |
1245 | const size_t my_align = alignment_of<aligning_element_type>::value; |
1246 | OtherType c1; |
1247 | ets_element<aligning_element_type> my_stack_element; |
1248 | OtherType c2; |
1249 | ets_element<aligning_element_type> my_stack_element2; |
1250 | struct { |
1251 | OtherType cxx; |
1252 | ets_element<aligning_element_type> my_struct_element; |
1253 | } mystruct1; |
1254 | tbb::internal::suppress_unused_warning(c1,c2); |
1255 | REMARK("using %s, c1 address == %lx (alignment %d), c2 address == %lx (alignment %d)\n" , name, &c1, align_val(&c1), &c2, align_val(&c2)); |
1256 | REMARK(" ---- my_align == %d\n" , (int)my_align); |
1257 | REMARK(" my_stack_element == %lx (alignment %d), my_stack_element2 == %lx (alignment %d)\n" , |
1258 | &my_stack_element, align_val(&my_stack_element), &my_stack_element2, align_val(&my_stack_element2)); |
1259 | if(is_between(&c1,&c2,&my_stack_element)) REMARK("my_struct_element is in the middle\n" ); |
1260 | if(is_between(&c1,&c2,&my_stack_element2)) REMARK("my_struct_element2 is in the middle\n" ); |
1261 | if(!is_between(&c1,&c2,&my_stack_element) && !is_between(&c1,&c2,&my_stack_element2)) REMARK("stack vars reorganized\n" ); |
1262 | REMARK(" structure field address == %lx, alignment %d\n" , |
1263 | mystruct1.my_struct_element.value(), |
1264 | align_val(mystruct1.my_struct_element.value()) |
1265 | ); |
1266 | ASSERT(tbb::internal::is_aligned(my_stack_element.value(), my_align), "Error in first stack alignment" ); |
1267 | ASSERT(tbb::internal::is_aligned(my_stack_element2.value(), my_align), "Error in second stack alignment" ); |
1268 | ASSERT(tbb::internal::is_aligned(mystruct1.my_struct_element.value(), my_align), "Error in struct element alignment" ); |
1269 | } |
1270 | |
1271 | //! Test situations where only default constructor or copy constructor is required. |
1272 | template<template<class> class Allocator> |
1273 | void TestInstantiation(const char *allocator_name) { |
1274 | REMARK("TestInstantiation<%s>\n" , allocator_name); |
1275 | // Test instantiation is possible when copy constructor is not required. |
1276 | tbb::enumerable_thread_specific<NoCopy, Allocator<NoCopy> > ets1; |
1277 | ets1.local(); |
1278 | ets1.combine_each(EmptyCombineEach<NoCopy>()); |
1279 | |
1280 | // Test instantiation when default constructor is not required, because exemplar is provided. |
1281 | HasNoDefaultConstructor x(SecretTag); |
1282 | tbb::enumerable_thread_specific<HasNoDefaultConstructor, Allocator<HasNoDefaultConstructor> > ets2(x); |
1283 | ets2.local(); |
1284 | ets2.combine(HasNoDefaultConstructorCombine()); |
1285 | |
1286 | // Test instantiation when default constructor is not required, because init function is provided. |
1287 | HasNoDefaultConstructorFinit f; |
1288 | tbb::enumerable_thread_specific<HasNoDefaultConstructor, Allocator<HasNoDefaultConstructor> > ets3(f); |
1289 | ets3.local(); |
1290 | ets3.combine(HasNoDefaultConstructorCombine()); |
1291 | |
1292 | #if __TBB_ETS_USE_CPP11 |
1293 | // Test instantiation with multiple arguments |
1294 | tbb::enumerable_thread_specific<HasSpecialAndMoveCtor, Allocator<HasSpecialAndMoveCtor> > ets4(SecretTag, 0x42, "meaningless" ); |
1295 | ets4.local(); |
1296 | ets4.combine_each(EmptyCombineEach<HasSpecialAndMoveCtor>()); |
1297 | // Test instantiation with one argument that should however use the variadic constructor |
1298 | tbb::enumerable_thread_specific<HasSpecialAndMoveCtor, Allocator<HasSpecialAndMoveCtor> > ets5(SecretTag); |
1299 | ets5.local(); |
1300 | ets5.combine_each(EmptyCombineEach<HasSpecialAndMoveCtor>()); |
1301 | // Test that move operations do not impose extra requirements |
1302 | // Default allocator is used. If it does not match Allocator, there will be elementwise move |
1303 | tbb::enumerable_thread_specific<HasSpecialAndMoveCtor> ets6( std::move(ets4) ); |
1304 | ets6.combine_each(EmptyCombineEach<HasSpecialAndMoveCtor>()); |
1305 | ets6 = std::move(ets5); |
1306 | #endif |
1307 | } |
1308 | |
1309 | class BigType { |
1310 | public: |
1311 | BigType() { /* avoid cl warning C4345 about default initialization of POD types */ } |
1312 | char my_data[12 * 1024 * 1024]; |
1313 | }; |
1314 | |
1315 | template<template<class> class Allocator> |
1316 | void TestConstructorWithBigType(const char *allocator_name) { |
1317 | typedef tbb::enumerable_thread_specific<BigType, Allocator<BigType> > CounterBigType; |
1318 | REMARK("TestConstructorWithBigType<%s>\n" , allocator_name); |
1319 | // Test default constructor |
1320 | CounterBigType MyCounters; |
1321 | // Create a local instance. |
1322 | typename CounterBigType::reference my_local = MyCounters.local(); |
1323 | my_local.my_data[0] = 'a'; |
1324 | // Test copy constructor |
1325 | CounterBigType MyCounters2(MyCounters); |
1326 | ASSERT(check_alignment(MyCounters2.local(), allocator_name).my_data[0]=='a', NULL); |
1327 | } |
1328 | |
1329 | int TestMain () { |
1330 | size_t tbb_allocator_mask; |
1331 | size_t cache_allocator_mask = tbb::internal::NFS_GetLineSize(); |
1332 | REMARK("estimatedCacheLineSize == %d, NFS_GetLineSize() returns %d\n" , |
1333 | (int)estimatedCacheLineSize, (int)tbb::internal::NFS_GetLineSize()); |
1334 | //TODO: use __TBB_alignof(T) to check for local() results instead of using internal knowledges of ets element padding |
1335 | if(tbb::tbb_allocator<int>::allocator_type() == tbb::tbb_allocator<int>::standard) { |
1336 | // scalable allocator is not available. |
1337 | tbb_allocator_mask = 1; |
1338 | REMARK("tbb::tbb_allocator is not available\n" ); |
1339 | } |
1340 | else { |
1341 | // this value is for large objects, but will be correct for small. |
1342 | tbb_allocator_mask = estimatedCacheLineSize; |
1343 | } |
1344 | AlignMask = cache_allocator_mask; |
1345 | TestInstantiation<tbb::cache_aligned_allocator>("tbb::cache_aligned_allocator" ); |
1346 | AlignMask = tbb_allocator_mask; |
1347 | TestInstantiation<tbb::tbb_allocator>("tbb::tbb_allocator" ); |
1348 | AlignMask = cache_allocator_mask; |
1349 | run_assignment_and_copy_constructor_tests<tbb::cache_aligned_allocator>("tbb::cache_aligned_allocator" ); |
1350 | AlignMask = tbb_allocator_mask; |
1351 | run_assignment_and_copy_constructor_tests<tbb::tbb_allocator>("tbb::tbb_allocator" ); |
1352 | run_segmented_iterator_tests(); |
1353 | flog_key_creation_and_deletion(); |
1354 | |
1355 | if (MinThread == 0) { |
1356 | run_serial_tests(); |
1357 | MinThread = 1; |
1358 | } |
1359 | if (MaxThread > 0) { |
1360 | AlignMask = cache_allocator_mask; |
1361 | run_parallel_tests<tbb::cache_aligned_allocator>("tbb::cache_aligned_allocator" ); |
1362 | AlignMask = tbb_allocator_mask; |
1363 | run_parallel_tests<tbb::tbb_allocator>("tbb::tbb_allocator" ); |
1364 | run_cross_type_tests(); |
1365 | } |
1366 | |
1367 | AlignMask = cache_allocator_mask; |
1368 | TestConstructorWithBigType<tbb::cache_aligned_allocator>("tbb::cache_aligned_allocator" ); |
1369 | AlignMask = tbb_allocator_mask; |
1370 | TestConstructorWithBigType<tbb::tbb_allocator>("tbb::tbb_allocator" ); |
1371 | |
1372 | allocate_ets_element_on_stack<int,char>("int vs. char" ); |
1373 | allocate_ets_element_on_stack<int,short>("int vs. short" ); |
1374 | allocate_ets_element_on_stack<int,char[3]>("int vs. char[3]" ); |
1375 | allocate_ets_element_on_stack<float,char>("float vs. char" ); |
1376 | allocate_ets_element_on_stack<float,short>("float vs. short" ); |
1377 | allocate_ets_element_on_stack<float,char[3]>("float vs. char[3]" ); |
1378 | |
1379 | return Harness::Done; |
1380 | } |
1381 | |