1/*
2 * VCluster_semantic_unit_test.hpp
3 *
4 * Created on: Feb 8, 2016
5 * Author: i-bird
6 */
7
8#define BOOST_TEST_DYN_LINK
9#include <boost/test/unit_test.hpp>
10#include "Grid/grid_util_test.hpp"
11#include "data_type/aggregate.hpp"
12#include "VCluster/cuda/VCluster_semantic_unit_tests_funcs.hpp"
13
14constexpr int NBX = 1;
15constexpr int NBX_ASYNC = 2;
16
17//! Example structure
18struct Aexample
19{
20 //! Example size_t
21 size_t a;
22
23 //! Example float
24 float b;
25
26 //! Example double
27 double c;
28};
29
30
31BOOST_AUTO_TEST_SUITE( VCluster_semantic_test )
32
33BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather)
34{
35 for (size_t i = 0 ; i < 100 ; i++)
36 {
37 Vcluster<> & vcl = create_vcluster();
38
39 if (vcl.getProcessUnitID() == 0 && i == 0)
40 std::cout << "Semantic gather test start" << std::endl;
41
42 if (vcl.getProcessingUnits() >= 32)
43 return;
44
45 //! [Gather the data on master]
46
47 openfpm::vector<size_t> v1;
48 v1.resize(vcl.getProcessUnitID());
49
50 for(size_t i = 0 ; i < vcl.getProcessUnitID() ; i++)
51 {v1.get(i) = 5;}
52
53 openfpm::vector<size_t> v2;
54
55 vcl.SGather(v1,v2,(i%vcl.getProcessingUnits()));
56
57 //! [Gather the data on master]
58
59 if (vcl.getProcessUnitID() == (i%vcl.getProcessingUnits()))
60 {
61 size_t n = vcl.getProcessingUnits();
62 BOOST_REQUIRE_EQUAL(v2.size(),n*(n-1)/2);
63
64 bool is_five = true;
65 for (size_t i = 0 ; i < v2.size() ; i++)
66 is_five &= (v2.get(i) == 5);
67
68 BOOST_REQUIRE_EQUAL(is_five,true);
69 }
70 }
71}
72
73BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_2)
74{
75 for (size_t i = 0 ; i < 100 ; i++)
76 {
77 Vcluster<> & vcl = create_vcluster();
78
79 if (vcl.getProcessingUnits() >= 32)
80 return;
81
82 //! [Gather the data on master complex]
83
84 openfpm::vector<size_t> v1;
85 v1.resize(vcl.getProcessUnitID());
86
87 for(size_t i = 0 ; i < vcl.getProcessUnitID() ; i++)
88 {v1.get(i) = 5;}
89
90 openfpm::vector<openfpm::vector<size_t>> v2;
91
92 vcl.SGather(v1,v2,0);
93
94 //! [Gather the data on master complex]
95
96 if (vcl.getProcessUnitID() == 0)
97 {
98 size_t n = vcl.getProcessingUnits();
99 BOOST_REQUIRE_EQUAL(v2.size(),n);
100
101 bool is_five = true;
102 for (size_t i = 0 ; i < v2.size() ; i++)
103 {
104 for (size_t j = 0 ; j < v2.get(i).size() ; j++)
105 is_five &= (v2.get(i).get(j) == 5);
106 }
107 BOOST_REQUIRE_EQUAL(is_five,true);
108
109 }
110
111 openfpm::vector<openfpm::vector<size_t>> v3;
112
113 vcl.SGather(v1,v3,1);
114
115 if (vcl.getProcessUnitID() == 1)
116 {
117 size_t n = vcl.getProcessingUnits();
118 BOOST_REQUIRE_EQUAL(v3.size(),n-1);
119
120 bool is_five = true;
121 for (size_t i = 0 ; i < v3.size() ; i++)
122 {
123 for (size_t j = 0 ; j < v3.get(i).size() ; j++)
124 is_five &= (v3.get(i).get(j) == 5);
125 }
126 BOOST_REQUIRE_EQUAL(is_five,true);
127
128 }
129 }
130}
131
132BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_3)
133{
134 for (size_t i = 0 ; i < 100 ; i++)
135 {
136 Vcluster<> & vcl = create_vcluster();
137
138 if (vcl.getProcessingUnits() >= 32)
139 {return;}
140
141 openfpm::vector<openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> > v1;
142
143 openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> v1_int;
144 aggregate<float, openfpm::vector<size_t>, Point_test<float>> aggr;
145 openfpm::vector<size_t> v1_int2;
146
147 v1_int2.add((size_t)7);
148 v1_int2.add((size_t)7);
149
150 aggr.template get<0>() = 7;
151 aggr.template get<1>() = v1_int2;
152 Point_test<float> p;
153 p.fill();
154 aggr.template get<2>() = p;
155
156 v1_int.add(aggr);
157 v1_int.add(aggr);
158 v1_int.add(aggr);
159
160 v1.add(v1_int);
161 v1.add(v1_int);
162 v1.add(v1_int);
163 v1.add(v1_int);
164
165 openfpm::vector<openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> > v2;
166
167 vcl.SGather(v1,v2,0);
168
169 if (vcl.getProcessUnitID() == 0)
170 {
171 size_t n = vcl.getProcessingUnits();
172
173 BOOST_REQUIRE_EQUAL(v2.size(),v1.size()*n);
174
175 bool is_seven = true;
176 for (size_t i = 0 ; i < v2.size() ; i++)
177 {
178 for (size_t j = 0 ; j < v2.get(i).size() ; j++)
179 {
180 is_seven &= (v2.get(i).template get<0>(j) == 7);
181
182 for (size_t k = 0; k < v2.get(i).template get<1>(j).size(); k++)
183 is_seven &= (v2.get(i).template get<1>(j).get(k) == 7);
184
185 Point_test<float> p = v2.get(i).template get<2>(j);
186
187 BOOST_REQUIRE(p.template get<0>() == 1);
188 BOOST_REQUIRE(p.template get<1>() == 2);
189 BOOST_REQUIRE(p.template get<2>() == 3);
190 BOOST_REQUIRE(p.template get<3>() == 4);
191
192 for (size_t l = 0 ; l < 3 ; l++)
193 p.template get<4>()[l] = 5;
194
195 for (size_t m = 0 ; m < 3 ; m++)
196 {
197 for (size_t n = 0 ; n < 3 ; n++)
198 {
199 p.template get<5>()[m][n] = 6;
200 }
201 }
202 }
203 }
204 BOOST_REQUIRE_EQUAL(is_seven,true);
205 }
206 }
207}
208
209BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_4)
210{
211 for (size_t i = 0 ; i < 100 ; i++)
212 {
213 Vcluster<> & vcl = create_vcluster();
214
215 if (vcl.getProcessingUnits() >= 32)
216 {return;}
217
218 size_t sz[] = {16,16};
219
220 grid_cpu<2,Point_test<float>> g1(sz);
221 g1.setMemory();
222 fill_grid<2>(g1);
223
224 openfpm::vector<grid_cpu<2,Point_test<float>>> v2;
225
226 vcl.SGather(g1,v2,0);
227
228 typedef Point_test<float> p;
229
230 if (vcl.getProcessUnitID() == 0)
231 {
232 size_t n = vcl.getProcessingUnits();
233 BOOST_REQUIRE_EQUAL(v2.size(),n);
234
235 bool match = true;
236 for (size_t i = 0 ; i < v2.size() ; i++)
237 {
238 auto it = v2.get(i).getIterator();
239
240 while (it.isNext())
241 {
242 grid_key_dx<2> key = it.get();
243
244 match &= (v2.get(i).template get<p::x>(key) == g1.template get<p::x>(key));
245 match &= (v2.get(i).template get<p::y>(key) == g1.template get<p::y>(key));
246 match &= (v2.get(i).template get<p::z>(key) == g1.template get<p::z>(key));
247 match &= (v2.get(i).template get<p::s>(key) == g1.template get<p::s>(key));
248
249 match &= (v2.get(i).template get<p::v>(key)[0] == g1.template get<p::v>(key)[0]);
250 match &= (v2.get(i).template get<p::v>(key)[1] == g1.template get<p::v>(key)[1]);
251 match &= (v2.get(i).template get<p::v>(key)[2] == g1.template get<p::v>(key)[2]);
252
253 match &= (v2.get(i).template get<p::t>(key)[0][0] == g1.template get<p::t>(key)[0][0]);
254 match &= (v2.get(i).template get<p::t>(key)[0][1] == g1.template get<p::t>(key)[0][1]);
255 match &= (v2.get(i).template get<p::t>(key)[0][2] == g1.template get<p::t>(key)[0][2]);
256 match &= (v2.get(i).template get<p::t>(key)[1][0] == g1.template get<p::t>(key)[1][0]);
257 match &= (v2.get(i).template get<p::t>(key)[1][1] == g1.template get<p::t>(key)[1][1]);
258 match &= (v2.get(i).template get<p::t>(key)[1][2] == g1.template get<p::t>(key)[1][2]);
259 match &= (v2.get(i).template get<p::t>(key)[2][0] == g1.template get<p::t>(key)[2][0]);
260 match &= (v2.get(i).template get<p::t>(key)[2][1] == g1.template get<p::t>(key)[2][1]);
261 match &= (v2.get(i).template get<p::t>(key)[2][2] == g1.template get<p::t>(key)[2][2]);
262
263 ++it;
264 }
265
266 }
267 BOOST_REQUIRE_EQUAL(match,true);
268 }
269 }
270}
271
272BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_5)
273{
274 for (size_t i = 0 ; i < 100 ; i++)
275 {
276 Vcluster<> & vcl = create_vcluster();
277
278 if (vcl.size() == 1)
279 {return;}
280
281 if (vcl.getProcessingUnits() >= 32)
282 {return;}
283
284 size_t sz[] = {16,16};
285 grid_cpu<2,Point_test<float>> g1(sz);
286 g1.setMemory();
287 fill_grid<2>(g1);
288 openfpm::vector<grid_cpu<2,Point_test<float>>> v1;
289
290 v1.add(g1);
291 v1.add(g1);
292 v1.add(g1);
293
294 openfpm::vector<grid_cpu<2,Point_test<float>>> v2;
295
296 vcl.SGather(v1,v2,1);
297
298 typedef Point_test<float> p;
299
300 if (vcl.getProcessUnitID() == 1)
301 {
302 size_t n = vcl.getProcessingUnits();
303 BOOST_REQUIRE_EQUAL(v2.size(),v1.size()*n);
304
305 bool match = true;
306 for (size_t i = 0 ; i < v2.size() ; i++)
307 {
308 auto it = v2.get(i).getIterator();
309
310 while (it.isNext())
311 {
312 grid_key_dx<2> key = it.get();
313
314 match &= (v2.get(i).template get<p::x>(key) == g1.template get<p::x>(key));
315 match &= (v2.get(i).template get<p::y>(key) == g1.template get<p::y>(key));
316 match &= (v2.get(i).template get<p::z>(key) == g1.template get<p::z>(key));
317 match &= (v2.get(i).template get<p::s>(key) == g1.template get<p::s>(key));
318
319 match &= (v2.get(i).template get<p::v>(key)[0] == g1.template get<p::v>(key)[0]);
320 match &= (v2.get(i).template get<p::v>(key)[1] == g1.template get<p::v>(key)[1]);
321 match &= (v2.get(i).template get<p::v>(key)[2] == g1.template get<p::v>(key)[2]);
322
323 match &= (v2.get(i).template get<p::t>(key)[0][0] == g1.template get<p::t>(key)[0][0]);
324 match &= (v2.get(i).template get<p::t>(key)[0][1] == g1.template get<p::t>(key)[0][1]);
325 match &= (v2.get(i).template get<p::t>(key)[0][2] == g1.template get<p::t>(key)[0][2]);
326 match &= (v2.get(i).template get<p::t>(key)[1][0] == g1.template get<p::t>(key)[1][0]);
327 match &= (v2.get(i).template get<p::t>(key)[1][1] == g1.template get<p::t>(key)[1][1]);
328 match &= (v2.get(i).template get<p::t>(key)[1][2] == g1.template get<p::t>(key)[1][2]);
329 match &= (v2.get(i).template get<p::t>(key)[2][0] == g1.template get<p::t>(key)[2][0]);
330 match &= (v2.get(i).template get<p::t>(key)[2][1] == g1.template get<p::t>(key)[2][1]);
331 match &= (v2.get(i).template get<p::t>(key)[2][2] == g1.template get<p::t>(key)[2][2]);
332
333 ++it;
334 }
335
336 }
337 BOOST_REQUIRE_EQUAL(match,true);
338 }
339 }
340}
341
342BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_6)
343{
344 for (size_t i = 0 ; i < 100 ; i++)
345 {
346 Vcluster<> & vcl = create_vcluster();
347
348 if (vcl.getProcessingUnits() >= 32)
349 {return;}
350
351 openfpm::vector<openfpm::vector<openfpm::vector<size_t>>> v1;
352 openfpm::vector<openfpm::vector<size_t>> v1_int;
353 openfpm::vector<size_t> v1_int2;
354
355 v1_int2.add((size_t)7);
356 v1_int2.add((size_t)7);
357
358 v1_int.add(v1_int2);
359 v1_int.add(v1_int2);
360 v1_int.add(v1_int2);
361
362 v1.add(v1_int);
363 v1.add(v1_int);
364 v1.add(v1_int);
365 v1.add(v1_int);
366
367 openfpm::vector<openfpm::vector<openfpm::vector<size_t>>> v2;
368
369 vcl.SGather(v1,v2,0);
370
371 if (vcl.getProcessUnitID() == 0)
372 {
373 size_t n = vcl.getProcessingUnits();
374
375 BOOST_REQUIRE_EQUAL(v2.size(),v1.size()*n);
376
377 bool is_seven = true;
378 for (size_t i = 0 ; i < v2.size() ; i++)
379 {
380 for (size_t j = 0 ; j < v2.get(i).size() ; j++)
381 {
382 for (size_t k = 0 ; k < v2.get(i).get(j).size() ; k++)
383 is_seven &= (v2.get(i).get(j).get(k) == 7);
384 }
385 }
386 BOOST_REQUIRE_EQUAL(is_seven,true);
387 }
388 }
389}
390
391BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_7)
392{
393 for (size_t i = 0 ; i < 100 ; i++)
394 {
395 Vcluster<> & vcl = create_vcluster();
396
397 if (vcl.getProcessingUnits() >= 32)
398 {return;}
399
400 openfpm::vector<Point_test<float>> v1;
401
402 Point_test<float> p1;
403 p1.fill();
404
405 v1.resize(vcl.getProcessUnitID());
406
407 for(size_t i = 0 ; i < vcl.getProcessUnitID() ; i++)
408 {v1.get(i) = p1;}
409
410 openfpm::vector<openfpm::vector<Point_test<float>>> v2;
411
412 vcl.SGather(v1,v2,0);
413
414 typedef Point_test<float> p;
415
416 if (vcl.getProcessUnitID() == 0)
417 {
418 size_t n = vcl.getProcessingUnits();
419 BOOST_REQUIRE_EQUAL(v2.size(),n);
420
421 bool match = true;
422
423 for (size_t i = 0 ; i < v2.size() ; i++)
424 {
425 for (size_t j = 0 ; j < v2.get(i).size() ; j++)
426 {
427 Point_test<float> p2 = v2.get(i).get(j);
428 //BOOST_REQUIRE(p2 == p1);
429
430 match &= (p2.template get<p::x>() == p1.template get<p::x>());
431 match &= (p2.template get<p::y>() == p1.template get<p::y>());
432 match &= (p2.template get<p::z>() == p1.template get<p::z>());
433 match &= (p2.template get<p::s>() == p1.template get<p::s>());
434
435 match &= (p2.template get<p::v>()[0] == p1.template get<p::v>()[0]);
436 match &= (p2.template get<p::v>()[1] == p1.template get<p::v>()[1]);
437 match &= (p2.template get<p::v>()[2] == p1.template get<p::v>()[2]);
438
439 match &= (p2.template get<p::t>()[0][0] == p1.template get<p::t>()[0][0]);
440 match &= (p2.template get<p::t>()[0][1] == p1.template get<p::t>()[0][1]);
441 match &= (p2.template get<p::t>()[0][2] == p1.template get<p::t>()[0][2]);
442 match &= (p2.template get<p::t>()[1][0] == p1.template get<p::t>()[1][0]);
443 match &= (p2.template get<p::t>()[1][1] == p1.template get<p::t>()[1][1]);
444 match &= (p2.template get<p::t>()[1][2] == p1.template get<p::t>()[1][2]);
445 match &= (p2.template get<p::t>()[2][0] == p1.template get<p::t>()[2][0]);
446 match &= (p2.template get<p::t>()[2][1] == p1.template get<p::t>()[2][1]);
447 match &= (p2.template get<p::t>()[2][2] == p1.template get<p::t>()[2][2]);
448 }
449 }
450 BOOST_REQUIRE_EQUAL(match,true);
451 }
452 }
453}
454
455BOOST_AUTO_TEST_CASE (Vcluster_semantic_gather_8)
456{
457 for (size_t i = 0 ; i < 100 ; i++)
458 {
459 Vcluster<> & vcl = create_vcluster();
460
461 if (vcl.getProcessingUnits() >= 32)
462 {return;}
463
464 openfpm::vector<Box<3,size_t>> v1;
465
466 Box<3,size_t> bx;
467 bx.setLow(0, 1);
468 bx.setLow(1, 2);
469 bx.setLow(2, 3);
470 bx.setHigh(0, 4);
471 bx.setHigh(1, 5);
472 bx.setHigh(2, 6);
473
474
475 v1.resize(vcl.getProcessUnitID());
476
477 for(size_t i = 0 ; i < vcl.getProcessUnitID() ; i++)
478 v1.get(i) = bx;
479
480 openfpm::vector<openfpm::vector<Box<3,size_t>>> v2;
481
482 vcl.SGather(v1,v2,0);
483
484 if (vcl.getProcessUnitID() == 0)
485 {
486 size_t n = vcl.getProcessingUnits();
487 BOOST_REQUIRE_EQUAL(v2.size(),n);
488
489 for (size_t i = 0 ; i < v2.size() ; i++)
490 {
491 for (size_t j = 0 ; j < v2.get(i).size() ; j++)
492 {
493 Box<3,size_t> b2 = v2.get(i).get(j);
494 BOOST_REQUIRE(bx == b2);
495 }
496 }
497 }
498 }
499}
500
501BOOST_AUTO_TEST_CASE (Vcluster_semantic_struct_gather)
502{
503 for (size_t i = 0 ; i < 100 ; i++)
504 {
505 Vcluster<> & vcl = create_vcluster();
506
507 if (vcl.getProcessingUnits() >= 32)
508 return;
509
510 openfpm::vector<Aexample> v1;
511 v1.resize(vcl.getProcessUnitID());
512
513 for(size_t i = 0 ; i < vcl.getProcessUnitID() ; i++)
514 {
515 v1.get(i).a = 5;
516 v1.get(i).b = 10.0;
517 v1.get(i).c = 11.0;
518 }
519
520 openfpm::vector<Aexample> v2;
521
522 vcl.SGather(v1,v2,(i%vcl.getProcessingUnits()));
523
524 if (vcl.getProcessUnitID() == (i%vcl.getProcessingUnits()))
525 {
526 size_t n = vcl.getProcessingUnits();
527 BOOST_REQUIRE_EQUAL(v2.size(),n*(n-1)/2);
528
529 bool is_correct = true;
530 for (size_t i = 0 ; i < v2.size() ; i++)
531 {
532 is_correct &= (v2.get(i).a == 5);
533 is_correct &= (v2.get(i).b == 10.0);
534 is_correct &= (v2.get(i).c == 11.0);
535 }
536
537 BOOST_REQUIRE_EQUAL(is_correct,true);
538 }
539 if (vcl.getProcessUnitID() == 0 && i == 99)
540 std::cout << "Semantic gather test stop" << std::endl;
541 }
542}
543
544
545BOOST_AUTO_TEST_CASE (Vcluster_semantic_layout_inte_gather)
546{
547 test_different_layouts<HeapMemory,memory_traits_inte>();
548 test_different_layouts<HeapMemory,memory_traits_lin>();
549}
550
551#define SSCATTER_MAX 7
552
553BOOST_AUTO_TEST_CASE (Vcluster_semantic_scatter)
554{
555 for (size_t i = 0 ; i < 100 ; i++)
556 {
557 Vcluster<> & vcl = create_vcluster();
558
559 if (vcl.getProcessingUnits() >= 32)
560 {return;}
561
562 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
563 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
564 nr = ((nr-1) * nr) / 2;
565
566 size_t n_elements = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
567
568 openfpm::vector<size_t> v1;
569 v1.resize(n_elements);
570
571 for(size_t i = 0 ; i < n_elements ; i++)
572 {v1.get(i) = 5;}
573
574 //! [Scatter the data from master]
575
576 openfpm::vector<size_t> v2;
577
578 openfpm::vector<size_t> prc;
579 openfpm::vector<size_t> sz;
580
581 // Scatter pattern
582 for (size_t i = 0 ; i < vcl.getProcessingUnits() ; i++)
583 {
584 sz.add(i % SSCATTER_MAX);
585 prc.add(i);
586 }
587
588 vcl.SScatter(v1,v2,prc,sz,(i%vcl.getProcessingUnits()));
589
590 //! [Scatter the data from master]
591
592 BOOST_REQUIRE_EQUAL(v2.size(),vcl.getProcessUnitID() % SSCATTER_MAX);
593
594 bool is_five = true;
595 for (size_t i = 0 ; i < v2.size() ; i++)
596 is_five &= (v2.get(i) == 5);
597
598 BOOST_REQUIRE_EQUAL(is_five,true);
599 }
600}
601
602
603BOOST_AUTO_TEST_CASE (Vcluster_semantic_struct_scatter)
604{
605 for (size_t i = 0 ; i < 100 ; i++)
606 {
607 Vcluster<> & vcl = create_vcluster();
608
609 if (vcl.getProcessingUnits() >= 32)
610 {return;}
611
612 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
613 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
614 nr = ((nr-1) * nr) / 2;
615
616 size_t n_elements = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
617
618 openfpm::vector<size_t> v1;
619 v1.resize(n_elements);
620
621 for(size_t i = 0 ; i < n_elements ; i++)
622 v1.get(i) = 5;
623
624 openfpm::vector<size_t> v2;
625
626 openfpm::vector<size_t> prc;
627 openfpm::vector<size_t> sz;
628
629 // Scatter pattern
630 for (size_t i = 0 ; i < vcl.getProcessingUnits() ; i++)
631 {
632 sz.add(i % SSCATTER_MAX);
633 prc.add(i);
634 }
635
636 vcl.SScatter(v1,v2,prc,sz,(i%vcl.getProcessingUnits()));
637
638 if (vcl.getProcessUnitID() == (i%vcl.getProcessingUnits()))
639 {
640 BOOST_REQUIRE_EQUAL(v2.size(),vcl.getProcessUnitID() % SSCATTER_MAX);
641
642 bool is_five = true;
643 for (size_t i = 0 ; i < v2.size() ; i++)
644 is_five &= (v2.get(i) == 5);
645
646 BOOST_REQUIRE_EQUAL(is_five,true);
647 }
648 }
649}
650
651template<unsigned int impl, typename VCluster_type, typename vector1, typename vector2, typename vector3>
652void scomm_unknown(VCluster_type & vcl, vector1 & v1, vector2 & v2, vector3 & prc_send, vector3 & prc_recv, vector3 & sz_recv)
653{
654 if (impl == NBX)
655 {
656 // Send and receive from the other processor v2 container the received data
657 // Because in this case v2 is an openfpm::vector<size_t>, all the received
658 // vector are concatenated one over the other. For example if the processor receive 3 openfpm::vector<size_t>
659 // each having 3,4,5 elements. v2 will be a vector of 12 elements
660 vcl.SSendRecv(v1,v2,prc_send,prc_recv,sz_recv);
661 }
662 else
663 {
664 vcl.SSendRecvAsync(v1,v2,prc_send,prc_recv,sz_recv);
665
666 vcl.progressCommunication();
667 usleep(1000);
668 vcl.progressCommunication();
669 usleep(10000);
670 vcl.progressCommunication();
671 usleep(1000);
672
673 vcl.SSendRecvWait(v1,v2,prc_send,prc_recv,sz_recv);
674 }
675}
676
677
678template<unsigned int impl>
679void Vcluster_semantic_sendrecv_all_unknown_impl()
680{
681 openfpm::vector<size_t> prc_recv2;
682 openfpm::vector<size_t> prc_recv3;
683
684 openfpm::vector<size_t> sz_recv2;
685 openfpm::vector<size_t> sz_recv3;
686
687 for (size_t i = 0 ; i < 100 ; i++)
688 {
689 Vcluster<> & vcl = create_vcluster();
690
691 if (vcl.getProcessUnitID() == 0 && i == 0)
692 std::cout << "Semantic sendrecv test start" << std::endl;
693
694
695 if (vcl.getProcessingUnits() >= 32)
696 {return;}
697
698 prc_recv2.clear();
699 prc_recv3.clear();
700 openfpm::vector<size_t> prc_send;
701 sz_recv2.clear();
702 sz_recv3.clear();
703
704 //! [dsde with complex objects1]
705
706 // A vector of vector we want to send each internal vector to one specified processor
707 openfpm::vector<openfpm::vector<size_t>> v1;
708
709 // We use this empty vector to receive data
710 openfpm::vector<size_t> v2;
711
712 // We use this empty vector to receive data
713 openfpm::vector<openfpm::vector<size_t>> v3;
714
715 // in this case each processor will send a message of different size to all the other processor
716 // but can also be a subset of processors
717 v1.resize(vcl.getProcessingUnits());
718
719 // We fill the send buffer with some sense-less data
720 for(size_t i = 0 ; i < v1.size() ; i++)
721 {
722 // each vector is filled with a different message size
723 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
724 v1.get(i).add(j);
725
726 // generate the sending list (in this case the sendinf list is all the other processor)
727 // but in general can be some of them and totally random
728 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
729 }
730
731 // Send and receive from the other processor v2 container the received data
732 // Because in this case v2 is an openfpm::vector<size_t>, all the received
733 // vector are concatenated one over the other. For example if the processor receive 3 openfpm::vector<size_t>
734 // each having 3,4,5 elements. v2 will be a vector of 12 elements
735 //vcl.SSendRecv(v1,v2,prc_send,prc_recv2,sz_recv2);
736 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
737
738 // Send and receive from the other processors v2 contain the received data
739 // Because in this case v2 is an openfpm::vector<openfpm::vector<size_t>>, all the vector from
740 // each processor will be collected. For example if the processor receive 3 openfpm::vector<size_t>
741 // each having 3,4,5 elements. v2 will be a vector of vector of 3 elements (openfpm::vector) and
742 // each element will be respectivly 3,4,5 elements
743
744 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
745
746 //! [dsde with complex objects1]
747
748 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
749 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
750 nr = ((nr-1) * nr) / 2;
751
752 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
753
754 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
755 size_t nc_check = (vcl.getProcessingUnits()-1) / SSCATTER_MAX;
756 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits()-1-nc_check);
757
758 bool match = true;
759 size_t s = 0;
760
761 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
762 {
763 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
764 {
765 match &= v2.get(s+j) == j;
766 }
767 s += sz_recv2.get(i);
768 }
769
770 BOOST_REQUIRE_EQUAL(match,true);
771
772 for (size_t i = 0 ; i < v3.size() ; i++)
773 {
774 for (size_t j = 0 ; j < v3.get(i).size() ; j++)
775 {
776 match &= v3.get(i).get(j) == j;
777 }
778 }
779
780 BOOST_REQUIRE_EQUAL(match,true);
781 }
782}
783
784void Vcluster_semantic_sendrecv_all_unknown_multiple_impl()
785{
786 openfpm::vector<size_t> prc_recv2[NQUEUE];
787 openfpm::vector<size_t> prc_recv3[NQUEUE];
788
789 openfpm::vector<size_t> sz_recv2[NQUEUE];
790 openfpm::vector<size_t> sz_recv3[NQUEUE];
791
792 openfpm::vector<size_t> prc_send[NQUEUE];
793 openfpm::vector<openfpm::vector<size_t>> v1[NQUEUE];
794 openfpm::vector<size_t> v2[NQUEUE];
795 openfpm::vector<openfpm::vector<size_t>> v3[NQUEUE];
796
797 for (size_t i = 0 ; i < 100 ; i++)
798 {
799 Vcluster<> & vcl = create_vcluster();
800
801 for (size_t k = 0 ; k < NQUEUE / 2 ; k++)
802 {
803 if (vcl.getProcessUnitID() == 0 && i == 0)
804 {std::cout << "Semantic sendrecv test start" << std::endl;}
805
806
807 if (vcl.getProcessingUnits() >= 32)
808 {return;}
809
810 prc_recv2[k].clear();
811 prc_recv3[k].clear();
812 prc_send[k].clear();
813 sz_recv2[k].clear();
814 sz_recv3[k].clear();
815
816 //! [dsde with complex objects1]
817
818 // A vector of vector we want to send each internal vector to one specified processor
819 v1[k].clear();
820
821 // We use this empty vector to receive data
822 v2[k].clear();
823
824 // We use this empty vector to receive data
825 v3[k].clear();
826
827 // in this case each processor will send a message of different size to all the other processor
828 // but can also be a subset of processors
829 v1[k].resize(vcl.getProcessingUnits());
830
831 // We fill the send buffer with some sense-less data
832 for(size_t i = 0 ; i < v1[k].size() ; i++)
833 {
834 // each vector is filled with a different message size
835 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
836 {v1[k].get(i).add(j);}
837
838 // generate the sending list (in this case the sendinf list is all the other processor)
839 // but in general can be some of them and totally random
840 prc_send[k].add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
841 }
842
843 // Send and receive from the other processor v2 container the received data
844 // Because in this case v2 is an openfpm::vector<size_t>, all the received
845 // vector are concatenated one over the other. For example if the processor receive 3 openfpm::vector<size_t>
846 // each having 3,4,5 elements. v2 will be a vector of 12 elements
847 //vcl.SSendRecv(v1,v2,prc_send,prc_recv2,sz_recv2);
848
849 vcl.SSendRecvAsync(v1[k],v2[k],prc_send[k],prc_recv2[k],sz_recv2[k]);
850
851 // Send and receive from the other processors v2 contain the received data
852 // Because in this case v2 is an openfpm::vector<openfpm::vector<size_t>>, all the vector from
853 // each processor will be collected. For example if the processor receive 3 openfpm::vector<size_t>
854 // each having 3,4,5 elements. v2 will be a vector of vector of 3 elements (openfpm::vector) and
855 // each element will be respectivly 3,4,5 elements
856
857 vcl.SSendRecvAsync(v1[k],v3[k],prc_send[k],prc_recv3[k],sz_recv3[k]);
858 }
859
860 vcl.progressCommunication();
861 usleep(1000);
862 vcl.progressCommunication();
863 usleep(10000);
864 vcl.progressCommunication();
865 usleep(1000);
866
867 for (size_t k = 0 ; k < NQUEUE / 2 ; k++)
868 {
869 vcl.SSendRecvWait(v1[k],v2[k],prc_send[k],prc_recv2[k],sz_recv2[k]);
870 vcl.SSendRecvWait(v1[k],v3[k],prc_send[k],prc_recv3[k],sz_recv3[k]);
871 }
872
873 //! [dsde with complex objects1]
874
875 for (size_t k = 0 ; k < NQUEUE / 2 ; k++)
876 {
877 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
878 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
879 nr = ((nr-1) * nr) / 2;
880
881 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
882
883 BOOST_REQUIRE_EQUAL(v2[k].size(),n_ele);
884 size_t nc_check = (vcl.getProcessingUnits()-1) / SSCATTER_MAX;
885 BOOST_REQUIRE_EQUAL(v3[k].size(),vcl.getProcessingUnits()-1-nc_check);
886
887 bool match = true;
888 size_t s = 0;
889
890 for (size_t i = 0 ; i < sz_recv2[k].size() ; i++)
891 {
892 for (size_t j = 0 ; j < sz_recv2[k].get(i); j++)
893 {
894 match &= v2[k].get(s+j) == j;
895 }
896 s += sz_recv2[k].get(i);
897 }
898
899 BOOST_REQUIRE_EQUAL(match,true);
900
901 for (size_t i = 0 ; i < v3[k].size() ; i++)
902 {
903 for (size_t j = 0 ; j < v3[k].get(i).size() ; j++)
904 {
905 match &= v3[k].get(i).get(j) == j;
906 }
907 }
908
909 BOOST_REQUIRE_EQUAL(match,true);
910 }
911 }
912}
913
914BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_all_unknown)
915{
916 Vcluster_semantic_sendrecv_all_unknown_impl<NBX>();
917}
918
919BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_all_unknown_async)
920{
921 Vcluster_semantic_sendrecv_all_unknown_impl<NBX_ASYNC>();
922}
923
924BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_all_multiple_unknown)
925{
926 Vcluster_semantic_sendrecv_all_unknown_multiple_impl();
927}
928
929template<unsigned int impl, typename VCluster_type, typename vector1, typename vector2, typename vector3>
930void scomm_known(VCluster_type & vcl, vector1 & v1, vector2 & v2, vector3 & prc_send, vector3 & prc_recv, vector3 & sz_recv)
931{
932 if (impl == NBX)
933 {
934 // Send and receive from the other processor v2 container the received data
935 // Because in this case v2 is an openfpm::vector<size_t>, all the received
936 // vector are concatenated one over the other. For example if the processor receive 3 openfpm::vector<size_t>
937 // each having 3,4,5 elements. v2 will be a vector of 12 elements
938 vcl.SSendRecv(v1,v2,prc_send,prc_recv,sz_recv,RECEIVE_KNOWN | KNOWN_ELEMENT_OR_BYTE);
939 }
940 else
941 {
942 vcl.SSendRecvAsync(v1,v2,prc_send,prc_recv,sz_recv,RECEIVE_KNOWN | KNOWN_ELEMENT_OR_BYTE);
943
944 vcl.progressCommunication();
945 usleep(1000);
946 vcl.progressCommunication();
947 usleep(1000);
948 vcl.progressCommunication();
949 usleep(1000);
950
951 vcl.SSendRecvWait(v1,v2,prc_send,prc_recv,sz_recv,RECEIVE_KNOWN | KNOWN_ELEMENT_OR_BYTE);
952 }
953}
954
955
956template<unsigned int impl>
957void Vcluster_semantic_sendrecv_receive_size_known_impl()
958{
959 openfpm::vector<size_t> prc_recv2;
960 openfpm::vector<size_t> prc_recv3;
961
962 openfpm::vector<size_t> sz_recv2;
963 openfpm::vector<size_t> sz_recv3;
964
965 for (size_t i = 0 ; i < 100 ; i++)
966 {
967 Vcluster<> & vcl = create_vcluster();
968
969 if (vcl.getProcessUnitID() == 0 && i == 0)
970 {std::cout << "Semantic sendrecv test start" << std::endl;}
971
972
973 if (vcl.getProcessingUnits() >= 32)
974 {return;}
975
976 openfpm::vector<size_t> prc_send;
977
978 openfpm::vector<openfpm::vector<size_t>> v1;
979 openfpm::vector<size_t> v2;
980 openfpm::vector<openfpm::vector<size_t>> v3;
981
982 v1.resize(vcl.getProcessingUnits());
983
984 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
985 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
986 nr = ((nr-1) * nr) / 2;
987
988 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
989
990 for(size_t i = 0 ; i < v1.size() ; i++)
991 {
992 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
993 v1.get(i).add(j);
994
995 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
996 }
997
998 // We receive to fill prc_recv2 and sz_recv2
999 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1000
1001 // carefull because SSendRecv does not fill prc_recv2 with processor that has a sending size of 0
1002 for(size_t i = 0 ; i < v1.size() ; i++)
1003 {
1004 if( i % SSCATTER_MAX == 0)
1005 {
1006 prc_recv2.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1007 sz_recv2.add(0);
1008 }
1009 }
1010
1011 // We reset v2 and we receive again saying that the processors are known and we know the elements
1012 v2.clear();
1013 //vcl.SSendRecv(v1,v2,prc_send,prc_recv2,sz_recv2,RECEIVE_KNOWN | KNOWN_ELEMENT_OR_BYTE);
1014 scomm_known<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1015 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1016
1017 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1018 size_t nc_check = (vcl.getProcessingUnits()-1) / SSCATTER_MAX;
1019 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits()-1-nc_check);
1020
1021 bool match = true;
1022 size_t s = 0;
1023
1024 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1025 {
1026 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1027 {
1028 match &= v2.get(s+j) == j;
1029 }
1030 s += sz_recv2.get(i);
1031 }
1032
1033 BOOST_REQUIRE_EQUAL(match,true);
1034
1035 for (size_t i = 0 ; i < v3.size() ; i++)
1036 {
1037 for (size_t j = 0 ; j < v3.get(i).size() ; j++)
1038 {
1039 match &= v3.get(i).get(j) == j;
1040 }
1041 }
1042
1043 BOOST_REQUIRE_EQUAL(match,true);
1044 }
1045}
1046
1047BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_receive_size_known)
1048{
1049 Vcluster_semantic_sendrecv_receive_size_known_impl<NBX>();
1050}
1051
1052BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_receive_size_known_async)
1053{
1054 Vcluster_semantic_sendrecv_receive_size_known_impl<NBX_ASYNC>();
1055}
1056
1057template<unsigned int impl, typename VCluster_type, typename vector1, typename vector2, typename vector3>
1058void scomm_known2(VCluster_type & vcl, vector1 & v1, vector2 & v2, vector3 & prc_send, vector3 & prc_recv, vector3 & sz_recv)
1059{
1060 if (impl == NBX)
1061 {
1062 // Send and receive from the other processor v2 container the received data
1063 // Because in this case v2 is an openfpm::vector<size_t>, all the received
1064 // vector are concatenated one over the other. For example if the processor receive 3 openfpm::vector<size_t>
1065 // each having 3,4,5 elements. v2 will be a vector of 12 elements
1066 vcl.SSendRecv(v1,v2,prc_send,prc_recv,sz_recv,RECEIVE_KNOWN);
1067 }
1068 else
1069 {
1070 vcl.SSendRecvAsync(v1,v2,prc_send,prc_recv,sz_recv,RECEIVE_KNOWN);
1071
1072 vcl.progressCommunication();
1073 usleep(1000);
1074 vcl.progressCommunication();
1075 usleep(1000);
1076 vcl.progressCommunication();
1077 usleep(1000);
1078
1079 vcl.SSendRecvWait(v1,v2,prc_send,prc_recv,sz_recv,RECEIVE_KNOWN);
1080 }
1081}
1082
1083template<unsigned int impl>
1084void Vcluster_semantic_sendrecv_receive_known_impl()
1085{
1086 openfpm::vector<size_t> prc_recv2;
1087 openfpm::vector<size_t> prc_recv3;
1088
1089 openfpm::vector<size_t> sz_recv2;
1090 openfpm::vector<size_t> sz_recv3;
1091
1092 for (size_t i = 0 ; i < 100 ; i++)
1093 {
1094 Vcluster<> & vcl = create_vcluster();
1095
1096 if (vcl.getProcessUnitID() == 0 && i == 0)
1097 {std::cout << "Semantic sendrecv test start" << std::endl;}
1098
1099
1100 if (vcl.getProcessingUnits() >= 32)
1101 {return;}
1102
1103 openfpm::vector<size_t> prc_send;
1104
1105 openfpm::vector<openfpm::vector<size_t>> v1;
1106 openfpm::vector<size_t> v2;
1107 openfpm::vector<openfpm::vector<size_t>> v3;
1108
1109 v1.resize(vcl.getProcessingUnits());
1110
1111 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
1112 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
1113 nr = ((nr-1) * nr) / 2;
1114
1115 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
1116
1117 for(size_t i = 0 ; i < v1.size() ; i++)
1118 {
1119 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
1120 v1.get(i).add(j);
1121
1122 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1123 }
1124
1125 // Receive to fill prc_recv2
1126 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1127
1128 // carefull because SSendRecv does not fill prc_recv2 with processor that has a sending size of 0
1129
1130 for(size_t i = 0 ; i < v1.size() ; i++)
1131 {
1132 if( i % SSCATTER_MAX == 0)
1133 {prc_recv2.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());}
1134 }
1135
1136 // Reset v2 and sz_recv2
1137
1138 v2.clear();
1139 sz_recv2.clear();
1140
1141 scomm_known2<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1142 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1143
1144 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1145 size_t nc_check = (vcl.getProcessingUnits()-1) / SSCATTER_MAX;
1146 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits()-1-nc_check);
1147
1148 bool match = true;
1149 size_t s = 0;
1150
1151 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1152 {
1153 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1154 {
1155 match &= v2.get(s+j) == j;
1156 }
1157 s += sz_recv2.get(i);
1158 }
1159
1160 BOOST_REQUIRE_EQUAL(match,true);
1161
1162 for (size_t i = 0 ; i < v3.size() ; i++)
1163 {
1164 for (size_t j = 0 ; j < v3.get(i).size() ; j++)
1165 {
1166 match &= v3.get(i).get(j) == j;
1167 }
1168 }
1169
1170 BOOST_REQUIRE_EQUAL(match,true);
1171 }
1172}
1173
1174BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_receive_known)
1175{
1176 Vcluster_semantic_sendrecv_receive_known_impl<NBX>();
1177}
1178
1179BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_receive_known_async)
1180{
1181 Vcluster_semantic_sendrecv_receive_known_impl<NBX_ASYNC>();
1182}
1183
1184template<unsigned int impl>
1185void Vcluster_semantic_struct_sendrecv_impl()
1186{
1187 for (size_t i = 0 ; i < 100 ; i++)
1188 {
1189 Vcluster<> & vcl = create_vcluster();
1190
1191 if (vcl.getProcessingUnits() >= 32)
1192 {return;}
1193
1194 openfpm::vector<size_t> prc_recv2;
1195 openfpm::vector<size_t> prc_recv3;
1196 openfpm::vector<size_t> prc_send;
1197 openfpm::vector<size_t> sz_recv2;
1198 openfpm::vector<size_t> sz_recv3;
1199 openfpm::vector<openfpm::vector<Box<3,size_t>>> v1;
1200 openfpm::vector<Box<3,size_t>> v2;
1201 openfpm::vector<openfpm::vector<Box<3,size_t>>> v3;
1202
1203 v1.resize(vcl.getProcessingUnits());
1204
1205 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
1206 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
1207 nr = ((nr-1) * nr) / 2;
1208
1209 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
1210
1211 for(size_t i = 0 ; i < v1.size() ; i++)
1212 {
1213 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
1214 {
1215 Box<3,size_t> b({j,j,j},{j,j,j});
1216 v1.get(i).add(b);
1217 }
1218
1219 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1220 }
1221
1222 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1223 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1224
1225 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1226 size_t nc_check = (vcl.getProcessingUnits()-1) / SSCATTER_MAX;
1227 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits()-1-nc_check);
1228
1229 bool match = true;
1230 size_t s = 0;
1231
1232 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1233 {
1234 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1235 {
1236 Box<3,size_t> b({j,j,j},{j,j,j});
1237 Box<3,size_t> bt = v2.get(s+j);
1238 match &= bt == b;
1239 }
1240 s += sz_recv2.get(i);
1241 }
1242
1243 BOOST_REQUIRE_EQUAL(match,true);
1244
1245 for (size_t i = 0 ; i < v3.size() ; i++)
1246 {
1247 for (size_t j = 0 ; j < v3.get(i).size() ; j++)
1248 {
1249 Box<3,size_t> b({j,j,j},{j,j,j});
1250 Box<3,size_t> bt = v3.get(i).get(j);
1251 match &= bt == b;
1252 }
1253 }
1254
1255 BOOST_REQUIRE_EQUAL(match,true);
1256 }
1257
1258 // Send and receive 0 and check
1259
1260 {
1261 Vcluster<> & vcl = create_vcluster();
1262
1263 openfpm::vector<size_t> prc_recv2;
1264 openfpm::vector<size_t> prc_send;
1265 openfpm::vector<size_t> sz_recv2;
1266 openfpm::vector<openfpm::vector<Box<3,size_t>>> v1;
1267 openfpm::vector<Box<3,size_t>> v2;
1268
1269 v1.resize(vcl.getProcessingUnits());
1270
1271
1272 for(size_t i = 0 ; i < v1.size() ; i++)
1273 {
1274 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1275 }
1276
1277 vcl.SSendRecv(v1,v2,prc_send,prc_recv2,sz_recv2);
1278
1279 BOOST_REQUIRE_EQUAL(v2.size(),0ul);
1280 BOOST_REQUIRE_EQUAL(prc_recv2.size(),0ul);
1281 BOOST_REQUIRE_EQUAL(sz_recv2.size(),0ul);
1282 }
1283}
1284
1285BOOST_AUTO_TEST_CASE (Vcluster_semantic_struct_sendrecv)
1286{
1287 Vcluster_semantic_struct_sendrecv_impl<NBX>();
1288}
1289
1290BOOST_AUTO_TEST_CASE (Vcluster_semantic_struct_sendrecv_async)
1291{
1292 Vcluster_semantic_struct_sendrecv_impl<NBX_ASYNC>();
1293}
1294
1295template<unsigned int impl>
1296void Vcluster_semantic_sendrecv_2_impl()
1297{
1298 for (size_t i = 0 ; i < 100 ; i++)
1299 {
1300 Vcluster<> & vcl = create_vcluster();
1301
1302 if (vcl.getProcessingUnits() >= 32)
1303 return;
1304
1305 openfpm::vector<size_t> prc_recv2;
1306 openfpm::vector<size_t> prc_recv3;
1307 openfpm::vector<size_t> prc_send;
1308 openfpm::vector<size_t> sz_recv2;
1309 openfpm::vector<size_t> sz_recv3;
1310
1311 openfpm::vector<openfpm::vector<aggregate<openfpm::vector<size_t>>> > v1;
1312 openfpm::vector<aggregate<openfpm::vector<size_t>>> v2;
1313 openfpm::vector<openfpm::vector<aggregate<openfpm::vector<size_t>>> > v3;
1314
1315 openfpm::vector<aggregate<openfpm::vector<size_t>>> v1_int;
1316 aggregate<openfpm::vector<size_t>> aggr;
1317 openfpm::vector<size_t> v1_int2;
1318
1319 v1_int2.add(7);
1320 v1_int2.add(7);
1321 v1_int2.add(7);
1322
1323 aggr.template get<0>() = v1_int2;
1324
1325 v1_int.add(aggr);
1326 v1_int.add(aggr);
1327 v1_int.add(aggr);
1328
1329 v1.resize(vcl.getProcessingUnits());
1330
1331 for(size_t i = 0 ; i < v1.size() ; i++)
1332 {
1333 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
1334 {
1335 v1.get(i).add(aggr);
1336 }
1337
1338 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1339 }
1340
1341 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
1342 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
1343 nr = ((nr-1) * nr) / 2;
1344
1345 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
1346
1347 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1348 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1349
1350 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1351
1352 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits());
1353
1354 bool match = true;
1355 bool is_seven = true;
1356 size_t s = 0;
1357
1358 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1359 {
1360 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1361 {
1362 for (size_t k = 0; k < v2.get(s+j).template get<0>().size(); k++)
1363 is_seven &= (v2.get(s+j).template get<0>().get(k) == 7);
1364 }
1365 s += sz_recv2.get(i);
1366 }
1367
1368 BOOST_REQUIRE_EQUAL(is_seven,true);
1369 BOOST_REQUIRE_EQUAL(match,true);
1370
1371 for (size_t i = 0 ; i < v3.size() ; i++)
1372 {
1373 for (size_t j = 0 ; j < v3.get(i).size(); j++)
1374 {
1375 for (size_t k = 0; k < v3.get(i).template get<0>(j).size(); k++)
1376 is_seven &= (v3.get(i).template get<0>(j).get(k) == 7);
1377 }
1378 }
1379
1380 BOOST_REQUIRE_EQUAL(is_seven,true);
1381 BOOST_REQUIRE_EQUAL(match,true);
1382 }
1383}
1384
1385BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_2)
1386{
1387 Vcluster_semantic_sendrecv_2_impl<NBX>();
1388}
1389
1390BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_2_async)
1391{
1392 Vcluster_semantic_sendrecv_2_impl<NBX_ASYNC>();
1393}
1394
1395template<unsigned int impl>
1396void Vcluster_semantic_sendrecv_3_impl()
1397{
1398 for (size_t i = 0 ; i < 100 ; i++)
1399 {
1400 Vcluster<> & vcl = create_vcluster();
1401
1402 if (vcl.getProcessingUnits() >= 32)
1403 return;
1404
1405 openfpm::vector<size_t> prc_recv2;
1406 openfpm::vector<size_t> prc_recv3;
1407 openfpm::vector<size_t> prc_send;
1408 openfpm::vector<size_t> sz_recv2;
1409 openfpm::vector<size_t> sz_recv3;
1410
1411 openfpm::vector<openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> > v1;
1412 openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> v2;
1413 openfpm::vector<openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> > v3;
1414
1415 openfpm::vector<aggregate<float, openfpm::vector<size_t>, Point_test<float>>> v1_int;
1416 aggregate<float, openfpm::vector<size_t>, Point_test<float>> aggr;
1417 openfpm::vector<size_t> v1_int2;
1418
1419 v1_int2.add((size_t)7);
1420 v1_int2.add((size_t)7);
1421
1422 aggr.template get<0>() = 7;
1423 aggr.template get<1>() = v1_int2;
1424
1425 typedef Point_test<float> p;
1426 p p1;
1427 p1.fill();
1428 aggr.template get<2>() = p1;
1429
1430 v1_int.add(aggr);
1431 v1_int.add(aggr);
1432 v1_int.add(aggr);
1433
1434 v1.resize(vcl.getProcessingUnits());
1435
1436 for(size_t i = 0 ; i < v1.size() ; i++)
1437 {
1438 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
1439 {
1440 v1.get(i).add(aggr);
1441 }
1442
1443 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1444 }
1445
1446 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
1447 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
1448 nr = ((nr-1) * nr) / 2;
1449
1450 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
1451
1452 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1453 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1454
1455 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1456
1457 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits());
1458
1459 bool match = true;
1460 bool is_seven = true;
1461 size_t s = 0;
1462
1463 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1464 {
1465 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1466 {
1467 is_seven &= (v2.get(s+j).template get<0>() == 7);
1468
1469 for (size_t k = 0; k < v2.get(s+j).template get<1>().size(); k++)
1470 is_seven &= (v2.get(s+j).template get<1>().get(k) == 7);
1471
1472 Point_test<float> p2 = v2.get(s+j).template get<2>();
1473
1474 match &= (p2.template get<p::x>() == p1.template get<p::x>());
1475 match &= (p2.template get<p::y>() == p1.template get<p::y>());
1476 match &= (p2.template get<p::z>() == p1.template get<p::z>());
1477 match &= (p2.template get<p::s>() == p1.template get<p::s>());
1478
1479 match &= (p2.template get<p::v>()[0] == p1.template get<p::v>()[0]);
1480 match &= (p2.template get<p::v>()[1] == p1.template get<p::v>()[1]);
1481 match &= (p2.template get<p::v>()[2] == p1.template get<p::v>()[2]);
1482
1483 match &= (p2.template get<p::t>()[0][0] == p1.template get<p::t>()[0][0]);
1484 match &= (p2.template get<p::t>()[0][1] == p1.template get<p::t>()[0][1]);
1485 match &= (p2.template get<p::t>()[0][2] == p1.template get<p::t>()[0][2]);
1486 match &= (p2.template get<p::t>()[1][0] == p1.template get<p::t>()[1][0]);
1487 match &= (p2.template get<p::t>()[1][1] == p1.template get<p::t>()[1][1]);
1488 match &= (p2.template get<p::t>()[1][2] == p1.template get<p::t>()[1][2]);
1489 match &= (p2.template get<p::t>()[2][0] == p1.template get<p::t>()[2][0]);
1490 match &= (p2.template get<p::t>()[2][1] == p1.template get<p::t>()[2][1]);
1491 match &= (p2.template get<p::t>()[2][2] == p1.template get<p::t>()[2][2]);
1492 }
1493 s += sz_recv2.get(i);
1494 }
1495
1496 BOOST_REQUIRE_EQUAL(is_seven,true);
1497 BOOST_REQUIRE_EQUAL(match,true);
1498
1499 for (size_t i = 0 ; i < v3.size() ; i++)
1500 {
1501 for (size_t j = 0 ; j < v3.get(i).size(); j++)
1502 {
1503 is_seven &= (v3.get(i).get(j).template get<0>() == 7);
1504
1505 for (size_t k = 0; k < v3.get(i).get(j).template get<1>().size(); k++)
1506 is_seven &= (v3.get(i).get(j).template get<1>().get(k) == 7);
1507
1508 Point_test<float> p2 = v3.get(i).get(j).template get<2>();
1509
1510 match &= (p2.template get<p::x>() == p1.template get<p::x>());
1511 match &= (p2.template get<p::y>() == p1.template get<p::y>());
1512 match &= (p2.template get<p::z>() == p1.template get<p::z>());
1513 match &= (p2.template get<p::s>() == p1.template get<p::s>());
1514
1515 match &= (p2.template get<p::v>()[0] == p1.template get<p::v>()[0]);
1516 match &= (p2.template get<p::v>()[1] == p1.template get<p::v>()[1]);
1517 match &= (p2.template get<p::v>()[2] == p1.template get<p::v>()[2]);
1518
1519 match &= (p2.template get<p::t>()[0][0] == p1.template get<p::t>()[0][0]);
1520 match &= (p2.template get<p::t>()[0][1] == p1.template get<p::t>()[0][1]);
1521 match &= (p2.template get<p::t>()[0][2] == p1.template get<p::t>()[0][2]);
1522 match &= (p2.template get<p::t>()[1][0] == p1.template get<p::t>()[1][0]);
1523 match &= (p2.template get<p::t>()[1][1] == p1.template get<p::t>()[1][1]);
1524 match &= (p2.template get<p::t>()[1][2] == p1.template get<p::t>()[1][2]);
1525 match &= (p2.template get<p::t>()[2][0] == p1.template get<p::t>()[2][0]);
1526 match &= (p2.template get<p::t>()[2][1] == p1.template get<p::t>()[2][1]);
1527 match &= (p2.template get<p::t>()[2][2] == p1.template get<p::t>()[2][2]);
1528 }
1529 }
1530
1531 BOOST_REQUIRE_EQUAL(is_seven,true);
1532 BOOST_REQUIRE_EQUAL(match,true);
1533 }
1534}
1535
1536BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_3)
1537{
1538 Vcluster_semantic_sendrecv_3_impl<NBX>();
1539}
1540
1541BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_3_async)
1542{
1543 Vcluster_semantic_sendrecv_3_impl<NBX_ASYNC>();
1544}
1545
1546template<unsigned int impl>
1547void Vcluster_semantic_sendrecv_4_impl()
1548{
1549 for (size_t i = 0 ; i < 100 ; i++)
1550 {
1551 Vcluster<> & vcl = create_vcluster();
1552
1553 if (vcl.getProcessingUnits() >= 32)
1554 return;
1555
1556 openfpm::vector<size_t> prc_recv2;
1557 openfpm::vector<size_t> prc_recv3;
1558 openfpm::vector<size_t> prc_send;
1559 openfpm::vector<size_t> sz_recv2;
1560 openfpm::vector<size_t> sz_recv3;
1561 openfpm::vector<openfpm::vector<aggregate<float,Point_test<float>>> > v1;
1562 openfpm::vector<aggregate<float,Point_test<float>> > v2;
1563 openfpm::vector<openfpm::vector<aggregate<float,Point_test<float>>> > v3;
1564
1565 v1.resize(vcl.getProcessingUnits());
1566
1567 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
1568 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
1569 nr = ((nr-1) * nr) / 2;
1570
1571 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
1572
1573 //Prepare an aggregate
1574 aggregate<float, Point_test<float> > aggr;
1575
1576 typedef Point_test<float> p;
1577
1578 p p1;
1579 p1.fill();
1580
1581 aggr.template get<0>() = 7;
1582 aggr.template get<1>() = p1;
1583
1584 //Fill v1 with aggregates
1585 for(size_t i = 0 ; i < v1.size() ; i++)
1586 {
1587 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
1588 {
1589 v1.get(i).add(aggr);
1590 }
1591
1592 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1593 }
1594
1595 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1596 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1597
1598 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1599 size_t nc_check = (vcl.getProcessingUnits()-1) / SSCATTER_MAX;
1600 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits()-1-nc_check);
1601 bool match = true;
1602 bool is_seven = true;
1603 size_t s = 0;
1604
1605 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1606 {
1607 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1608 {
1609 is_seven &= (v2.get(s+j).template get<0>() == 7);
1610
1611 Point_test<float> p2 = v2.get(s+j).template get<1>();
1612
1613 match &= (p2.template get<p::x>() == p1.template get<p::x>());
1614 match &= (p2.template get<p::y>() == p1.template get<p::y>());
1615 match &= (p2.template get<p::z>() == p1.template get<p::z>());
1616 match &= (p2.template get<p::s>() == p1.template get<p::s>());
1617
1618 match &= (p2.template get<p::v>()[0] == p1.template get<p::v>()[0]);
1619 match &= (p2.template get<p::v>()[1] == p1.template get<p::v>()[1]);
1620 match &= (p2.template get<p::v>()[2] == p1.template get<p::v>()[2]);
1621
1622 match &= (p2.template get<p::t>()[0][0] == p1.template get<p::t>()[0][0]);
1623 match &= (p2.template get<p::t>()[0][1] == p1.template get<p::t>()[0][1]);
1624 match &= (p2.template get<p::t>()[0][2] == p1.template get<p::t>()[0][2]);
1625 match &= (p2.template get<p::t>()[1][0] == p1.template get<p::t>()[1][0]);
1626 match &= (p2.template get<p::t>()[1][1] == p1.template get<p::t>()[1][1]);
1627 match &= (p2.template get<p::t>()[1][2] == p1.template get<p::t>()[1][2]);
1628 match &= (p2.template get<p::t>()[2][0] == p1.template get<p::t>()[2][0]);
1629 match &= (p2.template get<p::t>()[2][1] == p1.template get<p::t>()[2][1]);
1630 match &= (p2.template get<p::t>()[2][2] == p1.template get<p::t>()[2][2]);
1631 }
1632 s += sz_recv2.get(i);
1633 }
1634
1635 BOOST_REQUIRE_EQUAL(is_seven,true);
1636 BOOST_REQUIRE_EQUAL(match,true);
1637
1638 for (size_t i = 0 ; i < v3.size() ; i++)
1639 {
1640 for (size_t j = 0 ; j < v3.get(i).size() ; j++)
1641 {
1642 is_seven &= (v3.get(i).get(j).template get<0>() == 7);
1643
1644 Point_test<float> p2 = v3.get(i).get(j).template get<1>();
1645
1646 match &= (p2.template get<p::x>() == p1.template get<p::x>());
1647 match &= (p2.template get<p::y>() == p1.template get<p::y>());
1648 match &= (p2.template get<p::z>() == p1.template get<p::z>());
1649 match &= (p2.template get<p::s>() == p1.template get<p::s>());
1650
1651 match &= (p2.template get<p::v>()[0] == p1.template get<p::v>()[0]);
1652 match &= (p2.template get<p::v>()[1] == p1.template get<p::v>()[1]);
1653 match &= (p2.template get<p::v>()[2] == p1.template get<p::v>()[2]);
1654
1655 match &= (p2.template get<p::t>()[0][0] == p1.template get<p::t>()[0][0]);
1656 match &= (p2.template get<p::t>()[0][1] == p1.template get<p::t>()[0][1]);
1657 match &= (p2.template get<p::t>()[0][2] == p1.template get<p::t>()[0][2]);
1658 match &= (p2.template get<p::t>()[1][0] == p1.template get<p::t>()[1][0]);
1659 match &= (p2.template get<p::t>()[1][1] == p1.template get<p::t>()[1][1]);
1660 match &= (p2.template get<p::t>()[1][2] == p1.template get<p::t>()[1][2]);
1661 match &= (p2.template get<p::t>()[2][0] == p1.template get<p::t>()[2][0]);
1662 match &= (p2.template get<p::t>()[2][1] == p1.template get<p::t>()[2][1]);
1663 match &= (p2.template get<p::t>()[2][2] == p1.template get<p::t>()[2][2]);
1664 }
1665 }
1666
1667 BOOST_REQUIRE_EQUAL(is_seven,true);
1668 BOOST_REQUIRE_EQUAL(match,true);
1669 }
1670}
1671
1672BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_4)
1673{
1674 Vcluster_semantic_sendrecv_4_impl<NBX>();
1675}
1676
1677BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_4_async)
1678{
1679 Vcluster_semantic_sendrecv_4_impl<NBX_ASYNC>();
1680}
1681
1682template<unsigned int impl>
1683void Vcluster_semantic_sendrecv_5_impl()
1684{
1685 for (size_t i = 0 ; i < 100 ; i++)
1686 {
1687 Vcluster<> & vcl = create_vcluster();
1688
1689 if (vcl.getProcessingUnits() >= 32)
1690 return;
1691
1692 openfpm::vector<size_t> prc_recv2;
1693 openfpm::vector<size_t> prc_recv3;
1694 openfpm::vector<size_t> prc_send;
1695 openfpm::vector<size_t> sz_recv2;
1696 openfpm::vector<size_t> sz_recv3;
1697
1698 size_t sz[] = {16,16};
1699
1700 grid_cpu<2,Point_test<float>> g1(sz);
1701 g1.setMemory();
1702 fill_grid<2>(g1);
1703
1704 aggregate<grid_cpu<2,Point_test<float>>> aggr;
1705 aggr.template get<0>() = g1;
1706
1707
1708 openfpm::vector<openfpm::vector<aggregate<grid_cpu<2,Point_test<float>>>> > v1;
1709 openfpm::vector<aggregate<grid_cpu<2,Point_test<float>>> > v2;
1710 openfpm::vector<openfpm::vector<aggregate<grid_cpu<2,Point_test<float>>>> > v3;
1711
1712 v1.resize(vcl.getProcessingUnits());
1713
1714 for(size_t i = 0 ; i < v1.size() ; i++)
1715 {
1716 for (size_t j = 0 ; j < i % SSCATTER_MAX ; j++)
1717 {
1718 v1.get(i).add(aggr);
1719 }
1720
1721 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1722 }
1723
1724 size_t nc = vcl.getProcessingUnits() / SSCATTER_MAX;
1725 size_t nr = vcl.getProcessingUnits() - nc * SSCATTER_MAX;
1726 nr = ((nr-1) * nr) / 2;
1727
1728 size_t n_ele = nc * SSCATTER_MAX * (SSCATTER_MAX - 1) / 2 + nr;
1729
1730 scomm_unknown<impl>(vcl,v1,v2,prc_send,prc_recv2,sz_recv2);
1731
1732 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1733
1734 BOOST_REQUIRE_EQUAL(v2.size(),n_ele);
1735
1736 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits());
1737
1738 bool match = true;
1739 size_t s = 0;
1740 typedef Point_test<float> p;
1741
1742 for (size_t i = 0 ; i < sz_recv2.size() ; i++)
1743 {
1744 for (size_t j = 0 ; j < sz_recv2.get(i); j++)
1745 {
1746 grid_cpu<2,Point_test<float>> g2 = v2.get(s+j).template get<0>();
1747
1748 auto it = g2.getIterator();
1749
1750 while (it.isNext())
1751 {
1752 grid_key_dx<2> key = it.get();
1753
1754 match &= (g2.template get<p::x>(key) == g1.template get<p::x>(key));
1755 match &= (g2.template get<p::y>(key) == g1.template get<p::y>(key));
1756 match &= (g2.template get<p::z>(key) == g1.template get<p::z>(key));
1757 match &= (g2.template get<p::s>(key) == g1.template get<p::s>(key));
1758
1759 match &= (g2.template get<p::v>(key)[0] == g1.template get<p::v>(key)[0]);
1760 match &= (g2.template get<p::v>(key)[1] == g1.template get<p::v>(key)[1]);
1761 match &= (g2.template get<p::v>(key)[2] == g1.template get<p::v>(key)[2]);
1762
1763 match &= (g2.template get<p::t>(key)[0][0] == g1.template get<p::t>(key)[0][0]);
1764 match &= (g2.template get<p::t>(key)[0][1] == g1.template get<p::t>(key)[0][1]);
1765 match &= (g2.template get<p::t>(key)[0][2] == g1.template get<p::t>(key)[0][2]);
1766 match &= (g2.template get<p::t>(key)[1][0] == g1.template get<p::t>(key)[1][0]);
1767 match &= (g2.template get<p::t>(key)[1][1] == g1.template get<p::t>(key)[1][1]);
1768 match &= (g2.template get<p::t>(key)[1][2] == g1.template get<p::t>(key)[1][2]);
1769 match &= (g2.template get<p::t>(key)[2][0] == g1.template get<p::t>(key)[2][0]);
1770 match &= (g2.template get<p::t>(key)[2][1] == g1.template get<p::t>(key)[2][1]);
1771 match &= (g2.template get<p::t>(key)[2][2] == g1.template get<p::t>(key)[2][2]);
1772
1773 ++it;
1774 }
1775 }
1776 s += sz_recv2.get(i);
1777 }
1778 BOOST_REQUIRE_EQUAL(match,true);
1779
1780 for (size_t i = 0 ; i < v3.size() ; i++)
1781 {
1782 for (size_t j = 0 ; j < v3.get(i).size(); j++)
1783 {
1784 grid_cpu<2,Point_test<float>> g2 = v3.get(i).get(j).template get<0>();
1785
1786 auto it = g2.getIterator();
1787
1788 while (it.isNext())
1789 {
1790 grid_key_dx<2> key = it.get();
1791
1792 match &= (g2.template get<p::x>(key) == g1.template get<p::x>(key));
1793 match &= (g2.template get<p::y>(key) == g1.template get<p::y>(key));
1794 match &= (g2.template get<p::z>(key) == g1.template get<p::z>(key));
1795 match &= (g2.template get<p::s>(key) == g1.template get<p::s>(key));
1796
1797 match &= (g2.template get<p::v>(key)[0] == g1.template get<p::v>(key)[0]);
1798 match &= (g2.template get<p::v>(key)[1] == g1.template get<p::v>(key)[1]);
1799 match &= (g2.template get<p::v>(key)[2] == g1.template get<p::v>(key)[2]);
1800
1801 match &= (g2.template get<p::t>(key)[0][0] == g1.template get<p::t>(key)[0][0]);
1802 match &= (g2.template get<p::t>(key)[0][1] == g1.template get<p::t>(key)[0][1]);
1803 match &= (g2.template get<p::t>(key)[0][2] == g1.template get<p::t>(key)[0][2]);
1804 match &= (g2.template get<p::t>(key)[1][0] == g1.template get<p::t>(key)[1][0]);
1805 match &= (g2.template get<p::t>(key)[1][1] == g1.template get<p::t>(key)[1][1]);
1806 match &= (g2.template get<p::t>(key)[1][2] == g1.template get<p::t>(key)[1][2]);
1807 match &= (g2.template get<p::t>(key)[2][0] == g1.template get<p::t>(key)[2][0]);
1808 match &= (g2.template get<p::t>(key)[2][1] == g1.template get<p::t>(key)[2][1]);
1809 match &= (g2.template get<p::t>(key)[2][2] == g1.template get<p::t>(key)[2][2]);
1810
1811 ++it;
1812 }
1813 }
1814 }
1815 BOOST_REQUIRE_EQUAL(match,true);
1816 }
1817}
1818
1819BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_5)
1820{
1821 Vcluster_semantic_sendrecv_5_impl<NBX>();
1822}
1823
1824BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_5_async)
1825{
1826 Vcluster_semantic_sendrecv_5_impl<NBX_ASYNC>();
1827}
1828
1829template<unsigned int impl>
1830void Vcluster_semantic_sendrecv_6_impl()
1831{
1832 for (size_t i = 0 ; i < 100 ; i++)
1833 {
1834 Vcluster<> & vcl = create_vcluster();
1835
1836 if (vcl.getProcessingUnits() >= 32)
1837 return;
1838
1839 openfpm::vector<size_t> prc_recv2;
1840 openfpm::vector<size_t> prc_recv3;
1841 openfpm::vector<size_t> prc_send;
1842 openfpm::vector<size_t> sz_recv2;
1843 openfpm::vector<size_t> sz_recv3;
1844
1845 size_t sz[] = {8,10};
1846
1847 grid_cpu<2,Point_test<float>> g1(sz);
1848 g1.setMemory();
1849 fill_grid<2>(g1);
1850
1851 openfpm::vector<grid_cpu<2,Point_test<float>>> v1;
1852 openfpm::vector<grid_cpu<2,Point_test<float>>> v3;
1853
1854 v1.resize(vcl.getProcessingUnits());
1855
1856 for(size_t i = 0 ; i < v1.size() ; i++)
1857 {
1858 v1.get(i) = g1;
1859
1860 prc_send.add((i + vcl.getProcessUnitID()) % vcl.getProcessingUnits());
1861 }
1862
1863 scomm_unknown<impl>(vcl,v1,v3,prc_send,prc_recv3,sz_recv3);
1864
1865 BOOST_REQUIRE_EQUAL(v3.size(),vcl.getProcessingUnits());
1866
1867 bool match = true;
1868 typedef Point_test<float> p;
1869
1870 for (size_t i = 0 ; i < v3.size() ; i++)
1871 {
1872 for (size_t j = 0 ; j < v3.get(i).size(); j++)
1873 {
1874 grid_cpu<2,Point_test<float>> g2 = v3.get(i);
1875
1876 auto it = g2.getIterator();
1877
1878 while (it.isNext())
1879 {
1880 grid_key_dx<2> key = it.get();
1881
1882 match &= (g2.template get<p::x>(key) == g1.template get<p::x>(key));
1883 match &= (g2.template get<p::y>(key) == g1.template get<p::y>(key));
1884 match &= (g2.template get<p::z>(key) == g1.template get<p::z>(key));
1885 match &= (g2.template get<p::s>(key) == g1.template get<p::s>(key));
1886
1887 match &= (g2.template get<p::v>(key)[0] == g1.template get<p::v>(key)[0]);
1888 match &= (g2.template get<p::v>(key)[1] == g1.template get<p::v>(key)[1]);
1889 match &= (g2.template get<p::v>(key)[2] == g1.template get<p::v>(key)[2]);
1890
1891 match &= (g2.template get<p::t>(key)[0][0] == g1.template get<p::t>(key)[0][0]);
1892 match &= (g2.template get<p::t>(key)[0][1] == g1.template get<p::t>(key)[0][1]);
1893 match &= (g2.template get<p::t>(key)[0][2] == g1.template get<p::t>(key)[0][2]);
1894 match &= (g2.template get<p::t>(key)[1][0] == g1.template get<p::t>(key)[1][0]);
1895 match &= (g2.template get<p::t>(key)[1][1] == g1.template get<p::t>(key)[1][1]);
1896 match &= (g2.template get<p::t>(key)[1][2] == g1.template get<p::t>(key)[1][2]);
1897 match &= (g2.template get<p::t>(key)[2][0] == g1.template get<p::t>(key)[2][0]);
1898 match &= (g2.template get<p::t>(key)[2][1] == g1.template get<p::t>(key)[2][1]);
1899 match &= (g2.template get<p::t>(key)[2][2] == g1.template get<p::t>(key)[2][2]);
1900
1901 ++it;
1902 }
1903 }
1904 }
1905 BOOST_REQUIRE_EQUAL(match,true);
1906
1907 if (vcl.getProcessUnitID() == 0 && i == 99)
1908 std::cout << "Semantic sendrecv test start" << std::endl;
1909 }
1910}
1911
1912BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_6)
1913{
1914 Vcluster_semantic_sendrecv_6_impl<NBX>();
1915}
1916
1917BOOST_AUTO_TEST_CASE (Vcluster_semantic_sendrecv_6_async)
1918{
1919 Vcluster_semantic_sendrecv_6_impl<NBX_ASYNC>();
1920}
1921
1922BOOST_AUTO_TEST_SUITE_END()
1923
1924