]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | //---------------------------------------------------------------------------// |
2 | // Copyright (c) 2016 Jakub Szuppe <j.szuppe@gmail.com> | |
3 | // | |
4 | // Distributed under the Boost Software License, Version 1.0 | |
5 | // See accompanying file LICENSE_1_0.txt or copy at | |
6 | // http://www.boost.org/LICENSE_1_0.txt | |
7 | // | |
8 | // See http://boostorg.github.com/compute for more information. | |
9 | //---------------------------------------------------------------------------// | |
10 | ||
11 | // Undefining BOOST_COMPUTE_USE_OFFLINE_CACHE macro as we want to modify cached | |
12 | // parameters for copy algorithm without any undesirable consequences (like | |
13 | // saving modified values of those parameters). | |
14 | #ifdef BOOST_COMPUTE_USE_OFFLINE_CACHE | |
15 | #undef BOOST_COMPUTE_USE_OFFLINE_CACHE | |
16 | #endif | |
17 | ||
18 | #define BOOST_TEST_MODULE TestCopyTypeMismatch | |
19 | #include <boost/test/unit_test.hpp> | |
20 | ||
21 | #include <list> | |
22 | #include <vector> | |
23 | #include <string> | |
24 | #include <sstream> | |
25 | #include <iterator> | |
26 | #include <iostream> | |
27 | ||
28 | #include <boost/compute/svm.hpp> | |
29 | #include <boost/compute/system.hpp> | |
30 | #include <boost/compute/functional.hpp> | |
31 | #include <boost/compute/command_queue.hpp> | |
32 | #include <boost/compute/algorithm/copy.hpp> | |
33 | #include <boost/compute/async/future.hpp> | |
34 | #include <boost/compute/container/vector.hpp> | |
35 | #include <boost/compute/detail/device_ptr.hpp> | |
36 | #include <boost/compute/memory/svm_ptr.hpp> | |
37 | #include <boost/compute/detail/parameter_cache.hpp> | |
38 | ||
39 | #include "quirks.hpp" | |
40 | #include "check_macros.hpp" | |
41 | #include "context_setup.hpp" | |
42 | ||
43 | namespace bc = boost::compute; | |
44 | namespace compute = boost::compute; | |
45 | ||
46 | BOOST_AUTO_TEST_CASE(is_same_ignore_const) | |
47 | { | |
48 | BOOST_STATIC_ASSERT(( | |
49 | boost::compute::detail::is_same_value_type< | |
50 | std::vector<int>::iterator, | |
51 | compute::buffer_iterator<int> | |
52 | >::value | |
53 | )); | |
54 | BOOST_STATIC_ASSERT(( | |
55 | boost::compute::detail::is_same_value_type< | |
56 | std::vector<int>::const_iterator, | |
57 | compute::buffer_iterator<int> | |
58 | >::value | |
59 | )); | |
60 | BOOST_STATIC_ASSERT(( | |
61 | boost::compute::detail::is_same_value_type< | |
62 | std::vector<int>::iterator, | |
63 | compute::buffer_iterator<const int> | |
64 | >::value | |
65 | )); | |
66 | BOOST_STATIC_ASSERT(( | |
67 | boost::compute::detail::is_same_value_type< | |
68 | std::vector<int>::const_iterator, | |
69 | compute::buffer_iterator<const int> | |
70 | >::value | |
71 | )); | |
72 | } | |
73 | ||
74 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_double) | |
75 | { | |
76 | if(!device.supports_extension("cl_khr_fp64")) { | |
77 | std::cout << "skipping test: device does not support double" << std::endl; | |
78 | return; | |
79 | } | |
80 | ||
81 | using compute::double_; | |
82 | using compute::float_; | |
83 | ||
84 | float_ host[] = { 6.1f, 10.2f, 19.3f, 25.4f }; | |
85 | bc::vector<double_> device_vector(4, context); | |
86 | ||
87 | // copy host float data to double device vector | |
88 | bc::copy(host, host + 4, device_vector.begin(), queue); | |
89 | CHECK_RANGE_EQUAL(double_, 4, device_vector, (6.1f, 10.2f, 19.3f, 25.4f)); | |
90 | } | |
91 | ||
92 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_int) | |
93 | { | |
94 | using compute::int_; | |
95 | using compute::float_; | |
96 | ||
97 | float_ host[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
98 | bc::vector<int_> device_vector(4, context); | |
99 | ||
100 | // copy host float data to int device vector | |
101 | bc::copy(host, host + 4, device_vector.begin(), queue); | |
102 | CHECK_RANGE_EQUAL( | |
103 | int_, | |
104 | 4, | |
105 | device_vector, | |
106 | ( | |
107 | static_cast<int_>(6.1f), | |
108 | static_cast<int_>(-10.2f), | |
109 | static_cast<int_>(19.3f), | |
110 | static_cast<int_>(25.4f) | |
111 | ) | |
112 | ); | |
113 | } | |
114 | ||
115 | // HOST -> DEVICE | |
116 | ||
117 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_int_mapping_device_vector) | |
118 | { | |
119 | using compute::int_; | |
120 | using compute::uint_; | |
121 | using compute::float_; | |
122 | ||
123 | float_ host[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
124 | bc::vector<int_> device_vector(4, context); | |
125 | ||
126 | std::string cache_key = | |
127 | std::string("__boost_compute_copy_to_device_float_int"); | |
128 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
129 | bc::detail::parameter_cache::get_global_cache(device); | |
130 | ||
131 | // save | |
132 | uint_ map_copy_threshold = | |
133 | parameters->get(cache_key, "map_copy_threshold", 0); | |
134 | ||
135 | // force copy_to_device_map (mapping device vector to the host) | |
136 | parameters->set(cache_key, "map_copy_threshold", 1024); | |
137 | ||
138 | // copy host float data to int device vector | |
139 | bc::copy(host, host + 4, device_vector.begin(), queue); | |
140 | CHECK_RANGE_EQUAL( | |
141 | int_, | |
142 | 4, | |
143 | device_vector, | |
144 | ( | |
145 | static_cast<int_>(6.1f), | |
146 | static_cast<int_>(-10.2f), | |
147 | static_cast<int_>(19.3f), | |
148 | static_cast<int_>(25.4f) | |
149 | ) | |
150 | ); | |
151 | ||
152 | // restore | |
153 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
154 | } | |
155 | ||
156 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_int_convert_on_host) | |
157 | { | |
158 | using compute::int_; | |
159 | using compute::uint_; | |
160 | using compute::float_; | |
161 | ||
162 | std::string cache_key = | |
163 | std::string("__boost_compute_copy_to_device_float_int"); | |
164 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
165 | bc::detail::parameter_cache::get_global_cache(device); | |
166 | ||
167 | // save | |
168 | uint_ map_copy_threshold = | |
169 | parameters->get(cache_key, "map_copy_threshold", 0); | |
170 | uint_ direct_copy_threshold = | |
171 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
172 | ||
173 | // force copying by casting input data on host and performing | |
174 | // normal copy host->device (since types match now) | |
175 | parameters->set(cache_key, "map_copy_threshold", 0); | |
176 | parameters->set(cache_key, "direct_copy_threshold", 1024); | |
177 | ||
178 | float_ host[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
179 | bc::vector<int_> device_vector(4, context); | |
180 | ||
181 | // copy host float data to int device vector | |
182 | bc::copy(host, host + 4, device_vector.begin(), queue); | |
183 | CHECK_RANGE_EQUAL( | |
184 | int_, | |
185 | 4, | |
186 | device_vector, | |
187 | ( | |
188 | static_cast<int_>(6.1f), | |
189 | static_cast<int_>(-10.2f), | |
190 | static_cast<int_>(19.3f), | |
191 | static_cast<int_>(25.4f) | |
192 | ) | |
193 | ); | |
194 | ||
195 | // restore | |
196 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
197 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
198 | } | |
199 | ||
200 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_int_with_transform) | |
201 | { | |
202 | using compute::int_; | |
203 | using compute::uint_; | |
204 | using compute::float_; | |
205 | ||
206 | std::string cache_key = | |
207 | std::string("__boost_compute_copy_to_device_float_int"); | |
208 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
209 | bc::detail::parameter_cache::get_global_cache(device); | |
210 | ||
211 | // save | |
212 | uint_ map_copy_threshold = | |
213 | parameters->get(cache_key, "map_copy_threshold", 0); | |
214 | uint_ direct_copy_threshold = | |
215 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
216 | ||
217 | // force copying by mapping input data to the device memory | |
218 | // and using transform operation for casting & copying | |
219 | parameters->set(cache_key, "map_copy_threshold", 0); | |
220 | parameters->set(cache_key, "direct_copy_threshold", 0); | |
221 | ||
222 | float_ host[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
223 | bc::vector<int_> device_vector(4, context); | |
224 | ||
225 | // copy host float data to int device vector | |
226 | bc::copy(host, host + 4, device_vector.begin(), queue); | |
227 | CHECK_RANGE_EQUAL( | |
228 | int_, | |
229 | 4, | |
230 | device_vector, | |
231 | ( | |
232 | static_cast<int_>(6.1f), | |
233 | static_cast<int_>(-10.2f), | |
234 | static_cast<int_>(19.3f), | |
235 | static_cast<int_>(25.4f) | |
236 | ) | |
237 | ); | |
238 | ||
239 | // restore | |
240 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
241 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
242 | } | |
243 | ||
244 | BOOST_AUTO_TEST_CASE(copy_async_to_device_float_to_int) | |
245 | { | |
246 | using compute::int_; | |
247 | using compute::float_; | |
248 | ||
249 | float_ host[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
250 | bc::vector<int_> device_vector(4, context); | |
251 | ||
252 | // copy host float data to int device vector | |
253 | compute::future<void> future = | |
254 | bc::copy_async(host, host + 4, device_vector.begin(), queue); | |
255 | future.wait(); | |
256 | ||
257 | CHECK_RANGE_EQUAL( | |
258 | int_, | |
259 | 4, | |
260 | device_vector, | |
261 | ( | |
262 | static_cast<int_>(6.1f), | |
263 | static_cast<int_>(-10.2f), | |
264 | static_cast<int_>(19.3f), | |
265 | static_cast<int_>(25.4f) | |
266 | ) | |
267 | ); | |
268 | } | |
269 | ||
270 | BOOST_AUTO_TEST_CASE(copy_async_to_device_float_to_int_empty) | |
271 | { | |
272 | using compute::int_; | |
273 | using compute::float_; | |
274 | ||
275 | float_ host[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
276 | bc::vector<int_> device_vector(size_t(4), int_(1), queue); | |
277 | ||
278 | // copy nothing to int device vector | |
279 | compute::future<bc::vector<int_>::iterator > future = | |
280 | bc::copy_async(host, host, device_vector.begin(), queue); | |
281 | if(future.valid()) { | |
282 | future.wait(); | |
283 | } | |
284 | ||
285 | CHECK_RANGE_EQUAL( | |
286 | int_, | |
287 | 4, | |
288 | device_vector, | |
289 | ( | |
290 | int_(1), | |
291 | int_(1), | |
292 | int_(1), | |
293 | int_(1) | |
294 | ) | |
295 | ); | |
296 | } | |
297 | ||
298 | // Test copying from a std::list to a bc::vector. This differs from | |
299 | // the test copying from std::vector because std::list has non-contiguous | |
300 | // storage for its data values. | |
301 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_int_list_device_map) | |
302 | { | |
303 | using compute::int_; | |
304 | using compute::uint_; | |
305 | using compute::float_; | |
306 | ||
307 | std::string cache_key = | |
308 | std::string("__boost_compute_copy_to_device_float_int"); | |
309 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
310 | bc::detail::parameter_cache::get_global_cache(device); | |
311 | ||
312 | // save | |
313 | uint_ map_copy_threshold = | |
314 | parameters->get(cache_key, "map_copy_threshold", 0); | |
315 | ||
316 | // force copy_to_device_map (mapping device vector to the host) | |
317 | parameters->set(cache_key, "map_copy_threshold", 1024); | |
318 | ||
319 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
320 | std::list<float_> host(data, data + 4); | |
321 | bc::vector<int_> device_vector(4, context); | |
322 | ||
323 | // copy host float data to int device vector | |
324 | bc::copy(host.begin(), host.end(), device_vector.begin(), queue); | |
325 | CHECK_RANGE_EQUAL( | |
326 | int_, | |
327 | 4, | |
328 | device_vector, | |
329 | ( | |
330 | static_cast<int_>(6.1f), | |
331 | static_cast<int_>(-10.2f), | |
332 | static_cast<int_>(19.3f), | |
333 | static_cast<int_>(25.4f) | |
334 | ) | |
335 | ); | |
336 | ||
337 | // restore | |
338 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
339 | } | |
340 | ||
341 | // Test copying from a std::list to a bc::vector. This differs from | |
342 | // the test copying from std::vector because std::list has non-contiguous | |
343 | // storage for its data values. | |
344 | BOOST_AUTO_TEST_CASE(copy_to_device_float_to_int_list_convert_on_host) | |
345 | { | |
346 | using compute::int_; | |
347 | using compute::uint_; | |
348 | using compute::float_; | |
349 | ||
350 | std::string cache_key = | |
351 | std::string("__boost_compute_copy_to_device_float_int"); | |
352 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
353 | bc::detail::parameter_cache::get_global_cache(device); | |
354 | ||
355 | // save | |
356 | uint_ map_copy_threshold = | |
357 | parameters->get(cache_key, "map_copy_threshold", 0); | |
358 | uint_ direct_copy_threshold = | |
359 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
360 | ||
361 | // force copying by casting input data on host and performing | |
362 | // normal copy host->device (since types match now) | |
363 | parameters->set(cache_key, "map_copy_threshold", 0); | |
364 | parameters->set(cache_key, "direct_copy_threshold", 1024); | |
365 | ||
366 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
367 | std::list<float_> host(data, data + 4); | |
368 | bc::vector<int_> device_vector(4, context); | |
369 | ||
370 | // copy host float data to int device vector | |
371 | bc::copy(host.begin(), host.end(), device_vector.begin(), queue); | |
372 | CHECK_RANGE_EQUAL( | |
373 | int_, | |
374 | 4, | |
375 | device_vector, | |
376 | ( | |
377 | static_cast<int_>(6.1f), | |
378 | static_cast<int_>(-10.2f), | |
379 | static_cast<int_>(19.3f), | |
380 | static_cast<int_>(25.4f) | |
381 | ) | |
382 | ); | |
383 | ||
384 | // restore | |
385 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
386 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
387 | } | |
388 | ||
389 | // SVM requires OpenCL 2.0 | |
390 | #if defined(CL_VERSION_2_0) || defined(BOOST_COMPUTE_DOXYGEN_INVOKED) | |
391 | BOOST_AUTO_TEST_CASE(copy_to_device_svm_float_to_int_map) | |
392 | { | |
393 | REQUIRES_OPENCL_VERSION(2, 0); | |
394 | ||
395 | using compute::int_; | |
396 | using compute::uint_; | |
397 | using compute::float_; | |
398 | ||
399 | std::string cache_key = | |
400 | std::string("__boost_compute_copy_to_device_float_int"); | |
401 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
402 | bc::detail::parameter_cache::get_global_cache(device); | |
403 | ||
404 | // save | |
405 | uint_ map_copy_threshold = | |
406 | parameters->get(cache_key, "map_copy_threshold", 0); | |
407 | ||
408 | // force copy_to_device_map (mapping device vector to the host) | |
409 | parameters->set(cache_key, "map_copy_threshold", 1024); | |
410 | ||
411 | float_ host[] = { 5.1f, -10.3f, 19.4f, 26.7f }; | |
412 | compute::svm_ptr<int_> ptr = compute::svm_alloc<int_>(context, 4); | |
413 | ||
414 | // copy host float data to int device vector | |
415 | bc::copy(host, host + 4, ptr, queue); | |
416 | ||
417 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
418 | CHECK_HOST_RANGE_EQUAL( | |
419 | int_, | |
420 | 4, | |
421 | static_cast<int_*>(ptr.get()), | |
422 | ( | |
423 | static_cast<int_>(5.1f), | |
424 | static_cast<int_>(-10.3f), | |
425 | static_cast<int_>(19.4f), | |
426 | static_cast<int_>(26.7f) | |
427 | ) | |
428 | ); | |
429 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
430 | ||
431 | compute::svm_free(context, ptr); | |
432 | ||
433 | // restore | |
434 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
435 | } | |
436 | ||
437 | BOOST_AUTO_TEST_CASE(copy_to_device_svm_float_to_int_convert_on_host) | |
438 | { | |
439 | REQUIRES_OPENCL_VERSION(2, 0); | |
440 | ||
441 | if(bug_in_svmmemcpy(device)){ | |
442 | std::cerr << "skipping svmmemcpy test case" << std::endl; | |
443 | return; | |
444 | } | |
445 | ||
446 | using compute::int_; | |
447 | using compute::uint_; | |
448 | using compute::float_; | |
449 | ||
450 | std::string cache_key = | |
451 | std::string("__boost_compute_copy_to_device_float_int"); | |
452 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
453 | bc::detail::parameter_cache::get_global_cache(device); | |
454 | ||
455 | // save | |
456 | uint_ map_copy_threshold = | |
457 | parameters->get(cache_key, "map_copy_threshold", 0); | |
458 | uint_ direct_copy_threshold = | |
459 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
460 | ||
461 | // force copying by casting input data on host and performing | |
462 | // normal copy host->device (since types match now) | |
463 | parameters->set(cache_key, "map_copy_threshold", 0); | |
464 | parameters->set(cache_key, "direct_copy_threshold", 1024); | |
465 | ||
466 | float_ host[] = { 0.1f, 10.3f, 9.4f, -26.7f }; | |
467 | compute::svm_ptr<int_> ptr = compute::svm_alloc<int_>(context, 4); | |
468 | ||
469 | // copy host float data to int device vector | |
470 | bc::copy(host, host + 4, ptr, queue); | |
471 | ||
472 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
473 | CHECK_HOST_RANGE_EQUAL( | |
474 | int_, | |
475 | 4, | |
476 | static_cast<int_*>(ptr.get()), | |
477 | ( | |
478 | static_cast<int_>(0.1f), | |
479 | static_cast<int_>(10.3f), | |
480 | static_cast<int_>(9.4f), | |
481 | static_cast<int_>(-26.7f) | |
482 | ) | |
483 | ); | |
484 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
485 | ||
486 | compute::svm_free(context, ptr); | |
487 | ||
488 | // restore | |
489 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
490 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
491 | } | |
492 | ||
493 | BOOST_AUTO_TEST_CASE(copy_to_device_svm_float_to_int_with_transform) | |
494 | { | |
495 | REQUIRES_OPENCL_VERSION(2, 0); | |
496 | ||
497 | using compute::int_; | |
498 | using compute::uint_; | |
499 | using compute::float_; | |
500 | ||
501 | std::string cache_key = | |
502 | std::string("__boost_compute_copy_to_device_float_int"); | |
503 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
504 | bc::detail::parameter_cache::get_global_cache(device); | |
505 | ||
506 | // save | |
507 | uint_ map_copy_threshold = | |
508 | parameters->get(cache_key, "map_copy_threshold", 0); | |
509 | uint_ direct_copy_threshold = | |
510 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
511 | ||
512 | // force copying by mapping input data to the device memory | |
513 | // and using transform operation (copy kernel) for casting & copying | |
514 | parameters->set(cache_key, "map_copy_threshold", 0); | |
515 | parameters->set(cache_key, "direct_copy_threshold", 0); | |
516 | ||
517 | float_ host[] = { 4.1f, -11.3f, 219.4f, -26.7f }; | |
518 | compute::svm_ptr<int_> ptr = compute::svm_alloc<int_>(context, 4); | |
519 | ||
520 | // copy host float data to int device vector | |
521 | bc::copy(host, host + 4, ptr, queue); | |
522 | ||
523 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
524 | CHECK_HOST_RANGE_EQUAL( | |
525 | int_, | |
526 | 4, | |
527 | static_cast<int_*>(ptr.get()), | |
528 | ( | |
529 | static_cast<int_>(4.1f), | |
530 | static_cast<int_>(-11.3f), | |
531 | static_cast<int_>(219.4f), | |
532 | static_cast<int_>(-26.7f) | |
533 | ) | |
534 | ); | |
535 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
536 | ||
537 | compute::svm_free(context, ptr); | |
538 | ||
539 | // restore | |
540 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
541 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
542 | } | |
543 | ||
544 | BOOST_AUTO_TEST_CASE(copy_async_to_device_svm_float_to_int) | |
545 | { | |
546 | REQUIRES_OPENCL_VERSION(2, 0); | |
547 | ||
548 | using compute::int_; | |
549 | using compute::uint_; | |
550 | using compute::float_; | |
551 | ||
552 | float_ host[] = { 44.1f, -14.3f, 319.4f, -26.7f }; | |
553 | compute::svm_ptr<int_> ptr = compute::svm_alloc<int_>(context, 4); | |
554 | ||
555 | // copy host float data to int device vector | |
556 | compute::future<void> future = | |
557 | bc::copy_async(host, host + 4, ptr, queue); | |
558 | future.wait(); | |
559 | ||
560 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
561 | CHECK_HOST_RANGE_EQUAL( | |
562 | int_, | |
563 | 4, | |
564 | static_cast<int_*>(ptr.get()), | |
565 | ( | |
566 | static_cast<int_>(44.1f), | |
567 | static_cast<int_>(-14.3f), | |
568 | static_cast<int_>(319.4f), | |
569 | static_cast<int_>(-26.7f) | |
570 | ) | |
571 | ); | |
572 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
573 | ||
574 | compute::svm_free(context, ptr); | |
575 | } | |
576 | #endif | |
577 | ||
578 | // DEVICE -> DEVICE | |
579 | ||
580 | BOOST_AUTO_TEST_CASE(copy_on_device_float_to_int) | |
581 | { | |
582 | using compute::int_; | |
583 | using compute::float_; | |
584 | ||
585 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
586 | bc::vector<float_> device_fvector(data, data + 4, queue); | |
587 | bc::vector<int_> device_ivector(4, context); | |
588 | ||
589 | // copy device float vector to device int vector | |
590 | bc::copy( | |
591 | device_fvector.begin(), | |
592 | device_fvector.end(), | |
593 | device_ivector.begin(), | |
594 | queue | |
595 | ); | |
596 | ||
597 | CHECK_RANGE_EQUAL( | |
598 | int_, | |
599 | 4, | |
600 | device_ivector, | |
601 | ( | |
602 | static_cast<int_>(6.1f), | |
603 | static_cast<int_>(-10.2f), | |
604 | static_cast<int_>(19.3f), | |
605 | static_cast<int_>(25.4f) | |
606 | ) | |
607 | ); | |
608 | } | |
609 | ||
610 | BOOST_AUTO_TEST_CASE(copy_async_on_device_float_to_int) | |
611 | { | |
612 | using compute::int_; | |
613 | using compute::float_; | |
614 | ||
615 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
616 | bc::vector<float_> device_fvector(data, data + 4, queue); | |
617 | bc::vector<int_> device_ivector(4, context); | |
618 | ||
619 | // copy device float vector to device int vector | |
620 | compute::future<void> future = | |
621 | bc::copy_async( | |
622 | device_fvector.begin(), | |
623 | device_fvector.end(), | |
624 | device_ivector.begin(), | |
625 | queue | |
626 | ); | |
627 | future.wait(); | |
628 | ||
629 | CHECK_RANGE_EQUAL( | |
630 | int_, | |
631 | 4, | |
632 | device_ivector, | |
633 | ( | |
634 | static_cast<int_>(6.1f), | |
635 | static_cast<int_>(-10.2f), | |
636 | static_cast<int_>(19.3f), | |
637 | static_cast<int_>(25.4f) | |
638 | ) | |
639 | ); | |
640 | } | |
641 | ||
642 | BOOST_AUTO_TEST_CASE(copy_async_on_device_float_to_int_empty) | |
643 | { | |
644 | using compute::int_; | |
645 | using compute::float_; | |
646 | ||
647 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
648 | bc::vector<float_> device_fvector(data, data + 4, queue); | |
649 | bc::vector<int_> device_ivector(size_t(4), int_(1), queue); | |
650 | ||
651 | // copy device float vector to device int vector | |
652 | compute::future<void> future = | |
653 | bc::copy_async( | |
654 | device_fvector.begin(), | |
655 | device_fvector.begin(), | |
656 | device_ivector.begin(), | |
657 | queue | |
658 | ); | |
659 | if(future.valid()) { | |
660 | future.wait(); | |
661 | } | |
662 | ||
663 | CHECK_RANGE_EQUAL( | |
664 | int_, | |
665 | 4, | |
666 | device_ivector, | |
667 | ( | |
668 | int_(1), | |
669 | int_(1), | |
670 | int_(1), | |
671 | int_(1) | |
672 | ) | |
673 | ); | |
674 | } | |
675 | ||
676 | // SVM requires OpenCL 2.0 | |
677 | #if defined(CL_VERSION_2_0) || defined(BOOST_COMPUTE_DOXYGEN_INVOKED) | |
678 | BOOST_AUTO_TEST_CASE(copy_on_device_buffer_to_svm_float_to_int) | |
679 | { | |
680 | REQUIRES_OPENCL_VERSION(2, 0); | |
681 | ||
682 | using compute::int_; | |
683 | using compute::float_; | |
684 | ||
685 | float_ data[] = { 65.1f, -110.2f, -19.3f, 26.7f }; | |
686 | bc::vector<float_> device_vector(data, data + 4, queue); | |
687 | compute::svm_ptr<int_> ptr = compute::svm_alloc<int_>(context, 4); | |
688 | ||
689 | // copy host float data to int svm memory | |
690 | bc::copy(device_vector.begin(), device_vector.end(), ptr, queue); | |
691 | ||
692 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
693 | CHECK_HOST_RANGE_EQUAL( | |
694 | int_, | |
695 | 4, | |
696 | static_cast<int_*>(ptr.get()), | |
697 | ( | |
698 | static_cast<int_>(65.1f), | |
699 | static_cast<int_>(-110.2f), | |
700 | static_cast<int_>(-19.3f), | |
701 | static_cast<int_>(26.7f) | |
702 | ) | |
703 | ); | |
704 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
705 | ||
706 | compute::svm_free(context, ptr); | |
707 | } | |
708 | ||
709 | BOOST_AUTO_TEST_CASE(copy_on_device_svm_to_buffer_float_to_int) | |
710 | { | |
711 | REQUIRES_OPENCL_VERSION(2, 0); | |
712 | ||
713 | using compute::int_; | |
714 | using compute::float_; | |
715 | ||
716 | float_ data[] = { 6.1f, 11.2f, 19.3f, 6.7f }; | |
717 | bc::vector<int_> device_vector(4, context); | |
718 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
719 | ||
720 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
721 | for(size_t i = 0; i < 4; i++) { | |
722 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
723 | } | |
724 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
725 | ||
726 | // copy host float svm data to int device vector | |
727 | bc::copy(ptr, ptr + 4, device_vector.begin(), queue); | |
728 | ||
729 | CHECK_RANGE_EQUAL( | |
730 | int_, | |
731 | 4, | |
732 | device_vector, | |
733 | ( | |
734 | static_cast<int_>(6.1f), | |
735 | static_cast<int_>(11.2f), | |
736 | static_cast<int_>(19.3f), | |
737 | static_cast<int_>(6.7f) | |
738 | ) | |
739 | ); | |
740 | ||
741 | compute::svm_free(context, ptr); | |
742 | } | |
743 | ||
744 | BOOST_AUTO_TEST_CASE(copy_on_device_svm_to_svm_float_to_int) | |
745 | { | |
746 | REQUIRES_OPENCL_VERSION(2, 0); | |
747 | ||
748 | using compute::int_; | |
749 | using compute::float_; | |
750 | ||
751 | float_ data[] = { 0.1f, -10.2f, -1.3f, 2.7f }; | |
752 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
753 | compute::svm_ptr<int_> ptr2 = compute::svm_alloc<int_>(context, 4); | |
754 | ||
755 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
756 | for(size_t i = 0; i < 4; i++) { | |
757 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
758 | } | |
759 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
760 | ||
761 | // copy host float svm to int svm | |
762 | bc::copy(ptr, ptr + 4, ptr2, queue); | |
763 | ||
764 | queue.enqueue_svm_map(ptr2.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
765 | CHECK_HOST_RANGE_EQUAL( | |
766 | int_, | |
767 | 4, | |
768 | static_cast<int_*>(ptr2.get()), | |
769 | ( | |
770 | static_cast<int_>(0.1f), | |
771 | static_cast<int_>(-10.2f), | |
772 | static_cast<int_>(-1.3f), | |
773 | static_cast<int_>(2.7f) | |
774 | ) | |
775 | ); | |
776 | queue.enqueue_svm_unmap(ptr2.get()).wait(); | |
777 | ||
778 | compute::svm_free(context, ptr); | |
779 | compute::svm_free(context, ptr2); | |
780 | } | |
781 | ||
782 | BOOST_AUTO_TEST_CASE(copy_async_on_device_buffer_to_svm_float_to_int) | |
783 | { | |
784 | REQUIRES_OPENCL_VERSION(2, 0); | |
785 | ||
786 | using compute::int_; | |
787 | using compute::float_; | |
788 | ||
789 | float_ data[] = { 65.1f, -110.2f, -19.3f, 26.7f }; | |
790 | bc::vector<float_> device_vector(data, data + 4, queue); | |
791 | compute::svm_ptr<int_> ptr = compute::svm_alloc<int_>(context, 4); | |
792 | ||
793 | // copy host float data to int svm memory | |
794 | compute::future<bc::svm_ptr<int_> > future = | |
795 | bc::copy_async(device_vector.begin(), device_vector.end(), ptr, queue); | |
796 | future.wait(); | |
797 | ||
798 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
799 | CHECK_HOST_RANGE_EQUAL( | |
800 | int_, | |
801 | 4, | |
802 | static_cast<int_*>(ptr.get()), | |
803 | ( | |
804 | static_cast<int_>(65.1f), | |
805 | static_cast<int_>(-110.2f), | |
806 | static_cast<int_>(-19.3f), | |
807 | static_cast<int_>(26.7f) | |
808 | ) | |
809 | ); | |
810 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
811 | ||
812 | compute::svm_free(context, ptr); | |
813 | } | |
814 | ||
815 | BOOST_AUTO_TEST_CASE(copy_async_on_device_svm_to_buffer_float_to_int) | |
816 | { | |
817 | REQUIRES_OPENCL_VERSION(2, 0); | |
818 | ||
819 | using compute::int_; | |
820 | using compute::float_; | |
821 | ||
822 | float_ data[] = { 65.1f, -110.2f, -19.3f, 26.7f }; | |
823 | bc::vector<int_> device_vector(4, context); | |
824 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
825 | ||
826 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
827 | for(size_t i = 0; i < 4; i++) { | |
828 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
829 | } | |
830 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
831 | ||
832 | // copy host float svm data to int device vector | |
833 | compute::future<bc::vector<int_>::iterator > future = | |
834 | bc::copy_async(ptr, ptr + 4, device_vector.begin(), queue); | |
835 | future.wait(); | |
836 | ||
837 | CHECK_RANGE_EQUAL( | |
838 | int_, | |
839 | 4, | |
840 | device_vector, | |
841 | ( | |
842 | static_cast<int_>(65.1f), | |
843 | static_cast<int_>(-110.2f), | |
844 | static_cast<int_>(-19.3f), | |
845 | static_cast<int_>(26.7f) | |
846 | ) | |
847 | ); | |
848 | ||
849 | compute::svm_free(context, ptr); | |
850 | } | |
851 | ||
852 | BOOST_AUTO_TEST_CASE(copy_async_on_device_svm_to_svm_float_to_int) | |
853 | { | |
854 | REQUIRES_OPENCL_VERSION(2, 0); | |
855 | ||
856 | using compute::int_; | |
857 | using compute::float_; | |
858 | ||
859 | float_ data[] = { 0.1f, -10.2f, -1.3f, 2.7f }; | |
860 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
861 | compute::svm_ptr<int_> ptr2 = compute::svm_alloc<int_>(context, 4); | |
862 | ||
863 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
864 | for(size_t i = 0; i < 4; i++) { | |
865 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
866 | } | |
867 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
868 | ||
869 | // copy host float svm to int svm | |
870 | compute::future<bc::svm_ptr<int_> > future = | |
871 | bc::copy_async(ptr, ptr + 4, ptr2, queue); | |
872 | future.wait(); | |
873 | ||
874 | queue.enqueue_svm_map(ptr2.get(), 4 * sizeof(cl_int), CL_MAP_READ); | |
875 | CHECK_HOST_RANGE_EQUAL( | |
876 | int_, | |
877 | 4, | |
878 | static_cast<int_*>(ptr2.get()), | |
879 | ( | |
880 | static_cast<int_>(0.1f), | |
881 | static_cast<int_>(-10.2f), | |
882 | static_cast<int_>(-1.3f), | |
883 | static_cast<int_>(2.7f) | |
884 | ) | |
885 | ); | |
886 | queue.enqueue_svm_unmap(ptr2.get()).wait(); | |
887 | ||
888 | compute::svm_free(context, ptr); | |
889 | compute::svm_free(context, ptr2); | |
890 | } | |
891 | #endif | |
892 | ||
893 | // DEVICE -> HOST | |
894 | ||
895 | BOOST_AUTO_TEST_CASE(copy_to_host_float_to_int) | |
896 | { | |
897 | using compute::int_; | |
898 | using compute::float_; | |
899 | ||
900 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
901 | bc::vector<float_> device_vector(data, data + 4, queue); | |
902 | ||
903 | std::vector<int_> host_vector(4); | |
904 | // copy device float vector to int host vector | |
905 | bc::copy(device_vector.begin(), device_vector.end(), host_vector.begin(), queue); | |
906 | CHECK_HOST_RANGE_EQUAL( | |
907 | int_, | |
908 | 4, | |
909 | host_vector.begin(), | |
910 | ( | |
911 | static_cast<int_>(6.1f), | |
912 | static_cast<int_>(-10.2f), | |
913 | static_cast<int_>(19.3f), | |
914 | static_cast<int_>(25.4f) | |
915 | ) | |
916 | ); | |
917 | } | |
918 | ||
919 | BOOST_AUTO_TEST_CASE(copy_to_host_float_to_int_map) | |
920 | { | |
921 | using compute::int_; | |
922 | using compute::uint_; | |
923 | using compute::float_; | |
924 | ||
925 | std::string cache_key = | |
926 | std::string("__boost_compute_copy_to_host_float_int"); | |
927 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
928 | bc::detail::parameter_cache::get_global_cache(device); | |
929 | ||
930 | // save | |
931 | uint_ map_copy_threshold = | |
932 | parameters->get(cache_key, "map_copy_threshold", 0); | |
933 | ||
934 | // force copy_to_host_map (mapping device vector to the host) | |
935 | parameters->set(cache_key, "map_copy_threshold", 1024); | |
936 | ||
937 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
938 | bc::vector<float_> device_vector(data, data + 4, queue); | |
939 | ||
940 | std::vector<int_> host_vector(4); | |
941 | // copy device float vector to int host vector | |
942 | bc::copy(device_vector.begin(), device_vector.end(), host_vector.begin(), queue); | |
943 | CHECK_HOST_RANGE_EQUAL( | |
944 | int_, | |
945 | 4, | |
946 | host_vector.begin(), | |
947 | ( | |
948 | static_cast<int_>(6.1f), | |
949 | static_cast<int_>(-10.2f), | |
950 | static_cast<int_>(19.3f), | |
951 | static_cast<int_>(25.4f) | |
952 | ) | |
953 | ); | |
954 | ||
955 | // restore | |
956 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
957 | } | |
958 | ||
959 | BOOST_AUTO_TEST_CASE(copy_to_host_float_to_int_convert_on_host) | |
960 | { | |
961 | using compute::int_; | |
962 | using compute::uint_; | |
963 | using compute::float_; | |
964 | ||
965 | std::string cache_key = | |
966 | std::string("__boost_compute_copy_to_host_float_int"); | |
967 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
968 | bc::detail::parameter_cache::get_global_cache(device); | |
969 | ||
970 | // save | |
971 | uint_ map_copy_threshold = | |
972 | parameters->get(cache_key, "map_copy_threshold", 0); | |
973 | uint_ direct_copy_threshold = | |
974 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
975 | ||
976 | // force copying by copying input device vector to temporary | |
977 | // host vector of the same type and then copying from that temporary | |
978 | // vector to result using std::copy() | |
979 | parameters->set(cache_key, "map_copy_threshold", 0); | |
980 | parameters->set(cache_key, "direct_copy_threshold", 1024); | |
981 | ||
982 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
983 | bc::vector<float_> device_vector(data, data + 4, queue); | |
984 | ||
985 | std::vector<int_> host_vector(4); | |
986 | // copy device float vector to int host vector | |
987 | bc::copy(device_vector.begin(), device_vector.end(), host_vector.begin(), queue); | |
988 | CHECK_HOST_RANGE_EQUAL( | |
989 | int_, | |
990 | 4, | |
991 | host_vector.begin(), | |
992 | ( | |
993 | static_cast<int_>(6.1f), | |
994 | static_cast<int_>(-10.2f), | |
995 | static_cast<int_>(19.3f), | |
996 | static_cast<int_>(25.4f) | |
997 | ) | |
998 | ); | |
999 | ||
1000 | // restore | |
1001 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1002 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
1003 | } | |
1004 | ||
1005 | BOOST_AUTO_TEST_CASE(copy_to_host_float_to_int_convert_on_device) | |
1006 | { | |
1007 | using compute::int_; | |
1008 | using compute::uint_; | |
1009 | using compute::float_; | |
1010 | ||
1011 | std::string cache_key = | |
1012 | std::string("__boost_compute_copy_to_host_float_int"); | |
1013 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
1014 | bc::detail::parameter_cache::get_global_cache(device); | |
1015 | ||
1016 | // save | |
1017 | uint_ map_copy_threshold = | |
1018 | parameters->get(cache_key, "map_copy_threshold", 0); | |
1019 | uint_ direct_copy_threshold = | |
1020 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
1021 | ||
1022 | // force copying by mapping output data to the device memory | |
1023 | // and using transform operation for casting & copying | |
1024 | parameters->set(cache_key, "map_copy_threshold", 0); | |
1025 | parameters->set(cache_key, "direct_copy_threshold", 0); | |
1026 | ||
1027 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
1028 | bc::vector<float_> device_vector(data, data + 4, queue); | |
1029 | ||
1030 | std::vector<int_> host_vector(4); | |
1031 | // copy device float vector to int host vector | |
1032 | bc::copy(device_vector.begin(), device_vector.end(), host_vector.begin(), queue); | |
1033 | CHECK_HOST_RANGE_EQUAL( | |
1034 | int_, | |
1035 | 4, | |
1036 | host_vector.begin(), | |
1037 | ( | |
1038 | static_cast<int_>(6.1f), | |
1039 | static_cast<int_>(-10.2f), | |
1040 | static_cast<int_>(19.3f), | |
1041 | static_cast<int_>(25.4f) | |
1042 | ) | |
1043 | ); | |
1044 | ||
1045 | // restore | |
1046 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1047 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
1048 | } | |
1049 | ||
1050 | // Test copying from a bc::vector to a std::list . This differs from | |
1051 | // the test copying to std::vector because std::list has non-contiguous | |
1052 | // storage for its data values. | |
1053 | BOOST_AUTO_TEST_CASE(copy_to_host_list_float_to_int_map) | |
1054 | { | |
1055 | using compute::int_; | |
1056 | using compute::uint_; | |
1057 | using compute::float_; | |
1058 | ||
1059 | std::string cache_key = | |
1060 | std::string("__boost_compute_copy_to_host_float_int"); | |
1061 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
1062 | bc::detail::parameter_cache::get_global_cache(device); | |
1063 | ||
1064 | // save | |
1065 | uint_ map_copy_threshold = | |
1066 | parameters->get(cache_key, "map_copy_threshold", 0); | |
1067 | ||
1068 | // force copy_to_host_map (mapping device vector to the host) | |
1069 | parameters->set(cache_key, "map_copy_threshold", 1024); | |
1070 | ||
1071 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
1072 | bc::vector<float_> device_vector(data, data + 4, queue); | |
1073 | ||
1074 | std::list<int_> host_list(4); | |
1075 | // copy device float vector to int host vector | |
1076 | bc::copy(device_vector.begin(), device_vector.end(), host_list.begin(), queue); | |
1077 | ||
1078 | int_ expected[4] = { | |
1079 | static_cast<int_>(6.1f), | |
1080 | static_cast<int_>(-10.2f), | |
1081 | static_cast<int_>(19.3f), | |
1082 | static_cast<int_>(25.4f) | |
1083 | }; | |
1084 | BOOST_CHECK_EQUAL_COLLECTIONS( | |
1085 | host_list.begin(), host_list.end(), | |
1086 | expected, expected + 4 | |
1087 | ); | |
1088 | ||
1089 | // restore | |
1090 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1091 | } | |
1092 | ||
1093 | // Test copying from a bc::vector to a std::list . This differs from | |
1094 | // the test copying to std::vector because std::list has non-contiguous | |
1095 | // storage for its data values. | |
1096 | BOOST_AUTO_TEST_CASE(copy_to_host_list_float_to_int_covert_on_host) | |
1097 | { | |
1098 | using compute::int_; | |
1099 | using compute::uint_; | |
1100 | using compute::float_; | |
1101 | ||
1102 | std::string cache_key = | |
1103 | std::string("__boost_compute_copy_to_host_float_int"); | |
1104 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
1105 | bc::detail::parameter_cache::get_global_cache(device); | |
1106 | ||
1107 | // save | |
1108 | uint_ map_copy_threshold = | |
1109 | parameters->get(cache_key, "map_copy_threshold", 0); | |
1110 | uint_ direct_copy_threshold = | |
1111 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
1112 | ||
1113 | // force copying by copying input device vector to temporary | |
1114 | // host vector of the same type and then copying from that temporary | |
1115 | // vector to result using std::copy() | |
1116 | parameters->set(cache_key, "map_copy_threshold", 0); | |
1117 | parameters->set(cache_key, "direct_copy_threshold", 1024); | |
1118 | ||
1119 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
1120 | bc::vector<float_> device_vector(data, data + 4, queue); | |
1121 | ||
1122 | std::list<int_> host_list(4); | |
1123 | // copy device float vector to int host vector | |
1124 | bc::copy(device_vector.begin(), device_vector.end(), host_list.begin(), queue); | |
1125 | int_ expected[4] = { | |
1126 | static_cast<int_>(6.1f), | |
1127 | static_cast<int_>(-10.2f), | |
1128 | static_cast<int_>(19.3f), | |
1129 | static_cast<int_>(25.4f) | |
1130 | }; | |
1131 | BOOST_CHECK_EQUAL_COLLECTIONS( | |
1132 | host_list.begin(), host_list.end(), | |
1133 | expected, expected + 4 | |
1134 | ); | |
1135 | ||
1136 | // restore | |
1137 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1138 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
1139 | } | |
1140 | ||
1141 | BOOST_AUTO_TEST_CASE(copy_async_to_host_float_to_int) | |
1142 | { | |
1143 | using compute::int_; | |
1144 | using compute::float_; | |
1145 | ||
1146 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
1147 | bc::vector<float_> device_vector(data, data + 4, queue); | |
1148 | std::vector<int_> host_vector(device_vector.size()); | |
1149 | ||
1150 | // copy device float vector to host int vector | |
1151 | compute::future<void> future = | |
1152 | bc::copy_async( | |
1153 | device_vector.begin(), | |
1154 | device_vector.end(), | |
1155 | host_vector.begin(), | |
1156 | queue | |
1157 | ); | |
1158 | future.wait(); | |
1159 | ||
1160 | CHECK_HOST_RANGE_EQUAL( | |
1161 | int_, | |
1162 | 4, | |
1163 | host_vector.begin(), | |
1164 | ( | |
1165 | static_cast<int_>(6.1f), | |
1166 | static_cast<int_>(-10.2f), | |
1167 | static_cast<int_>(19.3f), | |
1168 | static_cast<int_>(25.4f) | |
1169 | ) | |
1170 | ); | |
1171 | } | |
1172 | ||
1173 | BOOST_AUTO_TEST_CASE(copy_async_to_host_float_to_int_empty) | |
1174 | { | |
1175 | using compute::int_; | |
1176 | using compute::float_; | |
1177 | ||
1178 | float_ data[] = { 6.1f, -10.2f, 19.3f, 25.4f }; | |
1179 | bc::vector<float_> device_vector(data, data + 4, queue); | |
1180 | std::vector<int_> host_vector(device_vector.size(), int_(1)); | |
1181 | ||
1182 | // copy device float vector to host int vector | |
1183 | compute::future<void> future = | |
1184 | bc::copy_async( | |
1185 | device_vector.begin(), | |
1186 | device_vector.begin(), | |
1187 | host_vector.begin(), | |
1188 | queue | |
1189 | ); | |
1190 | if(future.valid()) { | |
1191 | future.wait(); | |
1192 | } | |
1193 | ||
1194 | CHECK_HOST_RANGE_EQUAL( | |
1195 | int_, | |
1196 | 4, | |
1197 | host_vector.begin(), | |
1198 | ( | |
1199 | int_(1), | |
1200 | int_(1), | |
1201 | int_(1), | |
1202 | int_(1) | |
1203 | ) | |
1204 | ); | |
1205 | } | |
1206 | ||
1207 | // SVM requires OpenCL 2.0 | |
1208 | #if defined(CL_VERSION_2_0) || defined(BOOST_COMPUTE_DOXYGEN_INVOKED) | |
1209 | BOOST_AUTO_TEST_CASE(copy_to_host_svm_float_to_int_map) | |
1210 | { | |
1211 | REQUIRES_OPENCL_VERSION(2, 0); | |
1212 | ||
1213 | using compute::int_; | |
1214 | using compute::uint_; | |
1215 | using compute::float_; | |
1216 | ||
1217 | std::string cache_key = | |
1218 | std::string("__boost_compute_copy_to_host_float_int"); | |
1219 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
1220 | bc::detail::parameter_cache::get_global_cache(device); | |
1221 | ||
1222 | // save | |
1223 | uint_ map_copy_threshold = | |
1224 | parameters->get(cache_key, "map_copy_threshold", 0); | |
1225 | ||
1226 | // force copy_to_host_map (mapping device vector to the host) | |
1227 | parameters->set(cache_key, "map_copy_threshold", 1024); | |
1228 | ||
1229 | float_ data[] = { 6.1f, 1.2f, 1.3f, -66.7f }; | |
1230 | std::vector<int_> host_vector(4, 0); | |
1231 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
1232 | ||
1233 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
1234 | for(size_t i = 0; i < 4; i++) { | |
1235 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
1236 | } | |
1237 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
1238 | ||
1239 | // copy host float svm data to int host vector | |
1240 | bc::copy(ptr, ptr + 4, host_vector.begin(), queue); | |
1241 | ||
1242 | CHECK_HOST_RANGE_EQUAL( | |
1243 | int_, | |
1244 | 4, | |
1245 | host_vector.begin(), | |
1246 | ( | |
1247 | static_cast<int_>(6.1f), | |
1248 | static_cast<int_>(1.2f), | |
1249 | static_cast<int_>(1.3f), | |
1250 | static_cast<int_>(-66.7f) | |
1251 | ) | |
1252 | ); | |
1253 | ||
1254 | compute::svm_free(context, ptr); | |
1255 | ||
1256 | // restore | |
1257 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1258 | } | |
1259 | ||
1260 | BOOST_AUTO_TEST_CASE(copy_to_host_svm_float_to_int_convert_on_host) | |
1261 | { | |
1262 | REQUIRES_OPENCL_VERSION(2, 0); | |
1263 | ||
1264 | if(bug_in_svmmemcpy(device)){ | |
1265 | std::cerr << "skipping svmmemcpy test case" << std::endl; | |
1266 | return; | |
1267 | } | |
1268 | ||
1269 | using compute::int_; | |
1270 | using compute::uint_; | |
1271 | using compute::float_; | |
1272 | ||
1273 | std::string cache_key = | |
1274 | std::string("__boost_compute_copy_to_host_float_int"); | |
1275 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
1276 | bc::detail::parameter_cache::get_global_cache(device); | |
1277 | ||
1278 | // save | |
1279 | uint_ map_copy_threshold = | |
1280 | parameters->get(cache_key, "map_copy_threshold", 0); | |
1281 | uint_ direct_copy_threshold = | |
1282 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
1283 | ||
1284 | // force copying by copying input device vector to temporary | |
1285 | // host vector of the same type and then copying from that temporary | |
1286 | // vector to result using std::copy() | |
1287 | parameters->set(cache_key, "map_copy_threshold", 0); | |
1288 | parameters->set(cache_key, "direct_copy_threshold", 1024); | |
1289 | ||
1290 | float_ data[] = { 6.1f, 1.2f, 1.3f, 766.7f }; | |
1291 | std::vector<int_> host_vector(4, 0); | |
1292 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
1293 | ||
1294 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
1295 | for(size_t i = 0; i < 4; i++) { | |
1296 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
1297 | } | |
1298 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
1299 | ||
1300 | // copy host float svm data to int host vector | |
1301 | bc::copy(ptr, ptr + 4, host_vector.begin(), queue); | |
1302 | ||
1303 | CHECK_HOST_RANGE_EQUAL( | |
1304 | int_, | |
1305 | 4, | |
1306 | host_vector.begin(), | |
1307 | ( | |
1308 | static_cast<int_>(6.1f), | |
1309 | static_cast<int_>(1.2f), | |
1310 | static_cast<int_>(1.3f), | |
1311 | static_cast<int_>(766.7f) | |
1312 | ) | |
1313 | ); | |
1314 | ||
1315 | compute::svm_free(context, ptr); | |
1316 | ||
1317 | // restore | |
1318 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1319 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
1320 | } | |
1321 | ||
1322 | BOOST_AUTO_TEST_CASE(copy_to_host_svm_float_to_int_transform) | |
1323 | { | |
1324 | REQUIRES_OPENCL_VERSION(2, 0); | |
1325 | ||
1326 | using compute::int_; | |
1327 | using compute::uint_; | |
1328 | using compute::float_; | |
1329 | ||
1330 | std::string cache_key = | |
1331 | std::string("__boost_compute_copy_to_host_float_int"); | |
1332 | boost::shared_ptr<bc::detail::parameter_cache> parameters = | |
1333 | bc::detail::parameter_cache::get_global_cache(device); | |
1334 | ||
1335 | // save | |
1336 | uint_ map_copy_threshold = | |
1337 | parameters->get(cache_key, "map_copy_threshold", 0); | |
1338 | uint_ direct_copy_threshold = | |
1339 | parameters->get(cache_key, "direct_copy_threshold", 0); | |
1340 | ||
1341 | // force copying by copying input device vector to temporary | |
1342 | // host vector of the same type and then copying from that temporary | |
1343 | // vector to result using std::copy() | |
1344 | parameters->set(cache_key, "map_copy_threshold", 0); | |
1345 | parameters->set(cache_key, "direct_copy_threshold", 0); | |
1346 | ||
1347 | float_ data[] = { 0.1f, 11.2f, 1.3f, -66.7f }; | |
1348 | std::vector<int_> host_vector(4, 0); | |
1349 | compute::svm_ptr<float_> ptr = compute::svm_alloc<float_>(context, 4); | |
1350 | ||
1351 | queue.enqueue_svm_map(ptr.get(), 4 * sizeof(cl_int), CL_MAP_WRITE); | |
1352 | for(size_t i = 0; i < 4; i++) { | |
1353 | static_cast<float_*>(ptr.get())[i] = data[i]; | |
1354 | } | |
1355 | queue.enqueue_svm_unmap(ptr.get()).wait(); | |
1356 | ||
1357 | // copy host float svm data to int host vector | |
1358 | bc::copy(ptr, ptr + 4, host_vector.begin(), queue); | |
1359 | ||
1360 | CHECK_HOST_RANGE_EQUAL( | |
1361 | int_, | |
1362 | 4, | |
1363 | host_vector.begin(), | |
1364 | ( | |
1365 | static_cast<int_>(0.1f), | |
1366 | static_cast<int_>(11.2f), | |
1367 | static_cast<int_>(1.3f), | |
1368 | static_cast<int_>(-66.7f) | |
1369 | ) | |
1370 | ); | |
1371 | ||
1372 | compute::svm_free(context, ptr); | |
1373 | ||
1374 | // restore | |
1375 | parameters->set(cache_key, "map_copy_threshold", map_copy_threshold); | |
1376 | parameters->set(cache_key, "direct_copy_threshold", direct_copy_threshold); | |
1377 | } | |
1378 | #endif | |
1379 | ||
1380 | BOOST_AUTO_TEST_SUITE_END() |