vector_dist_gpu_unit_tests.cu 32.4 KB
Newer Older
incardon's avatar
incardon committed
1 2 3 4 5

#define BOOST_TEST_DYN_LINK
#include <boost/test/unit_test.hpp>
#include "VCluster/VCluster.hpp"
#include <Vector/vector_dist.hpp>
6
#include "Vector/tests/vector_dist_util_unit_tests.hpp"
incardon's avatar
incardon committed
7

incardon's avatar
incardon committed
8 9
#define SUB_UNIT_FACTOR 1024

incardon's avatar
incardon committed
10 11 12 13 14 15 16 17 18 19 20 21
template<unsigned int dim , typename vector_dist_type>
__global__ void move_parts_gpu_test(vector_dist_type vd)
{
	auto p = GET_PARTICLE(vd);

#pragma unroll
	for (int i = 0 ; i < dim ; i++)
	{
		vd.getPos(p)[i] += 0.05;
	}
}

incardon's avatar
incardon committed
22 23 24 25 26 27 28 29
BOOST_AUTO_TEST_SUITE( vector_dist_gpu_test )

void print_test(std::string test, size_t sz)
{
	if (create_vcluster().getProcessUnitID() == 0)
		std::cout << test << " " << sz << "\n";
}

incardon's avatar
incardon committed
30 31 32 33 34 35 36 37 38 39 40 41

__global__  void initialize_props(vector_dist_ker<3, float, aggregate<float, float [3], float[3]>> vd)
{
	auto p = GET_PARTICLE(vd);

	vd.template getProp<0>(p) = vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2];

	vd.template getProp<1>(p)[0] = vd.getPos(p)[0] + vd.getPos(p)[1];
	vd.template getProp<1>(p)[1] = vd.getPos(p)[0] + vd.getPos(p)[2];
	vd.template getProp<1>(p)[2] = vd.getPos(p)[1] + vd.getPos(p)[2];
}

incardon's avatar
Fixing  
incardon committed
42 43 44 45 46
template<typename T,typename CellList_type>
__global__  void calculate_force(vector_dist_ker<3, T, aggregate<T, T[3], T [3]>> vd,
		                         vector_dist_ker<3, T, aggregate<T, T[3], T [3]>> vd_sort,
		                         CellList_type cl,
		                         int rank)
incardon's avatar
incardon committed
47 48 49
{
	auto p = GET_PARTICLE(vd);

incardon's avatar
Fixing  
incardon committed
50
	Point<3,T> xp = vd.getPos(p);
incardon's avatar
incardon committed
51 52 53

    auto it = cl.getNNIterator(cl.getCell(xp));

incardon's avatar
Fixing  
incardon committed
54 55
    Point<3,T> force1({0.0,0.0,0.0});
    Point<3,T> force2({0.0,0.0,0.0});
incardon's avatar
incardon committed
56 57 58

    while (it.isNext())
    {
incardon's avatar
incardon committed
59 60
    	auto q1 = it.get_sort();
    	auto q2 = it.get();
incardon's avatar
incardon committed
61 62 63

    	if (q2 == p) {++it; continue;}

incardon's avatar
Fixing  
incardon committed
64 65
    	Point<3,T> xq_1 = vd_sort.getPos(q1);
    	Point<3,T> xq_2 = vd.getPos(q2);
incardon's avatar
incardon committed
66

incardon's avatar
Fixing  
incardon committed
67 68
    	Point<3,T> r1 = xq_1 - xp;
    	Point<3,T> r2 = xq_2 - xp;
incardon's avatar
incardon committed
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89

    	// Normalize

    	r1 /= r1.norm();
    	r2 /= r2.norm();

    	force1 += vd_sort.template getProp<0>(q1)*r1;
    	force2 += vd.template getProp<0>(q2)*r2;

    	++it;
    }

    vd.template getProp<1>(p)[0] = force1.get(0);
    vd.template getProp<1>(p)[1] = force1.get(1);
    vd.template getProp<1>(p)[2] = force1.get(2);

    vd.template getProp<2>(p)[0] = force2.get(0);
    vd.template getProp<2>(p)[1] = force2.get(1);
    vd.template getProp<2>(p)[2] = force2.get(2);
}

incardon's avatar
Fixing  
incardon committed
90 91
template<typename T, typename CellList_type>
__global__  void calculate_force_full_sort(vector_dist_ker<3, T, aggregate<T, T[3], T [3]>> vd,
incardon's avatar
incardon committed
92
		                         	 	   CellList_type cl, int rank)
93
{
incardon's avatar
incardon committed
94 95
	unsigned int p;
	GET_PARTICLE_SORT(p,cl);
incardon's avatar
incardon committed
96

incardon's avatar
Fixing  
incardon committed
97
	Point<3,T> xp = vd.getPos(p);
98 99 100

    auto it = cl.getNNIterator(cl.getCell(xp));

incardon's avatar
Fixing  
incardon committed
101
    Point<3,T> force1({0.0,0.0,0.0});
102 103 104

    while (it.isNext())
    {
incardon's avatar
incardon committed
105
    	auto q1 = it.get_sort();
106 107 108

    	if (q1 == p) {++it; continue;}

incardon's avatar
Fixing  
incardon committed
109
    	Point<3,T> xq_1 = vd.getPos(q1);
110

incardon's avatar
Fixing  
incardon committed
111
    	Point<3,T> r1 = xq_1 - xp;
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129

    	// Normalize

    	r1 /= r1.norm();

    	force1 += vd.template getProp<0>(q1)*r1;

    	++it;
    }

    vd.template getProp<1>(p)[0] = force1.get(0);
    vd.template getProp<1>(p)[1] = force1.get(1);
    vd.template getProp<1>(p)[2] = force1.get(2);
}

template<typename CellList_type, typename vector_type>
bool check_force(CellList_type & NN_cpu, vector_type & vd)
{
incardon's avatar
Fixing  
incardon committed
130 131
	typedef typename vector_type::stype St;

132 133 134 135 136 137 138 139
	auto it6 = vd.getDomainIterator();

	bool match = true;

	while (it6.isNext())
	{
		auto p = it6.get();

incardon's avatar
Fixing  
incardon committed
140
		Point<3,St> xp = vd.getPos(p);
141 142 143

		// Calculate on CPU

incardon's avatar
Fixing  
incardon committed
144
		Point<3,St> force({0.0,0.0,0.0});
145 146 147 148 149 150 151 152 153

		auto NNc = NN_cpu.getNNIterator(NN_cpu.getCell(xp));

		while (NNc.isNext())
		{
			auto q = NNc.get();

	    	if (q == p.getKey()) {++NNc; continue;}

incardon's avatar
Fixing  
incardon committed
154 155
	    	Point<3,St> xq_2 = vd.getPos(q);
	    	Point<3,St> r2 = xq_2 - xp;
156 157 158 159 160 161 162 163 164

	    	// Normalize

	    	r2 /= r2.norm();
	    	force += vd.template getProp<0>(q)*r2;

			++NNc;
		}

incardon's avatar
incardon committed
165 166 167
		match &= fabs(vd.template getProp<1>(p)[0] - vd.template getProp<2>(p)[0]) < 0.0003;
		match &= fabs(vd.template getProp<1>(p)[1] - vd.template getProp<2>(p)[1]) < 0.0003;
		match &= fabs(vd.template getProp<1>(p)[2] - vd.template getProp<2>(p)[2]) < 0.0003;
168

incardon's avatar
incardon committed
169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185
		match &= fabs(vd.template getProp<1>(p)[0] - force.get(0)) < 0.0003;
		match &= fabs(vd.template getProp<1>(p)[1] - force.get(1)) < 0.0003;
		match &= fabs(vd.template getProp<1>(p)[2] - force.get(2)) < 0.0003;

		if (match == false)
		{
			std::cout << "ERROR: " << vd.template getProp<1>(p)[0]  << "   " << vd.template getProp<2>(p)[0] << std::endl;
	                std::cout << "ERROR: " << vd.template getProp<1>(p)[1]  << "   " << vd.template getProp<2>(p)[1] << std::endl;
	                std::cout << "ERROR: " << vd.template getProp<1>(p)[2]  << "   " << vd.template getProp<2>(p)[2] << std::endl;

	                std::cout << "ERROR2: " << vd.template getProp<1>(p)[0] << "   " <<  force.get(0) << std::endl;
	                std::cout << "ERROR2: " << vd.template getProp<1>(p)[1] << "   " <<  force.get(1) << std::endl;
	                std::cout << "ERROR2: " << vd.template getProp<1>(p)[2] << "   " <<  force.get(2) << std::endl;


			break;
		}
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270

		++it6;
	}

	return match;
}

BOOST_AUTO_TEST_CASE( vector_dist_gpu_ghost_get )
{
	auto & v_cl = create_vcluster();

	if (v_cl.size() > 16)
	{return;}

	Box<3,float> domain({0.0,0.0,0.0},{1.0,1.0,1.0});

	// set the ghost based on the radius cut off (make just a little bit smaller than the spacing)
	Ghost<3,float> g(0.1);

	// Boundary conditions
	size_t bc[3]={PERIODIC,PERIODIC,PERIODIC};

	vector_dist_gpu<3,float,aggregate<float,float[3],float[3]>> vd(1000,domain,bc,g);

	auto it = vd.getDomainIterator();

	while (it.isNext())
	{
		auto p = it.get();

		vd.getPos(p)[0] = (float)rand() / RAND_MAX;
		vd.getPos(p)[1] = (float)rand() / RAND_MAX;
		vd.getPos(p)[2] = (float)rand() / RAND_MAX;

		vd.template getProp<0>(p) = vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2];

		vd.template getProp<1>(p)[0] = vd.getPos(p)[0] + vd.getPos(p)[1];
		vd.template getProp<1>(p)[1] = vd.getPos(p)[0] + vd.getPos(p)[2];
		vd.template getProp<1>(p)[2] = vd.getPos(p)[1] + vd.getPos(p)[2];

		vd.template getProp<2>(p)[0] = vd.getPos(p)[0] + 3.0*vd.getPos(p)[1];
		vd.template getProp<2>(p)[1] = vd.getPos(p)[0] + 3.0*vd.getPos(p)[2];
		vd.template getProp<2>(p)[2] = vd.getPos(p)[1] + 3.0*vd.getPos(p)[2];


		++it;
	}

	// Ok we redistribute the particles (CPU based)
	vd.map();

	vd.template ghost_get<0,1,2>();

	// Now we check the the ghost contain the correct information

	bool check = true;

	auto itg = vd.getDomainAndGhostIterator();

	while (itg.isNext())
	{
		auto p = itg.get();

		check &= (vd.template getProp<0>(p) == vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2]);

		check &= (vd.template getProp<1>(p)[0] == vd.getPos(p)[0] + vd.getPos(p)[1]);
		check &= (vd.template getProp<1>(p)[1] == vd.getPos(p)[0] + vd.getPos(p)[2]);
		check &= (vd.template getProp<1>(p)[2] == vd.getPos(p)[1] + vd.getPos(p)[2]);

		check &= (vd.template getProp<2>(p)[0] == vd.getPos(p)[0] + 3.0*vd.getPos(p)[1]);
		check &= (vd.template getProp<2>(p)[1] == vd.getPos(p)[0] + 3.0*vd.getPos(p)[2]);
		check &= (vd.template getProp<2>(p)[2] == vd.getPos(p)[1] + 3.0*vd.getPos(p)[2]);

		++itg;
	}

	size_t tot_s = vd.size_local_with_ghost();

	v_cl.sum(tot_s);
	v_cl.execute();

	// We check that we check something
	BOOST_REQUIRE(tot_s > 1000);
}

incardon's avatar
incardon committed
271 272 273
template<typename vector_type, typename CellList_type, typename CellList_type_cpu>
void check_cell_list_cpu_and_gpu(vector_type & vd, CellList_type & NN, CellList_type_cpu & NN_cpu)
{
incardon's avatar
Fixing  
incardon committed
274
	auto it5 = vd.getDomainIteratorGPU(32);
incardon's avatar
incardon committed
275

incardon's avatar
Fixing  
incardon committed
276
	calculate_force<typename vector_type::stype,decltype(NN.toKernel())><<<it5.wthr,it5.thr>>>(vd.toKernel(),vd.toKernel_sorted(),NN.toKernel(),create_vcluster().rank());
incardon's avatar
incardon committed
277 278 279 280 281 282

	vd.template deviceToHostProp<1,2>();

	bool test = check_force(NN_cpu,vd);
	BOOST_REQUIRE_EQUAL(test,true);

incardon's avatar
incardon committed
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
	// We reset the property 1 on device

	auto rst = vd.getDomainIterator();

	while (rst.isNext())
	{
		auto p = rst.get();

		vd.template getProp<1>(p)[0] = 0.0;
		vd.template getProp<1>(p)[1] = 0.0;
		vd.template getProp<1>(p)[2] = 0.0;

		++rst;
	}

	vd.template hostToDeviceProp<1>();

incardon's avatar
incardon committed
300 301
	// We do exactly the same test as before, but now we completely use the sorted version

incardon's avatar
Fixing  
incardon committed
302
	calculate_force_full_sort<typename vector_type::stype,decltype(NN.toKernel())><<<it5.wthr,it5.thr>>>(vd.toKernel_sorted(),NN.toKernel(),create_vcluster().rank());
incardon's avatar
incardon committed
303

incardon's avatar
incardon committed
304
	vd.template merge_sort<1>(NN);
incardon's avatar
incardon committed
305 306 307 308 309 310
	vd.template deviceToHostProp<1>();

	test = check_force(NN_cpu,vd);
	BOOST_REQUIRE_EQUAL(test,true);
}

incardon's avatar
incardon committed
311 312
template<typename CellList_type>
void vector_dist_gpu_test_impl()
incardon's avatar
incardon committed
313
{
incardon's avatar
incardon committed
314 315 316 317 318
	auto & v_cl = create_vcluster();

	if (v_cl.size() > 16)
	{return;}

incardon's avatar
incardon committed
319 320 321
	Box<3,float> domain({0.0,0.0,0.0},{1.0,1.0,1.0});

	// set the ghost based on the radius cut off (make just a little bit smaller than the spacing)
incardon's avatar
incardon committed
322
	Ghost<3,float> g(0.1);
incardon's avatar
incardon committed
323 324 325 326

	// Boundary conditions
	size_t bc[3]={NON_PERIODIC,NON_PERIODIC,NON_PERIODIC};

incardon's avatar
incardon committed
327
	vector_dist_gpu<3,float,aggregate<float,float[3],float[3]>> vd(10000,domain,bc,g);
incardon's avatar
incardon committed
328

incardon's avatar
incardon committed
329 330
	srand(55067*create_vcluster().rank());

incardon's avatar
incardon committed
331 332 333 334 335 336
	auto it = vd.getDomainIterator();

	while (it.isNext())
	{
		auto p = it.get();

incardon's avatar
incardon committed
337 338 339 340 341 342 343 344 345
		int x = rand();
		int y = rand();
		int z = rand();

		vd.getPos(p)[0] = (float)x / RAND_MAX;
		vd.getPos(p)[1] = (float)y / RAND_MAX;
		vd.getPos(p)[2] = (float)z / RAND_MAX;

		Point<3,float> xp = vd.getPos(p);
incardon's avatar
incardon committed
346 347 348 349

		++it;
	}

350
	// Ok we redistribute the particles (CPU based)
incardon's avatar
incardon committed
351
	vd.map();
incardon's avatar
incardon committed
352 353 354 355 356 357

	size_t size_l = vd.size_local();

	v_cl.sum(size_l);
	v_cl.execute();

incardon's avatar
incardon committed
358
	BOOST_REQUIRE_EQUAL(size_l,10000);
incardon's avatar
incardon committed
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382


	auto & ct = vd.getDecomposition();

	bool noOut = true;
	size_t cnt = 0;

	auto it2 = vd.getDomainIterator();

	while (it2.isNext())
	{
		auto p = it2.get();

		noOut &= ct.isLocal(vd.getPos(p));

		cnt++;
		++it2;
	}

	BOOST_REQUIRE_EQUAL(noOut,true);
	BOOST_REQUIRE_EQUAL(cnt,vd.size_local());

	// now we offload all the properties

incardon's avatar
incardon committed
383 384
	auto it3 = vd.getDomainIteratorGPU();

385
	// offload to device
incardon's avatar
incardon committed
386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408
	vd.hostToDevicePos();

	initialize_props<<<it3.wthr,it3.thr>>>(vd.toKernel());

	// now we check what we initialized

	vd.deviceToHostProp<0,1>();

	auto it4 = vd.getDomainIterator();

	while (it4.isNext())
	{
		auto p = it4.get();

		BOOST_REQUIRE_CLOSE(vd.template getProp<0>(p),vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2],0.01);

		BOOST_REQUIRE_CLOSE(vd.template getProp<1>(p)[0],vd.getPos(p)[0] + vd.getPos(p)[1],0.01);
		BOOST_REQUIRE_CLOSE(vd.template getProp<1>(p)[1],vd.getPos(p)[0] + vd.getPos(p)[2],0.01);
		BOOST_REQUIRE_CLOSE(vd.template getProp<1>(p)[2],vd.getPos(p)[1] + vd.getPos(p)[2],0.01);

		++it4;
	}

409 410 411
	// here we do a ghost_get
	vd.ghost_get<0>();

incardon's avatar
incardon committed
412
	// Double ghost get to check crashes
incardon's avatar
incardon committed
413 414
	vd.ghost_get<0>();

415 416 417 418
	// we re-offload what we received
	vd.hostToDevicePos();
	vd.template hostToDeviceProp<0>();

incardon's avatar
incardon committed
419
	auto NN = vd.template getCellListGPU<CellList_type>(0.1);
incardon's avatar
incardon committed
420
	auto NN_cpu = vd.getCellList(0.1);
421
	check_cell_list_cpu_and_gpu(vd,NN,NN_cpu);
incardon's avatar
incardon committed
422

incardon's avatar
incardon committed
423 424 425 426
	auto NN_up = vd.getCellListGPU(0.1);
	NN_up.clear();
	vd.updateCellList(NN_up);
	check_cell_list_cpu_and_gpu(vd,NN_up,NN_cpu);
incardon's avatar
incardon committed
427 428
}

incardon's avatar
incardon committed
429 430 431 432 433 434 435 436 437 438
BOOST_AUTO_TEST_CASE( vector_dist_gpu_test)
{
	vector_dist_gpu_test_impl<CellList_gpu<3,float,CudaMemory,shift_only<3, float>>>();
}

BOOST_AUTO_TEST_CASE( vector_dist_gpu_test_sparse)
{
	vector_dist_gpu_test_impl<CELLLIST_GPU_SPARSE<3,float>>();
}

incardon's avatar
incardon committed
439 440
template<typename St>
void vdist_calc_gpu_test()
incardon's avatar
Latest  
incardon committed
441 442 443 444 445 446
{
	auto & v_cl = create_vcluster();

	if (v_cl.size() > 16)
	{return;}

incardon's avatar
incardon committed
447 448 449 450 451 452 453 454
	Box<3,St> domain({0.0,0.0,0.0},{1.0,1.0,1.0});

	// set the ghost based on the radius cut off (make just a little bit smaller than the spacing)
	Ghost<3,St> g(0.1);

	// Boundary conditions
	size_t bc[3]={PERIODIC,PERIODIC,PERIODIC};

incardon's avatar
incardon committed
455 456
	//! [Create a gpu vector]

incardon's avatar
incardon committed
457
	vector_dist_gpu<3,St,aggregate<St,St[3],St[3]>> vd(1000,domain,bc,g);
incardon's avatar
incardon committed
458

incardon's avatar
incardon committed
459 460 461 462
	//! [Create a gpu vector]

	//! [Fill gpu vector and move to GPU]

incardon's avatar
incardon committed
463
	srand(v_cl.rank()*10000);
incardon's avatar
incardon committed
464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
	auto it = vd.getDomainIterator();

	while (it.isNext())
	{
		auto p = it.get();

		vd.getPos(p)[0] = (St)rand() / RAND_MAX;
		vd.getPos(p)[1] = (St)rand() / RAND_MAX;
		vd.getPos(p)[2] = (St)rand() / RAND_MAX;

		vd.template getProp<0>(p) = vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2];

		vd.template getProp<1>(p)[0] = vd.getPos(p)[0];
		vd.template getProp<1>(p)[1] = vd.getPos(p)[1];
		vd.template getProp<1>(p)[2] = vd.getPos(p)[2];

		vd.template getProp<2>(p)[0] = vd.getPos(p)[0] + vd.getPos(p)[1];
		vd.template getProp<2>(p)[1] = vd.getPos(p)[0] + vd.getPos(p)[2];
		vd.template getProp<2>(p)[2] = vd.getPos(p)[1] + vd.getPos(p)[2];

		++it;
	}

	// move on device
	vd.hostToDevicePos();
	vd.template hostToDeviceProp<0,1,2>();

	// Ok we redistribute the particles (GPU based)
	vd.map(RUN_ON_DEVICE);

incardon's avatar
incardon committed
494 495
	//! [Fill gpu vector and move to GPU]

incardon's avatar
incardon committed
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574
	vd.deviceToHostPos();
	vd.template deviceToHostProp<0,1,2>();

	// Reset the host part

	auto it3 = vd.getDomainIterator();

	while (it3.isNext())
	{
		auto p = it3.get();

		vd.getPos(p)[0] = 1.0;
		vd.getPos(p)[1] = 1.0;
		vd.getPos(p)[2] = 1.0;

		vd.template getProp<0>(p) = 0.0;

		vd.template getProp<0>(p) = 0.0;
		vd.template getProp<0>(p) = 0.0;
		vd.template getProp<0>(p) = 0.0;

		vd.template getProp<0>(p) = 0.0;
		vd.template getProp<0>(p) = 0.0;
		vd.template getProp<0>(p) = 0.0;

		++it3;
	}

	// we move from Device to CPU

	vd.deviceToHostPos();
	vd.template deviceToHostProp<0,1,2>();

	// Check

	auto it2 = vd.getDomainIterator();

	bool match = true;
	while (it2.isNext())
	{
		auto p = it2.get();

		match &= vd.template getProp<0>(p) == vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2];

		match &= vd.template getProp<1>(p)[0] == vd.getPos(p)[0];
		match &= vd.template getProp<1>(p)[1] == vd.getPos(p)[1];
		match &= vd.template getProp<1>(p)[2] == vd.getPos(p)[2];

		match &= vd.template getProp<2>(p)[0] == vd.getPos(p)[0] + vd.getPos(p)[1];
		match &= vd.template getProp<2>(p)[1] == vd.getPos(p)[0] + vd.getPos(p)[2];
		match &= vd.template getProp<2>(p)[2] == vd.getPos(p)[1] + vd.getPos(p)[2];

		++it2;
	}

	BOOST_REQUIRE_EQUAL(match,true);

	// count local particles

	size_t l_cnt = 0;
	size_t nl_cnt = 0;
	size_t n_out = 0;

	// Domain + ghost box
	Box<3,St> dom_ext = domain;
	dom_ext.enlarge(g);

	auto it5 = vd.getDomainIterator();
	count_local_n_local<3>(vd,it5,bc,domain,dom_ext,l_cnt,nl_cnt,n_out);

	BOOST_REQUIRE_EQUAL(n_out,0);
	BOOST_REQUIRE_EQUAL(l_cnt,vd.size_local());

	// we do 10 gpu steps (using a cpu vector to check that map and ghost get work as expented)

	for (size_t i = 0 ; i < 10 ; i++)
	{
		vd.map(RUN_ON_DEVICE);

incardon's avatar
incardon committed
575 576
		CUDA_SAFE(cudaGetLastError());

incardon's avatar
incardon committed
577 578 579 580 581
		vd.deviceToHostPos();
		vd.template deviceToHostProp<0,1,2>();

		// To test we copy on a cpu distributed vector and we do a map

incardon's avatar
incardon committed
582
		vector_dist<3,St,aggregate<St,St[3],St[3]>> vd_cpu(vd.getDecomposition().template duplicate_convert<HeapMemory,memory_traits_lin>(),0);
incardon's avatar
incardon committed
583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609

		auto itc = vd.getDomainIterator();

		while (itc.isNext())
		{
			auto p = itc.get();

			vd_cpu.add();

			vd_cpu.getLastPos()[0] = vd.getPos(p)[0];
			vd_cpu.getLastPos()[1] = vd.getPos(p)[1];
			vd_cpu.getLastPos()[2] = vd.getPos(p)[2];

			vd_cpu.template getLastProp<0>() = vd.template getProp<0>(p);

			vd_cpu.template getLastProp<1>()[0] = vd.template getProp<1>(p)[0];
			vd_cpu.template getLastProp<1>()[1] = vd.template getProp<1>(p)[1];
			vd_cpu.template getLastProp<1>()[2] = vd.template getProp<1>(p)[2];

			vd_cpu.template getLastProp<2>()[0] = vd.template getProp<2>(p)[0];
			vd_cpu.template getLastProp<2>()[1] = vd.template getProp<2>(p)[1];
			vd_cpu.template getLastProp<2>()[2] = vd.template getProp<2>(p)[2];

			++itc;
		}

		vd_cpu.template ghost_get<0,1,2>();
incardon's avatar
incardon committed
610 611 612

		//! [Fill the ghost on GPU]

incardon's avatar
incardon committed
613 614
		vd.template ghost_get<0,1,2>(RUN_ON_DEVICE);

incardon's avatar
incardon committed
615 616
		//! [Fill the ghost on GPU]

incardon's avatar
incardon committed
617 618 619 620 621 622 623 624 625 626 627
		vd.deviceToHostPos();
		vd.template deviceToHostProp<0,1,2>();

		match = true;

		// Particle on the gpu ghost and cpu ghost are not ordered in the same way so we have to reorder

		struct part
		{
			Point<3,St> xp;

incardon's avatar
incardon committed
628 629 630 631

			St prp0;
			St prp1[3];
			St prp2[3];
incardon's avatar
incardon committed
632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659

			bool operator<(const part & tmp) const
			{
				if (xp.get(0) < tmp.xp.get(0))
				{return true;}
				else if (xp.get(0) > tmp.xp.get(0))
				{return false;}

				if (xp.get(1) < tmp.xp.get(1))
				{return true;}
				else if (xp.get(1) > tmp.xp.get(1))
				{return false;}

				if (xp.get(2) < tmp.xp.get(2))
				{return true;}
				else if (xp.get(2) > tmp.xp.get(2))
				{return false;}

				return false;
			}
		};

		openfpm::vector<part> cpu_sort;
		openfpm::vector<part> gpu_sort;

		cpu_sort.resize(vd_cpu.size_local_with_ghost() - vd_cpu.size_local());
		gpu_sort.resize(vd.size_local_with_ghost() - vd.size_local());

incardon's avatar
incardon committed
660 661
		BOOST_REQUIRE_EQUAL(cpu_sort.size(),gpu_sort.size());

incardon's avatar
incardon committed
662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728
		size_t cnt = 0;

		auto itc2 = vd.getGhostIterator();
		while (itc2.isNext())
		{
			auto p = itc2.get();

			cpu_sort.get(cnt).xp.get(0) = vd_cpu.getPos(p)[0];
			gpu_sort.get(cnt).xp.get(0) = vd.getPos(p)[0];
			cpu_sort.get(cnt).xp.get(1) = vd_cpu.getPos(p)[1];
			gpu_sort.get(cnt).xp.get(1) = vd.getPos(p)[1];
			cpu_sort.get(cnt).xp.get(2) = vd_cpu.getPos(p)[2];
			gpu_sort.get(cnt).xp.get(2) = vd.getPos(p)[2];

			cpu_sort.get(cnt).prp0 = vd_cpu.template getProp<0>(p);
			gpu_sort.get(cnt).prp0 = vd.template getProp<0>(p);

			cpu_sort.get(cnt).prp1[0] = vd_cpu.template getProp<1>(p)[0];
			gpu_sort.get(cnt).prp1[0] = vd.template getProp<1>(p)[0];
			cpu_sort.get(cnt).prp1[1] = vd_cpu.template getProp<1>(p)[1];
			gpu_sort.get(cnt).prp1[1] = vd.template getProp<1>(p)[1];
			cpu_sort.get(cnt).prp1[2] = vd_cpu.template getProp<1>(p)[2];
			gpu_sort.get(cnt).prp1[2] = vd.template getProp<1>(p)[2];

			cpu_sort.get(cnt).prp2[0] = vd_cpu.template getProp<2>(p)[0];
			gpu_sort.get(cnt).prp2[0] = vd.template getProp<2>(p)[0];
			cpu_sort.get(cnt).prp2[1] = vd_cpu.template getProp<2>(p)[1];
			gpu_sort.get(cnt).prp2[1] = vd.template getProp<2>(p)[1];
			cpu_sort.get(cnt).prp2[2] = vd_cpu.template getProp<2>(p)[2];
			gpu_sort.get(cnt).prp2[2] = vd.template getProp<2>(p)[2];

			++cnt;
			++itc2;
		}

		cpu_sort.sort();
		gpu_sort.sort();

		for (size_t i = 0 ; i < cpu_sort.size() ; i++)
		{
			match &= cpu_sort.get(i).xp.get(0) == gpu_sort.get(i).xp.get(0);
			match &= cpu_sort.get(i).xp.get(1) == gpu_sort.get(i).xp.get(1);
			match &= cpu_sort.get(i).xp.get(2) == gpu_sort.get(i).xp.get(2);

			match &= cpu_sort.get(i).prp0 == gpu_sort.get(i).prp0;
			match &= cpu_sort.get(i).prp1[0] == gpu_sort.get(i).prp1[0];
			match &= cpu_sort.get(i).prp1[1] == gpu_sort.get(i).prp1[1];
			match &= cpu_sort.get(i).prp1[2] == gpu_sort.get(i).prp1[2];

			match &= cpu_sort.get(i).prp2[0] == gpu_sort.get(i).prp2[0];
			match &= cpu_sort.get(i).prp2[1] == gpu_sort.get(i).prp2[1];
			match &= cpu_sort.get(i).prp2[2] == gpu_sort.get(i).prp2[2];
		}

		BOOST_REQUIRE_EQUAL(match,true);

		// move particles on gpu

		auto ite = vd.getDomainIteratorGPU();
		move_parts_gpu_test<3,decltype(vd.toKernel())><<<ite.wthr,ite.thr>>>(vd.toKernel());
	}
}

BOOST_AUTO_TEST_CASE( vector_dist_map_on_gpu_test)
{
	vdist_calc_gpu_test<float>();
	vdist_calc_gpu_test<double>();
729
}
incardon's avatar
incardon committed
730

731 732 733
BOOST_AUTO_TEST_CASE(vector_dist_reduce)
{
	auto & v_cl = create_vcluster();
incardon's avatar
incardon committed
734 735 736 737

	if (v_cl.size() > 16)
	{return;}

incardon's avatar
Latest  
incardon committed
738 739 740 741 742 743
	Box<3,float> domain({0.0,0.0,0.0},{1.0,1.0,1.0});

	// set the ghost based on the radius cut off (make just a little bit smaller than the spacing)
	Ghost<3,float> g(0.1);

	// Boundary conditions
incardon's avatar
incardon committed
744
	size_t bc[3]={PERIODIC,PERIODIC,PERIODIC};
incardon's avatar
Latest  
incardon committed
745

746
	vector_dist_gpu<3,float,aggregate<float,double,int,size_t>> vd(5000*v_cl.size(),domain,bc,g);
incardon's avatar
Latest  
incardon committed
747 748 749

	auto it = vd.getDomainIterator();

750 751 752 753 754 755
	float fc = 1.0;
	double dc = 1.0;
	int ic = 1.0;
	size_t sc = 1.0;

	while(it.isNext())
incardon's avatar
Latest  
incardon committed
756 757 758
	{
		auto p = it.get();

759 760 761 762
		vd.template getProp<0>(p) = fc;
		vd.template getProp<1>(p) = dc;
		vd.template getProp<2>(p) = ic;
		vd.template getProp<3>(p) = sc;
incardon's avatar
Latest  
incardon committed
763

764 765 766 767
		fc += 1.0;
		dc += 1.0;
		ic += 1;
		sc += 1;
768

incardon's avatar
Latest  
incardon committed
769 770 771
		++it;
	}

772
	vd.template hostToDeviceProp<0,1,2,3>();
incardon's avatar
incardon committed
773

774 775 776 777
	float redf = reduce_local<0,_add_>(vd);
	double redd = reduce_local<1,_add_>(vd);
	int redi = reduce_local<2,_add_>(vd);
	size_t reds = reduce_local<3,_add_>(vd);
incardon's avatar
incardon committed
778

779 780 781 782
	BOOST_REQUIRE_EQUAL(redf,(vd.size_local()+1.0)*(vd.size_local())/2.0);
	BOOST_REQUIRE_EQUAL(redd,(vd.size_local()+1.0)*(vd.size_local())/2.0);
	BOOST_REQUIRE_EQUAL(redi,(vd.size_local()+1)*(vd.size_local())/2);
	BOOST_REQUIRE_EQUAL(reds,(vd.size_local()+1)*(vd.size_local())/2);
incardon's avatar
incardon committed
783

784 785 786 787
	float redf2 = reduce_local<0,_max_>(vd);
	double redd2 = reduce_local<1,_max_>(vd);
	int redi2 = reduce_local<2,_max_>(vd);
	size_t reds2 = reduce_local<3,_max_>(vd);
incardon's avatar
incardon committed
788 789 790 791 792

	BOOST_REQUIRE_EQUAL(redf2,vd.size_local());
	BOOST_REQUIRE_EQUAL(redd2,vd.size_local());
	BOOST_REQUIRE_EQUAL(redi2,vd.size_local());
	BOOST_REQUIRE_EQUAL(reds2,vd.size_local());
incardon's avatar
Latest  
incardon committed
793 794
}

795
void vector_dist_dlb_on_cuda_impl(size_t k,double r_cut)
incardon's avatar
incardon committed
796
{
incardon's avatar
Fixing  
incardon committed
797
	typedef vector_dist_gpu<3,double,aggregate<double,double[3],double[3]>> vector_type;
incardon's avatar
incardon committed
798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813

	Vcluster<> & v_cl = create_vcluster();

	if (v_cl.getProcessingUnits() > 8)
		return;

	Box<3,double> domain({0.0,0.0,0.0},{1.0,1.0,1.0});
	Ghost<3,double> g(0.1);
	size_t bc[3] = {PERIODIC,PERIODIC,PERIODIC};

	vector_type vd(0,domain,bc,g,DEC_GRAN(2048));

	// Only processor 0 initialy add particles on a corner of a domain

	if (v_cl.getProcessUnitID() == 0)
	{
814
		for(size_t i = 0 ; i < k ; i++)
incardon's avatar
incardon committed
815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837
		{
			vd.add();

			vd.getLastPos()[0] = ((double)rand())/RAND_MAX * 0.3;
			vd.getLastPos()[1] = ((double)rand())/RAND_MAX * 0.3;
			vd.getLastPos()[2] = ((double)rand())/RAND_MAX * 0.3;
		}
	}

	// Move to GPU
	vd.hostToDevicePos();
	vd.template hostToDeviceProp<0>();

	vd.map(RUN_ON_DEVICE);
	vd.template ghost_get<>(RUN_ON_DEVICE);

	// now move to CPU

	vd.deviceToHostPos();
	vd.template deviceToHostProp<0>();

	// Get the neighborhood of each particles

838
	auto VV = vd.getVerlet(r_cut);
incardon's avatar
incardon committed
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870

	// store the number of neighborhood for each particles

	auto it = vd.getDomainIterator();

	while (it.isNext())
	{
		auto p = it.get();

		vd.template getProp<0>(p) = VV.getNNPart(p.getKey());

		++it;
	}

	// Move to GPU
	vd.template hostToDeviceProp<0>();

	ModelSquare md;
	md.factor = 10;
	vd.addComputationCosts(md);
	vd.getDecomposition().decompose();
	vd.map(RUN_ON_DEVICE);

	vd.deviceToHostPos();
	// Move info to CPU for addComputationcosts

	vd.addComputationCosts(md);

	openfpm::vector<size_t> loads;
	size_t load = vd.getDecomposition().getDistribution().getProcessorLoad();
	v_cl.allGather(load,loads);
	v_cl.execute();
871

incardon's avatar
incardon committed
872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905
	for (size_t i = 0 ; i < loads.size() ; i++)
	{
		double load_f = load;
		double load_fc = loads.get(i);

		BOOST_REQUIRE_CLOSE(load_f,load_fc,7.0);
	}

	BOOST_REQUIRE(vd.size_local() != 0);

	Point<3,double> v({1.0,1.0,1.0});

	for (size_t i = 0 ; i < 25 ; i++)
	{
		// move particles to CPU and move the particles by 0.1

		vd.deviceToHostPos();

		auto it = vd.getDomainIterator();

		while (it.isNext())
		{
			auto p = it.get();

			vd.getPos(p)[0] += v.get(0) * 0.09;
			vd.getPos(p)[1] += v.get(1) * 0.09;
			vd.getPos(p)[2] += v.get(2) * 0.09;

			++it;
		}

		//Back to GPU
		vd.hostToDevicePos();
		vd.map(RUN_ON_DEVICE);
incardon's avatar
Fixing  
incardon committed
906
		vd.template ghost_get<0>(RUN_ON_DEVICE);
incardon's avatar
incardon committed
907
		vd.deviceToHostPos();
incardon's avatar
Fixing  
incardon committed
908 909 910 911 912 913
		vd.template deviceToHostProp<0,1,2>();

		// Check calc forces
		auto NN_gpu = vd.getCellListGPU(r_cut);
		auto NN_cpu = vd.getCellList(r_cut);
		check_cell_list_cpu_and_gpu(vd,NN_gpu,NN_cpu);
incardon's avatar
incardon committed
914

915
		auto VV2 = vd.getVerlet(r_cut);
incardon's avatar
incardon committed
916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934

		auto it2 = vd.getDomainIterator();

		bool match = true;
		while (it2.isNext())
		{
			auto p = it2.get();

			match &= vd.template getProp<0>(p) == VV2.getNNPart(p.getKey());

			++it2;
		}

		BOOST_REQUIRE_EQUAL(match,true);

		ModelSquare md;
		vd.addComputationCosts(md);
		vd.getDecomposition().redecompose(200);
		vd.map(RUN_ON_DEVICE);
incardon's avatar
incardon committed
935

incardon's avatar
incardon committed
936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
		BOOST_REQUIRE(vd.size_local() != 0);

		vd.template ghost_get<0>(RUN_ON_DEVICE);
		vd.deviceToHostPos();
		vd.template deviceToHostProp<0>();

		vd.addComputationCosts(md);

		openfpm::vector<size_t> loads;
		size_t load = vd.getDecomposition().getDistribution().getProcessorLoad();
		v_cl.allGather(load,loads);
		v_cl.execute();

		for (size_t i = 0 ; i < loads.size() ; i++)
		{
			double load_f = load;
			double load_fc = loads.get(i);

			BOOST_REQUIRE_CLOSE(load_f,load_fc,10.0);
		}
	}
}
incardon's avatar
incardon committed
958

959 960 961 962 963 964 965
BOOST_AUTO_TEST_CASE(vector_dist_dlb_on_cuda)
{
	vector_dist_dlb_on_cuda_impl(50000,0.01);
}

BOOST_AUTO_TEST_CASE(vector_dist_dlb_on_cuda2)
{
incardon's avatar
incardon committed
966
	if (create_vcluster().size() <= 3)
967 968 969 970 971 972 973 974 975 976 977 978 979 980
	{return;};

	vector_dist_dlb_on_cuda_impl(1000000,0.01);
}

BOOST_AUTO_TEST_CASE(vector_dist_dlb_on_cuda3)
{
	if (create_vcluster().size() < 8)
	{return;}

	vector_dist_dlb_on_cuda_impl(15000000,0.005);
}


incardon's avatar
incardon committed
981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086
BOOST_AUTO_TEST_CASE(vector_dist_keep_prop_on_cuda)
{
	typedef vector_dist_gpu<3,double,aggregate<double,double[3],double[3][3]>> vector_type;

	Vcluster<> & v_cl = create_vcluster();

	if (v_cl.getProcessingUnits() > 8)
		return;

	Box<3,double> domain({0.0,0.0,0.0},{1.0,1.0,1.0});
	Ghost<3,double> g(0.1);
	size_t bc[3] = {PERIODIC,PERIODIC,PERIODIC};

	vector_type vd(0,domain,bc,g,DEC_GRAN(2048));

	// Only processor 0 initialy add particles on a corner of a domain

	if (v_cl.getProcessUnitID() == 0)
	{
		for(size_t i = 0 ; i < 50000 ; i++)
		{
			vd.add();

			vd.getLastPos()[0] = ((double)rand())/RAND_MAX * 0.3;
			vd.getLastPos()[1] = ((double)rand())/RAND_MAX * 0.3;
			vd.getLastPos()[2] = ((double)rand())/RAND_MAX * 0.3;
		}
	}

	// Move to GPU
	vd.hostToDevicePos();
	vd.template hostToDeviceProp<0>();

	vd.map(RUN_ON_DEVICE);
	vd.template ghost_get<>(RUN_ON_DEVICE);

	// now move to CPU

	vd.deviceToHostPos();
	vd.template deviceToHostProp<0>();


	// store the number of neighborhood for each particles

	auto it = vd.getDomainIterator();

	while (it.isNext())
	{
		auto p = it.get();

		vd.template getProp<0>(p) = 0.0;

		vd.template getProp<1>(p)[0] = 1000.0;
		vd.template getProp<1>(p)[1] = 2000.0;
		vd.template getProp<1>(p)[2] = 3000.0;

		vd.template getProp<2>(p)[0][0] = 6000,0;
		vd.template getProp<2>(p)[0][1] = 7000.0;
		vd.template getProp<2>(p)[0][2] = 8000.0;
		vd.template getProp<2>(p)[1][0] = 9000.0;
		vd.template getProp<2>(p)[1][1] = 10000.0;
		vd.template getProp<2>(p)[1][2] = 11000.0;
		vd.template getProp<2>(p)[2][0] = 12000.0;
		vd.template getProp<2>(p)[2][1] = 13000.0;
		vd.template getProp<2>(p)[2][2] = 14000.0;

		++it;
	}

	// Move to GPU
	vd.template hostToDeviceProp<0,1,2>();

	ModelSquare md;
	md.factor = 10;
	vd.addComputationCosts(md);
	vd.getDecomposition().decompose();
	vd.map(RUN_ON_DEVICE);

	vd.deviceToHostPos();
	// Move info to CPU for addComputationcosts

	vd.addComputationCosts(md);

	openfpm::vector<size_t> loads;
	size_t load = vd.getDecomposition().getDistribution().getProcessorLoad();
	v_cl.allGather(load,loads);
	v_cl.execute();

	for (size_t i = 0 ; i < loads.size() ; i++)
	{
		double load_f = load;
		double load_fc = loads.get(i);

		BOOST_REQUIRE_CLOSE(load_f,load_fc,7.0);
	}

	BOOST_REQUIRE(vd.size_local() != 0);

	Point<3,double> v({1.0,1.0,1.0});

	int base = 0;

	for (size_t i = 0 ; i < 25 ; i++)
	{
		if (i % 2 == 0)
		{
incardon's avatar
incardon committed
1087 1088 1089
			// move particles to CPU and move the particles by 0.1

			vd.deviceToHostPos();
incardon's avatar
incardon committed
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138

			auto it = vd.getDomainIterator();

			while (it.isNext())
			{
				auto p = it.get();

				vd.getPos(p)[0] += v.get(0) * 0.09;
				vd.getPos(p)[1] += v.get(1) * 0.09;
				vd.getPos(p)[2] += v.get(2) * 0.09;

				++it;
			}

			//Back to GPU
			vd.hostToDevicePos();
			vd.map(RUN_ON_DEVICE);
			vd.template ghost_get<>(RUN_ON_DEVICE);
			vd.deviceToHostPos();
			vd.template deviceToHostProp<0,1,2>();

			ModelSquare md;
			vd.addComputationCosts(md);
			vd.getDecomposition().redecompose(200);
			vd.map(RUN_ON_DEVICE);

			BOOST_REQUIRE(vd.size_local() != 0);

			vd.template ghost_get<0>(RUN_ON_DEVICE);
			vd.deviceToHostPos();
			vd.template deviceToHostProp<0,1,2>();

			vd.addComputationCosts(md);

			openfpm::vector<size_t> loads;
			size_t load = vd.getDecomposition().getDistribution().getProcessorLoad();
			v_cl.allGather(load,loads);
			v_cl.execute();

			for (size_t i = 0 ; i < loads.size() ; i++)
			{
				double load_f = load;
				double load_fc = loads.get(i);

				BOOST_REQUIRE_CLOSE(load_f,load_fc,10.0);
			}
		}
		else
		{
incardon's avatar
incardon committed
1139
			vd.template deviceToHostProp<0,1,2>();
incardon's avatar
incardon committed
1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166

			auto it2 = vd.getDomainIterator();

			bool match = true;
			while (it2.isNext())
			{
				auto p = it2.get();

				vd.template getProp<0>(p) += 1;

				vd.template getProp<1>(p)[0] += 1.0;
				vd.template getProp<1>(p)[1] += 1.0;
				vd.template getProp<1>(p)[2] += 1.0;

				vd.template getProp<2>(p)[0][0] += 1.0;
				vd.template getProp<2>(p)[0][1] += 1.0;
				vd.template getProp<2>(p)[0][2] += 1.0;
				vd.template getProp<2>(p)[1][0] += 1.0;
				vd.template getProp<2>(p)[1][1] += 1.0;
				vd.template getProp<2>(p)[1][2] += 1.0;
				vd.template getProp<2>(p)[2][0] += 1.0;
				vd.template getProp<2>(p)[2][1] += 1.0;
				vd.template getProp<2>(p)[2][2] += 1.0;

				++it2;
			}

incardon's avatar
incardon committed
1167 1168
			vd.template hostToDeviceProp<0,1,2>();

incardon's avatar
incardon committed
1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
			++base;

			vd.template ghost_get<0,1,2>(RUN_ON_DEVICE | KEEP_PROPERTIES);
			vd.template deviceToHostProp<0,1,2>();

			// Check that the ghost contain the correct information

			auto itg = vd.getGhostIterator();

			while (itg.isNext())
			{
				auto p = itg.get();

				match &= vd.template getProp<0>(p) == base;

				match &= vd.template getProp<1>(p)[0] == base + 1000.0;
				match &= vd.template getProp<1>(p)[1] == base + 2000.0;
				match &= vd.template getProp<1>(p)[2] == base + 3000.0;

				match &= vd.template getProp<2>(p)[0][0] == base + 6000.0;
				match &= vd.template getProp<2>(p)[0][1] == base + 7000.0;
				match &= vd.template getProp<2>(p)[0][2] == base + 8000.0;
				match &= vd.template getProp<2>(p)[1][0] == base + 9000.0;
				match &= vd.template getProp<2>(p)[1][1] == base + 10000.0;
				match &= vd.template getProp<2>(p)[1][2] == base + 11000.0;
				match &= vd.template getProp<2>(p)[2][0] == base + 12000.0;
				match &= vd.template getProp<2>(p)[2][1] == base + 13000.0;
				match &= vd.template getProp<2>(p)[2][2] == base + 14000.0;

				++itg;
			}

			BOOST_REQUIRE_EQUAL(match,true);
		}
	}
}

incardon's avatar
incardon committed
1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318

BOOST_AUTO_TEST_CASE(vector_dist_compare_host_device)
{
	Box<3,double> domain({0.0,0.0,0.0},{1.0,1.0,1.0});
	Ghost<3,double> g(0.1);
	size_t bc[3] = {PERIODIC,PERIODIC,PERIODIC};

	if (create_vcluster().size() >= 16)
	{return;}

	vector_dist_gpu<3,double,aggregate<double,double[3],double[3][3]>> vdg(10000,domain,bc,g,DEC_GRAN(128));

	auto it = vdg.getDomainIterator();

	while (it.isNext())
	{
		auto p = it.get();

		vdg.getPos(p)[0] = (double)rand() / RAND_MAX;
		vdg.getPos(p)[1] = (double)rand() / RAND_MAX;
		vdg.getPos(p)[2] = (double)rand() / RAND_MAX;

		vdg.template getProp<0>(p) = (double)rand() / RAND_MAX;

		vdg.template getProp<1>(p)[0] = (double)rand() / RAND_MAX;
		vdg.template getProp<1>(p)[1] = (double)rand() / RAND_MAX;
		vdg.template getProp<1>(p)[2] = (double)rand() / RAND_MAX;

		vdg.template getProp<2>(p)[0][0] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[0][1] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[0][2] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[1][0] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[1][1] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[1][2] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[2][0] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[2][1] = (double)rand() / RAND_MAX;
		vdg.template getProp<2>(p)[2][2] = (double)rand() / RAND_MAX;

		++it;
	}

	vdg.map();

	vdg.hostToDeviceProp<0,1,2>();
	vdg.hostToDevicePos();

	bool test = vdg.compareHostAndDevicePos(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);

	vdg.getPos(100)[0] = 0.99999999;

	test = vdg.compareHostAndDevicePos(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,false);

	vdg.hostToDevicePos();
	vdg.getPos(100)[0] = 0.99999999;

	test = vdg.compareHostAndDevicePos(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);

	////////////////////////////////////////////////// PROP VECTOR

	test = vdg.compareHostAndDeviceProp<1>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);

	vdg.getProp<1>(103)[0] = 0.99999999;

	test = vdg.compareHostAndDeviceProp<1>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,false);

	vdg.hostToDeviceProp<1>();
	vdg.getProp<1>(103)[0] = 0.99999999;

	test = vdg.compareHostAndDeviceProp<1>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);

	////////////////////////////////////////////////// PROP scalar


	test = vdg.compareHostAndDeviceProp<0>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);

	vdg.getProp<0>(105) = 0.99999999;

	test = vdg.compareHostAndDeviceProp<0>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,false);

	vdg.hostToDeviceProp<0>();
	vdg.getProp<0>(105) = 0.99999999;

	test = vdg.compareHostAndDeviceProp<0>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);


	////////////////////////////////////////////////// PROP scalar


	test = vdg.compareHostAndDeviceProp<2>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);

	vdg.getProp<2>(108)[1][2] = 0.99999999;

	test = vdg.compareHostAndDeviceProp<2>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,false);

	vdg.hostToDeviceProp<2>();
	vdg.getProp<2>(108)[1][2] = 0.99999999;

	test = vdg.compareHostAndDeviceProp<2>(0.00001,0.00000001);
	BOOST_REQUIRE_EQUAL(test,true);
}


incardon's avatar
incardon committed
1319
BOOST_AUTO_TEST_SUITE_END()