vector_dist_gpu_unit_tests.cu 12.1 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
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
18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78

__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];
}

template<typename CellList_type>
__global__  void calculate_force(vector_dist_ker<3, float, aggregate<float, float[3], float [3]>> vd,
		                         vector_dist_ker<3, float, aggregate<float, float[3], float [3]>> vd_sort,
		                         CellList_type cl)
{
	auto p = GET_PARTICLE(vd);

	Point<3,float> xp = vd.getPos(p);

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

    auto cell = cl.getCell(xp);

    Point<3,float> force1({0.0,0.0,0.0});
    Point<3,float> force2({0.0,0.0,0.0});

    while (it.isNext())
    {
    	auto q1 = it.get();
    	auto q2 = it.get_orig();

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

    	Point<3,float> xq_1 = vd_sort.getPos(q1);
    	Point<3,float> xq_2 = vd.getPos(q2);

    	Point<3,float> r1 = xq_1 - xp;
    	Point<3,float> r2 = xq_2 - xp;

    	// 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);
}

79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
template<typename CellList_type>
__global__  void calculate_force_full_sort(vector_dist_ker<3, float, aggregate<float, float[3], float [3]>> vd,
		                         	 	   CellList_type cl)
{
	auto p = GET_PARTICLE(vd);

	Point<3,float> xp = vd.getPos(p);

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

    auto cell = cl.getCell(xp);

    Point<3,float> force1({0.0,0.0,0.0});

    while (it.isNext())
    {
    	auto q1 = it.get();

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

    	Point<3,float> xq_1 = vd.getPos(q1);

    	Point<3,float> r1 = xq_1 - xp;

    	// 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)
{
	auto it6 = vd.getDomainIterator();

	bool match = true;

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

		Point<3,float> xp = vd.getPos(p);

		// Calculate on CPU

		Point<3,float> force({0.0,0.0,0.0});

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

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

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

	    	Point<3,float> xq_2 = vd.getPos(q);
	    	Point<3,float> r2 = xq_2 - xp;

	    	// Normalize

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

			++NNc;
		}

		match &= fabs(vd.template getProp<1>(p)[0] - vd.template getProp<2>(p)[0]) < 0.0001;
		match &= fabs(vd.template getProp<1>(p)[1] - vd.template getProp<2>(p)[1]) < 0.0001;
		match &= fabs(vd.template getProp<1>(p)[2] - vd.template getProp<2>(p)[2]) < 0.0001;

		match &= fabs(vd.template getProp<1>(p)[0] - force.get(0)) < 0.0001;
		match &= fabs(vd.template getProp<1>(p)[1] - force.get(1)) < 0.0001;
		match &= fabs(vd.template getProp<1>(p)[2] - force.get(2)) < 0.0001;

		++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
245 246
BOOST_AUTO_TEST_CASE( vector_dist_gpu_test)
{
incardon's avatar
incardon committed
247 248 249 250 251
	auto & v_cl = create_vcluster();

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

incardon's avatar
incardon committed
252 253 254
	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
255
	Ghost<3,float> g(0.1);
incardon's avatar
incardon committed
256 257 258 259

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

incardon's avatar
incardon committed
260
	vector_dist_gpu<3,float,aggregate<float,float[3],float[3]>> vd(1000,domain,bc,g);
incardon's avatar
incardon committed
261 262 263 264 265 266 267 268 269 270 271 272 273 274

	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;

		++it;
	}

275
	// Ok we redistribute the particles (CPU based)
incardon's avatar
incardon committed
276
	vd.map();
incardon's avatar
incardon committed
277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309

	size_t size_l = vd.size_local();

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

	BOOST_REQUIRE_EQUAL(size_l,1000);


	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());

	vd.write("test_out_gpu");

	// now we offload all the properties

incardon's avatar
incardon committed
310 311
	auto it3 = vd.getDomainIteratorGPU();

312
	// offload to device
incardon's avatar
incardon committed
313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337
	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);

		//std::cout << "PROP 0 " << vd.template getProp<0>(p) << "   " << vd.getPos(p)[0] + vd.getPos(p)[1] + vd.getPos(p)[2] << std::endl;

		++it4;
	}

338 339 340 341 342 343 344
	// here we do a ghost_get
	vd.ghost_get<0>();

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

incardon's avatar
incardon committed
345 346 347 348 349 350 351 352 353
	auto NN = vd.getCellListGPU(0.1);
	auto NN_cpu = vd.getCellList(0.1);

	auto it5 = vd.getDomainIteratorGPU();

	calculate_force<decltype(NN.toKernel())><<<it5.wthr,it5.thr>>>(vd.toKernel(),vd.toKernel_sorted(),NN.toKernel());

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

354 355
	bool test = check_force(NN_cpu,vd);
	BOOST_REQUIRE_EQUAL(test,true);
incardon's avatar
incardon committed
356

357
	// We do exactly the same test as before, but now we completely use the sorted version
incardon's avatar
incardon committed
358

359
	calculate_force_full_sort<decltype(NN.toKernel())><<<it5.wthr,it5.thr>>>(vd.toKernel_sorted(),NN.toKernel());
incardon's avatar
incardon committed
360

361
	vd.template deviceToHostProp<1>();
incardon's avatar
incardon committed
362

363 364
	test = check_force(NN_cpu,vd);
	BOOST_REQUIRE_EQUAL(test,true);
incardon's avatar
incardon committed
365

366
	// check
incardon's avatar
incardon committed
367

368
	// Now we do a ghost_get from CPU
incardon's avatar
incardon committed
369

370
	// Than we offload on GPU
incardon's avatar
incardon committed
371

372
	// We construct a Cell-list
incardon's avatar
incardon committed
373

374
	// We calculate force on CPU and GPU to check if they match
incardon's avatar
incardon committed
375 376


incardon's avatar
incardon committed
377

incardon's avatar
incardon committed
378 379
}

incardon's avatar
Latest  
incardon committed
380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
BOOST_AUTO_TEST_CASE( vector_dist_map_on_gpu_test)
{
	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]={NON_PERIODIC,NON_PERIODIC,NON_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;

407 408 409 410 411 412 413 414 415 416
		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];

incardon's avatar
Latest  
incardon committed
417 418 419
		++it;
	}

420 421 422 423
	// move on device
	vd.hostToDevicePos();
	vd.hostToDeviceProp<0,1,2>();

incardon's avatar
incardon committed
424
	// Ok we redistribute the particles (GPU based)
incardon's avatar
incardon committed
425
	vd.map(RUN_ON_DEVICE);
incardon's avatar
incardon committed
426

427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
	// 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.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,float> 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());
incardon's avatar
incardon committed
496

497 498 499 500
	vd.deviceToHostPos();
	vd.deviceToHostProp<0,1,2>();

	vd.write("gpu_write_before_test");
incardon's avatar
incardon committed
501 502

	vd.ghost_get<0,1,2>(RUN_ON_DEVICE);
503 504 505 506 507

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

	vd.write("gpu_write_after_test");
incardon's avatar
Latest  
incardon committed
508 509
}

510

incardon's avatar
incardon committed
511
BOOST_AUTO_TEST_SUITE_END()