Pedro Roque / PHACT_src

config_device.c 57.9 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 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 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 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 1087 1088 1089 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 1139 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 1167 1168 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 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 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 
/*
 * config_device.c
 *
 *  Created on: 04/03/2016
 *      Author: Pedro
 */

#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

// for elapsed time calculation under windows
#if defined(WIN32) || defined(_WIN32) || defined(__WIN32) && !defined(__CYGWIN__)
#include <time.h>
#include <Winsock2.h>
#include <stdint.h>
#else
#include <sys/time.h>
#endif

#define CL_USE_DEPRECATED_OPENCL_1_2_APIS

#include "CL/cl.h"
#include "CL/cl_platform.h"

#include "config_device.h"
#include "bitmaps.h"
#include "kernels/cl_constraints.h"
#include "kernels/cl_variables.h"
#include "config.h"
#include "devices.h"
#include "constraints.h"
#include "split.h"
#include "utils/dev_errors.h"
#include "utils/benchmark.h"
#include "variables.h"

#if RUN_IN_CUDA
#include <cuda.h>
#include <builtin_types.h>
#endif

/*
 * Initialize device to do labeling, propagation and backtracking on device
 * dev_info - device_info structure about this device
 * dev_args - device_args structure about this device
 * filtering - if its the prefiltering phase
 */
void init_device(device_info *dev_info, device_args *dev_args, bool filtering) {

#if RUN_IN_CUDA

	if (dev_info->type == CL_DEVICE_TYPE_GPU) {

		// for elapse time calculation
		struct timeval start_cu, end_cu;
		char start_time_cu[40];
		char end_time_cu[40];
		char elapsed_time_cu[40];

		char cu_build_opt[3000];	// build options for NVCC
		char cu_cs_usage[200];		// to tell CUDA compiler which constraints should be compiled and which ones uses reification
		int max_com_size = 4000;
		char command_cp[max_com_size];	// command line execution command
		const char kernel_filename[] = "src/kernels/cl_explore.cl";			// name of the file containing the kernel that will be compiled by CUDA NVCC
		const char fatbin_filename[] = "src/kernels/cl_explore.fatbin";		// CUDA binary file after NVCC compilation for CUDA runtime compilation
		const char kernel_name[] = "explore";	// name of the kernel function
		CUresult err = cuInit(0);				// return value from CUDA calls
		unsigned int k, l;

		// get first CUDA device
		err = cuDeviceGet(&dev_args->device_cu, 0);
		if (err != CUDA_SUCCESS) {
			fprintf(stderr, "\nError: CUDA error on cuDeviceGet\n");

#if defined(WIN32) || defined(_WIN32) || defined(__WIN32) && !defined(__CYGWIN__)
			printf("\nPress any key to exit\n");
			int a = getchar();
#endif
			exit(0);
		}

		char name[100];
		err = cuDeviceGetName(name, 100, dev_args->device_cu);
		if (err != CUDA_SUCCESS) {
			fprintf(stderr, "\nError: CUDA error on cuDeviceGetName\n");

#if defined(WIN32) || defined(_WIN32) || defined(__WIN32) && !defined(__CYGWIN__)
			printf("\nPress any key to exit\n");
			int a = getchar();
#endif
			exit(0);
		}

		// add -save-temps to save temporary files during kernel compilation
		// add -cl-nv-verbose to enable verbose compilation
		sprintf(cu_build_opt, " -fatbin -ccbin g++ --cudart=static -gencode arch=compute_30,code=sm_30 -gencode arch=compute_30,code=compute_30"
				" -D CUDA_VERSION=1 -D __OPENCL_VERSION__=120 -D CL_N_VS=%d -D CL_N_CS=%d -D CL_N_VS_TO_LABEL=%d -D CL_SPLIT_VALUES_EXT=%d -D CL_N_VS_CS=%d"
				" -D CL_N_CS_VS=%d -D CL_BITS=%d -D CL_WORD=%d -D CL_MEM=%d -D CL_LABEL_M=%d -D CL_ASSIGN_M=%d -D CL_WORK=%d -D CL_OPT_M=%d -D CL_VAR_ID_TO_OPT=%d"
				" -D CL_D_TYPE=%d -D CL_D_MAX=%d -D CL_D_MIN=%d -D CL_STATS=%u -D CL_INTS_CONST=%d -D CL_B_DS_CONST=%d -D CL_VS_CONST=%d -D CL_CS_CONST=%d"
				" -D CL_N_SHARED_SS=%d -D CL_N_DEVS=%d -D PRINT_SOLS=%d -D CL_PRE_LABELING=%d -D CL_CS_IGNORE=%d -D CL_USE_BOOLEAN_VS=%d -D CL_FILTERING=%d"
				" -D CL_USE_N_BUFFERS=%d -D CL_N_TERMS=%d -D CL_CHECK_ERRORS=%d -D CL_VERIFY_SOLS=%d -D CL_FZN_SEQ=%d -D CL_FZN_SEQ_N_LABELS=%d"
				" -I src/ -I src/utils/ -I src/kernels/",
				N_VS, N_CS, dev_args->n_vs_to_label, dev_args->split_values_ext, dev_args->n_vs_cs, dev_args->n_cs_vs, CL_BITS_, CL_WORD_, dev_info->use_local_mem,
				LABEL_MODE, ASSIGN_MODE, WORK, OPT_MODE, VAR_ID_TO_OPT, DOMAIN_TYPE, D_MAX, D_MIN, PRINT_STATS, dev_args->ints_const, dev_args->b_ds_const,
				dev_args->cl_vs_const, dev_args->cl_cs_const, dev_args->n_shared_stores, N_DEVS, PRINT_SOLUTIONS, REV, CS_IGNORE, BOOLEAN_VS, filtering,
				dev_info->n_buffers, dev_info->n_terms, CL_CHECK_ERRORS, CL_VERIFY_SOLS, FZN_SEQ, FZN_SEQ_N_LABELS);

		// to tell OpenCL compiler which constraints should be compiled and which ones uses reification
		for (k = 0; k < N_C_TYPES; k++) {

#if COMPILE_ALL_CS == 0
			if (USE_CS[k] == 1) {
#endif
				sprintf(cu_cs_usage, " -D CS_%s=1", cs_get_type((c_kind)k));
				strcat(cu_build_opt, cu_cs_usage);

#if COMPILE_ALL_CS == 0
				if (USE_CS_REIFI[k] == 1) {
#endif
					sprintf(cu_cs_usage, " -D CS_R_%s=1", cs_get_type((c_kind)k));
					strcat(cu_build_opt, cu_cs_usage);

#if COMPILE_ALL_CS == 0
				}
			}
#endif
		}


#if DEBUG_IN_CUDA
		printf("\nDebug activated inside kernel (only available for Nvidia CPUs).\n\n");
		char cu_add[100];
		sprintf(cu_add, " -g -G -O0 ");
		strcat(cu_build_opt, cu_add);

		snprintf(command_cp, (unsigned long)max_com_size, "nvcc -x cu %s -o %s -v %s", kernel_filename, fatbin_filename, cu_build_opt);

		printf("\n\nNVCC command:\n%s\n\n", command_cp);
#else
		snprintf(command_cp, (unsigned long)max_com_size, "nvcc %s -x cu -o %s %s", kernel_filename, fatbin_filename, cu_build_opt);
#endif

		// for elapsed time calculation
		if (VERBOSE) {
			gettimeofday(&start_cu, NULL);
		}

		int nvcc_exit_status = system(command_cp);
		if (nvcc_exit_status) {
			fprintf(stderr, "\nError: nvcc exits with status code: %d\n", nvcc_exit_status);

#if defined(WIN32) || defined(_WIN32) || defined(__WIN32) && !defined(__CYGWIN__)
			printf("\nPress any key to exit\n");
			int a = getchar();
#endif

			exit(0);
		}

		if (VERBOSE) {
			gettimeofday(&end_cu, NULL);
			format_elapsed_time_s_ms(elapsed_time_cu, start_cu.tv_sec, start_cu.tv_usec, end_cu.tv_sec, end_cu.tv_usec);
			format_time_s_ms(start_time_cu, start_cu.tv_sec, start_cu.tv_usec);
			format_time_s_ms(end_time_cu, end_cu.tv_sec, end_cu.tv_usec);
			printf("%s...%s = %s (s.ms) -> CUDA kernel compiled for %s\n", start_time_cu, end_time_cu, elapsed_time_cu, name);
		}

		// for elapsed time calculation
		if (VERBOSE) {
			gettimeofday(&start_cu, NULL);
		}

		err = cuCtxCreate(&dev_args->context_cu, 0, dev_args->device_cu);
		cuda_check_error(err, "cuCtxCreate", name);

		if (VERBOSE) {
			gettimeofday(&end_cu, NULL);
			format_elapsed_time_s_ms(elapsed_time_cu, start_cu.tv_sec, start_cu.tv_usec, end_cu.tv_sec, end_cu.tv_usec);
			format_time_s_ms(start_time_cu, start_cu.tv_sec, start_cu.tv_usec);
			format_time_s_ms(end_time_cu, end_cu.tv_sec, end_cu.tv_usec);
			printf("%s...%s = %s (s.ms) -> CUDA context created\n", start_time_cu, end_time_cu, elapsed_time_cu);
		}

		// for elapsed time calculation
		if (VERBOSE) {
			gettimeofday(&start_cu, NULL);
		}

		err = cuModuleLoad(&dev_args->module_cu, fatbin_filename);
		cuda_check_error(err, "cuModuleLoad", name);

		if (VERBOSE) {
			gettimeofday(&end_cu, NULL);
			format_elapsed_time_s_ms(elapsed_time_cu, start_cu.tv_sec, start_cu.tv_usec, end_cu.tv_sec, end_cu.tv_usec);
			format_time_s_ms(start_time_cu, start_cu.tv_sec, start_cu.tv_usec);
			format_time_s_ms(end_time_cu, end_cu.tv_sec, end_cu.tv_usec);
			printf("%s...%s = %s (s.ms) -> CUDA kernel loaded\n", start_time_cu, end_time_cu, elapsed_time_cu);
		}

		// for elapsed time calculation
		if (VERBOSE) {
			gettimeofday(&start_cu, NULL);
		}

		err = cuModuleGetFunction(&dev_args->function_cu, dev_args->module_cu, kernel_name);
		cuda_check_error(err, "cuModuleGetFunction", name);

		if (VERBOSE) {
			gettimeofday(&end_cu, NULL);
			format_elapsed_time_s_ms(elapsed_time_cu, start_cu.tv_sec, start_cu.tv_usec, end_cu.tv_sec, end_cu.tv_usec);
			format_time_s_ms(start_time_cu, start_cu.tv_sec, start_cu.tv_usec);
			format_time_s_ms(end_time_cu, end_cu.tv_sec, end_cu.tv_usec);
			printf("%s...%s = %s (s.ms) -> CUDA function obtained\n", start_time_cu, end_time_cu, elapsed_time_cu);
		}

		// 0...cs_vs_idx	- each constraint list of constrained variables ids placed per constraint order
		// cs_vs_idx...cs_vs_idx+vs_cs_idx	- each variable list of constraints ids placed per variable order
		// cs_vs_idx+vs_cs_idx...cs_vs_idx+vs_cs_idx+n_const_cs	- each constraint list of constants placed per constraint order
		err = cuMemAlloc(&dev_args->ints_mem_cu, dev_args->ints_size);
		cuda_check_error(err, "cuMemAlloc ints_mem_cu", name);
		dev_args->ints = calloc(dev_args->ints_size / sizeof(cl_int), sizeof(cl_int));

		// buffer for cl_constr constant data
		err = cuMemAlloc(&dev_args->cl_cs_mem_cu, dev_args->cl_cs_size);
		cuda_check_error(err, "cuMemAlloc cl_cs_mem_cu", name);
		dev_args->cl_cs = malloc(dev_args->cl_cs_size);

		unsigned int n_vs_cs_cnt_cu = 0;

		// Fills constraints and variables per constraint buffer
		for (k = 0; k < N_CS; k++) {
			dev_args->cl_cs[k].kind = CS[k].kind;
			dev_args->cl_cs[k].n_c_vs = CS[k].n_c_vs;
			dev_args->cl_cs[k].n_c_consts = CS[k].n_c_consts;
			dev_args->cl_cs[k].v_idx = n_vs_cs_cnt_cu;
			dev_args->cl_cs[k].constant_val = CS[k].constant_val;
			dev_args->cl_cs[k].reif_var_id = CS[k].reif_v_id;
			dev_args->cl_cs[k].reified = CS[k].reified;
			dev_args->cl_cs[k].boolean = CS[k].boolean;
			dev_args->cl_cs[k].c_id = CS[k].c_id;

			for (l = 0; l < CS[k].n_c_vs; l++) {
				dev_args->ints[n_vs_cs_cnt_cu++] = CS[k].c_vs[l]->v_id;
			}
		}

		// Fills variables and constraints per variable buffer
		int n_cs_vs_cnt_cu = 0;

		if (DOMAIN_TYPE == BITMAP_) {

			err = cuMemAlloc(&dev_args->cl_vs_mem_cu, dev_args->cl_vs_size);
			cuda_check_error(err, "cuMemAlloc cl_vs_mem_cu", name);
			dev_args->cl_vs_bitmaps = malloc(dev_args->cl_vs_size);

			err = cuMemAlloc(&dev_args->b_ds_mem_cu, dev_args->b_ds_size);
			cuda_check_error(err, "cuMemAlloc b_ds_mem_cu", name);
			dev_args->b_ds = malloc(dev_args->b_ds_size);

			vs_copy_host_to_dev(dev_args->b_ds, VS, N_VS);

			for (k = 0; k < N_VS; k++) {
				dev_args->cl_vs_bitmaps[k].n_cs = VS[k].n_cs;
				dev_args->cl_vs_bitmaps[k].to_label = VS[k].to_label;
				dev_args->cl_vs_bitmaps[k].boolean = VS[k].boolean;
				dev_args->cl_vs_bitmaps[k].expanded = VS[k].expanded;
				dev_args->cl_vs_bitmaps[k].n_vals = VS[k].n_vals;

#if FZN_SEQ
				dev_args->cl_vs_bitmaps[k].label_h = VS[k].label_h;
				dev_args->cl_vs_bitmaps[k].assign_h = VS[k].assign_h;
#endif

				if (VS[k].n_cs > 0) {
					dev_args->cl_vs_bitmaps[k].c_idx = (unsigned int)n_cs_vs_cnt_cu;
					for (l = 0; l < VS[k].n_cs; l++) {
						dev_args->ints[dev_args->n_vs_cs + (unsigned int)n_cs_vs_cnt_cu] = (int)VS[k].cs[l]->c_id;
						n_cs_vs_cnt_cu++;
					}
				} else {
					dev_args->cl_vs_bitmaps[k].c_idx = 0;
				}
			}

		} else if (DOMAIN_TYPE == INTERVAL) {

			// buffer for cl_var constant data
			err = cuMemAlloc(&dev_args->cl_vs_mem_cu, dev_args->cl_vs_size);
			cuda_check_error(err, "cuMemAlloc cl_vs_mem_cu", name);
			dev_args->cl_vs_intervals = malloc(dev_args->cl_vs_size);

			for (k = 0; k < N_VS; k++) {
				dev_args->cl_vs_intervals[k].domain.s[0] = VS[k].domain_i.s[0];
				dev_args->cl_vs_intervals[k].domain.s[1] = VS[k].domain_i.s[1];

				dev_args->cl_vs_intervals[k].n_cs = VS[k].n_cs;
				dev_args->cl_vs_intervals[k].c_idx = (unsigned int)n_cs_vs_cnt_cu;
				dev_args->cl_vs_intervals[k].to_label = VS[k].to_label;
				dev_args->cl_vs_intervals[k].boolean = VS[k].boolean;
				dev_args->cl_vs_intervals[k].expanded = VS[k].expanded;
				dev_args->cl_vs_intervals[k].n_vals = VS[k].n_vals;

#if FZN_SEQ
				dev_args->cl_vs_intervals[k].label_h = VS[k].label_h;
				dev_args->cl_vs_intervals[k].assign_h = VS[k].assign_h;
#endif

				if (VS[k].n_cs > 0) {
					dev_args->cl_vs_intervals[k].c_idx = (unsigned int)n_cs_vs_cnt_cu;
					for (l = 0; l < VS[k].n_cs; l++) {
						dev_args->ints[dev_args->n_vs_cs + (unsigned int)n_cs_vs_cnt_cu] = (int)VS[k].cs[l]->c_id;
						n_cs_vs_cnt_cu++;
					}
				} else {
					dev_args->cl_vs_intervals[k].c_idx = 0;
				}
			}
		}

		// Fills constraint constant values, if existent
		unsigned int cs_consts_cnt_cu = 0;
		for (k = 0; k < N_CS; k++) {
			if (CS[k].n_c_consts > 0) {

				dev_args->cl_cs[k].const_idx = cs_consts_cnt_cu;

				for (l = 0; l < CS[k].n_c_consts; l++) {
					dev_args->ints[dev_args->n_vs_cs + dev_args->n_cs_vs + cs_consts_cnt_cu] = CS[k].c_consts[l];
					cs_consts_cnt_cu++;
				}
			} else {
				dev_args->cl_cs[k].const_idx = 0;
			}
		}

		// if not using local memory (using global memory only)
		if (!dev_info->use_local_mem) {

			// buffer for cl_vs_prop data
			err = cuMemAlloc(&dev_args->cl_vs_prop_mem_cu, dev_args->cl_vs_prop_size);
			cuda_check_error(err, "cuMemAlloc cl_vs_prop_mem_cu", name);

			// buffer for cl_vs_prop data
			err = cuMemAlloc(&dev_args->vs_id_to_prop_mem_cu, dev_args->vs_id_to_prop_size);
			cuda_check_error(err, "cuMemAlloc vs_id_to_prop_mem_cu", name);
		}

		// buffer for backtracking history
		// 0...(n_vs_to_label+1)*N_VS*wi_total - backtracking history
		err = cuMemAlloc(&dev_args->backtrack_mem1_cu, dev_args->backtrack_size);
		cuda_check_error(err, "cuMemAlloc backtrack_mem1_cu", name);

		// more buffers for backtracking
#if USE_MORE_BUFFERS
		if (dev_info->n_buffers > 1) {
			err = cuMemAlloc(&dev_args->backtrack_mem2_cu, dev_args->backtrack_size);
			cuda_check_error(err, "cuMemAlloc backtrack_mem2_cu", name);
		}
		if (dev_info->n_buffers > 2) {
			err = cuMemAlloc(&dev_args->backtrack_mem3_cu, dev_args->backtrack_size);
			cuda_check_error(err, "cuMemAlloc backtrack_mem3_cu", name);
		}
		if (dev_info->n_buffers > 3) {
			err = cuMemAlloc(&dev_args->backtrack_mem4_cu, dev_args->backtrack_size);
			cuda_check_error(err, "cuMemAlloc backtrack_mem4_cu", name);
		}
#endif

		// (dev_args->n_vs_to_label + 2) * dev_args->split_values_ext) * 2 - to use in kernel (hist_labeleds_id and hist_labeleds_n_vals)
		// n_terms * dev_args->wi_total - to use in propagators
		// D_MAX+1 - for ss generation
		err = cuMemAlloc(&dev_args->generic_mem_cu, dev_args->generic_size);
		cuda_check_error(err, "cuMemAlloc generic_mem_cu", name);
		dev_args->generic = calloc(dev_args->generic_size / sizeof(cl_int), sizeof(cl_int));
		// (dev_args->n_vs_to_label + 2) * dev_args->split_values_ext) * 2 - to use in kernel (hist_labeleds_id and hist_labeleds_n_vals)
		// n_terms * dev_args->wi_total - to use in propagators
		// D_MAX+1*wi_total - for ss generation
#if FZN_SEQ
		// if seq_search only:
		// D_MAX+1*wi_total + 1 - number of labeling heuristics that may be used
		// D_MAX+1*wi_total + 2... - list of labeling heuristics that may be used
		unsigned int i, j;
		i = (unsigned int)dev_args->generic_size / sizeof(cl_int);
		for (j = (unsigned int)FZN_SEQ_N_LABELS; j > 0; j--) {
			dev_args->generic[--i] = FZN_SEQ_LABELS[j - 1];
		}
#endif

		if (CS_IGNORE) {
			// N_CS * dev_args->wi_total - to flag CS_IGNORE
			err = cuMemAlloc(&dev_args->cs_ignore_mem_cu, dev_args->cs_ignore_size);
			cuda_check_error(err, "cuMemAlloc cs_ignore_mem_cu", name);
		}

		// if all solutions must be found
		if (WORK == CNT) {
			// buffer for atomics data (Most devices only have atomics for 32 bits variables)
			// 0 - first sub-search to explore
			// 1 - last sub-search to explore
			// 2 - n_ss
			// 3 - depth
			// 4 - WIs still working for work-sharing
			// 5 - 5+N_VS - n_repeat per variable
			// 5+N_VS...5+N_VS+N_WG*N_WI_WG - number of solutions found per work-item
			err = cuMemAlloc(&dev_args->atoms_mem_cu, dev_args->atoms_size);
			cuda_check_error(err, "cuMemAlloc atoms_mem_cu", name);
			dev_args->atoms = calloc(dev_args->atoms_size / sizeof(cl_uint), sizeof(cl_uint));

			// if only one solution must be found
		} else if (WORK == ONE) {
			// buffer for atomics data (Most devices only have atomics for 32 bits variables)
			// 0 - first sub-search to explore
			// 1 - last sub-search to explore
			// 2 - n_ss
			// 3 - depth
			// 4 - WIs still working for work-sharing
			// 5 - 5+N_VS - n_repeat per variable
			// 5+N_VS - solution found flag
			err = cuMemAlloc(&dev_args->atoms_mem_cu, dev_args->atoms_size);
			cuda_check_error(err, "cuMemAlloc atoms_mem_cu", name);
			dev_args->atoms = calloc(dev_args->atoms_size / sizeof(cl_uint), sizeof(cl_uint));

			// buffer for domains writable data
			// 0...N_VS - solution domains
			err = cuMemAlloc(&dev_args->domains_mem_cu, dev_args->domains_size);
			cuda_check_error(err, "cuMemAlloc domains_mem_cu", name);

			if (DOMAIN_TYPE == BITMAP_) {
				dev_args->bitmaps = malloc(dev_args->domains_size);

				// set buffer initial values to zero
				memset(dev_args->bitmaps, 0, dev_args->domains_size);

				err = cuMemcpyHtoD(dev_args->domains_mem_cu, dev_args->bitmaps, dev_args->domains_size);
				cuda_check_error(err, "cuMemcpyHtoD domains_mem_cu", name);

			} else if (DOMAIN_TYPE == INTERVAL) {

				dev_args->intervals = malloc(dev_args->domains_size);

				// set buffer initial values to zero
				memset(dev_args->intervals, 0, N_VS * 4);

				err = cuMemcpyHtoD(dev_args->domains_mem_cu, dev_args->intervals, dev_args->domains_size);
				cuda_check_error(err, "cuMemcpyHtoD domains_mem_cu", name);
			}

			// if optimization
		} else if (WORK == OPT) {
			// buffer for atomics data (Most devices only have atomics for 32 bits variables)
			// 0 - first sub-search to explore
			// 1 - last sub-search to explore
			// 2 - n_ss
			// 3 - depth
			// 4 - WIs still working for work-sharing
			// 5 - 5+N_VS - n_repeat per variable
			// 5+N_VS - solution found flag
			// 6+N_VS - Value to optimize
			// 7+N_VS - WIs still working for saving the best solution
			err = cuMemAlloc(&dev_args->atoms_mem_cu, dev_args->atoms_size);
			cuda_check_error(err, "cuMemAlloc atoms_mem_cu", name);
			dev_args->atoms = calloc(dev_args->atoms_size / sizeof(cl_uint), sizeof(cl_uint));

			// buffer for domains writable data
			// 0...N_VS*(D_MAX+1) - (D_MAX+1) solution stores because concurrency control
			err = cuMemAlloc(&dev_args->domains_mem_cu, dev_args->domains_size);
			cuda_check_error(err, "cuMemAlloc domains_mem_cu", name);

			if (DOMAIN_TYPE == BITMAP_) {
				dev_args->bitmaps = malloc(dev_args->domains_size);

				// set buffer initial values to zero
				memset(dev_args->bitmaps, 0, dev_args->domains_size);

				err = cuMemcpyHtoD(dev_args->domains_mem_cu, dev_args->bitmaps, dev_args->domains_size);
				cuda_check_error(err, "cuMemcpyHtoD domains_mem_cu", name);

			} else if (DOMAIN_TYPE == INTERVAL) {
				dev_args->intervals = malloc(dev_args->domains_size);

				// set buffer initial values to zero
				memset(dev_args->intervals, 0, N_VS * (D_MAX + 1) * 4);

				err = cuMemcpyHtoD(dev_args->domains_mem_cu, dev_args->intervals, dev_args->domains_size);
				cuda_check_error(err, "cuMemcpyHtoD domains_mem_cu", name);
			}
		}

		if (N_DEVS > 1) {

			// to count number of propagations done per work-item, for rank calculation
			err = cuMemAlloc(&dev_args->props_mem_cu, dev_args->props_size);
			cuda_check_error(err, "cuMemAlloc props_mem_cu", name);
			dev_args->props = calloc(dev_args->props_size / sizeof(cl_ulong), sizeof(cl_ulong));
		}

#if SHARED_SS > 0
		// for work-sharing after the ss in the block have finished
		err = cuMemAlloc(&dev_args->shared_stores_mem_cu, dev_args->shared_stores_size);
		cuda_check_error(err, "cuMemAlloc shared_stores_mem_cu", name);

		// flags for signaling the state of each work-sharing store
		// 0 - next shared SS to be picked
		// 1 - next shared SS to be filled
		// 2...number of SS already filled
		// 3..3+CL_N_SHARED_SS - V_ID that was labeled to generate this SS
		err = cuMemAlloc(&dev_args->shared_stores_flag_mem_cu, dev_args->shared_stores_flag_size);
		cuda_check_error(err, "cuMemAlloc shared_stores_flag_mem_cu", name);
		dev_args->shared_stores_flag = calloc(dev_args->shared_stores_flag_size / sizeof(cl_int), sizeof(cl_int));
#endif

		if (PRINT_STATS) {
			// 0 - nodes_fail
			// 1 - nodes_expl
			// 2 - backtracks
			// 3 - labels
			// 4 - pruning
			// 5 - props_ok
			// 6 - max_depth
			// ... repeat per work-item
			err = cuMemAlloc(&dev_args->stats_mem_cu, dev_args->stats_size);
			cuda_check_error(err, "cuMemAlloc stats_mem_cu", name);
			dev_args->stats = malloc(dev_args->stats_size);
			memset(dev_args->stats, 0, dev_args->stats_size);

			err = cuMemcpyHtoD(dev_args->stats_mem_cu, dev_args->stats, dev_args->stats_size);
			cuda_check_error(err, "cuMemcpyHtoD stats_mem_cu", name);
		}

		if (filtering) {
			// 0...N_VS - size of domains_mem buffer for the filtering result
			err = cuMemAlloc(&dev_args->filt_domains_mem_cu, dev_args->filt_domains_size);
			cuda_check_error(err, "cuMemAlloc filt_domains_mem_cu", name);

			if (DOMAIN_TYPE == BITMAP_) {
				dev_args->filt_bitmaps = malloc(dev_args->filt_domains_size);

			} else if (DOMAIN_TYPE == INTERVAL) {
				dev_args->filt_intervals = malloc(dev_args->filt_domains_size);
			}

			if (CS_IGNORE) {
				// 0...N_CS - size of filt_cs_mem buffer for the filtering
				err = cuMemAlloc(&dev_args->filt_cs_mem_cu, dev_args->filt_cs_size);
				cuda_check_error(err, "cuMemAlloc filt_cs_mem_cu", name);

				dev_args->filt_cs = malloc(dev_args->filt_cs_size);
			}
		}

		// write permanent data to device buffers
		err = cuMemcpyHtoD(dev_args->ints_mem_cu, dev_args->ints, dev_args->ints_size);
		cuda_check_error(err, "cuMemcpyHtoD ints_mem_cu", name);

		err = cuMemcpyHtoD(dev_args->generic_mem_cu, dev_args->generic, dev_args->generic_size);
		cuda_check_error(err, "cuMemcpyHtoD generic_mem_cu", name);

		if (DOMAIN_TYPE == BITMAP_) {
			err = cuMemcpyHtoD(dev_args->cl_vs_mem_cu, dev_args->cl_vs_bitmaps, dev_args->cl_vs_size);
			cuda_check_error(err, "cuMemcpyHtoD cl_vs_mem_cu", name);

			err = cuMemcpyHtoD(dev_args->b_ds_mem_cu, dev_args->b_ds, dev_args->b_ds_size);
			cuda_check_error(err, "cuMemcpyHtoD b_ds_mem_cu", name);

		} else if (DOMAIN_TYPE == INTERVAL) {
			err = cuMemcpyHtoD(dev_args->cl_vs_mem_cu, dev_args->cl_vs_intervals, dev_args->cl_vs_size);
			cuda_check_error(err, "cuMemcpyHtoD cl_vs_mem_cu", name);
		}
		err = cuMemcpyHtoD(dev_args->cl_cs_mem_cu, dev_args->cl_cs, dev_args->cl_cs_size);
		cuda_check_error(err, "cuMemcpyHtoD cl_cs_mem_cu", name);

		// Set OpenCL kernel arguments
		cl_uint arg_number_cu = 8;

#if USE_MORE_BUFFERS
		if (dev_info->n_buffers > 1) {
			arg_number_cu++;
		}
		if (dev_info->n_buffers > 2) {
			arg_number_cu++;
		}
		if (dev_info->n_buffers > 3) {
			arg_number_cu++;
		}
#endif

		if (CS_IGNORE) {
			arg_number_cu++;
		}

		if (WORK == ONE || WORK == OPT) {
			arg_number_cu++;
		}

		if (DOMAIN_TYPE == BITMAP_) {
			arg_number_cu++;
		}

		if (PRINT_STATS) {
			arg_number_cu++;
		}

		if (N_DEVS > 1) {
			arg_number_cu++;
		}
#if SHARED_SS > 0
		arg_number_cu += 2;
#endif

		if (filtering) {
			arg_number_cu++;

			if (CS_IGNORE) {
				arg_number_cu++;
			}
		}

		dev_args->kernel_args_cu = malloc(arg_number_cu * sizeof(CUdeviceptr));

		arg_number_cu = 0;

		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->atoms_mem_cu;
		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->ints_mem_cu;
		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->backtrack_mem1_cu;

		// more buffers for backtracking
#if USE_MORE_BUFFERS
		if (dev_info->n_buffers > 1) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->backtrack_mem2_cu;
		}
		if (dev_info->n_buffers > 2) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->backtrack_mem3_cu;
		}
		if (dev_info->n_buffers > 3) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->backtrack_mem4_cu;
		}
#endif

		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->generic_mem_cu;

		if (CS_IGNORE) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->cs_ignore_mem_cu;
		}

		if (WORK == ONE || WORK == OPT) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->domains_mem_cu;
		}

		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->cl_vs_mem_cu;
		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->cl_cs_mem_cu;

		// if using local memory
		if (dev_info->use_local_mem) {
			dev_args->shared_memory_size_cu = dev_args->cl_vs_prop_size;
			dev_args->shared_memory_size_cu += dev_args->vs_id_to_prop_size;
			// if not using local memory
		} else {
			dev_args->shared_memory_size_cu = 0;
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->cl_vs_prop_mem_cu;
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->vs_id_to_prop_mem_cu;
		}

		if (DOMAIN_TYPE == BITMAP_) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->b_ds_mem_cu;
		}

		if (PRINT_STATS) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->stats_mem_cu;
		}

		if (N_DEVS > 1) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->props_mem_cu;
		}
#if SHARED_SS > 0
		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->shared_stores_mem_cu;
		dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->shared_stores_flag_mem_cu;
#endif

		if (filtering) {
			dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->filt_domains_mem_cu;
			if (CS_IGNORE) {
				dev_args->kernel_args_cu[arg_number_cu++] = &dev_args->filt_cs_mem_cu;
			}
		}

	} else {

#endif

	// for elapse time calculation
	struct timeval start, end;
	char start_time[40];
	char end_time[40];
	char elapsed_time[40];

	dev_args->cq = NULL;		// Queue for buffering devices operations
	dev_args->kernel = NULL;	// kernel
	cl_int ret;					// Returned value from each OpenCL host function
	char build_opt[3000];	// build options for OpenCL compiler
	char cs_usage[200];		// to tell OpenCL compiler which constraints should be compiled and which ones uses reification
	unsigned int i, j;

	// for elapsed time calculation
	if (VERBOSE) {
		gettimeofday(&start, NULL);
	}

	cl_context_properties contextProperties[] = { CL_CONTEXT_PLATFORM, (cl_context_properties) dev_info->platform_id, 0 };
	dev_info->context = clCreateContext(contextProperties, 1, &dev_info->device_id, NULL, NULL, &ret);
	cl_check_error(ret, "clCreateContext", dev_info->dev_name);

	if (VERBOSE) {
		gettimeofday(&end, NULL);
		format_elapsed_time_s_ms(elapsed_time, start.tv_sec, start.tv_usec, end.tv_sec, end.tv_usec);
		format_time_s_ms(start_time, start.tv_sec, start.tv_usec);
		format_time_s_ms(end_time, end.tv_sec, end.tv_usec);
		printf("%s...%s = %s (s.ms) -> OpenCL context created for %s (%d)\n", start_time, end_time, elapsed_time, dev_info->dev_name, dev_info->dev_type_n);
	}

	// Load kernel source file
	FILE *fp;
	const char kernel_file[] = "src/kernels/cl_explore.cl";
	size_t src_size;
	char *src_str;
	fp = fopen(kernel_file, "r");
	if (!fp) {
		fprintf(stderr, "\nError: Failed to load OpenCL kernel at %s.\n", kernel_file);

#if defined(WIN32) || defined(_WIN32) || defined(__WIN32) && !defined(__CYGWIN__)
			printf("\nPress any key to exit\n");
			int a = getchar();
#endif

		exit(0);
	}
	src_str = (char*) malloc(0x1000000);
	src_size = fread(src_str, 1, 0x1000000, fp);
	fclose(fp);

	// kernel build options
	// some build options that may be added to AMD compiler (equal to AMD_OCL_BUILD_OPTIONS_APPEND=X on PHACT command line execution command):
	// -Werror -cl-opt-disable -cl-strict-aliasing -save-temps
	// oclgrind buil options:
	// --inst-counts --check-api --data-races --uninitialized --uniform-writes --constant-mem-size 65536 --global-mem-size 4231331840 --local-mem-size 49152
	sprintf(build_opt,
			" -Werror -D CUDA_VERSION=0 -cl-std=CL1.2 -D CL_N_VS=%d -D CL_N_CS=%d -D CL_N_VS_TO_LABEL=%d -D CL_SPLIT_VALUES_EXT=%d -D CL_N_VS_CS=%d -D CL_N_CS_VS=%d"
					" -D CL_BITS=%d -D CL_WORD=%d -D CL_MEM=%d -D CL_LABEL_M=%d -D CL_ASSIGN_M=%d -D CL_WORK=%d -D CL_OPT_M=%d -D CL_VAR_ID_TO_OPT=%d -D CL_D_TYPE=%d"
					" -D CL_D_MAX=%d -D CL_D_MIN=%d -D CL_STATS=%u -D CL_INTS_CONST=%d -D CL_B_DS_CONST=%d -D CL_VS_CONST=%d -D CL_CS_CONST=%d -D CL_N_SHARED_SS=%d"
					" -D CL_N_DEVS=%d -D PRINT_SOLS=%d -D CL_PRE_LABELING=%d -D CL_CS_IGNORE=%d -D CL_USE_BOOLEAN_VS=%d -D CL_FILTERING=%d -D CL_USE_N_BUFFERS=%d"
					" -D CL_N_TERMS=%d -D CL_CHECK_ERRORS=%d -D CL_VERIFY_SOLS=%d -D CL_FZN_SEQ=%d -D CL_FZN_SEQ_N_LABELS=%d -I src/ -I src/utils/ -I src/kernels/",
			N_VS, N_CS, dev_args->n_vs_to_label, dev_args->split_values_ext, dev_args->n_vs_cs, dev_args->n_cs_vs, CL_BITS_, CL_WORD_, dev_info->use_local_mem,
			LABEL_MODE, ASSIGN_MODE, WORK, OPT_MODE, VAR_ID_TO_OPT, DOMAIN_TYPE, D_MAX, D_MIN, PRINT_STATS, dev_args->ints_const, dev_args->b_ds_const,
			dev_args->cl_vs_const, dev_args->cl_cs_const, dev_args->n_shared_stores, N_DEVS, PRINT_SOLUTIONS, REV, CS_IGNORE, BOOLEAN_VS, filtering,
			dev_info->n_buffers, dev_info->n_terms, CL_CHECK_ERRORS, CL_VERIFY_SOLS, FZN_SEQ, FZN_SEQ_N_LABELS);

#if DEBUG
	if (dev_info->type == CL_DEVICE_TYPE_CPU) {
		printf("\nDebug activated inside kernel (only available for Intel CPUs).\n\n");
		char add[100];
		sprintf(add, " -g -s src/kernels/cl_explore.cl ");
		strcat(build_opt, add);
	}
#endif

#if CL_COMP_OPT == 0
	printf("\nThe OpenCL compiler optimizations are disabled.\n\n");
	char add[100];
	sprintf(add, " -cl-opt-disable ");
	strcat(build_opt, add);
#endif

#if COMPILE_ALL_CS == 1
	printf("All propagators are being compiled in kernel.\n\n");
#endif

	// to tell OpenCL compiler which constraints should be compiled and which ones uses reification
	for (i = 0; i < N_C_TYPES; i++) {

#if COMPILE_ALL_CS == 0
		if (USE_CS[i] == 1) {
#endif
			sprintf(cs_usage, " -D CS_%s=1", cs_get_type((c_kind) i));
			strcat(build_opt, cs_usage);

#if COMPILE_ALL_CS == 0
			if (USE_CS_REIFI[i] == 1) {
#endif
				sprintf(cs_usage, " -D CS_R_%s=1", cs_get_type((c_kind) i));
				strcat(build_opt, cs_usage);

#if COMPILE_ALL_CS == 0
			}
		}
#endif
	}

	// Creates a program object
	dev_info->prog = clCreateProgramWithSource(dev_info->context, 1, (const char**) &src_str, (const size_t*) &src_size, &ret);
	cl_check_error(ret, "clCreateProgramWithSource", dev_info->dev_name);
	free(src_str);

	// for elapsed time calculation
	if (VERBOSE) {
		gettimeofday(&start, NULL);
	}

	// Compiles and links the kernel and check for errors
	cl_check_build_error(clBuildProgram(dev_info->prog, 1, &dev_info->device_id, build_opt, NULL, NULL), &dev_info->prog, &dev_info->device_id,
			dev_info->dev_name);

	if (VERBOSE) {
		gettimeofday(&end, NULL);
		format_elapsed_time_s_ms(elapsed_time, start.tv_sec, start.tv_usec, end.tv_sec, end.tv_usec);
		format_time_s_ms(start_time, start.tv_sec, start.tv_usec);
		format_time_s_ms(end_time, end.tv_sec, end.tv_usec);
		printf("%s...%s = %s (s.ms) -> Kernel compiled for %s (%d)\n", start_time, end_time, elapsed_time, dev_info->dev_name, dev_info->dev_type_n);
	}

	// Create Command Queue
	dev_args->cq = clCreateCommandQueue(dev_info->context, dev_info->device_id, 0, &ret);
	cl_check_error(ret, "clCreateCommandQueue", dev_info->dev_name);

	// Create OpenCL kernels allowing it to be called from this source code
	dev_args->kernel = clCreateKernel(dev_info->prog, "explore", &ret);
	cl_check_error(ret, "clCreateKernel", dev_info->dev_name);

	// 0...cs_vs_idx	- each constraint list of constrained variables ids placed per constraint order
	// cs_vs_idx...cs_vs_idx+vs_cs_idx	- each variable list of constraints ids placed per variable order
	// cs_vs_idx+vs_cs_idx...cs_vs_idx+vs_cs_idx+n_const_cs	- each constraint list of constants placed per constraint order
	dev_args->ints_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_ONLY, dev_args->ints_size, NULL, &ret);
	cl_check_error(ret, "clCreateBuffer int_p_mem", dev_info->dev_name);
	dev_args->ints = calloc(dev_args->ints_size / sizeof(cl_int), sizeof(cl_int));

	// buffer for cl_constr constant data
	dev_args->cl_cs_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_ONLY, dev_args->cl_cs_size, NULL, &ret);
	cl_check_error(ret, "clCreateBuffer cl_cs_p_mem", dev_info->dev_name);
	dev_args->cl_cs = malloc(dev_args->cl_cs_size);

	// Fills constraints and variables per constraint buffer
	unsigned int n_vs_cs_cnt = 0;
	for (i = 0; i < N_CS; i++) {
		dev_args->cl_cs[i].kind = CS[i].kind;
		dev_args->cl_cs[i].n_c_vs = CS[i].n_c_vs;
		dev_args->cl_cs[i].n_c_consts = CS[i].n_c_consts;
		dev_args->cl_cs[i].v_idx = n_vs_cs_cnt;
		dev_args->cl_cs[i].constant_val = CS[i].constant_val;
		dev_args->cl_cs[i].reif_var_id = CS[i].reif_v_id;
		dev_args->cl_cs[i].reified = CS[i].reified;
		dev_args->cl_cs[i].boolean = CS[i].boolean;
		dev_args->cl_cs[i].c_id = CS[i].c_id;

		for (j = 0; j < CS[i].n_c_vs; j++) {
			dev_args->ints[n_vs_cs_cnt++] = CS[i].c_vs[j]->v_id;
		}
	}

	// Fills variables and constraints per variable buffer
	int n_cs_vs_cnt = 0;

	if (DOMAIN_TYPE == BITMAP_) {

		dev_args->cl_vs_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_ONLY, dev_args->cl_vs_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer cl_vs_mem", dev_info->dev_name);
		dev_args->cl_vs_bitmaps = malloc(dev_args->cl_vs_size);
		dev_args->b_ds_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_ONLY, dev_args->b_ds_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer bitmaps_aux_mem", dev_info->dev_name);
		dev_args->b_ds = malloc(dev_args->b_ds_size);

		vs_copy_host_to_dev(dev_args->b_ds, VS, N_VS);

		for (i = 0; i < N_VS; i++) {
			dev_args->cl_vs_bitmaps[i].n_cs = VS[i].n_cs;
			dev_args->cl_vs_bitmaps[i].to_label = VS[i].to_label;
			dev_args->cl_vs_bitmaps[i].boolean = VS[i].boolean;
			dev_args->cl_vs_bitmaps[i].expanded = VS[i].expanded;
			dev_args->cl_vs_bitmaps[i].n_vals = VS[i].n_vals;

#if FZN_SEQ
			dev_args->cl_vs_bitmaps[i].label_h = VS[i].label_h;
			dev_args->cl_vs_bitmaps[i].assign_h = VS[i].assign_h;
#endif

			if (VS[i].n_cs > 0) {
				dev_args->cl_vs_bitmaps[i].c_idx = (unsigned int) n_cs_vs_cnt;
				for (j = 0; j < VS[i].n_cs; j++) {
					dev_args->ints[dev_args->n_vs_cs + (unsigned int) n_cs_vs_cnt] = (int) VS[i].cs[j]->c_id;
					n_cs_vs_cnt++;
				}
			} else {
				dev_args->cl_vs_bitmaps[i].c_idx = 0;
			}
		}

	} else if (DOMAIN_TYPE == INTERVAL) {
		// buffer for cl_var constant data
		dev_args->cl_vs_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_ONLY, dev_args->cl_vs_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer cl_vs_p_mem", dev_info->dev_name);

		dev_args->cl_vs_intervals = malloc(dev_args->cl_vs_size);
		for (i = 0; i < N_VS; i++) {
			dev_args->cl_vs_intervals[i].domain.s[0] = VS[i].domain_i.s[0];
			dev_args->cl_vs_intervals[i].domain.s[1] = VS[i].domain_i.s[1];

			dev_args->cl_vs_intervals[i].n_cs = VS[i].n_cs;
			dev_args->cl_vs_intervals[i].c_idx = (unsigned int) n_cs_vs_cnt;
			dev_args->cl_vs_intervals[i].to_label = VS[i].to_label;
			dev_args->cl_vs_intervals[i].boolean = VS[i].boolean;
			dev_args->cl_vs_intervals[i].expanded = VS[i].expanded;
			dev_args->cl_vs_intervals[i].n_vals = VS[i].n_vals;

#if FZN_SEQ
			dev_args->cl_vs_intervals[i].label_h = VS[i].label_h;
			dev_args->cl_vs_intervals[i].assign_h = VS[i].assign_h;
#endif

			if (VS[i].n_cs > 0) {
				dev_args->cl_vs_intervals[i].c_idx = (unsigned int) n_cs_vs_cnt;
				for (j = 0; j < VS[i].n_cs; j++) {
					dev_args->ints[dev_args->n_vs_cs + (unsigned int) n_cs_vs_cnt] = (int) VS[i].cs[j]->c_id;
					n_cs_vs_cnt++;
				}
			} else {
				dev_args->cl_vs_intervals[i].c_idx = 0;
			}
		}
	}

	// Fills constraint constant values, if existent
	unsigned int cs_consts_cnt = 0;
	for (i = 0; i < N_CS; i++) {
		if (CS[i].n_c_consts > 0) {

			dev_args->cl_cs[i].const_idx = cs_consts_cnt;

			for (j = 0; j < CS[i].n_c_consts; j++) {
				dev_args->ints[dev_args->n_vs_cs + dev_args->n_cs_vs + cs_consts_cnt] = CS[i].c_consts[j];
				cs_consts_cnt++;
			}
		} else {
			dev_args->cl_cs[i].const_idx = 0;
		}
	}

	// if not using local memory (using global memory only)
	if (!dev_info->use_local_mem) {
		// buffer for cl_vs_prop data
		dev_args->cl_vs_prop_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->cl_vs_prop_size, NULL, &ret);

		// buffer for vs_id_to_prop data
		dev_args->vs_id_to_prop_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->vs_id_to_prop_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer vs_id_to_prop_mem", dev_info->dev_name);
	}

	// buffer for backtracking history
	// 0...(n_vs_to_label+1)*N_VS*wi_total - backtracking history
	dev_args->backtrack_mem1 = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->backtrack_size, NULL, &ret);
	cl_check_error(ret, "clCreateBuffer backtrack_mem1", dev_info->dev_name);

	// more buffers for backtracking
#if USE_MORE_BUFFERS
	if (dev_info->n_buffers > 1) {
		dev_args->backtrack_mem2 = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->backtrack_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer backtrack_mem2", dev_info->dev_name);
	}
	if (dev_info->n_buffers > 2) {
		dev_args->backtrack_mem3 = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->backtrack_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer backtrack_mem3", dev_info->dev_name);
	}
	if (dev_info->n_buffers > 3) {
		dev_args->backtrack_mem4 = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->backtrack_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer backtrack_mem4", dev_info->dev_name);
	}
#endif

	// (dev_args->n_vs_to_label + 2) * dev_args->split_values_ext) * 2 - to use in kernel (hist_labeleds_id and hist_labeleds_n_vals)
	// n_terms * dev_args->wi_total - to use in propagators
	// D_MAX+1 - for ss generation
	dev_args->generic_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->generic_size, NULL, &ret);
	cl_check_error(ret, "clCreateBuffer generic_mem", dev_info->dev_name);
	dev_args->generic = calloc(dev_args->generic_size / sizeof(cl_int), sizeof(cl_int));
	// (dev_args->n_vs_to_label + 2) * dev_args->split_values_ext) * 2 - to use in kernel (hist_labeleds_id and hist_labeleds_n_vals)
	// n_terms * dev_args->wi_total - to use in propagators
	// D_MAX+1*wi_total - for ss generation
#if FZN_SEQ
	// if seq_search only:
	// D_MAX+1*wi_total + 1... - list of labeling heuristics that may be used
	i = (unsigned int) dev_args->generic_size / sizeof(cl_int);
	for (j = (unsigned int) FZN_SEQ_N_LABELS; j > 0; j--) {
		dev_args->generic[--i] = FZN_SEQ_LABELS[j - 1];
	}
#endif

	if (CS_IGNORE) {
		// N_CS * dev_args->wi_total - to flag CS_IGNORE
		dev_args->cs_ignore_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->cs_ignore_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer cs_ignore_mem", dev_info->dev_name);
	}

	// if all solutions must be found
	if (WORK == CNT) {
		// buffer for atomics data (Most devices only have atomics for 32 bits variables)
		// 0 - first sub-search to explore
		// 1 - last sub-search to explore
		// 2 - n_ss
		// 3 - depth
		// 4 - WIs still working for work-sharing
		// 5 - 5+N_VS - n_repeat per variable
		// 5+N_VS...5+N_VS+N_WG*N_WI_WG - number of solutions found per work-item
		dev_args->atoms_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->atoms_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer atoms_p_mem", dev_info->dev_name);
		dev_args->atoms = calloc(dev_args->atoms_size / sizeof(cl_uint), sizeof(cl_uint));

		// if only one solution must be found
	} else if (WORK == ONE) {
		// buffer for atomics data (Most devices only have atomics for 32 bits variables)
		// 0 - first sub-search to explore
		// 1 - last sub-search to explore
		// 2 - n_ss
		// 3 - depth
		// 4 - WIs still working for work-sharing
		// 5 - 5+N_VS - n_repeat per variable
		// 5+N_VS - solution found flag
		dev_args->atoms_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->atoms_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer atoms_p_mem", dev_info->dev_name);
		dev_args->atoms = calloc(dev_args->atoms_size / sizeof(cl_uint), sizeof(cl_uint));

		// buffer for domains writable data
		// 0...N_VS - solution domains
		dev_args->domains_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->domains_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer domains_mem", dev_info->dev_name);

		if (DOMAIN_TYPE == BITMAP_) {
			dev_args->bitmaps = malloc(dev_args->domains_size);

			// set buffer initial values to zero
			memset(dev_args->bitmaps, 0, dev_args->domains_size);
			cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->domains_mem, CL_TRUE, 0, dev_args->domains_size, dev_args->bitmaps, 0, NULL, NULL),
					"clEnqueueWriteBuffer domains_mem", dev_info->dev_name);

		} else if (DOMAIN_TYPE == INTERVAL) {
			dev_args->intervals = malloc(dev_args->domains_size);

			// set buffer initial values to zero
			memset(dev_args->intervals, 0, N_VS * 4);
			cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->domains_mem, CL_TRUE, 0, dev_args->domains_size, dev_args->intervals, 0, NULL, NULL),
					"clEnqueueWriteBuffer domains_mem", dev_info->dev_name);
		}

		// if optimization
	} else if (WORK == OPT) {
		// buffer for atomics data (Most devices only have atomics for 32 bits variables)
		// 0 - first sub-search to explore
		// 1 - last sub-search to explore
		// 2 - n_ss
		// 3 - depth
		// 4 - WIs still working for work-sharing
		// 5 - 5+N_VS - n_repeat per variable
		// 5+N_VS - solution found flag
		// 6+N_VS - Value to optimize
		// 7+N_VS - WIs still working for saving the best solution
		dev_args->atoms_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->atoms_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer atoms_p_mem", dev_info->dev_name);
		dev_args->atoms = calloc(dev_args->atoms_size / sizeof(cl_uint), sizeof(cl_uint));

		// buffer for domains writable data
		// 0...N_VS*(D_MAX+1) - (D_MAX+1) solution stores because concurrency control
		dev_args->domains_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->domains_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer domains_mem", dev_info->dev_name);

		if (DOMAIN_TYPE == BITMAP_) {
			dev_args->bitmaps = malloc(dev_args->domains_size);

			// set buffer initial values to zero
			memset(dev_args->bitmaps, 0, dev_args->domains_size);
			cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->domains_mem, CL_TRUE, 0, dev_args->domains_size, dev_args->bitmaps, 0, NULL, NULL),
					"clEnqueueWriteBuffer domains_mem", dev_info->dev_name);

		} else if (DOMAIN_TYPE == INTERVAL) {
			dev_args->intervals = malloc(dev_args->domains_size);

			// set buffer initial values to zero
			memset(dev_args->intervals, 0, N_VS * (D_MAX + 1) * 4);
			cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->domains_mem, CL_TRUE, 0, dev_args->domains_size, dev_args->intervals, 0, NULL, NULL),
					"clEnqueueWriteBuffer domains_mem", dev_info->dev_name);
		}
	}

	if (N_DEVS > 1) {
		// to count number of propagations done per work-item, for rank calculation
		dev_args->props_mem = clCreateBuffer(dev_info->context, CL_MEM_WRITE_ONLY, dev_args->props_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer props_mem", dev_info->dev_name);
		dev_args->props = calloc(dev_args->props_size / sizeof(cl_ulong), sizeof(cl_ulong));
	}

#if SHARED_SS > 0
		// for work-sharing after the ss in the block have finished
		dev_args->shared_stores_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->shared_stores_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer shared_stores_mem", dev_info->dev_name);

		// flags for signaling the state of each work-sharing store
		// 0 - next shared SS to be picked
		// 1 - next shared SS to be filled
		// 2...number of SS already filled
		// 3..3+CL_N_SHARED_SS - V_ID that was labeled to generate this SS
		dev_args->shared_stores_flag_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->shared_stores_flag_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer shared_stores_flag_mem", dev_info->dev_name);
		dev_args->shared_stores_flag = calloc(dev_args->shared_stores_flag_size / sizeof(cl_int), sizeof(cl_int));
#endif

	if (PRINT_STATS) {
		// 0 - nodes_fail
		// 1 - nodes_expl
		// 2 - backtracks
		// 3 - labels
		// 4 - pruning
		// 5 - props_ok
		// 6 - max_depth
		// ... repeat per work-item
		dev_args->stats_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->stats_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer stats_mem", dev_info->dev_name);
		dev_args->stats = malloc(dev_args->stats_size);
		memset(dev_args->stats, 0, dev_args->stats_size);

		cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->stats_mem, CL_TRUE, 0, dev_args->stats_size, dev_args->stats, 0, NULL, NULL),
				"clEnqueueWriteBuffer stats_mem", dev_info->dev_name);
	}

	if (filtering) {
		// 0...N_VS - size of domains_mem buffer for the filtering result
		dev_args->filt_domains_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->filt_domains_size, NULL, &ret);
		cl_check_error(ret, "clCreateBuffer filt_domains_mem", dev_info->dev_name);

		if (DOMAIN_TYPE == BITMAP_) {
			dev_args->filt_bitmaps = malloc(dev_args->filt_domains_size);

		} else if (DOMAIN_TYPE == INTERVAL) {
			dev_args->filt_intervals = malloc(dev_args->filt_domains_size);
		}

		if (CS_IGNORE) {
			// 0...N_CS - size of filt_cs_mem buffer for the filtering
			dev_args->filt_cs_mem = clCreateBuffer(dev_info->context, CL_MEM_READ_WRITE, dev_args->filt_cs_size, NULL, &ret);
			cl_check_error(ret, "clCreateBuffer filt_cs_mem", dev_info->dev_name);

			dev_args->filt_cs = malloc(dev_args->filt_cs_size);
		}
	}

	// write permanent data to device buffers
	cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->ints_mem, CL_TRUE, 0, dev_args->ints_size, dev_args->ints, 0, NULL, NULL),
			"clEnqueueWriteBuffer int_p_mem", dev_info->dev_name);
	cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->generic_mem, CL_TRUE, 0, dev_args->generic_size, dev_args->generic, 0, NULL, NULL),
			"clEnqueueWriteBuffer generic_mem", dev_info->dev_name);
	if (DOMAIN_TYPE == BITMAP_) {
		cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->cl_vs_mem, CL_TRUE, 0, dev_args->cl_vs_size, dev_args->cl_vs_bitmaps, 0, NULL, NULL),
				"clEnqueueWriteBuffer cl_vs_mem", dev_info->dev_name);
		cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->b_ds_mem, CL_TRUE, 0, dev_args->b_ds_size, dev_args->b_ds, 0, NULL, NULL),
				"clEnqueueWriteBuffer bitmaps_aux_mem", dev_info->dev_name);
	} else if (DOMAIN_TYPE == INTERVAL) {
		cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->cl_vs_mem, CL_TRUE, 0, dev_args->cl_vs_size, dev_args->cl_vs_intervals, 0, NULL, NULL),
				"clEnqueueWriteBuffer cl_vs_p_mem", dev_info->dev_name);
	}
	cl_check_error(clEnqueueWriteBuffer(dev_args->cq, dev_args->cl_cs_mem, CL_TRUE, 0, dev_args->cl_cs_size, dev_args->cl_cs, 0, NULL, NULL),
			"clEnqueueWriteBuffer cl_cs_p_mem", dev_info->dev_name);

	// Set OpenCL kernel arguments
	cl_uint arg_number = 0;
	cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->atoms_mem), &dev_args->atoms_mem), "clSetKernelArg atoms_p_mem",
			dev_info->dev_name);
	cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->ints_mem), &dev_args->ints_mem), "clSetKernelArg int_p_mem",
			dev_info->dev_name);
	cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->backtrack_mem1), &dev_args->backtrack_mem1), "clSetKernelArg backtrack_mem1",
			dev_info->dev_name);

	// more buffers for backtracking
#if USE_MORE_BUFFERS
	if (dev_info->n_buffers > 1) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->backtrack_mem2), &dev_args->backtrack_mem2),
				"clSetKernelArg backtrack_mem2", dev_info->dev_name);
	}
	if (dev_info->n_buffers > 2) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->backtrack_mem3), &dev_args->backtrack_mem3),
				"clSetKernelArg backtrack_mem3", dev_info->dev_name);
	}
	if (dev_info->n_buffers > 3) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->backtrack_mem4), &dev_args->backtrack_mem4),
				"clSetKernelArg backtrack_mem4", dev_info->dev_name);
	}
#endif

	cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->generic_mem), &dev_args->generic_mem), "clSetKernelArg generic_mem",
			dev_info->dev_name);

	if (CS_IGNORE) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->cs_ignore_mem), &dev_args->cs_ignore_mem),
				"clSetKernelArg cs_ignore_mem", dev_info->dev_name);
	}

	if (WORK == ONE || WORK == OPT) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->domains_mem), &dev_args->domains_mem), "clSetKernelArg domains_mem",
				dev_info->dev_name);
	}

	cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->cl_vs_mem), &dev_args->cl_vs_mem), "clSetKernelArg cl_vs_mem",
			dev_info->dev_name);
	cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->cl_cs_mem), &dev_args->cl_cs_mem), "clSetKernelArg cl_cs_mem",
			dev_info->dev_name);
	// if using local memory
	if (dev_info->use_local_mem) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, dev_args->cl_vs_prop_size, NULL), "clSetKernelArg cl_vs_prop_mem", dev_info->dev_name);
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, dev_args->vs_id_to_prop_size, NULL), "clSetKernelArg vs_id_to_prop_mem",
				dev_info->dev_name);
		// if not using local memory
	} else {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->cl_vs_prop_mem), &dev_args->cl_vs_prop_mem),
				"clSetKernelArg cl_vs_prop_mem", dev_info->dev_name);
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->vs_id_to_prop_mem), &dev_args->vs_id_to_prop_mem),
				"clSetKernelArg vs_id_to_prop_mem", dev_info->dev_name);
	}

	if (DOMAIN_TYPE == BITMAP_) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->b_ds_mem), &dev_args->b_ds_mem), "clSetKernelArg bitmaps_aux_mem",
				dev_info->dev_name);
	}

	if (PRINT_STATS) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->stats_mem), &dev_args->stats_mem), "clSetKernelArg stats_mem",
				dev_info->dev_name);
	}

	if (N_DEVS > 1) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->props_mem), &dev_args->props_mem), "clSetKernelArg props_mem",
				dev_info->dev_name);
	}
#if SHARED_SS > 0
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->shared_stores_mem), &dev_args->shared_stores_mem), "clSetKernelArg shared_stores_mem",
				dev_info->dev_name);
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->shared_stores_flag_mem), &dev_args->shared_stores_flag_mem),
				"clSetKernelArg shared_stores_flag_mem", dev_info->dev_name);
#endif

	if (filtering) {
		cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->filt_domains_mem), &dev_args->filt_domains_mem),
				"clSetKernelArg filt_domains_mem", dev_info->dev_name);
		if (CS_IGNORE) {
			cl_check_error(clSetKernelArg(dev_args->kernel, arg_number++, sizeof(dev_args->filt_cs_mem), &dev_args->filt_cs_mem), "clSetKernelArg filt_cs_mem",
					dev_info->dev_name);
		}
	}

#if RUN_IN_CUDA
	}
#endif
}

/*
 * Clear device objects
 * dev_args - device_args structure about this device
 * dev_info - device_info structure about this device
 * filtering - if its the prefiltering phase
 */
void release_device(device_args *dev_args, device_info *dev_info, bool filtering) {

#if RUN_IN_CUDA

	if (dev_info->type == CL_DEVICE_TYPE_GPU) {

		CUresult err = cuInit(0);

		err = cuMemFree(dev_args->atoms_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree atoms_mem_cu", dev_info->dev_name);

		err = cuMemFree(dev_args->ints_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree ints_mem_cu", dev_info->dev_name);

		err = cuMemFree(dev_args->backtrack_mem1_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree backtrack_mem1_cu", dev_info->dev_name);

		// more buffers for backtracking
#if USE_MORE_BUFFERS
		if (dev_info->n_buffers > 1) {
			err = cuMemFree(dev_args->backtrack_mem2_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree backtrack_mem2_cu", dev_info->dev_name);
		}
		if (dev_info->n_buffers > 2) {
			err = cuMemFree(dev_args->backtrack_mem3_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree backtrack_mem3_cu", dev_info->dev_name);
		}
		if (dev_info->n_buffers > 3) {
			err = cuMemFree(dev_args->backtrack_mem4_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree backtrack_mem4_cu", dev_info->dev_name);
		}
#endif

		err = cuMemFree(dev_args->generic_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree generic_mem_cu", dev_info->dev_name);

		if (CS_IGNORE) {
			err = cuMemFree(dev_args->cs_ignore_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree cs_ignore_mem_cu", dev_info->dev_name);
		}

		if (WORK == ONE || WORK == OPT) {
			err = cuMemFree(dev_args->domains_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree domains_mem_cu", dev_info->dev_name);
		}

		err = cuMemFree(dev_args->cl_vs_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree cl_vs_mem_cu", dev_info->dev_name);

		err = cuMemFree(dev_args->cl_cs_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree cl_cs_mem_cu", dev_info->dev_name);

		// if not using local memory
		if (!dev_info->use_local_mem) {
			err = cuMemFree(dev_args->vs_id_to_prop_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree vs_id_to_prop_mem_cu", dev_info->dev_name);

			err = cuMemFree(dev_args->cl_vs_prop_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree cl_vs_prop_mem_cu", dev_info->dev_name);
		}

		if (PRINT_STATS) {
			err = cuMemFree(dev_args->stats_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree stats_mem_cu", dev_info->dev_name);
			free(dev_args->stats);
		}

#if SHARED_SS > 0
		err = cuMemFree(dev_args->shared_stores_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree shared_stores_mem_cu", dev_info->dev_name);

		err = cuMemFree(dev_args->shared_stores_flag_mem_cu);
		cuda_check_error(err, "cuMemAlloc cuMemFree shared_stores_flag_mem_cu", dev_info->dev_name);
#endif

		if (N_DEVS > 1) {
			err = cuMemFree(dev_args->props_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree props_mem_cu", dev_info->dev_name);
		}

		if (filtering) {
			err = cuMemFree(dev_args->filt_domains_mem_cu);
			cuda_check_error(err, "cuMemAlloc cuMemFree filt_domains_mem_cu", dev_info->dev_name);

			if (CS_IGNORE) {

				err = cuMemFree(dev_args->filt_cs_mem_cu);
				cuda_check_error(err, "cuMemAlloc cuMemFree filt_cs_mem_cu", dev_info->dev_name);
			}
		}

		free(dev_args->atoms);
		free(dev_args->ints);
		free(dev_args->generic);
		if (DOMAIN_TYPE == BITMAP_) {

			if (WORK == ONE || WORK == OPT) {
				free(dev_args->bitmaps);
			}
			free(dev_args->b_ds);
			free(dev_args->cl_vs_bitmaps);

		} else if (DOMAIN_TYPE == INTERVAL) {

			if (WORK == ONE || WORK == OPT) {
				free(dev_args->intervals);
			}
			free(dev_args->cl_vs_intervals);
		}
		free(dev_args->cl_cs);
#if SHARED_SS > 0
		free(dev_args->shared_stores_flag);
#endif

		if (N_DEVS > 1) {
			free(dev_args->props);
		}

		if (filtering) {
			if (DOMAIN_TYPE == BITMAP_) {
				free(dev_args->filt_bitmaps);
			} else {
				free(dev_args->filt_intervals);
			}
			if (CS_IGNORE) {
				free(dev_args->filt_cs);
			}
		}

		cuCtxDestroy(dev_args->context_cu);

	} else {

#endif

	cl_check_error(clFlush(dev_args->cq), "clFlush", dev_info->dev_name);
	cl_check_error(clFinish(dev_args->cq), "clFinish", dev_info->dev_name);

	cl_check_error(clReleaseKernel(dev_args->kernel), "clReleaseKernel", dev_info->dev_name);

	cl_check_error(clReleaseMemObject(dev_args->atoms_mem), "clReleaseMemObject atoms_p_mem", dev_info->dev_name);
	cl_check_error(clReleaseMemObject(dev_args->ints_mem), "clReleaseMemObject int_p_mem", dev_info->dev_name);

	cl_check_error(clReleaseMemObject(dev_args->backtrack_mem1), "clReleaseMemObject backtrack_mem1", dev_info->dev_name);

	// more buffers for backtracking
#if USE_MORE_BUFFERS
	if (dev_info->n_buffers > 1) {
		cl_check_error(clReleaseMemObject(dev_args->backtrack_mem2), "clReleaseMemObject backtrack_mem2", dev_info->dev_name);
	}
	if (dev_info->n_buffers > 2) {
		cl_check_error(clReleaseMemObject(dev_args->backtrack_mem3), "clReleaseMemObject backtrack_mem3", dev_info->dev_name);
	}
	if (dev_info->n_buffers > 3) {
		cl_check_error(clReleaseMemObject(dev_args->backtrack_mem4), "clReleaseMemObject backtrack_mem4", dev_info->dev_name);
	}
#endif

	cl_check_error(clReleaseMemObject(dev_args->generic_mem), "clReleaseMemObject generic_mem", dev_info->dev_name);

	if (CS_IGNORE) {
		cl_check_error(clReleaseMemObject(dev_args->cs_ignore_mem), "clReleaseMemObject cs_ignore_mem", dev_info->dev_name);
	}

	if (WORK == ONE || WORK == OPT) {
		cl_check_error(clReleaseMemObject(dev_args->domains_mem), "clReleaseMemObject domains_mem", dev_info->dev_name);
	}

	cl_check_error(clReleaseMemObject(dev_args->cl_vs_mem), "clReleaseMemObject cl_vs_mem", dev_info->dev_name);
	cl_check_error(clReleaseMemObject(dev_args->cl_cs_mem), "clReleaseMemObject cl_cs_mem", dev_info->dev_name);
	// if not using local memory
	if (!dev_info->use_local_mem) {
		cl_check_error(clReleaseMemObject(dev_args->vs_id_to_prop_mem), "clReleaseMemObject vs_id_to_prop_mem", dev_info->dev_name);
		cl_check_error(clReleaseMemObject(dev_args->cl_vs_prop_mem), "clReleaseMemObject cl_vs_prop_mem", dev_info->dev_name);
	}

	if (PRINT_STATS) {
		cl_check_error(clReleaseMemObject(dev_args->stats_mem), "clReleaseMemObject stats_mem", dev_info->dev_name);
		free(dev_args->stats);
	}

#if SHARED_SS > 0
		cl_check_error(clReleaseMemObject(dev_args->shared_stores_mem), "clReleaseMemObject shared_stores_mem", dev_info->dev_name);
		cl_check_error(clReleaseMemObject(dev_args->shared_stores_flag_mem), "clReleaseMemObject shared_stores_flag_mem", dev_info->dev_name);
#endif

	if (N_DEVS > 1) {
		cl_check_error(clReleaseMemObject(dev_args->props_mem), "clReleaseMemObject props_mem", dev_info->dev_name);
	}

	if (filtering) {
		cl_check_error(clReleaseMemObject(dev_args->filt_domains_mem), "clReleaseMemObject filt_domains_mem", dev_info->dev_name);
		if (CS_IGNORE) {
			cl_check_error(clReleaseMemObject(dev_args->filt_cs_mem), "clReleaseMemObject filt_cs_mem", dev_info->dev_name);
		}
	}

	free(dev_args->atoms);
	free(dev_args->ints);
	free(dev_args->generic);
	if (DOMAIN_TYPE == BITMAP_) {

		if (WORK == ONE || WORK == OPT) {
			free(dev_args->bitmaps);
		}
		free(dev_args->b_ds);
		free(dev_args->cl_vs_bitmaps);

	} else if (DOMAIN_TYPE == INTERVAL) {

		if (WORK == ONE || WORK == OPT) {
			free(dev_args->intervals);
		}
		free(dev_args->cl_vs_intervals);
	}
	free(dev_args->cl_cs);
#if SHARED_SS > 0
		free(dev_args->shared_stores_flag);
#endif

	if (N_DEVS > 1) {
		free(dev_args->props);
	}

	if (filtering) {
		if (DOMAIN_TYPE == BITMAP_) {
			free(dev_args->filt_bitmaps);
		} else {
			free(dev_args->filt_intervals);
		}
		if (CS_IGNORE) {
			free(dev_args->filt_cs);
		}
	}

	cl_check_error(clReleaseProgram(dev_info->prog), "clReleaseProgram", dev_info->dev_name);
	cl_check_error(clReleaseCommandQueue(dev_args->cq), "clReleaseCommandQueue", dev_info->dev_name);
	cl_check_error(clReleaseContext(dev_info->context), "clReleaseContext", dev_info->dev_name);

#if RUN_IN_CUDA
	}
#endif
}