1 // Copyright (c) 2017, Lawrence Livermore National Security, LLC. Produced at 2 // the Lawrence Livermore National Laboratory. LLNL-CODE-734707. All Rights 3 // reserved. See files LICENSE and NOTICE for details. 4 // 5 // This file is part of CEED, a collection of benchmarks, miniapps, software 6 // libraries and APIs for efficient high-order finite element and spectral 7 // element discretizations for exascale applications. For more information and 8 // source code availability see http://github.com/ceed. 9 // 10 // The CEED research is supported by the Exascale Computing Project 17-SC-20-SC, 11 // a collaborative effort of two U.S. Department of Energy organizations (Office 12 // of Science and the National Nuclear Security Administration) responsible for 13 // the planning and preparation of a capable exascale ecosystem, including 14 // software, applications, hardware, advanced system engineering and early 15 // testbed platforms, in support of the nation's exascale computing imperative. 16 17 // libCEED + PETSc Example: CEED BPs 3-6 with Multigrid 18 // 19 // This example demonstrates a simple usage of libCEED with PETSc to solve the 20 // CEED BP benchmark problems, see http://ceed.exascaleproject.org/bps. 21 // 22 // The code uses higher level communication protocols in DMPlex. 23 // 24 // Build with: 25 // 26 // make multigrid [PETSC_DIR=</path/to/petsc>] [CEED_DIR=</path/to/libceed>] 27 // 28 // Sample runs: 29 // 30 // multigrid -problem bp3 31 // multigrid -problem bp4 32 // multigrid -problem bp5 -ceed /cpu/self 33 // multigrid -problem bp6 -ceed /gpu/cuda 34 // 35 //TESTARGS -ceed {ceed_resource} -test -problem bp3 -degree 3 36 37 /// @file 38 /// CEED BPs 1-6 multigrid example using PETSc 39 const char help[] = "Solve CEED BPs using p-multigrid with PETSc and DMPlex\n"; 40 41 #include "bps.h" 42 43 int main(int argc, char **argv) { 44 PetscInt ierr; 45 MPI_Comm comm; 46 char filename[PETSC_MAX_PATH_LEN], 47 ceed_resource[PETSC_MAX_PATH_LEN] = "/cpu/self"; 48 double my_rt_start, my_rt, rt_min, rt_max; 49 PetscInt degree = 3, q_extra, *l_size, *xl_size, *g_size, dim = 3, fine_level, 50 mesh_elem[3] = {3, 3, 3}, num_comp_u = 1, num_levels = degree, *level_degrees; 51 PetscScalar *r; 52 PetscScalar eps = 1.0; 53 PetscBool test_mode, benchmark_mode, read_mesh, write_solution; 54 PetscLogStage solve_stage; 55 DM *dm, dm_orig; 56 SNES snes_dummy; 57 KSP ksp; 58 PC pc; 59 Mat *mat_O, *mat_pr, mat_coarse; 60 Vec *X, *X_loc, *mult, rhs, rhs_loc; 61 PetscMemType mem_type; 62 UserO *user_O; 63 UserProlongRestr *user_pr; 64 Ceed ceed; 65 CeedData *ceed_data; 66 CeedVector rhs_ceed, target; 67 CeedQFunction qf_error, qf_restrict, qf_prolong; 68 CeedOperator op_error; 69 BPType bp_choice; 70 CoarsenType coarsen; 71 72 ierr = PetscInitialize(&argc, &argv, NULL, help); 73 if (ierr) return ierr; 74 comm = PETSC_COMM_WORLD; 75 76 // Parse command line options 77 ierr = PetscOptionsBegin(comm, NULL, "CEED BPs in PETSc", NULL); CHKERRQ(ierr); 78 bp_choice = CEED_BP3; 79 ierr = PetscOptionsEnum("-problem", 80 "CEED benchmark problem to solve", NULL, 81 bp_types, (PetscEnum)bp_choice, (PetscEnum *)&bp_choice, 82 NULL); CHKERRQ(ierr); 83 num_comp_u = bp_options[bp_choice].num_comp_u; 84 test_mode = PETSC_FALSE; 85 ierr = PetscOptionsBool("-test", 86 "Testing mode (do not print unless error is large)", 87 NULL, test_mode, &test_mode, NULL); CHKERRQ(ierr); 88 benchmark_mode = PETSC_FALSE; 89 ierr = PetscOptionsBool("-benchmark", 90 "Benchmarking mode (prints benchmark statistics)", 91 NULL, benchmark_mode, &benchmark_mode, NULL); 92 CHKERRQ(ierr); 93 write_solution = PETSC_FALSE; 94 ierr = PetscOptionsBool("-write_solution", 95 "Write solution for visualization", 96 NULL, write_solution, &write_solution, NULL); 97 CHKERRQ(ierr); 98 ierr = PetscOptionsScalar("-eps", 99 "Epsilon parameter for Kershaw mesh transformation", 100 NULL, eps, &eps, NULL); 101 if (eps > 1 || eps <= 0) SETERRQ1(PETSC_COMM_WORLD, PETSC_ERR_ARG_OUTOFRANGE, 102 "-eps %D must be (0,1]", eps); 103 degree = test_mode ? 3 : 2; 104 ierr = PetscOptionsInt("-degree", "Polynomial degree of tensor product basis", 105 NULL, degree, °ree, NULL); CHKERRQ(ierr); 106 if (degree < 1) SETERRQ1(PETSC_COMM_WORLD, PETSC_ERR_ARG_OUTOFRANGE, 107 "-degree %D must be at least 1", degree); 108 q_extra = bp_options[bp_choice].q_extra; 109 ierr = PetscOptionsInt("-q_extra", "Number of extra quadrature points", 110 NULL, q_extra, &q_extra, NULL); CHKERRQ(ierr); 111 ierr = PetscOptionsString("-ceed", "CEED resource specifier", 112 NULL, ceed_resource, ceed_resource, 113 sizeof(ceed_resource), NULL); CHKERRQ(ierr); 114 coarsen = COARSEN_UNIFORM; 115 ierr = PetscOptionsEnum("-coarsen", 116 "Coarsening strategy to use", NULL, 117 coarsen_types, (PetscEnum)coarsen, 118 (PetscEnum *)&coarsen, NULL); CHKERRQ(ierr); 119 read_mesh = PETSC_FALSE; 120 ierr = PetscOptionsString("-mesh", "Read mesh from file", NULL, 121 filename, filename, sizeof(filename), &read_mesh); 122 CHKERRQ(ierr); 123 if (!read_mesh) { 124 PetscInt tmp = dim; 125 ierr = PetscOptionsIntArray("-cells","Number of cells per dimension", NULL, 126 mesh_elem, &tmp, NULL); CHKERRQ(ierr); 127 } 128 ierr = PetscOptionsEnd(); CHKERRQ(ierr); 129 130 // Set up libCEED 131 CeedInit(ceed_resource, &ceed); 132 CeedMemType mem_type_backend; 133 CeedGetPreferredMemType(ceed, &mem_type_backend); 134 135 // Setup DM 136 if (read_mesh) { 137 ierr = DMPlexCreateFromFile(PETSC_COMM_WORLD, filename, PETSC_TRUE, &dm_orig); 138 CHKERRQ(ierr); 139 } else { 140 ierr = DMPlexCreateBoxMesh(PETSC_COMM_WORLD, dim, PETSC_FALSE, mesh_elem, NULL, 141 NULL, NULL, PETSC_TRUE, &dm_orig); CHKERRQ(ierr); 142 } 143 144 { 145 DM dm_dist = NULL; 146 PetscPartitioner part; 147 148 ierr = DMPlexGetPartitioner(dm_orig, &part); CHKERRQ(ierr); 149 ierr = PetscPartitionerSetFromOptions(part); CHKERRQ(ierr); 150 ierr = DMPlexDistribute(dm_orig, 0, NULL, &dm_dist); CHKERRQ(ierr); 151 if (dm_dist) { 152 ierr = DMDestroy(&dm_orig); CHKERRQ(ierr); 153 dm_orig = dm_dist; 154 } 155 } 156 157 // Apply Kershaw mesh transformation 158 ierr = Kershaw(dm_orig, eps); CHKERRQ(ierr); 159 160 VecType vec_type; 161 switch (mem_type_backend) { 162 case CEED_MEM_HOST: vec_type = VECSTANDARD; break; 163 case CEED_MEM_DEVICE: { 164 const char *resolved; 165 CeedGetResource(ceed, &resolved); 166 if (strstr(resolved, "/gpu/cuda")) vec_type = VECCUDA; 167 else if (strstr(resolved, "/gpu/hip/occa")) 168 vec_type = VECSTANDARD; // https://github.com/CEED/libCEED/issues/678 169 else if (strstr(resolved, "/gpu/hip")) vec_type = VECHIP; 170 else vec_type = VECSTANDARD; 171 } 172 } 173 ierr = DMSetVecType(dm_orig, vec_type); CHKERRQ(ierr); 174 ierr = DMSetFromOptions(dm_orig); CHKERRQ(ierr); 175 176 // Allocate arrays for PETSc objects for each level 177 switch (coarsen) { 178 case COARSEN_UNIFORM: 179 num_levels = degree; 180 break; 181 case COARSEN_LOGARITHMIC: 182 num_levels = ceil(log(degree)/log(2)) + 1; 183 break; 184 } 185 ierr = PetscMalloc1(num_levels, &level_degrees); CHKERRQ(ierr); 186 fine_level = num_levels - 1; 187 188 switch (coarsen) { 189 case COARSEN_UNIFORM: 190 for (int i=0; i<num_levels; i++) level_degrees[i] = i + 1; 191 break; 192 case COARSEN_LOGARITHMIC: 193 for (int i=0; i<num_levels - 1; i++) level_degrees[i] = pow(2,i); 194 level_degrees[fine_level] = degree; 195 break; 196 } 197 ierr = PetscMalloc1(num_levels, &dm); CHKERRQ(ierr); 198 ierr = PetscMalloc1(num_levels, &X); CHKERRQ(ierr); 199 ierr = PetscMalloc1(num_levels, &X_loc); CHKERRQ(ierr); 200 ierr = PetscMalloc1(num_levels, &mult); CHKERRQ(ierr); 201 ierr = PetscMalloc1(num_levels, &user_O); CHKERRQ(ierr); 202 ierr = PetscMalloc1(num_levels, &user_pr); CHKERRQ(ierr); 203 ierr = PetscMalloc1(num_levels, &mat_O); CHKERRQ(ierr); 204 ierr = PetscMalloc1(num_levels, &mat_pr); CHKERRQ(ierr); 205 ierr = PetscMalloc1(num_levels, &l_size); CHKERRQ(ierr); 206 ierr = PetscMalloc1(num_levels, &xl_size); CHKERRQ(ierr); 207 ierr = PetscMalloc1(num_levels, &g_size); CHKERRQ(ierr); 208 209 // Setup DM and Operator Mat Shells for each level 210 for (CeedInt i=0; i<num_levels; i++) { 211 // Create DM 212 ierr = DMClone(dm_orig, &dm[i]); CHKERRQ(ierr); 213 ierr = DMGetVecType(dm_orig, &vec_type); CHKERRQ(ierr); 214 ierr = DMSetVecType(dm[i], vec_type); CHKERRQ(ierr); 215 PetscInt dim; 216 ierr = DMGetDimension(dm[i], &dim); CHKERRQ(ierr); 217 ierr = SetupDMByDegree(dm[i], level_degrees[i], num_comp_u, dim, 218 bp_options[bp_choice].enforce_bc, bp_options[bp_choice].bc_func); 219 CHKERRQ(ierr); 220 221 // Create vectors 222 ierr = DMCreateGlobalVector(dm[i], &X[i]); CHKERRQ(ierr); 223 ierr = VecGetLocalSize(X[i], &l_size[i]); CHKERRQ(ierr); 224 ierr = VecGetSize(X[i], &g_size[i]); CHKERRQ(ierr); 225 ierr = DMCreateLocalVector(dm[i], &X_loc[i]); CHKERRQ(ierr); 226 ierr = VecGetSize(X_loc[i], &xl_size[i]); CHKERRQ(ierr); 227 228 // Operator 229 ierr = PetscMalloc1(1, &user_O[i]); CHKERRQ(ierr); 230 ierr = MatCreateShell(comm, l_size[i], l_size[i], g_size[i], g_size[i], 231 user_O[i], &mat_O[i]); CHKERRQ(ierr); 232 ierr = MatShellSetOperation(mat_O[i], MATOP_MULT, 233 (void(*)(void))MatMult_Ceed); CHKERRQ(ierr); 234 ierr = MatShellSetOperation(mat_O[i], MATOP_GET_DIAGONAL, 235 (void(*)(void))MatGetDiag); CHKERRQ(ierr); 236 ierr = MatShellSetVecType(mat_O[i], vec_type); CHKERRQ(ierr); 237 238 // Level transfers 239 if (i > 0) { 240 // Interp 241 ierr = PetscMalloc1(1, &user_pr[i]); CHKERRQ(ierr); 242 ierr = MatCreateShell(comm, l_size[i], l_size[i-1], g_size[i], g_size[i-1], 243 user_pr[i], &mat_pr[i]); CHKERRQ(ierr); 244 ierr = MatShellSetOperation(mat_pr[i], MATOP_MULT, 245 (void(*)(void))MatMult_Prolong); 246 CHKERRQ(ierr); 247 ierr = MatShellSetOperation(mat_pr[i], MATOP_MULT_TRANSPOSE, 248 (void(*)(void))MatMult_Restrict); 249 CHKERRQ(ierr); 250 ierr = MatShellSetVecType(mat_pr[i], vec_type); CHKERRQ(ierr); 251 } 252 } 253 ierr = VecDuplicate(X[fine_level], &rhs); CHKERRQ(ierr); 254 255 // Print global grid information 256 if (!test_mode) { 257 PetscInt P = degree + 1, Q = P + q_extra; 258 259 const char *used_resource; 260 CeedGetResource(ceed, &used_resource); 261 262 ierr = VecGetType(X[0], &vec_type); CHKERRQ(ierr); 263 264 ierr = PetscPrintf(comm, 265 "\n-- CEED Benchmark Problem %d -- libCEED + PETSc + PCMG --\n" 266 " PETSc:\n" 267 " PETSc Vec Type : %s\n" 268 " libCEED:\n" 269 " libCEED Backend : %s\n" 270 " libCEED Backend MemType : %s\n" 271 " Mesh:\n" 272 " Number of 1D Basis Nodes (p) : %d\n" 273 " Number of 1D Quadrature Points (q) : %d\n" 274 " Global Nodes : %D\n" 275 " Owned Nodes : %D\n" 276 " DoF per node : %D\n" 277 " Multigrid:\n" 278 " Number of Levels : %d\n", 279 bp_choice+1, vec_type, used_resource, 280 CeedMemTypes[mem_type_backend], 281 P, Q, g_size[fine_level]/num_comp_u, l_size[fine_level]/num_comp_u, 282 num_comp_u, num_levels); CHKERRQ(ierr); 283 } 284 285 // Create RHS vector 286 ierr = VecDuplicate(X_loc[fine_level], &rhs_loc); CHKERRQ(ierr); 287 ierr = VecZeroEntries(rhs_loc); CHKERRQ(ierr); 288 ierr = VecGetArrayAndMemType(rhs_loc, &r, &mem_type); CHKERRQ(ierr); 289 CeedVectorCreate(ceed, xl_size[fine_level], &rhs_ceed); 290 CeedVectorSetArray(rhs_ceed, MemTypeP2C(mem_type), CEED_USE_POINTER, r); 291 292 // Set up libCEED operators on each level 293 ierr = PetscMalloc1(num_levels, &ceed_data); CHKERRQ(ierr); 294 for (int i=0; i<num_levels; i++) { 295 // Print level information 296 if (!test_mode && (i == 0 || i == fine_level)) { 297 ierr = PetscPrintf(comm," Level %D (%s):\n" 298 " Number of 1D Basis Nodes (p) : %d\n" 299 " Global Nodes : %D\n" 300 " Owned Nodes : %D\n", 301 i, (i? "fine" : "coarse"), level_degrees[i] + 1, 302 g_size[i]/num_comp_u, l_size[i]/num_comp_u); CHKERRQ(ierr); 303 } 304 ierr = PetscMalloc1(1, &ceed_data[i]); CHKERRQ(ierr); 305 ierr = SetupLibceedByDegree(dm[i], ceed, level_degrees[i], dim, q_extra, 306 dim, num_comp_u, g_size[i], xl_size[i], bp_options[bp_choice], 307 ceed_data[i], i==(fine_level), rhs_ceed, &target); 308 CHKERRQ(ierr); 309 } 310 311 // Gather RHS 312 CeedVectorTakeArray(rhs_ceed, MemTypeP2C(mem_type), NULL); 313 ierr = VecRestoreArrayAndMemType(rhs_loc, &r); CHKERRQ(ierr); 314 ierr = VecZeroEntries(rhs); CHKERRQ(ierr); 315 ierr = DMLocalToGlobal(dm[fine_level], rhs_loc, ADD_VALUES, rhs); CHKERRQ(ierr); 316 CeedVectorDestroy(&rhs_ceed); 317 318 // Create the restriction/interpolation QFunction 319 CeedQFunctionCreateIdentity(ceed, num_comp_u, CEED_EVAL_NONE, CEED_EVAL_INTERP, 320 &qf_restrict); 321 CeedQFunctionCreateIdentity(ceed, num_comp_u, CEED_EVAL_INTERP, CEED_EVAL_NONE, 322 &qf_prolong); 323 324 // Set up libCEED level transfer operators 325 ierr = CeedLevelTransferSetup(ceed, num_levels, num_comp_u, ceed_data, 326 level_degrees, 327 qf_restrict, qf_prolong); CHKERRQ(ierr); 328 329 // Create the error QFunction 330 CeedQFunctionCreateInterior(ceed, 1, bp_options[bp_choice].error, 331 bp_options[bp_choice].error_loc, &qf_error); 332 CeedQFunctionAddInput(qf_error, "u", num_comp_u, CEED_EVAL_INTERP); 333 CeedQFunctionAddInput(qf_error, "true_soln", num_comp_u, CEED_EVAL_NONE); 334 CeedQFunctionAddOutput(qf_error, "error", num_comp_u, CEED_EVAL_NONE); 335 336 // Create the error operator 337 CeedOperatorCreate(ceed, qf_error, CEED_QFUNCTION_NONE, CEED_QFUNCTION_NONE, 338 &op_error); 339 CeedOperatorSetField(op_error, "u", ceed_data[fine_level]->elem_restr_u, 340 ceed_data[fine_level]->basis_u, CEED_VECTOR_ACTIVE); 341 CeedOperatorSetField(op_error, "true_soln", 342 ceed_data[fine_level]->elem_restr_u_i, 343 CEED_BASIS_COLLOCATED, target); 344 CeedOperatorSetField(op_error, "error", ceed_data[fine_level]->elem_restr_u_i, 345 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE); 346 347 // Calculate multiplicity 348 for (int i=0; i<num_levels; i++) { 349 PetscScalar *x; 350 351 // CEED vector 352 ierr = VecZeroEntries(X_loc[i]); CHKERRQ(ierr); 353 ierr = VecGetArray(X_loc[i], &x); CHKERRQ(ierr); 354 CeedVectorSetArray(ceed_data[i]->x_ceed, CEED_MEM_HOST, CEED_USE_POINTER, x); 355 356 // Multiplicity 357 CeedElemRestrictionGetMultiplicity(ceed_data[i]->elem_restr_u, 358 ceed_data[i]->x_ceed); 359 CeedVectorSyncArray(ceed_data[i]->x_ceed, CEED_MEM_HOST); 360 361 // Restore vector 362 ierr = VecRestoreArray(X_loc[i], &x); CHKERRQ(ierr); 363 364 // Creat mult vector 365 ierr = VecDuplicate(X_loc[i], &mult[i]); CHKERRQ(ierr); 366 367 // Local-to-global 368 ierr = VecZeroEntries(X[i]); CHKERRQ(ierr); 369 ierr = DMLocalToGlobal(dm[i], X_loc[i], ADD_VALUES, X[i]); 370 CHKERRQ(ierr); 371 ierr = VecZeroEntries(X_loc[i]); CHKERRQ(ierr); 372 373 // Global-to-local 374 ierr = DMGlobalToLocal(dm[i], X[i], INSERT_VALUES, mult[i]); 375 CHKERRQ(ierr); 376 ierr = VecZeroEntries(X[i]); CHKERRQ(ierr); 377 378 // Multiplicity scaling 379 ierr = VecReciprocal(mult[i]); 380 } 381 382 // Set up Mat 383 for (int i=0; i<num_levels; i++) { 384 // User Operator 385 user_O[i]->comm = comm; 386 user_O[i]->dm = dm[i]; 387 user_O[i]->X_loc = X_loc[i]; 388 ierr = VecDuplicate(X_loc[i], &user_O[i]->Y_loc); CHKERRQ(ierr); 389 user_O[i]->x_ceed = ceed_data[i]->x_ceed; 390 user_O[i]->y_ceed = ceed_data[i]->y_ceed; 391 user_O[i]->op = ceed_data[i]->op_apply; 392 user_O[i]->ceed = ceed; 393 394 if (i > 0) { 395 // Prolongation/Restriction Operator 396 user_pr[i]->comm = comm; 397 user_pr[i]->dmf = dm[i]; 398 user_pr[i]->dmc = dm[i-1]; 399 user_pr[i]->loc_vec_c = X_loc[i-1]; 400 user_pr[i]->loc_vec_f = user_O[i]->Y_loc; 401 user_pr[i]->mult_vec = mult[i]; 402 user_pr[i]->ceed_vec_c = user_O[i-1]->x_ceed; 403 user_pr[i]->ceed_vec_f = user_O[i]->y_ceed; 404 user_pr[i]->op_prolong = ceed_data[i]->op_prolong; 405 user_pr[i]->op_restrict = ceed_data[i]->op_restrict; 406 user_pr[i]->ceed = ceed; 407 } 408 } 409 410 // Setup dummy SNES for AMG coarse solve 411 ierr = SNESCreate(comm, &snes_dummy); CHKERRQ(ierr); 412 ierr = SNESSetDM(snes_dummy, dm[0]); CHKERRQ(ierr); 413 ierr = SNESSetSolution(snes_dummy, X[0]); CHKERRQ(ierr); 414 415 // -- Jacobian matrix 416 ierr = DMSetMatType(dm[0], MATAIJ); CHKERRQ(ierr); 417 ierr = DMCreateMatrix(dm[0], &mat_coarse); CHKERRQ(ierr); 418 ierr = SNESSetJacobian(snes_dummy, mat_coarse, mat_coarse, NULL, 419 NULL); CHKERRQ(ierr); 420 421 // -- Residual evaluation function 422 ierr = SNESSetFunction(snes_dummy, X[0], FormResidual_Ceed, 423 user_O[0]); CHKERRQ(ierr); 424 425 // -- Form Jacobian 426 ierr = SNESComputeJacobianDefaultColor(snes_dummy, X[0], mat_O[0], 427 mat_coarse, NULL); CHKERRQ(ierr); 428 429 // Set up KSP 430 ierr = KSPCreate(comm, &ksp); CHKERRQ(ierr); 431 { 432 ierr = KSPSetType(ksp, KSPCG); CHKERRQ(ierr); 433 ierr = KSPSetNormType(ksp, KSP_NORM_NATURAL); CHKERRQ(ierr); 434 ierr = KSPSetTolerances(ksp, 1e-10, PETSC_DEFAULT, PETSC_DEFAULT, 435 PETSC_DEFAULT); CHKERRQ(ierr); 436 } 437 ierr = KSPSetFromOptions(ksp); CHKERRQ(ierr); 438 ierr = KSPSetOperators(ksp, mat_O[fine_level], mat_O[fine_level]); 439 CHKERRQ(ierr); 440 441 // Set up PCMG 442 ierr = KSPGetPC(ksp, &pc); CHKERRQ(ierr); 443 PCMGCycleType pcmg_cycle_type = PC_MG_CYCLE_V; 444 { 445 ierr = PCSetType(pc, PCMG); CHKERRQ(ierr); 446 447 // PCMG levels 448 ierr = PCMGSetLevels(pc, num_levels, NULL); CHKERRQ(ierr); 449 for (int i=0; i<num_levels; i++) { 450 // Smoother 451 KSP smoother; 452 PC smoother_pc; 453 ierr = PCMGGetSmoother(pc, i, &smoother); CHKERRQ(ierr); 454 ierr = KSPSetType(smoother, KSPCHEBYSHEV); CHKERRQ(ierr); 455 ierr = KSPChebyshevEstEigSet(smoother, 0, 0.1, 0, 1.1); CHKERRQ(ierr); 456 ierr = KSPChebyshevEstEigSetUseNoisy(smoother, PETSC_TRUE); CHKERRQ(ierr); 457 ierr = KSPSetOperators(smoother, mat_O[i], mat_O[i]); CHKERRQ(ierr); 458 ierr = KSPGetPC(smoother, &smoother_pc); CHKERRQ(ierr); 459 ierr = PCSetType(smoother_pc, PCJACOBI); CHKERRQ(ierr); 460 ierr = PCJacobiSetType(smoother_pc, PC_JACOBI_DIAGONAL); CHKERRQ(ierr); 461 462 // Work vector 463 if (i < num_levels - 1) { 464 ierr = PCMGSetX(pc, i, X[i]); CHKERRQ(ierr); 465 } 466 467 // Level transfers 468 if (i > 0) { 469 // Interpolation 470 ierr = PCMGSetInterpolation(pc, i, mat_pr[i]); CHKERRQ(ierr); 471 } 472 473 // Coarse solve 474 KSP coarse; 475 PC coarse_pc; 476 ierr = PCMGGetCoarseSolve(pc, &coarse); CHKERRQ(ierr); 477 ierr = KSPSetType(coarse, KSPPREONLY); CHKERRQ(ierr); 478 ierr = KSPSetOperators(coarse, mat_coarse, mat_coarse); CHKERRQ(ierr); 479 480 ierr = KSPGetPC(coarse, &coarse_pc); CHKERRQ(ierr); 481 ierr = PCSetType(coarse_pc, PCGAMG); CHKERRQ(ierr); 482 483 ierr = KSPSetOptionsPrefix(coarse, "coarse_"); CHKERRQ(ierr); 484 ierr = PCSetOptionsPrefix(coarse_pc, "coarse_"); CHKERRQ(ierr); 485 ierr = KSPSetFromOptions(coarse); CHKERRQ(ierr); 486 ierr = PCSetFromOptions(coarse_pc); CHKERRQ(ierr); 487 } 488 489 // PCMG options 490 ierr = PCMGSetType(pc, PC_MG_MULTIPLICATIVE); CHKERRQ(ierr); 491 ierr = PCMGSetNumberSmooth(pc, 3); CHKERRQ(ierr); 492 ierr = PCMGSetCycleType(pc, pcmg_cycle_type); CHKERRQ(ierr); 493 } 494 495 // First run, if benchmarking 496 if (benchmark_mode) { 497 ierr = KSPSetTolerances(ksp, 1e-10, PETSC_DEFAULT, PETSC_DEFAULT, 1); 498 CHKERRQ(ierr); 499 ierr = VecZeroEntries(X[fine_level]); CHKERRQ(ierr); 500 my_rt_start = MPI_Wtime(); 501 ierr = KSPSolve(ksp, rhs, X[fine_level]); CHKERRQ(ierr); 502 my_rt = MPI_Wtime() - my_rt_start; 503 ierr = MPI_Allreduce(MPI_IN_PLACE, &my_rt, 1, MPI_DOUBLE, MPI_MIN, comm); 504 CHKERRQ(ierr); 505 // Set maxits based on first iteration timing 506 if (my_rt > 0.02) { 507 ierr = KSPSetTolerances(ksp, 1e-10, PETSC_DEFAULT, PETSC_DEFAULT, 5); 508 CHKERRQ(ierr); 509 } else { 510 ierr = KSPSetTolerances(ksp, 1e-10, PETSC_DEFAULT, PETSC_DEFAULT, 20); 511 CHKERRQ(ierr); 512 } 513 } 514 515 // Timed solve 516 ierr = VecZeroEntries(X[fine_level]); CHKERRQ(ierr); 517 ierr = PetscBarrier((PetscObject)ksp); CHKERRQ(ierr); 518 519 // -- Performance logging 520 ierr = PetscLogStageRegister("Solve Stage", &solve_stage); CHKERRQ(ierr); 521 ierr = PetscLogStagePush(solve_stage); CHKERRQ(ierr); 522 523 // -- Solve 524 my_rt_start = MPI_Wtime(); 525 ierr = KSPSolve(ksp, rhs, X[fine_level]); CHKERRQ(ierr); 526 my_rt = MPI_Wtime() - my_rt_start; 527 528 529 // -- Performance logging 530 ierr = PetscLogStagePop(); 531 532 // Output results 533 { 534 KSPType ksp_type; 535 PCMGType pcmg_type; 536 KSPConvergedReason reason; 537 PetscReal rnorm; 538 PetscInt its; 539 ierr = KSPGetType(ksp, &ksp_type); CHKERRQ(ierr); 540 ierr = KSPGetConvergedReason(ksp, &reason); CHKERRQ(ierr); 541 ierr = KSPGetIterationNumber(ksp, &its); CHKERRQ(ierr); 542 ierr = KSPGetResidualNorm(ksp, &rnorm); CHKERRQ(ierr); 543 ierr = PCMGGetType(pc, &pcmg_type); CHKERRQ(ierr); 544 if (!test_mode || reason < 0 || rnorm > 1e-8) { 545 ierr = PetscPrintf(comm, 546 " KSP:\n" 547 " KSP Type : %s\n" 548 " KSP Convergence : %s\n" 549 " Total KSP Iterations : %D\n" 550 " Final rnorm : %e\n", 551 ksp_type, KSPConvergedReasons[reason], its, 552 (double)rnorm); CHKERRQ(ierr); 553 ierr = PetscPrintf(comm, 554 " PCMG:\n" 555 " PCMG Type : %s\n" 556 " PCMG Cycle Type : %s\n", 557 PCMGTypes[pcmg_type], 558 PCMGCycleTypes[pcmg_cycle_type]); CHKERRQ(ierr); 559 } 560 if (!test_mode) { 561 ierr = PetscPrintf(comm," Performance:\n"); CHKERRQ(ierr); 562 } 563 { 564 PetscReal max_error; 565 ierr = ComputeErrorMax(user_O[fine_level], op_error, X[fine_level], target, 566 &max_error); CHKERRQ(ierr); 567 PetscReal tol = 5e-2; 568 if (!test_mode || max_error > tol) { 569 ierr = MPI_Allreduce(&my_rt, &rt_min, 1, MPI_DOUBLE, MPI_MIN, comm); 570 CHKERRQ(ierr); 571 ierr = MPI_Allreduce(&my_rt, &rt_max, 1, MPI_DOUBLE, MPI_MAX, comm); 572 CHKERRQ(ierr); 573 ierr = PetscPrintf(comm, 574 " Pointwise Error (max) : %e\n" 575 " CG Solve Time : %g (%g) sec\n", 576 (double)max_error, rt_max, rt_min); CHKERRQ(ierr); 577 } 578 } 579 if (benchmark_mode && (!test_mode)) { 580 ierr = PetscPrintf(comm, 581 " DoFs/Sec in CG : %g (%g) million\n", 582 1e-6*g_size[fine_level]*its/rt_max, 583 1e-6*g_size[fine_level]*its/rt_min); 584 CHKERRQ(ierr); 585 } 586 } 587 588 if (write_solution) { 589 PetscViewer vtk_viewer_soln; 590 591 ierr = PetscViewerCreate(comm, &vtk_viewer_soln); CHKERRQ(ierr); 592 ierr = PetscViewerSetType(vtk_viewer_soln, PETSCVIEWERVTK); CHKERRQ(ierr); 593 ierr = PetscViewerFileSetName(vtk_viewer_soln, "solution.vtu"); CHKERRQ(ierr); 594 ierr = VecView(X[fine_level], vtk_viewer_soln); CHKERRQ(ierr); 595 ierr = PetscViewerDestroy(&vtk_viewer_soln); CHKERRQ(ierr); 596 } 597 598 // Cleanup 599 for (int i=0; i<num_levels; i++) { 600 ierr = VecDestroy(&X[i]); CHKERRQ(ierr); 601 ierr = VecDestroy(&X_loc[i]); CHKERRQ(ierr); 602 ierr = VecDestroy(&mult[i]); CHKERRQ(ierr); 603 ierr = VecDestroy(&user_O[i]->Y_loc); CHKERRQ(ierr); 604 ierr = MatDestroy(&mat_O[i]); CHKERRQ(ierr); 605 ierr = PetscFree(user_O[i]); CHKERRQ(ierr); 606 if (i > 0) { 607 ierr = MatDestroy(&mat_pr[i]); CHKERRQ(ierr); 608 ierr = PetscFree(user_pr[i]); CHKERRQ(ierr); 609 } 610 ierr = CeedDataDestroy(i, ceed_data[i]); CHKERRQ(ierr); 611 ierr = DMDestroy(&dm[i]); CHKERRQ(ierr); 612 } 613 ierr = PetscFree(level_degrees); CHKERRQ(ierr); 614 ierr = PetscFree(dm); CHKERRQ(ierr); 615 ierr = PetscFree(X); CHKERRQ(ierr); 616 ierr = PetscFree(X_loc); CHKERRQ(ierr); 617 ierr = PetscFree(mult); CHKERRQ(ierr); 618 ierr = PetscFree(mat_O); CHKERRQ(ierr); 619 ierr = PetscFree(mat_pr); CHKERRQ(ierr); 620 ierr = PetscFree(ceed_data); CHKERRQ(ierr); 621 ierr = PetscFree(user_O); CHKERRQ(ierr); 622 ierr = PetscFree(user_pr); CHKERRQ(ierr); 623 ierr = PetscFree(l_size); CHKERRQ(ierr); 624 ierr = PetscFree(xl_size); CHKERRQ(ierr); 625 ierr = PetscFree(g_size); CHKERRQ(ierr); 626 ierr = VecDestroy(&rhs); CHKERRQ(ierr); 627 ierr = VecDestroy(&rhs_loc); CHKERRQ(ierr); 628 ierr = MatDestroy(&mat_coarse); CHKERRQ(ierr); 629 ierr = KSPDestroy(&ksp); CHKERRQ(ierr); 630 ierr = SNESDestroy(&snes_dummy); CHKERRQ(ierr); 631 ierr = DMDestroy(&dm_orig); CHKERRQ(ierr); 632 CeedVectorDestroy(&target); 633 CeedQFunctionDestroy(&qf_error); 634 CeedQFunctionDestroy(&qf_restrict); 635 CeedQFunctionDestroy(&qf_prolong); 636 CeedOperatorDestroy(&op_error); 637 CeedDestroy(&ceed); 638 return PetscFinalize(); 639 } 640