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: Navier-Stokes 18 // 19 // This example demonstrates a simple usage of libCEED with PETSc to solve a 20 // Navier-Stokes problem. 21 // 22 // The code is intentionally "raw", using only low-level communication 23 // primitives. 24 // 25 // Build with: 26 // 27 // make [PETSC_DIR=</path/to/petsc>] [CEED_DIR=</path/to/libceed>] navierstokes 28 // 29 // Sample runs: 30 // 31 // ./navierstokes -ceed /cpu/self -problem density_current -petscspace_degree 1 32 // ./navierstokes -ceed /gpu/occa -problem advection -petscspace_degree 1 33 // 34 //TESTARGS -ceed {ceed_resource} -test -petscspace_degree 1 35 36 /// @file 37 /// Navier-Stokes example using PETSc 38 39 const char help[] = "Solve Navier-Stokes using PETSc and libCEED\n"; 40 41 #include <petscts.h> 42 #include <petscdmplex.h> 43 #include <ceed.h> 44 #include <stdbool.h> 45 #include <petscsys.h> 46 #include "common.h" 47 #include "advection.h" 48 #include "advection2d.h" 49 #include "densitycurrent.h" 50 51 // Problem Options 52 typedef enum { 53 NS_DENSITY_CURRENT = 0, 54 NS_ADVECTION = 1, 55 NS_ADVECTION2D = 2, 56 } problemType; 57 static const char *const problemTypes[] = { 58 "density_current", 59 "advection", 60 "advection2d", 61 "problemType","NS_",0 62 }; 63 64 typedef enum { 65 STAB_NONE = 0, 66 STAB_SU = 1, // Streamline Upwind 67 STAB_SUPG = 2, // Streamline Upwind Petrov-Galerkin 68 } StabilizationType; 69 static const char *const StabilizationTypes[] = { 70 "NONE", 71 "SU", 72 "SUPG", 73 "StabilizationType", "STAB_", NULL 74 }; 75 76 // Problem specific data 77 typedef struct { 78 CeedInt dim, qdatasize; 79 CeedQFunctionUser setup, ics, apply_rhs, apply_ifunction; 80 PetscErrorCode (*bc)(PetscInt, PetscReal, const PetscReal[], PetscInt, 81 PetscScalar[], void *); 82 const char *setup_loc, *ics_loc, *apply_rhs_loc, *apply_ifunction_loc; 83 const bool non_zero_time; 84 } problemData; 85 86 problemData problemOptions[] = { 87 [NS_DENSITY_CURRENT] = { 88 .dim = 3, 89 .qdatasize = 10, 90 .setup = Setup, 91 .setup_loc = Setup_loc, 92 .ics = ICsDC, 93 .ics_loc = ICsDC_loc, 94 .apply_rhs = DC, 95 .apply_rhs_loc = DC_loc, 96 .apply_ifunction = IFunction_DC, 97 .apply_ifunction_loc = IFunction_DC_loc, 98 .bc = Exact_DC, 99 .non_zero_time = false, 100 }, 101 [NS_ADVECTION] = { 102 .dim = 3, 103 .qdatasize = 10, 104 .setup = Setup, 105 .setup_loc = Setup_loc, 106 .ics = ICsAdvection, 107 .ics_loc = ICsAdvection_loc, 108 .apply_rhs = Advection, 109 .apply_rhs_loc = Advection_loc, 110 .apply_ifunction = IFunction_Advection, 111 .apply_ifunction_loc = IFunction_Advection_loc, 112 .bc = Exact_Advection, 113 .non_zero_time = true, 114 }, 115 [NS_ADVECTION2D] = { 116 .dim = 2, 117 .qdatasize = 5, 118 .setup = Setup2d, 119 .setup_loc = Setup2d_loc, 120 .ics = ICsAdvection2d, 121 .ics_loc = ICsAdvection2d_loc, 122 .apply_rhs = Advection2d, 123 .apply_rhs_loc = Advection2d_loc, 124 .apply_ifunction = IFunction_Advection2d, 125 .apply_ifunction_loc = IFunction_Advection2d_loc, 126 .bc = Exact_Advection2d, 127 .non_zero_time = true, 128 }, 129 }; 130 131 // PETSc user data 132 typedef struct User_ *User; 133 typedef struct Units_ *Units; 134 135 struct User_ { 136 MPI_Comm comm; 137 PetscInt outputfreq; 138 DM dm; 139 DM dmviz; 140 Mat interpviz; 141 Ceed ceed; 142 Units units; 143 CeedVector qceed, qdotceed, gceed; 144 CeedOperator op_rhs, op_ifunction; 145 Vec M; 146 char outputfolder[PETSC_MAX_PATH_LEN]; 147 PetscInt contsteps; 148 }; 149 150 struct Units_ { 151 // fundamental units 152 PetscScalar meter; 153 PetscScalar kilogram; 154 PetscScalar second; 155 PetscScalar Kelvin; 156 // derived units 157 PetscScalar Pascal; 158 PetscScalar JperkgK; 159 PetscScalar mpersquareds; 160 PetscScalar WpermK; 161 PetscScalar kgpercubicm; 162 PetscScalar kgpersquaredms; 163 PetscScalar Joulepercubicm; 164 }; 165 166 typedef struct SimpleBC_ *SimpleBC; 167 struct SimpleBC_ { 168 PetscInt nwall, nslip[3]; 169 PetscInt walls[10], slips[3][10]; 170 }; 171 172 // Essential BC dofs are encoded in closure indices as -(i+1). 173 static PetscInt Involute(PetscInt i) { 174 return i >= 0 ? i : -(i+1); 175 } 176 177 // Utility function to create local CEED restriction 178 static PetscErrorCode CreateRestrictionFromPlex(Ceed ceed, DM dm, CeedInt P, 179 CeedElemRestriction *Erestrict) { 180 181 PetscSection section; 182 PetscInt c, cStart, cEnd, Nelem, Ndof, *erestrict, eoffset, nfields, dim; 183 PetscErrorCode ierr; 184 Vec Uloc; 185 186 PetscFunctionBeginUser; 187 ierr = DMGetDimension(dm, &dim); CHKERRQ(ierr); 188 ierr = DMGetLocalSection(dm,§ion); CHKERRQ(ierr); 189 ierr = PetscSectionGetNumFields(section, &nfields); CHKERRQ(ierr); 190 PetscInt ncomp[nfields], fieldoff[nfields+1]; 191 fieldoff[0] = 0; 192 for (PetscInt f=0; f<nfields; f++) { 193 ierr = PetscSectionGetFieldComponents(section, f, &ncomp[f]); CHKERRQ(ierr); 194 fieldoff[f+1] = fieldoff[f] + ncomp[f]; 195 } 196 197 ierr = DMPlexGetHeightStratum(dm,0,&cStart,&cEnd); CHKERRQ(ierr); 198 Nelem = cEnd - cStart; 199 ierr = PetscMalloc1(Nelem*PetscPowInt(P, dim), &erestrict); CHKERRQ(ierr); 200 for (c=cStart,eoffset=0; c<cEnd; c++) { 201 PetscInt numindices, *indices, nnodes; 202 ierr = DMPlexGetClosureIndices(dm, section, section, c, &numindices, 203 &indices, NULL); CHKERRQ(ierr); 204 if (numindices % fieldoff[nfields]) SETERRQ1(PETSC_COMM_SELF, 205 PETSC_ERR_ARG_INCOMP, "Number of closure indices not compatible with Cell %D", 206 c); 207 nnodes = numindices / fieldoff[nfields]; 208 for (PetscInt i=0; i<nnodes; i++) { 209 // Check that indices are blocked by node and thus can be coalesced as a single field with 210 // fieldoff[nfields] = sum(ncomp) components. 211 for (PetscInt f=0; f<nfields; f++) { 212 for (PetscInt j=0; j<ncomp[f]; j++) { 213 if (Involute(indices[fieldoff[f]*nnodes + i*ncomp[f] + j]) 214 != Involute(indices[i*ncomp[0]]) + fieldoff[f] + j) 215 SETERRQ4(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP, 216 "Cell %D closure indices not interlaced for node %D field %D component %D", 217 c, i, f, j); 218 } 219 } 220 // Essential boundary conditions are encoded as -(loc+1), but we don't care so we decode. 221 PetscInt loc = Involute(indices[i*ncomp[0]]); 222 erestrict[eoffset++] = loc / fieldoff[nfields]; 223 } 224 ierr = DMPlexRestoreClosureIndices(dm, section, section, c, &numindices, 225 &indices, NULL); CHKERRQ(ierr); 226 } 227 if (eoffset != Nelem*PetscPowInt(P, dim)) SETERRQ3(PETSC_COMM_SELF, 228 PETSC_ERR_LIB,"ElemRestriction of size (%D,%D) initialized %D nodes", Nelem, 229 PetscPowInt(P, dim),eoffset); 230 ierr = DMGetLocalVector(dm, &Uloc); CHKERRQ(ierr); 231 ierr = VecGetLocalSize(Uloc, &Ndof); CHKERRQ(ierr); 232 ierr = DMRestoreLocalVector(dm, &Uloc); CHKERRQ(ierr); 233 CeedElemRestrictionCreate(ceed, CEED_INTERLACED, Nelem, PetscPowInt(P, dim), 234 Ndof/fieldoff[nfields], fieldoff[nfields], 235 CEED_MEM_HOST, CEED_COPY_VALUES, erestrict, Erestrict); 236 ierr = PetscFree(erestrict); CHKERRQ(ierr); 237 PetscFunctionReturn(0); 238 } 239 240 static int CreateVectorFromPetscVec(Ceed ceed, Vec p, CeedVector *v) { 241 PetscErrorCode ierr; 242 PetscInt m; 243 244 PetscFunctionBeginUser; 245 ierr = VecGetLocalSize(p, &m); CHKERRQ(ierr); 246 ierr = CeedVectorCreate(ceed, m, v); CHKERRQ(ierr); 247 PetscFunctionReturn(0); 248 } 249 250 static int VectorPlacePetscVec(CeedVector c, Vec p) { 251 PetscErrorCode ierr; 252 PetscInt mceed,mpetsc; 253 PetscScalar *a; 254 255 PetscFunctionBeginUser; 256 ierr = CeedVectorGetLength(c, &mceed); CHKERRQ(ierr); 257 ierr = VecGetLocalSize(p, &mpetsc); CHKERRQ(ierr); 258 if (mceed != mpetsc) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_INCOMP, 259 "Cannot place PETSc Vec of length %D in CeedVector of length %D",mpetsc,mceed); 260 ierr = VecGetArray(p, &a); CHKERRQ(ierr); 261 CeedVectorSetArray(c, CEED_MEM_HOST, CEED_USE_POINTER, a); 262 PetscFunctionReturn(0); 263 } 264 265 static PetscErrorCode DMPlexInsertBoundaryValues_NS(DM dm, 266 PetscBool insertEssential, Vec Qloc, PetscReal time, Vec faceGeomFVM, 267 Vec cellGeomFVM, Vec gradFVM) { 268 PetscErrorCode ierr; 269 Vec Qbc; 270 271 PetscFunctionBegin; 272 ierr = DMGetNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 273 ierr = VecAXPY(Qloc, 1., Qbc); CHKERRQ(ierr); 274 ierr = DMRestoreNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 275 PetscFunctionReturn(0); 276 } 277 278 // This is the RHS of the ODE, given as u_t = G(t,u) 279 // This function takes in a state vector Q and writes into G 280 static PetscErrorCode RHS_NS(TS ts, PetscReal t, Vec Q, Vec G, void *userData) { 281 PetscErrorCode ierr; 282 User user = *(User *)userData; 283 PetscScalar *q, *g; 284 Vec Qloc, Gloc; 285 286 // Global-to-local 287 PetscFunctionBeginUser; 288 ierr = DMGetLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 289 ierr = DMGetLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 290 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 291 ierr = DMGlobalToLocal(user->dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 292 ierr = DMPlexInsertBoundaryValues(user->dm, PETSC_TRUE, Qloc, 0.0, 293 NULL, NULL, NULL); CHKERRQ(ierr); 294 ierr = VecZeroEntries(Gloc); CHKERRQ(ierr); 295 296 // Ceed Vectors 297 ierr = VecGetArrayRead(Qloc, (const PetscScalar **)&q); CHKERRQ(ierr); 298 ierr = VecGetArray(Gloc, &g); CHKERRQ(ierr); 299 CeedVectorSetArray(user->qceed, CEED_MEM_HOST, CEED_USE_POINTER, q); 300 CeedVectorSetArray(user->gceed, CEED_MEM_HOST, CEED_USE_POINTER, g); 301 302 // Apply CEED operator 303 CeedOperatorApply(user->op_rhs, user->qceed, user->gceed, 304 CEED_REQUEST_IMMEDIATE); 305 306 // Restore vectors 307 ierr = VecRestoreArrayRead(Qloc, (const PetscScalar **)&q); CHKERRQ(ierr); 308 ierr = VecRestoreArray(Gloc, &g); CHKERRQ(ierr); 309 310 ierr = VecZeroEntries(G); CHKERRQ(ierr); 311 ierr = DMLocalToGlobal(user->dm, Gloc, ADD_VALUES, G); CHKERRQ(ierr); 312 313 // Inverse of the lumped mass matrix 314 ierr = VecPointwiseMult(G, G, user->M); // M is Minv 315 CHKERRQ(ierr); 316 317 ierr = DMRestoreLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 318 ierr = DMRestoreLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 319 PetscFunctionReturn(0); 320 } 321 322 static PetscErrorCode IFunction_NS(TS ts, PetscReal t, Vec Q, Vec Qdot, Vec G, 323 void *userData) { 324 PetscErrorCode ierr; 325 User user = *(User *)userData; 326 const PetscScalar *q, *qdot; 327 PetscScalar *g; 328 Vec Qloc, Qdotloc, Gloc; 329 330 // Global-to-local 331 PetscFunctionBeginUser; 332 ierr = DMGetLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 333 ierr = DMGetLocalVector(user->dm, &Qdotloc); CHKERRQ(ierr); 334 ierr = DMGetLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 335 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 336 ierr = DMGlobalToLocal(user->dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 337 ierr = DMPlexInsertBoundaryValues(user->dm, PETSC_TRUE, Qloc, 0.0, 338 NULL, NULL, NULL); CHKERRQ(ierr); 339 ierr = VecZeroEntries(Qdotloc); CHKERRQ(ierr); 340 ierr = DMGlobalToLocal(user->dm, Qdot, INSERT_VALUES, Qdotloc); CHKERRQ(ierr); 341 ierr = VecZeroEntries(Gloc); CHKERRQ(ierr); 342 343 // Ceed Vectors 344 ierr = VecGetArrayRead(Qloc, &q); CHKERRQ(ierr); 345 ierr = VecGetArrayRead(Qdotloc, &qdot); CHKERRQ(ierr); 346 ierr = VecGetArray(Gloc, &g); CHKERRQ(ierr); 347 CeedVectorSetArray(user->qceed, CEED_MEM_HOST, CEED_USE_POINTER, 348 (PetscScalar *)q); 349 CeedVectorSetArray(user->qdotceed, CEED_MEM_HOST, CEED_USE_POINTER, 350 (PetscScalar *)qdot); 351 CeedVectorSetArray(user->gceed, CEED_MEM_HOST, CEED_USE_POINTER, g); 352 353 // Apply CEED operator 354 CeedOperatorApply(user->op_ifunction, user->qceed, user->gceed, 355 CEED_REQUEST_IMMEDIATE); 356 357 // Restore vectors 358 ierr = VecRestoreArrayRead(Qloc, &q); CHKERRQ(ierr); 359 ierr = VecRestoreArrayRead(Qdotloc, &qdot); CHKERRQ(ierr); 360 ierr = VecRestoreArray(Gloc, &g); CHKERRQ(ierr); 361 362 ierr = VecZeroEntries(G); CHKERRQ(ierr); 363 ierr = DMLocalToGlobal(user->dm, Gloc, ADD_VALUES, G); CHKERRQ(ierr); 364 365 ierr = DMRestoreLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 366 ierr = DMRestoreLocalVector(user->dm, &Qdotloc); CHKERRQ(ierr); 367 ierr = DMRestoreLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 368 PetscFunctionReturn(0); 369 } 370 371 // User provided TS Monitor 372 static PetscErrorCode TSMonitor_NS(TS ts, PetscInt stepno, PetscReal time, 373 Vec Q, void *ctx) { 374 User user = ctx; 375 Vec Qloc; 376 char filepath[PETSC_MAX_PATH_LEN]; 377 PetscViewer viewer; 378 PetscErrorCode ierr; 379 380 // Set up output 381 PetscFunctionBeginUser; 382 // Print every 'outputfreq' steps 383 if (stepno % user->outputfreq != 0) 384 PetscFunctionReturn(0); 385 ierr = DMGetLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 386 ierr = PetscObjectSetName((PetscObject)Qloc, "StateVec"); CHKERRQ(ierr); 387 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 388 ierr = DMGlobalToLocal(user->dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 389 390 // Output 391 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-%03D.vtu", 392 user->outputfolder, stepno + user->contsteps); 393 CHKERRQ(ierr); 394 ierr = PetscViewerVTKOpen(PetscObjectComm((PetscObject)Q), filepath, 395 FILE_MODE_WRITE, &viewer); CHKERRQ(ierr); 396 ierr = VecView(Qloc, viewer); CHKERRQ(ierr); 397 if (user->dmviz) { 398 Vec Qrefined, Qrefined_loc; 399 char filepath_refined[PETSC_MAX_PATH_LEN]; 400 PetscViewer viewer_refined; 401 402 ierr = DMGetGlobalVector(user->dmviz, &Qrefined); CHKERRQ(ierr); 403 ierr = DMGetLocalVector(user->dmviz, &Qrefined_loc); CHKERRQ(ierr); 404 ierr = PetscObjectSetName((PetscObject)Qrefined_loc, "Refined"); 405 CHKERRQ(ierr); 406 ierr = MatInterpolate(user->interpviz, Q, Qrefined); CHKERRQ(ierr); 407 ierr = VecZeroEntries(Qrefined_loc); CHKERRQ(ierr); 408 ierr = DMGlobalToLocal(user->dmviz, Qrefined, INSERT_VALUES, Qrefined_loc); 409 CHKERRQ(ierr); 410 ierr = PetscSNPrintf(filepath_refined, sizeof filepath_refined, 411 "%s/nsrefined-%03D.vtu", 412 user->outputfolder, stepno + user->contsteps); 413 CHKERRQ(ierr); 414 ierr = PetscViewerVTKOpen(PetscObjectComm((PetscObject)Qrefined), 415 filepath_refined, 416 FILE_MODE_WRITE, &viewer_refined); CHKERRQ(ierr); 417 ierr = VecView(Qrefined_loc, viewer_refined); CHKERRQ(ierr); 418 ierr = DMRestoreLocalVector(user->dmviz, &Qrefined_loc); CHKERRQ(ierr); 419 ierr = DMRestoreGlobalVector(user->dmviz, &Qrefined); CHKERRQ(ierr); 420 ierr = PetscViewerDestroy(&viewer_refined); CHKERRQ(ierr); 421 } 422 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 423 ierr = DMRestoreLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 424 425 // Save data in a binary file for continuation of simulations 426 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-solution.bin", 427 user->outputfolder); CHKERRQ(ierr); 428 ierr = PetscViewerBinaryOpen(user->comm, filepath, FILE_MODE_WRITE, &viewer); 429 CHKERRQ(ierr); 430 ierr = VecView(Q, viewer); CHKERRQ(ierr); 431 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 432 433 // Save time stamp 434 // Dimensionalize time back 435 time /= user->units->second; 436 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-time.bin", 437 user->outputfolder); CHKERRQ(ierr); 438 ierr = PetscViewerBinaryOpen(user->comm, filepath, FILE_MODE_WRITE, &viewer); 439 CHKERRQ(ierr); 440 #if PETSC_VERSION_GE(3,13,0) 441 ierr = PetscViewerBinaryWrite(viewer, &time, 1, PETSC_REAL); 442 #else 443 ierr = PetscViewerBinaryWrite(viewer, &time, 1, PETSC_REAL, true); 444 #endif 445 CHKERRQ(ierr); 446 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 447 448 PetscFunctionReturn(0); 449 } 450 451 static PetscErrorCode ICs_PetscMultiplicity(CeedOperator op_ics, 452 CeedVector xcorners, CeedVector q0ceed, DM dm, Vec Qloc, Vec Q, 453 CeedElemRestriction restrictq, SetupContext ctxSetup, CeedScalar time) { 454 PetscErrorCode ierr; 455 CeedVector multlvec; 456 Vec Multiplicity, MultiplicityLoc; 457 458 ctxSetup->time = time; 459 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 460 ierr = VectorPlacePetscVec(q0ceed, Qloc); CHKERRQ(ierr); 461 CeedOperatorApply(op_ics, xcorners, q0ceed, CEED_REQUEST_IMMEDIATE); 462 ierr = VecZeroEntries(Q); CHKERRQ(ierr); 463 ierr = DMLocalToGlobal(dm, Qloc, ADD_VALUES, Q); CHKERRQ(ierr); 464 465 // Fix multiplicity for output of ICs 466 ierr = DMGetLocalVector(dm, &MultiplicityLoc); CHKERRQ(ierr); 467 CeedElemRestrictionCreateVector(restrictq, &multlvec, NULL); 468 ierr = VectorPlacePetscVec(multlvec, MultiplicityLoc); CHKERRQ(ierr); 469 CeedElemRestrictionGetMultiplicity(restrictq, multlvec); 470 CeedVectorDestroy(&multlvec); 471 ierr = DMGetGlobalVector(dm, &Multiplicity); CHKERRQ(ierr); 472 ierr = VecZeroEntries(Multiplicity); CHKERRQ(ierr); 473 ierr = DMLocalToGlobal(dm, MultiplicityLoc, ADD_VALUES, Multiplicity); 474 CHKERRQ(ierr); 475 ierr = VecPointwiseDivide(Q, Q, Multiplicity); CHKERRQ(ierr); 476 ierr = VecPointwiseDivide(Qloc, Qloc, MultiplicityLoc); CHKERRQ(ierr); 477 ierr = DMRestoreLocalVector(dm, &MultiplicityLoc); CHKERRQ(ierr); 478 ierr = DMRestoreGlobalVector(dm, &Multiplicity); CHKERRQ(ierr); 479 480 PetscFunctionReturn(0); 481 } 482 483 static PetscErrorCode ComputeLumpedMassMatrix(Ceed ceed, DM dm, 484 CeedElemRestriction restrictq, CeedBasis basisq, 485 CeedElemRestriction restrictqdi, CeedVector qdata, Vec M) { 486 PetscErrorCode ierr; 487 CeedQFunction qf_mass; 488 CeedOperator op_mass; 489 CeedVector mceed; 490 Vec Mloc; 491 CeedInt ncompq, qdatasize; 492 493 PetscFunctionBeginUser; 494 CeedElemRestrictionGetNumComponents(restrictq, &ncompq); 495 CeedElemRestrictionGetNumComponents(restrictqdi, &qdatasize); 496 // Create the Q-function that defines the action of the mass operator 497 CeedQFunctionCreateInterior(ceed, 1, Mass, Mass_loc, &qf_mass); 498 CeedQFunctionAddInput(qf_mass, "q", ncompq, CEED_EVAL_INTERP); 499 CeedQFunctionAddInput(qf_mass, "qdata", qdatasize, CEED_EVAL_NONE); 500 CeedQFunctionAddOutput(qf_mass, "v", ncompq, CEED_EVAL_INTERP); 501 502 // Create the mass operator 503 CeedOperatorCreate(ceed, qf_mass, NULL, NULL, &op_mass); 504 CeedOperatorSetField(op_mass, "q", restrictq, basisq, CEED_VECTOR_ACTIVE); 505 CeedOperatorSetField(op_mass, "qdata", restrictqdi, 506 CEED_BASIS_COLLOCATED, qdata); 507 CeedOperatorSetField(op_mass, "v", restrictq, basisq, CEED_VECTOR_ACTIVE); 508 509 ierr = DMGetLocalVector(dm, &Mloc); CHKERRQ(ierr); 510 ierr = VecZeroEntries(Mloc); CHKERRQ(ierr); 511 CeedElemRestrictionCreateVector(restrictq, &mceed, NULL); 512 ierr = VectorPlacePetscVec(mceed, Mloc); CHKERRQ(ierr); 513 514 { 515 // Compute a lumped mass matrix 516 CeedVector onesvec; 517 CeedElemRestrictionCreateVector(restrictq, &onesvec, NULL); 518 CeedVectorSetValue(onesvec, 1.0); 519 CeedOperatorApply(op_mass, onesvec, mceed, CEED_REQUEST_IMMEDIATE); 520 CeedVectorDestroy(&onesvec); 521 CeedOperatorDestroy(&op_mass); 522 CeedVectorDestroy(&mceed); 523 } 524 CeedQFunctionDestroy(&qf_mass); 525 526 ierr = VecZeroEntries(M); CHKERRQ(ierr); 527 ierr = DMLocalToGlobal(dm, Mloc, ADD_VALUES, M); CHKERRQ(ierr); 528 ierr = DMRestoreLocalVector(dm, &Mloc); CHKERRQ(ierr); 529 530 // Invert diagonally lumped mass vector for RHS function 531 ierr = VecReciprocal(M); CHKERRQ(ierr); 532 PetscFunctionReturn(0); 533 } 534 535 PetscErrorCode SetUpDM(DM dm, problemData *problem, const char *prefix, 536 SimpleBC bc, void *ctxSetup, PetscInt *degree) { 537 PetscErrorCode ierr; 538 539 PetscFunctionBeginUser; 540 { 541 // Configure the finite element space and boundary conditions 542 PetscFE fe; 543 PetscSpace fespace; 544 PetscInt ncompq = 5; 545 ierr = PetscFECreateDefault(PETSC_COMM_SELF,problem->dim, ncompq, 546 PETSC_FALSE, prefix, PETSC_DETERMINE, &fe); 547 CHKERRQ(ierr); 548 ierr = PetscObjectSetName((PetscObject)fe, "Q"); CHKERRQ(ierr); 549 ierr = DMAddField(dm,NULL,(PetscObject)fe); CHKERRQ(ierr); 550 ierr = DMCreateDS(dm); CHKERRQ(ierr); 551 /* Wall boundary conditions are zero velocity and zero flux for density and energy */ 552 { 553 PetscInt comps[3] = {1, 2, 3}; 554 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "wall", "Face Sets", 0, 555 3, comps, (void(*)(void))problem->bc, 556 bc->nwall, bc->walls, ctxSetup); CHKERRQ(ierr); 557 } 558 { 559 PetscInt comps[1] = {1}; 560 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "slipx", "Face Sets", 0, 561 1, comps, (void(*)(void))NULL, bc->nslip[0], 562 bc->slips[0], ctxSetup); CHKERRQ(ierr); 563 comps[0] = 2; 564 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "slipy", "Face Sets", 0, 565 1, comps, (void(*)(void))NULL, bc->nslip[1], 566 bc->slips[1], ctxSetup); CHKERRQ(ierr); 567 comps[0] = 3; 568 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "slipz", "Face Sets", 0, 569 1, comps, (void(*)(void))NULL, bc->nslip[2], 570 bc->slips[2], ctxSetup); CHKERRQ(ierr); 571 } 572 ierr = DMPlexSetClosurePermutationTensor(dm,PETSC_DETERMINE,NULL); 573 CHKERRQ(ierr); 574 ierr = PetscFEGetBasisSpace(fe, &fespace); CHKERRQ(ierr); 575 if (degree) { 576 ierr = PetscSpaceGetDegree(fespace, degree, NULL); CHKERRQ(ierr); 577 if (*degree < 1) SETERRQ1(PetscObjectComm((PetscObject)dm), 578 PETSC_ERR_ARG_OUTOFRANGE, 579 "Degree %D; must specify -petscspace_degree 1 (or greater)", *degree); 580 } 581 ierr = PetscFEDestroy(&fe); CHKERRQ(ierr); 582 } 583 { 584 // Empty name for conserved field (because there is only one field) 585 PetscSection section; 586 ierr = DMGetLocalSection(dm, §ion); CHKERRQ(ierr); 587 ierr = PetscSectionSetFieldName(section, 0, ""); CHKERRQ(ierr); 588 ierr = PetscSectionSetComponentName(section, 0, 0, "Density"); 589 CHKERRQ(ierr); 590 ierr = PetscSectionSetComponentName(section, 0, 1, "MomentumX"); 591 CHKERRQ(ierr); 592 ierr = PetscSectionSetComponentName(section, 0, 2, "MomentumY"); 593 CHKERRQ(ierr); 594 ierr = PetscSectionSetComponentName(section, 0, 3, "MomentumZ"); 595 CHKERRQ(ierr); 596 ierr = PetscSectionSetComponentName(section, 0, 4, "EnergyDensity"); 597 CHKERRQ(ierr); 598 } 599 PetscFunctionReturn(0); 600 } 601 602 int main(int argc, char **argv) { 603 PetscInt ierr; 604 MPI_Comm comm; 605 DM dm, dmcoord, dmviz, dmvizfine; 606 Mat interpviz; 607 TS ts; 608 TSAdapt adapt; 609 User user; 610 Units units; 611 char ceedresource[4096] = "/cpu/self"; 612 PetscInt cStart, cEnd, localNelem, lnodes, steps; 613 const PetscInt ncompq = 5; 614 PetscMPIInt rank; 615 PetscScalar ftime; 616 Vec Q, Qloc, Xloc; 617 Ceed ceed; 618 CeedInt numP, numQ; 619 CeedVector xcorners, qdata, q0ceed; 620 CeedBasis basisx, basisxc, basisq; 621 CeedElemRestriction restrictx, restrictxcoord, restrictq, restrictqdi; 622 CeedQFunction qf_setup, qf_ics, qf_rhs, qf_ifunction; 623 CeedOperator op_setup, op_ics; 624 CeedScalar Rd; 625 PetscScalar WpermK, Pascal, JperkgK, mpersquareds, kgpercubicm, 626 kgpersquaredms, Joulepercubicm; 627 problemType problemChoice; 628 problemData *problem = NULL; 629 StabilizationType stab; 630 PetscBool test, implicit; 631 PetscInt viz_refine = 0; 632 struct SimpleBC_ bc = { 633 .nwall = 6, 634 .walls = {1,2,3,4,5,6}, 635 }; 636 double start, cpu_time_used; 637 638 // Create the libCEED contexts 639 PetscScalar meter = 1e-2; // 1 meter in scaled length units 640 PetscScalar second = 1e-2; // 1 second in scaled time units 641 PetscScalar kilogram = 1e-6; // 1 kilogram in scaled mass units 642 PetscScalar Kelvin = 1; // 1 Kelvin in scaled temperature units 643 CeedScalar theta0 = 300.; // K 644 CeedScalar thetaC = -15.; // K 645 CeedScalar P0 = 1.e5; // Pa 646 CeedScalar N = 0.01; // 1/s 647 CeedScalar cv = 717.; // J/(kg K) 648 CeedScalar cp = 1004.; // J/(kg K) 649 CeedScalar g = 9.81; // m/s^2 650 CeedScalar lambda = -2./3.; // - 651 CeedScalar mu = 75.; // Pa s, dynamic viscosity 652 // mu = 75 is not physical for air, but is good for numerical stability 653 CeedScalar k = 0.02638; // W/(m K) 654 CeedScalar CtauS = 0.; // dimensionless 655 CeedScalar strong_form = 0.; // [0,1] 656 PetscScalar lx = 8000.; // m 657 PetscScalar ly = 8000.; // m 658 PetscScalar lz = 4000.; // m 659 CeedScalar rc = 1000.; // m (Radius of bubble) 660 PetscScalar resx = 1000.; // m (resolution in x) 661 PetscScalar resy = 1000.; // m (resolution in y) 662 PetscScalar resz = 1000.; // m (resolution in z) 663 PetscInt outputfreq = 10; // - 664 PetscInt contsteps = 0; // - 665 PetscInt degree; 666 PetscInt qextra = 2; // - 667 DMBoundaryType periodicity[] = {DM_BOUNDARY_NONE, DM_BOUNDARY_NONE, 668 DM_BOUNDARY_NONE 669 }; 670 PetscReal center[3], dc_axis[3] = {0, 0, 0}; 671 672 ierr = PetscInitialize(&argc, &argv, NULL, help); 673 if (ierr) return ierr; 674 675 // Allocate PETSc context 676 ierr = PetscCalloc1(1, &user); CHKERRQ(ierr); 677 ierr = PetscMalloc1(1, &units); CHKERRQ(ierr); 678 679 // Parse command line options 680 comm = PETSC_COMM_WORLD; 681 ierr = PetscOptionsBegin(comm, NULL, "Navier-Stokes in PETSc with libCEED", 682 NULL); CHKERRQ(ierr); 683 ierr = PetscOptionsString("-ceed", "CEED resource specifier", 684 NULL, ceedresource, ceedresource, 685 sizeof(ceedresource), NULL); CHKERRQ(ierr); 686 ierr = PetscOptionsBool("-test", "Run in test mode", 687 NULL, test=PETSC_FALSE, &test, NULL); CHKERRQ(ierr); 688 problemChoice = NS_DENSITY_CURRENT; 689 ierr = PetscOptionsEnum("-problem", "Problem to solve", NULL, 690 problemTypes, (PetscEnum)problemChoice, 691 (PetscEnum *)&problemChoice, NULL); CHKERRQ(ierr); 692 problem = &problemOptions[problemChoice]; 693 ierr = PetscOptionsEnum("-stab", "Stabilization method", NULL, 694 StabilizationTypes, (PetscEnum)(stab = STAB_NONE), 695 (PetscEnum *)&stab, NULL); CHKERRQ(ierr); 696 ierr = PetscOptionsBool("-implicit", "Use implicit (IFunction) formulation", 697 NULL, implicit=PETSC_FALSE, &implicit, NULL); 698 CHKERRQ(ierr); 699 { 700 PetscInt len; 701 PetscBool flg; 702 ierr = PetscOptionsIntArray("-bc_wall", 703 "Use wall boundary conditions on this list of faces", 704 NULL, bc.walls, 705 (len = sizeof(bc.walls) / sizeof(bc.walls[0]), 706 &len), &flg); CHKERRQ(ierr); 707 if (flg) bc.nwall = len; 708 for (PetscInt j=0; j<3; j++) { 709 const char *flags[3] = {"-bc_slip_x", "-bc_slip_y", "-bc_slip_z"}; 710 ierr = PetscOptionsIntArray(flags[j], 711 "Use slip boundary conditions on this list of faces", 712 NULL, bc.slips[j], 713 (len = sizeof(bc.slips[j]) / sizeof(bc.slips[j][0]), 714 &len), &flg); 715 CHKERRQ(ierr); 716 if (flg) bc.nslip[j] = len; 717 } 718 } 719 ierr = PetscOptionsInt("-viz_refine", 720 "Regular refinement levels for visualization", 721 NULL, viz_refine, &viz_refine, NULL); 722 CHKERRQ(ierr); 723 ierr = PetscOptionsScalar("-units_meter", "1 meter in scaled length units", 724 NULL, meter, &meter, NULL); CHKERRQ(ierr); 725 meter = fabs(meter); 726 ierr = PetscOptionsScalar("-units_second","1 second in scaled time units", 727 NULL, second, &second, NULL); CHKERRQ(ierr); 728 second = fabs(second); 729 ierr = PetscOptionsScalar("-units_kilogram","1 kilogram in scaled mass units", 730 NULL, kilogram, &kilogram, NULL); CHKERRQ(ierr); 731 kilogram = fabs(kilogram); 732 ierr = PetscOptionsScalar("-units_Kelvin", 733 "1 Kelvin in scaled temperature units", 734 NULL, Kelvin, &Kelvin, NULL); CHKERRQ(ierr); 735 Kelvin = fabs(Kelvin); 736 ierr = PetscOptionsScalar("-theta0", "Reference potential temperature", 737 NULL, theta0, &theta0, NULL); CHKERRQ(ierr); 738 ierr = PetscOptionsScalar("-thetaC", "Perturbation of potential temperature", 739 NULL, thetaC, &thetaC, NULL); CHKERRQ(ierr); 740 ierr = PetscOptionsScalar("-P0", "Atmospheric pressure", 741 NULL, P0, &P0, NULL); CHKERRQ(ierr); 742 ierr = PetscOptionsScalar("-N", "Brunt-Vaisala frequency", 743 NULL, N, &N, NULL); CHKERRQ(ierr); 744 ierr = PetscOptionsScalar("-cv", "Heat capacity at constant volume", 745 NULL, cv, &cv, NULL); CHKERRQ(ierr); 746 ierr = PetscOptionsScalar("-cp", "Heat capacity at constant pressure", 747 NULL, cp, &cp, NULL); CHKERRQ(ierr); 748 ierr = PetscOptionsScalar("-g", "Gravitational acceleration", 749 NULL, g, &g, NULL); CHKERRQ(ierr); 750 ierr = PetscOptionsScalar("-lambda", 751 "Stokes hypothesis second viscosity coefficient", 752 NULL, lambda, &lambda, NULL); CHKERRQ(ierr); 753 ierr = PetscOptionsScalar("-mu", "Shear dynamic viscosity coefficient", 754 NULL, mu, &mu, NULL); CHKERRQ(ierr); 755 ierr = PetscOptionsScalar("-k", "Thermal conductivity", 756 NULL, k, &k, NULL); CHKERRQ(ierr); 757 ierr = PetscOptionsScalar("-CtauS", 758 "Scale coefficient for tau (nondimensional)", 759 NULL, CtauS, &CtauS, NULL); CHKERRQ(ierr); 760 ierr = PetscOptionsScalar("-strong_form", 761 "Strong (1) or weak/integrated by parts (0) advection residual", 762 NULL, strong_form, &strong_form, NULL); 763 CHKERRQ(ierr); 764 ierr = PetscOptionsScalar("-lx", "Length scale in x direction", 765 NULL, lx, &lx, NULL); CHKERRQ(ierr); 766 ierr = PetscOptionsScalar("-ly", "Length scale in y direction", 767 NULL, ly, &ly, NULL); CHKERRQ(ierr); 768 ierr = PetscOptionsScalar("-lz", "Length scale in z direction", 769 NULL, lz, &lz, NULL); CHKERRQ(ierr); 770 ierr = PetscOptionsScalar("-rc", "Characteristic radius of thermal bubble", 771 NULL, rc, &rc, NULL); CHKERRQ(ierr); 772 ierr = PetscOptionsScalar("-resx","Target resolution in x", 773 NULL, resx, &resx, NULL); CHKERRQ(ierr); 774 ierr = PetscOptionsScalar("-resy","Target resolution in y", 775 NULL, resy, &resy, NULL); CHKERRQ(ierr); 776 ierr = PetscOptionsScalar("-resz","Target resolution in z", 777 NULL, resz, &resz, NULL); CHKERRQ(ierr); 778 PetscInt n = problem->dim; 779 ierr = PetscOptionsEnumArray("-periodicity", "Periodicity per direction", 780 NULL, DMBoundaryTypes, (PetscEnum *)periodicity, 781 &n, NULL); CHKERRQ(ierr); 782 n = problem->dim; 783 center[0] = 0.5 * lx; 784 center[1] = 0.5 * ly; 785 center[2] = 0.5 * lz; 786 ierr = PetscOptionsRealArray("-center", "Location of bubble center", 787 NULL, center, &n, NULL); CHKERRQ(ierr); 788 n = problem->dim; 789 ierr = PetscOptionsRealArray("-dc_axis", 790 "Axis of density current cylindrical anomaly, or {0,0,0} for spherically symmetric", 791 NULL, dc_axis, &n, NULL); CHKERRQ(ierr); 792 { 793 PetscReal norm = PetscSqrtReal(PetscSqr(dc_axis[0]) + 794 PetscSqr(dc_axis[1]) + PetscSqr(dc_axis[2])); 795 if (norm > 0) { 796 for (int i=0; i<3; i++) dc_axis[i] /= norm; 797 } 798 } 799 ierr = PetscOptionsInt("-output_freq", 800 "Frequency of output, in number of steps", 801 NULL, outputfreq, &outputfreq, NULL); CHKERRQ(ierr); 802 ierr = PetscOptionsInt("-continue", "Continue from previous solution", 803 NULL, contsteps, &contsteps, NULL); CHKERRQ(ierr); 804 ierr = PetscOptionsInt("-qextra", "Number of extra quadrature points", 805 NULL, qextra, &qextra, NULL); CHKERRQ(ierr); 806 PetscStrncpy(user->outputfolder, ".", 2); 807 ierr = PetscOptionsString("-of", "Output folder", 808 NULL, user->outputfolder, user->outputfolder, 809 sizeof(user->outputfolder), NULL); CHKERRQ(ierr); 810 ierr = PetscOptionsEnd(); CHKERRQ(ierr); 811 812 // Define derived units 813 Pascal = kilogram / (meter * PetscSqr(second)); 814 JperkgK = PetscSqr(meter) / (PetscSqr(second) * Kelvin); 815 mpersquareds = meter / PetscSqr(second); 816 WpermK = kilogram * meter / (pow(second,3) * Kelvin); 817 kgpercubicm = kilogram / pow(meter,3); 818 kgpersquaredms = kilogram / (PetscSqr(meter) * second); 819 Joulepercubicm = kilogram / (meter * PetscSqr(second)); 820 821 // Scale variables to desired units 822 theta0 *= Kelvin; 823 thetaC *= Kelvin; 824 P0 *= Pascal; 825 N *= (1./second); 826 cv *= JperkgK; 827 cp *= JperkgK; 828 Rd = cp - cv; 829 g *= mpersquareds; 830 mu *= Pascal * second; 831 k *= WpermK; 832 lx = fabs(lx) * meter; 833 ly = fabs(ly) * meter; 834 lz = fabs(lz) * meter; 835 rc = fabs(rc) * meter; 836 resx = fabs(resx) * meter; 837 resy = fabs(resy) * meter; 838 resz = fabs(resz) * meter; 839 for (int i=0; i<3; i++) center[i] *= meter; 840 841 const CeedInt dim = problem->dim, ncompx = problem->dim, 842 qdatasize = problem->qdatasize; 843 // Set up the libCEED context 844 struct SetupContext_ ctxSetup = { 845 .theta0 = theta0, 846 .thetaC = thetaC, 847 .P0 = P0, 848 .N = N, 849 .cv = cv, 850 .cp = cp, 851 .Rd = Rd, 852 .g = g, 853 .rc = rc, 854 .lx = lx, 855 .ly = ly, 856 .lz = lz, 857 .periodicity0 = periodicity[0], 858 .periodicity1 = periodicity[1], 859 .periodicity2 = periodicity[2], 860 .center[0] = center[0], 861 .center[1] = center[1], 862 .center[2] = center[2], 863 .dc_axis[0] = dc_axis[0], 864 .dc_axis[1] = dc_axis[1], 865 .dc_axis[2] = dc_axis[2], 866 .time = 0, 867 }; 868 869 { 870 const PetscReal scale[3] = {lx, ly, lz}; 871 ierr = DMPlexCreateBoxMesh(comm, dim, PETSC_FALSE, NULL, NULL, scale, 872 periodicity, PETSC_TRUE, &dm); 873 CHKERRQ(ierr); 874 } 875 if (1) { 876 DM dmDist = NULL; 877 PetscPartitioner part; 878 879 ierr = DMPlexGetPartitioner(dm, &part); CHKERRQ(ierr); 880 ierr = PetscPartitionerSetFromOptions(part); CHKERRQ(ierr); 881 ierr = DMPlexDistribute(dm, 0, NULL, &dmDist); CHKERRQ(ierr); 882 if (dmDist) { 883 ierr = DMDestroy(&dm); CHKERRQ(ierr); 884 dm = dmDist; 885 } 886 } 887 ierr = DMViewFromOptions(dm, NULL, "-dm_view"); CHKERRQ(ierr); 888 889 ierr = DMLocalizeCoordinates(dm); CHKERRQ(ierr); 890 ierr = DMSetFromOptions(dm); CHKERRQ(ierr); 891 ierr = SetUpDM(dm, problem, NULL, &bc, &ctxSetup, °ree); CHKERRQ(ierr); 892 if (!test) { 893 ierr = PetscPrintf(PETSC_COMM_WORLD, 894 "Degree of FEM Space: %D\n", 895 (PetscInt)degree); CHKERRQ(ierr); 896 } 897 dmviz = NULL; 898 interpviz = NULL; 899 if (viz_refine) { 900 ierr = DMPlexSetRefinementUniform(dm, PETSC_TRUE); CHKERRQ(ierr); 901 dmvizfine = NULL; 902 ierr = DMRefine(dm, MPI_COMM_NULL, &dmviz); CHKERRQ(ierr); 903 ierr = DMSetCoarseDM(dmviz, dm); CHKERRQ(ierr); 904 ierr = PetscOptionsSetValue(NULL,"-viz_petscspace_degree","1"); 905 CHKERRQ(ierr); 906 ierr = SetUpDM(dmviz, problem, "viz_", &bc, &ctxSetup, NULL); CHKERRQ(ierr); 907 ierr = DMCreateInterpolation(dm, dmviz, &interpviz, NULL); CHKERRQ(ierr); 908 for (PetscInt i = 1; i < viz_refine; i++) { 909 ierr = DMRefine(dmviz, MPI_COMM_NULL, &dmvizfine); CHKERRQ(ierr); 910 if (dmvizfine) { 911 ierr = DMSetCoarseDM(dmvizfine, dmviz); CHKERRQ(ierr); 912 ierr = PetscOptionsSetValue(NULL,"-viz_petscspace_degree","1"); 913 CHKERRQ(ierr); 914 ierr = SetUpDM(dmvizfine, problem, "viz_", &bc, &ctxSetup, NULL); CHKERRQ(ierr); 915 ierr = DMCreateInterpolation(dmviz, dmvizfine, &interpviz, NULL); CHKERRQ(ierr); 916 ierr = DMDestroy(&dmviz); CHKERRQ(ierr); 917 dmviz = dmvizfine; 918 } 919 } 920 } 921 ierr = DMCreateGlobalVector(dm, &Q); CHKERRQ(ierr); 922 ierr = DMGetLocalVector(dm, &Qloc); CHKERRQ(ierr); 923 ierr = VecGetSize(Qloc, &lnodes); CHKERRQ(ierr); 924 lnodes /= ncompq; 925 926 { 927 // Print grid information 928 CeedInt gdofs, odofs; 929 int comm_size; 930 char box_faces_str[PETSC_MAX_PATH_LEN] = "NONE"; 931 ierr = VecGetSize(Q, &gdofs); CHKERRQ(ierr); 932 ierr = VecGetLocalSize(Q, &odofs); CHKERRQ(ierr); 933 ierr = MPI_Comm_size(comm, &comm_size); CHKERRQ(ierr); 934 ierr = PetscOptionsGetString(NULL, NULL, "-dm_plex_box_faces", box_faces_str, 935 sizeof(box_faces_str), NULL); CHKERRQ(ierr); 936 if (!test) { 937 ierr = PetscPrintf(comm, "Global FEM dofs: %D (%D owned) on %d ranks\n", 938 gdofs, odofs, comm_size); CHKERRQ(ierr); 939 ierr = PetscPrintf(comm, "Local FEM nodes: %D\n", lnodes); CHKERRQ(ierr); 940 ierr = PetscPrintf(comm, "dm_plex_box_faces: %s\n", box_faces_str); 941 CHKERRQ(ierr); 942 } 943 944 } 945 946 // Set up global mass vector 947 ierr = VecDuplicate(Q,&user->M); CHKERRQ(ierr); 948 949 // Set up CEED 950 // CEED Bases 951 CeedInit(ceedresource, &ceed); 952 numP = degree + 1; 953 numQ = numP + qextra; 954 CeedBasisCreateTensorH1Lagrange(ceed, dim, ncompq, numP, numQ, CEED_GAUSS, 955 &basisq); 956 CeedBasisCreateTensorH1Lagrange(ceed, dim, ncompx, 2, numQ, CEED_GAUSS, 957 &basisx); 958 CeedBasisCreateTensorH1Lagrange(ceed, dim, ncompx, 2, numP, 959 CEED_GAUSS_LOBATTO, &basisxc); 960 961 ierr = DMGetCoordinateDM(dm, &dmcoord); CHKERRQ(ierr); 962 ierr = DMPlexSetClosurePermutationTensor(dmcoord, PETSC_DETERMINE, NULL); 963 CHKERRQ(ierr); 964 965 // CEED Restrictions 966 ierr = CreateRestrictionFromPlex(ceed, dm, degree+1, &restrictq); 967 CHKERRQ(ierr); 968 ierr = CreateRestrictionFromPlex(ceed, dmcoord, 2, &restrictx); CHKERRQ(ierr); 969 DMPlexGetHeightStratum(dm, 0, &cStart, &cEnd); CHKERRQ(ierr); 970 localNelem = cEnd - cStart; 971 CeedInt numQdim = CeedIntPow(numQ, dim); 972 CeedElemRestrictionCreateStrided(ceed, localNelem, numQdim, 973 localNelem*numQdim, qdatasize, 974 CEED_STRIDES_BACKEND, &restrictqdi); 975 CeedElemRestrictionCreateStrided(ceed, localNelem, PetscPowInt(numP, dim), 976 localNelem*PetscPowInt(numP, dim), ncompx, 977 CEED_STRIDES_BACKEND, &restrictxcoord); 978 979 ierr = DMGetCoordinatesLocal(dm, &Xloc); CHKERRQ(ierr); 980 ierr = CreateVectorFromPetscVec(ceed, Xloc, &xcorners); CHKERRQ(ierr); 981 982 // Create the CEED vectors that will be needed in setup 983 CeedInt Nqpts; 984 CeedBasisGetNumQuadraturePoints(basisq, &Nqpts); 985 CeedInt Ndofs = 1; 986 for (int d=0; d<3; d++) Ndofs *= numP; 987 CeedVectorCreate(ceed, qdatasize*localNelem*Nqpts, &qdata); 988 CeedElemRestrictionCreateVector(restrictq, &q0ceed, NULL); 989 990 // Create the Q-function that builds the quadrature data for the NS operator 991 CeedQFunctionCreateInterior(ceed, 1, problem->setup, problem->setup_loc, 992 &qf_setup); 993 CeedQFunctionAddInput(qf_setup, "dx", ncompx*dim, CEED_EVAL_GRAD); 994 CeedQFunctionAddInput(qf_setup, "weight", 1, CEED_EVAL_WEIGHT); 995 CeedQFunctionAddOutput(qf_setup, "qdata", qdatasize, CEED_EVAL_NONE); 996 997 // Create the Q-function that sets the ICs of the operator 998 CeedQFunctionCreateInterior(ceed, 1, problem->ics, problem->ics_loc, &qf_ics); 999 CeedQFunctionAddInput(qf_ics, "x", ncompx, CEED_EVAL_INTERP); 1000 CeedQFunctionAddOutput(qf_ics, "q0", ncompq, CEED_EVAL_NONE); 1001 1002 qf_rhs = NULL; 1003 if (problem->apply_rhs) { // Create the Q-function that defines the action of the RHS operator 1004 CeedQFunctionCreateInterior(ceed, 1, problem->apply_rhs, 1005 problem->apply_rhs_loc, &qf_rhs); 1006 CeedQFunctionAddInput(qf_rhs, "q", ncompq, CEED_EVAL_INTERP); 1007 CeedQFunctionAddInput(qf_rhs, "dq", ncompq*dim, CEED_EVAL_GRAD); 1008 CeedQFunctionAddInput(qf_rhs, "qdata", qdatasize, CEED_EVAL_NONE); 1009 CeedQFunctionAddInput(qf_rhs, "x", ncompx, CEED_EVAL_INTERP); 1010 CeedQFunctionAddOutput(qf_rhs, "v", ncompq, CEED_EVAL_INTERP); 1011 CeedQFunctionAddOutput(qf_rhs, "dv", ncompq*dim, CEED_EVAL_GRAD); 1012 } 1013 1014 qf_ifunction = NULL; 1015 if (problem->apply_ifunction) { // Create the Q-function that defines the action of the IFunction 1016 CeedQFunctionCreateInterior(ceed, 1, problem->apply_ifunction, 1017 problem->apply_ifunction_loc, &qf_ifunction); 1018 CeedQFunctionAddInput(qf_ifunction, "q", ncompq, CEED_EVAL_INTERP); 1019 CeedQFunctionAddInput(qf_ifunction, "dq", ncompq*dim, CEED_EVAL_GRAD); 1020 CeedQFunctionAddInput(qf_ifunction, "qdot", ncompq, CEED_EVAL_INTERP); 1021 CeedQFunctionAddInput(qf_ifunction, "qdata", qdatasize, CEED_EVAL_NONE); 1022 CeedQFunctionAddInput(qf_ifunction, "x", ncompx, CEED_EVAL_INTERP); 1023 CeedQFunctionAddOutput(qf_ifunction, "v", ncompq, CEED_EVAL_INTERP); 1024 CeedQFunctionAddOutput(qf_ifunction, "dv", ncompq*dim, CEED_EVAL_GRAD); 1025 } 1026 1027 // Create the operator that builds the quadrature data for the NS operator 1028 CeedOperatorCreate(ceed, qf_setup, NULL, NULL, &op_setup); 1029 CeedOperatorSetField(op_setup, "dx", restrictx, basisx, CEED_VECTOR_ACTIVE); 1030 CeedOperatorSetField(op_setup, "weight", CEED_ELEMRESTRICTION_NONE, 1031 basisx, CEED_VECTOR_NONE); 1032 CeedOperatorSetField(op_setup, "qdata", restrictqdi, 1033 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE); 1034 1035 // Create the operator that sets the ICs 1036 CeedOperatorCreate(ceed, qf_ics, NULL, NULL, &op_ics); 1037 CeedOperatorSetField(op_ics, "x", restrictx, basisxc, CEED_VECTOR_ACTIVE); 1038 CeedOperatorSetField(op_ics, "q0", restrictq, 1039 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE); 1040 1041 CeedElemRestrictionCreateVector(restrictq, &user->qceed, NULL); 1042 CeedElemRestrictionCreateVector(restrictq, &user->qdotceed, NULL); 1043 CeedElemRestrictionCreateVector(restrictq, &user->gceed, NULL); 1044 1045 if (qf_rhs) { // Create the RHS physics operator 1046 CeedOperator op; 1047 CeedOperatorCreate(ceed, qf_rhs, NULL, NULL, &op); 1048 CeedOperatorSetField(op, "q", restrictq, basisq, CEED_VECTOR_ACTIVE); 1049 CeedOperatorSetField(op, "dq", restrictq, basisq, CEED_VECTOR_ACTIVE); 1050 CeedOperatorSetField(op, "qdata", restrictqdi, 1051 CEED_BASIS_COLLOCATED, qdata); 1052 CeedOperatorSetField(op, "x", restrictx, basisx, xcorners); 1053 CeedOperatorSetField(op, "v", restrictq, basisq, CEED_VECTOR_ACTIVE); 1054 CeedOperatorSetField(op, "dv", restrictq, basisq, CEED_VECTOR_ACTIVE); 1055 user->op_rhs = op; 1056 } 1057 1058 if (qf_ifunction) { // Create the IFunction operator 1059 CeedOperator op; 1060 CeedOperatorCreate(ceed, qf_ifunction, NULL, NULL, &op); 1061 CeedOperatorSetField(op, "q", restrictq, basisq, CEED_VECTOR_ACTIVE); 1062 CeedOperatorSetField(op, "dq", restrictq, basisq, CEED_VECTOR_ACTIVE); 1063 CeedOperatorSetField(op, "qdot", restrictq, basisq, user->qdotceed); 1064 CeedOperatorSetField(op, "qdata", restrictqdi, 1065 CEED_BASIS_COLLOCATED, qdata); 1066 CeedOperatorSetField(op, "x", restrictx, basisx, xcorners); 1067 CeedOperatorSetField(op, "v", restrictq, basisq, CEED_VECTOR_ACTIVE); 1068 CeedOperatorSetField(op, "dv", restrictq, basisq, CEED_VECTOR_ACTIVE); 1069 user->op_ifunction = op; 1070 } 1071 1072 CeedQFunctionSetContext(qf_ics, &ctxSetup, sizeof ctxSetup); 1073 CeedScalar ctxNS[8] = {lambda, mu, k, cv, cp, g, Rd}; 1074 struct Advection2dContext_ ctxAdvection2d = { 1075 .CtauS = CtauS, 1076 .strong_form = strong_form, 1077 .stabilization = stab, 1078 }; 1079 switch (problemChoice) { 1080 case NS_DENSITY_CURRENT: 1081 if (qf_rhs) CeedQFunctionSetContext(qf_rhs, &ctxNS, sizeof ctxNS); 1082 if (qf_ifunction) CeedQFunctionSetContext(qf_ifunction, &ctxNS, 1083 sizeof ctxNS); 1084 break; 1085 case NS_ADVECTION: 1086 case NS_ADVECTION2D: 1087 if (qf_rhs) CeedQFunctionSetContext(qf_rhs, &ctxAdvection2d, 1088 sizeof ctxAdvection2d); 1089 if (qf_ifunction) CeedQFunctionSetContext(qf_ifunction, &ctxAdvection2d, 1090 sizeof ctxAdvection2d); 1091 } 1092 1093 // Set up PETSc context 1094 // Set up units structure 1095 units->meter = meter; 1096 units->kilogram = kilogram; 1097 units->second = second; 1098 units->Kelvin = Kelvin; 1099 units->Pascal = Pascal; 1100 units->JperkgK = JperkgK; 1101 units->mpersquareds = mpersquareds; 1102 units->WpermK = WpermK; 1103 units->kgpercubicm = kgpercubicm; 1104 units->kgpersquaredms = kgpersquaredms; 1105 units->Joulepercubicm = Joulepercubicm; 1106 1107 // Set up user structure 1108 user->comm = comm; 1109 user->outputfreq = outputfreq; 1110 user->contsteps = contsteps; 1111 user->units = units; 1112 user->dm = dm; 1113 user->dmviz = dmviz; 1114 user->interpviz = interpviz; 1115 user->ceed = ceed; 1116 1117 // Calculate qdata and ICs 1118 // Set up state global and local vectors 1119 ierr = VecZeroEntries(Q); CHKERRQ(ierr); 1120 1121 ierr = VectorPlacePetscVec(q0ceed, Qloc); CHKERRQ(ierr); 1122 1123 // Apply Setup Ceed Operators 1124 ierr = VectorPlacePetscVec(xcorners, Xloc); CHKERRQ(ierr); 1125 CeedOperatorApply(op_setup, xcorners, qdata, CEED_REQUEST_IMMEDIATE); 1126 ierr = ComputeLumpedMassMatrix(ceed, dm, restrictq, basisq, restrictqdi, qdata, 1127 user->M); CHKERRQ(ierr); 1128 1129 ierr = ICs_PetscMultiplicity(op_ics, xcorners, q0ceed, dm, Qloc, Q, restrictq, 1130 &ctxSetup, 0.0); 1131 if (1) { // Record boundary values from initial condition and override DMPlexInsertBoundaryValues() 1132 // We use this for the main simulation DM because the reference DMPlexInsertBoundaryValues() is very slow. If we 1133 // disable this, we should still get the same results due to the problem->bc function, but with potentially much 1134 // slower execution. 1135 Vec Qbc; 1136 ierr = DMGetNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 1137 ierr = VecCopy(Qloc, Qbc); CHKERRQ(ierr); 1138 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 1139 ierr = DMGlobalToLocal(dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 1140 ierr = VecAXPY(Qbc, -1., Qloc); CHKERRQ(ierr); 1141 ierr = DMRestoreNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 1142 ierr = PetscObjectComposeFunction((PetscObject)dm, 1143 "DMPlexInsertBoundaryValues_C",DMPlexInsertBoundaryValues_NS); CHKERRQ(ierr); 1144 } 1145 1146 MPI_Comm_rank(comm, &rank); 1147 if (!rank) {ierr = PetscMkdir(user->outputfolder); CHKERRQ(ierr);} 1148 // Gather initial Q values 1149 // In case of continuation of simulation, set up initial values from binary file 1150 if (contsteps) { // continue from existent solution 1151 PetscViewer viewer; 1152 char filepath[PETSC_MAX_PATH_LEN]; 1153 // Read input 1154 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-solution.bin", 1155 user->outputfolder); 1156 CHKERRQ(ierr); 1157 ierr = PetscViewerBinaryOpen(comm, filepath, FILE_MODE_READ, &viewer); 1158 CHKERRQ(ierr); 1159 ierr = VecLoad(Q, viewer); CHKERRQ(ierr); 1160 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 1161 } else { 1162 //ierr = DMLocalToGlobal(dm, Qloc, INSERT_VALUES, Q);CHKERRQ(ierr); 1163 } 1164 ierr = DMRestoreLocalVector(dm, &Qloc); CHKERRQ(ierr); 1165 1166 // Create and setup TS 1167 ierr = TSCreate(comm, &ts); CHKERRQ(ierr); 1168 ierr = TSSetDM(ts, dm); CHKERRQ(ierr); 1169 if (implicit) { 1170 ierr = TSSetType(ts, TSBDF); CHKERRQ(ierr); 1171 if (user->op_ifunction) { 1172 ierr = TSSetIFunction(ts, NULL, IFunction_NS, &user); CHKERRQ(ierr); 1173 } else { // Implicit integrators can fall back to using an RHSFunction 1174 ierr = TSSetRHSFunction(ts, NULL, RHS_NS, &user); CHKERRQ(ierr); 1175 } 1176 } else { 1177 if (!user->op_rhs) SETERRQ(comm,PETSC_ERR_ARG_NULL, 1178 "Problem does not provide RHSFunction"); 1179 ierr = TSSetType(ts, TSRK); CHKERRQ(ierr); 1180 ierr = TSRKSetType(ts, TSRK5F); CHKERRQ(ierr); 1181 ierr = TSSetRHSFunction(ts, NULL, RHS_NS, &user); CHKERRQ(ierr); 1182 } 1183 ierr = TSSetMaxTime(ts, 500. * units->second); CHKERRQ(ierr); 1184 ierr = TSSetExactFinalTime(ts, TS_EXACTFINALTIME_STEPOVER); CHKERRQ(ierr); 1185 ierr = TSSetTimeStep(ts, 1.e-2 * units->second); CHKERRQ(ierr); 1186 if (test) {ierr = TSSetMaxSteps(ts, 1); CHKERRQ(ierr);} 1187 ierr = TSGetAdapt(ts, &adapt); CHKERRQ(ierr); 1188 ierr = TSAdaptSetStepLimits(adapt, 1189 1.e-12 * units->second, 1190 1.e2 * units->second); CHKERRQ(ierr); 1191 ierr = TSSetFromOptions(ts); CHKERRQ(ierr); 1192 if (!contsteps) { // print initial condition 1193 if (!test) { 1194 ierr = TSMonitor_NS(ts, 0, 0., Q, user); CHKERRQ(ierr); 1195 } 1196 } else { // continue from time of last output 1197 PetscReal time; 1198 PetscInt count; 1199 PetscViewer viewer; 1200 char filepath[PETSC_MAX_PATH_LEN]; 1201 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-time.bin", 1202 user->outputfolder); CHKERRQ(ierr); 1203 ierr = PetscViewerBinaryOpen(comm, filepath, FILE_MODE_READ, &viewer); 1204 CHKERRQ(ierr); 1205 ierr = PetscViewerBinaryRead(viewer, &time, 1, &count, PETSC_REAL); 1206 CHKERRQ(ierr); 1207 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 1208 ierr = TSSetTime(ts, time * user->units->second); CHKERRQ(ierr); 1209 } 1210 if (!test) { 1211 ierr = TSMonitorSet(ts, TSMonitor_NS, user, NULL); CHKERRQ(ierr); 1212 } 1213 1214 // Solve 1215 start = MPI_Wtime(); 1216 ierr = PetscBarrier((PetscObject)ts); CHKERRQ(ierr); 1217 ierr = TSSolve(ts, Q); CHKERRQ(ierr); 1218 cpu_time_used = MPI_Wtime() - start; 1219 ierr = TSGetSolveTime(ts,&ftime); CHKERRQ(ierr); 1220 ierr = MPI_Allreduce(MPI_IN_PLACE, &cpu_time_used, 1, MPI_DOUBLE, MPI_MIN, 1221 comm); CHKERRQ(ierr); 1222 if (!test) { 1223 ierr = PetscPrintf(PETSC_COMM_WORLD, 1224 "Time taken for solution: %g\n", 1225 (double)cpu_time_used); CHKERRQ(ierr); 1226 } 1227 1228 // Get error 1229 if (problem->non_zero_time && !test) { 1230 Vec Qexact, Qexactloc; 1231 PetscReal norm; 1232 ierr = DMCreateGlobalVector(dm, &Qexact); CHKERRQ(ierr); 1233 ierr = DMGetLocalVector(dm, &Qexactloc); CHKERRQ(ierr); 1234 ierr = VecGetSize(Qexactloc, &lnodes); CHKERRQ(ierr); 1235 1236 ierr = ICs_PetscMultiplicity(op_ics, xcorners, q0ceed, dm, Qexactloc, Qexact, 1237 restrictq, &ctxSetup, ftime); CHKERRQ(ierr); 1238 1239 ierr = VecAXPY(Q, -1.0, Qexact); CHKERRQ(ierr); 1240 ierr = VecNorm(Q, NORM_MAX, &norm); CHKERRQ(ierr); 1241 CeedVectorDestroy(&q0ceed); 1242 ierr = PetscPrintf(PETSC_COMM_WORLD, 1243 "Max Error: %g\n", 1244 (double)norm); CHKERRQ(ierr); 1245 } 1246 1247 // Output Statistics 1248 ierr = TSGetStepNumber(ts,&steps); CHKERRQ(ierr); 1249 if (!test) { 1250 ierr = PetscPrintf(PETSC_COMM_WORLD, 1251 "Time integrator took %D time steps to reach final time %g\n", 1252 steps,(double)ftime); CHKERRQ(ierr); 1253 } 1254 1255 // Clean up libCEED 1256 CeedVectorDestroy(&qdata); 1257 CeedVectorDestroy(&user->qceed); 1258 CeedVectorDestroy(&user->qdotceed); 1259 CeedVectorDestroy(&user->gceed); 1260 CeedVectorDestroy(&xcorners); 1261 CeedBasisDestroy(&basisq); 1262 CeedBasisDestroy(&basisx); 1263 CeedBasisDestroy(&basisxc); 1264 CeedElemRestrictionDestroy(&restrictq); 1265 CeedElemRestrictionDestroy(&restrictx); 1266 CeedElemRestrictionDestroy(&restrictqdi); 1267 CeedElemRestrictionDestroy(&restrictxcoord); 1268 CeedQFunctionDestroy(&qf_setup); 1269 CeedQFunctionDestroy(&qf_ics); 1270 CeedQFunctionDestroy(&qf_rhs); 1271 CeedQFunctionDestroy(&qf_ifunction); 1272 CeedOperatorDestroy(&op_setup); 1273 CeedOperatorDestroy(&op_ics); 1274 CeedOperatorDestroy(&user->op_rhs); 1275 CeedOperatorDestroy(&user->op_ifunction); 1276 CeedDestroy(&ceed); 1277 1278 // Clean up PETSc 1279 ierr = VecDestroy(&Q); CHKERRQ(ierr); 1280 ierr = VecDestroy(&user->M); CHKERRQ(ierr); 1281 ierr = MatDestroy(&interpviz); CHKERRQ(ierr); 1282 ierr = DMDestroy(&dmviz); CHKERRQ(ierr); 1283 ierr = TSDestroy(&ts); CHKERRQ(ierr); 1284 ierr = DMDestroy(&dm); CHKERRQ(ierr); 1285 ierr = PetscFree(units); CHKERRQ(ierr); 1286 ierr = PetscFree(user); CHKERRQ(ierr); 1287 return PetscFinalize(); 1288 } 1289 1290