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 -degree 1 32 // ./navierstokes -ceed /gpu/occa -problem advection -degree 1 33 // 34 //TESTARGS -ceed {ceed_resource} -test -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 = false, 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", 260 mpetsc, mceed); 261 ierr = VecGetArray(p, &a); CHKERRQ(ierr); 262 CeedVectorSetArray(c, CEED_MEM_HOST, CEED_USE_POINTER, a); 263 PetscFunctionReturn(0); 264 } 265 266 static PetscErrorCode DMPlexInsertBoundaryValues_NS(DM dm, 267 PetscBool insertEssential, Vec Qloc, PetscReal time, Vec faceGeomFVM, 268 Vec cellGeomFVM, Vec gradFVM) { 269 PetscErrorCode ierr; 270 Vec Qbc; 271 272 PetscFunctionBegin; 273 ierr = DMGetNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 274 ierr = VecAXPY(Qloc, 1., Qbc); CHKERRQ(ierr); 275 ierr = DMRestoreNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 276 PetscFunctionReturn(0); 277 } 278 279 // This is the RHS of the ODE, given as u_t = G(t,u) 280 // This function takes in a state vector Q and writes into G 281 static PetscErrorCode RHS_NS(TS ts, PetscReal t, Vec Q, Vec G, void *userData) { 282 PetscErrorCode ierr; 283 User user = *(User *)userData; 284 PetscScalar *q, *g; 285 Vec Qloc, Gloc; 286 287 // Global-to-local 288 PetscFunctionBeginUser; 289 ierr = DMGetLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 290 ierr = DMGetLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 291 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 292 ierr = DMGlobalToLocal(user->dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 293 ierr = DMPlexInsertBoundaryValues(user->dm, PETSC_TRUE, Qloc, 0.0, 294 NULL, NULL, NULL); CHKERRQ(ierr); 295 ierr = VecZeroEntries(Gloc); CHKERRQ(ierr); 296 297 // Ceed Vectors 298 ierr = VecGetArrayRead(Qloc, (const PetscScalar **)&q); CHKERRQ(ierr); 299 ierr = VecGetArray(Gloc, &g); CHKERRQ(ierr); 300 CeedVectorSetArray(user->qceed, CEED_MEM_HOST, CEED_USE_POINTER, q); 301 CeedVectorSetArray(user->gceed, CEED_MEM_HOST, CEED_USE_POINTER, g); 302 303 // Apply CEED operator 304 CeedOperatorApply(user->op_rhs, user->qceed, user->gceed, 305 CEED_REQUEST_IMMEDIATE); 306 307 // Restore vectors 308 ierr = VecRestoreArrayRead(Qloc, (const PetscScalar **)&q); CHKERRQ(ierr); 309 ierr = VecRestoreArray(Gloc, &g); CHKERRQ(ierr); 310 311 ierr = VecZeroEntries(G); CHKERRQ(ierr); 312 ierr = DMLocalToGlobal(user->dm, Gloc, ADD_VALUES, G); CHKERRQ(ierr); 313 314 // Inverse of the lumped mass matrix 315 ierr = VecPointwiseMult(G, G, user->M); // M is Minv 316 CHKERRQ(ierr); 317 318 ierr = DMRestoreLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 319 ierr = DMRestoreLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 320 PetscFunctionReturn(0); 321 } 322 323 static PetscErrorCode IFunction_NS(TS ts, PetscReal t, Vec Q, Vec Qdot, Vec G, 324 void *userData) { 325 PetscErrorCode ierr; 326 User user = *(User *)userData; 327 const PetscScalar *q, *qdot; 328 PetscScalar *g; 329 Vec Qloc, Qdotloc, Gloc; 330 331 // Global-to-local 332 PetscFunctionBeginUser; 333 ierr = DMGetLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 334 ierr = DMGetLocalVector(user->dm, &Qdotloc); CHKERRQ(ierr); 335 ierr = DMGetLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 336 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 337 ierr = DMGlobalToLocal(user->dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 338 ierr = DMPlexInsertBoundaryValues(user->dm, PETSC_TRUE, Qloc, 0.0, 339 NULL, NULL, NULL); CHKERRQ(ierr); 340 ierr = VecZeroEntries(Qdotloc); CHKERRQ(ierr); 341 ierr = DMGlobalToLocal(user->dm, Qdot, INSERT_VALUES, Qdotloc); CHKERRQ(ierr); 342 ierr = VecZeroEntries(Gloc); CHKERRQ(ierr); 343 344 // Ceed Vectors 345 ierr = VecGetArrayRead(Qloc, &q); CHKERRQ(ierr); 346 ierr = VecGetArrayRead(Qdotloc, &qdot); CHKERRQ(ierr); 347 ierr = VecGetArray(Gloc, &g); CHKERRQ(ierr); 348 CeedVectorSetArray(user->qceed, CEED_MEM_HOST, CEED_USE_POINTER, 349 (PetscScalar *)q); 350 CeedVectorSetArray(user->qdotceed, CEED_MEM_HOST, CEED_USE_POINTER, 351 (PetscScalar *)qdot); 352 CeedVectorSetArray(user->gceed, CEED_MEM_HOST, CEED_USE_POINTER, g); 353 354 // Apply CEED operator 355 CeedOperatorApply(user->op_ifunction, user->qceed, user->gceed, 356 CEED_REQUEST_IMMEDIATE); 357 358 // Restore vectors 359 ierr = VecRestoreArrayRead(Qloc, &q); CHKERRQ(ierr); 360 ierr = VecRestoreArrayRead(Qdotloc, &qdot); CHKERRQ(ierr); 361 ierr = VecRestoreArray(Gloc, &g); CHKERRQ(ierr); 362 363 ierr = VecZeroEntries(G); CHKERRQ(ierr); 364 ierr = DMLocalToGlobal(user->dm, Gloc, ADD_VALUES, G); CHKERRQ(ierr); 365 366 ierr = DMRestoreLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 367 ierr = DMRestoreLocalVector(user->dm, &Qdotloc); CHKERRQ(ierr); 368 ierr = DMRestoreLocalVector(user->dm, &Gloc); CHKERRQ(ierr); 369 PetscFunctionReturn(0); 370 } 371 372 // User provided TS Monitor 373 static PetscErrorCode TSMonitor_NS(TS ts, PetscInt stepno, PetscReal time, 374 Vec Q, void *ctx) { 375 User user = ctx; 376 Vec Qloc; 377 char filepath[PETSC_MAX_PATH_LEN]; 378 PetscViewer viewer; 379 PetscErrorCode ierr; 380 381 // Set up output 382 PetscFunctionBeginUser; 383 // Print every 'outputfreq' steps 384 if (stepno % user->outputfreq != 0) 385 PetscFunctionReturn(0); 386 ierr = DMGetLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 387 ierr = PetscObjectSetName((PetscObject)Qloc, "StateVec"); CHKERRQ(ierr); 388 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 389 ierr = DMGlobalToLocal(user->dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 390 391 // Output 392 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-%03D.vtu", 393 user->outputfolder, stepno + user->contsteps); 394 CHKERRQ(ierr); 395 ierr = PetscViewerVTKOpen(PetscObjectComm((PetscObject)Q), filepath, 396 FILE_MODE_WRITE, &viewer); CHKERRQ(ierr); 397 ierr = VecView(Qloc, viewer); CHKERRQ(ierr); 398 if (user->dmviz) { 399 Vec Qrefined, Qrefined_loc; 400 char filepath_refined[PETSC_MAX_PATH_LEN]; 401 PetscViewer viewer_refined; 402 403 ierr = DMGetGlobalVector(user->dmviz, &Qrefined); CHKERRQ(ierr); 404 ierr = DMGetLocalVector(user->dmviz, &Qrefined_loc); CHKERRQ(ierr); 405 ierr = PetscObjectSetName((PetscObject)Qrefined_loc, "Refined"); 406 CHKERRQ(ierr); 407 ierr = MatInterpolate(user->interpviz, Q, Qrefined); CHKERRQ(ierr); 408 ierr = VecZeroEntries(Qrefined_loc); CHKERRQ(ierr); 409 ierr = DMGlobalToLocal(user->dmviz, Qrefined, INSERT_VALUES, Qrefined_loc); 410 CHKERRQ(ierr); 411 ierr = PetscSNPrintf(filepath_refined, sizeof filepath_refined, 412 "%s/nsrefined-%03D.vtu", 413 user->outputfolder, stepno + user->contsteps); 414 CHKERRQ(ierr); 415 ierr = PetscViewerVTKOpen(PetscObjectComm((PetscObject)Qrefined), 416 filepath_refined, 417 FILE_MODE_WRITE, &viewer_refined); CHKERRQ(ierr); 418 ierr = VecView(Qrefined_loc, viewer_refined); CHKERRQ(ierr); 419 ierr = DMRestoreLocalVector(user->dmviz, &Qrefined_loc); CHKERRQ(ierr); 420 ierr = DMRestoreGlobalVector(user->dmviz, &Qrefined); CHKERRQ(ierr); 421 ierr = PetscViewerDestroy(&viewer_refined); CHKERRQ(ierr); 422 } 423 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 424 ierr = DMRestoreLocalVector(user->dm, &Qloc); CHKERRQ(ierr); 425 426 // Save data in a binary file for continuation of simulations 427 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-solution.bin", 428 user->outputfolder); CHKERRQ(ierr); 429 ierr = PetscViewerBinaryOpen(user->comm, filepath, FILE_MODE_WRITE, &viewer); 430 CHKERRQ(ierr); 431 ierr = VecView(Q, viewer); CHKERRQ(ierr); 432 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 433 434 // Save time stamp 435 // Dimensionalize time back 436 time /= user->units->second; 437 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-time.bin", 438 user->outputfolder); CHKERRQ(ierr); 439 ierr = PetscViewerBinaryOpen(user->comm, filepath, FILE_MODE_WRITE, &viewer); 440 CHKERRQ(ierr); 441 #if PETSC_VERSION_GE(3,13,0) 442 ierr = PetscViewerBinaryWrite(viewer, &time, 1, PETSC_REAL); 443 #else 444 ierr = PetscViewerBinaryWrite(viewer, &time, 1, PETSC_REAL, true); 445 #endif 446 CHKERRQ(ierr); 447 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 448 449 PetscFunctionReturn(0); 450 } 451 452 static PetscErrorCode ICs_FixMultiplicity(CeedOperator op_ics, 453 CeedVector xcorners, CeedVector q0ceed, DM dm, Vec Qloc, Vec Q, 454 CeedElemRestriction restrictq, SetupContext ctxSetup, CeedScalar time) { 455 PetscErrorCode ierr; 456 CeedVector multlvec; 457 Vec Multiplicity, MultiplicityLoc; 458 459 ctxSetup->time = time; 460 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 461 ierr = VectorPlacePetscVec(q0ceed, Qloc); CHKERRQ(ierr); 462 CeedOperatorApply(op_ics, xcorners, q0ceed, CEED_REQUEST_IMMEDIATE); 463 ierr = VecZeroEntries(Q); CHKERRQ(ierr); 464 ierr = DMLocalToGlobal(dm, Qloc, ADD_VALUES, Q); CHKERRQ(ierr); 465 466 // Fix multiplicity for output of ICs 467 ierr = DMGetLocalVector(dm, &MultiplicityLoc); CHKERRQ(ierr); 468 CeedElemRestrictionCreateVector(restrictq, &multlvec, NULL); 469 ierr = VectorPlacePetscVec(multlvec, MultiplicityLoc); CHKERRQ(ierr); 470 CeedElemRestrictionGetMultiplicity(restrictq, multlvec); 471 CeedVectorDestroy(&multlvec); 472 ierr = DMGetGlobalVector(dm, &Multiplicity); CHKERRQ(ierr); 473 ierr = VecZeroEntries(Multiplicity); CHKERRQ(ierr); 474 ierr = DMLocalToGlobal(dm, MultiplicityLoc, ADD_VALUES, Multiplicity); 475 CHKERRQ(ierr); 476 ierr = VecPointwiseDivide(Q, Q, Multiplicity); CHKERRQ(ierr); 477 ierr = VecPointwiseDivide(Qloc, Qloc, MultiplicityLoc); CHKERRQ(ierr); 478 ierr = DMRestoreLocalVector(dm, &MultiplicityLoc); CHKERRQ(ierr); 479 ierr = DMRestoreGlobalVector(dm, &Multiplicity); CHKERRQ(ierr); 480 481 PetscFunctionReturn(0); 482 } 483 484 static PetscErrorCode ComputeLumpedMassMatrix(Ceed ceed, DM dm, 485 CeedElemRestriction restrictq, CeedBasis basisq, 486 CeedElemRestriction restrictqdi, CeedVector qdata, Vec M) { 487 PetscErrorCode ierr; 488 CeedQFunction qf_mass; 489 CeedOperator op_mass; 490 CeedVector mceed; 491 Vec Mloc; 492 CeedInt ncompq, qdatasize; 493 494 PetscFunctionBeginUser; 495 CeedElemRestrictionGetNumComponents(restrictq, &ncompq); 496 CeedElemRestrictionGetNumComponents(restrictqdi, &qdatasize); 497 // Create the Q-function that defines the action of the mass operator 498 CeedQFunctionCreateInterior(ceed, 1, Mass, Mass_loc, &qf_mass); 499 CeedQFunctionAddInput(qf_mass, "q", ncompq, CEED_EVAL_INTERP); 500 CeedQFunctionAddInput(qf_mass, "qdata", qdatasize, CEED_EVAL_NONE); 501 CeedQFunctionAddOutput(qf_mass, "v", ncompq, CEED_EVAL_INTERP); 502 503 // Create the mass operator 504 CeedOperatorCreate(ceed, qf_mass, NULL, NULL, &op_mass); 505 CeedOperatorSetField(op_mass, "q", restrictq, basisq, CEED_VECTOR_ACTIVE); 506 CeedOperatorSetField(op_mass, "qdata", restrictqdi, 507 CEED_BASIS_COLLOCATED, qdata); 508 CeedOperatorSetField(op_mass, "v", restrictq, basisq, CEED_VECTOR_ACTIVE); 509 510 ierr = DMGetLocalVector(dm, &Mloc); CHKERRQ(ierr); 511 ierr = VecZeroEntries(Mloc); CHKERRQ(ierr); 512 CeedElemRestrictionCreateVector(restrictq, &mceed, NULL); 513 ierr = VectorPlacePetscVec(mceed, Mloc); CHKERRQ(ierr); 514 515 { 516 // Compute a lumped mass matrix 517 CeedVector onesvec; 518 CeedElemRestrictionCreateVector(restrictq, &onesvec, NULL); 519 CeedVectorSetValue(onesvec, 1.0); 520 CeedOperatorApply(op_mass, onesvec, mceed, CEED_REQUEST_IMMEDIATE); 521 CeedVectorDestroy(&onesvec); 522 CeedOperatorDestroy(&op_mass); 523 CeedVectorDestroy(&mceed); 524 } 525 CeedQFunctionDestroy(&qf_mass); 526 527 ierr = VecZeroEntries(M); CHKERRQ(ierr); 528 ierr = DMLocalToGlobal(dm, Mloc, ADD_VALUES, M); CHKERRQ(ierr); 529 ierr = DMRestoreLocalVector(dm, &Mloc); CHKERRQ(ierr); 530 531 // Invert diagonally lumped mass vector for RHS function 532 ierr = VecReciprocal(M); CHKERRQ(ierr); 533 PetscFunctionReturn(0); 534 } 535 536 PetscErrorCode SetUpDM(DM dm, problemData *problem, PetscInt degree, 537 SimpleBC bc, void *ctxSetup) { 538 PetscErrorCode ierr; 539 540 PetscFunctionBeginUser; 541 { 542 // Configure the finite element space and boundary conditions 543 PetscFE fe; 544 PetscSpace fespace; 545 PetscInt ncompq = 5; 546 ierr = PetscFECreateLagrange(PETSC_COMM_SELF, problem->dim, ncompq, 547 PETSC_FALSE, degree, PETSC_DECIDE, 548 &fe); 549 ierr = PetscObjectSetName((PetscObject)fe, "Q"); CHKERRQ(ierr); 550 ierr = DMAddField(dm,NULL,(PetscObject)fe); CHKERRQ(ierr); 551 ierr = DMCreateDS(dm); CHKERRQ(ierr); 552 /* Wall boundary conditions are zero velocity and zero flux for density and energy */ 553 { 554 PetscInt comps[3] = {1, 2, 3}; 555 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "wall", "Face Sets", 0, 556 3, comps, (void(*)(void))problem->bc, 557 bc->nwall, bc->walls, ctxSetup); CHKERRQ(ierr); 558 } 559 { 560 PetscInt comps[1] = {1}; 561 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "slipx", "Face Sets", 0, 562 1, comps, (void(*)(void))NULL, bc->nslip[0], 563 bc->slips[0], ctxSetup); CHKERRQ(ierr); 564 comps[0] = 2; 565 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "slipy", "Face Sets", 0, 566 1, comps, (void(*)(void))NULL, bc->nslip[1], 567 bc->slips[1], ctxSetup); CHKERRQ(ierr); 568 comps[0] = 3; 569 ierr = DMAddBoundary(dm, DM_BC_ESSENTIAL, "slipz", "Face Sets", 0, 570 1, comps, (void(*)(void))NULL, bc->nslip[2], 571 bc->slips[2], ctxSetup); CHKERRQ(ierr); 572 } 573 ierr = DMPlexSetClosurePermutationTensor(dm,PETSC_DETERMINE,NULL); 574 CHKERRQ(ierr); 575 ierr = PetscFEGetBasisSpace(fe, &fespace); CHKERRQ(ierr); 576 ierr = PetscFEDestroy(&fe); CHKERRQ(ierr); 577 } 578 { 579 // Empty name for conserved field (because there is only one field) 580 PetscSection section; 581 ierr = DMGetLocalSection(dm, §ion); CHKERRQ(ierr); 582 ierr = PetscSectionSetFieldName(section, 0, ""); CHKERRQ(ierr); 583 ierr = PetscSectionSetComponentName(section, 0, 0, "Density"); 584 CHKERRQ(ierr); 585 ierr = PetscSectionSetComponentName(section, 0, 1, "MomentumX"); 586 CHKERRQ(ierr); 587 ierr = PetscSectionSetComponentName(section, 0, 2, "MomentumY"); 588 CHKERRQ(ierr); 589 ierr = PetscSectionSetComponentName(section, 0, 3, "MomentumZ"); 590 CHKERRQ(ierr); 591 ierr = PetscSectionSetComponentName(section, 0, 4, "EnergyDensity"); 592 CHKERRQ(ierr); 593 } 594 PetscFunctionReturn(0); 595 } 596 597 int main(int argc, char **argv) { 598 PetscInt ierr; 599 MPI_Comm comm; 600 DM dm, dmcoord, dmviz; 601 Mat interpviz; 602 TS ts; 603 TSAdapt adapt; 604 User user; 605 Units units; 606 char ceedresource[4096] = "/cpu/self"; 607 PetscInt cStart, cEnd, localNelem, lnodes, steps; 608 const PetscInt ncompq = 5; 609 PetscMPIInt rank; 610 PetscScalar ftime; 611 Vec Q, Qloc, Xloc; 612 Ceed ceed; 613 CeedInt numP, numQ; 614 CeedVector xcorners, qdata, q0ceed; 615 CeedBasis basisx, basisxc, basisq; 616 CeedElemRestriction restrictx, restrictxcoord, restrictq, restrictqdi; 617 CeedQFunction qf_setup, qf_ics, qf_rhs, qf_ifunction; 618 CeedOperator op_setup, op_ics; 619 CeedScalar Rd; 620 PetscScalar WpermK, Pascal, JperkgK, mpersquareds, kgpercubicm, 621 kgpersquaredms, Joulepercubicm; 622 problemType problemChoice; 623 problemData *problem = NULL; 624 StabilizationType stab; 625 PetscBool test, implicit; 626 PetscInt viz_refine = 0; 627 struct SimpleBC_ bc = { 628 .nwall = 6, 629 .walls = {1,2,3,4,5,6}, 630 }; 631 double start, cpu_time_used; 632 633 // Create the libCEED contexts 634 PetscScalar meter = 1e-2; // 1 meter in scaled length units 635 PetscScalar second = 1e-2; // 1 second in scaled time units 636 PetscScalar kilogram = 1e-6; // 1 kilogram in scaled mass units 637 PetscScalar Kelvin = 1; // 1 Kelvin in scaled temperature units 638 CeedScalar theta0 = 300.; // K 639 CeedScalar thetaC = -15.; // K 640 CeedScalar P0 = 1.e5; // Pa 641 CeedScalar N = 0.01; // 1/s 642 CeedScalar cv = 717.; // J/(kg K) 643 CeedScalar cp = 1004.; // J/(kg K) 644 CeedScalar g = 9.81; // m/s^2 645 CeedScalar lambda = -2./3.; // - 646 CeedScalar mu = 75.; // Pa s, dynamic viscosity 647 // mu = 75 is not physical for air, but is good for numerical stability 648 CeedScalar k = 0.02638; // W/(m K) 649 CeedScalar CtauS = 0.; // dimensionless 650 CeedScalar strong_form = 0.; // [0,1] 651 PetscScalar lx = 8000.; // m 652 PetscScalar ly = 8000.; // m 653 PetscScalar lz = 4000.; // m 654 CeedScalar rc = 1000.; // m (Radius of bubble) 655 PetscScalar resx = 1000.; // m (resolution in x) 656 PetscScalar resy = 1000.; // m (resolution in y) 657 PetscScalar resz = 1000.; // m (resolution in z) 658 PetscInt outputfreq = 10; // - 659 PetscInt contsteps = 0; // - 660 PetscInt degree = 1; // - 661 PetscInt qextra = 2; // - 662 DMBoundaryType periodicity[] = {DM_BOUNDARY_NONE, DM_BOUNDARY_NONE, 663 DM_BOUNDARY_NONE 664 }; 665 PetscReal center[3], dc_axis[3] = {0, 0, 0}; 666 667 ierr = PetscInitialize(&argc, &argv, NULL, help); 668 if (ierr) return ierr; 669 670 // Allocate PETSc context 671 ierr = PetscCalloc1(1, &user); CHKERRQ(ierr); 672 ierr = PetscMalloc1(1, &units); CHKERRQ(ierr); 673 674 // Parse command line options 675 comm = PETSC_COMM_WORLD; 676 ierr = PetscOptionsBegin(comm, NULL, "Navier-Stokes in PETSc with libCEED", 677 NULL); CHKERRQ(ierr); 678 ierr = PetscOptionsString("-ceed", "CEED resource specifier", 679 NULL, ceedresource, ceedresource, 680 sizeof(ceedresource), NULL); CHKERRQ(ierr); 681 ierr = PetscOptionsBool("-test", "Run in test mode", 682 NULL, test=PETSC_FALSE, &test, NULL); CHKERRQ(ierr); 683 problemChoice = NS_DENSITY_CURRENT; 684 ierr = PetscOptionsEnum("-problem", "Problem to solve", NULL, 685 problemTypes, (PetscEnum)problemChoice, 686 (PetscEnum *)&problemChoice, NULL); CHKERRQ(ierr); 687 problem = &problemOptions[problemChoice]; 688 ierr = PetscOptionsEnum("-stab", "Stabilization method", NULL, 689 StabilizationTypes, (PetscEnum)(stab = STAB_NONE), 690 (PetscEnum *)&stab, NULL); CHKERRQ(ierr); 691 ierr = PetscOptionsBool("-implicit", "Use implicit (IFunction) formulation", 692 NULL, implicit=PETSC_FALSE, &implicit, NULL); 693 CHKERRQ(ierr); 694 { 695 PetscInt len; 696 PetscBool flg; 697 ierr = PetscOptionsIntArray("-bc_wall", 698 "Use wall boundary conditions on this list of faces", 699 NULL, bc.walls, 700 (len = sizeof(bc.walls) / sizeof(bc.walls[0]), 701 &len), &flg); CHKERRQ(ierr); 702 if (flg) bc.nwall = len; 703 for (PetscInt j=0; j<3; j++) { 704 const char *flags[3] = {"-bc_slip_x", "-bc_slip_y", "-bc_slip_z"}; 705 ierr = PetscOptionsIntArray(flags[j], 706 "Use slip boundary conditions on this list of faces", 707 NULL, bc.slips[j], 708 (len = sizeof(bc.slips[j]) / sizeof(bc.slips[j][0]), 709 &len), &flg); 710 CHKERRQ(ierr); 711 if (flg) bc.nslip[j] = len; 712 } 713 } 714 ierr = PetscOptionsInt("-viz_refine", 715 "Regular refinement levels for visualization", 716 NULL, viz_refine, &viz_refine, NULL); 717 CHKERRQ(ierr); 718 ierr = PetscOptionsScalar("-units_meter", "1 meter in scaled length units", 719 NULL, meter, &meter, NULL); CHKERRQ(ierr); 720 meter = fabs(meter); 721 ierr = PetscOptionsScalar("-units_second","1 second in scaled time units", 722 NULL, second, &second, NULL); CHKERRQ(ierr); 723 second = fabs(second); 724 ierr = PetscOptionsScalar("-units_kilogram","1 kilogram in scaled mass units", 725 NULL, kilogram, &kilogram, NULL); CHKERRQ(ierr); 726 kilogram = fabs(kilogram); 727 ierr = PetscOptionsScalar("-units_Kelvin", 728 "1 Kelvin in scaled temperature units", 729 NULL, Kelvin, &Kelvin, NULL); CHKERRQ(ierr); 730 Kelvin = fabs(Kelvin); 731 ierr = PetscOptionsScalar("-theta0", "Reference potential temperature", 732 NULL, theta0, &theta0, NULL); CHKERRQ(ierr); 733 ierr = PetscOptionsScalar("-thetaC", "Perturbation of potential temperature", 734 NULL, thetaC, &thetaC, NULL); CHKERRQ(ierr); 735 ierr = PetscOptionsScalar("-P0", "Atmospheric pressure", 736 NULL, P0, &P0, NULL); CHKERRQ(ierr); 737 ierr = PetscOptionsScalar("-N", "Brunt-Vaisala frequency", 738 NULL, N, &N, NULL); CHKERRQ(ierr); 739 ierr = PetscOptionsScalar("-cv", "Heat capacity at constant volume", 740 NULL, cv, &cv, NULL); CHKERRQ(ierr); 741 ierr = PetscOptionsScalar("-cp", "Heat capacity at constant pressure", 742 NULL, cp, &cp, NULL); CHKERRQ(ierr); 743 ierr = PetscOptionsScalar("-g", "Gravitational acceleration", 744 NULL, g, &g, NULL); CHKERRQ(ierr); 745 ierr = PetscOptionsScalar("-lambda", 746 "Stokes hypothesis second viscosity coefficient", 747 NULL, lambda, &lambda, NULL); CHKERRQ(ierr); 748 ierr = PetscOptionsScalar("-mu", "Shear dynamic viscosity coefficient", 749 NULL, mu, &mu, NULL); CHKERRQ(ierr); 750 ierr = PetscOptionsScalar("-k", "Thermal conductivity", 751 NULL, k, &k, NULL); CHKERRQ(ierr); 752 ierr = PetscOptionsScalar("-CtauS", 753 "Scale coefficient for tau (nondimensional)", 754 NULL, CtauS, &CtauS, NULL); CHKERRQ(ierr); 755 ierr = PetscOptionsScalar("-strong_form", 756 "Strong (1) or weak/integrated by parts (0) advection residual", 757 NULL, strong_form, &strong_form, NULL); 758 CHKERRQ(ierr); 759 ierr = PetscOptionsScalar("-lx", "Length scale in x direction", 760 NULL, lx, &lx, NULL); CHKERRQ(ierr); 761 ierr = PetscOptionsScalar("-ly", "Length scale in y direction", 762 NULL, ly, &ly, NULL); CHKERRQ(ierr); 763 ierr = PetscOptionsScalar("-lz", "Length scale in z direction", 764 NULL, lz, &lz, NULL); CHKERRQ(ierr); 765 ierr = PetscOptionsScalar("-rc", "Characteristic radius of thermal bubble", 766 NULL, rc, &rc, NULL); CHKERRQ(ierr); 767 ierr = PetscOptionsScalar("-resx","Target resolution in x", 768 NULL, resx, &resx, NULL); CHKERRQ(ierr); 769 ierr = PetscOptionsScalar("-resy","Target resolution in y", 770 NULL, resy, &resy, NULL); CHKERRQ(ierr); 771 ierr = PetscOptionsScalar("-resz","Target resolution in z", 772 NULL, resz, &resz, NULL); CHKERRQ(ierr); 773 PetscInt n = problem->dim; 774 ierr = PetscOptionsEnumArray("-periodicity", "Periodicity per direction", 775 NULL, DMBoundaryTypes, (PetscEnum *)periodicity, 776 &n, NULL); CHKERRQ(ierr); 777 n = problem->dim; 778 center[0] = 0.5 * lx; 779 center[1] = 0.5 * ly; 780 center[2] = 0.5 * lz; 781 ierr = PetscOptionsRealArray("-center", "Location of bubble center", 782 NULL, center, &n, NULL); CHKERRQ(ierr); 783 n = problem->dim; 784 ierr = PetscOptionsRealArray("-dc_axis", 785 "Axis of density current cylindrical anomaly, or {0,0,0} for spherically symmetric", 786 NULL, dc_axis, &n, NULL); CHKERRQ(ierr); 787 { 788 PetscReal norm = PetscSqrtReal(PetscSqr(dc_axis[0]) + 789 PetscSqr(dc_axis[1]) + PetscSqr(dc_axis[2])); 790 if (norm > 0) { 791 for (int i=0; i<3; i++) dc_axis[i] /= norm; 792 } 793 } 794 ierr = PetscOptionsInt("-output_freq", 795 "Frequency of output, in number of steps", 796 NULL, outputfreq, &outputfreq, NULL); CHKERRQ(ierr); 797 ierr = PetscOptionsInt("-continue", "Continue from previous solution", 798 NULL, contsteps, &contsteps, NULL); CHKERRQ(ierr); 799 ierr = PetscOptionsInt("-degree", "Polynomial degree of finite elements", 800 NULL, degree, °ree, NULL); CHKERRQ(ierr); 801 ierr = PetscOptionsInt("-qextra", "Number of extra quadrature points", 802 NULL, qextra, &qextra, NULL); CHKERRQ(ierr); 803 PetscStrncpy(user->outputfolder, ".", 2); 804 ierr = PetscOptionsString("-of", "Output folder", 805 NULL, user->outputfolder, user->outputfolder, 806 sizeof(user->outputfolder), NULL); CHKERRQ(ierr); 807 ierr = PetscOptionsEnd(); CHKERRQ(ierr); 808 809 // Define derived units 810 Pascal = kilogram / (meter * PetscSqr(second)); 811 JperkgK = PetscSqr(meter) / (PetscSqr(second) * Kelvin); 812 mpersquareds = meter / PetscSqr(second); 813 WpermK = kilogram * meter / (pow(second,3) * Kelvin); 814 kgpercubicm = kilogram / pow(meter,3); 815 kgpersquaredms = kilogram / (PetscSqr(meter) * second); 816 Joulepercubicm = kilogram / (meter * PetscSqr(second)); 817 818 // Scale variables to desired units 819 theta0 *= Kelvin; 820 thetaC *= Kelvin; 821 P0 *= Pascal; 822 N *= (1./second); 823 cv *= JperkgK; 824 cp *= JperkgK; 825 Rd = cp - cv; 826 g *= mpersquareds; 827 mu *= Pascal * second; 828 k *= WpermK; 829 lx = fabs(lx) * meter; 830 ly = fabs(ly) * meter; 831 lz = fabs(lz) * meter; 832 rc = fabs(rc) * meter; 833 resx = fabs(resx) * meter; 834 resy = fabs(resy) * meter; 835 resz = fabs(resz) * meter; 836 for (int i=0; i<3; i++) center[i] *= meter; 837 838 const CeedInt dim = problem->dim, ncompx = problem->dim, 839 qdatasize = problem->qdatasize; 840 // Set up the libCEED context 841 struct SetupContext_ ctxSetup = { 842 .theta0 = theta0, 843 .thetaC = thetaC, 844 .P0 = P0, 845 .N = N, 846 .cv = cv, 847 .cp = cp, 848 .Rd = Rd, 849 .g = g, 850 .rc = rc, 851 .lx = lx, 852 .ly = ly, 853 .lz = lz, 854 .periodicity0 = periodicity[0], 855 .periodicity1 = periodicity[1], 856 .periodicity2 = periodicity[2], 857 .center[0] = center[0], 858 .center[1] = center[1], 859 .center[2] = center[2], 860 .dc_axis[0] = dc_axis[0], 861 .dc_axis[1] = dc_axis[1], 862 .dc_axis[2] = dc_axis[2], 863 .time = 0, 864 }; 865 866 // Create the mesh 867 { 868 const PetscReal scale[3] = {lx, ly, lz}; 869 ierr = DMPlexCreateBoxMesh(comm, dim, PETSC_FALSE, NULL, NULL, scale, 870 periodicity, PETSC_TRUE, &dm); 871 CHKERRQ(ierr); 872 } 873 874 // Distribute the mesh over processes 875 { 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 // Setup DM 890 ierr = DMLocalizeCoordinates(dm); CHKERRQ(ierr); 891 ierr = DMSetFromOptions(dm); CHKERRQ(ierr); 892 ierr = SetUpDM(dm, problem, degree, &bc, &ctxSetup); CHKERRQ(ierr); 893 894 // Print FEM space information 895 if (!test) { 896 ierr = PetscPrintf(PETSC_COMM_WORLD, 897 "Degree of FEM space: %D\n", 898 degree); CHKERRQ(ierr); 899 } 900 901 // Refine DM for high-order viz 902 dmviz = NULL; 903 interpviz = NULL; 904 if (viz_refine) { 905 DM dmhierarchy[viz_refine+1]; 906 907 ierr = DMPlexSetRefinementUniform(dm, PETSC_TRUE); CHKERRQ(ierr); 908 dmhierarchy[0] = dm; 909 for (PetscInt i = 0, d = degree; i < viz_refine; i++) { 910 Mat interp_next; 911 912 ierr = DMRefine(dmhierarchy[i], MPI_COMM_NULL, &dmhierarchy[i+1]); 913 CHKERRQ(ierr); 914 ierr = DMSetCoarseDM(dmhierarchy[i+1], dmhierarchy[i]); CHKERRQ(ierr); 915 d = (d + 1) / 2; 916 if (i + 1 == viz_refine) d = 1; 917 ierr = SetUpDM(dmhierarchy[i+1], problem, d, &bc, &ctxSetup); CHKERRQ(ierr); 918 ierr = DMCreateInterpolation(dmhierarchy[i], dmhierarchy[i+1], 919 &interp_next, NULL); CHKERRQ(ierr); 920 if (!i) interpviz = interp_next; 921 else { 922 Mat C; 923 ierr = MatMatMult(interp_next, interpviz, MAT_INITIAL_MATRIX, 924 PETSC_DECIDE, &C); CHKERRQ(ierr); 925 ierr = MatDestroy(&interp_next); CHKERRQ(ierr); 926 ierr = MatDestroy(&interpviz); CHKERRQ(ierr); 927 interpviz = C; 928 } 929 } 930 for (PetscInt i=1; i<viz_refine; i++) { 931 ierr = DMDestroy(&dmhierarchy[i]); CHKERRQ(ierr); 932 } 933 dmviz = dmhierarchy[viz_refine]; 934 } 935 ierr = DMCreateGlobalVector(dm, &Q); CHKERRQ(ierr); 936 ierr = DMGetLocalVector(dm, &Qloc); CHKERRQ(ierr); 937 ierr = VecGetSize(Qloc, &lnodes); CHKERRQ(ierr); 938 lnodes /= ncompq; 939 940 { 941 // Print grid information 942 CeedInt gdofs, odofs; 943 int comm_size; 944 char box_faces_str[PETSC_MAX_PATH_LEN] = "NONE"; 945 ierr = VecGetSize(Q, &gdofs); CHKERRQ(ierr); 946 ierr = VecGetLocalSize(Q, &odofs); CHKERRQ(ierr); 947 ierr = MPI_Comm_size(comm, &comm_size); CHKERRQ(ierr); 948 ierr = PetscOptionsGetString(NULL, NULL, "-dm_plex_box_faces", box_faces_str, 949 sizeof(box_faces_str), NULL); CHKERRQ(ierr); 950 if (!test) { 951 ierr = PetscPrintf(comm, "Global FEM dofs: %D (%D owned) on %d rank(s)\n", 952 gdofs, odofs, comm_size); CHKERRQ(ierr); 953 ierr = PetscPrintf(comm, "Local FEM nodes: %D\n", lnodes); CHKERRQ(ierr); 954 ierr = PetscPrintf(comm, "dm_plex_box_faces: %s\n", box_faces_str); 955 CHKERRQ(ierr); 956 } 957 958 } 959 960 // Set up global mass vector 961 ierr = VecDuplicate(Q, &user->M); CHKERRQ(ierr); 962 963 // Set up CEED 964 // CEED Bases 965 CeedInit(ceedresource, &ceed); 966 numP = degree + 1; 967 numQ = numP + qextra; 968 CeedBasisCreateTensorH1Lagrange(ceed, dim, ncompq, numP, numQ, CEED_GAUSS, 969 &basisq); 970 CeedBasisCreateTensorH1Lagrange(ceed, dim, ncompx, 2, numQ, CEED_GAUSS, 971 &basisx); 972 CeedBasisCreateTensorH1Lagrange(ceed, dim, ncompx, 2, numP, 973 CEED_GAUSS_LOBATTO, &basisxc); 974 975 ierr = DMGetCoordinateDM(dm, &dmcoord); CHKERRQ(ierr); 976 ierr = DMPlexSetClosurePermutationTensor(dmcoord, PETSC_DETERMINE, NULL); 977 CHKERRQ(ierr); 978 979 // CEED Restrictions 980 ierr = CreateRestrictionFromPlex(ceed, dm, degree+1, &restrictq); 981 CHKERRQ(ierr); 982 ierr = CreateRestrictionFromPlex(ceed, dmcoord, 2, &restrictx); CHKERRQ(ierr); 983 DMPlexGetHeightStratum(dm, 0, &cStart, &cEnd); CHKERRQ(ierr); 984 localNelem = cEnd - cStart; 985 CeedInt numQdim = CeedIntPow(numQ, dim); 986 CeedElemRestrictionCreateStrided(ceed, localNelem, numQdim, 987 localNelem*numQdim, qdatasize, 988 CEED_STRIDES_BACKEND, &restrictqdi); 989 CeedElemRestrictionCreateStrided(ceed, localNelem, PetscPowInt(numP, dim), 990 localNelem*PetscPowInt(numP, dim), ncompx, 991 CEED_STRIDES_BACKEND, &restrictxcoord); 992 993 ierr = DMGetCoordinatesLocal(dm, &Xloc); CHKERRQ(ierr); 994 ierr = CreateVectorFromPetscVec(ceed, Xloc, &xcorners); CHKERRQ(ierr); 995 996 // Create the CEED vectors that will be needed in setup 997 CeedInt Nqpts; 998 CeedBasisGetNumQuadraturePoints(basisq, &Nqpts); 999 CeedVectorCreate(ceed, qdatasize*localNelem*Nqpts, &qdata); 1000 CeedElemRestrictionCreateVector(restrictq, &q0ceed, NULL); 1001 1002 // Create the Q-function that builds the quadrature data for the NS operator 1003 CeedQFunctionCreateInterior(ceed, 1, problem->setup, problem->setup_loc, 1004 &qf_setup); 1005 CeedQFunctionAddInput(qf_setup, "dx", ncompx*dim, CEED_EVAL_GRAD); 1006 CeedQFunctionAddInput(qf_setup, "weight", 1, CEED_EVAL_WEIGHT); 1007 CeedQFunctionAddOutput(qf_setup, "qdata", qdatasize, CEED_EVAL_NONE); 1008 1009 // Create the Q-function that sets the ICs of the operator 1010 CeedQFunctionCreateInterior(ceed, 1, problem->ics, problem->ics_loc, &qf_ics); 1011 CeedQFunctionAddInput(qf_ics, "x", ncompx, CEED_EVAL_INTERP); 1012 CeedQFunctionAddOutput(qf_ics, "q0", ncompq, CEED_EVAL_NONE); 1013 1014 qf_rhs = NULL; 1015 if (problem->apply_rhs) { // Create the Q-function that defines the action of the RHS operator 1016 CeedQFunctionCreateInterior(ceed, 1, problem->apply_rhs, 1017 problem->apply_rhs_loc, &qf_rhs); 1018 CeedQFunctionAddInput(qf_rhs, "q", ncompq, CEED_EVAL_INTERP); 1019 CeedQFunctionAddInput(qf_rhs, "dq", ncompq*dim, CEED_EVAL_GRAD); 1020 CeedQFunctionAddInput(qf_rhs, "qdata", qdatasize, CEED_EVAL_NONE); 1021 CeedQFunctionAddInput(qf_rhs, "x", ncompx, CEED_EVAL_INTERP); 1022 CeedQFunctionAddOutput(qf_rhs, "v", ncompq, CEED_EVAL_INTERP); 1023 CeedQFunctionAddOutput(qf_rhs, "dv", ncompq*dim, CEED_EVAL_GRAD); 1024 } 1025 1026 qf_ifunction = NULL; 1027 if (problem->apply_ifunction) { // Create the Q-function that defines the action of the IFunction 1028 CeedQFunctionCreateInterior(ceed, 1, problem->apply_ifunction, 1029 problem->apply_ifunction_loc, &qf_ifunction); 1030 CeedQFunctionAddInput(qf_ifunction, "q", ncompq, CEED_EVAL_INTERP); 1031 CeedQFunctionAddInput(qf_ifunction, "dq", ncompq*dim, CEED_EVAL_GRAD); 1032 CeedQFunctionAddInput(qf_ifunction, "qdot", ncompq, CEED_EVAL_INTERP); 1033 CeedQFunctionAddInput(qf_ifunction, "qdata", qdatasize, CEED_EVAL_NONE); 1034 CeedQFunctionAddInput(qf_ifunction, "x", ncompx, CEED_EVAL_INTERP); 1035 CeedQFunctionAddOutput(qf_ifunction, "v", ncompq, CEED_EVAL_INTERP); 1036 CeedQFunctionAddOutput(qf_ifunction, "dv", ncompq*dim, CEED_EVAL_GRAD); 1037 } 1038 1039 // Create the operator that builds the quadrature data for the NS operator 1040 CeedOperatorCreate(ceed, qf_setup, NULL, NULL, &op_setup); 1041 CeedOperatorSetField(op_setup, "dx", restrictx, basisx, CEED_VECTOR_ACTIVE); 1042 CeedOperatorSetField(op_setup, "weight", CEED_ELEMRESTRICTION_NONE, 1043 basisx, CEED_VECTOR_NONE); 1044 CeedOperatorSetField(op_setup, "qdata", restrictqdi, 1045 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE); 1046 1047 // Create the operator that sets the ICs 1048 CeedOperatorCreate(ceed, qf_ics, NULL, NULL, &op_ics); 1049 CeedOperatorSetField(op_ics, "x", restrictx, basisxc, CEED_VECTOR_ACTIVE); 1050 CeedOperatorSetField(op_ics, "q0", restrictq, 1051 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE); 1052 1053 CeedElemRestrictionCreateVector(restrictq, &user->qceed, NULL); 1054 CeedElemRestrictionCreateVector(restrictq, &user->qdotceed, NULL); 1055 CeedElemRestrictionCreateVector(restrictq, &user->gceed, NULL); 1056 1057 if (qf_rhs) { // Create the RHS physics operator 1058 CeedOperator op; 1059 CeedOperatorCreate(ceed, qf_rhs, NULL, NULL, &op); 1060 CeedOperatorSetField(op, "q", restrictq, basisq, CEED_VECTOR_ACTIVE); 1061 CeedOperatorSetField(op, "dq", restrictq, basisq, CEED_VECTOR_ACTIVE); 1062 CeedOperatorSetField(op, "qdata", restrictqdi, 1063 CEED_BASIS_COLLOCATED, qdata); 1064 CeedOperatorSetField(op, "x", restrictx, basisx, xcorners); 1065 CeedOperatorSetField(op, "v", restrictq, basisq, CEED_VECTOR_ACTIVE); 1066 CeedOperatorSetField(op, "dv", restrictq, basisq, CEED_VECTOR_ACTIVE); 1067 user->op_rhs = op; 1068 } 1069 1070 if (qf_ifunction) { // Create the IFunction operator 1071 CeedOperator op; 1072 CeedOperatorCreate(ceed, qf_ifunction, NULL, NULL, &op); 1073 CeedOperatorSetField(op, "q", restrictq, basisq, CEED_VECTOR_ACTIVE); 1074 CeedOperatorSetField(op, "dq", restrictq, basisq, CEED_VECTOR_ACTIVE); 1075 CeedOperatorSetField(op, "qdot", restrictq, basisq, user->qdotceed); 1076 CeedOperatorSetField(op, "qdata", restrictqdi, 1077 CEED_BASIS_COLLOCATED, qdata); 1078 CeedOperatorSetField(op, "x", restrictx, basisx, xcorners); 1079 CeedOperatorSetField(op, "v", restrictq, basisq, CEED_VECTOR_ACTIVE); 1080 CeedOperatorSetField(op, "dv", restrictq, basisq, CEED_VECTOR_ACTIVE); 1081 user->op_ifunction = op; 1082 } 1083 1084 CeedQFunctionSetContext(qf_ics, &ctxSetup, sizeof ctxSetup); 1085 CeedScalar ctxNS[8] = {lambda, mu, k, cv, cp, g, Rd}; 1086 struct Advection2dContext_ ctxAdvection2d = { 1087 .CtauS = CtauS, 1088 .strong_form = strong_form, 1089 .stabilization = stab, 1090 }; 1091 switch (problemChoice) { 1092 case NS_DENSITY_CURRENT: 1093 if (qf_rhs) CeedQFunctionSetContext(qf_rhs, &ctxNS, sizeof ctxNS); 1094 if (qf_ifunction) CeedQFunctionSetContext(qf_ifunction, &ctxNS, 1095 sizeof ctxNS); 1096 break; 1097 case NS_ADVECTION: 1098 case NS_ADVECTION2D: 1099 if (qf_rhs) CeedQFunctionSetContext(qf_rhs, &ctxAdvection2d, 1100 sizeof ctxAdvection2d); 1101 if (qf_ifunction) CeedQFunctionSetContext(qf_ifunction, &ctxAdvection2d, 1102 sizeof ctxAdvection2d); 1103 } 1104 1105 // Set up PETSc context 1106 // Set up units structure 1107 units->meter = meter; 1108 units->kilogram = kilogram; 1109 units->second = second; 1110 units->Kelvin = Kelvin; 1111 units->Pascal = Pascal; 1112 units->JperkgK = JperkgK; 1113 units->mpersquareds = mpersquareds; 1114 units->WpermK = WpermK; 1115 units->kgpercubicm = kgpercubicm; 1116 units->kgpersquaredms = kgpersquaredms; 1117 units->Joulepercubicm = Joulepercubicm; 1118 1119 // Set up user structure 1120 user->comm = comm; 1121 user->outputfreq = outputfreq; 1122 user->contsteps = contsteps; 1123 user->units = units; 1124 user->dm = dm; 1125 user->dmviz = dmviz; 1126 user->interpviz = interpviz; 1127 user->ceed = ceed; 1128 1129 // Calculate qdata and ICs 1130 // Set up state global and local vectors 1131 ierr = VecZeroEntries(Q); CHKERRQ(ierr); 1132 1133 ierr = VectorPlacePetscVec(q0ceed, Qloc); CHKERRQ(ierr); 1134 1135 // Apply Setup Ceed Operators 1136 ierr = VectorPlacePetscVec(xcorners, Xloc); CHKERRQ(ierr); 1137 CeedOperatorApply(op_setup, xcorners, qdata, CEED_REQUEST_IMMEDIATE); 1138 ierr = ComputeLumpedMassMatrix(ceed, dm, restrictq, basisq, restrictqdi, qdata, 1139 user->M); CHKERRQ(ierr); 1140 1141 ierr = ICs_FixMultiplicity(op_ics, xcorners, q0ceed, dm, Qloc, Q, restrictq, 1142 &ctxSetup, 0.0); 1143 if (1) { // Record boundary values from initial condition and override DMPlexInsertBoundaryValues() 1144 // We use this for the main simulation DM because the reference DMPlexInsertBoundaryValues() is very slow. If we 1145 // disable this, we should still get the same results due to the problem->bc function, but with potentially much 1146 // slower execution. 1147 Vec Qbc; 1148 ierr = DMGetNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 1149 ierr = VecCopy(Qloc, Qbc); CHKERRQ(ierr); 1150 ierr = VecZeroEntries(Qloc); CHKERRQ(ierr); 1151 ierr = DMGlobalToLocal(dm, Q, INSERT_VALUES, Qloc); CHKERRQ(ierr); 1152 ierr = VecAXPY(Qbc, -1., Qloc); CHKERRQ(ierr); 1153 ierr = DMRestoreNamedLocalVector(dm, "Qbc", &Qbc); CHKERRQ(ierr); 1154 ierr = PetscObjectComposeFunction((PetscObject)dm, 1155 "DMPlexInsertBoundaryValues_C", DMPlexInsertBoundaryValues_NS); 1156 CHKERRQ(ierr); 1157 } 1158 1159 MPI_Comm_rank(comm, &rank); 1160 if (!rank) {ierr = PetscMkdir(user->outputfolder); CHKERRQ(ierr);} 1161 // Gather initial Q values 1162 // In case of continuation of simulation, set up initial values from binary file 1163 if (contsteps) { // continue from existent solution 1164 PetscViewer viewer; 1165 char filepath[PETSC_MAX_PATH_LEN]; 1166 // Read input 1167 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-solution.bin", 1168 user->outputfolder); 1169 CHKERRQ(ierr); 1170 ierr = PetscViewerBinaryOpen(comm, filepath, FILE_MODE_READ, &viewer); 1171 CHKERRQ(ierr); 1172 ierr = VecLoad(Q, viewer); CHKERRQ(ierr); 1173 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 1174 } 1175 ierr = DMRestoreLocalVector(dm, &Qloc); CHKERRQ(ierr); 1176 1177 // Create and setup TS 1178 ierr = TSCreate(comm, &ts); CHKERRQ(ierr); 1179 ierr = TSSetDM(ts, dm); CHKERRQ(ierr); 1180 if (implicit) { 1181 ierr = TSSetType(ts, TSBDF); CHKERRQ(ierr); 1182 if (user->op_ifunction) { 1183 ierr = TSSetIFunction(ts, NULL, IFunction_NS, &user); CHKERRQ(ierr); 1184 } else { // Implicit integrators can fall back to using an RHSFunction 1185 ierr = TSSetRHSFunction(ts, NULL, RHS_NS, &user); CHKERRQ(ierr); 1186 } 1187 } else { 1188 if (!user->op_rhs) SETERRQ(comm,PETSC_ERR_ARG_NULL, 1189 "Problem does not provide RHSFunction"); 1190 ierr = TSSetType(ts, TSRK); CHKERRQ(ierr); 1191 ierr = TSRKSetType(ts, TSRK5F); CHKERRQ(ierr); 1192 ierr = TSSetRHSFunction(ts, NULL, RHS_NS, &user); CHKERRQ(ierr); 1193 } 1194 ierr = TSSetMaxTime(ts, 500. * units->second); CHKERRQ(ierr); 1195 ierr = TSSetExactFinalTime(ts, TS_EXACTFINALTIME_STEPOVER); CHKERRQ(ierr); 1196 ierr = TSSetTimeStep(ts, 1.e-2 * units->second); CHKERRQ(ierr); 1197 if (test) {ierr = TSSetMaxSteps(ts, 1); CHKERRQ(ierr);} 1198 ierr = TSGetAdapt(ts, &adapt); CHKERRQ(ierr); 1199 ierr = TSAdaptSetStepLimits(adapt, 1200 1.e-12 * units->second, 1201 1.e2 * units->second); CHKERRQ(ierr); 1202 ierr = TSSetFromOptions(ts); CHKERRQ(ierr); 1203 if (!contsteps) { // print initial condition 1204 if (!test) { 1205 ierr = TSMonitor_NS(ts, 0, 0., Q, user); CHKERRQ(ierr); 1206 } 1207 } else { // continue from time of last output 1208 PetscReal time; 1209 PetscInt count; 1210 PetscViewer viewer; 1211 char filepath[PETSC_MAX_PATH_LEN]; 1212 ierr = PetscSNPrintf(filepath, sizeof filepath, "%s/ns-time.bin", 1213 user->outputfolder); CHKERRQ(ierr); 1214 ierr = PetscViewerBinaryOpen(comm, filepath, FILE_MODE_READ, &viewer); 1215 CHKERRQ(ierr); 1216 ierr = PetscViewerBinaryRead(viewer, &time, 1, &count, PETSC_REAL); 1217 CHKERRQ(ierr); 1218 ierr = PetscViewerDestroy(&viewer); CHKERRQ(ierr); 1219 ierr = TSSetTime(ts, time * user->units->second); CHKERRQ(ierr); 1220 } 1221 if (!test) { 1222 ierr = TSMonitorSet(ts, TSMonitor_NS, user, NULL); CHKERRQ(ierr); 1223 } 1224 1225 // Solve 1226 start = MPI_Wtime(); 1227 ierr = PetscBarrier((PetscObject)ts); CHKERRQ(ierr); 1228 ierr = TSSolve(ts, Q); CHKERRQ(ierr); 1229 cpu_time_used = MPI_Wtime() - start; 1230 ierr = TSGetSolveTime(ts, &ftime); CHKERRQ(ierr); 1231 ierr = MPI_Allreduce(MPI_IN_PLACE, &cpu_time_used, 1, MPI_DOUBLE, MPI_MIN, 1232 comm); CHKERRQ(ierr); 1233 if (!test) { 1234 ierr = PetscPrintf(PETSC_COMM_WORLD, 1235 "Time taken for solution (sec): %g\n", 1236 (double)cpu_time_used); CHKERRQ(ierr); 1237 } 1238 1239 // Get error 1240 if (problem->non_zero_time && !test) { 1241 Vec Qexact, Qexactloc; 1242 PetscReal norm; 1243 ierr = DMCreateGlobalVector(dm, &Qexact); CHKERRQ(ierr); 1244 ierr = DMGetLocalVector(dm, &Qexactloc); CHKERRQ(ierr); 1245 ierr = VecGetSize(Qexactloc, &lnodes); CHKERRQ(ierr); 1246 1247 ierr = ICs_FixMultiplicity(op_ics, xcorners, q0ceed, dm, Qexactloc, Qexact, 1248 restrictq, &ctxSetup, ftime); CHKERRQ(ierr); 1249 1250 ierr = VecAXPY(Q, -1.0, Qexact); CHKERRQ(ierr); 1251 ierr = VecNorm(Q, NORM_MAX, &norm); CHKERRQ(ierr); 1252 CeedVectorDestroy(&q0ceed); 1253 ierr = PetscPrintf(PETSC_COMM_WORLD, 1254 "Max Error: %g\n", 1255 (double)norm); CHKERRQ(ierr); 1256 // Clean up vectors 1257 ierr = DMRestoreLocalVector(dm, &Qexactloc); CHKERRQ(ierr); 1258 ierr = VecDestroy(&Qexact); CHKERRQ(ierr); 1259 } 1260 1261 // Output Statistics 1262 ierr = TSGetStepNumber(ts,&steps); CHKERRQ(ierr); 1263 if (!test) { 1264 ierr = PetscPrintf(PETSC_COMM_WORLD, 1265 "Time integrator took %D time steps to reach final time %g\n", 1266 steps, (double)ftime); CHKERRQ(ierr); 1267 } 1268 1269 // Clean up libCEED 1270 CeedVectorDestroy(&qdata); 1271 CeedVectorDestroy(&user->qceed); 1272 CeedVectorDestroy(&user->qdotceed); 1273 CeedVectorDestroy(&user->gceed); 1274 CeedVectorDestroy(&xcorners); 1275 CeedBasisDestroy(&basisq); 1276 CeedBasisDestroy(&basisx); 1277 CeedBasisDestroy(&basisxc); 1278 CeedElemRestrictionDestroy(&restrictq); 1279 CeedElemRestrictionDestroy(&restrictx); 1280 CeedElemRestrictionDestroy(&restrictqdi); 1281 CeedElemRestrictionDestroy(&restrictxcoord); 1282 CeedQFunctionDestroy(&qf_setup); 1283 CeedQFunctionDestroy(&qf_ics); 1284 CeedQFunctionDestroy(&qf_rhs); 1285 CeedQFunctionDestroy(&qf_ifunction); 1286 CeedOperatorDestroy(&op_setup); 1287 CeedOperatorDestroy(&op_ics); 1288 CeedOperatorDestroy(&user->op_rhs); 1289 CeedOperatorDestroy(&user->op_ifunction); 1290 CeedDestroy(&ceed); 1291 1292 // Clean up PETSc 1293 ierr = VecDestroy(&Q); CHKERRQ(ierr); 1294 ierr = VecDestroy(&user->M); CHKERRQ(ierr); 1295 ierr = MatDestroy(&interpviz); CHKERRQ(ierr); 1296 ierr = DMDestroy(&dmviz); CHKERRQ(ierr); 1297 ierr = TSDestroy(&ts); CHKERRQ(ierr); 1298 ierr = DMDestroy(&dm); CHKERRQ(ierr); 1299 ierr = PetscFree(units); CHKERRQ(ierr); 1300 ierr = PetscFree(user); CHKERRQ(ierr); 1301 return PetscFinalize(); 1302 } 1303 1304