1 #include <petsc/private/dmpleximpl.h> /*I "petscdmplex.h" I*/ 2 #include <petsclandau.h> /*I "petsclandau.h" I*/ 3 #include <petscts.h> 4 #include <petscdmforest.h> 5 #include <petscdmcomposite.h> 6 7 /* Landau collision operator */ 8 9 /* relativistic terms */ 10 #if defined(PETSC_USE_REAL_SINGLE) 11 #define SPEED_OF_LIGHT 2.99792458e8F 12 #define C_0(v0) (SPEED_OF_LIGHT/v0) /* needed for relativistic tensor on all architectures */ 13 #else 14 #define SPEED_OF_LIGHT 2.99792458e8 15 #define C_0(v0) (SPEED_OF_LIGHT/v0) /* needed for relativistic tensor on all architectures */ 16 #endif 17 18 #define PETSC_THREAD_SYNC 19 #include "land_tensors.h" 20 21 #if defined(PETSC_HAVE_OPENMP) 22 #include <omp.h> 23 #endif 24 25 /* vector padding not supported */ 26 #define LANDAU_VL 1 27 28 static PetscErrorCode LandauGPUMapsDestroy(void *ptr) 29 { 30 P4estVertexMaps *maps = (P4estVertexMaps*)ptr; 31 PetscErrorCode ierr; 32 PetscFunctionBegin; 33 // free device data 34 if (maps[0].deviceType != LANDAU_CPU) { 35 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 36 if (maps[0].deviceType == LANDAU_KOKKOS) { 37 ierr = LandauKokkosDestroyMatMaps(maps, maps[0].numgrids);CHKERRQ(ierr); // imples Kokkos does 38 } // else could be CUDA 39 #elif defined(PETSC_HAVE_CUDA) 40 if (maps[0].deviceType == LANDAU_CUDA) { 41 ierr = LandauCUDADestroyMatMaps(maps, maps[0].numgrids);CHKERRQ(ierr); 42 } else SETERRQ1(PETSC_COMM_SELF, PETSC_ERR_PLIB, "maps->deviceType %D ?????",maps->deviceType); 43 #endif 44 } 45 // free host data 46 for (PetscInt grid=0 ; grid < maps[0].numgrids ; grid++) { 47 ierr = PetscFree(maps[grid].c_maps);CHKERRQ(ierr); 48 ierr = PetscFree(maps[grid].gIdx);CHKERRQ(ierr); 49 } 50 ierr = PetscFree(maps);CHKERRQ(ierr); 51 52 PetscFunctionReturn(0); 53 } 54 static PetscErrorCode energy_f(PetscInt dim, PetscReal time, const PetscReal x[], PetscInt Nf_dummy, PetscScalar *u, void *actx) 55 { 56 PetscReal v2 = 0; 57 PetscFunctionBegin; 58 /* compute v^2 / 2 */ 59 for (int i = 0; i < dim; ++i) v2 += x[i]*x[i]; 60 /* evaluate the Maxwellian */ 61 u[0] = v2/2; 62 PetscFunctionReturn(0); 63 } 64 65 /* needs double */ 66 static PetscErrorCode gamma_m1_f(PetscInt dim, PetscReal time, const PetscReal x[], PetscInt Nf_dummy, PetscScalar *u, void *actx) 67 { 68 PetscReal *c2_0_arr = ((PetscReal*)actx); 69 double u2 = 0, c02 = (double)*c2_0_arr, xx; 70 71 PetscFunctionBegin; 72 /* compute u^2 / 2 */ 73 for (int i = 0; i < dim; ++i) u2 += x[i]*x[i]; 74 /* gamma - 1 = g_eps, for conditioning and we only take derivatives */ 75 xx = u2/c02; 76 #if defined(PETSC_USE_DEBUG) 77 u[0] = PetscSqrtReal(1. + xx); 78 #else 79 u[0] = xx/(PetscSqrtReal(1. + xx) + 1.) - 1.; // better conditioned. -1 might help condition and only used for derivative 80 #endif 81 PetscFunctionReturn(0); 82 } 83 84 /* 85 LandauFormJacobian_Internal - Evaluates Jacobian matrix. 86 87 Input Parameters: 88 . globX - input vector 89 . actx - optional user-defined context 90 . dim - dimension 91 92 Output Parameters: 93 . J0acP - Jacobian matrix filled, not created 94 */ 95 static PetscErrorCode LandauFormJacobian_Internal(Vec a_X, Mat JacP, const PetscInt dim, PetscReal shift, void *a_ctx) 96 { 97 LandauCtx *ctx = (LandauCtx*)a_ctx; 98 PetscErrorCode ierr; 99 PetscInt numCells[LANDAU_MAX_GRIDS],Nq,Nb; 100 PetscQuadrature quad; 101 PetscReal Eq_m[LANDAU_MAX_SPECIES]; // could be static data w/o quench (ex2) 102 PetscScalar *cellClosure=NULL; 103 const PetscScalar *xdata=NULL; 104 PetscDS prob; 105 PetscContainer container; 106 P4estVertexMaps *maps; 107 Mat subJ[LANDAU_MAX_GRIDS*LANDAU_MAX_BATCH_SZ]; 108 109 PetscFunctionBegin; 110 PetscValidHeaderSpecific(a_X,VEC_CLASSID,1); 111 PetscValidHeaderSpecific(JacP,MAT_CLASSID,2); 112 PetscValidPointer(ctx,5); 113 /* check for matrix container for GPU assembly. Support CPU assembly for debugging */ 114 if (ctx->plex[0] == NULL) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"Plex not created"); 115 ierr = PetscLogEventBegin(ctx->events[10],0,0,0,0);CHKERRQ(ierr); 116 ierr = DMGetDS(ctx->plex[0], &prob);CHKERRQ(ierr); // same DS for all grids 117 ierr = PetscObjectQuery((PetscObject) JacP, "assembly_maps", (PetscObject *) &container);CHKERRQ(ierr); 118 if (container) { 119 if (!ctx->gpu_assembly) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"GPU matrix container but no GPU assembly"); 120 ierr = PetscContainerGetPointer(container, (void **) &maps);CHKERRQ(ierr); 121 if (!maps) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"empty GPU matrix container"); 122 for (PetscInt i=0;i<ctx->num_grids*ctx->batch_sz;i++) subJ[i] = NULL; 123 } else { 124 if (ctx->gpu_assembly) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"No GPU matrix container but GPU assembly"); 125 for (PetscInt tid=0 ; tid<ctx->batch_sz ; tid++) { 126 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 127 ierr = DMCreateMatrix(ctx->plex[grid], &subJ[ LAND_PACK_IDX(tid,grid) ]);CHKERRQ(ierr); 128 } 129 } 130 maps = NULL; 131 } 132 // get dynamic data (Eq is odd, for quench and Spitzer test) for CPU assembly and raw data for Jacobian GPU assembly. Get host numCells[], Nq (yuck) 133 ierr = PetscFEGetQuadrature(ctx->fe[0], &quad);CHKERRQ(ierr); 134 ierr = PetscQuadratureGetData(quad, NULL, NULL, &Nq, NULL, NULL);CHKERRQ(ierr); Nb = Nq; 135 if (Nq >LANDAU_MAX_NQ) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"Order too high. Nq = %D > LANDAU_MAX_NQ (%D)",Nq,LANDAU_MAX_NQ); 136 // get metadata for collecting dynamic data 137 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 138 PetscInt cStart, cEnd; 139 if (ctx->plex[grid] == NULL) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"Plex not created"); 140 ierr = DMPlexGetHeightStratum(ctx->plex[grid], 0, &cStart, &cEnd);CHKERRQ(ierr); 141 numCells[grid] = cEnd - cStart; // grids can have different topology 142 } 143 ierr = PetscLogEventEnd(ctx->events[10],0,0,0,0);CHKERRQ(ierr); 144 if (shift==0) { /* create dynamic point data: f_alpha for closure of each cell (cellClosure[nbatch,ngrids,ncells[g],f[Nb,ns[g]]]) or xdata */ 145 DM pack; 146 ierr = VecGetDM(a_X, &pack);CHKERRQ(ierr); 147 if (!pack) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "pack has no DM"); 148 ierr = PetscLogEventBegin(ctx->events[1],0,0,0,0);CHKERRQ(ierr); 149 ierr = MatZeroEntries(JacP);CHKERRQ(ierr); 150 for (PetscInt fieldA=0;fieldA<ctx->num_species;fieldA++) { 151 Eq_m[fieldA] = ctx->Ez * ctx->t_0 * ctx->charges[fieldA] / (ctx->v_0 * ctx->masses[fieldA]); /* normalize dimensionless */ 152 if (dim==2) Eq_m[fieldA] *= 2 * PETSC_PI; /* add the 2pi term that is not in Landau */ 153 } 154 if (!ctx->gpu_assembly) { 155 Vec *locXArray,*globXArray; 156 PetscScalar *cellClosure_it; 157 PetscInt cellClosure_sz=0,nDMs,Nf[LANDAU_MAX_GRIDS]; 158 PetscSection section[LANDAU_MAX_GRIDS],globsection[LANDAU_MAX_GRIDS]; 159 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 160 ierr = DMGetLocalSection(ctx->plex[grid], §ion[grid]);CHKERRQ(ierr); 161 ierr = DMGetGlobalSection(ctx->plex[grid], &globsection[grid]);CHKERRQ(ierr); 162 ierr = PetscSectionGetNumFields(section[grid], &Nf[grid]);CHKERRQ(ierr); 163 } 164 /* count cellClosure size */ 165 ierr = DMCompositeGetNumberDM(pack,&nDMs);CHKERRQ(ierr); 166 for (PetscInt grid=0 ; grid<ctx->num_grids ; grid++) cellClosure_sz += Nb*Nf[grid]*numCells[grid]; 167 ierr = PetscMalloc1(cellClosure_sz*ctx->batch_sz,&cellClosure);CHKERRQ(ierr); 168 cellClosure_it = cellClosure; 169 ierr = PetscMalloc(sizeof(*locXArray)*nDMs, &locXArray);CHKERRQ(ierr); 170 ierr = PetscMalloc(sizeof(*globXArray)*nDMs, &globXArray);CHKERRQ(ierr); 171 ierr = DMCompositeGetLocalAccessArray(pack, a_X, nDMs, NULL, locXArray);CHKERRQ(ierr); 172 ierr = DMCompositeGetAccessArray(pack, a_X, nDMs, NULL, globXArray);CHKERRQ(ierr); 173 for (PetscInt b_id = 0 ; b_id < ctx->batch_sz ; b_id++) { // OpenMP (once) 174 for (PetscInt grid=0 ; grid<ctx->num_grids ; grid++) { 175 Vec locX = locXArray[ LAND_PACK_IDX(b_id,grid) ], globX = globXArray[ LAND_PACK_IDX(b_id,grid) ], locX2; 176 PetscInt cStart, cEnd, ei; 177 ierr = VecDuplicate(locX,&locX2);CHKERRQ(ierr); 178 ierr = DMGlobalToLocalBegin(ctx->plex[grid], globX, INSERT_VALUES, locX2);CHKERRQ(ierr); 179 ierr = DMGlobalToLocalEnd (ctx->plex[grid], globX, INSERT_VALUES, locX2);CHKERRQ(ierr); 180 ierr = DMPlexGetHeightStratum(ctx->plex[grid], 0, &cStart, &cEnd);CHKERRQ(ierr); 181 for (ei = cStart ; ei < cEnd; ++ei) { 182 PetscScalar *coef = NULL; 183 ierr = DMPlexVecGetClosure(ctx->plex[grid], section[grid], locX2, ei, NULL, &coef);CHKERRQ(ierr); 184 ierr = PetscMemcpy(cellClosure_it,coef,Nb*Nf[grid]*sizeof(*cellClosure_it));CHKERRQ(ierr); /* change if LandauIPReal != PetscScalar */ 185 ierr = DMPlexVecRestoreClosure(ctx->plex[grid], section[grid], locX2, ei, NULL, &coef);CHKERRQ(ierr); 186 cellClosure_it += Nb*Nf[grid]; 187 } 188 ierr = VecDestroy(&locX2);CHKERRQ(ierr); 189 } 190 } 191 if (cellClosure_it-cellClosure != cellClosure_sz*ctx->batch_sz) SETERRQ2(PETSC_COMM_SELF, PETSC_ERR_PLIB, "iteration wrong %D != cellClosure_sz = %D",cellClosure_it-cellClosure,cellClosure_sz*ctx->batch_sz); 192 ierr = DMCompositeRestoreLocalAccessArray(pack, a_X, nDMs, NULL, locXArray);CHKERRQ(ierr); 193 ierr = DMCompositeRestoreAccessArray(pack, a_X, nDMs, NULL, globXArray);CHKERRQ(ierr); 194 ierr = PetscFree(locXArray);CHKERRQ(ierr); 195 ierr = PetscFree(globXArray);CHKERRQ(ierr); 196 xdata = NULL; 197 } else { 198 PetscMemType mtype; 199 ierr = VecGetArrayReadAndMemType(a_X,&xdata,&mtype);CHKERRQ(ierr); 200 if (mtype!=PETSC_MEMTYPE_HOST && ctx->deviceType == LANDAU_CPU) { 201 SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"CPU run with device data: use -mat_type aij"); 202 } 203 cellClosure = NULL; 204 } 205 ierr = PetscLogEventEnd(ctx->events[1],0,0,0,0);CHKERRQ(ierr); 206 } else xdata = cellClosure = NULL; 207 208 /* do it */ 209 if (ctx->deviceType == LANDAU_CUDA || ctx->deviceType == LANDAU_KOKKOS) { 210 if (ctx->deviceType == LANDAU_CUDA) { 211 #if defined(PETSC_HAVE_CUDA) 212 ierr = LandauCUDAJacobian(ctx->plex,Nq,ctx->batch_sz,ctx->num_grids,numCells,Eq_m,cellClosure,xdata,&ctx->SData_d,ctx->subThreadBlockSize,shift,ctx->events,ctx->mat_offset, ctx->species_offset, subJ, JacP);CHKERRQ(ierr); 213 #else 214 SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-landau_device_type %s not built","cuda"); 215 #endif 216 } else if (ctx->deviceType == LANDAU_KOKKOS) { 217 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 218 ierr = LandauKokkosJacobian(ctx->plex,Nq,ctx->batch_sz,ctx->num_grids,numCells,Eq_m,cellClosure,xdata,&ctx->SData_d,ctx->subThreadBlockSize,shift,ctx->events,ctx->mat_offset, ctx->species_offset, subJ,JacP);CHKERRQ(ierr); 219 #else 220 SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-landau_device_type %s not built","kokkos"); 221 #endif 222 } 223 } else { /* CPU version */ 224 PetscTabulation *Tf; // used for CPU and print info. Same on all grids and all species 225 PetscInt ip_offset[LANDAU_MAX_GRIDS+1], ipf_offset[LANDAU_MAX_GRIDS+1], elem_offset[LANDAU_MAX_GRIDS+1],IPf_sz_glb,IPf_sz_tot,num_grids=ctx->num_grids,Nf[LANDAU_MAX_GRIDS]; 226 PetscReal *ff, *dudx, *dudy, *dudz, *invJ_a = (PetscReal*)ctx->SData_d.invJ, *xx = (PetscReal*)ctx->SData_d.x, *yy = (PetscReal*)ctx->SData_d.y, *zz = (PetscReal*)ctx->SData_d.z, *ww = (PetscReal*)ctx->SData_d.w; 227 PetscReal Eq_m[LANDAU_MAX_SPECIES], invMass[LANDAU_MAX_SPECIES], nu_alpha[LANDAU_MAX_SPECIES], nu_beta[LANDAU_MAX_SPECIES]; 228 PetscSection section[LANDAU_MAX_GRIDS],globsection[LANDAU_MAX_GRIDS]; 229 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 230 ierr = DMGetLocalSection(ctx->plex[grid], §ion[grid]);CHKERRQ(ierr); 231 ierr = DMGetGlobalSection(ctx->plex[grid], &globsection[grid]);CHKERRQ(ierr); 232 ierr = PetscSectionGetNumFields(section[grid], &Nf[grid]);CHKERRQ(ierr); 233 } 234 /* count IPf size, etc */ 235 ierr = PetscDSGetTabulation(prob, &Tf);CHKERRQ(ierr); // Bf, &Df same for all grids 236 const PetscReal *const BB = Tf[0]->T[0], * const DD = Tf[0]->T[1]; 237 ip_offset[0] = ipf_offset[0] = elem_offset[0] = 0; 238 for (PetscInt grid=0 ; grid<num_grids ; grid++) { 239 PetscInt nfloc = ctx->species_offset[grid+1] - ctx->species_offset[grid]; 240 elem_offset[grid+1] = elem_offset[grid] + numCells[grid]; 241 ip_offset[grid+1] = ip_offset[grid] + numCells[grid]*Nq; 242 ipf_offset[grid+1] = ipf_offset[grid] + Nq*nfloc*numCells[grid]; 243 } 244 IPf_sz_glb = ipf_offset[num_grids]; 245 IPf_sz_tot = IPf_sz_glb*ctx->batch_sz; 246 if (shift==0.0) { /* compute dynamic data f and df and init data for Jacobian */ 247 #if defined(PETSC_HAVE_THREADSAFETY) 248 double starttime, endtime; 249 starttime = MPI_Wtime(); 250 #endif 251 ierr = PetscLogEventBegin(ctx->events[8],0,0,0,0);CHKERRQ(ierr); 252 for (PetscInt fieldA=0;fieldA<ctx->num_species;fieldA++) { 253 invMass[fieldA] = ctx->m_0/ctx->masses[fieldA]; 254 Eq_m[fieldA] = ctx->Ez * ctx->t_0 * ctx->charges[fieldA] / (ctx->v_0 * ctx->masses[fieldA]); /* normalize dimensionless */ 255 if (dim==2) Eq_m[fieldA] *= 2 * PETSC_PI; /* add the 2pi term that is not in Landau */ 256 nu_alpha[fieldA] = PetscSqr(ctx->charges[fieldA]/ctx->m_0)*ctx->m_0/ctx->masses[fieldA]; 257 nu_beta[fieldA] = PetscSqr(ctx->charges[fieldA]/ctx->epsilon0)*ctx->lnLam / (8*PETSC_PI) * ctx->t_0*ctx->n_0/PetscPowReal(ctx->v_0,3); 258 } 259 ierr = PetscMalloc4(IPf_sz_tot, &ff, IPf_sz_tot, &dudx, IPf_sz_tot, &dudy, dim==3 ? IPf_sz_tot : 0, &dudz);CHKERRQ(ierr); 260 // F df/dx 261 for (PetscInt tid = 0 ; tid < ctx->batch_sz*elem_offset[num_grids] ; tid++) { // for each element 262 const PetscInt b_Nelem = elem_offset[num_grids], b_elem_idx = tid%b_Nelem, b_id = tid/b_Nelem; // b_id == OMP thd_id in batch 263 // find my grid: 264 PetscInt grid = 0; 265 while (b_elem_idx >= elem_offset[grid+1]) grid++; // yuck search for grid 266 { 267 const PetscInt loc_nip = numCells[grid]*Nq, loc_Nf = ctx->species_offset[grid+1] - ctx->species_offset[grid], loc_elem = b_elem_idx - elem_offset[grid]; 268 const PetscInt moffset = LAND_MOFFSET(b_id,grid,ctx->batch_sz,ctx->num_grids,ctx->mat_offset); //b_id*b_N + ctx->mat_offset[grid]; 269 PetscScalar *coef, coef_buff[LANDAU_MAX_SPECIES*LANDAU_MAX_NQ]; 270 PetscReal *invJe = &invJ_a[(ip_offset[grid] + loc_elem*Nq)*dim*dim]; // ingJ is static data on batch 0 271 PetscInt b,f,q; 272 if (cellClosure) { 273 coef = &cellClosure[b_id*IPf_sz_glb + ipf_offset[grid] + loc_elem*Nb*loc_Nf]; // this is const 274 } else { 275 coef = coef_buff; 276 for (f = 0; f < loc_Nf; ++f) { 277 LandauIdx *const Idxs = &maps[grid].gIdx[loc_elem][f][0]; 278 for (b = 0; b < Nb; ++b) { 279 PetscInt idx = Idxs[b]; 280 if (idx >= 0) { 281 coef[f*Nb+b] = xdata[idx+moffset]; 282 } else { 283 idx = -idx - 1; 284 coef[f*Nb+b] = 0; 285 for (q = 0; q < maps[grid].num_face; q++) { 286 PetscInt id = maps[grid].c_maps[idx][q].gid; 287 PetscScalar scale = maps[grid].c_maps[idx][q].scale; 288 coef[f*Nb+b] += scale*xdata[id+moffset]; 289 } 290 } 291 } 292 } 293 } 294 /* get f and df */ 295 for (PetscInt qi = 0; qi < Nq; qi++) { 296 const PetscReal *invJ = &invJe[qi*dim*dim]; 297 const PetscReal *Bq = &BB[qi*Nb]; 298 const PetscReal *Dq = &DD[qi*Nb*dim]; 299 PetscReal u_x[LANDAU_DIM]; 300 /* get f & df */ 301 for (f = 0; f < loc_Nf; ++f) { 302 const PetscInt idx = b_id*IPf_sz_glb + ipf_offset[grid] + f*loc_nip + loc_elem*Nq + qi; 303 PetscInt b, e; 304 PetscReal refSpaceDer[LANDAU_DIM]; 305 ff[idx] = 0.0; 306 for (int d = 0; d < LANDAU_DIM; ++d) refSpaceDer[d] = 0.0; 307 for (b = 0; b < Nb; ++b) { 308 const PetscInt cidx = b; 309 ff[idx] += Bq[cidx]*PetscRealPart(coef[f*Nb+cidx]); 310 for (int d = 0; d < dim; ++d) { 311 refSpaceDer[d] += Dq[cidx*dim+d]*PetscRealPart(coef[f*Nb+cidx]); 312 } 313 } 314 for (int d = 0; d < LANDAU_DIM; ++d) { 315 for (e = 0, u_x[d] = 0.0; e < LANDAU_DIM; ++e) { 316 u_x[d] += invJ[e*dim+d]*refSpaceDer[e]; 317 } 318 } 319 dudx[idx] = u_x[0]; 320 dudy[idx] = u_x[1]; 321 #if LANDAU_DIM==3 322 dudz[idx] = u_x[2]; 323 #endif 324 } 325 } // q 326 } // grid 327 } // grid*batch 328 ierr = PetscLogEventEnd(ctx->events[8],0,0,0,0);CHKERRQ(ierr); 329 #if defined(PETSC_HAVE_THREADSAFETY) 330 endtime = MPI_Wtime(); 331 if (ctx->stage) ctx->times[LANDAU_F_DF] += (endtime - starttime); 332 #endif 333 } // Jacobian setup 334 /* doit it */ 335 for (PetscInt tid = 0 ; tid < ctx->batch_sz*elem_offset[num_grids] ; tid++) { // for each element 336 const PetscInt b_Nelem = elem_offset[num_grids]; 337 const PetscInt b_elem_idx = tid%b_Nelem, b_id = tid/b_Nelem; 338 PetscInt grid = 0; 339 #if defined(PETSC_HAVE_THREADSAFETY) 340 double starttime, endtime; 341 starttime = MPI_Wtime(); 342 #endif 343 while (b_elem_idx >= elem_offset[grid+1]) grid++; 344 { 345 const PetscInt loc_Nf = ctx->species_offset[grid+1] - ctx->species_offset[grid], loc_elem = b_elem_idx - elem_offset[grid]; 346 const PetscInt moffset = LAND_MOFFSET(b_id,grid,ctx->batch_sz,ctx->num_grids,ctx->mat_offset)/* ; b_id*b_N + ctx->mat_offset[grid] */, totDim = loc_Nf*Nq, elemMatSize = totDim*totDim; 347 PetscScalar *elemMat; 348 const PetscReal *invJe = &invJ_a[(ip_offset[grid] + loc_elem*Nq)*dim*dim]; 349 ierr = PetscMalloc1(elemMatSize, &elemMat);CHKERRQ(ierr); 350 ierr = PetscMemzero(elemMat, elemMatSize*sizeof(*elemMat));CHKERRQ(ierr); 351 ierr = PetscLogEventBegin(ctx->events[4],0,0,0,0);CHKERRQ(ierr); 352 for (PetscInt qj = 0; qj < Nq; ++qj) { 353 const PetscInt jpidx_glb = ip_offset[grid] + qj + loc_elem * Nq; 354 PetscReal g0[LANDAU_MAX_SPECIES], g2[LANDAU_MAX_SPECIES][LANDAU_DIM], g3[LANDAU_MAX_SPECIES][LANDAU_DIM][LANDAU_DIM]; // could make a LANDAU_MAX_SPECIES_GRID ~ number of ions - 1 355 PetscInt d,d2,dp,d3,IPf_idx; 356 if (shift==0.0) { // Jacobian 357 const PetscReal * const invJj = &invJe[qj*dim*dim]; 358 PetscReal gg2[LANDAU_MAX_SPECIES][LANDAU_DIM],gg3[LANDAU_MAX_SPECIES][LANDAU_DIM][LANDAU_DIM], gg2_temp[LANDAU_DIM], gg3_temp[LANDAU_DIM][LANDAU_DIM]; 359 const PetscReal vj[3] = {xx[jpidx_glb], yy[jpidx_glb], zz ? zz[jpidx_glb] : 0}, wj = ww[jpidx_glb]; 360 // create g2 & g3 361 for (d=0;d<LANDAU_DIM;d++) { // clear accumulation data D & K 362 gg2_temp[d] = 0; 363 for (d2=0;d2<LANDAU_DIM;d2++) gg3_temp[d][d2] = 0; 364 } 365 /* inner beta reduction */ 366 IPf_idx = 0; 367 for (PetscInt grid_r = 0, f_off = 0, ipidx = 0; grid_r < ctx->num_grids ; grid_r++, f_off = ctx->species_offset[grid_r]) { // IPf_idx += nip_loc_r*Nfloc_r 368 PetscInt nip_loc_r = numCells[grid_r]*Nq, Nfloc_r = Nf[grid_r]; 369 for (PetscInt ei_r = 0, loc_fdf_idx = 0; ei_r < numCells[grid_r]; ++ei_r) { 370 for (PetscInt qi = 0; qi < Nq; qi++, ipidx++, loc_fdf_idx++) { 371 const PetscReal wi = ww[ipidx], x = xx[ipidx], y = yy[ipidx]; 372 PetscReal temp1[3] = {0, 0, 0}, temp2 = 0; 373 #if LANDAU_DIM==2 374 PetscReal Ud[2][2], Uk[2][2], mask = (PetscAbs(vj[0]-x) < 100*PETSC_SQRT_MACHINE_EPSILON && PetscAbs(vj[1]-y) < 100*PETSC_SQRT_MACHINE_EPSILON) ? 0. : 1.; 375 LandauTensor2D(vj, x, y, Ud, Uk, mask); 376 #else 377 PetscReal U[3][3], z = zz[ipidx], mask = (PetscAbs(vj[0]-x) < 100*PETSC_SQRT_MACHINE_EPSILON && PetscAbs(vj[1]-y) < 100*PETSC_SQRT_MACHINE_EPSILON && PetscAbs(vj[2]-z) < 100*PETSC_SQRT_MACHINE_EPSILON) ? 0. : 1.; 378 if (ctx->use_relativistic_corrections) { 379 LandauTensor3DRelativistic(vj, x, y, z, U, mask, C_0(ctx->v_0)); 380 } else { 381 LandauTensor3D(vj, x, y, z, U, mask); 382 } 383 #endif 384 for (int f = 0; f < Nfloc_r ; ++f) { 385 const PetscInt idx = b_id*IPf_sz_glb + ipf_offset[grid_r] + f*nip_loc_r + ei_r*Nq + qi; // IPf_idx + f*nip_loc_r + loc_fdf_idx; 386 temp1[0] += dudx[idx]*nu_beta[f+f_off]*invMass[f+f_off]; 387 temp1[1] += dudy[idx]*nu_beta[f+f_off]*invMass[f+f_off]; 388 #if LANDAU_DIM==3 389 temp1[2] += dudz[idx]*nu_beta[f+f_off]*invMass[f+f_off]; 390 #endif 391 temp2 += ff[idx]*nu_beta[f+f_off]; 392 } 393 temp1[0] *= wi; 394 temp1[1] *= wi; 395 #if LANDAU_DIM==3 396 temp1[2] *= wi; 397 #endif 398 temp2 *= wi; 399 #if LANDAU_DIM==2 400 for (d2 = 0; d2 < 2; d2++) { 401 for (d3 = 0; d3 < 2; ++d3) { 402 /* K = U * grad(f): g2=e: i,A */ 403 gg2_temp[d2] += Uk[d2][d3]*temp1[d3]; 404 /* D = -U * (I \kron (fx)): g3=f: i,j,A */ 405 gg3_temp[d2][d3] += Ud[d2][d3]*temp2; 406 } 407 } 408 #else 409 for (d2 = 0; d2 < 3; ++d2) { 410 for (d3 = 0; d3 < 3; ++d3) { 411 /* K = U * grad(f): g2 = e: i,A */ 412 gg2_temp[d2] += U[d2][d3]*temp1[d3]; 413 /* D = -U * (I \kron (fx)): g3 = f: i,j,A */ 414 gg3_temp[d2][d3] += U[d2][d3]*temp2; 415 } 416 } 417 #endif 418 } // qi 419 } // ei_r 420 IPf_idx += nip_loc_r*Nfloc_r; 421 } /* grid_r - IPs */ 422 if (IPf_idx != IPf_sz_glb) SETERRQ2(PETSC_COMM_SELF, PETSC_ERR_PLIB, "IPf_idx != IPf_sz %D %D",IPf_idx,IPf_sz_glb); 423 // add alpha and put in gg2/3 424 for (PetscInt fieldA = 0, f_off = ctx->species_offset[grid]; fieldA < loc_Nf; ++fieldA) { 425 for (d2 = 0; d2 < dim; d2++) { 426 gg2[fieldA][d2] = gg2_temp[d2]*nu_alpha[fieldA+f_off]; 427 for (d3 = 0; d3 < dim; d3++) { 428 gg3[fieldA][d2][d3] = -gg3_temp[d2][d3]*nu_alpha[fieldA+f_off]*invMass[fieldA+f_off]; 429 } 430 } 431 } 432 /* add electric field term once per IP */ 433 for (PetscInt fieldA = 0, f_off = ctx->species_offset[grid] ; fieldA < loc_Nf; ++fieldA) { 434 gg2[fieldA][dim-1] += Eq_m[fieldA+f_off]; 435 } 436 /* Jacobian transform - g2, g3 */ 437 for (PetscInt fieldA = 0; fieldA < loc_Nf; ++fieldA) { 438 for (d = 0; d < dim; ++d) { 439 g2[fieldA][d] = 0.0; 440 for (d2 = 0; d2 < dim; ++d2) { 441 g2[fieldA][d] += invJj[d*dim+d2]*gg2[fieldA][d2]; 442 g3[fieldA][d][d2] = 0.0; 443 for (d3 = 0; d3 < dim; ++d3) { 444 for (dp = 0; dp < dim; ++dp) { 445 g3[fieldA][d][d2] += invJj[d*dim + d3]*gg3[fieldA][d3][dp]*invJj[d2*dim + dp]; 446 } 447 } 448 g3[fieldA][d][d2] *= wj; 449 } 450 g2[fieldA][d] *= wj; 451 } 452 } 453 } else { // mass 454 PetscReal wj = ww[jpidx_glb]; 455 /* Jacobian transform - g0 */ 456 for (PetscInt fieldA = 0; fieldA < loc_Nf ; ++fieldA) { 457 if (dim==2) { 458 g0[fieldA] = wj * shift * 2. * PETSC_PI; // move this to below and remove g0 459 } else { 460 g0[fieldA] = wj * shift; // move this to below and remove g0 461 } 462 } 463 } 464 /* FE matrix construction */ 465 { 466 PetscInt fieldA,d,f,d2,g; 467 const PetscReal *BJq = &BB[qj*Nb], *DIq = &DD[qj*Nb*dim]; 468 /* assemble - on the diagonal (I,I) */ 469 for (fieldA = 0; fieldA < loc_Nf ; fieldA++) { 470 for (f = 0; f < Nb ; f++) { 471 const PetscInt i = fieldA*Nb + f; /* Element matrix row */ 472 for (g = 0; g < Nb; ++g) { 473 const PetscInt j = fieldA*Nb + g; /* Element matrix column */ 474 const PetscInt fOff = i*totDim + j; 475 if (shift==0.0) { 476 for (d = 0; d < dim; ++d) { 477 elemMat[fOff] += DIq[f*dim+d]*g2[fieldA][d]*BJq[g]; 478 for (d2 = 0; d2 < dim; ++d2) { 479 elemMat[fOff] += DIq[f*dim + d]*g3[fieldA][d][d2]*DIq[g*dim + d2]; 480 } 481 } 482 } else { // mass 483 elemMat[fOff] += BJq[f]*g0[fieldA]*BJq[g]; 484 } 485 } 486 } 487 } 488 } 489 } /* qj loop */ 490 ierr = PetscLogEventEnd(ctx->events[4],0,0,0,0);CHKERRQ(ierr); 491 #if defined(PETSC_HAVE_THREADSAFETY) 492 endtime = MPI_Wtime(); 493 if (ctx->stage) ctx->times[LANDAU_KERNEL] += (endtime - starttime); 494 #endif 495 /* assemble matrix */ 496 if (!container) { 497 PetscInt cStart; 498 ierr = PetscLogEventBegin(ctx->events[6],0,0,0,0);CHKERRQ(ierr); 499 ierr = DMPlexGetHeightStratum(ctx->plex[grid], 0, &cStart, NULL);CHKERRQ(ierr); 500 ierr = DMPlexMatSetClosure(ctx->plex[grid], section[grid], globsection[grid], subJ[ LAND_PACK_IDX(b_id,grid) ], loc_elem + cStart, elemMat, ADD_VALUES);CHKERRQ(ierr); 501 ierr = PetscLogEventEnd(ctx->events[6],0,0,0,0);CHKERRQ(ierr); 502 } else { // GPU like assembly for debugging 503 PetscInt fieldA,idx,q,f,g,d,nr,nc,rows0[LANDAU_MAX_Q_FACE],cols0[LANDAU_MAX_Q_FACE]={0},rows[LANDAU_MAX_Q_FACE],cols[LANDAU_MAX_Q_FACE]; 504 PetscScalar vals[LANDAU_MAX_Q_FACE*LANDAU_MAX_Q_FACE],row_scale[LANDAU_MAX_Q_FACE],col_scale[LANDAU_MAX_Q_FACE]={0}; 505 /* assemble - from the diagonal (I,I) in this format for DMPlexMatSetClosure */ 506 for (fieldA = 0; fieldA < loc_Nf ; fieldA++) { 507 LandauIdx *const Idxs = &maps[grid].gIdx[loc_elem][fieldA][0]; 508 for (f = 0; f < Nb ; f++) { 509 idx = Idxs[f]; 510 if (idx >= 0) { 511 nr = 1; 512 rows0[0] = idx; 513 row_scale[0] = 1.; 514 } else { 515 idx = -idx - 1; 516 nr = maps[grid].num_face; 517 for (q = 0; q < maps[grid].num_face; q++) { 518 rows0[q] = maps[grid].c_maps[idx][q].gid; 519 row_scale[q] = maps[grid].c_maps[idx][q].scale; 520 } 521 } 522 for (g = 0; g < Nb; ++g) { 523 idx = Idxs[g]; 524 if (idx >= 0) { 525 nc = 1; 526 cols0[0] = idx; 527 col_scale[0] = 1.; 528 } else { 529 idx = -idx - 1; 530 nc = maps[grid].num_face; 531 for (q = 0; q < maps[grid].num_face; q++) { 532 cols0[q] = maps[grid].c_maps[idx][q].gid; 533 col_scale[q] = maps[grid].c_maps[idx][q].scale; 534 } 535 } 536 const PetscInt i = fieldA*Nb + f; /* Element matrix row */ 537 const PetscInt j = fieldA*Nb + g; /* Element matrix column */ 538 const PetscScalar Aij = elemMat[i*totDim + j]; 539 for (q = 0; q < nr; q++) rows[q] = rows0[q] + moffset; 540 for (d = 0; d < nc; d++) cols[d] = cols0[d] + moffset; 541 for (q = 0; q < nr; q++) { 542 for (d = 0; d < nc; d++) { 543 vals[q*nc + d] = row_scale[q]*col_scale[d]*Aij; 544 } 545 } 546 ierr = MatSetValues(JacP,nr,rows,nc,cols,vals,ADD_VALUES);CHKERRQ(ierr); 547 } 548 } 549 } 550 } 551 if (loc_elem==-1) { 552 PetscErrorCode ierr2; 553 ierr2 = PetscPrintf(ctx->comm,"CPU Element matrix\n");CHKERRQ(ierr2); 554 for (int d = 0; d < totDim; ++d) { 555 for (int f = 0; f < totDim; ++f) {ierr2 = PetscPrintf(ctx->comm," %12.5e", PetscRealPart(elemMat[d*totDim + f]));CHKERRQ(ierr2);} 556 ierr2 = PetscPrintf(ctx->comm,"\n");CHKERRQ(ierr2); 557 } 558 exit(12); 559 } 560 ierr = PetscFree(elemMat);CHKERRQ(ierr); 561 } /* grid */ 562 } /* outer element & batch loop */ 563 if (shift==0.0) { // mass 564 ierr = PetscFree4(ff, dudx, dudy, dudz);CHKERRQ(ierr); 565 } 566 if (!container) { // move nest matrix to global JacP 567 for (PetscInt b_id = 0 ; b_id < ctx->batch_sz ; b_id++) { // OpenMP 568 for (PetscInt grid=0 ; grid<ctx->num_grids ; grid++) { 569 const PetscInt moffset = LAND_MOFFSET(b_id,grid,ctx->batch_sz,ctx->num_grids,ctx->mat_offset); // b_id*b_N + ctx->mat_offset[grid]; 570 PetscInt nloc, nzl, colbuf[1024], row; 571 const PetscInt *cols; 572 const PetscScalar *vals; 573 Mat B = subJ[ LAND_PACK_IDX(b_id,grid) ]; 574 ierr = MatAssemblyBegin(B, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 575 ierr = MatAssemblyEnd(B, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 576 ierr = MatGetSize(B, &nloc, NULL);CHKERRQ(ierr); 577 for (int i=0 ; i<nloc ; i++) { 578 ierr = MatGetRow(B,i,&nzl,&cols,&vals);CHKERRQ(ierr); 579 if (nzl>1024) SETERRQ1(PetscObjectComm((PetscObject) B), PETSC_ERR_PLIB, "Row too big: %D",nzl); 580 for (int j=0; j<nzl; j++) colbuf[j] = cols[j] + moffset; 581 row = i + moffset; 582 ierr = MatSetValues(JacP,1,&row,nzl,colbuf,vals,ADD_VALUES);CHKERRQ(ierr); 583 ierr = MatRestoreRow(B,i,&nzl,&cols,&vals);CHKERRQ(ierr); 584 } 585 ierr = MatDestroy(&B);CHKERRQ(ierr); 586 } 587 } 588 } 589 } /* CPU version */ 590 591 ierr = MatAssemblyBegin(JacP, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 592 ierr = MatAssemblyEnd(JacP, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 593 /* clean up */ 594 if (cellClosure) { 595 ierr = PetscFree(cellClosure);CHKERRQ(ierr); 596 } 597 if (xdata) { 598 ierr = VecRestoreArrayReadAndMemType(a_X,&xdata);CHKERRQ(ierr); 599 } 600 PetscFunctionReturn(0); 601 } 602 603 #if defined(LANDAU_ADD_BCS) 604 static void zero_bc(PetscInt dim, PetscInt Nf, PetscInt NfAux, 605 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 606 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 607 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar uexact[]) 608 { 609 uexact[0] = 0; 610 } 611 #endif 612 613 #define MATVEC2(__a,__x,__p) {int i,j; for (i=0.; i<2; i++) {__p[i] = 0; for (j=0.; j<2; j++) __p[i] += __a[i][j]*__x[j]; }} 614 static void CircleInflate(PetscReal r1, PetscReal r2, PetscReal r0, PetscInt num_sections, PetscReal x, PetscReal y, 615 PetscReal *outX, PetscReal *outY) 616 { 617 PetscReal rr = PetscSqrtReal(x*x + y*y), outfact, efact; 618 if (rr < r1 + PETSC_SQRT_MACHINE_EPSILON) { 619 *outX = x; *outY = y; 620 } else { 621 const PetscReal xy[2] = {x,y}, sinphi=y/rr, cosphi=x/rr; 622 PetscReal cth,sth,xyprime[2],Rth[2][2],rotcos,newrr; 623 if (num_sections==2) { 624 rotcos = 0.70710678118654; 625 outfact = 1.5; efact = 2.5; 626 /* rotate normalized vector into [-pi/4,pi/4) */ 627 if (sinphi >= 0.) { /* top cell, -pi/2 */ 628 cth = 0.707106781186548; sth = -0.707106781186548; 629 } else { /* bottom cell -pi/8 */ 630 cth = 0.707106781186548; sth = .707106781186548; 631 } 632 } else if (num_sections==3) { 633 rotcos = 0.86602540378443; 634 outfact = 1.5; efact = 2.5; 635 /* rotate normalized vector into [-pi/6,pi/6) */ 636 if (sinphi >= 0.5) { /* top cell, -pi/3 */ 637 cth = 0.5; sth = -0.866025403784439; 638 } else if (sinphi >= -.5) { /* mid cell 0 */ 639 cth = 1.; sth = .0; 640 } else { /* bottom cell +pi/3 */ 641 cth = 0.5; sth = 0.866025403784439; 642 } 643 } else if (num_sections==4) { 644 rotcos = 0.9238795325112; 645 outfact = 1.5; efact = 3; 646 /* rotate normalized vector into [-pi/8,pi/8) */ 647 if (sinphi >= 0.707106781186548) { /* top cell, -3pi/8 */ 648 cth = 0.38268343236509; sth = -0.923879532511287; 649 } else if (sinphi >= 0.) { /* mid top cell -pi/8 */ 650 cth = 0.923879532511287; sth = -.38268343236509; 651 } else if (sinphi >= -0.707106781186548) { /* mid bottom cell + pi/8 */ 652 cth = 0.923879532511287; sth = 0.38268343236509; 653 } else { /* bottom cell + 3pi/8 */ 654 cth = 0.38268343236509; sth = .923879532511287; 655 } 656 } else { 657 cth = 0.; sth = 0.; rotcos = 0; efact = 0; 658 } 659 Rth[0][0] = cth; Rth[0][1] =-sth; 660 Rth[1][0] = sth; Rth[1][1] = cth; 661 MATVEC2(Rth,xy,xyprime); 662 if (num_sections==2) { 663 newrr = xyprime[0]/rotcos; 664 } else { 665 PetscReal newcosphi=xyprime[0]/rr, rin = r1, rout = rr - rin; 666 PetscReal routmax = r0*rotcos/newcosphi - rin, nroutmax = r0 - rin, routfrac = rout/routmax; 667 newrr = rin + routfrac*nroutmax; 668 } 669 *outX = cosphi*newrr; *outY = sinphi*newrr; 670 /* grade */ 671 PetscReal fact,tt,rs,re, rr = PetscSqrtReal(PetscSqr(*outX) + PetscSqr(*outY)); 672 if (rr > r2) { rs = r2; re = r0; fact = outfact;} /* outer zone */ 673 else { rs = r1; re = r2; fact = efact;} /* electron zone */ 674 tt = (rs + PetscPowReal((rr - rs)/(re - rs),fact) * (re-rs)) / rr; 675 *outX *= tt; 676 *outY *= tt; 677 } 678 } 679 680 static PetscErrorCode GeometryDMLandau(DM base, PetscInt point, PetscInt dim, const PetscReal abc[], PetscReal xyz[], void *a_ctx) 681 { 682 LandauCtx *ctx = (LandauCtx*)a_ctx; 683 PetscReal r = abc[0], z = abc[1]; 684 if (ctx->inflate) { 685 PetscReal absR, absZ; 686 absR = PetscAbs(r); 687 absZ = PetscAbs(z); 688 CircleInflate(ctx->i_radius[0],ctx->e_radius,ctx->radius[0],ctx->num_sections,absR,absZ,&absR,&absZ); // wrong: how do I know what grid I am on? 689 r = (r > 0) ? absR : -absR; 690 z = (z > 0) ? absZ : -absZ; 691 } 692 xyz[0] = r; 693 xyz[1] = z; 694 if (dim==3) xyz[2] = abc[2]; 695 696 PetscFunctionReturn(0); 697 } 698 699 /* create DMComposite of meshes for each species group */ 700 static PetscErrorCode LandauDMCreateVMeshes(MPI_Comm comm_self, const PetscInt dim, const char prefix[], LandauCtx *ctx, DM pack) 701 { 702 PetscErrorCode ierr; 703 704 PetscFunctionBegin; 705 { /* p4est, quads */ 706 /* Create plex mesh of Landau domain */ 707 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 708 PetscReal radius = ctx->radius[grid]; 709 if (!ctx->sphere) { 710 PetscInt cells[] = {2,2,2}; 711 PetscReal lo[] = {-radius,-radius,-radius}, hi[] = {radius,radius,radius}; 712 DMBoundaryType periodicity[3] = {DM_BOUNDARY_NONE, dim==2 ? DM_BOUNDARY_NONE : DM_BOUNDARY_NONE, DM_BOUNDARY_NONE}; 713 if (dim==2) { lo[0] = 0; cells[0] = /* cells[1] = */ 1; } 714 ierr = DMPlexCreateBoxMesh(comm_self, dim, PETSC_FALSE, cells, lo, hi, periodicity, PETSC_TRUE, &ctx->plex[grid]);CHKERRQ(ierr); // todo: make composite and create dm[grid] here 715 ierr = DMLocalizeCoordinates(ctx->plex[grid]);CHKERRQ(ierr); /* needed for periodic */ 716 if (dim==3) {ierr = PetscObjectSetName((PetscObject) ctx->plex[grid], "cube");CHKERRQ(ierr);} 717 else {ierr = PetscObjectSetName((PetscObject) ctx->plex[grid], "half-plane");CHKERRQ(ierr);} 718 } else if (dim==2) { // sphere is all wrong. should just have one inner radius 719 PetscInt numCells,cells[16][4],i,j; 720 PetscInt numVerts; 721 PetscReal inner_radius1 = ctx->i_radius[grid], inner_radius2 = ctx->e_radius; 722 PetscReal *flatCoords = NULL; 723 PetscInt *flatCells = NULL, *pcell; 724 if (ctx->num_sections==2) { 725 #if 1 726 numCells = 5; 727 numVerts = 10; 728 int cells2[][4] = { {0,1,4,3}, 729 {1,2,5,4}, 730 {3,4,7,6}, 731 {4,5,8,7}, 732 {6,7,8,9} }; 733 for (i = 0; i < numCells; i++) for (j = 0; j < 4; j++) cells[i][j] = cells2[i][j]; 734 ierr = PetscMalloc2(numVerts * 2, &flatCoords, numCells * 4, &flatCells);CHKERRQ(ierr); 735 { 736 PetscReal (*coords)[2] = (PetscReal (*) [2]) flatCoords; 737 for (j = 0; j < numVerts-1; j++) { 738 PetscReal z, r, theta = -PETSC_PI/2 + (j%3) * PETSC_PI/2; 739 PetscReal rad = (j >= 6) ? inner_radius1 : (j >= 3) ? inner_radius2 : ctx->radius[grid]; 740 z = rad * PetscSinReal(theta); 741 coords[j][1] = z; 742 r = rad * PetscCosReal(theta); 743 coords[j][0] = r; 744 } 745 coords[numVerts-1][0] = coords[numVerts-1][1] = 0; 746 } 747 #else 748 numCells = 4; 749 numVerts = 8; 750 static int cells2[][4] = {{0,1,2,3}, 751 {4,5,1,0}, 752 {5,6,2,1}, 753 {6,7,3,2}}; 754 for (i = 0; i < numCells; i++) for (j = 0; j < 4; j++) cells[i][j] = cells2[i][j]; 755 ierr = loc2(numVerts * 2, &flatCoords, numCells * 4, &flatCells);CHKERRQ(ierr); 756 { 757 PetscReal (*coords)[2] = (PetscReal (*) [2]) flatCoords; 758 PetscInt j; 759 for (j = 0; j < 8; j++) { 760 PetscReal z, r; 761 PetscReal theta = -PETSC_PI/2 + (j%4) * PETSC_PI/3.; 762 PetscReal rad = ctx->radius[grid] * ((j < 4) ? 0.5 : 1.0); 763 z = rad * PetscSinReal(theta); 764 coords[j][1] = z; 765 r = rad * PetscCosReal(theta); 766 coords[j][0] = r; 767 } 768 } 769 #endif 770 } else if (ctx->num_sections==3) { 771 numCells = 7; 772 numVerts = 12; 773 int cells2[][4] = { {0,1,5,4}, 774 {1,2,6,5}, 775 {2,3,7,6}, 776 {4,5,9,8}, 777 {5,6,10,9}, 778 {6,7,11,10}, 779 {8,9,10,11} }; 780 for (i = 0; i < numCells; i++) for (j = 0; j < 4; j++) cells[i][j] = cells2[i][j]; 781 ierr = PetscMalloc2(numVerts * 2, &flatCoords, numCells * 4, &flatCells);CHKERRQ(ierr); 782 { 783 PetscReal (*coords)[2] = (PetscReal (*) [2]) flatCoords; 784 for (j = 0; j < numVerts; j++) { 785 PetscReal z, r, theta = -PETSC_PI/2 + (j%4) * PETSC_PI/3; 786 PetscReal rad = (j >= 8) ? inner_radius1 : (j >= 4) ? inner_radius2 : ctx->radius[grid]; 787 z = rad * PetscSinReal(theta); 788 coords[j][1] = z; 789 r = rad * PetscCosReal(theta); 790 coords[j][0] = r; 791 } 792 } 793 } else if (ctx->num_sections==4) { 794 numCells = 10; 795 numVerts = 16; 796 int cells2[][4] = { {0,1,6,5}, 797 {1,2,7,6}, 798 {2,3,8,7}, 799 {3,4,9,8}, 800 {5,6,11,10}, 801 {6,7,12,11}, 802 {7,8,13,12}, 803 {8,9,14,13}, 804 {10,11,12,15}, 805 {12,13,14,15}}; 806 for (i = 0; i < numCells; i++) for (j = 0; j < 4; j++) cells[i][j] = cells2[i][j]; 807 ierr = PetscMalloc2(numVerts * 2, &flatCoords, numCells * 4, &flatCells);CHKERRQ(ierr); 808 { 809 PetscReal (*coords)[2] = (PetscReal (*) [2]) flatCoords; 810 for (j = 0; j < numVerts-1; j++) { 811 PetscReal z, r, theta = -PETSC_PI/2 + (j%5) * PETSC_PI/4; 812 PetscReal rad = (j >= 10) ? inner_radius1 : (j >= 5) ? inner_radius2 : ctx->radius[grid]; 813 z = rad * PetscSinReal(theta); 814 coords[j][1] = z; 815 r = rad * PetscCosReal(theta); 816 coords[j][0] = r; 817 } 818 coords[numVerts-1][0] = coords[numVerts-1][1] = 0; 819 } 820 } else { 821 numCells = 0; 822 numVerts = 0; 823 } 824 for (j = 0, pcell = flatCells; j < numCells; j++, pcell += 4) { 825 pcell[0] = cells[j][0]; pcell[1] = cells[j][1]; 826 pcell[2] = cells[j][2]; pcell[3] = cells[j][3]; 827 } 828 ierr = DMPlexCreateFromCellListPetsc(comm_self,2,numCells,numVerts,4,ctx->interpolate,flatCells,2,flatCoords,&ctx->plex[grid]);CHKERRQ(ierr); 829 ierr = PetscFree2(flatCoords,flatCells);CHKERRQ(ierr); 830 ierr = PetscObjectSetName((PetscObject) ctx->plex[grid], "semi-circle");CHKERRQ(ierr); 831 } else SETERRQ(ctx->comm, PETSC_ERR_PLIB, "Velocity space meshes does not support cubed sphere"); 832 833 ierr = DMSetFromOptions(ctx->plex[grid]);CHKERRQ(ierr); 834 } // grid loop 835 ierr = PetscObjectSetOptionsPrefix((PetscObject)pack,prefix);CHKERRQ(ierr); 836 ierr = DMSetFromOptions(pack);CHKERRQ(ierr); 837 838 { /* convert to p4est (or whatever), wait for discretization to create pack */ 839 char convType[256]; 840 PetscBool flg; 841 ierr = PetscOptionsBegin(ctx->comm, prefix, "Mesh conversion options", "DMPLEX");CHKERRQ(ierr); 842 ierr = PetscOptionsFList("-dm_landau_type","Convert DMPlex to another format (p4est)","plexland.c",DMList,DMPLEX,convType,256,&flg);CHKERRQ(ierr); 843 ierr = PetscOptionsEnd();CHKERRQ(ierr); 844 if (flg) { 845 ctx->use_p4est = PETSC_TRUE; /* flag for Forest */ 846 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 847 DM dmforest; 848 ierr = DMConvert(ctx->plex[grid],convType,&dmforest);CHKERRQ(ierr); 849 if (dmforest) { 850 PetscBool isForest; 851 ierr = PetscObjectSetOptionsPrefix((PetscObject)dmforest,prefix);CHKERRQ(ierr); 852 ierr = DMIsForest(dmforest,&isForest);CHKERRQ(ierr); 853 if (isForest) { 854 if (ctx->sphere && ctx->inflate) { 855 ierr = DMForestSetBaseCoordinateMapping(dmforest,GeometryDMLandau,ctx);CHKERRQ(ierr); 856 } 857 ierr = DMDestroy(&ctx->plex[grid]);CHKERRQ(ierr); 858 ctx->plex[grid] = dmforest; // Forest for adaptivity 859 } else SETERRQ(ctx->comm, PETSC_ERR_PLIB, "Converted to non Forest?"); 860 } else SETERRQ(ctx->comm, PETSC_ERR_PLIB, "Convert failed?"); 861 } 862 } else ctx->use_p4est = PETSC_FALSE; /* flag for Forest */ 863 } 864 } /* non-file */ 865 ierr = DMSetDimension(pack, dim);CHKERRQ(ierr); 866 ierr = PetscObjectSetName((PetscObject) pack, "Mesh");CHKERRQ(ierr); 867 ierr = DMSetApplicationContext(pack, ctx);CHKERRQ(ierr); 868 869 PetscFunctionReturn(0); 870 } 871 872 static PetscErrorCode SetupDS(DM pack, PetscInt dim, PetscInt grid, LandauCtx *ctx) 873 { 874 PetscErrorCode ierr; 875 PetscInt ii,i0; 876 char buf[256]; 877 PetscSection section; 878 879 PetscFunctionBegin; 880 for (ii = ctx->species_offset[grid], i0 = 0 ; ii < ctx->species_offset[grid+1] ; ii++, i0++) { 881 if (ii==0) ierr = PetscSNPrintf(buf, 256, "e"); 882 else {ierr = PetscSNPrintf(buf, 256, "i%D", ii);CHKERRQ(ierr);} 883 /* Setup Discretization - FEM */ 884 ierr = PetscFECreateDefault(PETSC_COMM_SELF, dim, 1, PETSC_FALSE, NULL, PETSC_DECIDE, &ctx->fe[ii]);CHKERRQ(ierr); 885 ierr = PetscObjectSetName((PetscObject) ctx->fe[ii], buf);CHKERRQ(ierr); 886 ierr = DMSetField(ctx->plex[grid], i0, NULL, (PetscObject) ctx->fe[ii]);CHKERRQ(ierr); 887 } 888 ierr = DMCreateDS(ctx->plex[grid]);CHKERRQ(ierr); 889 ierr = DMGetSection(ctx->plex[grid], §ion);CHKERRQ(ierr); 890 for (PetscInt ii = ctx->species_offset[grid], i0 = 0 ; ii < ctx->species_offset[grid+1] ; ii++, i0++) { 891 if (ii==0) ierr = PetscSNPrintf(buf, 256, "se"); 892 else ierr = PetscSNPrintf(buf, 256, "si%D", ii); 893 ierr = PetscSectionSetComponentName(section, i0, 0, buf);CHKERRQ(ierr); 894 } 895 PetscFunctionReturn(0); 896 } 897 898 /* Define a Maxwellian function for testing out the operator. */ 899 900 /* Using cartesian velocity space coordinates, the particle */ 901 /* density, [1/m^3], is defined according to */ 902 903 /* $$ n=\int_{R^3} dv^3 \left(\frac{m}{2\pi T}\right)^{3/2}\exp [- mv^2/(2T)] $$ */ 904 905 /* Using some constant, c, we normalize the velocity vector into a */ 906 /* dimensionless variable according to v=c*x. Thus the density, $n$, becomes */ 907 908 /* $$ n=\int_{R^3} dx^3 \left(\frac{mc^2}{2\pi T}\right)^{3/2}\exp [- mc^2/(2T)*x^2] $$ */ 909 910 /* Defining $\theta=2T/mc^2$, we thus find that the probability density */ 911 /* for finding the particle within the interval in a box dx^3 around x is */ 912 913 /* f(x;\theta)=\left(\frac{1}{\pi\theta}\right)^{3/2} \exp [ -x^2/\theta ] */ 914 915 typedef struct { 916 PetscReal v_0; 917 PetscReal kT_m; 918 PetscReal n; 919 PetscReal shift; 920 } MaxwellianCtx; 921 922 static PetscErrorCode maxwellian(PetscInt dim, PetscReal time, const PetscReal x[], PetscInt Nf_dummy, PetscScalar *u, void *actx) 923 { 924 MaxwellianCtx *mctx = (MaxwellianCtx*)actx; 925 PetscInt i; 926 PetscReal v2 = 0, theta = 2*mctx->kT_m/(mctx->v_0*mctx->v_0); /* theta = 2kT/mc^2 */ 927 PetscFunctionBegin; 928 /* compute the exponents, v^2 */ 929 for (i = 0; i < dim; ++i) v2 += x[i]*x[i]; 930 /* evaluate the Maxwellian */ 931 u[0] = mctx->n*PetscPowReal(PETSC_PI*theta,-1.5)*(PetscExpReal(-v2/theta)); 932 if (mctx->shift!=0.) { 933 v2 = 0; 934 for (i = 0; i < dim-1; ++i) v2 += x[i]*x[i]; 935 v2 += (x[dim-1]-mctx->shift)*(x[dim-1]-mctx->shift); 936 /* evaluate the shifted Maxwellian */ 937 u[0] += mctx->n*PetscPowReal(PETSC_PI*theta,-1.5)*(PetscExpReal(-v2/theta)); 938 } 939 PetscFunctionReturn(0); 940 } 941 942 /*@ 943 LandauAddMaxwellians - Add a Maxwellian distribution to a state 944 945 Collective on X 946 947 Input Parameters: 948 . dm - The mesh (local) 949 + time - Current time 950 - temps - Temperatures of each species (global) 951 . ns - Number density of each species (global) 952 - grid - index into current grid - just used for offset into temp and ns 953 + actx - Landau context 954 955 Output Parameter: 956 . X - The state (local to this grid) 957 958 Level: beginner 959 960 .keywords: mesh 961 .seealso: LandauCreateVelocitySpace() 962 @*/ 963 PetscErrorCode LandauAddMaxwellians(DM dm, Vec X, PetscReal time, PetscReal temps[], PetscReal ns[], PetscInt grid, PetscInt b_id, void *actx) 964 { 965 LandauCtx *ctx = (LandauCtx*)actx; 966 PetscErrorCode (*initu[LANDAU_MAX_SPECIES])(PetscInt, PetscReal, const PetscReal [], PetscInt, PetscScalar [], void *); 967 PetscErrorCode ierr,ii,i0; 968 PetscInt dim; 969 MaxwellianCtx *mctxs[LANDAU_MAX_SPECIES], data[LANDAU_MAX_SPECIES]; 970 971 PetscFunctionBegin; 972 ierr = DMGetDimension(dm, &dim);CHKERRQ(ierr); 973 if (!ctx) { ierr = DMGetApplicationContext(dm, &ctx);CHKERRQ(ierr); } 974 for (ii = ctx->species_offset[grid], i0 = 0 ; ii < ctx->species_offset[grid+1] ; ii++, i0++) { 975 mctxs[i0] = &data[i0]; 976 data[i0].v_0 = ctx->v_0; // v_0 same for all grids 977 data[i0].kT_m = ctx->k*temps[ii]/ctx->masses[ii]; /* kT/m */ 978 data[i0].n = ns[ii] * (1+(double)b_id/100.0); // make solves a little different to mimic application, n[0] use for Conner-Hastie 979 initu[i0] = maxwellian; 980 data[i0].shift = 0; 981 } 982 data[0].shift = ctx->electronShift; 983 /* need to make ADD_ALL_VALUES work - TODO */ 984 ierr = DMProjectFunction(dm, time, initu, (void**)mctxs, INSERT_ALL_VALUES, X);CHKERRQ(ierr); 985 PetscFunctionReturn(0); 986 } 987 988 /* 989 LandauSetInitialCondition - Addes Maxwellians with context 990 991 Collective on X 992 993 Input Parameters: 994 . dm - The mesh 995 - grid - index into current grid - just used for offset into temp and ns 996 + actx - Landau context with T and n 997 998 Output Parameter: 999 . X - The state 1000 1001 Level: beginner 1002 1003 .keywords: mesh 1004 .seealso: LandauCreateVelocitySpace(), LandauAddMaxwellians() 1005 */ 1006 static PetscErrorCode LandauSetInitialCondition(DM dm, Vec X, PetscInt grid, PetscInt b_id, void *actx) 1007 { 1008 LandauCtx *ctx = (LandauCtx*)actx; 1009 PetscErrorCode ierr; 1010 PetscFunctionBegin; 1011 if (!ctx) { ierr = DMGetApplicationContext(dm, &ctx);CHKERRQ(ierr); } 1012 ierr = VecZeroEntries(X);CHKERRQ(ierr); 1013 ierr = LandauAddMaxwellians(dm, X, 0.0, ctx->thermal_temps, ctx->n, grid, b_id, ctx);CHKERRQ(ierr); 1014 PetscFunctionReturn(0); 1015 } 1016 1017 // adapt a level once. Forest in/out 1018 static PetscErrorCode adaptToleranceFEM(PetscFE fem, Vec sol, PetscInt type, PetscInt grid, LandauCtx *ctx, DM *newForest) 1019 { 1020 DM forest, plex, adaptedDM = NULL; 1021 PetscDS prob; 1022 PetscBool isForest; 1023 PetscQuadrature quad; 1024 PetscInt Nq, *Nb, cStart, cEnd, c, dim, qj, k; 1025 DMLabel adaptLabel = NULL; 1026 PetscErrorCode ierr; 1027 1028 PetscFunctionBegin; 1029 forest = ctx->plex[grid]; 1030 ierr = DMCreateDS(forest);CHKERRQ(ierr); 1031 ierr = DMGetDS(forest, &prob);CHKERRQ(ierr); 1032 ierr = DMGetDimension(forest, &dim);CHKERRQ(ierr); 1033 ierr = DMIsForest(forest, &isForest);CHKERRQ(ierr); 1034 if (!isForest) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"! Forest"); 1035 ierr = DMConvert(forest, DMPLEX, &plex);CHKERRQ(ierr); 1036 ierr = DMPlexGetHeightStratum(plex,0,&cStart,&cEnd);CHKERRQ(ierr); 1037 ierr = DMLabelCreate(PETSC_COMM_SELF,"adapt",&adaptLabel);CHKERRQ(ierr); 1038 ierr = PetscFEGetQuadrature(fem, &quad);CHKERRQ(ierr); 1039 ierr = PetscQuadratureGetData(quad, NULL, NULL, &Nq, NULL, NULL);CHKERRQ(ierr); 1040 if (Nq >LANDAU_MAX_NQ) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"Order too high. Nq = %D > LANDAU_MAX_NQ (%D)",Nq,LANDAU_MAX_NQ); 1041 ierr = PetscDSGetDimensions(prob, &Nb);CHKERRQ(ierr); 1042 if (type==4) { 1043 for (c = cStart; c < cEnd; c++) { 1044 ierr = DMLabelSetValue(adaptLabel, c, DM_ADAPT_REFINE);CHKERRQ(ierr); 1045 } 1046 ierr = PetscInfo1(sol, "Phase:%s: Uniform refinement\n","adaptToleranceFEM");CHKERRQ(ierr); 1047 } else if (type==2) { 1048 PetscInt rCellIdx[8], eCellIdx[64], iCellIdx[64], eMaxIdx = -1, iMaxIdx = -1, nr = 0, nrmax = (dim==3) ? 8 : 2; 1049 PetscReal minRad = PETSC_INFINITY, r, eMinRad = PETSC_INFINITY, iMinRad = PETSC_INFINITY; 1050 for (c = 0; c < 64; c++) { eCellIdx[c] = iCellIdx[c] = -1; } 1051 for (c = cStart; c < cEnd; c++) { 1052 PetscReal tt, v0[LANDAU_MAX_NQ*3], detJ[LANDAU_MAX_NQ]; 1053 ierr = DMPlexComputeCellGeometryFEM(plex, c, quad, v0, NULL, NULL, detJ);CHKERRQ(ierr); 1054 for (qj = 0; qj < Nq; ++qj) { 1055 tt = PetscSqr(v0[dim*qj+0]) + PetscSqr(v0[dim*qj+1]) + PetscSqr(((dim==3) ? v0[dim*qj+2] : 0)); 1056 r = PetscSqrtReal(tt); 1057 if (r < minRad - PETSC_SQRT_MACHINE_EPSILON*10.) { 1058 minRad = r; 1059 nr = 0; 1060 rCellIdx[nr++]= c; 1061 ierr = PetscInfo4(sol, "\t\tPhase: adaptToleranceFEM Found first inner r=%e, cell %D, qp %D/%D\n", r, c, qj+1, Nq);CHKERRQ(ierr); 1062 } else if ((r-minRad) < PETSC_SQRT_MACHINE_EPSILON*100. && nr < nrmax) { 1063 for (k=0;k<nr;k++) if (c == rCellIdx[k]) break; 1064 if (k==nr) { 1065 rCellIdx[nr++]= c; 1066 ierr = PetscInfo5(sol, "\t\t\tPhase: adaptToleranceFEM Found another inner r=%e, cell %D, qp %D/%D, d=%e\n", r, c, qj+1, Nq, r-minRad);CHKERRQ(ierr); 1067 } 1068 } 1069 if (ctx->sphere) { 1070 if ((tt=r-ctx->e_radius) > 0) { 1071 PetscInfo2(sol, "\t\t\t %D cell r=%g\n",c,tt); 1072 if (tt < eMinRad - PETSC_SQRT_MACHINE_EPSILON*100.) { 1073 eMinRad = tt; 1074 eMaxIdx = 0; 1075 eCellIdx[eMaxIdx++] = c; 1076 } else if (eMaxIdx > 0 && (tt-eMinRad) <= PETSC_SQRT_MACHINE_EPSILON && c != eCellIdx[eMaxIdx-1]) { 1077 eCellIdx[eMaxIdx++] = c; 1078 } 1079 } 1080 if ((tt=r-ctx->i_radius[grid]) > 0) { 1081 if (tt < iMinRad - 1.e-5) { 1082 iMinRad = tt; 1083 iMaxIdx = 0; 1084 iCellIdx[iMaxIdx++] = c; 1085 } else if (iMaxIdx > 0 && (tt-iMinRad) <= PETSC_SQRT_MACHINE_EPSILON && c != iCellIdx[iMaxIdx-1]) { 1086 iCellIdx[iMaxIdx++] = c; 1087 } 1088 } 1089 } 1090 } 1091 } 1092 for (k=0;k<nr;k++) { 1093 ierr = DMLabelSetValue(adaptLabel, rCellIdx[k], DM_ADAPT_REFINE);CHKERRQ(ierr); 1094 } 1095 if (ctx->sphere) { 1096 for (c = 0; c < eMaxIdx; c++) { 1097 ierr = DMLabelSetValue(adaptLabel, eCellIdx[c], DM_ADAPT_REFINE);CHKERRQ(ierr); 1098 ierr = PetscInfo3(sol, "\t\tPhase:%s: refine sphere e cell %D r=%g\n","adaptToleranceFEM",eCellIdx[c],eMinRad);CHKERRQ(ierr); 1099 } 1100 for (c = 0; c < iMaxIdx; c++) { 1101 ierr = DMLabelSetValue(adaptLabel, iCellIdx[c], DM_ADAPT_REFINE);CHKERRQ(ierr); 1102 ierr = PetscInfo3(sol, "\t\tPhase:%s: refine sphere i cell %D r=%g\n","adaptToleranceFEM",iCellIdx[c],iMinRad);CHKERRQ(ierr); 1103 } 1104 } 1105 ierr = PetscInfo4(sol, "Phase:%s: Adaptive refine origin cells %D,%D r=%g\n","adaptToleranceFEM",rCellIdx[0],rCellIdx[1],minRad);CHKERRQ(ierr); 1106 } else if (type==0 || type==1 || type==3) { /* refine along r=0 axis */ 1107 PetscScalar *coef = NULL; 1108 Vec coords; 1109 PetscInt csize,Nv,d,nz; 1110 DM cdm; 1111 PetscSection cs; 1112 ierr = DMGetCoordinatesLocal(forest, &coords);CHKERRQ(ierr); 1113 ierr = DMGetCoordinateDM(forest, &cdm);CHKERRQ(ierr); 1114 ierr = DMGetLocalSection(cdm, &cs);CHKERRQ(ierr); 1115 for (c = cStart; c < cEnd; c++) { 1116 PetscInt doit = 0, outside = 0; 1117 ierr = DMPlexVecGetClosure(cdm, cs, coords, c, &csize, &coef);CHKERRQ(ierr); 1118 Nv = csize/dim; 1119 for (nz = d = 0; d < Nv; d++) { 1120 PetscReal z = PetscRealPart(coef[d*dim + (dim-1)]), x = PetscSqr(PetscRealPart(coef[d*dim + 0])) + ((dim==3) ? PetscSqr(PetscRealPart(coef[d*dim + 1])) : 0); 1121 x = PetscSqrtReal(x); 1122 if (x < PETSC_MACHINE_EPSILON*10. && PetscAbs(z)<PETSC_MACHINE_EPSILON*10.) doit = 1; /* refine origin */ 1123 else if (type==0 && (z < -PETSC_MACHINE_EPSILON*10. || z > ctx->re_radius+PETSC_MACHINE_EPSILON*10.)) outside++; /* first pass don't refine bottom */ 1124 else if (type==1 && (z > ctx->vperp0_radius1 || z < -ctx->vperp0_radius1)) outside++; /* don't refine outside electron refine radius */ 1125 else if (type==3 && (z > ctx->vperp0_radius2 || z < -ctx->vperp0_radius2)) outside++; /* don't refine outside ion refine radius */ 1126 if (x < PETSC_MACHINE_EPSILON*10.) nz++; 1127 } 1128 ierr = DMPlexVecRestoreClosure(cdm, cs, coords, c, &csize, &coef);CHKERRQ(ierr); 1129 if (doit || (outside<Nv && nz)) { 1130 ierr = DMLabelSetValue(adaptLabel, c, DM_ADAPT_REFINE);CHKERRQ(ierr); 1131 } 1132 } 1133 ierr = PetscInfo1(sol, "Phase:%s: RE refinement\n","adaptToleranceFEM");CHKERRQ(ierr); 1134 } 1135 ierr = DMDestroy(&plex);CHKERRQ(ierr); 1136 ierr = DMAdaptLabel(forest, adaptLabel, &adaptedDM);CHKERRQ(ierr); 1137 ierr = DMLabelDestroy(&adaptLabel);CHKERRQ(ierr); 1138 *newForest = adaptedDM; 1139 if (adaptedDM) { 1140 if (isForest) { 1141 ierr = DMForestSetAdaptivityForest(adaptedDM,NULL);CHKERRQ(ierr); // ???? 1142 } else exit(33); // ??????? 1143 ierr = DMConvert(adaptedDM, DMPLEX, &plex);CHKERRQ(ierr); 1144 ierr = DMPlexGetHeightStratum(plex,0,&cStart,&cEnd);CHKERRQ(ierr); 1145 ierr = PetscInfo2(sol, "\tPhase: adaptToleranceFEM: %D cells, %d total quadrature points\n",cEnd-cStart,Nq*(cEnd-cStart));CHKERRQ(ierr); 1146 ierr = DMDestroy(&plex);CHKERRQ(ierr); 1147 } else *newForest = NULL; 1148 PetscFunctionReturn(0); 1149 } 1150 1151 // forest goes in (ctx->plex[grid]), plex comes out 1152 static PetscErrorCode adapt(PetscInt grid, LandauCtx *ctx, Vec *uu) 1153 { 1154 PetscErrorCode ierr; 1155 PetscInt adaptIter; 1156 1157 PetscFunctionBegin; 1158 PetscInt type, limits[5] = {(grid==0) ? ctx->numRERefine : 0, (grid==0) ? ctx->nZRefine1 : 0, ctx->numAMRRefine[grid], (grid==0) ? ctx->nZRefine2 : 0,ctx->postAMRRefine[grid]}; 1159 for (type=0;type<5;type++) { 1160 for (adaptIter = 0; adaptIter<limits[type];adaptIter++) { 1161 DM newForest = NULL; 1162 ierr = adaptToleranceFEM(ctx->fe[0], *uu, type, grid, ctx, &newForest);CHKERRQ(ierr); 1163 if (newForest) { 1164 ierr = DMDestroy(&ctx->plex[grid]);CHKERRQ(ierr); 1165 ierr = VecDestroy(uu);CHKERRQ(ierr); 1166 ierr = DMCreateGlobalVector(newForest,uu);CHKERRQ(ierr); 1167 ierr = PetscObjectSetName((PetscObject) *uu, "uAMR");CHKERRQ(ierr); 1168 ierr = LandauSetInitialCondition(newForest, *uu, grid, 0, ctx);CHKERRQ(ierr); 1169 ctx->plex[grid] = newForest; 1170 } else { 1171 exit(4); // can happen with no AMR and post refinement 1172 } 1173 } 1174 } 1175 PetscFunctionReturn(0); 1176 } 1177 1178 static PetscErrorCode ProcessOptions(LandauCtx *ctx, const char prefix[]) 1179 { 1180 PetscErrorCode ierr; 1181 PetscBool flg, sph_flg; 1182 PetscInt ii,nt,nm,nc,num_species_grid[LANDAU_MAX_GRIDS]; 1183 PetscReal v0_grid[LANDAU_MAX_GRIDS]; 1184 DM dummy; 1185 1186 PetscFunctionBegin; 1187 ierr = DMCreate(ctx->comm,&dummy);CHKERRQ(ierr); 1188 /* get options - initialize context */ 1189 ctx->verbose = 1; // should be 0 for silent compliance 1190 #if defined(PETSC_HAVE_THREADSAFETY) 1191 ctx->batch_sz = PetscNumOMPThreads; 1192 #else 1193 ctx->batch_sz = 1; 1194 #endif 1195 ctx->batch_view_idx = 0; 1196 ctx->interpolate = PETSC_TRUE; 1197 ctx->gpu_assembly = PETSC_TRUE; 1198 ctx->aux_bool = PETSC_FALSE; 1199 ctx->electronShift = 0; 1200 ctx->M = NULL; 1201 ctx->J = NULL; 1202 /* geometry and grids */ 1203 ctx->sphere = PETSC_FALSE; 1204 ctx->inflate = PETSC_FALSE; 1205 ctx->aux_bool = PETSC_FALSE; 1206 ctx->use_p4est = PETSC_FALSE; 1207 ctx->num_sections = 3; /* 2, 3 or 4 */ 1208 for (PetscInt grid=0;grid<LANDAU_MAX_GRIDS;grid++) { 1209 ctx->radius[grid] = 5.; /* thermal radius (velocity) */ 1210 ctx->numAMRRefine[grid] = 5; 1211 ctx->postAMRRefine[grid] = 0; 1212 ctx->species_offset[grid+1] = 1; // one species default 1213 num_species_grid[grid] = 0; 1214 ctx->plex[grid] = NULL; /* cache as expensive to Convert */ 1215 } 1216 ctx->species_offset[0] = 0; 1217 ctx->re_radius = 0.; 1218 ctx->vperp0_radius1 = 0; 1219 ctx->vperp0_radius2 = 0; 1220 ctx->nZRefine1 = 0; 1221 ctx->nZRefine2 = 0; 1222 ctx->numRERefine = 0; 1223 num_species_grid[0] = 1; // one species default 1224 /* species - [0] electrons, [1] one ion species eg, duetarium, [2] heavy impurity ion, ... */ 1225 ctx->charges[0] = -1; /* electron charge (MKS) */ 1226 ctx->masses[0] = 1/1835.469965278441013; /* temporary value in proton mass */ 1227 ctx->n[0] = 1; 1228 ctx->v_0 = 1; /* thermal velocity, we could start with a scale != 1 */ 1229 ctx->thermal_temps[0] = 1; 1230 /* constants, etc. */ 1231 ctx->epsilon0 = 8.8542e-12; /* permittivity of free space (MKS) F/m */ 1232 ctx->k = 1.38064852e-23; /* Boltzmann constant (MKS) J/K */ 1233 ctx->lnLam = 10; /* cross section ratio large - small angle collisions */ 1234 ctx->n_0 = 1.e20; /* typical plasma n, but could set it to 1 */ 1235 ctx->Ez = 0; 1236 ctx->subThreadBlockSize = 1; /* for device and maybe OMP */ 1237 ctx->numConcurrency = 0; /* for device (matrix solver hint) */ 1238 for (PetscInt grid=0;grid<LANDAU_NUM_TIMERS;grid++) ctx->times[grid] = 0; 1239 ctx->use_matrix_mass = PETSC_FALSE; /* fast but slightly fragile */ 1240 ctx->use_relativistic_corrections = PETSC_FALSE; 1241 ctx->use_energy_tensor_trick = PETSC_FALSE; /* Use Eero's trick for energy conservation v --> grad(v^2/2) */ 1242 ctx->SData_d.w = NULL; 1243 ctx->SData_d.x = NULL; 1244 ctx->SData_d.y = NULL; 1245 ctx->SData_d.z = NULL; 1246 ctx->SData_d.invJ = NULL; 1247 ierr = PetscOptionsBegin(ctx->comm, prefix, "Options for Fokker-Plank-Landau collision operator", "none");CHKERRQ(ierr); 1248 { 1249 char opstring[256]; 1250 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 1251 ctx->deviceType = LANDAU_KOKKOS; 1252 ierr = PetscStrcpy(opstring,"kokkos");CHKERRQ(ierr); 1253 #if defined(PETSC_HAVE_CUDA) 1254 ctx->subThreadBlockSize = 16; 1255 #endif 1256 #elif defined(PETSC_HAVE_CUDA) 1257 ctx->deviceType = LANDAU_CUDA; 1258 ierr = PetscStrcpy(opstring,"cuda");CHKERRQ(ierr); 1259 #else 1260 ctx->deviceType = LANDAU_CPU; 1261 ierr = PetscStrcpy(opstring,"cpu");CHKERRQ(ierr); 1262 ctx->subThreadBlockSize = 0; 1263 #endif 1264 ierr = PetscOptionsString("-dm_landau_device_type","Use kernels on 'cpu', 'cuda', or 'kokkos'","plexland.c",opstring,opstring,256,NULL);CHKERRQ(ierr); 1265 ierr = PetscStrcmp("cpu",opstring,&flg);CHKERRQ(ierr); 1266 if (flg) { 1267 ctx->deviceType = LANDAU_CPU; 1268 ctx->subThreadBlockSize = 0; 1269 } else { 1270 ierr = PetscStrcmp("cuda",opstring,&flg);CHKERRQ(ierr); 1271 if (flg) { 1272 ctx->deviceType = LANDAU_CUDA; 1273 ctx->subThreadBlockSize = 0; 1274 } else { 1275 ierr = PetscStrcmp("kokkos",opstring,&flg);CHKERRQ(ierr); 1276 if (flg) ctx->deviceType = LANDAU_KOKKOS; 1277 else SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-dm_landau_device_type %s",opstring); 1278 } 1279 } 1280 } 1281 ierr = PetscOptionsReal("-dm_landau_electron_shift","Shift in thermal velocity of electrons","none",ctx->electronShift,&ctx->electronShift, NULL);CHKERRQ(ierr); 1282 ierr = PetscOptionsInt("-dm_landau_verbose", "Level of verbosity output", "plexland.c", ctx->verbose, &ctx->verbose, NULL);CHKERRQ(ierr); 1283 ierr = PetscOptionsInt("-dm_landau_batch_size", "Number of 'vertices' to batch", "ex2.c", ctx->batch_sz, &ctx->batch_sz, NULL);CHKERRQ(ierr); 1284 if (LANDAU_MAX_BATCH_SZ < ctx->batch_sz) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"LANDAU_MAX_BATCH_SZ %D < ctx->batch_sz %D",LANDAU_MAX_BATCH_SZ,ctx->batch_sz); 1285 ierr = PetscOptionsInt("-dm_landau_batch_view_idx", "Index of batch for diagnostics like plotting", "ex2.c", ctx->batch_view_idx, &ctx->batch_view_idx, NULL);CHKERRQ(ierr); 1286 if (ctx->batch_view_idx >= ctx->batch_sz) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"-ctx->batch_view_idx %D > ctx->batch_sz %D",ctx->batch_view_idx,ctx->batch_sz); 1287 ierr = PetscOptionsReal("-dm_landau_Ez","Initial parallel electric field in unites of Conner-Hastie critical field","plexland.c",ctx->Ez,&ctx->Ez, NULL);CHKERRQ(ierr); 1288 ierr = PetscOptionsReal("-dm_landau_n_0","Normalization constant for number density","plexland.c",ctx->n_0,&ctx->n_0, NULL);CHKERRQ(ierr); 1289 ierr = PetscOptionsReal("-dm_landau_ln_lambda","Cross section parameter","plexland.c",ctx->lnLam,&ctx->lnLam, NULL);CHKERRQ(ierr); 1290 ierr = PetscOptionsBool("-dm_landau_use_mataxpy_mass", "Use fast but slightly fragile MATAXPY to add mass term", "plexland.c", ctx->use_matrix_mass, &ctx->use_matrix_mass, NULL);CHKERRQ(ierr); 1291 ierr = PetscOptionsBool("-dm_landau_use_relativistic_corrections", "Use relativistic corrections", "plexland.c", ctx->use_relativistic_corrections, &ctx->use_relativistic_corrections, NULL);CHKERRQ(ierr); 1292 ierr = PetscOptionsBool("-dm_landau_use_energy_tensor_trick", "Use Eero's trick of using grad(v^2/2) instead of v as args to Landau tensor to conserve energy with relativistic corrections and Q1 elements", "plexland.c", ctx->use_energy_tensor_trick, &ctx->use_energy_tensor_trick, NULL);CHKERRQ(ierr); 1293 1294 /* get num species with temperature, set defaults */ 1295 for (ii=1;ii<LANDAU_MAX_SPECIES;ii++) { 1296 ctx->thermal_temps[ii] = 1; 1297 ctx->charges[ii] = 1; 1298 ctx->masses[ii] = 1; 1299 ctx->n[ii] = 1; 1300 } 1301 nt = LANDAU_MAX_SPECIES; 1302 ierr = PetscOptionsRealArray("-dm_landau_thermal_temps", "Temperature of each species [e,i_0,i_1,...] in keV (must be set to set number of species)", "plexland.c", ctx->thermal_temps, &nt, &flg);CHKERRQ(ierr); 1303 if (flg) { 1304 PetscInfo1(dummy, "num_species set to number of thermal temps provided (%D)\n",nt); 1305 ctx->num_species = nt; 1306 } else SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"-dm_landau_thermal_temps ,t1,t2,.. must be provided to set the number of species"); 1307 for (ii=0;ii<ctx->num_species;ii++) ctx->thermal_temps[ii] *= 1.1604525e7; /* convert to Kelvin */ 1308 nm = LANDAU_MAX_SPECIES-1; 1309 ierr = PetscOptionsRealArray("-dm_landau_ion_masses", "Mass of each species in units of proton mass [i_0=2,i_1=40...]", "plexland.c", &ctx->masses[1], &nm, &flg);CHKERRQ(ierr); 1310 if (flg && nm != ctx->num_species-1) { 1311 SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"num ion masses %D != num species %D",nm,ctx->num_species-1); 1312 } 1313 nm = LANDAU_MAX_SPECIES; 1314 ierr = PetscOptionsRealArray("-dm_landau_n", "Number density of each species = n_s * n_0", "plexland.c", ctx->n, &nm, &flg);CHKERRQ(ierr); 1315 if (flg && nm != ctx->num_species) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"wrong num n: %D != num species %D",nm,ctx->num_species); 1316 for (ii=0;ii<LANDAU_MAX_SPECIES;ii++) ctx->masses[ii] *= 1.6720e-27; /* scale by proton mass kg */ 1317 ctx->masses[0] = 9.10938356e-31; /* electron mass kg (should be about right already) */ 1318 ctx->m_0 = ctx->masses[0]; /* arbitrary reference mass, electrons */ 1319 nc = LANDAU_MAX_SPECIES-1; 1320 ierr = PetscOptionsRealArray("-dm_landau_ion_charges", "Charge of each species in units of proton charge [i_0=2,i_1=18,...]", "plexland.c", &ctx->charges[1], &nc, &flg);CHKERRQ(ierr); 1321 if (flg && nc != ctx->num_species-1) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"num charges %D != num species %D",nc,ctx->num_species-1); 1322 for (ii=0;ii<LANDAU_MAX_SPECIES;ii++) ctx->charges[ii] *= 1.6022e-19; /* electron/proton charge (MKS) */ 1323 /* geometry and grids */ 1324 nt = LANDAU_MAX_GRIDS; 1325 ierr = PetscOptionsIntArray("-dm_landau_num_species_grid","Number of species on each grid: [ 1, ....] or [S, 0 ....] for single grid","plexland.c", num_species_grid, &nt, &flg);CHKERRQ(ierr); 1326 if (flg) { 1327 ctx->num_grids = nt; 1328 for (ii=nt=0;ii<ctx->num_grids;ii++) nt += num_species_grid[ii]; 1329 if (ctx->num_species != nt) SETERRQ4(ctx->comm,PETSC_ERR_ARG_WRONG,"-dm_landau_num_species_grid: sum %D != num_species = %D. %D grids (check that number of grids <= LANDAU_MAX_GRIDS = %D)",nt,ctx->num_species,ctx->num_grids,LANDAU_MAX_GRIDS); 1330 } else { 1331 ctx->num_grids = 1; // go back to a single grid run 1332 num_species_grid[0] = ctx->num_species; 1333 } 1334 for (ctx->species_offset[0] = ii = 0; ii < ctx->num_grids ; ii++) ctx->species_offset[ii+1] = ctx->species_offset[ii] + num_species_grid[ii]; 1335 if (ctx->species_offset[ctx->num_grids] != ctx->num_species) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"ctx->species_offset[ctx->num_grids] %D != ctx->num_species = %D ???????????",ctx->species_offset[ctx->num_grids],ctx->num_species); 1336 for (PetscInt grid = 0; grid < ctx->num_grids ; grid++) { 1337 int iii = ctx->species_offset[grid]; // normalize with first (arbitrary) species on grid 1338 v0_grid[grid] = PetscSqrtReal(ctx->k*ctx->thermal_temps[iii]/ctx->masses[iii]); /* arbitrary units for non-dimensionalization: mean velocity in 1D of first species on grid */ 1339 } 1340 ii = 0; 1341 ierr = PetscOptionsInt("-dm_landau_v0_grid", "Index of grid to use for setting v_0 (electrons are default). Not recommended to change", "plexland.c", ii, &ii, NULL);CHKERRQ(ierr); 1342 ctx->v_0 = v0_grid[ii]; /* arbitrary units for non dimensionalization: global mean velocity in 1D of electrons */ 1343 ctx->t_0 = 8*PETSC_PI*PetscSqr(ctx->epsilon0*ctx->m_0/PetscSqr(ctx->charges[0]))/ctx->lnLam/ctx->n_0*PetscPowReal(ctx->v_0,3); /* note, this t_0 makes nu[0,0]=1 */ 1344 /* domain */ 1345 nt = LANDAU_MAX_GRIDS; 1346 ierr = PetscOptionsRealArray("-dm_landau_domain_radius","Phase space size in units of thermal velocity of grid","plexland.c",ctx->radius,&nt, &flg);CHKERRQ(ierr); 1347 if (flg && nt < ctx->num_grids) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"-dm_landau_domain_radius: given %D radius != number grids %D",nt,ctx->num_grids); 1348 for (PetscInt grid = 0; grid < ctx->num_grids ; grid++) { 1349 if (flg && ctx->radius[grid] <= 0) { /* negative is ratio of c */ 1350 if (ctx->radius[grid] == 0) ctx->radius[grid] = 0.75; 1351 else ctx->radius[grid] = -ctx->radius[grid]; 1352 ctx->radius[grid] = ctx->radius[grid]*SPEED_OF_LIGHT/ctx->v_0; // use any species on grid to normalize (v_0 same for all on grid) 1353 ierr = PetscInfo2(dummy, "Change domain radius to %g for grid %D\n",ctx->radius[grid],grid);CHKERRQ(ierr); 1354 } 1355 ctx->radius[grid] *= v0_grid[grid]/ctx->v_0; // scale domain by thermal radius relative to v_0 1356 } 1357 /* amr parametres */ 1358 nt = LANDAU_MAX_GRIDS; 1359 ierr = PetscOptionsIntArray("-dm_landau_amr_levels_max", "Number of AMR levels of refinement around origin, after (RE) refinements along z", "plexland.c", ctx->numAMRRefine, &nt, &flg);CHKERRQ(ierr); 1360 if (flg && nt < ctx->num_grids) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"-dm_landau_amr_levels_max: given %D != number grids %D",nt,ctx->num_grids); 1361 nt = LANDAU_MAX_GRIDS; 1362 ierr = PetscOptionsIntArray("-dm_landau_amr_post_refine", "Number of levels to uniformly refine after AMR", "plexland.c", ctx->postAMRRefine, &nt, &flg);CHKERRQ(ierr); 1363 for (ii=1;ii<ctx->num_grids;ii++) ctx->postAMRRefine[ii] = ctx->postAMRRefine[0]; // all grids the same now 1364 ierr = PetscOptionsInt("-dm_landau_amr_re_levels", "Number of levels to refine along v_perp=0, z>0", "plexland.c", ctx->numRERefine, &ctx->numRERefine, &flg);CHKERRQ(ierr); 1365 ierr = PetscOptionsInt("-dm_landau_amr_z_refine1", "Number of levels to refine along v_perp=0", "plexland.c", ctx->nZRefine1, &ctx->nZRefine1, &flg);CHKERRQ(ierr); 1366 ierr = PetscOptionsInt("-dm_landau_amr_z_refine2", "Number of levels to refine along v_perp=0", "plexland.c", ctx->nZRefine2, &ctx->nZRefine2, &flg);CHKERRQ(ierr); 1367 ierr = PetscOptionsReal("-dm_landau_re_radius","velocity range to refine on positive (z>0) r=0 axis for runaways","plexland.c",ctx->re_radius,&ctx->re_radius, &flg);CHKERRQ(ierr); 1368 ierr = PetscOptionsReal("-dm_landau_z_radius1","velocity range to refine r=0 axis (for electrons)","plexland.c",ctx->vperp0_radius1,&ctx->vperp0_radius1, &flg);CHKERRQ(ierr); 1369 ierr = PetscOptionsReal("-dm_landau_z_radius2","velocity range to refine r=0 axis (for ions) after origin AMR","plexland.c",ctx->vperp0_radius2, &ctx->vperp0_radius2, &flg);CHKERRQ(ierr); 1370 /* spherical domain (not used) */ 1371 ierr = PetscOptionsInt("-dm_landau_num_sections", "Number of tangential section in (2D) grid, 2, 3, of 4", "plexland.c", ctx->num_sections, &ctx->num_sections, NULL);CHKERRQ(ierr); 1372 ierr = PetscOptionsBool("-dm_landau_sphere", "use sphere/semi-circle domain instead of rectangle", "plexland.c", ctx->sphere, &ctx->sphere, &sph_flg);CHKERRQ(ierr); 1373 ierr = PetscOptionsBool("-dm_landau_inflate", "With sphere, inflate for curved edges", "plexland.c", ctx->inflate, &ctx->inflate, &flg);CHKERRQ(ierr); 1374 ierr = PetscOptionsReal("-dm_landau_e_radius","Electron thermal velocity, used for circular meshes","plexland.c",ctx->e_radius, &ctx->e_radius, &flg);CHKERRQ(ierr); 1375 if (flg && !sph_flg) ctx->sphere = PETSC_TRUE; /* you gave me an e radius but did not set sphere, user error really */ 1376 if (!flg) { 1377 ctx->e_radius = 1.5*PetscSqrtReal(8*ctx->k*ctx->thermal_temps[0]/ctx->masses[0]/PETSC_PI)/ctx->v_0; 1378 } 1379 nt = LANDAU_MAX_GRIDS; 1380 ierr = PetscOptionsRealArray("-dm_landau_i_radius","Ion thermal velocity, used for circular meshes","plexland.c",ctx->i_radius, &nt, &flg);CHKERRQ(ierr); 1381 if (flg && !sph_flg) ctx->sphere = PETSC_TRUE; 1382 if (!flg) { 1383 ctx->i_radius[0] = 1.5*PetscSqrtReal(8*ctx->k*ctx->thermal_temps[1]/ctx->masses[1]/PETSC_PI)/ctx->v_0; // need to correct for ion grid domain 1384 } 1385 if (flg && ctx->num_grids != nt) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"-dm_landau_i_radius: %D != num_species = %D",nt,ctx->num_grids); 1386 if (ctx->sphere && ctx->e_radius <= ctx->i_radius[0]) SETERRQ3(ctx->comm,PETSC_ERR_ARG_WRONG,"bad radii: %g < %g < %g",ctx->i_radius[0],ctx->e_radius,ctx->radius[0]); 1387 /* processing options */ 1388 ierr = PetscOptionsInt("-dm_landau_sub_thread_block_size", "Number of threads in Kokkos integration point subblock", "plexland.c", ctx->subThreadBlockSize, &ctx->subThreadBlockSize, NULL);CHKERRQ(ierr); 1389 ierr = PetscOptionsBool("-dm_landau_gpu_assembly", "Assemble Jacobian on GPU", "plexland.c", ctx->gpu_assembly, &ctx->gpu_assembly, NULL);CHKERRQ(ierr); 1390 ierr = PetscOptionsInt("-dm_landau_mat_hint", "An integer hint attached to matrix for solvers", "plexland.c", ctx->numConcurrency, &ctx->numConcurrency, NULL);CHKERRQ(ierr); 1391 1392 ierr = PetscOptionsEnd();CHKERRQ(ierr); 1393 for (ii=ctx->num_species;ii<LANDAU_MAX_SPECIES;ii++) ctx->masses[ii] = ctx->thermal_temps[ii] = ctx->charges[ii] = 0; 1394 if (ctx->verbose > 0) { 1395 ierr = PetscPrintf(ctx->comm, "masses: e=%10.3e; ions in proton mass units: %10.3e %10.3e ...\n",ctx->masses[0],ctx->masses[1]/1.6720e-27,ctx->num_species>2 ? ctx->masses[2]/1.6720e-27 : 0);CHKERRQ(ierr); 1396 ierr = PetscPrintf(ctx->comm, "charges: e=%10.3e; charges in elementary units: %10.3e %10.3e\n", ctx->charges[0],-ctx->charges[1]/ctx->charges[0],ctx->num_species>2 ? -ctx->charges[2]/ctx->charges[0] : 0);CHKERRQ(ierr); 1397 ierr = PetscPrintf(ctx->comm, "n: e: %10.3e i: %10.3e %10.3e\n", ctx->n[0],ctx->n[1],ctx->num_species>2 ? ctx->n[2] : 0);CHKERRQ(ierr); 1398 ierr = PetscPrintf(ctx->comm, "thermal T (K): e=%10.3e i=%10.3e %10.3e. v_0=%10.3e (%10.3ec) n_0=%10.3e t_0=%10.3e, %s, %s, %D batched\n", ctx->thermal_temps[0], ctx->thermal_temps[1], (ctx->num_species>2) ? ctx->thermal_temps[2] : 0, ctx->v_0, ctx->v_0/SPEED_OF_LIGHT, ctx->n_0, ctx->t_0, ctx->use_relativistic_corrections ? "relativistic" : "classical", ctx->use_energy_tensor_trick ? "Use trick" : "Intuitive",ctx->batch_sz);CHKERRQ(ierr); 1399 ierr = PetscPrintf(ctx->comm, "Domain radius (AMR levels) grid %D: %g (%D) ",0,ctx->radius[0],ctx->numAMRRefine[0]);CHKERRQ(ierr); 1400 for (ii=1;ii<ctx->num_grids;ii++) PetscPrintf(ctx->comm, ", %D: %10.3e (%D) ",ii,ctx->radius[ii],ctx->numAMRRefine[ii]); 1401 ierr = PetscPrintf(ctx->comm,"\n");CHKERRQ(ierr); 1402 } 1403 ierr = DMDestroy(&dummy);CHKERRQ(ierr); 1404 { 1405 PetscMPIInt rank; 1406 ierr = MPI_Comm_rank(ctx->comm, &rank);CHKERRMPI(ierr); 1407 ctx->stage = 0; 1408 ierr = PetscLogEventRegister("Landau Create", DM_CLASSID, &ctx->events[13]);CHKERRQ(ierr); /* 13 */ 1409 ierr = PetscLogEventRegister("Landau Operator", DM_CLASSID, &ctx->events[11]);CHKERRQ(ierr); /* 11 */ 1410 ierr = PetscLogEventRegister("Landau Jacobian", DM_CLASSID, &ctx->events[0]);CHKERRQ(ierr); /* 0 */ 1411 ierr = PetscLogEventRegister("Landau Mass", DM_CLASSID, &ctx->events[9]);CHKERRQ(ierr); /* 9 */ 1412 ierr = PetscLogEventRegister(" Preamble", DM_CLASSID, &ctx->events[10]);CHKERRQ(ierr); /* 10 */ 1413 ierr = PetscLogEventRegister(" static IP Data", DM_CLASSID, &ctx->events[7]);CHKERRQ(ierr); /* 7 */ 1414 ierr = PetscLogEventRegister(" dynamic IP-Jac", DM_CLASSID, &ctx->events[1]);CHKERRQ(ierr); /* 1 */ 1415 ierr = PetscLogEventRegister(" Kernel-init", DM_CLASSID, &ctx->events[3]);CHKERRQ(ierr); /* 3 */ 1416 ierr = PetscLogEventRegister(" Jac-f-df (GPU)", DM_CLASSID, &ctx->events[8]);CHKERRQ(ierr); /* 8 */ 1417 ierr = PetscLogEventRegister(" Kernel (GPU)", DM_CLASSID, &ctx->events[4]);CHKERRQ(ierr); /* 4 */ 1418 ierr = PetscLogEventRegister(" Copy to CPU", DM_CLASSID, &ctx->events[5]);CHKERRQ(ierr); /* 5 */ 1419 ierr = PetscLogEventRegister(" CPU assemble", DM_CLASSID, &ctx->events[6]);CHKERRQ(ierr); /* 6 */ 1420 ierr = PetscLogEventRegister(" GPU ass. setup", DM_CLASSID, &ctx->events[2]);CHKERRQ(ierr); /* 2 */ 1421 1422 if (rank) { /* turn off output stuff for duplicate runs - do we need to add the prefix to all this? */ 1423 ierr = PetscOptionsClearValue(NULL,"-snes_converged_reason");CHKERRQ(ierr); 1424 ierr = PetscOptionsClearValue(NULL,"-ksp_converged_reason");CHKERRQ(ierr); 1425 ierr = PetscOptionsClearValue(NULL,"-snes_monitor");CHKERRQ(ierr); 1426 ierr = PetscOptionsClearValue(NULL,"-ksp_monitor");CHKERRQ(ierr); 1427 ierr = PetscOptionsClearValue(NULL,"-ts_monitor");CHKERRQ(ierr); 1428 ierr = PetscOptionsClearValue(NULL,"-ts_adapt_monitor");CHKERRQ(ierr); 1429 ierr = PetscOptionsClearValue(NULL,"-dm_landau_amr_dm_view");CHKERRQ(ierr); 1430 ierr = PetscOptionsClearValue(NULL,"-dm_landau_amr_vec_view");CHKERRQ(ierr); 1431 ierr = PetscOptionsClearValue(NULL,"-dm_landau_mass_dm_view");CHKERRQ(ierr); 1432 ierr = PetscOptionsClearValue(NULL,"-dm_landau_mass_view");CHKERRQ(ierr); 1433 ierr = PetscOptionsClearValue(NULL,"-dm_landau_jacobian_view");CHKERRQ(ierr); 1434 ierr = PetscOptionsClearValue(NULL,"-dm_landau_mat_view");CHKERRQ(ierr); 1435 ierr = PetscOptionsClearValue(NULL,"-");CHKERRQ(ierr); 1436 ierr = PetscOptionsClearValue(NULL,"-info");CHKERRQ(ierr); 1437 } 1438 } 1439 PetscFunctionReturn(0); 1440 } 1441 1442 static PetscErrorCode CreateStaticGPUData(PetscInt dim, LandauCtx *ctx) 1443 { 1444 PetscErrorCode ierr; 1445 PetscSection section[LANDAU_MAX_GRIDS],globsection[LANDAU_MAX_GRIDS]; 1446 PetscQuadrature quad; 1447 const PetscReal *quadWeights; 1448 PetscInt q,eidx,fieldA,numCells[LANDAU_MAX_GRIDS],Nq,Nb,Nf[LANDAU_MAX_GRIDS]; 1449 PetscTabulation *Tf; 1450 PetscDS prob; 1451 1452 PetscFunctionBegin; 1453 ierr = DMGetDS(ctx->plex[0], &prob);CHKERRQ(ierr); // same DS for all grids 1454 ierr = PetscDSGetTabulation(prob, &Tf);CHKERRQ(ierr); // Bf, &Df same for all grids 1455 /* DS, Tab and quad is same on all grids */ 1456 if (ctx->plex[0] == NULL) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"Plex not created"); 1457 ierr = PetscFEGetQuadrature(ctx->fe[0], &quad);CHKERRQ(ierr); 1458 ierr = PetscQuadratureGetData(quad, NULL, NULL, &Nq, NULL, &quadWeights);CHKERRQ(ierr); Nb = Nq; 1459 if (Nq >LANDAU_MAX_NQ) SETERRQ2(ctx->comm,PETSC_ERR_ARG_WRONG,"Order too high. Nq = %D > LANDAU_MAX_NQ (%D)",Nq,LANDAU_MAX_NQ); 1460 /* setup each grid */ 1461 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 1462 PetscInt cStart, cEnd; 1463 if (ctx->plex[grid] == NULL) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"Plex not created"); 1464 ierr = DMPlexGetHeightStratum(ctx->plex[grid], 0, &cStart, &cEnd);CHKERRQ(ierr); 1465 numCells[grid] = cEnd - cStart; // grids can have different topology 1466 ierr = DMGetLocalSection(ctx->plex[grid], §ion[grid]);CHKERRQ(ierr); 1467 ierr = DMGetGlobalSection(ctx->plex[grid], &globsection[grid]);CHKERRQ(ierr); 1468 ierr = PetscSectionGetNumFields(section[grid], &Nf[grid]);CHKERRQ(ierr); 1469 } 1470 #define MAP_BF_SIZE (64*LANDAU_DIM*LANDAU_DIM*LANDAU_MAX_Q_FACE*LANDAU_MAX_SPECIES) 1471 /* create GPU assembly data */ 1472 if (ctx->gpu_assembly) { /* we need GPU object with GPU assembly */ 1473 PetscContainer container; 1474 PetscScalar elemMatrix[LANDAU_MAX_NQ*LANDAU_MAX_NQ*LANDAU_MAX_SPECIES*LANDAU_MAX_SPECIES], *elMat; 1475 pointInterpolationP4est pointMaps[MAP_BF_SIZE][LANDAU_MAX_Q_FACE]; 1476 P4estVertexMaps *maps; 1477 /* create GPU asssembly data */ 1478 ierr = PetscInfo1(ctx->plex[0], "Make GPU maps %D\n",1);CHKERRQ(ierr); 1479 ierr = PetscLogEventBegin(ctx->events[2],0,0,0,0);CHKERRQ(ierr); 1480 ierr = PetscMalloc(sizeof(*maps)*ctx->num_grids, &maps);CHKERRQ(ierr); 1481 ierr = PetscContainerCreate(PETSC_COMM_SELF, &container);CHKERRQ(ierr); 1482 ierr = PetscContainerSetPointer(container, (void *)maps);CHKERRQ(ierr); 1483 ierr = PetscContainerSetUserDestroy(container, LandauGPUMapsDestroy);CHKERRQ(ierr); 1484 ierr = PetscObjectCompose((PetscObject) ctx->J, "assembly_maps", (PetscObject) container);CHKERRQ(ierr); 1485 ierr = PetscContainerDestroy(&container);CHKERRQ(ierr); 1486 1487 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 1488 PetscInt cStart, cEnd, ej, Nfloc = Nf[grid], totDim = Nfloc*Nq; 1489 ierr = DMPlexGetHeightStratum(ctx->plex[grid], 0, &cStart, &cEnd);CHKERRQ(ierr); 1490 // make maps 1491 maps[grid].d_self = NULL; 1492 maps[grid].num_elements = numCells[grid]; 1493 maps[grid].num_face = (PetscInt)(pow(Nq,1./((double)dim))+.001); // Q 1494 maps[grid].num_face = (PetscInt)(pow(maps[grid].num_face,(double)(dim-1))+.001); // Q^2 1495 maps[grid].num_reduced = 0; 1496 maps[grid].deviceType = ctx->deviceType; 1497 maps[grid].numgrids = ctx->num_grids; 1498 // count reduced and get 1499 ierr = PetscMalloc(maps[grid].num_elements * sizeof(*maps[grid].gIdx), &maps[grid].gIdx);CHKERRQ(ierr); 1500 for (fieldA=0;fieldA<Nf[grid];fieldA++) { 1501 for (ej = cStart, eidx = 0 ; ej < cEnd; ++ej, ++eidx) { 1502 for (q = 0; q < Nb; ++q) { 1503 PetscInt numindices,*indices; 1504 PetscScalar *valuesOrig = elMat = elemMatrix; 1505 ierr = PetscMemzero(elMat, totDim*totDim*sizeof(*elMat));CHKERRQ(ierr); 1506 elMat[ (fieldA*Nb + q)*totDim + fieldA*Nb + q] = 1; 1507 ierr = DMPlexGetClosureIndices(ctx->plex[grid], section[grid], globsection[grid], ej, PETSC_TRUE, &numindices, &indices, NULL, (PetscScalar **) &elMat);CHKERRQ(ierr); 1508 for (PetscInt f = 0 ; f < numindices ; ++f) { // look for a non-zero on the diagonal 1509 if (PetscAbs(PetscRealPart(elMat[f*numindices + f])) > PETSC_MACHINE_EPSILON) { 1510 // found it 1511 if (PetscAbs(PetscRealPart(elMat[f*numindices + f] - 1.)) < PETSC_MACHINE_EPSILON) { 1512 maps[grid].gIdx[eidx][fieldA][q] = (LandauIdx)indices[f]; // normal vertex 1.0 1513 } else { //found a constraint 1514 int jj = 0; 1515 PetscReal sum = 0; 1516 const PetscInt ff = f; 1517 maps[grid].gIdx[eidx][fieldA][q] = -maps[grid].num_reduced - 1; // gid = -(idx+1): idx = -gid - 1 1518 do { // constraints are continous in Plex - exploit that here 1519 int ii; 1520 for (ii = 0, pointMaps[maps[grid].num_reduced][jj].scale = 0; ii < maps[grid].num_face; ii++) { // DMPlex puts them all together 1521 if (ff + ii < numindices) { 1522 pointMaps[maps[grid].num_reduced][jj].scale += PetscRealPart(elMat[f*numindices + ff + ii]); 1523 } 1524 } 1525 sum += pointMaps[maps[grid].num_reduced][jj].scale; 1526 if (pointMaps[maps[grid].num_reduced][jj].scale == 0) pointMaps[maps[grid].num_reduced][jj].gid = -1; // 3D has Q and Q^2 interps -- all contiguous??? 1527 else pointMaps[maps[grid].num_reduced][jj].gid = indices[f]; 1528 } while (++jj < maps[grid].num_face && ++f < numindices); // jj is incremented if we hit the end 1529 while (jj++ < maps[grid].num_face) { 1530 pointMaps[maps[grid].num_reduced][jj].scale = 0; 1531 pointMaps[maps[grid].num_reduced][jj].gid = -1; 1532 } 1533 if (PetscAbs(sum-1.0) > 10*PETSC_MACHINE_EPSILON) { // debug 1534 int d,f; 1535 PetscReal tmp = 0; 1536 PetscPrintf(PETSC_COMM_SELF,"\t\t%D.%D.%D) ERROR total I = %22.16e (LANDAU_MAX_Q_FACE=%d, #face=%D)\n",eidx,q,fieldA,sum,LANDAU_MAX_Q_FACE,maps[grid].num_face); 1537 for (d = 0, tmp = 0; d < numindices; ++d) { 1538 if (tmp!=0 && PetscAbs(tmp-1.0) > 10*PETSC_MACHINE_EPSILON) {ierr = PetscPrintf(PETSC_COMM_WORLD,"%3D) %3D: ",d,indices[d]);CHKERRQ(ierr);} 1539 for (f = 0; f < numindices; ++f) { 1540 tmp += PetscRealPart(elMat[d*numindices + f]); 1541 } 1542 if (tmp!=0) {ierr = PetscPrintf(ctx->comm," | %22.16e\n",tmp);CHKERRQ(ierr);} 1543 } 1544 } 1545 maps[grid].num_reduced++; 1546 if (maps[grid].num_reduced>=MAP_BF_SIZE) SETERRQ2(PETSC_COMM_SELF, PETSC_ERR_PLIB, "maps[grid].num_reduced %d > %d",maps[grid].num_reduced,MAP_BF_SIZE); 1547 } 1548 break; 1549 } 1550 } 1551 // cleanup 1552 ierr = DMPlexRestoreClosureIndices(ctx->plex[grid], section[grid], globsection[grid], ej, PETSC_TRUE, &numindices, &indices, NULL, (PetscScalar **) &elMat);CHKERRQ(ierr); 1553 if (elMat != valuesOrig) {ierr = DMRestoreWorkArray(ctx->plex[grid], numindices*numindices, MPIU_SCALAR, &elMat);CHKERRQ(ierr);} 1554 } 1555 } 1556 } 1557 // allocate and copy point datamaps[grid].gIdx[eidx][field][q] 1558 ierr = PetscMalloc(maps[grid].num_reduced * sizeof(*maps[grid].c_maps), &maps[grid].c_maps);CHKERRQ(ierr); 1559 for (ej = 0; ej < maps[grid].num_reduced; ++ej) { 1560 for (q = 0; q < maps[grid].num_face; ++q) { 1561 maps[grid].c_maps[ej][q].scale = pointMaps[ej][q].scale; 1562 maps[grid].c_maps[ej][q].gid = pointMaps[ej][q].gid; 1563 } 1564 } 1565 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 1566 if (ctx->deviceType == LANDAU_KOKKOS) { 1567 ierr = LandauKokkosCreateMatMaps(maps, pointMaps, Nf, Nq, grid);CHKERRQ(ierr); // imples Kokkos does 1568 } // else could be CUDA 1569 #endif 1570 #if defined(PETSC_HAVE_CUDA) 1571 if (ctx->deviceType == LANDAU_CUDA) { 1572 ierr = LandauCUDACreateMatMaps(maps, pointMaps, Nf, Nq, grid);CHKERRQ(ierr); 1573 } 1574 #endif 1575 } /* grids */ 1576 ierr = PetscLogEventEnd(ctx->events[2],0,0,0,0);CHKERRQ(ierr); 1577 } // GPU assembly 1578 { /* create static point data, Jacobian called first, only one vertex copy */ 1579 PetscReal *invJe,*ww,*xx,*yy,*zz=NULL,*invJ_a; 1580 PetscInt outer_ipidx, outer_ej,grid, nip_glb = 0; 1581 PetscFE fe; 1582 1583 ierr = PetscLogEventBegin(ctx->events[7],0,0,0,0);CHKERRQ(ierr); 1584 ierr = PetscInfo(ctx->plex[0], "Initialize static data\n");CHKERRQ(ierr); 1585 for (PetscInt grid=0;grid<ctx->num_grids;grid++) nip_glb += Nq*numCells[grid]; 1586 /* collect f data, first time is for Jacobian, but make mass now */ 1587 if (ctx->verbose > 0) { 1588 PetscInt ncells = 0, N; 1589 ierr = MatGetSize(ctx->J,&N,NULL);CHKERRQ(ierr); 1590 for (PetscInt grid=0;grid<ctx->num_grids;grid++) ncells += numCells[grid]; 1591 ierr = PetscPrintf(ctx->comm,"%D) %s %D IPs, %D cells total, Nb=%D, Nq=%D, dim=%D, Tab: Nb=%D Nf=%D Np=%D cdim=%D N=%D\n", 1592 0,"FormLandau",nip_glb,ncells, Nb, Nq, dim, Nb, ctx->num_species, Nb, dim, N);CHKERRQ(ierr); 1593 } 1594 ierr = PetscMalloc4(nip_glb,&ww,nip_glb,&xx,nip_glb,&yy,nip_glb*dim*dim,&invJ_a);CHKERRQ(ierr); 1595 if (dim==3) { 1596 ierr = PetscMalloc1(nip_glb,&zz);CHKERRQ(ierr); 1597 } 1598 if (ctx->use_energy_tensor_trick) { 1599 ierr = PetscFECreateDefault(PETSC_COMM_SELF, dim, 1, PETSC_FALSE, NULL, PETSC_DECIDE, &fe);CHKERRQ(ierr); 1600 ierr = PetscObjectSetName((PetscObject) fe, "energy");CHKERRQ(ierr); 1601 } 1602 /* init each grids static data - no batch */ 1603 for (grid=0, outer_ipidx=0, outer_ej=0 ; grid < ctx->num_grids ; grid++) { // OpenMP (once) 1604 Vec v2_2 = NULL; // projected function: v^2/2 for non-relativistic, gamma... for relativistic 1605 PetscSection e_section; 1606 DM dmEnergy; 1607 PetscInt cStart, cEnd, ej; 1608 1609 ierr = DMPlexGetHeightStratum(ctx->plex[grid], 0, &cStart, &cEnd);CHKERRQ(ierr); 1610 // prep energy trick, get v^2 / 2 vector 1611 if (ctx->use_energy_tensor_trick) { 1612 PetscErrorCode (*energyf[1])(PetscInt, PetscReal, const PetscReal [], PetscInt, PetscScalar [], void *) = {ctx->use_relativistic_corrections ? gamma_m1_f : energy_f}; 1613 Vec glob_v2; 1614 PetscReal *c2_0[1], data[1] = {PetscSqr(C_0(ctx->v_0))}; 1615 1616 ierr = DMClone(ctx->plex[grid], &dmEnergy);CHKERRQ(ierr); 1617 ierr = PetscObjectSetName((PetscObject) dmEnergy, "energy");CHKERRQ(ierr); 1618 ierr = DMSetField(dmEnergy, 0, NULL, (PetscObject)fe);CHKERRQ(ierr); 1619 ierr = DMCreateDS(dmEnergy);CHKERRQ(ierr); 1620 ierr = DMGetSection(dmEnergy, &e_section);CHKERRQ(ierr); 1621 ierr = DMGetGlobalVector(dmEnergy,&glob_v2);CHKERRQ(ierr); 1622 ierr = PetscObjectSetName((PetscObject) glob_v2, "trick");CHKERRQ(ierr); 1623 c2_0[0] = &data[0]; 1624 ierr = DMProjectFunction(dmEnergy, 0., energyf, (void**)c2_0, INSERT_ALL_VALUES, glob_v2);CHKERRQ(ierr); 1625 ierr = DMGetLocalVector(dmEnergy, &v2_2);CHKERRQ(ierr); 1626 ierr = VecZeroEntries(v2_2);CHKERRQ(ierr); /* zero BCs so don't set */ 1627 ierr = DMGlobalToLocalBegin(dmEnergy, glob_v2, INSERT_VALUES, v2_2);CHKERRQ(ierr); 1628 ierr = DMGlobalToLocalEnd (dmEnergy, glob_v2, INSERT_VALUES, v2_2);CHKERRQ(ierr); 1629 ierr = DMViewFromOptions(dmEnergy,NULL, "-energy_dm_view");CHKERRQ(ierr); 1630 ierr = VecViewFromOptions(glob_v2,NULL, "-energy_vec_view");CHKERRQ(ierr); 1631 ierr = DMRestoreGlobalVector(dmEnergy, &glob_v2);CHKERRQ(ierr); 1632 } 1633 /* append part of the IP data for each grid */ 1634 for (ej = 0 ; ej < numCells[grid]; ++ej, ++outer_ej) { 1635 PetscScalar *coefs = NULL; 1636 PetscReal vj[LANDAU_MAX_NQ*LANDAU_DIM],detJj[LANDAU_MAX_NQ], Jdummy[LANDAU_MAX_NQ*LANDAU_DIM*LANDAU_DIM], c0 = C_0(ctx->v_0), c02 = PetscSqr(c0); 1637 invJe = invJ_a + outer_ej*Nq*dim*dim; 1638 ierr = DMPlexComputeCellGeometryFEM(ctx->plex[grid], ej+cStart, quad, vj, Jdummy, invJe, detJj);CHKERRQ(ierr); 1639 if (ctx->use_energy_tensor_trick) { 1640 ierr = DMPlexVecGetClosure(dmEnergy, e_section, v2_2, ej+cStart, NULL, &coefs);CHKERRQ(ierr); 1641 } 1642 /* create static point data */ 1643 for (PetscInt qj = 0; qj < Nq; qj++, outer_ipidx++) { 1644 const PetscInt gidx = outer_ipidx; 1645 const PetscReal *invJ = &invJe[qj*dim*dim]; 1646 ww [gidx] = detJj[qj] * quadWeights[qj]; 1647 if (dim==2) ww [gidx] *= vj[qj * dim + 0]; /* cylindrical coordinate, w/o 2pi */ 1648 // get xx, yy, zz 1649 if (ctx->use_energy_tensor_trick) { 1650 double refSpaceDer[3],eGradPhi[3]; 1651 const PetscReal * const DD = Tf[0]->T[1]; 1652 const PetscReal *Dq = &DD[qj*Nb*dim]; 1653 for (int d = 0; d < 3; ++d) refSpaceDer[d] = eGradPhi[d] = 0.0; 1654 for (int b = 0; b < Nb; ++b) { 1655 for (int d = 0; d < dim; ++d) refSpaceDer[d] += Dq[b*dim+d]*PetscRealPart(coefs[b]); 1656 } 1657 xx[gidx] = 1e10; 1658 if (ctx->use_relativistic_corrections) { 1659 double dg2_c2 = 0; 1660 //for (int d = 0; d < dim; ++d) refSpaceDer[d] *= c02; 1661 for (int d = 0; d < dim; ++d) dg2_c2 += PetscSqr(refSpaceDer[d]); 1662 dg2_c2 *= (double)c02; 1663 if (dg2_c2 >= .999) { 1664 xx[gidx] = vj[qj * dim + 0]; /* coordinate */ 1665 yy[gidx] = vj[qj * dim + 1]; 1666 if (dim==3) zz[gidx] = vj[qj * dim + 2]; 1667 PetscPrintf(ctx->comm,"Error: %12.5e %D.%D) dg2/c02 = %12.5e x= %12.5e %12.5e %12.5e\n",PetscSqrtReal(xx[gidx]*xx[gidx] + yy[gidx]*yy[gidx] + zz[gidx]*zz[gidx]), ej, qj, dg2_c2, xx[gidx],yy[gidx],zz[gidx]); 1668 } else { 1669 PetscReal fact = c02/PetscSqrtReal(1. - dg2_c2); 1670 for (int d = 0; d < dim; ++d) refSpaceDer[d] *= fact; 1671 // could test with other point u' that (grad - grad') * U (refSpaceDer, refSpaceDer') == 0 1672 } 1673 } 1674 if (xx[gidx] == 1e10) { 1675 for (int d = 0; d < dim; ++d) { 1676 for (int e = 0 ; e < dim; ++e) { 1677 eGradPhi[d] += invJ[e*dim+d]*refSpaceDer[e]; 1678 } 1679 } 1680 xx[gidx] = eGradPhi[0]; 1681 yy[gidx] = eGradPhi[1]; 1682 if (dim==3) zz[gidx] = eGradPhi[2]; 1683 } 1684 } else { 1685 xx[gidx] = vj[qj * dim + 0]; /* coordinate */ 1686 yy[gidx] = vj[qj * dim + 1]; 1687 if (dim==3) zz[gidx] = vj[qj * dim + 2]; 1688 } 1689 } /* q */ 1690 if (ctx->use_energy_tensor_trick) { 1691 ierr = DMPlexVecRestoreClosure(dmEnergy, e_section, v2_2, ej+cStart, NULL, &coefs);CHKERRQ(ierr); 1692 } 1693 } /* ej */ 1694 if (ctx->use_energy_tensor_trick) { 1695 ierr = DMRestoreLocalVector(dmEnergy, &v2_2);CHKERRQ(ierr); 1696 ierr = DMDestroy(&dmEnergy);CHKERRQ(ierr); 1697 } 1698 } /* grid */ 1699 if (ctx->use_energy_tensor_trick) { 1700 ierr = PetscFEDestroy(&fe);CHKERRQ(ierr); 1701 } 1702 /* cache static data */ 1703 if (ctx->deviceType == LANDAU_CUDA || ctx->deviceType == LANDAU_KOKKOS) { 1704 #if defined(PETSC_HAVE_CUDA) || defined(PETSC_HAVE_KOKKOS_KERNELS) 1705 PetscReal invMass[LANDAU_MAX_SPECIES],nu_alpha[LANDAU_MAX_SPECIES], nu_beta[LANDAU_MAX_SPECIES]; 1706 for (PetscInt grid = 0; grid < ctx->num_grids ; grid++) { 1707 for (PetscInt ii=ctx->species_offset[grid];ii<ctx->species_offset[grid+1];ii++) { 1708 invMass[ii] = ctx->m_0/ctx->masses[ii]; 1709 nu_alpha[ii] = PetscSqr(ctx->charges[ii]/ctx->m_0)*ctx->m_0/ctx->masses[ii]; 1710 nu_beta[ii] = PetscSqr(ctx->charges[ii]/ctx->epsilon0)*ctx->lnLam / (8*PETSC_PI) * ctx->t_0*ctx->n_0/PetscPowReal(ctx->v_0,3); 1711 } 1712 } 1713 if (ctx->deviceType == LANDAU_CUDA) { 1714 #if defined(PETSC_HAVE_CUDA) 1715 ierr = LandauCUDAStaticDataSet(ctx->plex[0], Nq, ctx->batch_sz, ctx->num_grids, numCells, ctx->species_offset, ctx->mat_offset, 1716 nu_alpha, nu_beta, invMass, invJ_a, xx, yy, zz, ww, &ctx->SData_d);CHKERRQ(ierr); 1717 #else 1718 SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-landau_device_type %s not built","cuda"); 1719 #endif 1720 } else if (ctx->deviceType == LANDAU_KOKKOS) { 1721 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 1722 ierr = LandauKokkosStaticDataSet(ctx->plex[0], Nq, ctx->batch_sz, ctx->num_grids, numCells, ctx->species_offset, ctx->mat_offset, 1723 nu_alpha, nu_beta, invMass,invJ_a,xx,yy,zz,ww,&ctx->SData_d);CHKERRQ(ierr); 1724 #else 1725 SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-landau_device_type %s not built","kokkos"); 1726 #endif 1727 } 1728 #endif 1729 /* free */ 1730 ierr = PetscFree4(ww,xx,yy,invJ_a);CHKERRQ(ierr); 1731 if (dim==3) { 1732 ierr = PetscFree(zz);CHKERRQ(ierr); 1733 } 1734 } else { /* CPU version, just copy in, only use part */ 1735 ctx->SData_d.w = (void*)ww; 1736 ctx->SData_d.x = (void*)xx; 1737 ctx->SData_d.y = (void*)yy; 1738 ctx->SData_d.z = (void*)zz; 1739 ctx->SData_d.invJ = (void*)invJ_a; 1740 } 1741 ierr = PetscLogEventEnd(ctx->events[7],0,0,0,0);CHKERRQ(ierr); 1742 } // initialize 1743 PetscFunctionReturn(0); 1744 } 1745 1746 /*@C 1747 LandauCreateVelocitySpace - Create a DMPlex velocity space mesh 1748 1749 Collective on comm 1750 1751 Input Parameters: 1752 + comm - The MPI communicator 1753 . dim - velocity space dimension (2 for axisymmetric, 3 for full 3X + 3V solver) 1754 - prefix - prefix for options (not tested) 1755 1756 Output Parameter: 1757 . pack - The DM object representing the mesh 1758 + X - A vector (user destroys) 1759 - J - Optional matrix (object destroys) 1760 1761 Level: beginner 1762 1763 .keywords: mesh 1764 .seealso: DMPlexCreate(), LandauDestroyVelocitySpace() 1765 @*/ 1766 PetscErrorCode LandauCreateVelocitySpace(MPI_Comm comm, PetscInt dim, const char prefix[], Vec *X, Mat *J, DM *pack) 1767 { 1768 PetscErrorCode ierr; 1769 LandauCtx *ctx; 1770 Vec Xsub[LANDAU_MAX_GRIDS]; 1771 1772 PetscFunctionBegin; 1773 if (dim!=2 && dim!=3) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "Only 2D and 3D supported"); 1774 if (LANDAU_DIM != dim) SETERRQ2(PETSC_COMM_SELF, PETSC_ERR_PLIB, "dim %D != LANDAU_DIM %d",dim,LANDAU_DIM); 1775 ierr = PetscNew(&ctx);CHKERRQ(ierr); 1776 ctx->comm = comm; /* used for diagnostics and global errors */ 1777 /* process options */ 1778 ierr = ProcessOptions(ctx,prefix);CHKERRQ(ierr); 1779 if (dim==2) ctx->use_relativistic_corrections = PETSC_FALSE; 1780 /* Create Mesh */ 1781 ierr = DMCompositeCreate(PETSC_COMM_SELF,pack);CHKERRQ(ierr); 1782 ierr = PetscLogEventBegin(ctx->events[13],0,0,0,0);CHKERRQ(ierr); 1783 ierr = LandauDMCreateVMeshes(PETSC_COMM_SELF, dim, prefix, ctx, *pack);CHKERRQ(ierr); // creates grids (Forest of AMR) 1784 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 1785 /* create FEM */ 1786 ierr = SetupDS(ctx->plex[grid],dim,grid,ctx);CHKERRQ(ierr); 1787 /* set initial state */ 1788 ierr = DMCreateGlobalVector(ctx->plex[grid],&Xsub[grid]);CHKERRQ(ierr); 1789 ierr = PetscObjectSetName((PetscObject) Xsub[grid], "u_orig");CHKERRQ(ierr); 1790 /* initial static refinement, no solve */ 1791 ierr = LandauSetInitialCondition(ctx->plex[grid], Xsub[grid], grid, 0, ctx);CHKERRQ(ierr); 1792 /* forest refinement - forest goes in (if forest), plex comes out */ 1793 if (ctx->use_p4est) { 1794 DM plex; 1795 ierr = adapt(grid,ctx,&Xsub[grid]);CHKERRQ(ierr); // forest goes in, plex comes out 1796 ierr = DMViewFromOptions(ctx->plex[grid],NULL,"-dm_landau_amr_dm_view");CHKERRQ(ierr); // need to differentiate - todo 1797 ierr = VecViewFromOptions(Xsub[grid], NULL, "-dm_landau_amr_vec_view");CHKERRQ(ierr); 1798 // convert to plex, all done with this level 1799 ierr = DMConvert(ctx->plex[grid], DMPLEX, &plex);CHKERRQ(ierr); 1800 ierr = DMDestroy(&ctx->plex[grid]);CHKERRQ(ierr); 1801 ctx->plex[grid] = plex; 1802 } 1803 #if !defined(LANDAU_SPECIES_MAJOR) 1804 ierr = DMCompositeAddDM(*pack,ctx->plex[grid]);CHKERRQ(ierr); 1805 #else 1806 for (PetscInt b_id=0;b_id<ctx->batch_sz;b_id++) { // add batch size DMs for this species grid 1807 ierr = DMCompositeAddDM(*pack,ctx->plex[grid]);CHKERRQ(ierr); 1808 } 1809 #endif 1810 ierr = DMSetApplicationContext(ctx->plex[grid], ctx);CHKERRQ(ierr); 1811 } 1812 #if !defined(LANDAU_SPECIES_MAJOR) 1813 // stack the batched DMs, could do it all here!!! b_id=0 1814 for (PetscInt b_id=1;b_id<ctx->batch_sz;b_id++) { 1815 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 1816 ierr = DMCompositeAddDM(*pack,ctx->plex[grid]);CHKERRQ(ierr); 1817 } 1818 } 1819 #endif 1820 // create ctx->mat_offset 1821 ctx->mat_offset[0] = 0; 1822 for (PetscInt grid=0 ; grid < ctx->num_grids ; grid++) { 1823 PetscInt n; 1824 ierr = VecGetLocalSize(Xsub[grid],&n);CHKERRQ(ierr); 1825 ctx->mat_offset[grid+1] = ctx->mat_offset[grid] + n; 1826 } 1827 // creat Jac 1828 ierr = DMSetApplicationContext(*pack, ctx);CHKERRQ(ierr); 1829 ierr = DMSetFromOptions(*pack);CHKERRQ(ierr); 1830 ierr = DMCreateMatrix(*pack, &ctx->J);CHKERRQ(ierr); 1831 ierr = MatSetOption(ctx->J, MAT_IGNORE_ZERO_ENTRIES, PETSC_TRUE);CHKERRQ(ierr); 1832 ierr = MatSetOption(ctx->J, MAT_STRUCTURALLY_SYMMETRIC, PETSC_TRUE);CHKERRQ(ierr); 1833 ierr = PetscObjectSetName((PetscObject)ctx->J, "Jac");CHKERRQ(ierr); 1834 if (J) *J = ctx->J; 1835 // construct X, copy data in 1836 ierr = DMCreateGlobalVector(*pack,X);CHKERRQ(ierr); 1837 for (PetscInt grid=0 ; grid < ctx->num_grids ; grid++) { 1838 PetscInt n; 1839 ierr = VecGetLocalSize(Xsub[grid],&n);CHKERRQ(ierr); 1840 for (PetscInt b_id = 0 ; b_id < ctx->batch_sz ; b_id++) { 1841 PetscScalar const *values; 1842 const PetscInt moffset = LAND_MOFFSET(b_id,grid,ctx->batch_sz,ctx->num_grids,ctx->mat_offset); 1843 ierr = LandauSetInitialCondition(ctx->plex[grid], Xsub[grid], grid, b_id, ctx);CHKERRQ(ierr); 1844 ierr = VecGetArrayRead(Xsub[grid],&values);CHKERRQ(ierr); 1845 for (int i=0, idx = moffset; i<n; i++, idx++) { 1846 ierr = VecSetValue(*X,idx,values[i],INSERT_VALUES);CHKERRQ(ierr); 1847 } 1848 ierr = VecRestoreArrayRead(Xsub[grid],&values);CHKERRQ(ierr); 1849 } 1850 } 1851 // cleanup 1852 for (PetscInt grid=0 ; grid < ctx->num_grids ; grid++) { 1853 ierr = VecDestroy(&Xsub[grid]);CHKERRQ(ierr); 1854 } 1855 /* check for correct matrix type */ 1856 if (ctx->gpu_assembly) { /* we need GPU object with GPU assembly */ 1857 PetscBool flg; 1858 if (ctx->deviceType == LANDAU_CUDA) { 1859 ierr = PetscObjectTypeCompareAny((PetscObject)ctx->J,&flg,MATSEQAIJCUSPARSE,MATMPIAIJCUSPARSE,MATAIJCUSPARSE,"");CHKERRQ(ierr); 1860 if (!flg) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"must use '-dm_mat_type aijcusparse -dm_vec_type cuda' for GPU assembly and Cuda or use '-dm_landau_device_type cpu'"); 1861 } else if (ctx->deviceType == LANDAU_KOKKOS) { 1862 ierr = PetscObjectTypeCompareAny((PetscObject)ctx->J,&flg,MATSEQAIJKOKKOS,MATMPIAIJKOKKOS,MATAIJKOKKOS,"");CHKERRQ(ierr); 1863 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 1864 if (!flg) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"must use '-dm_mat_type aijkokkos -dm_vec_type kokkos' for GPU assembly and Kokkos or use '-dm_landau_device_type cpu'"); 1865 #else 1866 if (!flg) SETERRQ(ctx->comm,PETSC_ERR_ARG_WRONG,"must configure with '--download-kokkos-kernels' for GPU assembly and Kokkos or use '-dm_landau_device_type cpu'"); 1867 #endif 1868 } 1869 } 1870 // create AMR GPU assembly maps and static GPU data 1871 ierr = CreateStaticGPUData(dim,ctx);CHKERRQ(ierr); 1872 1873 ierr = PetscLogEventEnd(ctx->events[13],0,0,0,0);CHKERRQ(ierr); 1874 PetscFunctionReturn(0); 1875 } 1876 1877 /*@ 1878 LandauDestroyVelocitySpace - Destroy a DMPlex velocity space mesh 1879 1880 Collective on dm 1881 1882 Input/Output Parameters: 1883 . dm - the dm to destroy 1884 1885 Level: beginner 1886 1887 .keywords: mesh 1888 .seealso: LandauCreateVelocitySpace() 1889 @*/ 1890 PetscErrorCode LandauDestroyVelocitySpace(DM *dm) 1891 { 1892 PetscErrorCode ierr,ii; 1893 LandauCtx *ctx; 1894 PetscFunctionBegin; 1895 ierr = DMGetApplicationContext(*dm, &ctx);CHKERRQ(ierr); 1896 ierr = MatDestroy(&ctx->M);CHKERRQ(ierr); 1897 ierr = MatDestroy(&ctx->J);CHKERRQ(ierr); 1898 for (ii=0;ii<ctx->num_species;ii++) { 1899 ierr = PetscFEDestroy(&ctx->fe[ii]);CHKERRQ(ierr); 1900 } 1901 if (ctx->deviceType == LANDAU_CUDA) { 1902 #if defined(PETSC_HAVE_CUDA) 1903 ierr = LandauCUDAStaticDataClear(&ctx->SData_d);CHKERRQ(ierr); 1904 #else 1905 SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-landau_device_type %s not built","cuda"); 1906 #endif 1907 } else if (ctx->deviceType == LANDAU_KOKKOS) { 1908 #if defined(PETSC_HAVE_KOKKOS_KERNELS) 1909 ierr = LandauKokkosStaticDataClear(&ctx->SData_d);CHKERRQ(ierr); 1910 #else 1911 SETERRQ1(ctx->comm,PETSC_ERR_ARG_WRONG,"-landau_device_type %s not built","kokkos"); 1912 #endif 1913 } else { 1914 if (ctx->SData_d.x) { /* in a CPU run */ 1915 PetscReal *invJ = (PetscReal*)ctx->SData_d.invJ, *xx = (PetscReal*)ctx->SData_d.x, *yy = (PetscReal*)ctx->SData_d.y, *zz = (PetscReal*)ctx->SData_d.z, *ww = (PetscReal*)ctx->SData_d.w; 1916 ierr = PetscFree4(ww,xx,yy,invJ);CHKERRQ(ierr); 1917 if (zz) { 1918 ierr = PetscFree(zz);CHKERRQ(ierr); 1919 } 1920 } 1921 } 1922 1923 if (ctx->times[LANDAU_MATRIX_TOTAL] > 0) { // OMP timings 1924 ierr = PetscPrintf(ctx->comm, "TSStep N 1.0 %10.3e\n",ctx->times[LANDAU_EX2_TSSOLVE]);CHKERRQ(ierr); 1925 ierr = PetscPrintf(ctx->comm, "2: Solve: %10.3e with %D threads\n",ctx->times[LANDAU_EX2_TSSOLVE] - ctx->times[LANDAU_MATRIX_TOTAL],ctx->batch_sz);CHKERRQ(ierr); 1926 ierr = PetscPrintf(ctx->comm, "3: Landau: %10.3e\n",ctx->times[LANDAU_MATRIX_TOTAL]);CHKERRQ(ierr); 1927 ierr = PetscPrintf(ctx->comm, "Landau Jacobian %D 1.0 %10.3e\n",(PetscInt)ctx->times[LANDAU_JACOBIAN_COUNT],ctx->times[LANDAU_JACOBIAN]);CHKERRQ(ierr); 1928 ierr = PetscPrintf(ctx->comm, "Landau Operator N 1.0 %10.3e\n",ctx->times[LANDAU_OPERATOR]);CHKERRQ(ierr); 1929 ierr = PetscPrintf(ctx->comm, "Landau Mass N 1.0 %10.3e\n",ctx->times[LANDAU_MASS]);CHKERRQ(ierr); 1930 ierr = PetscPrintf(ctx->comm, " Jac-f-df (GPU) N 1.0 %10.3e\n",ctx->times[LANDAU_F_DF]);CHKERRQ(ierr); 1931 ierr = PetscPrintf(ctx->comm, " Kernel (GPU) N 1.0 %10.3e\n",ctx->times[LANDAU_KERNEL]);CHKERRQ(ierr); 1932 ierr = PetscPrintf(ctx->comm, "MatLUFactorNum X 1.0 %10.3e\n",ctx->times[KSP_FACTOR]);CHKERRQ(ierr); 1933 ierr = PetscPrintf(ctx->comm, "MatSolve X 1.0 %10.3e\n",ctx->times[KSP_SOLVE]);CHKERRQ(ierr); 1934 } 1935 for (PetscInt grid=0 ; grid < ctx->num_grids ; grid++) { 1936 ierr = DMDestroy(&ctx->plex[grid]);CHKERRQ(ierr); 1937 } 1938 PetscFree(ctx); 1939 ierr = DMDestroy(dm);CHKERRQ(ierr); 1940 PetscFunctionReturn(0); 1941 } 1942 1943 /* < v, ru > */ 1944 static void f0_s_den(PetscInt dim, PetscInt Nf, PetscInt NfAux, 1945 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 1946 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 1947 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar *f0) 1948 { 1949 PetscInt ii = (PetscInt)PetscRealPart(constants[0]); 1950 f0[0] = u[ii]; 1951 } 1952 1953 /* < v, ru > */ 1954 static void f0_s_mom(PetscInt dim, PetscInt Nf, PetscInt NfAux, 1955 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 1956 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 1957 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar *f0) 1958 { 1959 PetscInt ii = (PetscInt)PetscRealPart(constants[0]), jj = (PetscInt)PetscRealPart(constants[1]); 1960 f0[0] = x[jj]*u[ii]; /* x momentum */ 1961 } 1962 1963 static void f0_s_v2(PetscInt dim, PetscInt Nf, PetscInt NfAux, 1964 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 1965 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 1966 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar *f0) 1967 { 1968 PetscInt i, ii = (PetscInt)PetscRealPart(constants[0]); 1969 double tmp1 = 0.; 1970 for (i = 0; i < dim; ++i) tmp1 += x[i]*x[i]; 1971 f0[0] = tmp1*u[ii]; 1972 } 1973 1974 static PetscErrorCode gamma_n_f(PetscInt dim, PetscReal time, const PetscReal x[], PetscInt Nf, PetscScalar *u, void *actx) 1975 { 1976 const PetscReal *c2_0_arr = ((PetscReal*)actx); 1977 const PetscReal c02 = c2_0_arr[0]; 1978 1979 PetscFunctionBegin; 1980 for (int s = 0 ; s < Nf ; s++) { 1981 PetscReal tmp1 = 0.; 1982 for (int i = 0; i < dim; ++i) tmp1 += x[i]*x[i]; 1983 #if defined(PETSC_USE_DEBUG) 1984 u[s] = PetscSqrtReal(1. + tmp1/c02);// u[0] = PetscSqrtReal(1. + xx); 1985 #else 1986 { 1987 PetscReal xx = tmp1/c02; 1988 u[s] = xx/(PetscSqrtReal(1. + xx) + 1.); // better conditioned = xx/(PetscSqrtReal(1. + xx) + 1.) 1989 } 1990 #endif 1991 } 1992 PetscFunctionReturn(0); 1993 } 1994 1995 /* < v, ru > */ 1996 static void f0_s_rden(PetscInt dim, PetscInt Nf, PetscInt NfAux, 1997 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 1998 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 1999 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar *f0) 2000 { 2001 PetscInt ii = (PetscInt)PetscRealPart(constants[0]); 2002 f0[0] = 2.*PETSC_PI*x[0]*u[ii]; 2003 } 2004 2005 /* < v, ru > */ 2006 static void f0_s_rmom(PetscInt dim, PetscInt Nf, PetscInt NfAux, 2007 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 2008 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 2009 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar *f0) 2010 { 2011 PetscInt ii = (PetscInt)PetscRealPart(constants[0]); 2012 f0[0] = 2.*PETSC_PI*x[0]*x[1]*u[ii]; 2013 } 2014 2015 static void f0_s_rv2(PetscInt dim, PetscInt Nf, PetscInt NfAux, 2016 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 2017 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 2018 PetscReal t, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar *f0) 2019 { 2020 PetscInt ii = (PetscInt)PetscRealPart(constants[0]); 2021 f0[0] = 2.*PETSC_PI*x[0]*(x[0]*x[0] + x[1]*x[1])*u[ii]; 2022 } 2023 2024 /*@ 2025 LandauPrintNorms - collects moments and prints them 2026 2027 Collective on dm 2028 2029 Input Parameters: 2030 + X - the state 2031 - stepi - current step to print 2032 2033 Level: beginner 2034 2035 .keywords: mesh 2036 .seealso: LandauCreateVelocitySpace() 2037 @*/ 2038 PetscErrorCode LandauPrintNorms(Vec X, PetscInt stepi) 2039 { 2040 PetscErrorCode ierr; 2041 LandauCtx *ctx; 2042 PetscDS prob; 2043 DM pack; 2044 PetscInt cStart, cEnd, dim, ii, i0, nDMs; 2045 PetscScalar xmomentumtot=0, ymomentumtot=0, zmomentumtot=0, energytot=0, densitytot=0, tt[LANDAU_MAX_SPECIES]; 2046 PetscScalar xmomentum[LANDAU_MAX_SPECIES], ymomentum[LANDAU_MAX_SPECIES], zmomentum[LANDAU_MAX_SPECIES], energy[LANDAU_MAX_SPECIES], density[LANDAU_MAX_SPECIES]; 2047 Vec *globXArray; 2048 2049 PetscFunctionBegin; 2050 ierr = VecGetDM(X, &pack);CHKERRQ(ierr); 2051 if (!pack) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "Vector has no DM"); 2052 ierr = DMGetDimension(pack, &dim);CHKERRQ(ierr); 2053 if (dim!=2 && dim!=3) SETERRQ1(PETSC_COMM_SELF, PETSC_ERR_PLIB, "dim= %D",dim); 2054 ierr = DMGetApplicationContext(pack, &ctx);CHKERRQ(ierr); 2055 if (!ctx) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "no context"); 2056 /* print momentum and energy */ 2057 ierr = DMCompositeGetNumberDM(pack,&nDMs);CHKERRQ(ierr); 2058 if (nDMs != ctx->num_grids*ctx->batch_sz) SETERRQ2(PETSC_COMM_WORLD, PETSC_ERR_PLIB, "#DM wrong %D %D",nDMs,ctx->num_grids*ctx->batch_sz); 2059 ierr = PetscMalloc(sizeof(*globXArray)*nDMs, &globXArray);CHKERRQ(ierr); 2060 ierr = DMCompositeGetAccessArray(pack, X, nDMs, NULL, globXArray);CHKERRQ(ierr); 2061 for (PetscInt grid = 0; grid < ctx->num_grids ; grid++) { 2062 Vec Xloc = globXArray[ LAND_PACK_IDX(ctx->batch_view_idx,grid) ]; 2063 ierr = DMGetDS(ctx->plex[grid], &prob);CHKERRQ(ierr); 2064 for (ii=ctx->species_offset[grid],i0=0;ii<ctx->species_offset[grid+1];ii++,i0++) { 2065 PetscScalar user[2] = { (PetscScalar)i0, (PetscScalar)ctx->charges[ii]}; 2066 ierr = PetscDSSetConstants(prob, 2, user);CHKERRQ(ierr); 2067 if (dim==2) { /* 2/3X + 3V (cylindrical coordinates) */ 2068 ierr = PetscDSSetObjective(prob, 0, &f0_s_rden);CHKERRQ(ierr); 2069 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2070 density[ii] = tt[0]*ctx->n_0*ctx->charges[ii]; 2071 ierr = PetscDSSetObjective(prob, 0, &f0_s_rmom);CHKERRQ(ierr); 2072 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2073 zmomentum[ii] = tt[0]*ctx->n_0*ctx->v_0*ctx->masses[ii]; 2074 ierr = PetscDSSetObjective(prob, 0, &f0_s_rv2);CHKERRQ(ierr); 2075 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2076 energy[ii] = tt[0]*0.5*ctx->n_0*ctx->v_0*ctx->v_0*ctx->masses[ii]; 2077 zmomentumtot += zmomentum[ii]; 2078 energytot += energy[ii]; 2079 densitytot += density[ii]; 2080 ierr = PetscPrintf(ctx->comm, "%3D) species-%D: charge density= %20.13e z-momentum= %20.13e energy= %20.13e",stepi,ii,PetscRealPart(density[ii]),PetscRealPart(zmomentum[ii]),PetscRealPart(energy[ii]));CHKERRQ(ierr); 2081 } else { /* 2/3Xloc + 3V */ 2082 ierr = PetscDSSetObjective(prob, 0, &f0_s_den);CHKERRQ(ierr); 2083 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2084 density[ii] = tt[0]*ctx->n_0*ctx->charges[ii]; 2085 ierr = PetscDSSetObjective(prob, 0, &f0_s_mom);CHKERRQ(ierr); 2086 user[1] = 0; 2087 ierr = PetscDSSetConstants(prob, 2, user);CHKERRQ(ierr); 2088 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2089 xmomentum[ii] = tt[0]*ctx->n_0*ctx->v_0*ctx->masses[ii]; 2090 user[1] = 1; 2091 ierr = PetscDSSetConstants(prob, 2, user);CHKERRQ(ierr); 2092 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2093 ymomentum[ii] = tt[0]*ctx->n_0*ctx->v_0*ctx->masses[ii]; 2094 user[1] = 2; 2095 ierr = PetscDSSetConstants(prob, 2, user);CHKERRQ(ierr); 2096 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2097 zmomentum[ii] = tt[0]*ctx->n_0*ctx->v_0*ctx->masses[ii]; 2098 if (ctx->use_relativistic_corrections) { 2099 /* gamma * M * f */ 2100 if (ii==0 && grid==0) { // do all at once 2101 Vec Mf, globGamma, *globMfArray, *globGammaArray; 2102 PetscErrorCode (*gammaf[1])(PetscInt, PetscReal, const PetscReal [], PetscInt, PetscScalar [], void *) = {gamma_n_f}; 2103 PetscReal *c2_0[1], data[1]; 2104 2105 ierr = VecDuplicate(X,&globGamma);CHKERRQ(ierr); 2106 ierr = VecDuplicate(X,&Mf);CHKERRQ(ierr); 2107 ierr = PetscMalloc(sizeof(*globMfArray)*nDMs, &globMfArray);CHKERRQ(ierr); 2108 ierr = PetscMalloc(sizeof(*globMfArray)*nDMs, &globGammaArray);CHKERRQ(ierr); 2109 /* M * f */ 2110 ierr = MatMult(ctx->M,X,Mf);CHKERRQ(ierr); 2111 /* gamma */ 2112 ierr = DMCompositeGetAccessArray(pack, globGamma, nDMs, NULL, globGammaArray);CHKERRQ(ierr); 2113 for (PetscInt grid = 0; grid < ctx->num_grids ; grid++) { // yes a grid loop in a grid loop to print nice, need to fix for batching 2114 Vec v1 = globGammaArray[ LAND_PACK_IDX(ctx->batch_view_idx,grid) ]; 2115 data[0] = PetscSqr(C_0(ctx->v_0)); 2116 c2_0[0] = &data[0]; 2117 ierr = DMProjectFunction(ctx->plex[grid], 0., gammaf, (void**)c2_0, INSERT_ALL_VALUES, v1);CHKERRQ(ierr); 2118 } 2119 ierr = DMCompositeRestoreAccessArray(pack, globGamma, nDMs, NULL, globGammaArray);CHKERRQ(ierr); 2120 /* gamma * Mf */ 2121 ierr = DMCompositeGetAccessArray(pack, globGamma, nDMs, NULL, globGammaArray);CHKERRQ(ierr); 2122 ierr = DMCompositeGetAccessArray(pack, Mf, nDMs, NULL, globMfArray);CHKERRQ(ierr); 2123 for (PetscInt grid = 0; grid < ctx->num_grids ; grid++) { // yes a grid loop in a grid loop to print nice 2124 PetscInt Nf = ctx->species_offset[grid+1] - ctx->species_offset[grid], N, bs; 2125 Vec Mfsub = globMfArray[ LAND_PACK_IDX(ctx->batch_view_idx,grid) ], Gsub = globGammaArray[ LAND_PACK_IDX(ctx->batch_view_idx,grid) ], v1, v2; 2126 // get each component 2127 ierr = VecGetSize(Mfsub,&N);CHKERRQ(ierr); 2128 ierr = VecCreate(ctx->comm,&v1);CHKERRQ(ierr); 2129 ierr = VecSetSizes(v1,PETSC_DECIDE,N/Nf);CHKERRQ(ierr); 2130 ierr = VecCreate(ctx->comm,&v2);CHKERRQ(ierr); 2131 ierr = VecSetSizes(v2,PETSC_DECIDE,N/Nf);CHKERRQ(ierr); 2132 ierr = VecSetFromOptions(v1);CHKERRQ(ierr); // ??? 2133 ierr = VecSetFromOptions(v2);CHKERRQ(ierr); 2134 // get each component 2135 ierr = VecGetBlockSize(Gsub,&bs);CHKERRQ(ierr); 2136 if (bs != Nf) SETERRQ2(PETSC_COMM_SELF, PETSC_ERR_PLIB, "bs %D != num_species %D in Gsub",bs,Nf); 2137 ierr = VecGetBlockSize(Mfsub,&bs);CHKERRQ(ierr); 2138 if (bs != Nf) SETERRQ2(PETSC_COMM_SELF, PETSC_ERR_PLIB, "bs %D != num_species %D",bs,Nf); 2139 for (int i=0, ix=ctx->species_offset[grid] ; i<Nf ; i++, ix++) { 2140 PetscScalar val; 2141 ierr = VecStrideGather(Gsub,i,v1,INSERT_VALUES);CHKERRQ(ierr); 2142 ierr = VecStrideGather(Mfsub,i,v2,INSERT_VALUES);CHKERRQ(ierr); 2143 ierr = VecDot(v1,v2,&val);CHKERRQ(ierr); 2144 energy[ix] = PetscRealPart(val)*ctx->n_0*ctx->v_0*ctx->v_0*ctx->masses[ix]; 2145 } 2146 ierr = VecDestroy(&v1);CHKERRQ(ierr); 2147 ierr = VecDestroy(&v2);CHKERRQ(ierr); 2148 } /* grids */ 2149 ierr = DMCompositeRestoreAccessArray(pack, globGamma, nDMs, NULL, globGammaArray);CHKERRQ(ierr); 2150 ierr = DMCompositeRestoreAccessArray(pack, Mf, nDMs, NULL, globMfArray);CHKERRQ(ierr); 2151 ierr = PetscFree(globGammaArray);CHKERRQ(ierr); 2152 ierr = PetscFree(globMfArray);CHKERRQ(ierr); 2153 ierr = VecDestroy(&globGamma);CHKERRQ(ierr); 2154 ierr = VecDestroy(&Mf);CHKERRQ(ierr); 2155 } 2156 } else { 2157 ierr = PetscDSSetObjective(prob, 0, &f0_s_v2);CHKERRQ(ierr); 2158 ierr = DMPlexComputeIntegralFEM(ctx->plex[grid],Xloc,tt,ctx);CHKERRQ(ierr); 2159 energy[ii] = 0.5*tt[0]*ctx->n_0*ctx->v_0*ctx->v_0*ctx->masses[ii]; 2160 } 2161 ierr = PetscPrintf( ctx->comm, "%3D) species %D: density=%20.13e, x-momentum=%20.13e, y-momentum=%20.13e, z-momentum=%20.13e, energy=%21.13e", 2162 stepi,ii,PetscRealPart(density[ii]),PetscRealPart(xmomentum[ii]),PetscRealPart(ymomentum[ii]),PetscRealPart(zmomentum[ii]),PetscRealPart(energy[ii]));CHKERRQ(ierr); 2163 xmomentumtot += xmomentum[ii]; 2164 ymomentumtot += ymomentum[ii]; 2165 zmomentumtot += zmomentum[ii]; 2166 energytot += energy[ii]; 2167 densitytot += density[ii]; 2168 } 2169 if (ctx->num_species>1) PetscPrintf(ctx->comm, "\n"); 2170 } 2171 } 2172 ierr = DMCompositeRestoreAccessArray(pack, X, nDMs, NULL, globXArray);CHKERRQ(ierr); 2173 ierr = PetscFree(globXArray);CHKERRQ(ierr); 2174 /* totals */ 2175 ierr = DMPlexGetHeightStratum(ctx->plex[0],0,&cStart,&cEnd);CHKERRQ(ierr); 2176 if (ctx->num_species>1) { 2177 if (dim==2) { 2178 ierr = PetscPrintf(ctx->comm, "\t%3D) Total: charge density=%21.13e, momentum=%21.13e, energy=%21.13e (m_i[0]/m_e = %g, %D cells on electron grid)", 2179 stepi,(double)PetscRealPart(densitytot),(double)PetscRealPart(zmomentumtot),(double)PetscRealPart(energytot),(double)(ctx->masses[1]/ctx->masses[0]),cEnd-cStart);CHKERRQ(ierr); 2180 } else { 2181 ierr = PetscPrintf(ctx->comm, "\t%3D) Total: charge density=%21.13e, x-momentum=%21.13e, y-momentum=%21.13e, z-momentum=%21.13e, energy=%21.13e (m_i[0]/m_e = %g, %D cells)", 2182 stepi,(double)PetscRealPart(densitytot),(double)PetscRealPart(xmomentumtot),(double)PetscRealPart(ymomentumtot),(double)PetscRealPart(zmomentumtot),(double)PetscRealPart(energytot),(double)(ctx->masses[1]/ctx->masses[0]),cEnd-cStart);CHKERRQ(ierr); 2183 } 2184 } else { 2185 ierr = PetscPrintf(ctx->comm, " -- %D cells",cEnd-cStart);CHKERRQ(ierr); 2186 } 2187 if (ctx->verbose > 1) {ierr = PetscPrintf(ctx->comm,", %D sub (vector) threads\n",ctx->subThreadBlockSize);CHKERRQ(ierr);} 2188 else {ierr = PetscPrintf(ctx->comm,"\n");CHKERRQ(ierr);} 2189 PetscFunctionReturn(0); 2190 } 2191 2192 /* < v, u > */ 2193 static void g0_1(PetscInt dim, PetscInt Nf, PetscInt NfAux, 2194 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 2195 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 2196 PetscReal t, PetscReal u_tShift, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar g0[]) 2197 { 2198 g0[0] = 1.; 2199 } 2200 2201 /* < v, u > */ 2202 static void g0_r(PetscInt dim, PetscInt Nf, PetscInt NfAux, 2203 const PetscInt uOff[], const PetscInt uOff_x[], const PetscScalar u[], const PetscScalar u_t[], const PetscScalar u_x[], 2204 const PetscInt aOff[], const PetscInt aOff_x[], const PetscScalar a[], const PetscScalar a_t[], const PetscScalar a_x[], 2205 PetscReal t, PetscReal u_tShift, const PetscReal x[], PetscInt numConstants, const PetscScalar constants[], PetscScalar g0[]) 2206 { 2207 g0[0] = 2.*PETSC_PI*x[0]; 2208 } 2209 2210 /*@ 2211 LandauCreateMassMatrix - Create mass matrix for Landau 2212 2213 Collective on pack 2214 2215 Input Parameters: 2216 . pack - the DM object 2217 2218 Output Parameters: 2219 . Amat - The mass matrix (optional), mass matrix is added to the DM context 2220 2221 Level: beginner 2222 2223 .keywords: mesh 2224 .seealso: LandauCreateVelocitySpace() 2225 @*/ 2226 PetscErrorCode LandauCreateMassMatrix(DM pack, Mat *Amat) 2227 { 2228 DM mass_pack,massDM[LANDAU_MAX_GRIDS]; 2229 PetscDS prob; 2230 PetscInt ii,dim,N1=1,N2; 2231 PetscErrorCode ierr; 2232 LandauCtx *ctx; 2233 Mat packM,subM[LANDAU_MAX_GRIDS]; 2234 2235 PetscFunctionBegin; 2236 PetscValidHeaderSpecific(pack,DM_CLASSID,1); 2237 if (Amat) PetscValidPointer(Amat,2); 2238 ierr = DMGetApplicationContext(pack, &ctx);CHKERRQ(ierr); 2239 if (!ctx) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "no context"); 2240 ierr = DMGetDimension(pack, &dim);CHKERRQ(ierr); 2241 ierr = DMCompositeCreate(PetscObjectComm((PetscObject) pack),&mass_pack);CHKERRQ(ierr); 2242 /* create pack mass matrix */ 2243 for (PetscInt grid=0, ix=0 ; grid<ctx->num_grids ; grid++) { 2244 ierr = DMClone(ctx->plex[grid], &massDM[grid]);CHKERRQ(ierr); 2245 ierr = DMCopyFields(ctx->plex[grid], massDM[grid]);CHKERRQ(ierr); 2246 ierr = DMCreateDS(massDM[grid]);CHKERRQ(ierr); 2247 ierr = DMGetDS(massDM[grid], &prob);CHKERRQ(ierr); 2248 for (ix=0, ii=ctx->species_offset[grid];ii<ctx->species_offset[grid+1];ii++,ix++) { 2249 if (dim==3) {ierr = PetscDSSetJacobian(prob, ix, ix, g0_1, NULL, NULL, NULL);CHKERRQ(ierr);} 2250 else {ierr = PetscDSSetJacobian(prob, ix, ix, g0_r, NULL, NULL, NULL);CHKERRQ(ierr);} 2251 } 2252 #if !defined(LANDAU_SPECIES_MAJOR) 2253 ierr = DMCompositeAddDM(mass_pack,massDM[grid]);CHKERRQ(ierr); 2254 #else 2255 for (PetscInt b_id=0;b_id<ctx->batch_sz;b_id++) { // add batch size DMs for this species grid 2256 ierr = DMCompositeAddDM(mass_pack,massDM[grid]);CHKERRQ(ierr); 2257 } 2258 #endif 2259 ierr = DMCreateMatrix(massDM[grid], &subM[grid]);CHKERRQ(ierr); 2260 } 2261 #if !defined(LANDAU_SPECIES_MAJOR) 2262 // stack the batched DMs 2263 for (PetscInt b_id=1;b_id<ctx->batch_sz;b_id++) { 2264 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 2265 ierr = DMCompositeAddDM(mass_pack, massDM[grid]);CHKERRQ(ierr); 2266 } 2267 } 2268 #endif 2269 ierr = PetscOptionsInsertString(NULL,"-dm_preallocate_only"); 2270 ierr = DMSetFromOptions(mass_pack);CHKERRQ(ierr); 2271 ierr = DMCreateMatrix(mass_pack, &packM);CHKERRQ(ierr); 2272 ierr = PetscOptionsInsertString(NULL,"-dm_preallocate_only false"); 2273 ierr = MatSetOption(packM, MAT_IGNORE_ZERO_ENTRIES, PETSC_TRUE);CHKERRQ(ierr); 2274 ierr = MatSetOption(packM, MAT_STRUCTURALLY_SYMMETRIC, PETSC_TRUE);CHKERRQ(ierr); 2275 ierr = DMDestroy(&mass_pack);CHKERRQ(ierr); 2276 /* make mass matrix for each block */ 2277 for (PetscInt grid=0;grid<ctx->num_grids;grid++) { 2278 Vec locX; 2279 DM plex = massDM[grid]; 2280 ierr = DMGetLocalVector(plex, &locX);CHKERRQ(ierr); 2281 /* Mass matrix is independent of the input, so no need to fill locX */ 2282 ierr = DMPlexSNESComputeJacobianFEM(plex, locX, subM[grid], subM[grid], ctx);CHKERRQ(ierr); 2283 ierr = DMRestoreLocalVector(plex, &locX);CHKERRQ(ierr); 2284 ierr = DMDestroy(&massDM[grid]);CHKERRQ(ierr); 2285 } 2286 ierr = MatGetSize(ctx->J, &N1, NULL);CHKERRQ(ierr); 2287 ierr = MatGetSize(packM, &N2, NULL);CHKERRQ(ierr); 2288 if (N1 != N2) SETERRQ2(PetscObjectComm((PetscObject) pack), PETSC_ERR_PLIB, "Incorrect matrix sizes: |Jacobian| = %D, |Mass|=%D",N1,N2); 2289 /* assemble block diagonals */ 2290 for (PetscInt grid=0 ; grid<ctx->num_grids ; grid++) { 2291 Mat B = subM[grid]; 2292 PetscInt nloc, nzl, colbuf[1024], row; 2293 ierr = MatGetSize(B, &nloc, NULL);CHKERRQ(ierr); 2294 for (PetscInt b_id = 0 ; b_id < ctx->batch_sz ; b_id++) { 2295 const PetscInt moffset = LAND_MOFFSET(b_id,grid,ctx->batch_sz,ctx->num_grids,ctx->mat_offset); 2296 const PetscInt *cols; 2297 const PetscScalar *vals; 2298 for (int i=0 ; i<nloc ; i++) { 2299 ierr = MatGetRow(B,i,&nzl,&cols,&vals);CHKERRQ(ierr); 2300 if (nzl>1024) SETERRQ1(PetscObjectComm((PetscObject) pack), PETSC_ERR_PLIB, "Row too big: %D",nzl); 2301 for (int j=0; j<nzl; j++) colbuf[j] = cols[j] + moffset; 2302 row = i + moffset; 2303 ierr = MatSetValues(packM,1,&row,nzl,colbuf,vals,INSERT_VALUES);CHKERRQ(ierr); 2304 ierr = MatRestoreRow(B,i,&nzl,&cols,&vals);CHKERRQ(ierr); 2305 } 2306 } 2307 } 2308 // cleanup 2309 for (PetscInt grid=0 ; grid<ctx->num_grids ; grid++) { 2310 ierr = MatDestroy(&subM[grid]);CHKERRQ(ierr); 2311 } 2312 ierr = MatAssemblyBegin(packM,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2313 ierr = MatAssemblyEnd(packM,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 2314 ierr = PetscObjectSetName((PetscObject)packM, "mass");CHKERRQ(ierr); 2315 ierr = MatViewFromOptions(packM,NULL,"-dm_landau_mass_view");CHKERRQ(ierr); 2316 ctx->M = packM; /* this could be a noop, a = a */ 2317 if (Amat) *Amat = packM; 2318 PetscFunctionReturn(0); 2319 } 2320 2321 /*@ 2322 LandauIFunction - TS residual calculation 2323 2324 Collective on ts 2325 2326 Input Parameters: 2327 + TS - The time stepping context 2328 . time_dummy - current time (not used) 2329 - X - Current state 2330 + X_t - Time derivative of current state 2331 . actx - Landau context 2332 2333 Output Parameter: 2334 . F - The residual 2335 2336 Level: beginner 2337 2338 .keywords: mesh 2339 .seealso: LandauCreateVelocitySpace(), LandauIJacobian() 2340 @*/ 2341 PetscErrorCode LandauIFunction(TS ts, PetscReal time_dummy, Vec X, Vec X_t, Vec F, void *actx) 2342 { 2343 PetscErrorCode ierr; 2344 LandauCtx *ctx=(LandauCtx*)actx; 2345 PetscInt dim; 2346 DM pack; 2347 #if defined(PETSC_HAVE_THREADSAFETY) 2348 double starttime, endtime; 2349 #endif 2350 2351 PetscFunctionBegin; 2352 ierr = TSGetDM(ts,&pack);CHKERRQ(ierr); 2353 ierr = DMGetApplicationContext(pack, &ctx);CHKERRQ(ierr); 2354 if (!ctx) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "no context"); 2355 if (ctx->stage) { 2356 ierr = PetscLogStagePush(ctx->stage);CHKERRQ(ierr); 2357 } 2358 ierr = PetscLogEventBegin(ctx->events[11],0,0,0,0);CHKERRQ(ierr); 2359 ierr = PetscLogEventBegin(ctx->events[0],0,0,0,0);CHKERRQ(ierr); 2360 #if defined(PETSC_HAVE_THREADSAFETY) 2361 starttime = MPI_Wtime(); 2362 #endif 2363 ierr = DMGetDimension(pack, &dim);CHKERRQ(ierr); 2364 if (!ctx->aux_bool) { 2365 ierr = PetscInfo3(ts, "Create Landau Jacobian t=%g X=%p %s\n",time_dummy,X_t,ctx->aux_bool ? " -- seems to be in line search" : "");CHKERRQ(ierr); 2366 ierr = LandauFormJacobian_Internal(X,ctx->J,dim,0.0,(void*)ctx);CHKERRQ(ierr); 2367 ierr = MatViewFromOptions(ctx->J, NULL, "-dm_landau_jacobian_view");CHKERRQ(ierr); 2368 ctx->aux_bool = PETSC_TRUE; 2369 } else { 2370 ierr = PetscInfo(ts, "Skip forming Jacobian, has not changed (should check norm)\n");CHKERRQ(ierr); 2371 } 2372 /* mat vec for op */ 2373 ierr = MatMult(ctx->J,X,F);CHKERRQ(ierr);CHKERRQ(ierr); /* C*f */ 2374 /* add time term */ 2375 if (X_t) { 2376 ierr = MatMultAdd(ctx->M,X_t,F,F);CHKERRQ(ierr); 2377 } 2378 #if defined(PETSC_HAVE_THREADSAFETY) 2379 if (ctx->stage) { 2380 endtime = MPI_Wtime(); 2381 ctx->times[LANDAU_OPERATOR] += (endtime - starttime); 2382 ctx->times[LANDAU_JACOBIAN] += (endtime - starttime); 2383 ctx->times[LANDAU_JACOBIAN_COUNT] += 1; 2384 } 2385 #endif 2386 ierr = PetscLogEventEnd(ctx->events[0],0,0,0,0);CHKERRQ(ierr); 2387 ierr = PetscLogEventEnd(ctx->events[11],0,0,0,0);CHKERRQ(ierr); 2388 if (ctx->stage) { 2389 ierr = PetscLogStagePop();CHKERRQ(ierr); 2390 #if defined(PETSC_HAVE_THREADSAFETY) 2391 ctx->times[LANDAU_MATRIX_TOTAL] += (endtime - starttime); 2392 #endif 2393 } 2394 PetscFunctionReturn(0); 2395 } 2396 2397 /*@ 2398 LandauIJacobian - TS Jacobian construction 2399 2400 Collective on ts 2401 2402 Input Parameters: 2403 + TS - The time stepping context 2404 . time_dummy - current time (not used) 2405 - X - Current state 2406 + U_tdummy - Time derivative of current state (not used) 2407 . shift - shift for du/dt term 2408 - actx - Landau context 2409 2410 Output Parameter: 2411 . Amat - Jacobian 2412 + Pmat - same as Amat 2413 2414 Level: beginner 2415 2416 .keywords: mesh 2417 .seealso: LandauCreateVelocitySpace(), LandauIFunction() 2418 @*/ 2419 PetscErrorCode LandauIJacobian(TS ts, PetscReal time_dummy, Vec X, Vec U_tdummy, PetscReal shift, Mat Amat, Mat Pmat, void *actx) 2420 { 2421 PetscErrorCode ierr; 2422 LandauCtx *ctx=NULL; 2423 PetscInt dim; 2424 DM pack; 2425 #if defined(PETSC_HAVE_THREADSAFETY) 2426 double starttime, endtime; 2427 #endif 2428 PetscFunctionBegin; 2429 ierr = TSGetDM(ts,&pack);CHKERRQ(ierr); 2430 ierr = DMGetApplicationContext(pack, &ctx);CHKERRQ(ierr); 2431 if (!ctx) SETERRQ(PETSC_COMM_SELF, PETSC_ERR_PLIB, "no context"); 2432 if (Amat!=Pmat || Amat!=ctx->J) SETERRQ(ctx->comm, PETSC_ERR_PLIB, "Amat!=Pmat || Amat!=ctx->J"); 2433 ierr = DMGetDimension(pack, &dim);CHKERRQ(ierr); 2434 /* get collision Jacobian into A */ 2435 if (ctx->stage) { 2436 ierr = PetscLogStagePush(ctx->stage);CHKERRQ(ierr); 2437 } 2438 ierr = PetscLogEventBegin(ctx->events[11],0,0,0,0);CHKERRQ(ierr); 2439 ierr = PetscLogEventBegin(ctx->events[9],0,0,0,0);CHKERRQ(ierr); 2440 #if defined(PETSC_HAVE_THREADSAFETY) 2441 starttime = MPI_Wtime(); 2442 #endif 2443 ierr = PetscInfo2(ts, "Adding just mass to Jacobian t=%g, shift=%g\n",(double)time_dummy,(double)shift);CHKERRQ(ierr); 2444 if (shift==0.0) SETERRQ(ctx->comm, PETSC_ERR_PLIB, "zero shift"); 2445 if (!ctx->aux_bool) SETERRQ(ctx->comm, PETSC_ERR_PLIB, "wrong state"); 2446 if (!ctx->use_matrix_mass) { 2447 ierr = LandauFormJacobian_Internal(X,ctx->J,dim,shift,(void*)ctx);CHKERRQ(ierr); 2448 ierr = MatViewFromOptions(ctx->J, NULL, "-dm_landau_mat_view");CHKERRQ(ierr); 2449 } else { /* add mass */ 2450 ierr = MatAXPY(Pmat,shift,ctx->M,SAME_NONZERO_PATTERN);CHKERRQ(ierr); 2451 } 2452 ctx->aux_bool = PETSC_FALSE; 2453 #if defined(PETSC_HAVE_THREADSAFETY) 2454 if (ctx->stage) { 2455 endtime = MPI_Wtime(); 2456 ctx->times[LANDAU_OPERATOR] += (endtime - starttime); 2457 ctx->times[LANDAU_MASS] += (endtime - starttime); 2458 } 2459 #endif 2460 ierr = PetscLogEventEnd(ctx->events[9],0,0,0,0);CHKERRQ(ierr); 2461 ierr = PetscLogEventEnd(ctx->events[11],0,0,0,0);CHKERRQ(ierr); 2462 if (ctx->stage) { 2463 ierr = PetscLogStagePop();CHKERRQ(ierr); 2464 #if defined(PETSC_HAVE_THREADSAFETY) 2465 ctx->times[LANDAU_MATRIX_TOTAL] += (endtime - starttime); 2466 #endif 2467 } 2468 PetscFunctionReturn(0); 2469 } 2470