int main(int argc, char *argv[]) { GRID *g; MAP *map, *map0; INT n = 4; DOF *u; static INT pre_refines = 0; static char *fn = "../test/cube.dat"; /*static char *fn = "../albert-files/cube5.dat";*/ phgOptionsPreset("-dof_type P1"); phgOptionsRegisterInt("-pre_refines", "Pre-refinements", &pre_refines); phgInit(&argc, &argv); g = phgNewGrid(-1); if (!phgImport(g, fn, FALSE)) phgError(1, "can't read file \"%s\".\n", fn); phgRefineAllElements(g, pre_refines); phgPartitionGrid(g); u = phgDofNew(g, DOF_DEFAULT, 1, "u", DofNoAction); map0 = phgMapCreate(u, NULL); phgInfo(-1, "\n"); phgInfo(-1, "Test special map with size = 0 on all but one process:\n"); map = phgMapCreateSimpleMap(g->comm, g->rank == g->nprocs - 1 ? n : 0, n); phgInfo(-1, "---- DOF rows, special columns:\n"); do_test(map0, map); phgInfo(-1, "---- Special rows, DOF columns:\n"); do_test(map, map0); phgInfo(-1, "---- Special rows, special columns:\n"); do_test(map, map); phgMapDestroy(&map); phgInfo(-1, "\n"); phgInfo(-1, "Test serial map:\n"); map = phgMapCreateSimpleMap(g->comm, n, n); phgInfo(-1, "---- DOF rows, serial columns:\n"); do_test(map0, map); phgInfo(-1, "---- serial rows, DOF columns:\n"); do_test(map, map0); phgInfo(-1, "---- Serial rows, serial columns:\n"); do_test(map, map); phgMapDestroy(&map); phgMapDestroy(&map0); phgDofFree(&u); phgFreeGrid(&g); phgFinalize(); return 0; }
int main(int argc, char *argv[]) { GRID *g; SIMPLEX *e; DOF **u, **p, **T, **gradu, *eu, *ep, *egradu, *ediv, *eT, *dH = NULL; FLOAT Time, *dt, res, non_du, non_dp, non_dT; INT tstep = 0, nelem; char mesh_file[100], hostname[256], data_file[100], data_u[100], data_p[100], data_T[100], data_Crd[100]; size_t mem, mem_peak; int verb; double tt[3], tt1[3]; /* ---------- NS ---------- */ NSSolver *ns = NULL; SURF_BAS *surf_bas = NULL; LAYERED_MESH *gL = NULL; //GEO_INFO *geo = NULL; /* MG_BLOCK_DOFS *bk = NULL; */ /* ================================================================================ * * Initialize Grid & parameters * * ================================================================================ */ /* Global (static) options */ Unused(verb); ns_params = phgParametersCreate(); phgInit(&argc, &argv); phgOptionsShowUsed(); g = phgNewGrid(-1); //phgSetPeriodicity(g, ns_params->periodicity); phgImportSetBdryMapFunc(my_bc_map); if (ns_params->resume) { phgResumeStage(g, &Time, &tstep, mesh_file, data_file); phgPrintf("================================\n\n"); phgPrintf("* RESUME from time:%E, tstep:%d\n", Time, tstep); phgPrintf("* mesh:%s\n", mesh_file); phgPrintf("* data:%s\n", data_file); phgPrintf("================================\n"); if (!phgImport(g, mesh_file, FALSE)) phgError(1, "can't read file \"%s\".\n", ns_params->fn); } else { phgPrintf("Using mesh: %s\n", ns_params->fn); if (!phgImport(g, ns_params->fn, FALSE)) phgError(1, "can't read file \"%s\".\n", ns_params->fn); } checkBdry(g); elapsed_time(g, FALSE, 0.); /* reset timer */ gethostname(hostname, sizeof(hostname)); printf("#%5d# runing PID %5d on %s \n", phgRank, getpid(), hostname); NsSolver_Options(); phgPrintf(" Pre-refine & repartition "); phgRefineAllElements(g, ns_params->pre_refines); /* Set Reynolds number */ Time = ns_params->time_start; /* default: 0 */ setFuncTime(Time); setFlowParameter(ns_params->Re, ns_params->nu, Time); /* ================================================================================ * * build ice grid * * ================================================================================ */ iceInit(g, &gL); phgPrintf("Geometry initialization done!\n"); phgExportVTK(g, "ice_domain.vtk", NULL, NULL); /* ================================================================================ * * Create INS solver * * ================================================================================ */ /* Note: pointers u, p, gradu, dt * DIRECTLY access private member of INS solver. */ phgPrintf(" Create INS solver"); tstep = 1; /* time step start at 1 */ setFuncTime(Time); /* in file ins-bc.c: static */ //data_sur_T = read_txt_data("./LAS_temp_for_mesh.txt"); ns = phgNSCreate(g, ns_params); //get_mask_bot(ns); ns->time[1] = Time; u = ns->u; p = ns->p; T = ns->T; gradu = ns->gradu; dH = ns->dH; dt = ns->dt; /* direct accses ns */ dt[0] = ns_params->dt0; ns->gL = gL; //ns->bk = bk = NULL; //init_line_block(T[1], gL); /* Use line block */ elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); /* Init height & depth */ if (gL != NULL){ get_height_depth(ns); } /* surf bases */ //surf_bas = ns->surf_bas; #if 0 DOF *beta = phgDofNew(g, DOF_P1, 1, "beta", func_beta); phgExportEnsight(g, "check", beta, NULL); phgFinalize(); exit(1); #endif /* ------------------------------------------------------------ * * Resume dof data. * * ------------------------------------------------------------ */ if (ns_params->resume) { FILE *fp = NULL; char fname[1000]; phgResumeStage(g, &Time, &tstep, mesh_file, data_file); /* resume coord */ { const FLOAT *v = DofData(ns->coord); int i, k; //sprintf(data_Crd, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat.Crd", tstep - 1); sprintf(data_Crd, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat.Crd", 2); assert(ns->coord->type == DOF_P1); load_dof_data3(g, ns->coord, data_Crd, mesh_file); for (i = 0; i < g->nvert; i++) for (k = 0; k < Dim; k++) g->verts[i][k] = *(v++); phgGeomInit_(g, TRUE); } #if 1 /* resmue u_{n-1} */ //sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", tstep - 2); sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", 1); DATA_FILE_SURFIX; sprintf(fname, "%s.p%03d", data_u, g->rank); if ((fp = fopen(fname, "r")) == NULL) { phgPrintf("* u_{%d} unavailable.\n", tstep - 2); } else { fclose(fp); phgDofCopy(u[1], &u[0], NULL, "u_{n}"); phgDofCopy(p[1], &p[0], NULL, "p_{n}"); phgDofCopy(T[1], &T[0], NULL, "T_{n}"); gradu[0] = NULL; load_dof_data3(g, u[0], data_u, mesh_file); load_dof_data3(g, p[0], data_p, mesh_file); load_dof_data3(g, T[0], data_T, mesh_file); phgPrintf(" Resume u_ {%5d}[%8d]:%24.12E p_ {%5d}[%8d]:%24.12E\n", tstep - 2, DofGetDataCountGlobal(u[0]), phgDofNormL2(u[0]), tstep - 2, DofGetDataCountGlobal(p[0]), phgDofNormL2(p[0])); phgPrintf(" Resume T_{%5d}[%8d]:%24.12E\n", tstep - 2, DofGetDataCountGlobal(T[0]), phgDofNormL2(T[0])); phgDofGradient(u[0], &gradu[0], NULL, "gradu_{n}"); phgDofSetFunction(u[0], DofInterpolation); phgDofSetFunction(p[0], DofInterpolation); //phgDofSetBdryDataByFunction(u[0], func_u, SETFLOW); DOF_SCALE(u[0], "resume"); DOF_SCALE(p[0], "resume"); DOF_SCALE(T[0], "resume"); DOF_SCALE(gradu[0], "resume"); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); } /* resmue u_{n} */ //sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", tstep - 1); sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", 2); DATA_FILE_SURFIX; sprintf(fname, "%s.p%03d", data_u, g->rank); if ((fp = fopen(fname, "r")) == NULL) { phgError(1, "read Dof data %s failed!\n", data_file); } else { fclose(fp); load_dof_data3(g, u[1], data_u, mesh_file); load_dof_data3(g, p[1], data_p, mesh_file); load_dof_data3(g, T[1], data_T, mesh_file); phgPrintf(" Resume u_ {%5d}[%8d]:%24.12E p_ {%5d}[%8d]:%24.12E\n", tstep - 1, DofGetDataCountGlobal(u[1]), phgDofNormL2(u[1]), tstep - 1, DofGetDataCountGlobal(p[1]), phgDofNormL2(p[1])); phgPrintf(" Resume T_{%5d}[%8d]:%24.12E\n", tstep - 1, DofGetDataCountGlobal(T[1]), phgDofNormL2(T[1])); phgDofGradient(u[1], &gradu[1], NULL, "gradu_{n+1}"); phgDofSetFunction(u[1], DofInterpolation); phgDofSetFunction(p[1], DofInterpolation); //phgDofSetBdryDataByFunction(u[1], func_u, SETFLOW); DOF_SCALE(u[1], "resume"); DOF_SCALE(p[1], "resume"); DOF_SCALE(T[1], "resume"); DOF_SCALE(gradu[1], "resume"); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); } #endif /* Re init height & depth */ if (gL != NULL) { get_height_depth(ns); build_layered_mesh_height(g, gL); check_height(g, gL); } /* reconstruct last time step */ //Time -= dt[0]; Time = tstep - 1; ns->time[1] = Time; ns->time[0] = Time; setFuncTime(Time); #if 0 phgExportEnsightT(g, OUTPUT_DIR "/ins_" NS_PROBLEM , tstep, tstep, u[1], p[1], NULL); phgFinalize(); return 0; #endif /* debug exit */ } /* end of resume */ /* Init temp field */ DOF_SCALE(ns->beta, "test"); if (ns_params->solve_temp && tstep == 1) { phgPrintf("Init temp field!\n"); phgNSTempInit(ns); } /* ================================================================================ * * * Main loop: * 1. Steady state: adaptive refinement. * 2. Time dependent: time advance. * * ================================================================================ */ while (TRUE) { FLOAT all_memory_usage = phgMemoryUsage(g, NULL)/(1024.0*1024.0); if (all_memory_usage > 10000) break; get_surf_bot_elev(ns); DOF *grad_surf_elev = phgDofGradient(ns->surf_elev_P1, NULL, NULL, "gradu_surf_elev"); ns->grad_surf_elev = grad_surf_elev; //phgPrintf("modify mask bot!!\n"); //modify_mask_bot(ns); //modify_mask_bot(ns); //phgDofSetDataByValue(u[1], 0.); //phgDofSetDataByValue(p[1], 0.); ns->surf_bas = get_surface_bases(g, DOF_P2); surf_bas = ns->surf_bas; static BOOLEAN initialized = FALSE; FLOAT time_end = ns_params->time_end; elapsed_time(g, FALSE, 0.); /* reset timer */ phgGetTime(tt); if (Fabs(time_end - Time) < 1e-12) { phgPrintf("\n=======\nTime reach end: %lf, exit.\n", Time); phgPrintf("Time End %f\n", time_end); break; } if (tstep > ns_params->max_tstep) { phgPrintf("\n=======\nTime step reach end: %d, exit.\n", tstep); break; } #if 0 /* use time t^{n+1} */ dt[-1] = dt[0]; if (Time + dt[0] > time_end) dt[0] = time_end - Time; Time += dt[0]; setFuncTime(Time); #endif phgPrintf("\n==========\ntime: %lf, step:%d\n", (double)Time, tstep); phgPrintf(" %d DOF (u:%d, p:%d), %d elements, %d submesh%s, load imbalance: %lg\n", DofGetDataCountGlobal(u[1]) + DofGetDataCountGlobal(p[1]), DofGetDataCountGlobal(u[1]), DofGetDataCountGlobal(p[1]), g->nleaf_global, g->nprocs, g->nprocs > 1 ? "es" : "", (double)g->lif); /* save mesh */ if (ns_params->record && tstep % ns_params->step_span == 0) { phgResumeLogUpdate(g, &Time, &tstep, ns_params->fn, NULL); } if (!initialized) { /* reset mem_peak */ phgMemoryUsageReset(); initialized = TRUE; } /* ------------------------------------------------------------ * * Time marching * * ------------------------------------------------------------ */ /* update variale, * call this routine after time update and/or grid change.*/ phgNSTimeAdvance(ns, Time, tstep); phgPrintf(" update solution"); elapsed_time(g, TRUE, 0.); if (ns_params->pin_node) phgNSPinNode(ns); /* -------------------------------------------------------------------------------- * * Step 3. * * Momentum Equations. * * -------------------------------------------------------------------------------- */ int mask_iter = 0, inverse_iter = 0; INT IF_CHANGE_MASK; #if 1 while (TRUE) { // iteration for ice shelf mask updating and inversion if ((inverse_iter % 2) == 0) ns->set_dirichlet_bc = 0; /* newton bc for the surface, i.e. normal simulation */ else ns->set_dirichlet_bc = 1; /* constrain the surface with obs. vel. for inversion */ phgPrintf("----------------------------\n"); phgPrintf("ice shelf mask iteration: %d\n", mask_iter); phgPrintf("----------------------------\n"); elapsed_time(g, FALSE, 0.); /* reset timer */ /* * non-linear iteration. * */ int max_nonstep = 0, newton_start = 0; assert(ns_params->utype == DOF_P2); /* For nonlinear iter */ int nonstep = 0; non_du = non_dp = non_dT = 1e+10; DOF *u_last = phgDofCopy(u[1], NULL, NULL, "u_last"); DOF *p_last = phgDofCopy(p[1], NULL, NULL, "p_last"); FLOAT non_res_last = 1.; LTYPE ltype_last = PICARD; /* First step, change max non step. */ if (tstep == 1) { if (ns_params->max_nonstep0 > 0) max_nonstep = ns_params->max_nonstep0; else max_nonstep = ns_params->max_nonstep; if (ns_params->newton_start0 > 0) newton_start = ns_params->newton_start0; else newton_start = ns_params->newton_start; phgPrintf(" * Set max nonstep to %d for first step.\n", max_nonstep); phgPrintf(" * Set Newton start to %d for first step.\n", newton_start); } else { max_nonstep = ns_params->max_nonstep; newton_start = ns_params->newton_start; } while (TRUE) { phgPrintf("\n ==================\n"); phgPrintf(" Non-linear interation step: %d\n", nonstep); /* Init const viscosity */ if (ns_params->start_const_vis && tstep == 0 && nonstep == 0 && mask_iter == 0) { phgPrintf("* vis: const\n"); ns->viscosity_type = VIS_CONST; } else { phgPrintf("* vis: strain\n"); ns->viscosity_type = VIS_STRAIN; } sayHello("Non linear solve begin"); phgNSInitSolverU(ns); if (inverse_iter < 1) get_viscosity(ns); /* initiate viscosity field at the beginning * afterwards use the updated viscosity field */ if (nonstep < newton_start) ns->ltype = PICARD; else ns->ltype = NEWTON; phgPrintf(" Build RHS: "); phgNSBuildSolverURHS(ns, 1, nonstep, Time); //phgNSBuildSolverURHS(ns); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); ns->non_res = res = phgVecNorm2(ns->solver_u->rhs, 0, NULL); phgPrintf(" nonlinear residual: %24.12E\n", res); /* Restore Picard if no improvement */ #if 1 if (ltype_last == NEWTON && res > non_res_last * .75) { phgPrintf(" !!! Newton step failed, use Picard to run again\n"); ns->ltype = PICARD; max_nonstep += 5; /* Add more Picard steps */ /* resotre dofs: * Fix me: temprature */ phgDofCopy(u_last, &u[1], NULL, "u_{n+1}"); phgDofCopy(p_last, &p[1], NULL, "p_{n+1}"); phgDofGradient(u[1], &gradu[1], NULL, "gradu_{n+1}"); phgNSBuildSolverURHS(ns, 1, nonstep, Time); ns->non_res = res = phgVecNorm2(ns->solver_u->rhs, 0, NULL); phgPrintf(" nonlinear residual: %24.12E\n", res); } #endif /* save non res */ non_res_last = res; ltype_last = ns->ltype; /* build matrices */ //if (ns_params->use_PCD) //phgNSInitPc(ns); phgPrintf(" Build matrices:\n"); phgNSBuildSolverUMat(ns, 1, nonstep, Time); //phgNSBuildSolverUMat(ns); phgPrintf(" done "); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); #if 0 if (ns_params->use_PCD) { phgPrintf(" Build Pc: \n"); phgNSBuildPc(ns); phgPrintf(" done "); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); } #endif //elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); /* * solve equation and update (u, p) * */ phgPrintf("solver tol: %E\n", ns->solver_u->rtol); phgSolverSolve(ns->solver_u, TRUE, ns->du, ns->dp, NULL); // elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); //get_water_pressure(ns); #if USE_NODAL_LOADS get_nodal_force(ns); phgMatDestroy(&ns->matF0); phgMatDestroy(&ns->matB0); phgMatDestroy(&ns->matBt0); phgMatDestroy(&ns->matC0); //phgSolverDestroy(&ns->solver_u0); ns->solver_u0 = NULL; #endif //get_contact(ns); #if USE_SLIDING_BC rotate_dof_bases(ns->du, surf_bas, FALSE); #endif phgPrintf(" solver_u: nits = %d, resid = %0.4lg ", ns->solver_u->nits, ns->solver_u->residual); //elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); /* save dofs */ phgDofCopy(u[1], &u_last, NULL, "u_last"); phgDofCopy(p[1], &p_last, NULL, "p_last"); phgExportVTK(g, "u_test.vtk", ns->u[1], NULL); /* nonlinear correction */ phgDofAXPY(1.0, ns->du, &u[1]); phgDofAXPY(1.0, ns->dp, &p[1]); assert(u[1]->type == ns_params->utype); assert(p[1]->type == ns_params->ptype); #if USE_SLIDING_BC //dof_set_normal_data(u[1], surf_bas); #else #endif /* non_du = phgDofNormL2(ns->du); */ /* non_dp = phgDofNormL2(ns->dp); */ non_du = phgDofNormInftyVec(ns->du); non_dp = phgDofNormInftyVec(ns->dp); phgPrintf(" \n du: %24.12E dp: %24.12E\n", non_du, non_dp); phgPrintf(" u: [%24.12E, %24.12E]\n", phgDofMinValVec(u[1]), phgDofMaxValVec(u[1])); phgPrintf(" p: [%24.12E, %24.12E]\n", phgDofMinValVec(p[1]), phgDofMaxValVec(p[1])); phgDofGradient(u[1], &gradu[1], NULL, "gradu_{n+1}"); //get_avg_gu(ns); //gradu[1] = ns->avg_gu; //elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); //if (ns_params->use_PCD) //phgNSDestroyPc(ns); // /* evolution of u */ //DOF_SCALE(u[1], "after solve"); //DOF_SCALE(p[1], "after solve"); #if 0 # warning check solution U,p sprintf(vtk_file, OUTPUT_DIR "non_%02d_u.vtk", nonstep); phgExportVTK(g, vtk_file, u[1], p[1], NULL); phgExportEnsightT(g, OUTPUT_DIR "ins_" NS_PROBLEM , nonstep, nonstep, u[1], p[1], T[1], ns->du, ns->dp, ns->dT, NULL); #endif #if 0 if (FALSE && nonstep % ns_params->step_span == 0) { phgPrintf(" Output solution to ensight "); phgExportEnsightT(g, OUTPUT_DIR "/ins_" NS_PROBLEM , nonstep, nonstep, u[1], p[1], T[1], NULL); /* ensight */ sprintf(vtk_file, OUTPUT_DIR "non_%02d_T.vtk", nonstep); phgExportVTK(g, vtk_file , u[1], p[1], T[1], ns->du, NULL); elapsed_time(g, TRUE, 0.); //ice_monitor(ns, nonstep); } #endif /* Linearized */ #if 0 if (!ns_params->non_linear && nonstep >= 0) { phgPrintf(" Linearized iteration converges.\n"); break; } #endif phgGetTime(tt1); phgPrintf(" time usage of current non step: %lfs\n", (double)(tt1[2] - tt[2])); nonstep++; /* * Nonliner iteration break, * converge for characteristic value. * Velocity: 100 m/a * Pressure: 1e8 Pa * * */ INT u_convergence; if (mask_iter < 1) u_convergence = check_u_convergence0(ns, u[1], p[1], u_last, p_last, ns_params->u_tol0, ns_params->u_tol0); else u_convergence = check_u_convergence(ns, u[1], p[1], u_last, p_last, ns_params->u_tol, ns_params->u_tol); //elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); const FLOAT U0 = 100; const FLOAT P0 = 1e8; if ( nonstep >= ns_params->min_nonstep && ns->viscosity_type != VIS_CONST && u_convergence || nonstep > ns_params->max_nonstep) { if (nonstep > ns_params->max_nonstep) { phgPrintf(" Non-linear iteration reach max step," " results may be inaccrate!\n"); #if 1 && USE_NODAL_LOADS /* get_nodal_force(ns); phgMatDestroy(&ns->matF0); phgMatDestroy(&ns->matB0); phgMatDestroy(&ns->matBt0); phgMatDestroy(&ns->matC0); //phgSolverDestroy(&ns->solver_u0); ns->solver_u0 = NULL; */ phgMatDestroy(&ns->matF); phgMatDestroy(&ns->matB); phgMatDestroy(&ns->matBt); phgMatDestroy(&ns->matC); phgSolverDestroy(&ns->solver_u); ns->solver_u = NULL; phgMapDestroy(&ns->Vmap); phgMapDestroy(&ns->Pmap); phgDofFree(&ns->u_shape); phgDofFree(&ns->p_shape); #endif break; } else { phgPrintf(" Non-linear iteration converges.\n"); #if 1 && USE_NODAL_LOADS /* get_nodal_force(ns); phgMatDestroy(&ns->matF0); phgMatDestroy(&ns->matB0); phgMatDestroy(&ns->matBt0); phgMatDestroy(&ns->matC0); //phgSolverDestroy(&ns->solver_u0); ns->solver_u0 = NULL; */ phgMatDestroy(&ns->matF); phgMatDestroy(&ns->matB); phgMatDestroy(&ns->matBt); phgMatDestroy(&ns->matC); phgSolverDestroy(&ns->solver_u); ns->solver_u = NULL; phgMapDestroy(&ns->Vmap); phgMapDestroy(&ns->Pmap); phgDofFree(&ns->u_shape); phgDofFree(&ns->p_shape); #endif break; } } phgNSDestroySolverU(ns); } /* solve */ phgDofFree(&u_last); phgDofFree(&p_last); //phgPrintf("Save Dofs\n"); //save_dof_data3(g, u[1], OUTPUT_DIR"u.dat"); //save_dof_data3(g, p[1], OUTPUT_DIR"p.dat"); #elif 0 /* Set velocity, for debugging */ ns->viscosity_type = VIS_STRAIN; phgDofSetDataByFunction(u[1], func_u0); //phgDofSetDataByFunction(p[1], func_p0); phgDofGradient(u[1], &gradu[1], NULL, "gradu_{n+1}"); #else phgPrintf("Load Dofs\n"); load_dof_data3(g, u[1], OUTPUT_DIR"u.dat", NULL); load_dof_data3(g, p[1], OUTPUT_DIR"p.dat", NULL); #endif /* Project to continuous gradu */ #if 1 phgPrintf("\n ======================= \n"); phgPrintf("project velocity gradient"); phgPrintf("\n======================== \n"); proj_gradu(ns, ns->gradu[1]); #endif //DOF_SCALE(gradu[1], "grad u"); //DOF_SCALE(ns->Gradu, "Grad u"); phgPrintf("\n----------------------------\n"); phgPrintf("check ice shelf mask status"); phgPrintf("\n----------------------------\n"); get_mask_bot(ns); if (!(ns_params->another_run_with_updated_mask)) { mask_iter = 1; phgPrintf("manually stops another run with updated mask!\n\n\n"); IF_CHANGE_MASK = if_update_shelf_mask(ns); IF_CHANGE_MASK = 0; get_mask_bot(ns); if (tstep % ns_params->step_span == 0) { phgPrintf("Save water and nodal forces to VTK \n"); DOF *water_P1 = phgDofCopy(ns->water_force, NULL, DOF_P1, NULL); DOF *nodal_P1 = phgDofCopy(ns->nodal_force, NULL, DOF_P1, NULL); sprintf(vtk_file, MASK_OUTPUT_DIR "MASK_%05d.vtk", tstep); phgExportVTK(g, vtk_file, water_P1,nodal_P1,ns->contact_force,ns->mask_bot,NULL); phgDofFree(&water_P1);phgDofFree(&nodal_P1); } } if (mask_iter < 1) { IF_CHANGE_MASK = if_update_shelf_mask(ns); } if ((ns_params->another_run_with_updated_mask)){ if (IF_CHANGE_MASK == 0) { phgPrintf("The lower surface mask remains unchanged. Stop the iteration of mask updating !\n"); if (tstep % ns_params->step_span == 0) { get_mask_bot(ns); phgPrintf("Save water and nodal forces to VTK \n"); DOF *water_P1 = phgDofCopy(ns->water_force, NULL, DOF_P1, NULL); DOF *nodal_P1 = phgDofCopy(ns->nodal_force, NULL, DOF_P1, NULL); sprintf(vtk_file, MASK_OUTPUT_DIR "MASK_%05d.vtk", tstep); phgExportVTK(g, vtk_file, water_P1,nodal_P1,ns->contact_force,ns->mask_bot,NULL); phgDofFree(&water_P1);phgDofFree(&nodal_P1); } phgDofFree(&ns->nodal_force); phgDofFree(&ns->water_force); phgDofFree(&ns->contact_force); break; } } get_strain_rate(ns); if (ns->set_dirichlet_bc) { //DOF *u_d = phgDofCopy(ns->u[1], NULL, DOF_P2, NULL); DOF *eu_d = phgDofCopy(ns->strain_rate, NULL, DOF_P1, NULL); ns->eu_d = eu_d; } else { //DOF *u_n = phgDofCopy(ns->u[1], NULL, DOF_P2, NULL); DOF *eu_n = phgDofCopy(ns->strain_rate, NULL, DOF_P1, NULL); ns->eu_n = eu_n; } DOF *visc_old = phgDofCopy(ns->viscosity, NULL, DOF_P1, NULL); if (inverse_iter > 0) update_viscosity_inversion(ns); FLOAT tol = 0.01; INT visc_convergence = check_visc_convergence(ns, visc_old, tol); if((mask_iter >= 1) && (inverse_iter > 0) && (visc_convergence == 1)) { phgPrintf("\n-------------------------\n"); phgPrintf("ice shelf mask updated \n"); phgPrintf("--------------------------\n"); break; } if (mask_iter < 1) if (tstep % ns_params->step_span == 0) { phgPrintf("Save water and nodal forces to VTK \n"); DOF *water_P1 = phgDofCopy(ns->water_force, NULL, DOF_P1, NULL); DOF *nodal_P1 = phgDofCopy(ns->nodal_force, NULL, DOF_P1, NULL); sprintf(vtk_file, MASK_OUTPUT_DIR "MASK_%05d.vtk", tstep); phgExportVTK(g, vtk_file, water_P1,nodal_P1,ns->contact_force,ns->mask_bot,NULL); phgDofFree(&water_P1);phgDofFree(&nodal_P1); } phgDofFree(&ns->nodal_force); phgDofFree(&ns->water_force); phgDofFree(&ns->contact_force); //phgDofFree(&ns->mask_bot); //phgDofFree(&ns->water_pressure); //phgDofFree(&ns->stress_nn); // phgPrintf("Free stress_nn !\n"); //phgDofFree(&ns->stress); //phgDofFree(&ns->water_pressure1); //phgDofFree(&ns->stress_nn1); //phgDofFree(&ns->stress1); //phgDofFree(&ns->avg_gu); mask_iter++; inverse_iter++; } proj_gradu(ns, ns->gradu[1]); get_stress(ns, ns->gradu[1], ns->p[1]); #if 0 if (1) { int i, k; DOF *Gu[DDim], *gu[DDim], *guDG0, *stress[DDim]; guDG0 = phgDofCopy(ns->gradu[1], NULL, DOF_P0, NULL); for (k = 0; k < DDim; k++) { FLOAT *vGu; INT n; char name[1000]; sprintf(name, "Gu%d", k); Gu[k] = phgDofNew(g, DOF_P1, 1, name, DofNoAction); vGu = ns->Gradu->data; /* DOF_P1 */ n = DofGetDataCount(Gu[k]); for (i = 0; i < n; i++) Gu[k]->data[i] = vGu[i * DDim + k]; sprintf(name, "stress%d", k); stress[k] = phgDofNew(g, DOF_P1, 1, name, DofNoAction); vGu = ns->stress->data; /* DOF_P1 */ n = DofGetDataCount(stress[k]); for (i = 0; i < n; i++) stress[k]->data[i] = vGu[i * DDim + k]; sprintf(name, "gu%d", k); gu[k] = phgDofNew(g, DOF_P0, 1, name, DofNoAction); vGu = guDG0->data; /* DOF_P0 */ n = DofGetDataCount(gu[k]); for (i = 0; i < n; i++) gu[k]->data[i] = vGu[i * DDim + k]; } phgExportVTK(g, "stress.vtk", stress[0], stress[1], stress[2], stress[3], stress[4], stress[5], stress[6], stress[7], stress[8], NULL); phgExportVTK(g, "Gu.vtk", Gu[0], Gu[1], Gu[2], Gu[3], Gu[4], Gu[5], Gu[6], Gu[7], Gu[8], gu[0], gu[1], gu[2], gu[3], gu[4], gu[5], gu[6], gu[7], gu[8], NULL); // phgFinalize(); } #endif /* -------------------------------------------------------------------------------- * * Step 4. * * Solve temperature. * * -------------------------------------------------------------------------------- */ #if 1 if (ns_params->solve_temp) { phgPrintf("\n ==================\n"); phgPrintf(" Temperature solve \n"); phgPrintf(" ==================\n\n"); phgPrintf(" T type: %s\n", T[1]->type->name); elapsed_time(g, FALSE, 0.); /* reset timer */ phgNSInitSolverT(ns); phgPrintf(" Build Mat: "); phgNSBuildSolverTMat(ns, FALSE); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); phgPrintf(" Build RHS: "); phgNSBuildSolverTRHS(ns, FALSE); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); phgNSSolverTBuildConstrain(ns); phgDofCopy(ns->T[1], &ns->dT, NULL, "dT"); phgNSSolverTSolve(ns, FALSE); elapsed_time(g, TRUE, phgPerfGetMflops(g, NULL, NULL)); phgNSDestroySolverT(ns); //find_melt_region(ns); DOF_SCALE(ns->T[1], "after solve"); phgDofAXPY(-1.0, ns->T[1], &ns->dT); non_dT = phgDofNormInftyVec(ns->dT); phgPrintf(" dT: %24.12E\n", non_dT); DOF *temp_diff = phgDofCopy(ns->T[1], NULL, NULL, "Td"); { FLOAT *vt = temp_diff->data; const FLOAT *vh = ns->depth_P2->data; INT i, n = DofGetDataCount(temp_diff); for (i = 0; i < n; i++, vh++, vt++) *vt = TEMP_WATER - BETA_MELT * (*vh) *LEN_SCALING - (*vt); } } else { phgPrintf("Temp not updated.\n"); non_dT = 0.; } #endif /* ------------------------------------------------------------ * * Error check * * ------------------------------------------------------------ */ #if 0 if (!ns_params->compute_error) { eu = ep = egradu = eT = NULL; ediv = phgDofDivergence(u[1], NULL, NULL, "err div u"); phgPrintf( " normL2(u, p) = (%20.12E, %20.12E)\n" " normH1(u) = (%20.12E)\n" " normDiv(u) = (%20.12E)\n" " normGadu = (%20.12E)\n", dofNormL2(u[1]), dofNormL2(p[1]), dofNormL2(gradu[1]), dofNormL2(ediv), dofNormL2(ns->gradu[1])); elapsed_time(g, TRUE, 0.); } else { /* ----Error check------- */ phgPrintf(" Errors: \n"); eu = phgDofCopy(u[1], NULL, ns_params->utype, "erru"); ep = phgDofCopy(p[1], NULL, ns_params->ptype, "errp"); eT = phgDofCopy(T[1], NULL, NULL, "errT"); egradu = phgDofCopy(gradu[1], NULL, NULL, "err grad u"); ediv = phgDofDivergence(u[1], NULL, NULL, "err div u"); adjust_time(- (1. - ns_params->Theta) * dt[0]); restore_time(); phgPrintf(" errL2(u, p) = (%20.12E, %20.12E)\n" " errH1(u) = (%20.12E)\n" " errDiv(u) = (%20.12E)\n", dofNormL2(eu), dofNormL2(ep), dofNormL2(egradu), dofNormL2(ediv)); elapsed_time(g, TRUE, 0.); phgDofFree(&egradu); dof_norm_L2(eT); } #endif getPecletNum(g, u[1], ns_params->nu, 6); mem = phgMemoryUsage(g, &mem_peak); phgPrintf(" Memory usage: current %0.4lgMB, peak %0.4lgMB\n", (double)mem / (1024.0 * 1024.0), (double)mem_peak / (1024.0 * 1024.0)); /* ------------------------------------------------------------ * * Move mesh * * ------------------------------------------------------------ */ FLOAT ice_volume_last = get_ice_volume(g); // DOF *dH_last = phgDofCopy(ns->dH, NULL, NULL, "dH_last"); if (ns_params->solve_height) { if (gL != NULL) { /* Unstructed layered mesh */ phgPrintf("Move mesh.\n"); get_surf_dH(ns); //DOF *dH_fem = phgDofCopy(ns->dH, NULL, NULL, NULL); phgExportVTK(g, "dH_fem1.vtk", ns->dH, NULL); get_smooth_surface_values(ns, ns->dH, 0); get_smooth_surface_values(ns, ns->dH, 1); phgExportVTK(g, "dH_fem.vtk", ns->dH, NULL); /* save_free_surface_elev(ns, 0); save_free_surface_elev(ns, 1); save_free_surface_velo(ns, 0, 0); save_free_surface_velo(ns, 1, 0); save_free_surface_velo(ns, 2, 0); save_free_surface_velo(ns, 0, 1); save_free_surface_velo(ns, 1, 1); save_free_surface_velo(ns, 2, 1); if (phgRank == 0) { system("python get_upper_ds.py"); system("python get_lower_ds.py"); } load_dH_from_file(ns, ns->dH, 0); load_dH_from_file(ns, ns->dH, 1); //DOF *dH_fdm = phgDofCopy(ns->dH, NULL, NULL, NULL); //phgDofAXPY(-1, dH_fem, &dH_fdm); //phgExportVTK(g, "dH_diff.vtk", dH_fdm, NULL); *///phgExportVTK(g, "dH_fdm.vtk", ns->dH, NULL); get_moved_coord(ns, tstep); phgExportVTK(g, "dH_fem2.vtk", NULL); move_mesh(ns); //phgExportVTK(g, "moved_geo.vtk", NULL); //check_height(g, gL); } else { /* Structed layered mesh */ struct_mesh_update(ns, tstep, Time); } phgDofGradient(u[1], &gradu[1], NULL, "gradu_{n+1}"); phgDofGradient(u[0], &gradu[0], NULL, "gradu_{n}"); } else { phgPrintf("Mesh not moved.\n"); } phgPrintf("\n-----------------------------------------\n"); phgPrintf(" dH: [%4.2E, %4.2E]\n", phgDofMinValVec(ns->dH), phgDofMaxValVec(ns->dH)); phgPrintf("------------------------------------------\n\n"); FLOAT ice_volume = get_ice_volume(g); //DOF *u_P1 = phgDofCopy(u[1], NULL, DOF_P1, NULL); //phgExportVTK(g, "u_P1.vtk", u_P1, NULL); #if 1 //INT dH_convergence = check_surf_convergence(ns, dH_last); //if (dH_convergence == 1) FLOAT dVdt = fabs(ice_volume-ice_volume_last)/ice_volume/dt[0]; if (Time > 10 && dVdt < ns_params->s_tol) { phgPrintf("-----------------------------------------------------\n"); phgPrintf("The ice domain reaches a steady state! Model stops! dVdt %e", dVdt); phgPrintf("\n-----------------------------------------------------\n"); break; } else { phgPrintf("\n--------------------------------------------------------------\n"); phgPrintf("The ice domain is still in an unsteady state! Model continues! dVdt %e\n", dVdt); phgPrintf("---------------------------------------------------------------\n\n"); } #endif update_grounding_line(ns, tstep); // after we update the geometry, we need to check the ice shelf mask again /* ------------------------------------------------------------ * * Output solution * * ------------------------------------------------------------ */ if (tstep % ns_params->step_span == 0) { //ice_monitor(ns, tstep); #if 1 /* Temp check */ phgPrintf(" Output solution to VTK"); /* phgExportEnsightT(g, OUTPUT_DIR "/ins_" NS_PROBLEM , Time, tstep, */ /* u[1], p[1], T[1], ns->depth_P1, */ /* NULL); /\* ensight *\/ */ DOF *u_P1 = phgDofCopy(u[1], NULL, DOF_P1, NULL); //phgExportVTK(g,"results.vtk",u_P1,p[1],NULL); sprintf(vtk_file, dH_OUTPUT_DIR "ice_%05d.vtk", tstep); phgExportVTK(g, vtk_file, u_P1, NULL); phgDofFree(&u_P1); sprintf(vtk_file, dH_OUTPUT_DIR "dH_%05d.vtk", tstep); phgExportVTK(g, vtk_file, ns->dH, NULL); sprintf(vtk_file, dH_OUTPUT_DIR "stress_%05d.vtk", tstep); phgExportVTK(g, vtk_file, ns->stress, NULL); //sprintf(vtk_file, OUTPUT_DIR "mask_%05d.vtk", tstep); //phgExportVTK(g, vtk_file, ns->mask_bot, NULL); elapsed_time(g, TRUE, 0.); #endif /* Save coord data */ #if 1 if (tstep % ns_params->step_span_resume == 0) { { //sprintf(data_Crd, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat.Crd", tstep); sprintf(data_Crd, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat.Crd", 2); assert(ns->coord->type == DOF_P1); save_dof_data3(g, ns->coord, data_Crd); } if (ns_params->record) { /* save dof data for time step {n}, {n+1} */ //sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", tstep - 1); sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", 1); DATA_FILE_SURFIX; save_dof_data3(g, u[0], data_u); save_dof_data3(g, p[0], data_p); save_dof_data3(g, T[0], data_T); phgPrintf(" Save u_ {%5d}[%8d]:%24.12E p_ {%5d}[%8d]:%24.12E\n", tstep - 1, DofGetDataCountGlobal(u[0]), phgDofNormL2(u[0]), tstep - 1, DofGetDataCountGlobal(p[0]), phgDofNormL2(p[0])); phgPrintf(" Save T_{%5d}[%8d]:%24.12E\n", tstep - 1, DofGetDataCountGlobal(T[0]), phgDofNormL2(T[0])); DOF_SCALE(u[0], "save"); DOF_SCALE(p[0], "save"); DOF_SCALE(T[0], "save"); DOF_SCALE(gradu[0], "save"); //sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", tstep); sprintf(data_file, OUTPUT_DIR "/ins_" NS_PROBLEM "_%05d.dat", 2); DATA_FILE_SURFIX; save_dof_data3(g, u[1], data_u); save_dof_data3(g, p[1], data_p); save_dof_data3(g, T[1], data_T); phgPrintf(" Save u_ {%5d}[%8d]:%24.12E p_ {%5d}[%8d]:%24.12E\n", tstep, DofGetDataCountGlobal(u[1]), phgDofNormL2(u[1]), tstep, DofGetDataCountGlobal(p[1]), phgDofNormL2(p[1])); phgPrintf(" Save T_{%5d}[%8d]:%24.12E\n", tstep, DofGetDataCountGlobal(T[1]), phgDofNormL2(T[1])); DOF_SCALE(u[1], "save"); DOF_SCALE(p[1], "save"); DOF_SCALE(T[1], "save"); DOF_SCALE(gradu[1], "save"); phgResumeLogUpdate(g, NULL, NULL, NULL, data_file); } /* end of record */ if (gL != NULL) { //check_height(g, gL); } sayHello("After record final solution data"); } #endif } //phgDofFree(&dH_last); /* clean up */ //phgDofFree(&ediv); //phgDofFree(&eu); //phgDofFree(&ep); //phgDofFree(&eT); /* ---------------------------------------------------------------------- * * Compute drag force FD * * ---------------------------------------------------------------------- * */ phgGetTime(tt1); phgPrintf(" total time usage of current time step: %lfs\n", (double)(tt1[2] - tt[2])); if (mem_peak > 1024 * (size_t)ns_params->mem_max * 1024) { phgPrintf("\n=======\nMem usage reach max, exit.\n"); break; } tstep++; #if 1 /* use time t^{n} */ dt[-1] = dt[0]; if (Time + dt[0] > time_end) dt[0] = time_end - Time; Time += dt[0]; setFuncTime(Time); #endif phgDofFree(&surf_bas->dof); phgDofFree(&ns->surf_bas->dof); } /* end of time advaning */ /* destroy line block */ //destroy_line_block(&ns->bk); /* destroy reused solver */ if (ns->solver_u != NULL) { if (ns_params->use_PCD) phgNSDestroyPc(ns); phgNSDestroySolverU(ns); } phgNSFinalize(&ns); phgFreeGrid(&g); phgFinalize(); phgFree(ns_params); return 0; }
int main(int argc, char *argv[]) { MAT *A, *B; VEC *U = NULL, *x; FLOAT *C; SOLVER *solver, *solver1 = NULL; INT i, j, k, n, *pvt, N = 1000, K = 2; char *main_opts = NULL, *sub_opts = NULL; phgOptionsRegisterInt("-n", "N value", &N); phgOptionsRegisterInt("-k", "K value", &K); phgOptionsRegisterString("-main_solver_opts", "Options for the main solver", &main_opts); phgOptionsRegisterString("-sub_solver_opts", "Options for the subsolver", &sub_opts); /* a direct solver is preferable for the sparse matrix */ phgOptionsPreset("-solver mumps"); phgInit(&argc, &argv); phgPrintf( "----------------------------------------------------------------------------\n" "This code solves (A+UU^t)x=b using the Sherman-Morrison-Woodbury formula.\n" "Note: may use the following to disable use of the Sherman-Morrison-Woodbury\n" "algorithm and change to the default solver instead:\n" " -preonly_pc_type solver -preonly_pc_opts \"-solver_maxit 2000\"\n" "----------------------------------------------------------------------------\n" ); phgPrintf("Generating the linear system: N = %"dFMT", K = %"dFMT"\n", N, K); /* A is a distributed NxN SPD tridiagonal matrix (A = [-1, 2, -1]) */ n = N / phgNProcs + (phgRank < (N % phgNProcs) ? 1 : 0); A = phgMatCreate(phgComm, n, N); phgPrintf(" Generating matrix A.\n"); for (i = 0; i < n; i++) { /* diagonal */ phgMatAddEntry(A, i, i, 2.0); /* diagonal - 1 */ if (i > 0) phgMatAddEntry(A, i, i - 1, -1.0); else if (phgRank > 0) phgMatAddLGEntry(A, i, A->rmap->partition[phgRank] - 1, -1.0); /* diagonal + 1 */ if (i < n - 1) phgMatAddEntry(A, i, i + 1, -1.0); else if (phgRank < phgNProcs - 1) phgMatAddLGEntry(A, i, A->rmap->partition[phgRank] + n, -1.0); } phgMatAssemble(A); /* U is a K-component vector */ U = phgMapCreateVec(A->rmap, K); phgVecRandomize(U, 123); /* solver1 is the solver for A */ phgOptionsPush(); phgOptionsSetOptions(sub_opts); solver1 = phgMat2Solver(SOLVER_DEFAULT, A); phgOptionsPop(); /* x is a scratch vector */ x = phgMapCreateVec(A->rmap, 1); /* C is a KxK dense matrix, pvt is an integer array, they store the LU * factorization of (I + U^t*inv(A)*U) */ phgPrintf(" Generating the dense matrix I+U^t*inv(A)*U.\n"); C = phgCalloc(K * K, sizeof(*C)); pvt = phgAlloc(K * sizeof(*pvt)); for (i = 0; i < K; i++) { for (j = 0; j < n; j++) { solver1->rhs->data[j] = U->data[i * n + j]; x->data[j] = 0.0; } solver1->rhs->assembled = TRUE; phgSolverVecSolve(solver1, FALSE, x); for (j = 0; j < K; j++) for (k = 0; k < n; k++) C[i * K + j] += U->data[j * n + k] * x->data[k]; } #if USE_MPI if (U->map->nprocs > 1) { FLOAT *tmp = phgAlloc(K * K * sizeof(*tmp)); MPI_Allreduce(C, tmp, K * K, PHG_MPI_FLOAT, MPI_SUM, U->map->comm); phgFree(C); C = tmp; } #endif /* USE_MPI */ for (i = 0; i < K; i++) C[i * K + i] += 1.0; phgPrintf(" Factorizing the dense matrix I+U^t*inv(A)*U.\n"); phgSolverDenseLU(K, C, pvt); /* B is a matrix-free matrix representing A + U*U^t, B->mv_data is used * to pass A, U, solver1, C and pvt to callback functions */ B = phgMapCreateMatrixFreeMat(A->rmap, A->cmap, funcB, /* arguments carried over to CB functions */ A, U, solver1, C, pvt, NULL); /* solver is a PreOnly solver for B whose pc_proc is set to sherman(). * * Note: can also use pcg, gmres, or petsc for this solver, in this case * the solution obtained with the Sherman-Morisson formula is iteratively * refined. */ phgOptionsPush(); phgOptionsSetOptions("-solver preonly"); phgOptionsSetOptions(main_opts); solver = phgMat2Solver(SOLVER_DEFAULT, B); phgSolverSetPC(solver, solver, sherman); phgOptionsPop(); for (i = 0; i < n; i++) x->data[i] = 1.0; phgMatVec(MAT_OP_N, 1.0, B, x, 0.0, &solver->rhs); phgPrintf("Solving the linear system.\n"); /* reset initial solution to zero */ memset(x->data, 0, n * sizeof(*x->data)); phgSolverVecSolve(solver, TRUE, x); for (i = 0; i < n; i++) solver->rhs->data[i] = 1.0; phgVecAXPBY(-1.0, solver->rhs, 1.0, &x); phgPrintf("Checking the result: |x - x_exact| / |x_exact| = %lg\n", (double)phgVecNorm2(x, 0, NULL) / sqrt((double)N)); phgSolverDestroy(&solver); phgSolverDestroy(&solver1); phgMatDestroy(&A); phgMatDestroy(&B); phgVecDestroy(&U); phgVecDestroy(&x); phgFree(C); phgFree(pvt); phgFinalize(); return 0; }
int main(int argc, char *argv[]) { ELEMENT *e; GRID *g; DOF *u, *v, *u_hp, *v_hp; HP_TYPE *hp; MAP *map; char *fn = "cube.dat"; char *dof_u = "P2", *dof_v = "P1"; INT step = 0, pre_refines = 0; phgOptionsRegisterFilename("-mesh_file", "Mesh file", &fn); phgOptionsRegisterInt("-pre_refines", "Pre-refinements", &pre_refines); phgOptionsRegisterString("-dof_u", "DOF type for u", &dof_u); phgOptionsRegisterString("-dof_v", "DOF type for v", &dof_v); phgInit(&argc, &argv); g = phgNewGrid(-1); if (!phgImport(g, fn, FALSE)) phgError(1, "can't read file \"%s\".\n", fn); phgRefineAllElements(g, pre_refines); phgOptionsSetHandler("-dof_type", dof_u); u = phgDofNew(g, DOF_DEFAULT, 1, "u", DofInterpolation); phgOptionsSetHandler("-dof_type", dof_v); v = phgDofNew(g, DOF_DEFAULT, 1, "v", DofInterpolation); phgPrintf("u->type = %s, v->type = %s\n", u->type->name, v->type->name); hp = phgHPNew(g, HP_HB); u_hp = phgHPDofNew(g, hp, 1, "u_hp", DofInterpolation); phgHPFree(&hp); hp = phgHPNew(g, HP_HC); v_hp = phgHPDofNew(g, hp, 1, "v_hp", DofInterpolation); phgHPFree(&hp); while (TRUE) { if (phgBalanceGrid(g, 1.1, 1, NULL, 0.)) phgPrintf("Repartition mesh, %d submeshes, load imbalance: %lg\n", g->nprocs, (double)g->lif); phgPrintf("Testing map with non HP DOFs:\n"); map = phgMapCreate(u, v, NULL); phgPrintf(" nlocal = %d, nglobal = %d\n", map->nlocal, map->nglobal); phgMapDestroy(&map); phgPrintf("Testing map with HP DOFs:\n"); ForAllElements(g, e) e->hp_order = 1 + GlobalElement(g, e->index) % 4; phgHPSetup(u_hp->hp, FALSE); ForAllElements(g, e) e->hp_order = 1 + (3 - GlobalElement(g, e->index) % 4); phgHPSetup(v_hp->hp, FALSE); map = phgMapCreate(u_hp, v_hp, NULL); phgPrintf(" nlocal = %d, nglobal = %d\n", map->nlocal, map->nglobal); phgMapDestroy(&map); phgPrintf("Testing map with HP and non HP DOFs:\n"); map = phgMapCreate(u, u_hp, v, v_hp, NULL); phgPrintf(" nlocal = %d, nglobal = %d\n", map->nlocal, map->nglobal); phgMapDestroy(&map); if (++step >= 1) break; phgRefineAllElements(g, 1); } phgDofFree(&u_hp); phgDofFree(&v_hp); phgDofFree(&u); phgDofFree(&v); phgFreeGrid(&g); phgFinalize(); return 0; }
int main(int argc, char *argv[]) { GRID *g; DOF *u_h; MAT *A, *A0, *B; MAP *map; INT i; size_t nnz, mem, mem_peak; VEC *x, *y0, *y1, *y2; double t0, t1, dnz, dnz1, mflops, mop; char *fn = "../test/cube.dat"; FLOAT mem_max = 300; INT refine = 0; phgOptionsRegisterFilename("-mesh_file", "Mesh file", (char **)&fn); phgOptionsRegisterInt("-loop_count", "Loop count", &loop_count); phgOptionsRegisterInt("-refine", "Refinement level", &refine); phgOptionsRegisterFloat("-mem_max", "Maximum memory", &mem_max); phgInit(&argc, &argv); g = phgNewGrid(-1); if (!phgImport(g, fn, FALSE)) phgError(1, "can't read file \"%s\".\n", fn); phgRefineAllElements(g, refine); u_h = phgDofNew(g, DOF_DEFAULT, 1, "u_h", DofNoAction); while (TRUE) { phgPrintf("\n"); if (phgBalanceGrid(g, 1.2, 1, NULL, 0.)) phgPrintf("Repartition mesh, %d submeshes, load imbalance: %lg\n", g->nprocs, (double)g->lif); map = phgMapCreate(u_h, NULL); A = phgMapCreateMat(map, map); A->handle_bdry_eqns = TRUE; build_matrix(A, u_h); phgMatAssemble(A); /* Note: A is unsymmetric (A' != A) if boundary entries not removed */ phgMatRemoveBoundaryEntries(A); #if 0 /* test block matrix operation */ A0 = phgMatCreateBlockMatrix(g->comm, 1, 1, &A, NULL); #else A0 = A; #endif phgPrintf("%d DOF, %d elems, %d submeshes, matrix size: %d, LIF: %lg\n", DofGetDataCountGlobal(u_h), g->nleaf_global, g->nprocs, A->rmap->nglobal, (double)g->lif); /* test PHG mat-vec multiply */ x = phgMapCreateVec(A->cmap, 1); y1 = phgMapCreateVec(A->rmap, 1); phgVecRandomize(x, 123); phgMatVec(MAT_OP_N, 1.0, A0, x, 0.0, &y1); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); for (i = 0; i < loop_count; i++) { phgMatVec(MAT_OP_N, 1.0, A0, x, 0.0, &y1); } t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); y0 = phgVecCopy(y1, NULL); nnz = A->nnz_d + A->nnz_o; #if USE_MPI dnz1 = nnz; MPI_Reduce(&dnz1, &dnz, 1, MPI_DOUBLE, MPI_SUM, 0, g->comm); #else dnz = nnz; #endif mop = loop_count * (dnz + dnz - A->rmap->nlocal) * 1e-6; phgPrintf("\n"); t1 -= t0; phgPrintf(" PHG: time %0.4lf, nnz %0.16lg, %0.2lfMF (%0.2lfMF)\n", t1, dnz, mop / (t1 == 0 ? 1. : t1), mflops); /* test trans(A)*x */ phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); for (i = 0; i < loop_count; i++) { phgMatVec(MAT_OP_T, 1.0, A0, x, 0.0, &y1); } t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); t1 -= t0; phgPrintf(" A'*x: time %0.4lf, nnz %0.16lg, %0.2lfMF (%0.2lfMF), " "err: %le\n", t1, dnz, mop / (t1 == 0 ? 1. : t1), mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y0, 1.0, &y1), 0, NULL)); /* time A * trans(A) */ phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); B = phgMatMat(MAT_OP_N, MAT_OP_N, 1.0, A, A, 0.0, NULL); t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); nnz = B->nnz_d + B->nnz_o; #if USE_MPI dnz1 = nnz; MPI_Reduce(&dnz1, &dnz, 1, MPI_DOUBLE, MPI_SUM, 0, g->comm); #else dnz = nnz; #endif /* compare B*x <--> A*A*x */ y2 = phgMatVec(MAT_OP_N, 1.0, B, x, 0.0, NULL); phgMatVec(MAT_OP_N, 1.0, A0, y0, 0.0, &y1); phgMatDestroy(&B); t1 -= t0; phgPrintf(" A*A: time %0.4lf, nnz %0.16lg, %0.2lfMF, err: %le\n", t1, dnz, mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y1, 1.0, &y2), 0, NULL)); #if USE_PETSC { Mat ma, mb; MatInfo info; Vec va, vb, vc; PetscScalar *vec; ma = phgPetscCreateMatAIJ(A); MatGetVecs(ma, PETSC_NULL, &va); VecDuplicate(va, &vb); VecGetArray(va, &vec); memcpy(vec, x->data, x->map->nlocal * sizeof(*vec)); VecRestoreArray(va, &vec); MatMult(ma, va, vb); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); for (i = 0; i < loop_count; i++) { MatMult(ma, va, vb); } t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); VecGetArray(vb, &vec); memcpy(y1->data, vec, x->map->nlocal * sizeof(*vec)); VecRestoreArray(vb, &vec); MatGetInfo(ma, MAT_GLOBAL_SUM, &info); /*phgPrintf(" --------------------------------------------" "-------------------------\n");*/ phgPrintf("\n"); t1 -= t0; dnz = info.nz_used; phgPrintf(" PETSc: time %0.4lf, nnz %0.16lg, %0.2lfMF (%0.2lfMF), " "err: %le\n", t1, dnz, mop / (t1==0 ? 1.:t1), mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y0, 1.0, &y1), 0, NULL)); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); for (i = 0; i < loop_count; i++) { MatMultTranspose(ma, va, vb); } t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); VecGetArray(vb, &vec); memcpy(y1->data, vec, x->map->nlocal * sizeof(*vec)); VecRestoreArray(vb, &vec); t1 -= t0; phgPrintf(" A'*x: time %0.4lf, nnz %0.16lg, %0.2lfMF (%0.2lfMF), " "err: %le\n", t1, dnz, mop / (t1==0 ? 1.:t1), mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y0, 1.0, &y1), 0, NULL)); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); MatMatMult(ma, ma, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &mb); t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); t1 -= t0; MatGetInfo(mb, MAT_GLOBAL_SUM, &info); dnz = info.nz_used; VecDuplicate(va, &vc); /* compare B*x <--> A*A*x */ MatMult(ma, vb, vc); MatMult(mb, va, vb); VecGetArray(vb, &vec); memcpy(y1->data, vec, x->map->nlocal * sizeof(*vec)); VecRestoreArray(vb, &vec); VecGetArray(vc, &vec); memcpy(y2->data, vec, x->map->nlocal * sizeof(*vec)); VecRestoreArray(vc, &vec); phgPrintf(" A*A: time %0.4lf, nnz %0.16lg, %0.2lfMF, err: %le\n", t1, dnz, mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y1, 1.0, &y2), 0, NULL)); phgPetscMatDestroy(&mb); phgPetscMatDestroy(&ma); phgPetscVecDestroy(&va); phgPetscVecDestroy(&vb); phgPetscVecDestroy(&vc); } #endif /* USE_PETSC */ #if USE_HYPRE { HYPRE_IJMatrix ma; HYPRE_IJVector va, vb, vc; HYPRE_ParCSRMatrix par_ma; hypre_ParCSRMatrix *par_mb; HYPRE_ParVector par_va, par_vb, par_vc; HYPRE_Int offset, *ni, start, end; assert(sizeof(INT)==sizeof(int) && sizeof(FLOAT)==sizeof(double)); setup_hypre_mat(A, &ma); ni = phgAlloc(2 * A->rmap->nlocal * sizeof(*ni)); offset = A->cmap->partition[A->cmap->rank]; for (i = 0; i < A->rmap->nlocal; i++) ni[i] = i + offset; HYPRE_IJVectorCreate(g->comm, offset, offset + A->rmap->nlocal - 1, &va); HYPRE_IJVectorCreate(g->comm, offset, offset + A->rmap->nlocal - 1, &vb); HYPRE_IJVectorCreate(g->comm, offset, offset + A->rmap->nlocal - 1, &vc); HYPRE_IJVectorSetObjectType(va, HYPRE_PARCSR); HYPRE_IJVectorSetObjectType(vb, HYPRE_PARCSR); HYPRE_IJVectorSetObjectType(vc, HYPRE_PARCSR); HYPRE_IJVectorSetMaxOffProcElmts(va, 0); HYPRE_IJVectorSetMaxOffProcElmts(vb, 0); HYPRE_IJVectorSetMaxOffProcElmts(vc, 0); HYPRE_IJVectorInitialize(va); HYPRE_IJVectorInitialize(vb); HYPRE_IJVectorInitialize(vc); HYPRE_IJMatrixGetObject(ma, (void **)(void *)&par_ma); HYPRE_IJVectorGetObject(va, (void **)(void *)&par_va); HYPRE_IJVectorGetObject(vb, (void **)(void *)&par_vb); HYPRE_IJVectorGetObject(vc, (void **)(void *)&par_vc); HYPRE_IJVectorSetValues(va, A->cmap->nlocal, ni, (double *)x->data); HYPRE_IJVectorAssemble(va); HYPRE_IJVectorAssemble(vb); HYPRE_IJVectorAssemble(vc); HYPRE_IJMatrixGetRowCounts(ma, A->cmap->nlocal, ni, ni + A->rmap->nlocal); for (i = 0, nnz = 0; i < A->rmap->nlocal; i++) nnz += ni[A->rmap->nlocal + i]; #if USE_MPI dnz1 = nnz; MPI_Reduce(&dnz1, &dnz, 1, MPI_DOUBLE, MPI_SUM, 0, g->comm); #else dnz = nnz; #endif HYPRE_ParCSRMatrixMatvec(1.0, par_ma, par_va, 0.0, par_vb); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); for (i = 0; i < loop_count; i++) { HYPRE_ParCSRMatrixMatvec(1.0, par_ma, par_va, 0.0, par_vb); } t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); HYPRE_IJVectorGetValues(vb, A->rmap->nlocal, ni, (double*)y1->data); /*phgPrintf(" --------------------------------------------" "-------------------------\n");*/ phgPrintf("\n"); t1 -= t0; phgPrintf(" HYPRE: time %0.4lf, nnz %0.16lg, %0.2lfMF (%0.2lfMF), " "err: %le\n", t1, dnz, mop / (t1==0 ? 1.:t1), mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y0, 1.0, &y1), 0, NULL)); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); for (i = 0; i < loop_count; i++) { HYPRE_ParCSRMatrixMatvecT(1.0, par_ma, par_va, 0.0, par_vb); } t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); HYPRE_IJVectorGetValues(vb, A->rmap->nlocal, ni, (double*)y1->data); t1 -= t0; phgPrintf(" A'*x: time %0.4lf, nnz %0.16lg, %0.2lfMF (%0.2lfMF), " "err: %le\n", t1, dnz, mop / (t1==0 ? 1.:t1), mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y0, 1.0, &y1), 0, NULL)); phgPerfGetMflops(g, NULL, NULL); /* reset flops counter */ t0 = phgGetTime(NULL); /* Note: 'HYPRE_ParCSRMatrix' is currently typedef'ed to * 'hypre_ParCSRMatrix *' */ par_mb = hypre_ParMatmul((hypre_ParCSRMatrix *)par_ma, (hypre_ParCSRMatrix *)par_ma); t1 = phgGetTime(NULL); mflops = phgPerfGetMflops(g, NULL, NULL); start = hypre_ParCSRMatrixFirstRowIndex(par_mb); end = hypre_ParCSRMatrixLastRowIndex(par_mb) + 1; for (i = start, nnz = 0; i < end; i++) { HYPRE_Int ncols; hypre_ParCSRMatrixGetRow(par_mb, i, &ncols, NULL, NULL); hypre_ParCSRMatrixRestoreRow(par_mb, i, &ncols, NULL, NULL); nnz += ncols; } #if USE_MPI dnz1 = nnz; MPI_Reduce(&dnz1, &dnz, 1, MPI_DOUBLE, MPI_SUM, 0, g->comm); #else dnz = nnz; #endif /* compare B*x <--> A*A*x */ HYPRE_ParCSRMatrixMatvec(1.0, par_ma, par_vb, 0.0, par_vc); HYPRE_ParCSRMatrixMatvec(1.0, (void *)par_mb, par_va, 0.0, par_vb); HYPRE_IJVectorGetValues(vb, A->rmap->nlocal, ni, (double*)y1->data); HYPRE_IJVectorGetValues(vc, A->rmap->nlocal, ni, (double*)y2->data); hypre_ParCSRMatrixDestroy((par_mb)); t1 -= t0; phgPrintf(" A*A: time %0.4lf, nnz %0.16lg, %0.2lfMF, err: %le\n", t1, dnz, mflops, (double)phgVecNorm2(phgVecAXPBY(-1.0, y1, 1.0, &y2), 0, NULL)); phgFree(ni); HYPRE_IJMatrixDestroy(ma); HYPRE_IJVectorDestroy(va); HYPRE_IJVectorDestroy(vb); HYPRE_IJVectorDestroy(vc); } #endif /* USE_HYPRE */ if (A0 != A) phgMatDestroy(&A0); #if 0 if (A->rmap->nglobal > 1000) { VEC *v = phgMapCreateVec(A->rmap, 3); for (i = 0; i < v->map->nlocal; i++) { v->data[i + 0 * v->map->nlocal] = 1 * (i + v->map->partition[g->rank]); v->data[i + 1 * v->map->nlocal] = 2 * (i + v->map->partition[g->rank]); v->data[i + 2 * v->map->nlocal] = 3 * (i + v->map->partition[g->rank]); } phgMatDumpMATLAB(A, "A", "A.m"); phgVecDumpMATLAB(v, "v", "v.m"); phgFinalize(); exit(0); } #endif phgMatDestroy(&A); phgVecDestroy(&x); phgVecDestroy(&y0); phgVecDestroy(&y1); phgVecDestroy(&y2); phgMapDestroy(&map); mem = phgMemoryUsage(g, &mem_peak); dnz = mem / (1024.0 * 1024.0); dnz1 = mem_peak / (1024.0 * 1024.0); /*phgPrintf(" --------------------------------------------" "-------------------------\n");*/ phgPrintf("\n"); phgPrintf(" Memory: current %0.4lgMB, peak %0.4lgMB\n", dnz, dnz1); #if 0 { static int loop_count = 0; if (++loop_count == 4) break; } #endif if (mem_peak > 1024 * (size_t)1024 * mem_max) break; phgRefineAllElements(g, 1); } phgDofFree(&u_h); phgFreeGrid(&g); phgFinalize(); return 0; }
int main(int argc, char *argv[]) { GRID *g; DOF *u, *v, *w, *u0, *v0, *w0; SOLVER *solver; INT i, j; char *fn = "cube.dat"; phgVerbosity = 0; phgInit(&argc, &argv); /*phgPause(0);*/ g = phgNewGrid(-1); if (!phgImport(g, fn, FALSE)) phgError(1, "can't read file \"%s\".\n", fn); for (i = 0; i < 4; i++) phgRefineAllElements(g, 1); phgBalanceGrid(g, 1.0, 0, NULL, 0.); for (i = 0; i < 8; i++) { phgRefineRandomElements(g, "25%"); phgBalanceGrid(g, 1.2, -1, NULL, 0.); } phgCheckConformity(g); u = phgDofNew(g, DOF_P1, 1, "u", DofNoAction); v = phgDofNew(g, DOF_ND1, 1, "v", DofNoAction); w = phgDofNew(g, DOF_P1, 1, "w", DofNoAction); phgDofSetDataByValue(u, 0.0); phgDofSetDataByValue(v, 0.0); phgDofSetDataByValue(w, 0.0); u0 = phgDofNew(g, u->type, u->dim, "u0", DofNoAction); v0 = phgDofNew(g, v->type, v->dim, "v0", DofNoAction); w0 = phgDofNew(g, w->type, w->dim, "w0", DofNoAction); solver = phgSolverCreate(SOLVER_DEFAULT, u, v, w, NULL); for (i = 0; i < DofGetDataCount(u); i++) { j = phgSolverMapD2L(solver, 0, i); u0->data[i] = (FLOAT)(phgSolverMapL2G(solver, j) % 3); phgSolverAddMatrixEntry(solver, j, j, 1.0); phgSolverAddRHSEntry(solver, j, u0->data[i]); } for (i = 0; i < DofGetDataCount(v); i++) { j = phgSolverMapD2L(solver, 1, i); v0->data[i] = (FLOAT)(phgSolverMapL2G(solver, j) % 5); phgSolverAddMatrixEntry(solver, j, j, 1.0); phgSolverAddRHSEntry(solver, j, v0->data[i]); } for (i = 0; i < DofGetDataCount(w); i++) { j = phgSolverMapD2L(solver, 2, i); w0->data[i] = (FLOAT)(phgSolverMapL2G(solver, j) % 7); phgSolverAddMatrixEntry(solver, j, j, 1.0); phgSolverAddRHSEntry(solver, j, w0->data[i]); } phgSolverSolve(solver, TRUE, u, v, w, NULL); phgSolverDestroy(&solver); phgDofAXPY(-1.0, u0, &u); phgDofAXPY(-1.0, v0, &v); phgDofAXPY(-1.0, w0, &w); phgPrintf("Error: u = %lg, v = %lg, w = %lg\n", (double)phgDofNormInftyVec(u), (double)phgDofNormInftyVec(v), (double)phgDofNormInftyVec(w)); phgDofFree(&u); phgDofFree(&v); phgDofFree(&w); phgDofFree(&u0); phgDofFree(&v0); phgDofFree(&w0); phgFreeGrid(&g); phgFinalize(); return 0; }