MFEM  v4.6.0
Finite element discretization library
ex1p.cpp
Go to the documentation of this file.
1 // MFEM Example 1 - Parallel Version
2 // AmgX Modification
3 //
4 // Compile with: make ex1p
5 //
6 // AmgX sample runs:
7 // mpirun -np 4 ex1p
8 // mpirun -np 4 ex1p -d cuda
9 // mpirun -np 10 ex1p --amgx-file amg_pcg.json --amgx-mpi-teams
10 // mpirun -np 4 ex1p --amgx-file amg_pcg.json
11 //
12 // Description: This example code demonstrates the use of MFEM to define a
13 // simple finite element discretization of the Laplace problem
14 // -Delta u = 1 with homogeneous Dirichlet boundary conditions.
15 // Specifically, we discretize using a FE space of the specified
16 // order, or if order < 1 using an isoparametric/isogeometric
17 // space (i.e. quadratic for quadratic curvilinear mesh, NURBS for
18 // NURBS mesh, etc.)
19 //
20 // The example highlights the use of mesh refinement, finite
21 // element grid functions, as well as linear and bilinear forms
22 // corresponding to the left-hand side and right-hand side of the
23 // discrete linear system. We also cover the explicit elimination
24 // of essential boundary conditions, static condensation, and the
25 // optional connection to the GLVis tool for visualization.
26 
27 #include "mfem.hpp"
28 #include <fstream>
29 #include <iostream>
30 
31 using namespace std;
32 using namespace mfem;
33 
34 #ifndef MFEM_USE_AMGX
35 #error This example requires that MFEM is built with MFEM_USE_AMGX=YES
36 #endif
37 
38 int main(int argc, char *argv[])
39 {
40  // 1. Initialize MPI and HYPRE.
41  Mpi::Init(argc, argv);
42  int num_procs = Mpi::WorldSize();
43  int myid = Mpi::WorldRank();
44  Hypre::Init();
45 
46  // 2. Parse command-line options.
47  const char *mesh_file = "../../data/star.mesh";
48  int order = 1;
49  bool static_cond = false;
50  bool pa = false;
51  const char *device_config = "cpu";
52  bool visualization = true;
53  bool amgx_lib = true;
54  bool amgx_mpi_teams = false;
55  const char* amgx_json_file = ""; // JSON file for AmgX
56  int ndevices = 1;
57 
58  OptionsParser args(argc, argv);
59  args.AddOption(&mesh_file, "-m", "--mesh",
60  "Mesh file to use.");
61  args.AddOption(&order, "-o", "--order",
62  "Finite element order (polynomial degree) or -1 for"
63  " isoparametric space.");
64  args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
65  "--no-static-condensation", "Enable static condensation.");
66  args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
67  "--no-partial-assembly", "Enable Partial Assembly.");
68  args.AddOption(&amgx_lib, "-amgx", "--amgx-lib", "-no-amgx",
69  "--no-amgx-lib", "Use AmgX in example.");
70  args.AddOption(&amgx_json_file, "--amgx-file", "--amgx-file",
71  "AMGX solver config file (overrides --amgx-solver, --amgx-verbose)");
72  args.AddOption(&amgx_mpi_teams, "--amgx-mpi-teams", "--amgx-mpi-teams",
73  "--amgx-mpi-gpu-exclusive", "--amgx-mpi-gpu-exclusive",
74  "Create MPI teams when using AmgX to load balance between ranks and GPUs.");
75  args.AddOption(&device_config, "-d", "--device",
76  "Device configuration string, see Device::Configure().");
77  args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
78  "--no-visualization",
79  "Enable or disable GLVis visualization.");
80  args.AddOption(&ndevices, "-nd","--gpus-per-node-in-teams-mode",
81  "Number of GPU devices per node (Only used if amgx_mpi_teams is true).");
82 
83  args.Parse();
84  if (!args.Good())
85  {
86  if (myid == 0)
87  {
88  args.PrintUsage(cout);
89  }
90  return 1;
91  }
92  if (myid == 0)
93  {
94  args.PrintOptions(cout);
95  }
96 
97  // 3. Enable hardware devices such as GPUs, and programming models such as
98  // CUDA, OCCA, RAJA and OpenMP based on command line options.
99  Device device(device_config);
100  if (myid == 0) { device.Print(); }
101 
102  // 4. Read the (serial) mesh from the given mesh file on all processors. We
103  // can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
104  // and volume meshes with the same code.
105  Mesh mesh(mesh_file, 1, 1);
106  int dim = mesh.Dimension();
107 
108  // 5. Refine the serial mesh on all processors to increase the resolution. In
109  // this example we do 'ref_levels' of uniform refinement. We choose
110  // 'ref_levels' to be the largest number that gives a final mesh with no
111  // more than 10,000 elements.
112  {
113  int ref_levels =
114  (int)floor(log(10000./mesh.GetNE())/log(2.)/dim);
115  for (int l = 0; l < ref_levels; l++)
116  {
117  mesh.UniformRefinement();
118  }
119  }
120 
121  // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
122  // this mesh further in parallel to increase the resolution. Once the
123  // parallel mesh is defined, the serial mesh can be deleted.
124  ParMesh pmesh(MPI_COMM_WORLD, mesh);
125  mesh.Clear();
126  {
127  int par_ref_levels = 2;
128  for (int l = 0; l < par_ref_levels; l++)
129  {
130  pmesh.UniformRefinement();
131  }
132  }
133 
134  // 7. Define a parallel finite element space on the parallel mesh. Here we
135  // use continuous Lagrange finite elements of the specified order. If
136  // order < 1, we instead use an isoparametric/isogeometric space.
138  bool delete_fec;
139  if (order > 0)
140  {
141  fec = new H1_FECollection(order, dim);
142  delete_fec = true;
143  }
144  else if (pmesh.GetNodes())
145  {
146  fec = pmesh.GetNodes()->OwnFEC();
147  delete_fec = false;
148  if (myid == 0)
149  {
150  cout << "Using isoparametric FEs: " << fec->Name() << endl;
151  }
152  }
153  else
154  {
155  fec = new H1_FECollection(order = 1, dim);
156  delete_fec = true;
157  }
158  ParFiniteElementSpace fespace(&pmesh, fec);
159  HYPRE_BigInt size = fespace.GlobalTrueVSize();
160  if (myid == 0)
161  {
162  cout << "Number of finite element unknowns: " << size << endl;
163  }
164 
165  // 8. Determine the list of true (i.e. parallel conforming) essential
166  // boundary dofs. In this example, the boundary conditions are defined
167  // by marking all the boundary attributes from the mesh as essential
168  // (Dirichlet) and converting them to a list of true dofs.
169  Array<int> ess_tdof_list;
170  if (pmesh.bdr_attributes.Size())
171  {
172  Array<int> ess_bdr(pmesh.bdr_attributes.Max());
173  ess_bdr = 1;
174  fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
175  }
176 
177  // 9. Set up the parallel linear form b(.) which corresponds to the
178  // right-hand side of the FEM linear system, which in this case is
179  // (1,phi_i) where phi_i are the basis functions in fespace.
180  ParLinearForm b(&fespace);
181  ConstantCoefficient one(1.0);
182  b.AddDomainIntegrator(new DomainLFIntegrator(one));
183  b.Assemble();
184 
185  // 10. Define the solution vector x as a parallel finite element grid function
186  // corresponding to fespace. Initialize x with initial guess of zero,
187  // which satisfies the boundary conditions.
188  ParGridFunction x(&fespace);
189  x = 0.0;
190 
191  // 11. Set up the parallel bilinear form a(.,.) on the finite element space
192  // corresponding to the Laplacian operator -Delta, by adding the Diffusion
193  // domain integrator.
194  ParBilinearForm a(&fespace);
195  if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
196  a.AddDomainIntegrator(new DiffusionIntegrator(one));
197 
198  // 12. Assemble the parallel bilinear form and the corresponding linear
199  // system, applying any necessary transformations such as: parallel
200  // assembly, eliminating boundary conditions, applying conforming
201  // constraints for non-conforming AMR, static condensation, etc.
202  if (static_cond) { a.EnableStaticCondensation(); }
203  a.Assemble();
204 
205  OperatorPtr A;
206  Vector B, X;
207  a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
208 
209  // 13. Solve the linear system A X = B.
210  // * With full assembly, use the BoomerAMG preconditioner from hypre.
211  // * If AmgX is available solve using amg preconditioner.
212  // * With partial assembly, use Jacobi smoothing, for now.
213  Solver *prec = NULL;
214  if (pa)
215  {
216  if (UsesTensorBasis(fespace))
217  {
218  prec = new OperatorJacobiSmoother(a, ess_tdof_list);
219  }
220 
221  CGSolver cg(MPI_COMM_WORLD);
222  cg.SetRelTol(1e-12);
223  cg.SetMaxIter(2000);
224  cg.SetPrintLevel(1);
225  if (prec) { cg.SetPreconditioner(*prec); }
226  cg.SetOperator(*A);
227  cg.Mult(B, X);
228  delete prec;
229  }
230  else if (amgx_lib && strcmp(amgx_json_file,"") == 0)
231  {
232  MFEM_VERIFY(!amgx_mpi_teams,
233  "Please add JSON file to try AmgX with MPI teams mode");
234 
235  bool amgx_verbose = false;
236  prec = new AmgXSolver(MPI_COMM_WORLD, AmgXSolver::PRECONDITIONER,
237  amgx_verbose);
238 
239  CGSolver cg(MPI_COMM_WORLD);
240  cg.SetRelTol(1e-12);
241  cg.SetMaxIter(2000);
242  cg.SetPrintLevel(1);
243  if (prec) { cg.SetPreconditioner(*prec); }
244  cg.SetOperator(*A);
245  cg.Mult(B, X);
246  delete prec;
247 
248  }
249  else if (amgx_lib && strcmp(amgx_json_file,"") != 0)
250  {
251  AmgXSolver amgx;
252  amgx.ReadParameters(amgx_json_file, AmgXSolver::EXTERNAL);
253 
254  if (amgx_mpi_teams)
255  {
256  // Forms MPI teams to load balance between MPI ranks and GPUs
257  amgx.InitMPITeams(MPI_COMM_WORLD, ndevices);
258  }
259  else
260  {
261  // Assumes each MPI rank is paired with a GPU
262  amgx.InitExclusiveGPU(MPI_COMM_WORLD);
263  }
264 
265  amgx.SetOperator(*A.As<HypreParMatrix>());
266  amgx.SetConvergenceCheck(true);
267  amgx.Mult(B, X);
268 
269  // Release MPI communicators and resources created by AmgX
270  amgx.Finalize();
271  }
272  else
273  {
274  prec = new HypreBoomerAMG;
275 
276  CGSolver cg(MPI_COMM_WORLD);
277  cg.SetRelTol(1e-12);
278  cg.SetMaxIter(2000);
279  cg.SetPrintLevel(1);
280  if (prec) { cg.SetPreconditioner(*prec); }
281  cg.SetOperator(*A);
282  cg.Mult(B, X);
283  delete prec;
284  }
285 
286  // 14. Recover the parallel grid function corresponding to X. This is the
287  // local finite element solution on each processor.
288  a.RecoverFEMSolution(X, b, x);
289 
290  // 15. Save the refined mesh and the solution in parallel. This output can
291  // be viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
292  {
293  ostringstream mesh_name, sol_name;
294  mesh_name << "mesh." << setfill('0') << setw(6) << myid;
295  sol_name << "sol." << setfill('0') << setw(6) << myid;
296 
297  ofstream mesh_ofs(mesh_name.str().c_str());
298  mesh_ofs.precision(8);
299  pmesh.Print(mesh_ofs);
300 
301  ofstream sol_ofs(sol_name.str().c_str());
302  sol_ofs.precision(8);
303  x.Save(sol_ofs);
304  }
305 
306  // 16. Send the solution by socket to a GLVis server.
307  if (visualization)
308  {
309  char vishost[] = "localhost";
310  int visport = 19916;
311  socketstream sol_sock(vishost, visport);
312  sol_sock << "parallel " << num_procs << " " << myid << "\n";
313  sol_sock.precision(8);
314  sol_sock << "solution\n" << pmesh << x << flush;
315  }
316 
317  // 17. Free the used memory.
318  if (delete_fec)
319  {
320  delete fec;
321  }
322 
323  return 0;
324 }
Class for domain integration L(v) := (f, v)
Definition: lininteg.hpp:108
virtual void GetEssentialTrueDofs(const Array< int > &bdr_attr_is_ess, Array< int > &ess_tdof_list, int component=-1)
Definition: pfespace.cpp:1031
int visport
Conjugate gradient method.
Definition: solvers.hpp:493
A coefficient that is constant across space and time.
Definition: coefficient.hpp:84
void PrintOptions(std::ostream &out) const
Print the options.
Definition: optparser.cpp:331
int Dimension() const
Dimension of the reference space used within the elements.
Definition: mesh.hpp:1020
void PrintUsage(std::ostream &out) const
Print the usage message.
Definition: optparser.cpp:462
Pointer to an Operator of a specified type.
Definition: handle.hpp:33
virtual void Mult(const Vector &b, Vector &x) const
Operator application: y=A(x).
Definition: solvers.cpp:718
T Max() const
Find the maximal element in the array, using the comparison operator < for class T.
Definition: array.cpp:68
void Print(std::ostream &out=mfem::out)
Print the configuration of the MFEM virtual device object.
Definition: device.cpp:279
bool Good() const
Return true if the command line options were parsed successfully.
Definition: optparser.hpp:159
Abstract parallel finite element space.
Definition: pfespace.hpp:28
void InitMPITeams(const MPI_Comm &comm, const int nDevs)
Definition: amgxsolver.cpp:149
STL namespace.
bool UsesTensorBasis(const FiniteElementSpace &fes)
Return true if the mesh contains only one topology and the elements are tensor elements.
Definition: fespace.hpp:1306
void ReadParameters(const std::string config, CONFIG_SRC source)
Definition: amgxsolver.cpp:186
The BoomerAMG solver in hypre.
Definition: hypre.hpp:1590
void SetConvergenceCheck(bool setConvergenceCheck_=true)
Add a check for convergence after applying Mult.
Definition: amgxsolver.cpp:193
Class for parallel linear form.
Definition: plinearform.hpp:26
virtual void SetPrintLevel(int print_lvl)
Legacy method to set the level of verbosity of the solver output.
Definition: solvers.cpp:71
void Parse()
Parse the command-line options. Note that this function expects all the options provided through the ...
Definition: optparser.cpp:151
char vishost[]
Jacobi smoothing for a given bilinear form (no matrix necessary).
Definition: solvers.hpp:302
double b
Definition: lissajous.cpp:42
void UniformRefinement(int i, const DSTable &, int *, int *, int *)
Definition: mesh.cpp:10232
void SetMaxIter(int max_it)
Definition: solvers.hpp:201
virtual const char * Name() const
Definition: fe_coll.hpp:80
HYPRE_BigInt GlobalTrueVSize() const
Definition: pfespace.hpp:281
Array< int > bdr_attributes
A list of all unique boundary attributes used by the Mesh.
Definition: mesh.hpp:275
void SetRelTol(double rtol)
Definition: solvers.hpp:199
Collection of finite elements from the same family in multiple dimensions. This class is used to matc...
Definition: fe_coll.hpp:26
virtual void SetOperator(const Operator &op)
Definition: amgxsolver.cpp:859
void AddOption(bool *var, const char *enable_short_name, const char *enable_long_name, const char *disable_short_name, const char *disable_long_name, const char *description, bool required=false)
Add a boolean option and set &#39;var&#39; to receive the value. Enable/disable tags are used to set the bool...
Definition: optparser.hpp:82
HYPRE_Int HYPRE_BigInt
int main(int argc, char *argv[])
Definition: ex1p.cpp:69
virtual void Save(std::ostream &out) const
Definition: pgridfunc.cpp:909
int GetNE() const
Returns number of elements.
Definition: mesh.hpp:1086
double a
Definition: lissajous.cpp:41
void InitExclusiveGPU(const MPI_Comm &comm)
Definition: amgxsolver.cpp:120
OpType * As() const
Return the Operator pointer statically cast to a specified OpType. Similar to the method Get()...
Definition: handle.hpp:104
virtual void Mult(const Vector &b, Vector &x) const
Operator application: y=A(x).
Definition: amgxsolver.cpp:902
int dim
Definition: ex24.cpp:53
Class for parallel bilinear form.
int Size() const
Return the logical size of the array.
Definition: array.hpp:141
void Clear()
Clear the contents of the Mesh.
Definition: mesh.hpp:678
virtual void SetOperator(const Operator &op)
Also calls SetOperator for the preconditioner if there is one.
Definition: solvers.hpp:507
Vector data type.
Definition: vector.hpp:58
virtual void SetPreconditioner(Solver &pr)
This should be called before SetOperator.
Definition: solvers.cpp:173
Arbitrary order H1-conforming (continuous) finite elements.
Definition: fe_coll.hpp:259
void GetNodes(Vector &node_coord) const
Definition: mesh.cpp:8302
void Print(std::ostream &out=mfem::out) const override
Definition: pmesh.cpp:4825
Base class for solvers.
Definition: operator.hpp:682
Class for parallel grid function.
Definition: pgridfunc.hpp:32
The MFEM Device class abstracts hardware devices such as GPUs, as well as programming models such as ...
Definition: device.hpp:121
Wrapper for hypre&#39;s ParCSR matrix class.
Definition: hypre.hpp:343
Class for parallel meshes.
Definition: pmesh.hpp:32