DistributedSCHUR.cc

// -*- tab-width: 2; indent-tabs-mode: nil; coding: utf-8-with-signature -*-
//-----------------------------------------------------------------------------
// Copyright 2000-2026 CEA (www.cea.fr) IFPEN (www.ifpenergiesnouvelles.com)
// See the top-level COPYRIGHT file for details.
// SPDX-License-Identifier: Apache-2.0
//-----------------------------------------------------------------------------
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
/*
 * Ce fichier est basé sur le travail sur la bibliothèque AMGCL (version mars 2026)
 * qui peut être trouvée à https://github.com/ddemidov/amgcl.
 *
 * Copyright (c) 2012-2022 Denis Demidov <dennis.demidov@gmail.com>
 * SPDX-License-Identifier: MIT
 */
/*---------------------------------------------------------------------------*/
/*---------------------------------------------------------------------------*/
 
// Pour Eigen
#pragma GCC diagnostic ignored "-Wdeprecated-copy"
#pragma GCC diagnostic ignored "-Wint-in-bool-context"
 
#include <iostream>
#include <iterator>
#include <iomanip>
#include <fstream>
#include <vector>
#include <numeric>
#include <cmath>
 
#include <boost/range/iterator_range.hpp>
#include <boost/scope_exit.hpp>
 
#if defined(SOLVER_BACKEND_CUDA)
#  include "arccore/alina/CudaBackend.h"
#  include "arccore/alina/relaxation_cusparse_ilu0.h"
typedef Arcane::Alina::backend::cuda<double> Backend;
#else
#  ifndef SOLVER_BACKEND_BUILTIN
#    define SOLVER_BACKEND_BUILTIN
#  endif
#include "arccore/alina/BuiltinBackend.h"
typedef Arcane::Alina::BuiltinBackend<double> Backend;
#endif
 
#include "arccore/alina/IO.h"
#include "arccore/alina/Adapters.h"
#include "arccore/alina/AMG.h"
#include "arccore/alina/CoarseningRuntime.h"
#include "arccore/alina/RelaxationRuntime.h"
#include "arccore/alina/DistributedPreconditionedSolver.h"
#include "arccore/alina/DistributedSchurPressureCorrection.h"
#include "arccore/alina/DistributedPreconditioner.h"
#include "arccore/alina/DistributedSubDomainDeflation.h"
#include "arccore/alina/DistributedSolverRuntime.h"
#include "arccore/alina/DistributedDirectSolverRuntime.h"
#include "arccore/alina/Profiler.h"
 
#include "arccore/common/internal/ProgramOptions.h"
 
using namespace Arcane;
using namespace Arcane::Alina;
 
using Alina::precondition;
 
//---------------------------------------------------------------------------
std::vector<ptrdiff_t>
read_problem(const Alina::mpi_communicator& world,
             const std::string& A_file,
             const std::string& rhs_file,
             const std::string& part_file,
             std::vector<ptrdiff_t>& ptr,
             std::vector<ptrdiff_t>& col,
             std::vector<double>& val,
             std::vector<double>& rhs)
{
  // Lit la partition
  ptrdiff_t n, m;
  std::vector<ptrdiff_t> domain(world.size + 1, 0);
  std::vector<int> part;
 
  std::tie(n, m) = Alina::IO::mm_reader(part_file)(part);
  for (int p : part) {
    ++domain[p + 1];
    precondition(p < world.size, "MPI world does not correspond to partition");
  }
  std::partial_sum(domain.begin(), domain.end(), domain.begin());
 
  ptrdiff_t chunk_beg = domain[world.rank];
  ptrdiff_t chunk_end = domain[world.rank + 1];
  ptrdiff_t chunk = chunk_end - chunk_beg;
 
  // Réordonne les inconnues
  std::vector<ptrdiff_t> order(n);
  for (ptrdiff_t i = 0; i < n; ++i)
    order[i] = domain[part[i]]++;
 
  std::rotate(domain.begin(), domain.end() - 1, domain.end());
  domain[0] = 0;
 
  // Lit le bloc de la matrice
  {
    using namespace Arcane::Alina::IO;
 
    std::ifstream A(A_file.c_str(), std::ios::binary);
    precondition(A, "Failed to open matrix file (" + A_file + ")");
 
    std::ifstream b(rhs_file.c_str(), std::ios::binary);
    precondition(b, "Failed to open rhs file (" + rhs_file + ")");
 
    ptrdiff_t rows;
    precondition(read(A, rows), "File I/O error");
    precondition(rows == n, "Matrix and partition have incompatible sizes");
 
    ptr.clear();
    ptr.reserve(chunk + 1);
    ptr.push_back(0);
 
    std::vector<ptrdiff_t> gptr(n + 1);
    precondition(read(A, gptr), "File I/O error");
 
    size_t col_beg = sizeof(rows) + sizeof(gptr[0]) * (n + 1);
    size_t val_beg = col_beg + sizeof(col[0]) * gptr.back();
    size_t rhs_beg = 2 * sizeof(ptrdiff_t);
 
    // Compte les non-zéros locaux
    for (ptrdiff_t i = 0; i < n; ++i)
      if (part[i] == world.rank)
        ptr.push_back(gptr[i + 1] - gptr[i]);
 
    std::partial_sum(ptr.begin(), ptr.end(), ptr.begin());
 
    col.clear();
    col.reserve(ptr.back());
    val.clear();
    val.reserve(ptr.back());
    rhs.clear();
    rhs.reserve(chunk);
 
    // Lit les bandes de la matrice et du membre de droite
    for (ptrdiff_t i = 0; i < n; ++i) {
      if (part[i] != world.rank)
        continue;
 
      ptrdiff_t c;
      A.seekg(col_beg + gptr[i] * sizeof(c));
      for (ptrdiff_t j = gptr[i], e = gptr[i + 1]; j < e; ++j) {
        precondition(read(A, c), "File I/O error (1)");
        col.push_back(order[c]);
      }
    }
 
    for (ptrdiff_t i = 0; i < n; ++i) {
      if (part[i] != world.rank)
        continue;
 
      double v;
      A.seekg(val_beg + gptr[i] * sizeof(v));
      for (ptrdiff_t j = gptr[i], e = gptr[i + 1]; j < e; ++j) {
        precondition(read(A, v), "File I/O error (2)");
        val.push_back(v);
      }
    }
 
    for (ptrdiff_t i = 0; i < n; ++i) {
      if (part[i] != world.rank)
        continue;
 
      double f;
      b.seekg(rhs_beg + i * sizeof(f));
      precondition(read(b, f), "File I/O error (3)");
      rhs.push_back(f);
    }
  }
 
  return domain;
}
 
//---------------------------------------------------------------------------
int main(int argc, char* argv[])
{
  auto& prof = Alina::Profiler::globalProfiler();
 
  int provided;
  MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
  BOOST_SCOPE_EXIT(void)
  {
    MPI_Finalize();
  }
  BOOST_SCOPE_EXIT_END
 
  Alina::mpi_communicator world(MPI_COMM_WORLD);
 
  if (world.rank == 0)
    std::cout << "World size: " << world.size << std::endl;
 
  // Lit la configuration depuis la ligne de commande
  namespace po = Arcane::ProgramOptions;
  using std::string;
  po::options_description desc("Options");
 
  desc.add_options()("help,h", "show help")(
  "matrix,A",
  po::value<string>()->required(),
  "La matrice du système au format binaire")(
  "rhs,f",
  po::value<string>(),
  "Le membre de droite au format binaire")(
  "part,s",
  po::value<string>()->required(),
  "Partitionnement du problème au format MatrixMarket")(
  "pmask,m",
  po::value<string>(),
  "Le masque de pression au format binaire. Ou, si le paramètre a "
  "la forme '%n:m', alors chaque variable (n+i*m)-ième est traitée comme une pression.")(
  "params,P",
  po::value<string>(),
  "fichier de paramètres au format json")(
  "prm,p",
  po::value<std::vector<string>>()->multitoken(),
  "Paramètres spécifiés sous forme de paires nom=valeur. "
  "Peut être fourni plusieurs fois. Exemples :\n"
  "  -p solver.tol=1e-3\n"
  "  -p precond.coarse_enough=300");
 
  po::variables_map vm;
  po::store(po::parse_command_line(argc, argv, desc), vm);
 
  if (vm.count("help")) {
    if (world.rank == 0)
      std::cout << desc << std::endl;
    return 0;
  }
 
  po::notify(vm);
 
  Alina::PropertyTree prm;
  if (vm.count("params"))
    prm.read_json(vm["params"].as<string>());
 
  if (vm.count("prm")) {
    for (const string& v : vm["prm"].as<std::vector<string>>()) {
      prm.putKeyValue(v);
    }
  }
 
  prof.tic("read problem");
  std::vector<ptrdiff_t> ptr;
  std::vector<ptrdiff_t> col;
  std::vector<double> val;
  std::vector<double> rhs;
 
  std::vector<ptrdiff_t> domain = read_problem(
  world,
  vm["matrix"].as<string>(), vm["rhs"].as<string>(), vm["part"].as<string>(),
  ptr, col, val, rhs);
 
  ptrdiff_t chunk = domain[world.rank + 1] - domain[world.rank];
  prof.toc("read problem");
 
  std::vector<char> pm;
  if (vm.count("pmask")) {
    std::string pmask = vm["pmask"].as<string>();
    prm.put("precond.pmask_size", chunk);
 
    switch (pmask[0]) {
    case '%':
    case '<':
    case '>':
      prm.put("precond.pmask_pattern", pmask);
      break;
    default:
      precondition(false, "Pressure mask may only be set with a pattern");
    }
  }
 
  std::function<double(ptrdiff_t, unsigned)> dv = Alina::constant_deflation(1);
  prm.put("precond.psolver.num_def_vec", 1);
  prm.put("precond.psolver.def_vec", &dv);
 
  Backend::params bprm;
 
#if defined(SOLVER_BACKEND_VEXCL)
  vex::Context ctx(vex::Filter::Env);
  std::cout << ctx << std::endl;
  bprm.q = ctx;
#elif defined(SOLVER_BACKEND_CUDA)
  cusparseCreate(&bprm.cusparse_handle);
#endif
 
  auto f = Backend::copy_vector(rhs, bprm);
  auto x = Backend::create_vector(chunk, bprm);
 
  Alina::backend::clear(*x);
 
  MPI_Barrier(world);
 
  prof.tic("setup");
  typedef DistributedPreconditionedSolver<
    DistributedSchurPressureCorrection<
      DistributedPreconditionedSolver<
        DistributedBlockPreconditioner<
         RelaxationAsPreconditioner<Backend, RelaxationRuntime>>,
        DistributedSolverRuntime<Backend>>,
      DistributedSubDomainDeflation<
        AMG<Backend, CoarseningRuntime, RelaxationRuntime>,
        DistributedSolverRuntime<Backend>,
        DistributedDirectSolverRuntime<Backend>>>,
    DistributedSolverRuntime<Backend>>
  Solver;
 
  Solver solve(world, std::tie(chunk, ptr, col, val), prm, bprm);
  double tm_setup = prof.toc("setup");
 
  prof.tic("solve");
  Alina::SolverResult r = solve(*f, *x);
  double tm_solve = prof.toc("solve");
 
  if (world.rank == 0) {
    std::cout << "Iters: " << r.nbIteration() << std::endl
              << "Error: " << r.residual() << std::endl
              << prof << std::endl;
  }
}