LBFluid.cpp

Go to the documentation of this file.

/*
 * Copyright (C) 2021-2023 The ESPResSo project
 *
 * This file is part of ESPResSo.
 *
 * ESPResSo is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * ESPResSo is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
#include <config/config.hpp>
 
#ifdef ESPRESSO_WALBERLA
 
#include "LBFluid.hpp"
#include "LBWalberlaNodeState.hpp"
#include "WalberlaCheckpoint.hpp"
 
#include "core/BoxGeometry.hpp"
#include "core/lb/LBWalberla.hpp"
#include "core/lees_edwards/lees_edwards.hpp"
#include "core/lees_edwards/protocols.hpp"
#include "core/system/System.hpp"
 
#include <script_interface/code_info/CodeInfo.hpp>
#include <script_interface/communication.hpp>
 
#include <walberla_bridge/LatticeWalberla.hpp>
#include <walberla_bridge/lattice_boltzmann/LeesEdwardsPack.hpp>
#include <walberla_bridge/lattice_boltzmann/lb_walberla_init.hpp>
#include <walberla_bridge/utils/ResourceManager.hpp>
#include <walberla_bridge/walberla_init.hpp>
 
#include <utils/Vector.hpp>
#include <utils/matrix.hpp>
#include <utils/mpi/reduce_optional.hpp>
 
#include <boost/mpi.hpp>
#include <boost/mpi/collectives/all_reduce.hpp>
#include <boost/mpi/collectives/broadcast.hpp>
 
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <filesystem>
#include <functional>
#include <memory>
#include <optional>
#include <sstream>
#include <stdexcept>
#include <string>
#include <unordered_map>
#include <vector>
 
namespace ScriptInterface::walberla {
 
std::unordered_map<std::string, int> const LBVTKHandle::obs_map = {
    {"density", static_cast<int>(OutputVTK::density)},
    {"velocity_vector", static_cast<int>(OutputVTK::velocity_vector)},
    {"pressure_tensor", static_cast<int>(OutputVTK::pressure_tensor)},
};
 
Variant LBFluid::do_call_method(std::string const &name,
                                VariantMap const &params) {
  if (name == "activate") {
    auto &system = get_system();
    context()->parallel_try_catch(
        [&]() { lb_throw_if_expired(m_mpi_cart_comm_observer); });
    system.lb.set<::LB::LBWalberla>(m_instance, m_lb_params);
    system.lb.update_collision_model();
    m_is_active = true;
    return {};
  }
  if (name == "deactivate") {
    get_system().lb.reset();
    m_is_active = false;
    return {};
  }
  if (not name.starts_with("get_")) {
    context()->parallel_try_catch(
        [&]() { lb_throw_if_expired(m_mpi_cart_comm_observer); });
  }
  if (name == "add_force_at_pos") {
    auto const &box_geo = *get_system().box_geo;
    auto const pos = get_value<Utils::Vector3d>(params, "pos");
    auto const f = get_value<Utils::Vector3d>(params, "force");
    auto const folded_pos = box_geo.folded_position(pos);
    m_instance->add_force_at_pos(folded_pos * m_conv_dist, f * m_conv_force);
    return {};
  }
  if (name == "get_interpolated_velocity") {
    auto const pos = get_value<Utils::Vector3d>(params, "pos");
    return get_interpolated_velocity(pos);
  }
  if (name == "get_boundary_force_from_shape") {
    return get_boundary_force_from_shape(
        get_value<std::vector<int>>(params, "raster"));
  }
  if (name == "get_boundary_force") {
    return get_boundary_force();
  }
  if (name == "get_pressure_tensor") {
    return get_average_pressure_tensor();
  }
  if (name == "load_checkpoint") {
    auto const path = get_value<std::filesystem::path>(params, "path");
    auto const mode = get_value<int>(params, "mode");
    load_checkpoint(path, mode);
    return {};
  }
  if (name == "save_checkpoint") {
    auto const path = get_value<std::filesystem::path>(params, "path");
    auto const mode = get_value<int>(params, "mode");
    save_checkpoint(path, mode);
    return {};
  }
  if (name == "clear_boundaries") {
    m_instance->clear_boundaries();
    get_system().on_lb_boundary_conditions_change();
    return {};
  }
  if (name == "add_boundary_from_shape") {
    m_instance->update_boundary_from_shape(
        get_value<std::vector<int>>(params, "raster"),
        get_value<std::vector<double>>(params, "values"));
    return {};
  }
  if (name == "get_lattice_speed") {
    return 1. / m_conv_speed;
  }
 
  return Base::do_call_method(name, params);
}
 
void LBFluid::make_instance(VariantMap const &params) {
  auto const visc = get_value<double>(params, "kinematic_viscosity");
  auto const dens = get_value<double>(params, "density");
  auto const gpu = get_value_or(params, "gpu", false);
  auto const precision = get_value_or(params, "single_precision", gpu);
  auto const lb_lattice = m_lattice->lattice();
  auto const lb_visc = m_conv_visc * visc;
  auto const lb_dens = m_conv_dens * dens;
  auto *make_new_instance = &new_lb_walberla_cpu;
  if (gpu) {
    std::vector<std::string> required_features;
    required_features.emplace_back("CUDA");
    CodeInfo::check_features(required_features);
#ifdef ESPRESSO_CUDA
    auto const blocks_per_mpi_rank = get_value<Utils::Vector3i>(
        m_lattice->get_parameter("blocks_per_mpi_rank"));
    if (blocks_per_mpi_rank != Utils::Vector3i{{1, 1, 1}}) {
      throw std::runtime_error(
          "Using more than one block per MPI rank is not supported for GPU LB");
    }
    make_new_instance = &new_lb_walberla_gpu;
#endif
  }
  m_instance = make_new_instance(lb_lattice, lb_visc, lb_dens, precision);
}
 
void LBFluid::do_construct(VariantMap const &params) {
  m_lattice = get_value<std::shared_ptr<LatticeWalberla>>(params, "lattice");
  m_vtk_writers =
      get_value_or<decltype(m_vtk_writers)>(params, "vtk_writers", {});
  auto const tau = get_value<double>(params, "tau");
  auto const agrid = get_value<double>(m_lattice->get_parameter("agrid"));
  auto const visc = get_value<double>(params, "kinematic_viscosity");
  auto const dens = get_value<double>(params, "density");
  auto const kT = get_value<double>(params, "kT");
  auto const ext_f = get_value<Utils::Vector3d>(params, "ext_force_density");
  m_lb_params = std::make_shared<::LB::LBWalberlaParams>(agrid, tau);
  m_is_active = false;
  auto const seed = get_value<int>(params, "seed");
  context()->parallel_try_catch([&]() {
    if (tau <= 0.) {
      throw std::domain_error("Parameter 'tau' must be > 0");
    }
    m_conv_dist = 1. / agrid;
    m_conv_visc = Utils::int_pow<1>(tau) / Utils::int_pow<2>(agrid);
    m_conv_energy = Utils::int_pow<2>(tau) / Utils::int_pow<2>(agrid);
    m_conv_dens = Utils::int_pow<3>(agrid);
    m_conv_speed = Utils::int_pow<1>(tau) / Utils::int_pow<1>(agrid);
    m_conv_press = Utils::int_pow<2>(tau) * Utils::int_pow<1>(agrid);
    m_conv_force = Utils::int_pow<2>(tau) / Utils::int_pow<1>(agrid);
    m_conv_force_dens = Utils::int_pow<2>(tau) * Utils::int_pow<2>(agrid);
    auto const lb_visc = m_conv_visc * visc;
    auto const lb_dens = m_conv_dens * dens;
    auto const lb_kT = m_conv_energy * kT;
    auto const lb_ext_f = m_conv_force_dens * ext_f;
    if (seed < 0) {
      throw std::domain_error("Parameter 'seed' must be >= 0");
    }
    if (lb_kT < 0.) {
      throw std::domain_error("Parameter 'kT' must be >= 0");
    }
    if (lb_dens <= 0.) {
      throw std::domain_error("Parameter 'density' must be > 0");
    }
    if (lb_visc < 0.) {
      throw std::domain_error("Parameter 'kinematic_viscosity' must be >= 0");
    }
    make_instance(params);
    m_mpi_cart_comm_observer = ::walberla::get_mpi_cart_comm_observer();
    m_instance->set_collision_model(lb_kT, seed);
    m_instance->set_external_force(lb_ext_f);
    m_instance->ghost_communication();
    for (auto &vtk : m_vtk_writers) {
      vtk->attach_to_lattice(m_instance, get_lattice_to_md_units_conversion());
    }
  });
}
 
Variant
LBFluid::get_boundary_force_from_shape(std::vector<int> const &raster) const {
  auto const local =
      m_instance->get_boundary_force_from_shape(raster) / m_conv_force;
  return mpi_reduce_sum(context()->get_comm(), local);
}
 
Variant LBFluid::get_boundary_force() const {
  auto const local = m_instance->get_boundary_force() / m_conv_force;
  return mpi_reduce_sum(context()->get_comm(), local);
}
 
std::vector<Variant> LBFluid::get_average_pressure_tensor() const {
  auto const local = m_instance->get_pressure_tensor() / m_conv_press;
  auto const tensor_flat = mpi_reduce_sum(context()->get_comm(), local);
  auto tensor = Utils::Matrix<double, 3, 3>{};
  std::ranges::copy(tensor_flat, tensor.m_data.begin());
  return std::vector<Variant>{tensor.row<0>().as_vector(),
                              tensor.row<1>().as_vector(),
                              tensor.row<2>().as_vector()};
}
 
Variant LBFluid::get_interpolated_velocity(Utils::Vector3d const &pos) const {
  auto const &box_geo = *get_system().box_geo;
  auto const lb_pos = box_geo.folded_position(pos) * m_conv_dist;
  auto const result = m_instance->get_velocity_at_pos(lb_pos);
  return Utils::Mpi::reduce_optional(context()->get_comm(), result) /
         m_conv_speed;
}
 
void LBFluid::load_checkpoint(std::filesystem::path const &path, int mode) {
  auto &lb_obj = *m_instance;
 
  auto const read_metadata = [&lb_obj](CheckpointFile &cpfile) {
    auto const expected_grid_size = lb_obj.get_lattice().get_grid_dimensions();
    auto const expected_pop_size = lb_obj.stencil_size();
    Utils::Vector3i read_grid_size;
    std::size_t read_pop_size;
    cpfile.read(read_grid_size);
    cpfile.read(read_pop_size);
    if (read_grid_size != expected_grid_size) {
      std::stringstream message;
      message << "grid dimensions mismatch, read [" << read_grid_size << "], "
              << "expected [" << expected_grid_size << "].";
      throw std::runtime_error(message.str());
    }
    if (read_pop_size != expected_pop_size) {
      throw std::runtime_error("population size mismatch, read " +
                               std::to_string(read_pop_size) + ", expected " +
                               std::to_string(expected_pop_size) + ".");
    }
  };
 
  auto const read_data = [&lb_obj](CheckpointFile &cpfile) {
    auto const grid_size = lb_obj.get_lattice().get_grid_dimensions();
    auto const i_max = grid_size[0];
    auto const j_max = grid_size[1];
    auto const k_max = grid_size[2];
    LBWalberlaNodeState cpnode;
    cpnode.populations.resize(lb_obj.stencil_size());
    for (int i = 0; i < i_max; i++) {
      for (int j = 0; j < j_max; j++) {
        for (int k = 0; k < k_max; k++) {
          auto const ind = Utils::Vector3i{{i, j, k}};
          cpfile.read(cpnode.populations);
          cpfile.read(cpnode.last_applied_force);
          cpfile.read(cpnode.is_boundary);
          if (cpnode.is_boundary) {
            cpfile.read(cpnode.slip_velocity);
          }
          lb_obj.set_node_population(ind, cpnode.populations);
          lb_obj.set_node_last_applied_force(ind, cpnode.last_applied_force);
          if (cpnode.is_boundary) {
            lb_obj.set_node_velocity_at_boundary(ind, cpnode.slip_velocity);
          }
        }
      }
    }
  };
 
  auto const on_success = [&lb_obj]() {
    lb_obj.ghost_communication();
    lb_obj.reallocate_ubb_field();
  };
 
  load_checkpoint_common(*context(), "LB", path, mode, read_metadata, read_data,
                         on_success);
}
 
void LBFluid::save_checkpoint(std::filesystem::path const &path, int mode) {
  auto &lb_obj = *m_instance;
 
  auto const write_metadata = [&lb_obj,
                               mode](std::shared_ptr<CheckpointFile> cpfile_ptr,
                                     Context const &context) {
    auto const grid_size = lb_obj.get_lattice().get_grid_dimensions();
    auto const pop_size = lb_obj.stencil_size();
    if (context.is_head_node()) {
      cpfile_ptr->write(grid_size);
      cpfile_ptr->write(pop_size);
      unit_test_handle(mode);
    }
  };
 
  auto const on_failure = [](std::shared_ptr<CheckpointFile> const &,
                             Context const &context) {
    if (context.is_head_node()) {
      auto failure = true;
      boost::mpi::broadcast(context.get_comm(), failure, 0);
    }
  };
 
  auto const write_data = [&lb_obj,
                           mode](std::shared_ptr<CheckpointFile> cpfile_ptr,
                                 Context const &context) {
    auto const get_node_checkpoint =
        [&](Utils::Vector3i const &ind) -> std::optional<LBWalberlaNodeState> {
      auto const pop = lb_obj.get_node_population(ind);
      auto const laf = lb_obj.get_node_last_applied_force(ind);
      auto const lbb = lb_obj.get_node_is_boundary(ind);
      auto const vbb = lb_obj.get_node_velocity_at_boundary(ind);
      if (pop and laf and lbb and ((*lbb) ? vbb.has_value() : true)) {
        LBWalberlaNodeState cpnode;
        cpnode.populations = *pop;
        cpnode.last_applied_force = *laf;
        cpnode.is_boundary = *lbb;
        if (*lbb) {
          cpnode.slip_velocity = *vbb;
        }
        return {cpnode};
      }
      return std::nullopt;
    };
 
    auto failure = false;
    auto const &comm = context.get_comm();
    auto const is_head_node = context.is_head_node();
    auto const unit_test_mode = (mode != static_cast<int>(CptMode::ascii)) and
                                (mode != static_cast<int>(CptMode::binary));
    auto const grid_size = lb_obj.get_lattice().get_grid_dimensions();
    auto const i_max = grid_size[0];
    auto const j_max = grid_size[1];
    auto const k_max = grid_size[2];
    LBWalberlaNodeState cpnode;
    for (int i = 0; i < i_max; i++) {
      for (int j = 0; j < j_max; j++) {
        for (int k = 0; k < k_max; k++) {
          auto const ind = Utils::Vector3i{{i, j, k}};
          auto const result = get_node_checkpoint(ind);
          if (!unit_test_mode) {
            assert(1 == boost::mpi::all_reduce(comm, static_cast<int>(!!result),
                                               std::plus<>()) &&
                   "Incorrect number of return values");
          }
          if (is_head_node) {
            if (result) {
              cpnode = *result;
            } else {
              comm.recv(boost::mpi::any_source, 42, cpnode);
            }
            auto &cpfile = *cpfile_ptr;
            cpfile.write(cpnode.populations);
            cpfile.write(cpnode.last_applied_force);
            cpfile.write(cpnode.is_boundary);
            if (cpnode.is_boundary) {
              cpfile.write(cpnode.slip_velocity);
            }
            boost::mpi::broadcast(comm, failure, 0);
          } else {
            if (result) {
              comm.send(0, 42, *result);
            }
            boost::mpi::broadcast(comm, failure, 0);
            if (failure) {
              return;
            }
          }
        }
      }
    }
  };
 
  save_checkpoint_common(*context(), "LB", path, mode, write_metadata,
                         write_data, on_failure);
}
 
} // namespace ScriptInterface::walberla
 
#endif // ESPRESSO_WALBERLA