Homec4science
Diffusion akantu (bugfixes/flatten-internal-that-are-missing) Browse

rAKA/src/synchronizer/grid_synchronizer.cc
5cf490a6c1b7bugfixes/flatten…

Last Change
Raw File

grid_synchronizer.cc
View Options

/**
* @file grid_synchronizer.cc
*
* @author Aurelia Isabel Cuba Ramos <aurelia.cubaramos@epfl.ch>
* @author Nicolas Richart <nicolas.richart@epfl.ch>
*
* @date creation: Mon Oct 03 2011
* @date last modification: Tue Nov 07 2017
*
* @brief implementation of the grid synchronizer
*
* @section LICENSE
*
* Copyright (©) 2010-2018 EPFL (Ecole Polytechnique Fédérale de Lausanne)
* Laboratory (LSMS - Laboratoire de Simulation en Mécanique des Solides)
*
* Akantu is free software: you can redistribute it and/or modify it under the
* terms of the GNU Lesser General Public License as published by the Free
* Software Foundation, either version 3 of the License, or (at your option) any
* later version.
*
* Akantu 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 Lesser General Public License for more
* details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with Akantu. If not, see <http://www.gnu.org/licenses/>.
*
*/
/* -------------------------------------------------------------------------- */
#include "grid_synchronizer.hh"
#include "aka_grid_dynamic.hh"
#include "communicator.hh"
#include "fe_engine.hh"
#include "integration_point.hh"
#include "mesh.hh"
#include "mesh_io.hh"
#include <iostream>
/* -------------------------------------------------------------------------- */
namespace akantu {
/* -------------------------------------------------------------------------- */
template <class E>
void GridSynchronizer::createGridSynchronizer(const SpatialGrid<E> & grid) {
AKANTU_DEBUG_IN();
const Communicator & comm = this->mesh.getCommunicator();
UInt nb_proc = comm.getNbProc();
UInt my_rank = comm.whoAmI();
if (nb_proc == 1)
return;
UInt spatial_dimension = this->mesh.getSpatialDimension();
BBox my_bounding_box(spatial_dimension);
const auto & lower = grid.getLowerBounds();
const auto & upper = grid.getUpperBounds();
const auto & spacing = grid.getSpacing();
my_bounding_box.getLowerBounds() = lower - spacing;
my_bounding_box.getUpperBounds() = upper + spacing;
AKANTU_DEBUG_INFO(
"Exchange of bounding box to detect the overlapping regions.");
auto && bboxes = my_bounding_box.allGather(comm);
std::vector<bool> intersects_proc(nb_proc);
std::fill(intersects_proc.begin(), intersects_proc.end(), true);
Matrix<Int> first_cells(spatial_dimension, nb_proc);
Matrix<Int> last_cells(spatial_dimension, nb_proc);
std::map<UInt, ElementTypeMapArray<UInt>> element_per_proc;
// check the overlapping between my box and the one from other processors
for (UInt p = 0; p < nb_proc; ++p) {
if (p == my_rank)
continue;
const auto & proc_bounding_box = bboxes[p];
auto intersection = my_bounding_box.intersection(proc_bounding_box);
Vector<Int> first_cell_p = first_cells(p);
Vector<Int> last_cell_p = last_cells(p);
intersects_proc[p] = intersection;
if (intersects_proc[p]) {
for (UInt s = 0; s < spatial_dimension; ++s) {
first_cell_p(s) = grid.getCellID(intersection.getLowerBounds()(s), s);
last_cell_p(s) = grid.getCellID(intersection.getUpperBounds()(s), s);
}
}
// create the list of cells in the overlapping
using CellID = typename SpatialGrid<E>::CellID;
std::vector<CellID> cell_ids;
if (intersects_proc[p]) {
AKANTU_DEBUG_INFO("I intersects with processor " << p);
CellID cell_id(spatial_dimension);
// for (UInt i = 0; i < spatial_dimension; ++i) {
// if(first_cell_p[i] != 0) --first_cell_p[i];
// if(last_cell_p[i] != 0) ++last_cell_p[i];
// }
for (Int fd = first_cell_p(0); fd <= last_cell_p(0); ++fd) {
cell_id.setID(0, fd);
if (spatial_dimension == 1) {
cell_ids.push_back(cell_id);
} else {
for (Int sd = first_cell_p(1); sd <= last_cell_p(1); ++sd) {
cell_id.setID(1, sd);
if (spatial_dimension == 2) {
cell_ids.push_back(cell_id);
} else {
for (Int ld = first_cell_p(2); ld <= last_cell_p(2); ++ld) {
cell_id.setID(2, ld);
cell_ids.push_back(cell_id);
}
}
}
}
}
// get the list of elements in the cells of the overlapping
std::set<Element> to_send;
for (auto & cur_cell_id : cell_ids) {
auto & cell = grid.getCell(cur_cell_id);
for (auto & element : cell) {
to_send.insert(element);
}
}
AKANTU_DEBUG_INFO("I have prepared " << to_send.size()
<< " elements to send to processor "
<< p);
auto & scheme = this->getCommunications().createSendScheme(p);
std::stringstream sstr;
sstr << "element_per_proc_" << p;
element_per_proc.emplace(
std::piecewise_construct, std::forward_as_tuple(p),
std::forward_as_tuple(sstr.str(), id, memory_id));
ElementTypeMapArray<UInt> & elempproc = element_per_proc[p];
for (auto elem : to_send) {
ElementType type = elem.type;
UInt nb_nodes_per_element = mesh.getNbNodesPerElement(type);
// /!\ this part must be slow due to the access in the
// ElementTypeMapArray<UInt>
if (!elempproc.exists(type, _not_ghost))
elempproc.alloc(0, nb_nodes_per_element, type, _not_ghost);
Vector<UInt> global_connect(nb_nodes_per_element);
Vector<UInt> local_connect = mesh.getConnectivity(type).begin(
nb_nodes_per_element)[elem.element];
for (UInt i = 0; i < nb_nodes_per_element; ++i) {
global_connect(i) = mesh.getNodeGlobalId(local_connect(i));
AKANTU_DEBUG_ASSERT(
global_connect(i) < mesh.getNbGlobalNodes(),
"This global node send in the connectivity does not seem correct "
<< global_connect(i) << " corresponding to "
<< local_connect(i) << " from element " << elem.element);
}
elempproc(type).push_back(global_connect);
scheme.push_back(elem);
}
}
}
AKANTU_DEBUG_INFO("I have finished to compute intersection,"
<< " no it's time to communicate with my neighbors");
/**
* Sending loop, sends the connectivity asynchronously to all concerned proc
*/
std::vector<CommunicationRequest> isend_requests;
Tensor3<UInt> space(2, _max_element_type, nb_proc);
for (UInt p = 0; p < nb_proc; ++p) {
if (p == my_rank)
continue;
if (not intersects_proc[p])
continue;
Matrix<UInt> info_proc = space(p);
auto & elempproc = element_per_proc[p];
UInt count = 0;
for (auto type : elempproc.elementTypes(_all_dimensions, _not_ghost)) {
Array<UInt> & conn = elempproc(type, _not_ghost);
Vector<UInt> info = info_proc((UInt)type);
info[0] = (UInt)type;
info[1] = conn.size() * conn.getNbComponent();
AKANTU_DEBUG_INFO(
"I have " << conn.size() << " elements of type " << type
<< " to send to processor " << p << " (communication tag : "
<< Tag::genTag(my_rank, count, DATA_TAG) << ")");
isend_requests.push_back(
comm.asyncSend(info, p, Tag::genTag(my_rank, count, SIZE_TAG)));
if (info[1] != 0)
isend_requests.push_back(comm.asyncSend<UInt>(
conn, p, Tag::genTag(my_rank, count, DATA_TAG)));
++count;
}
Vector<UInt> info = info_proc((UInt)_not_defined);
info[0] = (UInt)_not_defined;
info[1] = 0;
isend_requests.push_back(
comm.asyncSend(info, p, Tag::genTag(my_rank, count, SIZE_TAG)));
}
/**
* Receives the connectivity and store them in the ghosts elements
*/
MeshAccessor mesh_accessor(mesh);
auto & global_nodes_ids = mesh_accessor.getNodesGlobalIds();
auto & nodes_type = mesh_accessor.getNodesFlags();
std::vector<CommunicationRequest> isend_nodes_requests;
Vector<UInt> nb_nodes_to_recv(nb_proc);
UInt nb_total_nodes_to_recv = 0;
UInt nb_current_nodes = global_nodes_ids.size();
NewNodesEvent new_nodes;
NewElementsEvent new_elements;
std::map<UInt, std::vector<UInt>> ask_nodes_per_proc;
for (UInt p = 0; p < nb_proc; ++p) {
nb_nodes_to_recv(p) = 0;
if (p == my_rank)
continue;
if (!intersects_proc[p])
continue;
auto & scheme = this->getCommunications().createRecvScheme(p);
ask_nodes_per_proc.emplace(std::piecewise_construct,
std::forward_as_tuple(p),
std::forward_as_tuple(0));
auto & ask_nodes = ask_nodes_per_proc[p];
UInt count = 0;
ElementType type = _not_defined;
do {
Vector<UInt> info(2);
comm.receive(info, p, Tag::genTag(p, count, SIZE_TAG));
type = (ElementType)info[0];
if (type == _not_defined)
break;
UInt nb_nodes_per_element = mesh.getNbNodesPerElement(type);
UInt nb_element = info[1] / nb_nodes_per_element;
Array<UInt> tmp_conn(nb_element, nb_nodes_per_element);
tmp_conn.clear();
if (info[1] != 0)
comm.receive<UInt>(tmp_conn, p, Tag::genTag(p, count, DATA_TAG));
AKANTU_DEBUG_INFO("I will receive "
<< nb_element << " elements of type "
<< ElementType(info[0]) << " from processor " << p
<< " (communication tag : "
<< Tag::genTag(p, count, DATA_TAG) << ")");
auto & ghost_connectivity = mesh_accessor.getConnectivity(type, _ghost);
auto & ghost_counter = mesh_accessor.getGhostsCounters(type, _ghost);
UInt nb_ghost_element = ghost_connectivity.size();
Element element{type, 0, _ghost};
Vector<UInt> conn(nb_nodes_per_element);
for (UInt el = 0; el < nb_element; ++el) {
UInt nb_node_to_ask_for_elem = 0;
for (UInt n = 0; n < nb_nodes_per_element; ++n) {
UInt gn = tmp_conn(el, n);
UInt ln = global_nodes_ids.find(gn);
AKANTU_DEBUG_ASSERT(gn < mesh.getNbGlobalNodes(),
"This global node seems not correct "
<< gn << " from element " << el << " node "
<< n);
if (ln == UInt(-1)) {
global_nodes_ids.push_back(gn);
nodes_type.push_back(NodeFlag::_pure_ghost); // pure ghost node
ln = nb_current_nodes;
new_nodes.getList().push_back(ln);
++nb_current_nodes;
ask_nodes.push_back(gn);
++nb_node_to_ask_for_elem;
}
conn[n] = ln;
}
// all the nodes are already known locally, the element should
// already exists
auto c = UInt(-1);
if (nb_node_to_ask_for_elem == 0) {
c = ghost_connectivity.find(conn);
element.element = c;
}
if (c == UInt(-1)) {
element.element = nb_ghost_element;
++nb_ghost_element;
ghost_connectivity.push_back(conn);
ghost_counter.push_back(1);
new_elements.getList().push_back(element);
} else {
++ghost_counter(c);
}
scheme.push_back(element);
}
count++;
} while (type != _not_defined);
AKANTU_DEBUG_INFO("I have "
<< ask_nodes.size()
<< " missing nodes for elements coming from processor "
<< p << " (communication tag : "
<< Tag::genTag(my_rank, 0, ASK_NODES_TAG) << ")");
ask_nodes.push_back(UInt(-1));
isend_nodes_requests.push_back(
comm.asyncSend(ask_nodes, p, Tag::genTag(my_rank, 0, ASK_NODES_TAG)));
nb_nodes_to_recv(p) = ask_nodes.size() - 1;
nb_total_nodes_to_recv += nb_nodes_to_recv(p);
}
comm.waitAll(isend_requests);
comm.freeCommunicationRequest(isend_requests);
/**
* Sends requested nodes to proc
*/
auto & nodes = const_cast<Array<Real> &>(mesh.getNodes());
UInt nb_nodes = nodes.size();
std::vector<CommunicationRequest> isend_coordinates_requests;
std::map<UInt, Array<Real>> nodes_to_send_per_proc;
for (UInt p = 0; p < nb_proc; ++p) {
if (p == my_rank || !intersects_proc[p])
continue;
Array<UInt> asked_nodes;
CommunicationStatus status;
AKANTU_DEBUG_INFO("Waiting list of nodes to send to processor "
<< p << "(communication tag : "
<< Tag::genTag(p, 0, ASK_NODES_TAG) << ")");
comm.probe<UInt>(p, Tag::genTag(p, 0, ASK_NODES_TAG), status);
UInt nb_nodes_to_send = status.size();
asked_nodes.resize(nb_nodes_to_send);
AKANTU_DEBUG_INFO("I have " << nb_nodes_to_send - 1
<< " nodes to send to processor " << p
<< " (communication tag : "
<< Tag::genTag(p, 0, ASK_NODES_TAG) << ")");
AKANTU_DEBUG_INFO("Getting list of nodes to send to processor "
<< p << " (communication tag : "
<< Tag::genTag(p, 0, ASK_NODES_TAG) << ")");
comm.receive(asked_nodes, p, Tag::genTag(p, 0, ASK_NODES_TAG));
nb_nodes_to_send--;
asked_nodes.resize(nb_nodes_to_send);
nodes_to_send_per_proc.emplace(std::piecewise_construct,
std::forward_as_tuple(p),
std::forward_as_tuple(0, spatial_dimension));
auto & nodes_to_send = nodes_to_send_per_proc[p];
auto node_it = nodes.begin(spatial_dimension);
for (UInt n = 0; n < nb_nodes_to_send; ++n) {
UInt ln = global_nodes_ids.find(asked_nodes(n));
AKANTU_DEBUG_ASSERT(ln != UInt(-1), "The node ["
<< asked_nodes(n)
<< "] requested by proc " << p
<< " was not found locally!");
nodes_to_send.push_back(node_it + ln);
}
if (nb_nodes_to_send != 0) {
AKANTU_DEBUG_INFO("Sending the "
<< nb_nodes_to_send << " nodes to processor " << p
<< " (communication tag : "
<< Tag::genTag(p, 0, SEND_NODES_TAG) << ")");
isend_coordinates_requests.push_back(comm.asyncSend(
nodes_to_send, p, Tag::genTag(my_rank, 0, SEND_NODES_TAG)));
}
#if not defined(AKANTU_NDEBUG)
else {
AKANTU_DEBUG_INFO("No nodes to send to processor " << p);
}
#endif
}
comm.waitAll(isend_nodes_requests);
comm.freeCommunicationRequest(isend_nodes_requests);
nodes.resize(nb_total_nodes_to_recv + nb_nodes);
for (UInt p = 0; p < nb_proc; ++p) {
if ((p != my_rank) && (nb_nodes_to_recv(p) > 0)) {
AKANTU_DEBUG_INFO("Receiving the "
<< nb_nodes_to_recv(p) << " nodes from processor " << p
<< " (communication tag : "
<< Tag::genTag(p, 0, SEND_NODES_TAG) << ")");
Vector<Real> nodes_to_recv(nodes.storage() + nb_nodes * spatial_dimension,
nb_nodes_to_recv(p) * spatial_dimension);
comm.receive(nodes_to_recv, p, Tag::genTag(p, 0, SEND_NODES_TAG));
nb_nodes += nb_nodes_to_recv(p);
}
#if not defined(AKANTU_NDEBUG)
else {
if (p != my_rank) {
AKANTU_DEBUG_INFO("No nodes to receive from processor " << p);
}
}
#endif
}
comm.waitAll(isend_coordinates_requests);
comm.freeCommunicationRequest(isend_coordinates_requests);
mesh.sendEvent(new_nodes);
mesh.sendEvent(new_elements);
AKANTU_DEBUG_OUT();
}
/* -------------------------------------------------------------------------- */
template void GridSynchronizer::createGridSynchronizer<IntegrationPoint>(
const SpatialGrid<IntegrationPoint> & grid);
template void GridSynchronizer::createGridSynchronizer<Element>(
const SpatialGrid<Element> & grid);
} // namespace akantu
c4science · Help