我在 2 台不同的机器上运行一个程序。在一个它工作正常没有问题。另一方面，它会导致分段错误。通过调试，我已经找出了故障发生的位置，但我无法找出它发生的逻辑原因。
在一个函数中，我有以下代码：
pass_particles(particle_grid, particle_properties, input_data, coll_eros_track, collision_number_part, world, grid_rank_lookup, grid_locations);
cout<<"done passing particles"<<endl;
函数pass_particles如下所示：
void pass_particles(map<int,map<int,Particle> > & particle_grid, std::vector<Particle_props> & particle_properties, User_input& input_data, data_tracking & coll_eros_track, vector<int> & collision_number_part, mpi::communicator & world, std::map<int,int> & grid_rank_lookup, map<int,std::vector<double> > & grid_locations)
{
//cout<<"east-west"<<endl;
//east-west exchange (x direction)
map<int, vector<Particle> > particles_to_be_sent_east;
map<int, vector<Particle> > particles_to_be_sent_west;
vector<Particle> particles_received_east;
vector<Particle> particles_received_west;
int counter_x_sent=0;
int counter_x_received=0;
for(grid_iter=particle_grid.begin();grid_iter!=particle_grid.end();grid_iter++)
{
map<int,Particle>::iterator part_iter;
for (part_iter=grid_iter->second.begin();part_iter!=grid_iter->second.end();)
{
if (particle_properties[part_iter->second.global_part_num()].particle_in_box()[grid_iter->first])
{
//decide if a particle has left the box...need to consider whether particle was already outside the box
if ((part_iter->second.position().x()<(grid_locations[grid_iter->first][0]) && part_iter->second.position().x()>(grid_locations[grid_iter->first-input_data.z_numboxes()][0]))
|| (input_data.periodic_walls_x() && (grid_iter->first-floor(grid_iter->first/(input_data.xz_numboxes()))*input_data.xz_numboxes()<input_data.z_numboxes()) && (part_iter->second.position().x()>(grid_locations[input_data.total_boxes()-1][0]))))
{
particles_to_be_sent_west[grid_iter->first].push_back(part_iter->second);
particle_properties[particle_grid[grid_iter->first][part_iter->first].global_part_num()].particle_in_box()[grid_iter->first]=false;
counter_sent++;
counter_x_sent++;
}
else if ((part_iter->second.position().x()>(grid_locations[grid_iter->first][1]) && part_iter->second.position().x()<(grid_locations[grid_iter->first+input_data.z_numboxes()][1]))
|| (input_data.periodic_walls_x() && (grid_iter->first-floor(grid_iter->first/(input_data.xz_numboxes()))*input_data.xz_numboxes())>input_data.xz_numboxes()-input_data.z_numboxes()-1) && (part_iter->second.position().x()<(grid_locations[0][1])))
{
particles_to_be_sent_east[grid_iter->first].push_back(part_iter->second);
particle_properties[particle_grid[grid_iter->first][part_iter->first].global_part_num()].particle_in_box()[grid_iter->first]=false;
counter_sent++;
counter_x_sent++;
}
//select particles in overlap areas to send to neighboring cells
else if ((part_iter->second.position().x()>(grid_locations[grid_iter->first][0]) && part_iter->second.position().x()<(grid_locations[grid_iter->first][0]+input_data.diam_large())))
{
particles_to_be_sent_west[grid_iter->first].push_back(part_iter->second);
counter_sent++;
counter_x_sent++;
}
else if ((part_iter->second.position().x()<(grid_locations[grid_iter->first][1]) && part_iter->second.position().x()>(grid_locations[grid_iter->first][1]-input_data.diam_large())))
{
particles_to_be_sent_east[grid_iter->first].push_back(part_iter->second);
counter_sent++;
counter_x_sent++;
}
++part_iter;
}
else if (particles_received_current[grid_iter->first].find(part_iter->first)!=particles_received_current[grid_iter->first].end())
{
if ((part_iter->second.position().x()>(grid_locations[grid_iter->first][0]) && part_iter->second.position().x()<(grid_locations[grid_iter->first][0]+input_data.diam_large())))
{
particles_to_be_sent_west[grid_iter->first].push_back(part_iter->second);
counter_sent++;
counter_x_sent++;
}
else if ((part_iter->second.position().x()<(grid_locations[grid_iter->first][1]) && part_iter->second.position().x()>(grid_locations[grid_iter->first][1]-input_data.diam_large())))
{
particles_to_be_sent_east[grid_iter->first].push_back(part_iter->second);
counter_sent++;
counter_x_sent++;
}
part_iter++;
}
else
{
particle_grid[grid_iter->first].erase(part_iter++);
counter_removed++;
}
}
}
world.barrier();
mpi::request reqs_x_send[particles_to_be_sent_west.size()+particles_to_be_sent_east.size()];
vector<multimap<int,int> > box_sent_x_info;
box_sent_x_info.resize(world.size());
vector<multimap<int,int> > box_received_x_info;
box_received_x_info.resize(world.size());
int counter_x_reqs=0;
//send particles
for(grid_iter_vec=particles_to_be_sent_west.begin();grid_iter_vec!=particles_to_be_sent_west.end();grid_iter_vec++)
{
if (grid_iter_vec->second.size()!=0)
{
//send a particle. 50 will be "west" tag
if (input_data.periodic_walls_x() && (grid_iter_vec->first-floor(grid_iter_vec->first/(input_data.xz_numboxes()))*input_data.xz_numboxes()<input_data.z_numboxes()))
{
reqs_x_send[counter_x_reqs++]=world.isend(grid_rank_lookup[grid_iter_vec->first + input_data.z_numboxes()*(input_data.x_numboxes()-1)], grid_iter_vec->first + input_data.z_numboxes()*(input_data.x_numboxes()-1), particles_to_be_sent_west[grid_iter_vec->first]);
box_sent_x_info[grid_rank_lookup[grid_iter_vec->first + input_data.z_numboxes()*(input_data.x_numboxes()-1)]].insert(pair<int,int>(world.rank(), grid_iter_vec->first + input_data.z_numboxes()*(input_data.x_numboxes()-1)));
}
else if (!(grid_iter_vec->first-floor(grid_iter_vec->first/(input_data.xz_numboxes()))*input_data.xz_numboxes()<input_data.z_numboxes()))
{
reqs_x_send[counter_x_reqs++]=world.isend(grid_rank_lookup[grid_iter_vec->first - input_data.z_numboxes()], grid_iter_vec->first - input_data.z_numboxes(), particles_to_be_sent_west[grid_iter_vec->first]);
box_sent_x_info[grid_rank_lookup[grid_iter_vec->first - input_data.z_numboxes()]].insert(pair<int,int>(world.rank(),grid_iter_vec->first - input_data.z_numboxes()));
}
}
}
for(grid_iter_vec=particles_to_be_sent_east.begin();grid_iter_vec!=particles_to_be_sent_east.end();grid_iter_vec++)
{
if (grid_iter_vec->second.size()!=0)
{
//send a particle. 60 will be "east" tag
if (input_data.periodic_walls_x() && (grid_iter_vec->first-floor(grid_iter_vec->first/(input_data.xz_numboxes())*input_data.xz_numboxes())>input_data.xz_numboxes()-input_data.z_numboxes()-1))
{
reqs_x_send[counter_x_reqs++]=world.isend(grid_rank_lookup[grid_iter_vec->first - input_data.z_numboxes()*(input_data.x_numboxes()-1)], 2000000000-(grid_iter_vec->first - input_data.z_numboxes()*(input_data.x_numboxes()-1)), particles_to_be_sent_east[grid_iter_vec->first]);
box_sent_x_info[grid_rank_lookup[grid_iter_vec->first - input_data.z_numboxes()*(input_data.x_numboxes()-1)]].insert(pair<int,int>(world.rank(),2000000000-(grid_iter_vec->first - input_data.z_numboxes()*(input_data.x_numboxes()-1))));
}
else if (!(grid_iter_vec->first-floor(grid_iter_vec->first/(input_data.xz_numboxes())*input_data.xz_numboxes())>input_data.xz_numboxes()-input_data.z_numboxes()-1))
{
reqs_x_send[counter_x_reqs++]=world.isend(grid_rank_lookup[grid_iter_vec->first + input_data.z_numboxes()], 2000000000-(grid_iter_vec->first + input_data.z_numboxes()), particles_to_be_sent_east[grid_iter_vec->first]);
box_sent_x_info[grid_rank_lookup[grid_iter_vec->first + input_data.z_numboxes()]].insert(pair<int,int>(world.rank(), 2000000000-(grid_iter_vec->first + input_data.z_numboxes())));
}
}
}
counter=0;
for (int i=0;i<world.size();i++)
{
//if (world.rank()!=i)
//{
reqs[counter++]=world.isend(i,1000000000,box_sent_x_info[i]);
reqs[counter++]=world.irecv(i,1000000000,box_received_x_info[i]);
//}
}
mpi::wait_all(reqs, reqs + world.size()*2);
//receive particles
//receive west particles
for (int j=0;j<world.size();j++)
{
multimap<int,int>::iterator received_info_iter;
for (received_info_iter=box_received_x_info[j].begin();received_info_iter!=box_received_x_info[j].end();received_info_iter++)
{
//receive the message
if (received_info_iter->second<1000000000)
{
//receive the message
world.recv(received_info_iter->first,received_info_iter->second,particles_received_west);
//loop through all the received particles and add them to the particle_grid for this processor
for (unsigned int i=0;i<particles_received_west.size();i++)
{
particle_grid[received_info_iter->second].insert(pair<int,Particle>(particles_received_west[i].global_part_num(),particles_received_west[i]));
if(particles_received_west[i].position().x()>grid_locations[received_info_iter->second][0] && particles_received_west[i].position().x()<grid_locations[received_info_iter->second][1])
{
particle_properties[particles_received_west[i].global_part_num()].particle_in_box()[received_info_iter->second]=true;
}
counter_received++;
counter_x_received++;
}
}
else
{
//receive the message
world.recv(received_info_iter->first,received_info_iter->second,particles_received_east);
//loop through all the received particles and add them to the particle_grid for this processor
for (unsigned int i=0;i<particles_received_east.size();i++)
{
particle_grid[2000000000-received_info_iter->second].insert(pair<int,Particle>(particles_received_east[i].global_part_num(),particles_received_east[i]));
if(particles_received_east[i].position().x()>grid_locations[2000000000-received_info_iter->second][0] && particles_received_east[i].position().x()<grid_locations[2000000000-received_info_iter->second][1])
{
particle_properties[particles_received_east[i].global_part_num()].particle_in_box()[2000000000-received_info_iter->second]=true;
}
counter_received++;
counter_x_received++;
}
}
}
}
mpi::wait_all(reqs_y_send, reqs_y_send + particles_to_be_sent_bottom.size()+particles_to_be_sent_top.size());
mpi::wait_all(reqs_z_send, reqs_z_send + particles_to_be_sent_south.size()+particles_to_be_sent_north.size());
mpi::wait_all(reqs_x_send, reqs_x_send + particles_to_be_sent_west.size()+particles_to_be_sent_east.size());
cout<<"x sent "<<counter_x_sent<<" and received "<<counter_x_received<<" from rank "<<world.rank()<<endl;
cout<<"rank "<<world.rank()<<" sent "<<counter_sent<<" and received "<<counter_received<<" and removed "<<counter_removed<<endl;
cout<<"done passing"<<endl;
}
我只发布了一些代码（所以忽略一些变量可能看起来未定义的事实，因为它们在我没有发布的代码的一部分中）
当我运行代码（在它失败的机器上）时，我得到done passing但不是done passing particles
我不知道什么可能导致被调用函数的末尾和调用函数的下一行之间的分段错误，以及为什么它会发生在一台机器上而不是另一台机器上。
如果在函数的末尾和调用方的后续行之间崩溃，则可能会在局部变量的析构函数中崩溃。 您需要在调试器中运行程序，以找出哪个对象的析构函数正在崩溃。
有几种可能性：
- 您实际上正在返回，但 cout 由操作系统缓冲，因此您看不到"完成传递粒子"，因为应用程序首先崩溃。
- 您有一些本地类，该类具有一个析构函数，该析构函数是 seg 错误。
尝试在调试器中运行它，以找出它实际崩溃的位置。
编辑：
既然你已经提到你正在使用 gcc，请添加 -g 标志并使用 gdb 运行它。然后 Gdb 会准确地告诉你哪里出了问题（可能是空取消引用）。
以防万
一以后有人回到这个问题。我更新到最新版本的boost mpi（当时），1.50，这个问题消失了。没有太多的解决方案，但它奏效了。