|
core 1.5.0
Shared MPI block decomposition, asynchronous ghost-layer exchange, SDF geometry, load balancing (header-only C++20)
|
The collectives of a distributed run. More...
#include <anderson.hpp>
Public Attributes | |
| std::function< void(double *data, int count)> | sumAll |
| std::function< void(void *data, std::size_t bytes)> | broadcast |
| int | rank = 0 |
| this rank in the communicator; rank 0's decisions win (design §6.1) | |
The collectives of a distributed run.
Both callables empty = a single rank (serial, or np = 1 without them). sumAll(data, n) replaces data[0..n) by its global sum on every rank (MPI_Allreduce SUM); broadcast(data, bytes) copies rank 0's bytes to every rank (MPI_Bcast). solver/anderson_mpi.hpp builds one from an MPI_Comm.
Definition at line 44 of file anderson.hpp.
Definition at line 45 of file anderson.hpp.
Referenced by peclet::core::solver::andersonComm().
Definition at line 46 of file anderson.hpp.
Referenced by peclet::core::solver::andersonComm().
| int peclet::core::solver::AndersonComm::rank = 0 |
this rank in the communicator; rank 0's decisions win (design §6.1)
Definition at line 47 of file anderson.hpp.
Referenced by peclet::core::solver::andersonComm(), and peclet::core::solver::AndersonCore::stepFailed().