mpl

KO | EN

gitlite — search

#react #llm #go #machine-learning #deep-learning #docker #nodejs #java #windows #kubernetes #awesome-list #pytorch #ios #openai

(★ 186)

A C++17 message passing library based on MPI

File Explorer

Download Latest Version (.zip)

- - build-with-IntelMPI.yml
  - build-with-mpich.yml
  - build-with-openmpi.yml
- mplConfig.cmake
- PVS-Studio.cmake
- PVS-Studio.cmake.LICENSE
- - data_types.md
  - Doxyfile.in
  - README.md
  - style.css
- - auxiliary.rst
  - cmake_integration.rst
  - communicator.rst
  - conf.py.in
  - constants.rst
  - data_types.rst
  - design.rst
  - environmental_management.rst
  - error.rst
  - file.rst
  - file_error_handling.cc
  - grid.rst
  - group.rst
  - index.rst
  - info.rst
  - installation.rst
  - interoperability.rst
  - layouts.rst
  - reduction_operations.rst
  - tags.rst
- CMakeLists.txt
- - auxiliary.html
  - cmake_integration.html
  - communicator.html
  - constants.html
  - data_types.html
  - design.html
  - environmental_management.html
  - error.html
  - file.html
  - genindex.html
  - grid.html
  - group.html
  - index.html
  - info.html
  - installation.html
  - interoperability.html
  - layouts.html
  - objects.inv
  - reduction_operations.html
  - search.html
  - searchindex.js
  - tags.html
- .nojekyll
- mpl_parallel_2018.pdf
- arrays.cc
- blocking.cc
- blocking_vector.cc
- CMakeLists.txt
- collective.cc
- communicator.cc
- custom_initialization.cc
- distributed_grid.cc
- distributed_grid_scatter_gather.cc
- file.cc
- gather.cc
- gatherv.cc
- heat_equation_Jacobi_method.cc
- heat_equation_successive_over-relaxation.cc
- hello_world.cc
- intercommunicator.cc
- iterators.cc
- layouts.cc
- matrix_gather.cc
- nonblocking.cc
- nonblocking_mult.cc
- parallel_sort_mpi.c
- parallel_sort_mpl.cc
- probe.cc
- process_creation.cc
- process_creation_client.cc
- process_creation_multiple.cc
- reduce_lcm.cc
- reduce_min_loc.cc
- standard_types.cc
- stl_container.cc
- struct.cc
- subarray.cc
- vibrating_string_mpi.c
- vibrating_string_mpl.cc
- cartesian_communicator.hpp
- comm_group.hpp
- command_line.hpp
- datatype.hpp
- displacements.hpp
- distributed_graph_communicator.hpp
- distributed_grid.hpp
- environment.hpp
- error.hpp
- file.hpp
- flat_memory.hpp
- graph_communicator.hpp
- info.hpp
- layout.hpp
- message.hpp
- mpl.hpp
- operator.hpp
- ranks.hpp
- request.hpp
- status.hpp
- tag.hpp
- topology_communicator.hpp
- utility.hpp
- vector.hpp
- CMakeLists.txt
- test_cartesian_communicator.cc
- test_communicator.cc
- test_communicator_allgather.cc
- test_communicator_allgatherv.cc
- test_communicator_allreduce.cc
- test_communicator_alltoall.cc
- test_communicator_alltoallv.cc
- test_communicator_barrier.cc
- test_communicator_bcast.cc
- test_communicator_exscan.cc
- test_communicator_gather.cc
- test_communicator_gatherv.cc
- test_communicator_init_send_init_recv.cc
- test_communicator_isend_irecv.cc
- test_communicator_mprobe_mrecv.cc
- test_communicator_probe.cc
- test_communicator_reduce.cc
- test_communicator_reduce_scatter.cc
- test_communicator_reduce_scatter_block.cc
- test_communicator_scan.cc
- test_communicator_scatter.cc
- test_communicator_scatterv.cc
- test_communicator_send_recv.cc
- test_communicator_sendrecv.cc
- test_displacements.cc
- test_dist_graph_communicator.cc
- test_file.cc
- test_graph_communicator.cc
- test_group.cc
- test_helper.hpp
- test_info.cc
- test_initialization.cc
- test_inter_communicator.cc
- test_mpi_communicator.cc
.clang-format
.clang-tidy
.gitignore
CMakeLists.txt
COPYING
README.md

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# MPL - A message passing library

[![build-with-openmpi](https://github.com/rabauke/mpl/actions/workflows/build-with-openmpi.yml/badge.svg)](https://github.com/rabauke/mpl/actions/workflows/build-with-openmpi.yml)
[![build-with-mpich](https://github.com/rabauke/mpl/actions/workflows/build-with-mpich.yml/badge.svg)](https://github.com/rabauke/mpl/actions/workflows/build-with-mpich.yml)
[![build-with-IntelMPI](https://github.com/rabauke/mpl/actions/workflows/build-with-IntelMPI.yml/badge.svg)](https://github.com/rabauke/mpl/actions/workflows/build-with-IntelMPI.yml)
[![License](https://img.shields.io/badge/License-BSD_3--Clause-blue.svg)](https://opensource.org/licenses/BSD-3-Clause)

MPL is a message passing library written in C++17 based on the
[Message Passing Interface](http://mpi-forum.org/) (MPI) standard. 
Since the C++ API has been dropped from the MPI standard in version 
3.1, it is the aim of MPL to provide a modern C++ message passing 
library for high performance computing.

MPL will neither bring all functions of the C language MPI-API to C++
nor provide a direct mapping of the C API to some C++ functions and
classes. The library's focus lies on the MPI core message passing
functions, ease of use, type safety, and elegance. The aim of MPL is to
provide an idiomatic C++ message passing library without introducing a
significant overhead compared to utilizing MPI via its plain C-API.
This library is most useful for developers who have at least some basic
knowledge of the Message Passing Interface standard and would like to
utilize it via a more user-friendly interface in modern C++. Unlike
[Boost.MPI](https://www.boost.org/doc/libs/1_77_0/doc/html/mpi.html),
MPL does not rely on an external serialization library and has a
negligible run-time overhead.

## Supported features

MPL assumes that the underlying MPI implementation supports the 
[version 3.1](https://www.mpi-forum.org/docs/mpi-3.1/mpi31-report.pdf) 
of the Message Passing Interface standard.  Future versions of MPL 
may also employ features of the new 
[version 4.0](https://www.mpi-forum.org/docs/mpi-4.0/mpi40-report.pdf) 
or later MPI versions.

MPL gives currently access via a convenient C++ interface to the 
following features of the Message Passing Interface standard:

* environmental management (implicit initialization and finalization, timers, but no error handling).
* point-to-point communication (blocking and non-blocking),
* collective communication (blocking and non-blocking),
* derived data types (happens automatically for many custom data types or via the `base_struct_builder` helper class and the layout classes of MPL),
* communicator- and group-management,
* process topologies (cartesian and graph topologies),
* inter-communicators,
* dynamic process creation and
* file i/o.

Currently, the following MPI features are not yet supported by MPL:

* error handling and
* one-sided communication.

Although MPL covers a subset of the MPI functionality only, it has 
probably the largest MPI-feature coverage among all alternative C++ 
interfaces to MPI.

## Hello parallel world

MPL is built on top of the Message Passing Interface (MPI) standard.  Therefore, 
MPL shares many concepts known from the MPI standard, e.g., the concept of a
communicator.  Communicators manage the message exchange between different processes, 
i.e., messages are sent and received with the help of a communicator.

The MPL environment provides a global default communicator `comm_world`, which will 
be used in the following Hello-World program.  The program prints out some information 
about each process:
* its rank, 
* the total number of processes and 
* the computer's name the process is running on.

If there are two or more processes, a message is sent from process 0 to process 1, 
which is also printed.
 
```C++
#include <cstdlib>
#include <iostream>
// include MPL header file
#include <mpl/mpl.hpp>

int main() {
  // get a reference to communicator "world"
  const mpl::communicator &comm_world{mpl::environment::comm_world()};
  // each process prints a message containing the processor name, the rank
  // in communicator world and the size of communicator world
  // output may depend on the underlying MPI implementation
  std::cout << "Hello world! I am running on \"" << mpl::environment::processor_name()
            << "\". My rank is " << comm_world.rank() << " out of " << comm_world.size()
            << " processes.\n";
  // if there are two or more processes send a message from process 0 to process 1
  if (comm_world.size() >= 2) {
    if (comm_world.rank() == 0) {
      std::string message{"Hello world!"};
      comm_world.send(message, 1);  // send message to rank 1
    } else if (comm_world.rank() == 1) {
      std::string message;
      comm_world.recv(message, 0);  // receive message from rank 0
      std::cout << "got: \"" << message << "\"\n";
    }
  }
  return EXIT_SUCCESS;
}
```

## Documentation

For further documentation see the
[documentation](https://rabauke.github.io/mpl/html/), the blog posts

* [MPL – A message passing library](https://www.numbercrunch.de/blog/2015/08/mpl-a-message-passing-library/),
  * [MPL – Collective communication](https://www.numbercrunch.de/blog/2015/09/mpl-collective-communication/),
  * [MPL – Data types](https://www.numbercrunch.de/blog/2015/09/mpl-data-types/),

the presentation

* [Message Passing mit modernem C++](https://rabauke.github.io/mpl/mpl_parallel_2018.pdf) (German only),

the book

* [Parallel Programming for Science and Engineering](https://theartofhpc.com/pcse.html) by Victor Eijkhout, or

the workshop material

* [SMU O'Donnell Data Science and Research Computing Institute Parallel C++ Workshop](https://southernmethodistuniversity.github.io/parallel_cpp/intro.html)

and the files in the `examples` directory of the source package.

mpl

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

mpl

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

Download README