HPCmatlab software framework

About HPCmatlab

HPCmatlab is a framework for fast prototyping of parallel applications on Matlab. It uses the Message Passing Interface (MPI) for sending and receiving messages among different Matlab processes. The users can use mex functions with syntax similar to the MPI standard syntax.

This version of HPCmatlab is compatible with Matlab versions up to R2015a only. It is not compatible with Matlab versions R2015b onward. We are currently adding support for newer Matlab versions.

Version 1.0

This version includes our MPI functions.

Documentation

The User Guide for this version is included in form of the PDF file "HPCmatlab_1.0_User_Guide.pdf" under doc/

For users new to parallel programming, we recommend trying out the simple send-receive examples first to get started with the framework.

For details on implementation and performance results, refer our conference paper

URL: http://dx.doi.org/10.1016/j.procs.2016.05.467

A PDF copy of this paper can also be found under doc/. ([CC BY-NC-ND 4.0] (https://creativecommons.org/licenses/by-nc-nd/4.0/ "CC BY-NC-ND 4.0"))

Examples

The examples of simple programs using point-to-point, collective and one-sided communications are included under examples/ directory. They also have sample job submission files for SLURM.

Installation and Use

GitHub download page: https://github.com/xinchenguo/HPCmatlab

1. Make sure Matlab's mex compiler is in your $PATH. Also, mpicc and mpicxx should be in $PATH with all environment variables set up for MPI libraries.

2. Run make to compile the MEX functions from source code. The MEX functions will be placed under matlab/ directory.

3. There are two ways to add the path to these functions to Matlab's search path:

• Before running your script, set the MATLABPATH environment variable, e.g.

export MATLABPATH=/path/to/HPCmatlab/matlab:$MATLABPATH

• At the start of your Matlab script, use Matlab's addpath function. Refer examples for how to do this.

4. Use the default MPI command (mpirun,mpiexec,srun, etc.) of a cluster to start MATLAB with HPCmatlab framework.
   • mpirun -n 2 matlab -r "test,exit"
   • Refer to your cluster's user guide or consult your system admin for the proper MPI command

Contact

Use our GitHub page for questions, bug reports, comments, feedback and suggestions:

https://github.com/xinchenguo/HPCmatlab/issues

Citation

@article{Guo20161461,
title = "HPCmatlab: A Framework for Fast Prototyping of Parallel Applications in Matlab ",
journal = "Procedia Computer Science ",
volume = "80",
number = "",
pages = "1461 - 1472",
year = "2016",
note = "International Conference on Computational Science 2016, \{ICCS\} 2016, 6-8 June 2016, San Diego, California, \{USA\} ",
issn = "1877-0509",
doi = "http://dx.doi.org/10.1016/j.procs.2016.05.467",
url = "http://www.sciencedirect.com/science/article/pii/S1877050916309516",
author = "Xinchen Guo and Mukul Dave and Mohamed Sayeed",
keywords = "Parallel Programming",
keywords = "Message Passing Interface",
keywords = "Matlab",
keywords = "MEX Functions",
keywords = "Parallel I/O "
}