The examples below use the threaded MPI model for the Adapteva Epiphany architecture, specifically the Parallella platform. All applications are distributed across the 16 Epiphany cores.

The codes were developed by the US Army Research Laboratory, Computational Sciences Division, Computing Architectures Branch.

The examples were tested on the 2015 Parallella image.

In order to build and test this example, at present you need to download and install a preview release of the libcoprthr_mpi library that has been made to work with the COPRTHR SDK version 1.6.1. The preview library is presently free to download and use for experimentation (non-commercial use).

-JAR Updated 6/11/15

This software demonstrates a high performance implementation of Cannon's Algorithm for matrix multiplication for two-dimensional network meshes. The communication is blocking and uses the MPI_Sendrecv_replace routine for communication between cores.

View ./cannon source code or the README.

This demonstrates a 2D FFT with a 2D domain decomposition and uses the MPI_Sendrecv_replace for communication

View ./fft2d source code.

This example demonstrates a high performance gravitational particle interaction (N-body simulation) using MPI_Sendrecv_replace for convenience.

View ./nbody source code or the README.

This is a high performance example of a finite difference scheme using a 5-point stencil update where a solution is iterated until convergence and boundary data is communicated between neighboring cores over the two-dimensional domain decomposition.

View ./stencil source code.