CUDA/Ada
========

CUDA/Ada is an Ada binding to NVIDIA's CUDA parallel computing platform and
programming model. This project was developed during the course of the master
seminar "Progam Analysis and Transformation" at the University of Applied
Sciences Rapperswil.


Licence
-------
--------------------------------------------------------------------------------
Copyright (C) 2011 Reto Buerki <reet@codelabs.ch>
Copyright (C) 2011 Adrian-Ken Rueegsegger <ken@codelabs.ch>
University of Applied Sciences Rapperswil

This program is free software: you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free Software
Foundation, either version 3 of the License, or (at your option) any later
version.
--------------------------------------------------------------------------------


Documentation
-------------
The paper about CUDA/Ada can be found here:
http://www.codelabs.ch/cuda-ada/cuda-ada-article.pdf.


Download
--------
The current development version of CUDA/Ada is available through its git
repository:

  $ git clone http://git.codelabs.ch/git/cuda-ada.git

A browsable version of the repository is also available here:
http://git.codelabs.ch/?p=cuda-ada.git.


Build
-----
To compile CUDA/Ada on your system, you need to have the following software
installed:

* GNAT compiler:
  http://www.gnu.org/software/gnat/gnat.html

* Ahven (Test-Framework):
  http://ahven.stronglytyped.org/

* CUDA Toolkit for Linux 10.10:
  http://developer.nvidia.com/cuda-downloads

* NVIDIA Developer Drivers for Linux:
  http://developer.nvidia.com/cuda-downloads


Testing
-------
CUDA/Ada contains an unit test suite which can be run by entering the following
command:

  $ make tests

Of course, you need a CUDA aware GPU for this to work.


Benchmarks
----------
CUDA/Ada provides benchmarking code that measures matrix addition and
multiplication in Ada, CUDA/Ada and native CUDA C. The benchmarks can be run by
issuing the following command:

  $ make perf COUNT=10

This will print the cumulated execution times of ten successive matrix
operations for the three different implementations.