The purpose of this project is to study the common serial implementations of matrix multiplication and to study the performance benefits of parallel and approximate solutions. Matrix multiplication is a very well defined problem, the results can be easily confirmed and benchmarked using a set of defined matrices for all experiments.

The project utilizes the programming language Python and uses Numpy to perform matrix operations. The software matrix multiplication algorithms developed consist of parallel implementations of matrix multiplication algorithms utilizing the CPU and GPU and approximate algorithms for performing matrix multiplications generated using evolutionary strategies and evaluated within a tolerance of error.

The following is a summary of the various source codes in the repository.

codes/

• This contains the main code for the project, the soure files setup.py and test_framework.py contain the code for installing the dependencies and executing the test framework to evaluate the benchmark impementations against the tests.
• The testplans are YAML documents that end in .yml and contain the descriptions of the test plans to execute, which describes how the benchmark codes are evaluated.

benchmarks/

• This contains each of the matrix multiplication benchmark implementations these include the naive, baseline, parallel, and approximate implementations.
• The helpers.py source file contains many useful helper functions which are common to each of the benchmark implemetations.

analysis/

• This contains the R code that was used to analyze all of the test results and generate the tables and plots used for the final report.

reports/

• Each of the reports for the project including the final report, use the provided Makefile to generate the PDF reports from the LaTeX source files.

1. In order to build the reports containing the results of this project please execute the Makefile in the reports/ folder, by default this will generate all of the progress reports for this project.

2. In order to execute the benchmarks and evaluate the various implementations of matrix multiplication you will need to execute the test framework in the codes/ folder. The test framework follows the instructions outlined in the testplan.yml document which specifies the tests to execute against the benchmark codes.

3. In order to execute the test framework you will need to install the depenencies. Navigate to the codes/ folder and then execute the setup script which the following command, this will install all needed dependencies.

   sudo python2 setup.py develop

4. After installing the dependencies you can execute the test framework using the following command, if you want verbosity you can add the argument --verbose. As well, if you would like to save the results to a CSV file the argument --save= can be used. Lastly, it is recommended that you execute the simple test plan which requires much less time to complete than the complete test plan.

   python2 test_framework.py --benchmarks=benchmarks simpleplan.yml

5. If you would like to view the various matrix multiplication benchmarks you can see the implementations in the benchmarks/ directory.

All of the source code in this repository, is licensed under the GNU General Public License, Version 3