Syr2k Optimizations in Chapel and C

Note: There are plenty of temp files and messy things; this repo has yet to be edited for presentation.

This page is a supplement to my undergraduate thesis, which is also included in this repository in thesis.pdf.

In this folder we have C and Chapel codes that are various optimizations of the dense matrix codes syr2k from PolyBench 4.0. We also have several recorded timing results.

The official results used in my thesis are in the official-thesis folder.

Note:

• Keller is the name of our research machine with two E5-2620 chips
• Goldwasser is the name of our intel core i5-6500 machine

Running The Codes

$ python3 run-timing.py filename-you-want

Play around with the configuration at the top to customize which tests run and how many times.

Getting The Tools

The results will (obviously) vary by architecture. Hopefully the results we obtained on our specific architectures are reproducible.

The instructions below work for Ubuntu 18.04 Bionic Beaver.

Getting the right C compiler

Hopefully the standard gcc works. If not, dig through the Makefile and see what tools you need.

Getting Chapel Working

To get chapel working you need to install docker.

1. Install docker from here. You should follow the 'install via repository section'.

2. Let docker run without root permissions: see here

Checking Correctness

For C programs:

1. Edit the Makefile by reading the comments in there.

2. Do this:

$ cp c-codes/syr2k.backup.c syr2k.c
$ make syr2k PLY_SIZE=-DMEDIUM_DATASET 
$ ./syr2k 2> original_output_on_med
$ cp c-codes/syr2k.somethingelse.c syr2k.c
$ make syr2k PLY_SIZE=-DMEDIUM_DATASET 
$ ./syr2k 2> somethingelse_output_on_med
$ diff original_output_on_med somethingelse_output_on_med

Note that we could replace MEDIUM with SMALL, MEDIUM, LARGE, HAVERLARGE to get further checks.

For Chapel programs:

1. Uncomment the print statement at the bottom of the code.
2. Compare the values with the output from the original syr2k Chapel version, which is correct.