Add Benchmarks

[jbush001]

Umbrella bug for porting open source parallel benchmarks to compare with other architectures or evaluate tradeoffs. Many existing benchmarks may not be practical or appropriate, but may serve as a starting point for a simpler one:

• They will probably need to be scaled down to run in a bounded amount of time in Verilog simulation.
• They may need modification to explicitly take advantage of Nyuzi resources (vectorization & multithreading)

NAS Parallel Benchmarks (https://www.nas.nasa.gov/publications/npb.html) Based on computational fluid dynamics problems. These are available in C and use MPI and OpenMP communication libraries. These expect 64 bit arithmetic, which makes them unsuitable as-is. These are relatively small snippets of real world problems.

PARSEC (http://parsec.cs.princeton.edu/overview.htm). Contains 13 programs which simulate full real-world workloads. These have a bunch of library dependencies like GLib, GNU Scientific Library, and Mesa 3D, which would make porting directly a large project.

LINPACK/HPL (http://www.netlib.org/benchmark/hpl/) Heavily floating point focused. The lack of a full floating point library and double precision arithmetic may pose problems for using directly.

[jbush001]

This document gives an overview of HPC benchmarks: https://apps.dtic.mil/dtic/tr/fulltext/u2/a612837.pdf

[jbush001]

OpenDwarfs: https://github.com/vtsynergy/OpenDwarfs
SHOC: https://github.com/vetter/shoc
Rodinia: http://lava.cs.virginia.edu/Rodinia/download_links.htm