The original STREAM benchmark was developed by John D. McCalpin and is copyright
1991-2005. 

This version was adapted by Lars Bergstrom in 2011-2012 with much feedback from
Twitter, Metarstation, and a variety of other end users.

First, you will need to install libnuma on your machine. This library is not suppored on
OSX, but is available via any package manager on Linux or can be locally installed from:
http://oss.sgi.com/projects/libnuma/

Then, set the number of threads equal to the number of cores on your machine:
export OMP_NUM_THREADS=48

Finally, compile using GCC:
gcc -O3 -std=c99 -fopenmp -lnuma -DN=80000000 -DNTIMES=100 stream.c -o stream-gcc

Variants:
- To simulate non-NUMA aware access, define the constant: -DNON_NUMA
This constant will cause the memory to be touched by a thread on one node and then
for all of the stream operations to happen from another node.
- To show what happens when you perform only accesses that do not coincide with
a prior cache block, compile with -DSTRIDE=8 (assuming 64-byte cache lines and 8-byte
doubles).