This repository contains the benchmark code supplementing my blog post on a matrix-matrix multiplication benchmark on Intel Xeon and Xeon Phi.
The following instructions are for *nix-based systems. The code should also work on Windows, but is not tested.
First run the script
$> ./init.sh
to download and extract the sparse matrices. This will pull about 350MB of data and thus may take a while.
Second, run the executables via
$> ./run_mkl_xeon
$> ./run_mkl_knl 0 # for DDR4
$> ./run_mkl_knl 1 # for MCDRAM
$> ./run_mkl_knc
This assumes that your Intel compiler variables are set accordingly. Note that for the Xeon Phi KNC you have to cross-compile on the host and run the script on the device.
The folder results/ contains the results I obtained on different systems. Feel free to paste your results in there in replot the graphs via
$> Rscript plot.R
The free open source statistics package R is required for the plots.
The code is provided under a permissive MIT/X11-style license. See file LICENSE.txt for details.