ogs6-btl

OGS6 matrix benchmarking copy from Eigen3-BTL

Compilation

It is better to compile the code outside the source tree. In the ogs6-btl/ directory create build directory

mkdir build
cd build

Call cmake to prepare makefiles

cmake ../btl

And, finally, compile benchmarks executing

make

Extra libraries

To create benchmarks for any of the interfaces present in btl/libs/ corresponding libraries has to be found in "usual" system directories like /usr/lib/, or environment variables for specific library location has to be provided to cmake. To create benchmarks for ATLAS libraries installed in $HOME/local/lib/, for example, call cmake as:

ATLASDIR=$HOME/local/lib cmake ../btl

Running benchmarks

To run the benchmarks use ctest: In the build/ directory

ctest -V

See further documentation in btl's README file in the ogs6-btl/btl/ directory.

Postprocessing

After the benchmarks are done the results are written into the build/ directory in libs/<library interface>/bench_.*.dat files. In the results/ directory create another directory corresponding to your cpu (on linux you can find out cpu information with cat /proc/cpuinfo) like [email protected]/. Copy the above bench_.*.dat files there

cp build/libs/*/bench_*.dat results/<your_cpuinfo>/

Visualization

There is an R-script in results/ directory, which reads all benchmarks into a data frame. It provides simple plotting functions using the ggplot2 package for visualisation.

A typical R-session could be:

source('processResults.R')  # Assuming you are in the results/ directory.
bs = createTable(".")       # Use different path pointing to the results.
b = selectMaxMflops(bb)     # Remove duplicate entries.

# Use one of plotXXX functions for plotting, e.g.:
plotAllBenchmarks(b)
plotBenchmarks(b, logy = TRUE)
plotSmallSizeBenchmarks(b)