Small Repository Providing a data storing and visualization Tool for the Alpaka 2 Benchmarks (https://github.com/psychocoderHPC/alpaka2)

Dependencies (only for bench_viz):

pip install matplotlib (might have to use --break-system-packages)

for manual local use:

chmod +x ./bin/bench_store
./bin/bench_store --args	
chmod +x ./bin/bench_viz
./bin/bench_viz --args

install:

chmod +x ./install
./install

will add both skripts to your ~/.bashrc and make them executable
you can now use bench_store and bench_viz from the command line

How to use it

Storing data into .json files:

bench_store $dir1 ... $dirN --output-dir $pathToYourOutputDir

note that --output-dir is optional, otherwise the .json will be dumped at the current working directory.

Files inside the directories specified with $dir1 to $dirN should be of the format: $tag_$nr.

Tag is an optional specifier to group data sets and nr is the number of elements used (X-axis argument).

The $nr is necessary whereas files without $tag will be marked with the unspecified tag. Not using a _ will also mark them as unspecified.

Visualizing created .json files:

bench_viz <json_file> --byTag $tag OR --byAcc $substringOfAccelerator OR --byKernel $kernel_name

substrings for tag, accelerator or kernelName are also allowed (if its not ambigious)

example 1:

bench_viz $pathTo/local_float.json --byAcc cpus 

will make plots for the specified Accelerator CpuSerial for every Kernel and Tag

example 2:

bench_viz $pathTo/local_float.json --byKernel dot

will make plots for the dotKernel for every Accelerator and Tag

example 3:

bench_viz $pathTo/local_float.json --byTag hal

will make plots for the specified Tag: hal, for every Kernel and Accelerator

example 4:

bench_viz $pathTo/local_float.json

will output a scatterplot (due to expected data heterogeneity) across all Tags for every kernel and Accelerator (this is highly uncouraged)