This tool consists of three executables:

run_benchmarks.py

Pofiles the executable passed as an argument and puts the output into files. Additionally the programming model, GPU vendor and subdirectory where the results should be stored have to specified

Example

python3 run_benchmarks.py --gpu nvidia --model cuda -o measurements_cuda_triad <cuda_triad_executable>

create_profiles.py

Processes the collected metrics and stores them into files for each kernel seperatly.

Example

python3 create_profiles.py --gpu nvidia --model cuda -o results_cuda_triad <directory with profiles creater by runbenchmarks.py>

signature.py

Pass an arbitrary amount of files created by create_profiles.py. Creates signature plots from that.

Example usage

python signature.py --name <name of plot> <base-kernel-file> <arbitrarily many kernel files>

This will create a plot with the specified name with all outputs realative to the first passed signature file. Additionally thread/team counts will be printed to the terminal.