CoolerSpace Benchmarker

This is the benchmarker for the CoolerSpace library. The CoolerSpace GitHub repository can be found here.

Image Credits

We have used several Unsplash photographs to benchmark the efficacy of CoolerSpace. Please refer to this file for links to the original photographs and credits.

What's included in this repository?

We have included 4 different benchmarker programs in this repository.

1. benchmarker.py: This program benchmarks compilation time, optimization time, and runtime of CoolerSpace on several programs in the programs directory. Test files are output to the results directory.
2. numba_programs/benchmark.py: This program benchmarks equivalent programs written in numba. Generates the numba.csv results file.
3. cupy_programs/benchmark.py: This program benchmarks equivalent programs written in CuPy. Generates the cupy.csv results file.
4. colourscience_programs/benchmark.py: This program benchmarks equivalent programs written in ColourScience. Generates the cs.csv results file.

The cases variable in each of these benchmark programs can be modified to choose which programs to benchmark at what resolutions. The TRIAL_COUNT variable is used to set the number of trials that are run for every test case.

Dependencies

Nvidia CUDA dependencies are required depending on which version of ONNX Runtime you have installed. Please refer to this website to look up which version of CUDA and cuDNN you need to run benchmarker.py.