- supports multiple domain size (matrix size)
- computes average execution time (by running the program N times)
- computes speedup
- computes strong scaling
- computes weak scaling
- saves computation results into CSV files (easy to plot!)
- verbose logging
Run ./util/hpc-perf-evaluation.py for OpenMP source codes.
Run ./util/hpc-perf-evaluation-cuda.py for CUDA source codes.
Results will be available in the data folder as CSV files.
Made for problems involving matrixes, but built for general-purpose uses.
Program timings must be managed and printed by the program itself, since a wrapper script like that one cannot separately consider the serial and the parallel portions of the execution time.