Performance analysis connects measured simulation cost with an explanation of where time is spent. The project separates benchmark samples, profiling, and report generation:
$ cp config.example config $ make bench $ make profile $ make report
Edit config before running the sweep. Keep grid size, steps, precision, boundary, and snapshot frequency fixed when comparing execution backends or variants.
Benchmarks record timing and throughput. Profiles investigate expensive operations, while the traffic model estimates effective memory bandwidth. make report combines the selected measurements under results/reports/, with PDF output when available.
Small grids can expose launch or barrier overhead; large grids can pressure memory. Measurements identify the change, while profiles and counters help determine its cause.