Cachegrind can be used to get consistent performance results for your benchmarks, allowing you to for example run your benchmarks in noisy VMs or CI runners.
This repository includes a script to help you generate a single combined performance metric, and get more consistency. It is designed to run on Linux.
See the code for usage.
- Switched to parsing the Cachegrind output file, rather than parsing the stderr output.
- Fixed a bug that inflated L3 hit counts by double-counting RAM hits as L3 hits.
- Changed cost multiplier for RAM from 30 to 35 to account for the changes from the bug fix.