Benchmarking with Cachegrind

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.

Changelog

Dec 22, 2020

• 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.