garybernhardt

This tool is used to compare microbenchmarks across two versions of code. It's
paranoid about nulling out timing error, so the numbers should be meaningful.
It runs the benchmarks many times, scaling the iterations up if the benchmark
is extremely short, and it nulls out its own timing overhead while doing so. It
reports results graphically with a text interface in the terminal.

You first run it with --record, which generates a JSON dotfile with runtimes
for each of your benchmarks. Then you change the code and run again with
--compare, which re-runs and generates comparison plots between your recorded
and current times. In the example output, I did a --record on the master