A bit misleading "run your binary for at least a minute or more" wording
Closed this issue · 1 comments
Hi!
In WGUNDERWOOD/tex-fmt#22 (comment) we found an issue that people are running their PGO training workloads according to the recommendations from the cargo-pgo's README file "for at least a minute". I completely understand your intention why did you write such a recommendation in the README file? However, as we see, this recommendation can be misleading for people - such a recommendation can be at least useless in some cases.
Maybe we can think about changing the wording a bit to something else. Honestly, I don't know what to write instead since choosing the proper PGO training workload is more like an art than a simple step-by-step algorithm. Or we can just remove it at all from the README file :)
It's true that the minimum duration recommendation is more applicable when sampling is used, however cargo-pgo uses instrumentation, so that's a bit different indeed. I guess that the main point is to exercise all the important parts of the codebase (in the coverage sense).
Feel free to send a PR that rewords this.