Zulip topic.

Plan for parallel-rustc

  1. Gather the performance data of the feature internally
  2. Make a “go/no-go” decision based on the data

Overall conclusions

The overall conclusions was that before shipping we should improve the documentation. To that end, we are trying to gather major design questions in a hackmd document. Here is a link to the start of a discussion about documentation.

Gathered data

The most important data on our plan are recorded on hackmd.

Experiments on other work-load are also welcome. The data are reported in the following topics on Zulip#t-compiler/wg-parallel-rustc:

We also discussed performance tradeoffs. Towards the end of the meeting, we noted some research on what sorts of compilation time hits are noticebale to users, which is relevant to the “single core overhead” we are willing to tolerate from using locks.