Geometric mean of perf tests
Recent patch !7051 (closed) was an attempt to get a tiny perf improvement. (You don't have to look there.) Its output is captured in https://gitlab.haskell.org/ghc/ghc/-/jobs/864936#L6180, which will link you to the allocation changes. You will see that there seems to be a small net benefit -- just what I was hoping for. But it would be nice to know for sure that this is a net benefit, as might be observed by having the geometric mean of a class of perf tests at the bottom.
In today's call, @bgamari thought this would be relatively easy to add to our CI magic.