nikic wrote: > Update: it seems to be worse after all, which is unexpected, since it > achieved `~1.5%` faster lexing locally. > > The reason I opened this PR in the first place is because the cycles count > was reduced in almost all averages on > [llvm-compile-time-tracker](https://llvm-compile-time-tracker.com/?config=Overview&stat=cycles&remote=Thibault-Monnier) > (see `cf26169083` commit overview). However, I do admit that the results > look quite noisy, so I'll wait for someone with more experience to give their > opinion. > > EDIT: Seeing other experiments I made, it seems the `cycles` metric is too > noisy for any meaningful comparison anyway.
Yes, that's why there is the red warning at the top, though it's only present on the commit-to-commit comparison page: > Warning: The cycles metric is very noisy and likely not meaningful for > comparisons between specific revisions. Statistically significant changes are indicated with red/green coloring. https://github.com/llvm/llvm-project/pull/171052 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
