https://gcc.gnu.org/bugzilla/show_bug.cgi?id=30920
--- Comment #2 from Eric Gallager <egallager at gcc dot gnu.org> --- (In reply to Ian Blanes from comment #0) > The current splaying implementation does a bottom up splaying but starting > at the root (should do a top down). It works but performs worst. One example > of this is that while in the implementation described in > http://www.cs.cmu.edu/~sleator/papers/self-adjusting.pdf the root node ends > at most two levels down, in the current implementation it can end at an > arbiratry depth (example in the mailing list). So the property of keeping > recently used nodes up doesn't work. > > Interestingly it doesn't hurt performance so much. Aactually it improves the > performance compared to the previous version (previous bottom up > implementation) because of the recursive nature of the former one. > > (patch and in depth explanation here) > http://gcc.gnu.org/ml/gcc/2006-11/msg00074.html > http://gcc.gnu.org/ml/gcc/2006-07/msg00210.html Linked threads point to this other thread: https://gcc.gnu.org/ml/gcc-patches/2005-10/msg00833.html It says that Brian Makin is the original author, but I'm not seeing his email address as cc-able here on Bugzilla... does he have a new one?