http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60381
Alexandre Oliva <aoliva at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |ASSIGNED Last reconfirmed| |2014-03-05 Component|middle-end |debug Assignee|unassigned at gcc dot gnu.org |aoliva at gcc dot gnu.org Ever confirmed|0 |1 --- Comment #4 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- Mine. Wow, I didn't see that one coming. I thought we only have links from preserved values to other preserved values, but this proves this is not the case. A reversal of 208220 will certainly fix that, but I'm assessing the impact of that on the pr59992 fix, and looking into combining the two table cleanups at cselib_reset_table into a single pass. I have a hunch there's some remaining complexity from back when we actually visited each insn twice and had to roll back cselib table changes.