https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68799
Bill Schmidt <wschmidt at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Last reconfirmed| |2015-12-11 Ever confirmed|0 |1 --- Comment #4 from Bill Schmidt <wschmidt at gcc dot gnu.org> --- Confirmed while building in doko's environment. However, I haven't been able to fully sort out the problem by looking at the SLSR detail dumps. It appears to involve two conditional candidates that depend on the same PHI basis. The first one is processed correctly, introducing a new PHI basis for it, but when we look at the second one it appears that we can no longer map from the PHI statement to the PHI candidate. It may have to do with using split_edge to place the add statements that feed the new PHI basis, but off the cuff I don't see anything wrong with that. Matthias, can you please incorporate th3 debug patch from the previous comment into your source so I can gather more information at the time of the failure? It won't change behavior of the compiler unless -fdump-tree-slsr* is specified.