https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79165
--- Comment #20 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Some other benchmarks spiked at same time (and recovered with the removal of
the bad linemap):
* "doduc" (from 3.5s up to to 4.s)
(with the bad linemap, get_next_line is 15% of the profile (in perf)
without the bad linemap, get_next_line is only 0.3% of the profile)
* "induct" (from 2.85s up to 3.275s)
* "protein", possibly
* "induct2"
* plus some C++ testcases appeared to spike at the same time:
* "tramp3d":
* "boost wave"
* "botan"
* "DLV"
