[Bug tree-optimization/69347] [6 regression] excessive compile time with -O2

["law at redhat dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69347

--- Comment #13 from Jeffrey A. Law <law at redhat dot com> ---
The memory usage is a trivial problem.  The FSM bits do this in a couple
places:

      vec_alloc (next_path, n_basic_blocks_for_fn (cfun));

gcc-5 did this too and I dutifully copied it when those bits moved around for
gcc-6.

What did change in gcc-6 is how often we call into the FSM bits -- we rely on
the FSM approach fairly heavily in gcc-6 and it exposed that lameness.

All the uses are of the safe_push variety, so we can shove in any constant and
it'll adjust appropriately.  Doing so drops the usage more than an order of
magnitude.