[Bug middle-end/65743] [5 regression] LTO+FDO build of Firefox crashes at startup

["hubicka at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65743

--- Comment #2 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Indeed, the compiler seems to think that the call is always bogus.

Considering ensureHolder/2875306 with 77 size
 to be inlined into getOwnPropertyDescriptor/2875397 in unknown:-1
 Estimated badness is -0.000068, frequency 1.00.
 Called 13697x
    Badness calculation for getOwnPropertyDescriptor/2875397 ->
ensureHolder/2875306
      size growth 67, time 23 inline hints: same_scc in_scc known_hot
big_speedup
      -0.000548: guessed profile. frequency 1.000000, count 13697 caller count
13697 time w/o inlining 78.000000, time w inlining 60.000000 overall growth
1129 (current) 602 (original) 658
      Adjusted by hints -0.000068
                Accounting size:51.50, time:18.78 on predicate:(true)
Processing frequency ensureHolder
  Called by getOwnPropertyDescriptor that is normal or hot
Processing frequency SetProxyExtra
  Called by ensureHolder that is normal or hot
discovered direct call to non-function in ensureHolder/18119817, making it
__builtin_unreachable
converting indirect call in ensureHolder to direct call to
__builtin_unreachable
Speculative indirect call ensureHolder/18119817 => createHolder/18119819 has
turned out to have contradicting known target __builtin_unreachable
Polymorphic call context combine:    Outer type (dynamic):struct XrayTraits (or
a derived type) offset 0

So this seems to be ipa-prop bug. It shouldnot prove the call to be bogus
(ipa-devirt machinery seems happy about the call)