[Bug tree-optimization/81945] [8 Regression] ICE in operator[], at vec.h:749

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81945

--- Comment #3 from amker at gcc dot gnu.org ---
Don't know transformation done by graphite.  In this case, graphite0 has an
additional function dump:
;; Function at._loopfn.1 (at._loopfn.1, funcdef_no=2, decl_uid=1951,
cgraph_uid=1, symbol_order=2)

at._loopfn.1 (void * .paral_data_param)
{
  //...
}

which is generated from the original function at?  Also the new function
contains loop and loop->orig_loop_num is copied from data LFN_LOOP_DIST_ALIAS. 
It also looks like the distributed loop is handled by graphite/parloops.  In
general I think it's okay, only waste some compilation time because transformed
loop will be removed by vectorizer.  But leaking loop->orig_loop_num
information to newly created function is wrong.