[Bug rtl-optimization/23490] [3.4/4.0/4.1 Regression] Long compile time for array initializer with inlined constructor

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

------- Comment #9 from hubicka at gcc dot gnu dot org  2005-11-03 20:06 -------
For reload CSE there is --param max-cselib-memory-locations=10
that cuts it's time to 1%, overall.
Since this testcase is sort of designed to excercise the worst case behaviour,
I think it is not too bad...


alias analysis has:
/* Maximum length of pbi->mem_set_list before we start dropping
   new elements on the floor.  */
#define MAX_MEM_SET_LIST_LEN    100
perhaps unifying all these lists would make sense.  Would that be applicable
for 4.1? (Ie I would have to pick some common value, like 300 - cselib defaults
to 500)

CSE has:
      /* If we have processed 1,000 insns, flush the hash table to
         avoid extreme quadratic behavior.  We must not include NOTEs
         in the count since there may be more of them when generating
         debugging information.  If we clear the table at different
         times, code generated with -g -O might be different than code
         generated with -O but not -g.

         ??? This is a real kludge and needs to be done some other way.
         Perhaps for 2.9.  */

I am not sure if I can fit memory specific limit in this (it would be nice to
simply throw away old entries rather than flushing it all), or we can just keep
it.
None of these algorithms would grow ad infimum and manifest worse behaviour
than in this testcase, so I would not consider it P2

Honza


-- 

hubicka at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23490