[Bug middle-end/33966] Revision 129625 caused 11% slowdown on 200.sixtrack

[Bug middle-end/33966] Revision 129625 caused 11% slowdown on 200.sixtrack

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33966

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.6.0

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-01-28 03:15:52 UTC ---
This is all fixed in 4.6.0, we only have one partition now and use an alias
oracle for that one partition.  So closing as fixed.

[Bug middle-end/33966] Revision 129625 caused 11% slowdown on 200.sixtrack

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2007-11-01 11:40 -------
Current partitioning has several problems:

 1) Accounting only accounts for memory references through pointers.

 2) Accounting for the number of VOPs a statement will generate and accounting
    for how many accesses a memory symbol will experience are mixed (thus not
    allowing to weight these factors differently easily)

 3) The goal the pscore function tries to set at is not clear (partly due to
    reason 2).

I have some ideas/patches on this side, but it's late in stage3 and we'll
get (performance) disruption by the PR33870 fix on the partitioning side
as well.


-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33966

[Bug middle-end/33966] Revision 129625 caused 11% slowdown on 200.sixtrack

[rguenth at gcc dot gnu dot org]

------- Comment #2 from rguenth at gcc dot gnu dot org  2007-11-01 11:36 -------
Note I didn't experience any sixtrack regression (-O3 and -O3 -funroll-loops
-ffast-math) on Opteron.  This is very likely a partitioning issue, as the
--param max-aliased-vops is different for -O2 vs. -O3 and sixtrack likely
"lived on the edge" with the -O2 setting.

The patch mainly changed which variables with the same partitioning score are
considered for partitioning.  So if at all, this is a partitioning/scoring bug.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|rguenther at suse dot de    |rguenth at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33966

[Bug middle-end/33966] Revision 129625 caused 11% slowdown on 200.sixtrack

[ubizjak at gmail dot com]

------- Comment #1 from ubizjak at gmail dot com  2007-11-01 10:23 -------
PR Header should say 11%, not 1%.


-- 

ubizjak at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Revision 129625 caused 1%   |Revision 129625 caused 11%
                   |slowdown on 200.sixtrack    |slowdown on 200.sixtrack


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33966