[Bug middle-end/54114] New: variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[Bug middle-end/54114] New: variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[jimis at gmx dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

             Bug #: 54114
           Summary: variable-tracking performance regression from
                    4.8-20120610 to 4.8-20120701
    Classification: Unclassified
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jimis@gmx.net


Compiling the XPath.cpp from xalanc shows significant performance regression
because of variable tracking, between GCC 4.8-20120610 to 4.8-20120701.
Tracking it down is hard since var-tracking.c changed a lot in the past month,
CC'ing lxo for better ideas. :-)

It seems to apply mostly to C++ code, see the graph and the last annotation at
http://teras-ics.mooo.com:8003/

[Bug middle-end/54114] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[jimis at gmx dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

--- Comment #1 from jimis <jimis at gmx dot net> 2012-07-28 20:49:55 UTC ---
Sorry guys my machine died an ugly death, so the web interface for my profiling
runs is unfortunately offline. :-( From memory I remember that this regression
is more than 0.5s overhead out of ~5s. For more details I'll have to repair my
equipment.

[Bug middle-end/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2012-07-30
            Version|unknown                     |4.8.0
   Target Milestone|---                         |4.8.0
            Summary|variable-tracking           |[4.8 Regression]
                   |performance regression from |variable-tracking
                   |4.8-20120610 to             |performance regression from
                   |4.8-20120701                |4.8-20120610 to
                   |                            |4.8-20120701
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-07-30 09:11:00 UTC ---
Please attach preprocessed source.

[Bug middle-end/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[jimis at gmx dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

--- Comment #3 from jimis <jimis at gmx dot net> 2012-07-30 12:18:49 UTC ---
Created attachment 27894
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27894
XPatch.cpp preprocessed source from xalanc

Hi thanks for your patience, I resurrected my PC so now I have access to my
files, and the webinterface is up for anyone caring to see the function
breakdown (just check footnote [6] in front page). Attached is the preprocessed
source, I compile with

$CC1PLUS -quiet -D_GNU_SOURCE -D LINUX -D _REENTRANT -D XALAN_INMEM_MSG_LOADER 
XPath.ii  -g -O2 -fno-elide-constructors -fPIC -o /dev/null

[Bug middle-end/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

--- Comment #4 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2012-08-07 20:28:05 UTC ---
It seems quite clear to me that the new cost arises due to
clobber_overlapping_mems, a relatively expensive operation that fixes a debug
info correctness bug and that we just didn't do before.  I'm afraid I can't
think of a way to do it in a way that is significantly more efficient, so we'll
have to choose between this additional cost, or living with debug locations
that are no longer valid due to overlapping stores :-(

I'm tempted to mark this as WONTFIX, but I thought I'd leave this open for a
bit in case anyone else wants to give it a try.

[Bug middle-end/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
             Status|WAITING                     |NEW

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-19 14:04:51 UTC ---
Confirmed.

[Bug middle-end/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |jakub at gcc dot gnu.org
         Resolution|                            |WONTFIX

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-12-18 11:01:24 UTC ---
 variable tracking       :   0.08 ( 1%) usr   0.01 ( 3%) sys   0.11 ( 1%) wall 
  3165 kB ( 3%) ggc
 var-tracking dataflow   :   0.36 ( 4%) usr   0.00 ( 0%) sys   0.38 ( 4%) wall 
   226 kB ( 0%) ggc
 var-tracking emit       :   0.31 ( 3%) usr   0.00 ( 0%) sys   0.36 ( 4%) wall 
  2186 kB ( 2%) ggc

for me, I'd call this insignificant, given that it was really a debug info
correctness fix.
We have PR54402 which tracks cases where the slowdown is actually significant.

[Bug middle-end/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

--- Comment #7 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2013-01-18 10:58:47 UTC ---
Author: aoliva
Date: Fri Jan 18 10:58:15 2013
New Revision: 195291

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=195291
Log:
PR debug/54114
PR debug/54402
PR debug/49888
* var-tracking.c (negative_power_of_two_p): New.
(global_get_addr_cache, local_get_addr_cache): New.
(get_addr_from_global_cache, get_addr_from_local_cache): New.
(vt_canonicalize_addr): Rewrite using the above.  Adjust the
heading comment.
(vt_stack_offset_p): Remove.
(vt_canon_true_dep): Always canonicalize loc's address.
(clobber_overlapping_mems): Make sure we have a MEM.
(local_get_addr_clear_given_value): New.
(val_reset): Clear local cached entries.
(compute_bb_dataflow): Create and release the local cache.
Disable duplicate MEMs clobbering.
(emit_notes_in_bb): Clobber MEMs likewise.
(vt_emit_notes): Create and release the local cache.
(vt_initialize, vt_finalize): Create and release the global
cache, respectively.
* alias.c (rtx_equal_for_memref_p): Compare operands of
ENTRY_VALUEs.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/alias.c
    trunk/gcc/var-tracking.c

[Bug lto/54114] [4.8 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[aoliva at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Alexandre Oliva <aoliva at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |NEW
          Component|middle-end                  |lto
         Resolution|WONTFIX                     |

--- Comment #8 from Alexandre Oliva <aoliva at gcc dot gnu.org> 2013-01-18 11:06:38 UTC ---
This should be much better now.  There still ought to be a better way to deal
with MEM clobbering than comparing to each one at every write, but we don't
have that yet.

[Bug debug/54114] [4.8/4.9 Regression] variable-tracking performance regression from 4.8-20120610 to 4.8-20120701

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|lto                         |debug
   Target Milestone|4.8.0                       |4.9.0
            Summary|[4.8 Regression]            |[4.8/4.9 Regression]
                   |variable-tracking           |variable-tracking
                   |performance regression from |performance regression from
                   |4.8-20120610 to             |4.8-20120610 to
                   |4.8-20120701                |4.8-20120701

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-21 18:56:44 UTC ---
Something to look for 4.9 then.

[Bug debug/54114] VTA compile-time performance could be improved

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at redhat dot com
            Summary|[4.8/4.9 Regression]        |VTA compile-time
                   |variable-tracking           |performance could be
                   |performance regression from |improved
                   |4.8-20120610 to             |
                   |4.8-20120701                |

--- Comment #10 from Jeffrey A. Law <law at redhat dot com> ---
Removing regression marker based on c#7 and c#8.  Sounds like there's still
work that can be done here, but that the regression aspect is significantly
improved.

[Bug debug/54114] VTA compile-time performance could be improved

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.0                       |4.9.1

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.0 has been released

[Bug debug/54114] VTA compile-time performance could be improved

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.1                       |4.9.2

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.1 has been released.

[Bug debug/54114] VTA compile-time performance could be improved

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.2                       |4.9.3

--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.2 has been released.

[Bug debug/54114] VTA compile-time performance could be improved

[dsl at dsl dot pp.ua]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Eliot <dsl at dsl dot pp.ua> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dsl at dsl dot pp.ua

--- Comment #14 from Eliot <dsl at dsl dot pp.ua> ---
Comment on attachment 27894
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=27894
XPatch.cpp preprocessed source from xalanc

http://legup.eecg.utoronto.ca

[Bug debug/54114] VTA compile-time performance could be improved

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.3 has been released.

[Bug debug/54114] VTA compile-time performance could be improved

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.3                       |4.9.4

[Bug debug/54114] VTA compile-time performance could be improved

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54114

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |compile-time-hog
   Target Milestone|4.9.4                       |---