[Bug tree-optimization/37534] New: [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[Bug tree-optimization/37534] New: [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[luisgpm at linux dot vnet dot ibm dot com]

A degradation of ~17% occured in SPEC2000's ppc64 187.facerec benchmark right
after the IRA patch went in (revision 139590 -
http://gcc.gnu.org/viewcvs?view=rev&revision=139590).

The attached code is as far as i could go without removing all the bad
instruction sequences and yet making it smaller than the whole function.

While the code generated by r139589 has 174 load instructions and 59 store
instructions, the one generated by r139590 hs 215 loads and 103 stores. A
difference of about 40 instructions for each.

If we look into the complete function, the difference jumps to about 250 more
load/store instructions. The logs show that r139590 has more spills than
r139589.


-- 
           Summary: [4.4 Regression] IRA causes 17% degradation in
                    187.facerec benchmark
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: luisgpm at linux dot vnet dot ibm dot com
 GCC build triplet: powerpc64-linux
  GCC host triplet: powerpc64-linux
GCC target triplet: powerpc64-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug tree-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #1 from luisgpm at linux dot vnet dot ibm dot com  2008-09-15 23:46 -------
Created an attachment (id=16332)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16332&action=view)
187.facerec problematic source code

Reduced source file.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug tree-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2008-09-15 23:49 -------
The module FaceRecTypes is not in the attached  fortran code so the source
cannot be compiled.


-- 

pinskia at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug tree-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2008-09-15 23:50 -------
A simple fix is to remove that use statement and also change Params definition
to be:
        Integer, Intent(In)    :: Params


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[pinskia at gcc dot gnu dot org]

------- Comment #4 from pinskia at gcc dot gnu dot org  2008-09-16 01:04 -------
The first thing I noticed (this for both with/without IRA), is that we are
moving from cr and then comparing right away:
+       mfcr %r0,1
+       rlwinm %r0,%r0,30,1
...
+       cmpdi %cr7,%r0,0

Why is that?

The other thing I noticed for the tree level (still without/with IRA), is that
FRAME.98 is not SRA'ed because we take the address of one of the elements even
though we know that cannot be done on the normal pointer tricks on it (see PR
17217 which I had forgot to reopen when the code was disabled).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[pinskia at gcc dot gnu dot org]

------- Comment #5 from pinskia at gcc dot gnu dot org  2008-09-16 01:14 -------
The mfcr comes from:
        Do NewX = CoordX(IX, IY)-MaxDev, CoordX(IX, IY)+MaxDev, Step


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[pinskia at gcc dot gnu dot org]

------- Comment #6 from pinskia at gcc dot gnu dot org  2008-09-16 01:30 -------
(In reply to comment #5)
> The mfcr comes from:
>         Do NewX = CoordX(IX, IY)-MaxDev, CoordX(IX, IY)+MaxDev, Step

Filed that as PR 37536.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[luisgpm at linux dot vnet dot ibm dot com]

------- Comment #7 from luisgpm at linux dot vnet dot ibm dot com  2008-09-16 02:26 -------
Created an attachment (id=16333)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16333&action=view)
187.facerec problematic source code

Updated source file without compilation dependencies.


-- 

luisgpm at linux dot vnet dot ibm dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #16332|0                           |1
        is obsolete|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |missed-optimization, ra
   Target Milestone|---                         |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[mmitchel at gcc dot gnu dot org]

------- Comment #8 from mmitchel at gcc dot gnu dot org  2008-10-22 03:10 -------
What is the current performance status?


-- 

mmitchel at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[vmakarov at redhat dot com]

------- Comment #9 from vmakarov at redhat dot com  2008-10-22 14:00 -------
The performance status is still the same.  But I am working on it and actually
biggest part of my work time (about 90%) I spent on performance problems
particular this one.  They are the most hard to solve.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[bonzini at gnu dot org]

------- Comment #10 from bonzini at gnu dot org  2009-02-03 09:45 -------
How is it now?


-- 

bonzini at gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bonzini at gnu dot org
             Status|UNCONFIRMED                 |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[vmakarov at redhat dot com]

------- Comment #11 from vmakarov at redhat dot com  2009-02-03 14:48 -------
I have a patch (a new spill heuristic) which makes facerec even faster with IRA
on power6.  The patch is in

http://gcc.gnu.org/ml/gcc-patches/2008-11/msg00368.html

Unfortunately the new spill heuristic results in worse SPECINT2000 scores for
x86, x86_64.  So I did not committed the patch.

Facerec has one very hot (80%) function localmove::graphRoutines.f90 and
there is only one hot loop in the function.  Although the loop is
pretty big because of inlining TopCostFct.  The loop contains a few
if-statements and several big switch-statements.  But only a small part of
the loop body is hot.  Static branch prediction makes very inaccurate results
in this case and the old RA accidentally generates a better code than IRA.

 This is classical example of heuristic approach drawback.  You never
get best code using one heuristic for all programs.  Sometimes the old
RA will generate better code especially when the static branch probability
prediction is bad.  What are goal should be is to achieve
better or not worse code in "average", i.e. for some credible benchmark like
SPECFP2000 and IRA is successful in this. 

I am still working on some better spill heuristics.  Classical graph coloring
spill heuristics are bad and IMHO they should be based on register pressure not
on number of allocnos conflicting with allocno being spilled.

I don't think the problem will be solved for gcc4.4 but it might be solved for
gcc 4.5.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[bonzini at gnu dot org]

------- Comment #12 from bonzini at gnu dot org  2009-02-03 14:56 -------
Maybe we can approach the problem from the other side, for example marking I/O
functions as cold, or marking I/O functions as cold so that
PRED_FLAG_FIRST_MATCH hits in predict.def?


-- 

bonzini at gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|0000-00-00 00:00:00         |2009-02-03 14:56:47
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4 Regression] IRA causes 17% degradation in 187.facerec benchmark

[bonzini at gnu dot org]

-- 

bonzini at gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|2009-02-03 14:56:47         |2009-02-03 14:57:02
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4/4.5 Regression] IRA causes 17% degradation in 187.facerec benchmark

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.0                       |4.4.1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4/4.5 Regression] IRA causes 17% degradation in 187.facerec benchmark

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.1                       |4.4.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4/4.5 Regression] IRA causes 17% degradation in 187.facerec benchmark

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.2                       |4.4.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4/4.5 Regression] IRA causes 17% degradation in 187.facerec benchmark

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.3                       |4.4.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4/4.5/4.6 Regression] IRA causes 17% degradation in 187.facerec benchmark

[jakub at gcc dot gnu dot org]

-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.4                       |4.4.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

[Bug rtl-optimization/37534] [4.4/4.5/4.6 Regression] IRA causes 17% degradation in 187.facerec benchmark

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

--- Comment #14 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-02-20 15:41:32 UTC ---
It would also be interesting if with profile-feedback and/or LTO this
particular problem vanishes, given that both should result in better
static branch predictions.

[Bug rtl-optimization/37534] [4.4/4.5/4.6 Regression] IRA causes 17% degradation in 187.facerec benchmark

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING
                 CC|                            |dje at gcc dot gnu.org,
                   |                            |steven at gcc dot gnu.org

--- Comment #13 from Steven Bosscher <steven at gcc dot gnu.org> 2011-02-20 15:20:37 UTC ---
Ping?

No reconfirmation, no progress, no movement of _any_ kind for two years. Can
anyone tell whether there is still a regression here?

[Bug rtl-optimization/37534] [4.4/4.5/4.6 Regression] IRA causes 17% degradation in 187.facerec benchmark

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37534

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.4.5                       |4.4.6