[Bug fortran/30367] New: gfortran compile times excessive with -Wall

[Bug fortran/30367] New: gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

Hi,

Gfortran successfully compiles our FUN3D Fortran 90 code base (~600KLOC) and
does so in comparable time to other compilers.  When -Wall is turned on,
however, the compile times of certain large routines (e.g., the 12KLOC
Adjoint/residual_turbulent.f90), the compilation time increases from a few
minutes to an hour.

Unfortunately, the code is U.S. Person export controlled, but Steve Kargl
successfully obtained a copy in the past (and could request an updated,
autoconf'd version by sending a request to fun_support@larc.nasa.gov).

Sorry about the lame bug report, but I'll be glad to furnish whatever further
details I can given the export restriction.

Regards,
--
Bil Kleb
1.757.864.4364
http://fun3d.larc.nasa.gov
http://funit.rubyforge.org


-- 
           Summary: gfortran compile times excessive with -Wall
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: Bil dot Kleb at NASA dot gov
 GCC build triplet: http://quatramaran.ens.fr/~coudert/gfortran/gfortran-
                    linux.tar.g
  GCC host triplet: Linux 2.6.9-5.ELsmp #1 SMP Wed Jan 5 19:30:39 EST 2005
                    i686 i686
GCC target triplet: Same


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[jvdelisle at gcc dot gnu dot org]

------- Comment #1 from jvdelisle at gcc dot gnu dot org  2007-01-09 06:25 -------
Are there a lot of warnings emitted when compiled with -Wall ? or None?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #2 from Bil dot Kleb at NASA dot gov  2007-01-09 13:15 -------
No warnings with -Wall -- we've been busy cleaning the last few months.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #3 from Bil dot Kleb at NASA dot gov  2007-01-13 12:35 -------
I've just tried to duplicate the compile time difference with -Wall and found
that the -Wall was a red herring -- it always takes 169 minutes to compile our
code with gfortran, -Wall or not.

This is about 8 times longer than any other compiler, even with the SGI
compiler running on hardware that is an order of magnitude slower than the
hardware gfortran is running on.

Our current compiler guantlet includes: Absoft-10.0.3, NAG-5.1.282, g95-0.91,
PGI-6.2.4, Sun-8.3.35_2, gfortran-20070103, ifort-9.1.040, SGI-7.41,
LaheyX-6.10a, and LaheyX-8.00a.  (Where -O0 is the only compiler option
specified for all compilers.)

Gfortran's compilation time is mainly due to a three routines that take on the
order of 1/2 hour each. These routines are long (~10KLOC) and compute large
Jacobian matrices formed from many temporary variables.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[kargl at troutmask dot apl dot washington dot edu]

------- Comment #4 from kargl at troutmask dot apl dot washington dot edu  2007-01-13 17:23 -------
Subject: Re:  gfortran compile times excessive with
 -Wall

Bil dot Kleb at NASA dot gov wrote:
> 
> Gfortran's compilation time is mainly due to a three routines that take on the
> order of 1/2 hour each. These routines are long (~10KLOC) and compute large
> Jacobian matrices formed from many temporary variables.
> 

Do these routines use a large number of modules?
Any possibility of sending one of the problematic
routines to us to profile the compiler?  Can you
run "gfortran -S -fmem-report -ftime-report" with
one of the files and attach the info here?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #5 from Bil dot Kleb at NASA dot gov  2007-01-13 21:11 -------
Created an attachment (id=12896)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12896&action=view)
Entire compilation log with -fmem-report -ftime-report turned on

The trouble routines are part of the Adjoint code:

 residual_turbpart.f90
 residual_turbulent.f90
 precond_turbpart.f90
 precond_turbulent.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #6 from Bil dot Kleb at NASA dot gov  2007-01-13 21:12 -------
Created an attachment (id=12897)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12897&action=view)
Lakos-style use associations for residual_turbpart.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #8 from Bil dot Kleb at NASA dot gov  2007-01-13 21:13 -------
Created an attachment (id=12899)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12899&action=view)
Lakos-style use associations for precond_turbpart.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #7 from Bil dot Kleb at NASA dot gov  2007-01-13 21:13 -------
Created an attachment (id=12898)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12898&action=view)
Lakos-style use associations for residual_turbulent.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #9 from Bil dot Kleb at NASA dot gov  2007-01-13 21:14 -------
Created an attachment (id=12900)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12900&action=view)
Lakos-style use associations for precond_turbulent.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #10 from Bil dot Kleb at NASA dot gov  2007-01-13 21:17 -------
Based on the Lakos-style analysis of the dependencies for these particular
routines, it /might/ be possible to supply self-contained source; but it will
take a while to check with the powers that be.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive with -Wall

[Bil dot Kleb at NASA dot gov]

------- Comment #11 from Bil dot Kleb at NASA dot gov  2007-01-13 21:28 -------
Created an attachment (id=12902)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12902&action=view)
An excerpt from precond_turbulent.f90

Here's an excerpt from one of the offenders.  Note: the module only has a
single routine and it has a ridiculous amount of local variables...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive with -Wall

[pinskia at gcc dot gnu dot org]

------- Comment #12 from pinskia at gcc dot gnu dot org  2007-01-13 21:29 -------
 global alloc          :2249.91 (100%) usr   0.28 (76%) sys2250.16 (100%) wall 
  3897 kB (13%) ggc
 flow 2                :   0.04 ( 0%) usr   0.00 ( 0%) sys   0.03 ( 0%) wall   
 447 kB ( 2%) ggc
 reg stack             :   0.01 ( 0%) usr   0.00 ( 0%) sys   0.01 ( 0%) wall   
   7 kB ( 0%) ggc
 final                 :   0.12 ( 0%) usr   0.00 ( 0%) sys   0.12 ( 0%) wall   
   0 kB ( 0%) ggc
 TOTAL                 :2251.56             0.37          2251.91             
29616 kB
Extra diagnostic checks enabled; compiler may run slowly.
Configure with --disable-checking to disable checks.


But I note you have checking enabled so can you recompile GCC with
--enable-checking=release?


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|fortran                     |rtl-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug fortran/30367] gfortran compile times excessive for routines with large numbers of local variables

[Bil dot Kleb at NASA dot gov]

------- Comment #13 from Bil dot Kleb at NASA dot gov  2007-01-13 21:47 -------
Unfortunately, I am working from a binary release I got from 

 http://gcc.gnu.org/wiki/GFortranBinaries.

Specifically, I have the 03-Jan-2007 the 32-bit binary from:

 http://quatramaran.ens.fr/~coudert/gfortran/gfortran-linux.tar.gz

Is there an alternative binary available, or will I have to compile from
source?


-- 

Bil dot Kleb at NASA dot gov changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|rtl-optimization            |fortran
            Summary|gfortran compile times      |gfortran compile times
                   |excessive with -Wall        |excessive for routines with
                   |                            |large numbers of local
                   |                            |variables


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[jvdelisle at gcc dot gnu dot org]

------- Comment #14 from jvdelisle at gcc dot gnu dot org  2007-01-13 22:13 -------
The binaries will be rebuilt in a few days.


-- 

jvdelisle at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|fortran                     |rtl-optimization


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[Bil dot Kleb at NASA dot gov]

------- Comment #15 from Bil dot Kleb at NASA dot gov  2007-01-23 08:44 -------
I tried out the one dated 23-jan-07, but it was apparently also configured with
checks:

 Extra diagnostic checks enabled; compiler may run slowly.
 Configure with --disable-checking to disable checks.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[jvdelisle at gcc dot gnu dot org]

------- Comment #16 from jvdelisle at gcc dot gnu dot org  2007-01-24 02:21 -------
The build scripts are automated and I don't think I have access to tweak those
so I sent a note to FX to see if he can do so


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[Bil dot Kleb at NASA dot gov]

------- Comment #17 from Bil dot Kleb at NASA dot gov  2007-01-24 03:16 -------
I managed to compile from SVN source with --disable-checking, and now the
compilation times for the offending routines are on the order of a second!

I will attach the new make log in case you'd like to compare with the first.
(Note: our code has undergone small changes in the interim.)  


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[Bil dot Kleb at NASA dot gov]

------- Comment #18 from Bil dot Kleb at NASA dot gov  2007-01-24 03:18 -------
Created an attachment (id=12946)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=12946&action=view)
A make log with --disable-checking gfortran


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[pinskia at gcc dot gnu dot org]

------- Comment #19 from pinskia at gcc dot gnu dot org  2007-01-24 03:35 -------
I think we can declare this as invalid then, as all compile times for the files
are under 3 seconds.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[pinskia at gcc dot gnu dot org]

------- Comment #20 from pinskia at gcc dot gnu dot org  2007-01-24 03:41 -------
I should mention that the extra compile time spend with a --enable-checking
build is internal checking in the compiler can be expensive at some points
causing these compile time explosions.  For an example, a checking alogorthim
might be 2^x while the normal code is x^2.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[Bil dot Kleb at NASA dot gov]

------- Comment #21 from Bil dot Kleb at NASA dot gov  2007-01-24 03:50 -------
So, can we publish builds that yield reasonable compile times? I.e., gcc
configured with --disable-checking.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[pinskia at gcc dot gnu dot org]

------- Comment #22 from pinskia at gcc dot gnu dot org  2007-01-24 04:17 -------
(In reply to comment #21)
> So, can we publish builds that yield reasonable compile times? I.e., gcc
> configured with --disable-checking.

The builds on gcc.gnu.org/wiki are not published by the FSF.  The FSF does not
publish any binaries really.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367

[Bug rtl-optimization/30367] gfortran compile times excessive for routines with large numbers of local variables

[fxcoudert at gcc dot gnu dot org]

------- Comment #23 from fxcoudert at gcc dot gnu dot org  2007-01-24 08:19 -------
Starting tomorrow, the binaries linked from the gfortran wiki page will be
compiled with --enable-checking=release, which should go much faster.

Thanks to Jerry (jvdelisle) for bringing that PR to my attention.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30367