wrong code with -fforce-addr

[Manfred Schwarb]

Hi,

I have a rather nasty optimization issue with gfortran (as of yesterday). As I 
think it could
be an optimization issue and not an gfortran frontend issue, I post it here,
the fortran list was not able to help so far.

I narrowed my problem to one single fortran function. If I compile this 
function with
"gfortran -O2 -march=pentium4" the output is OK, using
"gfortran -O2 -march=pentium4 -fforce-addr" produces wrong output.

When reducing optimization level to "-O1", the issue vanishes for all tried 
flag combinations.
If I comment out a particular non-functional line in this code, the issue goes 
away:

if (always_true) then
  
else
   this_line_commented_out_makes_it_working_again
   
endif

Further data points:
- output of -fdump-tree-optimized is the same for both the working and the 
broken case (of the 
original function, not the line commenting test).
- there are differences in assembler output though.
- the code calls C functions.
- I could reproduce this issue on an different, non-pentium4 machine using the 
same flags.
- the flag -march=pentium4 is necessary to trigger the issue.


I'm in a loss where to search for the real cause. Has anybody a hint how to 
proceed further?
Should I post the code of this function here on the list (~300 LOC)?

Thanks,
Manfred

Re: wrong code with -fforce-addr

[Manfred Schwarb]

> > Hi,
> >
> > I have a rather nasty optimization issue with gfortran (as of
> yesterday). As I think it could
> > be an optimization issue and not an gfortran frontend issue, I post it
> here,
> > the fortran list was not able to help so far.
> >
> > I narrowed my problem to one single fortran function. If I compile this
> function with
> > "gfortran -O2 -march=pentium4" the output is OK, using
> > "gfortran -O2 -march=pentium4 -fforce-addr" produces wrong output.
> >
> > When reducing optimization level to "-O1", the issue vanishes for all
> tried flag combinations.
> > If I comment out a particular non-functional line in this code, the
> issue goes away:
> >
> > if (always_true) then
> >   
> > else
> >this_line_commented_out_makes_it_working_again
> >
> > endif
> >
> > Further data points:
> > - output of -fdump-tree-optimized is the same for both the working and
> the broken case (of the
> > original function, not the line commenting test).
> > - there are differences in assembler output though.
> > - the code calls C functions.
> > - I could reproduce this issue on an different, non-pentium4 machine
> using the same flags.
> > - the flag -march=pentium4 is necessary to trigger the issue.
> >
> >
> > I'm in a loss where to search for the real cause. Has anybody a hint how
> to proceed further?
> > Should I post the code of this function here on the list (~300 LOC)?
> 
> You should file a bugreport with bugzilla instead. 
> http://gcc.gnu.org/bugzilla
> 
> Thanks,
> Richard.

It's http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33638

I first hoped to get help to narrow things a bit further, as my
problem is really a bit strange, so I decided to first write an email.
I simply assigned it now to the middle-end, but of course it could 
also be a gfortran issue.

Thanks,
Manfred



-- 
Ist Ihr Browser Vista-kompatibel? Jetzt die neuesten 
Browser-Versionen downloaden: http://www.gmx.net/de/go/browser

Re: wrong code with -fforce-addr

[Manfred Schwarb]

> On Wed, Oct 03, 2007 at 12:24:27PM +0200, Manfred Schwarb wrote:
> > I'm in a loss where to search for the real cause. Has anybody a hint
> > how to proceed further?
> 
> Sounds like weird-but-somewhat-determinist behaviour you can get when
> you do out-of-bounds access on the stack or this kind of problems.
> Did you try valgrind or your os-of-choice equivalent on the code?
> 
>   OG.

I retested with valgrind just now, there was actually one problem
in an unrelated function ("Conditional jump or move depends on 
uninitialised value"). I fixed this now with proper initialization.

After this correction the codes passes cleanly with valgrind and efence.

But this did not help, I still do get wrong code with "-fforce-addr",
the behaviour did not change at all.

Thanks,
Manfred



-- 
Ist Ihr Browser Vista-kompatibel? Jetzt die neuesten 
Browser-Versionen downloaden: http://www.gmx.net/de/go/browser