-mips4 f77 compilation problem under 6.2

-mips4 f77 compilation problem under 6.2

Post by Alexis Cousei » Thu, 01 Aug 1996 04:00:00




> Since upgrading to IRIX 6.2, we're getting bus errors with
> -mips4 compilations that we never used to get before.  These
> only occur with optimization (-O3);  when we compile with -g, the
> bus errors don't occur, which makes it a bit tough to track them
> down.  The problems don't occur with -mips2 executables (or
> with -mips1 executables, compiled under 5.2).  Before upgrading
> to 6.2 from 6.01, we never saw this problem either.

> According to our "developers' friend" at SGI, the following patches
> have been released for the 6.2 Fortran compiler:

>   1262: released      6.2      libblas autoblock problem
>   1294: released      6.2      Fix to Fortran to be able to compile with -g opt
>   1345: released      6.2      f77 namelist read variable fix
>   1378: released      6.2      Complex Arithmetic Broken in 6.2 libftn

> None of these descriptions match my problem, but we've seen that these
> descriptions are not always exhaustive.  So -- though this is basically
> a shot in the dark -- I thought I'd ask whether anyone out there
> knows whether any of these patches might possibly also include
> patches to the code optimizer that might fix my problem.

Not sure I'd really blame it on the compiler. -mips4 is also -64 by default,
and that changes the entire addressing model, and that can reveal * little
things lurking in the app.

Tried -mips4 -n32? At least that'll give you the same addressing model as your
-mips2 -32 code...

Your 6.01 code, though *was* -64 code if it was -mips4. But the compilers have
changed quite a lot.

--
Alexis Cousein
Sales Support Engineer
Silicon Graphics NV/SA  (Belgium)

 
 
 

-mips4 f77 compilation problem under 6.2

Post by Peter Shenk » Thu, 01 Aug 1996 04:00:00


Since upgrading to IRIX 6.2, we're getting bus errors with
-mips4 compilations that we never used to get before.  These
only occur with optimization (-O3);  when we compile with -g, the
bus errors don't occur, which makes it a bit tough to track them
down.  The problems don't occur with -mips2 executables (or
with -mips1 executables, compiled under 5.2).  Before upgrading
to 6.2 from 6.01, we never saw this problem either.

According to our "developers' friend" at SGI, the following patches
have been released for the 6.2 Fortran compiler:

  1262: released      6.2      libblas autoblock problem
  1294: released      6.2      Fix to Fortran to be able to compile with -g opt
  1345: released      6.2      f77 namelist read variable fix
  1378: released      6.2      Complex Arithmetic Broken in 6.2 libftn

None of these descriptions match my problem, but we've seen that these
descriptions are not always exhaustive.  So -- though this is basically
a shot in the dark -- I thought I'd ask whether anyone out there
knows whether any of these patches might possibly also include
patches to the code optimizer that might fix my problem.

BTW, we do have 1294 (only) installed.

        -P.

--
********************* (Note new snail-mail address.) **********************
* Peter S. Shenkin, Chemistry, Columbia U., 3000 Broadway, Mail Code 3153,*

MacroModel WWW page: http://www.cc.columbia.edu/cu/chemistry/mmod/mmod.html

 
 
 

-mips4 f77 compilation problem under 6.2

Post by Benjamin Z. Goldste » Fri, 09 Aug 1996 04:00:00





>>Not sure I'd really blame it on the compiler. -mips4 is also -64 by default,
>>and that changes the entire addressing model, and that can reveal * little
>>things lurking in the app.

>>Tried -mips4 -n32? At least that'll give you the same addressing model as your
>>-mips2 -32 code...

>>Your 6.01 code, though *was* -64 code if it was -mips4. But the compilers have
>>changed quite a lot.
>Yes, we previously had been generating -mips4 -64 code that ran
>correctly.  We resolved a bunch of 64-bit-addressing problems several
>years back when we first ported to 6.0.1.
>So we're back to possibly blaming it on the compiler -- specifically,
>the code optimizer, since the "-g" version runs just fine.
>Unfortunately, it's hard for us to try "-n32", since all our
>-mips4 libraries (and there are a lot of them) are compiled
>-64 at the moment.  So if this does sound familiar to someone,
>and there's a patch that might help, I'd like to know about it.
>(My next attempt will be to lower the optimization level....)

I have often found that the highest optimization levels aren't as reliable
as the basic ones.  I often found that the performance differences are
minimal, too.  You might try using the "-g3" option instead of the "-g"
option.

Best regards,
--
Benjamin Z. Golds*

 
 
 

1. Answer to: Bus Error with -mips4 IRIX 6.2 Fortran compilation

I recently posted a note to the effect that since upgrading to 6.2
from 6.0.1, our -mips4 -64 executables are giving Bus errors, when
they never did before.  I mentioned that this only occurred at high
optimization;  under -g, there was no problem, which made things hard
to track down.

Nevertheless, I did eventually track it down, using -O3 -g3 and cvd.
The cause was programmer error (surprise, surprise :-) ).  I'm
embarrassed to say that I called a subroutine with no arguments
when the subroutine definition required an argument list.  No
arguments were actually accessed during this call, which, I
suppose, is why there was never a problem before, on SGI or
any other platform.  But it was certainly an error in my code.

Thanks to Alexis Cousein and Omar Stradella, both of SGI, for
help.

        -P.

--
********************* (Note new snail-mail address.) **********************
* Peter S. Shenkin, Chemistry, Columbia U., 3000 Broadway, Mail Code 3153,*

MacroModel WWW page: http://www.cc.columbia.edu/cu/chemistry/mmod/mmod.html

2. limit on third party conduits?

3. gcc compilation problems under IRIX 6.2

4. Little problem with Emacs 19.22

5. Problems w/ gcc compilation under irix 6.2 on R10k?

6. Mercury Interactive Address

7. IRIX 6.2 /usr/lib32/mips4 .so files

8. zoom vfp24k internal

9. xntp compilation error on 6.2

10. Compilation difficulties (Irix 6.2)

11. f77 on Irix 6.2, ``-noerror_limit'' ??

12. f77 and NAMELIST bug in IRIX 6.2

13. f77, Irix 6.2, -mp, assertion failure