[Mitgcm-support] RE: SUN compiler at Scripps

[mitgcm-support at dev.mitgcm.org]

Patrick,

 They should try and get the compiler upgraded to some
Workshop 6.0 version if possible. The revision they have
SC5.0 ..... was never that robust.

Chris

-----Original Message-----
From: heimbach at fjord.ucsd.edu [mailto:heimbach at fjord.ucsd.edu]
Sent: Thursday, April 04, 2002 4:41 PM
To: support at mitgcm.org; dstammer at ucsd.edu
Subject: Re: SUN compiler at Scripps


Detlef,

switch from f77 to f90 works fine on MIT Sun's.
You seem to have a compiler problem at Scripps.

Versions/libraries used:
* at MIT
  (SunOS slough 5.8 Generic_108528-07 sun4u sparc SUNW,Sun-Blade-1000):
  Sun WorkShop 6 update 2 Fortran 95 6.2 2001/05/15
  f90comp: Sun WorkShop 6 update 2 Fortran 95 6.2 2001/05/15
  /opt/SUNWspro/WS6U2/lib/v9/libmtsk.a
  loader: Software Generation Utilities - Solaris-ELF (4.0)

* at SIO
  (SunOS rain 5.8 Generic_108528-13 sun4u sparc SUNW,Ultra-80)
  f90: WorkShop Compilers 5.0 98/12/21 FORTRAN 90 2.0
  f90comp: WorkShop Compilers 5.0 98/12/21 FORTRAN 90 2.0
  /opt/SUNWspro/SC5.0/lib/v9/
  loader: Solaris Link Editors: 5.8-1.272

Patrick

Patrick Heimbach wrote:
>
> Attempt to witch from f77 to f90 on SUN fails so far:
> ("standard" compiler options, but O3 changed to O0:)
> f90 -stackvar -explicitpar -e -u -noautopar
> -xtypemap=real:64,double:64,integer:32 -fsimple=0 -dalign -O0 -xarch=v9
> -c
>
> *  f90: Internal Error, interface, iexp:3756
>    in cost_weights (fix: temporarily switched off)
> *  in mdsio_writemeta need to replace I/O format
>    '(x,...)' by '(1x,...)'  (fixed)
>    (compiler claims it expects integer at "x", then exits)
> *  at linking:
>    -xtypemap=real:64,double:64,integer:32 (command line)
>    ld: fatal: Symbol referencing errors. No output written to mitgcmuv
>    (replacing by "old" -r8 -i4 no more supported)
>
> Patrick Heimbach wrote:
> >
> > Correction for IBM SP3 (blue horizon):
> > Gradient checks devastating AND
> > NaN's for 240 timestep (10 day) run.
> > Occurrence of NaN's at timestep 204 (vs 214 on Scripps SUN).
> > Suggests a similar structure of the problem on these platforms.
> > On IBM -O option had been removed, -g option instead.
> >
> > Patrick Heimbach wrote:
> > >
> > > verification/natl_box_adjoint/
> > > with ecco_c44_e19 and latest merged code:
> > >
> > > - OK under Linux
> > >   (Linux twain.lcs.mit.edu 2.4.7-10 #1 Thu Sep 6 17:27:27 EDT 2001
i686)
> > > - OK on MIT SUN's, e.g. fjord
> > >   (SunOS fjord.mit.edu 5.8 Generic_108528-09 sun4u sparc
> > > SUNW,Sun-Blade-1000)
> > > - NaN's on Scripps SUN's after ~50 timesteps with current compiler
> > > settings
> > >   (SunOS rain.ucsd.edu 5.8 Generic_108528-13 sun4u sparc
SUNW,Ultra-80)
> > > - No NaN's on IBM SP3 (blue horizon), but devastating
> > >   gradient checks
> > > NB: code is identical, scp-ed between different platforms
> > >
> > > Trouble with "-g" option on Scripps SUN's:
> > >
> > > * Using -xtypemap=real:64,double:64,integer:32 with -g:
> > > f77: Warning: illegal option -xtypemap=real:64,double:64,integer:32
> > > ignored
> > > f77: Warning: Optimizer level changed from 0 to 3 to support
> > > parallelized code
> > > f77: Warning: -parallel conflicts with -g. -parallel turned off.
> > > ild: (bad file) Input file /usr/lib/libm.a contains 32-bit
relocatable,
> > > but producing a 64-bit file.
> > >
> > > * Using -r8 -i4 with -g:
> > > f77: Warning: Use -xtypemap for obsolescent -r8 or -dbl; see f77 man
> > > page.
> > > f77: Warning: Optimizer level changed from 0 to 3 to support
> > > parallelized code
> > > f77: Warning: -parallel conflicts with -g. -parallel turned off.
> > > ild: (bad file) Input file /usr/lib/libm.a contains 32-bit
relocatable,
> > > but producing a 64-bit file.
> > >
> > > * Switching off -stackvar -explicitpar :
> > >   Note: IEEE floating-point exception flags raised:
> > >     Inexact;  Underflow;  Invalid Operation;
> > >   (code goes wild)
>
> --
> _/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
> Patrick Heimbach           Scripps Institution of Oceanography
> Tel: +1/858/822-4062   Physical Oceanography Research Division
> Fax: +1/858/534-9820                 Nierenberg Hall, Room 445
> http://www.sio.ucsd.edu               8605 La Jolla Shores Dr.
> mailto:heimbach at fjord.ucsd.edu         La Jolla, CA 92093, USA