Martin,

You can go to

http://www.frontier.iarc.uaf.edu/~hsimmons/test1_ibm.tar.gz

to find my files with the mods necessary to compile and run test1 on
our IBM.  I ran with -02 optimization and MPI using 8 cpus over 2 cpu
nodes (4 cpus/node), and results looked reasonable.  Specific details that
may be important:

This machine has pwr3 architecture, and I have version 7.1.0.0 of the
compiler. Notice that I modified the compile script so that when I compile
test case 1 with 02 optimization, I deprecate only mpp_domains to be
compiled using only the default optimization. This is accomplished using a
sed command to modify the Makefile.

--------------------------------------------------------------------
$MKMF -c "$MACROS" -m Makefile -p $EXPERIMENT.exe -t $TEMPLATE $UTILITIES
$DRIVER $CORE $BC $MOM_PARAMS $MOM_SOURCES $INCLUDE_DIR || exit(-1)

# replace FFLAGS with FFLAGS2 in Makefile so that mpp_domains is compiled
without optimization
\mv Makefile tmp
cat tmp | sed 's/^\(.*\)FFLAGS\(.*mpp_domains.F90.*\)/\1FFLAGS2\2/' >  Makefile
make
--------------------------------------------------------------------------

I have not remounted the assault on our IBM p690, which is a newer, more
modern machine (pwr4). When I tried several months ago, I ran into
internal compiler errors. The p690 has version 8.1.0.1 of the compiler,
but it may have been patched since I complained about the internal
compiler errors.

Let me know if test1_ibm_tar.gz file presents any problems or if you have
any other suggestions or questions. As always, site paths will need to be
changed in fms_site_paths.

Harper

PS, here are the contents of
http://www.frontier.iarc.uaf.edu/~hsimmons/test1_ibm.tar.gz

-------------------------------------------------------
tar -cvf test1_ibm.tar bin/*hls* preprocessing/grid_generator
runscripts/fms_site_paths runscripts/test1
a bin/template.hls.02.ibm.mpp 2 blocks.
a bin/template.hls.03.ibm 1 blocks.
a bin/template.hls.03.ibm.mpp 1 blocks.
a bin/template.hls.ibm 1 blocks.
a bin/template.hls.ibm.mpp 2 blocks.
a preprocessing/grid_generator
a preprocessing/grid_generator/topog.F90 76 blocks.
a preprocessing/grid_generator/grid_gen.F90 2 blocks.
a preprocessing/grid_generator/build_grid_generator.csh 3 blocks.
a preprocessing/grid_generator/edit_grid.F90 13 blocks.
a preprocessing/grid_generator/grid_compare_driver.F90 6 blocks.
a preprocessing/grid_generator/grids.f90 99 blocks.
a preprocessing/grid_generator/grid_io.f90 25 blocks.
a preprocessing/grid_generator/grid_compare.F90 10 blocks.
a preprocessing/grid_generator/CVS
a preprocessing/grid_generator/CVS/Tag 1 blocks.
a preprocessing/grid_generator/CVS/Root 1 blocks.
a preprocessing/grid_generator/CVS/Entries 1 blocks.
a preprocessing/grid_generator/CVS/Repository 1 blocks.
a runscripts/fms_site_paths 1 blocks.
a runscripts/test1
a runscripts/test1/mom4_experiment_paths 2 blocks.
a runscripts/test1/mom4_run.csh 23 blocks.
a runscripts/test1/README 9 blocks.
a runscripts/test1/mom4_compile.csh 8 blocks.
a runscripts/test1/CVS
a runscripts/test1/CVS/Tag 1 blocks.
a runscripts/test1/CVS/Root 1 blocks.
a runscripts/test1/CVS/Entries 1 blocks.
a runscripts/test1/CVS/Repository 1 blocks.
a runscripts/test1/preprocessing
a runscripts/test1/preprocessing/generate_grid.csh 2 blocks.
a runscripts/test1/preprocessing/CVS
a runscripts/test1/preprocessing/CVS/Tag 1 blocks.
a runscripts/test1/preprocessing/CVS/Root 1 blocks.
a runscripts/test1/preprocessing/CVS/Entries 1 blocks.
a runscripts/test1/preprocessing/CVS/Repository 1 blocks.
a runscripts/test1/preprocessing/grid_output 20 blocks.
a runscripts/test1/tmp_llrun 1 blocks.
a runscripts/test1/tmp_llrun.out 2132 blocks.
a runscripts/test1/tmp_llrun~ 1 blocks.
a runscripts/test1/tmp_llrun.err 369 blocks.


=====================================================================
 International Arctic Research Center, University of Alaska, Fairbanks
 903 Koyukuk Drive, Fairbanks, Alaska 99775-7340
---------------------------------------------------------------------
fax  : (907) 474-2643
tele : (907) 474-5729
email: [EMAIL PROTECTED]
web  : http://www.frontier.iarc.uaf.edu/~hsimmons

On Fri, 31 Jan 2003, Martin Schmidt wrote:

> Hi Harper,
>
> MOM4 with an IBM is becoming more and more a nightmare. I am not able to run the
> testcases. Unfortunately I have deleted the previous beta.
>
> May be you can help me a little bit.
>
> xlf7 is not able to compile the fms code! It is running out of ressources.
>
> xlf8 does the job, but now the behaviour of the TRANSFER function is changed! This
> influences the function lowercase and lcase defined and used at different locations
> in the fms-stuff. As a result the model bombs very early with the message that the
> fms warning level is undefined. Indeed warning is amputated to w. Making prints and
> the optimisation level influences the result. As a workaround, I have written my
> own lowercase fuinction avoiding to use TRANSFER.
>
> Otherwise my list reads:
>
> xlf version 7.1 is not able to compile mpp_domains (no more ressources)
>
> Generally: #endif + comment shoud be C-like:
> #endif /* co */
>
> nsclock.c
> long delta_t=0;                       /* clock resolution in nanoseconds */
>
> udunits.F90
> #ifdef __aix
> #define UTOPEN utopen_
> #define UTDEC utdec_
> #define UTTIME  uttime_
> #define UTORIGIN utorigin_
> #define UTCALTIME utcaltime_
> #define UTMAKE utmake_
> #define UTCVT utcvt_
> #define UTCLS utcls_
> #endif
>
> ocean_util.F90
> #ifdef __aix
> # define flush(unit) flush_(unit)
> #endif
> subroutine flush_buffers
>
> mpp.F90
> #ifdef __aix
> #define FLUSH( FLUSH_(
> #endif
>
> uppercase und lowercase lcase in fms.f90 fms_io.f90 diag_manager.f90 rewritten
>
> topog.F90 layout(2) double definition
>   integer :: jb, jt, il, ir, nx1, ny1, istrt
>
> mkmf improved, my old ideas -WF can be improved.
>
> -D --> -WF,-D
> bind options with field indicees in the scripts with " "
> set MACROS = '-Duse_netCDF
> "-Dij_bounds=iscomp-halo:iecomp+halo,jscomp-halo:jecomp+halo"
> "-D___=iscomp:iecomp,jscomp:jecomp"'
>
> Now the aix branche in mkmf is:
>
> if ( $opt_c ) {
>    if ( $Config{archname} eq "aix" ) {
>       $opt_c .= ' -D__aix';
> #AIX fortran (xlf) requires -WF, in front, comma delimiter, no spaces
>       my $cppdefs_xlf = $opt_c;
>       $cppdefs_xlf =~ s/,/\\,/g; # escape any commas already there
>       $cppdefs_xlf =~ s/-D/-WF,-D/g; # replace -D with -WF,-D
>       $cppdefs_xlf =~ s/\"-WF,/-WF,\"/g; # "-WF, with -WF,"
>       &print_formatted_list("CPPDEFS_XLF = $cppdefs_xlf");
>       $compile_cmd{'.F'}   = q/$(FC) $(CPPDEFS_XLF) $(FFLAGS) -c/;
>       $compile_cmd{'.F90'} = q/$(FC) $(CPPDEFS_XLF) $(FFLAGS) -c/;
>    }
>    &print_formatted_list("CPPDEFS = $opt_c") if $opt_c;
> }
> The script to generate gridgenaror works with this mkmf.
>
> namelist topog : no space!
>
> Now I can compile the gridgenerator and MOM4. Anyway, I have to remove most of the
> entries from diag_table. Otherwise I get messages, that variables are not defined
> as input variables. Now the code runs over this hill but breakes down reading the
> grid:
>
> FATAL: NETCDF ERROR: Invalid dimension id or name
>
> My gridfile and that on the server have the same size and structure.
> Why do I need OCCAM_1degree.nc to generate this file? Test1 should have a flat
> bottom. Any idea where to continue?
>
> Greetings and a nice weekend,
> Martin
>


-------------------------------------------------------
This SF.NET email is sponsored by:
SourceForge Enterprise Edition + IBM + LinuxWorld = Something 2 See!
http://www.vasoftware.com
_______________________________________________
FMS-mom4 mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/fms-mom4

Reply via email to