Hi Steve and all other mom4 users,
In response to Steve's email,
I have successfully ran test cases 1 and 2 on a SGI Origin 3400 and on a
Compaq alpha ES45 system.
For the SGI
I was looking at running a different levels of optimisation, which
resulted in a SGI compiler bug report, to do with the
mom4_beta2/shared/diag_manager/diag_output.f90 file, for which the below
is a code fragment.
If you have two files, sub.f90 and main.f90, containing
sub.f90
=============================================================
module sub_mod
public :: diag_output_init
contains
subroutine diag_output_init ( time_id )
integer , intent(out) :: time_id
time_id = 15
print *,'JLR !!!!! ',time_id
end subroutine diag_output_init
end module sub_mod
main.f90
============================================================
program main
use sub_mod, only: diag_output_init
type file_type
character(len=128) :: name
integer :: time_axis_id
end type file_type
type(file_type) :: files(1)
CALL diag_output_init ( files(1)%time_axis_id )
end program main
The code works OK for -O1 -IPA, but NOT for -O2 -IPA
This resulted in a compiler bug report, with the following response,
Summary: When IPA inlines the procedure diag_output_init
into main, the parameter time_id transformed into
an internal representation that the optimizer is
not expecting. Consequently, alias analysis fails,
and the dead code eliminator deletes "time_id = 15".
Workaround: Compile with the flag
-INLINE:never=DIAG_OUTPUT_INIT.in.SUB_MOD
to prevent IPA from inlining this procedure.
The engineer working the bug hopes to have a fix in the
7.4.1m compilers (note that the 7.4 compilers are not
released yet).
I havn't checked the accuracy at running at this level of optimisation,
just that the code will compile and run.
For the Compaq ES45 system,
Test 1 ran ok, gave results essentially the same as the SGI results.
Test 2 ran ok, the results differ, even the taux and tauy values differ, I
am still working on this.
I had to make some modification to the source code, for problems with
1) multiple definitions of the same variable
2) some elements on namelists being declared private
3) problems with an integer being declared after it was used to declare an
array (if move the integer :: nt statement up a few lines in the source
code).
I also had problems with the optimistion levels with two files,
mpp_domains.F90 and fms_io.F90
The default optimisation level of the compaq compiler is O4, but these
routines will not work if compiled above O1.
I also had trouble getting the udunits library compiled correctly
(compiled, but gave erronous results). I needed to change the file
udunits-1.11.7/src/port/misc/stdlib.h line 6 from
# define UD_SYSTEM_STDLIB_H "///usr/include.dtk/stdlib.h"
to
# define UD_SYSTEM_STDLIB_H "///usr/include/stdlib.h"
I can email anyone interested the changes to the source code and a
template mkmf.template file for the compaq system.
Cheers
Jason
Dr Jason Roberts
TPAC
GPO Box 252-37, Hobart, Tasmania, 7001, AUSTRALIA
Ph: (03) 6226 2990 Int'l Ph: +61 3 6226 2990
Fax:(03) 6226 2973 Int'l Fax:+61 3 6226 2973
Email: [EMAIL PROTECTED]
On Fri, 24 Jan 2003, Stephen Griffies wrote:
>
> Hi,
>
> To follow-up on Harper's email, I wish to poll the mom4-users to know
> who is successfully running mom4-beta1 or mom4-beta2 experiments on
> their machines, even SGIs, and who has simply accepted defeat for the
> meantime? Please reply to this email list with your experiences, either
> good or poor, with as much detail as you can.
>
> As you know, developing a modern climate model code that can run on
> various platforms is tough. To better focus our efforts, we desperately
> need to hear from the user community to gauge the success, or lack, of
> your efforts to port mom4/FMS to your machine. Any and all of your
> input will be valuable as we aim to release more flexible code.
>
> I look forward to hearing from you (PLEASE reply to this email-list
> rather than solely to me).
>
> Thanks,
> Steve
>
> --
> Dr. Stephen M. Griffies phone: +1-609-452-6672
> Geophysical Fluid Dynamics Laboratory FAX: +1-609-987-5063
> P.O. Box 308, Forrestal Campus Site B email: [EMAIL PROTECTED]
> Princeton, NJ 08542 USA http://www.gfdl.noaa.gov
-------------------------------------------------------
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
