[MITgcm-support] Re: More problems

Patrick Heimbach heimbach at MIT.EDU
Mon Nov 8 20:20:19 EST 2004


Paola,

looking at your earlier mail, I see that you are ommitting
conversion to big endian, i.e. you only have (corrected)
   DEFINES=' -DWORDLENGTH=1'
whereas we usually(always?) set
   DEFINES=' -DWORDLENGTH=1 -D_BYTESWAPIO'
to convert to big endian (and remain platform independent)
when running the model.

So I think what's happening is that rdmds.m assumes
big endian (default), but you've generated little endian output.
You can change this by setting ieee='l' in rdmds.m
to look at little endian output.

The cleaner way though is to stick to the conventions
and do everything in big endian,
i.e. 
o produce big endian input (for which your open
  statement below contains the correct 'b' parameter),
o run big endian (using -D_BYTESWAPIO)
o happily look at big endian output.
That all works well on the altix.

-Patrick

PS:
I think that ifort 8.X now also has a compiler option
'-convert big_endian' which would allow you to ommit 
the CPP option -D_BYTESWAPIO



Quoting Paola Cessi <pcessi at ucsd.edu>:

> Hi Patrick - sorry to bug you more, but I don't seem to be able to read 
> the output files (.data and .meta).
> 
> Since ORNL does not have matlab I transfer the files to another 64 bits 
> (little endian) machine (an AMD Opteron running LinuxRH8) that has matlab 
> and then read them  using rdmds.m. Is there something I need to do with 
> rdmds in order to read them properly?
> 
> Also, when generating the input files with gendata.m are these the correct 
> format options for the Intel Fortran 8 on an ALtix (which is 64 bits - 
> little endian)?
> 
> fid=fopen('Qnet.circle','w','b'); fwrite(fid,Q,'real*8'); fclose(fid);
> 
> paola
> 
> 


--------------------------------------------------------
Patrick Heimbach   Massachusetts Institute of Technology
FON: +1/617/253-5259                  EAPS, Room 54-1518
FAX: +1/617/253-4464             77 Massachusetts Avenue
mailto:heimbach at mit.edu               Cambridge MA 02139
http://www.mit.edu/~heimbach/                        USA




More information about the MITgcm-support mailing list