[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 19990723: NetCDF conversion problems on Crays

Hi John,

[Russ asked me to reply to you.]

> To: address@hidden
> cc: address@hidden
> From: address@hidden (John Sheldon)
> Subject: NetCDF conversion problems on Crays
> Organization: .
> Keywords: 199907231930.NAA00955

In the above message, you wrote:

> A couple more netCDF conversion questions on Crays that are probably
> already known, but I'd like to verify...
> 
> Here at GFDL we have a Cray T90 
> ('uname -a'=sn7005 sn7005 10.0.0.2 roo.8 CRAY TS)
> and a T3E 
> ('uname -a=sn6602 t3e 2.0.3.32 unicosmk CRAY T3E)
> 
> We've noticed that the following conversions don't work:
> 
> - - on the T90, FORTRAN DOUBLEPRECISION (ie, 128-bit) is garbled in
>   the netCDF file.  This is not too big a deal, since almost nobody
>   uses DOUBLEPRECISION.  Still...

The netCDF package's largest floating-point value is a 32-bit, IEEE
float---so 128-bit floating-point values are out-of-the-question.
Sorry.

> - - on the T3E, INTEGER*4 (stored internally as 32-bits) is garbled,
>   and generally provokes an error from netCDF, providing (thankfully)
>   a valuable reminder not to do that.  But suppose someone wanted to...
> 
> - - on the T3E, REAL*4 (also stored internally as 32-bits) is garbled.
>   This is more problematic, since we do have some codes that use REAL*4,
>   and this seems less likely to generate a tell-tale error.
> 
> So, am I interpreting this right?  Are there any plans to include
> these conversions in some later release?

In order to answer these questions, I'll need to know more about the
Cray environments.  Would you please send me the output from executing
the netCDF-package configure script on the two systems as well as the
files:

    libsrc/ncconfig.h
    fortran/nfconfig.inc

for the two systems.

--------
Steve Emmerson   <http://www.unidata.ucar.edu>