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

Re: [Fwd: Re: NetCDF for platforms with 8-byte floats and doubles]

>To: address@hidden, address@hidden
>From: Helen Fairhurst <address@hidden>
>Subject: Re: 20021114: NetCDF for platforms with 8-byte floats and doubles
>Organization: UK Met Office
>Keywords: Cray T3E

Helen,

> Matt Huddleston has also been in touch with you recently 
> about installing NetCDF on our Cray T3E. Just reporting 
> in to say that we think we have now overcome the problems 
> we were experiencing with NetCDF and 8-byte reals on the T3E. 
> 
> For future information, this is what we did 
> 
> - edited "f90/netcdf_overloads.f90" to remove all 
> FourByteReal definitions (as per Steve's suggestion) 
> 
> - made sure we had the copy of "f90/netcdf_test.f90" 
> from http://www.unidata.ucar.edu/glimpse/netcdf/4454   
> Then edited "f90/netcdf_test.f90" to change all FourByteReal 
> declarations to EightByteReal (for fillVal and validRange) 
> and change the definition for "P" to nf90_double instead of 
> nf90_float. [Perhaps all the definitions of nf90_float and/or 
> nf90_real in "f90/netcdf_test.f90" should be changed to 
> nf90_double, but it seems to work OK with just this.] 
> 
> We also found that we should NOT use "libsrc/ncx_cray.c" 
> (this is only for Cray vector architectures). In our final 
> build we included the version of "f90/netcdf_variables.f90" 
> that deals with the nf90_Inquire_Variable(...) issue mentioned 
> on the known problems page for the current release.
> 
> 
> Following this, the results from f90/test only differed in 
> the following respects (in line with the edits above) 
> 
> $ 
> $ diff f90/example_good.cdl f90/example.cdl
> 8c8
> <       float P(frtime, lat, lon) ;
> ---
> >       double P(frtime, lat, lon) ;
> 11,12c11,12
> <               P:valid_range = 0.f, 1500.f ;
> <               P:_FillValue = -9999.f ;
> ---
> >               P:valid_range = 0., 1500. ;
> >               P:_FillValue = -9999. ;
> $ 
> 
> 
> All other tests were fine. Matt will be running some further 
> tests against "real" data to check that everything is OK. Matt 
> and I would like to thank you both very much for your help with 
> this. Many thanks and best regards, 

Thanks very much for sending the above info, and sorry it has taken so
long to acknowledge your work.  Now that we know this works, we'll try
to get it into our web pages and the next release of the software, so
other T3E users will have an easier time with it and can benefit from
your (and Matt's) contributions.

--Russ