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

Re: 980205: Error message... configure fails on CONVEX

> Subject: Re: 980205: Error message... configure fails on CONVEX
> Glenn P. Davis wrote:
>
> > >To: address@hidden
> > >From: Juan Antonio Prego Ortiz <address@hidden>
> > >Subject: Error message...
> > >Keywords: 199802051250.FAA03966
> > >ConvexOS abaco.cedex.es C4620 11.1 convex
> > >netcdf-3.3.1
> >
> > For some reason which I have not been able to determine,
> > the configure script is reporting that your machine does not
> > use IEEE floating point. I'm almost certain that it does.
> >
> > If it is the case that this machine uses IEEE floating point,
> > you can comment out the
> > macro
> > #define NO_IEEE_FLOAT
> > in libsrc/ncconfig.h
> > to manually override this.
> >
> > If the machine is not IEEE, then there is more work involved.
> >
> > If the machine uses CRAY1 floating point format,
> > most of the work is done. You can start with adding
> > #define _CRAY to ncx.c to turn on CRAY floating point conversions.
> >
> > The only other machine floating point format supported is VAX,
> > which is unlikey to be of help in this situation.
> >
> > -glenn
>
>     Dear Glenn:
>
>     Your suggestion has been succesful for compiling the code. Now we have
library
> which works. But a second problem appeared: the netcdf files are not machine
> independent anymore.
> This is the ncdump output on the CONVEX:
>
> netcdf 8202con_atm_50 {
> dimensions:
>         x = 51 ;
>         y = 52 ;
>         z = 4 ;
> variables:
>         double nav_lon(y, x) ;
>                 nav_lon:units = "degrees_east" ;
>                 nav_lon:long_name = "Longitude" ;
>         double nav_lat(y, x) ;
>                 nav_lat:units = "degrees_north" ;
>                 nav_lat:long_name = "Latitude" ;
>         float mask(y, x) ;
>                 mask:units = "-" ;
>                 mask:long_name = "Land-sea mask" ;
>                 mask:short_name = "mask" ;
>                 mask:missing_value = 123456.8f ;
>         double orog(y, x) ;
>                 orog:units = "m" ;
>                 orog:long_name = "Orography" ;
>                 orog:short_name = "orog" ;
>                 orog:missing_value = 123456.789 ;
>
>
> and this is the ncdump output in the HP workstation:
>
> etcdf 8202con_atm_50 {
> dimensions:
>         x = 51 ;
>         y = 52 ;
>         z = 4 ;
> variables:
>         double nav_lon(y, x) ;
>                 nav_lon:units = "degrees_east" ;
>                 nav_lon:long_name = "Longitude" ;
>         double nav_lat(y, x) ;
>                 nav_lat:units = "degrees_north" ;
>                 nav_lat:long_name = "Latitude" ;
>         float mask(y, x) ;
>                 mask:units = "-" ;
>                 mask:long_name = "Land-sea mask" ;
>                 mask:short_name = "mask" ;
>                 mask:missing_value = 493827.2f ;
>         double orog(y, x) ;
>                 orog:units = "m" ;
>                 orog:long_name = "Orography" ;
>                 orog:short_name = "orog" ;
>                 orog:missing_value = 493827.156 ;
>
>     As you can see the missing_value is 4 times larger on the HP than on the
> CONVEX. We have noted that this factor appears in the standard binary files
as
> well. Could you give us a hint on how to make NETCDF files machine
independent
> again?
>
>     Thanks in advance, Juan

Juan:

Just to be scientific,
we will need to isolate whether this is a problem with the netcdf library,
or with ncdump, and whether the problem is on the Convex or the HP.

You say you were sucessful in compiling the library on the Convex.
Were you able to successfully run the netcdf tests on the Convex?
This is done by running 'make test' in the src directory.
The tests in libsrc an nctest partial check that the binary form
of the files is what it should be, and there is extensive cross checking
of ncdump/ncgen.

Looking at your symptoms, I'm wondering if the system problem that caused
'configure' to fail is causing some other problems.

Can you send me a copy of the following files from your system
/usr/include/float.h (if it exists)
/usr/include/limits.h

-glenn