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

Re: 20050115:netCDF 64 bit 32 bit & HDF5->netCDF

Lou,

> Hmm, interesting.  I believe these are all big-endian processors.
> So there have been no tests by Unidata of netCDF on 64-bit little-endian
> processors yet (w/ Solaris, SuSE, or RedHat -- which are the only ones
> I know of to date that will run on x86_65).
> 
> But from the rest of your email (below) it sounds like some user
> _might_ have done this already -- at least you've not heard anything
> to the contrary.

I tested on Solaris 10 installed on an AMD Opteron, which I believe is
a 64-bit little-endian architecture.

> My reason for asking is that we are about to try a migration of GMT
> from x86_32 to x86_64 running RedHat Enterprise, and a requirement
> for GMT is to have netCDF installed (since several of GMT's data
> formats are in a simple netCDF format).  NetCDF seems to compile
> AOK on it using gcc (by default -m64), and GMT compiles AOK after
> that.  We haven't tried a test of a netCDF file via GMT yet though.
> I thought I'd ask some stupid questions first! :)  A reasonable test
> might be to create a GMT-netCDF file on x86_32 and see if it read OK
> on x86_64, and then the other way round, right?

Yes, but part of that testing is already done when you install netCDF
and run "make test".  There is a sample binary netcdf file
libsrc/test_nc.sav that "make test" compares bit-for-bit with a newly
created netCDF file written by the test program in libsrc/t_nc.c, to
test binary compatibility.  If there were any difference that
depended on platforms, "make test" would fail in this directory.  But
we know "make test" has run successfully on numerous platforms with
all sorts of architectures, so I'm very confident that it will
work OK on your 64-bit little-endian machine.

--Russ