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

Re: 20010821: building netCDF on 64-bit machines

>To: address@hidden
>From: Crystal Shaw <address@hidden>
>Subject: Re:  20010821: building netCDF on 64-bit machines
>Organization: NCSA
>Keywords: itanium, 64-bit platforms

Crystal,

> You have been very helpful! In the nc_test directory, got stuck in the
> nc_get_vars_text. Most of the comparisons went thru very fast until it
> hits get_vars_text. Need to work with the system people to get the
> debugger.

Even without the debugger, you can stick print statements in the
function test_nc_get_vars_text(void) in nc_test/test_get.c to see
where it's getting stuck.  By just running "make test" in that
directory after a few times, you should be able to narrow the problem
down to a single loop or a small block of statements.

Another thing you could try is to turn off the optimizer or at least
lower the level of optimization, using a C compiler flag something
like -O0 or -O1 (I don't actually know what compiler flags your
compiler uses).  In my experience, bugs in new compilers often occur
at higher optimization levels, since vendors are initially trying to
achieve speed at the expense of correctness.  You might just try
compiling libsrc/putget.c and nc_test/test_get.c with no optimization
at all and see if the problem goes away, or check other compiler flags
to see if there's any special optimizations or shortcuts you can turn
off in compiling one or both of those files.

> The nctest directory is core dumping, the last print out is 
>  Testing ncvarputg. It seems related to the nc_test problem???

Yes, the nc_test tests are fairly rigorous tests of the netCDF-3
interface.  The nctest directory tests the netCDF-2 interface, which
is built on top of the netCDF-3 interface, so if the nc_test test is
failing, the nctest test would probably fail too.  The fact that
nctest is failing makes it more likely the problem is in compiling
libsrc/putget.c, since that would be used by nctest also.

> The fortran/test and f90/test got many errors, does it mean the library
> wasn't compiled correctly? Will try to compile the library again.

Since the Fortran interface is just a thin layer over the netCDF-3 C
interface, that would also likely fail if libsrc/putget.c were
incorrectly compiled.

> I know it is like walking in the dark, any suggestions are appreciated.
> 
> IF I could get an account on the system, would you or someone else be
> able to help us port, so we could get wrf going there, assuming there
> is still going to be a supercomputing conference.

If after trying the above you are still having troubles with the port,
and we had some reason to believe it's not a compiler bug, we could
try to help isolate the problem.  But debugging a new compiler for the
vendors can be a great sink of time ...

--Russ

_____________________________________________________________________

Russ Rew                                         UCAR Unidata Program
address@hidden                     http://www.unidata.ucar.edu