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

20031212: netCDF beta with ifc and icc



Timothy,

>Date: Fri, 12 Dec 2003 15:06:31 -0600
>From: Timothy J Campbell <address@hidden>
>Organization: PET Component One
>To: Steve Emmerson <address@hidden>
>Subject: Re: 20031212: netCDF beta with ifc and icc

The above message contained the following:

> Thanks for the quick response.  The "O"s you referred to in FFLAGS and 
> F90FLAGS were typos in the email.  I have tried using the "-Vaxlib" and 
> "-dynamic-libcxa" options.  There is no difference in the compile or tests.
> 
> It seems everything is compiling fine.  The failures in the testing all 
> have to do with putting and getting floats.  For example,
> 
> *** Testing nc_put_var_float ...
>          FAILURE at line 2831 of test_put.c: Numeric conversion not 
> representable
>          FAILURE at line 2878 of test_put.c: Numeric conversion not 
> representable
>          FAILURE at line 697 of test_put.c: Var value read not that expected
> varid: 9, var_name: ir, index: 1, expect: 2.14748e+09, got: -2.14748e+09
>          FAILURE at line 697 of test_put.c: Var value read not that expected
> varid: 21, var_name: i2, index: 1, expect: 2.14748e+09, got: -2.14748e+09
>          FAILURE at line 697 of test_put.c: Var value read not that expected
> varid: 27, var_name: i3, index: 1, expect: 2.14748e+09, got: -2.14748e+09
>          FAILURE at line 697 of test_put.c: Var value read not that expected
> varid: 33, var_name: i4, index: 1, expect: 2.14748e+09, got: -2.14748e+09
> 
> I am unable to make any sense from the "Numeric conversion ..." message. 

The "Numeric conversion" messages from the compiler have to do with
interpreting a floating-point literal-value in the C code that is close
to or at the limit of values representable by the compiler (we're
talking really big values here).

> As for the values written/read something is simply causing a sign 
> change.  Based on these failures do you have any suggestions on what I 
> should examine next?

You've tried everything that I can think of.

It is entirely possible that the netCDF library is OK and that the
"problem" you discovered has more to do with limitations of the compiler
than with correctness of the build.  We've seen this before.  The
"solution" in this case is to try the netCDF library on some of your own
data and in some of your own programs and closely monitor the results.
It it works OK, then it's likely to be OK.

If you want to try the build, then go ahead and do the "make install".

Please let me know what you discover.

> Thanks,
> Tim

Regards,
Steve Emmerson