Re: 19991202: Fortran interface failure on Nec SX5

Marc,

> To: address@hidden
> cc: address@hidden
> From: Marc Guyon <address@hidden>
> Subject: netcdf on Nec SX5?
> Organization: .
> Keywords: 199912021756.KAA15510

In the above message, you wrote:

> I'm trying to build the netcdf library on our new platform, a Nec
> SX5... I find on your site the recommendation for the Nec SX4 and I
> have tried to adapt them :
> 
> 1. To build a library with real and integer on 4 bytes, I use :
> 
>     CC=cc
>     CFLAGS=-h2
>     FC=f90  
>     FFLAGS=-dw -dW
> 
> 2. To build a library with real on 8 bytes and integer on 4 bytes, I use :
> 
>     CC=cc
>     CFLAGS=-h2
>     FC=f90
>     FFLAGS=-dw -dW -Wf"-A dbl4"
> 
> 3. To build a library with real and integer on 8 bytes, I use :
> 
>     CC=cc
>     CFLAGS=-h2
>     FC=f90
>     FFLAGS=-ew
> 
> I have compiled with many warning and specially very worried message 
> such as :
> 
> "fort-v2compat.c", line 616: warning: improper integer precision : op "="
>  
> It seems that I have a lot of troubles when an instruction involves
> different kind of types... But, in the 3 cases, I succeeded to obtain
> the libnetcdf.a, libnetcdf++.a, ncgen, ncdump and nctst... The ftest,
> t_nc, nctest, nc_test work well, like the test programs of ncdump
> and nctst... But, I have a lot failures with nf_test that tests the
> FORTRAN interface.
>
> In the cases (1) and (2), I have theses messages :
> 
> 
> ********************************my messages******************************
> 
> *** Testing nf_get_var1_int1 ...
>  Range error: No error
>  Range error: No error
>  Range error: No error
>  Range error: No error

Failure of the nf_test/test target indicates that the Fortran interface
probably will not work correctly.

> 
>    ###   1844   FAILURES TESTING nf_get_var1_int1! ###
> *** Testing nf_get_var_int1 ...
>  Range error: No error
>  Range error: No error
>  Range error: No error
>  Range error: No error
> 
> 
>    ###   146   FAILURES TESTING nf_get_var_int1! ###
> *** Testing nf_get_vara_int1 ...
>  Range error: No error
>  Range error: No error
>  Range error: No error
>  Range error: No error
> 
> 
>    ###   496   FAILURES TESTING nf_get_vara_int1! ###
> *** Testing nf_get_vars_int1 ...
>  Range error: No error
>  Range error: No error
>  Range error: No error
>  Range error: No error
> 
> 
>    ###   690   FAILURES TESTING nf_get_vars_int1! ###
> *** Testing nf_get_varm_int1 ...
>  Range error: No error
>  Range error: No error
>  Range error: No error
>  Range error: No error
> 
> 
>    ###   966   FAILURES TESTING nf_get_varm_int1! ###
> *** Testing nf_get_att_int1 ...
>  Range error: No error
>  Range error: No error
> 
> 
>    ###   4   FAILURES TESTING nf_get_att_int1! ###
> *** Testing nf_set_fill ...
>  Total number of failures:   4146
> 
> 
> In the case (3), I have 6412 failures...
> 
> ************************end**************************************
> 
> On your support email archive, I find some notes on how building 
> netcdf on SX4:
> 
> ************************your archive******************************
> 
>   Russ Rew                                         UCAR Unidata Program
> address@hidden                     http://www.unidata.ucar.edu
> 
> 
> SUPER-UX unix 7.2  SX-4 (sx4.hstc.necsyl.com)
>     Wed Sep 17 13:30:59 MDT 1997
>     /home/cust/ncar1/netcdf-3.3.2b/{lib,lib_float2}
> 
>     CC="cc -Xa"
>     CFLAGS=-h2
>     FC=f90
>     FFLAGS=""
>     CXX=""
> 
>         Passes libsrc/t_nc.
>         Passes nctest.
>         nc_test:
>                 30 failures in nc_test for _long functions.
>                 Okay to ignore: cpp problem sets up improper ranges for test.
>                 (cpp problem reported to NEC)
>         fortran/ftest works.
>         Compiling nftest: Many (ignoreable) warn(82)
>         nf_test:
>                 6435 failures
>                 get_int1: Range error: No error
>                 get_int2: Range error: No error
>                         Okay to ignore.
>                 put_real: Var value read not that expected
>                         ???
>         Passes ncdump test.
>         Minor roundoff errors in test of ncgen.
>         <               i:d = -1.e+308, ...
>         ---
>         >               i:d = -9.99999999999999e+307, ...
> 
>         There are SX specific hacks in nc_test and ncx.m4 to deal with
>         the broken C preprocessor.
> 
>         
>     Cray compatibility options:
>     CC="cc -Xa -hfloat2 -hint64"
>     CFLAGS=-h2
>     FC="f90 -float2"
>     FFLAGS=""
>     CXX=""
>     FLDFLAGS="-Wl -int64"       # needed for the interlanguage linkage
>     LIBS=-lm                    # needed for ie3_fl2(), fl2_ie3()
> 
>         Passes libsrc/t_nc.
>         Passes nctest.
>         Passes nc_test.
>         Passes fortran/ftest.
>         nf_test:
>                 get_int1: Range error: No error
>                 get_int2: Range error: No error
>                 Total number of failures:   6412
>            Okay to ignore - int1 and int2 types are integer
>         Minor roundoff errors in test of ncdump.
>                 <   7, 8, 9.99999999999989e+29 ;
>                 ---
>                 >   7, 8, 9.99999999999977e+29 ;
>         Minor roundoff errors in test of ncgen.
>                 < i:d = -9.99999999999881e+307, 0., 9.99999999999881e+307 ;
>                 ---
>                 > i:d = -9.9999999999975e+307, 0., 9.9999999999975e+307 ;
>                 ...
> 
> ***********************end of your archive**********************************
> 
> 
> I have the same message with ncgen!!!
> 
> 
> So, I would like to know your feeling about the result of the test of
> ncgen and about the number of failures of the nf_test?...

The "ncgen" errors can probably be safely ignored.  The "nf_test" errors
indicate serious problems with the netCDF-3 Fortran interface.

> It seems
> that I have the same number of failure that in the case of Cray
> compatibility options? Do you know if somebody has succeeded to reduce
> the number of failures on Nec computer? what is your opinion about
> the reliability of my libnetcdf.a and ncgen, maybe the people who
> has installed netcdf on the SX4 have observed the same messages then
> netcdf has been used without any problem for application?

I don't know anyone who uses the netCDF package on an SX4.  I'm not
saying that they don't exist (I'm sure they do) only that I don't know
them.

In order to track down the cause of your problem would you please do the
following:

    1.  To to the top-level source directory.

    2.  Perform steps 3 through 8 near the end of the file INSTALL for 
        your "case 1" above.  Where the sending of a file is indicated,
        please send them directly to me.

> many thank for your attention
> Marc Guyon
> 
>                        _\|//-
>                        ( 0-0 )
> +-------------------o00--(_)--00o-----------------------------+
> |                                                             |
> |                   Marc Guyon   IDRIS                        |
> +------------------------------+------------------------------+
> | Institut du Developpement &  | Batiment 506                 |
> | des Ressources en            | BP 167                       |
> | Informatique Scientifique    | F 91403 Orsay Cedex          |
> | Equipe Support Utilisateurs  | France                       |
> +------------------------------+------------------------------+
> | Phone:  + 33 1 69 35 85 68     Fax: + 33 1 69 85 37 75      |
> |                                                             |
> |                    mailto:address@hidden                    |
> |                                                             |
> |                   .oooO                                     |
> |                   (   )   Oooo.                             |
> +--------------------\ (----(   )-----------------------------+
>                       \_)    ) /
>                             (_/

--------
Steve Emmerson   <http://www.unidata.ucar.edu>