Re: building netcdf-3 on CRAY T90

> The results of my initial attempts at that were in an earlier mail
> (starting to pile up, isn't it?).  The "make all" step had a problem
> (probably unimportant) in that it tried to access "fortran" - here's
> the "make all" log:
>

> sh-56 sh: fortran: not found.
> Make: "cd fortran && \
>     echo "Making \`all' in directory `pwd`" && \
>     echo "" && \
>     make all || exit 1": Error code 1
> Make: "subdir=`dirname fortran/all`; \
> target=`basename fortran/all`; \
> make SUBDIR=$subdir TGET=$target subdir_target": Error code 1
> cmd-2436 make: Stop.
> ====================

You're right, this is an oversight in the Makefile.
We are working on netcdf-3 fortran here...

> I figured this was just an oversight in the Makefile and went ahead
> with the "make test" and got this:
> ===================
> Making `test' in directory
/t90home1/b/jps/unidata.ucar.edu/pub/netcdf/netcdf-3.2a/src/libsrc
>
>         c89 -c    -DDEBUG t_nc.c
> FORCE: No action was taken.
>         c89   t_nc.o libnetcdf.a  -o t_nc
>         ./t_nc
> Assertion failed: csp->mant & 0x800000000000, file ncx.c, line 1388
> Make: "./t_nc" terminated due to signal 6 (core dumped)
> Make: "cd libsrc && \
>     echo "Making \`test' in directory `pwd`" && \
>     echo "" && \
>     make test || exit 1": Error code 1
> Make: "subdir=`dirname libsrc/test`; \
> target=`basename libsrc/test`; \
> make SUBDIR=$subdir TGET=$target subdir_target": Error code 1
> cmd-2436 make: Stop.
> [1]  + Exit 1               make test >& make_test.log
> ===================

This error indicates that 'configure' thinks this is NO_IEEE_FLOAT
If it is IEEE, delete the line in ncconfig.h that #defines NO_IEEE_FLOAT and
recompile. We'll try to figure out why configure doesn't think that it is
IEEE later...


>
> > In order to achieve your original goal of better fill performance,
> > change NFILL in putget.c from 16 to (NC_PG_CHUNK/8). (I'll make this easier
> > to do in a future release.)
>
> Is this a suggestion for 2.4.3 in the absence of the 3.2a C-core?

No. It is for 3.2.

> You mentioned earlier that version 3 already has a lot of optimization in
> it, but I do see that the 3.2 version of putget.c has NFILL set to
> 16....

The default assumes a different balance of space over time than you want.

-glenn