Test of NXcompress

[Ray Osborn]

As you can tell from the CVS messages, I've added NXcompress to the Fortran
90 interface.  I've now tested NXcompress using all three interfaces on
Alpha/VMS and Linux, using Absoft Pro Fortran.  Actually, it's impossible to
tell whether the compression is really working using the NAPI test routines,
since the compressed array is so small that it doesn't make much difference
to the overall size.  However, it runs without any errors.   More
significantly, I wrote a LRMECS data file of about 0.4MB using all three
compression algorithms.  Both the LZW and Huffmann options reduced the file
by a factor of 3 while RLE only reduced it by just over 5%.  I didn't do any
timing tests, but I didn't notice a significant increase in completion time.
It would be interesting to hear what the time difference is with much larger
files.

I noticed a minor inconsistency in the external references to NXfcompress in
NAPIF.F and its name mangling in NAPI.H, and I've added NX_COMP_NONE to the
parameter list to be equal to COMP_CODE_NONE as defined in HDF.  Otherwise,
I think we have a working compression scheme.  However, I still need to add
default compression routines to NXUmodule.f90 before this version is
complete.

Any other experiences?  Thanks to Mark for adding this so smoothly.

Ray
-- 
Dr Ray Osborn                Tel: +1 (630) 252-9011
Materials Science Division   Fax: +1 (630) 252-7777
Argonne National Laboratory  E-mail: ROsborn at anl.gov
Argonne, IL 60439-4845