Sparse benchmark, version 0.9.7, released 17 Nov 2000.
Jack Dongarra, Victor Eijkhout, Henk van der Vorst
copyright 2000

This is a benchmark of Conjugate Gradient methods,
using sparse data storage.

For questions and comments, mail sparsebench@cs.utk.edu.

==== Executive summary ====
Issue the following command:
    configure

Then: 
    Test -m <machine name>
to generate test matrices and run the benchmark on them. After this, you can 
get a performance summary by
    Analyze -m <machine name>
If you are happy with your results, mail them off to UT by
    Report -m <machine name>

==== More options ====
If you have gnuplot and ghostview installed, try, for instance,
    Analyze -m <machine name> -d -c crs
for a plot of the performance of the various crs-storage methods.
This may give you an indication that you are running completely
'in cache', in which case you can run larger problems with
    Test -m <machine name> -s "<bunch of larger sizes>"

You can test more than one machine and store all results by using
    Test -m <name>
    Analyze -m <name>
    Report -m <name>
(don't forget to issue 'configure' for each machine).
If you've tested several machines, compare them with
    Compare
Remark. Every time you run the benchmark, the test matrices 
are generated, which takes time. This amount of time can not be helped.
By default, the matrices are deleted after use.
If you use `Test -d', the test matrices are saved to disk after use.
This saves considerable time if you rerun the benchmark, but it will
cost you a few hundred megabyte.

==== Further reading ====
For more information about running the benchmark codes,
see the file install.ps.
For more information about the concepts behind the benchmark
and a list of published results, see the file bench.ps.

==== Change History ====
0.9.4 : first version put out on netlib
0.9.5 : minor changes in README, Test, Analyze,
        call to 'configure.more' removed.