This directory contains a simple system for collecting and displaying data
on the performance of various linear solver methods on some problems.
This is the approach:

    The program uses the standard arguments for solvers, accellerators,
    etc.  As long as the program uses these standard arguments, it can
    be used with this script

The program must generate output using the routines SVDBHeader() and
SVDBHistory(); these generate output that this program can understand.
The example programs use these and may be used with this script.  The
file SVDBOpen() opens the appropriate files.

By default, SVDBOpen used "-fdb name" for the directory file (see below)
and "-fdbres name" for the residual file.  "-fdbtitle string" specifies
a description string for the run (a default string will be supplied 
otherwise).

The "database":
Since there is no standard database system for Unix systems, we need to 
provide some simple substitute.  The system used here is to use a directory
to contain all of the data.  This directory contains a "directory" file that
contains one line for each run, holding a description and a summary of the
run, and the file name of the file that contains the residual history.
This allows us to use grep and awk to search for runs to graph/print.

The comparisons:
We want to combine the following queries
   1 or more accellerators
   1 or more preconditioners
   1 or more problems/size
We want to filter these to
   Sort on iterations, time, final residual (and perhaps take top 10)
   Select range of iterations, time, final residual
   Also, we'd like to aggregate these (to form, say, a pie chart of 
   percentage of time successful).

Finally, we want to generate output for other tools, including
matlab format lists.
