## Synthetic, rigged benchmark

To run the benchmarks, make sure you have compiled cyclobench and
installed GAP.

Then run (for example):

    $ ./run_bench.sh 100000 20 200 10

To generate the benchmark data by runnning 100,000 dot products for
each implementation, starting with a field order of 20, up to 200,
with a step of 10. So we'll run a benchmark for each order in 20, 30,
..., 200.

To plot the graphs, you can use:

   $ cd results
   $ ../plot.py -x 0 "\$n\$" -y 1 "time taken (ms)" \
   -f "Antic 0.2.1" antic \
   -f "cyclotomic sparse 0.1.0" sparse \
   -f "cyclotomic dense 0.1.0" dense \
   -f "cyclotomic structure 0.1.0" structure \
   -f "GAP 4.11.0" gap

which specifies the x and y axes, the filenames with the data, and the
legend labels to use for the data coming from each file.

This should generate a PNG file and tex file with the results graphed.

## Real-world benchmark

The better benchmark to run is the character inner product one. First,
generate the test data (irreducible characters and one random
character from one group):

    $ gap -q < char.g > test_data

GAP will also run the calculation and say how long it took:

    GAP time elapsed: 378.918407ms

Now run cyclobench in character mode:

    $ cyclobench character < test_data
    character product benchmark, reading from stdin
    time elapsed (ms):
    219

It will probably take a very long time to read the data (I haven't
optimized the parsing code at all - it works by running a Haskell
program and piping text), but will then print the time taken to run
the actual computation, after parsing.