These benchmarks were written for the post Day 3 of Prolog. Contributions using other libraries or languages are welcome.
First, clone or download this repository to a location of your choice.
Install SWI-Prolog (or your Prolog implementation of choice). The following steps will run the benchmark.
$ cd constraint-programming-benchmarks/prolog/
$ swipl
?- ['sudoku.pl'].
?- bench(1).
You should see something like this when executed:
?- bench(100).
100 iterations taking 7805 msec
true.
Install Scala. The following steps will run the benchmark.
$ cd constraint-programming-benchmarks/scala/
$ sbt
> run
You should see something like this when executed:
> run
[info] 0% Scenario{vm=java, trial=0, benchmark=Sudoku, length=100} 97656735.29 ns; ?=42751912.25 ns @ 10 trials
[info]
[info] ms
[info] 97.7
[info]
[info] vm: java
[info] trial: 0
[info] benchmark: Sudoku
[info] length: 100
[success] Total time: 24 s, completed Aug 6, 2012 3:10:35 AM
Many thanks to the scala benchmarking template and caliper.
You'll notice I'm only doing one iteration of the benchmark when calling bench
and in Benchmark.scala. I was testing with 100 iterations, but I noticed that
the time for the Scala benchmark wouldn't increase when I changed it from 1 to
even 1000 iterations. There must be some kind of optimization the JVM is doing
that either caliper or the scala benchmarking template isn't accounting for, but
solving that is farther than I want to go right now.
I compromised by running the Scala benchmark multiple times (each time it takes the average of 10 runs) and averaging those averages. They all fell within a relatively narrow band, so hopefully it is accurate enough.