[Mono-list] SciMark2 benchmark in F#
Jon Harrop
jon at ffconsultancy.com
Tue Jan 20 19:43:49 EST 2009
On Tuesday 20 January 2009 23:02:55 you wrote:
> The scimark benchmark posted here is a *lot* more useful because it's the
> exact same code running on different platforms and versions of the .NET
> framework.
Indeed.
Actually, I had applied some simple low-level optimizations to the F# code to
make it slightly faster on .NET. This is a (small) source of bias so perhaps
I should revert to the simplest unoptimized version. Variations might also be
interesting, e.g. doing the FFT over complex numbers represented as structs.
If you like, I can probably dig out or make up some more F# benchmarks. In
fact, I have an F# version of my ray tracer benchmark.
Also, you may like to write some C# benchmarks based upon the freely-available
linear algebra library Alglib:
http://www.alglib.net
I benchmarked their eigensolvers recently and was pleasantly surprised to find
that they are only ~4x slower (on .NET) than vendor tuned libraries like the
Intel MKL. Hugely impressive, IMHO. These would make good benchmarks because
the hot paths are relatively long.
--
Dr Jon Harrop, Flying Frog Consultancy Ltd.
http://www.ffconsultancy.com/?e
More information about the Mono-list
mailing list