[Mono-list] Java / Mono performance diff (convolution benchmark)
jonathan.shore at gmail.com
Fri Jan 29 17:46:44 EST 2010
In my haste to do the prior benchmark, as Jon Harrop pointed out, there was some dead code (or rather dead work that was not going to be used). I wrote a new benchmark based on real code that I use. In this case a multivariate 1D FIR convolution. So as not to include the kitchen sink of dependencies with my libraries wrote some simple substitutes. Be aware that because of this the code is non-idiomatic C# and the same is true for the Java version. Not trying to win design points with this test.
Again, this algorithm is array bound. But the test has no code that can be optimised away (that I am aware of) unlike the prior test.
The performance results are:
Mono: 2 min, 2 secs
Java: 0 min, 58 secs
The mailer will not allow posts with more than a few K, so will send in a followup message (awaiting moderation).
As I've not been able to build glib on OSX, have not yet been able to try this with the LLVM option.
More information about the Mono-list