[Mono-list] Re: Time problems on Mono

[shudo@computer.org]

> From: shudo@computer.org

> > > Perhaps it would be nice to port LINPACK test to C# and see the differences
> > > between runtimes
> >
> > There is a work arround about this: 
> > http://www.shudo.net/jit/perf/
>
> Yes.  I have ported the Java version of Linpack benchmark to C#.
> The porting work was reasonably easy due to similarity between the two
> languages.
>
> Now, the comparison page has the source code.

A guy suggested to me testing a development version of Mono
and I tried.


From: Miguel de Icaza <miguel@novell.com>

> >     Would it be possible for your benchmarks to try also Mono 1.1.3?  I
> > would like to see how our new JIT features stack up.
>
> Also, in Mono 1.1.3, you might want to try our new SSAPRE code:
> '
> 	mono -O=ssapre
>
> Or the standard optimizations:
>
> 	mono -O=all

Mono 1.1.3 results a worse score with `mono -O=all' as follows:


% mono Linpack.exe 1000
Problem size: 1000
Mflop/s: 314.211  Time: 2.13 secs (2.12808227539063 sec)  Norm Res: 8.57  Precision: 2.22044604925031E-16

% mono -O=all Linpack.exe 1000
Problem size: 1000
Mflop/s: 178.26  Time: 3.75 secs (3.75108337402344 sec)  Norm Res: 8.57  Precision: 2.22044604925031E-16

% mono -O=ssapre Linpack.exe 1000
Problem size: 1000
Mflop/s: 312.413  Time: 2.14 secs (2.14032745361328 sec)  Norm Res: 8.57  Precision: 2.22044604925031E-16


The running time (2 or 3 seconds) is rather short on a recent PC.
Did the time for all optimizations take much time for such a short run?

Anyway, now the benchmark page has the source code Linpack.cs and you
can try it by yourself.


  Kazuyuki Shudo
  Grid Technology Research Center
  National Institute of Advanced Industrial Science and Technology (AIST)