[Mono-list] Performance on a Geode-based system

[Paolo Molaro]

On 11/21/02 Fergus Henderson wrote:
> > Most of the benchmarks in mono are as fast as gcc up to about 50%
> > slower. There are a few cases where gcc can optimize code a lot better,
> > but usual application code is just about 0-10% slower with mono.
> 
> I presume you are comparing C# code compiled with mcsc and run with
> mono to C code compiled with gcc.
> 
> My experience with Mercury and Microsoft's .NET CLR implementation is that
> Mercury code compiled for .NET generally runs about 5-10 times slower.
> 
> Your mileage may vary.

Of course, it depends how well the IL code generated matches to the C
code compiled by gcc. C# code matches quite well to equivalent C code, I
guess Mercury needs to generate IL code that gives the same output than
the C code, but that is structured differently.
Or it may be a case where the optimizations done by gcc are really
effective. If you can provide the IL code for a method that is slow with
the CLR versus the same C code, we can see what optimizations can be
added to the jitter (if at all) to make it faster.

lupus

-- 
-----------------------------------------------------------------
lupus@debian.org                                     debian/rules
lupus@ximian.com                             Monkeys do it better