[Mono-devel-list] Performance w/ boxing

[Ben Maurer]

Well, given that no objects are promoted past gen0 with the MS GC, I am
not sure how theirs is any better. (in a real life test, where there
were other allocations, their GC would obviously have an edge). Then
again, I will admit I do not know much about GCs ;-). 

Those optimizations you brought up sound interesting. I would love to
hear more about them. Could you give some pointers as to what exactly
the problems are and how one would go about coding the solution? Also,
would the effects here be seen globally?

-- Ben



>>> Paolo Molaro <lupus at ximian.com> 02/19/04 10:31 AM >>>
On 02/19/04 Ben Maurer wrote:
> Today I posted a blog entry about the performance of boxing. Someone
> posted that MS was a few times faster on the test. My first reaction
was
> that this was a GC problem, however given the nature of the test case,
> the generational GC does not give MS much of an advantage. I am not

You obviously have no idea of how a generational GC works.

> I would be interested if any of the runtime guys have any ideas about
> how to speed this up. The code is below:

Of course we have ideas, we actually think about the cause of
performance
issues.
Do you want a 50+% speedup over the current time? Use a generational GC.
A 10% speedup? Make stack unwinding work across the managed->unmanaged
boundary with any compiler.
Another 5-10% speedup? Get the jit to produce better code (especially
better register allocation).

lupus

-- 
-----------------------------------------------------------------
lupus at debian.org                                     debian/rules
lupus at ximian.com                             Monkeys do it better
_______________________________________________
Mono-devel-list mailing list
Mono-devel-list at lists.ximian.com
http://lists.ximian.com/mailman/listinfo/mono-devel-list