Re: [DotGNU]CVM to machine code

[Paolo Molaro]

On 07/27/03 Gopal V wrote:
> > I'm just not seeing that. What I am seeing is that it's equal in
> > execution time. Of course, that was what Gopal was talking about.
> 
> I think Paolo might be testing against an outdated version before the
> the generic unroller , interface table dispatch and indirections removal
> for callvirt . The new release ie 0.5.10 was released only today ...
> *Oof*, I almost believed the 3 times faster comment .. :)

As said in my mail I used the cvs version from yesterday. 
So, the 3x comment still stands, just from Rich's
numbers it's 2 times faster and it was run without optimizations.
The actual number may change with the cpu, I guess it can change from
2.7 to 3.3 times faster, since mono can take advantage of cpu-specific
instructions, so if your processor has them it would be in the upper
range, other wise it can be *just* 2.7-2.8 times faster than ilrun.

> So CVM delivers the first output fast and cumulative favours the JIT ...
> Hmm.. but I had thought mono would be faster in the total execution
> for pnetmark as well ... 8686000 calls to a single method and that 
> sort of thing.

Please read the pnetmark sources or Rhys comments, you apparently don't
know how that benchmark works.

lupus

-- 
-----------------------------------------------------------------
address@hidden                                     debian/rules
address@hidden                             Monkeys do it better