Re: [DotGNU]CVM to machine code

[Paolo Molaro]

On 07/26/03 Rich Baumann wrote:
> On Sat, 2003-07-26 at 14:19, Paolo Molaro wrote:
> > Total jit time for pnetmark is 0.26 seconds, for a test that lasts about
> > 50 seconds: it's irrelevant.
> 
> I get different results. Using today's daily mono tarball and a fresh
> from cvs pnet I get the following pnetmark results:
> 
> $ time ./mono ~/dotgnu/pnetmark-0.0.6/src/pnetmark.exe

You didn't enable all the optimizations (use the -O=all switch to
mono). It's like benchmarking the C code produced by gcc without using
-O2. You should also always run several times (at least 3) to factor out
interference from other programs running on the system (that should be
idle).

> real    0m25.214s
> user    0m23.330s
> sys     0m0.060s
[...]
> real    0m24.198s
> user    0m23.630s
> sys     0m0.070s

See what Rhys said, or read the sources, total time is meaningless for
pnetmark, it's supposed to be almost the same. Also, if you run multiple
times, it varies a lot, ilrun goes from 48 to 54 for me, for example.

> > > 5 flavours of ilrun builds , your perfomance might vary.
> > 
> > Mono is about 3 times faster than the fastest ilrun build, the others
> > are 5 or more times slower.
> 
> 
> 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.

Nope, read above or Rhys' explanation, the execution time is supposed to
be mostly the same.

lupus

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