[Mono-devel-list] Re: Performance Test

[Paolo Molaro]

On 07/01/03 David Jeske wrote:
> My apologies, I accidentally typed "mint" instead of mini, which led
> me to mis-speak.  Let me re-state: "mint core-dumps trying to run the
> program". I know mint has some issues, is it going to be working again
> someday, or is it going to die in favor of encouraging people to port
> mini?

mint should be considered just as an aid in getting familiar with the
mono runtime and a port to new architecture. Using it to run benchmarks
is an exercise in futility, IMHO, but that obviously won't stop
people:-) That said, I'm going to do only minimal maintainance to mint,
since it doesn't make much sense to use it when you have a JIT
available. I should also note that Bernie Solomon has patches to speed
it up and I guess several fixes, too for his port to hppa: we're just
waiting for him to submit them for review and inclusion in cvs:-)

> C:\neo\scriptperf>mono scriptperf.exe

Note: not all the optimizations are enabled by default yet, so you might
want to run mono with the -O=all option (you need to put it in quotes
under windows, because the ugly shell gets confused by the '=').
-O=all gives better results for all the tests except index_test and
index_string_test: I haven't investigated the generated code on that 
one yet, it may be related to boxing or some code inside Hashtable.
I also just committed a change to cvs that speeds up boxing operations:
this results in better numbers from 5 to 10 % for index_test and
index_string_test for me. On my PII 400, index_test ran with mono is
much closer to the MS runtime speed, dunno why its almost 2x slower for
you, since the 3x difference is index_string is consistent with what I
get.

lupus

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