[Mono-list] CLI-Grande Test Results

[Ben Maurer]

Hello Marcus,

Thank you, as always, for your interest in performance. I really do
appreciate your dedication.

On Wed, 2004-06-09 at 02:42, Marcus wrote:
> I observed that using full optimization (-O=all) with Mono sometimes made 
> results worse.
This may be due to abcrem. That optimization generally takes alot of
time. I think at one point, we talked about removing abcrem from -O=all.
We should consider this again.

To see the effect of abcrem, try running --compile-all on corlib with
-O=all and -O=all,-abcrem. The difference is quite large.

IMHO, these tests should have some sort of `warm up period' in which
they invoke the methods so that they can be jitted. If we want to test
the speed of jitting, we should do so in another benchmark. It would be
interesting to see the effects of different -O options on --compile-all
on a few assemblies.

> Also, the Create:Array tests run so much faster with full 
> optimization that I am not sure how to interpret the results.
Maybe the abcrem optimization was effective there?


Could you give us any insight to the performance of these benchmarks
over time? I know you run them every so often, and would love to hear
about what has and has not helped the benchmarks.

Someday, I hope we can get benchmarks run automatically on a server,
like our NUnit tests are now, so that we can see trends over time. Your
informal reports can be a first step in this process.

Thanks again,
-- Ben