HashBits, a lazy way
John M McIntosh
johnmci at mac.com
Sat Jul 12 18:09:04 UTC 2003
> From: Stephen Pair <stephen at p...>
> I wonder how this will translate to real world performance? 5%
> improvement? 10%? Since real code doesn't allocate a bunch of objects
> in a tight loop like this, I imagine that the % of total cycles spent
> in
> allocation must be lower than the 20% in this test case. It would be
> nice to have some real world benchmarks to compare...building the
> interpreter is nice, but how about picking a decent sized package and
> installing it in the image (from disk)? Since, people do tend to do
> that quite a lot.
If I look bounce atoms, 100, with higher performance I see these
overall CPU usage numbers
52% interpreter
8.2% related to mac screen updating
5.5% for copyLoopNoSource
4.28% to instantiate (makepoint, pushfloat, allocate/recycle context,
BlockCopy)
2.8 mark/trace
... etc
Tuning with better code, plus no-fill takes the 4.28% to 2.96% So it's
1.5% in this example.
I've also done a test allocating 50 ByteArrays in a loop and see a 10%
faster allocation rate, with the
allocation C routine running at least 30% faster.
This testing still is doing the hash because I wasn't sure how to
implement the change. But I see
Lex in message http://groups.yahoo.com/group/squeak/message/65587 has
an answer for me.
--
========================================================================
===
John M. McIntosh <johnmci at smalltalkconsulting.com> 1-800-477-2659
Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com
========================================================================
===
More information about the Squeak-dev
mailing list
|