Report from a novice VM h4x0r.
John M McIntosh
johnmci at mac.com
Thu Apr 1 02:55:16 UTC 2004
On Mar 31, 2004, at 5:50 PM, Alan Grimes wrote:
> Performance wise, I have not been very careful testing but I beleive
> that there are enough "gems" in it to yield a 35-40% improvment in GC
> code and a 1-2% improvment in opcode performance as measured by
> TinyBenchmarks.
Mmm 35-40% in the GC would be welcome. The last refactoring I did moved
most/all working
variables into powerpc registers and avoided any calling outside of the
inlined method. Although the
routines involved appear to be separate when non-inlined, they infact
fold into single procedures.
A change I made to Slang then ensures any globals that fold into a
single procedure then become local variables. Yes
Andreas did trip over this 'feature', but adding a single accessor
method and telling Slang not to toss it during optimization,
then ensures the variable becomes a global, which is more expensive to
use. These changes made a 30ish percent improvement
on the powerpc GC times.
Having looked at what's there now from time to time, I'd be very
interested to see what you
are proposing in this area.
--
========================================================================
===
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
|