[Q] Speed comparison 2.8 vs 3.8

Eddie Cottongim cottonsqueak at earthlink.net
Fri Sep 2 06:34:01 UTC 2005


One other thing, 3.9a has an optimized StrikeFontSet>>displayString:  
on:  from:  to:  at:  kern:  baselineY:  that makes a huge difference 
(factor of 2 or 3) and we don't want to lose it if possible. Something 
like the new morphs menu for example can take seconds to open without it.

Eddie Cottongim wrote:

> Yoshiki,
>
> One minor problem - it seems to break balloon help. All the balloons 
> are empty.
>
> The problem is somewhere in to the GrafPort displayScanner methods. I 
> couldn't immediately figure out how to fix it without hurting the 
> speed. Reverting those two methods will fix the balloons, but make 
> editing slow again.
>
> Oh, and I tried your change on a slow machine, and its a noticable 
> improvement for text editing of any size there, not just monster 
> paragraphs.
>
> Thanks again,
> Eddie
>
>
> Eddie Cottongim wrote:
>
>> It seems to have solved the big paragraph problem. It is very smooth 
>> now, appears to be as good as 3.7.
>>
>> As for the browser test (which doesn't have much to do with this, no 
>> really big paragraphs, but I tried anyway) 3.8-6665 gets around a 10% 
>> improvement. 3.9a-6681 gets a smaller improvement, around 1% (less 
>> than the margin of error but it did seem to be repeatable). It might 
>> be because the default preferences are different, or because the 
>> other speed improvements in 3.9a cause this code to be exercised 
>> less. In either case its an improvement.
>>
>> Sounds very promising!
>>
>> Thanks,
>> Eddie
>>
>>
>> Yoshiki Ohshima wrote:
>>
>>>  Eddie,
>>>
>>>  Ok.  I'm not exactly confortable with this patch, but it trys to do
>>> something clever to go to primitive when possible.
>>>
>>>  Can you run the benchmark that you reported the 3X, 6X differences
>>> with this installed into 3.8-6665 or similar image, and see what you
>>> get.
>>>
>>>  Thanks!
>>>
>>> -- Yoshiki
>>>  
>>>
>>> ------------------------------------------------------------------------ 
>>>
>>>
>>>
>>>  
>>>
>>
>>
>>
>
>
>




More information about the Squeak-dev mailing list