On Fri, May 14, 2010 at 1:03 PM, Levente Uzonyi <leves at elte.hu> wrote:

> On Fri, 14 May 2010, Eliot Miranda wrote:
>
>  On Fri, May 14, 2010 at 10:28 AM, Levente Uzonyi <leves at elte.hu> wrote:
>>
>>  On Fri, 14 May 2010, christophe.jalady at free.fr wrote:
>>>
>>>  Just print this two scripts:
>>>
>>>>
>>>> |s|
>>>> s := WideString with: (Character value: 16r55E4).
>>>> [100000 timesRepeat: [s hash]] timeToRun.
>>>>
>>>> |s|
>>>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>>>> 16rE4)
>>>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>>>> [100000 timesRepeat: [s hash]] timeToRun.
>>>>
>>>> I've seen that "ByteString>>stringHash:initialHash" use a primitive.
>>>> Could
>>>> we
>>>> have such optimisation for WideString ?
>>>>
>>>>
>>> Sure. There are (at least) four ways to do it:
>>> 1) Update the implementation of the primitive to support wordarrays. Then
>>> build a vm with it and finally change the image side code. This would be
>>> the
>>> best IMHO. Is there a reason why the primitive is not implemented this
>>> way?
>>> 2) Use NativeBoost to write a primitive for this method. This is much
>>> easier than building a vm/plugin, though you have to write the code in
>>> X86
>>> assembly. This would be a cool demo for NativeBoost's primitive writing
>>> capabilities. (Much better than the Fibonacci-number calculator. :))
>>> 3) Use Exupery, I think it can optimize methods like this.
>>> 4) Use the Cog VM. ;) (I wonder how fast this method is with Cog.)
>>>
>>>
>> 2.66GHz Intel Core i7 MacBook Pro (my new tool!!!!).  Current Cog, my 3.8
>> derived work image:
>>
>> |s|
>> s := WideString with: (Character value: 16r55E4).
>> [100000 timesRepeat: [s hash]] timeToRun. 16
>>
>> |s|
>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>> 16rE4)
>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>> [100000 timesRepeat: [s hash]] timeToRun. 10
>>
>> |s|
>> s := WideString with: (Character value: 16r55E4).
>> [100000000 timesRepeat: [s hash]] timeToRun. 15725
>>
>> |s|
>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>> 16rE4)
>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>> [100000000 timesRepeat: [s hash]] timeToRun. 11078
>>
>>
>> Current 4.1/Squeak 4.1.1beta2U.app:
>> |s|
>> s := WideString with: (Character value: 16r55E4).
>> [100000 timesRepeat: [s hash]] timeToRun. 98
>>
>> |s|
>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>> 16rE4)
>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>> [100000 timesRepeat: [s hash]] timeToRun. 31
>>
>> |s|
>> s := WideString with: (Character value: 16r55E4).
>> [100000000 timesRepeat: [s hash]] timeToRun. 100660
>>
>> |s|
>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>> 16rE4)
>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>> [100000000 timesRepeat: [s hash]] timeToRun. 29476
>>
>> Current 4.1/current Cog
>> |s|
>> s := WideString with: (Character value: 16r55E4).
>> [100000 timesRepeat: [s hash]] timeToRun. 14
>>
>> |s|
>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>> 16rE4)
>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>> [100000 timesRepeat: [s hash]] timeToRun. 10
>>
>> |s|
>> s := WideString with: (Character value: 16r55E4).
>> [100000000 timesRepeat: [s hash]] timeToRun. 14946
>>
>> |s|
>> s := ByteString with: (Character value: 16rE4) with: (Character value:
>> 16rE4)
>> with: (Character value: 16rE4) with: (Character value: 16rE4).
>> [100000000 timesRepeat: [s hash]] timeToRun. 10171
>>
>> So quite a bit faster :)
>>
>
> Thanks, it looks really nice. I wonder why are the results with 4.1 better
> than with your 3.8 derived image.
>

The code looks the same and so I expect the times are close enough to be in
the noise.  I didn't shut my machine down to a consistent state so other
tasks could have affected things, etc.


>
> Levente
>
>
>> best
>> Eliot
>>
>>
>>
>>> Levente
>>>
>>>
>>>  Christophe.
>>>>
>>>>
>>>>
>>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20100514/4c630a5a/attachment.htm