[self-interest] Performance figures interesting

[Michael Latta]

Good point that the bytecodes for the Self VM are not necessarily the same
as the Smalltalk VM.  But since the benchmark is not really measuring
bytecodes but a certain amount of computation, the numbers still indicate a
good measure of the efficiency of the system for that micro benchmark.

It is interesting that the Sparc VM was 3x the PPC system.  Was this due to
the Sparc hardware like register windows or more due to the effort put into
the two compilers?

Is PPC the primary target for Klien or will you be targeting x86 now that
Macs are using that chip?

Michael




-----Original Message-----
From: self-interest at yahoogroups.com [mailto:self-interest at yahoogroups.com]
On Behalf Of David Ungar
Sent: Tuesday, May 09, 2006 9:30 PM
To: self-interest at yahoogroups.com
Subject: Re: [self-interest] Performance figures interesting

I recently ran some numbers; the SPARC Self VM was about 3x more  
efficient,
clock-for-clock, than the PPC version.
Lets of reasons....
I also suspect that the Self version of the benchmark may be  
"running" many more bytecodes
than the Smalltalk one--but haven't looked at this very carefully.

- David


On May 9, 2006, at 9:11 PM, Michael Latta wrote:

> Io looks interesting, but misses a few things I really like in  
> Smalltalk
> (user defined control structures using blocks primarily).  Having  
> compact
> syntax for things like collect: detect: and to:by:do: or any I  
> choose to
> make up for my application is very high on my list.  I can use  
> Smalltalk to
> build a domain specific language by just modeling it.
>
> The figure for Self is using the Sun VM that is about 10 years old  
> with some
> cleanup since.  I am glad you can get better numbers in your VM.   
> There may
> be something in the microbenchmark that is pathological to the Self  
> VM way
> of doing things.  If it never saw it as a hotspot it may have  
> interpreted
> the entire benchmark even with repeated executions.
>
> Michael
>
>
> -----Original Message-----
> From: self-interest at yahoogroups.com [mailto:self- 
> interest at yahoogroups.com]
> On Behalf Of Steve Dekorte
> Sent: Tuesday, May 09, 2006 8:53 PM
> To: self-interest at yahoogroups.com
> Subject: Re: [self-interest] Performance figures interesting
>
>
> On 9 May 2006, at 02:08 pm, Michael Latta wrote:
>> PowerMac G5 2.0Ghz w/4GB 400Mhz memory
>> Self ... 879023 sends/sec'
>
> That doesn't seem right. On a 2.5Ghz G5 w/1GB, Io can do 3.7 million
> sends per second (by send, I mean a dynamic lookup and a block/method
> activation).
>
> This page shows numbers for Ruby and Python on the same machine:
>
> 	http://www.iolanguage.com/about/speed/basic/ppc/
>
> And for x86:
>
> 	http://www.iolanguage.com/about/speed/basic/x86/
>
> -- Steve
>
>
>
>
> Yahoo! Groups Links
>
>
>
>
>
>
>
>
>
>
> Yahoo! Groups Links
>
>
>
>
>
>
>



 
Yahoo! Groups Links