Re: v8 first impressions

"David Griswold" <[email protected]> Mon, 22 Sep 2008 14:30:31 -0700
Newsgroups gmane.comp.lang.smalltalk.strongtalk
Message-ID <[email protected]>
------=_Part_168_21000192.1222119031710
Content-Type: text/plain; charset=ISO-8859-1

Yes, it's really a shame that original development on the system stopped
just as the new back-end was starting to run programs about the size of
Richards.  It can do quite a bit more optimization than the current
Strongtalk back-end, although it still isn't a real optimizing compiler by
any means.  But there is a lot of low-hanging fruit that it can take
advantage of.
Unfortunately, as I recall it is still far from finished, and it would be a
serious project for a real compiler guy to get it done.  But kudos for
getting any numbers out of it at all!  Maybe at some point someone will do
something with it.-Dave

On Mon, Sep 22, 2008 at 2:15 PM, talksmall <[email protected]>wrote:

>
> Just a slight update to my earlier post about the relative performance
> of various VMs.
>
> With the new back-end in place the performance of the Strongtalk VM
> improves substantially. I didn't include it in my previous test,
> because I honestly didn't think it was stable enough to complete a
> run. Still, I gave it a try and it surprised me.
>
> A 100 run average time with the new back-end produces an average time
> of 4.36ms. That's about 3 times as fast as the VW and V8 figures.
>
> Running with the new back-end is pretty unstable (I had to try a
> couple of times to get a complete run and paste the results in a
> spreadsheet without the VM crashing with an access violation), but it
> does offer the prospect of substantial performance improvements.
>
> Regards, Steve
>
> On Sep 8, 1:47 am, talksmall <[email protected]> wrote:
> > I ran a comparison of the Richards benchmark for which both Javascript
> > and Smalltalk versions are available.
> >
> > The version released with V8 isn't exactly comparable to the standard
> > version since it uses a count of 1000 rather than 10000. With that
> > tweaked to match the standard count I ran a comparison of V8 against a
> > selection of Smalltalk VMs - VisualWorks NC, Squeak, GNU Smalltalk and
> > Strongtalk.
> >
> > I customised the Smalltalk tests to follow the same methodology as the
> > V8 benchmarks - iterating the test for a second and then dividing by
> > the number of iterations.
> >
> > The results are interesting
> >
> > Strongtalk:    7.335ms
> > VW:           13.013ms
> > V8:           13.648ms
> > GNUSmalltalk: 91.090ms
> > Squeak:      102.600ms
> >
> > All of the tests were run on a Vista 64 Quad core machine at 2.66 GHz,
> > though in each case I was running a 32 bit VM (since not all the VMs
> > have 64 bit builds).
> >
> > Regards, Steve
> >
>

--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Strongtalk-general" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to [email protected]
For more options, visit this group at http://groups.google.com/group/strongtalk-general?hl=en
-~----------~----~----~----~------~----~------~--~---


------=_Part_168_21000192.1222119031710
Content-Type: text/html; charset=ISO-8859-1

<div dir="ltr">Yes, it&#39;s really a shame that original development on the system stopped just as the new back-end was starting to run programs about the size of Richards. &nbsp;It can do quite a bit more optimization than the current Strongtalk back-end, although it still isn&#39;t a real optimizing compiler by any means. &nbsp;But there is a lot of low-hanging fruit that it can take advantage of.<div>
<br></div><div>Unfortunately, as I recall it is still far from finished, and it would be a serious project for a real compiler guy to get it done. &nbsp;But kudos for getting any numbers out of it at all! &nbsp;Maybe at some point someone will do something with it.<div>
-Dave<br><br><div class="gmail_quote">On Mon, Sep 22, 2008 at 2:15 PM, talksmall <span dir="ltr">&lt;<a href="mailto:[email protected]">[email protected]</a>&gt;</span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<br>
Just a slight update to my earlier post about the relative performance<br>
of various VMs.<br>
<br>
With the new back-end in place the performance of the Strongtalk VM<br>
improves substantially. I didn&#39;t include it in my previous test,<br>
because I honestly didn&#39;t think it was stable enough to complete a<br>
run. Still, I gave it a try and it surprised me.<br>
<br>
A 100 run average time with the new back-end produces an average time<br>
of 4.36ms. That&#39;s about 3 times as fast as the VW and V8 figures.<br>
<br>
Running with the new back-end is pretty unstable (I had to try a<br>
couple of times to get a complete run and paste the results in a<br>
spreadsheet without the VM crashing with an access violation), but it<br>
does offer the prospect of substantial performance improvements.<br>
<br>
Regards, Steve<br>
<div><div></div><div class="Wj3C7c"><br>
On Sep 8, 1:47&nbsp;am, talksmall &lt;<a href="mailto:[email protected]">[email protected]</a>&gt; wrote:<br>
&gt; I ran a comparison of the Richards benchmark for which both Javascript<br>
&gt; and Smalltalk versions are available.<br>
&gt;<br>
&gt; The version released with V8 isn&#39;t exactly comparable to the standard<br>
&gt; version since it uses a count of 1000 rather than 10000. With that<br>
&gt; tweaked to match the standard count I ran a comparison of V8 against a<br>
&gt; selection of Smalltalk VMs - VisualWorks NC, Squeak, GNU Smalltalk and<br>
&gt; Strongtalk.<br>
&gt;<br>
&gt; I customised the Smalltalk tests to follow the same methodology as the<br>
&gt; V8 benchmarks - iterating the test for a second and then dividing by<br>
&gt; the number of iterations.<br>
&gt;<br>
&gt; The results are interesting<br>
&gt;<br>
&gt; Strongtalk: &nbsp; &nbsp;7.335ms<br>
&gt; VW: &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 13.013ms<br>
&gt; V8: &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 13.648ms<br>
&gt; GNUSmalltalk: 91.090ms<br>
&gt; Squeak: &nbsp; &nbsp; &nbsp;102.600ms<br>
&gt;<br>
&gt; All of the tests were run on a Vista 64 Quad core machine at 2.66 GHz,<br>
&gt; though in each case I was running a 32 bit VM (since not all the VMs<br>
&gt; have 64 bit builds).<br>
&gt;<br>
&gt; Regards, Steve<br>
<br>
</div></div></blockquote></div><br></div></div></div><br>
--~--~---------~--~----~------------~-------~--~----~<br>
You received this message because you are subscribed to the Google Groups &quot;Strongtalk-general&quot; group. <br> To post to this group, send email to [email protected] <br> To unsubscribe from this group, send email to [email protected] <br> For more options, visit this group at http://groups.google.com/group/strongtalk-general?hl=en <br> -~----------~----~----~----~------~----~------~--~---<br>
<br>

------=_Part_168_21000192.1222119031710--