From: Carter Cheng <cartercheng@...>
Date: 2011-10-27T19:27:57+09:00
Subject: [ruby-core:40450] Re: JIT development for MRI

--f46d043bdfb828c5a904b045393f
Content-Type: text/plain; charset=ISO-8859-1

Thanks Charlie for the input.

On Thu, Oct 27, 2011 at 2:50 AM, Charles Oliver Nutter
<headius@headius.com>wrote:

> Depending on the difficulty of implementing a JIT for MRI, it would be
> interesting to see an incremental approach...first just a simple
> translation of bytecoded operations to assembly, then optimizing those
> operations within a given body, then exploring cross-call
> optimizations like tracing or profiled inlining. No need to boil the
> ocean at first if you can start to show useful performance gains from
> a simple JIT.
>
> - Charlie
>
> On Thu, Oct 27, 2011 at 3:02 AM, Carter Cheng <cartercheng@gmail.com>
> wrote:
> > Thanks Koichi. I may make an attempt to implement some sort of JIT for
> MRI-
> > that being said it is hard to predict the performance prior to completing
> > the code and running tests (or I am not so good an engineer that I do
> have a
> > deep intuition of this).
> >
> > On Thu, Oct 27, 2011 at 12:14 AM, SASADA Koichi <ko1@atdot.net> wrote:
> >>
> >> (2011/10/27 15:28), Carter Cheng wrote:
> >> > Thanks Koichi. How do profiling based approaches differ from trace
> >> > recording
> >> > in this context?
> >>
> >> Trace based approach is completely difference, I think.  And I don't
> >> know which is suitable for Ruby.
> >>
> >> (I guest that on small footprint programs, trace based one is good.  And
> >> ruby is...  But I don't have any more experience)
> >>
> >> --
> >> // SASADA Koichi at atdot dot net
> >>
> >
> >
>
>

--f46d043bdfb828c5a904b045393f
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

Thanks Charlie for the input.<br><br><div class=3D"gmail_quote">On Thu, Oct=
 27, 2011 at 2:50 AM, Charles Oliver Nutter <span dir=3D"ltr">&lt;<a href=
=3D"mailto:headius@headius.com">headius@headius.com</a>&gt;</span> wrote:<b=
r><blockquote class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:=
1px #ccc solid;padding-left:1ex;">
Depending on the difficulty of implementing a JIT for MRI, it would be<br>
interesting to see an incremental approach...first just a simple<br>
translation of bytecoded operations to assembly, then optimizing those<br>
operations within a given body, then exploring cross-call<br>
optimizations like tracing or profiled inlining. No need to boil the<br>
ocean at first if you can start to show useful performance gains from<br>
a simple JIT.<br>
<br>
- Charlie<br>
<div><div></div><div class=3D"h5"><br>
On Thu, Oct 27, 2011 at 3:02 AM, Carter Cheng &lt;<a href=3D"mailto:carterc=
heng@gmail.com">cartercheng@gmail.com</a>&gt; wrote:<br>
&gt; Thanks Koichi. I may make an attempt to implement some sort of JIT for=
 MRI-<br>
&gt; that being said it is hard to predict the performance prior to complet=
ing<br>
&gt; the code and running tests (or I am not so good an engineer that I do =
have a<br>
&gt; deep intuition of this).<br>
&gt;<br>
&gt; On Thu, Oct 27, 2011 at 12:14 AM, SASADA Koichi &lt;<a href=3D"mailto:=
ko1@atdot.net">ko1@atdot.net</a>&gt; wrote:<br>
&gt;&gt;<br>
&gt;&gt; (2011/10/27 15:28), Carter Cheng wrote:<br>
&gt;&gt; &gt; Thanks Koichi. How do profiling based approaches differ from =
trace<br>
&gt;&gt; &gt; recording<br>
&gt;&gt; &gt; in this context?<br>
&gt;&gt;<br>
&gt;&gt; Trace based approach is completely difference, I think. =A0And I d=
on&#39;t<br>
&gt;&gt; know which is suitable for Ruby.<br>
&gt;&gt;<br>
&gt;&gt; (I guest that on small footprint programs, trace based one is good=
. =A0And<br>
&gt;&gt; ruby is... =A0But I don&#39;t have any more experience)<br>
&gt;&gt;<br>
&gt;&gt; --<br>
&gt;&gt; // SASADA Koichi at atdot dot net<br>
&gt;&gt;<br>
&gt;<br>
&gt;<br>
<br>
</div></div></blockquote></div><br>

--f46d043bdfb828c5a904b045393f--