[prev in list] [next in list] [prev in thread] [next in thread] 

List:       perl5-porters
Subject:    Re: Storable for bytecode?
From:       "Joshua ben Jore" <twists () gmail ! com>
Date:       2007-12-26 2:55:51
Message-ID: dc5c751d0712251855j3e9ffe8cm40301352a87a7ed2 () mail ! gmail ! com
[Download RAW message or body]

On Dec 24, 2007 12:14 PM, Andy Armstrong <andy@hexten.net> wrote:
> On 24 Dec 2007, at 19:46, Nicholas Clark wrote:
> > I'm sure that being on a plane back from "Vienna" had something to
> > do with
> > this - eg lack of oxygen, or somesuch, but I wondered:
>
> [snip]
>
> While the crazy talk is flying I was wondering the other day whether
> there's any scope for doing JIT things with the bytecode. Your
> previous post intersects with that idea somewhat. Are there typically
> other (expensive) optimisations that a long running program could do
> to itself?

I've been pondering whether I can use the data collected by
Runops::Trace by our QA department to do interesting JIT. Mostly I
realize I just don't know what the universe of interesting JIT is.
What I've come up with on my own is possibly looking for call sites
that use functions that turn out to purely functional or perhaps have
branching or loops that can be unrolled or conditions removed.

Or perhaps providing annotation information back to subsequent
compilations for manual optimizations. I dunno. There's a lot of
potential interesting work and a lot of background to assimilate to
know what to possibly do first.

Work is sponsoring a week of free hacking the second or third week in
January. I'd thought about introducing a new opcode that would add the
TEMP flag to SVs in lexicals for their final use so perhaps the values
could be stolen instead of being copied.

Josh
[prev in list] [next in list] [prev in thread] [next in thread]