-------- Message original -------- Sujet: Re: Message via your Google Profile: Inline caches in JavaScript Date : Tue, 5 Mar 2013 17:59:49 +0100 De : Vyacheslav Egorov me@mrale.ph Pour : Érick Lavoie erick.lavoie@gmail.com

Hi Érick!

This paper was in fact forwarded to me internally couple of days ago, but I could not access it back then, because your site was down. Now I see that site is back up and I'll definitely read it. I did read through your slides from the talk in Mozilla though.

In its current state I view my approach more like the way of explanation, then the actual way of implementation. I wish I was a researcher, then I would probably have time to turn my ideas into an actual implementation :-)

It is not mentioned in that post, but IC code outlined there was also released as part of mö.js code https://code.google.com/p/moe-js/source/browse/#git%2Ftracing%2Fengine%253Fs... [though its not used by the optimizing backend itself, but it was used] and a different POV on my intentions was given in the talk at JSConf EU 2012 http://mrale.ph/blog/2012/11/18/jsconfeu2012-talk.html

Unfortunately since then a lot of my brainpower is spent on the Dart VM so I did not have time to explore some additional ideas, that I have.

But you might be interested to hear that V8 is now moving in a direction, that I dreamed about ever since Crankshaft was implemented: towards implementing IC stubs in hydrogen (high level IR) instead of a hand written assembly. Some stubs were already rewritten. It is not the same as implementing stubs in JS itself, but this is very close :-) So the future is happening.

On Tue, Mar 5, 2013 at 5:31 PM, Érick Lavoie erick.lavoie@gmail.com wrote:

Hi Vyacheslav, I just stumbled upon your blog post "http://mrale.ph/blog/2012/06/03/explaining-js-vms-in-js-inline-caches.html", in which you present a way to implement inline caches in JavaScript. I independently came up last fall with the same technique after playing with V8 and trying to find a fast way to implement them for Photon, a Metacircular VM for JavaScript aimed at run-time instrumentation. I was part of the original Tachyon team and am still in the same lab so you might recognize a common nomenclature for our VMs ;-). I wrote a paper that explains the general architecture of the system and two specific optimizations for inline caches and object representations that you might find interesting. I would be more than glad to have your comments (ideas, improvements, etc.) on it: http://ericklavoie.com/pubs/RESoLVE13.pdf I will present this work at the RESoLVE'13 workshop that will be held in Houston, Texas as part of ASPLOS 2013 if you can make it. Note that the workshop presentation does not count as a publication and we will submit an improved version of the paper to OOPSLA at the end of March. Regards, Erick

---

This message was sent to you from your Google profile. The sender does not have your email address.

If you no longer wish to receive messages from your Google profile, you may edit your settings.