harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Egor Pasko <egor.pa...@gmail.com>
Subject Re: Is it important to improve optimizations of Harmony's JIT compiler?
Date Tue, 19 May 2009 17:31:02 GMT
On the 0x5AF day of Apache Harmony niejiutao@gmail.com wrote:
> Hi all,
>
> I'm interested in compiler optimizations and have some thoughts on
> improving the IR and optimizations of Jitrino.OPT.  However, it seems
> that the development on Jitrino.OPT is not very active (only 617 lines
> of changes of source code in the past five months), so I wonder whether
> improving the optimizations of JIT compiler is important for Harmony
> community.  If it's not, why? (Note that Harmony is not very competitive
> to other JVMs on performance, http://www.stefankrause.net/wp/?p=9)  If
> it is, are there any existing plans or any people working on this?

The biggest problem with JIT performance improvements that I do not
know how to deal with now is the need of a good performance lab.

This means a diverse pool of high-end machines and specific profiling
software. Add to this at least one of those performance experts (aka
men in black) who make a fair time commitment for spotting
bottlenecks, catching regressions, maintaining the lab.

Measuring performance on a random laptop just does not work.

I recall Alexey Shipilev had thoughts of setting up a lab like this
(at home?:), but I did not hear the status from him for quite a while.
 
-- 
Egor Pasko


Mime
View raw message