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: [drlvm]A subject to profiling instrumenting
Date Thu, 31 Aug 2006 10:33:50 GMT
On the 0x1D6 day of Apache Harmony ustczz@gmail.com wrote:
> How do you think, should patterns depend on certain JIT and GC
> > tightly? I suspect they depend heavily on GC, and not-so-much on
> > JIT. Though, I should say that JET uses a very cheap register
> > allocation mechanism, so it should negatively influence on the number
> > of memory accesses, and heavily. If you are more oriented on
> > performance you should be rather using OPT for that. But, yet, I do
> > not mind if you start with a more easy-to-use JET :)
> 
> 
> I think memory access pattern depends heavily on GC. I want to use
> JET to instrument some profile to get the access sequence of OBJECTS.
> Making use of GC to analyze the sequence is what to do next. I think you
> are an expert on GC. Am I right? :-)

How did you know? No, I am rather a JIT guy ;0)

> I don`t choose the OPT, because I think JET is much more easier and faster
> than OPT. We shouldn`t waste too much time in instrumenting. Is it?

Yes, but the hottest fragments of code (an, yes, most critical
mem-accesses) are compiled by OPT. Currently, I do not know, what is better
1) trace all accesses or 
2) just the hottest series of accesses 
(I'll read the papers, ask Mikhail and, maybe, figure it out). 
If assumption (1) is true, then JET is worth to start, but if (2) is
the only reasonable target, then, you should obviously get on with OPT.

> Are you trying to implement some known techniques or is your work a
> > subject of ongoing research? What papers can I read on this to be more
> > acquainted with what you are doing?
> 
> Yes, I want to first implement Chilimbi`s work in DRLVM. The following list
> mainly line out the techniques:
> 1.Efficient Representations and Abstractions for Quantifying and Exploiting
> Data Reference Locality.
> 2.Bursty Tracing: A Framework for Low-Overhead Temporal Profiling
> 3.Dynamic Hot Data Stream Prefetching for General-Purpose Programs
> 4.Profile-guided Proactive Garbage Collection for Locality Optimization

OK, I saw this list somewhere :) I'll take a look.

> If you want to know it more quickly, just reading the first three paper.

Thanks!
BTW, In old times I read "Exploring and Optimizing Itanium2 Cache(s)
Performance for Scientific Computing" but that looks too advanced for
JIT or GC...

-- 
Egor Pasko, Intel Managed Runtime Division


---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org


Mime
View raw message