Re: [rvm-research] Contribute back to Jikes RVM!

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Hello Laurence/Colin,

    I am looking to implement Primitive read barriers to track all the reads
of Java application objects. Just wanted to check if the read barriers have
been implemented by someone after the previous post ? If not,Laurence I see
in one of your previous reply that there are other ways to collect these
stats. Do you infer using the tools like Valgrind? Can you help me by giving
more details about other ways of collecting these stats ?

Regards,
Jagadish.

Colin(Du Li) wrote:
> 
> Dear Laurence,
> 
> Thanks so much for your thorough explanation!
> It's quite interesting for me to try to get the last class of barriers
> done. We can discuss more later.
> Thanks again.
> 
> Du Li
> 
> Laurence Hellyer wrote:
>> 
>> 
>> On 12 Jan 2010, at 23:56, Colin(Du Li) wrote:
>> 
>>> 
>>> Dear Laurence,
>>> 
>>> I assume you also implemented primitive read barriers. When can it be
>>> committed?
>>> Thanks a lot!
>>> 
>>> Du Li
>> 
>> Hi,
>> 
>> Sorry, but for my work I merely implemented primitive write barriers. 
>> Are you wanting primitive read barriers for collecting statistics or for
>> another purpose? (Collecting statistics can be done in many other ways)
>> 
>> Jennifer Sartor (University of Texas) has been actively working on
>> Arraylets and presumably this has involved implementing primitive read
>> barriers (at least on array's).
>> 
>>>From my experience of implementing primitive write barriers the main
difficulty is not so much modifying the compilers to emit the barrier but
catching all places inside the RVM that write to the heap via Magic or JNI
(field reflection code for starters).  When creating my primitive write
barriers to ensure correctness  I write protected the heap and my primitive
write barriers then wrote to a virtual memory alias to write the value into
the heap - any RVM code that wrote to the heap directly caused a SIGSEGV
(see [1] for a presentation on this).  At the ISMM 2009 Wild and Crazy
session I expanded upon this idea to suggest a read protected heap for
debugging primitive read barriers.
>> 
>> The big draw back of this approach is that *all* access to the heap
>> (including status words, TIB's and alignment gaps) need some sort of
>> barrier which leads to extra implementation effort.  Daniel Frampton
>> (Australian National University) has proposed (and I believe implemented)
>> an alternative debugging tool that uses ValGrind and can selectively
>> protect individual heap fields.
>> 
>> If there is enough desire for primitive read barriers perhaps we can get
>> a collaboration going to implement this last class of barriers in
>> JikesRVM and MMTk.
>> 
>> Kind regards
>> Laurence
>> 
>> [1]
>> http://www.cs.kent.ac.uk/people/rpg/lh243/conferences/index.html#MM-NET09
>> 
>>> 
>>> 
>>> Laurence Hellyer wrote:
>>>> 
>>>> 
>>>> For those wanting to use primitive write barriers hopefully this might
>>>> help:
>>>> 
>>>> In the SVN head, r15805 commited the last of necessary changes to the
>>>> compilers etc and supports primitive write barriers.
>>>> 
>>>> The best place to get started with using the primitive write barrier
>>>> code
>>>> is to look at commit r15806, this commit added the
>>>> UsePrimitiveWirteBarriers GC algorithm.  UsePrimitiveWriteBarriers is
>>>> really an extension of the SemiSpace GC that uses the primitive write
>>>> barriers to unconditionally make primitive writes to the heap  - the
>>>> only
>>>> real use of this GC is to test that the compilers are correctly calling
>>>> the primitive write barriers.
>>>> 
>>>> If you have further questions I will be happy to try to answer them.  A
>>>> couple of points to bear in mind when starting off:
>>>> 
>>>> i) The primitive write barriers are for all primitive writes to the
>>>> heap
>>>> (objects and arrays).  There are no barriers on static primitives
>>>> (although I can't think why you would want one)
>>>> ii) There currently is no support within the MMTk scripting language
>>>> for
>>>> testing primitive write barriers
>>>> 
>>>> Kind regards
>>>> Laurence
>>>> 
>>>> Laurence Hellyer
>>>> Research Student
>>>> School of Computing
>>>> University of Kent
>>>> 
>>>> More info: http://www.cs.kent.ac.uk/people/rpg/lh243/
>>>> 
>>>> 
>>>> ------------------------------------------------------------------------------
>>>> This SF.Net email is sponsored by the Verizon Developer Community
>>>> Take advantage of Verizon's best-in-class app development support
>>>> A streamlined, 14 day to market process makes app distribution fast and
>>>> easy
>>>> Join now and get one step closer to millions of Verizon customers
>>>> http://p.sf.net/sfu/verizon-dev2dev 
>>>> _______________________________________________
>>>> Jikesrvm-researchers mailing list
>>>> Jik...@li...
>>>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>>>> 
>>>> 
>>> 
>>> -- 
>>> View this message in context:
>>> http://old.nabble.com/-rvm-research--Contribute-back-to-Jikes-RVM%21-tp26760794p27137132.html
>>> Sent from the jikesrvm-researchers mailing list archive at Nabble.com.
>>> 
>>> 
>>> ------------------------------------------------------------------------------
>>> This SF.Net email is sponsored by the Verizon Developer Community
>>> Take advantage of Verizon's best-in-class app development support
>>> A streamlined, 14 day to market process makes app distribution fast and
>>> easy
>>> Join now and get one step closer to millions of Verizon customers
>>> http://p.sf.net/sfu/verizon-dev2dev 
>>> _______________________________________________
>>> Jikesrvm-researchers mailing list
>>> Jik...@li...
>>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>> 
>> 
>> Laurence Hellyer
>> Research Student
>> School of Computing
>> University of Kent
>> 
>> More info: http://www.cs.kent.ac.uk/people/rpg/lh243/
>> 
>> 
>> ------------------------------------------------------------------------------
>> This SF.Net email is sponsored by the Verizon Developer Community
>> Take advantage of Verizon's best-in-class app development support
>> A streamlined, 14 day to market process makes app distribution fast and
>> easy
>> Join now and get one step closer to millions of Verizon customers
>> http://p.sf.net/sfu/verizon-dev2dev 
>> _______________________________________________
>> Jikesrvm-researchers mailing list
>> Jik...@li...
>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>> 
>> 
> 
> 

-- 
View this message in context: http://old.nabble.com/-rvm-research--Contribute-back-to-Jikes-RVM%21-tp26760794p32599846.html
Sent from the jikesrvm-researchers mailing list archive at Nabble.com.