Re: [rvm-research] Escape analysis

Hi Richard,

Not sure if this helps, but we've implemented a redundant barrier 
elimination analysis that eliminates a barrier if an earlier barrier in 
the region makes the later barrier unnecessary. The analysis is part of 
our Octet patch on the Research Archive ( 
http://sourceforge.net/p/jikesrvm/research-archive/43/), and Appendix B 
of the Octet paper ( 
http://www.cse.ohio-state.edu/~mikebond/octet-oopsla-2013.pdf 
<http://www.cse.ohio-state.edu/%7Emikebond/octet-oopsla-2013.pdf>) 
describes the analysis.

A while back, we tried to use Jikes's escape analysis to help us remove 
barriers for definitely non-escaping objects, but as far as we could 
tell, it was giving us unsound results (maybe we were forcing it to run 
on non-SSA IR when it should only run on SSA IR).

Cheers,
Mike

On 10/15/2013 12:12 AM, Richard Jones wrote:
> Hi Sam
>
> It's an optimisation that involves removing certain kinds of barrier, 
> both read and write. While it is important not  to remove barriers 
> that should not be removed, it may not be disastrous to fail to remove 
> some that could have been removed. Is this enough info?
>
> Thanks
> Richard
>
> On 14 Oct 2013, at 19:29, Sam Guyer <[email protected] 
> <mailto:[email protected]>> wrote:
>
>> Richard
>>
>> What are you trying to do? Is high precision important?
>>
>> On Oct 14, 2013, at 1:38 PM, Richard Jones <[email protected] 
>> <mailto:[email protected]>> wrote:
>>
>>>
>>> I'm interested in some fairly simple escape analysis. While Jikes 
>>> RVM seems to provide a framework for analysing escapement and 
>>> applying transformations, this seems to depend on SSA, e.g. see 
>>> org.jikesrvm.compilers.opt.escape.EscapeTransformations.perform(). 
>>> However, the project status page indicates that SSA form in the 
>>> compilers is disabled because of bugs. Looking further (e.g. 
>>> http://sourceforge.net/p/jikesrvm/bugs/514/), it seems that a lot of 
>>> effort was spent quite a few years ago on tracking the bugs down 
>>> without success. Has anyone worked / is anyone working further on this?
>>>
>>> - Richard
>>> ------------------------------------------------------------------------------
>>> October Webinars: Code for Performance
>>> Free Intel webinars can help you accelerate application performance.
>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the 
>>> most from
>>> the latest Intel processors and coprocessors. See abstracts and 
>>> register >
>>> http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Jikesrvm-researchers mailing list
>>> [email protected] 
>>> <mailto:[email protected]>
>>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>> ------------------------------------------------------------------------------
>> October Webinars: Code for Performance
>> Free Intel webinars can help you accelerate application performance.
>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the 
>> most from
>> the latest Intel processors and coprocessors. See abstracts and 
>> register >
>> http://pubads.g.doubleclick.net/gampad/clk?id=60135031&iu=/4140/ostg.clktrk_______________________________________________
>> Jikesrvm-researchers mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers
>

------------------------------------------------------------------------------
October Webinars: Code for Performance
Free Intel webinars can help you accelerate application performance.
Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from 
the latest Intel processors and coprocessors. See abstracts and register >
http://pubads.g.doubleclick.net/gampad/clk?id=60135991&iu=/4140/ostg.clktrk

_______________________________________________
Jikesrvm-researchers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers