Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ptp-dev] OpenMP Refactorings/Transformations that developers would like to have in PTP

Nick,
1) I don't think I ever went any further with the OpenMP features, I think I was going to use the EclipseCon talk as an excuse to work on it, then got busy with other things.
I am also not an OpenMP expert, so yes, the folks there at NCSA/UIUC can perhaps suggest some ideas.
2) Don't have any specific plans for analysis new features in PLDT. Have been working on some IBM-internal extensions for other API sets.
Re: merging with Codan, it's probably a good idea but as before I haven't had the time to investigate. Originally the analysis we did for MPI barrier analysis as well as OpenMP provided some structures that were more generally useful but didn't get added to CDT. I did manage to keep the generic stuff separate from the MPI and OpenMP-specific uses, however.
The 2008 EclipseCon talk "Static Analysis in PTP with CDT" is probably the best description of that. http://www.eclipsecon.org/2008/?page=sub/&id=373

>but that wraps stuff in its own OMPCFG and PAST nodes.
CDT's AST has evolved and been extended some since that code was written. I remember thinking during one of the updates to keep it functioning as CDT APIs changed, that perhaps alot of that could be simplified and perhaps even the OMP-specific classes could be removed because CDT now has some node classes that do that. Maybe it was pragma AST notes, not sure. I agree that Codan is more active but you are right we have some graphs etc in PLDT that provide some more features. Merging them does seem worth looking into, I just don't have the time at the moment. I am happy to help/advise if you want to!


...Beth

Beth Tibbitts
Eclipse Parallel Tools Platform http://eclipse.org/ptp
IBM STG - High Performance Computing Tools
Mailing Address: IBM Corp., 745 West New Circle Road, Lexington, KY 40511


Inactive hide details for nchen.dev---03/08/2011 07:10:09 PM---Hi Beth Thanks for the encouragement :-)nchen.dev---03/08/2011 07:10:09 PM---Hi Beth Thanks for the encouragement :-)


    From:

nchen.dev@xxxxxxx

    To:

Parallel Tools Platform general developers <ptp-dev@xxxxxxxxxxx>

    Date:

03/08/2011 07:10 PM

    Subject:

Re: [ptp-dev] OpenMP Refactorings/Transformations that developers would like to have in PTP

    Sent by:

ptp-dev-bounces@xxxxxxxxxxx




Hi Beth

Thanks for the encouragement :-)

There are two things that I need with:

1) I did a search for "OpenMP" on the ptp-dev list looking for features that people might be interested in. I stumbled upon your post about your EclipesCon talk (which was unfortunately canceled) in <
http://dev.eclipse.org/mhonarc/lists/ptp-dev/msg02389.html>. I was wondering if there are any features that you were planning to implement. Like I said in my previous e-mail, I'm not an expert OpenMP user so I am interested in suggestions on features/tools that OpenMP developers would really love to have in an IDE. I'll also talk to Marc Snir and the folks at NCSA on the UIUC campus for ideas.

2) What are the future plans of the analysis in PLDT? I see some overlap between what it and Codan has. Are there plans to merge the two? It would help to know so I know which one to use (for C refactorings). Right now Codan seems more active but PLDT has some nice features like its concurrency analysis for OpenMP but that wraps stuff in its own OMPCFG and PAST nodes.

Thanks again.

--
Nick

On Mar 7, 2011, at 7:15 AM, Beth Tibbitts wrote:

> Nick,
> That's terrific! Let me know if I can help you find your way around what's there now.
> OpenMP could certainly use some attention to add some new features.
>
> ...Beth
>
> Beth Tibbitts
> Eclipse Parallel Tools Platform
http://eclipse.org/ptp
> IBM STG - High Performance Computing Tools
> Mailing Address: IBM Corp., 745 West New Circle Road, Lexington, KY 40511
>
> <graycol.gif>nchen.dev---03/07/2011 02:35:16 AM---Hi (Cross-posting on ptp-dev and ptp-user since developers and users of PTP might be interested in t
>
> <ecblank.gif>
> From:
> <ecblank.gif>
> nchen.dev@xxxxxxx
> <ecblank.gif>
> To:
> <ecblank.gif>
> ptp-dev@xxxxxxxxxxx, ptp-user@xxxxxxxxxxx
> <ecblank.gif>
> Date:
> <ecblank.gif>
> 03/07/2011 02:35 AM
> <ecblank.gif>
> Subject:
> <ecblank.gif>
> [ptp-dev] OpenMP Refactorings/Transformations that developers would like to have in PTP
> <ecblank.gif>
> Sent by:
> <ecblank.gif>
> ptp-dev-bounces@xxxxxxxxxxx
>
>
>
> Hi
>
> (Cross-posting on ptp-dev and ptp-user since developers and users of PTP might be interested in this)
>
> First let me introduce myself. I'm Nick, a PhD student working with Ralph Johnson in the Software Engineering group at the University of Illinois at Urbana-Champaign. I'm looking at using PTP to implement a set of transformations/refactorings that might be useful for OpenMP developers. Right now, I'm concentrating on implementing transformations (through the LTK refactoring framework) that will reduce/eliminate false sharing. This includes transformations such as splitting scalar[1], expand record[1] and selection of runtime scheduling parameters[2].
>
> What are some other refactorings/transformations that you would like to see implemented? I'm not an OpenMP expert so I would greatly appreciate any suggestions from expert developers on features that could help make their lives easier while working with OpenMP.
>
> I also welcome any other suggestions on other features that you would like to see in PTP for OpenMP. I noticed that the PLDT plug-ins for OpenMP were contributed in 2006 and have been undergoing constant API evolutions to keep up with CDT and the Eclipse platform over the years; however, nothing new has been added. So if you have any other features that you would like, please let me know and I might be able to implement them as well.
>
> Thanks in advance!
>
> --
> Nick Chen
>
> [1] False Sharing and Spatial Locality in Multiprocessor Caches <
http://iacoma.cs.uiuc.edu/iacoma-papers/false_sharing.pdf>
> [2] False Sharing Elimination by Selection of Runtime Scheduling Parameters <
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.43.8785>
> _______________________________________________
> ptp-dev mailing list
> ptp-dev@xxxxxxxxxxx
>
https://dev.eclipse.org/mailman/listinfo/ptp-dev
>
>
> _______________________________________________
> ptp-dev mailing list
> ptp-dev@xxxxxxxxxxx
>
https://dev.eclipse.org/mailman/listinfo/ptp-dev

_______________________________________________
ptp-dev mailing list
ptp-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-dev


GIF image

GIF image


Back to the top