To access the code, go to the 'code' tab which will give you a handle into
the SVN repository.
-Brad
On Thu, 19 Jun 2014, Rafael Asenjo Plaza wrote:
Hi Abhinav,
Here is the sourceforge repository for Chapel:
http://sourceforge.net/projects/chapel/
Thanks,
Rafa.
El 19/06/2014, a las 19:33, Abhinav Sarje <[email protected]> escribió:
Hi Rafael,
Thanks for the prompt reply. I could not find how to get access to the
repository - I saw only the release version download. Could you please point me
to the repo which I could checkout?
Thanks,
Abhinav.
On Thu, Jun 19, 2014 at 1:56 AM, Rafael Asenjo Plaza <[email protected]> wrote:
Hi Abhinav,
The BulkTransferStride mod is currently working only for distributed array
assignments, sort of A[d1]=B[d2], were A and B are block or cyclic distributed
arrays and d1 and d2 are the considered domains. I’ve not taken a look at
SSCA2, but I have not spotted such kind of assignments in
trunk/test/studies/hpcc/FFT. You can find an implementation of the FFT that
takes advantage of the bulkTransferStride here:
test/optimizations/bulkcomm/asenjo/redistBlockToCyclic
If you compile FFT_assignment.chpl as follows:
chpl -o fft FFT_assignment.chpl -s useBulkTransferStride -s
debugBlockDistBulkTransfer
you’ll see messages like
In BlockArr.doiBulkTransfer
In BlockArr.doiBulkTransferToDR()
pointing out that BulkTransferStride is in use.
There is another flag, debugCyclicDistBulkTransfer, for debugging
BulkTransferStride for cyclically distributed arrays.
Probably you have already read this paper, but just in case you haven't, here
is the link:
Global Data Re-allocation via Communication Aggregation in Chapel
Thanks,
Rafa.
El 18/06/2014, a las 21:57, Abhinav Sarje <[email protected]> escribió:
Hi,
I am trying to run the benchmarks, specifically FFT (from HPCC) and SSCA2, which are
bundled with Chapel. I want to enable the use of strided transfers, but have been unable
to get any run to use them. I am using the "-suseBulkTransferStride" option.
The stride transfer functions are being generated by the chpl compiler, as I see from the
saved C code.
Executing the benchmarks completes successfully, but never calls these strided
transfer functions. I have tried with up to 16 locales. I am using this with
GASNet.
Could someone please let me know what I am missing or doing wrong? Any help
appreciated.
Thanks,
Abhinav.
------------------------------------------------------------------------------
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast. Scalable. Simple. Ideal for Dirty Data.
Leverages Graph Analysis for Fast Processing & Easy Data Exploration
http://p.sf.net/sfu/hpccsystems_______________________________________________
Chapel-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-users
__
Rafael Asenjo Plaza
Dept. Arquitectura de Computadores
Complejo Tecnologico Campus de Teatinos
E-29071 MALAGA (SPAIN)
Tel: +34 95 213 27 91
Fax: +34 95 213 27 90
http://www.ac.uma.es/~asenjo
__
Rafael Asenjo Plaza
Dept. Arquitectura de Computadores
Complejo Tecnologico Campus de Teatinos
E-29071 MALAGA (SPAIN)
Tel: +34 95 213 27 91
Fax: +34 95 213 27 90
http://www.ac.uma.es/~asenjo
------------------------------------------------------------------------------
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast. Scalable. Simple. Ideal for Dirty Data.
Leverages Graph Analysis for Fast Processing & Easy Data Exploration
http://p.sf.net/sfu/hpccsystems
_______________________________________________
Chapel-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-users
