[ITK-dev] [ITK] New Highly Parallel Build System, the POWER8

[Will Schroeder]

On Fri, Apr 24, 2015 at 9:34 AM, Bradley Lowekamp <blowekamp at mail.nih.gov>
wrote:

> I hope were are able to improve ITK threading performance with this
> system. But due to the tweaky-ness of this type of performance and not
> having direct access to the system to easily run performance analysis, I
> and a little unclear how to best utilize it.


Since I have recently been working on coarse grained parallelism lately, I
have been using the vtkSMPTools abstraction that Berk Geveci and his team
designed. Under the hood it uses one of Intel TBB, X-Kaapi, raw threads,
serial, and soon OpenMP (picked at compile time). It is fantastically easy
to use, with built in load balancing (via thread stealing). I have
tremendous respect for what Berk's team is doing...fine-grained (e.g., GPU
via VTKm), coarse grained (vtkSMPTools), and data parallel (advanced
execution architectures and data models in VTK). Berk is PI and involved in
several projects with the DOE labs as they move towards exascale computing
and really we should be tapping that community more. For example,
VTK/ParaView/Catalyst are running nicely on hundreds of thousands of
processors in some hero runs.

As another example, we designed a threaded contouring algorithm (Flying
Edges) that just knocks the pants off of any other isocontouring algorithm
I've seen (we've submitted a paper). It uses this vtkSMPTools abstraction
and we were able to easily run it on this POWER8 system (and other
systems). It was really cool to run the algorithm on a 2048^3 dataset and
see all 160 cores pegged :-)

W


-- 
William J. Schroeder, PhD
Kitware, Inc.
28 Corporate Drive
Clifton Park, NY 12065
will.schroeder at kitware.com
http://www.kitware.com
(518) 881-4902
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-developers/attachments/20150424/a1e059fa/attachment.html>