CGAL users discussion list

["Maddox, Willie B. IV" <>]

Laurent,

One example of a CUDA-CGAL implementation that my group would find useful 
would be in finding the intersection of a large list of CGAL objects (e.g. 
10^5 Ray_3 objects) with an AABB_Triangle_Tree.  The Idea would be to 
"somehow" load the AABB_Tree and the list of objects into device memory, and 
then distribute the objects across the CUDA cores such that multiple 
intersection tests can be computed in parallel.

Using the cmake "findCUDA" module, I've been playing around with a few 
CUDA-CGAL test programs.  In one test, I attempt to pass a single CGAL 
Point_2 object to device memory, make a copy, and then pass it back to the 
host.  Unfortunately, CUDA doesn't seem to understand the CGAL Point_2 
datatype and I am not even sure if it should.  

There are also a few other "gotchas" that I have encountered. For example, 
the memory allocated on the host for a Delaunay_triangulation_2 object does 
not change with the size of the triangulation.  On my system the value is 
always "memSize = sizeof(dt2)" = 184 Bytes regardless of the number of 
vertices. This becomes a problem when the compiler attempts to allocate 
device memory via "cudaMalloc(d_dt2, &dt2, memSize)".

Our current implementation of CGAL utilizes CPU parallel processing. While it 
is faster than serial processing, it is still not as fast as we need it to 
be.  Utilizing our existing CUDA cores would be an attractive alternative on 
many levels (i.e. maintenance, size, cost, power consumption, etc.) to a CPU 
cluster.

-Will

-----Original Message-----
From: 

 
[mailto:]
 On Behalf Of Laurent Rineau (CGAL/GeometryFactory)
Sent: Tuesday, February 11, 2014 7:12 AM
To: 

Subject: Re: [cgal-discuss] CGAL+CUDA?

Le Tuesday 11 February 2014 02:07:52 Maddox, Willie B. IV a écrit :
> Are there plans to add CUDA functionality to CGAL in the near future? 

No, there are no short-term plan. In which parts of CGAL would you see 
benefits of using CUDA?

There was a work done around 2012 by the MPII, to introduce the use of CUDA 
in the internals of the CGAL Algebraic Kernel, but that work seems stalled.

--
Laurent Rineau, PhD
R&D Engineer at GeometryFactory           http://www.geometryfactory.com/
Release Manager of the CGAL Project       http://www.cgal.org/


-- 
You are currently subscribed to cgal-discuss.
To unsubscribe or access the archives, go to
https://sympa.inria.fr/sympa/info/cgal-discuss