Re: OpenMP in oct files

[Daniel Kraft]

Hi again,

On 2013-09-23 13:13, Mike Puglia wrote:
> I'm certainly no expert in this, so I'm not sure I can answer your
> questions well, but I had almost the same problems getting openmp to
> work in Octave as well.  I don't know how the matrix classes are
> written, but I assume when you start accessing a matrix from multiple
> threads at the same time there will be a problem with pointers flying
> all over at least.  I tried to parallelize a few very simple for loops
> that acted on a (octave) matrix class and all I can say is the result
> was unstable.  Although all of the for loops were pretty similar, I had
> cases where the parallel loop worked flawlessly with the expected
> accelerations, I had cases where the loops worked flawlessly but with no
> acceleration (though multiple threads were active) and I had cases where
> the results were incorrect.  I don't think I had any problems with
> crashing however.  After looking into it I'm pretty sure that I wasn't
> making any of the basic mistakes for working with openmp.  All of the
> loops were pretty simple and should have had the same behavior (i.e.
> they all work or they all don't work, they are all accelerated or they
> are all not accelerated).  That was not the case however.  I'm sure
> someone with better knowledge of the matrix classes can explain why, but
> I eventually gave up on it.

thanks for confirming my results. :)  I read already multiple times that
Octave is far from being thread-safe, thus my suspicion is indeed that
this is the problem.  (Even though it seems to me that the basic matrix
classes are effectively hard to make *not* thread-safe.)

My code should of course be fine for parallelisation (no wrong data
dependencies), but I think I'll just try to rewrite the relevant
sections without using Octave routines at all and see if that helps.
With Fortran it should be more or less straight-forward to replace the
Octave classes, and it works very well to link to Fortran from oct-files
(I already did that in the past successfully).

> Regarding CUDA, I'm just starting to work on that myself.  Did you use
> the VS2010 build by any chance?  If not, how did you link with the nvcc
> compiler output?

Take a look at [1], which is the seminar report I wrote about my
experiments.  I have to admit that I never used CUDA for anything "real"
so far (apart from those toy experiments).  I worked on GNU/Linux with
nvcc.  The basic idea is the same, just compile the CUDA code with nvcc
into a shared library and use it from the oct-file code.

  [1] http://www.domob.eu/research/OctaveGPU.pdf

Yours,
Daniel

-- 
http://www.domob.eu/
--
Done:  Arc-Bar-Cav-Hea-Kni-Ran-Rog-Sam-Tou-Val-Wiz
To go: Mon-Pri