Re: Parallel Octave

[CdeMills]

Max Brister wrote
> 
> 
> The level of parallelism Jean is talking about is possible without
> parallelizing the interpreter (which would be a lot of work). This is
> possible so long as all the parallel code is implemented in C or C++,
> and not in Octave.
> 
> It should also be possible to make an octave forge package that
> provides parallel implementations of functions like eig using PETSEC
> or SLEPC. It looks like there are already packages in octave forge
> that provide access to mpi in Octave (parallel and openmpi_ext).
> 
> 
Indeed. There is a short-term target, using parallel version of some libs;
and long-term target, making the interpreter thread-aware. 

I was talking about C++ constructs like

do_something ()
if (! error_state)
  do_the_next_thing ()
else
  ::error ("do_something failed")
endif

There are a few variables, like error_state, which are set and writen at
many places in the C++ code. Such variables should be made thread specific.

Regards

Pascal 


--
View this message in context: 
http://octave.1599824.n4.nabble.com/Parallel-Octave-tp4530463p4536871.html
Sent from the Octave - Maintainers mailing list archive at Nabble.com.