CGAL users discussion list

[Marc Glisse <>]

On Tue, 1 May 2012, Ioannis wrote:

> > It is not possible to make reference-counting thread-safe without slowing
> > down single-threaded applications, so this hasn't been done yet. It is
> > safe to use different Gmpq in different threads (as opposed to Gmpzf which
> > may only be used in a single thread), but not to share one between
> > threads.
>
> > You could try to use a type that does not use reference counting
> > (mpq_class (from gmpxx.h) instead of Gmpq).
>
> I am using CCAL 4.0 with with Exact_predicates_exact_constructions kernel
> and my multithreaded code fails occasionally.
> Is the thread unsafe reference counting still an issue in the new version?
> The same problem I have with multithreading and the Cartesian<Cgal::Gmpq>
> kernel.

Yes, the paragraph you quoted is still valid. I guess we'll have to 
provide an option to make reference-counting thread-safe (and slower) at 
some point, for Gmpq it shouldn't be hard, especially if we use C++11 and 
don't try to squeeze the last microsecond of performance out of it. For 
other types (including the kernel objects in 
Exact_predicates_exact_constructions), it might be a bit harder as the 
interface wasn't written with atomicity in mind.

> Is the Simple_cartesian<mpq_class> the only alternative solution and how
> slow could that be comparing to the other kernels?

Compared to Cartesian<Gmpq>, and assuming objects aren't copied around all 
the time, it should be about the same. Compared to 
Exact_predicates_exact_constructions, it can be quite a bit slower.

> I saw in an older post that some filtering could speed it up but I am not
> sure what does that mean in this context and how can I use it in my code.

http://www.cgal.org/Manual/latest/doc_html/cgal_manual/Kernel_23_ref/Class_Filtered_kernel.html

--
Marc Glisse