On Thu, Feb 5, 2009 at 17:10, Andreas Klöckner <[email protected]>wrote:

> On Donnerstag 05 Februar 2009, Nicholas Tung wrote:
> >  It appears that calling get_function twice, despite returning different
> > Python objects, will not return the same CUDA function object (CUfunction
> > is actually a pointer, ick...).
> >
> > Consider the attached patch to add a "reprepare" method such that only
> the
> > block size or shared memory is reset (I have not tested the shared memory
> > code, but it's the same as in the prepare method). I think this is a
> likely
> > scenario; it's unlikely the format of the arguments would change between
> > invocations.
>
> I've merged an edited form of your warning, but I don't see the point of
> the
> reprepare method. Instead, I'd suggest using set_shared and set_block_shape
> directly.


right, good idea. Perhaps this could be added to the documentation, along
with to_device and from_device?

thanks,
Nicholas
_______________________________________________
PyCuda mailing list
[email protected]
http://tiker.net/mailman/listinfo/pycuda_tiker.net

Reply via email to