Hi,
This is to tell you that I think the common gpu ndarray it ready. We
already integrated it in part in Theano, but we didn't finish
converting all old operation in the new one.
Andreas, do you want to look at it?
How would you like to add it in PyCUDA/PyOpenCL? As we don't have yet
all the fonctionality of the array in them, what about allowing to use
both, and to make the gpu ndarray have the same attribute for the gpu
data only for contigous data? That way, people will be able to reuse
it with the current PyCUDA/PyOpenCL implementation.
Here is the list of operation integrated:
https://github.com/abergeron/compyte/wiki/Comparison-of-existing-implementa…
The code is in this branch: https://github.com/abergeron/compyte/tree/reorg
Fred
p.s. I don't anouce it yet on Theano mailing list as I need to "fix" a
overhead issue that make it much slower then the old back-end. This is
Theano specific stuff. We swap between the C and Python context too
frequently.
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 01/11/2013 16:31, Isaac Gerg wrote:
> So back to my original question, does anyone know when will pycuda
> get 2d FFT support?
Probably never :) PyCUDA is a nice and simple wrapper around CUDA
rather than a full environment for scientific computing with CUDA.
This is the same reason why there is no wrapper around cuBLAS.
If none of the alternatives presented thus far are suitable then you
should consider wrapping CUFFT with ctypes. It is not particularly
difficult and is a viable route to both good performance and
compatibility with Python 3.x.
Regards, Freddie.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG/MacGPG2 v2.0.20 (Darwin)
Comment: GPGTools - http://gpgtools.org
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
iEYEARECAAYFAlJ0HTgACgkQ/J9EM/uoqVepMgCfcdGRYI/5CnZeV5j4IilMumra
rZUAn2tAM+xoG+y2Yv7q4vb4jCH3fYye
=+2D9
-----END PGP SIGNATURE-----
So back to my original question, does anyone know when will pycuda get 2d
FFT support?
On Thu, Oct 31, 2013 at 6:12 PM, Bogdan Opanchuk <mantihor(a)gmail.com> wrote:
> Hi Isaac,
>
> It does not need one, it's a pure Python module. Just run "pip install
> reikna".
>
> On Fri, Nov 1, 2013 at 9:05 AM, Isaac Gerg <isaac.gerg(a)gergltd.com> wrote:
> > There's no windows 64bit installer.
> >
> > On Oct 31, 2013 5:44 PM, "Bogdan Opanchuk" <mantihor(a)gmail.com> wrote:
> >>
> >> Hi Isaac,
> >>
> >> You can try my package Reikna (http://reikna.publicfields.net). The
> >> FFT there is somewhat slower than the CUFFT one, but it works with
> >> Python 3.
> >>
> >> On Thu, Oct 31, 2013 at 11:54 PM, Isaac Gerg <isaac.gerg(a)gergltd.com>
> >> wrote:
> >> > They have no support for python 3.2 64 bit :(
> >> >
> >> > On Oct 31, 2013 5:44 AM, "Ahmed Fasih" <wuzzyview(a)gmail.com> wrote:
> >> >>
> >> >> scikits.cuda has both high-level and low-level wrappers from PyCUDA
> to
> >> >> CUFFT for all your FFT needs: use the former to do standard tasks
> >> >> conveniently, and the latter to make full use of the CUFFT API.
> >> >> http://lebedov.github.io/scikits.cuda/reference.html
> >> >>
> >> >>
> >> >> On Wed, Oct 30, 2013 at 10:42 PM, Isaac Gerg <isaac.gerg(a)gergltd.com
> >
> >> >> wrote:
> >> >>>
> >> >>> When will pycuda get 2d fft support?
> >> >>>
> >> >>> Isaac
> >> >>>
> >> >>> _______________________________________________
> >> >>> PyCUDA mailing list
> >> >>> PyCUDA(a)tiker.net
> >> >>> http://lists.tiker.net/listinfo/pycuda
> >> >>>
> >> >>
> >> >
> >> > _______________________________________________
> >> > PyCUDA mailing list
> >> > PyCUDA(a)tiker.net
> >> > http://lists.tiker.net/listinfo/pycuda
> >> >
>
