FFT support

[jakirkham]

Just opening this to note interest in having FFT support. Is this something that is planned at some point?

[nouiz]

In Theano, we wrapped cufft. We won't manually implement an fft here. You can use skcuda cufft wrapper with libgpuarray as we done in Theano. Do you want to use cuda or OpenCL? Do you know and OpenCL try implementation? Le sam. 24 juin 2017 23:36, jakirkham <[email protected]> a écrit :

…

Just opening this to note interest in having FFT support. Is this something that is planned at some point? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#463>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-5znXE5ruAOkEWqggkdsvldSifxLks5sHdXKgaJpZM4OEgPZ> .

[jakirkham]

I see. Wasn't sure exactly what all was being considered in scope for this project. Generally got the impression that GPU array computations from Theano were being offloaded to libgpuarray/pygpu.

Ideally was wanting to use both CUDA and OpenCL. As far as OpenCL implementations, clFFT comes to mind.

[jakirkham]

To be clear, wasn't expecting FFTs to be implemented in this library. More than happy to have an interface to cuFFT and clFFT as was done with BLAS.

[nouiz]

I agree, an interface that wrap cufft and clfft fit in this project. Contribution welcome. It is currently not on our short term project we plan to work on.

…

On Mon, Jun 26, 2017 at 11:04 AM jakirkham ***@***.***> wrote: To be clear, wasn't expecting FFTs to be implemented in this library. More than happy to have an interface to cuFFT and clFFT as was done with BLAS. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#463 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-_OvCYTcB72ls3JAAq-IgMi7q49Fks5sH8iAgaJpZM4OEgPZ> .