request: MATLAB/octave wrappers

[ahbarnett]

Just lodging the request that we build some MATLAB/octave wrappers some day soon. Calling this from MATLAB would be really cool, as a drop-in replacement for FINUFFT.

[janden]

MATLAB may be doable, but as far as I know, Octave doesn't support GPU arrays natively (there's apparently a package called ocl, but I haven't played around with it) . I also don't know if MATLAB has a MEX interface for GPU arrays. My guess, however, is that it is significantly more painful to implement than the CPU MEX interface. If you think there's a large enough need, we could look into it.

[ahbarnett]

Matlab w/ Parallel Toolbox can interface to gpuArray objects directly, in a MEX-like fashion:

https://www.mathworks.com/help/parallel-computing/run-mex-functions-containing-cuda-code.html

A self-contained few days of interfacing there, adapting our FINUFFT matlab interfaces, a nice project for someone.

[janden]

All right, sure, but would you use it? I don't think it's a good idea to add a feature until we have a user (just one would be enough). Otherwise it's just adding to our maintenance (and testing) burden without much of a benefit.

[ahbarnett]

Dan Fortunato wants to do semi-langragian advection PDEs on the sphere using NUFFT to reinterpolate back to grids. I suggested cuFINUFFT. He works in Matlab mostly. Maybe he can write the interface? Or maybe Mathworks can license cuFINUFFT and write wrapper themselves. Either way, matlab opens up users. Just a thought.

…

On Mon, Sep 28, 2020 at 8:49 AM Joakim Andén ***@***.***> wrote: All right, sure, but would you use it? I don't think it's a good idea to add a feature until we have a user (just one would be enough). Otherwise it's just adding to our maintenance (and testing) burden without much of a benefit. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#59 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACNZRSXC7NPM5NOPIHHPDWLSICA4XANCNFSM4RU6VLCQ> .

-- *---------------------------------------------------------------------~^`^~._.~' |\ Alex H. Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942

[janden]

All right. If he would use it, it could make sense to add, I agree. Does he already use the gpuArray stuff in Matlab?

[ahbarnett]

I don't know. I can ask him @danfortunato

…

On Wed, Sep 30, 2020 at 4:32 AM Joakim Andén ***@***.***> wrote: All right. If he would use it, it could make sense to add, I agree. Does he already use the gpuArray stuff in Matlab? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#59 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACNZRSQIVE2NACVYL6NGCWTSILUKXANCNFSM4RU6VLCQ> .

-- *---------------------------------------------------------------------~^`^~._.~' |\ Alex H. Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942

[danfortunato]

Hi guys. I haven't played around with the gpuArray stuff in MATLAB (no NVIDIA chip in my laptop sadly). I don't have a pressing need for the cuFINUFFT wrappers, but I'm happy to help implement this. The semi-Lagrangian stuff doesn't need to be in MATLAB though (and maybe it shouldn't be).

[ahbarnett]

If you write it in py then cufinufft is ready to go.. !

…

On Thu, Oct 1, 2020 at 4:42 PM Dan Fortunato ***@***.***> wrote: Hi guys. I haven't played around with the gpuArray stuff in MATLAB (no NVIDIA chip in my laptop sadly). I don't have a pressing need for the cuFINUFFT wrappers, but I'm happy to help implement this. The semi-Lagrangian stuff doesn't need to be in MATLAB though (and maybe it shouldn't be). — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#59 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACNZRSU3G2UHNHKTQMLGWVDSITSUFANCNFSM4RU6VLCQ> .

-- *---------------------------------------------------------------------~^`^~._.~' |\ Alex H. Barnett Center for Computational Mathematics, Flatiron Institute | \ http://users.flatironinstitute.org/~ahb 646-876-5942

[tangjinchuan]

Hi, do we have an OpenCL implementation? I am a code contributor to Octave gpuArray (octave-ocl). If we have an OpenCL implementation like VKFFT. I am happy to work with my students to implement it in an undergraduate degree project.

Best wishes,
Jinchuan Tang

[janden]

Hi Jinchuan,

This is currently a CUDA-only implementation. We haven't looked into OpenCL, but I imagine this would be quite a significant undertaking, so don't think this is likely in the near future.

[tangjinchuan]

Dear Joakim, No problem. Maybe with the help of GPT-4/5, things will get easier in the distant future.

…

On Sun, 19 Nov 2023 at 06:46, Joakim Andén ***@***.***> wrote: Hi Jinchuan, This is currently a CUDA-only implementation. We haven't looked into OpenCL, but I imagine this would be quite a significant undertaking, so don't think this is likely in the near future. — Reply to this email directly, view it on GitHub <#59 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABQBUVG7QQWAGPU6HTPX6YDYFE3GBAVCNFSM4RU6VLC2U5DIOJSWCZC7NNSXTN2JONZXKZKDN5WW2ZLOOQ5TCOBRG43DMOBWGE2Q> . You are receiving this because you commented.Message ID: ***@***.***>

[mikecjz]

Referencing my example solution from the finufft repo flatironinstitute/finufft#432.