Array "equivalent" on the GPU?

[Pontus Stenetorp]

Everyone,

Do we have some sort of GPGPUArray type that would allow us to switch
(reasonably) seamlessly between CPU and GPU-based computing?
Supporting most of the semantics of the Array provided in base,
multiplication, summing, element-wise operations, etc.

The reason that I am writing this is that the issue of Julia and GPGPU
computing for end-users has come up with my colleagues over the last
few months and I suspect that Torch, Theano, et al. are ahead of us in
this aspect. Which, if it is the case, is unacceptable and something
that I would love to help out. We must not allow a GPGPU gap!

Pontus

[Tim Holy]

I can't speak for any packages other than CUDArt. It provides a few operations
on arrays (https://github.com/JuliaGPU/CUDArt.jl/blob/master/src/arrays.jl),
but nothing comprehensive. Mostly stuff like copy!, fill!, and similar.

--Tim

[Eric Forgy]

I'm looking at ArrayFire.jl that seems like it would fit the bill (but you probably know that :))

[Valentin Churavy]

Simon just started work on https://github.com/JuliaGPU/GPUArrays.jl and contributions and design input are welcome!