Simon

Can you point me to where this could be done? It wasn't easy for me to find the place for such additions, since everything seems to be done via the...

I would start one step earlier: Define a sensible GPU buffer/array type and operations like broadcast, converts between backends (if possible, no copy), and a solid set of helper functions...

Lol, the assertslow is there for a reason, since scalar indexing into a gpuarray is very slow and should not be done! You can do `allowslow(true)` to disable that error....

Seems like the code in Base doesn't route us to the `Base._unsafe_getindex!` we overload. Was this not the case in CuArrays?

Ah, that segfaulted at some point, so I was using the Base fallback with CLArrays! I should try if that's still the case! I have an idea why this might...

clFFT (which is probably also their CPU backend) is quite likely the culprit here ;) I coincidentally just tried out clFFT directly and got a very similar result: `maximum(abs(a -...

CUDA backend seems to be in the same range: ``` Julia julia> maximum(abs(af-Array(bf))) 0.0001501969f0 ```

>I wonder if it would be easier to port James’s kernel to OpenCL versus We can just port it to julia in a way that it works with both CLArrays...

That sounds like it actually messes with ImageMagick! See: https://github.com/JuliaIO/ImageMagick.jl/issues/25

If the ArrayFire C lib actually uses ImageMagick, it wouldn't be a big surprise if that messes things up. ImageMagick has a super messy OS global initialization procedure -.-