Rewrite calls to eg. Base.sin

[maleadt]

We should figure out a way to support users calling Base.sin instead of CUDAnative.sin. One possible option would be to hook inference, overriding method calls when operating under @cuda. The same mechanism could be used to dispatch on the hardware generation.

[maleadt]

xref JuliaGPU/CUDAnative.jl#61

[cdsousa]

Any hope this can be sorted out in the near future? This issue is preventing me from using third-party libraries that I guess would work perfectly fine in CUDA.jl.
(edit: wrong example given, see next comment) For example, I'm calling some function of a library with a StaticArrays vector as an argument, and that function internally calls a normalize function that uses sqrt. I don't see how can I work around this.

[cdsousa]

Sorry, my bad. I see that it actually works, I was getting another problem but thinking it was this one!
Anyway, there is still a problem with sin giving a warning.

[maleadt]

Hopefully the new compiler interfaces in 1.6 will make this possible, yes. But still months away at the least.