Implement sorting with AK

[christiangnrd]

AK does not (yet?) support a dims argument. Porting over the CUDA sort to KernelAbstractions will require some more thought as both sorting algorithms use dynamic shared memory.

[maleadt]

Porting over the CUDA sort to KernelAbstractions will require some more thought as both sorting algorithms use dynamic shared memory.

I'd punt on that, and start by using KA.jl for sorting without dims. Porting that is "wasted" effort better spent on adding dims support to KA.jl, I'd guess.

[christiangnrd]

I'd punt on that, and start by using KA.jl for sorting without dims. Porting that is "wasted" effort better spent on adding dims support to KA.jl, I'd guess.

@maleadt When you write KA (presumably KernelAbstractions) do you mean AK (AcceleratedKernels)?

[maleadt]

@maleadt When you write KA (presumably KernelAbstractions) do you mean AK (AcceleratedKernels)?

Sorry, yeah

[maleadt]

Superseded by #790, which implements sorting for every GPU back-end on top of AcceleratedKernels.jl, released in GPUArrays 12. Thanks!