Allow sorting arrays of more than 2^30 elements

[pawel-tarasiuk-quantumz]

This allows sorting arrays of more than 2^30 elements.

In merge_sort.jl, half_size_group is currently converted to Int32. As a result, sorting arrays with more than 2^30 elements raises a DivideError.

The expected behavior is that arrays should be sortable up to the limits of available memory.

Proposed fix skips the Int32 conversion when the number of elements exceeds 2^30.

MWE (works with proposed changes):

using AcceleratedKernels
using CUDA

A = CUDA.rand(2^30 + 1)
AcceleratedKernels.sort!(A)

@show issorted(Vector(A))

Current result:

ERROR: LoadError: DivideError: integer division error
Stacktrace:
  [1] div
    @ ./int.jl:295 [inlined]
  [2] div
    @ ./div.jl:345 [inlined]
  [3] div
    @ ./div.jl:49 [inlined]
  [4] merge_sort!(v::CuArray{…}, backend::CUDABackend; lt::Function, by::Function, rev::Nothing, order::Base.Order.ForwardOrdering, block_size::Int64, temp::Nothing)
    @ AcceleratedKernels .../AcceleratedKernels.jl/src/sort/merge_sort.jl:177
  [5] merge_sort!
    @ .../AcceleratedKernels.jl/src/sort/merge_sort.jl:139 [inlined]
  [6] #_sort_impl!#49
    @ .../AcceleratedKernels.jl/src/sort/sort.jl:100 [inlined]
  [7] _sort_impl!
    @ .../AcceleratedKernels.jl/src/sort/sort.jl:81 [inlined]
  [8] #sort!#48
    @ .../AcceleratedKernels.jl/src/sort/sort.jl:74 [inlined]
  [9] sort!
    @ .../AcceleratedKernels.jl/src/sort/sort.jl:70 [inlined]
 [10] sort!(v::CuArray{Float32, 1, CUDA.DeviceMemory})
    @ AcceleratedKernels .../AcceleratedKernels.jl/src/sort/sort.jl:70
 [11] top-level scope
    @ .../mwe.jl:5
 [12] include(fname::String)
    @ Main ./sysimg.jl:38
 [13] top-level scope
    @ REPL[1]:1
in expression starting at .../mwe.jl:5
Some type information was truncated. Use `show(err)` to see complete types.