Initial KernelAbstractions.jl integration.

[maleadt]

Demo:

using Test
using OpenCL, pocl_jll
using KernelAbstractions, OpenCL.OpenCLKernels

backend = OpenCLBackend()

@kernel function copy_kernel!(A, @Const(B))
    I = @index(Global)
    @inbounds A[I] = B[I]
end

function mycopy!(A, B)
    backend = get_backend(A)
    @assert size(A) == size(B)
    @assert get_backend(B) == backend

    kernel = copy_kernel!(backend)
    kernel(A, B, ndrange = length(A))
end

A = KernelAbstractions.zeros(backend, Float64, 128, 128)
B = KernelAbstractions.ones(backend, Float64, 128, 128)
mycopy!(A, B)
KernelAbstractions.synchronize(backend)
@test A == B

julia> @test A == B
Test Passed

cc @vchuravy

[codecov]

Codecov Report

Attention: Patch coverage is 60.97561% with 32 lines in your changes missing coverage. Please review.

Project coverage is 72.72%. Comparing base (daaf3cf) to head (d677ee2).

Files with missing lines	Patch %	Lines
src/OpenCLKernels.jl	61.03%	30 Missing ⚠️
src/array.jl	60.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #233      +/-   ##
==========================================
+ Coverage   71.60%   72.72%   +1.12%     
==========================================
  Files           7        8       +1     
  Lines         324      407      +83     
==========================================
+ Hits          232      296      +64     
- Misses         92      111      +19     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[vchuravy]

Excellent!

[vchuravy]

The last can of worms is probably atomics...

[maleadt]

Hopefully that shouldn't too hard, I have them wrapped already from oneAPI.jl: https://github.com/JuliaGPU/OpenCL.jl/blob/master/lib/intrinsics/src/atomic.jl
How do these need to be hooked up to KA.jl?

EDIT: Atomix.jl apparently, so I guess this is a matter of duplicating https://github.com/JuliaConcurrent/Atomix.jl/tree/main/lib/AtomixCUDA