Add support for KernelAbstraction 0.9

[vchuravy]

No description provided.

[maleadt]

i128 failure was a typo in the skip list. 1.6 issues remain, and are different.

julia> using oneAPI, oneAPI.oneAPIKernels, KernelAbstractions

julia> backend = oneAPIBackend()
oneAPIBackend()

julia> @kernel function index_linear_global(A)
              I = @index(Global, Linear)
              A[I] = I
       end
index_linear_global (generic function with 4 methods)

julia> A = allocate(backend, Int, 16, 16)
16×16 oneMatrix{Int64, oneAPI.oneL0.DeviceBuffer}:
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0
 0  0  0  0  0  0  0  0  0  0  0  0  0  0  0  0

julia> index_linear_global(backend, 8)(A, ndrange=length(A))
ERROR: GPU compilation of MethodInstance for gpu_index_linear_global(::KernelAbstractions.CompilerMetadata{KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIteration.DynamicCheck, Nothing, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, KernelAbstractions.NDIteration.NDRange{1, KernelAbstractions.NDIteration.DynamicSize, KernelAbstractions.NDIteration.StaticSize{(8,)}, CartesianIndices{1, Tuple{Base.OneTo{Int64}}}, Nothing}}, ::oneDeviceMatrix{Int64, 1}) failed
KernelError: kernel returns a value of type `Union{}`

[maleadt]

Hmm, it's simply the call to __validindex that fails:

CodeInfo(
    @ /home/tim/Julia/pkg/KernelAbstractions/src/macros.jl:80 within `gpu_index_linear_global'
1 ─     (KernelAbstractions.__validindex)(__ctx__)::Union{}
└──     unreachable

@vchuravy Any thoughts?

EDIT: Oh, I wonder if this is an issue with the world age overlay table.

[vchuravy]

Yeah you likely have to do https://github.com/JuliaGPU/CUDA.jl/blob/5c51766d0a9e7819ea79f314e37ed6a8a5d24369/src/CUDAKernels.jl#L149-L175 if you want to support 1.6

[maleadt]

That's unexpected. But I don't feel like debugging this, so 🤷

[codecov]

Codecov Report

Patch coverage: 64.13% and project coverage change: -0.86 ⚠️

Comparison is base (e821aa4) 86.30% compared to head (1b3a2b3) 85.44%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #300      +/-   ##
==========================================
- Coverage   86.30%   85.44%   -0.86%     
==========================================
  Files          39       40       +1     
  Lines        1825     1917      +92     
==========================================
+ Hits         1575     1638      +63     
- Misses        250      279      +29     

Impacted Files	Coverage Δ
src/oneAPI.jl	80.00% <ø> (ø)
src/oneAPIKernels.jl	64.13% <64.13%> (ø)

... and 2 files with indirect coverage changes

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.