ConcreteArrays of non-primitive isbits types

[albertomercurio]

If I run the following code it fails

a = Iterators.product(1:3, 1:3) |> collect
Reactant.ConcreteRArray(a)

ERROR: MethodError: no method matching primitive_type(::Type{Tuple{Int64, Int64}})
The function `primitive_type` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  primitive_type(::Type{UInt16})
   @ Reactant ~/.julia/packages/Reactant/IgTfV/src/xla/Utils.jl:44
  primitive_type(::Type{UInt8})
   @ Reactant ~/.julia/packages/Reactant/IgTfV/src/xla/Utils.jl:44
  primitive_type(::Type{Reactant.F8E5M2FNUZ})
   @ Reactant ~/.julia/packages/Reactant/IgTfV/src/xla/Utils.jl:44
  ...

Stacktrace:
 [1] Reactant.XLA.PJRT.Buffer(client::Reactant.XLA.PJRT.Client, array::Matrix{Tuple{…}}, device::Reactant.XLA.PJRT.Device)
   @ Reactant.XLA.PJRT ~/.julia/packages/Reactant/IgTfV/src/xla/PJRT/Buffer.jl:12
 [2] Reactant.XLA.PJRT.AsyncBuffer(::Reactant.XLA.PJRT.Client, ::Vararg{Any}; kwargs::@Kwargs{})
   @ Reactant.XLA.PJRT ~/.julia/packages/Reactant/IgTfV/src/xla/PJRT/AsyncBuffer.jl:8
 [3] (::Reactant.Sharding.NoSharding)(client::Reactant.XLA.PJRT.Client, device::Reactant.XLA.PJRT.Device, x::Matrix{…})
   @ Reactant.Sharding ~/.julia/packages/Reactant/IgTfV/src/Sharding.jl:193
 [4] ConcretePJRTArray(data::Matrix{…}; client::Nothing, idx::Nothing, device::Nothing, sharding::Reactant.Sharding.NoSharding)
   @ Reactant ~/.julia/packages/Reactant/IgTfV/src/Types.jl:230
 [5] ConcretePJRTArray(data::Matrix{Tuple{Int64, Int64}})
   @ Reactant ~/.julia/packages/Reactant/IgTfV/src/Types.jl:222
 [6] top-level scope
   @ ~/GitHub/gcs_misc/notebooks/loop_spin_expval.jl:633
Some type information was truncated. Use `show(err)` to see complete types.

While it works if I convert it to a CuArray instead

julia> CuArray(a)
3×3 CuArray{Tuple{Int64, Int64}, 2, CUDA.DeviceMemory}:
 (1, 1)  (1, 2)  (1, 3)
 (2, 1)  (2, 2)  (2, 3)
 (3, 1)  (3, 2)  (3, 3)

Also, using to_rarray doesn't do anything

julia> Reactant.to_rarray(a)
3×3 Matrix{Tuple{Int64, Int64}}:
 (1, 1)  (1, 2)  (1, 3)
 (2, 1)  (2, 2)  (2, 3)
 (3, 1)  (3, 2)  (3, 3)

[albertomercurio]

Hello, any thoughts on this?

[wsmoses]

sorry yeah, currently only concretearrays of primitive types are supported. we may extend this later, but its nevertheless a limitation for now.

[albertomercurio]

Ok I see. It would be great in order to support custom kernels. It works on CUDA but not here.

Thanks anyway for the information.