repeat does not work with KnetArray

[johnberg1]

I was using repeat() function to replicate a Float64 array and the following code seems to be working:
a = randn(256, 5)
repeat(a,16,16)

but when I cast the array to a KnetArray type and try again,
a = KnetArray{}(randn(256, 5))
repeat(a,16,16)

I encounter the following error:

MethodError: no method matching repeat(::KnetArray{Float64,2}, ::Int64, ::Int64) Closest candidates are: repeat(!Matched::Union{GPUArrays.AbstractGPUArray{T,2}, GPUArrays.AbstractGPUArray{T,1}} where T, ::Int64, ::Int64) at C:\Users\canbe\.julia\packages\GPUArrays\Ck0bk\src\host\base.jl:3 repeat(!Matched::Union{AbstractArray{T,2}, AbstractArray{T,1}} where T, ::Integer, ::Integer) at abstractarraymath.jl:275 repeat(!Matched::AbstractArray, ::Integer...) at abstractarraymath.jl:273 ...

It looks like repeat() does not support KnetArray type. I'm using Julia Version 1.5.2 and Knet v1.4.3

[denizyuret]

As a temporary workaround until I fix this, I would recommend:

KnetArray(repeat(CuArray(a),16,16)))

Or use CuArray as your array type, might be a bit slower but supports more operations.

[Alexander-Barth]

Apparently, CUDA (as of 3.9.0) does not have such kernel either:

using CUDA; repeat(cu(ones(2,2)),1,1,3)
┌ Warning: Performing scalar indexing on task Task (runnable) @0x00007f7d738a0010.
│ Invocation of getindex resulted in scalar indexing of a GPU array.
│ This is typically caused by calling an iterating implementation of a method.
│ Such implementations *do not* execute on the GPU, but very slowly on the CPU,
│ and therefore are only permitted from the REPL for prototyping purposes.
│ If you did intend to index this array, annotate the caller with @allowscalar.

My workaround is to use an array of indices:

KnetArray(ones(2,2))[:,:,ones(Int,3)]