Support conv_direct! on custom datatypes

[adrhill]

Closes #490, #405.

As described in #490, undefined array elements can't be indexed on all datatypes:

julia> v1 = Array{Float32}(undef, 3)
3-element Vector{Float32}:
 2.7761215f-21
 1.0f-45
 2.7765319f-21

julia> v2 = Array{Set}(undef, 3)
3-element Vector{Set}:
 #undef
 #undef
 #undef

julia> v1[1] # works
2.7761215f-21

julia> v2[1] # doesn't work
ERROR: UndefRefError: access to undefined reference
Stacktrace:
 [1] getindex(A::Vector{Set}, i1::Int64)
   @ Base ./essentials.jl:13
 [2] top-level scope
   @ REPL[34]:1

Such indexing is used in the right-hand side of conv_direct! on main

NNlib.jl/src/impl/conv_direct.jl

Line 111 in 627374c

y[w_idx, h_idx, d_idx, c_out, batch] = alpha*dotprod + beta*y[w_idx, h_idx, d_idx, c_out, batch]

which means that conv_direct! doesn't support all input datatypes.

However, assignments work, which is way the problem can be solved by first setting y to zero, as suggested here by @ToucheSir.

[adrhill]

Looks like failures on nightly are unrelated.

[adrhill]

@mcabbott could a patch release be tagged for this?

[pxl-th]

Tagged 0.9.18 release: 62f6074

[adrhill]

Thanks!

[Seelengrab]

As I mentioned in the original issue, naive fill!/fill doesn't necessarily work with e.g. BigInt:

julia> a = Vector{BigInt}(undef, 2)
2-element Vector{BigInt}:
 #undef
 #undef

julia> fill!(a, zero(BigInt))
2-element Vector{BigInt}:
 0
 0

julia> a[1] ===a[2]
true

I don't know how important these kinds of usecases are for NNlib.jl though.

[adrhill]

Is that problematic here?

julia> a = Vector{BigInt}(undef, 2)
2-element Vector{BigInt}:
 #undef
 #undef

julia> fill!(a, zero(BigInt))
2-element Vector{BigInt}:
 0
 0

julia> for (i, ai) in enumerate(a)
         a[i] = i + 0 * a[i]
       end

julia> a
2-element Vector{BigInt}:
 1
 2