Minimizing sum fails

[cossio]

T = randn(10,10)
ps = Flux.params(T)
opt=ADAM()
for iter = 1:100
    gs = gradient(ps) do
        sum(T)
    end
    Flux.update!(opt, ps, gs)
end

Throws the following error:

ArgumentError: Cannot setindex! to 0.0009999999900000003 for an AbstractFill with value 1.0.

Stacktrace:
[1] setindex! at /home/cossio/.julia/packages/FillArrays/tE9Xq/src/FillArrays.jl:41 [inlined]
[2] _setindex! at ./abstractarray.jl:1176 [inlined]
[3] setindex! at ./abstractarray.jl:1153 [inlined]
[4] macro expansion at ./broadcast.jl:932 [inlined]
[5] macro expansion at ./simdloop.jl:77 [inlined]
[6] copyto! at ./broadcast.jl:931 [inlined]
[7] copyto! at ./broadcast.jl:886 [inlined]
[8] materialize! at ./broadcast.jl:848 [inlined]
[9] materialize!(::FillArrays.Fill{Float64,2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}}, ::Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2},Nothing,typeof(*),Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2},Nothing,typeof(/),Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2},Nothing,typeof(/),Tuple{Array{Float64,2},Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{0},Nothing,typeof(-),Tuple{Int64,Float64}}}},Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2},Nothing,typeof(+),Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2},Nothing,typeof(sqrt),Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{2},Nothing,typeof(/),Tuple{Array{Float64,2},Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{0},Nothing,typeof(-),Tuple{Int64,Float64}}}}}},Float64}}}},Float64}}) at ./broadcast.jl:845
[10] apply!(::ADAM, ::Array{Float64,2}, ::FillArrays.Fill{Float64,2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}}) at /home/cossio/.julia/packages/Flux/05b38/src/optimise/optimisers.jl:177
[11] update!(::ADAM, ::Array{Float64,2}, ::FillArrays.Fill{Float64,2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}}) at /home/cossio/.julia/packages/Flux/05b38/src/optimise/train.jl:23
[12] update!(::ADAM, ::Params, ::Zygote.Grads) at /home/cossio/.julia/packages/Flux/05b38/src/optimise/train.jl:29
[13] top-level scope at In[89]:8
[14] include_string(::Function, ::Module, ::String, ::String) at ./loading.jl:1091

[cossio]

https://discourse.julialang.org/t/gradient-of-sum/55389

[CarloLucibello]

I think we just have to handle Fills properly in apply!

[DhairyaLGandhi]

This is likely a regression in NNlib or Zygote, because we haven't broken the optimisation in here

[cossio]

I don't think Zygote is the issue, since the gradient is properly computed. Maybe apply! should dispatch on the type of gs, and mutate only if appropriate.

For instance, something like

apply!(opt, x, g::AbstractArray) = apply!(opt, x, Array(g)) # copy g if it is not mutable
apply!(opt, x, g::Array) = ... # normal code here that mutates g

[DhairyaLGandhi]

I think its that Fill doesn't want to eagerly collect, which is great, but it can lead to edge cases