remove @adjoint for hv/h/v/cat

[mzgubic]

Closes #1217

Seems fine locally, let's see what integration tests say.

[ToucheSir]

GPU failure on https://github.com/FluxML/Zygote.jl/blob/528e0be677d1feb9ccf6fc4ab298f4d8a106de10/test/cuda.jl#L134looks real, but is masked on CI by an odd show-related error. Running standalone:

julia> gradient((x,y) -> sum(vcat(x,y)), r, 5)
ERROR: MethodError: no method matching ndims(::Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{0}, Nothing, typeof(identity), Tuple{CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}}})
Closest candidates are:
  ndims(::CUDA.CUSPARSE.CuSparseMatrix) at ~/.julia/packages/CUDA/DfvRa/lib/cusparse/array.jl:225
  ndims(::CUDA.CUSPARSE.CuSparseDeviceMatrixBSR) at ~/.julia/packages/CUDA/DfvRa/lib/cusparse/device.jl:64
  ndims(::Base.Iterators.ProductIterator) at iterators.jl:1015
  ...
Stacktrace:
  [1] check_reducedims(R::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}, A::Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{0}, Nothing, typeof(identity), Tuple{CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}}})
    @ Base ./reducedim.jl:244
  [2] mapreducedim!(f::typeof(identity), op::typeof(Base.add_sum), R::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}, A::Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{0}, Nothing, typeof(identity), Tuple{CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}}}; init::Float32)
    @ CUDA ~/.julia/packages/CUDA/DfvRa/src/mapreduce.jl:172
  [3] _mapreduce(f::typeof(identity), op::typeof(Base.add_sum), As::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}; dims::Colon, init::Nothing)
    @ GPUArrays ~/.julia/packages/GPUArrays/Hyss4/src/host/mapreduce.jl:69
  [4] mapreduce(::Function, ::Function, ::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}; dims::Function, init::Nothing)
    @ GPUArrays ~/.julia/packages/GPUArrays/Hyss4/src/host/mapreduce.jl:31
  [5] mapreduce(::Function, ::Function, ::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer})
    @ GPUArrays ~/.julia/packages/GPUArrays/Hyss4/src/host/mapreduce.jl:31
  [6] _sum(f::Function, a::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}, ::Colon; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Base ./reducedim.jl:999
  [7] _sum(f::Function, a::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}, ::Colon)
    @ Base ./reducedim.jl:999
  [8] _sum(a::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}, ::Colon; kw::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Base ./reducedim.jl:998
  [9] _sum(a::CuArray{Float32, 0, CUDA.Mem.DeviceBuffer}, ::Colon)
    @ Base ./reducedim.jl:998
 [10] #sum#772
    @ ./reducedim.jl:994 [inlined]
 [11] sum
    @ ./reducedim.jl:994 [inlined]
 [12] (::ChainRules.var"#1342#1345"{Base.RefValue{Int64}, CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, Val{1}})(project::ChainRulesCore.ProjectTo{Float64, NamedTuple{(), Tuple{}}}, sizeX::Tuple{})
    @ ChainRules ~/.julia/packages/ChainRules/EyLkg/src/rulesets/Base/array.jl:310
 [13] map
    @ ./tuple.jl:247 [inlined]
 [14] (::ChainRules.var"#vcat_pullback#1344"{Tuple{ChainRulesCore.ProjectTo{AbstractArray, NamedTuple{(:element, :axes), Tuple{ChainRulesCore.ProjectTo{Float32, NamedTuple{(), Tuple{}}}, Tuple{Base.OneTo{Int64}}}}}, ChainRulesCore.ProjectTo{Float64, NamedTuple{(), Tuple{}}}}, Tuple{Tuple{Int64}, Tuple{}}, Val{1}})(ȳ::CuArray{Float32, 1, CUDA.Mem.DeviceBuffer})
    @ ChainRules ~/.julia/packages/ChainRules/EyLkg/src/rulesets/Base/array.jl:295
 [15] ZBack
    @ ~/.julia/dev/Zygote/src/compiler/chainrules.jl:205 [inlined]
 [16] Pullback
    @ ./REPL[6]:1 [inlined]
 [17] (::typeof(∂(#3)))(Δ::Float32)
    @ Zygote ~/.julia/dev/Zygote/src/compiler/interface2.jl:0
 [18] (::Zygote.var"#60#61"{typeof(∂(#3))})(Δ::Float32)
    @ Zygote ~/.julia/dev/Zygote/src/compiler/interface.jl:45
 [19] gradient(::Function, ::CuArray{Float32, 1, CUDA.Mem.DeviceBuffer}, ::Vararg{Any})
    @ Zygote ~/.julia/dev/Zygote/src/compiler/interface.jl:97
 [20] top-level scope
    @ REPL[6]:1
 [21] top-level scope
    @ ~/.julia/packages/CUDA/DfvRa/src/initialization.jl:52

E: looks like this is JuliaGPU/GPUArrays.jl#362 as mentioned in #1061 (comment).

[mcabbott]

Never got back to JuliaGPU/GPUArrays.jl#362 . But the rule could now use @allowscalar via GPUArraysCore instead.

[mzgubic]

DynamicPPL seems to be the same failures as just running current HEAD on julia 1.8, so this is probably good to go? TuringLang/DynamicPPL.jl#423

[devmotion]

DynamicPPL errors on the current HEAD with Julia 1.8 are fixed by updates of Tracker and Bijectors.

[ToucheSir]

Another good chunk of adjoints gone!