Assigning a number to a CuArray slice via broadcasting fails

[ali-ramadhan]

It seems that assigning a number to a CuArray slice via broadcast fails now. It must have worked in the past (probably in v1.0.2) one of my tests just broke but maybe this is the new intended behavior.

I'm using CuArrays v1.1.0, CUDAnative v2.2.1, and CUDAdrv v3.1.0.

Minimal working example:

julia> C = rand(2, 2, 2) |> CuArray
2×2×2 CuArray{Float64,3}:
[:, :, 1] =
 0.0908594  0.995228
 0.27418    0.945454
​
[:, :, 2] =
 0.4299  0.600029
 0.8766  0.601922
​
julia> C[:, 2, :] .= 100
ERROR: MethodError: no method matching CuArray(::Int64)
Closest candidates are:
  CuArray(::CuArrays.CUSOLVER.CuQRPackedQ) at /home/gridsan/aramadhan/.julia/packages/CuArrays/eFBar/src/solver/highlevel.jl:22
  CuArray(::AbstractArray{T,N}) where {T, N} at /home/gridsan/aramadhan/.julia/packages/CuArrays/eFBar/src/array.jl:116
  CuArray(::LinearAlgebra.UniformScaling, ::Tuple{Int64,Int64}) at /home/gridsan/aramadhan/.julia/packages/GPUArrays/joKVJ/src/construction.jl:30
  ...
Stacktrace:
 [1] map(::Type, ::Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}}) at ./tuple.jl:167
 [2] adapt_structure(::CUDAnative.Adaptor, ::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}) at /home/gridsan/aramadhan/.julia/packages/CuArrays/eFBar/src/subarray.jl:61
 [3] adapt(::CUDAnative.Adaptor, ::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}) at /home/gridsan/aramadhan/.julia/packages/Adapt/aeQPS/src/Adapt.jl:6
 [4] cudaconvert(::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}) at /home/gridsan/aramadhan/.julia/packages/CUDAnative/nItlk/src/execution.jl:209
 [5] _broadcast_getindex_evalf at ./broadcast.jl:578 [inlined]
 [6] _broadcast_getindex at ./broadcast.jl:551 [inlined]
 [7] (::getfield(Base.Broadcast, Symbol("##19#20")){Base.Broadcast.Broadcasted{Base.Broadcast.Style{Tuple},Nothing,typeof(CUDAnative.cudaconvert),Tuple{Tuple{CuArrays.CuKernelState,SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}},typeof(identity),Tuple{Int64}}}}}})(::Int64) at ./broadcast.jl:953
 [8] ntuple at ./tuple.jl:161 [inlined]
 [9] copy at ./broadcast.jl:953 [inlined]
 [10] materialize at ./broadcast.jl:753 [inlined]
 [11] macro expansion at /home/gridsan/aramadhan/.julia/packages/CUDAnative/nItlk/src/execution.jl:172 [inlined]
 [12] macro expansion at ./gcutils.jl:87 [inlined]
 [13] macro expansion at /home/gridsan/aramadhan/.julia/packages/CUDAnative/nItlk/src/execution.jl:171 [inlined]
 [14] _gpu_call(::CuArrays.CuArrayBackend, ::Function, ::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}, ::Tuple{SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}},typeof(identity),Tuple{Int64}}}, ::Tuple{Tuple{Int64},Tuple{Int64}}) at /home/gridsan/aramadhan/.julia/packages/CuArrays/eFBar/src/gpuarray_interface.jl:60
 [15] gpu_call(::Function, ::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}, ::Tuple{SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}},typeof(identity),Tuple{Int64}}}, ::Int64) at /home/gridsan/aramadhan/.julia/packages/GPUArrays/joKVJ/src/abstract_gpu_interface.jl:151
 [16] gpu_call(::Function, ::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}, ::Tuple{SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false},Base.Broadcast.Broadcasted{Nothing,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}},typeof(identity),Tuple{Int64}}}) at /home/gridsan/aramadhan/.julia/packages/GPUArrays/joKVJ/src/abstract_gpu_interface.jl:128
 [17] copyto! at /home/gridsan/aramadhan/.julia/packages/GPUArrays/joKVJ/src/broadcast.jl:48 [inlined]
 [18] copyto! at /home/gridsan/aramadhan/.julia/packages/GPUArrays/joKVJ/src/broadcast.jl:60 [inlined]
 [19] materialize!(::SubArray{Float64,2,CuArray{Float64,3},Tuple{Base.Slice{Base.OneTo{Int64}},Int64,Base.Slice{Base.OneTo{Int64}}},false}, ::Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{0},Nothing,typeof(identity),Tuple{Int64}}) at ./broadcast.jl:756
 [20] top-level scope at none:0

[maleadt]

Seems to work here:

julia> C = rand(2, 2, 2) |> CuArray
2×2×2 CuArray{Float64,3}:
[:, :, 1] =
 0.142639  0.148232
 0.628564  0.232154

[:, :, 2] =
 0.441277  0.874009 
 0.899314  0.0589816

julia> C[:, 2, :] .= 100
2×2 view(::CuArray{Float64,3}, :, 2, :) with eltype Float64:
 100.0  100.0
 100.0  100.0

something that got fixed on master maybe? I'll be tagging a release soon.

[maleadt]

Yeah that's #382

[ali-ramadhan]

Ah awesome, thank you! I should have tried master before opening an issue.