Skip to content
This repository has been archived by the owner on Mar 12, 2021. It is now read-only.

Missing constructor CuArray{Float32,4}(::UndefInitializer, ::Tuple{Int64,Int64,Int64}) #408

Closed
baggepinnen opened this issue Sep 3, 2019 · 3 comments
Closed

Missing constructor CuArray{Float32,4}(::UndefInitializer, ::Tuple{Int64,Int64,Int64}) #408

baggepinnen opened this issue Sep 3, 2019 · 3 comments
Labels
bug

Comments

@baggepinnen
Copy link
Contributor

The following throws an error

using Flux, CuArrays
julia> Z = gpu(Conv((3,3), 1=>1))(gpu(randn(5,5,1,1)))
Tracked 3×3×1×1 CuArray{Float32,4}:
[:, :, 1, 1] =
  1.85926   -4.03764  -1.79426
  0.348474  -3.98205  -1.1394 
 -0.766892  -1.29553  -1.96865

julia> Z = mapslices(Z, dims=(1,2,4)) do Zs
           Zs
       end
ERROR: MethodError: no method matching CuArray{Float32,4}(::UndefInitializer, ::Tuple{Int64,Int64,Int64})
Closest candidates are:
  CuArray{Float32,4}(::UndefInitializer, ::Tuple{Vararg{Int64,N}}) where {T, N} at /home/fredrikb/.julia/packages/CuArrays/wXQp8/src/array.jl:38
  CuArray{Float32,4}(::CUDAdrv.Mem.Buffer, ::Tuple{Vararg{Int64,N}}; offset, own) where {T, N} at /home/fredrikb/.julia/packages/CuArrays/wXQp8/src/array.jl:11
  CuArray{Float32,4}(::UndefInitializer, ::Integer...) where {T, N} at /home/fredrikb/.julia/packages/CuArrays/wXQp8/src/array.jl:42
  ...
Stacktrace:
 [1] similar(::Type{CuArray{Float32,4}}, ::Tuple{Int64,Int64,Int64}) at ./abstractarray.jl:672
 [2] convert(::Type{CuArray{Float32,4}}, ::Array{Float32,3}) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/construction.jl:80
 [3] convert(::Type{CuArray{Float32,4}}, ::TrackedArray{…,CuArray{Float32,3}}) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/construction.jl:71
 [4] gpu_convert at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/indexing.jl:119 [inlined]
 [5] _unsafe_setindex!(::IndexLinear, ::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Vararg{Union{Real, AbstractArray},N} where N) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/indexing.jl:131
 [6] _setindex!(::IndexLinear, ::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Base.OneTo{Int64}, ::Vararg{Union{Real, AbstractArray},N} where N) at ./multidimensional.jl:712
 [7] setindex!(::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Base.OneTo{Int64}, ::Int64, ::Vararg{Any,N} where N) at ./abstractarray.jl:1074
 [8] concatenate_setindex!(::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Vararg{Any,N} where N) at ./abstractarray.jl:2060
 [9] #mapslices#109(::Tuple{Int64,Int64,Int64}, ::typeof(mapslices), ::getfield(Main, Symbol("##21#22")), ::TrackedArray{…,CuArray{Float32,4}}) at ./abstractarray.jl:2026
 [10] (::getfield(Base, Symbol("#kw##mapslices")))(::NamedTuple{(:dims,),Tuple{Tuple{Int64,Int64,Int64}}}, ::typeof(mapslices), ::Function, ::TrackedArray{…,CuArray{Float32,4}}) at ./none:0
 [11] top-level scope at none:0

Naively defining the suggested method does not help

julia> CuArray{T,N}(::UndefInitializer, dims::Tuple) where {T,N} =
         CuArray{T,N}(alloc(prod(dims)*sizeof(T)), dims...)

julia> Z = mapslices(Z, dims=(1,2,4)) do Zs
           Zs
       end
ERROR: MethodError: no method matching CuArray{Float32,4}(::CUDAdrv.Mem.DeviceBuffer, ::Int64, ::Int64, ::Int64)
Closest candidates are:
  CuArray{Float32,4}(::UndefInitializer, ::Integer...) where {T, N} at /home/fredrikb/.julia/packages/CuArrays/wXQp8/src/array.jl:42
  CuArray{Float32,4}(::Function, ::Any...) where {T, N} at /home/fredrikb/.julia/packages/CuArrays/wXQp8/src/array.jl:56
  CuArray{Float32,4}(::UniformScaling, ::Integer, ::Integer) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/construction.jl:34
  ...
Stacktrace:
 [1] CuArray{Float32,4}(::UndefInitializer, ::Tuple{Int64,Int64,Int64}) at /home/fredrikb/arl/megafauna/explore.jl:248
 [2] similar(::Type{CuArray{Float32,4}}, ::Tuple{Int64,Int64,Int64}) at ./abstractarray.jl:672
 [3] convert(::Type{CuArray{Float32,4}}, ::Array{Float32,3}) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/construction.jl:80
 [4] convert(::Type{CuArray{Float32,4}}, ::TrackedArray{…,CuArray{Float32,3}}) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/construction.jl:71
 [5] gpu_convert at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/indexing.jl:119 [inlined]
 [6] _unsafe_setindex!(::IndexLinear, ::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Vararg{Union{Real, AbstractArray},N} where N) at /home/fredrikb/.julia/packages/GPUArrays/J4c3Q/src/indexing.jl:131
 [7] _setindex!(::IndexLinear, ::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Base.OneTo{Int64}, ::Vararg{Union{Real, AbstractArray},N} where N) at ./multidimensional.jl:712
 [8] setindex!(::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Base.OneTo{Int64}, ::Int64, ::Vararg{Any,N} where N) at ./abstractarray.jl:1074
 [9] concatenate_setindex!(::CuArray{Float32,4}, ::TrackedArray{…,CuArray{Float32,3}}, ::Base.OneTo{Int64}, ::Vararg{Any,N} where N) at ./abstractarray.jl:2060
 [10] #mapslices#109(::Tuple{Int64,Int64,Int64}, ::typeof(mapslices), ::getfield(Main, Symbol("##31#32")), ::TrackedArray{…,CuArray{Float32,4}}) at ./abstractarray.jl:2026
 [11] (::getfield(Base, Symbol("#kw##mapslices")))(::NamedTuple{(:dims,),Tuple{Tuple{Int64,Int64,Int64}}}, ::typeof(mapslices), ::Function, ::TrackedArray{…,CuArray{Float32,4}}) at ./none:0
 [12] top-level scope at none:0

Details on Julia:

julia> versioninfo()
Julia Version 1.3.0-alpha.0
Commit 6c11e7c2c4 (2019-07-23 01:46 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-6.0.1 (ORCJIT, skylake)
Environment:
  JULIA_NUM_THREADS = 6

(v1.3) pkg> st
    Status `~/.julia/environments/v1.3/Project.toml`
  [c52e3926] Atom v0.9.1
  [70b36510] AutomaticDocstrings v0.1.1 [`~/.julia/dev/AutomaticDocstrings`]
  [aae01518] BandedMatrices v0.10.1
  [07f847a3] BasisFunctionExpansions v1.0.1
  [6e4b80f9] BenchmarkTools v0.4.2
  [5d742f6a] CSVFiles v0.15.0
  [3895d2a7] CUDAapi v1.1.0
  [8f4d0f93] Conda v1.3.0
  [3abffc1c] ControlSystemIdentification v0.1.5 [`~/.julia/dev/ControlSystemIdentification`]
  [a6e380b2] ControlSystems v0.5.3 [`~/.julia/dev/ControlSystems`]
  [3a865a2d] CuArrays v1.2.1
  [717857b8] DSP v0.6.0
  [a93c6f00] DataFrames v0.19.2
  [864edb3b] DataStructures v0.17.0
  [9cd584b6] DifferentialDynamicProgramming v0.4.0 [`~/.julia/dev/DifferentialDynamicProgramming`]
  [09c11fbc] DiskDataProviders v0.1.0 [`~/.julia/dev/DiskDataProviders`]
  [7a1cc6ca] FFTW v0.3.0
  [5789e2e9] FileIO v1.0.7
  [587475ba] Flux v0.9.0 [`~/.julia/dev/Flux`]
  [f6369f11] ForwardDiff v0.10.3
  [28b8d3ca] GR v0.41.0
  [f67ccb44] HDF5 v0.12.2
  [cd3eb016] HTTP v0.8.5
  [7073ff75] IJulia v1.19.0
  [6a3955dd] ImageFiltering v0.6.5
  [6218d12a] ImageMagick v0.7.5
  [916415d5] Images v0.18.0
  [d0351b0e] InspectDR v0.3.6
  [e5e0dc1b] Juno v0.7.2
  [1902f260] Knet v1.2.4
  [26dcc766] LPVSpectral v0.1.2
  [82ba68a1] LazyWAVFiles v0.1.0 [`~/.julia/dev/LazyWAVFiles`]
  [98b081ad] Literate v2.0.3
  [43172d54] LiterateWeave v0.1.0 [`~/.julia/dev/LiterateWeave`]
  [33e6dc65] MKL v0.0.0 #master (https://github.com/JuliaComputing/MKL.jl)
  [f0e99cf1] MLBase v0.8.0
  [cc2ba9b6] MLDataUtils v0.5.0
  [0987c9cc] MonteCarloMeasurements v0.4.0 [`~/.julia/dev/MonteCarloMeasurements`]
  [6f286f6a] MultivariateStats v0.6.0
  [872c559c] NNlib v0.6.0
  [1dea7af3] OrdinaryDiffEq v5.14.0
  [58dd65bb] Plotly v0.2.0
  [f0f68f2c] PlotlyJS v0.12.5
  [91a5bcdd] Plots v0.26.2
  [438e738f] PyCall v1.91.2
  [d330b81b] PyPlot v2.8.1
  [a75dfa10] Robotlib v0.3.2+ [`~/.julia/dev/Robotlib`]
  [90137ffa] StaticArrays v0.11.0
  [2913bbd2] StatsBase v0.30.0
  [f3b207a7] StatsPlots v0.10.2
  [bd369af6] Tables v0.2.11
  [5e47fb64] TestImages v0.5.1
  [b8865327] UnicodePlots v1.1.0
  [8149f6b0] WAV v1.0.2
  [44d3d7a6] Weave v0.9.1
  [cd998857] Yota v0.1.3+ [`~/.julia/dev/Yota`]
  [e88e6eb3] Zygote v0.3.4 [`~/.julia/dev/Zygote`]
  [700de1a5] ZygoteRules v0.1.0 [`~/.julia/dev/ZygoteRules`]
@baggepinnen baggepinnen added the bug label Sep 3, 2019
@maleadt
Copy link
Member

maleadt commented Sep 3, 2019

There is a dimension mismatch in these calls (4-dim CuArray, 3-dim indices), and these calls do not work with Array either:

julia> Array{Float32,4}(undef, (1,2,3))
ERROR: MethodError: no method matching Array{Float32,4}(::UndefInitializer, ::Tuple{Int64,Int64,Int64})

julia> similar(Array{Float32,4}, (1,2,3))
ERROR: MethodError: no method matching Array{Float32,4}(::UndefInitializer, ::Tuple{Int64,Int64,Int64})

julia> convert(Array{Float32,4}, Array{Float32,3}(undef, (1,2,3)))
ERROR: MethodError: no method matching Array{Float32,4}(::Array{Float32,3})

So not a missing constructor or method. The call to mapslices also works with pure CuArray:

julia> Z = CuArray{Float32,4}(undef, (5,5,1,1))
5×5×1×1 CuArray{Float32,4}:
[:, :, 1, 1] =
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0

julia> Z = mapslices(Z, dims=(1,2,4)) do Zs
                  Zs
              end
5×5×1×1 CuArray{Float32,4}:
[:, :, 1, 1] =
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0
 0.0  0.0  0.0  0.0  0.0

So this looks like an issue with TrackedArray instead?

@baggepinnen
Copy link
Contributor Author

Ah sorry, my bad. I'll open an issue on tracker instead!

@baggepinnen
Copy link
Contributor Author

Ref: FluxML/Tracker.jl#45

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Assignees
No one assigned
Labels
bug
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@maleadt @baggepinnen