sum(CUDA.rand(3,3)) broken

[denizyuret]

In latest master CUDA.rand returns a ReshapedArray and sum doesn't seem to like this.

[maleadt]

Works for me.

julia> using CUDA

julia> sum(CUDA.rand(3,3))
4.0282273f0

julia> CUDA.rand(3,3)
3×3 CuArray{Float32,2}:
 0.280607  0.476438    0.71696
 0.960326  0.103435    0.878106
 0.201858  0.00404511  0.880829

julia> typeof(ans)
CuArray{Float32,2}

[denizyuret]

Do you have the latest master for both CUDA and GPUArrays? I have CUDA at 0f1496e GPUArrays at f5272f9

julia> using Pkg

julia> pkg"dev CUDA GPUArrays"
Path `/userfiles/dyuret/.julia/dev/CUDA` exists and looks like the correct package. Using existing path.
Path `/userfiles/dyuret/.julia/dev/GPUArrays` exists and looks like the correct package. Using existing path.
  Resolving package versions...
Updating `/dev/shm/dyuret/.julia/environments/v1.5/Project.toml`
  [052768ef] ~ CUDA v2.0.1 ⇒ v2.0.0 `/userfiles/dyuret/.julia/dev/CUDA`
  [0c68f7d7] ~ GPUArrays v6.0.1 ⇒ v6.0.1 `/userfiles/dyuret/.julia/dev/GPUArrays`
Updating `/dev/shm/dyuret/.julia/environments/v1.5/Manifest.toml`
  [b99e7846] - BinaryProvider v0.5.10
  [052768ef] ~ CUDA v2.0.1 ⇒ v2.0.0 `/userfiles/dyuret/.julia/dev/CUDA`
  [0c68f7d7] ~ GPUArrays v6.0.1 ⇒ v6.0.1 `/userfiles/dyuret/.julia/dev/GPUArrays`

julia> using CUDA
[ Info: Precompiling CUDA [052768ef-5323-5732-b1bb-66c8b64840ba]

julia> sum(CUDA.rand(3,3))
ERROR: MethodError: no method matching cached_compilation(::Dict{UInt64,Any}, ::typeof(CUDA.cufunction_compile), ::typeof(CUDA.cufu\
nction_link), ::GPUCompiler.FunctionSpec{typeof(CUDA.partial_mapreduce_grid),Tuple{typeof(identity),typeof(Base.add_sum),Float32,Ca\
rtesianIndices{2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}},CartesianIndices{2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}},Val{tru\
e},CuDeviceArray{Float32,3,1},CuDeviceArray{Float32,2,1}}})
Stacktrace:
 [1] cufunction(::typeof(CUDA.partial_mapreduce_grid), ::Type{Tuple{typeof(identity),typeof(Base.add_sum),Float32,CartesianIndices{\
2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}},CartesianIndices{2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}},Val{true},CuDeviceArra\
y{Float32,3,1},CuDeviceArray{Float32,2,1}}}; name::Nothing, kwargs::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tupl\
e{}}}) at /userfiles/dyuret/.julia/dev/CUDA/src/compiler/execution.jl:297
 [2] cufunction(::typeof(CUDA.partial_mapreduce_grid), ::Type{Tuple{typeof(identity),typeof(Base.add_sum),Float32,CartesianIndices{\
2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}},CartesianIndices{2,Tuple{Base.OneTo{Int64},Base.OneTo{Int64}}},Val{true},CuDeviceArra\
y{Float32,3,1},CuDeviceArray{Float32,2,1}}}) at /userfiles/dyuret/.julia/dev/CUDA/src/compiler/execution.jl:294
 [3] mapreducedim!(::typeof(identity), ::typeof(Base.add_sum), ::CuArray{Float32,2}, ::CuArray{Float32,2}; init::Float32) at /userf\
iles/dyuret/.julia/dev/CUDA/src/mapreduce.jl:196
 [4] _mapreduce(::typeof(identity), ::typeof(Base.add_sum), ::CuArray{Float32,2}; dims::Colon, init::Nothing) at /userfiles/dyuret/\
.julia/dev/GPUArrays/src/host/mapreduce.jl:62
 [5] #mapreduce#15 at /userfiles/dyuret/.julia/dev/GPUArrays/src/host/mapreduce.jl:28 [inlined]
 [6] mapreduce at /userfiles/dyuret/.julia/dev/GPUArrays/src/host/mapreduce.jl:28 [inlined]
 [7] _sum at ./reducedim.jl:727 [inlined]
 [8] _sum at ./reducedim.jl:726 [inlined]
 [9] #sum#627 at ./reducedim.jl:722 [inlined]
 [10] sum(::CuArray{Float32,2}) at ./reducedim.jl:722
 [11] top-level scope at none:1

julia> pkg"st"
Status `/dev/shm/dyuret/.julia/environments/v1.5/Project.toml`
  [6710c13c] AutoGrad v1.2.3
  [fbb218c0] BSON v0.2.6
  [6e4b80f9] BenchmarkTools v0.5.0
  [052768ef] CUDA v2.0.0 `/userfiles/dyuret/.julia/dev/CUDA`
  [864edb3b] DataStructures v0.18.7
  [5789e2e9] FileIO v1.4.3
  [0c68f7d7] GPUArrays v6.0.1 `/userfiles/dyuret/.julia/dev/GPUArrays`
  [7073ff75] IJulia v1.21.4
  [6218d12a] ImageMagick v1.1.6
  [916415d5] Images v0.23.0
  [c8e1da08] IterTools v1.3.0
  [4138dd39] JLD v0.10.0
  [033835bb] JLD2 v0.2.3
  [682c06a0] JSON v0.21.1
  [1902f260] Knet v1.4.3 `/userfiles/dyuret/.julia/dev/Knet`
  [eb30cadb] MLDatasets v0.5.2
  [0db19996] NBInclude v2.2.0
  [872c559c] NNlib v0.7.5
  [91a5bcdd] Plots v1.6.10
  [438e738f] PyCall v1.92.1
  [295af30f] Revise v3.1.4
  [276daf66] SpecialFunctions v0.10.3
  [b8865327] UnicodePlots v1.3.0
  [a5390f91] ZipFile v0.9.2

[maleadt]

Yes, I am. The error you post here now is different, and because you're not using the Manifest or otherwise correctly matching up dependencies.

[denizyuret]

Yes, I typically do: ``` pkg"dev CUDA GPUArrays Knet" pkg"resolve" pkg"up" ``` And keep working. I am not sure how to juggle the Manifest files of multiple packages, e.g. when I am developing Knet and trying to test it on the latest version of CUDA/GPUArrays.

…

On Fri, Oct 9, 2020 at 4:44 PM Tim Besard ***@***.***> wrote: Yes, I am. The error you post here now is different, and because you're not using the Manifest or otherwise correctly matching up dependencies. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#480 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAN43JVIVU2LP44OLM7ZXRTSJ4HVTANCNFSM4SJ33Q7Q> .

[maleadt]

You also need GPUCompiler#master currently.

[denizyuret]

Thanks! I will take a look at the Manifest files from now on ;)

[maleadt]

Glad it's working again :-) I try not to have to rely on Manifests too much, but some times there's a breaking change in e.g. GPUCompiler that I don't immediately want to tag, so it can be tested a bit.