Error using CuArrays in Colab

New to Julia
1

I get the following error while trying to use CuArrays on GoogleColab:

pkg"add CuArrays; precompile;"
using CuArrays

mgpu = cu(mcpu)
@benchmark mgpu*mgpu

The error message is:

Resolving package versions...
 Installed Requires ─────────── v0.5.2
 Installed MacroTools ───────── v0.5.2
 Installed AbstractFFTs ─────── v0.4.1
 Installed GPUArrays ────────── v2.0.0
 Installed Adapt ────────────── v1.0.0
 Installed CUDAapi ──────────── v2.0.0
 Installed DataStructures ───── v0.17.5
 Installed CuArrays ─────────── v1.4.5
 Installed CEnum ────────────── v0.2.0
 Installed TimerOutputs ─────── v0.5.2
 Installed NNlib ────────────── v0.6.0
 Installed OrderedCollections ─ v1.1.0
 Installed CUDAnative ───────── v2.5.5
 Installed CUDAdrv ──────────── v4.0.4
 Installed LLVM ─────────────── v1.3.2
  Updating `~/.julia/environments/v1.0/Project.toml`
  [3a865a2d] + CuArrays v1.4.5
  Updating `~/.julia/environments/v1.0/Manifest.toml`
  [621f4979] + AbstractFFTs v0.4.1
  [79e6a3ab] + Adapt v1.0.0
  [fa961155] + CEnum v0.2.0
  [3895d2a7] + CUDAapi v2.0.0
  [c5f51814] + CUDAdrv v4.0.4
  [be33ccc6] + CUDAnative v2.5.5
  [3a865a2d] + CuArrays v1.4.5
  [864edb3b] + DataStructures v0.17.5
  [0c68f7d7] + GPUArrays v2.0.0
  [929cbde3] + LLVM v1.3.2
  [1914dd2f] + MacroTools v0.5.2
  [872c559c] + NNlib v0.6.0
  [bac558e1] + OrderedCollections v1.1.0
  [ae029012] + Requires v0.5.2
  [a759f4b9] + TimerOutputs v0.5.2
Precompiling project...
Precompiling CuArrays
┌ Info: Precompiling CuArrays [3a865a2d-5b23-5a0f-bc46-62713ec82fae]
└ @ Base loading.jl:1192
┌ Info: CuArrays.jl failed to initialize and will be unavailable (set JULIA_CUDA_SILENT or JULIA_CUDA_VERBOSE to silence or expand this message)
└ @ CuArrays /root/.julia/packages/CuArrays/Dzwot/src/CuArrays.jl:144
SYSTEM: show(lasterr) caused an error

Stacktrace:
 [1] macro expansion at /root/.julia/packages/CUDAdrv/3EzC1/src/error.jl:123 [inlined]
 [2] cuDeviceGet(::Base.RefValue{Int32}, ::Int64) at /root/.julia/packages/CUDAdrv/3EzC1/src/libcuda.jl:30
 [3] Type at /root/.julia/packages/CUDAdrv/3EzC1/src/devices.jl:25 [inlined]
 [4] initialize at /root/.julia/packages/CUDAnative/2WQzk/src/init.jl:40 [inlined]
 [5] maybe_initialize(::Symbol) at /root/.julia/packages/CUDAnative/2WQzk/src/init.jl:33
 [6] macro expansion at /root/.julia/packages/CUDAdrv/3EzC1/src/error.jl:119 [inlined]
 [7] cuMemAlloc_v2(::Base.RefValue{CUDAdrv.CuPtr{Nothing}}, ::Int64) at /root/.julia/packages/CUDAdrv/3EzC1/src/libcuda.jl:312
 [8] alloc at /root/.julia/packages/CUDAdrv/3EzC1/src/memory.jl:64 [inlined]
 [9] macro expansion at /root/.julia/packages/TimerOutputs/Tf7lx/src/TimerOutput.jl:214 [inlined]
 [10] macro expansion at /root/.julia/packages/CuArrays/Dzwot/src/memory.jl:55 [inlined]
 [11] macro expansion at ./util.jl:213 [inlined]
 [12] actual_alloc(::Int64) at /root/.julia/packages/CuArrays/Dzwot/src/memory.jl:54
 [13] actual_alloc at /root/.julia/packages/CuArrays/Dzwot/src/memory/binned.jl:53 [inlined]
 [14] macro expansion at /root/.julia/packages/CuArrays/Dzwot/src/memory/binned.jl:194 [inlined]
 [15] macro expansion at /root/.julia/packages/TimerOutputs/Tf7lx/src/TimerOutput.jl:214 [inlined]
 [16] pool_alloc(::Int64, ::Int64) at /root/.julia/packages/CuArrays/Dzwot/src/memory/binned.jl:193
 [17] (::getfield(CuArrays.BinnedPool, Symbol("##9#10")){Int64,Int64,Set{CuArrays.BinnedPool.Block},Array{CuArrays.BinnedPool.Block,1}})() at /root/.julia/packages/CuArrays/Dzwot/src/memory/binned.jl:291
 [18] lock(::getfield(CuArrays.BinnedPool, Symbol("##9#10")){Int64,Int64,Set{CuArrays.BinnedPool.Block},Array{CuArrays.BinnedPool.Block,1}}, ::ReentrantLock) at ./lock.jl:101
 [19] alloc(::Int64) at /root/.julia/packages/CuArrays/Dzwot/src/memory/binned.jl:290
 [20] macro expansion at /root/.julia/packages/TimerOutputs/Tf7lx/src/TimerOutput.jl:214 [inlined]
 [21] macro expansion at /root/.julia/packages/CuArrays/Dzwot/src/memory.jl:121 [inlined]
 [22] macro expansion at ./util.jl:213 [inlined]
 [23] alloc at /root/.julia/packages/CuArrays/Dzwot/src/memory.jl:120 [inlined]
 [24] CuArray{Float32,2,P} where P(::UndefInitializer, ::Tuple{Int64,Int64}) at /root/.julia/packages/CuArrays/Dzwot/src/array.jl:90
 [25] convert at /root/.julia/packages/CuArrays/Dzwot/src/array.jl:98 [inlined]
 [26] adapt_storage at /root/.julia/packages/CuArrays/Dzwot/src/array.jl:240 [inlined]
 [27] adapt_structure at /root/.julia/packages/Adapt/aeQPS/src/Adapt.jl:9 [inlined]
 [28] adapt at /root/.julia/packages/Adapt/aeQPS/src/Adapt.jl:6 [inlined]
 [29] cu(::Array{Float64,2}) at /root/.julia/packages/CuArrays/Dzwot/src/array.jl:288
 [30] top-level scope at In[2]:3rd_place_medal: ```

How do I fix this?
2

@maleadt Can you please comment on how to fix this?

3

Well, the error gives you a suggestion to get more information (set the env var before importing), so please do that first. There’s also been a new version that includes a fix for initialization on Colab, so be sure to update packages and and try again.

4

I tried with the latest release and it seems to work fine. Well, I come asking for I am a beginner. Sorry if some of my posts are redundant. I find it difficult to seek help with anything related to GPU programming in Julia. There is very little documentation and very little help.

5

What do you mean by that? We do our best to support all users and their questions, but you posted this outside of the GPU domain so I didn’t get a notification. Documentation is lacking, sure, but we’re working on that: https://juliagpu.gitlab.io/CUDA.jl/
For quick questions, feel free to hop on the gpu Slack channel and ask there.

1 Like
6
  1. To begin with I use the tutorial of CuArrays. There, it is pretty straightforward that one should add and use the packages. There are no additional variables’ settings specified. These pop-up on trying to run the code on colab.
  2. For the longest time, I was clueless about running Julia on Colab. For, the procedure mentioned here https://discourse.julialang.org/t/julia-on-google-colab-free-gpu-accelerated-shareable-notebooks/15319?u=ennvvy was broken. I had to wait for a couple of weeks to be able to fix this issue.

I am not finding flaws here with the community. I am only reporting the problems that I have faced, hoping the feedback leads to better user experience eventually.