Check for CUDA availability at run time.

Project.toml

@@ -5,7 +5,7 @@ version = "0.9.0"
 [deps]
 AbstractTrees = "1520ce14-60c1-5f80-bbc7-55ef81b5835c"
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
-CUDAapi = "3895d2a7-ec45-59b8-82bb-cfc6a382f9b3"
+CUDAdrv = "c5f51814-7f29-56b8-a69c-e4d8f6be1fde"
 CodecZlib = "944b1d66-785c-5afd-91f1-9de20f533193"
 Colors = "5ae59095-9a9b-59fe-a467-6f913c188581"
 CuArrays = "3a865a2d-5b23-5a0f-bc46-62713ec82fae"

src/Flux.jl

@@ -21,19 +21,9 @@ export SGD, Descent, ADAM, Momentum, Nesterov, RMSProp,
   ADAMW, RADAM, InvDecay, ExpDecay, WeightDecay
 
 
-allow_cuda() = parse(Bool, get(ENV, "FLUX_USE_CUDA", "true"))
+ENV["CUDA_INIT_SILENT"] = true
-const consider_cuda = allow_cuda()
+using CUDAdrv, CuArrays
-
+const use_cuda = Ref(false)
-using CUDAapi
-const use_cuda = consider_cuda && has_cuda()
-if use_cuda
-  try
-    using CuArrays
-  catch
-    @error "CUDA is installed, but CuArrays.jl fails to load. Please fix the issue, or load Flux with FLUX_USE_CUDA=false."
-    rethrow()
-  end
-end
 
 include("utils.jl")
 include("onehot.jl")
@@ -49,21 +39,19 @@ include("data/Data.jl")
 
 include("deprecations.jl")
 
-if use_cuda
+include("cuda/cuda.jl")
-  include("cuda/cuda.jl")
-end
 
 function __init__()
-  # check if the GPU usage conditions that are baked in the precompilation image
+  if !CUDAdrv.functional()
-  # match the current situation, and force a recompilation if not.
+    @warn "CUDA available, but CUDAdrv.jl failed to load"
-  if (allow_cuda() != consider_cuda) || (consider_cuda && has_cuda() != use_cuda)
+  elseif length(devices()) == 0
-      cachefile = if VERSION >= v"1.3-"
+    @warn "CUDA available, but no GPU detected"
-          Base.compilecache_path(Base.PkgId(Flux))
+  elseif !CuArrays.functional()
-      else
+    @warn "CUDA GPU available, but CuArrays.jl failed to load"
-          abspath(DEPOT_PATH[1], Base.cache_file_entry(Base.PkgId(Flux)))
+  elseif !CuArrays.has_cudnn()
-      end
+    @warn "CUDA GPU available, but CuArrays.jl did not find libcudnn"
-      rm(cachefile)
+  else
-      error("Your set-up changed, and Flux.jl needs to be reconfigured. Please load the package again.")
+    use_cuda[] = true
   end
 end
 

src/cuda/cuda.jl

@@ -2,12 +2,8 @@ module CUDA
 
 using ..CuArrays
 
-if CuArrays.libcudnn !== nothing  # TODO: use CuArrays.has_cudnn()
+using CuArrays: CUDNN
-  using CuArrays: CUDNN
+include("curnn.jl")
-  include("curnn.jl")
+include("cudnn.jl")
-  include("cudnn.jl")
-else
-  @warn "CUDNN is not installed, some functionality will not be available."
-end
 
 end

src/functor.jl

@@ -73,13 +73,7 @@ end
 
 cpu(m) = fmap(x -> adapt(Array, x), m)
 
-const gpu_adaptor = if use_cuda
+gpu(x) = use_cuda[] ? fmap(CuArrays.cu, x) : x
-  CuArrays.cu
-else
-  identity
-end
-
-gpu(x) = fmap(gpu_adaptor, x)
 
 # Precision
 

src/onehot.jl

@@ -37,12 +37,10 @@ import Adapt: adapt, adapt_structure
 
 adapt_structure(T, xs::OneHotMatrix) = OneHotMatrix(xs.height, adapt(T, xs.data))
 
-if use_cuda
+import .CuArrays: CuArray, cudaconvert
-  import .CuArrays: CuArray, cudaconvert
+import Base.Broadcast: BroadcastStyle, ArrayStyle
-  import Base.Broadcast: BroadcastStyle, ArrayStyle
+BroadcastStyle(::Type{<:OneHotMatrix{<:CuArray}}) = ArrayStyle{CuArray}()
-  BroadcastStyle(::Type{<:OneHotMatrix{<:CuArray}}) = ArrayStyle{CuArray}()
+cudaconvert(x::OneHotMatrix{<:CuArray}) = OneHotMatrix(x.height, cudaconvert(x.data))
-  cudaconvert(x::OneHotMatrix{<:CuArray}) = OneHotMatrix(x.height, cudaconvert(x.data))
-end
 
 """
     onehot(l, labels[, unk])

test/cuda/cuda.jl

@@ -53,8 +53,6 @@ end
   @test y[3,:] isa CuArray
 end
 
-if CuArrays.libcudnn != nothing
+@info "Testing Flux/CUDNN"
-  @info "Testing Flux/CUDNN"
+include("cudnn.jl")
-  include("cudnn.jl")
+include("curnn.jl")
-  include("curnn.jl")
-end