Merge branch 'master' of ssh://github.com/jpsamaroo/DaggerGPU.jl

Author: jpsamaroo

.github/workflows/TagBot.yml

@@ -0,0 +1,11 @@
+name: TagBot
+on:
+  schedule:
+    - cron: 0 * * * *
+jobs:
+  TagBot:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: JuliaRegistries/TagBot@v1
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}

Manifest.toml

@@ -5,7 +5,7 @@ uuid = "2a0f44e3-6c83-55bd-87e4-b1978d98bd5f"
 
 [[Dagger]]
 deps = ["Distributed", "LinearAlgebra", "MemPool", "Profile", "Random", "Serialization", "SharedArrays", "SparseArrays", "Statistics", "StatsBase"]
-git-tree-sha1 = "e77f451e4c1f9acbf794cb6377ec42130ff10f56"
+git-tree-sha1 = "8262f275c3acf3787fc70d1bd99aec011a02ddb8"
 repo-rev = "jps/compute-resource"
 repo-url = "https://github.com/JuliaParallel/Dagger.jl.git"
 uuid = "d58978e5-989f-55fb-8d15-ea34adc7bf54"

Project.toml

@@ -8,7 +8,9 @@ Dagger = "d58978e5-989f-55fb-8d15-ea34adc7bf54"
 Requires = "ae029012-a4dd-5104-9daa-d747884805df"
 
 [compat]
-julia = "1.0"
+Dagger = "0.8"
+Requires = "1.0"
+julia = "1"
 
 [extras]
 CuArrays = "3a865a2d-5b23-5a0f-bc46-62713ec82fae"

src/DaggerGPU.jl

@@ -17,6 +17,8 @@ end
 
 processor(kind::Symbol) = processor(Val(kind))
 processor(::Val) = Dagger.ThreadProc
+cancompute(kind::Symbol) = cancompute(Val(kind))
+cancompute(::Val) = false
 
 function __init__()
     @require CuArrays="3a865a2d-5b23-5a0f-bc46-62713ec82fae" begin

src/cuarrays.jl

@@ -7,10 +7,11 @@ end
 
 @gpuproc(CuArrayProc, CuArray)
 
+processor(::Val{:CUDA}) = CuArrayProc
+cancompute(::Val{:CUDA}) = CUDAapi.has_cuda()
 
 push!(Dagger.PROCESSOR_CALLBACKS, proc -> begin
     if CUDAapi.has_cuda()
-        @eval processor(::Val{:CUDA}) = CuArrayProc
-        return CuArrayProc(first(devices()))
+        return CuArrayProc(first(CUDAdrv.devices()))
     end
 end)

src/rocarrays.jl

@@ -7,10 +7,11 @@ end
 
 @gpuproc(ROCArrayProc, ROCArray)
 
+processor(::Val{:ROC}) = ROCArrayProc
+cancompute(::Val{:ROC}) = ROCArrays.configured
 
 push!(Dagger.PROCESSOR_CALLBACKS, proc -> begin
     if ROCArrays.configured
-        @eval processor(::Val{:ROC}) = ROCArrayProc
         return ROCArrayProc(HSARuntime.get_default_agent())
     end
 end)

test/runtests.jl

@@ -12,8 +12,14 @@ end
 cuproc = DaggerGPU.processor(:CUDA)
 rocproc = DaggerGPU.processor(:ROC)
 
-cuproc === Dagger.ThreadProc && @warn "No CUDA devices available"
-rocproc === Dagger.ThreadProc && @warn "No ROCm devices available"
+if !DaggerGPU.cancompute(:CUDA)
+    @warn "No CUDA devices available, falling back to ThreadProc"
+    cuproc = Dagger.ThreadProc
+end
+if !DaggerGPU.cancompute(:ROC)
+    @warn "No ROCm devices available, falling back to ThreadProc"
+    rocproc = Dagger.ThreadProc
+end
 
 as = [delayed(x->x+1)(1) for i in 1:10]
 b = delayed((xs...)->[sum(xs)])(as...)