Merge pull request #54 from SciML/cuda

allow CUDA

Author: ChrisRackauckas

src/ArrayInterface.jl

@@ -551,6 +551,16 @@ function __init__()
     end
   end
 
+  @require CUDA="3a865a2d-5b23-5a0f-bc46-62713ec82fae" begin
+    @require Adapt="79e6a3ab-5dfb-504d-930d-738a2a938a0e" begin
+      include("cuarrays2.jl")
+    end
+    @require DiffEqBase="2b5f629d-d688-5b77-993f-72d75c75574e" begin
+      # actually do QR
+      lu_instance(A::CUDA.CuMatrix{T}) where T = CUDA.CUSOLVER.CuQR(similar(A, 0, 0), similar(A, 0))
+    end
+  end
+
   @require BandedMatrices="aae01518-5342-5314-be14-df237901396f" begin
     function findstructralnz(x::BandedMatrices.BandedMatrix)
       l,u=BandedMatrices.bandwidths(x)

src/cuarrays2.jl

@@ -0,0 +1,13 @@
+fast_scalar_indexing(::Type{<:CUDA.CuArray}) = false
+@inline allowed_getindex(x::CUDA.CuArray,i...) = CUDA.@allowscalar(x[i...])
+@inline allowed_setindex!(x::CUDA.CuArray,v,i...) = (CUDA.@allowscalar(x[i...] = v))
+
+function Base.setindex(x::CUDA.CuArray,v,i::Int)
+  _x = copy(x)
+  allowed_setindex!(_x,v,i)
+  _x
+end
+
+function restructure(x::CUDA.CuArray,y)
+  reshape(Adapt.adapt(parameterless_type(x),y),size(x)...)
+end