Add workaround for CUDA.jl issue

Add CUDA package as a dependency so that we can engage in a little type
piracy to workaround CUDA.jl issue #1559.  Once CUDA.jl has resolved the
issue, we can drop CUDA as a dependency and remove the type piracy.  For
more details see:

  JuliaGPU/CUDA.jl#1559

Project.toml

@@ -5,6 +5,7 @@ version = "0.1.0"
 
 [deps]
 AbstractFFTs = "621f4979-c628-5d54-868e-fcf4e3e8185c"
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 FFTW = "7a1cc6ca-52ef-59f5-83cd-3a7055c09341"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 

src/fftfdr.jl

@@ -1,10 +1,16 @@
 using AbstractFFTs
+using LinearAlgebra
+
 # We need to depend on FFTW for access to its flags constants because
 # AbstractFFTs defines the planning flags to be "a bitwise-or of FFTW planner
 # flags".  For more details, see:
 # https://github.com/JuliaMath/AbstractFFTs.jl/issues/71
 import FFTW
-using LinearAlgebra
+
+# Type piracy to workaround CUDA.jl issue #1559.  For more details, see:
+# https://github.com/JuliaGPU/CUDA.jl/issues/1559
+import CUDA: CuArray
+AbstractFFTs.plan_brfft(A::CuArray, d::Integer, region; kwargs...) = plan_brfft(A, d, region)
 
 function phasor(k::Integer, n::Integer, r::Float32, N::Integer)
     cispi(-2*k*n*r/N)