Switch to CUDACore (#57)

* Switch to CUDACore

* Bump StridedViews version whoops

* Use cuRAND for tests

Author: kshyatt

Project.toml

@@ -1,6 +1,6 @@
 name = "Strided"
 uuid = "5e0ebb24-38b0-5f93-81fe-25c709ecae67"
-version = "2.3.6"
+version = "2.4.0"
 authors = ["Lukas Devos <lukas.devos@ugent.be>", "Maarten Van Damme <maartenvd1994@gmail.com>", "Jutho Haegeman <jutho.haegeman@ugent.be>"]
 
 [deps]
@@ -12,35 +12,37 @@ TupleTools = "9d95972d-f1c8-5527-a6e0-b4b365fa01f6"
 AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
 GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
-CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 
 [extensions]
 StridedAMDGPUExt = "AMDGPU"
 StridedJLArraysExt = "JLArrays"
 StridedGPUArraysExt = "GPUArrays"
-StridedCUDAExt = "CUDA"
+StridedCUDACoreExt = "CUDACore"
 
 [compat]
 AMDGPU = "2"
 Aqua = "0.8"
-CUDA = "5"
+CUDACore = "6"
+cuRAND = "6"
 JLArrays = "0.3.1"
 GPUArrays = "11.4.1"
 LinearAlgebra = "1.6"
 Random = "1.6"
-StridedViews = "0.4.6"
+StridedViews = "0.5"
 Test = "1.6"
 TupleTools = "1.6"
 julia = "1.6"
 
 [extras]
 AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
-CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
+CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
+cuRAND = "20fd9a0b-12d5-4c2f-a8af-7c34e9e60431"
 JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
 GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [targets]
-test = ["Test", "Random", "Aqua", "AMDGPU", "CUDA", "GPUArrays", "JLArrays"]
+test = ["Test", "Random", "Aqua", "AMDGPU", "CUDACore", "cuRAND", "GPUArrays", "JLArrays"]

ext/StridedCUDAExt.jl ext/StridedCUDACoreExt.jlext/StridedCUDAExt.jl renamed to ext/StridedCUDACoreExt.jl

@@ -1,8 +1,8 @@
-module StridedCUDAExt
+module StridedCUDACoreExt
 
-using Strided, StridedViews, CUDA
-using CUDA: Adapt, KernelAdaptor
-using CUDA: GPUArrays
+using Strided, StridedViews, CUDACore
+using CUDACore: Adapt, KernelAdaptor
+using CUDACore: GPUArrays
 
 const ALL_FS = Union{typeof(adjoint), typeof(conj), typeof(identity), typeof(transpose)}
 

test/cuda.jl

@@ -1,7 +1,7 @@
 for T in (Float32, Float64, Complex{Float32}, Complex{Float64})
     @testset "Copy with CuStridedView: $T, $f1, $f2" for f2 in (identity, conj, adjoint, transpose), f1 in (identity, conj, transpose, adjoint)
         for m1 in (0, 16, 32), m2 in (0, 16, 32)
-            A1 = CUDA.randn(T, (m1, m2))
+            A1 = cuRAND.randn(T, (m1, m2))
             A2 = similar(A1)
             zA1 = CuMatrix(f1(zeros(T, (m1, m2))))
             zA2 = CuMatrix(f2(zeros(T, (m1, m2))))
@@ -10,16 +10,16 @@ for T in (Float32, Float64, Complex{Float32}, Complex{Float64})
             B1 = f1(StridedView(A1c))
             B2 = f2(StridedView(A2c))
             axes(f1(A1)) == axes(f2(A2)) || continue
-            @test collect(CuMatrix(copy!(f2(A2), f1(A1)))) == CUDA.Adapt.adapt(Vector{T}, copy!(B2, B1))
+            @test collect(CuMatrix(copy!(f2(A2), f1(A1)))) == CUDACore.Adapt.adapt(Vector{T}, copy!(B2, B1))
             @test copy!(zA1, f1(A1)) == copy!(zA2, B1)
             A3 = CuArray(randn(T, (m1, m2)))
             A3c = copy(A3)
             B3 = f1(StridedView(A3c))
             @. B1 = 2 * B1 - B3 / 3 # test copyto! of Broadcasted
             @. A1 = 2 * A1 - A3 / 3 # test copyto! of Broadcasted
-            @test CUDA.Adapt.adapt(Vector{T}, f1(A1)) == CUDA.Adapt.adapt(Vector{T}, B1)
+            @test CUDACore.Adapt.adapt(Vector{T}, f1(A1)) == CUDACore.Adapt.adapt(Vector{T}, B1)
             x = rand(T)
-            @test f1(StridedView(CUDA.Adapt.adapt(Vector{T}, fill!(A1c, x)))) == CUDA.Adapt.adapt(Vector{T}, fill!(B1, x))
+            @test f1(StridedView(CUDACore.Adapt.adapt(Vector{T}, fill!(A1c, x)))) == CUDACore.Adapt.adapt(Vector{T}, fill!(B1, x))
         end
     end
 end

test/runtests.jl

@@ -4,7 +4,7 @@ using Random
 using Strided
 using Strided: StridedView
 using Aqua
-using JLArrays, AMDGPU, CUDA, GPUArrays
+using JLArrays, AMDGPU, CUDACore, cuRAND, GPUArrays
 
 Random.seed!(1234)
 
@@ -32,7 +32,7 @@ if !is_buildkite
     Aqua.test_all(Strided; piracies = false)
 end
 
-if CUDA.functional()
+if CUDACore.functional()
     include("cuda.jl")
 end