JuliaGPU · maleadt · Apr 21, 2026 · Apr 17, 2026 · Apr 17, 2026 · Apr 17, 2026
diff --git a/.buildkite/pipeline.yml b/.buildkite/pipeline.yml
@@ -348,7 +348,7 @@ steps:
           - JuliaCI/julia#v1:
               version: "1.12"
           - JuliaCI/julia-test#v1:
-              test_args: "--gpu=0,1 --quickfail core libraries/cublas/xt"
+              test_args: "--quickfail core libraries/cublas/xt"
           - JuliaCI/julia-coverage#v1:
               dirs:
                 - src

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
diff --git a/examples/peakflops.jl b/examples/peakflops.jl
@@ -23,7 +23,7 @@ function kernel_100fma(a, b, c, out)
     return
 end
 
-function peakflops(n::Integer=5000, dev::CuDevice=CuDevice(0))
+function peakflops(n::Integer=1024, dev::CuDevice=CuDevice(0))
     device!(dev) do
         dims = (n, n)
         a = round.(rand(Float32, dims) * 100)

diff --git a/lib/cublas/test/Project.toml b/lib/cublas/test/Project.toml
@@ -3,7 +3,11 @@ Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
 BFloat16s = "ab4f0b2a-ad5b-11e8-123f-65d77653426b"
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuBLAS = "182d3088-87b7-4494-8cad-fc6afaa545bc"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cublas/test/runtests.jl b/lib/cublas/test/runtests.jl
@@ -1,11 +1,11 @@
-include("setup.jl")
+using cuBLAS
+using ParallelTestRunner
 
-@testset verbose=true "cuBLAS" begin
-    for (root, _, files) in walkdir(@__DIR__)
-        for file in sort(files)
-            endswith(file, ".jl") || continue
-            file in ("setup.jl", "runtests.jl") && continue
-            include(joinpath(root, file))
-        end
-    end
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
+
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuBLAS, ARGS; init_code, testsuite)
diff --git a/lib/cudnn/test/Project.toml b/lib/cudnn/test/Project.toml
@@ -2,7 +2,11 @@
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
 NNlib = "872c559c-99b0-510c-b3b7-b6c96a88d5cd"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuDNN = "02a925ec-e4fe-4b08-9a7e-0d78e3d38ccd"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cudnn/test/runtests.jl b/lib/cudnn/test/runtests.jl
@@ -1,20 +1,11 @@
-include("setup.jl")
-@test cuDNN.functional()
+using cuDNN
+using ParallelTestRunner
 
-@testset verbose=true "cuDNN" begin
-
-# include all tests
-for entry in readdir(@__DIR__)
-    endswith(entry, ".jl") || continue
-    entry in ["runtests.jl", "setup.jl"] && continue
-
-    # generate a testset
-    name = splitext(entry)[1]
-    @eval begin
-        @testset $name begin
-            include($entry)
-        end
-    end
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
 
-end
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuDNN, ARGS; init_code, testsuite)
diff --git a/lib/cufft/test/Project.toml b/lib/cufft/test/Project.toml
@@ -4,5 +4,9 @@ Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 FFTW = "7a1cc6ca-52ef-59f5-83cd-3a7055c09341"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuFFT = "533571aa-0936-420e-b4be-9c66f5f626ca"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cufft/test/ensure_increasing.jl b/lib/cufft/test/ensure_increasing.jl
@@ -0,0 +1,10 @@
+using cuFFT
+
+@testset "ensure_increasing" begin
+    @test (1,2,3) == cuFFT.ensure_increasing((1,2,3))
+    @test (1,2,3) == cuFFT.ensure_increasing((2,1,3))
+    @test (1,2,3) == cuFFT.ensure_increasing((1,3,2))
+    @test (1,2,3) == cuFFT.ensure_increasing((3,1,2))
+    @test (1,2,3) == cuFFT.ensure_increasing((2,3,1))
+    @test (10,20,30) == cuFFT.ensure_increasing((30,20,10))
+end
diff --git a/lib/cufft/test/runtests.jl b/lib/cufft/test/runtests.jl
@@ -1,16 +1,11 @@
-include("setup.jl")
-@test cuFFT.version() isa VersionNumber
+using cuFFT
+using ParallelTestRunner
 
-@testset "ensure_increasing" begin
-    @test (1,2,3) == cuFFT.ensure_increasing((1,2,3))
-    @test (1,2,3) == cuFFT.ensure_increasing((2,1,3))
-    @test (1,2,3) == cuFFT.ensure_increasing((1,3,2))
-    @test (1,2,3) == cuFFT.ensure_increasing((3,1,2))
-    @test (1,2,3) == cuFFT.ensure_increasing((2,3,1))
-    @test (10,20,30) == cuFFT.ensure_increasing((30,20,10))
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
 
-include("complex.jl")
-include("real.jl")
-include("integer.jl")
-include("issues.jl")
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuFFT, ARGS; init_code, testsuite)
diff --git a/lib/curand/test/Project.toml b/lib/curand/test/Project.toml
@@ -1,5 +1,9 @@
 [deps]
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuRAND = "20fd9a0b-12d5-4c2f-a8af-7c34e9e60431"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/curand/test/runtests.jl b/lib/curand/test/runtests.jl
@@ -1,5 +1,11 @@
-include("setup.jl")
-@test cuRAND.version() isa VersionNumber
+using cuRAND
+using ParallelTestRunner
 
-include("high_level.jl")
-include("native.jl")
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
+end
+
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuRAND, ARGS; init_code, testsuite)
diff --git a/lib/cusolver/test/Project.toml b/lib/cusolver/test/Project.toml
@@ -2,8 +2,12 @@
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuBLAS = "182d3088-87b7-4494-8cad-fc6afaa545bc"
 cuSOLVER = "887afef0-6a32-4de5-add4-7827692ba8fc"
 cuSPARSE = "b26da814-b3bc-49ef-b0ee-c816305aa060"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cusolver/test/runtests.jl b/lib/cusolver/test/runtests.jl
@@ -1,11 +1,11 @@
-include("setup.jl")
+using cuSOLVER
+using ParallelTestRunner
 
-@testset verbose=true "cuSOLVER" begin
-    for (root, _, files) in walkdir(@__DIR__)
-        for file in sort(files)
-            endswith(file, ".jl") || continue
-            file in ("setup.jl", "runtests.jl") && continue
-            include(joinpath(root, file))
-        end
-    end
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
+
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuSOLVER, ARGS; init_code, testsuite)
diff --git a/lib/cusparse/test/Project.toml b/lib/cusparse/test/Project.toml
@@ -4,10 +4,14 @@ CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 InteractiveUtils = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
 KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 SparseMatricesCSR = "a0a7dd2c-ebf4-11e9-1f05-cf50bc540ca1"
 SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
 StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuSPARSE = "b26da814-b3bc-49ef-b0ee-c816305aa060"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cusparse/test/array.jl b/lib/cusparse/test/array.jl
@@ -8,9 +8,10 @@
     @test ndims(d_x)  == 1
     dense_d_x = CuVector(x)
     dense_d_x2 = CuVector(d_x)
+    ctx = :module => @__MODULE__
     CUDACore.@allowscalar begin
-        @test sprint(show, d_x) == replace(sprint(show, x), "SparseVector{Float64, Int64}"=>"cuSPARSE.CuSparseVector{Float64, Int32}", "sparsevec(["=>"sparsevec(Int32[")
-        @test sprint(show, MIME"text/plain"(), d_x) == replace(sprint(show, MIME"text/plain"(), x), "SparseVector{Float64, Int64}"=>"CuSparseVector{Float64, Int32}", "sparsevec(["=>"sparsevec(Int32[")
+        @test sprint(show, d_x; context=ctx) == replace(sprint(show, x; context=ctx), "SparseVector{Float64, Int64}"=>"cuSPARSE.CuSparseVector{Float64, Int32}", "sparsevec(["=>"sparsevec(Int32[")
+        @test sprint(show, MIME"text/plain"(), d_x; context=ctx) == replace(sprint(show, MIME"text/plain"(), x; context=ctx), "SparseVector{Float64, Int64}"=>"CuSparseVector{Float64, Int32}", "sparsevec(["=>"sparsevec(Int32[")
         @test Array(d_x[:])        == x[:]
         @test d_x[firstindex(d_x)] == x[firstindex(x)]
         @test d_x[div(end, 2)]     == x[div(end, 2)]
@@ -45,8 +46,8 @@
     @test size(d_x,3) == 1
     @test ndims(d_x)  == 2
     CUDACore.@allowscalar begin
-        @test sprint(show, d_x) == sprint(show, SparseMatrixCSC(d_x))
-        @test sprint(show, MIME"text/plain"(), d_x) == replace(sprint(show, MIME"text/plain"(), x), "SparseMatrixCSC{Float64, Int64}"=>"CuSparseMatrixCSC{Float64, Int32}")
+        @test sprint(show, d_x; context=ctx) == sprint(show, SparseMatrixCSC(d_x); context=ctx)
+        @test sprint(show, MIME"text/plain"(), d_x; context=ctx) == replace(sprint(show, MIME"text/plain"(), x; context=ctx), "SparseMatrixCSC{Float64, Int64}"=>"CuSparseMatrixCSC{Float64, Int32}")
         @test Array(d_x[:])        == x[:]
         @test d_x[:, :]            == x[:, :]
         @test d_tx[:, :]           == transpose(x)[:, :]
@@ -247,7 +248,7 @@
     # to hit the CuSparseArrayCSR path
     CUDACore.unsafe_free!(repeat(d_x, 1, 1, 3))
     CUDACore.@allowscalar begin
-        @test startswith(sprint(show, MIME"text/plain"(), repeat(d_x, 1, 1, 2)), "$m×$n×2 CuSparseArrayCSR{Float64, Int32, 3} with $(2*nnz(d_x)) stored entries:\n")
+        @test startswith(sprint(show, MIME"text/plain"(), repeat(d_x, 1, 1, 2); context=ctx), "$m×$n×2 CuSparseArrayCSR{Float64, Int32, 3} with $(2*nnz(d_x)) stored entries:\n")
     end
     @test length(d_x) == m*n
     @test_throws ArgumentError copyto!(d_y,d_x)

diff --git a/lib/cusparse/test/runtests.jl b/lib/cusparse/test/runtests.jl
@@ -1,12 +1,11 @@
-include("setup.jl")
-@test cuSPARSE.version() isa VersionNumber
+using cuSPARSE
+using ParallelTestRunner
 
-@testset verbose=true "cuSPARSE" begin
-    for (root, _, files) in walkdir(@__DIR__)
-        for file in sort(files)
-            endswith(file, ".jl") || continue
-            file in ("setup.jl", "runtests.jl") && continue
-            include(joinpath(root, file))
-        end
-    end
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
+
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuSPARSE, ARGS; init_code, testsuite)
diff --git a/lib/custatevec/test/Project.toml b/lib/custatevec/test/Project.toml
@@ -2,5 +2,9 @@
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuStateVec = "92f7fd98-d22e-4c0d-85a8-6ade11b672fb"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/custatevec/test/runtests.jl b/lib/custatevec/test/runtests.jl
@@ -1,13 +1,11 @@
-include("setup.jl")
-@test cuStateVec.functional()
+using cuStateVec
+using ParallelTestRunner
 
-@testset verbose=true "cuStateVec" begin
-    include("errors.jl")
-    include("apply_matrix.jl")
-    include("apply_pauli.jl")
-    include("measure.jl")
-    include("collapse.jl")
-    include("utilities.jl")
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
 
-include("multigpu.jl")
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuStateVec, ARGS; init_code, testsuite)
diff --git a/lib/cutensor/test/Project.toml b/lib/cutensor/test/Project.toml
@@ -2,6 +2,10 @@
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuTENSOR = "011b41b2-24ef-40a8-b3eb-fa098493e9e1"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cutensor/test/kernel_cache.jl b/lib/cutensor/test/kernel_cache.jl
@@ -0,0 +1,14 @@
+using cuTENSOR
+using CUDACore
+
+if CUDACore.runtime_version() >= v"11.8" && capability(device()) >= v"8.0"
+    @testset "kernel cache" begin
+        mktempdir() do dir
+            cd(dir) do
+                cuTENSOR.write_cache!("kernelCache.bin")
+                @test isfile("kernelCache.bin")
+                cuTENSOR.read_cache!("kernelCache.bin")
+            end
+        end
+    end
+end
diff --git a/lib/cutensor/test/runtests.jl b/lib/cutensor/test/runtests.jl
@@ -1,27 +1,11 @@
-include("setup.jl")
-@test cuTENSOR.functional()
+using cuTENSOR
+using ParallelTestRunner
 
-@testset verbose=true "cuTENSOR" begin
-
-include("base.jl")
-
-include("elementwise_binary.jl")
-include("elementwise_trinary.jl")
-include("permutations.jl")
-include("contractions.jl")
-include("reductions.jl")
-
-# we should have some kernels in the cache after this
-if CUDACore.runtime_version() >= v"11.8" && capability(device()) >= v"8.0"
-@testset "kernel cache" begin
-    mktempdir() do dir
-    cd(dir) do
-        cuTENSOR.write_cache!("kernelCache.bin")
-        @test isfile("kernelCache.bin")
-        cuTENSOR.read_cache!("kernelCache.bin")
-    end
-    end
-end
+const init_code = quote
+    include(joinpath(@__DIR__, "setup.jl"))
 end
 
-end
+testsuite = find_tests(@__DIR__)
+delete!(testsuite, "setup")
+
+runtests(cuTENSOR, ARGS; init_code, testsuite)
diff --git a/lib/cutensornet/test/Project.toml b/lib/cutensornet/test/Project.toml
@@ -2,6 +2,10 @@
 CUDACore = "bd0ed864-bdfe-4181-a5ed-ce625a5fdea2"
 CUDA_Runtime_jll = "76a88914-d11a-5bdc-97e0-2f5a05c973a2"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuTENSOR = "011b41b2-24ef-40a8-b3eb-fa098493e9e1"
 cuTensorNet = "448d79b3-4b49-4e06-a5ea-00c62c0dc3db"
+
+[compat]
+ParallelTestRunner = "2"
diff --git a/lib/cutensornet/test/contractions.jl b/lib/cutensornet/test/contractions.jl
@@ -27,7 +27,7 @@ m = 16
         aligns_in = UInt32.([256, 256, 256])
         aligns_out = UInt32(256)
         ctn = CuTensorNetwork(elty, modes_in, extents_in, [C_NULL, C_NULL, C_NULL], Int32[0, 0, 0], Int32.(modesD), extentsD, C_NULL)
-        @testset for max_ws_size in [2^28, 2^32]
+        @testset for max_ws_size in [2^28]
             @testset for tuning in [NoAutoTune(), AutoTune()]
                 ctn.input_arrs = raw_data_in
                 info = rehearse_contraction(ctn, max_ws_size)