trying to make the ci happy

src/GPU/KernelLauncher.jl

@@ -39,13 +39,13 @@ Launches a GPU kernel using the specified backend. This interface provides a gen
 mechanism for running GPU-accelerated computations across different GPU backends.
 
 # Arguments
-- `kernel::AbstractGPUKernel`: The GPU kernel to be launched.
+- `::AbstractGPUKernel`: The GPU kernel to be launched.
 
 # Notes
 This function must be implemented for specific GPU kernels. If not implemented,
 an error will be thrown.
 """
-function launch!(kernel::AbstractKernel)
+function launch!(::AbstractKernel)
     throw(ErrorException("A concrete implementation of launch! is required"))
 end
 
@@ -94,12 +94,5 @@ getbackend(kernel::LazyKernel) = kernel.backend
 
 
 ### GPU Backend ###
-
-"""
-    BackendCUDA <: AbstractGPUBackend
-
-Represents the CUDA backend for GPU acceleration. This type serves as a concrete
-implementation of `AbstractGPUBackend` for executing GPU computations using CUDA.
-"""
 struct BackendCUDA <: AbstractBackend end
 struct BackendCPU <: AbstractBackend end

test/GPU/test_adapt.jl

@@ -44,7 +44,7 @@ function nodes_gpu_kernel(nodes, dh, cv)
 end
 
 @testset "Adapt" begin
-    dh, cv = generate_problem()
+    dh, cv = generate_Bilinear_problem()
     cpudofs = dofs_cpu(dh, cv) |> cu
     ncells = dh |> get_grid |> getncells
     nbasefunctions = cv |> getnbasefunctions

test/GPU/test_iterator.jl

@@ -91,10 +91,10 @@ function get_cpu_kefe(dh, cellvalues)
     return kes |> cu, fes |> cu
 end
 
-@testset "Test iterators" begin
-    dh, cellvalues = generate_problem()
+@testset "Test shared memory iterators" begin
+    dh, cellvalues = generate_Bilinear_problem()
     n_basefuncs = getnbasefunctions(cellvalues)
-    # 1. Test that dofs for each cell in the grid are correctly computed
+    # 1 Test that dofs for each cell in the grid are correctly computed
     ncells = dh |> get_grid |> getncells
     dofs = CUDA.fill(Int32(0), n_basefuncs, ncells)
     correct_dofs = getalldofs(dh)
@@ -109,3 +109,23 @@ end
     @test all(abs.(kes_gpu .- kes_cpu) .< 1.0e-3) #TODO: This needs further investigation
     @test all(fes_gpu .≈ fes_cpu)
 end
+
+
+@testset "Test global memory iterators" begin
+    dh, cellvalues = generate_Biquadratic_problem()
+    n_basefuncs = getnbasefunctions(cellvalues)
+    # 1 Test that dofs for each cell in the grid are correctly computed
+    ncells = dh |> get_grid |> getncells
+    dofs = CUDA.fill(Int32(0), n_basefuncs, ncells)
+    correct_dofs = getalldofs(dh)
+    init_kernel(BackendCUDA, ncells, n_basefuncs, dof_kernel_kernel!, (dofs, dh, n_basefuncs)) |> launch!
+    @test all(dofs .≈ correct_dofs)
+
+    # 2. Test that local ke and fe are correctly computed
+    kes_gpu = CUDA.fill(0.0f0, ncells, n_basefuncs, n_basefuncs)
+    fes_gpu = CUDA.fill(0.0f0, ncells, n_basefuncs)
+    init_kernel(BackendCUDA, ncells, n_basefuncs, localkefe_kernel!, (kes_gpu, fes_gpu, cellvalues, dh)) |> launch!
+    kes_cpu, fes_cpu = get_cpu_kefe(dh, cellvalues)
+    @test all(abs.(kes_gpu .- kes_cpu) .< 1.0e-1) #TODO: This needs further investigation
+    @test all(fes_gpu .≈ fes_cpu)
+end

test/GPU/test_utils.jl

@@ -1,4 +1,4 @@
-function generate_problem()
+function generate_Bilinear_problem()
     left = Tensor{1, 2, Float32}((0, -0))
 
     right = Tensor{1, 2, Float32}((rand(10.0:100000.0), rand(10.0:100000.0)))
@@ -21,3 +21,27 @@ function generate_problem()
 
     return dh, cellvalues
 end
+
+function generate_Biquadratic_problem()
+    left = Tensor{1, 2, Float32}((0, -0))
+
+    right = Tensor{1, 2, Float32}((rand(10.0:100000.0), rand(10.0:100000.0)))
+
+    grid_dims = (rand(100:1000), rand(100:1000)) # to make sure the problem is big enough to use `CUDAGlobalCellIterator`
+
+    grid = generate_grid(Quadrilateral, grid_dims, left, right)
+
+    ip = Lagrange{RefQuadrilateral, 2}() # define the interpolation function (i.e. Biquadratic lagrange)
+
+    qr = QuadratureRule{RefQuadrilateral}(Float32, 3) # 3x3 quadrature rule
+
+    cellvalues = CellValues(Float32, qr, ip)
+
+    dh = DofHandler(grid)
+
+    add!(dh, :u, ip)
+
+    close!(dh)
+
+    return dh, cellvalues
+end