Add OpenCL.jl backend for CPU-based GPU debugging

.github/workflows/CI.yml

@@ -34,6 +34,7 @@ jobs:
         group:
           - CPU
           - JLArrays
+          - OpenCL
     uses: "SciML/.github/.github/workflows/tests.yml@master"
     with:
       julia-version: "${{ matrix.version }}"

Project.toml

@@ -30,13 +30,15 @@ AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
 Metal = "dde4c033-4e86-420c-a63e-0dd931031962"
+OpenCL = "08131aa3-fb12-5dee-8b74-c09406e224a2"
 oneAPI = "8f75cd03-7ff8-4ecb-9b8f-daf728133b1b"
 
 [extensions]
 AMDGPUExt = ["AMDGPU"]
 CUDAExt = ["CUDA"]
 JLArraysExt = ["JLArrays"]
 MetalExt = ["Metal"]
+OpenCLExt = ["OpenCL"]
 oneAPIExt = ["oneAPI"]
 
 [compat]
@@ -53,6 +55,7 @@ KernelAbstractions = "0.9"
 LinearSolve = "3"
 Metal = "1"
 MuladdMacro = "0.2"
+OpenCL = "0.9, 0.10"
 Parameters = "0.12"
 RecursiveArrayTools = "3"
 SciMLBase = "2.92"
@@ -70,5 +73,7 @@ CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 GPUArraysCore = "46192b85-c4d5-4398-a991-12ede77f4527"
 JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
 Metal = "dde4c033-4e86-420c-a63e-0dd931031962"
-oneAPI = "8f75cd03-7ff8-4ecb-9b8f-daf728133b1b"
+OpenCL = "08131aa3-fb12-5dee-8b74-c09406e224a2"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+oneAPI = "8f75cd03-7ff8-4ecb-9b8f-daf728133b1b"
+pocl_jll = "627d6b7a-bbe6-5189-83e7-98cc0a5aeadd"

README.md

@@ -36,6 +36,7 @@ SciML's GPU support extends to a wide array of hardware, including:
 | AMD              | ROCm                | [AMDGPU.jl](https://github.com/JuliaGPU/AMDGPU.jl) | `AMDGPU.ROCBackend()`    |
 | Intel            | OneAPI              | [OneAPI.jl](https://github.com/JuliaGPU/oneAPI.jl) | `oneAPI.oneAPIBackend()` |
 | Apple (M-Series) | Metal               | [Metal.jl](https://github.com/JuliaGPU/Metal.jl)   | `Metal.MetalBackend()`   |
+| CPU (fast kernels) | OpenCL           | [OpenCL.jl](https://github.com/JuliaGPU/OpenCL.jl) | `OpenCL.CLBackend()`     |
 
 For this tutorial we will demonstrate the CUDA backend for NVIDIA GPUs, though any of the other GPUs can be
 used by simply swapping out the `backend` choice.

ext/OpenCLExt.jl

@@ -0,0 +1,11 @@
+module OpenCLExt
+using OpenCL
+import DiffEqGPU
+
+using .OpenCL
+import .OpenCL: OpenCLBackend
+
+DiffEqGPU.maxthreads(::OpenCLBackend) = 256
+DiffEqGPU.maybe_prefer_blocks(::OpenCLBackend) = OpenCLBackend()
+
+end

test/Project.toml

@@ -8,15 +8,17 @@ ForwardDiff = "f6369f11-7733-5829-9624-2563aa707210"
 GPUArraysCore = "46192b85-c4d5-4398-a991-12ede77f4527"
 JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
+OpenCL = "08131aa3-fb12-5dee-8b74-c09406e224a2"
 Optimization = "7f7a1694-90dd-40f0-9382-eb1efda571ba"
 OptimizationOptimisers = "42dfb2eb-d2b4-4451-abcd-913932933ac1"
 OrdinaryDiffEq = "1dea7af3-3e70-54e6-95c3-0bf5283fa5ed"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 SafeTestsets = "1bc83da4-3b8d-516f-aca4-4fe02f6d838f"
+SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 StochasticDiffEq = "789caeaf-c7a9-5a7d-9973-96adeb23e2a0"
-SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 TOML = "fa267f1f-6049-4f14-aa54-33bafae1ed76"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 Zygote = "e88e6eb3-aa80-5325-afca-941959d7151f"
+pocl_jll = "627d6b7a-bbe6-5189-83e7-98cc0a5aeadd"

test/utils.jl

@@ -11,6 +11,9 @@ elseif GROUP == "oneAPI"
 elseif GROUP == "Metal"
     using Metal
     Metal.MetalBackend()
+elseif GROUP == "OpenCL"
+    using OpenCL, pocl_jll
+    OpenCL.OpenCLBackend()
 elseif GROUP == "JLArrays"
     using JLArrays
     JLArrays.JLBackend()