Add ROCKernels backend (#209)

* Approx compare in gamma/erf/erfc tests

Co-authored-by: Valentin Churavy <[email protected]>

Author: jpsamaroo

.buildkite/pipeline.yml

@@ -1,5 +1,5 @@
 steps:
-  - label: "Julia v1"
+  - label: "CUDA Julia v1"
     plugins:
       - JuliaCI/julia#v1:
           version: "1"
@@ -19,10 +19,10 @@ steps:
       KERNELABSTRACTIONS_TEST_BACKEND: "CUDA"
     timeout_in_minutes: 60
 
-  - label: "Julia 1.6-nightly"
+  - label: "CUDA Julia 1.6"
     plugins:
       - JuliaCI/julia#v1:
-          version: "1.6-nightly"
+          version: "1.6"
       - JuliaCI/julia-coverage#v1:
           codecov: true
           dirs:
@@ -39,7 +39,7 @@ steps:
       KERNELABSTRACTIONS_TEST_BACKEND: "CUDA"
     timeout_in_minutes: 60
 
-  - label: "Julia nightly"
+  - label: "CUDA Julia nightly"
     plugins:
       - JuliaCI/julia#v1:
           version: "nightly"
@@ -59,6 +59,25 @@ steps:
       KERNELABSTRACTIONS_TEST_BACKEND: "CUDA"
     timeout_in_minutes: 60
 
+  - label: "ROCm Julia 1.6"
+    plugins:
+      - JuliaCI/julia#v1:
+          version: "1.6"
+      - JuliaCI/julia-coverage#v1:
+          codecov: true
+          dirs:
+            - src
+            - lib
+    commands:
+      - julia .ci/develop.jl
+      - julia .ci/test.jl
+    agents:
+      queue: "juliagpu"
+      rocm: "*"
+    env:
+      KERNELABSTRACTIONS_TEST_BACKEND: "ROCM"
+    timeout_in_minutes: 60
+
 env:
   JULIA_PKG_SERVER: "" # it often struggles with our large artifacts
   SECRET_CODECOV_TOKEN: "c5pjPUwULD2L8ss0gRtjCPiagRlTQ11TdbZP6gIhTPeA/gN5w5/7JvDCg36UpKER6FXnQDeBGGkQafHiLdBaH/FWQ2B2VKErtBarIBJa2zWvKu8mYs9PJzw/qLGT2sMXI9kcao63H6/HAwbslJcY0a5Mg+SwM3M05XqSHgnrHMnbBXysKP6VzFEIX7uoyEKOnoWDj8rGJKFYLW2DBRtd6Yc23ESfFXPAqbS7sgXxwQHKzz20FMQBJUmbiDIzPlk3k2n2TvgAWQ0VNK0e4/UooMbULL3UjY4oaMOF0XpJAnWlmvGgy8gEnZKSVp3ieXy/Ubu7BWwH/BT59wDy6LuDxA==;U2FsdGVkX18Wil69f7qJYu6yU5iNx+Zq8akUcOp+McU1CR4Jw4QBsrUKIF4W4uK+/752FQo40BwFsfnIC8CJ/Q=="

.ci/develop.jl

@@ -1,11 +1,17 @@
 import Pkg
 
+Pkg.update()
+
 root_directory = dirname(@__DIR__)
 
 kernelabstractions = Pkg.PackageSpec(path = root_directory)
 cudakernels = Pkg.PackageSpec(path = joinpath(root_directory, "lib", "CUDAKernels"))
 
 Pkg.develop(kernelabstractions)
 Pkg.develop(cudakernels)
-
+if !(VERSION < v"1.6-")
+    rockernels = Pkg.PackageSpec(path = joinpath(root_directory, "lib", "ROCKernels"))
+    Pkg.develop(rockernels)
+end
+Pkg.build()
 Pkg.precompile()

.ci/test.jl

@@ -4,5 +4,8 @@ pkgs = [
     "KernelAbstractions",
     "CUDAKernels",
 ]
+if !(VERSION < v"1.6-")
+    push!(pkgs, "ROCKernels")
+end
 
 Pkg.test(pkgs; coverage = true)

lib/CUDAKernels/Project.toml

@@ -18,9 +18,3 @@ Cassette = "0.3.3"
 KernelAbstractions = "0.6"
 SpecialFunctions = "0.10, 1.0"
 StaticArrays = "0.12, 1.0"
-
-[extras]
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-
-[targets]
-test = ["Test"]

lib/CUDAKernels/test/Project.toml

@@ -0,0 +1,9 @@
+[deps]
+InteractiveUtils = "b77e0a4c-d291-57a0-90e8-8db25a27a240"
+KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
+MPI = "da04e1cc-30fd-572f-bb4f-1f8673147195"
+MacroTools = "1914dd2f-81c6-5fcd-8719-6d5c9610ff09"
+SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
+CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"

lib/CUDAKernels/test/runtests.jl

@@ -1,2 +1,27 @@
+using KernelAbstractions
+using CUDA
 using CUDAKernels
 using Test
+
+include(joinpath(dirname(pathof(KernelAbstractions)), "..", "test", "testsuite.jl"))
+
+if parse(Bool, get(ENV, "CI", "false"))
+    default = "CPU"
+else
+    default = "CUDA"
+end
+
+backend = get(ENV, "KERNELABSTRACTIONS_TEST_BACKEND", default)
+
+if backend != "CUDA"
+    @info "CUDA backend not selected"
+    exit()
+end
+
+CUDA.versioninfo()
+if CUDA.functional(true)
+    CUDA.allowscalar(false)
+    Testsuite.testsuite(CUDADevice, backend, CUDA, CuArray, CUDA.CuDeviceArray)
+else
+    error("No CUDA GPUs available!")
+end

lib/ROCKernels/Project.toml

@@ -0,0 +1,21 @@
+name = "ROCKernels"
+uuid = "7eb9e9f0-4bd3-4c4c-8bef-26bd9629d9b9"
+authors = ["Valentin Churavy <[email protected]>", "Julian P Samaroo <[email protected]>"]
+version = "0.1.0"
+
+[deps]
+AMDGPU = "21141c5a-9bdb-4563-92ae-f87d6854732e"
+Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"
+Cassette = "7057c7e9-c182-5462-911a-8362d720325c"
+KernelAbstractions = "63c18a36-062a-441e-b654-da1e3ab1ce7c"
+SpecialFunctions = "276daf66-3868-5448-9aa4-cd146d93841b"
+StaticArrays = "90137ffa-7385-5640-81b9-e52037218182"
+
+[compat]
+AMDGPU = "0.2.5"
+Adapt = "0.4, 1.0, 2.0, 3.0"
+Cassette = "0.3.3"
+KernelAbstractions = "0.6"
+SpecialFunctions = "0.10, 1.0"
+StaticArrays = "0.12, 1.0"
+julia = "1.6"