CI: use the shared TestGroups workflow

.github/workflows/ci.yml

@@ -7,6 +7,7 @@ on:
       - 'release-'
     tags: '*'
   pull_request:
+    types: [opened, synchronize, reopened, ready_for_review, closed]
   workflow_dispatch:
 
 concurrency:
@@ -15,62 +16,26 @@ concurrency:
 
 jobs:
   test:
-    name: Julia ${{ matrix.version }} - ${{ matrix.os }} - ${{ matrix.arch }}
-    runs-on: ${{ matrix.os }}
-    timeout-minutes: 90
-    strategy:
-      fail-fast: false
-      matrix:
-        version:
-          - 'lts' # julia lts
-          - '1' # latest stable 1.x release of Julia
-        os:
-          - ubuntu-latest
-          - macOS-latest
-          - windows-latest
-        arch:
-          - x64
-    steps:
-      - uses: actions/checkout@v7
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: ${{ matrix.version }}
-          arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v3
-      - uses: julia-actions/julia-buildpkg@latest
-      - uses: julia-actions/julia-runtest@latest
-        env:
-          JULIA_NUM_THREADS: 4
-      - uses: julia-actions/julia-processcoverage@v1
-      - uses: codecov/codecov-action@v7
-        with:
-          token: ${{ secrets.CODECOV_TOKEN }}
-          file: lcov.info
+    name: "Tests"
+    uses: "QuantumKitHub/QuantumKitHubActions/.github/workflows/TestGroups.yml@main"
+    with:
+      fast: ${{ github.event.pull_request.draft == true }}
+      nthreads: 4
+      timeout-minutes: 90
+      # Enzyme plays a large role here, so also test macOS and Windows on PRs, once ubuntu passes
+      extra-os-on-prs: true
+      staged: ${{ github.event_name == 'pull_request' }}
+    secrets:
+      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+
   test-nightly:
     needs: test
-    name: Julia nightly - ${{ matrix.os }} - ${{ matrix.arch }}
-    runs-on: ${{ matrix.os }}
-    timeout-minutes: 90
-    strategy:
-      fail-fast: false
-      matrix:
-        version:
-          - 'nightly'
-        os:
-          - ubuntu-latest
-          - macOS-latest
-          - windows-latest
-        arch:
-          - x64
-    steps:
-      - uses: actions/checkout@v7
-      - uses: julia-actions/setup-julia@v2
-        with:
-          version: ${{ matrix.version }}
-          arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v3
-      - uses: julia-actions/julia-buildpkg@latest
-      - uses: julia-actions/julia-runtest@latest
-        continue-on-error: true
-        env:
-          JULIA_NUM_THREADS: 4
+    name: "Tests (nightly)"
+    uses: "QuantumKitHub/QuantumKitHubActions/.github/workflows/Tests.yml@main"
+    with:
+      julia-version: "nightly"
+      nthreads: 4
+      timeout-minutes: 90
+      coverage: false
+    secrets:
+      CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}

Project.toml

@@ -58,6 +58,7 @@ LinearAlgebra = "1.6"
 Logging = "1.6"
 Mooncake = "0.5"
 PackageExtensionCompat = "1"
+ParallelTestRunner = "2"
 PrecompileTools = "1.1"
 Preferences = "1.4"
 PtrArrays = "1.2"
@@ -86,10 +87,11 @@ EnzymeTestUtils = "12d8515a-0907-448a-8884-5fe00fdf1c5a"
 JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
 Logging = "56ddb016-857b-54e1-b83d-db4d58db5568"
 Mooncake = "da2b9cff-9c12-43a0-ae48-6db2b0edb7d6"
+ParallelTestRunner = "d3525ed8-44d0-4b2c-a655-542cee43accc"
 Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
 TBLIS = "48530278-0828-4a49-9772-0f3830dfa1e9"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 cuTENSOR = "011b41b2-24ef-40a8-b3eb-fa098493e9e1"
 
 [targets]
-test = ["Test", "Random", "DynamicPolynomials", "ChainRulesTestUtils", "ChainRulesCore", "cuRAND", "CUDACore", "cuTENSOR", "Aqua", "Logging", "Bumper", "Mooncake", "Enzyme", "EnzymeTestUtils", "Adapt", "JLArrays", "AMDGPU", "TBLIS"]
+test = ["Test", "Random", "DynamicPolynomials", "ChainRulesTestUtils", "ChainRulesCore", "cuRAND", "CUDACore", "cuTENSOR", "Aqua", "Logging", "Bumper", "Mooncake", "Enzyme", "EnzymeTestUtils", "Adapt", "JLArrays", "AMDGPU", "TBLIS", "ParallelTestRunner"]

test/core/aqua.jl

@@ -0,0 +1,3 @@
+using TensorOperations
+using Aqua
+Aqua.test_all(TensorOperations)

test/butensor.jl → test/extensions/bumper.jl

@@ -1,7 +1,7 @@
 @testset "@butensor dependency check" begin
     @test_throws ArgumentError begin
         ex = :(@butensor A[a, b, c, d] := B[a, b, c, d])
-        macroexpand(Main, ex)
+        macroexpand(@__MODULE__, ex)
     end
 end
 

test/cutensor.jl → test/extensions/cutensor.jl

@@ -1,7 +1,7 @@
 @testset "@cutensor dependency check" begin
     @test_throws ArgumentError begin
         ex = :(@cutensor A[a, b, c, d] := B[a, b, c, d])
-        macroexpand(Main, ex)
+        macroexpand(@__MODULE__, ex)
     end
 end
 

test/gpu.jl → test/extensions/gpuarrays.jl

@@ -6,6 +6,7 @@ using TupleTools
 using JLArrays
 using VectorInterface
 using CUDACore, AMDGPU
+using cuTENSOR # contractions of `CuArray`s require the cuTENSOR extension
 
 test_result(a::AbstractArray, b::AbstractArray; kwargs...) =
     isapprox(collect(a), collect(b); kwargs...)

test/extensions/tblis_unsupported.jl

@@ -0,0 +1,11 @@
+# the Windows build of `tblis_jll` aborts the process from inside the library, so
+# the extension refuses to call into it and there is nothing else to test
+using TensorOperations
+using TBLIS
+using Test
+
+A = randn(Float64, (3, 4))
+@test_throws ArgumentError tensoradd!(
+    zeros(Float64, (4, 3)), A, ((2, 1), ()), false, 1, 0,
+    TensorOperations.TBLISBackend()
+)

test/runtests.jl

@@ -1,109 +1,68 @@
+using ParallelTestRunner
 using TensorOperations
-using LinearAlgebra
-using Test
-using Random
-Random.seed!(1234567)
 
-using TensorOperations: IndexError
-using TensorOperations: BaseCopy, BaseView, StridedNative, StridedBLAS
-using TensorOperations: DefaultAllocator, ManualAllocator, BufferAllocator
+testsuite = find_tests(@__DIR__)
 
-precision(::Type{<:Union{Float32, Complex{Float32}}}) = 1.0e-2
-precision(::Type{<:Union{Float64, Complex{Float64}}}) = 1.0e-8
+# --fast is passed by the CI on draft PRs, but there are no slow tests to skip
+args = parse_args(ARGS; custom = ["fast"])
+filter_tests!(testsuite, args)
 
-# https://github.com/QuantumKitHub/TensorOperations.jl/issues/280: the generated code has to
-# keep the user's `LineNumberNode`s -- Julia only emits a coverage counter for lines that a
-# `LineNumberNode` points at -- and drop the ones synthesized by the parser, which would
-# otherwise re-attribute the statements that follow them to a line in TensorOperations itself.
-# Only `LineNumberNode`s in block-statement position matter for that: elsewhere (most notably
-# the mandatory 2nd argument of a `:macrocall`) they are structurally required and left alone.
-function statementlinenumbernodes(ex, acc = LineNumberNode[])
-    if ex isa Expr
-        if ex.head === :block
-            for e in ex.args
-                e isa LineNumberNode && push!(acc, e)
-            end
-        end
-        foreach(e -> statementlinenumbernodes(e, acc), ex.args)
-    end
-    return acc
+# don't run all tests on GPU, only the GPU specific ones
+if get(ENV, "BUILDKITE", "false") == "true"
+    filter!(in(("extensions/cutensor", "extensions/gpuarrays")) ∘ first, testsuite)
 end
+# the Windows build of TBLIS can only be tested for refusing to run
+delete!(testsuite, Sys.iswindows() ? "extensions/tblis" : "extensions/tblis_unsupported")
 
-# don't run all tests on GPU, only the GPU
-# specific ones
-is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
-if !is_buildkite
-    @testset "tensoropt" verbose = true begin
-        include("tensoropt.jl")
-    end
-    @testset "auxiliary" verbose = true begin
-        include("auxiliary.jl")
-    end
-    @testset "macro keywords" verbose = true begin
-        include("macro_kwargs.jl")
-    end
-    @testset "method syntax" verbose = true begin
-        include("methods.jl")
-    end
-    @testset "macro with index notation" verbose = true begin
-        include("tensor.jl")
-    end
-    @testset "allocator" verbose = true begin
-        include("allocator.jl")
-    end
-    @testset "ad" verbose = false begin
-        include("ad.jl")
-    end
-    @testset "mooncake" verbose = false begin
-        include("mooncake.jl")
-    end
-    is_apple_ci = Sys.isapple() && get(ENV, "CI", "false") == "true"
-    if !is_apple_ci
-        @testset "enzyme" verbose = false begin
-            include("enzyme.jl")
-        end
-    end
+# The cuTENSOR and Bumper tests first check that the macros requiring them throw when they are
+# not loaded, so they run on a fresh worker rather than one that may have loaded them already
+const fresh_worker_tests = ("extensions/cutensor", "extensions/bumper")
+const fresh_workers = Channel{ParallelTestRunner.PTRWorker}(Inf)
+function test_worker(name)
+    name in fresh_worker_tests || return nothing
+    worker = addworker()
+    put!(fresh_workers, worker)
+    return worker
 end
 
-# note: cuTENSOR should not be loaded before this point
-# as there is a test which requires it to be loaded after
-@testset "cuTENSOR extension" verbose = true begin
-    include("cutensor.jl")
-end
-@testset "GPUArrays" verbose = true begin
-    include("gpu.jl")
-end
+const init_code = quote
+    using TensorOperations
+    using LinearAlgebra
+    using Test
+    using Random
+    Random.seed!(1234567)
 
-if !is_buildkite
-    # note: Bumper should not be loaded before this point
-    # as there is a test which requires it to be loaded after
-    @testset "Bumper extension" verbose = true begin
-        include("butensor.jl")
-    end
+    const is_buildkite = get(ENV, "BUILDKITE", "false") == "true"
 
-    if Sys.iswindows()
-        # the Windows build of `tblis_jll` aborts the process from inside the library, so
-        # the extension refuses to call into it and there is nothing else to test
-        using TBLIS
-        @testset "TBLIS extension (unsupported platform)" begin
-            A = randn(Float64, (3, 4))
-            @test_throws ArgumentError tensoradd!(
-                zeros(Float64, (4, 3)), A, ((2, 1), ()), false, 1, 0,
-                TensorOperations.TBLISBackend()
-            )
-        end
-    else
-        @testset "TBLIS extension" verbose = true begin
-            include("tblis.jl")
-        end
-    end
+    using TensorOperations: IndexError
+    using TensorOperations: BaseCopy, BaseView, StridedNative, StridedBLAS
+    using TensorOperations: DefaultAllocator, ManualAllocator, BufferAllocator
 
-    @testset "Polynomials" begin
-        include("polynomials.jl")
-    end
+    precision(::Type{<:Union{Float32, Complex{Float32}}}) = 1.0e-2
+    precision(::Type{<:Union{Float64, Complex{Float64}}}) = 1.0e-8
 
-    @testset "Aqua" verbose = true begin
-        using Aqua
-        Aqua.test_all(TensorOperations)
+    # https://github.com/QuantumKitHub/TensorOperations.jl/issues/280: the generated code has to
+    # keep the user's `LineNumberNode`s -- Julia only emits a coverage counter for lines that a
+    # `LineNumberNode` points at -- and drop the ones synthesized by the parser, which would
+    # otherwise re-attribute the statements that follow them to a line in TensorOperations itself.
+    # Only `LineNumberNode`s in block-statement position matter for that: elsewhere (most notably
+    # the mandatory 2nd argument of a `:macrocall`) they are structurally required and left alone.
+    function statementlinenumbernodes(ex, acc = LineNumberNode[])
+        if ex isa Expr
+            if ex.head === :block
+                for e in ex.args
+                    e isa LineNumberNode && push!(acc, e)
+                end
+            end
+            foreach(e -> statementlinenumbernodes(e, acc), ex.args)
+        end
+        return acc
     end
 end
+
+try
+    runtests(TensorOperations, args; testsuite, init_code, test_worker)
+finally
+    close(fresh_workers)
+    foreach(ParallelTestRunner.Malt.stop, fresh_workers)
+end