Debug ci hangs

.buildkite/pipeline.yml

@@ -21,3 +21,40 @@ steps:
     env:
       OCL_ICD_FILENAMES: "libnvidia-opencl.so.1"
     timeout_in_minutes: 60
+
+  - group: "macos"
+    key: "julia"
+    steps:
+      - label: "{{matrix.julia}} macos {{matrix.arch}}"
+        plugins:
+          - JuliaCI/julia#v1:
+              version: "{{matrix.julia}}"
+          - JuliaCI/julia-coverage#v1:
+              codecov: true
+        commands: |
+          julia --project -e '
+            using Pkg
+
+            println("--- :julia: Instantiating project")
+            Pkg.add("pocl_jll")
+            Pkg.add("InteractiveUtils")
+            Pkg.develop(path="lib/intrinsics")
+
+            println("+++ :julia: Running tests")
+            using InteractiveUtils
+            InteractiveUtils.versioninfo()
+            Pkg.test(; coverage=true, test_args=`--platform=pocl --verbose --jobs=2`)'
+        agents:
+          queue: "juliaecosystem"
+          os: "macos"
+          arch: "{{matrix.arch}}"
+        if: build.message !~ /\[skip tests\]/
+        timeout_in_minutes: 120
+        matrix:
+          setup:
+            julia:
+              - "1.10"
+              - "1.12"
+            arch:
+              - "x86_64"
+              - "aarch64"

.github/workflows/Test.yml

@@ -23,15 +23,15 @@ jobs:
       fail-fast: false
       matrix:
         version: ['1.10', '1.12']
-        os: [ubuntu-24.04, ubuntu-24.04-arm, macOS-15, macOS-15-intel, windows-2025]
+        os: [ubuntu-24.04-arm, macOS-15, macOS-15-intel, windows-2022]
         arch: [x64, arm64]
-        pocl: [jll, local]
+        pocl: [jll]
         memory_backend: [usm, svm, buffer]
         exclude:
           # unsupported combinations
           - os: ubuntu-24.04
             arch: arm64
-          - os: windows-2025
+          - os: windows-2022
             arch: arm64
           - os: ubuntu-24.04-arm
             arch: x64
@@ -45,7 +45,7 @@ jobs:
             pocl: local
           - os: macOS-15
             pocl: local
-          - os: windows-2025
+          - os: windows-2022
             pocl: local
           # Remove these exclusions once macOS CI hangs are fixed
           - os: macOS-15-intel
@@ -150,7 +150,9 @@ jobs:
         uses: julia-actions/julia-runtest@v1
         if: runner.os != 'Windows'
         with:
-          test_args: '--quickfail --platform=pocl'
+          test_args: '--quickfail --platform=pocl --verbose --jobs=2'
+        env:
+          POCL_WORK_GROUP_METHOD: cbs
 
       - name: Setup BusyBox
         if: runner.os == 'Windows'

test/intrinsics.jl

@@ -15,28 +15,28 @@ const ispocl = cl.platform().name == "Portable Computing Language"
 @testset "intrinsics" begin
 
 @testset "barrier" begin
+@time "Barrier Local Mem fence" @on_device barrier(OpenCL.LOCAL_MEM_FENCE)
+@time "Barrier global Mem fence" @on_device barrier(OpenCL.GLOBAL_MEM_FENCE)
+@time "Barrier both Mem fence" @on_device barrier(OpenCL.LOCAL_MEM_FENCE | OpenCL.GLOBAL_MEM_FENCE)
 
-@on_device barrier(OpenCL.LOCAL_MEM_FENCE)
-@on_device barrier(OpenCL.GLOBAL_MEM_FENCE)
-@on_device barrier(OpenCL.LOCAL_MEM_FENCE | OpenCL.GLOBAL_MEM_FENCE)
+@time "WorkGroup Barrier Local Mem fence" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE)
+@time "WorkGroup Barrier global Mem fence" @on_device work_group_barrier(OpenCL.GLOBAL_MEM_FENCE)
+@time "WorkGroup Barrier image Mem fence" @on_device work_group_barrier(OpenCL.IMAGE_MEM_FENCE)
 
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE)
-@on_device work_group_barrier(OpenCL.GLOBAL_MEM_FENCE)
-@on_device work_group_barrier(OpenCL.IMAGE_MEM_FENCE)
+@time "WorkGroup Barrier L/G Mem fence" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE | OpenCL.GLOBAL_MEM_FENCE)
+@time "WorkGroup Barrier L/I Mem fence" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE | OpenCL.IMAGE_MEM_FENCE)
+@time "WorkGroup Barrier L/G/I Mem fence" @on_device work_group_barrier(OpenCL.GLOBAL_MEM_FENCE | OpenCL.LOCAL_MEM_FENCE | OpenCL.IMAGE_MEM_FENCE)
 
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE | OpenCL.GLOBAL_MEM_FENCE)
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE | OpenCL.IMAGE_MEM_FENCE)
-@on_device work_group_barrier(OpenCL.GLOBAL_MEM_FENCE | OpenCL.LOCAL_MEM_FENCE | OpenCL.IMAGE_MEM_FENCE)
-
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_work_item)
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_work_group)
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_device)
-cl.memory_backend() isa cl.SVMBackend && @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_all_svm_devices)
-@on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_sub_group)
+@time "WorkGroup Barrier Local Mem fence, work item scope" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_work_item)
+@time "WorkGroup Barrier Local Mem fence, workgroup scope" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_work_group)
+@time "WorkGroup Barrier Local Mem fence, device scope" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_device)
+@time "Skipped" cl.memory_backend() isa cl.SVMBackend && @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_all_svm_devices)
+@time "WorkGroup Barrier Local Mem fence, subgroup scope" @on_device work_group_barrier(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_scope_sub_group)
 
 end
 
 @testset "mem_fence" begin
+@warn "mem_fence"
 
 @on_device mem_fence(OpenCL.LOCAL_MEM_FENCE)
 @on_device mem_fence(OpenCL.GLOBAL_MEM_FENCE)
@@ -53,6 +53,7 @@ end
 end
 
 @testset "atomic_work_item_fence" begin
+@warn "atomic_work_item_fence"
 
 @on_device atomic_work_item_fence(OpenCL.LOCAL_MEM_FENCE, OpenCL.memory_order_relaxed, OpenCL.memory_scope_work_item)
 @on_device atomic_work_item_fence(OpenCL.GLOBAL_MEM_FENCE, OpenCL.memory_order_acquire, OpenCL.memory_scope_work_group)
@@ -64,6 +65,7 @@ cl.memory_backend() isa cl.SVMBackend && @on_device atomic_work_item_fence(OpenC
 end
 
 @testset "math" begin
+@warn "math"
 
 @testset "unary - $T" for T in float_types
     @testset "$f" for f in [
@@ -101,6 +103,7 @@ end
             hypot,
             (^),
         ]
+    @warn "binary - $T, $f"
         x = rand(T)
         y = rand(T)
         broken = ispocl && T == Float16 && f == atan
@@ -112,6 +115,7 @@ end
     @testset "$f" for f in [
             fma,
         ]
+    @warn "ternary - $T, $f"
         x = rand(T)
         y = rand(T)
         z = rand(T)
@@ -128,6 +132,8 @@ end
             OpenCL.rint,
             OpenCL.rsqrt,
         ]
+    @warn "OpenCL-specific unary - $T, $f"
+
         x = rand(T)
         broken = ispocl && T == Float16 && !(f in [OpenCL.rint, OpenCL.rsqrt])
         @test call_on_device(f, x) isa Real broken = broken  # Just check it doesn't error
@@ -146,6 +152,8 @@ end
             OpenCL.nextafter,
             OpenCL.powr,
         ]
+    @warn "OpenCL-specific binary - $T, $f"
+
         x = rand(T)
         y = rand(T)
         broken = ispocl && T == Float16 && !(f in [OpenCL.maxmag, OpenCL.minmag])
@@ -159,6 +167,8 @@ end
     x = rand(T)
     y = rand(T)
     z = rand(T)
+    @warn "OpenCL-specific ternary - $T"
+
     @test call_on_device(OpenCL.mad, x, y, z) ≈ x * y + z
 end
 

test/runtests.jl

@@ -46,7 +46,12 @@ do_verbose, _ = extract_flag!(ARGS, "--verbose")
 do_quickfail, _ = extract_flag!(ARGS, "--quickfail")
 
 include("setup.jl")     # make sure everything is precompiled
+
 @info "System information:\n" * sprint(io->OpenCL.versioninfo(io))
+Sys.isapple() && @info read(`launchctl limit maxfiles`, String)
+Sys.isapple() && @info read(`ulimit -a`, String)
+Sys.isapple() && @info read(`df -YIh`, String)
+Sys.isapple() && @info read(`vm_stat`, String)
 
 @info "Running $jobs tests in parallel. If this is too many, specify the `--jobs` argument to the tests, or set the JULIA_CPU_THREADS environment variable."