Merge pull request #928

Setup AMD MI300 CI runners on Alps

Author: luraess

README.md

@@ -2,7 +2,7 @@
 
 *AMD GPU (ROCm) programming in Julia*
 
-[![][docs-stable-img]][docs-stable-url] [![][docs-dev-img]][docs-dev-url] [![][buildkite-img]][buildkite-url] [![][codecov-img]][codecov-url] [![][benchmark-img]][benchmark-url]
+[![][docs-stable-img]][docs-stable-url] [![][docs-dev-img]][docs-dev-url] [![][buildkite-img]][buildkite-url] [![][cscs-ci-img]][cscs-ci-url] [![][codecov-img]][codecov-url] [![][benchmark-img]][benchmark-url]
 
 [![][doi-img]][doi-url] [![][downloads-img]][downloads-url]
 
@@ -27,6 +27,9 @@
 [benchmark-img]: https://img.shields.io/badge/benchmarks-Chart-yellowgreen
 [benchmark-url]: https://amdgpu.juliagpu.org/bench/
 
+[cscs-ci-img]: https://gitlab.com/cscs-ci/ci-testing/webhook-ci/mirrors/3266148763123078/6677263153381401/badges/main/pipeline.svg?ignore_skipped=true
+[cscs-ci-url]: https://gitlab.com/cscs-ci/ci-testing/webhook-ci/mirrors/3266148763123078/6677263153381401/-/pipelines
+
 ## Quick start
 
 AMDGPU.jl can be installed with the Julia package manager.

ci/cscs-mi300.yml

@@ -0,0 +1,27 @@
+include:
+  - remote: 'https://gitlab.com/cscs-ci/recipes/-/raw/master/templates/v2/.ci-ext.yml'
+
+unit_test:
+  extends: .baremetal-runner-beverin-mi300
+  variables:
+    JULIA: /users/lraess/julia_amd/juliaup/julia-1.12.6+0.x64.linux.gnu/bin/julia
+    UENV: b2550889de318ab5
+    SLURM_JOB_NUM_NODES: 1
+    SLURM_NTASKS_PER_NODE: 1
+    SLURM_GPUS_PER_TASK: 2
+    SLURM_TIMELIMIT: "00:45:00"
+    JULIA_NUM_THREADS: 4
+    JULIA_DEPOT_PATH: "${CI_PROJECT_DIR}/.julia"
+    JULIA_AMDGPU_CORE_MUST_LOAD: "1"
+    JULIA_AMDGPU_HIP_MUST_LOAD: "1"
+    JULIA_AMDGPU_DISABLE_ARTIFACTS: "1"
+  script:
+    - srun -n 1 --uenv $UENV --view=default $JULIA --project -e '
+        println("Instantiating project");
+        using Pkg;
+        Pkg.activate(pwd());
+        Pkg.instantiate();
+        using AMDGPU;
+        AMDGPU.versioninfo();
+        println("Running tests");
+        Pkg.test("AMDGPU"; test_args=`--jobs=32`);'

test/hip_core_tests.jl

@@ -24,7 +24,7 @@ end
 
 if length(AMDGPU.devices()) > 1
     @testset "HIP Peer Access" begin
-        dev1, dev2, _ = AMDGPU.devices()
+        dev1, dev2 = AMDGPU.devices()[1:2]
         @test AMDGPU.HIP.can_access_peer(dev1, dev2) isa Bool
     end
 end