[ci] Adding ability to run kernel-specified test runners

[geomin12]

Motivation

For expansion of TheRock's test machines, we are adding test machines specifically for oem kernels and allowing users to use these machines via labels

Users will be able to trigger options via: workflow_dispatch and pull request labels

Technical Details

I ran apt install linux-oem-24.04a on a couple of Linux gfx1151 machines, then created an oem label and added logic to replace test runners with the oem test runner if label was specified

Test Plan

Testing all via CI

Test Result

First normal CI run (cancelled):
Output:

OUTPUT linux_variants=[{"test-runs-on": "linux-gfx120X-gpu-rocm", "family": "gfx120X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx120X-all"}, {"test-runs-on": "linux-mi325-1gpu-ossci-rocm-frac", "test-runs-on-multi-gpu": "linux-mi325-4gpu-ossci-rocm", "benchmark-runs-on": "linux-mi325-1gpu-ossci-rocm-frac", "family": "gfx94X-dcgpu", "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx94X-dcgpu"}, {"test-runs-on": "linux-gfx1151-gpu-rocm", "test-runs-on-kernel": {"oem": "linux-strix-halo-gpu-rocm-oem"}, "family": "gfx1151", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx1151"}, {"test-runs-on": "", "family": "gfx110X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx110X-all"}]
OUTPUT linux_test_labels=[]
OUTPUT windows_variants=[{"test-runs-on": "", "family": "gfx120X-all", "bypass_tests_for_releases": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "windows-release", "artifact_group": "gfx120X-all"}, {"test-runs-on": "windows-gfx1151-gpu-rocm", "benchmark-runs-on": "windows-gfx1151-gpu-rocm", "family": "gfx1151", "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "windows-release", "artifact_group": "gfx1151"}, {"test-runs-on": "windows-gfx110X-gpu-rocm", "family": "gfx110X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "windows-release", "artifact_group": "gfx110X-all"}]
OUTPUT windows_test_labels=[]
OUTPUT enable_build_jobs=true
OUTPUT test_type=smoke

OEM kernel label added run:
output:

OUTPUT linux_variants=[{"test-runs-on": "", "family": "gfx110X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx110X-all"}, {"test-runs-on": "linux-strix-halo-gpu-rocm-oem", "test-runs-on-kernel": {"oem": "linux-strix-halo-gpu-rocm-oem"}, "family": "gfx1151", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx1151"}, {"test-runs-on": "", "test-runs-on-multi-gpu": "", "benchmark-runs-on": "linux-mi325-1gpu-ossci-rocm-frac", "family": "gfx94X-dcgpu", "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx94X-dcgpu"}, {"test-runs-on": "", "family": "gfx120X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx120X-all"}]
OUTPUT linux_test_labels=[]
OUTPUT windows_variants=[{"test-runs-on": "", "family": "gfx110X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "windows-release", "artifact_group": "gfx110X-all"}, {"test-runs-on": "", "benchmark-runs-on": "windows-gfx1151-gpu-rocm", "family": "gfx1151", "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "windows-release", "artifact_group": "gfx1151"}, {"test-runs-on": "", "family": "gfx120X-all", "bypass_tests_for_releases": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "windows-release", "artifact_group": "gfx120X-all"}]
OUTPUT windows_test_labels=[]
OUTPUT enable_build_jobs=true
OUTPUT test_type=smoke

Workflow dispatch OEM kernel run
output:

OUTPUT linux_variants=[{"test-runs-on": "", "family": "gfx120X-all", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx120X-all"}, {"test-runs-on": "linux-strix-halo-gpu-rocm-oem", "test-runs-on-kernel": {"oem": "linux-strix-halo-gpu-rocm-oem"}, "family": "gfx1151", "bypass_tests_for_releases": true, "sanity_check_only_for_family": true, "build_variant_label": "release", "build_variant_suffix": "", "build_variant_cmake_preset": "", "artifact_group": "gfx1151"}]
OUTPUT linux_test_labels=[]
OUTPUT windows_variants=[]
OUTPUT windows_test_labels=[]
OUTPUT enable_build_jobs=true
OUTPUT test_type=smoke

with final test results in this PR run

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[HereThereBeDragons]

can we adjust the PR label from kernel:oem to test_runner:oem?

[HereThereBeDragons]

if this option is only for PRs, you dont need those here as you can get the pr label content from the github event

[geomin12]

I plan on leaving this option in case developers want to test without waiting for build!

[HereThereBeDragons]

before merging: can we have one ci run with it?

[geomin12]

before merging: can we have one ci run with it?

this run (https://github.com/ROCm/TheRock/actions/runs/21492661988/job/61927151095?pr=2902) tests using the oem label which is noted here: