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

.github/workflows/ci.yml

@@ -43,6 +43,10 @@ on:
       windows_use_prebuilt_artifacts:
         type: boolean
         description: "If enabled, the CI will pull Windows artifacts using artifact_run_id and only run tests"
+      additional_label_options:
+        type: string
+        description: "If enabled, these additional labels options will be applied. ex: kernel:oem"
+        default: ""
       artifact_run_id:
         type: string
         description: "If provided, the tests will run on this artifact ID"

.github/workflows/setup.yml

@@ -75,6 +75,7 @@ jobs:
           INPUT_WINDOWS_AMDGPU_FAMILIES: ${{ github.event.inputs.windows_amdgpu_families }}
           WINDOWS_TEST_LABELS: ${{ github.event.inputs.windows_test_labels }}
           WINDOWS_USE_PREBUILT_ARTIFACTS: ${{ github.event.inputs.windows_use_prebuilt_artifacts }}
+          ADDITIONAL_LABEL_OPTIONS: ${{ github.event.inputs.additional_label_options }}
           BUILD_VARIANT: ${{ inputs.build_variant }}
           MULTI_ARCH: ${{ inputs.multi_arch }}
           # Variable comes from ROCm organization variable 'ROCM_THEROCK_TEST_RUNNERS'

build_tools/github_actions/amdgpu_family_matrix.py

@@ -77,6 +77,9 @@
     "gfx1151": {
         "linux": {
             "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,
             "build_variants": ["release"],

build_tools/github_actions/configure_ci.py

@@ -224,13 +224,27 @@ def should_ci_run_given_modified_paths(paths: Optional[Iterable[str]]) -> bool:
 
 def get_pr_labels(args) -> List[str]:
     """Gets a list of labels applied to a pull request."""
-    data = json.loads(args.get("pr_labels"))
+    data = json.loads(args.get("pr_labels", "{}"))
     labels = []
     for label in data.get("labels", []):
         labels.append(label["name"])
     return labels
 
 
+def get_workflow_dispatch_additional_label_options(args) -> List[str]:
+    """Gets a list of additional label options from workflow_dispatch."""
+    additional_label_options = args.get(
+        "workflow_dispatch_additional_label_options", ""
+    )
+    if additional_label_options:
+        return [
+            label.strip()
+            for label in additional_label_options.split(",")
+            if label.strip()
+        ]
+    return []
+
+
 def filter_known_names(
     requested_names: List[str], name_type: str, target_matrix=None
 ) -> List[str]:
@@ -608,6 +622,33 @@ def matrix_generator(
                     artifact_group += f"-{build_variant_suffix}"
                 matrix_row["artifact_group"] = artifact_group
 
+                # We retrieve labels from both PR and workflow_dispatch to customize the build and test jobs
+                label_options = []
+                label_options.extend(get_pr_labels(base_args))
+                label_options.extend(
+                    get_workflow_dispatch_additional_label_options(base_args)
+                )
+                for label in label_options:
+                    # If a specific test kernel type was specified, we use that kernel-enabled test runners
+                    # We disable the other machines that do not have the specified kernel type
+                    # If a kernel test label was added, we set the test-runs-on accordingly to kernel-specific test machines
+                    if "kernel" in label:
+                        _, kernel_type = label.split(":")
+                        # If the architecture has a valid kernel machine, we set it here
+                        if (
+                            "test-runs-on-kernel" in platform_info
+                            and kernel_type in platform_info["test-runs-on-kernel"]
+                        ):
+                            matrix_row["test-runs-on"] = platform_info[
+                                "test-runs-on-kernel"
+                            ][kernel_type]
+                        # Otherwise, we disable the test runner for this architecture
+                        else:
+                            matrix_row["test-runs-on"] = ""
+                            if "test-runs-on-multi-gpu" in platform_info:
+                                matrix_row["test-runs-on-multi-gpu"] = ""
+                        break
+
                 matrix_output.append(matrix_row)
 
     print(f"Generated build matrix: {str(matrix_output)}")
@@ -766,6 +807,9 @@ def format_variants(variants):
     base_args["workflow_dispatch_windows_test_labels"] = os.getenv(
         "WINDOWS_TEST_LABELS", ""
     )
+    base_args["workflow_dispatch_additional_label_options"] = os.getenv(
+        "ADDITIONAL_LABEL_OPTIONS", ""
+    )
     base_args["build_variant"] = os.getenv("BUILD_VARIANT", "release")
     base_args["multi_arch"] = os.environ.get("MULTI_ARCH", "false") == "true"
 

build_tools/github_actions/tests/configure_ci_test.py

@@ -293,6 +293,26 @@ def test_invalid_test_label_linux_pull_request_matrix_generator(self):
         )
         self.assertEqual(linux_test_labels, [])
 
+    def test_kernel_test_label_linux_pull_request_matrix_generator(self):
+        base_args = {
+            "pr_labels": '{"labels":[{"name":"kernel:oem"}]}',
+            "build_variant": "release",
+        }
+        linux_target_output, linux_test_labels = configure_ci.matrix_generator(
+            is_pull_request=True,
+            is_workflow_dispatch=False,
+            is_push=False,
+            is_schedule=False,
+            base_args=base_args,
+            families={},
+            platform="linux",
+        )
+        self.assertGreaterEqual(len(linux_target_output), 1)
+        self.assert_target_output_is_valid(
+            target_output=linux_target_output, allow_xfail=False
+        )
+        self.assertEqual(linux_test_labels, [])
+
     def test_main_linux_branch_push_matrix_generator(self):
         base_args = {"branch_name": "main", "build_variant": "release"}
         linux_target_output, linux_test_labels = configure_ci.matrix_generator(

docs/development/ci_behavior_manipulation.md

@@ -16,6 +16,7 @@ However, if additional options are wanted, you can add a label to manipulate the
 - `run-all-archs-ci`: The CI will build all possible architectures
 - `gfx...`: A build and test (if a test machine is available) is added to the CI matrix for the specified gfx family. (ex: `gfx120X`, `gfx950`)
 - `test:...`: The full test will run only for the specified label and other labeled projects (ex: `test:rocthrust`, `test:hipblaslt`)
+- `kernel:...`: The CI will run tests on only kernel-specific enabled test machines (ex: `kernel:oem`)
 
 ## Workflow dispatch behavior