Add Arm64 CPU Phase 1A CI bootstrap (#22123)

Co-authored-by: Ma Mingfei <mingfei.ma@intel.com>

test/srt/cpu/test_server_args_backend.py

@@ -0,0 +1,35 @@
+import unittest
+from unittest.mock import patch
+
+from sglang.srt.server_args import ServerArgs
+
+
+class TestServerArgsCPUBackend(unittest.TestCase):
+    def _make_server_args(self, attention_backend=None):
+        server_args = ServerArgs.__new__(ServerArgs)
+        server_args.device = "cpu"
+        server_args.attention_backend = attention_backend
+        server_args.sampling_backend = None
+        return server_args
+
+    @patch("sglang.srt.server_args.is_host_cpu_arm64", return_value=True)
+    def test_arm_cpu_defaults_to_torch_native(self, _mock_is_arm64):
+        server_args = self._make_server_args()
+
+        ServerArgs._handle_cpu_backends(server_args)
+
+        self.assertEqual(server_args.attention_backend, "torch_native")
+        self.assertEqual(server_args.sampling_backend, "pytorch")
+
+    @patch("sglang.srt.server_args.is_host_cpu_arm64", return_value=False)
+    def test_x86_cpu_defaults_to_intel_amx(self, _mock_is_arm64):
+        server_args = self._make_server_args()
+
+        ServerArgs._handle_cpu_backends(server_args)
+
+        self.assertEqual(server_args.attention_backend, "intel_amx")
+        self.assertEqual(server_args.sampling_backend, "pytorch")
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/run_suite.py

@@ -44,6 +44,21 @@ suite_amd = {
     # by test/run_suite.py using the registry system.
 }
 
+# Keep the Arm64 bootstrap suite limited to hosted-runner-safe unit kernels.
+# `test_extend.py`, `test_mamba.py`, and `test_mla.py` still hit the
+# x86-specific BF16 BRGEMM/VNNI path on Arm and need dedicated fallbacks.
+suite_arm64 = {
+    "per-commit-cpu-arm64": [
+        TestFile("cpu/test_activation.py"),
+        TestFile("cpu/test_decode.py"),
+        TestFile("cpu/test_norm.py"),
+        TestFile("cpu/test_qwen3.py"),
+        TestFile("cpu/test_rope.py"),
+        TestFile("cpu/test_server_args_backend.py"),
+        TestFile("cpu/test_topk.py"),
+    ],
+}
+
 # Add Intel Xeon tests
 suite_xeon = {
     "per-commit-cpu": [
@@ -66,6 +81,7 @@ suite_xeon = {
         TestFile("cpu/test_qkv_proj_with_rope.py"),
         TestFile("cpu/test_qwen3.py"),
         TestFile("cpu/test_rope.py"),
+        TestFile("cpu/test_server_args_backend.py"),
         TestFile("cpu/test_shared_expert.py"),
         TestFile("cpu/test_topk.py"),
     ],
@@ -83,6 +99,7 @@ suite_xpu = {
 }
 
 suites.update(suite_amd)
+suites.update(suite_arm64)
 suites.update(suite_xeon)
 suites.update(suite_xpu)