[AMD] ci: add label-gated extra-a tier (kv_canary + mock_model unit tests) (#27822)

test/registered/kv_canary/test_self_unit_buffer_alloc.py

@@ -12,10 +12,11 @@ from sglang.srt.kv_canary.pool_patcher.buffer_alloc import (
     make_row_source,
     resolve_real_kv_read_bytes,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=10, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=10, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _config(mode: RealKvHashMode) -> CanaryConfig:

test/registered/kv_canary/test_self_unit_endpoint.py

@@ -21,11 +21,12 @@ from sglang.srt.kv_canary.expected_inputs import ExpectedInputs
 from sglang.srt.kv_canary.state import (
     ViolationLog,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import DEFAULT_DEVICE
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=20, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=20, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _make_endpoint(*, device, kernel_kind=CanaryLaunchTag.HEAD_K_FULL, swa_lut=None):

test/registered/kv_canary/test_self_unit_future_tensor.py

@@ -6,10 +6,11 @@ from typing import cast
 import torch
 
 from sglang.srt.kv_canary.runner.future_tensor import FutureTensors
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=20, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=20, suite="extra-a-test-1-gpu-small-amd")
 
 
 class _FakeEvent:

test/registered/kv_canary/test_self_unit_perturb.py

@@ -27,7 +27,7 @@ from sglang.srt.kv_canary.perturb.utils import (
     flip_first_byte_in_source,
     pick_target_group,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import (
     DEFAULT_DEVICE,
     make_buffer_group,
@@ -41,6 +41,7 @@ if TYPE_CHECKING:
     from sglang.srt.mem_cache.base_prefix_cache import BasePrefixCache
 
 register_cuda_ci(est_time=10, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=10, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestParseTargetGroupKind(CustomTestCase):

test/registered/kv_canary/test_self_unit_plan_input.py

@@ -6,7 +6,7 @@ from types import SimpleNamespace
 import torch
 
 from sglang.srt.kv_canary.plan_input import PlanInput
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import (
     DEFAULT_DEVICE,
     make_forward_batch,
@@ -14,6 +14,7 @@ from sglang.test.kv_canary.fixtures import (
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=30, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=30, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _make_static_plan_input(*, bs_capacity: int, device) -> PlanInput:

test/registered/kv_canary/test_self_unit_pool_patcher.py

@@ -15,7 +15,7 @@ from sglang.jit_kernel.kv_canary.verify import (
 )
 from sglang.srt.kv_canary.buffer_group import PoolKind
 from sglang.srt.kv_canary.pool_patcher.api import attach_canary_buffers
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import (
     DEFAULT_DEVICE,
     make_base_config,
@@ -25,6 +25,7 @@ from sglang.test.kv_canary.fixtures import (
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=45, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=45, suite="extra-a-test-1-gpu-small-amd")
 
 
 class PoolPatcherHelper:

test/registered/kv_canary/test_self_unit_pool_patcher_utils.py

@@ -3,10 +3,11 @@ from __future__ import annotations
 import unittest
 
 from sglang.srt.kv_canary.pool_patcher.utils import wrap_method
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=10, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=10, suite="extra-a-test-1-gpu-small-amd")
 
 
 class _FakeObj:

test/registered/kv_canary/test_self_unit_radix_walker.py

@@ -6,11 +6,12 @@ import torch
 
 from sglang.srt.kv_canary.radix_cache_walker import walk_radix_cache_for_canary
 from sglang.srt.mem_cache.swa_radix_cache import SWARadixCache, TreeNode
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import DEFAULT_DEVICE, make_radix_cache
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=30, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=30, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestSelfUnitRadixWalker(CustomTestCase):

test/registered/kv_canary/test_self_unit_req_to_expected_token_ids_manager.py

@@ -10,11 +10,12 @@ from sglang.srt.kv_canary.req_to_expected_token_ids_manager import (
     compute_req_all_ids_info,
     populate_req_to_expected_token_ids,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import DEFAULT_DEVICE, make_forward_batch
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=15, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=15, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _make_req(*, origin: list[int], output: list[int]) -> SimpleNamespace:

test/registered/kv_canary/test_self_unit_runner_health.py

@@ -11,7 +11,7 @@ from sglang.srt.kv_canary.config import CanaryConfig
 from sglang.srt.kv_canary.runner import stats_logger as stats_logger_module
 from sglang.srt.kv_canary.runner.health_checker import KernelRunCounterHealthChecker
 from sglang.srt.kv_canary.state import CanaryDeviceState
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.runner_test_base import (
     CanaryManagerTestCase,
     make_config,
@@ -20,6 +20,7 @@ from sglang.test.kv_canary.runner_test_base import (
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=45, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=45, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestSelfUnitManagerHealth(CanaryManagerTestCase):

test/registered/kv_canary/test_self_unit_runner_per_forward.py

@@ -12,7 +12,7 @@ from sglang.srt.kv_canary import endpoint as endpoint_module
 from sglang.srt.kv_canary.expected_inputs import ExpectedInputs
 from sglang.srt.kv_canary.runner import kernel_launcher as kernel_launcher_module
 from sglang.srt.kv_canary.state import ViolationLog
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import make_buffer_group, make_forward_batch
 from sglang.test.kv_canary.runner_test_base import (
     CanaryManagerTestCase,
@@ -21,6 +21,7 @@ from sglang.test.kv_canary.runner_test_base import (
 )
 
 register_cuda_ci(est_time=45, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=45, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestManagerPerForward(CanaryManagerTestCase):

test/registered/kv_canary/test_self_unit_runner_swa_divergence.py

@@ -15,12 +15,13 @@ from sglang.srt.kv_canary.runner.swa_divergence import (
     SwaDivergenceReporter,
     compute_swa_full_idx_divergence,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import make_buffer_group
 from sglang.test.kv_canary.runner_test_base import CanaryManagerTestCase, make_manager
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=45, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=45, suite="extra-a-test-1-gpu-small-amd")
 
 _DEVICE = torch.device("cuda")
 

test/registered/kv_canary/test_self_unit_runner_sweep.py

@@ -4,7 +4,7 @@ import unittest
 from unittest.mock import patch
 
 from sglang.srt.kv_canary import endpoint as endpoint_module
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import (
     make_forward_batch,
     make_radix_cache,
@@ -17,6 +17,7 @@ from sglang.test.kv_canary.runner_test_base import (
 )
 
 register_cuda_ci(est_time=45, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=45, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _run_one_cycle(manager, forward_batch) -> None:

test/registered/kv_canary/test_self_unit_sweep_plan_builder.py

@@ -5,7 +5,7 @@ import unittest
 import torch
 
 from sglang.srt.kv_canary.sweep_plan_builder import build_verify_plan_radix_sweep
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import (
     DEFAULT_DEVICE,
     make_radix_cache,
@@ -14,6 +14,7 @@ from sglang.test.kv_canary.fixtures import (
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=30, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=30, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestSelfUnitSweepPlanBuilder(CustomTestCase):

test/registered/kv_canary/test_self_unit_token_oracle.py

@@ -9,11 +9,12 @@ from sglang.srt.kv_canary.expected_inputs import ExpectedInputs
 from sglang.srt.kv_canary.token_oracle.oracle import HashOracle
 from sglang.srt.kv_canary.token_oracle.oracle_manager import TokenOracleManager
 from sglang.srt.model_executor.forward_batch_info import ForwardMode
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kv_canary.fixtures import DEFAULT_DEVICE
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=1, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=1, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestTokenOracleManager(CustomTestCase):

test/registered/kv_canary/test_self_unit_violation.py

@@ -15,10 +15,11 @@ from sglang.srt.kv_canary.runner.violation_reporter import (
     ViolationReporter,
     _format_violation,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=5, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=5, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _make_row(

test/registered/mock_model/test_self_unit_canary_mock_wiring.py

@@ -12,11 +12,12 @@ from sglang.srt.model_executor.forward_batch_info import (
     ForwardMode,
     _stable_hash_str_to_i64,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.mock_model.utils import mock_model_server_args, mock_model_server_env
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=60, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=60, suite="extra-a-test-1-gpu-small-amd")
 
 
 @dataclasses.dataclass

test/registered/mock_model/test_self_unit_install.py

@@ -9,10 +9,11 @@ os.environ["SGLANG_KV_CANARY_ENABLE_TOKEN_ORACLE"] = "1"
 from sglang.srt.kv_canary.token_oracle.install import install_token_oracle_from_env
 from sglang.srt.kv_canary.token_oracle.oracle import HashOracle
 from sglang.srt.layers.sampler import _CUSTOM_SAMPLER_FACTORIES
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=60, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=60, suite="extra-a-test-1-gpu-small-amd")
 
 
 def _make_server_args(*, sampling_backend: str) -> SimpleNamespace:

test/registered/mock_model/test_self_unit_oracle.py

@@ -10,10 +10,11 @@ from sglang.srt.kv_canary.token_oracle.oracle import (
     HashOracle,
     _splitmix64_tensor,
 )
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=60, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=60, suite="extra-a-test-1-gpu-small-amd")
 
 
 _U64_MASK: int = (1 << 64) - 1

test/registered/mock_model/test_self_unit_oracle_torch_vs_ref.py

@@ -7,10 +7,11 @@ import torch
 
 from sglang.jit_kernel.kv_canary.verify_ref import splitmix64
 from sglang.srt.kv_canary.token_oracle.oracle import HashOracle
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=30, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=30, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestHashOracleTorchVsRef(CustomTestCase):

test/registered/mock_model/test_self_unit_sampler_hookpoint.py

@@ -18,10 +18,11 @@ from sglang.srt.kv_canary.token_oracle.oracle import HashOracle
 from sglang.srt.kv_canary.token_oracle.sampler import install_oracle_sampler
 from sglang.srt.layers.sampler import _CUSTOM_SAMPLER_FACTORIES
 from sglang.srt.server_args import SAMPLING_BACKEND_CHOICES
-from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_cuda_ci(est_time=60, stage="extra-a", runner_config="1-gpu-small")
+register_amd_ci(est_time=60, suite="extra-a-test-1-gpu-small-amd")
 
 
 class TestInstallOracleSampler(CustomTestCase):

test/run_suite.py

@@ -43,6 +43,14 @@ PER_COMMIT_SUITES = {
         "stage-c-test-4-gpu-amd",
         "stage-c-test-large-8-gpu-amd",
         "stage-c-test-large-8-gpu-amd-mi35x",
+        # extra-a: label-gated PR opt-in suite in pr-test-amd-extra.yml
+        # (mirror of CUDA extra-a; tests stay tagged per-commit but only
+        # dispatch when the PR carries the `run-ci-extra` label). Only the
+        # 1-gpu-small mock-model / kv_canary *unit* tests are onboarded so
+        # far; the canary *e2e* tests (1-/2-gpu-large) need the canary JIT
+        # kernel ported to ROCm first, so those suites are intentionally
+        # not yet registered for AMD.
+        "extra-a-test-1-gpu-small-amd",
     ],
     HWBackend.MUSA: [],
     HWBackend.CUDA: [