[Kernel] Reclassify kernel tests by ops group + move helpers out of the package (RFC #29630) (#32128)

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

test/registered/kernels/ops/attention/test_c128_v2.py

@@ -7,7 +7,10 @@ import pytest
 import torch
 import triton
 
-from sglang.kernels.jit.tests.deepseek_v4.common import (
+from sglang.kernels.ops.attention.dsv4 import compress_forward
+from sglang.srt.utils import get_device
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.kernels.deepseek_v4.common import (
     LegacyContext,
     PagedContext,
     make_legacy_context,
@@ -15,9 +18,6 @@ from sglang.kernels.jit.tests.deepseek_v4.common import (
     make_state_pool,
     to_seq_extend,
 )
-from sglang.kernels.ops.attention.dsv4 import compress_forward
-from sglang.srt.utils import get_device
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=30, stage="base-b-kernel-unit", runner_config="1-gpu-large")
 register_amd_ci(est_time=30, suite="nightly-amd-kernel-1-gpu", nightly=True)

test/registered/kernels/ops/attention/test_c4_v2.py

@@ -7,7 +7,10 @@ import pytest
 import torch
 import triton
 
-from sglang.kernels.jit.tests.deepseek_v4.common import (
+from sglang.kernels.ops.attention.dsv4 import compress_forward
+from sglang.srt.utils import get_device
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.kernels.deepseek_v4.common import (
     LegacyContext,
     PagedContext,
     make_legacy_context,
@@ -15,9 +18,6 @@ from sglang.kernels.jit.tests.deepseek_v4.common import (
     make_state_pool,
     to_seq_extend,
 )
-from sglang.kernels.ops.attention.dsv4 import compress_forward
-from sglang.srt.utils import get_device
-from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 
 register_cuda_ci(est_time=30, stage="base-b-kernel-unit", runner_config="1-gpu-large")
 register_amd_ci(est_time=30, suite="nightly-amd-kernel-1-gpu", nightly=True)

test/registered/kernels/ops/attention/test_deepseek_v4_compress_state_runtime_shapes.py

@@ -89,16 +89,16 @@ from typing import Callable, Iterable, Literal, Optional
 import pytest
 import torch
 
-from sglang.kernels.jit.tests.deepseek_v4.common import (
-    make_legacy_context,
-    to_seq_extend,
-)
 from sglang.kernels.ops.attention.dsv4 import (
     CompressorDecodePlan,
     CompressorPrefillPlan,
     compress_forward,
 )
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.kernels.deepseek_v4.common import (
+    make_legacy_context,
+    to_seq_extend,
+)
 from sglang.utils import is_in_ci
 
 register_cuda_ci(est_time=30, stage="base-b-kernel-unit", runner_config="1-gpu-large")

test/registered/kernels/ops/communication/test_custom_all_reduce.py

@@ -28,7 +28,6 @@ import torch
 import torch.distributed as dist
 
 import sglang.srt.distributed.parallel_state as ps
-from sglang.kernels.jit.tests.utils import multigpu_pytest_main
 from sglang.kernels.jit.utils import cache_once, get_ci_test_range
 from sglang.kernels.ops.communication.all_reduce import (
     AllReduceAlgo,
@@ -39,6 +38,7 @@ from sglang.srt.distributed.device_communicators.custom_all_reduce_v2 import (
     CustomAllReduceV2,
 )
 from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.kernels.utils import multigpu_pytest_main
 
 register_cuda_ci(
     est_time=300,

test/registered/kernels/ops/communication/test_symm_mem_all_gather.py

@@ -26,13 +26,13 @@ import torch
 import torch.distributed as dist
 
 import sglang.srt.distributed.parallel_state as ps
-from sglang.kernels.jit.tests.utils import multigpu_pytest_main
 from sglang.kernels.jit.utils import cache_once, get_ci_test_range
 from sglang.srt.distributed.device_communicators.triton_symm_mem_ag import (
     all_gather_inner,
     create_state,
 )
 from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.kernels.utils import multigpu_pytest_main
 
 register_cuda_ci(est_time=240, stage="extra-b", runner_config="8-gpu-h200")
 # Nightly is not redundant here: it sets SGLANG_JIT_KERNEL_RUN_FULL_TESTS=1 to expand get_ci_test_range sweeps.

test/registered/kernels/ops/communication/test_tp_qknorm.py

@@ -14,7 +14,6 @@ import torch.distributed as dist
 import triton
 
 import sglang.srt.distributed.parallel_state as ps
-from sglang.kernels.jit.tests.utils import multigpu_pytest_main
 from sglang.kernels.jit.utils import cache_once
 from sglang.kernels.ops.communication.all_reduce import (
     fused_parallel_qknorm,
@@ -26,6 +25,7 @@ from sglang.srt.distributed.device_communicators.custom_all_reduce_v2 import (
     CustomAllReduceV2,
 )
 from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.kernels.utils import multigpu_pytest_main
 
 register_cuda_ci(
     est_time=300,