[Kernel] Migrate top-level srt/layers stray kernels to sglang.kernels (RFC #29630, Phase 2.5, 3/7) (#30787)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

test/manual/layers/test_fused_gate_sigmoid_mul_add.py

@@ -3,7 +3,7 @@ import itertools
 import pytest
 import torch
 
-from sglang.srt.layers.elementwise import fused_gate_sigmoid_mul_add
+from sglang.kernels.ops.layernorm.elementwise import fused_gate_sigmoid_mul_add
 
 DTYPES = [torch.float16, torch.bfloat16]
 TOKEN_COUNTS = [1, 2, 4, 8, 16, 64, 512, 1024, 2048, 4096, 8192]

test/manual/layers/test_fused_sigmoid_mul.py

@@ -3,7 +3,7 @@ import itertools
 import pytest
 import torch
 
-from sglang.srt.layers.elementwise import fused_sigmoid_mul
+from sglang.kernels.ops.layernorm.elementwise import fused_sigmoid_mul
 
 DTYPES = [torch.float16, torch.bfloat16]
 TOKEN_COUNTS = [1, 2, 4, 8, 16, 64, 512, 1024, 2048, 4096, 8192]

test/registered/jit/deepseek_v4/test_fp4_indexer.py

@@ -11,14 +11,14 @@ from sglang.jit_kernel.dsv4 import (
     fused_q_indexer_rope_hadamard_fp4_quant,
 )
 from sglang.jit_kernel.hadamard import hadamard_transform
+from sglang.kernels.ops.attention.deepseek_v4_rope import (
+    apply_rotary_emb_triton,
+    precompute_freqs_cis,
+)
 from sglang.srt.layers.attention.dsv4.fp4_indexer import (
     quantize_fp4_indexer_tensor,
     store_fp4_index_k_cache,
 )
-from sglang.srt.layers.deepseek_v4_rope import (
-    apply_rotary_emb_triton,
-    precompute_freqs_cis,
-)
 from sglang.test.ci.ci_register import register_cuda_ci
 
 register_cuda_ci(est_time=60, stage="base-b-kernel-unit", runner_config="1-gpu-large")

test/registered/kernels/test_gemma4_fused_routing.py

@@ -26,7 +26,7 @@ pytestmark = pytest.mark.skipif(
 
 @pytest.fixture(scope="module")
 def fused_routing():
-    from sglang.srt.layers.gemma4_fused_ops import gemma4_fused_routing
+    from sglang.kernels.ops.layernorm.gemma4_fused_ops import gemma4_fused_routing
 
     return gemma4_fused_routing
 

test/registered/kernels/test_mhc_kernels.py

@@ -1,8 +1,8 @@
 import pytest
 import torch
 
-import sglang.srt.layers.mhc as mhc
-from sglang.srt.layers.mhc import mhc_fused_post_pre, mhc_post, mhc_pre
+import sglang.kernels.ops.layernorm.mhc as mhc
+from sglang.kernels.ops.layernorm.mhc import mhc_fused_post_pre, mhc_post, mhc_pre
 from sglang.test.ci.ci_register import register_cuda_ci
 
 register_cuda_ci(est_time=30, stage="base-b", runner_config="1-gpu-large")