[Performance] Use forward_native for Conv3dLayer and add UT (#8375)

What this PR does / why we need it?
switch Ascend conv3d forward_oot to use forward_native and  add ut

Does this PR introduce any user-facing change?
No

How was this patch tested?
by CI

---------

Signed-off-by: zouyizhou <zouyizhou@huawei.com>

tests/ut/_310p/ops/test_conv_310.py

@@ -0,0 +1,35 @@
+from unittest.mock import MagicMock, patch
+
+import pytest
+import torch
+from vllm.config import set_current_vllm_config
+from vllm.model_executor.layers.conv import Conv3dLayer
+
+from vllm_ascend._310p.ops.conv import AscendConv3dLayer310
+
+
+@pytest.fixture(autouse=True)
+def default_vllm_config():
+    mock_config = MagicMock()
+    mock_config.compilation_config.custom_ops = ["all"]
+    with set_current_vllm_config(mock_config):
+        yield mock_config
+
+
+def test_conv3d_310_forward_oot_uses_forward_native():
+    layer = AscendConv3dLayer310(
+        in_channels=2,
+        out_channels=4,
+        kernel_size=(2, 2, 2),
+        stride=(2, 2, 2),
+        bias=True,
+        params_dtype=torch.float32,
+    )
+    x = torch.randn(1, 2, 4, 4, 4, dtype=torch.float32)
+    expected = torch.randn(1, 4, 2, 2, 2, dtype=torch.float32)
+
+    with patch.object(Conv3dLayer, "forward_native", autospec=True, return_value=expected) as mock_forward_native:
+        out = layer.forward_oot(x)
+
+    mock_forward_native.assert_called_once_with(layer, x)
+    assert out is expected

vllm_ascend/_310p/ops/conv.py

@@ -0,0 +1,27 @@
+#
+# Copyright (c) 2026 Huawei Technologies Co., Ltd. All Rights Reserved.
+# This file is a part of the vllm-ascend project.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+
+import torch
+
+from vllm_ascend.ops.conv import AscendConv3dLayer
+
+
+class AscendConv3dLayer310(AscendConv3dLayer):
+    def forward_oot(self, x: torch.Tensor) -> torch.Tensor:
+        # 310P should avoid the aclnn BatchMatMulV2 Conv3D path used by
+        # AscendConv3dLayer and keep vLLM's native Conv3d dispatch behavior.
+        return super().forward_native(x)

vllm_ascend/utils.py

@@ -662,6 +662,7 @@ def register_ascend_customop(vllm_config: VllmConfig | None = None):
     if is_310p():
         from vllm_ascend._310p.fused_moe.fused_moe import AscendFusedMoE310, AscendSharedFusedMoE310
         from vllm_ascend._310p.ops.activation import AscendSiluAndMul310
+        from vllm_ascend._310p.ops.conv import AscendConv3dLayer310
         from vllm_ascend._310p.ops.layernorm import (
             AscendGemmaRMSNorm310,
             AscendRMSNorm310,
@@ -686,6 +687,7 @@ def register_ascend_customop(vllm_config: VllmConfig | None = None):
                 "ParallelLMHead": AscendParallelLMHead310,
                 "VocabParallelEmbedding": AscendVocabParallelEmbedding310,
                 "MMEncoderAttention": AscendMMEncoderAttention310,
+                "Conv3dLayer": AscendConv3dLayer310,
             }
         )