[Feat] enable hierarchical communication for mc2 ops on A2 (#3015)

Currently, when in A2, setting the environment variables
`HCCL_INTRA_PCIE_ENABLE=1` and `HCCL_INTRA_ROCE_ENABLE=0` can reduce
cross-machine communication traffic and significantly improve
communication performance.

For more details, please refer to
[document](https://www.hiascend.com/document/detail/zh/Pytorch/710/apiref/torchnpuCustomsapi/context/torch_npu-npu_moe_distribute_dispatch_v2.md)

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: realliujiaxu <realliujiaxu@163.com>

vllm_ascend/torchair/ops/torchair_fused_moe.py

@@ -51,7 +51,8 @@ from vllm_ascend.torchair.utils import npu_stream_switch, npu_wait_tensor
 from vllm_ascend.utils import (AscendSocVersion, dispose_tensor,
                                get_all_reduce_merge_state,
                                get_ascend_soc_version,
-                               get_rm_router_logits_state, is_310p)
+                               get_rm_router_logits_state, is_310p,
+                               is_hierarchical_communication_enabled)
 
 
 def torchair_fused_experts_with_mc2(
@@ -78,6 +79,10 @@ def torchair_fused_experts_with_mc2(
 
     # NOTE: Currently, when in A3, we need to pass in some extra param into dispatch & combine
     a3_need_extra_args = get_ascend_soc_version() == AscendSocVersion.A3
+    # NOTE: When in A2, setting the environment variables HCCL_INTRA_PCIE_ENABLE=1 and
+    # HCCL_INTRA_ROCE_ENABLE=0 can reduce cross-machine communication traffic and significantly
+    # improve communication performance.
+    need_expert_scale = is_hierarchical_communication_enabled()
 
     enable_dispatch_v2 = hasattr(torch_npu, "npu_moe_distribute_dispatch_v2")
 
@@ -108,6 +113,10 @@ def torchair_fused_experts_with_mc2(
         stage1_kwargs.update({
             "x_active_mask": mc2_mask,
         })
+    if need_expert_scale:
+        stage1_kwargs.update({
+            "expert_scales": topk_weights.to(torch.float32),
+        })
 
     kwargs_mc2.update(stage1_kwargs)
 
@@ -116,8 +125,8 @@ def torchair_fused_experts_with_mc2(
     ) if enable_dispatch_v2 else torch_npu.npu_moe_distribute_dispatch(
         **kwargs_mc2)
     # comm_stream.wait_stream(torch.npu.current_stream())
-    expand_x, dynamic_scale, assist_info_for_combine, expert_token_nums, ep_recv_counts = output[
-        0:5]
+    expand_x, dynamic_scale, assist_info_for_combine, expert_token_nums, \
+        ep_recv_counts, _, expand_scales = output[0:7]
 
     if shared_experts is not None:
         with npu_stream_switch("moe_secondary", 0):
@@ -167,6 +176,7 @@ def torchair_fused_experts_with_mc2(
         "group_ep": moe_all_to_all_group_name,
         "ep_world_size": ep_world_size,
         "ep_rank_id": ep_rank_id,
+        "expand_scales": expand_scales,
     }
     if enable_dispatch_v2:
         stage3_kwargs.update({