Support Qwen3 MoE context parallel (#18233)

Co-authored-by: Shunkang <182541032+Shunkangz@users.noreply.github.co>
Co-authored-by: Jiying Dong <87510204+dongjiyingdjy@users.noreply.github.com>

python/sglang/srt/managers/schedule_policy.py

@@ -35,6 +35,7 @@ import torch
 
 from sglang.srt.dllm.config import DllmConfig
 from sglang.srt.layers.attention.nsa.utils import is_nsa_prefill_cp_in_seq_split
+from sglang.srt.layers.utils.cp_utils import is_prefill_context_parallel_enabled
 from sglang.srt.managers.schedule_batch import Req, ScheduleBatch
 from sglang.srt.mem_cache.base_prefix_cache import (
     BasePrefixCache,
@@ -432,6 +433,7 @@ class PrefillAdder:
         )
         self.nsa_prefill_cp_in_seq_split = is_nsa_prefill_cp_in_seq_split()
         self.max_running_requests = max_running_requests
+        self.prefill_context_parallel_enabled = is_prefill_context_parallel_enabled()
         self.prefill_max_requests = prefill_max_requests
         self.prefill_delayer_single_pass = prefill_delayer_single_pass
         self.max_prefill_bs = max_prefill_bs
@@ -738,7 +740,9 @@ class PrefillAdder:
         # TODO support cp with multiple requests
         # Enabling context parallelism currently presents precision issues;
         # therefore, the prefill-batch setting is temporarily set to 1.
-        if self.nsa_prefill_cp_in_seq_split and len(self.can_run_list) >= 1:
+        if (
+            self.nsa_prefill_cp_in_seq_split or self.prefill_context_parallel_enabled
+        ) and len(self.can_run_list) >= 1:
             return AddReqResult.OTHER
 
         if (x := self.prefill_max_requests) is not None and len(self.can_run_list) >= x: