Fix notify_dispatch: using warp 0 to issue send

Signed-off-by: wzc.wuzhicheng <wzc.wuzhicheng@linux.alibaba.com>

csrc/kernels/internode.cu

@@ -254,10 +254,12 @@ notify_dispatch(const int* num_tokens_per_rank, int* moe_recv_counter_mapped, in
         // TODO: overlap EP barrier and NVL cleaning
         for (int i = 0; i < kNumRDMARanks; ++i) {
             if (i != rdma_rank) {
-                nvshmemi_ibgda_put_nbi_warp<true>(reinterpret_cast<uint64_t>(rdma_recv_num_tokens_mixed.recv_buffer(rdma_rank)), 
-                                                reinterpret_cast<uint64_t>(rdma_recv_num_tokens_mixed.send_buffer(i)),
-                                                (NUM_MAX_NVL_PEERS + num_rdma_experts + 1) * sizeof(int),
-                                                translate_dst_rdma_rank<kLowLatencyMode>(i, nvl_rank), 0, lane_id, 0);
+                if (warp_id == 0) {
+                    nvshmemi_ibgda_put_nbi_warp<true>(reinterpret_cast<uint64_t>(rdma_recv_num_tokens_mixed.recv_buffer(rdma_rank)),
+                                                    reinterpret_cast<uint64_t>(rdma_recv_num_tokens_mixed.send_buffer(i)),
+                                                    (NUM_MAX_NVL_PEERS + num_rdma_experts + 1) * sizeof(int),
+                                                    translate_dst_rdma_rank<kLowLatencyMode>(i, nvl_rank), 0, lane_id, 0);
+                }
             } else { 
                 UNROLLED_WARP_COPY(1, lane_id, NUM_MAX_NVL_PEERS + num_rdma_experts + 1, 
                                     rdma_recv_num_tokens_mixed.recv_buffer(rdma_rank),