Fix compilation

csrc/kernels/internode_ll.cu

@@ -501,7 +501,7 @@ __forceinline__ __device__ void logfmt_check_amaxmin(uint8_t* meta_buffer, float
     }
 
     const auto& casted = warp_reduce_and<kNumSendUnrolls>(enable_cast) ? 1u << (lane_id / kNumRecvUnrolls): 0u;
-    const auto& num_casted_prefix = std::__popcount(warp_reduce_or<kNumRecvUnrolls, true>(casted) & ((1u << (lane_id / kNumRecvUnrolls)) - 1));
+    const auto& num_casted_prefix = __popc(warp_reduce_or<kNumRecvUnrolls, true>(casted) & ((1u << (lane_id / kNumRecvUnrolls)) - 1));
 
     if (lane_id < kNumLanes and lane_id % kNumRecvUnrolls == 0)
         shared_cast_info[lane_id / kNumRecvUnrolls] = (num_casted_prefix << 1) | (casted ? 1u : 0u);