在多卡训练过程中出现cudaStreamSynchronize misaligned address错误
Created by: yeyupiaoling
- 系统:Ubuntu 16.04
- 显卡:2张 2080ti
- 版本:PaddlePaddle-GPU 1.8.3.post107
F0823 14:54:47.975309 16507 all_reduce_op_handle.cc:192] cudaStreamSynchronize misaligned address
*** Check failure stack trace: ***
@ 0x7ff2ccc25d4d google::LogMessage::Fail()
@ 0x7ff2ccc297fc google::LogMessage::SendToLog()
@ 0x7ff2ccc25873 google::LogMessage::Flush()
@ 0x7ff2ccc2ad0e google::LogMessageFatal::~LogMessageFatal()
@ 0x7ff2cfedb35c paddle::framework::details::AllReduceOpHandle::SyncNCCLAllReduce()
@ 0x7ff2cfedb3f4 paddle::framework::details::AllReduceOpHandle::NCCLAllReduceFunc()
@ 0x7ff2cfedbf66 paddle::framework::details::AllReduceOpHandle::AllReduceFunc()
@ 0x7ff2cfe3bf1c paddle::framework::details::FusedAllReduceOpHandle::FusedAllReduceFunc()
@ 0x7ff2cfe3d4fe paddle::framework::details::FusedAllReduceOpHandle::RunImpl()
@ 0x7ff2cfe77c31 paddle::framework::details::FastThreadedSSAGraphExecutor::RunOpSync()
@ 0x7ff2cfe7572f paddle::framework::details::FastThreadedSSAGraphExecutor::RunOp()
@ 0x7ff2cfe759f4 _ZNSt17_Function_handlerIFvvESt17reference_wrapperISt12_Bind_simpleIFS1_ISt5_BindIFZN6paddle9framework7details28FastThreadedSSAGraphExecutor10RunOpAsyncEPSt13unordered_mapIPNS6_12OpHandleBaseESt6atomicIiESt4hashISA_ESt8equal_toISA_ESaISt4pairIKSA_SC_EEESA_RKSt10shared_ptrINS5_13BlockingQueueImEEEEUlvE_vEEEvEEEE9_M_invokeERKSt9_Any_data
@ 0x7ff2ccc83213 std::_Function_handler<>::_M_invoke()
@ 0x7ff2cca7ee67 std::__future_base::_State_base::_M_do_set()
@ 0x7ff3105d8a99 __pthread_once_slow
@ 0x7ff2cfe71bc2 _ZNSt13__future_base11_Task_stateISt5_BindIFZN6paddle9framework7details28FastThreadedSSAGraphExecutor10RunOpAsyncEPSt13unordered_mapIPNS4_12OpHandleBaseESt6atomicIiESt4hashIS8_ESt8equal_toIS8_ESaISt4pairIKS8_SA_EEES8_RKSt10shared_ptrINS3_13BlockingQueueImEEEEUlvE_vEESaIiEFvvEE6_M_runEv
@ 0x7ff2cca812c4 _ZZN10ThreadPoolC1EmENKUlvE_clEv
@ 0x7ff3011d7421 execute_native_thread_routine_compat
@ 0x7ff3105d16ba start_thread
@ 0x7ff30fbf74dd clone
@ (nil) (unknown)