Created by: HugoLian

升级之后GPU运行原有任务（只是每天训练数据的差异，理论上不会有数据差异），报错：

I0718 14:57:56.227458  1672 Util.cpp:166] commandline:  --use_gpu=True --trainer_count=1 
I0718 14:58:02.015449  1672 GradientMachine.cpp:94] Initing parameters..
I0718 14:58:05.849926  1672 GradientMachine.cpp:101] Init parameters done.
F0718 14:58:24.706127  1672 Matrix.cpp:653] Not supported
*** Check failure stack trace: ***
    @     0x7f9fe05c868d  google::LogMessage::Fail()
    @     0x7f9fe05cc13c  google::LogMessage::SendToLog()
    @     0x7f9fe05c81b3  google::LogMessage::Flush()
    @     0x7f9fe05cd64e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f9fe03df2c6  paddle::GpuMatrix::mul()
    @     0x7f9fe02e3235  paddle::FullyConnectedLayer::forward()
    @     0x7f9fe01da79d  paddle::NeuralNetwork::forward()
    @     0x7f9fe01b1ee3  paddle::GradientMachine::forwardBackward()
    @     0x7f9fe05a44e4  GradientMachine::forwardBackward()
    @     0x7f9fe013cff9  _wrap_GradientMachine_forwardBackward
    @     0x7fa0173053d3  PyEval_EvalFrameEx
    @     0x7fa017307160  PyEval_EvalCodeEx
    @     0x7fa0173054d1  PyEval_EvalFrameEx
    @     0x7fa017307160  PyEval_EvalCodeEx
    @     0x7fa0173054d1  PyEval_EvalFrameEx
    @     0x7fa017307160  PyEval_EvalCodeEx
    @     0x7fa0173054d1  PyEval_EvalFrameEx
    @     0x7fa017307160  PyEval_EvalCodeEx
    @     0x7fa0173054d1  PyEval_EvalFrameEx
    @     0x7fa017307160  PyEval_EvalCodeEx
    @     0x7fa017307272  PyEval_EvalCode
    @     0x7fa01732165c  run_mod
    @     0x7fa017321730  PyRun_FileExFlags
    @     0x7fa017322c3c  PyRun_SimpleFileExFlags
    @     0x7fa0173344fc  Py_Main
    @       0x38bfc21b45  (unknown)
Thread [140325559519040] Forwarding __fc_layer_0__, 
*** Aborted at 1531897104 (unix time) try "date -d @1531897104" if you are using GNU date ***
PC: @                0x0 (unknown)
*** SIGABRT (@0x1f400000688) received by PID 1672 (TID 0x7fa01720c740) from PID 1672; stack trace: ***
    @       0x38c040f130 (unknown)
    @       0x38bfc359d9 (unknown)
    @       0x38bfc370e8 (unknown)
    @     0x7f9fe05d2bcb google::FindSymbol()
    @     0x7f9fe05d358a google::GetSymbolFromObjectFile()
    @     0x7f9fe05d3c52 google::SymbolizeAndDemangle()
    @     0x7f9fe05d1458 google::DumpStackTrace()
    @     0x7f9fe05d1516 google::DumpStackTraceAndExit()
    @     0x7f9fe05c868d google::LogMessage::Fail()
    @     0x7f9fe05cc13c google::LogMessage::SendToLog()
    @     0x7f9fe05c81b3 google::LogMessage::Flush()
    @     0x7f9fe05cd64e google::LogMessageFatal::~LogMessageFatal()
    @     0x7f9fe03df2c6 paddle::GpuMatrix::mul()
    @     0x7f9fe02e3235 paddle::FullyConnectedLayer::forward()
    @     0x7f9fe01da79d paddle::NeuralNetwork::forward()
    @     0x7f9fe01b1ee3 paddle::GradientMachine::forwardBackward()
    @     0x7f9fe05a44e4 GradientMachine::forwardBackward()
    @     0x7f9fe013cff9 _wrap_GradientMachine_forwardBackward
    @     0x7fa0173053d3 PyEval_EvalFrameEx
    @     0x7fa017307160 PyEval_EvalCodeEx
    @     0x7fa0173054d1 PyEval_EvalFrameEx
    @     0x7fa017307160 PyEval_EvalCodeEx
    @     0x7fa0173054d1 PyEval_EvalFrameEx
    @     0x7fa017307160 PyEval_EvalCodeEx
    @     0x7fa0173054d1 PyEval_EvalFrameEx
    @     0x7fa017307160 PyEval_EvalCodeEx
    @     0x7fa0173054d1 PyEval_EvalFrameEx
    @     0x7fa017307160 PyEval_EvalCodeEx
    @     0x7fa017307272 PyEval_EvalCode
    @     0x7fa01732165c run_mod
    @     0x7fa017321730 PyRun_FileExFlags
    @     0x7fa017322c3c PyRun_SimpleFileExFlags
Aborted

改用CPU运行时有时可以跑一些batch，多数情况下报错并卡死在这里：

请问这可能是因为什么呢？ 另外之前的版本trainer_count > 1是可以正常使用的，新版本>1时不能训练么？