Created by: peterzhang2029

Currently, an error will occur when using hsigmoid_layer with 'use_gpu=True'.

I1127 11:19:25.492727 31440 Util.cpp:166] commandline:  --use_gpu=True --trainer_count=1
I1127 11:19:29.783864 31440 GradientMachine.cpp:94] Initing parameters..
I1127 11:19:29.974581 31440 GradientMachine.cpp:101] Init parameters done.
F1127 11:19:31.461540 31440 Matrix.h:501] Not implemeted
*** Check failure stack trace: ***
    @     0x7fadd9f1996d  google::LogMessage::Fail()
    @     0x7fadd9f1d41c  google::LogMessage::SendToLog()
    @     0x7fadd9f19493  google::LogMessage::Flush()
    @     0x7fadd9f1e92e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7fadd9cbab80  paddle::Matrix::addByBitCode()
    @     0x7fadd9b967b2  paddle::HierarchicalSigmoidLayer::forward()
    @     0x7fadd9c1fedd  paddle::NeuralNetwork::forward()
    @     0x7fadd9c20bf3  paddle::GradientMachine::forwardBackward()
    @     0x7fadd9ef5674  GradientMachine::forwardBackward()
    @     0x7fadd9a980b9  _wrap_GradientMachine_forwardBackward
    @     0x7fae0ee013a3  PyEval_EvalFrameEx
    @     0x7fae0ee03130  PyEval_EvalCodeEx
    @     0x7fae0ee014a1  PyEval_EvalFrameEx