Created by: yeyupiaoling

在场景文字识别的例子中：https://github.com/PaddlePaddle/models/tree/develop/scene_text_recognition 训练是一定要GPU，但是预测是可以不用的，如这个代码：https://github.com/PaddlePaddle/models/blob/9264e8cebb0219a9cd765511b8a7c9236f0b1da8/scene_text_recognition/network_conf.py#L81-L89 但是在真正的实际上，在预测使用CPU是报错：

F0403 08:36:29.233067  3738 Allocator.h:54] Check failed: posix_memalign(&ptr, 4096ul, size) == 0 (12 vs. 0) 
*** Check failure stack trace: ***
    @     0x7f2cf5f3e2cd  google::LogMessage::Fail()
    @     0x7f2cf5f40618  google::LogMessage::SendToLog()
    @     0x7f2cf5f3dddb  google::LogMessage::Flush()
    @     0x7f2cf5f414ee  google::LogMessageFatal::~LogMessageFatal()
    @     0x7f2cf5e9ca91  paddle::PoolAllocator::alloc()
    @     0x7f2cf5e95d25  paddle::CpuMemoryHandle::CpuMemoryHandle()
    @     0x7f2cf5e50c93  paddle::CpuMatrix::CpuMatrix()
    @     0x7f2cf5e50e1e  paddle::Matrix::create()
    @     0x7f2cf5e61518  paddle::Matrix::resizeOrCreate()
    @     0x7f2cf5c49fa7  paddle::Layer::resetSpecifyOutput()
    @     0x7f2cf5c4a212  paddle::Layer::resetOutput()
    @     0x7f2cf5d1b331  paddle::BatchNormalizationLayer::forward()
    @     0x7f2cf5b3fe3f  paddle::NeuralNetwork::forward()
    @     0x7f2cf5ac0d86  _wrap_GradientMachine_forward
    @           0x4bc3fa  PyEval_EvalFrameEx
    @           0x4c136f  PyEval_EvalFrameEx
    @           0x4b9ab6  PyEval_EvalCodeEx
    @           0x4c1e6f  PyEval_EvalFrameEx
    @           0x4d4c9d  (unknown)
    @           0x4bc9b6  PyEval_EvalFrameEx
    @           0x4d4c9d  (unknown)
    @           0x4bc9b6  PyEval_EvalFrameEx
    @           0x4b9ab6  PyEval_EvalCodeEx
    @           0x4c16e7  PyEval_EvalFrameEx
    @           0x4c136f  PyEval_EvalFrameEx
    @           0x4c136f  PyEval_EvalFrameEx
    @           0x4b9ab6  PyEval_EvalCodeEx
    @           0x4eb30f  (unknown)
    @           0x4e5422  PyRun_FileExFlags
    @           0x4e3cd6  PyRun_SimpleFileExFlags
    @           0x493ae2  Py_Main
    @     0x7f2d1325b830  __libc_start_main