Multi-card mode does not work

[BigFishMaster]

I used multi-card mode as follow in my experiment.

    if parallel:
        places = fluid.layers.get_places()
        pd = fluid.layers.ParallelDo(places)
        with pd.do():
            img_ = pd.read_input(image)
            label_ = pd.read_input(label)
            prediction, avg_cost, accuracy = net_conf(img_, label_, class_dim)
            for o in [avg_cost, accuracy]:
                pd.write_output(o)

        avg_cost, accuracy = pd()
        # get mean loss and acc through every devices.
        avg_cost = fluid.layers.mean(x=avg_cost)
        accuracy = fluid.layers.mean(x=accuracy)
    else:
        prediction, avg_cost, accuracy = net_conf(image, label, class_dim)

I got ERRORs below. Can anyone fix this bug for me ?

Traceback (most recent call last):
  File "resnet50_parallel.py", line 192, in <module>
    train(learning_rate=0.1, batch_size=24, num_passes=30, init_model=None)
  File "resnet50_parallel.py", line 158, in train
    exe.run(feed=feeder.feed(data))
  File "/home/ssd5/code/code_paddle/compile/code/resnet50/python-gcc482-paddle/lib/python2.7/site-packages/paddle/v2/fluid/executor.py", line 273, in run
    self.executor.run(program.desc, scope, 0, True, True)
paddle.v2.fluid.core.EnforceNotMet: holder_ should not be null
Tensor not initialized yet when Tensor::type() is called. at [/paddle/paddle/framework/tensor.h:119]
PaddlePaddle Call Stacks: 
0       0x7fd6d2d09846p paddle::platform::EnforceNotMet::EnforceNotMet(std::__exception_ptr::exception_ptr, char const*, int) + 486
1       0x7fd6d2d0ade6p paddle::framework::Tensor::type() const + 150
2       0x7fd6d36ca8b9p paddle::framework::LoDTensor::MergeLoDTensor(std::vector<paddle::framework::LoDTensor const*, std::allocator<paddle::framework::LoDTensor const*> > const&, boost::variant<paddle::platform::CUDAPlace, paddle::platform::CPUPlace, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_>) + 153
3       0x7fd6d35172f8p paddle::operators::ParallelDoOp::Run(paddle::framework::Scope const&, boost::variant<paddle::platform::CUDAPlace, paddle::platform::CPUPlace, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_, boost::detail::variant::void_> const&) const + 2936
4       0x7fd6d2db583ap paddle::framework::Executor::Run(paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool) + 1194
5       0x7fd6d2d227fbp void pybind11::cpp_function::initialize<pybind11::cpp_function::initialize<void, paddle::framework::Executor, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool, pybind11::name, pybind11::is_method, pybind11::sibling>(void (paddle::framework::Executor::*)(paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool), pybind11::name const&, pybind11::is_method const&, pybind11::sibling const&)::{lambda(paddle::framework::Executor*, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool)#1}, void, paddle::framework::Executor*, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool, pybind11::name, pybind11::is_method, pybind11::sibling>(pybind11::cpp_function::initialize<void, paddle::framework::Executor, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool, pybind11::name, pybind11::is_method, pybind11::sibling>(void (paddle::framework::Executor::*)(paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool), pybind11::name const&, pybind11::is_method const&, pybind11::sibling const&)::{lambda(paddle::framework::Executor*, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool)#1}&&, void (*)(paddle::framework::Executor*, paddle::framework::ProgramDesc const&, paddle::framework::Scope*, int, bool, bool), pybind11::name const&, pybind11::is_method const&, pybind11::sibling const&)::{lambda(pybind11::detail::function_call&)#3}::_FUN(pybind11::detail::function_call) + 555
6       0x7fd6d2d1c034p pybind11::cpp_function::dispatcher(_object*, _object*, _object*) + 2596
7       0x7fd7275f255fp PyEval_EvalFrameEx + 29855
8       0x7fd7275f486dp PyEval_EvalCodeEx + 2061
9       0x7fd7275f19fcp PyEval_EvalFrameEx + 26940
10      0x7fd7275f486dp PyEval_EvalCodeEx + 2061
11      0x7fd7275f19fcp PyEval_EvalFrameEx + 26940
12      0x7fd7275f486dp PyEval_EvalCodeEx + 2061
13      0x7fd7275f49a2p PyEval_EvalCode + 50
14      0x7fd72761d782p PyRun_FileExFlags + 146
15      0x7fd72761eaf9p PyRun_SimpleFileExFlags + 217
16      0x7fd72763482dp Py_Main + 3149
17      0x7fd726831bd5p __libc_start_main + 245
18            0x4007a1p

[kuke]

I met the same error in GPU mode, but seems that ParallelDo can work in CPU mode.