Tensorflow segmentation fault running multirow tensor in C++ api

Question

We are developing a method to load and run a .pb file within our code and have run into an odd segmentation fault which we cannot figure out how to debug. See the following code:

extern "C" void run_graph_(char* path){

tensorflow::Status s;
tensorflow::GraphDef gdef;
s = tensorflow::ReadBinaryProto(tensorflow::Env::Default(), path, &gdef);
if (!s.ok()) 
{
    std::cout << "error: " << s.ToString() << "\n";     
}
else
{
    std::cout << "complete: " << s.ToString() << "\n"; 
}

for (int i = 0; i < gdef.node_size(); i++)
{
    if (gdef.node(i).name() == "nn_output" || gdef.node(i).name() == "nn_input")
        {
            gdef.node(i).PrintDebugString();
        }
    }

tensorflow::Session* session;
s = tensorflow::NewSession(tensorflow::SessionOptions(), &session);
s = session->Create(gdef);

std::vector<tensorflow::Tensor> outputs;
tensorflow::Input::Initializer a({{8.81474371e-01,   8.90273631e-01,   1.72516504e-01, 5.55829580e-04,   9.59886648e-01,   1.20462446e-02, 1.98175483e-03,   1.38663768e-02}});
std::cout << a.tensor.DebugString() << std::endl;

std::cout << "run" << std::endl;
s = session->Run({{"nn_input", a.tensor}}, {"nn_output"}, {}, &outputs);
std::cout << "ran" << std::endl;
if (!s.ok()) 
{
    std::cout << "error: " << s.ToString() << "\n";     
}
if( outputs.size() > 0 )
{
    std::cout << "output: " << outputs[0].DebugString() << std::endl; // Tensor<type: float shape: [] values: 30>
}

// Free any resources used by the session
session->Close();
}

If we run this code, it works fine and we get the following output:

complete: OK
name: "nn_input"
op: "Placeholder"
attr {
  key: "dtype"
  value {
    type: DT_DOUBLE
  }
}
attr {
  key: "shape"
  value {
    shape {
      dim {
        size: -1
      }
      dim {
        size: 8
      }
    }
  }
}
name: "nn_output"
op: "Add"
input: "MatMul_4"
input: "Variable_4/read"
attr {
  key: "T"
  value {
    type: DT_DOUBLE
  }
}
Tensor<type: double shape: [1,8] values: [0.881474371 0.890273631 0.172516504]...>
run
ran
output: Tensor<type: double shape: [1,2] values: [0.016238948144062765 0.18446201154785777]>

However if we change the input tensor to have more than 1 row, for example:

tensorflow::Input::Initializer a({{8.81474371e-01,   8.90273631e-01,   1.72516504e-01, 5.55829580e-04,   9.59886648e-01,   1.20462446e-02, 1.98175483e-03,   1.38663768e-02}, {5.58892306e-01,   8.76436973e-01,   1.78063232e-01, 1.27910966e-01,   2.23700877e-01,   1.81518624e-01, 2.75920096e-01,   1.19812048e-02}, {3.69272573e-01,   6.75944584e-01,   8.35322158e-01, 3.89913301e-02,   4.26433973e-01,   2.94287142e-01, 1.08364415e-01,   3.93754220e-02}, {4.27477538e-01,   6.70533752e-01,   4.24946841e-02, 4.63300582e-02,   1.07105860e-02,   1.92844420e-02, 2.40294042e-02,   2.53449079e-03}, {2.66025541e-01,   9.75372575e-01,   6.53711732e-02, 7.52453239e-02,   2.75028037e-02,   1.23259054e-01, 1.29876045e-01,   1.75538941e-01}});

we get a seg fault when it tried to run the graph with the following output:

complete: OK
name: "nn_input"
op: "Placeholder"
attr {
  key: "dtype"
  value {
    type: DT_DOUBLE
  }
}
attr {
  key: "shape"
  value {
    shape {
      dim {
        size: -1
      }
      dim {
        size: 8
      }
    }
  }
}
name: "nn_output"
op: "Add"
input: "MatMul_4"
input: "Variable_4/read"
attr {
  key: "T"
  value {
    type: DT_DOUBLE
  }
}
Tensor<type: double shape: [5,8] values: [0.881474371 0.890273631 0.172516504]...>
run

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0  0x7fcd55f2f26f in ???
#1  0x7fcd5488b77a in ???
#2  0x7fcd548bebb3 in ???
#3  0x7fcd549032c3 in ???
#4  0x7fcd549204ef in ???
#5  0x7fcd52beccfc in ???
#6  0x7fcd52bbbcda in ???
#7  0x7fcd52bad633 in ???
#8  0x7fcd52818fc0 in ???
#9  0x7fcd52816d09 in ???
#10  0x7fcd5206706e in execute_native_thread_routine
        at ../../../.././libstdc++-v3/src/c++11/thread.cc:83
#11  0x7fcd56d46e24 in ???
#12  0x7fcd55ff234c in ???
#13  0xffffffffffffffff in ???
Segmentation fault

We are using the C++ on CentOS 7. I believe we should be able to run a multidimensional tensor through this graph as it works in python. Any help is appreciated.