segmentation fault during LoadNetwork if tensorflow c library is also used

Li__Rachel · ‎04-13-2020

Environment: ubuntu 16.04 (I installed using apt-get)

The following code has segmentation fault only if `TF_NumDims` is used in a dummy function. dummy function is not used anywhere in the code.

The code is seg faulted at line `ExecutableNetwork executable_network = ie.LoadNetwork(network, device_name, openvino_config);`

include <iostream>
#include <inference_engine.hpp>
#include <opencv2/opencv.hpp>
#include <tensorflow/c/c_api.h>

int64_t dummy (TF_Tensor* tensor) {
  return TF_NumDims(tensor); // with this line, segmentation fault at LoadNetwork
//  return 0; // with this line, there is no segmentationf fault
}

int main(int argc, char** argv) {
  std::cout << "Hello world!\n";
  using namespace InferenceEngine;
    try {
        const std::string device_name = "CPU";
        const std::string input_model = "./ml_runtime-test-data/resnet_v1_50.xml";
        const std::string input_param = "./ml_runtime-test-data/resnet_v1_50.bin";

        Core ie;

        CNNNetwork network = ie.ReadNetwork(input_model, input_param);
        network.setBatchSize(1);

        InputInfo::Ptr input_info = network.getInputsInfo().begin()->second;
        std::string input_name = network.getInputsInfo().begin()->first;

        input_info->getPreProcess().setResizeAlgorithm(RESIZE_BILINEAR);
        input_info->setLayout(Layout::NHWC);
        input_info->setPrecision(Precision::U8);

        DataPtr output_info = network.getOutputsInfo().begin()->second;
        std::string output_name = network.getOutputsInfo().begin()->first;

        output_info->setPrecision(Precision::FP32);

        const std::map< std::string, std::string > openvino_config;
        ExecutableNetwork executable_network = ie.LoadNetwork(network, device_name, openvino_config);

        InferRequest infer_request = executable_network.CreateInferRequest();

        std::cout << input_name << std::endl;
        std::cout << output_name << std::endl;

        cv::Mat mat = cv::imread("./ml_runtime-test-data/golden-retriever-puppy.jpg", cv::IMREAD_COLOR);
        size_t channels = mat.channels();
        size_t height = mat.size().height;
        size_t width = mat.size().width;

        size_t strideH = mat.step.buf[0];
        size_t strideW = mat.step.buf[1];

        InferenceEngine::TensorDesc tDesc(InferenceEngine::Precision::U8,
                                          {1, channels, height, width},
                                          InferenceEngine::Layout::NHWC);

        Blob::Ptr imgBlob = InferenceEngine::make_shared_blob<uint8_t>(tDesc, mat.data);

        infer_request.SetBlob(input_name, imgBlob);  // infer_request accepts input blob of any size
        infer_request.Infer();
        Blob::Ptr output = infer_request.GetBlob(output_name);

        InferenceEngine::TBlob<float32_t>& tblob = dynamic_cast<InferenceEngine::TBlob<float32_t>&>(*output);
        InferenceEngine::SizeVector dims = tblob.getTensorDesc().getDims();
        size_t input_rank = dims.size();
        size_t batchSize = dims[0];
        std::vector<unsigned> indexes(tblob.size());

        float32_t* data = tblob.data();

        std::iota(std::begin(indexes), std::end(indexes), 0);
        std::partial_sort(std::begin(indexes), std::begin(indexes) + 5, std::end(indexes),
                          [&data](unsigned l, unsigned r) {
                              return data > data;
                          });

        std::cout << indexes.at(0) << std::endl;
    } catch (const std::exception & ex) {
        std::cerr << ex.what() << std::endl;
    }

  return 0;
}

Here is back trace.

[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Hello world!
[New Thread 0x7fffe65a5700 (LWP 13802)]
[New Thread 0x7fffe5da4700 (LWP 13803)]

Thread 2 "test" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffe65a5700 (LWP 13802)]
0x00007fffe913e68c in ?? () from /opt/intel/openvino_2020.1.023/deployment_tools/inference_engine/lib/intel64/libMKLDNNPlugin.so
(gdb) bt
#0  0x00007fffe913e68c in ?? () from /opt/intel/openvino_2020.1.023/deployment_tools/inference_engine/lib/intel64/libMKLDNNPlugin.so
#1  0x00007fffef136cf2 in ?? () from /opt/intel/openvino_2020.1.023/deployment_tools/inference_engine/lib/intel64/libinference_engine.so
#2  0x00007fffef1362d9 in ?? () from /opt/intel/openvino_2020.1.023/deployment_tools/inference_engine/lib/intel64/libinference_engine.so
#3  0x00007fffede97a99 in __pthread_once_slow (once_control=0x8464098, init_routine=0x7ffff7b07750 <__once_proxy>) at pthread_once.c:116
#4  0x00007fffef13730a in ?? () from /opt/intel/openvino_2020.1.023/deployment_tools/inference_engine/lib/intel64/libinference_engine.so
#5  0x00007fffef13650e in ?? () from /opt/intel/openvino_2020.1.023/deployment_tools/inference_engine/lib/intel64/libinference_engine.so
#6  0x00007ffff7b088e0 in ?? () from /usr/lib/x86_64-linux-gnu/libstdc++.so.6
#7  0x00007fffede906ba in start_thread (arg=0x7fffe65a5700) at pthread_create.c:333
#8  0x00007fffedbc641d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109

Li__Rachel · ‎04-13-2020

I am able to repeat the above steps in a different machine and got the same segmentation fault.