nvinfer1::IExecutionContext cannot destory

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Description

nvinfer1::iexecutioncontext cannot destroy

Environment

TensorRT Version: 8.0
GPU Type: nvidia xavier nx
Nvidia Driver Version: jetpack 4.6
CUDA Version: 10.2
CUDNN Version: 8.0
Operating System + Version: ubuntu 18.04
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag):

Relevant Files

I call loadModel first, and then call the unloadModel function immediately. The program reports a Segmentation fault, which is reported in execution_context->destroy(). I try to put execution_context->destroy() in the loadModel function, and it works normally. There is no operation between loadModel and unloadModel, why is this happening?
bool nvidia_normal::UnloadModel() {
execution_context->destroy();
qDebug()<<“22”;
checkRuntime(cudaStreamDestroy(stream));
stream==nullptr;
checkRuntime(cudaFreeHost(input_data_host));
input_data_host=nullptr;
checkRuntime(cudaFreeHost(output_data_host));
output_data_host=nullptr;
checkRuntime(cudaFree(input_data_device));
input_data_device=nullptr;
checkRuntime(cudaFree(output_data_device));
output_data_device=nullptr;
engine->destroy();
runtime->destroy();
}

Steps To Reproduce

Please include:

  • Exact steps/commands to build your repro
  • Exact steps/commands to run your repro
  • Full traceback of errors encountered
2

class nvidia_normal : public AlgModelInterface
{
public:
nvidia_normal();
virtual bool LoadModel(mode_param param);
virtual bool Detect(cv::Mat image, QList &list);
virtual bool UnloadModel();
bool classification{false};
private:

int input_batch;
 int input_channel;
 int input_height;
 int input_width;
 int input_numel;
 float* input_data_host;
 float* input_data_device;
 int output_numbox;
 int output_numprob;
 int num_classes;
 int output_numel;
 float* output_data_host;
 float* output_data_device;
 nvinfer1::IRuntime* runtime;
 nvinfer1::ICudaEngine *engine;
 nvinfer1::IExecutionContext * execution_context;
 cudaStream_t stream;

};
This is the definition of the class

3

This is the calling code, which does not do anything in between, but the destory can be run in the loadModel function, and when I call the unload function, an error will be reported：
nvidia_normal *model =new nvidia_normal();
model->LoadModel(modelparam);
model->UnloadModel();

4

Hi,

This looks like a Jetson issue. Please refer to the below samples in case useful.

For any further assistance, we will move this post to to Jetson related forum.

Thanks!