mEngine->createExecutionContextWithoutDeviceMemory() crashed

Description

createExecutionContextWithoutDeviceMemory() may not work well when deserializing a engine file to engine.

Environment

TensorRT Version: 7.2.2.3
GPU Type: GeForce RTX 3080
Nvidia Driver Version: 461.40-desktop-win10-64bit-international-dch-whql
CUDA Version: 11.1
CUDNN Version: 8.0.5
Operating System + Version: Windows 10

Steps To Reproduce

Anything works well when i build a engine from onnx model file(this model is ok, works well before), and process inference;
But:

  1. build a engine from onnx model file;
  2. serialize the engine to disk as binary file;
  3. deserialize this binary file to engine;
  4. mEngine->createExecutionContextWithoutDeviceMemory() crashed.

    stack_20210127173036

many thanks!

Hi @liangjia1989,

Could you please share us issue reproduce script. Which could be helpful to debug.

Thank you.