create more than one engine will cause crash

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

In my project, there are two networks need to inference In the whole pipeline. So I parser the two model into different engines, contexts, runtime and all that related class have two instances .But when I run it, it will crash in parsering the origin model. The first one will build OK, and the second one will failed to crash , It is the same in reverse models. Why did it happen ? Is there any class may be singleton that I can not create twice or some state need to clear?

Thanks.

2

Hello,

to help us debug, can you please share a small repro that demonstrates the symptoms you are seeing? AFAIK, TRT is multi-thread friendly.

regards,
NVES

3

Hi，NVES,

Thanks for your reply.I upload the project to the github [url]https://github.com/lewes6369/tensorRTWrapper[/url]. please take a look to the sample named “runTwoNets”.

PS: If exporting the engine to file and load from file. Then I can create more than one engine, otherwise will failed to crash.

Thanks.

4

Hello,

apologize for the delay. Hello, can you provide details on the platforms you are using?

Linux distro and version
GPU type
nvidia driver version
CUDA version
CUDNN version
Python version [if using python]
Tensorflow version
TensorRT version

5

Hi,NVES

the platforms info as below:

Ubuntu 16.04
TensorRT 4.0.1.6
CUDA 9.2
GPU NV GTX 1060

Used C++ and no tf.

6

Hello,

Our engineers recommend trying with TensorRT 6.0? We believe this was fixed with a change to track inputs to make sure they are only accessed by their own networks.

regards,
NVIDIA Enterprise Support

7

Is there a version of TensorRT 6.0 for the TX2? Perhaps included in some Jetpack version?