Libflattenconcat.so undefined symbol

I am trying to load libflattenconcat.so in python application which I am moving from a Jetson to my PC, running ubuntu 18.04 LTS. I was able to download TensorRT source and build the pluggin, which I then copied to where the application was expecting it. When I use the line:

ctypes.CDLL("/opt/libflattenconcat.so")

I get the error:

/opt/libflattenconcat.so: undefined symbol: _ZN8nvinfer16plugin16throwCublasErrorEPKcS2_iiS2_

Any idea what that means? Thanks.

Hi,
Please refer to the installation steps from the below link if in case you are missing on anything
https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html
Also, we suggest you to use TRT NGC containers to avoid any system dependency related issues.
https://ngc.nvidia.com/catalog/containers/nvidia:tensorrt

Thanks!

I did use those steps to install TensorRT and I have verified the installation. I’ve run some samples ok. I will try the containers, but I am still hopeful someone can help me understand what the symbol in that error is. I do have the CUDA Toolkit installed.

I would like to be able to setup the environment without using a container, ideally.

Hi @loophole64,

Could you please provide us following environment details.
Also please let us know which method are you following for the installation.

TensorRT Version :
GPU Type :
Nvidia Driver Version :
CUDA Version :
CUDNN Version :
Operating System + Version :
Python Version (if applicable) :

Thank you.

TensorRT Version : dpkg shows 7.2.3-1+cuda11.1
GPU Type : NVIDIA GeForce GTX 1060 3GB
Nvidia Driver Version : nvidia-smi shows 465.19.01

CUDA Version : nvidia-smi shows 11.3. dpkg shows mix of 11.1, 11.3, 10.2. Could this be an issue? I thought I had installed 11.1. nvcc --version shows:
Cuda compilation tools, release 11.1, V11.1.105
Build cuda_11.1.TC455_06.29190527_0
/usr/local/cuda appears linked to 11.1

CUDNN Version : dpkg shows 8.1.1.33-1+cuda11.2
Operating System + Version : Ubuntu 18.04 LTS
Python Version (if applicable) : Python 3.6.9

It looks like maybe I have some version mismatches. CuDNN looks like it wants CUDA 11.2 which isn’t installed. I installed this stuff a few times as I was figuring it out, so I probably mismatched versions. Maybe I should I reinstall CUDA 11.2, 11.1, CuDNN 8 and TensorRT 7?

Hi @loophole64,

Pleas make sure all dependencies are installed. Please refer following doc to check support matrix.
https://docs.nvidia.com/deeplearning/tensorrt/support-matrix/index.html

Thank you.

Thank you @spolisetty, that support matrix is very helpfull. I am going to remove and reinstall everything and I will let you know how it turns out. ^_^

1 Like

TensorRT Version : 7.2.3-1+cuda11.1
GPU Type : NVIDIA GeForce GTX 1060 3GB
Nvidia Driver Version : 465.19.01
CUDA Version : 11.1 and 11.2 are installed.
CUDNN Version : 8.1.1
Operating System + Version : Ubuntu 18.04 LTS
Python Version (if applicable) : 3.6.9

These are all consistant with the matrix for TensorRT 7.2.3 but I am still getting the same error.

libflattenconcat.so was built from the 7.2.3.4 repository.

The error seems to indicate an issue with libnvinfer and cublas. I’m unsure how to check my Cublas version, but my understanding is that it is installed as part of Cuda, is that correct?

Hi @loophole64,

Sorry for the delayed response. This looks like environment setup related. We request you to post your query here, you may get better help here.

Thank you.

1 Like