Setting up cudnn and libs properly?

Accelerated Computing CUDA CUDA Setup and Installation
1

Howdy,

I just installed tensorrt. When trying to build sampleMNIST I realized I didn’t have cudnn installed (I got a “libcudnn.so.7 not found” while building), so I installed it. It now builds, but when I try to run it I get:

./sample_mnist: error while loading shared libraries: libcudnn.so.7: cannot open shared object file: No such file or directory

I can get around this by adding /usr/local/cuda/lib64 to my LD_LIBRARY_PATH, but I’m not sure that’s the right approach. Also, ldconfig tells me:

/sbin/ldconfig.real: /usr/local/cuda-8.0/targets/x86_64-linux/lib/libcudnn.so.7 is not a symbolic link

I can fix this with

sudo ln -sf /usr/local/cuda-8.0/targets/x86_64-linux/lib/libcudnn.so.7.1.3 /usr/local/cuda-8.0/targets/x86_64-linux/lib/libcudnn.so.7

But it’s possible I’ve bungled my installation somehow.

2

I must have, because now importing nvinfer gives me:

ImportError: libnvinfer.so.4: cannot open shared object file: No such file or directory

I have that file in /home/ml/tensorrt/TensorRT-3.0.4/targets/x86_64-linux-gnu/lib, symlinked by /home/ml/tensorrt/TensorRT-3.0.4/lib (which is also in my LD_LIBRARY_PATH)