Crosscompiling, TensorRT/CuDNN

I am trying to crosscompile some C++ software for a Drive PX2.

Using the SDK installed by DriveInstall in a Ubuntu 16.04 VM everything works out of the box.

find_package(TensorRT REQUIRED)
find_package(CuDNN REQUIRED)

results in

-- Found TensorRT:
--  - Includes: /usr/local/nvidia/tensorrt/targets/aarch64-linux-gnu/include
--  - Libraries: /usr/local/nvidia/tensorrt/targets/aarch64-linux-gnu/lib/;/usr/local/nvidia/tensorrt/targets/aarch64-linux-gnu/lib/
-- Found libcudnn:
--  - Includes: /usr/local/cuda-9.0/targets/aarch64-linux/include
--  - Libraries: /usr/local/cuda-9.0/targets/aarch64-linux/lib/

I set up a new host VM with Ubuntu 16.04 using the current version SDK Manager (Version:
It looks like TensorRT and CuDNN (libs for cross compiling) are not getting installed on the host with this version. What am i doing wrong?

Dear emm7,
We will look into it and update you.

Dear emm7,
Did you select only Host side installation in SDKmanager? Also, Could you please delete cache (~/.nvsdkm folder) and try installing again and attach the following log files here.

Thanks. I set up a new clean VM and ran a full install (Host and Autochauffeur, skipping the flash step). I attached ~/.nvsdkm/sdkm.log, the (zipped) ~/.nvsdkm/logs directory and the exported logs from the sdk manager.

I then deleted ~/.nvsdkm, am currently re-running the install and will add the new logs afterwards. (175 KB) (54.4 KB) (93.9 KB)

Dear emm7,
If the installation is successful you can find tensorRT at /usr/local/cuda-9.2/dl and /usr/local/cuda-9.2/dl/targets/aarch64-linux-gnu/lib contains the cross compiled libraries of TensorRT, cudnn. Could you check and confirm