Compiling Autoware v1.12.0 with cuda v10.2

edit

Hello,

I have a pre-compiled Autoware v1.12.0 on my system which I compiled when I had Cuda 9 on my Ubuntu 16.04 system. Basic Autoware functionalities came up, but I could not utilize my GPU HW. Referring NVIDIA support pointed me to update Cuda owing to my GPU HW. I have a Turing GPU with compute capability 7.5. NVIDIA guidelines for such hardware is to use Cuda v10.x. Accordingly I have installed Cuda v10.2 with cuDNN v7.5.x on an Ubuntu 16.04 OS. However since then, I am trying to recompile Autoware v1.12.0 which is failing at lidar_euclidean_cluster_detect package due to a missing file: /usr/local/cuda/include/thrust/detail/dispatch/is_trivial_copy.h.

Error message is:

No rule to make target '/usr/local/cuda/include/thrust/detail/dispatch/is_trivial_copy.h', needed by 'CMakeFiles/gpu_euclidean_clustering.dir/nodes/lidar_euclidean_cluster_detect/gpu_euclidean_clustering_generated_gpu_euclidean_clustering.cu.o'.  Stop.

Has anyone installed Autoware v1.12.0 with Cuda v10.x? Autoware recommends Cuda v9.0 for Autoware on OS Ubuntu 16.04, but based on my GPU HW, NVIDIA lays out requirement for Cuda v10.x. Any suggestions on how to get out of this paradox?

Thanks