Cuda build error

– The CUDA compiler identification is unknown
– Check for working CUDA compiler: /xhome/cuda/cuda_nvcc-linux-aarch64-12.0.76-archive/bin/nvcc
– Check for working CUDA compiler: /xhome/cuda/cuda_nvcc-linux-aarch64-12.0.76-archive/bin/nvcc – broken
CMake Error at /home/make_3.10/cmake-3.10.3-Linux-x86_64/share/cmake-3.10/Modules/CMakeTestCUDACompiler.cmake:46 (message):
The CUDA compiler

"/xhome/cuda/cuda_nvcc-linux-aarch64-12.0.76-archive/bin/nvcc"

is not able to compile a simple test program.

It fails with the following output:

if i am pointing cross complie path to dynamic one i am getting above error for hello.cu complialtion in cmake.

can anyone help me on this.

thanks .

Are you running this on a Nano? If so, the CUDA release on a Nano only works for 10.x version. You would have to change your compiled CUDA to a 10.x release installed from JetPack/SDK Manager. Note that JetPack/SDKM is a GUI installer, and the actual software which flashes to a Jetson (or gets installed as an optional package, of which CUDA is one), have their release versions tied together (getting one implies knowing the release of the other). See:

To see your current L4T release use “head -n 1 /etc/nv_tegra_release”. You’ll want an Ubuntu 18.04 host PC for the Nano releases. Note that if you have a JetPack/SDKM release which is too new for the L4T release, then you can start sdkmanager like this to make older releases available:
sdkmanager --archivedversions

Hi,

CUDA 12 cannot work on JetPack 4.x release.
Please use CUDA 10.2 from JetPack instead.

Thanks.

hi Currently i am trying to build cross compile .cu and .cpp combined binary to run on NX .

i am using cmake verison 3.10

NVIDIA cuda verison 10.2

and i want to include .cu files as shared library.

Error i am getting:

/lib/ld-linux-aarch64.so.1: No such file or directory

sdk/build/cmake/…/…/…/bsp/rootfs/usr/include: Permission denied

Tries i have done :
Install cuda and tried to build CPP_integartion in local host complied successfully .

normal helloworld.cu also complies successfully in local host .

but while doing cross compilation for .cu and .ccp files i am not able to build .cu files

can anyone help me on this

thanks

Hi,

If you want to cross-compile a CUDA app on a desktop, please install the host component in JetPack.
It includes that required cross package for cross-compiling.

Thanks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.