I have installed CUDA 5.0 and NVCC on my Ubuntu virtual machine and have had problems compiling even a basic CUDA C program. The error is as follows:
user@ubuntu:~/CUDA$ nvcc helloworld.cu -o helloworld.o -target-cpu-arch=ARM -ccbin=/usr/bin/arm-linux-gnueabi-gcc-4.6 --machine=32
/usr/lib/gcc/arm-linux-gnueabi/4.6/../../../../arm-linux-gnueabi/bin/ld: skipping incompatible /usr/local/cuda-5.0/bin/../lib/libcudart.so when searching for -lcudart
/usr/lib/gcc/arm-linux-gnueabi/4.6/../../../../arm-linux-gnueabi/bin/ld: skipping incompatible /usr/lib/libcudart.so when searching for -lcudart
/usr/lib/gcc/arm-linux-gnueabi/4.6/../../../../arm-linux-gnueabi/bin/ld: cannot find -lcudart
collect2: ld returned 1 exit status
I have tried to research this problem and came across this link: skipping incompatible libcudart.so when searching for -lcudart
And so I followed the advice that was offered on that link and added
/usr/local/cuda-5.0/lib64
and
/usr/local/cuda-5.0/lib
to my LD_LIBRARY_PATH environment variable and now this is the result of the
user@ubuntu:~/CUDA$ echo $LD_LIBRARY_PATH
/usr/local/cuda-5.0/lib:/usr/local/cuda-5.0/lib64
user@ubuntu:~/CUDA$
However, the problem still persists, please help.
CUDA 5.5 was the first CUDA release to support ARM v7. The CUDA 5.0 toolkit you installed does not have support, which is why it can't find the right version of the CUDA runtime library.
It turns out that the CUDA installer I was using from NVIDIA will not allow me to cross compile for my CARMA board, but it has to be downloaded from the manufacturer SECO.
May be, it's a problem with Cuda Toolkits. The first the Cuda Toolkit 5.0 doesn't support Arm. The second the Cuda Toolkit 6.5 cross compile lost libcudart.so. I solved this problem is copying libcudart.so from the Cuda Toolkit 6.5 by Arm.
add /etc/ld.so.conf this:
/usr/local/cuda-5.0/lib /usr/local/cuda-5.0/lib64
and run command: ldconfig your compile will find cudart lib
来源：https://stackoverflow.com/questions/30701263/nvcc-cuda-cross-compiling-cannot-find-lcudart