Run CUDA on jetson nano

Hi.
We are trying to run CUDA app on jetson nano.
We have customized kernel 4.9.140-jetson-nano and custom rootfs (not related to Tegra_Linux_Sample-Root-Filesystem_R32.5.0).
We have installed tegra drivers to our rootfs from Tegra210_Linux_R32.5.0_aarch64.tbz2 and got perfectly working H264 codec. Now we want to get CUDA working.
We have installed CUDA 10.0 with cross-compillation tools to the PC using JetPack.
We have built /usr/local/cuda-10.0/samples/1_Utilities/deviceQuery and link with libcudart_static.a
As result we got output:
./deviceQuery
./deviceQuery Starting…

 CUDA Device Query (Runtime API) version (CUDART static linking)

NVMAP_IOC_QUERY_HEAP_PARAMS failed [Inappropriate ioctl for device]
cudaGetDeviceCount returned 30
-> unknown error
Result = FAIL

What libraries do we need to install to get the sample working?
We don’t want to install ALL libraries related to CUDA, we want to install ONLY libraries related to this sample (deviceQuery) to get it working.
Thanks.

Hi,

For R32.5, you will need to install CUDA 10.2 for compatibility.
Would you mind to give it a try?

Thanks.

Thanks for fast reply.
I’ve tried and got:

./deviceQuery 
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

NVMAP_IOC_QUERY_HEAP_PARAMS failed [Inappropriate ioctl for device]
cudaGetDeviceCount returned 999
-> unknown error
Result = FAIL

The issue is solved by replacing customized kernel with one from R32.4.2