cudErrorNoKernelImageForDevice: no kernel image is available for execution on the device

[nicetaker]

Hello, I'm having the following problem:
when I run ./cukd_sample and other testing code，it will have some problem:

terminate called after throwing an instance of thrust: system:: system_error'
what(): parallel_for failed: cudErrorNoKernelImageForDevice: no kernel image is available for execution on the device.

The full output is below:
•/cukd_sample

generating 1000000 uniform random points
allocated memory for the world space bounding box ... calling builder...
terminate called after throwing an instance of thrust: system:: system_error'
what(): parallel_for failed: cudErrorNoKernelImageForDevice: no kernel image is available for execution on the device

My code environment is as follows:
wls2,ubuntu20.04，
cuda11.4,cudnn8.6.0,
4090

I compiled the code as with cmake as follows commod:
mkdir build
cd build
cmake ..
make

plz advise me on how ti fix this;
thankx!

[ingowald]

It means that you've compiled the code for a different GPU architecture than the one you're actually using. Ie., you're probably compiling for arch 70 (volta), but using something newer like a 4080 (arch=86) that doesn't support "70".

First, look up what GPU you are using, and what the CUDA arch for that GPU is; then in the cmake dialog set the "CMAKE_CUDA_ARCHITECTURE" to this value.

I'll also soon push a change where it lets cmake autodetect (newer versions can do that), but for now that should work.

[nicetaker]

Thank you very much！

I see your latest update.
It can work now.
it really helps me.
thank you again