Fatal error: cuda.h: No such file or directory

[andyluo7]

I am building the project with "make build" command.
The following error occurred.
Should I modify the Makefile or any environment parameter?

[ 20%] Building CXX object lwis/CMakeFiles/lwis.dir/src/lwis.cpp.o
In file included from /home/aluo/inference_results_v0.5/closed/NVIDIA/code/harness/lwis/src/lwis.cpp:17:0:
/home/aluo/inference_results_v0.5/closed/NVIDIA/code/harness/lwis/include/lwis.hpp:31:10: fatal error: cuda.h: No such file or directory
#include <cuda.h>
^~~~~~~~
compilation terminated.
lwis/CMakeFiles/lwis.dir/build.make:62: recipe for target 'lwis/CMakeFiles/lwis.dir/src/lwis.cpp.o' failed
make[4]: *** [lwis/CMakeFiles/lwis.dir/src/lwis.cpp.o] Error 1
make[4]: Leaving directory '/home/aluo/inference_results_v0.5/closed/NVIDIA/build/harness'
CMakeFiles/Makefile2:122: recipe for target 'lwis/CMakeFiles/lwis.dir/all' failed
make[3]: *** [lwis/CMakeFiles/lwis.dir/all] Error 2
make[3]: Leaving directory '/home/aluo/inference_results_v0.5/closed/NVIDIA/build/harness'
Makefile:83: recipe for target 'all' failed
make[2]: *** [all] Error 2
make[2]: Leaving directory '/home/aluo/inference_results_v0.5/closed/NVIDIA/build/harness'
Makefile:280: recipe for target 'build_harness' failed
make[1]: *** [build_harness] Error 2
make[1]: Leaving directory '/home/aluo/inference_results_v0.5/closed/NVIDIA'
Makefile:228: recipe for target 'build' failed
make: *** [build] Error 2

[nvpohanh]

@andyluo7 Did you run make build inside the container? If so, could you try nvidia-smi and ls /usr/local/cuda/include | grep cuda.h to see if nvidia-docker is running correctly? Thanks

[andyluo7]

@nvpohanh , no, i did not run it in the container. I try to run the inference on AGX Xavier.
Should I run it in container?

[nvpohanh]

Ah, sorry, didn't know that this was on Xavier. No, you don't need docker on Xavier.

Which JetPack did you install? We were using JetPack 4.3 DP.

[andyluo7]

@nvpohanh, I used the latest Jetpack 4.4 DP. Is there any difference?
There are multiple cuda.h files in the system.
I wonder if I should change the Makefile to add the include path.

[nvpohanh]

Yes, maybe they changed the path. Could you add that to /usr/local/cuda/include/cuda.h?

[andyluo7]

I have to hard coded #include </usr/local/cuda/include/cuda.h> in the .h and .c files in /inference_results_v0.5/closed/NVIDIA/code/harness/lwis to make it work.