How to pass the compilation options of the host compiler to NVCC

Accelerated Computing CUDA CUDA Programming and Performance
1

I have set up some CXXFLAGS, and when compiling .cpp files, these paths are valid. However, when compiling .cu files, I cannot find the header file, and the header file paths are all in CXXFLAGS. What should I configure to solve this problem

| /home/zxy/build2/tmp/work/armv8a-oe4t-linux/python3-pytorch/2.4.1/recipe-sysroot-native/usr/local/cuda-12.6/bin/..//include/crt/host_config.h:218:10: fatal error: features.h: No such file or directory
|   218 | #include <features.h> /* for __THROW */
|       |          ^~~~~~~~~~~~
| compilation terminated.
| Compiling       src/device/common.cu
2

As per documentation:

to specify an include search path, use either of these:

--include-path path
-I path

If the question is really about CMake’s handling of include paths, sorry, can’t help with that.

3

What is the nvcc command line? If you want to use environment variables, see 1. Introduction — NVIDIA CUDA Compiler Driver 12.9 documentation

4

Thanks, using the nvcc command directly should be relatively simple, but my question may involve cmake’s passing and handling of parameters, and I need to study it