cmake add_custom_command() is not passing the ccbin flag to nvcc for the Fatbin and PTX cuda samples

[philmcm]

Using Version 13.3 of CUDA Toolkit on Fedora 44. The default GCC/G++ is version 16 and is unsupported by the cuda toolkit (> 15 not supported)

After generation of makefiles using the following command which passes older GCC/G++ compiler versions:
cmake -D CMAKE_C_COMPILER=gcc-15 -D CMAKE_CXX_COMPILER=g++-15 -D CMAKE_CUDA_HOST_COMPILER:STRING=g++-15 ..

nvcc is not receiving the correct compiler version via the cmake custom commands in a few samples. This only affects the samples involving the cmake custom commands used by fatbin and PTX related code. All other samples compile fine and run correctly.

I've introduced a workaround to the each of the affected CMakeLists.txt files to pass the compiler version:

Changed from:
add_custom_command(
...
COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}
...
)

To:
COMMAND ${CMAKE_CUDA_COMPILER} ${INCLUDES} ${ALL_CCFLAGS} -Wno-deprecated-gpu-targets -ccbin=${CMAKE_CUDA_HOST_COMPILER} ${GENCODE_FLAGS} -o ${CUDA_FATBIN_FILE} -fatbin ${CUDA_KERNEL_SOURCE}

Although I think there might be a more appropriate option to pass in this case.

The affected examples are:
0_Introduction/
matrixMulDrv
simpleDrvRuntime
simpleTextureDrv
vectorAddDrv
vectorAddMMAP
2_Concepts_and_Techniques/
threadMigration
3_CUDA_Features/
memMapIPCDrv
ptxjit