support to use different host compiler for CUDA about umpire HOT 5 CLOSED

I believe that the correct thing to do would be to specify -DCMAKE_CXX_COMPILER=/path/to/gcc5/bin. CMake should then correctly set the host compiler.

What version of CMake are you using?

from umpire.

I am using cmake3.12.1. I tried this by setting -DCMAKE_CXX_COMPILER=/path/to/gcc5/bin. But enable_language(cuda) seems to still use the gcc8 I have in my environment. I will try to load gcc5 to my environment as well.

from umpire.

If you set CMAKE_CXX_COMPILER to be the g++ 5 executable, does it work?

from umpire.

Hi David,
set CMAKE_CXX_COMPILER to g++5 does not help.
set CMAKE_CUDA_HOST_COMPILER to g++5 fixed the issue.
I think it is CMake's issue by not picking up CUDA_NVCC_FLAGS in the enable_language(cuda) function.
I will close this issue, thank you for you help!

from umpire.

I overlooked this the first time, but the flag variable for CUDA flags changed - it's now CMAKE_CUDA_FLAGS. Maybe that would be another fix.

from umpire.