I don't have CUDA-compliant hardware, so I can't test what I'm writing for its support, I'll ask Gentoo Bugzilla members about it after posting my ebuild there when it will be clean enough...
Currently nvidia-cuda-toolkit is available through Portage, and its ebuild seems to create a file /etc/env.d/99cuda containing:
Code:PATH=/opt/cuda/bin
ROOTPATH=/opt/cuda/bin
LDPATH=/opt/cuda/lib
MANPATH=/opt/cuda/man
It is then used by init scripts. PATH is added to users' PATH vars, and LDPATH is used to populate /etc/ld.so.conf.
But in ugene_globals.pri I see:
Code:UGENE_CUDA_LIB_DIR = $$(CUDA_LIB_PATH)
UGENE_CUDA_INC_DIR = $$(CUDA_INC_PATH)
So these 2 vars are still needed, but during build-time only, while CUDA_BIN_PATH and LD_LIBRARY_PATH aren't at all, right?