x11-drivers/nvidia-drivers-290.10-r1 installs libcuda.so into ${CL_ROOT} and an explicit 'eselect opencl set nvidia' call is necessary to make CUDA usable. This is weird, as OpenCL and CUDA, while similar, are two different, independent APIs. Also, it is presently very unlikely that we will get CUDA implementations other than the one from NVIDIA. What's the reason for coupling OpenCL and CUDA in this way? Can't libcuda.so go into /usr/lib*, and the opencl stuff in ${CL_ROOT}?
(In reply to comment #0) > x11-drivers/nvidia-drivers-290.10-r1 installs libcuda.so into ${CL_ROOT} and an > explicit 'eselect opencl set nvidia' call is necessary to make CUDA usable. > This is weird, as OpenCL and CUDA, while similar, are two different, > independent APIs. Also, it is presently very unlikely that we will get CUDA > implementations other than the one from NVIDIA. > > What's the reason for coupling OpenCL and CUDA in this way? Can't libcuda.so > go into /usr/lib*, and the opencl stuff in ${CL_ROOT}? Yeah, currently there's no reason for libcuda.so to go into CL_ROOT. I've reverted that change +*nvidia-drivers-290.10-r2 (15 Feb 2012) +*nvidia-drivers-295.20-r1 (15 Feb 2012) + + 15 Feb 2012; Kacper Kowalik <xarthisius@gentoo.org> + +nvidia-drivers-290.10-r2.ebuild, +nvidia-drivers-295.20-r1.ebuild, + -nvidia-drivers-290.10-r1.ebuild, -nvidia-drivers-295.20.ebuild: + Move libcuda and libnvidia-compiler back to /usr/lib since they're not + creating any conflicts now. That will allow to use CUDA without explicit + eselect opencl. Fixes bug 403495 by Michal Januszewski <spock@gentoo.org>
