To allow compiled in virtuals we need to only build for actual sm archs.

CUDA currently doesn't support building for compute_ and having compiled in virtuals ( using separable compilation ). So we need to transition everything over to sm_

This merge request has been queued for testing. Test results may be viewed on:

• CDash (master)
• Buildbot (master) (only visible from inside Kitware)

Branch-at-master: d7ce601c

Do: merge

mentioned in commit de79765c

merged