cuda - How can I clear the L2 cache (and the TLB) of a GPU? -

i have discrete nvidia gpu (say, kepler or maxwell). want clear l2 cache before kernel scheduled, not taint test results.

i allocate large slab of memory , read sequentially lot of that's someplace far away, , work. i'd rather simpler...

notes:

• i'm interested in how in opencl, albeit less so.
• ptx inlining acceptable (but i'd rather write proper code).