CUDA 9.0 deprecated non-sync shuffle instructions

[ProfFan]

This PR fixes the following Warning:

ptxas /tmp/tmpxft_000010c5_00000000-5_WarpingSolver.ptx, line 677;
warning : Instruction 'shfl' without '.sync' is deprecated since PTX ISA version 6.0 and
will be discontinued in a future PTX ISA version

Tested on my Mac with LLVM 6.0 and latest Opt/Terra

I did not see performance regressions with the demo example applications.

Best,
Fan

[Mx7f]

Thanks! Unfortunately the .sync. operators weren't introduced until after CUDA 7.5 (Opt's currently recommended version). So either we need to detect the CUDA version and switch what code we use, or just wait to merge this until we get pre-built terra binaries compatible with CUDA 9 (terralang/terra#299 (comment)).

[ProfFan]

Bump @Mx7f :)