explore -cl-uniform-work-group-size compiler option

[pkienzle]

Docs says:

This requires that the global work-size be a multiple of the work-group size specified to clEnqueueNDRangeKernel. Allow optimizations that are made possible by this restriction.

I believe we are already doing this since some drivers were failing when the length of the q array was not a multiple of 32. Instead we check whether we are beyond the end of the array when we enter the kernel. If other drivers can benefit from this with a speed improvement or smaller code size it should be easy to implement.