read __prof_trigger in kernel?

The CUDA 6.5 documentation says __prof_trigger() is of type void, ie does not return a value.
Is it possible (perhaps via PTX???) to read the hardware performance counters.
They can be increased (using __prof_trigger) by a kernel, but can that kernel read them back?

Any comments or suggestions would be most welcome
Bill
http://www.cs.ucl.ac.uk/staff/W.Langdon/