Write out process.

<p>The thread id indicates (based on the computation _popc(B)) whether a given thread is writing out an element less than the pivot or greater than or equal to the pivot. The values and , which are maintained in shared memory and updated incrementally, indicate the location in the global array of the location of the last element that is less than the pivot and greater than or equal to the pivot. This operation involves at most two coalesced memory writes.</p>