__any_sync for exchanging data between threads in a warp: broadcast

Note: srcLane is the lane index within a group of width w

int value = __any_sync(mask,v,srcLane,w)
Broadcasts the value of variable v in thread srcLane to the other threads; which threads take part is determined by the bits of mask

int value = __any_sync(mask,tid,2,w);
// Here thread 2's tid value (which is also 2) is broadcast, so value is 2 for every thread that reaches this point; if tid meant something else, each other thread would use the corresponding value
Output: `shfl:  2  2  2  2  2  2  2  2 10 10 10 10 10 10 10 10`
The 10 appears because that thread's actual tid is 10, but within its group its lane index is 2

__shfl_xor_sync(FULL_MASK, tid, laneMask, WIDTH);
Bitwise XOR: setting laneMask to 1 swaps the tid values of each pair of adjacent threads