The point of that code, using atomic, was to show how slow it is not to enhance performance when threaded… see the benchmark output:
Atomic operation is 10x slower… benchmark increments a single location in an array 1000 times
Quite some discussion like this and this, that atomic adds are very slow