[adrian_b]

This method is fine, but with it you must always take care to choose bufsize as a power of two, otherwise the overhead of having to do integer divisions for all queue operations may be noticeable.

[arinlen]

> but with it you must always take care to choose bufsize as a power of two, otherwise the overhead of having to do integer divisions

Keep in mind that one of the reasons behind the arbitrary "power of two" requirement is that the method proposed by the author abuses a single atomic integer to hold two shorts representing the indices for the head and tail elements.

I don't think it makes much sense to talk about overhead given this method requires doing a bunch of bitmasks to operate over the high and low short.

[adrian_b]

The overhead due to bit masks and rotations/shifts is negligible on almost all CPUs.

Larger CPUs can do many such operations during every clock cycle and even in very small CPUs they are usually single-cycle operations.

Depending on the CPU, integer divisions are 10 to 100 times slower, so in a completely different range of overhead.