From what I understand, this is the way you'd do it with hardware registers (maintain the read and write indices each with one extra MSB to detect the difference between full/empty).

We've been using similar code in PortAudio since the late 90s[0]. I'm pretty sure Phil Burk got the idea from his hardware work.

[0] https://app.assembla.com/spaces/portaudio/git/source/master/...