Are these stronger guarantees actually needed?

[thirtythreeforty]

This library is awesome, thanks!

Quick performance question. This code:

waitfree-mpsc-queue/mpsc.c

Lines 10 to 12 in 551e2b2

	#define memory_order_release memory_order_seq_cst
	#define memory_order_acquire memory_order_seq_cst
	#define memory_order_relaxed memory_order_seq_cst

makes sure that all atomic operations happen with the maximum amount of synchronization. Is this required, or is this code just erring on the side of caution?

[dbittman]

It's erring on the side of caution. I have not fully verified the correctness of the relaxed ordering (though I believe it's right, according to a reading of the standard that I did when I wrote it). I have found that on x86 it makes little difference, and on armv8 there is a very small (but statistically significant) performance increase when using the relaxed orderings. Additionally, there are lines in the relaxed orderings that could possibly be relaxed further, if of course the code is correct with the current relaxed orderings.