https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103699
--- Comment #17 from Petr <kobalicek.petr at gmail dot com> --- Guys thanks a lot for your feedback. Is the may_alias annotation guaranteed to behave as expected in the future versions of GCC too, or it's just too much UB that it's better to do unaligned reads with memcpy? What I like on my existing solution is that I can specify alignment, so for example I can say that this 64-bit load is 4-byte aligned, and that hints a compiler, I'm not sure how to hint that with memcpy. So far annotating the code with may_alias works for me, it passes tests, but I'm kinda unsure whether this is a good way to express unaligned memory access considering I have faced this issue.
