This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix DCE REG_LIBCALL note moving from noop move insns (PR rtl-optimization/33644)
> But omitting those side effects is, from the POV of a conforming
> program, fine. Everywhere except in the dynamic linker during TLS
> setup, at least. Can't we get away with marking it const or pure?
Usually you do in that sort of situation. An analogy is a memoized
function: technically it's not const or pure, but in practice you can
treat it as so.