On Wed, Sep 29, 2021 at 01:13:46PM +0100, Jonathan Wakely via Gcc-patches wrote:
> But I think that's OK, as I think the built-in is
> smart enough to be a no-op for types with no padding.
Yes. The only effect it will have is that during the initial optimization
passes the variable/parameter will be addressable where without the call and
__addressof it wouldn't be; but as soon as __builtin_clear_padding is folded
to nothing if there is no padding or something if there is (which happens
quite early) and TODO_update_address_taken is done at the end
of some pass, it will no longer be addressable (unless something different
keeps taking its address).
Jakub
