> Yes, but I'd expect them to be optimized away (well, hopefully). OK, but you cannot reasonably expose everything in GENERIC/GIMPLE, for example the mask-and-shift operations to extract bitfields in reverse SSO, only the RTL expander has the (quite complex) logic and I doubt you want to teach the GIMPLE optimizers about this low-level mess.
> Anwyay, and I thought use of the feature would be rare so that "every load > and store" is still very few? Experience has shown that, when users start to use it, they use it extensively (too extensively if you ask me) so the implementation cannot be too dumb. -- Eric Botcazou
