[Bug target/114252] Introducing bswapsi reduces code performance

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114252

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Note I do understand what you are saying, just the middle-end in detecting and
using __builtin_bswap32 does what it does everywhere else - it checks whether
the target implements the operation.

The middle-end doesn't try to actually compare costs (it has no idea of the
bswapsi costs), and it most definitely doesn't see how AVR is special in
having only QImode registers and thus the created SImode load (which the
target supports!) will end up as four registers.  To me a 'bswap' on
AVR never makes sense since whatever is swapped will be _always_ available
as a set of byte registers.

That's why I question AVR exposing bswapsi to the middle-end rather than
suggesting the middle-end should maybe see whether AVR has any regs of
HImode or larger.  Note that would break for targets that could eventually
do a load-multiple byteswapped to a set of QImode regs (guess there's no
such one in GCC at least), but it's the only heuristic that might work here.

The only thing that maybe would make sense with AVR exposing bswapsi is
users calling __builtin_bswap but since it always expands as a libcall
even that makes no sense.

So my preferred fix would be to remove bswapsi from avr.md?

Does it benefit from recognizing bswap done with shifts on an int?