[Bug tree-optimization/78821] GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78821

--- Comment #23 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Uroš Bizjak from comment #22)
> (In reply to Jakub Jelinek from comment #21)
> > Created attachment 42660 [details]
> > gcc8-pr78821-i386.patch
> > 
> > Untested patch for the -mmovbe movbew loads/stores.  Note, is there any
> > particular reason why we don't use for AT&T syntax w/l/q suffixes for movbe?
> > At least looking at binutils testsuite, it covers both movbe without suffix
> > and movbe[wlq].  On the other side, for bswap insn there are no suffixes.
> 
> No particular reason - we should add suffixes.

Ok, will add them incrementally.

> OTOH, proposed patch should use enabled attribute to enable xchg, and split
> the pattern after reload to lose FLAGS_REG clobber for xchg and movbe
> aliternatives.

I believe enabled attribute can't depend on 
optimize_function_for_size_p (cfun).
Splitting after reload is something I can certainly do.