https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108440
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I think it is a fuzzy area.
But as documented above simplify_rotate, those cases with two & (B - 1) masks
already result in rotate with & (B - 1) on the rotate count and then on various
targets that masking is simply ignored if the underlying hardware instruction
handles "out of bounds" rotate counts the only sane way.
So I bet we could add the & (B - 1) even for these cases as well as for the
planned
case where Y <= B rather than just Y < B.
