[RFC] Consider lrotate const rotation in vectorizer

[Segher Boessenkool]

On Wed, Jul 17, 2019 at 01:32:50PM +0200, Richard Biener wrote:
> On Wed, Jul 17, 2019 at 12:54 PM Jakub Jelinek <jakub@redhat.com> wrote:
> Ick.  I wouldn't even touch SHIFT_COUNT_TRUNCATED with a 10-foot pole
> here.  And for rotates we can simply always truncate constant amounts to
> the rotated operands width, no?  For non-constant ones I fear targets
> would need to support both to get reliable expansion.

SHIFT_COUNT_TRUNCATED has no meaning for rotate instructions, as far as
I can see.  Mathematically it doesn't, and are there any CPUs that fail
for it for no reason?  Any archs where some rotate amounts give
undefined results?  So if this is true, SHIFT_COUNT_TRUNCATED can be
treated as always *on* for rotates.


Segher