https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112117
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2023-10-30
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. Peeling a single iteration might look like the easiest way here,
the trick is to see we can then eliminate an add and a PHI node in the loop.
Note with the current pass ordering this would come quite late since the
required store-motion is performed only before PRE. It could also be
seen as hoisting opportunity, transforming
# a_lsm.7_16 = PHI <c.0_1(3), a_lsm.7_6(2)>
_3 = c.0_1 + a_lsm.7_16;
into
tem1 = c.0_1 + c.0_1;
tem2 = c.0_1 + a_lsm.7_6;
# _3 = PHI <tem1(3), tem2(2)>
but the rest wouldn't reduce as nicely to a multiplication then.
