https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122715
ktkachov at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |ktkachov at gcc dot gnu.org
Last reconfirmed| |2025-11-17
Summary|X86 Missed optimization of |Missed optimization of
|_mm_avg_*(x, x) --> x |IFN_AVG_FLOOR (x, x) --> x
Ever confirmed|0 |1
Keywords| |missed-optimization
Status|UNCONFIRMED |NEW
--- Comment #1 from ktkachov at gcc dot gnu.org ---
Confirmed, an equivalent aarch64 testcase is:
#include <arm_neon.h>
uint32x4_t
foo (uint32x4_t a, uint32x4_t b)
{
return vhaddq_u32 (a, a);
}
It can be fixed by adding folding of IFN_AVG_FLOOR (x, x) in match.pd IMO
