https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118994
Bug ID: 118994
Summary: GCC fails to optimize (a >> 1) + (b >> 1) + ((a | b) &
1) to PAVGB/PAVGW (or equivalent instruction)
Product: gcc
Version: 14.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: john_platts at hotmail dot com
Target Milestone: ---
SSE2 PAVGB/PAVGW, NEON/SVE2 SRHADD/URHADD, AltiVec
vavgsb/vavgsh/vavgsw/vavgub/vavguh/vavguw, and LSX/LASX vavgr is equivalent to
the following expression for each lane of an integer vector:
(a >> 1) + (b >> 1) + ((a | b) & 1)
A snippet that shows that GCC fails to optimize the above expression down to
PAVGB/PAVGW (or an equivalent instruction) where supported on the target can be
found at https://godbolt.org/z/4o6c5e1hj.
