https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92135
Bug ID: 92135
Summary: Implement popcountsi expansion for arm
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: ktkachov at gcc dot gnu.org
Target Milestone: ---
Target: arm*
On aarch64 we can do SImode and DImode popcount operations using the AdvSIMD
CNT instruction. As the comment in aarch64.md says:
;; Pop count be done via the "CNT" instruction in AdvSIMD.
;;
;; MOV v.1d, x0
;; CNT v1.8b, v.8b
;; ADDV b2, v1.8b
;; MOV w0, v2.b[0]
We should be able to do a similar thing on arm, using the VCNT instruction.
This just needs implementing as an expansion in arm.md.
int
foocnt (int a)
{
return __builtin_popcount (a);
}
is the trivial testcase.
Haven't thought too much about the DImode case, but perhaps that can also be
accelerated in similar ways
