https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123133
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|middle-end |target
Severity|normal |enhancement
Status|UNCONFIRMED |NEW
Last reconfirmed| |2025-12-15
Ever confirmed|0 |1
Target| |aarch64
--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
on aarch64 we have:
;; positive extension: (needed tie breaker) unsigned cost: 4; signed cost: 4
While x86_64 we have:
;; positive extension: unsigned cost: 1; signed cost: 4
I think the cost for zero-extend from SI to DI should be lower cost than sign
extend because it is just a "mov" would in many cases be optimized away and/or
be a mov instruction which then will be free with many cores.
