https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118505
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |pinskia at gcc dot gnu.org
Target| |aarch64
Keywords| |missed-optimization
Ever confirmed|0 |1
Status|UNCONFIRMED |NEW
Component|rtl-optimization |tree-optimization
Last reconfirmed| |2025-01-16
Target Milestone|--- |15.0
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the question is doing:
if (a)
{
b = c * c;
...
}
else
{
... (repeated the same as the ... above)
}
is faster than say:
t = c * c;
b = a ? t : a;
And is it dependent on the type. In this case it is a floating point type.
Does changing real_t to say int change the performance way way or another?
if so we should also check the type of lhs.
One other thing we could do here is add a target hook to get a decent cost
model here and pass the type and tree code down.
Note poor_ifcvt_candidate_code most likely should include RDIV_EXPR (but that
is not the issue here).
