https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119147
Bug ID: 119147
Summary: 525.x264_r is approx. slower with LTO+PGO than without
(at -Ofast -march-native)
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: middle-end
Assignee: unassigned at gcc dot gnu.org
Reporter: hubicka at gcc dot gnu.org
Target Milestone: ---
This seems to be at least partly caused by fact that ipa-cp does not clone
function with no hot calls. This is wrong. Since the function itself may spend
a lot of time, we do not want to give up sine it is called just few times.
The cost model should consider the expected peedup after cloning. I.e.
time_benefit multiplied by sum of counts of call edges.
