https://gcc.gnu.org/g:1efd525e828d5c30d0962d05ba81f9d7c8176ca7
commit r16-291-g1efd525e828d5c30d0962d05ba81f9d7c8176ca7
Author: Jan Hubicka <hubi...@ucw.cz>
Date: Tue Apr 29 22:43:45 2025 +0200
Fix cs_interesting_for_ipcp_p wrt flag_profile_partial_training.
As noticed by Martin Jambor, I introduced a bug while simplifying
cs_interesting_for_ipcp_p and reversed condition for
flag_profile_partial_training. Also I noticed that we probably want to
consider calls with unintialized counts for cloning so the pass does
somehting
with -fno-guess-branch-probability even thugh this is probably not very
useful
in practice.
gcc/ChangeLog:
* ipa-cp.cc (cs_interesting_for_ipcp_p): Fix handling of
uninitialized
counts and 0 IPA cost wrt flag_profile_partial_training.
Diff:
---
gcc/ipa-cp.cc | 10 ++++------
1 file changed, 4 insertions(+), 6 deletions(-)
diff --git a/gcc/ipa-cp.cc b/gcc/ipa-cp.cc
index b4b96997d750..f7e5aa9bfd5c 100644
--- a/gcc/ipa-cp.cc
+++ b/gcc/ipa-cp.cc
@@ -545,14 +545,12 @@ cs_interesting_for_ipcp_p (cgraph_edge *e)
return true;
/* If local (possibly guseed or adjusted 0 profile) claims edge is
not executed, do not propagate. */
- if (!e->count.nonzero_p ())
+ if (e->count.initialized_p () && !e->count.nonzero_p ())
return false;
- /* If IPA profile says edge is executed zero times, but zero
- is quality is ADJUSTED, still consider it for cloning in
- case we have partial training. */
+ /* If we have zero IPA profile, still consider edge for cloning
+ in case we do partial training. */
if (e->count.ipa ().initialized_p ()
- && opt_for_fn (e->callee->decl,flag_profile_partial_training)
- && e->count.nonzero_p ())
+ && !opt_for_fn (e->callee->decl,flag_profile_partial_training))
return false;
return true;
}
