https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80581
--- Comment #4 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
This is an roundoff issue. While summing the times it happens that n*1000/1000
> n. The following patch should silence it unless they cummulate for very
large value (which I will likely want to investigate then)
Index: ipa-inline-analysis.c
===================================================================
--- ipa-inline-analysis.c (revision 247436)
+++ ipa-inline-analysis.c (working copy)
@@ -3422,7 +3422,7 @@ estimate_node_size_and_time (struct cgra
min_size = (*info->entry)[0].size;
gcc_checking_assert (size >= 0);
gcc_checking_assert (time >= 0);
- gcc_checking_assert (nonspecialized_time >= time);
+ gcc_checking_assert ((nonspecialized_time - time).to_int () >= 0);
if (info->loop_iterations
&& !evaluate_predicate (info->loop_iterations, possible_truths))
I will commit it after regtesting.