Comment # 11 on bug 1202027 from Martin Li������ka

Ok, so non-LTO bootstrap has a smaller diff, but still ~180 arc counters are
different:

$ diff -U0 /tmp/p1 /tmp/p2 | grep '\[count:' | wc -l
374

and I verified first different symbol assembly is:

 ;; Function htab_find_slot_with_hash (htab_find_slot_with_hash, funcdef_no=42,
decl_uid=3888, cgraph_uid=43, symbol_order=45)
...
where 4 edges are different:

 Read edge from 2 to 3, count:72182393
 Read edge from 3 to 4, count:15675
 Read edge from 5 to 6, count:0
 Read edge from 6 to 30, count:0
 Read edge from 7 to 8, count:15675
-Read edge from 8 to 9, count:25778269
+Read edge from 8 to 9, count:25752577
 Read edge from 10 to 14, count:0
-Read edge from 10 to 11, count:46404124
-Read edge from 11 to 12, count:46404124
+Read edge from 10 to 11, count:46429816
+Read edge from 11 to 12, count:46429816
 Read edge from 13 to 30, count:9482767
-Read edge from 17 to 18, count:50140237
+Read edge from 17 to 18, count:50145567
 Read edge from 21 to 22, count:0
-Read edge from 21 to 23, count:68578907
+Read edge from 21 to 23, count:68659229
 Read edge from 22 to 16, count:0
 Read edge from 25 to 30, count:6910902
 Read edge from 27 to 28, count:0
 Read edge from 27 to 29, count:55788724

which leads likely to a different inline decision. Note
htab_find_slot_with_hash is the non-template hash table used e.g. in libcpp.