Hi, when splitting functions, tree-inline determined correctly entry count of the new function part, but then in case entry block of new function part is in a loop it scales body which is not suposed to happen.
Bootstrapped/regtested x86_64-linux, comitted. * tree-inline.cc (copy_cfg_body): Fix profile of split functions. diff --git a/gcc/tree-inline.cc b/gcc/tree-inline.cc index b2570517343..dee2dfc2620 100644 --- a/gcc/tree-inline.cc +++ b/gcc/tree-inline.cc @@ -3084,7 +3084,7 @@ copy_cfg_body (copy_body_data * id, /* Register specific tree functions. */ gimple_register_cfg_hooks (); - /* If we are inlining just region of the function, make sure to connect + /* If we are offlining region of the function, make sure to connect new entry to ENTRY_BLOCK_PTR_FOR_FN (cfun). Since new entry can be part of loop, we must compute frequency and probability of ENTRY_BLOCK_PTR_FOR_FN (cfun) based on the frequencies and @@ -3093,12 +3093,14 @@ copy_cfg_body (copy_body_data * id, { edge e; edge_iterator ei; - den = profile_count::zero (); + ENTRY_BLOCK_PTR_FOR_FN (cfun)->count = profile_count::zero (); FOR_EACH_EDGE (e, ei, new_entry->preds) if (!e->src->aux) - den += e->count (); - ENTRY_BLOCK_PTR_FOR_FN (cfun)->count = den; + ENTRY_BLOCK_PTR_FOR_FN (cfun)->count += e->count (); + /* Do not scale - the profile of offlined region should + remain unchanged. */ + num = den = profile_count::one (); } profile_count::adjust_for_ipa_scaling (&num, &den);