Hi,
as noticed by Martin's new heatmaps default_function_sections disables
test.unlikely and text.startup sections during LTO. This was meant to
be only with tp_first_run profiling. The hot section is still useful
since it does "poor man's clustering" so I kept only the test to disable
startup subsection with corrected fix.
Bootstrapped/regtested x86_64-linux and checked that it works well with
heatmap.
Honza
* varasm.c (default_function_section): Fix confused tests for
tp_first_run reordering.
Index: varasm.c
===================================================================
--- varasm.c (revision 279167)
+++ varasm.c (working copy)
@@ -589,9 +589,13 @@ default_function_section (tree decl, enu
where we can split away unnecessary parts of static constructors. */
if (startup && freq != NODE_FREQUENCY_UNLIKELY_EXECUTED)
{
- /* If we do have a profile or(and) LTO phase is executed, we do not need
- these ELF section. */
- if (!in_lto_p || !flag_profile_values)
+ /* During LTO the tp_first_run profiling will naturally place all
+ initialization code first. Using separate section is counter-productive
+ because startup only code may call functions which are no longer
+ startup only. */
+ if (!in_lto_p
+ || !cgraph_node::get (decl)->tp_first_run
+ || !opt_for_fn (decl, flag_profile_reorder_functions))
return get_named_text_section (decl, ".text.startup", NULL);
else
return NULL;
@@ -607,10 +611,7 @@ default_function_section (tree decl, enu
case NODE_FREQUENCY_UNLIKELY_EXECUTED:
return get_named_text_section (decl, ".text.unlikely", NULL);
case NODE_FREQUENCY_HOT:
- /* If we do have a profile or(and) LTO phase is executed, we do not
need
- these ELF section. */
- if (!in_lto_p || !flag_profile_values)
- return get_named_text_section (decl, ".text.hot", NULL);
+ return get_named_text_section (decl, ".text.hot", NULL);
/* FALLTHRU */
default:
return NULL;
