This patch add a new flag to let user to tell compiler that the
AutoFDO profile is accurate. So the compiler will assume function
without any sample is UNLIKELY_EXECUTED. This could save 10%~20% text
section size.
Bootstrapped and passed regression test.
OK for google-4_8 branch?
Thanks,
Dehao
Index: gcc/common.opt
===================================================================
--- gcc/common.opt (revision 203546)
+++ gcc/common.opt (working copy)
@@ -942,6 +942,10 @@ Common Joined RejectNegative Var(auto_profile_file
Use sample profile information for call graph node weights. The profile
file is specified in the argument.
+fauto-profile-accurate
+Common Report Var(flag_auto_profile_accurate) Optimization
+Whether to assume the sample profile is accurate.
+
; -fcheck-bounds causes gcc to generate array bounds checks.
; For C, C++ and ObjC: defaults off.
; For Java: defaults to on.
Index: gcc/predict.c
===================================================================
--- gcc/predict.c (revision 203546)
+++ gcc/predict.c (working copy)
@@ -2866,7 +2866,9 @@ compute_function_frequency (void)
|| (flag_auto_profile && profile_status == PROFILE_GUESSED))
{
int flags = flags_from_decl_or_type (current_function_decl);
- if (lookup_attribute ("cold", DECL_ATTRIBUTES (current_function_decl))
+ if (profile_info && flag_auto_profile_accurate)
+ node->frequency = NODE_FREQUENCY_UNLIKELY_EXECUTED;
+ else if (lookup_attribute ("cold", DECL_ATTRIBUTES
(current_function_decl))
!= NULL)
node->frequency = NODE_FREQUENCY_UNLIKELY_EXECUTED;
else if (lookup_attribute ("hot", DECL_ATTRIBUTES
(current_function_decl))
