Summary: GCC incorrectly puts function in .text.unlikely
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization

Created attachment 21875
Test case

If you compile the attached test case (x.c) with -O2 on ToT sources you will
find that init_target_chars is put into .text.unlikely instead of .text.  I
have verified this on IA64 HP-UX and Linux and on X86 Linux.  I do not think
this function should be put into .text.unlikely in this case and GCC 4.5 does
not do so.

It looks like this started with Honza's partial inlining change (r161382).

The following patch seems to fix the problem but I am not sure if it is the
correct fix or not.  Should node->frequency always be set in this case or
should the original value be preserved if no attribute is seen.  I do not
believe node->frequency is uninitialized because if I change the frequeny
enum in cgraph.h to make NODE_FREQUENCY_NORMAL a 0 value, the frequency
is still set to NODE_FREQUENCY_UNLIKELY_EXECUTED (unless I change predict.c).

Index: predict.c
--- predict.c   (revision 164573)
+++ predict.c   (working copy)
@@ -2204,6 +2204,8 @@ compute_function_frequency (void)
       else if (DECL_STATIC_CONSTRUCTOR (current_function_decl)
               || DECL_STATIC_DESTRUCTOR (current_function_decl))
         node->frequency = NODE_FREQUENCY_EXECUTED_ONCE;
+      else 
+        node->frequency = NODE_FREQUENCY_NORMAL;

Configure bugmail:
------- You are receiving this mail because: -------
You are on the CC list for the bug.

Reply via email to