https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66334

--- Comment #11 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to H.J. Lu from comment #10)
> (In reply to Vladimir Makarov from comment #9)
> > 
> > I will work on the patch and commit it on next week.
> > 
> > Thanks.
> 
> I tried this patch:
> 
> https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;
> h=ab377c74f283f3db51b4e250b9c7acecc32e8ff8
> 
> It seems to work.

Thanks, H.J.  Your patch fixes given problem but I'd like to play more safe. 
May be there are some analogous situations with explicit ebx setting.

So I am going to submit the following patch today or tomorrow.

Index: ChangeLog
===================================================================
--- ChangeLog   (revision 225200)
+++ ChangeLog   (working copy)
@@ -1,3 +1,10 @@
+2015-07-06  Vladimir Makarov  <vmaka...@redhat.com>
+
+       PR middle-end/66334
+       * ira-lives.c (process_bb_node_lives): Make conflicts with PIC
+       hard regno live at the start of BB with incoming abnormal edges.
+       * lra-lives.c (process_bb_lives): Ditto.
+
 2015-06-30  Vladimir Makarov  <vmaka...@redhat.com>

        PR debug/66691
Index: ira-lives.c
===================================================================
--- ira-lives.c (revision 225134)
+++ ira-lives.c (working copy)
@@ -1346,7 +1346,21 @@ process_bb_node_lives (ira_loop_tree_nod
             allocate such regs in this case.  */
          if (!cfun->has_nonlocal_label && bb_has_abnormal_call_pred (bb))
            for (px = 0; px < FIRST_PSEUDO_REGISTER; px++)
-             if (call_used_regs[px])
+             if (call_used_regs[px]
+#ifdef REAL_PIC_OFFSET_TABLE_REGNUM
+                 /* We should create a conflict of PIC pseudo with
+                    PIC hard reg as PIC hard reg can have a wrong
+                    value after jump described by the abnormal edge.
+                    In this case we can not allocate PIC hard reg to
+                    PIC pseudo as PIC pseudo will also have a wrong
+                    value.  This code is not critical as LRA can fix
+                    it but it is better to have the right allocation
+                    earlier.  */
+                 || (px == REAL_PIC_OFFSET_TABLE_REGNUM
+                     && pic_offset_table_rtx != NULL_RTX
+                     && REGNO (pic_offset_table_rtx) >= FIRST_PSEUDO_REGISTER)
+#endif
+                 )
                make_hard_regno_born (px);
        }

Index: lra-lives.c
===================================================================
--- lra-lives.c (revision 225200)
+++ lra-lives.c (working copy)
@@ -957,7 +957,18 @@ process_bb_lives (basic_block bb, int &c
         allocate such regs in this case.  */
       if (!cfun->has_nonlocal_label && bb_has_abnormal_call_pred (bb))
        for (px = 0; px < FIRST_PSEUDO_REGISTER; px++)
-         if (call_used_regs[px])
+         if (call_used_regs[px]
+#ifdef REAL_PIC_OFFSET_TABLE_REGNUM
+             /* We should create a conflict of PIC pseudo with PIC
+                hard reg as PIC hard reg can have a wrong value after
+                jump described by the abnormal edge.  In this case we
+                can not allocate PIC hard reg to PIC pseudo as PIC
+                pseudo will also have a wrong value.  */
+             || (px == REAL_PIC_OFFSET_TABLE_REGNUM
+                 && pic_offset_table_rtx != NULL_RTX
+                 && REGNO (pic_offset_table_rtx) >= FIRST_PSEUDO_REGISTER)
+#endif
+             )
            make_hard_regno_born (px, false);
     }

Reply via email to