[Bug rtl-optimization/51447] [4.6/4.7/4.8 Regression] global register variable definition incorrectly removed as dead code

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51447

--- Comment #12 from Steven Bosscher <steven at gcc dot gnu.org> 2012-11-11 22:05:37 UTC ---
(In reply to comment #7)
> Created attachment 26100 [details]
> gcc47-pr51447.patch
> 
> While that fixes the testcase at -O1, at -O2 and above it still fails. 
> Apparently ud_dce pass removes it.

I think DCE should never delete an insn setting a global reg. So in
addition to your patch, we'd need the following patch:

Index: dce.c
===================================================================
--- dce.c       (revision 193411)
+++ dce.c       (working copy)
@@ -98,6 +98,11 @@ deletable_insn_p (rtx insn, bool fast, b
   rtx body, x;
   int i;

+  /* If INSN sets a global_reg, leave it untouched.  */
+  for (df_ref *def_rec = DF_INSN_DEFS (insn); *def_rec; def_rec++)
+    if (global_regs[DF_REF_REGNO (*def_rec)])
+      return false;
+  
   if (CALL_P (insn)
       /* We cannot delete calls inside of the recursive dce because
         this may cause basic blocks to be deleted and this messes up