Re: GCC build failed for native with your patch on 2002-03-14T14:15:04Z.

[Jan Hubicka <jh at suse dot cz>]

> With your recent patch, GCC does not compile on:
>  native
> Attached is build output for those targets.
> 
> The build failures are new.
> 
> For more information, see <http://people.redhat.com/geoffk/gcc-regression/>.
...
> +
> +	* recog.c (split_insn): Use delete_insn_and_edges.


Hi,
the problem is caused by the split_insn fix I made for darwin.  The problem is
that I've made split_insn to actually remove insn instead of patching it out
and I had to change split_insns check to use same way as other loops do -
verify whether we got to NEXT_INSN (bb->end).  This unfortuantely does not work
as the NEXT_INSN (bb->end) may be changed by try_split called via split_insn in
case new barrier is created.

This patch cleans up the exit condition so I hope it will work reliably now.
I've got past the problematic point in the bootstrap, but it is still in progress.

I've also found unrelated bug in try_split - delete_related_insns may actually
clobber the CFG.

OK for mainline/branch once testing finishes?
I am just starting the PPC bootstrap as well.

In case I will not get any negative reactions I will commit it as obvious to
the mainline to avoid breakage of bootstrap on i386.

I apologize for the breakage.
Honza

Thu Mar 14 17:48:38 CET 2002  Jan Hubicka  <jh@suse.cz>
	* emit-rtl.c (try_split): Use delete_insns.
	* recog.c (split_all_insns): Fix terminating condition.

Index: emit-rtl.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/emit-rtl.c,v
retrieving revision 1.256
diff -c -3 -p -r1.256 emit-rtl.c
*** emit-rtl.c	2002/03/09 12:53:17	1.256
--- emit-rtl.c	2002/03/14 16:42:14
*************** try_split (pat, trial, last)
*** 3076,3082 ****
  
  	  tem = emit_insn_after (seq, trial);
  
! 	  delete_related_insns (trial);
  	  if (has_barrier)
  	    emit_barrier_after (tem);
  
--- 3076,3082 ----
  
  	  tem = emit_insn_after (seq, trial);
  
! 	  delete_insn (trial);
  	  if (has_barrier)
  	    emit_barrier_after (tem);
  
Index: recog.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/recog.c,v
retrieving revision 1.141
diff -c -3 -p -r1.141 recog.c
*** recog.c	2002/03/14 10:10:10	1.141
--- recog.c	2002/03/14 17:08:10
*************** split_all_insns (upd_life)
*** 2735,2748 ****
      {
        basic_block bb = BASIC_BLOCK (i);
        rtx insn, next;
  
!       for (insn = bb->head; insn ; insn = next)
  	{
  	  rtx last;
  
  	  /* Can't use `next_real_insn' because that might go across
  	     CODE_LABELS and short-out basic blocks.  */
  	  next = NEXT_INSN (insn);
  	  last = split_insn (insn);
  	  if (last)
  	    {
--- 2735,2750 ----
      {
        basic_block bb = BASIC_BLOCK (i);
        rtx insn, next;
+       bool finish = false;
  
!       for (insn = bb->head; !finish ; insn = next)
  	{
  	  rtx last;
  
  	  /* Can't use `next_real_insn' because that might go across
  	     CODE_LABELS and short-out basic blocks.  */
  	  next = NEXT_INSN (insn);
+ 	  finish = (insn == bb->end);
  	  last = split_insn (insn);
  	  if (last)
  	    {
*************** split_all_insns (upd_life)
*** 2756,2768 ****
  	      changed = 1;
  	      insn = last;
  	    }
- 
- 	  if (next == NEXT_INSN (bb->end))
- 	    break;
  	}
- 
-       if (insn == NULL)
- 	abort ();
      }
  
    if (changed)
--- 2758,2764 ----