🚀 go-pugleaf

Bug in scheduler - abort() in sched_analyze_2 () at line #1223 in sched.c

I am using gcc from 25-March-92 configured for i386v4.

For the function cppp.c/collect_expansion() compiled -O -fschedule-insns.
(Note I have enabled scheduling - i.e. defined functional units).
We have the following sequence.


(insn 663 662 664 (set (cc0)
       (compare (reg/v:SI 31)
           (reg/v:SI 72))) -1 (nil)
   (nil))

(note 664 663 665 "" NOTE_INSN_DELETED)

(insn 665 664 667 (set (mem/s:QI (plus:SI (reg/v:SI 79)
               (const_int 5)))
       (eq:QI (cc0)
           (const_int 0))) 135 {seq} (nil)
   (nil))


sched.c/sched_analyze_2()  is called with :
 insn = insn 665   and   x = cc0
And this causes an abort in sched_analyze_2() as follows:

	/* User of CC0 depends on immediately preceding insn.
	   All notes are removed from the list of insns to schedule before we
	   reach here, so the previous insn must be the setter of cc0.  */
	if (GET_CODE (PREV_INSN (insn)) != INSN)
	  abort ();

The comment states that the note (664) is remove before we get here,
but from what I can see this is done by the procedure unlink_notes().
Now unlink_notes() is called after sched_analyze() in schedule_block(), 
and sched_analyze() calls sched_analyze_2().

Regards,
Tevi.

tevi@iilcad.intel.com
972-4-355675