CHECK_DEAD_NOTES abort in sched1 for sh-elf

["Sanjiv Kumar Gupta, Noida" <sanjivg at noida dot hcltech dot com>]

Hi,

For the following test case,

struct frame {
    int stereo;
    int single;
};

extern int a ();

int do_layer2(struct frame *fr)
{
  int clip=0;
  int j;
  int stereo = fr->stereo;
  int single = fr->single;

  if(stereo == 1 || single == 3)
    single = 0;

    for (j=0;j<3;j++)
    {
      if(single >= 0)
      {
        clip += a();
      }
    }
  return clip;
}

CHECK_DEAD_NOTES is getting an extra REG_DEAD note after
sched1 calls update_life_info.

Here are the rtl dumps.

test.c.20.life
---------------
(insn 114 113 115 0 test.c:16 (set (reg:SI 181)
        (const_int -1 [0xffffffff])) 123 {movsi_ie} (nil)
    (nil))

(insn 115 114 116 0 test.c:16 (parallel [
            (set (reg:SI 180)
                (neg:SI (plus:SI (reg:SI 147 t)
                        (reg:SI 181))))
            (set (reg:SI 147 t)
                (ne:SI (ior:SI (reg:SI 147 t)
                        (reg:SI 181))
                    (const_int 0 [0x0])))
        ]) 94 {negc} (insn_list 113 (insn_list 114 (nil)))
    (expr_list:REG_DEAD (reg:SI 181)
        (expr_list:REG_UNUSED (reg:SI 147 t)
            (nil))))

test.c.21.combine
-----------------
(note 114 113 115 0 NOTE_INSN_DELETED)

(insn 115 114 116 0 test.c:16 (set (reg:SI 182)
        (eq:SI (reg:SI 147 t)
            (const_int 1 [0x1]))) 212 {movt} (insn_list 113 (nil))
    (nil))

test.c.24.sched
-----------------
(insn:HI 115 113 116 0 test.c:16 (set (reg:SI 182)
        (eq:SI (reg:SI 147 t)
            (const_int 1 [0x1]))) 212 {movt} (insn_list 113 (nil))
    (expr_list:REG_DEAD (reg:SI 147 t)		--> extra REG_DEAD note. <--

        (nil)))

I am not sure whether this is a problem with distribute_notes in combiner,
or update_life_info in sched1.

recompute_reg_usage before sched1 does not pass PROP_DEATH_NOTES flag to
update_life_info. But if I do that it solves the problem.

The bug can be reproduced by enabling the sched1 in sh.h.

Any pointers to this would be a great help.

Thanks
Sanjiv