> From: Eric Botcazou <botca...@adacore.com>
> Date: Fri, 11 Sep 2020 13:09:48 +0200
> > @@ -2618,6 +2643,16 @@ fill_slots_from_thread (rtx_jump_insn *insn, rtx
> > condition, lose = 1;
> > mark_set_resources (trial, &set, 0, MARK_SRC_DEST_CALL);
> > mark_referenced_resources (trial, &needed, true);
> > + if (filter_flags)
> > + {
> > + mark_set_resources (trial, &fset, 0, MARK_SRC_DEST_CALL);
> > +
> > + /* Groups of flags-register setters with users should not
> > + affect opportunities to move flags-register-setting insns
> > + (clobbers) into the delay-slot. */
> > + CLEAR_HARD_REG_BIT (needed.regs, targetm.flags_regnum);
> > + CLEAR_HARD_REG_BIT (fset.regs, targetm.flags_regnum);
> > + }
>
> Don't you need to put the new block before mark_referenced_resources (as I
> did
> in fill_simple_delay_slots) in case the needed insn reads the flags register?
I think the simplest answer is that wouldn't be good; it'd leave the
"CLEAR_HARD_REG_BIT (needed.regs, targetm.flags_regnum);" without
effect.
Note also fill_simple_delay_slots is moving insns from the before the
delay-slotting insn, while this is from a "tread" (or arm, or leg)
"after".
brgds, H-P
