HAO CHEN GUI <guih...@linux.ibm.com> writes:
> Hi Richard,
>
> 在 2024/6/17 17:04, Richard Sandiford 写道:
>> I don't think we should keep the single_set condition after this change.
>> insn_cost can handle all instructions.
>
> Just tested with removing single_set condition. It causes some regressions.
> If the new_rtl is a debug insn, it still can do the replacement as it can
> be recog and it's not a single_set rtl. After removing single_set condition,
> the "ok" (recog) will be reset to false in the "if" block as the cost of
> debug insn is unknown. So the replacement won't be done.
>
> bool ok = recog (attempt, use_change);
> if (ok && !prop.changed_mem_p () && !use_insn->is_asm ())
> {
> bool strict_p = !prop.likely_profitable_p ();
> if (!change_is_worthwhile (use_change, strict_p))
> {
> if (dump_file)
> fprintf (dump_file, "change not profitable");
> ok = false;
> }
> }
>
> if (!ok)
> {
> /* The pattern didn't match, but if all uses of SRC folded to
> constants, we can add a REG_EQUAL note for the result, if there
> isn't one already. */
> if (!prop.folded_to_constants_p ())
We shouldn't be trying to cost changes to debug insns, so I think
the extra condition should be !use_insn->is_debug_insn ().
Thanks,
Richard
