https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121218
--- Comment #13 from rguenther at suse dot de <rguenther at suse dot de> --- On Wed, 23 Jul 2025, rsandifo at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121218 > > --- Comment #12 from Richard Sandiford <rsandifo at gcc dot gnu.org> --- > On the REG_DEAD thing: REG_DEAD notes are not kept up-to-date, so any pass > that > wants to use them needs to ask for them to be recomputed. I see. I might have chased a wrong thing with the REG_DEAD note then, it appears correctly in the next pass which likely recomputes them. > The REG_DEAD usage > in scalar_chain::convert_insn_common suggests that STV should be calling > df_note_add_problem. Though that only updates an existing note and that could be removed if we do not need to keep those up to date. In any case the issue needs still to be tracked down given I wasn't yet able to pin-point a wrong transform. -fdisable-rtl-stv[12] doesn't fix it btw, but there are a few insn patterns guarded by TARGET_STV, like the move pattern enabling some more alternatives and two define_insn_and_split.
