Hi Jakub, On Fri, Feb 10, 2017 at 08:50:36PM +0100, Jakub Jelinek wrote: > On the following testcases combiner during notes distribution sees a > REG_DEAD note and decides to remove the setter thereof. That register > has remembered a last value on that insn though, and it is a pseudo set > multiple times. Later on we combine some further insns into > pseudo = pseudo op something and use the last known value of the pseudo at > the already deleted insn (0) and therefore simplify it into pseudo = > something. > > Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for > trunk? > > 2017-02-10 Jakub Jelinek <ja...@redhat.com> > > PR rtl-optimization/79388 > PR rtl-optimization/79450 > * combine.c (distribute_notes): When removing TEM_INSN for which > corresponding dest has last value recorded, invalidate that last > value. > > * gcc.c-torture/execute/pr79388.c: New test. > * gcc.c-torture/execute/pr79450.c: New test. > > --- gcc/combine.c.jj 2017-01-30 09:31:48.000000000 +0100 > +++ gcc/combine.c 2017-02-10 17:05:57.500482518 +0100 > @@ -14288,6 +14288,11 @@ distribute_notes (rtx notes, rtx_insn *f > NULL_RTX, NULL_RTX, NULL_RTX); > distribute_links (LOG_LINKS (tem_insn)); > > + unsigned int regno = REGNO (XEXP (note, 0)); > + reg_stat_type *rsp = ®_stat[regno]; > + if (rsp->last_set == tem_insn) > + record_value_for_reg (XEXP (note, 0), NULL, > NULL_RTX); > + > SET_INSN_DELETED (tem_insn); > if (tem_insn == i2) > i2 = NULL;
This looks terribly ad hoc, but everything similar is handled the same way. Some day it will need to be made more robust, but the patch is okay for now. Thanks! Segher
