Re: [PATCH] [i386] Fix _mm256_zeroupper to notify LRA that vzeroupper will kill sse registers. [PR target/82735]

[Hongtao Liu <crazylht@gmail.com>]

On Tue, Jun 1, 2021 at 10:22 AM Hongtao Liu <crazylht@gmail.com> wrote:
>
> On Thu, May 27, 2021 at 6:50 PM Richard Sandiford
> <richard.sandiford@arm.com> wrote:
> >
> > Jakub Jelinek <jakub@redhat.com> writes:
> > > On Thu, May 27, 2021 at 01:07:09PM +0800, Hongtao Liu via Gcc-patches wrote:
> > >> +  /* Flag used for call_insn indicates it's a fake call.  */
> > >> +  RTX_FLAG (insn, used) = 1;
> > >
> > >> +      /* CALL_INSN use "used" flag to indicate it's a fake call.  */
> > >> +      if (i == STACK_POINTER_REGNUM
> > >> +      && !RTX_FLAG (insn_info->insn, used))
> > >
> > >> -      && ! SIBLING_CALL_P (insn))
> > >> +      && ! SIBLING_CALL_P (insn)
> > >> +      && !RTX_FLAG (insn, used))
> > >
> > >> -      /* For all other RTXes clear the used flag on the copy.  */
> > >> -      RTX_FLAG (copy, used) = 0;
> > >> +      /* For all other RTXes clear the used flag on the copy.
> > >> +     CALL_INSN use "used" flag to indicate it's a fake call.  */
> > >> +      if (!INSN_P (orig))
> > >> +    RTX_FLAG (copy, used) = 0;
> > >>        break;
> > >>      }
> > >>    return copy;
> > >> @@ -57,7 +57,8 @@ requires_stack_frame_p (rtx_insn *insn, HARD_REG_SET prologue_used,
> > >>    HARD_REG_SET hardregs;
> > >>    unsigned regno;
> > >>
> > >> -  if (CALL_P (insn))
> > >> +  /* CALL_INSN use "used" flag to indicate it's a fake call.  */
> > >> +  if (CALL_P (insn) && !RTX_FLAG (insn, used))
> > >>      return !SIBLING_CALL_P (insn);
> > >
> > > Please define a macro for this in rtl.h (and mention it above used;
> > > member too in a comment, see all the other comments in there), like:
> > > /* 1 if RTX is a call_insn for a fake call.  */
> > > #define FAKE_CALL_P(RTX)                                      \
> > >   (RTL_FLAG_CHECK1 ("FAKE_CALL_P", (RTX), CALL_INSN)->used)
> Changed.
> > > Though, I'm also not sure if used can be actually used for this,
> > > because it is used e.g. in emit-rtl.c for verification of RTL sharing.
> >
> > I thought it should be OK, since:
> >
> > - copy_rtx_if_shared_1 and mark_used_flags do nothing for insns
> > - verify_rtx_sharing is only called for parts of an insn, rather than
> >   an insn itself
> >
> > I guess an alternative would be to add a new rtx_code for fake call
> > insns and use CALL_P to test for both.  However, that would lose the
> > property that the default behaviour is conservatively correct
> > (even for direct checks of CALL_INSN), so the flag IMO seems better.
> >
> > Thanks,
> > Richard
> >
> > > Though, it seems no other rtl flag is free for CALL_INSN.
> > > Could this fake call flag sit on the CALL rtx instead?
> > >
> > >       Jakub
>
> Updated separate patch for the middle-end part.

gcc/ChangeLog

        PR target/82735
        * df-scan.c (df_get_call_refs): When call_insn is a fake call,
        it won't use stack pointer reg.
        * final.c (leaf_function_p): When call_insn is a fake call, it
        won't affect caller as a leaf function.
        * reg-stack.c (callee_clobbers_any_stack_reg): New.
        (subst_stack_regs): When call_insn doesn't clobber any stack
        reg, don't clear the arguments.
        * rtl.c (shallow_copy_rtx): Don't clear flag used when orig is
        a insn.
        * shrink-wrap.c (requires_stack_frame_p): No need for stack
        frame for a fake call.
        * rtl.h (FAKE_CALL_P): New macro.

>
> --
> BR,
> Hongtao



-- 
BR,
Hongtao