On Mon, Jun 2, 2025 at 4:45 PM Richard Sandiford
<richard.sandif...@arm.com> wrote:
>
> "H.J. Lu" <hjl.to...@gmail.com> writes:
> > On Mon, Jun 2, 2025 at 4:07 PM Richard Sandiford
> > <richard.sandif...@arm.com> wrote:
> >>
> >> "H.J. Lu" <hjl.to...@gmail.com> writes:
> >> > Since not all CALL instructions in RTL passes have a REG_CALL_DECL note,
> >> > update get_call_fndecl to also check function symbol for function
> >> > declaration so that it can be used on CALL instructions like
> >> >
> >> > (call_insn 39 38 61 7 (set (reg:SI 0 ax)
> >> > (call (mem:QI (symbol_ref:DI ("foo") [flags 0x3]
> >> > <function_decl 0x7fffe96da900 foo>) [0 foo S1 A8])
> >> > (const_int 0 [0]))) "pr92080-15.c":24:9 1480 {*call_value}
> >> > (expr_list:REG_DEAD (reg:SI 5 di)
> >> > (expr_list:REG_DEAD (reg:SI 4 si)
> >> > (expr_list:REG_EH_REGION (const_int 0 [0])
> >> > (nil))))
> >> > (expr_list:SI (use (reg:SI 5 di))
> >> > (expr_list:SI (use (reg:SI 4 si))
> >> > (nil))))
> >> >
> >> > PR other/120494
> >> > * rtlanal.cc (get_call_fndecl): Also check function symbol to
> >> > get function declaration.
> >>
> >> What's your use case for this? I think we should instead move to making
> >> the notes more generally available, since looking at the call rtx won't
> >> work for targets that need indirect calls (e.g. for -mlong-calls).
> >
> > I am working on a pass which needs to check if CALL insn is a recursive
> > call.
>
> I assume false negatives are allowed? That is, "false" means "might be
> recursive, might not"?
Yes. For my pass, it only has performance impact.
> > Currently I have
> >
> > * Get the declaration of the function called by INSN. If INDIRECT_P
> > is true, also get indirect declaration. */
> >
> > static tree
> > get_call_fndecl_from_rtx (const rtx_insn *insn, bool indirect_p = false)
> > {
> > rtx note = find_reg_note (insn, REG_CALL_DECL, NULL_RTX);
> > if (note)
> > {
> > rtx datum = XEXP (note, 0);
> > if (datum)
> > return SYMBOL_REF_DECL (datum);
> > }
> > rtx call = get_call_rtx_from (insn);
> > rtx fnaddr = XEXP (call, 0);
> > rtx addr = XEXP (fnaddr, 0);
> > if (GET_CODE (addr) == SYMBOL_REF)
> > {
> > tree fndecl = SYMBOL_REF_DECL (addr);
> > if (fndecl && TREE_CODE (fndecl) == FUNCTION_DECL)
> > return fndecl;
> > }
> > return indirect_p ? MEM_EXPR (fnaddr) : nullptr;
> > }
> >
> > if (CALL_P (insn))
> > {
> > tree fndecl = get_call_fndecl_from_rtx (insn);
> > if (fndecl == current_function_decl
> > && decl_binds_to_current_def_p (fndecl))
> > {
> > recursive_call_p = true;
> > break;
> > }
> > }
> >
> > If get_call_fndecl works without a REG_CALL_DECL note, I can use it
> > instead of writing my own.
>
> I think we should consider removing the flag_ipa_ra guards around
> the code that adds REG_CALL_DECLs.
>
> Thanks,
> Richard
--
H.J.
