[2/2] Mark untyped calls and handle them specially [PR98689]

This patch fixes a regression introduced by the rtl-ssa patches.
It was seen on HPPA but it might be latent elsewhere.

The problem is that the traditional way of expanding an untyped_call
is to emit sequences like:

   (call (mem (symbol_ref "foo")))
   (set (reg pseudo1) (reg result1))
   ...
   (set (reg pseudon) (reg resultn))

The ABI specifies that result1..resultn are clobbered by the call but
nothing in the RTL indicates that result1..resultn are the results
of the call.  Normally, using a clobbered value gives undefined results,
but in this case the results are well-defined and matter for correctness.

This seems like a niche case, so I think it would be better to mark
it explicitly rather than try to detect it heuristically.

Note that in expand_builtin_apply we already have an rtx_insn *,
so it doesn't matter whether we call emit_call_insn or emit_insn.
Calling emit_insn seems more natural now that the gen_* call
has been split out.  It also matches later code in the function.

Tested on aarch64-linux-gnu, arm-linux-gnueabihf, armeb-eabi and
x86_64-linux-gnu.  OK to install?

Richard


gcc/
	PR rtl-optimization/98689
	* reg-notes.def (UNTYPED_CALL): New note.
	* combine.c (distribute_notes): Handle it.
	* emit-rtl.c (try_split): Likewise.
	* rtlanal.c (rtx_properties::try_to_add_insn): Likewise.  Assume
	that calls with the note implicitly set all return value registers.
	* builtins.c (expand_builtin_apply): Add a REG_UNTYPED_CALL
	to untyped_calls.
---
 gcc/builtins.c    |  8 ++++++--
 gcc/combine.c     |  1 +
 gcc/emit-rtl.c    |  1 +
 gcc/reg-notes.def | 15 +++++++++++++++
 gcc/rtlanal.c     |  9 +++++++++
 5 files changed, 32 insertions(+), 2 deletions(-)

On Fri, Apr 16, 2021 at 12:02 PM Richard Sandiford via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> This patch fixes a regression introduced by the rtl-ssa patches.
> It was seen on HPPA but it might be latent elsewhere.
>
> The problem is that the traditional way of expanding an untyped_call
> is to emit sequences like:
>
>    (call (mem (symbol_ref "foo")))
>    (set (reg pseudo1) (reg result1))
>    ...
>    (set (reg pseudon) (reg resultn))
>
> The ABI specifies that result1..resultn are clobbered by the call but
> nothing in the RTL indicates that result1..resultn are the results
> of the call.  Normally, using a clobbered value gives undefined results,
> but in this case the results are well-defined and matter for correctness.
>
> This seems like a niche case, so I think it would be better to mark
> it explicitly rather than try to detect it heuristically.
>
> Note that in expand_builtin_apply we already have an rtx_insn *,
> so it doesn't matter whether we call emit_call_insn or emit_insn.
> Calling emit_insn seems more natural now that the gen_* call
> has been split out.  It also matches later code in the function.
>
> Tested on aarch64-linux-gnu, arm-linux-gnueabihf, armeb-eabi and
> x86_64-linux-gnu.  OK to install?

OK.  Does DF handle this correctly?

I wonder why we cannot simply do sth like

(call_insn (set (parallel [ (reg:...) (reg:...) ] ) (mem (symbol_ref...)))

or so?  But I guess we'd have to alter all targets for this...

Thanks,
Richard.

> Richard
>
>
> gcc/
>         PR rtl-optimization/98689
>         * reg-notes.def (UNTYPED_CALL): New note.
>         * combine.c (distribute_notes): Handle it.
>         * emit-rtl.c (try_split): Likewise.
>         * rtlanal.c (rtx_properties::try_to_add_insn): Likewise.  Assume
>         that calls with the note implicitly set all return value registers.
>         * builtins.c (expand_builtin_apply): Add a REG_UNTYPED_CALL
>         to untyped_calls.
> ---
>  gcc/builtins.c    |  8 ++++++--
>  gcc/combine.c     |  1 +
>  gcc/emit-rtl.c    |  1 +
>  gcc/reg-notes.def | 15 +++++++++++++++
>  gcc/rtlanal.c     |  9 +++++++++
>  5 files changed, 32 insertions(+), 2 deletions(-)
>
> diff --git a/gcc/builtins.c b/gcc/builtins.c
> index 196dda3fa5e..d30c4eb62fc 100644
> --- a/gcc/builtins.c
> +++ b/gcc/builtins.c
> @@ -2490,8 +2490,12 @@ expand_builtin_apply (rtx function, rtx arguments, rtx argsize)
>    if (targetm.have_untyped_call ())
>      {
>        rtx mem = gen_rtx_MEM (FUNCTION_MODE, function);
> -      emit_call_insn (targetm.gen_untyped_call (mem, result,
> -                                               result_vector (1, result)));
> +      rtx_insn *seq = targetm.gen_untyped_call (mem, result,
> +                                               result_vector (1, result));
> +      for (rtx_insn *insn = seq; insn; insn = NEXT_INSN (insn))
> +       if (CALL_P (insn))
> +         add_reg_note (insn, REG_UNTYPED_CALL, NULL_RTX);
> +      emit_insn (seq);
>      }
>    else if (targetm.have_call_value ())
>      {
> diff --git a/gcc/combine.c b/gcc/combine.c
> index 6cb5b0ca102..9063a07bd00 100644
> --- a/gcc/combine.c
> +++ b/gcc/combine.c
> @@ -14337,6 +14337,7 @@ distribute_notes (rtx notes, rtx_insn *from_insn, rtx_insn *i3, rtx_insn *i2,
>         case REG_SETJMP:
>         case REG_TM:
>         case REG_CALL_DECL:
> +       case REG_UNTYPED_CALL:
>         case REG_CALL_NOCF_CHECK:
>           /* These notes must remain with the call.  It should not be
>              possible for both I2 and I3 to be a call.  */
> diff --git a/gcc/emit-rtl.c b/gcc/emit-rtl.c
> index 1113c58c49e..07e908624a0 100644
> --- a/gcc/emit-rtl.c
> +++ b/gcc/emit-rtl.c
> @@ -3964,6 +3964,7 @@ try_split (rtx pat, rtx_insn *trial, int last)
>           break;
>
>         case REG_CALL_DECL:
> +       case REG_UNTYPED_CALL:
>           gcc_assert (call_insn != NULL_RTX);
>           add_reg_note (call_insn, REG_NOTE_KIND (note), XEXP (note, 0));
>           break;
> diff --git a/gcc/reg-notes.def b/gcc/reg-notes.def
> index d7f0bfdab0b..995052ebc28 100644
> --- a/gcc/reg-notes.def
> +++ b/gcc/reg-notes.def
> @@ -233,6 +233,21 @@ REG_NOTE (STACK_CHECK)
>     insn.  This note is a SYMBOL_REF.  */
>  REG_NOTE (CALL_DECL)
>
> +/* Indicates that the call is an untyped_call.  These calls are special
> +   in that they set all of the target ABI's return value registers to a
> +   defined value without explicitly saying so.  For example, a typical
> +   untyped_call sequence has the form:
> +
> +       (call (mem (symbol_ref "foo")))
> +       (set (reg pseudo1) (reg result1))
> +       ...
> +       (set (reg pseudon) (reg resultn))
> +
> +   The ABI specifies that result1..resultn are clobbered by the call,
> +   but the RTL does not indicate that result1..resultn are the results
> +   of the call.  */
> +REG_NOTE (UNTYPED_CALL)
> +
>  /* Indicate that a call should not be verified for control-flow consistency.
>     The target address of the call is assumed as a valid address and no check
>     to validate a branch to the target address is needed.  The call is marked
> diff --git a/gcc/rtlanal.c b/gcc/rtlanal.c
> index 170420a610b..67a49e65fd8 100644
> --- a/gcc/rtlanal.c
> +++ b/gcc/rtlanal.c
> @@ -2333,6 +2333,15 @@ rtx_properties::try_to_add_insn (const rtx_insn *insn, bool include_notes)
>               *ref_iter++ = rtx_obj_reference (regno, flags,
>                                                reg_raw_mode[regno], 0);
>         }
> +      /* Untyped calls implicitly set all function value registers.
> +        Again, we add them first in case the main pattern contains
> +        a fixed-form clobber.  */
> +      if (find_reg_note (insn, REG_UNTYPED_CALL, NULL_RTX))
> +       for (unsigned int regno = 0; regno < FIRST_PSEUDO_REGISTER; ++regno)
> +         if (targetm.calls.function_value_regno_p (regno)
> +             && ref_iter != ref_end)
> +           *ref_iter++ = rtx_obj_reference (regno, rtx_obj_flags::IS_WRITE,
> +                                            reg_raw_mode[regno], 0);
>        if (ref_iter != ref_end && !RTL_CONST_CALL_P (insn))
>         {
>           auto mem_flags = rtx_obj_flags::IS_READ;

Richard Biener via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> On Fri, Apr 16, 2021 at 12:02 PM Richard Sandiford via Gcc-patches
> <gcc-patches@gcc.gnu.org> wrote:
>>
>> This patch fixes a regression introduced by the rtl-ssa patches.
>> It was seen on HPPA but it might be latent elsewhere.
>>
>> The problem is that the traditional way of expanding an untyped_call
>> is to emit sequences like:
>>
>>    (call (mem (symbol_ref "foo")))
>>    (set (reg pseudo1) (reg result1))
>>    ...
>>    (set (reg pseudon) (reg resultn))
>>
>> The ABI specifies that result1..resultn are clobbered by the call but
>> nothing in the RTL indicates that result1..resultn are the results
>> of the call.  Normally, using a clobbered value gives undefined results,
>> but in this case the results are well-defined and matter for correctness.
>>
>> This seems like a niche case, so I think it would be better to mark
>> it explicitly rather than try to detect it heuristically.
>>
>> Note that in expand_builtin_apply we already have an rtx_insn *,
>> so it doesn't matter whether we call emit_call_insn or emit_insn.
>> Calling emit_insn seems more natural now that the gen_* call
>> has been split out.  It also matches later code in the function.
>>
>> Tested on aarch64-linux-gnu, arm-linux-gnueabihf, armeb-eabi and
>> x86_64-linux-gnu.  OK to install?
>
> OK.  Does DF handle this correctly?

Kind-of.  DF creates explicit clobbers for all call-clobbered registers
and later uses use the results of these clobbers.  I guess consumers handle
that leniently in practice and don't treat the result as undefined.

But yeah, it would probably be better to add a proper set in DF too.

> I wonder why we cannot simply do sth like
>
> (call_insn (set (parallel [ (reg:...) (reg:...) ] ) (mem (symbol_ref...)))
>
> or so?  But I guess we'd have to alter all targets for this...

Yeah, that would be better.  Targets don't need to define untyped_call
if they just have one return value register, and in that case the
builtins.c code will use a normal call_value that describes the
operation properly.  But the other targets would need special code
to handle this.

Thanks,
Richard