lower-subreg, expr: Mitigate inefficiencies derived from "(clobber (reg X))" followed by "(set (subreg (reg X)) (...))"

  Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
data flow consistent, but it also increases register allocation pressure
and thus often creates many unwanted register-to-register moves that
cannot be optimized away.  It seems just analogous to partial register
stall which is a famous problem on processors that do register renaming.

In my opinion, when the register to be clobbered is a composite of hard
ones, we should clobber the individual elements separetely, otherwise
clear the entire to zero prior to use as the "init-regs" pass does (like
partial register stall workarounds on x86 CPUs).  Such redundant zero
constant assignments will be removed later in the "cprop_hardreg" pass.

This patch may give better output code quality for the reasons above,
especially on architectures that don't have DFmode hard registers
(On architectures with such hard registers, this patch changes virtually
nothing).

For example (Espressif ESP8266, Xtensa without FP hard regs):

    /* example */
    double _Complex conjugate(double _Complex z) {
      __imag__(z) *= -1;
      return z;
    }

    ;; before
    conjugate:
        movi.n  a6, -1
        slli    a6, a6, 31
        mov.n   a8, a2
        mov.n   a9, a3
        mov.n   a7, a4
        xor     a6, a5, a6
        mov.n   a2, a8
        mov.n   a3, a9
        mov.n   a4, a7
        mov.n   a5, a6
        ret.n

    ;; after
    conjugate:
        movi.n  a6, -1
        slli    a6, a6, 31
        xor     a6, a5, a6
        mov.n   a5, a6
        ret.n

gcc/ChangeLog:

	* lower-subreg.cc (resolve_simple_move):
	Add zero clear of the entire register immediately after
	the clobber.
	* expr.cc (emit_move_complex_parts):
	Change to clobber the real and imaginary parts separately
	instead of the whole complex register if possible.
---
 gcc/expr.cc         | 26 ++++++++++++++++++++------
 gcc/lower-subreg.cc |  7 ++++++-
 2 files changed, 26 insertions(+), 7 deletions(-)
  

Takayuki 'January June' Suwa via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
> data flow consistent, but it also increases register allocation pressure
> and thus often creates many unwanted register-to-register moves that
> cannot be optimized away.

There are two things here:

- If emit_move_complex_parts emits a clobber of a hard register,
  then that's probably a bug/misfeature.  The point of the clobber is
  to indicate that the register has no useful contents.  That's useful
  for wide pseudos that are written to in parts, since it avoids the
  need to track the liveness of each part of the pseudo individually.
  But it shouldn't be necessary for hard registers, since subregs of
  hard registers are simplified to hard registers wherever possible
  (which on most targets is "always").

  So I think the emit_move_complex_parts clobber should be restricted
  to !HARD_REGISTER_P, like the lower-subreg clobber is.  If that helps
  (if only partly) then it would be worth doing as its own patch.

- I think it'd be worth looking into more detail why a clobber makes
  a difference to register pressure.  A clobber of a pseudo register R
  shouldn't make R conflict with things that are live at the point of
  the clobber.

>  It seems just analogous to partial register
> stall which is a famous problem on processors that do register renaming.
>
> In my opinion, when the register to be clobbered is a composite of hard
> ones, we should clobber the individual elements separetely, otherwise
> clear the entire to zero prior to use as the "init-regs" pass does (like
> partial register stall workarounds on x86 CPUs).  Such redundant zero
> constant assignments will be removed later in the "cprop_hardreg" pass.

I don't think we should rely on the zero being optimised away later.

Emitting the zero also makes it harder for the register allocator
to elide the move.  For example, if we have:

  (set (subreg:SI (reg:DI P) 0) (reg:SI R0))
  (set (subreg:SI (reg:DI P) 4) (reg:SI R1))

then there is at least a chance that the RA could assign hard registers
R0:R1 to P, which would turn the moves into nops.  If we emit:

  (set (reg:DI P) (const_int 0))

beforehand then that becomes impossible, since R0 and R1 would then
conflict with P.

TBH I'm surprised we still run init_regs for LRA.  I thought there was
a plan to stop doing that, but perhaps I misremember.

Thanks,
Richard

> This patch may give better output code quality for the reasons above,
> especially on architectures that don't have DFmode hard registers
> (On architectures with such hard registers, this patch changes virtually
> nothing).
>
> For example (Espressif ESP8266, Xtensa without FP hard regs):
>
>     /* example */
>     double _Complex conjugate(double _Complex z) {
>       __imag__(z) *= -1;
>       return z;
>     }
>
>     ;; before
>     conjugate:
>         movi.n  a6, -1
>         slli    a6, a6, 31
>         mov.n   a8, a2
>         mov.n   a9, a3
>         mov.n   a7, a4
>         xor     a6, a5, a6
>         mov.n   a2, a8
>         mov.n   a3, a9
>         mov.n   a4, a7
>         mov.n   a5, a6
>         ret.n
>
>     ;; after
>     conjugate:
>         movi.n  a6, -1
>         slli    a6, a6, 31
>         xor     a6, a5, a6
>         mov.n   a5, a6
>         ret.n
>
> gcc/ChangeLog:
>
> 	* lower-subreg.cc (resolve_simple_move):
> 	Add zero clear of the entire register immediately after
> 	the clobber.
> 	* expr.cc (emit_move_complex_parts):
> 	Change to clobber the real and imaginary parts separately
> 	instead of the whole complex register if possible.
> ---
>  gcc/expr.cc         | 26 ++++++++++++++++++++------
>  gcc/lower-subreg.cc |  7 ++++++-
>  2 files changed, 26 insertions(+), 7 deletions(-)
>
> diff --git a/gcc/expr.cc b/gcc/expr.cc
> index 80bb1b8a4c5..9732e8fd4e5 100644
> --- a/gcc/expr.cc
> +++ b/gcc/expr.cc
> @@ -3775,15 +3775,29 @@ emit_move_complex_push (machine_mode mode, rtx x, rtx y)
>  rtx_insn *
>  emit_move_complex_parts (rtx x, rtx y)
>  {
> -  /* Show the output dies here.  This is necessary for SUBREGs
> -     of pseudos since we cannot track their lifetimes correctly;
> -     hard regs shouldn't appear here except as return values.  */
> -  if (!reload_completed && !reload_in_progress
> -      && REG_P (x) && !reg_overlap_mentioned_p (x, y))
> -    emit_clobber (x);
> +  rtx_insn *re_insn, *im_insn;
>  
>    write_complex_part (x, read_complex_part (y, false), false, true);
> +  re_insn = get_last_insn ();
>    write_complex_part (x, read_complex_part (y, true), true, false);
> +  im_insn = get_last_insn ();
> +
> +  /* Show the output dies here.  This is necessary for SUBREGs
> +     of pseudos since we cannot track their lifetimes correctly.  */
> +  if (can_create_pseudo_p ()
> +      && REG_P (x) && ! reg_overlap_mentioned_p (x, y))
> +    {
> +      /* Hard regs shouldn't appear here except as return values.  */
> +      if (HARD_REGISTER_P (x) && REG_NREGS (x) % 2 == 0)
> +	{
> +	  emit_insn_before (gen_clobber (SET_DEST (PATTERN (re_insn))),
> +			    re_insn);
> +	  emit_insn_before (gen_clobber (SET_DEST (PATTERN (im_insn))),
> +			    im_insn);
> +	}
> +      else
> +	emit_insn_before (gen_clobber (x), re_insn);
> +    }
>  
>    return get_last_insn ();
>  }
> diff --git a/gcc/lower-subreg.cc b/gcc/lower-subreg.cc
> index 03e9326c663..4ff0a7d1556 100644
> --- a/gcc/lower-subreg.cc
> +++ b/gcc/lower-subreg.cc
> @@ -1086,7 +1086,12 @@ resolve_simple_move (rtx set, rtx_insn *insn)
>        unsigned int i;
>  
>        if (REG_P (dest) && !HARD_REGISTER_NUM_P (REGNO (dest)))
> -	emit_clobber (dest);
> +	{
> +	  emit_clobber (dest);
> +	  /* We clear the entire of dest with zero after the clobber,
> +	     similar to the "init-regs" pass.  */
> +	  emit_move_insn (dest, CONST0_RTX (GET_MODE (dest)));
> +	}
>  
>        for (i = 0; i < words; ++i)
>  	{
  

Thanks for your response.

On 2022/08/03 16:52, Richard Sandiford wrote:
> Takayuki 'January June' Suwa via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>> Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
>> data flow consistent, but it also increases register allocation pressure
>> and thus often creates many unwanted register-to-register moves that
>> cannot be optimized away.
> 
> There are two things here:
> 
> - If emit_move_complex_parts emits a clobber of a hard register,
>   then that's probably a bug/misfeature.  The point of the clobber is
>   to indicate that the register has no useful contents.  That's useful
>   for wide pseudos that are written to in parts, since it avoids the
>   need to track the liveness of each part of the pseudo individually.
>   But it shouldn't be necessary for hard registers, since subregs of
>   hard registers are simplified to hard registers wherever possible
>   (which on most targets is "always").
> 
>   So I think the emit_move_complex_parts clobber should be restricted
>   to !HARD_REGISTER_P, like the lower-subreg clobber is.  If that helps
>   (if only partly) then it would be worth doing as its own patch.
> 
> - I think it'd be worth looking into more detail why a clobber makes
>   a difference to register pressure.  A clobber of a pseudo register R
>   shouldn't make R conflict with things that are live at the point of
>   the clobber.

I agree with its worth.
In fact, aside from other ports, on the xtensa one, RA in code with frequent D[FC]mode pseudos is terribly bad.
For example, in __muldc3 on libgcc2, the size of the stack frame reserved will almost double depending on whether or not this patch is applied.

> 
>>  It seems just analogous to partial register
>> stall which is a famous problem on processors that do register renaming.
>>
>> In my opinion, when the register to be clobbered is a composite of hard
>> ones, we should clobber the individual elements separetely, otherwise
>> clear the entire to zero prior to use as the "init-regs" pass does (like
>> partial register stall workarounds on x86 CPUs).  Such redundant zero
>> constant assignments will be removed later in the "cprop_hardreg" pass.
> 
> I don't think we should rely on the zero being optimised away later.
> 
> Emitting the zero also makes it harder for the register allocator
> to elide the move.  For example, if we have:
> 
>   (set (subreg:SI (reg:DI P) 0) (reg:SI R0))
>   (set (subreg:SI (reg:DI P) 4) (reg:SI R1))
> 
> then there is at least a chance that the RA could assign hard registers
> R0:R1 to P, which would turn the moves into nops.  If we emit:
> 
>   (set (reg:DI P) (const_int 0))
> 
> beforehand then that becomes impossible, since R0 and R1 would then
> conflict with P.

Ah, surely, as you pointed out for targets where "(reg: DI)" corresponds to one hard register.

> 
> TBH I'm surprised we still run init_regs for LRA.  I thought there was
> a plan to stop doing that, but perhaps I misremember.

Sorry I am not sure about the status of LRA... because the xtensa port is still using reload.

As conclusion, trying to tweak the common code side may have been a bit premature.
I'll consider if I can deal with those issues on the side of the target-specific code.

> 
> Thanks,
> Richard
> 
>> This patch may give better output code quality for the reasons above,
>> especially on architectures that don't have DFmode hard registers
>> (On architectures with such hard registers, this patch changes virtually
>> nothing).
>>
>> For example (Espressif ESP8266, Xtensa without FP hard regs):
>>
>>     /* example */
>>     double _Complex conjugate(double _Complex z) {
>>       __imag__(z) *= -1;
>>       return z;
>>     }
>>
>>     ;; before
>>     conjugate:
>>         movi.n  a6, -1
>>         slli    a6, a6, 31
>>         mov.n   a8, a2
>>         mov.n   a9, a3
>>         mov.n   a7, a4
>>         xor     a6, a5, a6
>>         mov.n   a2, a8
>>         mov.n   a3, a9
>>         mov.n   a4, a7
>>         mov.n   a5, a6
>>         ret.n
>>
>>     ;; after
>>     conjugate:
>>         movi.n  a6, -1
>>         slli    a6, a6, 31
>>         xor     a6, a5, a6
>>         mov.n   a5, a6
>>         ret.n
>>
>> gcc/ChangeLog:
>>
>> 	* lower-subreg.cc (resolve_simple_move):
>> 	Add zero clear of the entire register immediately after
>> 	the clobber.
>> 	* expr.cc (emit_move_complex_parts):
>> 	Change to clobber the real and imaginary parts separately
>> 	instead of the whole complex register if possible.
>> ---
>>  gcc/expr.cc         | 26 ++++++++++++++++++++------
>>  gcc/lower-subreg.cc |  7 ++++++-
>>  2 files changed, 26 insertions(+), 7 deletions(-)
>>
>> diff --git a/gcc/expr.cc b/gcc/expr.cc
>> index 80bb1b8a4c5..9732e8fd4e5 100644
>> --- a/gcc/expr.cc
>> +++ b/gcc/expr.cc
>> @@ -3775,15 +3775,29 @@ emit_move_complex_push (machine_mode mode, rtx x, rtx y)
>>  rtx_insn *
>>  emit_move_complex_parts (rtx x, rtx y)
>>  {
>> -  /* Show the output dies here.  This is necessary for SUBREGs
>> -     of pseudos since we cannot track their lifetimes correctly;
>> -     hard regs shouldn't appear here except as return values.  */
>> -  if (!reload_completed && !reload_in_progress
>> -      && REG_P (x) && !reg_overlap_mentioned_p (x, y))
>> -    emit_clobber (x);
>> +  rtx_insn *re_insn, *im_insn;
>>  
>>    write_complex_part (x, read_complex_part (y, false), false, true);
>> +  re_insn = get_last_insn ();
>>    write_complex_part (x, read_complex_part (y, true), true, false);
>> +  im_insn = get_last_insn ();
>> +
>> +  /* Show the output dies here.  This is necessary for SUBREGs
>> +     of pseudos since we cannot track their lifetimes correctly.  */
>> +  if (can_create_pseudo_p ()
>> +      && REG_P (x) && ! reg_overlap_mentioned_p (x, y))
>> +    {
>> +      /* Hard regs shouldn't appear here except as return values.  */
>> +      if (HARD_REGISTER_P (x) && REG_NREGS (x) % 2 == 0)
>> +	{
>> +	  emit_insn_before (gen_clobber (SET_DEST (PATTERN (re_insn))),
>> +			    re_insn);
>> +	  emit_insn_before (gen_clobber (SET_DEST (PATTERN (im_insn))),
>> +			    im_insn);
>> +	}
>> +      else
>> +	emit_insn_before (gen_clobber (x), re_insn);
>> +    }
>>  
>>    return get_last_insn ();
>>  }
>> diff --git a/gcc/lower-subreg.cc b/gcc/lower-subreg.cc
>> index 03e9326c663..4ff0a7d1556 100644
>> --- a/gcc/lower-subreg.cc
>> +++ b/gcc/lower-subreg.cc
>> @@ -1086,7 +1086,12 @@ resolve_simple_move (rtx set, rtx_insn *insn)
>>        unsigned int i;
>>  
>>        if (REG_P (dest) && !HARD_REGISTER_NUM_P (REGNO (dest)))
>> -	emit_clobber (dest);
>> +	{
>> +	  emit_clobber (dest);
>> +	  /* We clear the entire of dest with zero after the clobber,
>> +	     similar to the "init-regs" pass.  */
>> +	  emit_move_insn (dest, CONST0_RTX (GET_MODE (dest)));
>> +	}
>>  
>>        for (i = 0; i < words; ++i)
>>  	{
  

On 8/3/2022 1:52 AM, Richard Sandiford via Gcc-patches wrote:
> Takayuki 'January June' Suwa via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>> Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
>> data flow consistent, but it also increases register allocation pressure
>> and thus often creates many unwanted register-to-register moves that
>> cannot be optimized away.
> There are two things here:
>
> - If emit_move_complex_parts emits a clobber of a hard register,
>    then that's probably a bug/misfeature.  The point of the clobber is
>    to indicate that the register has no useful contents.  That's useful
>    for wide pseudos that are written to in parts, since it avoids the
>    need to track the liveness of each part of the pseudo individually.
>    But it shouldn't be necessary for hard registers, since subregs of
>    hard registers are simplified to hard registers wherever possible
>    (which on most targets is "always").
>
>    So I think the emit_move_complex_parts clobber should be restricted
>    to !HARD_REGISTER_P, like the lower-subreg clobber is.  If that helps
>    (if only partly) then it would be worth doing as its own patch.
Agreed.

>
> - I think it'd be worth looking into more detail why a clobber makes
>    a difference to register pressure.  A clobber of a pseudo register R
>    shouldn't make R conflict with things that are live at the point of
>    the clobber.
Also agreed.

>
>>   It seems just analogous to partial register
>> stall which is a famous problem on processors that do register renaming.
>>
>> In my opinion, when the register to be clobbered is a composite of hard
>> ones, we should clobber the individual elements separetely, otherwise
>> clear the entire to zero prior to use as the "init-regs" pass does (like
>> partial register stall workarounds on x86 CPUs).  Such redundant zero
>> constant assignments will be removed later in the "cprop_hardreg" pass.
> I don't think we should rely on the zero being optimised away later.
>
> Emitting the zero also makes it harder for the register allocator
> to elide the move.  For example, if we have:
>
>    (set (subreg:SI (reg:DI P) 0) (reg:SI R0))
>    (set (subreg:SI (reg:DI P) 4) (reg:SI R1))
>
> then there is at least a chance that the RA could assign hard registers
> R0:R1 to P, which would turn the moves into nops.  If we emit:
>
>    (set (reg:DI P) (const_int 0))
>
> beforehand then that becomes impossible, since R0 and R1 would then
> conflict with P.
>
> TBH I'm surprised we still run init_regs for LRA.  I thought there was
> a plan to stop doing that, but perhaps I misremember.
I have vague memories of dealing with some of this nonsense a few 
release cycles.  I don't recall all the details, but init-regs + 
lower-subreg + regcprop + splitting all conspired to generate poor code 
on the MIPS targets.  See pr87761, though it doesn't include all my 
findings -- I can't recall if I walked through the entire tortured 
sequence in the gcc-patches discussion or not.

I ended up working around in the mips backend in conjunction with some 
changes to regcprop IIRC.

Jeff
  

Jeff Law via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> On 8/3/2022 1:52 AM, Richard Sandiford via Gcc-patches wrote:
>> Takayuki 'January June' Suwa via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>>> Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
>>> data flow consistent, but it also increases register allocation pressure
>>> and thus often creates many unwanted register-to-register moves that
>>> cannot be optimized away.
>> There are two things here:
>>
>> - If emit_move_complex_parts emits a clobber of a hard register,
>>    then that's probably a bug/misfeature.  The point of the clobber is
>>    to indicate that the register has no useful contents.  That's useful
>>    for wide pseudos that are written to in parts, since it avoids the
>>    need to track the liveness of each part of the pseudo individually.
>>    But it shouldn't be necessary for hard registers, since subregs of
>>    hard registers are simplified to hard registers wherever possible
>>    (which on most targets is "always").
>>
>>    So I think the emit_move_complex_parts clobber should be restricted
>>    to !HARD_REGISTER_P, like the lower-subreg clobber is.  If that helps
>>    (if only partly) then it would be worth doing as its own patch.
> Agreed.
>
>>
>> - I think it'd be worth looking into more detail why a clobber makes
>>    a difference to register pressure.  A clobber of a pseudo register R
>>    shouldn't make R conflict with things that are live at the point of
>>    the clobber.
> Also agreed.
>
>>
>>>   It seems just analogous to partial register
>>> stall which is a famous problem on processors that do register renaming.
>>>
>>> In my opinion, when the register to be clobbered is a composite of hard
>>> ones, we should clobber the individual elements separetely, otherwise
>>> clear the entire to zero prior to use as the "init-regs" pass does (like
>>> partial register stall workarounds on x86 CPUs).  Such redundant zero
>>> constant assignments will be removed later in the "cprop_hardreg" pass.
>> I don't think we should rely on the zero being optimised away later.
>>
>> Emitting the zero also makes it harder for the register allocator
>> to elide the move.  For example, if we have:
>>
>>    (set (subreg:SI (reg:DI P) 0) (reg:SI R0))
>>    (set (subreg:SI (reg:DI P) 4) (reg:SI R1))
>>
>> then there is at least a chance that the RA could assign hard registers
>> R0:R1 to P, which would turn the moves into nops.  If we emit:
>>
>>    (set (reg:DI P) (const_int 0))
>>
>> beforehand then that becomes impossible, since R0 and R1 would then
>> conflict with P.
>>
>> TBH I'm surprised we still run init_regs for LRA.  I thought there was
>> a plan to stop doing that, but perhaps I misremember.
> I have vague memories of dealing with some of this nonsense a few 
> release cycles.  I don't recall all the details, but init-regs + 
> lower-subreg + regcprop + splitting all conspired to generate poor code 
> on the MIPS targets.  See pr87761, though it doesn't include all my 
> findings -- I can't recall if I walked through the entire tortured 
> sequence in the gcc-patches discussion or not.
>
> I ended up working around in the mips backend in conjunction with some 
> changes to regcprop IIRC.

Thanks for the pointer, hadn't seen that.  And yeah, for the early-ish
passes, I guess the interaction between lower-subreg and init-regs is
important too, not just the interaction between lower-subreg and RA.
It probably also ties into the problems with overly-scalarised register
moves, like in PR 106106.

So maybe I was being too optimistic :-)

Richard
  

Takayuki 'January June' Suwa <jjsuwa_sys3175@yahoo.co.jp> writes:
> Thanks for your response.
>
> On 2022/08/03 16:52, Richard Sandiford wrote:
>> Takayuki 'January June' Suwa via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>>> Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
>>> data flow consistent, but it also increases register allocation pressure
>>> and thus often creates many unwanted register-to-register moves that
>>> cannot be optimized away.
>> 
>> There are two things here:
>> 
>> - If emit_move_complex_parts emits a clobber of a hard register,
>>   then that's probably a bug/misfeature.  The point of the clobber is
>>   to indicate that the register has no useful contents.  That's useful
>>   for wide pseudos that are written to in parts, since it avoids the
>>   need to track the liveness of each part of the pseudo individually.
>>   But it shouldn't be necessary for hard registers, since subregs of
>>   hard registers are simplified to hard registers wherever possible
>>   (which on most targets is "always").
>> 
>>   So I think the emit_move_complex_parts clobber should be restricted
>>   to !HARD_REGISTER_P, like the lower-subreg clobber is.  If that helps
>>   (if only partly) then it would be worth doing as its own patch.
>> 
>> - I think it'd be worth looking into more detail why a clobber makes
>>   a difference to register pressure.  A clobber of a pseudo register R
>>   shouldn't make R conflict with things that are live at the point of
>>   the clobber.
>
> I agree with its worth.
> In fact, aside from other ports, on the xtensa one, RA in code with frequent D[FC]mode pseudos is terribly bad.
> For example, in __muldc3 on libgcc2, the size of the stack frame reserved will almost double depending on whether or not this patch is applied.

Yeah, that's a lot.

>>>  It seems just analogous to partial register
>>> stall which is a famous problem on processors that do register renaming.
>>>
>>> In my opinion, when the register to be clobbered is a composite of hard
>>> ones, we should clobber the individual elements separetely, otherwise
>>> clear the entire to zero prior to use as the "init-regs" pass does (like
>>> partial register stall workarounds on x86 CPUs).  Such redundant zero
>>> constant assignments will be removed later in the "cprop_hardreg" pass.
>> 
>> I don't think we should rely on the zero being optimised away later.
>> 
>> Emitting the zero also makes it harder for the register allocator
>> to elide the move.  For example, if we have:
>> 
>>   (set (subreg:SI (reg:DI P) 0) (reg:SI R0))
>>   (set (subreg:SI (reg:DI P) 4) (reg:SI R1))
>> 
>> then there is at least a chance that the RA could assign hard registers
>> R0:R1 to P, which would turn the moves into nops.  If we emit:
>> 
>>   (set (reg:DI P) (const_int 0))
>> 
>> beforehand then that becomes impossible, since R0 and R1 would then
>> conflict with P.
>
> Ah, surely, as you pointed out for targets where "(reg: DI)" corresponds to one hard register.

I was thinking here about the case where (reg:DI …) corresponds to
2 hard registers.  Each subreg move is then a single hard register
copy, but assigning P to the combination R0:R1 can remove both of
the subreg moves.

>> TBH I'm surprised we still run init_regs for LRA.  I thought there was
>> a plan to stop doing that, but perhaps I misremember.
>
> Sorry I am not sure about the status of LRA... because the xtensa port is still using reload.

Ah, hadn't realised that.  If you have time to work on it, it would be
really good to move over to LRA.  There are plans to remove old reload.

It might be that old reload *does* treat a pseudo clobber as a conflict.
I can't remember now.  If so, then zeroing the register wouldn't be
too bad (for old reload only).

> As conclusion, trying to tweak the common code side may have been a bit premature.
> I'll consider if I can deal with those issues on the side of the target-specific code.

It's likely to be at least partly a target-independent issue, so tweaking
the common code makes sense in principle.

Does adding !HARD_REGISTER_P (x) to:

  /* Show the output dies here.  This is necessary for SUBREGs
     of pseudos since we cannot track their lifetimes correctly;
     hard regs shouldn't appear here except as return values.  */
  if (!reload_completed && !reload_in_progress
      && REG_P (x) && !reg_overlap_mentioned_p (x, y))
    emit_clobber (x);

in emit_move_complex_parts help?  If so, I think we should do at
least that much.

Thanks,
Richard
  

(sorry repost due to the lack of cc here)
Hi!

On 2022/08/04 18:49, Richard Sandiford wrote:
> Takayuki 'January June' Suwa <jjsuwa_sys3175@yahoo.co.jp> writes:
>> Thanks for your response.
>>
>> On 2022/08/03 16:52, Richard Sandiford wrote:
>>> Takayuki 'January June' Suwa via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
>>>> Emitting "(clobber (reg X))" before "(set (subreg (reg X)) (...))" keeps
>>>> data flow consistent, but it also increases register allocation pressure
>>>> and thus often creates many unwanted register-to-register moves that
>>>> cannot be optimized away.
>>>
>>> There are two things here:
>>>
>>> - If emit_move_complex_parts emits a clobber of a hard register,
>>>   then that's probably a bug/misfeature.  The point of the clobber is
>>>   to indicate that the register has no useful contents.  That's useful
>>>   for wide pseudos that are written to in parts, since it avoids the
>>>   need to track the liveness of each part of the pseudo individually.
>>>   But it shouldn't be necessary for hard registers, since subregs of
>>>   hard registers are simplified to hard registers wherever possible
>>>   (which on most targets is "always").
>>>
>>>   So I think the emit_move_complex_parts clobber should be restricted
>>>   to !HARD_REGISTER_P, like the lower-subreg clobber is.  If that helps
>>>   (if only partly) then it would be worth doing as its own patch.
>>>
>>> - I think it'd be worth looking into more detail why a clobber makes
>>>   a difference to register pressure.  A clobber of a pseudo register R
>>>   shouldn't make R conflict with things that are live at the point of
>>>   the clobber.
>>
>> I agree with its worth.
>> In fact, aside from other ports, on the xtensa one, RA in code with frequent D[FC]mode pseudos is terribly bad.
>> For example, in __muldc3 on libgcc2, the size of the stack frame reserved will almost double depending on whether or not this patch is applied.
> 
> Yeah, that's a lot.

So lots, but almost double might be an overstatement :)

BTW after some quick experimentation, I found that turning on -fsplit-wide-types-early would roughly (but not completely) solve the problem.  Surely, the output was not so bad in the past...

> 
>>>>  It seems just analogous to partial register
>>>> stall which is a famous problem on processors that do register renaming.
>>>>
>>>> In my opinion, when the register to be clobbered is a composite of hard
>>>> ones, we should clobber the individual elements separetely, otherwise
>>>> clear the entire to zero prior to use as the "init-regs" pass does (like
>>>> partial register stall workarounds on x86 CPUs).  Such redundant zero
>>>> constant assignments will be removed later in the "cprop_hardreg" pass.
>>>
>>> I don't think we should rely on the zero being optimised away later.
>>>
>>> Emitting the zero also makes it harder for the register allocator
>>> to elide the move.  For example, if we have:
>>>
>>>   (set (subreg:SI (reg:DI P) 0) (reg:SI R0))
>>>   (set (subreg:SI (reg:DI P) 4) (reg:SI R1))
>>>
>>> then there is at least a chance that the RA could assign hard registers
>>> R0:R1 to P, which would turn the moves into nops.  If we emit:
>>>
>>>   (set (reg:DI P) (const_int 0))
>>>
>>> beforehand then that becomes impossible, since R0 and R1 would then
>>> conflict with P.
>>
>> Ah, surely, as you pointed out for targets where "(reg: DI)" corresponds to one hard register.
> 
> I was thinking here about the case where (reg:DI …) corresponds to
> 2 hard registers.  Each subreg move is then a single hard register
> copy, but assigning P to the combination R0:R1 can remove both of
> the subreg moves.
> 
>>> TBH I'm surprised we still run init_regs for LRA.  I thought there was
>>> a plan to stop doing that, but perhaps I misremember.
>>
>> Sorry I am not sure about the status of LRA... because the xtensa port is still using reload.
> 
> Ah, hadn't realised that.  If you have time to work on it, it would be
> really good to move over to LRA.  There are plans to remove old reload.

Alas you do overestimate me :) I've only been working about the GCC development for a little over a year.
Well it's a lie that I'm not interested in it, but too much for me.

> 
> It might be that old reload *does* treat a pseudo clobber as a conflict.
> I can't remember now.  If so, then zeroing the register wouldn't be
> too bad (for old reload only).
> 
>> As conclusion, trying to tweak the common code side may have been a bit premature.
>> I'll consider if I can deal with those issues on the side of the target-specific code.
> 
> It's likely to be at least partly a target-independent issue, so tweaking
> the common code makes sense in principle.
> 
> Does adding !HARD_REGISTER_P (x) to:
> 
>   /* Show the output dies here.  This is necessary for SUBREGs
>      of pseudos since we cannot track their lifetimes correctly;
>      hard regs shouldn't appear here except as return values.  */
>   if (!reload_completed && !reload_in_progress
>       && REG_P (x) && !reg_overlap_mentioned_p (x, y))
>     emit_clobber (x);
> 
> in emit_move_complex_parts help?  If so, I think we should do at

Probably yes.  Quick test says the abovementioned mod makes the ad-hoc fix I posted earlier (https://gcc.gnu.org/pipermail/gcc-patches/2022-June/596626.html) a thing of the past.

> least that much.
> 
> Thanks,
> Richard
  

On 8/4/2022 3:49 AM, Richard Sandiford via Gcc-patches wrote:
>
>>> TBH I'm surprised we still run init_regs for LRA.  I thought there was
>>> a plan to stop doing that, but perhaps I misremember.
>> Sorry I am not sure about the status of LRA... because the xtensa port is still using reload.
> Ah, hadn't realised that.  If you have time to work on it, it would be
> really good to move over to LRA.  There are plans to remove old reload.
Definitely worth investigating.  With the cc0 removal done I think the 
last blocker for removing the old reload pass is gone.   We just need to 
get the remaining targets converted to LRA.

> It might be that old reload *does* treat a pseudo clobber as a conflict.
> I can't remember now.  If so, then zeroing the register wouldn't be
> too bad (for old reload only).
No idea anymore either.  I'd be a bit surprised since IIRC the main 
purpose was to tell the old uninit warning code that the entire object 
was set by the subsequent libcall sequence.  But all that code is long-gone.

Which I think raises a question.  Do we even need those CLOBBERSs anymore?


>
>> As conclusion, trying to tweak the common code side may have been a bit premature.
>> I'll consider if I can deal with those issues on the side of the target-specific code.
> It's likely to be at least partly a target-independent issue, so tweaking
> the common code makes sense in principle.
>
> Does adding !HARD_REGISTER_P (x) to:
>
>    /* Show the output dies here.  This is necessary for SUBREGs
>       of pseudos since we cannot track their lifetimes correctly;
>       hard regs shouldn't appear here except as return values.  */
>    if (!reload_completed && !reload_in_progress
>        && REG_P (x) && !reg_overlap_mentioned_p (x, y))
>      emit_clobber (x);
>
> in emit_move_complex_parts help?  If so, I think we should do at
> least that much.
If we can't remove the CLOBBERs entirely, then this sounds like a good 
thing, even if it doesn't help this specific case.

jeff
  

On 8/4/2022 6:35 AM, Takayuki 'January June' Suwa via Gcc-patches wrote:
> So lots, but almost double might be an overstatement :)
>
> BTW after some quick experimentation, I found that turning on -fsplit-wide-types-early would roughly (but not completely) solve the problem.  Surely, the output was not so bad in the past...
It could have been.


>> Ah, hadn't realised that.  If you have time to work on it, it would be
>> really good to move over to LRA.  There are plans to remove old reload.
> Alas you do overestimate me :) I've only been working about the GCC development for a little over a year.
> Well it's a lie that I'm not interested in it, but too much for me.
It may actually be trivial -- change TARGET_LRA_P to be 
hook_bool_void_true in the xtensa port, then rebuild & test. In the 
couple conversions I've done it's exposed a very small number of issues 
that were trivially resolved.  Given the what I've seen from you I would 
expect it's within your capabilities and there's folks here that can 
help if you do run into problems.

jeff