Re: [PATCH] i386: fix ix86_hardreg_mov_ok with lra_in_progress

[Hongtao Liu]

On Mon, May 6, 2024 at 3:40 PM Kong, Lingling  wrote:
>
> Hi,
> Originally eliminate_regs_in_insn will transform
> (parallel [
>   (set (reg:QI 130)
> (plus:QI (subreg:QI (reg:DI 19 frame) 0)
>   (const_int 96)))
>   (clobber (reg:CC 17 flag))]) {*addqi_1}
> to
> (set (reg:QI 130)
>   (subreg:QI (reg:DI 19 frame) 0)) {*movqi_internal}
> when verify_changes.
>
> But with No Flags add, it transforms
> (set (reg:QI 5 di)
>   (plus:QI (subreg:QI (reg:DI 19 frame) 0)
>(const_int 96))) {*addqi_1_nf}
> to
> (set (reg:QI 5 di)
>  (subreg:QI (reg:DI 19 frame) 0)) {*addqi_1_nf}.
> there is no extra clobbers at the end, and its dest reg just is a hardreg. 
> For ix86_hardreg_mov_ok, it returns false. So it fails to update insn and 
> causes the ICE when transform to movqi_internal.
>
> But actually it is ok and safe for ix86_hardreg_mov_ok when lra_in_progress.
>
> And tested the spec2017, the performance was not affected.
> Bootstrapped and regtested on x86_64-pc-linux-gnu. OK for trunk?
Ok.
>
> gcc/ChangeLog:
>
> * config/i386/i386.cc (ix86_hardreg_mov_ok): Relax
> hard reg mov restriction when lra in progress.
> ---
>  gcc/config/i386/i386.cc | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/gcc/config/i386/i386.cc b/gcc/config/i386/i386.cc index 
> 4d6b2b98761..ca4348a18bf 100644
> --- a/gcc/config/i386/i386.cc
> +++ b/gcc/config/i386/i386.cc
> @@ -20357,7 +20357,8 @@ ix86_hardreg_mov_ok (rtx dst, rtx src)
>? standard_sse_constant_p (src, GET_MODE (dst))
>: x86_64_immediate_operand (src, GET_MODE (dst)))
>&& ix86_class_likely_spilled_p (REGNO_REG_CLASS (REGNO (dst)))
> -  && !reload_completed)
> +  && !reload_completed
> +  && !lra_in_progress)
>  return false;
>return true;
>  }
> --
> 2.31.1
>


-- 
BR,
Hongtao

[PATCH] i386: fix ix86_hardreg_mov_ok with lra_in_progress

[Kong, Lingling]

Hi,
Originally eliminate_regs_in_insn will transform 
(parallel [
  (set (reg:QI 130)
(plus:QI (subreg:QI (reg:DI 19 frame) 0)
  (const_int 96)))
  (clobber (reg:CC 17 flag))]) {*addqi_1} 
to 
(set (reg:QI 130) 
  (subreg:QI (reg:DI 19 frame) 0)) {*movqi_internal}
when verify_changes.

But with No Flags add, it transforms
(set (reg:QI 5 di)
  (plus:QI (subreg:QI (reg:DI 19 frame) 0)
   (const_int 96))) {*addqi_1_nf}
to
(set (reg:QI 5 di)
 (subreg:QI (reg:DI 19 frame) 0)) {*addqi_1_nf}.
there is no extra clobbers at the end, and its dest reg just is a hardreg. For 
ix86_hardreg_mov_ok, it returns false. So it fails to update insn and causes 
the ICE when transform to movqi_internal.

But actually it is ok and safe for ix86_hardreg_mov_ok when lra_in_progress.

And tested the spec2017, the performance was not affected.
Bootstrapped and regtested on x86_64-pc-linux-gnu. OK for trunk?

gcc/ChangeLog:

* config/i386/i386.cc (ix86_hardreg_mov_ok): Relax
hard reg mov restriction when lra in progress.
---
 gcc/config/i386/i386.cc | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/gcc/config/i386/i386.cc b/gcc/config/i386/i386.cc index 
4d6b2b98761..ca4348a18bf 100644
--- a/gcc/config/i386/i386.cc
+++ b/gcc/config/i386/i386.cc
@@ -20357,7 +20357,8 @@ ix86_hardreg_mov_ok (rtx dst, rtx src)
   ? standard_sse_constant_p (src, GET_MODE (dst))
   : x86_64_immediate_operand (src, GET_MODE (dst)))
   && ix86_class_likely_spilled_p (REGNO_REG_CLASS (REGNO (dst)))
-  && !reload_completed)
+  && !reload_completed
+  && !lra_in_progress)
 return false;
   return true;
 }
--
2.31.1