diff mbox series

i386: Move vzeroupper pass from after reload pass to after postreload_cse [PR112760]

Message ID ZW+i2GmeDAwxA0ZB@tucnak
State New
Headers show
Series i386: Move vzeroupper pass from after reload pass to after postreload_cse [PR112760] | expand

Commit Message

Jakub Jelinek Dec. 5, 2023, 10:23 p.m. UTC 
Hi!

Regardless of the outcome of the REG_UNUSED discussions, I think
it is a good idea to move the vzeroupper pass one pass later.
As can be seen in the multiple PRs and as postreload.cc documents,
reload/LRA is known to create dead statements quite often, which
is the reason why we have postreload_cse pass at all.
Doing vzeroupper pass before such cleanup means the pass including
df_analyze for it needs to process more instructions than needed
and because mode switching adds note problem, also higher chance of
having stale REG_UNUSED notes.
And, I really don't see why vzeroupper can't wait until those cleanups
are done.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2023-12-05  Jakub Jelinek  <jakub@redhat.com>

	PR rtl-optimization/112760
	* config/i386/i386-passes.def (pass_insert_vzeroupper): Insert
	after pass_postreload_cse rather than pass_reload.
	* config/i386/i386-features.cc (rest_of_handle_insert_vzeroupper):
	Adjust comment for it.

	* gcc.dg/pr112760.c: New test.


	Jakub

Comments

Hongtao Liu Dec. 6, 2023, 1:20 a.m. UTC | #1 
On Wed, Dec 6, 2023 at 6:23 AM Jakub Jelinek <jakub@redhat.com> wrote:
>
> Hi!
>
> Regardless of the outcome of the REG_UNUSED discussions, I think
> it is a good idea to move the vzeroupper pass one pass later.
> As can be seen in the multiple PRs and as postreload.cc documents,
> reload/LRA is known to create dead statements quite often, which
> is the reason why we have postreload_cse pass at all.
> Doing vzeroupper pass before such cleanup means the pass including
> df_analyze for it needs to process more instructions than needed
> and because mode switching adds note problem, also higher chance of
> having stale REG_UNUSED notes.
> And, I really don't see why vzeroupper can't wait until those cleanups
> are done.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
LGTM.
>
> 2023-12-05  Jakub Jelinek  <jakub@redhat.com>
>
>         PR rtl-optimization/112760
>         * config/i386/i386-passes.def (pass_insert_vzeroupper): Insert
>         after pass_postreload_cse rather than pass_reload.
>         * config/i386/i386-features.cc (rest_of_handle_insert_vzeroupper):
>         Adjust comment for it.
>
>         * gcc.dg/pr112760.c: New test.
>
> --- gcc/config/i386/i386-passes.def.jj  2023-01-16 11:52:15.960735877 +0100
> +++ gcc/config/i386/i386-passes.def     2023-12-05 19:15:01.748279329 +0100
> @@ -24,7 +24,7 @@ along with GCC; see the file COPYING3.
>     REPLACE_PASS (PASS, INSTANCE, TGT_PASS)
>   */
>
> -  INSERT_PASS_AFTER (pass_reload, 1, pass_insert_vzeroupper);
> +  INSERT_PASS_AFTER (pass_postreload_cse, 1, pass_insert_vzeroupper);
>    INSERT_PASS_AFTER (pass_combine, 1, pass_stv, false /* timode_p */);
>    /* Run the 64-bit STV pass before the CSE pass so that CONST0_RTX and
>       CONSTM1_RTX generated by the STV pass can be CSEed.  */
> --- gcc/config/i386/i386-features.cc.jj 2023-11-02 07:49:15.029894060 +0100
> +++ gcc/config/i386/i386-features.cc    2023-12-05 19:15:48.658620698 +0100
> @@ -2627,10 +2627,11 @@ convert_scalars_to_vector (bool timode_p
>  static unsigned int
>  rest_of_handle_insert_vzeroupper (void)
>  {
> -  /* vzeroupper instructions are inserted immediately after reload to
> -     account for possible spills from 256bit or 512bit registers.  The pass
> -     reuses mode switching infrastructure by re-running mode insertion
> -     pass, so disable entities that have already been processed.  */
> +  /* vzeroupper instructions are inserted immediately after reload and
> +     postreload_cse to clean up after it a little bit to account for possible
> +     spills from 256bit or 512bit registers.  The pass reuses mode switching
> +     infrastructure by re-running mode insertion pass, so disable entities
> +     that have already been processed.  */
>    for (int i = 0; i < MAX_386_ENTITIES; i++)
>      ix86_optimize_mode_switching[i] = 0;
>
> --- gcc/testsuite/gcc.dg/pr112760.c.jj  2023-12-01 13:46:57.444746529 +0100
> +++ gcc/testsuite/gcc.dg/pr112760.c     2023-12-01 13:46:36.729036971 +0100
> @@ -0,0 +1,22 @@
> +/* PR rtl-optimization/112760 */
> +/* { dg-do run } */
> +/* { dg-options "-O2 -fno-dce -fno-guess-branch-probability --param=max-cse-insns=0" } */
> +/* { dg-additional-options "-m8bit-idiv -mavx" { target i?86-*-* x86_64-*-* } } */
> +
> +unsigned g;
> +
> +__attribute__((__noipa__)) unsigned short
> +foo (unsigned short a, unsigned short b)
> +{
> +  unsigned short x = __builtin_add_overflow_p (a, g, (unsigned short) 0);
> +  g -= g / b;
> +  return x;
> +}
> +
> +int
> +main ()
> +{
> +  unsigned short x = foo (40, 6);
> +  if (x != 0)
> +    __builtin_abort ();
> +}
>
>         Jakub
>
diff mbox series

Patch

--- gcc/config/i386/i386-passes.def.jj	2023-01-16 11:52:15.960735877 +0100
+++ gcc/config/i386/i386-passes.def	2023-12-05 19:15:01.748279329 +0100
@@ -24,7 +24,7 @@  along with GCC; see the file COPYING3.
    REPLACE_PASS (PASS, INSTANCE, TGT_PASS)
  */
 
-  INSERT_PASS_AFTER (pass_reload, 1, pass_insert_vzeroupper);
+  INSERT_PASS_AFTER (pass_postreload_cse, 1, pass_insert_vzeroupper);
   INSERT_PASS_AFTER (pass_combine, 1, pass_stv, false /* timode_p */);
   /* Run the 64-bit STV pass before the CSE pass so that CONST0_RTX and
      CONSTM1_RTX generated by the STV pass can be CSEed.  */
--- gcc/config/i386/i386-features.cc.jj	2023-11-02 07:49:15.029894060 +0100
+++ gcc/config/i386/i386-features.cc	2023-12-05 19:15:48.658620698 +0100
@@ -2627,10 +2627,11 @@  convert_scalars_to_vector (bool timode_p
 static unsigned int
 rest_of_handle_insert_vzeroupper (void)
 {
-  /* vzeroupper instructions are inserted immediately after reload to
-     account for possible spills from 256bit or 512bit registers.  The pass
-     reuses mode switching infrastructure by re-running mode insertion
-     pass, so disable entities that have already been processed.  */
+  /* vzeroupper instructions are inserted immediately after reload and
+     postreload_cse to clean up after it a little bit to account for possible
+     spills from 256bit or 512bit registers.  The pass reuses mode switching
+     infrastructure by re-running mode insertion pass, so disable entities
+     that have already been processed.  */
   for (int i = 0; i < MAX_386_ENTITIES; i++)
     ix86_optimize_mode_switching[i] = 0;
 
--- gcc/testsuite/gcc.dg/pr112760.c.jj	2023-12-01 13:46:57.444746529 +0100
+++ gcc/testsuite/gcc.dg/pr112760.c	2023-12-01 13:46:36.729036971 +0100
@@ -0,0 +1,22 @@ 
+/* PR rtl-optimization/112760 */
+/* { dg-do run } */
+/* { dg-options "-O2 -fno-dce -fno-guess-branch-probability --param=max-cse-insns=0" } */
+/* { dg-additional-options "-m8bit-idiv -mavx" { target i?86-*-* x86_64-*-* } } */
+
+unsigned g;
+
+__attribute__((__noipa__)) unsigned short
+foo (unsigned short a, unsigned short b)
+{
+  unsigned short x = __builtin_add_overflow_p (a, g, (unsigned short) 0);
+  g -= g / b;
+  return x;
+}
+
+int
+main ()
+{
+  unsigned short x = foo (40, 6);
+  if (x != 0)
+    __builtin_abort ();
+}