diff mbox series

[v2,2/3] x86: Move and slightly improve memset_erms

Message ID 20220629221204.1242709-2-goldstein.w.n@gmail.com
State Superseded
Headers
Series [v2,1/3] x86: Add definition for __wmemset_chk AVX2 RTM in ifunc impl list |

Checks

Context Check Description
dj/TryBot-apply_patch success Patch applied to master at the time it was sent

Commit Message

Noah Goldstein June 29, 2022, 10:12 p.m. UTC 
  Implementation wise:
    1. Remove the VZEROUPPER as memset_{impl}_unaligned_erms does not
       use the L(stosb) label that was previously defined.

    2. Don't give the hotpath (fallthrough) to zero size.

Code positioning wise:

Move memset_{chk}_erms to its own file.  Leaving it in between the
memset_{impl}_unaligned both adds unnecessary complexity to the
file and wastes space in a relatively hot cache section.
---
 sysdeps/x86_64/multiarch/Makefile             |  1 +
 sysdeps/x86_64/multiarch/memset-erms.S        | 25 +++++++++++++++
 .../multiarch/memset-vec-unaligned-erms.S     | 31 -------------------
 3 files changed, 26 insertions(+), 31 deletions(-)
 create mode 100644 sysdeps/x86_64/multiarch/memset-erms.S

Comments

H.J. Lu June 29, 2022, 10:18 p.m. UTC | #1 
On Wed, Jun 29, 2022 at 3:12 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
>
> Implementation wise:
>     1. Remove the VZEROUPPER as memset_{impl}_unaligned_erms does not
>        use the L(stosb) label that was previously defined.
>
>     2. Don't give the hotpath (fallthrough) to zero size.
>
> Code positioning wise:
>
> Move memset_{chk}_erms to its own file.  Leaving it in between the
> memset_{impl}_unaligned both adds unnecessary complexity to the
> file and wastes space in a relatively hot cache section.
> ---
>  sysdeps/x86_64/multiarch/Makefile             |  1 +
>  sysdeps/x86_64/multiarch/memset-erms.S        | 25 +++++++++++++++
>  .../multiarch/memset-vec-unaligned-erms.S     | 31 -------------------
>  3 files changed, 26 insertions(+), 31 deletions(-)
>  create mode 100644 sysdeps/x86_64/multiarch/memset-erms.S
>
> diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
> index 62a4d96fb8..18cea04423 100644
> --- a/sysdeps/x86_64/multiarch/Makefile
> +++ b/sysdeps/x86_64/multiarch/Makefile
> @@ -30,6 +30,7 @@ sysdep_routines += \
>    memset-avx2-unaligned-erms-rtm \
>    memset-avx512-no-vzeroupper \
>    memset-avx512-unaligned-erms \
> +  memset-erms \
>    memset-evex-unaligned-erms \
>    memset-sse2-unaligned-erms \
>    rawmemchr-avx2 \
> diff --git a/sysdeps/x86_64/multiarch/memset-erms.S b/sysdeps/x86_64/multiarch/memset-erms.S
> new file mode 100644
> index 0000000000..1fce0c9fcc
> --- /dev/null
> +++ b/sysdeps/x86_64/multiarch/memset-erms.S
> @@ -0,0 +1,25 @@

Need copyright notice.

> +#include <sysdep.h>
> +
> +#if defined USE_MULTIARCH && IS_IN (libc)
> +       .text
> +ENTRY (__memset_chk_erms)
> +       cmp     %RDX_LP, %RCX_LP
> +       jb      HIDDEN_JUMPTARGET (__chk_fail)
> +END (__memset_chk_erms)
> +
> +/* Only used to measure performance of REP STOSB.  */
> +ENTRY (__memset_erms)
> +       /* Skip zero length.  */
> +       test    %RDX_LP, %RDX_LP
> +       jz       L(stosb_return_zero)
> +       mov     %RDX_LP, %RCX_LP
> +       movzbl  %sil, %eax
> +       mov     %RDI_LP, %RDX_LP
> +       rep stosb
> +       mov     %RDX_LP, %RAX_LP
> +       ret
> +L(stosb_return_zero):
> +       movq    %rdi, %rax
> +       ret
> +END (__memset_erms)
> +#endif
> diff --git a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
> index abc12d9cda..905d0fa464 100644
> --- a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
> +++ b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
> @@ -156,37 +156,6 @@ L(entry_from_wmemset):
>  #if defined USE_MULTIARCH && IS_IN (libc)
>  END (MEMSET_SYMBOL (__memset, unaligned))
>
> -# if VEC_SIZE == 16
> -ENTRY (__memset_chk_erms)
> -       cmp     %RDX_LP, %RCX_LP
> -       jb      HIDDEN_JUMPTARGET (__chk_fail)
> -END (__memset_chk_erms)
> -
> -/* Only used to measure performance of REP STOSB.  */
> -ENTRY (__memset_erms)
> -       /* Skip zero length.  */
> -       test    %RDX_LP, %RDX_LP
> -       jnz      L(stosb)
> -       movq    %rdi, %rax
> -       ret
> -# else
> -/* Provide a hidden symbol to debugger.  */
> -       .hidden MEMSET_SYMBOL (__memset, erms)
> -ENTRY (MEMSET_SYMBOL (__memset, erms))
> -# endif
> -L(stosb):
> -       mov     %RDX_LP, %RCX_LP
> -       movzbl  %sil, %eax
> -       mov     %RDI_LP, %RDX_LP
> -       rep stosb
> -       mov     %RDX_LP, %RAX_LP
> -       VZEROUPPER_RETURN
> -# if VEC_SIZE == 16
> -END (__memset_erms)
> -# else
> -END (MEMSET_SYMBOL (__memset, erms))
> -# endif
> -
>  # if defined SHARED && IS_IN (libc)
>  ENTRY_CHK (MEMSET_CHK_SYMBOL (__memset_chk, unaligned_erms))
>         cmp     %RDX_LP, %RCX_LP
> --
> 2.34.1
>
Noah Goldstein June 29, 2022, 11:09 p.m. UTC | #2 
On Wed, Jun 29, 2022 at 3:19 PM H.J. Lu <hjl.tools@gmail.com> wrote:
>
> On Wed, Jun 29, 2022 at 3:12 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote:
> >
> > Implementation wise:
> >     1. Remove the VZEROUPPER as memset_{impl}_unaligned_erms does not
> >        use the L(stosb) label that was previously defined.
> >
> >     2. Don't give the hotpath (fallthrough) to zero size.
> >
> > Code positioning wise:
> >
> > Move memset_{chk}_erms to its own file.  Leaving it in between the
> > memset_{impl}_unaligned both adds unnecessary complexity to the
> > file and wastes space in a relatively hot cache section.
> > ---
> >  sysdeps/x86_64/multiarch/Makefile             |  1 +
> >  sysdeps/x86_64/multiarch/memset-erms.S        | 25 +++++++++++++++
> >  .../multiarch/memset-vec-unaligned-erms.S     | 31 -------------------
> >  3 files changed, 26 insertions(+), 31 deletions(-)
> >  create mode 100644 sysdeps/x86_64/multiarch/memset-erms.S
> >
> > diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
> > index 62a4d96fb8..18cea04423 100644
> > --- a/sysdeps/x86_64/multiarch/Makefile
> > +++ b/sysdeps/x86_64/multiarch/Makefile
> > @@ -30,6 +30,7 @@ sysdep_routines += \
> >    memset-avx2-unaligned-erms-rtm \
> >    memset-avx512-no-vzeroupper \
> >    memset-avx512-unaligned-erms \
> > +  memset-erms \
> >    memset-evex-unaligned-erms \
> >    memset-sse2-unaligned-erms \
> >    rawmemchr-avx2 \
> > diff --git a/sysdeps/x86_64/multiarch/memset-erms.S b/sysdeps/x86_64/multiarch/memset-erms.S
> > new file mode 100644
> > index 0000000000..1fce0c9fcc
> > --- /dev/null
> > +++ b/sysdeps/x86_64/multiarch/memset-erms.S
> > @@ -0,0 +1,25 @@
>
> Need copyright notice.

Fixed in V3.
>
> > +#include <sysdep.h>
> > +
> > +#if defined USE_MULTIARCH && IS_IN (libc)
> > +       .text
> > +ENTRY (__memset_chk_erms)
> > +       cmp     %RDX_LP, %RCX_LP
> > +       jb      HIDDEN_JUMPTARGET (__chk_fail)
> > +END (__memset_chk_erms)
> > +
> > +/* Only used to measure performance of REP STOSB.  */
> > +ENTRY (__memset_erms)
> > +       /* Skip zero length.  */
> > +       test    %RDX_LP, %RDX_LP
> > +       jz       L(stosb_return_zero)
> > +       mov     %RDX_LP, %RCX_LP
> > +       movzbl  %sil, %eax
> > +       mov     %RDI_LP, %RDX_LP
> > +       rep stosb
> > +       mov     %RDX_LP, %RAX_LP
> > +       ret
> > +L(stosb_return_zero):
> > +       movq    %rdi, %rax
> > +       ret
> > +END (__memset_erms)
> > +#endif
> > diff --git a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
> > index abc12d9cda..905d0fa464 100644
> > --- a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
> > +++ b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
> > @@ -156,37 +156,6 @@ L(entry_from_wmemset):
> >  #if defined USE_MULTIARCH && IS_IN (libc)
> >  END (MEMSET_SYMBOL (__memset, unaligned))
> >
> > -# if VEC_SIZE == 16
> > -ENTRY (__memset_chk_erms)
> > -       cmp     %RDX_LP, %RCX_LP
> > -       jb      HIDDEN_JUMPTARGET (__chk_fail)
> > -END (__memset_chk_erms)
> > -
> > -/* Only used to measure performance of REP STOSB.  */
> > -ENTRY (__memset_erms)
> > -       /* Skip zero length.  */
> > -       test    %RDX_LP, %RDX_LP
> > -       jnz      L(stosb)
> > -       movq    %rdi, %rax
> > -       ret
> > -# else
> > -/* Provide a hidden symbol to debugger.  */
> > -       .hidden MEMSET_SYMBOL (__memset, erms)
> > -ENTRY (MEMSET_SYMBOL (__memset, erms))
> > -# endif
> > -L(stosb):
> > -       mov     %RDX_LP, %RCX_LP
> > -       movzbl  %sil, %eax
> > -       mov     %RDI_LP, %RDX_LP
> > -       rep stosb
> > -       mov     %RDX_LP, %RAX_LP
> > -       VZEROUPPER_RETURN
> > -# if VEC_SIZE == 16
> > -END (__memset_erms)
> > -# else
> > -END (MEMSET_SYMBOL (__memset, erms))
> > -# endif
> > -
> >  # if defined SHARED && IS_IN (libc)
> >  ENTRY_CHK (MEMSET_CHK_SYMBOL (__memset_chk, unaligned_erms))
> >         cmp     %RDX_LP, %RCX_LP
> > --
> > 2.34.1
> >
>
>
> --
> H.J.
diff mbox series

Patch

diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index 62a4d96fb8..18cea04423 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -30,6 +30,7 @@  sysdep_routines += \
   memset-avx2-unaligned-erms-rtm \
   memset-avx512-no-vzeroupper \
   memset-avx512-unaligned-erms \
+  memset-erms \
   memset-evex-unaligned-erms \
   memset-sse2-unaligned-erms \
   rawmemchr-avx2 \
diff --git a/sysdeps/x86_64/multiarch/memset-erms.S b/sysdeps/x86_64/multiarch/memset-erms.S
new file mode 100644
index 0000000000..1fce0c9fcc
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/memset-erms.S
@@ -0,0 +1,25 @@ 
+#include <sysdep.h>
+
+#if defined USE_MULTIARCH && IS_IN (libc)
+	.text
+ENTRY (__memset_chk_erms)
+	cmp	%RDX_LP, %RCX_LP
+	jb	HIDDEN_JUMPTARGET (__chk_fail)
+END (__memset_chk_erms)
+
+/* Only used to measure performance of REP STOSB.  */
+ENTRY (__memset_erms)
+	/* Skip zero length.  */
+	test	%RDX_LP, %RDX_LP
+	jz	 L(stosb_return_zero)
+	mov	%RDX_LP, %RCX_LP
+	movzbl	%sil, %eax
+	mov	%RDI_LP, %RDX_LP
+	rep stosb
+	mov	%RDX_LP, %RAX_LP
+	ret
+L(stosb_return_zero):
+	movq	%rdi, %rax
+	ret
+END (__memset_erms)
+#endif
diff --git a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
index abc12d9cda..905d0fa464 100644
--- a/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
+++ b/sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S
@@ -156,37 +156,6 @@  L(entry_from_wmemset):
 #if defined USE_MULTIARCH && IS_IN (libc)
 END (MEMSET_SYMBOL (__memset, unaligned))
 
-# if VEC_SIZE == 16
-ENTRY (__memset_chk_erms)
-	cmp	%RDX_LP, %RCX_LP
-	jb	HIDDEN_JUMPTARGET (__chk_fail)
-END (__memset_chk_erms)
-
-/* Only used to measure performance of REP STOSB.  */
-ENTRY (__memset_erms)
-	/* Skip zero length.  */
-	test	%RDX_LP, %RDX_LP
-	jnz	 L(stosb)
-	movq	%rdi, %rax
-	ret
-# else
-/* Provide a hidden symbol to debugger.  */
-	.hidden	MEMSET_SYMBOL (__memset, erms)
-ENTRY (MEMSET_SYMBOL (__memset, erms))
-# endif
-L(stosb):
-	mov	%RDX_LP, %RCX_LP
-	movzbl	%sil, %eax
-	mov	%RDI_LP, %RDX_LP
-	rep stosb
-	mov	%RDX_LP, %RAX_LP
-	VZEROUPPER_RETURN
-# if VEC_SIZE == 16
-END (__memset_erms)
-# else
-END (MEMSET_SYMBOL (__memset, erms))
-# endif
-
 # if defined SHARED && IS_IN (libc)
 ENTRY_CHK (MEMSET_CHK_SYMBOL (__memset_chk, unaligned_erms))
 	cmp	%RDX_LP, %RCX_LP