[PATCH] Little memset_explicit optimisation

[PATCH] Little memset_explicit optimisation

[David CARLIER]

[-- Attachment #1: Type: text/plain, Size: 1 bytes --]



[-- Attachment #2: 0001-memzero_explicit-optimisation-for-size.patch --]
[-- Type: application/octet-stream, Size: 720 bytes --]

From c6bb758902b3b358de9a98bb4156c306685f59d7 Mon Sep 17 00:00:00 2001
From: David Carlier <devnexen@gmail.com>
Date: Sat, 17 Nov 2018 14:13:13 +0000
Subject: [PATCH] memzero_explicit, optimisation for size.

Using the return value of memset for save/load sake.

Signed-off-by: David Carlier <devnexen@gmail.com>
---
 lib/string.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/lib/string.c b/lib/string.c
index 38e4ca08e757..92da04a0213b 100644
--- a/lib/string.c
+++ b/lib/string.c
@@ -720,7 +720,7 @@ EXPORT_SYMBOL(memset);
  */
 void memzero_explicit(void *s, size_t count)
 {
-	memset(s, 0, count);
+	s = memset(s, 0, count);
 	barrier_data(s);
 }
 EXPORT_SYMBOL(memzero_explicit);
-- 
2.19.1

Re: [PATCH] Little memset_explicit optimisation

[Andy Shevchenko]

On Sat, Nov 24, 2018 at 12:35:43PM +0000, David CARLIER wrote:
> 

Hmm... Can we see the difference in assembly generation?


-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] Little memset_explicit optimisation

[Joey Pabalinas]

[-- Attachment #1: Type: text/plain, Size: 742 bytes --]

On Sat, Nov 24, 2018 at 12:35:43PM +0000, David CARLIER wrote:
> Using the return value of memset for save/load sake.
> 
> Signed-off-by: David Carlier <devnexen@gmail.com>
> ---
>  lib/string.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/lib/string.c b/lib/string.c
> index 38e4ca08e757..92da04a0213b 100644
> --- a/lib/string.c
> +++ b/lib/string.c
> @@ -720,7 +720,7 @@ EXPORT_SYMBOL(memset);
>   */
>  void memzero_explicit(void *s, size_t count)
>  {
> -       memset(s, 0, count);
> +       s = memset(s, 0, count);
>         barrier_data(s);
>  }
>  EXPORT_SYMBOL(memzero_explicit);

Could you elaborate on the optimization that this patch performs?

-- 
Cheers,
Joey Pabalinas

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH] Little memset_explicit optimisation

[David CARLIER]

Sorry I m not used yet at all to LKML rules.

So here a slight difference in assembly generated between the two
versions (amd64) :
`
     .loc 1 7 7
     leaq    -12(%rbp), %rax
     movq    %rax, -8(%rbp)
-    .loc 1 11 2
+    .loc 1 9 6
     movq    -8(%rbp), %rax
     movl    $4, %edx
     movl    $0, %esi
     movq    %rax, %rdi
     call    memset@PLT
+    movq    %rax, -8(%rbp)
     .loc 1 13 23
     movq    -8(%rbp), %rax
     movl    (%rax), %eax
`
On Mon, 26 Nov 2018 at 11:37, Joey Pabalinas <joeypabalinas@gmail.com> wrote:
>
> On Sat, Nov 24, 2018 at 12:35:43PM +0000, David CARLIER wrote:
> > Using the return value of memset for save/load sake.
> >
> > Signed-off-by: David Carlier <devnexen@gmail.com>
> > ---
> >  lib/string.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/lib/string.c b/lib/string.c
> > index 38e4ca08e757..92da04a0213b 100644
> > --- a/lib/string.c
> > +++ b/lib/string.c
> > @@ -720,7 +720,7 @@ EXPORT_SYMBOL(memset);
> >   */
> >  void memzero_explicit(void *s, size_t count)
> >  {
> > -       memset(s, 0, count);
> > +       s = memset(s, 0, count);
> >         barrier_data(s);
> >  }
> >  EXPORT_SYMBOL(memzero_explicit);
>
> Could you elaborate on the optimization that this patch performs?
>
> --
> Cheers,
> Joey Pabalinas

Re: Re: [PATCH] Little memset_explicit optimisation

[Joey Pabalinas]

[-- Attachment #1: Type: text/plain, Size: 1041 bytes --]

On Mon, Nov 26, 2018 at 07:36:19PM +0000, David CARLIER wrote:
> Sorry I m not used yet at all to LKML rules.
> 
> So here a slight difference in assembly generated between the two
> versions (amd64) :
> `
>      .loc 1 7 7
>      leaq    -12(%rbp), %rax
>      movq    %rax, -8(%rbp)
> -    .loc 1 11 2
> +    .loc 1 9 6
>      movq    -8(%rbp), %rax
>      movl    $4, %edx
>      movl    $0, %esi
>      movq    %rax, %rdi
>      call    memset@PLT
> +    movq    %rax, -8(%rbp)
>      .loc 1 13 23
>      movq    -8(%rbp), %rax
>      movl    (%rax), %eax

What is the advantage of having the added `movq %rax, -8(%rbp)` here?

The next instruction is `movq -8(%rbp), %rax` and nothing afterwords
uses the value stored in `-8(%rbp)`.

Also, is this compiled without optimization? Take a looks at the
assembly in a small test case with -O1 (making sure to use the target
variable so it isn't optimized out) and compare the assembly generated
with and without that assignment.

-- 
Cheers,
Joey Pabalinas

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]