[PATCH] lib: string.c: A speed optimized for strncpy

[PATCH] lib: string.c: A speed optimized for strncpy

[Rickard Strandqvist]

This variant is in my tests about 7-10% faster, and I also think
it is perhaps even clearer code than before.

I assume that more will do the testing, I do not know if we should do tests
on different types of hardware as well, my test was on a new Intel I7.

Rickard Strandqvist (1):
  lib: string.c:  A speed optimized for strncpy

 lib/string.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

-- 
1.7.10.4

[PATCH] lib: string.c: A speed optimized for strncpy

[Rickard Strandqvist]

This variant is in my tests about 7-10% faster, and also think
it is perhaps even clearer code than before.

Signed-off-by: Rickard Strandqvist <rickard_strandqvist@spectrumdigital.se>
---
 lib/string.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/lib/string.c b/lib/string.c
index f3c6ff5..6961229 100644
--- a/lib/string.c
+++ b/lib/string.c
@@ -123,12 +123,12 @@ char *strncpy(char *dest, const char *src, size_t count)
 {
 	char *tmp = dest;
 
-	while (count) {
-		if ((*tmp = *src) != 0)
-			src++;
-		tmp++;
-		count--;
-	}
+	while (count && (*tmp++ = *src++))
+		--count;
+
+	while (count--)
+		*tmp++ = '\0';
+
 	return dest;
 }
 EXPORT_SYMBOL(strncpy);
-- 
1.7.10.4

Re: [PATCH] lib: string.c: A speed optimized for strncpy

[Joe Perches]

On Sun, 2014-10-05 at 15:29 +0200, Rickard Strandqvist wrote:
> This variant is in my tests about 7-10% faster, and also think
> it is perhaps even clearer code than before.
[]
> diff --git a/lib/string.c b/lib/string.c
[]
> @@ -123,12 +123,12 @@ char *strncpy(char *dest, const char *src, size_t count)
>  {
>  	char *tmp = dest;
>  
> -	while (count) {
> -		if ((*tmp = *src) != 0)
> -			src++;
> -		tmp++;
> -		count--;
> -	}
> +	while (count && (*tmp++ = *src++))
> +		--count;
> +
> +	while (count--)
> +		*tmp++ = '\0';

Perhaps it could be faster to use memset.
It might depend on the value of count.

{
	while (count && (*tmp++ = *src++))
		count--;

	if (count > 0)
		memset(tmp, 0, count);
}

Re: [PATCH] lib: string.c: A speed optimized for strncpy

[Rickard Strandqvist]

Hi

Yes, it can be faster, even if it is as you say, probably a difference
depending on the size of the count.
And even greater need to test this on a variety of hardware :-/

But I try to do my test with the memset variant to.

Kind regards
Rickard Strandqvist


2014-10-05 17:36 GMT+02:00 Joe Perches <joe@perches.com>:
> On Sun, 2014-10-05 at 15:29 +0200, Rickard Strandqvist wrote:
>> This variant is in my tests about 7-10% faster, and also think
>> it is perhaps even clearer code than before.
> []
>> diff --git a/lib/string.c b/lib/string.c
> []
>> @@ -123,12 +123,12 @@ char *strncpy(char *dest, const char *src, size_t count)
>>  {
>>       char *tmp = dest;
>>
>> -     while (count) {
>> -             if ((*tmp = *src) != 0)
>> -                     src++;
>> -             tmp++;
>> -             count--;
>> -     }
>> +     while (count && (*tmp++ = *src++))
>> +             --count;
>> +
>> +     while (count--)
>> +             *tmp++ = '\0';
>
> Perhaps it could be faster to use memset.
> It might depend on the value of count.
>
> {
>         while (count && (*tmp++ = *src++))
>                 count--;
>
>         if (count > 0)
>                 memset(tmp, 0, count);
> }
>

Re: [PATCH] lib: string.c: A speed optimized for strncpy

[Dan Carpenter]

On Sun, Oct 05, 2014 at 06:01:43PM +0200, Rickard Strandqvist wrote:
> Hi
> 
> Yes, it can be faster, even if it is as you say, probably a difference
> depending on the size of the count.
> And even greater need to test this on a variety of hardware :-/

Most architectures (the notable exception is ARM) have an their own
optimized strncpy() function.  Probably strzcpy() should just call it.

char *strzcpy(char *dest, const char *src, size_t count)
{
	strncpy(dest, src, count);
	if (count)
		dest[count - 1] = '\0';
	return dest;
}

regards,
dan carpenter