Fwd: Alignment check in strlen

Fwd: Alignment check in strlen

[Jonas Wagner]

[-- Attachment #1.1: Type: text/plain, Size: 517 bytes --]

Dear all,

attached is a patch that makes the alignment check in strlen more
consistent with the one in memcpy, and hopefully faster.

In src/string/memcpy.c, I find the following:

#define ALIGN (sizeof(size_t)-1)
if (((uintptr_t)d & ALIGN) != ((uintptr_t)s & ALIGN))
  goto misaligned;

In src/string/strlen.c, a different check was used instead:

#define ALIGN (sizeof(size_t))
for (; (uintptr_t)s % ALIGN; s++) if (!*s) return s-a;

I do not think there is any particular reason for this difference.

Best,
Jonas

[-- Attachment #1.2: Type: text/html, Size: 1166 bytes --]

[-- Attachment #2: 0001-strlen-use-bitwise-AND-for-alignment-test-instead-of.patch --]
[-- Type: application/octet-stream, Size: 1018 bytes --]

From d2b8aff9b799a2402e7f7769bdfa511cc246768b Mon Sep 17 00:00:00 2001
From: Jonas Wagner <jonas.wagner@epfl.ch>
Date: Wed, 20 Feb 2013 18:50:01 +0100
Subject: [PATCH] strlen: use bitwise AND for alignment test, instead of %.

---
 src/string/strlen.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/string/strlen.c b/src/string/strlen.c
index d6f8631..92784ea 100644
--- a/src/string/strlen.c
+++ b/src/string/strlen.c
@@ -3,7 +3,7 @@
 #include <stdint.h>
 #include <limits.h>
 
-#define ALIGN (sizeof(size_t))
+#define ALIGN (sizeof(size_t)-1)
 #define ONES ((size_t)-1/UCHAR_MAX)
 #define HIGHS (ONES * (UCHAR_MAX/2+1))
 #define HASZERO(x) ((x)-ONES & ~(x) & HIGHS)
@@ -12,7 +12,7 @@ size_t strlen(const char *s)
 {
 	const char *a = s;
 	const size_t *w;
-	for (; (uintptr_t)s % ALIGN; s++) if (!*s) return s-a;
+	for (; (uintptr_t)s & ALIGN; s++) if (!*s) return s-a;
 	for (w = (const void *)s; !HASZERO(*w); w++);
 	for (s = (const void *)w; *s; s++);
 	return s-a;
-- 
1.7.10.4

Re: Fwd: Alignment check in strlen

[Szabolcs Nagy]

* Jonas Wagner <jonas.wagner@epfl.ch> [2013-02-20 19:05:28 +0100]:
> attached is a patch that makes the alignment check in strlen more
> consistent with the one in memcpy, and hopefully faster.
> 

there is already a patch set to make string.h functions
more consistent

that addresses more issues but needs clean up

see febr 4 and 5
http://www.openwall.com/lists/musl/2013/02/

so yes at some point these should be fixed..

> In src/string/memcpy.c, I find the following:
> 
> #define ALIGN (sizeof(size_t)-1)
> if (((uintptr_t)d & ALIGN) != ((uintptr_t)s & ALIGN))
>   goto misaligned;
> 
> In src/string/strlen.c, a different check was used instead:
> 
> #define ALIGN (sizeof(size_t))
> for (; (uintptr_t)s % ALIGN; s++) if (!*s) return s-a;
> 
> I do not think there is any particular reason for this difference.
> 
> Best,
> Jonas