SSSE3 strcpy/stpcpy for x86-64

This patch adds SSSE3 strcpy/stpcpy. I got up to 4X speed up on Core 2
and Core i7.  I disabled it on Atom since SSSE3 version is slower for
shorter (<64byte) data.

ChangeLog

@@ -1,3 +1,22 @@
+2009-06-30  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* string/stpncpy.c (STPNCPY): New.  Defined if not defined.
+	(__stpncpy): Renamed to ...
+	(STPNCPY): This.
+	(stpncpy): Create alias only if STPNCPY is not defined.
+	* string/strncpy.c (STRNCPY): New.  Defined to strncpy if not
+	defined.
+	(strncpy): Renamed to ...
+	(STRNCPY): This.
+	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
+	 stpncpy-c strncpy-c for string.
+	* sysdeps/x86_64/multiarch/stpcpy.S: New file.
+	* sysdeps/x86_64/multiarch/stpncpy-c.c: New file.
+	* sysdeps/x86_64/multiarch/stpncpy.S: New file.
+	* sysdeps/x86_64/multiarch/strcpy.S: New file.
+	* sysdeps/x86_64/multiarch/strncpy-c.c: New file.
+	* sysdeps/x86_64/multiarch/strncpy.S: New file.
+
 2009-07-02  Ulrich Drepper  <drepper@redhat.com>
 
 	* malloc/malloc.c [ATOMIC_FASTBINS] (_int_free): Add full barrier when

string/stpncpy.c

@@ -28,17 +28,19 @@
 # include <sys/types.h>
 #endif
 
-#ifndef weak_alias
+#ifndef STPNCPY
-# define __stpncpy stpncpy
+# ifdef weak_alias
+#  define STPNCPY	__stpncpy
+weak_alias (__stpncpy, stpncpy)
+# else
+#  define STPNCPY	stpncpy
+# endif
 #endif
 
 /* Copy no more than N characters of SRC to DEST, returning the address of
    the terminating '\0' in DEST, if any, or else DEST + N.  */
 char *
-__stpncpy (dest, src, n)
+STPNCPY (char *dest, const char *src, size_t n)
-     char *dest;
-     const char *src;
-     size_t n;
 {
   char c;
   char *s = dest;
@@ -96,5 +98,4 @@ __stpncpy (dest, src, n)
 }
 #ifdef weak_alias
 libc_hidden_def (__stpncpy)
-weak_alias (__stpncpy, stpncpy)
 #endif

string/strncpy.c

@@ -21,11 +21,12 @@
 
 #undef strncpy
 
+#ifndef STRNCPY
+#define STRNCPY strncpy
+#endif
+
 char *
-strncpy (s1, s2, n)
+STRNCPY (char *s1, const char *s2, size_t n)
-     char *s1;
-     const char *s2;
-     size_t n;
 {
   reg_char c;
   char *s = s1;

sysdeps/x86_64/multiarch/Makefile

@@ -4,5 +4,5 @@ gen-as-const-headers += ifunc-defines.sym
 endif
 
 ifeq ($(subdir),string)
-sysdep_routines += strncmp-c
+sysdep_routines += stpncpy-c strncpy-c strncmp-c
 endif

sysdeps/x86_64/multiarch/stpcpy.S

@@ -0,0 +1,7 @@
+#define USE_AS_STPCPY
+#define STRCPY __stpcpy
+#include "strcpy.S"
+
+weak_alias (__stpcpy, stpcpy)
+libc_hidden_def (__stpcpy)
+libc_hidden_builtin_def (stpcpy)

sysdeps/x86_64/multiarch/stpncpy-c.c

@@ -0,0 +1,8 @@
+#define STPNCPY __stpncpy_sse2
+#ifdef SHARED
+#undef libc_hidden_def
+#define libc_hidden_def(name) \
+  __hidden_ver1 (__stpncpy_sse2, __GI___stpncpy, __stpncpy_sse2);
+#endif
+
+#include "stpncpy.c"

sysdeps/x86_64/multiarch/stpncpy.S

@@ -0,0 +1,6 @@
+#define STRCPY __stpncpy
+#define USE_AS_STPCPY
+#define USE_AS_STRNCPY
+#include "strcpy.S"
+
+weak_alias (__stpncpy, stpncpy)

sysdeps/x86_64/multiarch/strncpy-c.c

@@ -0,0 +1,8 @@
+#define STRNCPY __strncpy_sse2
+#ifdef SHARED
+#undef libc_hidden_builtin_def
+#define libc_hidden_builtin_def(name) \
+  __hidden_ver1 (__strncpy_sse2, __GI_strncpy, __strncpy_sse2);
+#endif
+
+#include "strncpy.c"

sysdeps/x86_64/multiarch/strncpy.S

@@ -0,0 +1,3 @@
+#define STRCPY strncpy
+#define USE_AS_STRNCPY
+#include "strcpy.S"