wcsmbs: optimize wcsncat

This patch rewrites wcsncat using wcslen, wcsnlen, and wmemcpy.  This is
similar to the optimization done on strncat by 3eb38795db and e80514b5a8.

Checked on x86_64-linux-gnu.

	* wcsmbs/wcsncat.c (wcsncat): Rewrite using wcslen, wcsnlen, and
	wmemcpy.

2 files changed, 10 insertions, 46 deletions

diff --git a/ChangeLog b/ChangeLog
index 2406275..d51188c 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,8 @@
 2019-02-27  Adhemerval Zanella  <adhemerval.zanella@linaro.org>
 
+	* wcsmbs/wcsncat.c (wcsncat): Rewrite using wcslen, wcsnlen, and
+	wmemcpy.
+
 	* wcsmbs/wcscpy.c (__wcpcpy): Rewrite using wcslen and wmemcpy.
 
 	* include/wchar.h (__wcscpy): New prototype.
diff --git a/wcsmbs/wcsncat.c b/wcsmbs/wcsncat.c
index 65e9b22..cb6fe71 100644
--- a/wcsmbs/wcsncat.c
+++ b/wcsmbs/wcsncat.c
@@ -26,54 +26,15 @@
 wchar_t *
 WCSNCAT (wchar_t *dest, const wchar_t *src, size_t n)
 {
-  wchar_t c;
-  wchar_t * const s = dest;
+  wchar_t *ret = dest;
 
-  /* Find the end of DEST.  */
-  do
-    c = *dest++;
-  while (c != L'\0');
+  /* Find the end of dest.  */
+  dest += __wcslen (dest);
 
-  /* Make DEST point before next character, so we can increment
-     it while memory is read (wins on pipelined cpus).	*/
-  dest -= 2;
+  size_t ds = __wcsnlen (src, n);
 
-  if (n >= 4)
-    {
-      size_t n4 = n >> 2;
-      do
-	{
-	  c = *src++;
-	  *++dest = c;
-	  if (c == L'\0')
-	    return s;
-	  c = *src++;
-	  *++dest = c;
-	  if (c == L'\0')
-	    return s;
-	  c = *src++;
-	  *++dest = c;
-	  if (c == L'\0')
-	    return s;
-	  c = *src++;
-	  *++dest = c;
-	  if (c == L'\0')
-	    return s;
-	} while (--n4 > 0);
-      n &= 3;
-    }
+  dest[ds] = L'\0';
+  __wmemcpy (dest, src, ds);
 
-  while (n > 0)
-    {
-      c = *src++;
-      *++dest = c;
-      if (c == L'\0')
-	return s;
-      n--;
-    }
-
-  if (c != L'\0')
-    *++dest = L'\0';
-
-  return s;
+  return ret;
 }