bca718
From 1026f1e15a84134fd19f58c98af85ec9474f7722 Mon Sep 17 00:00:00 2001
bca718
From: Stefan Liebler <stli@linux.vnet.ibm.com>
bca718
Date: Thu, 8 Oct 2015 11:21:50 +0200
bca718
Subject: [PATCH 19/30] S390: Optimize strncmp and wcsncmp. (AND COMMON-CODE
bca718
 WCSNCMP - picked form upstream)
bca718
bca718
upstream-commit-id: cee82e70ccb7b2f054cd781b0a603ae244523e72
bca718
https://www.sourceware.org/ml/libc-alpha/2015-07/msg00087.html
bca718
bca718
common-code wcsncmp:
bca718
upstream-commit-id: 920a0395ba9fa5949ec87aaf5daa0259da16749d
bca718
https://www.sourceware.org/ml/libc-alpha/2015-04/msg00098.html
bca718
bca718
This patch provides optimized versions of strncmp and wcsncmp with the z13
bca718
vector instructions.
bca718
bca718
ChangeLog:
bca718
bca718
	* sysdeps/s390/multiarch/strncmp-c.c: New File.
bca718
	* sysdeps/s390/multiarch/strncmp-vx.S: Likewise.
bca718
	* sysdeps/s390/multiarch/strncmp.c: Likewise.
bca718
	* sysdeps/s390/multiarch/wcsncmp-c.c: Likewise.
bca718
	* sysdeps/s390/multiarch/wcsncmp-vx.S: Likewise.
bca718
	* sysdeps/s390/multiarch/wcsncmp.c: Likewise.
bca718
	* sysdeps/s390/multiarch/Makefile (sysdep_routines): Add strncmp and
bca718
	wcsncmp functions.
bca718
	* sysdeps/s390/multiarch/ifunc-impl-list.c
bca718
	(__libc_ifunc_impl_list): Add ifunc test for strncmp, wcsncmp.
bca718
	* wcsmbs/wcsncmp.c (WCSNCMP): Define and use macro.
bca718
	* benchtests/bench-strncmp.c: Add wcsncmp support.
bca718
	* benchtests/bench-wcsncmp.c: New File.
bca718
	* benchtests/Makefile (wcsmbs-bench): Add wcsncmp.
bca718
---
bca718
 benchtests/Makefile                      |   2 +-
bca718
 benchtests/bench-strncmp.c               | 104 +++++++++++++-----
bca718
 benchtests/bench-wcsncmp.c               |  20 ++++
bca718
 localedata/tests-mbwc/dat_wcsncmp.c      |  18 ++--
bca718
 localedata/tests-mbwc/tst_wcsncmp.c      |   2 +
bca718
 string/strncmp.c                         |   2 +-
bca718
 string/test-strncmp.c                    | 175 +++++++++++++++++++++---------
bca718
 sysdeps/s390/multiarch/Makefile          |   6 +-
bca718
 sysdeps/s390/multiarch/ifunc-impl-list.c |   3 +
bca718
 sysdeps/s390/multiarch/strncmp-c.c       |  28 +++++
bca718
 sysdeps/s390/multiarch/strncmp-vx.S      | 137 ++++++++++++++++++++++++
bca718
 sysdeps/s390/multiarch/strncmp.c         |  30 ++++++
bca718
 sysdeps/s390/multiarch/wcsncmp-c.c       |  25 +++++
bca718
 sysdeps/s390/multiarch/wcsncmp-vx.S      | 177 +++++++++++++++++++++++++++++++
bca718
 sysdeps/s390/multiarch/wcsncmp.c         |  27 +++++
bca718
 wcsmbs/Makefile                          |   2 +-
bca718
 wcsmbs/test-wcsncmp-ifunc.c              |  20 ++++
bca718
 wcsmbs/test-wcsncmp.c                    |   2 +
bca718
 wcsmbs/wcsncmp.c                         |  41 +++----
bca718
 19 files changed, 716 insertions(+), 105 deletions(-)
bca718
 create mode 100644 benchtests/bench-wcsncmp.c
bca718
 create mode 100644 sysdeps/s390/multiarch/strncmp-c.c
bca718
 create mode 100644 sysdeps/s390/multiarch/strncmp-vx.S
bca718
 create mode 100644 sysdeps/s390/multiarch/strncmp.c
bca718
 create mode 100644 sysdeps/s390/multiarch/wcsncmp-c.c
bca718
 create mode 100644 sysdeps/s390/multiarch/wcsncmp-vx.S
bca718
 create mode 100644 sysdeps/s390/multiarch/wcsncmp.c
bca718
 create mode 100644 wcsmbs/test-wcsncmp-ifunc.c
bca718
 create mode 100644 wcsmbs/test-wcsncmp.c
bca718
bca718
diff --git a/benchtests/Makefile b/benchtests/Makefile
bca718
index f6333eb..f6342da 100644
bca718
--- a/benchtests/Makefile
bca718
+++ b/benchtests/Makefile
bca718
@@ -39,7 +39,7 @@ string-bench := bcopy bzero memccpy memchr memcmp memcpy memmem memmove \
bca718
 		strncasecmp strncat strncmp strncpy strnlen strpbrk strrchr \
bca718
 		strspn strstr strcpy_chk stpcpy_chk memrchr strsep strtok
bca718
 wcsmbs-bench := wcslen wcsnlen wcscpy wcpcpy wcsncpy wcpncpy wcscat wcsncat \
bca718
-		wcsncmp
bca718
+		wcsncmp wcsncmp
bca718
 string-bench-all := $(string-bench) ${wcsmbs-bench}
bca718
 
bca718
 stdlib-bench := strtod
bca718
diff --git a/benchtests/bench-strncmp.c b/benchtests/bench-strncmp.c
bca718
index 25df3db..496ed68 100644
bca718
--- a/benchtests/bench-strncmp.c
bca718
+++ b/benchtests/bench-strncmp.c
bca718
@@ -17,17 +17,66 @@
bca718
    <http://www.gnu.org/licenses/>.  */
bca718
 
bca718
 #define TEST_MAIN
bca718
-#define TEST_NAME "strncmp"
bca718
+#ifdef WIDE
bca718
+# define TEST_NAME "wcsncmp"
bca718
+#else
bca718
+# define TEST_NAME "strncmp"
bca718
+#endif /* !WIDE */
bca718
 #include "bench-string.h"
bca718
 
bca718
-typedef int (*proto_t) (const char *, const char *, size_t);
bca718
-int simple_strncmp (const char *, const char *, size_t);
bca718
-int stupid_strncmp (const char *, const char *, size_t);
bca718
+#ifdef WIDE
bca718
+# include <wchar.h>
bca718
+
bca718
+# define L(str) L##str
bca718
+# define STRNCMP wcsncmp
bca718
+# define SIMPLE_STRNCMP simple_wcsncmp
bca718
+# define STUPID_STRNCMP stupid_wcsncmp
bca718
+# define CHAR wchar_t
bca718
+# define CHARBYTES 4
bca718
+/* Wcsncmp uses signed semantics for comparison, not unsigned.
bca718
+   Avoid using substraction since possible overflow.  */
bca718
+int
bca718
+simple_wcsncmp (const CHAR *s1, const CHAR *s2, size_t n)
bca718
+{
bca718
+  wchar_t c1, c2;
bca718
+  while (n--)
bca718
+    {
bca718
+      c1 = *s1++;
bca718
+      c2 = *s2++;
bca718
+      if (c1 == L ('\0') || c1 != c2)
bca718
+	return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
+    }
bca718
+  return 0;
bca718
+}
bca718
 
bca718
-IMPL (stupid_strncmp, 0)
bca718
-IMPL (simple_strncmp, 0)
bca718
-IMPL (strncmp, 1)
bca718
+int
bca718
+stupid_wcsncmp (const CHAR *s1, const CHAR *s2, size_t n)
bca718
+{
bca718
+  wchar_t c1, c2;
bca718
+  size_t ns1 = wcsnlen (s1, n) + 1, ns2 = wcsnlen (s2, n) + 1;
bca718
+
bca718
+  n = ns1 < n ? ns1 : n;
bca718
+  n = ns2 < n ? ns2 : n;
bca718
 
bca718
+  while (n--)
bca718
+    {
bca718
+      c1 = *s1++;
bca718
+      c2 = *s2++;
bca718
+      if (c1 != c2)
bca718
+	return c1 > c2 ? 1 : -1;
bca718
+    }
bca718
+  return 0;
bca718
+}
bca718
+
bca718
+#else
bca718
+# define L(str) str
bca718
+# define STRNCMP strncmp
bca718
+# define SIMPLE_STRNCMP simple_strncmp
bca718
+# define STUPID_STRNCMP stupid_strncmp
bca718
+# define CHAR char
bca718
+# define CHARBYTES 1
bca718
+
bca718
+/* Strncmp uses unsigned semantics for comparison.  */
bca718
 int
bca718
 simple_strncmp (const char *s1, const char *s2, size_t n)
bca718
 {
bca718
@@ -49,9 +98,16 @@ stupid_strncmp (const char *s1, const char *s2, size_t n)
bca718
   while (n-- && (ret = *(unsigned char *) s1++ - * (unsigned char *) s2++) == 0);
bca718
   return ret;
bca718
 }
bca718
+#endif /* !WIDE */
bca718
+
bca718
+typedef int (*proto_t) (const CHAR *, const CHAR *, size_t);
bca718
+
bca718
+IMPL (STUPID_STRNCMP, 0)
bca718
+IMPL (SIMPLE_STRNCMP, 0)
bca718
+IMPL (STRNCMP, 1)
bca718
 
bca718
 static void
bca718
-do_one_test (impl_t *impl, const char *s1, const char *s2, size_t n,
bca718
+do_one_test (impl_t *impl, const CHAR *s1, const CHAR *s2, size_t n,
bca718
 	     int exp_result)
bca718
 {
bca718
   size_t i, iters = INNER_LOOP_ITERS;
bca718
@@ -74,12 +130,12 @@ do_test_limit (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
 	 int exp_result)
bca718
 {
bca718
   size_t i, align_n;
bca718
-  char *s1, *s2;
bca718
+  CHAR *s1, *s2;
bca718
 
bca718
   if (n == 0)
bca718
     {
bca718
-      s1 = (char*)(buf1 + page_size);
bca718
-      s2 = (char*)(buf2 + page_size);
bca718
+      s1 = (CHAR*)(buf1 + page_size);
bca718
+      s2 = (CHAR*)(buf2 + page_size);
bca718
       printf ("Length %4zd/%4zd:", len, n);
bca718
 
bca718
       FOR_EACH_IMPL (impl, 0)
bca718
@@ -92,16 +148,16 @@ do_test_limit (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
 
bca718
   align1 &= 15;
bca718
   align2 &= 15;
bca718
-  align_n = (page_size - n) & 15;
bca718
+  align_n = (page_size - n * CHARBYTES) & 15;
bca718
 
bca718
-  s1 = (char*)(buf1 + page_size - n);
bca718
-  s2 = (char*)(buf2 + page_size - n);
bca718
+  s1 = (CHAR*)(buf1 + page_size - n * CHARBYTES);
bca718
+  s2 = (CHAR*)(buf2 + page_size - n * CHARBYTES);
bca718
 
bca718
   if (align1 < align_n)
bca718
-    s1 -= (align_n - align1);
bca718
+    s1 = (CHAR *) ((char *) s1 - (align_n - align1));
bca718
 
bca718
   if (align2 < align_n)
bca718
-    s2 -= (align_n - align2);
bca718
+    s2 = (CHAR *) ((char *) s2 - (align_n - align2));
bca718
 
bca718
   for (i = 0; i < n; i++)
bca718
     s1[i] = s2[i] = 1 + 23 * i % max_char;
bca718
@@ -129,24 +185,24 @@ do_test (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
 	 int exp_result)
bca718
 {
bca718
   size_t i;
bca718
-  char *s1, *s2;
bca718
+  CHAR *s1, *s2;
bca718
 
bca718
   if (n == 0)
bca718
     return;
bca718
 
bca718
-  align1 &= 7;
bca718
-  if (align1 + n + 1 >= page_size)
bca718
+  align1 &= 63;
bca718
+  if (align1 + (n + 1) * CHARBYTES >= page_size)
bca718
     return;
bca718
 
bca718
   align2 &= 7;
bca718
-  if (align2 + n + 1 >= page_size)
bca718
+  if (align2 + (n + 1) * CHARBYTES >= page_size)
bca718
     return;
bca718
 
bca718
-  s1 = (char*)(buf1 + align1);
bca718
-  s2 = (char*)(buf2 + align2);
bca718
+  s1 = (CHAR*)(buf1 + align1);
bca718
+  s2 = (CHAR*)(buf2 + align2);
bca718
 
bca718
   for (i = 0; i < n; i++)
bca718
-    s1[i] = s2[i] = 1 + 23 * i % max_char;
bca718
+    s1[i] = s2[i] = 1 + (23 << ((CHARBYTES - 1) * 8)) * i % max_char;
bca718
 
bca718
   s1[n] = 24 + exp_result;
bca718
   s2[n] = 23;
bca718
@@ -162,7 +218,7 @@ do_test (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
   printf ("Length %4zd/%4zd, alignment %2zd/%2zd:", len, n, align1, align2);
bca718
 
bca718
   FOR_EACH_IMPL (impl, 0)
bca718
-    do_one_test (impl, (char*)s1, (char*)s2, n, exp_result);
bca718
+    do_one_test (impl, s1, s2, n, exp_result);
bca718
 
bca718
   putchar ('\n');
bca718
 }
bca718
diff --git a/benchtests/bench-wcsncmp.c b/benchtests/bench-wcsncmp.c
bca718
new file mode 100644
bca718
index 0000000..8720060
bca718
--- /dev/null
bca718
+++ b/benchtests/bench-wcsncmp.c
bca718
@@ -0,0 +1,20 @@
bca718
+/* Measure wcsncmp functions.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#define WIDE 1
bca718
+#include "bench-strncmp.c"
bca718
diff --git a/localedata/tests-mbwc/dat_wcsncmp.c b/localedata/tests-mbwc/dat_wcsncmp.c
bca718
index 167ce48..f468a8b 100644
bca718
--- a/localedata/tests-mbwc/dat_wcsncmp.c
bca718
+++ b/localedata/tests-mbwc/dat_wcsncmp.c
bca718
@@ -33,7 +33,7 @@ TST_WCSNCMP tst_wcsncmp_loc [] = {
bca718
       },
bca718
       { /*input.*/ { { 0x00D1,0x00D2,0x00D3,0x0000 },
bca718
 		     { 0x0000,0x00D2,0x00D3,0x0000 }, 3 },  /* #06 */
bca718
-	/*expect*/ { 0,1,0x00D1,			},
bca718
+	/*expect*/ { 0,1,1,			},
bca718
       },
bca718
       { /*input.*/ { { 0x00D1,0x00D2,0x00D3,0x0000 },
bca718
 		     { 0x00D1,0x00D2,0x00D9,0x0000 }, 2 },  /* #07 */
bca718
@@ -41,11 +41,11 @@ TST_WCSNCMP tst_wcsncmp_loc [] = {
bca718
       },
bca718
       { /*input.*/ { { 0x00D1,0x00D2,0x00D3,0x0000 },
bca718
 		     { 0x00D1,0x00D2,0x00D9,0x0000 }, 3 },  /* #08 */
bca718
-	/*expect*/ { 0,1,-0x0006,			},
bca718
+	/*expect*/ { 0,1,-1,			},
bca718
       },
bca718
       { /*input.*/ { { 0x00D1,0x00D2,0x00D3,0x0000 },
bca718
 		     { 0x00D1,0x00D2,0x0000	   }, 4 },  /* #09 */
bca718
-	/*expect*/ { 0,1,0x00D3,			},
bca718
+	/*expect*/ { 0,1,1,			},
bca718
       },
bca718
       { .is_last = 1 }
bca718
     }
bca718
@@ -75,7 +75,7 @@ TST_WCSNCMP tst_wcsncmp_loc [] = {
bca718
       },
bca718
       { /*input.*/ { { 0x0041,0x0042,0x0043,0x0000 },
bca718
 		     { 0x0000,0x0042,0x0043,0x0000 }, 3 },  /* #06 */
bca718
-	/*expect*/ { 0,1,0x0041,			},
bca718
+	/*expect*/ { 0,1,1,			},
bca718
       },
bca718
       { /*input.*/ { { 0x0041,0x0042,0x0043,0x0000 },
bca718
 		     { 0x0041,0x0042,0x0049,0x0000 }, 2 },  /* #07 */
bca718
@@ -83,11 +83,11 @@ TST_WCSNCMP tst_wcsncmp_loc [] = {
bca718
       },
bca718
       { /*input.*/ { { 0x0041,0x0042,0x0043,0x0000 },
bca718
 		     { 0x0041,0x0042,0x0049,0x0000 }, 3 },  /* #08 */
bca718
-	/*expect*/ { 0,1,-0x0006,			},
bca718
+	/*expect*/ { 0,1,-1,			},
bca718
       },
bca718
       { /*input.*/ { { 0x0041,0x0042,0x0043,0x0000 },
bca718
 		     { 0x0041,0x0042,0x0000	   }, 4 },  /* #09 */
bca718
-	/*expect*/ { 0,1,0x0043,			},
bca718
+	/*expect*/ { 0,1,1,			},
bca718
       },
bca718
       { .is_last = 1 }
bca718
     }
bca718
@@ -117,7 +117,7 @@ TST_WCSNCMP tst_wcsncmp_loc [] = {
bca718
       },
bca718
       { /*input.*/ { { 0x3041,0x3042,0x3043,0x0000 },
bca718
 		     { 0x0000,0x3042,0x3043,0x0000 }, 3 },  /* #06 */
bca718
-	/*expect*/ { 0,1,0x3041,			},
bca718
+	/*expect*/ { 0,1,1,			},
bca718
       },
bca718
       { /*input.*/ { { 0x3041,0x3042,0x3043,0x0000 },
bca718
 		     { 0x3041,0x3042,0x3049,0x0000 }, 2 },  /* #07 */
bca718
@@ -125,11 +125,11 @@ TST_WCSNCMP tst_wcsncmp_loc [] = {
bca718
       },
bca718
       { /*input.*/ { { 0x3041,0x3042,0x3043,0x0000 },
bca718
 		     { 0x3041,0x3042,0x3049,0x0000 }, 3 },  /* #08 */
bca718
-	/*expect*/ { 0,1,-0x0006,			},
bca718
+	/*expect*/ { 0,1,-1,			},
bca718
       },
bca718
       { /*input.*/ { { 0x3041,0x3042,0x3043,0x0000 },
bca718
 		     { 0x3041,0x3042,0x0000	   }, 4 },  /* #09 */
bca718
-	/*expect*/ { 0,1,0x3043,			},
bca718
+	/*expect*/ { 0,1,1,			},
bca718
       },
bca718
       { .is_last = 1 }
bca718
     }
bca718
diff --git a/localedata/tests-mbwc/tst_wcsncmp.c b/localedata/tests-mbwc/tst_wcsncmp.c
bca718
index d046ecd..e378efb 100644
bca718
--- a/localedata/tests-mbwc/tst_wcsncmp.c
bca718
+++ b/localedata/tests-mbwc/tst_wcsncmp.c
bca718
@@ -24,6 +24,8 @@ tst_wcsncmp (FILE * fp, int debug_flg)
bca718
       ws2 = TST_INPUT (wcsncmp).ws2;
bca718
       n = TST_INPUT (wcsncmp).n;
bca718
       ret = wcsncmp (ws1, ws2, n);
bca718
+      ret = (ret > 0 ? 1 : ret < 0 ? -1 : 0);
bca718
+
bca718
 
bca718
       if (debug_flg)
bca718
 	{
bca718
diff --git a/string/strncmp.c b/string/strncmp.c
bca718
index d79305a..bd52138 100644
bca718
--- a/string/strncmp.c
bca718
+++ b/string/strncmp.c
bca718
@@ -21,7 +21,7 @@
bca718
 #undef strncmp
bca718
 
bca718
 #ifndef STRNCMP
bca718
-#define STRNCMP strncmp
bca718
+# define STRNCMP strncmp
bca718
 #endif
bca718
 
bca718
 /* Compare no more than N characters of S1 and S2,
bca718
diff --git a/string/test-strncmp.c b/string/test-strncmp.c
bca718
index 7169593..950bf24 100644
bca718
--- a/string/test-strncmp.c
bca718
+++ b/string/test-strncmp.c
bca718
@@ -1,5 +1,5 @@
bca718
 /* Test and measure strncmp functions.
bca718
-   Copyright (C) 1999-2012 Free Software Foundation, Inc.
bca718
+   Copyright (C) 1999-2015 Free Software Foundation, Inc.
bca718
    This file is part of the GNU C Library.
bca718
    Written by Jakub Jelinek <jakub@redhat.com>, 1999.
bca718
 
bca718
@@ -18,18 +18,82 @@
bca718
    <http://www.gnu.org/licenses/>.  */
bca718
 
bca718
 #define TEST_MAIN
bca718
-#define TEST_NAME "strncmp"
bca718
+#ifdef WIDE
bca718
+# define TEST_NAME "wcsncmp"
bca718
+#else
bca718
+# define TEST_NAME "strncmp"
bca718
+#endif
bca718
 #include "test-string.h"
bca718
 
bca718
-typedef int (*proto_t) (const char *, const char *, size_t);
bca718
-int simple_strncmp (const char *, const char *, size_t);
bca718
-int stupid_strncmp (const char *, const char *, size_t);
bca718
+#ifdef WIDE
bca718
+# include <wchar.h>
bca718
+
bca718
+# define L(str) L##str
bca718
+# define STRNCMP wcsncmp
bca718
+# define STRCPY wcscpy
bca718
+# define STRDUP wcsdup
bca718
+# define MEMCPY wmemcpy
bca718
+# define SIMPLE_STRNCMP simple_wcsncmp
bca718
+# define STUPID_STRNCMP stupid_wcsncmp
bca718
+# define CHAR wchar_t
bca718
+# define UCHAR wchar_t
bca718
+# define CHARBYTES 4
bca718
+# define CHAR__MAX WCHAR_MAX
bca718
+# define CHAR__MIN WCHAR_MIN
bca718
+
bca718
+/* Wcsncmp uses signed semantics for comparison, not unsigned.
bca718
+   Avoid using substraction since possible overflow */
bca718
+int
bca718
+simple_wcsncmp (const CHAR *s1, const CHAR *s2, size_t n)
bca718
+{
bca718
+  wchar_t c1, c2;
bca718
+
bca718
+  while (n--)
bca718
+    {
bca718
+      c1 = *s1++;
bca718
+      c2 = *s2++;
bca718
+      if (c1 == L('\0') || c1 != c2)
bca718
+	return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
+    }
bca718
+  return 0;
bca718
+}
bca718
 
bca718
-IMPL (stupid_strncmp, 0)
bca718
-IMPL (simple_strncmp, 0)
bca718
-IMPL (strncmp, 1)
bca718
 
bca718
 int
bca718
+stupid_wcsncmp (const CHAR *s1, const CHAR *s2, size_t n)
bca718
+{
bca718
+  wchar_t c1, c2;
bca718
+  size_t ns1 = wcsnlen (s1, n) + 1, ns2 = wcsnlen (s2, n) + 1;
bca718
+
bca718
+  n = ns1 < n ? ns1 : n;
bca718
+  n = ns2 < n ? ns2 : n;
bca718
+
bca718
+  while (n--)
bca718
+    {
bca718
+      c1 = *s1++;
bca718
+      c2 = *s2++;
bca718
+      if (c1 != c2)
bca718
+	return c1 > c2 ? 1 : -1;
bca718
+    }
bca718
+  return 0;
bca718
+}
bca718
+
bca718
+#else
bca718
+# define L(str) str
bca718
+# define STRNCMP strncmp
bca718
+# define STRCPY strcpy
bca718
+# define STRDUP strdup
bca718
+# define MEMCPY memcpy
bca718
+# define SIMPLE_STRNCMP simple_strncmp
bca718
+# define STUPID_STRNCMP stupid_strncmp
bca718
+# define CHAR char
bca718
+# define UCHAR unsigned char
bca718
+# define CHARBYTES 1
bca718
+# define CHAR__MAX CHAR_MAX
bca718
+# define CHAR__MIN CHAR_MIN
bca718
+
bca718
+/* Strncmp uses unsigned semantics for comparison. */
bca718
+int
bca718
 simple_strncmp (const char *s1, const char *s2, size_t n)
bca718
 {
bca718
   int ret = 0;
bca718
@@ -50,9 +114,16 @@ stupid_strncmp (const char *s1, const char *s2, size_t n)
bca718
   while (n-- && (ret = *(unsigned char *) s1++ - * (unsigned char *) s2++) == 0);
bca718
   return ret;
bca718
 }
bca718
+#endif
bca718
+
bca718
+typedef int (*proto_t) (const CHAR *, const CHAR *, size_t);
bca718
+
bca718
+IMPL (STUPID_STRNCMP, 0)
bca718
+IMPL (SIMPLE_STRNCMP, 0)
bca718
+IMPL (STRNCMP, 1)
bca718
 
bca718
 static int
bca718
-check_result (impl_t *impl, const char *s1, const char *s2, size_t n,
bca718
+check_result (impl_t *impl, const CHAR *s1, const CHAR *s2, size_t n,
bca718
 	     int exp_result)
bca718
 {
bca718
   int result = CALL (impl, s1, s2, n);
bca718
@@ -70,7 +141,7 @@ check_result (impl_t *impl, const char *s1, const char *s2, size_t n,
bca718
 }
bca718
 
bca718
 static void
bca718
-do_one_test (impl_t *impl, const char *s1, const char *s2, size_t n,
bca718
+do_one_test (impl_t *impl, const CHAR *s1, const CHAR *s2, size_t n,
bca718
 	     int exp_result)
bca718
 {
bca718
   if (check_result (impl, s1, s2, n, exp_result) < 0)
bca718
@@ -82,12 +153,12 @@ do_test_limit (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
 	 int exp_result)
bca718
 {
bca718
   size_t i, align_n;
bca718
-  char *s1, *s2;
bca718
+  CHAR *s1, *s2;
bca718
 
bca718
   if (n == 0)
bca718
     {
bca718
-      s1 = (char*)(buf1 + page_size);
bca718
-      s2 = (char*)(buf2 + page_size);
bca718
+      s1 = (CHAR*) (buf1 + page_size);
bca718
+      s2 = (CHAR*) (buf2 + page_size);
bca718
 
bca718
       FOR_EACH_IMPL (impl, 0)
bca718
 	do_one_test (impl, s1, s2, n, 0);
bca718
@@ -97,16 +168,16 @@ do_test_limit (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
 
bca718
   align1 &= 15;
bca718
   align2 &= 15;
bca718
-  align_n = (page_size - n) & 15;
bca718
+  align_n = (page_size - n * CHARBYTES) & 15;
bca718
 
bca718
-  s1 = (char*)(buf1 + page_size - n);
bca718
-  s2 = (char*)(buf2 + page_size - n);
bca718
+  s1 = (CHAR*) (buf1 + page_size - n * CHARBYTES);
bca718
+  s2 = (CHAR*) (buf2 + page_size - n * CHARBYTES);
bca718
 
bca718
   if (align1 < align_n)
bca718
-    s1 -= (align_n - align1);
bca718
+    s1 = (CHAR *) ((char *) s1 - (align_n - align1));
bca718
 
bca718
   if (align2 < align_n)
bca718
-    s2 -= (align_n - align2);
bca718
+    s2 = (CHAR *) ((char *) s2 - (align_n - align2));
bca718
 
bca718
   for (i = 0; i < n; i++)
bca718
     s1[i] = s2[i] = 1 + 23 * i % max_char;
bca718
@@ -130,24 +201,24 @@ do_test (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
 	 int exp_result)
bca718
 {
bca718
   size_t i;
bca718
-  char *s1, *s2;
bca718
+  CHAR *s1, *s2;
bca718
 
bca718
   if (n == 0)
bca718
     return;
bca718
 
bca718
-  align1 &= 7;
bca718
-  if (align1 + n + 1 >= page_size)
bca718
+  align1 &= 63;
bca718
+  if (align1 + (n + 1) * CHARBYTES >= page_size)
bca718
     return;
bca718
 
bca718
   align2 &= 7;
bca718
-  if (align2 + n + 1 >= page_size)
bca718
+  if (align2 + (n + 1) * CHARBYTES >= page_size)
bca718
     return;
bca718
 
bca718
-  s1 = (char*)(buf1 + align1);
bca718
-  s2 = (char*)(buf2 + align2);
bca718
+  s1 = (CHAR*) (buf1 + align1);
bca718
+  s2 = (CHAR*) (buf2 + align2);
bca718
 
bca718
   for (i = 0; i < n; i++)
bca718
-    s1[i] = s2[i] = 1 + 23 * i % max_char;
bca718
+    s1[i] = s2[i] = 1 + (23 << ((CHARBYTES - 1) * 8)) * i % max_char;
bca718
 
bca718
   s1[n] = 24 + exp_result;
bca718
   s2[n] = 23;
bca718
@@ -161,19 +232,20 @@ do_test (size_t align1, size_t align2, size_t len, size_t n, int max_char,
bca718
     s2[n - 1] -= exp_result;
bca718
 
bca718
   FOR_EACH_IMPL (impl, 0)
bca718
-    do_one_test (impl, (char*)s1, (char*)s2, n, exp_result);
bca718
+    do_one_test (impl, s1, s2, n, exp_result);
bca718
 }
bca718
 
bca718
 static void
bca718
-do_page_test (size_t offset1, size_t offset2, char *s2)
bca718
+do_page_test (size_t offset1, size_t offset2, CHAR *s2)
bca718
 {
bca718
-  char *s1;
bca718
+  CHAR *s1;
bca718
   int exp_result;
bca718
 
bca718
-  if (offset1 >= page_size || offset2 >= page_size)
bca718
+  if (offset1 * CHARBYTES >= page_size || offset2 * CHARBYTES >= page_size)
bca718
     return;
bca718
 
bca718
-  s1 = (char *) (buf1 + offset1);
bca718
+  s1 = (CHAR *) buf1;
bca718
+  s1 += offset1;
bca718
   s2 += offset2;
bca718
 
bca718
   exp_result= *s1;
bca718
@@ -191,8 +263,8 @@ do_random_tests (void)
bca718
   size_t i, j, n, align1, align2, pos, len1, len2, size;
bca718
   int result;
bca718
   long r;
bca718
-  unsigned char *p1 = buf1 + page_size - 512;
bca718
-  unsigned char *p2 = buf2 + page_size - 512;
bca718
+  UCHAR *p1 = (UCHAR *) (buf1 + page_size - 512 * CHARBYTES);
bca718
+  UCHAR *p2 = (UCHAR *) (buf2 + page_size - 512 * CHARBYTES);
bca718
 
bca718
   for (n = 0; n < ITERATIONS; n++)
bca718
     {
bca718
@@ -240,7 +312,7 @@ do_random_tests (void)
bca718
 	}
bca718
 
bca718
       result = 0;
bca718
-      memcpy (p2 + align2, p1 + align1, pos);
bca718
+      MEMCPY (p2 + align2, p1 + align1, pos);
bca718
       if (pos < len1)
bca718
 	{
bca718
 	  if (p2[align2 + pos] == p1[align1 + pos])
bca718
@@ -263,7 +335,7 @@ do_random_tests (void)
bca718
 
bca718
       FOR_EACH_IMPL (impl, 1)
bca718
 	{
bca718
-	  r = CALL (impl, (char*)(p1 + align1), (char*)(p2 + align2), size);
bca718
+	  r = CALL (impl, (CHAR *)(p1 + align1), (CHAR *)(p2 + align2), size);
bca718
 	  /* Test whether on 64-bit architectures where ABI requires
bca718
 	     callee to promote has the promotion been done.  */
bca718
 	  asm ("" : "=g" (r) : "0" (r));
bca718
@@ -282,19 +354,26 @@ do_random_tests (void)
bca718
 static void
bca718
 check1 (void)
bca718
 {
bca718
-  char *s1 = (char *)(buf1 + 0xb2c);
bca718
-  char *s2 = (char *)(buf1 + 0xfd8);
bca718
-  size_t i;
bca718
+  CHAR *s1 = (CHAR *)(buf1 + 0xb2c);
bca718
+  CHAR *s2 = (CHAR *)(buf1 + 0xfd8);
bca718
+  size_t i, offset;
bca718
   int exp_result;
bca718
 
bca718
-  strcpy(s1, "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrs");
bca718
-  strcpy(s2, "abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijkLMNOPQRSTUV");
bca718
+  strcpy(s1, L("abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrs"));
bca718
+  strcpy(s2, L("abcdefghijklmnopqrstuvwxyzabcdefghijklmnopqrstuvwxyzabcdefghijkLMNOPQRSTUV"));
bca718
+
bca718
+  /* Check possible overflow bug for wcsncmp */
bca718
+  s1[4] = CHAR__MAX;
bca718
+  s2[4] = CHAR__MIN;
bca718
 
bca718
-  for (i = 0; i < 80; i++)
bca718
+  for (offset = 0; offset < 6; offset++)
bca718
     {
bca718
-      exp_result = simple_strncmp (s1, s2, i);
bca718
-      FOR_EACH_IMPL (impl, 0)
bca718
-	 check_result (impl, s1, s2, i, exp_result);
bca718
+      for (i = 0; i < 80; i++)
bca718
+	{
bca718
+	  exp_result = SIMPLE_STRNCMP (s1 + offset, s2 + offset, i);
bca718
+	  FOR_EACH_IMPL (impl, 0)
bca718
+	    check_result (impl, s1 + offset, s2 + offset, i, exp_result);
bca718
+	}
bca718
     }
bca718
 }
bca718
 
bca718
@@ -302,17 +381,17 @@ static void
bca718
 check2 (void)
bca718
 {
bca718
   size_t i;
bca718
-  char *s1, *s2;
bca718
+  CHAR *s1, *s2;
bca718
 
bca718
-  s1 = (char *) buf1;
bca718
-  for (i = 0; i < page_size - 1; i++)
bca718
+  s1 = (CHAR *) buf1;
bca718
+  for (i = 0; i < (page_size / CHARBYTES) - 1; i++)
bca718
     s1[i] = 23;
bca718
   s1[i] = 0;
bca718
 
bca718
-  s2 = strdup (s1);
bca718
+  s2 = STRDUP (s1);
bca718
 
bca718
   for (i = 0; i < 64; ++i)
bca718
-    do_page_test (3990 + i, 2635, s2);
bca718
+    do_page_test ((3988 / CHARBYTES) + i, (2636 / CHARBYTES), s2);
bca718
 
bca718
   free (s2);
bca718
 }
bca718
diff --git a/sysdeps/s390/multiarch/Makefile b/sysdeps/s390/multiarch/Makefile
bca718
index d8fbd55..d77bee5 100644
bca718
--- a/sysdeps/s390/multiarch/Makefile
bca718
+++ b/sysdeps/s390/multiarch/Makefile
bca718
@@ -7,7 +7,8 @@ sysdep_routines += strlen strlen-vx strlen-c \
bca718
 		   stpncpy stpncpy-vx stpncpy-c \
bca718
 		   strcat strcat-vx strcat-c \
bca718
 		   strncat strncat-vx strncat-c \
bca718
-		   strcmp strcmp-vx
bca718
+		   strcmp strcmp-vx \
bca718
+		   strncmp strncmp-vx strncmp-c
bca718
 endif
bca718
 
bca718
 ifeq ($(subdir),wcsmbs)
bca718
@@ -19,5 +20,6 @@ sysdep_routines += wcslen wcslen-vx wcslen-c \
bca718
 		   wcpncpy wcpncpy-vx wcpncpy-c \
bca718
 		   wcscat wcscat-vx wcscat-c \
bca718
 		   wcsncat wcsncat-vx wcsncat-c \
bca718
-		   wcscmp wcscmp-vx wcscmp-c
bca718
+		   wcscmp wcscmp-vx wcscmp-c \
bca718
+		   wcsncmp wcsncmp-vx wcsncmp-c
bca718
 endif
bca718
diff --git a/sysdeps/s390/multiarch/ifunc-impl-list.c b/sysdeps/s390/multiarch/ifunc-impl-list.c
bca718
index 196d3ec..5bfc493 100644
bca718
--- a/sysdeps/s390/multiarch/ifunc-impl-list.c
bca718
+++ b/sysdeps/s390/multiarch/ifunc-impl-list.c
bca718
@@ -106,6 +106,9 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
bca718
   IFUNC_VX_IMPL (strcmp);
bca718
   IFUNC_VX_IMPL (wcscmp);
bca718
 
bca718
+  IFUNC_VX_IMPL (strncmp);
bca718
+  IFUNC_VX_IMPL (wcsncmp);
bca718
+
bca718
 #endif /* HAVE_S390_VX_ASM_SUPPORT */
bca718
 
bca718
   return i;
bca718
diff --git a/sysdeps/s390/multiarch/strncmp-c.c b/sysdeps/s390/multiarch/strncmp-c.c
bca718
new file mode 100644
bca718
index 0000000..75da859
bca718
--- /dev/null
bca718
+++ b/sysdeps/s390/multiarch/strncmp-c.c
bca718
@@ -0,0 +1,28 @@
bca718
+/* Default strncmp implementation for S/390.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
bca718
+# define STRNCMP  __strncmp_c
bca718
+# ifdef SHARED
bca718
+#  undef libc_hidden_builtin_def
bca718
+#  define libc_hidden_builtin_def(name)			\
bca718
+  __hidden_ver1 (__strncmp_c, __GI_strncmp, __strncmp_c);
bca718
+# endif /* SHARED */
bca718
+
bca718
+# include <string/strncmp.c>
bca718
+#endif /* HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc */
bca718
diff --git a/sysdeps/s390/multiarch/strncmp-vx.S b/sysdeps/s390/multiarch/strncmp-vx.S
bca718
new file mode 100644
bca718
index 0000000..36e99b8
bca718
--- /dev/null
bca718
+++ b/sysdeps/s390/multiarch/strncmp-vx.S
bca718
@@ -0,0 +1,137 @@
bca718
+/* Vector optimized 32/64 bit S/390 version of strncmp.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
bca718
+
bca718
+# include "sysdep.h"
bca718
+# include "asm-syntax.h"
bca718
+
bca718
+	.text
bca718
+
bca718
+/* int strncmp (const char *s1, const char *s2, size_t n)
bca718
+   Compare at most n characters of two strings.
bca718
+
bca718
+   Register usage:
bca718
+   -r0=tmp
bca718
+   -r1=tmp
bca718
+   -r2=s1
bca718
+   -r3=s2
bca718
+   -r4=n
bca718
+   -r5=current_len
bca718
+   -v16=part of s1
bca718
+   -v17=part of s2
bca718
+   -v18=index of unequal
bca718
+*/
bca718
+ENTRY(__strncmp_vx)
bca718
+	.machine "z13"
bca718
+	.machinemode "zarch_nohighgprs"
bca718
+
bca718
+# if !defined __s390x__
bca718
+	llgfr	%r4,%r4
bca718
+# endif /* !defined __s390x__ */
bca718
+
bca718
+	clgije	%r4,0,.Lend_equal /* Nothing to do if n == 0,  */
bca718
+	lghi	%r5,0		/* current_len = 0.  */
bca718
+
bca718
+.Lloop:
bca718
+	vlbb	%v16,0(%r5,%r2),6 /* Load s1 to block boundary.  */
bca718
+	vlbb	%v17,0(%r5,%r3),6 /* Load s2 to block boundary.  */
bca718
+	lcbb	%r0,0(%r5,%r2),6 /* Get loaded byte count of s1.  */
bca718
+	jo	.Llt16_1	/* Jump away if vr is not fully loaded.  */
bca718
+	lcbb	%r1,0(%r5,%r3),6 /* Get loaded byte count of s2.  */
bca718
+	jo	.Llt16_2	/* Jump away if vr is not fully loaded.  */
bca718
+	aghi	%r5,16		/* Both vrs are fully loaded.  */
bca718
+	clgrjhe	%r5,%r4,.Llastcmp /* If current_len >= n ->last compare.  */
bca718
+	vfenezbs %v18,%v16,%v17	/* Compare not equal with zero search.  */
bca718
+	jno	.Lfound
bca718
+
bca718
+	vlbb	%v16,0(%r5,%r2),6
bca718
+	vlbb	%v17,0(%r5,%r3),6
bca718
+	lcbb	%r0,0(%r5,%r2),6
bca718
+	jo	.Llt16_1
bca718
+	lcbb	%r1,0(%r5,%r3),6
bca718
+	jo	.Llt16_2
bca718
+	aghi	%r5,16
bca718
+	clgrjhe	%r5,%r4,.Llastcmp
bca718
+	vfenezbs %v18,%v16,%v17
bca718
+	jno	.Lfound
bca718
+
bca718
+	vlbb	%v16,0(%r5,%r2),6
bca718
+	vlbb	%v17,0(%r5,%r3),6
bca718
+	lcbb	%r0,0(%r5,%r2),6
bca718
+	jo	.Llt16_1
bca718
+	lcbb	%r1,0(%r5,%r3),6
bca718
+	jo	.Llt16_2
bca718
+	aghi	%r5,16
bca718
+	clgrjhe	%r5,%r4,.Llastcmp
bca718
+	vfenezbs %v18,%v16,%v17
bca718
+	jno	.Lfound
bca718
+
bca718
+	vlbb	%v16,0(%r5,%r2),6
bca718
+	vlbb	%v17,0(%r5,%r3),6
bca718
+	lcbb	%r0,0(%r5,%r2),6
bca718
+	jo	.Llt16_1
bca718
+	lcbb	%r1,0(%r5,%r3),6
bca718
+	jo	.Llt16_2
bca718
+	aghi	%r5,16
bca718
+	clgrjhe	%r5,%r4,.Llastcmp
bca718
+	vfenezbs %v18,%v16,%v17
bca718
+	jno	.Lfound
bca718
+	j	.Lloop
bca718
+
bca718
+.Llt16_1:
bca718
+	lcbb	%r1,0(%r5,%r3),6 /* Get loaded byte count ofs2.  */
bca718
+.Llt16_2:
bca718
+	clr	%r0,%r1		/* Compare logical.  */
bca718
+	locrh	%r0,%r1		/* Compute minimum of bytes loaded.  */
bca718
+	algfr	%r5,%r0		/* Add smallest loaded bytes to current_len.  */
bca718
+	clgrj	%r5,%r4,10,.Llastcmp /* If current_len >= n ->last compare.  */
bca718
+	vfenezbs %v18,%v16,%v17	/* Compare not equal with zero search.  */
bca718
+	vlgvb	%r1,%v18,7	/* Get not equal index or 16 if all equal.  */
bca718
+	clrjl	%r1,%r0,.Lfound /* Jump away if miscompare is within
bca718
+				    loaded bytes (index < loaded-bytes) */
bca718
+	j	.Lloop
bca718
+
bca718
+.Llastcmp:
bca718
+	/* Use comparision result only if located within first n characters.
bca718
+	   %r0: loaded byte count in vreg;
bca718
+	   %r5: current_len;
bca718
+	   %r4: n;
bca718
+	   (current_len - n): [0...16[
bca718
+	   First ignored match index: loaded bytes - (current_len-n): ]0...16]
bca718
+	*/
bca718
+	slgr	%r5,%r4		/* %r5 = current_len - n.  */
bca718
+	slr	%r0,%r5		/* %r0 = first ignored match index.  */
bca718
+	vfenezbs %v18,%v16,%v17	/* Compare not equal with zero search.  */
bca718
+	vlgvb	%r1,%v18,7	/* Get not equal index or 16 if all equal.  */
bca718
+	clrjl	%r1,%r0,.Lfound /* Jump away if miscompare is within
bca718
+				    loaded bytes and below n bytes.  */
bca718
+	j	.Lend_equal	/* Miscompare after n-bytes -> end equal.  */
bca718
+
bca718
+.Lfound:
bca718
+	/* Difference or end of string.  */
bca718
+	je	.Lend_equal
bca718
+	lghi	%r2,1
bca718
+	lghi	%r1,-1
bca718
+	locgrl	%r2,%r1
bca718
+	br	%r14
bca718
+.Lend_equal:
bca718
+	lghi	%r2,0
bca718
+	br	%r14
bca718
+END(__strncmp_vx)
bca718
+#endif /* HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc */
bca718
diff --git a/sysdeps/s390/multiarch/strncmp.c b/sysdeps/s390/multiarch/strncmp.c
bca718
new file mode 100644
bca718
index 0000000..1e1e05a
bca718
--- /dev/null
bca718
+++ b/sysdeps/s390/multiarch/strncmp.c
bca718
@@ -0,0 +1,30 @@
bca718
+/* Multiple versions of strncmp.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
bca718
+# include <string.h>
bca718
+# include <ifunc-resolve.h>
bca718
+
bca718
+
bca718
+# undef strcmp
bca718
+extern __typeof (strncmp) __strncmp;
bca718
+s390_vx_libc_ifunc2 (__strncmp, strncmp)
bca718
+
bca718
+#else
bca718
+# include <string/strncmp.c>
bca718
+#endif /* !(defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc) */
bca718
diff --git a/sysdeps/s390/multiarch/wcsncmp-c.c b/sysdeps/s390/multiarch/wcsncmp-c.c
bca718
new file mode 100644
bca718
index 0000000..058cd0c
bca718
--- /dev/null
bca718
+++ b/sysdeps/s390/multiarch/wcsncmp-c.c
bca718
@@ -0,0 +1,25 @@
bca718
+/* Default wcsncmp implementation for S/390.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
bca718
+# define WCSNCMP  __wcsncmp_c
bca718
+
bca718
+# include <wchar.h>
bca718
+extern __typeof (wcsncmp) __wcsncmp_c;
bca718
+# include <wcsmbs/wcsncmp.c>
bca718
+#endif
bca718
diff --git a/sysdeps/s390/multiarch/wcsncmp-vx.S b/sysdeps/s390/multiarch/wcsncmp-vx.S
bca718
new file mode 100644
bca718
index 0000000..9a44424
bca718
--- /dev/null
bca718
+++ b/sysdeps/s390/multiarch/wcsncmp-vx.S
bca718
@@ -0,0 +1,177 @@
bca718
+/* Vector optimized 32/64 bit S/390 version of wcsncmp.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
bca718
+
bca718
+# include "sysdep.h"
bca718
+# include "asm-syntax.h"
bca718
+
bca718
+	.text
bca718
+
bca718
+/* int wcsncmp (const wchar_t *s1, const wchar_t *s2, size_t n)
bca718
+   Compare at most n characters of two strings.
bca718
+
bca718
+   Register usage:
bca718
+   -r0=tmp
bca718
+   -r1=tmp
bca718
+   -r2=s1
bca718
+   -r3=s2
bca718
+   -r4=n
bca718
+   -r5=current_len
bca718
+   -v16=part of s1
bca718
+   -v17=part of s2
bca718
+   -v18=index of unequal
bca718
+*/
bca718
+ENTRY(__wcsncmp_vx)
bca718
+	.machine "z13"
bca718
+	.machinemode "zarch_nohighgprs"
bca718
+
bca718
+# if !defined __s390x__
bca718
+	llgfr	%r4,%r4
bca718
+# endif /* !defined __s390x__ */
bca718
+
bca718
+	clgije	%r4,0,.Lend_equal /* Nothing to do if n == 0.  */
bca718
+
bca718
+	/* Check range of n and convert to byte-count.  */
bca718
+# ifdef __s390x__
bca718
+	tmhh	%r4,49152	/* Test bit 0 or 1 of maxlen.  */
bca718
+	lghi	%r1,-4		/* Max byte-count is 18446744073709551612.  */
bca718
+# else
bca718
+	tmlh	%r4,49152	/* Test bit 0 or 1 of maxlen.  */
bca718
+	llilf	%r1,4294967292	/* Max byte-count is 4294967292.  */
bca718
+# endif /* !__s390x__ */
bca718
+	sllg	%r4,%r4,2	/* Convert character-count to byte-count.  */
bca718
+	locgrne	%r4,%r1		/* Use max byte-count, if bit 0/1 was one.  */
bca718
+
bca718
+	/* Check first character without vector load.  */
bca718
+	lghi	%r5,4		/* current_len = 4 bytes.  */
bca718
+	/* Check s1/2[0].  */
bca718
+	lt	%r0,0(%r2)
bca718
+	l	%r1,0(%r3)
bca718
+	je	.Lend_cmp_one_char
bca718
+	crjne	%r0,%r1,.Lend_cmp_one_char
bca718
+
bca718
+.Lloop:
bca718
+	vlbb	%v17,0(%r5,%r3),6 /* Load s2 to block boundary.  */
bca718
+	vlbb	%v16,0(%r5,%r2),6 /* Load s1 to block boundary.  */
bca718
+	lcbb	%r0,0(%r5,%r2),6 /* Get loaded byte count of s1.  */
bca718
+	jo	.Llt16_1	/* Jump away if vector not fully loaded.  */
bca718
+	lcbb	%r1,0(%r5,%r3),6 /* Get loaded byte count of s2.  */
bca718
+	jo	.Llt16_2	/* Jump away if vector not fully loaded.  */
bca718
+	aghi	%r5,16		/* Both vectors are fully loaded.  */
bca718
+	vfenezfs %v18,%v16,%v17	/* Compare not equal with zero search.  */
bca718
+	clgrjhe	%r5,%r4,.Llastcmp /* If current_len >= n ->last compare.  */
bca718
+	jno	.Lfound
bca718
+
bca718
+	vlbb	%v17,0(%r5,%r3),6
bca718
+	vlbb	%v16,0(%r5,%r2),6
bca718
+	lcbb	%r0,0(%r5,%r2),6
bca718
+	jo	.Llt16_1
bca718
+	lcbb	%r1,0(%r5,%r3),6
bca718
+	jo	.Llt16_2
bca718
+	aghi	%r5,16
bca718
+	vfenezfs %v18,%v16,%v17
bca718
+	clgrjhe	%r5,%r4,.Llastcmp
bca718
+	jno	.Lfound
bca718
+
bca718
+	vlbb	%v17,0(%r5,%r3),6
bca718
+	vlbb	%v16,0(%r5,%r2),6
bca718
+	lcbb	%r0,0(%r5,%r2),6
bca718
+	jo	.Llt16_1
bca718
+	lcbb	%r1,0(%r5,%r3),6
bca718
+	jo	.Llt16_2
bca718
+	aghi	%r5,16
bca718
+	vfenezfs %v18,%v16,%v17
bca718
+	clgrjhe	%r5,%r4,.Llastcmp
bca718
+	jno	.Lfound
bca718
+
bca718
+	vlbb	%v17,0(%r5,%r3),6
bca718
+	vlbb	%v16,0(%r5,%r2),6
bca718
+	lcbb	%r0,0(%r5,%r2),6
bca718
+	jo	.Llt16_1
bca718
+	lcbb	%r1,0(%r5,%r3),6
bca718
+	jo	.Llt16_2
bca718
+	aghi	%r5,16
bca718
+	vfenezfs %v18,%v16,%v17
bca718
+	clgrjhe	%r5,%r4,.Llastcmp
bca718
+	jno	.Lfound
bca718
+
bca718
+	j	.Lloop
bca718
+
bca718
+.Llt16_1:
bca718
+	lcbb	%r1,0(%r5,%r3),6 /* Get loaded byte count of s2.  */
bca718
+.Llt16_2:
bca718
+	clr	%r0,%r1		/* Compare logical.  */
bca718
+	locrh	%r0,%r1		/* Compute minimum of bytes loaded.  */
bca718
+	nill	%r0,65532	/* Align bytes loaded to full characters.  */
bca718
+	jz	.Lcmp_one_char	/* Jump away if no full char is available.  */
bca718
+.Llt_cmp:
bca718
+	algfr	%r5,%r0		/* Add smallest loaded bytes to current_len.  */
bca718
+	vfenezfs %v18,%v16,%v17	/* Compare not equal with zero search.  */
bca718
+	clgrj	%r5,%r4,10,.Llastcmp /* If current_len >= n -> last compare  */
bca718
+	vlgvb	%r1,%v18,7	/* Get not equal index or 16 if all equal.  */
bca718
+	clrjl	%r1,%r0,.Lfound /* Jump away if miscompare is within
bca718
+				   loaded bytes; (index < loaded-bytes) */
bca718
+	j	.Lloop
bca718
+
bca718
+.Lcmp_one_char:
bca718
+	/* At least one of both strings is not 4-byte aligned
bca718
+	   and there is no full character before next block-boundary.
bca718
+	   Compare one character to get over the boundary and
bca718
+	   proceed with normal loop!  */
bca718
+	vlef	%v16,0(%r5,%r2),0 /* Load one character.  */
bca718
+	lghi	%r0,4		/* Loaded byte count is 4.  */
bca718
+	vlef	%v17,0(%r5,%r3),0
bca718
+	j	.Llt_cmp	/* Proceed with comparision.  */
bca718
+
bca718
+.Llastcmp:
bca718
+	/* Use comparision result only if located within first n characters.
bca718
+	   %r0: loaded byte count in vreg;
bca718
+	   %r5: current_len;
bca718
+	   %r4: n;
bca718
+	   (current_len - n): [0...16[
bca718
+	   First ignored match index: loaded bytes - (current_len-n): ]0...16]
bca718
+	*/
bca718
+	slgr	%r5,%r4		/* %r5 = current_len - n.  */
bca718
+	slr	%r0,%r5		/* %r0 = first ignored match index.  */
bca718
+	vlgvb	%r4,%v18,7	/* Get not equal index or 16 if all equal.  */
bca718
+	clrjl	%r4,%r0,.Lfound2 /* Jump away if miscompare is within
bca718
+				     loaded bytes and below n bytes.  */
bca718
+.Lend_equal:
bca718
+	lghi	%r2,0
bca718
+	br	%r14
bca718
+
bca718
+.Lfound:
bca718
+	/* Difference or end of string.  */
bca718
+	/* vfenezf found an unequal element or zero.
bca718
+	   This instruction compares unsigned words, but wchar_t is signed.
bca718
+	   Thus we have to compare the found element again.  */
bca718
+	vlgvb	%r4,%v18,7	/* Extract not equal byte-index.  */
bca718
+.Lfound2:
bca718
+	srl	%r4,2		/* And convert it to character-index.  */
bca718
+	vlgvf	%r0,%v16,0(%r4)	/* Load character-values.  */
bca718
+	vlgvf	%r1,%v17,0(%r4)
bca718
+.Lend_cmp_one_char:
bca718
+	cr	%r0,%r1
bca718
+	je	.Lend_equal
bca718
+	lghi	%r2,1
bca718
+	lghi	%r1,-1
bca718
+	locgrl	%r2,%r1
bca718
+	br	%r14
bca718
+END(__wcsncmp_vx)
bca718
+#endif /* HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc */
bca718
diff --git a/sysdeps/s390/multiarch/wcsncmp.c b/sysdeps/s390/multiarch/wcsncmp.c
bca718
new file mode 100644
bca718
index 0000000..0d79661
bca718
--- /dev/null
bca718
+++ b/sysdeps/s390/multiarch/wcsncmp.c
bca718
@@ -0,0 +1,27 @@
bca718
+/* Multiple versions of wcsncmp.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#if defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc
bca718
+# include <wchar.h>
bca718
+# include <ifunc-resolve.h>
bca718
+
bca718
+s390_vx_libc_ifunc2 (__wcsncmp, wcsncmp)
bca718
+
bca718
+#else
bca718
+# include <wcsmbs/wcsncmp.c>
bca718
+#endif /* !(defined HAVE_S390_VX_ASM_SUPPORT && !defined NOT_IN_libc) */
bca718
diff --git a/wcsmbs/Makefile b/wcsmbs/Makefile
bca718
index 44b1502..611b2c9 100644
bca718
--- a/wcsmbs/Makefile
bca718
+++ b/wcsmbs/Makefile
bca718
@@ -41,7 +41,7 @@ routines := wcscat wcschr wcscmp wcscpy wcscspn wcsdup wcslen wcsncat \
bca718
 	    isoc99_swscanf isoc99_vswscanf \
bca718
 	    mbrtoc16 c16rtomb
bca718
 
bca718
-strop-tests :=  wcscmp wmemcmp wcslen wcschr wcsrchr wcscpy wcsnlen \
bca718
+strop-tests :=  wcscmp wcsncmp wmemcmp wcslen wcschr wcsrchr wcscpy wcsnlen \
bca718
 		wcpcpy wcsncpy wcpncpy wcscat wcsncat
bca718
 tests := tst-wcstof wcsmbs-tst1 tst-wcsnlen tst-btowc tst-mbrtowc \
bca718
 	 tst-wcrtomb tst-wcpncpy tst-mbsrtowcs tst-wchar-h tst-mbrtowc2 \
bca718
diff --git a/wcsmbs/test-wcsncmp-ifunc.c b/wcsmbs/test-wcsncmp-ifunc.c
bca718
new file mode 100644
bca718
index 0000000..35176f0
bca718
--- /dev/null
bca718
+++ b/wcsmbs/test-wcsncmp-ifunc.c
bca718
@@ -0,0 +1,20 @@
bca718
+/* Test and measure IFUNC implementations of wcsncmp function.
bca718
+   Copyright (C) 2015 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#define TEST_IFUNC 1
bca718
+#include "test-wcsncmp.c"
bca718
diff --git a/wcsmbs/test-wcsncmp.c b/wcsmbs/test-wcsncmp.c
bca718
new file mode 100644
bca718
index 0000000..07757d8
bca718
--- /dev/null
bca718
+++ b/wcsmbs/test-wcsncmp.c
bca718
@@ -0,0 +1,2 @@
bca718
+#define WIDE 1
bca718
+#include "../string/test-strncmp.c"
bca718
diff --git a/wcsmbs/wcsncmp.c b/wcsmbs/wcsncmp.c
bca718
index 7f1704f..1522b6f 100644
bca718
--- a/wcsmbs/wcsncmp.c
bca718
+++ b/wcsmbs/wcsncmp.c
bca718
@@ -18,53 +18,56 @@
bca718
 
bca718
 #include <wchar.h>
bca718
 
bca718
+#ifndef WCSNCMP
bca718
+# define WCSNCMP wcsncmp
bca718
+#endif
bca718
 
bca718
 /* Compare no more than N characters of S1 and S2,
bca718
    returning less than, equal to or greater than zero
bca718
    if S1 is lexicographically less than, equal to or
bca718
    greater than S2.  */
bca718
 int
bca718
-wcsncmp (s1, s2, n)
bca718
+WCSNCMP (s1, s2, n)
bca718
      const wchar_t *s1;
bca718
      const wchar_t *s2;
bca718
      size_t n;
bca718
 {
bca718
-  wint_t c1 = L'\0';
bca718
-  wint_t c2 = L'\0';
bca718
+  wchar_t c1 = L'\0';
bca718
+  wchar_t c2 = L'\0';
bca718
 
bca718
   if (n >= 4)
bca718
     {
bca718
       size_t n4 = n >> 2;
bca718
       do
bca718
 	{
bca718
-	  c1 = (wint_t) *s1++;
bca718
-	  c2 = (wint_t) *s2++;
bca718
+	  c1 = *s1++;
bca718
+	  c2 = *s2++;
bca718
 	  if (c1 == L'\0' || c1 != c2)
bca718
-	    return c1 - c2;
bca718
-	  c1 = (wint_t) *s1++;
bca718
-	  c2 = (wint_t) *s2++;
bca718
+	    return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
+	  c1 = *s1++;
bca718
+	  c2 = *s2++;
bca718
 	  if (c1 == L'\0' || c1 != c2)
bca718
-	    return c1 - c2;
bca718
-	  c1 = (wint_t) *s1++;
bca718
-	  c2 = (wint_t) *s2++;
bca718
+	    return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
+	  c1 = *s1++;
bca718
+	  c2 = *s2++;
bca718
 	  if (c1 == L'\0' || c1 != c2)
bca718
-	    return c1 - c2;
bca718
-	  c1 = (wint_t) *s1++;
bca718
-	  c2 = (wint_t) *s2++;
bca718
+	    return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
+	  c1 = *s1++;
bca718
+	  c2 = *s2++;
bca718
 	  if (c1 == L'\0' || c1 != c2)
bca718
-	    return c1 - c2;
bca718
+	    return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
 	} while (--n4 > 0);
bca718
       n &= 3;
bca718
     }
bca718
 
bca718
   while (n > 0)
bca718
     {
bca718
-      c1 = (wint_t) *s1++;
bca718
-      c2 = (wint_t) *s2++;
bca718
+      c1 = *s1++;
bca718
+      c2 = *s2++;
bca718
       if (c1 == L'\0' || c1 != c2)
bca718
-	return c1 - c2;
bca718
+	return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
bca718
       n--;
bca718
     }
bca718
 
bca718
-  return c1 - c2;
bca718
+  return 0;
bca718
 }
bca718
-- 
bca718
2.3.0
bca718