S390: Optimize strncmp and wcsncmp.

This patch provides optimized versions of strncmp and wcsncmp with the z13
vector instructions.

ChangeLog:

	* sysdeps/s390/multiarch/strncmp-c.c: New File.
	* sysdeps/s390/multiarch/strncmp-vx.S: Likewise.
	* sysdeps/s390/multiarch/strncmp.c: Likewise.
	* sysdeps/s390/multiarch/wcsncmp-c.c: Likewise.
	* sysdeps/s390/multiarch/wcsncmp-vx.S: Likewise.
	* sysdeps/s390/multiarch/wcsncmp.c: Likewise.
	* sysdeps/s390/multiarch/Makefile (sysdep_routines): Add strncmp and
	wcsncmp functions.
	* sysdeps/s390/multiarch/ifunc-impl-list.c
	(__libc_ifunc_impl_list): Add ifunc test for strncmp, wcsncmp.
	* wcsmbs/wcsncmp.c (WCSNCMP): Define and use macro.
	* benchtests/bench-strncmp.c: Add wcsncmp support.
	* benchtests/bench-wcsncmp.c: New File.
	* benchtests/Makefile (wcsmbs-bench): Add wcsncmp.

1 files changed, 85 insertions, 25 deletions

diff --git a/benchtests/bench-strncmp.c b/benchtests/bench-strncmp.c
index 1d06e5bb6e..3b264145a8 100644
--- a/benchtests/bench-strncmp.c
+++ b/benchtests/bench-strncmp.c
@@ -17,17 +17,68 @@
    <http://www.gnu.org/licenses/>.  */
 
 #define TEST_MAIN
-#define TEST_NAME "strncmp"
+#ifdef WIDE
+# define TEST_NAME "wcsncmp"
+#else
+# define TEST_NAME "strncmp"
+#endif /* !WIDE */
 #include "bench-string.h"
 
-typedef int (*proto_t) (const char *, const char *, size_t);
-int simple_strncmp (const char *, const char *, size_t);
-int stupid_strncmp (const char *, const char *, size_t);
+#ifdef WIDE
+# include <wchar.h>
 
-IMPL (stupid_strncmp, 0)
-IMPL (simple_strncmp, 0)
-IMPL (strncmp, 1)
+# define L(str) L##str
+# define STRNCMP wcsncmp
+# define SIMPLE_STRNCMP simple_wcsncmp
+# define STUPID_STRNCMP stupid_wcsncmp
+# define CHAR wchar_t
+# define CHARBYTES 4
 
+/* Wcsncmp uses signed semantics for comparison, not unsigned.
+   Avoid using substraction since possible overflow.  */
+int
+simple_wcsncmp (const CHAR *s1, const CHAR *s2, size_t n)
+{
+  wchar_t c1, c2;
+
+  while (n--)
+    {
+      c1 = *s1++;
+      c2 = *s2++;
+      if (c1 == L ('\0') || c1 != c2)
+	return c1 > c2 ? 1 : (c1 < c2 ? -1 : 0);
+    }
+  return 0;
+}
+
+int
+stupid_wcsncmp (const CHAR *s1, const CHAR *s2, size_t n)
+{
+  wchar_t c1, c2;
+  size_t ns1 = wcsnlen (s1, n) + 1, ns2 = wcsnlen (s2, n) + 1;
+
+  n = ns1 < n ? ns1 : n;
+  n = ns2 < n ? ns2 : n;
+
+  while (n--)
+    {
+      c1 = *s1++;
+      c2 = *s2++;
+      if (c1 != c2)
+	return c1 > c2 ? 1 : -1;
+    }
+  return 0;
+}
+
+#else
+# define L(str) str
+# define STRNCMP strncmp
+# define SIMPLE_STRNCMP simple_strncmp
+# define STUPID_STRNCMP stupid_strncmp
+# define CHAR char
+# define CHARBYTES 1
+
+/* Strncmp uses unsigned semantics for comparison.  */
 int
 simple_strncmp (const char *s1, const char *s2, size_t n)
 {
@@ -46,12 +97,21 @@ stupid_strncmp (const char *s1, const char *s2, size_t n)
 
   n = ns1 < n ? ns1 : n;
   n = ns2 < n ? ns2 : n;
-  while (n-- && (ret = *(unsigned char *) s1++ - * (unsigned char *) s2++) == 0);
+  while (n-- && (ret = *(unsigned char *) s1++ - *(unsigned char *) s2++) == 0);
   return ret;
 }
 
+#endif /* !WIDE */
+
+typedef int (*proto_t) (const CHAR *, const CHAR *, size_t);
+
+IMPL (STUPID_STRNCMP, 0)
+IMPL (SIMPLE_STRNCMP, 0)
+IMPL (STRNCMP, 1)
+
+
 static void
-do_one_test (impl_t *impl, const char *s1, const char *s2, size_t n,
+do_one_test (impl_t *impl, const CHAR *s1, const CHAR *s2, size_t n,
 	     int exp_result)
 {
   size_t i, iters = INNER_LOOP_ITERS;
@@ -74,12 +134,12 @@ do_test_limit (size_t align1, size_t align2, size_t len, size_t n, int max_char,
 	 int exp_result)
 {
   size_t i, align_n;
-  char *s1, *s2;
+  CHAR *s1, *s2;
 
   if (n == 0)
     {
-      s1 = (char*)(buf1 + page_size);
-      s2 = (char*)(buf2 + page_size);
+      s1 = (CHAR *) (buf1 + page_size);
+      s2 = (CHAR *) (buf2 + page_size);
       printf ("Length %4zd/%4zd:", len, n);
 
       FOR_EACH_IMPL (impl, 0)
@@ -92,16 +152,16 @@ do_test_limit (size_t align1, size_t align2, size_t len, size_t n, int max_char,
 
   align1 &= 15;
   align2 &= 15;
-  align_n = (page_size - n) & 15;
+  align_n = (page_size - n * CHARBYTES) & 15;
 
-  s1 = (char*)(buf1 + page_size - n);
-  s2 = (char*)(buf2 + page_size - n);
+  s1 = (CHAR *) (buf1 + page_size - n * CHARBYTES);
+  s2 = (CHAR *) (buf2 + page_size - n * CHARBYTES);
 
   if (align1 < align_n)
-    s1 -= (align_n - align1);
+    s1 = (CHAR *) ((char *) s1 - (align_n - align1));
 
   if (align2 < align_n)
-    s2 -= (align_n - align2);
+    s2 = (CHAR *) ((char *) s2 - (align_n - align2));
 
   for (i = 0; i < n; i++)
     s1[i] = s2[i] = 1 + 23 * i % max_char;
@@ -129,24 +189,24 @@ do_test (size_t align1, size_t align2, size_t len, size_t n, int max_char,
 	 int exp_result)
 {
   size_t i;
-  char *s1, *s2;
+  CHAR *s1, *s2;
 
   if (n == 0)
     return;
 
-  align1 &= 7;
-  if (align1 + n + 1 >= page_size)
+  align1 &= 63;
+  if (align1 + (n + 1) * CHARBYTES >= page_size)
     return;
 
   align2 &= 7;
-  if (align2 + n + 1 >= page_size)
+  if (align2 + (n + 1) * CHARBYTES >= page_size)
     return;
 
-  s1 = (char*)(buf1 + align1);
-  s2 = (char*)(buf2 + align2);
+  s1 = (CHAR *) (buf1 + align1);
+  s2 = (CHAR *) (buf2 + align2);
 
   for (i = 0; i < n; i++)
-    s1[i] = s2[i] = 1 + 23 * i % max_char;
+    s1[i] = s2[i] = 1 + (23 << ((CHARBYTES - 1) * 8)) * i % max_char;
 
   s1[n] = 24 + exp_result;
   s2[n] = 23;
@@ -162,7 +222,7 @@ do_test (size_t align1, size_t align2, size_t len, size_t n, int max_char,
   printf ("Length %4zd/%4zd, alignment %2zd/%2zd:", len, n, align1, align2);
 
   FOR_EACH_IMPL (impl, 0)
-    do_one_test (impl, (char*)s1, (char*)s2, n, exp_result);
+    do_one_test (impl, s1, s2, n, exp_result);
 
   putchar ('\n');
 }