wcsmbs: optimize wcscat

This patch rewrites wcscat using wcslen and wcscpy.  This is similar to
the optimization done on strcat by 6e46de42fe.

The strcpy changes are mainly to add the internal alias to avoid PLT
calls.

Checked on x86_64-linux-gnu and a build against the affected
architectures.

	* include/wchar.h (__wcscpy): New prototype.
	* sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c
	(__wcscpy): Route internal symbol to generic implementation.
	* sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy.c (wcscpy):
	Add internal __wcscpy alias.
	* sysdeps/powerpc/powerpc64/multiarch/wcscpy.c (wcscpy): Likewise.
	* sysdeps/s390/wcscpy.c (wcscpy): Likewise.
	* sysdeps/x86_64/multiarch/wcscpy.c (wcscpy): Likewise.
	* wcsmbs/wcscpy.c (wcscpy): Add
	* sysdeps/x86_64/multiarch/wcscpy-c.c (WCSCPY): Adjust macro to
	use generic implementation.
	* wcsmbs/wcscat.c (wcscat): Rewrite using wcslen and wcscpy.

10 files changed, 67 insertions, 58 deletions

diff --git a/ChangeLog b/ChangeLog
index 5ff8f8a..b963736 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,18 @@
 2019-02-27  Adhemerval Zanella  <adhemerval.zanella@linaro.org>
 
+	* include/wchar.h (__wcscpy): New prototype.
+	* sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c
+	(__wcscpy): Route internal symbol to generic implementation.
+	* sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy.c (wcscpy):
+	Add internal __wcscpy alias.
+	* sysdeps/powerpc/powerpc64/multiarch/wcscpy.c (wcscpy): Likewise.
+	* sysdeps/s390/wcscpy.c (wcscpy): Likewise.
+	* sysdeps/x86_64/multiarch/wcscpy.c (wcscpy): Likewise.
+	* wcsmbs/wcscpy.c (wcscpy): Add
+	* sysdeps/x86_64/multiarch/wcscpy-c.c (WCSCPY): Adjust macro to
+	use generic implementation.
+	* wcsmbs/wcscat.c (wcscat): Rewrite using wcslen and wcscpy.
+
 	* wcsmbs/wcpncpy.c (__wcpcpy): Rewrite using wcslen, wmemcpy, and
 	wmemset.
 
diff --git a/include/wchar.h b/include/wchar.h
index 614073b..2cb4495 100644
--- a/include/wchar.h
+++ b/include/wchar.h
@@ -182,6 +182,10 @@ extern size_t __wcsnrtombs (char *__restrict __dst,
 			    size_t __nwc, size_t __len,
 			    __mbstate_t *__restrict __ps)
      attribute_hidden;
+extern wchar_t *__wcscpy (wchar_t *__restrict __dest,
+			  const wchar_t *__restrict __src)
+			  attribute_hidden __nonnull ((1, 2));
+libc_hidden_proto (__wcscpy)
 extern wchar_t *__wcsncpy (wchar_t *__restrict __dest,
 			   const wchar_t *__restrict __src, size_t __n);
 extern wchar_t *__wcpcpy (wchar_t *__dest, const wchar_t *__src);
diff --git a/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c b/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c
index 52b692b..31e0d81 100644
--- a/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c
+++ b/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy-ppc32.c
@@ -17,10 +17,11 @@
 
 #include <wchar.h>
 
-#if IS_IN (libc)
-# define WCSCPY  __wcscpy_ppc
-#endif
-
 extern __typeof (wcscpy) __wcscpy_ppc;
 
+#define WCSCPY  __wcscpy_ppc
 #include <wcsmbs/wcscpy.c>
+
+#ifdef SHARED
+__hidden_ver1 (__wcscpy_ppc, __GI___wcscpy, __wcscpy_ppc);
+#endif
diff --git a/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy.c b/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy.c
index ecca37d..e879846 100644
--- a/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy.c
+++ b/sysdeps/powerpc/powerpc32/power4/multiarch/wcscpy.c
@@ -17,20 +17,20 @@
    <http://www.gnu.org/licenses/>.  */
 
 #if IS_IN (libc)
+# define wcscpy __redirect_wcscpy
 # include <wchar.h>
-# include <shlib-compat.h>
+# undef wcscpy
 # include "init-arch.h"
 
-extern __typeof (wcscpy) __wcscpy_ppc attribute_hidden;
-extern __typeof (wcscpy) __wcscpy_power6 attribute_hidden;
-extern __typeof (wcscpy) __wcscpy_power7 attribute_hidden;
+extern __typeof (__redirect_wcscpy) __wcscpy_ppc attribute_hidden;
+extern __typeof (__redirect_wcscpy) __wcscpy_power6 attribute_hidden;
+extern __typeof (__redirect_wcscpy) __wcscpy_power7 attribute_hidden;
 
-libc_ifunc (wcscpy,
-	     (hwcap & PPC_FEATURE_HAS_VSX)
-             ? __wcscpy_power7 :
-	       (hwcap & PPC_FEATURE_ARCH_2_05)
-	       ? __wcscpy_power6
-             : __wcscpy_ppc);
-#else
-#include <wcsmbs/wcscpy.c>
+libc_ifunc_redirected (__redirect_wcscpy, wcscpy,
+		       (hwcap & PPC_FEATURE_HAS_VSX)
+		       ? __wcscpy_power7 :
+			 (hwcap & PPC_FEATURE_ARCH_2_05)
+			 ? __wcscpy_power6
+		       : __wcscpy_ppc);
+weak_alias (wcscpy, __wcscpy)
 #endif
diff --git a/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c b/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c
index 3cea9a4..3f918b2 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/wcscpy.c
@@ -16,21 +16,20 @@
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
-#if IS_IN (libc)
-# include <wchar.h>
-# include <shlib-compat.h>
-# include "init-arch.h"
+#define __wcscpy __redirect___wcscpy
+#include <wchar.h>
+#undef __wcscpy
+#include <shlib-compat.h>
+#include "init-arch.h"
 
 extern __typeof (wcscpy) __wcscpy_ppc attribute_hidden;
 extern __typeof (wcscpy) __wcscpy_power6 attribute_hidden;
 extern __typeof (wcscpy) __wcscpy_power7 attribute_hidden;
 
-libc_ifunc (wcscpy,
-	     (hwcap & PPC_FEATURE_HAS_VSX)
-             ? __wcscpy_power7 :
-	       (hwcap & PPC_FEATURE_ARCH_2_05)
-	       ? __wcscpy_power6
-             : __wcscpy_ppc);
-#else
-#include <wcsmbs/wcscpy.c>
-#endif
+libc_ifunc_redirected (__redirect___wcscpy, __wcscpy,
+		       (hwcap & PPC_FEATURE_HAS_VSX)
+		       ? __wcscpy_power7 :
+		         (hwcap & PPC_FEATURE_ARCH_2_05)
+		         ? __wcscpy_power6
+	               : __wcscpy_ppc);
+weak_alias (__wcscpy, wcscpy)
diff --git a/sysdeps/s390/wcscpy.c b/sysdeps/s390/wcscpy.c
index 2e8ef50..a569f91 100644
--- a/sysdeps/s390/wcscpy.c
+++ b/sysdeps/s390/wcscpy.c
@@ -30,9 +30,11 @@ extern __typeof (wcscpy) WCSCPY_C attribute_hidden;
 extern __typeof (wcscpy) WCSCPY_Z13 attribute_hidden;
 # endif
 
-s390_libc_ifunc_expr (wcscpy, wcscpy,
+s390_libc_ifunc_expr (wcscpy, __wcscpy,
 		      (HAVE_WCSCPY_Z13 && (hwcap & HWCAP_S390_VX))
 		      ? WCSCPY_Z13
 		      : WCSCPY_DEFAULT
 		      )
+weak_alias (__wcscpy, wcscpy)
+libc_hidden_def (__wcscpy)
 #endif
diff --git a/sysdeps/x86_64/multiarch/wcscpy-c.c b/sysdeps/x86_64/multiarch/wcscpy-c.c
index a51a83a..26d6984 100644
--- a/sysdeps/x86_64/multiarch/wcscpy-c.c
+++ b/sysdeps/x86_64/multiarch/wcscpy-c.c
@@ -1,5 +1,5 @@
 #if IS_IN (libc)
-# define wcscpy  __wcscpy_sse2
+# define WCSCPY  __wcscpy_sse2
 #endif
 
-#include "wcsmbs/wcscpy.c"
+#include <wcsmbs/wcscpy.c>
diff --git a/sysdeps/x86_64/multiarch/wcscpy.c b/sysdeps/x86_64/multiarch/wcscpy.c
index 101a585..96151b4 100644
--- a/sysdeps/x86_64/multiarch/wcscpy.c
+++ b/sysdeps/x86_64/multiarch/wcscpy.c
@@ -19,9 +19,9 @@
 
 /* Define multiple versions only for the definition in libc.  */
 #if IS_IN (libc)
-# define wcscpy __redirect_wcscpy
+# define __wcscpy __redirect_wcscpy
 # include <wchar.h>
-# undef wcscpy
+# undef __wcscpy
 
 # define SYMBOL_NAME wcscpy
 # include <init-arch.h>
@@ -40,5 +40,10 @@ IFUNC_SELECTOR (void)
   return OPTIMIZE (sse2);
 }
 
-libc_ifunc_redirected (__redirect_wcscpy, wcscpy, IFUNC_SELECTOR ());
+libc_ifunc_redirected (__redirect_wcscpy, __wcscpy, IFUNC_SELECTOR ());
+weak_alias (__wcscpy, wcscpy)
+# ifdef SHARED
+__hidden_ver1 (__wcscpy, __GI___wcscpy, __redirect_wcscpy)
+  __attribute__((visibility ("hidden"))) __attribute_copy__ (wcscpy);
+# endif
 #endif
diff --git a/wcsmbs/wcscat.c b/wcsmbs/wcscat.c
index 6a25b20..1a9d667 100644
--- a/wcsmbs/wcscat.c
+++ b/wcsmbs/wcscat.c
@@ -26,26 +26,7 @@
 wchar_t *
 __wcscat (wchar_t *dest, const wchar_t *src)
 {
-  wchar_t *s1 = dest;
-  const wchar_t *s2 = src;
-  wchar_t c;
-
-  /* Find the end of the string.  */
-  do
-    c = *s1++;
-  while (c != L'\0');
-
-  /* Make S1 point before the next character, so we can increment
-     it while memory is read (wins on pipelined cpus).	*/
-  s1 -= 2;
-
-  do
-    {
-      c = *s2++;
-      *++s1 = c;
-    }
-  while (c != L'\0');
-
+  __wcscpy (dest + __wcslen (dest), src);
   return dest;
 }
 #ifndef WCSCAT
diff --git a/wcsmbs/wcscpy.c b/wcsmbs/wcscpy.c
index 7a34c77..636bf6b 100644
--- a/wcsmbs/wcscpy.c
+++ b/wcsmbs/wcscpy.c
@@ -20,13 +20,13 @@
 #include <wchar.h>
 
 
-#ifndef WCSCPY
-# define WCSCPY wcscpy
+#ifdef WCSCPY
+# define __wcscpy WCSCPY
 #endif
 
 /* Copy SRC to DEST.  */
 wchar_t *
-WCSCPY (wchar_t *dest, const wchar_t *src)
+__wcscpy (wchar_t *dest, const wchar_t *src)
 {
   wint_t c;
   wchar_t *wcp;
@@ -58,3 +58,7 @@ WCSCPY (wchar_t *dest, const wchar_t *src)
 
   return dest;
 }
+#ifndef WCSCPY
+weak_alias (__wcscpy, wcscpy)
+libc_hidden_def (__wcscpy)
+#endif