Merge commit 'origin/master' into fedora/master

45 files changed, 4437 insertions, 358 deletions

diff --git a/ChangeLog b/ChangeLog
index 6f7026debf..a2fbd61fe6 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,111 @@
+2009-07-26  Ulrich Drepper  <drepper@redhat.com>
+
+	* sysdeps/x86_64/tst-xmmymm.sh: New file.  Check whether any of the
+	functions used in ld.so modify xmm/ymm registers.
+	* sysdeps/x86_64/Makefile:  Hook new test up.
+	* sysdeps/x86_64/rtld-memchr.c: New file.
+	* sysdeps/x86_64/rtld-memcmp.c: New file.
+	* sysdeps/x86_64/rtld-rawmemchr.c: New file.
+	* sysdeps/x86_64/rtld-strchr.S: New file.
+	* sysdeps/x86_64/rtld-strcmp.S: New file.
+	* sysdeps/x86_64/rtld-strlen.S: New file.
+	* sysdeps/x86_64/multiarch/rtld-rawmemchr.c: New file.
+	* sysdeps/x86_64/multiarch/rtld-strlen.S: New file.
+
+2009-07-26  H.J. Lu  <hongjiu.lu@intel.com>
+
+	* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Remove
+	strncmp-c.
+	* sysdeps/x86_64/multiarch/strcmp.S (aftertail): Removed.
+	(exit): Likewise.
+	(Byte1): Likewise.
+	(Byte2): Likewise.
+	(Byte3): Likewise.
+	(Byte4): Likewise.
+	(Byte5): Likewise.
+	(Byte6): Likewise.
+	(next_8_bytes): Likewise.
+	(Byte0): Remove commented out codes.
+	(unaligned_table): Align jump table at 8 bytes.
+	Add _sse4_2 to all labels.  Always include "../strcmp.S".
+	* sysdeps/x86_64/multiarch/strncmp-c.c: Removed.
+	* sysdeps/x86_64/strcmp.S: Add SSE2 support.
+	* sysdeps/x86_64/strncmp.S: New file.
+
+2009-07-26  Ulrich Drepper  <drepper@redhat.com>
+
+	[BZ #10422]
+	* sysdeps/unix/sysv/linux/eventfd.c: Add compatibility for old
+	kernels, dropped when eventfd2 support was added.
+	* sysdeps/unix/sysv/linux/signalfd.c: Add compatibility for old
+	kernels, dropped when signalfd4 support was added.
+	* sysdeps/unix/sysv/linux/kernel-features.h: More CLOEXEC syscalls
+	added, name them.
+
+	[BZ #10452]
+	* resolv/res_send.c (send_dg): Pass full SERVFAIL, NOTIMP, REFUSED
+	replies up.
+
+	* elf/elf.h: Define NT_GNU_GOLD_VERSION.
+
+2009-07-25  Ulrich Drepper  <drepper@redhat.com>
+
+	* sysdeps/x86_64/multiarch/strcmp.S: Some more optimizations for
+	modern processor versions.  Patch by H.J. Lu <hongjiu.lu@intel.com>.
+
+	[BZ #10448]
+	* sysdeps/posix/getaddrinfo.c (gaih_inet): If NSS module contains no
+	callback we must touch the status to avoid using stale value.
+
+	* sysdeps/x86_64/multiarch/strcmp.S: Exclude unused code from being
+	compiled in.
+
+2009-07-24  Ulrich Drepper  <drepper@redhat.com>
+
+	* sysdeps/unix/sysv/linux/configure.in: Don't automatically include
+	/lib/modules/* headers anymore.  We have sane headers in the standard
+	place now.
+
+2009-06-16  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+
+	* sysdeps/s390/dl-procinfo.c (_dl_s390_cap_flags): "hpage",
+	"etf3enh" and "highgprs" added.
+	(_dl_s390_platforms): "z10" added.
+	* sysdeps/s390/dl-procinfo.h (_DL_HWCAP_COUNT, _DL_PLATFORMS_COUNT):
+	Increased for the new entries.
+	(HWCAP enum): HWCAP_S390_HPAGE, HWCAP_S390_ETF3EH and
+	HWCAP_S390_HIGH_GPRS added.
+
+	* sysdeps/s390/s390-64/Makefile: Adjusted to build the new modules.
+	* sysdeps/s390/s390-64/iso-8859-1_cp037_z900.c: New file.
+	* sysdeps/s390/s390-64/utf16-utf32-z9.c: New file.
+	* sysdeps/s390/s390-64/utf8-utf16-z9.c: New file.
+	* sysdeps/s390/s390-64/utf8-utf32-z9.c: New file.
+
 2009-07-23  Ulrich Drepper  <drepper@redhat.com>
 
+	* sysdeps/x86_64/cacheinfo.c [USE_MULTIARCH]: Rearrange code to
+	avoid additional cpuid instructions.  Most of the information is
+	stored somewhere.
+
+	* sysdeps/unix/sysv/linux/i386/sysconf.c (intel_02_known): Add more
+	cache descriptors.
+	* sysdeps/x86_64/cacheinfo.c (intel_02_known): Likewise.
+
+	* sysdeps/x86_64/multiarch/init-arch.c (__init_cpu_features): Reset
+	SSSE3 bit for Atoms.
+	* sysdeps/x86_64/multiarch/strcpy.S: New need to perform Atom test
+	here anymore.
+
+	* posix/tst-rfc3484.c (do_test): Initialize entire sockaddr_in
+	structure before copying it to avoid warning.
+	* posix/tst-rfc3484-2.c (do_test): Likewise.
+	* posix/tst-rfc3484-3.c (do_test): Likewise.
+
+	[BZ #10416]
+	* include/unistd.h: Make header file suitable for C++ test cases.
+	Patch by Duncan Simpson <dps@simpson.demon.co.uk>.
+
 	* sysdeps/unix/sysv/linux/i386/makecontext.S: Ensure we preserve the
 	stack alignment in the exit code.
 
diff --git a/elf/elf.h b/elf/elf.h
index 7efdedefb4..ce6de07e91 100644
--- a/elf/elf.h
+++ b/elf/elf.h
@@ -1054,6 +1054,9 @@ typedef struct
    The descriptor consists of any nonzero number of bytes.  */
 #define NT_GNU_BUILD_ID	3
 
+/* Version note generated by GNU gold containing a version string.  */
+#define NT_GNU_GOLD_VERSION	4
+
 
 /* Move records.  */
 typedef struct
diff --git a/include/unistd.h b/include/unistd.h
index 72d7e2e88c..ccba893abe 100644
--- a/include/unistd.h
+++ b/include/unistd.h
@@ -1,6 +1,8 @@
 #ifndef _UNISTD_H
 # include <posix/unistd.h>
 
+__BEGIN_DECLS
+
 libc_hidden_proto (_exit, __noreturn__)
 libc_hidden_proto (alarm)
 libc_hidden_proto (confstr)
@@ -174,4 +176,6 @@ extern int __have_sock_cloexec;
    unless it is really necessary.  */
 #define __have_pipe2 __have_sock_cloexec
 
+__END_DECLS
+
 #endif
diff --git a/nptl/ChangeLog b/nptl/ChangeLog
index 1f24aa5849..c485435e82 100644
--- a/nptl/ChangeLog
+++ b/nptl/ChangeLog
@@ -1,3 +1,17 @@
+2009-07-26  Ulrich Drepper  <drepper@redhat.com>
+
+	[BZ #10418]
+	* pthread_mutex_lock.c (pthread_mutex_lock): Use _rel instead of of
+	_acq variants of cmpxchg.
+	* pthread_mutex_timedlock.c (pthread_mutex_timedlock): Likewise.
+
+2009-07-23  Ulrich Drepper  <drepper@redhat.com>
+
+	* sysdeps/x86_64/configure.in: New file.
+
+	* sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S: Fix error
+	path when not using absolute timeout futex.
+
 2009-07-20  Ulrich Drepper  <drepper@redhat.com>
 
 	* sysdeps/unix/sysv/linux/x86_64/pthread_cond_wait.S: Minor
diff --git a/nptl/pthread_mutex_lock.c b/nptl/pthread_mutex_lock.c
index 406e588fdb..a0ff881faf 100644
--- a/nptl/pthread_mutex_lock.c
+++ b/nptl/pthread_mutex_lock.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2002-2007, 2008 Free Software Foundation, Inc.
+/* Copyright (C) 2002-2007, 2008, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
 
@@ -160,7 +160,7 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
 #endif
 
 	      newval
-		= atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+		= atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 						       newval, oldval);
 
 	      if (newval != oldval)
@@ -285,7 +285,7 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
 #ifdef NO_INCR
 	newval |= FUTEX_WAITERS;
 #endif
-	oldval = atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+	oldval = atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 						      newval, 0);
 
 	if (oldval != 0)
@@ -420,7 +420,7 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
 	    oldprio = ceiling;
 
 	    oldval
-	      = atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+	      = atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 #ifdef NO_INCR
 						     ceilval | 2,
 #else
@@ -434,7 +434,7 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
 	    do
 	      {
 		oldval
-		  = atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+		  = atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 							 ceilval | 2,
 							 ceilval | 1);
 
@@ -445,7 +445,7 @@ __pthread_mutex_lock_full (pthread_mutex_t *mutex)
 		  lll_futex_wait (&mutex->__data.__lock, ceilval | 2,
 				  PTHREAD_MUTEX_PSHARED (mutex));
 	      }
-	    while (atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+	    while (atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 							ceilval | 2, ceilval)
 		   != ceilval);
 	  }
diff --git a/nptl/pthread_mutex_timedlock.c b/nptl/pthread_mutex_timedlock.c
index 8d0db79d58..2c6ff114da 100644
--- a/nptl/pthread_mutex_timedlock.c
+++ b/nptl/pthread_mutex_timedlock.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2002-2007, 2008 Free Software Foundation, Inc.
+/* Copyright (C) 2002-2007, 2008, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
 
@@ -126,7 +126,7 @@ pthread_mutex_timedlock (mutex, abstime)
 	      int newval = id | (oldval & FUTEX_WAITERS);
 
 	      newval
-		= atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+		= atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 						       newval, oldval);
 	      if (newval != oldval)
 		{
@@ -246,7 +246,7 @@ pthread_mutex_timedlock (mutex, abstime)
 	      }
 	  }
 
-	oldval = atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+	oldval = atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 						      id, 0);
 
 	if (oldval != 0)
@@ -404,7 +404,7 @@ pthread_mutex_timedlock (mutex, abstime)
 	    oldprio = ceiling;
 
 	    oldval
-	      = atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+	      = atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 						     ceilval | 1, ceilval);
 
 	    if (oldval == ceilval)
@@ -413,7 +413,7 @@ pthread_mutex_timedlock (mutex, abstime)
 	    do
 	      {
 		oldval
-		  = atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+		  = atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 							 ceilval | 2,
 							 ceilval | 1);
 
@@ -456,7 +456,7 @@ pthread_mutex_timedlock (mutex, abstime)
 					  PTHREAD_MUTEX_PSHARED (mutex));
 		  }
 	      }
-	    while (atomic_compare_and_exchange_val_acq (&mutex->__data.__lock,
+	    while (atomic_compare_and_exchange_val_rel (&mutex->__data.__lock,
 							ceilval | 2, ceilval)
 		   != ceilval);
 	  }
diff --git a/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S b/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S
index e12790cb96..7486825d5f 100644
--- a/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S
+++ b/nptl/sysdeps/unix/sysv/linux/x86_64/pthread_cond_timedwait.S
@@ -551,12 +551,12 @@ __pthread_cond_timedwait:
 	jne	53b
 
 	cmpq	24(%rsp), %r9
-	jbe	45b
+	jbe	15f
 
 	cmpq	%rax, %r9
 	ja	39b
 
-	cmpq	$-ETIMEDOUT, %r14
+15:	cmpq	$-ETIMEDOUT, %r14
 	jne	8b
 
 	jmp	99b
diff --git a/nptl/sysdeps/x86_64/configure b/nptl/sysdeps/x86_64/configure
new file mode 100644
index 0000000000..b959168843
--- /dev/null
+++ b/nptl/sysdeps/x86_64/configure
@@ -0,0 +1,36 @@
+# This file is generated from configure.in by Autoconf.  DO NOT EDIT!
+ # Local configure fragment for sysdeps/i386.
+
+{ echo "$as_me:$LINENO: checking for .cfi_personality and .cfi_lsda pseudo-ops" >&5
+echo $ECHO_N "checking for .cfi_personality and .cfi_lsda pseudo-ops... $ECHO_C" >&6; }
+if test "${libc_cv_asm_cfi_personality+set}" = set; then
+  echo $ECHO_N "(cached) $ECHO_C" >&6
+else
+    cat > conftest.s <<EOF
+${libc_cv_dot_text}
+foo:
+	.cfi_startproc
+	.cfi_personality 0, foo
+	.cfi_lsda 0, foo
+	.cfi_endproc
+EOF
+  if { ac_try='${CC-cc} $ASFLAGS -c conftest.s 1>&5'
+  { (eval echo "$as_me:$LINENO: \"$ac_try\"") >&5
+  (eval $ac_try) 2>&5
+  ac_status=$?
+  echo "$as_me:$LINENO: \$? = $ac_status" >&5
+  (exit $ac_status); }; }; then
+    libc_cv_asm_cfi_personality=yes
+  else
+    libc_cv_asm_cfi_personality=no
+  fi
+  rm -f conftest*
+
+fi
+{ echo "$as_me:$LINENO: result: $libc_cv_asm_cfi_personality" >&5
+echo "${ECHO_T}$libc_cv_asm_cfi_personality" >&6; }
+if test x"$libc_cv_asm_cfi_personality" != xyes; then
+  { { echo "$as_me:$LINENO: error: assembler too old, .cfi_personality support missing" >&5
+echo "$as_me: error: assembler too old, .cfi_personality support missing" >&2;}
+   { (exit 1); exit 1; }; }
+fi
diff --git a/nptl/sysdeps/x86_64/configure.in b/nptl/sysdeps/x86_64/configure.in
new file mode 100644
index 0000000000..0ba0cc3726
--- /dev/null
+++ b/nptl/sysdeps/x86_64/configure.in
@@ -0,0 +1,23 @@
+GLIBC_PROVIDES dnl See aclocal.m4 in the top level source directory.
+# Local configure fragment for sysdeps/i386.
+
+AC_CACHE_CHECK([for .cfi_personality and .cfi_lsda pseudo-ops],
+	       libc_cv_asm_cfi_personality, [dnl
+  cat > conftest.s <<EOF
+${libc_cv_dot_text}
+foo:
+	.cfi_startproc
+	.cfi_personality 0, foo
+	.cfi_lsda 0, foo
+	.cfi_endproc
+EOF
+  if AC_TRY_COMMAND(${CC-cc} $ASFLAGS -c conftest.s 1>&AS_MESSAGE_LOG_FD); then
+    libc_cv_asm_cfi_personality=yes
+  else
+    libc_cv_asm_cfi_personality=no
+  fi
+  rm -f conftest*
+])
+if test x"$libc_cv_asm_cfi_personality" != xyes; then
+  AC_MSG_ERROR([assembler too old, .cfi_personality support missing])
+fi
diff --git a/posix/tst-rfc3484-2.c b/posix/tst-rfc3484-2.c
index c85fdd0742..bf5f6cff7e 100644
--- a/posix/tst-rfc3484-2.c
+++ b/posix/tst-rfc3484-2.c
@@ -82,6 +82,8 @@ do_test (void)
   struct sockaddr_in so1;
   so1.sin_family = AF_INET;
   so1.sin_addr.s_addr = h (0xc0a85f19);
+  /* Clear the rest of the structure to avoid warnings.  */
+  memset (so1.sin_zero, '\0', sizeof (so1.sin_zero));
 
   struct sockaddr_in sa1;
   sa1.sin_family = AF_INET;
diff --git a/posix/tst-rfc3484-3.c b/posix/tst-rfc3484-3.c
index 3aa4563c0c..8eba74e48e 100644
--- a/posix/tst-rfc3484-3.c
+++ b/posix/tst-rfc3484-3.c
@@ -113,6 +113,8 @@ do_test (void)
   struct sockaddr_in so;
   so.sin_family = AF_INET;
   so.sin_addr.s_addr = h (0x0aa85f19);
+  /* Clear the rest of the structure to avoid warnings.  */
+  memset (so.sin_zero, '\0', sizeof (so.sin_zero));
 
   for (int i = 0; i < naddrs; ++i)
     {
diff --git a/posix/tst-rfc3484.c b/posix/tst-rfc3484.c
index 15d0c94a5e..26835cf8b2 100644
--- a/posix/tst-rfc3484.c
+++ b/posix/tst-rfc3484.c
@@ -102,6 +102,8 @@ do_test (void)
   struct sockaddr_in so;
   so.sin_family = AF_INET;
   so.sin_addr.s_addr = h (0xc0a85f19);
+  /* Clear the rest of the structure to avoid warnings.  */
+  memset (so.sin_zero, '\0', sizeof (so.sin_zero));
 
   for (int i = 0; i < naddrs; ++i)
     {
diff --git a/resolv/res_send.c b/resolv/res_send.c
index 971a4afb6f..4c14db1bf5 100644
--- a/resolv/res_send.c
+++ b/resolv/res_send.c
@@ -1278,14 +1278,10 @@ send_dg(res_state statp,
 				? *thisanssiz : *thisresplen);
 
 			if (recvresp1 || (buf2 != NULL && recvresp2))
-			  {
-			    *resplen2 = 1;
-			    return resplen;
-			  }
+			  return resplen;
 			if (buf2 != NULL)
 			  {
 			    /* We are waiting for a possible second reply.  */
-			    resplen = 1;
 			    if (hp->id == anhp->id)
 			      recvresp1 = 1;
 			    else
diff --git a/sysdeps/posix/getaddrinfo.c b/sysdeps/posix/getaddrinfo.c
index d346c621fb..a788d18fee 100644
--- a/sysdeps/posix/getaddrinfo.c
+++ b/sysdeps/posix/getaddrinfo.c
@@ -833,6 +833,8 @@ gaih_inet (const char *name, const struct gaih_service *service,
 			       && inet6_status != NSS_STATUS_UNAVAIL)
 			status = inet6_status;
 		    }
+		  else
+		    status = NSS_STATUS_UNAVAIL;
 		}
 
 	      if (nss_next_action (nip, status) == NSS_ACTION_RETURN)
diff --git a/sysdeps/s390/dl-procinfo.c b/sysdeps/s390/dl-procinfo.c
index 32c6aef951..d51d7b2379 100644
--- a/sysdeps/s390/dl-procinfo.c
+++ b/sysdeps/s390/dl-procinfo.c
@@ -1,5 +1,5 @@
 /* Data for s390 version of processor capability information.
-   Copyright (C) 2006 Free Software Foundation, Inc.
+   Copyright (C) 2006, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Martin Schwidefsky <schwidefsky@de.ibm.com>, 2006.
 
@@ -47,11 +47,11 @@
 #if !defined PROCINFO_DECL && defined SHARED
   ._dl_s390_cap_flags
 #else
-PROCINFO_CLASS const char _dl_s390_cap_flags[7][6]
+PROCINFO_CLASS const char _dl_s390_cap_flags[10][8]
 #endif
 #ifndef PROCINFO_DECL
 = {
-     "esan3", "zarch", "stfle", "msa", "ldisp", "eimm", "dfp"
+     "esan3", "zarch", "stfle", "msa", "ldisp", "eimm", "dfp", "hpage", "etf3enh", "highgprs"
   }
 #endif
 #if !defined SHARED || defined PROCINFO_DECL
@@ -63,11 +63,11 @@ PROCINFO_CLASS const char _dl_s390_cap_flags[7][6]
 #if !defined PROCINFO_DECL && defined SHARED
   ._dl_s390_platforms
 #else
-PROCINFO_CLASS const char _dl_s390_platforms[4][7]
+PROCINFO_CLASS const char _dl_s390_platforms[5][7]
 #endif
 #ifndef PROCINFO_DECL
 = {
-     "g5", "z900", "z990", "z9-109"
+    "g5", "z900", "z990", "z9-109", "z10"
   }
 #endif
 #if !defined SHARED || defined PROCINFO_DECL
diff --git a/sysdeps/s390/dl-procinfo.h b/sysdeps/s390/dl-procinfo.h
index 178d7cc017..0a7ebd3be9 100644
--- a/sysdeps/s390/dl-procinfo.h
+++ b/sysdeps/s390/dl-procinfo.h
@@ -1,5 +1,5 @@
 /* s390 version of processor capability information handling macros.
-   Copyright (C) 2006 Free Software Foundation, Inc.
+   Copyright (C) 2006, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Martin Schwidefsky <schwidefsky@de.ibm.com>, 2006.
 
@@ -22,9 +22,9 @@
 #define _DL_PROCINFO_H	1
 #include <ldsodefs.h>
 
-#define _DL_HWCAP_COUNT 7
+#define _DL_HWCAP_COUNT 10
 
-#define _DL_PLATFORMS_COUNT	4
+#define _DL_PLATFORMS_COUNT	5
 
 /* The kernel provides up to 32 capability bits with elf_hwcap.  */
 #define _DL_FIRST_PLATFORM	32
@@ -45,6 +45,9 @@ enum
   HWCAP_S390_LDISP = 1 << 4,
   HWCAP_S390_EIMM = 1 << 5,
   HWCAP_S390_DFP = 1 << 6,
+  HWCAP_S390_HPAGE = 1 << 7,
+  HWCAP_S390_ETF3EH = 1 << 8,
+  HWCAP_S390_HIGH_GPRS = 1 << 9,
 };
 
 #define HWCAP_IMPORTANT (HWCAP_S390_ZARCH | HWCAP_S390_LDISP \
diff --git a/sysdeps/s390/s390-64/Makefile b/sysdeps/s390/s390-64/Makefile
index 0a5051449d..1814f37abd 100644
--- a/sysdeps/s390/s390-64/Makefile
+++ b/sysdeps/s390/s390-64/Makefile
@@ -9,3 +9,70 @@ CFLAGS-rtld.c += -Wno-uninitialized -Wno-unused
 CFLAGS-dl-load.c += -Wno-unused
 CFLAGS-dl-reloc.c += -Wno-unused
 endif
+
+ifeq ($(subdir),iconvdata)
+ISO-8859-1_CP037_Z900-routines := iso-8859-1_cp037_z900
+ISO-8859-1_CP037_Z900-map := gconv.map
+
+UTF8_UTF32_Z9-routines := utf8-utf32-z9
+UTF8_UTF32_Z9-map := gconv.map
+
+UTF16_UTF32_Z9-routines := utf16-utf32-z9
+UTF16_UTF32_Z9-map := gconv.map
+
+UTF8_UTF16_Z9-routines := utf8-utf16-z9
+UTF8_UTF16_Z9-map := gconv.map
+
+s390x-iconv-modules = ISO-8859-1_CP037_Z900 UTF8_UTF16_Z9 UTF16_UTF32_Z9 UTF8_UTF32_Z9
+
+extra-modules-left += $(s390x-iconv-modules)
+include extra-module.mk
+
+extra-objs      += $(addsuffix .so, $(s390x-iconv-modules))
+install-others  += $(patsubst %, $(inst_gconvdir)/%.so, $(s390x-iconv-modules))
+
+distribute += iso-8859-1_cp037_z900.c utf8-utf32-z9.c utf16-utf32-z9.c utf8-utf16-z9.c
+
+$(patsubst %, $(inst_gconvdir)/%.so, $(s390x-iconv-modules)) : \
+$(inst_gconvdir)/%.so: $(objpfx)%.so $(+force)
+	$(do-install-program)
+
+$(objpfx)gconv-modules-s390: gconv-modules $(+force)
+	cp $< $@
+	echo >> $@
+	echo "# S/390 hardware accelerated modules" >> $@
+	echo -n "module	ISO-8859-1//		IBM037//	" >> $@
+	echo "	ISO-8859-1_CP037_Z900	1" >> $@
+	echo -n "module	IBM037//		ISO-8859-1//	" >> $@
+	echo "	ISO-8859-1_CP037_Z900	1" >> $@
+	echo -n "module	ISO-10646/UTF8/		UTF-32//	" >> $@
+	echo "	UTF8_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-32BE//		ISO-10646/UTF8/	" >> $@
+	echo "	UTF8_UTF32_Z9		1" >> $@
+	echo -n "module	ISO-10646/UTF8/		UTF-32BE//	" >> $@
+	echo "	UTF8_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-16BE//		UTF-32//	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-32BE//		UTF-16//	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	INTERNAL		UTF-16//	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-32BE//		UTF-16BE//	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	INTERNAL		UTF-16BE//	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-16BE//		UTF-32BE//	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-16BE//		INTERNAL	" >> $@
+	echo "	UTF16_UTF32_Z9		1" >> $@
+	echo -n "module	UTF-16BE//		ISO-10646/UTF8/	" >> $@
+	echo "	UTF8_UTF16_Z9		1" >> $@
+	echo -n "module	ISO-10646/UTF8/		UTF-16//	" >> $@
+	echo "	UTF8_UTF16_Z9		1" >> $@
+	echo -n "module	ISO-10646/UTF8/		UTF-16BE//	" >> $@
+	echo "	UTF8_UTF16_Z9		1" >> $@
+
+$(inst_gconvdir)/gconv-modules: $(objpfx)gconv-modules-s390 $(+force)
+	$(do-install)
+
+endif
diff --git a/sysdeps/s390/s390-64/iso-8859-1_cp037_z900.c b/sysdeps/s390/s390-64/iso-8859-1_cp037_z900.c
new file mode 100644
index 0000000000..d4c4931f22
--- /dev/null
+++ b/sysdeps/s390/s390-64/iso-8859-1_cp037_z900.c
@@ -0,0 +1,238 @@
+/* Conversion between ISO 8859-1 and IBM037.
+
+   This module uses the Z900 variant of the Translate One To One
+   instruction.
+   Copyright (C) 1997-2009 Free Software Foundation, Inc.
+
+   Author: Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+   Based on the work by Ulrich Drepper  <drepper@cygnus.com>, 1997.
+
+   Thanks to Daniel Appich who covered the relevant performance work
+   in his diploma thesis.
+
+   This is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   This is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+
+// conversion table from ISO-8859-1 to IBM037
+static const unsigned char table_iso8859_1_to_cp037[256]
+__attribute__ ((aligned (8))) =
+{
+  [0x00] = 0x00, [0x01] = 0x01, [0x02] = 0x02, [0x03] = 0x03,
+  [0x04] = 0x37, [0x05] = 0x2D, [0x06] = 0x2E, [0x07] = 0x2F,
+  [0x08] = 0x16, [0x09] = 0x05, [0x0A] = 0x25, [0x0B] = 0x0B,
+  [0x0C] = 0x0C, [0x0D] = 0x0D, [0x0E] = 0x0E, [0x0F] = 0x0F,
+  [0x10] = 0x10, [0x11] = 0x11, [0x12] = 0x12, [0x13] = 0x13,
+  [0x14] = 0x3C, [0x15] = 0x3D, [0x16] = 0x32, [0x17] = 0x26,
+  [0x18] = 0x18, [0x19] = 0x19, [0x1A] = 0x3F, [0x1B] = 0x27,
+  [0x1C] = 0x1C, [0x1D] = 0x1D, [0x1E] = 0x1E, [0x1F] = 0x1F,
+  [0x20] = 0x40, [0x21] = 0x5A, [0x22] = 0x7F, [0x23] = 0x7B,
+  [0x24] = 0x5B, [0x25] = 0x6C, [0x26] = 0x50, [0x27] = 0x7D,
+  [0x28] = 0x4D, [0x29] = 0x5D, [0x2A] = 0x5C, [0x2B] = 0x4E,
+  [0x2C] = 0x6B, [0x2D] = 0x60, [0x2E] = 0x4B, [0x2F] = 0x61,
+  [0x30] = 0xF0, [0x31] = 0xF1, [0x32] = 0xF2, [0x33] = 0xF3,
+  [0x34] = 0xF4, [0x35] = 0xF5, [0x36] = 0xF6, [0x37] = 0xF7,
+  [0x38] = 0xF8, [0x39] = 0xF9, [0x3A] = 0x7A, [0x3B] = 0x5E,
+  [0x3C] = 0x4C, [0x3D] = 0x7E, [0x3E] = 0x6E, [0x3F] = 0x6F,
+  [0x40] = 0x7C, [0x41] = 0xC1, [0x42] = 0xC2, [0x43] = 0xC3,
+  [0x44] = 0xC4, [0x45] = 0xC5, [0x46] = 0xC6, [0x47] = 0xC7,
+  [0x48] = 0xC8, [0x49] = 0xC9, [0x4A] = 0xD1, [0x4B] = 0xD2,
+  [0x4C] = 0xD3, [0x4D] = 0xD4, [0x4E] = 0xD5, [0x4F] = 0xD6,
+  [0x50] = 0xD7, [0x51] = 0xD8, [0x52] = 0xD9, [0x53] = 0xE2,
+  [0x54] = 0xE3, [0x55] = 0xE4, [0x56] = 0xE5, [0x57] = 0xE6,
+  [0x58] = 0xE7, [0x59] = 0xE8, [0x5A] = 0xE9, [0x5B] = 0xBA,
+  [0x5C] = 0xE0, [0x5D] = 0xBB, [0x5E] = 0xB0, [0x5F] = 0x6D,
+  [0x60] = 0x79, [0x61] = 0x81, [0x62] = 0x82, [0x63] = 0x83,
+  [0x64] = 0x84, [0x65] = 0x85, [0x66] = 0x86, [0x67] = 0x87,
+  [0x68] = 0x88, [0x69] = 0x89, [0x6A] = 0x91, [0x6B] = 0x92,
+  [0x6C] = 0x93, [0x6D] = 0x94, [0x6E] = 0x95, [0x6F] = 0x96,
+  [0x70] = 0x97, [0x71] = 0x98, [0x72] = 0x99, [0x73] = 0xA2,
+  [0x74] = 0xA3, [0x75] = 0xA4, [0x76] = 0xA5, [0x77] = 0xA6,
+  [0x78] = 0xA7, [0x79] = 0xA8, [0x7A] = 0xA9, [0x7B] = 0xC0,
+  [0x7C] = 0x4F, [0x7D] = 0xD0, [0x7E] = 0xA1, [0x7F] = 0x07,
+  [0x80] = 0x20, [0x81] = 0x21, [0x82] = 0x22, [0x83] = 0x23,
+  [0x84] = 0x24, [0x85] = 0x15, [0x86] = 0x06, [0x87] = 0x17,
+  [0x88] = 0x28, [0x89] = 0x29, [0x8A] = 0x2A, [0x8B] = 0x2B,
+  [0x8C] = 0x2C, [0x8D] = 0x09, [0x8E] = 0x0A, [0x8F] = 0x1B,
+  [0x90] = 0x30, [0x91] = 0x31, [0x92] = 0x1A, [0x93] = 0x33,
+  [0x94] = 0x34, [0x95] = 0x35, [0x96] = 0x36, [0x97] = 0x08,
+  [0x98] = 0x38, [0x99] = 0x39, [0x9A] = 0x3A, [0x9B] = 0x3B,
+  [0x9C] = 0x04, [0x9D] = 0x14, [0x9E] = 0x3E, [0x9F] = 0xFF,
+  [0xA0] = 0x41, [0xA1] = 0xAA, [0xA2] = 0x4A, [0xA3] = 0xB1,
+  [0xA4] = 0x9F, [0xA5] = 0xB2, [0xA6] = 0x6A, [0xA7] = 0xB5,
+  [0xA8] = 0xBD, [0xA9] = 0xB4, [0xAA] = 0x9A, [0xAB] = 0x8A,
+  [0xAC] = 0x5F, [0xAD] = 0xCA, [0xAE] = 0xAF, [0xAF] = 0xBC,
+  [0xB0] = 0x90, [0xB1] = 0x8F, [0xB2] = 0xEA, [0xB3] = 0xFA,
+  [0xB4] = 0xBE, [0xB5] = 0xA0, [0xB6] = 0xB6, [0xB7] = 0xB3,
+  [0xB8] = 0x9D, [0xB9] = 0xDA, [0xBA] = 0x9B, [0xBB] = 0x8B,
+  [0xBC] = 0xB7, [0xBD] = 0xB8, [0xBE] = 0xB9, [0xBF] = 0xAB,
+  [0xC0] = 0x64, [0xC1] = 0x65, [0xC2] = 0x62, [0xC3] = 0x66,
+  [0xC4] = 0x63, [0xC5] = 0x67, [0xC6] = 0x9E, [0xC7] = 0x68,
+  [0xC8] = 0x74, [0xC9] = 0x71, [0xCA] = 0x72, [0xCB] = 0x73,
+  [0xCC] = 0x78, [0xCD] = 0x75, [0xCE] = 0x76, [0xCF] = 0x77,
+  [0xD0] = 0xAC, [0xD1] = 0x69, [0xD2] = 0xED, [0xD3] = 0xEE,
+  [0xD4] = 0xEB, [0xD5] = 0xEF, [0xD6] = 0xEC, [0xD7] = 0xBF,
+  [0xD8] = 0x80, [0xD9] = 0xFD, [0xDA] = 0xFE, [0xDB] = 0xFB,
+  [0xDC] = 0xFC, [0xDD] = 0xAD, [0xDE] = 0xAE, [0xDF] = 0x59,
+  [0xE0] = 0x44, [0xE1] = 0x45, [0xE2] = 0x42, [0xE3] = 0x46,
+  [0xE4] = 0x43, [0xE5] = 0x47, [0xE6] = 0x9C, [0xE7] = 0x48,
+  [0xE8] = 0x54, [0xE9] = 0x51, [0xEA] = 0x52, [0xEB] = 0x53,
+  [0xEC] = 0x58, [0xED] = 0x55, [0xEE] = 0x56, [0xEF] = 0x57,
+  [0xF0] = 0x8C, [0xF1] = 0x49, [0xF2] = 0xCD, [0xF3] = 0xCE,
+  [0xF4] = 0xCB, [0xF5] = 0xCF, [0xF6] = 0xCC, [0xF7] = 0xE1,
+  [0xF8] = 0x70, [0xF9] = 0xDD, [0xFA] = 0xDE, [0xFB] = 0xDB,
+  [0xFC] = 0xDC, [0xFD] = 0x8D, [0xFE] = 0x8E, [0xFF] = 0xDF
+};
+
+// conversion table from IBM037 to ISO-8859-1
+static const unsigned char table_cp037_iso8859_1[256]
+__attribute__ ((aligned (8))) =
+{
+  [0x00] = 0x00, [0x01] = 0x01, [0x02] = 0x02, [0x03] = 0x03,
+  [0x04] = 0x9C, [0x05] = 0x09, [0x06] = 0x86, [0x07] = 0x7F,
+  [0x08] = 0x97, [0x09] = 0x8D, [0x0A] = 0x8E, [0x0B] = 0x0B,
+  [0x0C] = 0x0C, [0x0D] = 0x0D, [0x0E] = 0x0E, [0x0F] = 0x0F,
+  [0x10] = 0x10, [0x11] = 0x11, [0x12] = 0x12, [0x13] = 0x13,
+  [0x14] = 0x9D, [0x15] = 0x85, [0x16] = 0x08, [0x17] = 0x87,
+  [0x18] = 0x18, [0x19] = 0x19, [0x1A] = 0x92, [0x1B] = 0x8F,
+  [0x1C] = 0x1C, [0x1D] = 0x1D, [0x1E] = 0x1E, [0x1F] = 0x1F,
+  [0x20] = 0x80, [0x21] = 0x81, [0x22] = 0x82, [0x23] = 0x83,
+  [0x24] = 0x84, [0x25] = 0x0A, [0x26] = 0x17, [0x27] = 0x1B,
+  [0x28] = 0x88, [0x29] = 0x89, [0x2A] = 0x8A, [0x2B] = 0x8B,
+  [0x2C] = 0x8C, [0x2D] = 0x05, [0x2E] = 0x06, [0x2F] = 0x07,
+  [0x30] = 0x90, [0x31] = 0x91, [0x32] = 0x16, [0x33] = 0x93,
+  [0x34] = 0x94, [0x35] = 0x95, [0x36] = 0x96, [0x37] = 0x04,
+  [0x38] = 0x98, [0x39] = 0x99, [0x3A] = 0x9A, [0x3B] = 0x9B,
+  [0x3C] = 0x14, [0x3D] = 0x15, [0x3E] = 0x9E, [0x3F] = 0x1A,
+  [0x40] = 0x20, [0x41] = 0xA0, [0x42] = 0xE2, [0x43] = 0xE4,
+  [0x44] = 0xE0, [0x45] = 0xE1, [0x46] = 0xE3, [0x47] = 0xE5,
+  [0x48] = 0xE7, [0x49] = 0xF1, [0x4A] = 0xA2, [0x4B] = 0x2E,
+  [0x4C] = 0x3C, [0x4D] = 0x28, [0x4E] = 0x2B, [0x4F] = 0x7C,
+  [0x50] = 0x26, [0x51] = 0xE9, [0x52] = 0xEA, [0x53] = 0xEB,
+  [0x54] = 0xE8, [0x55] = 0xED, [0x56] = 0xEE, [0x57] = 0xEF,
+  [0x58] = 0xEC, [0x59] = 0xDF, [0x5A] = 0x21, [0x5B] = 0x24,
+  [0x5C] = 0x2A, [0x5D] = 0x29, [0x5E] = 0x3B, [0x5F] = 0xAC,
+  [0x60] = 0x2D, [0x61] = 0x2F, [0x62] = 0xC2, [0x63] = 0xC4,
+  [0x64] = 0xC0, [0x65] = 0xC1, [0x66] = 0xC3, [0x67] = 0xC5,
+  [0x68] = 0xC7, [0x69] = 0xD1, [0x6A] = 0xA6, [0x6B] = 0x2C,
+  [0x6C] = 0x25, [0x6D] = 0x5F, [0x6E] = 0x3E, [0x6F] = 0x3F,
+  [0x70] = 0xF8, [0x71] = 0xC9, [0x72] = 0xCA, [0x73] = 0xCB,
+  [0x74] = 0xC8, [0x75] = 0xCD, [0x76] = 0xCE, [0x77] = 0xCF,
+  [0x78] = 0xCC, [0x79] = 0x60, [0x7A] = 0x3A, [0x7B] = 0x23,
+  [0x7C] = 0x40, [0x7D] = 0x27, [0x7E] = 0x3D, [0x7F] = 0x22,
+  [0x80] = 0xD8, [0x81] = 0x61, [0x82] = 0x62, [0x83] = 0x63,
+  [0x84] = 0x64, [0x85] = 0x65, [0x86] = 0x66, [0x87] = 0x67,
+  [0x88] = 0x68, [0x89] = 0x69, [0x8A] = 0xAB, [0x8B] = 0xBB,
+  [0x8C] = 0xF0, [0x8D] = 0xFD, [0x8E] = 0xFE, [0x8F] = 0xB1,
+  [0x90] = 0xB0, [0x91] = 0x6A, [0x92] = 0x6B, [0x93] = 0x6C,
+  [0x94] = 0x6D, [0x95] = 0x6E, [0x96] = 0x6F, [0x97] = 0x70,
+  [0x98] = 0x71, [0x99] = 0x72, [0x9A] = 0xAA, [0x9B] = 0xBA,
+  [0x9C] = 0xE6, [0x9D] = 0xB8, [0x9E] = 0xC6, [0x9F] = 0xA4,
+  [0xA0] = 0xB5, [0xA1] = 0x7E, [0xA2] = 0x73, [0xA3] = 0x74,
+  [0xA4] = 0x75, [0xA5] = 0x76, [0xA6] = 0x77, [0xA7] = 0x78,
+  [0xA8] = 0x79, [0xA9] = 0x7A, [0xAA] = 0xA1, [0xAB] = 0xBF,
+  [0xAC] = 0xD0, [0xAD] = 0xDD, [0xAE] = 0xDE, [0xAF] = 0xAE,
+  [0xB0] = 0x5E, [0xB1] = 0xA3, [0xB2] = 0xA5, [0xB3] = 0xB7,
+  [0xB4] = 0xA9, [0xB5] = 0xA7, [0xB6] = 0xB6, [0xB7] = 0xBC,
+  [0xB8] = 0xBD, [0xB9] = 0xBE, [0xBA] = 0x5B, [0xBB] = 0x5D,
+  [0xBC] = 0xAF, [0xBD] = 0xA8, [0xBE] = 0xB4, [0xBF] = 0xD7,
+  [0xC0] = 0x7B, [0xC1] = 0x41, [0xC2] = 0x42, [0xC3] = 0x43,
+  [0xC4] = 0x44, [0xC5] = 0x45, [0xC6] = 0x46, [0xC7] = 0x47,
+  [0xC8] = 0x48, [0xC9] = 0x49, [0xCA] = 0xAD, [0xCB] = 0xF4,
+  [0xCC] = 0xF6, [0xCD] = 0xF2, [0xCE] = 0xF3, [0xCF] = 0xF5,
+  [0xD0] = 0x7D, [0xD1] = 0x4A, [0xD2] = 0x4B, [0xD3] = 0x4C,
+  [0xD4] = 0x4D, [0xD5] = 0x4E, [0xD6] = 0x4F, [0xD7] = 0x50,
+  [0xD8] = 0x51, [0xD9] = 0x52, [0xDA] = 0xB9, [0xDB] = 0xFB,
+  [0xDC] = 0xFC, [0xDD] = 0xF9, [0xDE] = 0xFA, [0xDF] = 0xFF,
+  [0xE0] = 0x5C, [0xE1] = 0xF7, [0xE2] = 0x53, [0xE3] = 0x54,
+  [0xE4] = 0x55, [0xE5] = 0x56, [0xE6] = 0x57, [0xE7] = 0x58,
+  [0xE8] = 0x59, [0xE9] = 0x5A, [0xEA] = 0xB2, [0xEB] = 0xD4,
+  [0xEC] = 0xD6, [0xED] = 0xD2, [0xEE] = 0xD3, [0xEF] = 0xD5,
+  [0xF0] = 0x30, [0xF1] = 0x31, [0xF2] = 0x32, [0xF3] = 0x33,
+  [0xF4] = 0x34, [0xF5] = 0x35, [0xF6] = 0x36, [0xF7] = 0x37,
+  [0xF8] = 0x38, [0xF9] = 0x39, [0xFA] = 0xB3, [0xFB] = 0xDB,
+  [0xFC] = 0xDC, [0xFD] = 0xD9, [0xFE] = 0xDA, [0xFF] = 0x9F
+};
+
+/* Definitions used in the body of the `gconv' function.  */
+#define CHARSET_NAME		"ISO-8859-1//"
+#define FROM_LOOP		iso8859_1_to_cp037_z900
+#define TO_LOOP			cp037_to_iso8859_1_z900
+#define DEFINE_INIT		1
+#define DEFINE_FINI		1
+#define MIN_NEEDED_FROM		1
+#define MIN_NEEDED_TO		1
+
+/* The Z900 variant of troo forces us to always specify a test
+   character which ends the translation.  So if we run into the
+   situation where the translation has been interrupted due to the
+   test character we translate the character by hand and jump back
+   into the instruction.  */
+
+#define TROO_LOOP(TABLE)						\
+  {									\
+    register const unsigned char test asm ("0") = 0;			\
+    register const unsigned char *pTable asm ("1") = TABLE;		\
+    register unsigned char *pOutput asm ("2") = outptr;			\
+    register uint64_t length asm ("3");					\
+    const unsigned char* pInput = inptr;				\
+    uint64_t tmp;							\
+									\
+    length = (inend - inptr < outend - outptr				\
+	      ? inend - inptr : outend - outptr);			\
+									\
+    asm volatile ("0:                        \n\t"			\
+                  "  troo    %0,%1           \n\t"			\
+                  "  jz      1f              \n\t"			\
+                  "  jo      0b              \n\t"			\
+                  "  llgc    %3,0(%1)        \n\t"			\
+                  "  la      %3,0(%3,%4)     \n\t"			\
+                  "  mvc     0(1,%0),0(%3)   \n\t"			\
+                  "  aghi    %1,1            \n\t"			\
+                  "  aghi    %0,1            \n\t"			\
+                  "  aghi    %2,-1           \n\t"			\
+                  "  j       0b              \n\t"			\
+                  "1:                        \n"			\
+									\
+     : "+a" (pOutput), "+a" (pInput), "+d" (length), "=&a" (tmp)        \
+     : "a" (pTable), "d" (test)						\
+     : "cc");								\
+									\
+    inptr = pInput;							\
+    outptr = pOutput;							\
+  }
+
+/* First define the conversion function from ISO 8859-1 to CP037.  */
+#define MIN_NEEDED_INPUT	MIN_NEEDED_FROM
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_TO
+#define LOOPFCT			FROM_LOOP
+#define BODY TROO_LOOP (table_iso8859_1_to_cp037)
+
+#include <iconv/loop.c>
+
+
+/* Next, define the conversion function from CP037 to ISO 8859-1.  */
+#define MIN_NEEDED_INPUT	MIN_NEEDED_TO
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_FROM
+#define LOOPFCT			TO_LOOP
+#define BODY TROO_LOOP (table_cp037_iso8859_1);
+
+#include <iconv/loop.c>
+
+
+/* Now define the toplevel functions.  */
+#include <iconv/skeleton.c>
diff --git a/sysdeps/s390/s390-64/utf16-utf32-z9.c b/sysdeps/s390/s390-64/utf16-utf32-z9.c
new file mode 100644
index 0000000000..868dea68ca
--- /dev/null
+++ b/sysdeps/s390/s390-64/utf16-utf32-z9.c
@@ -0,0 +1,325 @@
+/* Conversion between UTF-16 and UTF-32 BE/internal.
+
+   This module uses the Z9-109 variants of the Convert Unicode
+   instructions.
+   Copyright (C) 1997-2009 Free Software Foundation, Inc.
+
+   Author: Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+   Based on the work by Ulrich Drepper  <drepper@cygnus.com>, 1997.
+
+   Thanks to Daniel Appich who covered the relevant performance work
+   in his diploma thesis.
+
+   This is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   This is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <dl-procinfo.h>
+#include <gconv.h>
+
+/* UTF-32 big endian byte order mark.  */
+#define BOM_UTF32               0x0000feffu
+
+/* UTF-16 big endian byte order mark.  */
+#define BOM_UTF16	        0xfeff
+
+#define DEFINE_INIT		0
+#define DEFINE_FINI		0
+#define MIN_NEEDED_FROM		2
+#define MAX_NEEDED_FROM		4
+#define MIN_NEEDED_TO		4
+#define FROM_LOOP		from_utf16_loop
+#define TO_LOOP			to_utf16_loop
+#define FROM_DIRECTION		(dir == from_utf16)
+#define PREPARE_LOOP							\
+  enum direction dir = ((struct utf16_data *) step->__data)->dir;	\
+  int emit_bom = ((struct utf16_data *) step->__data)->emit_bom;	\
+									\
+  if (emit_bom && !data->__internal_use					\
+      && data->__invocation_counter == 0)				\
+    {									\
+      if (dir == to_utf16)						\
+	{								\
+          /* Emit the UTF-16 Byte Order Mark.  */			\
+          if (__builtin_expect (outbuf + 2 > outend, 0))		\
+	    return __GCONV_FULL_OUTPUT;					\
+									\
+	  put16u (outbuf, BOM_UTF16);					\
+	  outbuf += 2;							\
+	}								\
+      else								\
+	{								\
+          /* Emit the UTF-32 Byte Order Mark.  */			\
+	  if (__builtin_expect (outbuf + 4 > outend, 0))		\
+	    return __GCONV_FULL_OUTPUT;					\
+									\
+	  put32u (outbuf, BOM_UTF32);					\
+	  outbuf += 4;							\
+	}								\
+    }
+
+/* Direction of the transformation.  */
+enum direction
+{
+  illegal_dir,
+  to_utf16,
+  from_utf16
+};
+
+struct utf16_data
+{
+  enum direction dir;
+  int emit_bom;
+};
+
+
+extern int gconv_init (struct __gconv_step *step);
+int
+gconv_init (struct __gconv_step *step)
+{
+  /* Determine which direction.  */
+  struct utf16_data *new_data;
+  enum direction dir = illegal_dir;
+  int emit_bom;
+  int result;
+
+  emit_bom = (__strcasecmp (step->__to_name, "UTF-32//") == 0
+	      || __strcasecmp (step->__to_name, "UTF-16//") == 0);
+
+  if (__strcasecmp (step->__from_name, "UTF-16BE//") == 0
+      && (__strcasecmp (step->__to_name, "UTF-32//") == 0
+	  || __strcasecmp (step->__to_name, "UTF-32BE//") == 0
+	  || __strcasecmp (step->__to_name, "INTERNAL") == 0))
+    {
+      dir = from_utf16;
+    }
+  else if ((__strcasecmp (step->__to_name, "UTF-16//") == 0
+	    || __strcasecmp (step->__to_name, "UTF-16BE//") == 0)
+	   && (__strcasecmp (step->__from_name, "UTF-32BE//") == 0
+	       || __strcasecmp (step->__from_name, "INTERNAL") == 0))
+    {
+      dir = to_utf16;
+    }
+
+  result = __GCONV_NOCONV;
+  if (dir != illegal_dir)
+    {
+      new_data = (struct utf16_data *) malloc (sizeof (struct utf16_data));
+
+      result = __GCONV_NOMEM;
+      if (new_data != NULL)
+	{
+	  new_data->dir = dir;
+	  new_data->emit_bom = emit_bom;
+	  step->__data = new_data;
+
+	  if (dir == from_utf16)
+	    {
+	      step->__min_needed_from = MIN_NEEDED_FROM;
+	      step->__max_needed_from = MIN_NEEDED_FROM;
+	      step->__min_needed_to = MIN_NEEDED_TO;
+	      step->__max_needed_to = MIN_NEEDED_TO;
+	    }
+	  else
+	    {
+	      step->__min_needed_from = MIN_NEEDED_TO;
+	      step->__max_needed_from = MIN_NEEDED_TO;
+	      step->__min_needed_to = MIN_NEEDED_FROM;
+	      step->__max_needed_to = MIN_NEEDED_FROM;
+	    }
+
+	  step->__stateful = 0;
+
+	  result = __GCONV_OK;
+	}
+    }
+
+  return result;
+}
+
+
+extern void gconv_end (struct __gconv_step *data);
+void
+gconv_end (struct __gconv_step *data)
+{
+  free (data->__data);
+}
+
+/* The macro for the hardware loop.  This is used for both
+   directions.  */
+#define HARDWARE_CONVERT(INSTRUCTION)					\
+  {									\
+    register const unsigned char* pInput asm ("8") = inptr;		\
+    register unsigned long long inlen asm ("9") = inend - inptr;	\
+    register unsigned char* pOutput asm ("10") = outptr;		\
+    register unsigned long long outlen asm("11") = outend - outptr;	\
+    uint64_t cc = 0;							\
+									\
+    asm volatile ("0: " INSTRUCTION "  \n\t"				\
+                  "   jo     0b        \n\t"				\
+		  "   ipm    %2        \n"			        \
+		  : "+a" (pOutput), "+a" (pInput), "+d" (cc),		\
+		    "+d" (outlen), "+d" (inlen)				\
+		  :							\
+		  : "cc", "memory");					\
+									\
+    inptr = pInput;							\
+    outptr = pOutput;							\
+    cc >>= 28;								\
+									\
+    if (cc == 1)							\
+      {									\
+	result = __GCONV_FULL_OUTPUT;					\
+	break;								\
+      }									\
+    else if (cc == 2)							\
+      {									\
+	result = __GCONV_ILLEGAL_INPUT;					\
+	break;								\
+      }									\
+  }
+
+/* Conversion function from UTF-16 to UTF-32 internal/BE.  */
+
+#define MIN_NEEDED_INPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_INPUT	MAX_NEEDED_FROM
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_TO
+#define LOOPFCT			FROM_LOOP
+/* The software routine is copied from utf-16.c (minus bytes
+   swapping).  */
+#define BODY								\
+  {									\
+    if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH)				\
+      {									\
+	HARDWARE_CONVERT ("cu24 %0, %1, 1");				\
+	if (inptr != inend)						\
+	  {								\
+	    /* Check if the third byte is				\
+	       a valid start of a UTF-16 surrogate.  */			\
+	    if (inend - inptr == 3 && (inptr[3] & 0xfc) != 0xdc)	\
+	      STANDARD_FROM_LOOP_ERR_HANDLER (3);			\
+									\
+	    result = __GCONV_INCOMPLETE_INPUT;				\
+	    break;							\
+	  }								\
+	continue;							\
+      }									\
+									\
+    uint16_t u1 = get16 (inptr);					\
+									\
+    if (__builtin_expect (u1 < 0xd800, 1) || u1 > 0xdfff)		\
+      {									\
+	/* No surrogate.  */						\
+	put32 (outptr, u1);						\
+	inptr += 2;							\
+      }									\
+    else								\
+      {									\
+	/* It's a surrogate character.  At least the first word says	\
+	   it is.  */							\
+	if (__builtin_expect (inptr + 4 > inend, 0))			\
+	  {								\
+	    /* We don't have enough input for another complete input	\
+	       character.  */						\
+	    result = __GCONV_INCOMPLETE_INPUT;				\
+	    break;							\
+	  }								\
+									\
+	inptr += 2;							\
+	uint16_t u2 = get16 (inptr);					\
+	if (__builtin_expect (u2 < 0xdc00, 0)				\
+	    || __builtin_expect (u2 > 0xdfff, 0))			\
+	  {								\
+	    /* This is no valid second word for a surrogate.  */	\
+	    inptr -= 2;							\
+	    STANDARD_FROM_LOOP_ERR_HANDLER (2);				\
+	  }								\
+									\
+	put32 (outptr, ((u1 - 0xd7c0) << 10) + (u2 - 0xdc00));		\
+	inptr += 2;							\
+      }									\
+    outptr += 4;							\
+  }
+#define LOOP_NEED_FLAGS
+#include <iconv/loop.c>
+
+/* Conversion from UTF-32 internal/BE to UTF-16.  */
+
+#define MIN_NEEDED_INPUT	MIN_NEEDED_TO
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_OUTPUT	MAX_NEEDED_FROM
+#define LOOPFCT			TO_LOOP
+/* The software routine is copied from utf-16.c (minus bytes
+   swapping).  */
+#define BODY								\
+  {									\
+    if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH)				\
+      {									\
+	HARDWARE_CONVERT ("cu42 %0, %1");				\
+									\
+	if (inptr != inend)						\
+	  {								\
+	    result = __GCONV_INCOMPLETE_INPUT;				\
+	    break;							\
+	  }								\
+	continue;							\
+      }									\
+									\
+    uint32_t c = get32 (inptr);						\
+									\
+    if (__builtin_expect (c <= 0xd7ff, 1)				\
+	|| (c >=0xdc00 && c <= 0xffff))					\
+      {									\
+        /* Two UTF-16 chars.  */					\
+        put16 (outptr, c);						\
+      }									\
+    else if (__builtin_expect (c >= 0x10000, 1)				\
+	     && __builtin_expect (c <= 0x10ffff, 1))			\
+      {									\
+	/* Four UTF-16 chars.  */					\
+        uint16_t zabcd = ((c & 0x1f0000) >> 16) - 1;			\
+	uint16_t out;							\
+									\
+	/* Generate a surrogate character.  */				\
+	if (__builtin_expect (outptr + 4 > outend, 0))			\
+	  {								\
+	    /* Overflow in the output buffer.  */			\
+	    result = __GCONV_FULL_OUTPUT;				\
+	    break;							\
+	  }								\
+									\
+	out = 0xd800;							\
+	out |= (zabcd & 0xff) << 6;					\
+	out |= (c >> 10) & 0x3f;					\
+	put16 (outptr, out);						\
+	outptr += 2;							\
+									\
+	out = 0xdc00;							\
+	out |= c & 0x3ff;						\
+	put16 (outptr, out);						\
+      }									\
+    else								\
+      {									\
+        STANDARD_TO_LOOP_ERR_HANDLER (4);				\
+      }									\
+    outptr += 2;							\
+    inptr += 4;								\
+  }
+#define LOOP_NEED_FLAGS
+#include <iconv/loop.c>
+
+#include <iconv/skeleton.c>
diff --git a/sysdeps/s390/s390-64/utf8-utf16-z9.c b/sysdeps/s390/s390-64/utf8-utf16-z9.c
new file mode 100644
index 0000000000..531d3ebd4b
--- /dev/null
+++ b/sysdeps/s390/s390-64/utf8-utf16-z9.c
@@ -0,0 +1,463 @@
+/* Conversion between UTF-16 and UTF-32 BE/internal.
+
+   This module uses the Z9-109 variants of the Convert Unicode
+   instructions.
+   Copyright (C) 1997-2009 Free Software Foundation, Inc.
+
+   Author: Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+   Based on the work by Ulrich Drepper  <drepper@cygnus.com>, 1997.
+
+   Thanks to Daniel Appich who covered the relevant performance work
+   in his diploma thesis.
+
+   This is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   This is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <dl-procinfo.h>
+#include <gconv.h>
+
+/* UTF-16 big endian byte order mark.  */
+#define BOM_UTF16	0xfeff
+
+#define DEFINE_INIT		0
+#define DEFINE_FINI		0
+#define MIN_NEEDED_FROM		1
+#define MAX_NEEDED_FROM		4
+#define MIN_NEEDED_TO		2
+#define MAX_NEEDED_TO		4
+#define FROM_LOOP		from_utf8_loop
+#define TO_LOOP			to_utf8_loop
+#define FROM_DIRECTION		(dir == from_utf8)
+#define PREPARE_LOOP							\
+  enum direction dir = ((struct utf8_data *) step->__data)->dir;	\
+  int emit_bom = ((struct utf8_data *) step->__data)->emit_bom;		\
+									\
+  if (emit_bom && !data->__internal_use					\
+      && data->__invocation_counter == 0)				\
+    {									\
+      /* Emit the UTF-16 Byte Order Mark.  */				\
+      if (__builtin_expect (outbuf + 2 > outend, 0))			\
+	return __GCONV_FULL_OUTPUT;					\
+									\
+      put16u (outbuf, BOM_UTF16);					\
+      outbuf += 2;							\
+    }
+
+/* Direction of the transformation.  */
+enum direction
+{
+  illegal_dir,
+  to_utf8,
+  from_utf8
+};
+
+struct utf8_data
+{
+  enum direction dir;
+  int emit_bom;
+};
+
+
+extern int gconv_init (struct __gconv_step *step);
+int
+gconv_init (struct __gconv_step *step)
+{
+  /* Determine which direction.  */
+  struct utf8_data *new_data;
+  enum direction dir = illegal_dir;
+  int emit_bom;
+  int result;
+
+  emit_bom = (__strcasecmp (step->__to_name, "UTF-16//") == 0);
+
+  if (__strcasecmp (step->__from_name, "ISO-10646/UTF8/") == 0
+      && (__strcasecmp (step->__to_name, "UTF-16//") == 0
+	  || __strcasecmp (step->__to_name, "UTF-16BE//") == 0))
+    {
+      dir = from_utf8;
+    }
+  else if (__strcasecmp (step->__from_name, "UTF-16BE//") == 0
+	   && __strcasecmp (step->__to_name, "ISO-10646/UTF8/") == 0)
+    {
+      dir = to_utf8;
+    }
+
+  result = __GCONV_NOCONV;
+  if (dir != illegal_dir)
+    {
+      new_data = (struct utf8_data *) malloc (sizeof (struct utf8_data));
+
+      result = __GCONV_NOMEM;
+      if (new_data != NULL)
+	{
+	  new_data->dir = dir;
+	  new_data->emit_bom = emit_bom;
+	  step->__data = new_data;
+
+	  if (dir == from_utf8)
+	    {
+	      step->__min_needed_from = MIN_NEEDED_FROM;
+	      step->__max_needed_from = MIN_NEEDED_FROM;
+	      step->__min_needed_to = MIN_NEEDED_TO;
+	      step->__max_needed_to = MIN_NEEDED_TO;
+	    }
+	  else
+	    {
+	      step->__min_needed_from = MIN_NEEDED_TO;
+	      step->__max_needed_from = MIN_NEEDED_TO;
+	      step->__min_needed_to = MIN_NEEDED_FROM;
+	      step->__max_needed_to = MIN_NEEDED_FROM;
+	    }
+
+	  step->__stateful = 0;
+
+	  result = __GCONV_OK;
+	}
+    }
+
+  return result;
+}
+
+
+extern void gconv_end (struct __gconv_step *data);
+void
+gconv_end (struct __gconv_step *data)
+{
+  free (data->__data);
+}
+
+/* The macro for the hardware loop.  This is used for both
+   directions.  */
+#define HARDWARE_CONVERT(INSTRUCTION)					\
+  {									\
+    register const unsigned char* pInput asm ("8") = inptr;		\
+    register unsigned long long inlen asm ("9") = inend - inptr;	\
+    register unsigned char* pOutput asm ("10") = outptr;		\
+    register unsigned long long outlen asm("11") = outend - outptr;	\
+    uint64_t cc = 0;							\
+									\
+    asm volatile ("0: " INSTRUCTION "  \n\t"				\
+                  "   jo     0b        \n\t"				\
+		  "   ipm    %2        \n"			        \
+		  : "+a" (pOutput), "+a" (pInput), "+d" (cc),		\
+		    "+d" (outlen), "+d" (inlen)				\
+		  :							\
+		  : "cc", "memory");					\
+									\
+    inptr = pInput;							\
+    outptr = pOutput;							\
+    cc >>= 28;								\
+									\
+    if (cc == 1)							\
+      {									\
+	result = __GCONV_FULL_OUTPUT;					\
+	break;								\
+      }									\
+    else if (cc == 2)							\
+      {									\
+	result = __GCONV_ILLEGAL_INPUT;					\
+	break;								\
+      }									\
+  }
+
+/* Conversion function from UTF-8 to UTF-16.  */
+
+#define MIN_NEEDED_INPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_INPUT	MAX_NEEDED_FROM
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_TO
+#define LOOPFCT			FROM_LOOP
+/* The software implementation is based on the code in gconv_simple.c.  */
+#define BODY								\
+  {									\
+    if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH)				\
+      {									\
+	HARDWARE_CONVERT ("cu12 %0, %1, 1");				\
+									\
+	if (inptr != inend)						\
+	  {								\
+	    int i;							\
+	    for (i = 1; inptr + i < inend; ++i)				\
+	      if ((inptr[i] & 0xc0) != 0x80)				\
+		break;							\
+								\
+	    if (__builtin_expect (inptr + i == inend, 1))		\
+	      {								\
+		result = __GCONV_INCOMPLETE_INPUT;			\
+		break;							\
+	      }								\
+	    STANDARD_FROM_LOOP_ERR_HANDLER (i);				\
+	  }								\
+	continue;							\
+    }									\
+									\
+    /* Next input byte.  */						\
+    uint16_t ch = *inptr;						\
+									\
+    if (__builtin_expect (ch < 0x80, 1))				\
+      {									\
+	/* One byte sequence.  */					\
+	++inptr;							\
+      }									\
+    else								\
+      {									\
+	uint_fast32_t cnt;						\
+	uint_fast32_t i;						\
+									\
+	if (ch >= 0xc2 && ch < 0xe0)					\
+	  {								\
+	    /* We expect two bytes.  The first byte cannot be 0xc0	\
+	       or 0xc1, otherwise the wide character could have been	\
+	       represented using a single byte.  */			\
+	    cnt = 2;							\
+	    ch &= 0x1f;							\
+	  }								\
+        else if (__builtin_expect ((ch & 0xf0) == 0xe0, 1))		\
+	  {								\
+	    /* We expect three bytes.  */				\
+	    cnt = 3;							\
+	    ch &= 0x0f;							\
+	  }								\
+	else if (__builtin_expect ((ch & 0xf8) == 0xf0, 1))		\
+	  {								\
+	    /* We expect four bytes.  */				\
+	    cnt = 4;							\
+	    ch &= 0x07;							\
+	  }								\
+	else								\
+	  {								\
+	    /* Search the end of this ill-formed UTF-8 character.  This	\
+	       is the next byte with (x & 0xc0) != 0x80.  */		\
+	    i = 0;							\
+	    do								\
+	      ++i;							\
+	    while (inptr + i < inend					\
+		   && (*(inptr + i) & 0xc0) == 0x80			\
+		   && i < 5);						\
+									\
+	  errout:							\
+	    STANDARD_FROM_LOOP_ERR_HANDLER (i);				\
+	  }								\
+									\
+	if (__builtin_expect (inptr + cnt > inend, 0))			\
+	  {								\
+	    /* We don't have enough input.  But before we report	\
+	       that check that all the bytes are correct.  */		\
+	    for (i = 1; inptr + i < inend; ++i)				\
+	      if ((inptr[i] & 0xc0) != 0x80)				\
+		break;							\
+									\
+	    if (__builtin_expect (inptr + i == inend, 1))		\
+	      {								\
+		result = __GCONV_INCOMPLETE_INPUT;			\
+		break;							\
+	      }								\
+									\
+	    goto errout;						\
+	  }								\
+									\
+	if (cnt == 4)							\
+	  {								\
+	    /* For 4 byte UTF-8 chars two UTF-16 chars (high and	\
+	       low) are needed.  */					\
+	    uint16_t zabcd, high, low;					\
+									\
+	    if (__builtin_expect (outptr + 4 > outend, 0))		\
+	      {								\
+		/* Overflow in the output buffer.  */			\
+		result = __GCONV_FULL_OUTPUT;				\
+		break;							\
+	      }								\
+									\
+	    /* See Principles of Operations cu12.  */			\
+	    zabcd = (((inptr[0] & 0x7) << 2) |				\
+                     ((inptr[1] & 0x30) >> 4)) - 1;			\
+									\
+	    /* z-bit must be zero after subtracting 1.  */		\
+	    if (zabcd & 0x10)						\
+	      STANDARD_FROM_LOOP_ERR_HANDLER (4)			\
+									\
+	    high = (uint16_t)(0xd8 << 8);       /* high surrogate id */ \
+	    high |= zabcd << 6;	                        /* abcd bits */	\
+	    high |= (inptr[1] & 0xf) << 2;              /* efgh bits */	\
+	    high |= (inptr[2] & 0x30) >> 4;               /* ij bits */	\
+									\
+	    low = (uint16_t)(0xdc << 8);         /* low surrogate id */ \
+	    low |= ((uint16_t)inptr[2] & 0xc) << 6;       /* kl bits */	\
+	    low |= (inptr[2] & 0x3) << 6;                 /* mn bits */	\
+	    low |= inptr[3] & 0x3f;                   /* opqrst bits */	\
+									\
+	    put16 (outptr, high);					\
+	    outptr += 2;						\
+	    put16 (outptr, low);					\
+	    outptr += 2;						\
+	    inptr += 4;							\
+	    continue;							\
+	  }								\
+	else								\
+	  {								\
+	    /* Read the possible remaining bytes.  */			\
+	    for (i = 1; i < cnt; ++i)					\
+	      {								\
+		uint16_t byte = inptr[i];				\
+									\
+		if ((byte & 0xc0) != 0x80)				\
+		  /* This is an illegal encoding.  */			\
+		  break;						\
+									\
+		ch <<= 6;						\
+		ch |= byte & 0x3f;					\
+	      }								\
+	    inptr += cnt;						\
+									\
+	  }								\
+      }									\
+    /* Now adjust the pointers and store the result.  */		\
+    *((uint16_t *) outptr) = ch;					\
+    outptr += sizeof (uint16_t);					\
+  }
+
+#define LOOP_NEED_FLAGS
+#include <iconv/loop.c>
+
+/* Conversion from UTF-16 to UTF-8.  */
+
+#define MIN_NEEDED_INPUT	MIN_NEEDED_TO
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_OUTPUT	MAX_NEEDED_FROM
+#define LOOPFCT			TO_LOOP
+/* The software routine is based on the functionality of the S/390
+   hardware instruction (cu21) as described in the Principles of
+   Operation.  */
+#define BODY								\
+  {									\
+    if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH)				\
+      {									\
+	HARDWARE_CONVERT ("cu21 %0, %1");				\
+	if (inptr != inend)						\
+	  {								\
+	    /* Check if the third byte is				\
+	       a valid start of a UTF-16 surrogate.  */			\
+	    if (inend - inptr == 3 && (inptr[3] & 0xfc) != 0xdc)	\
+	      STANDARD_TO_LOOP_ERR_HANDLER (3);				\
+									\
+	    result = __GCONV_INCOMPLETE_INPUT;				\
+	    break;							\
+	  }								\
+	continue;							\
+      }									\
+									\
+    uint16_t c = get16 (inptr);						\
+									\
+    if (__builtin_expect (c <= 0x007f, 1))				\
+      {									\
+	/* Single byte UTF-8 char.  */					\
+	*outptr = c & 0xff;						\
+	outptr++;							\
+      }									\
+    else if (c >= 0x0080 && c <= 0x07ff)				\
+      {									\
+        /* Two byte UTF-8 char.  */					\
+									\
+	if (__builtin_expect (outptr + 2 > outend, 0))			\
+	  {								\
+	    /* Overflow in the output buffer.  */			\
+	    result = __GCONV_FULL_OUTPUT;				\
+	    break;							\
+	  }								\
+									\
+        outptr[0] = 0xc0;						\
+        outptr[0] |= c >> 6;						\
+									\
+        outptr[1] = 0x80;						\
+        outptr[1] |= c & 0x3f;						\
+									\
+	outptr += 2;							\
+      }									\
+    else if (c >= 0x0800 && c <= 0xd7ff)				\
+      {									\
+	/* Three byte UTF-8 char.  */					\
+									\
+	if (__builtin_expect (outptr + 3 > outend, 0))			\
+	  {								\
+	    /* Overflow in the output buffer.  */			\
+	    result = __GCONV_FULL_OUTPUT;				\
+	    break;							\
+	  }								\
+	outptr[0] = 0xe0;						\
+	outptr[0] |= c >> 12;						\
+									\
+	outptr[1] = 0x80;						\
+	outptr[1] |= (c >> 6) & 0x3f;					\
+									\
+	outptr[2] = 0x80;						\
+	outptr[2] |= c & 0x3f;						\
+									\
+	outptr += 3;							\
+      }									\
+    else if (c >= 0xd800 && c <= 0xdbff)				\
+      {									\
+        /* Four byte UTF-8 char.  */					\
+	uint16_t low, uvwxy;						\
+									\
+	if (__builtin_expect (outptr + 4 > outend, 0))			\
+	  {								\
+	    /* Overflow in the output buffer.  */			\
+	    result = __GCONV_FULL_OUTPUT;				\
+	    break;							\
+	  }								\
+	inptr += 2;							\
+	if (__builtin_expect (inptr + 2 > inend, 0))			\
+	  {								\
+	    result = __GCONV_INCOMPLETE_INPUT;				\
+	    break;							\
+	  }								\
+									\
+	low = get16 (inptr);						\
+									\
+	if ((low & 0xfc00) != 0xdc00)					\
+	  {								\
+	    inptr -= 2;							\
+	    STANDARD_TO_LOOP_ERR_HANDLER (2);				\
+	  }								\
+	uvwxy = ((c >> 6) & 0xf) + 1;					\
+	outptr[0] = 0xf0;						\
+	outptr[0] |= uvwxy >> 2;					\
+									\
+	outptr[1] = 0x80;						\
+	outptr[1] |= (uvwxy << 4) & 0x30;				\
+	outptr[1] |= (c >> 2) & 0x0f;					\
+									\
+	outptr[2] = 0x80;						\
+	outptr[2] |= (c & 0x03) << 4;					\
+	outptr[2] |= (low >> 6) & 0x0f;					\
+									\
+	outptr[3] = 0x80;						\
+	outptr[3] |= low & 0x3f;					\
+									\
+	outptr += 4;							\
+      }									\
+    else								\
+      {									\
+        STANDARD_TO_LOOP_ERR_HANDLER (2);				\
+      }									\
+    inptr += 2;								\
+  }
+#define LOOP_NEED_FLAGS
+#include <iconv/loop.c>
+
+#include <iconv/skeleton.c>
diff --git a/sysdeps/s390/s390-64/utf8-utf32-z9.c b/sysdeps/s390/s390-64/utf8-utf32-z9.c
new file mode 100644
index 0000000000..17ef8bc890
--- /dev/null
+++ b/sysdeps/s390/s390-64/utf8-utf32-z9.c
@@ -0,0 +1,508 @@
+/* Conversion between UTF-8 and UTF-32 BE/internal.
+
+   This module uses the Z9-109 variants of the Convert Unicode
+   instructions.
+   Copyright (C) 1997-2009 Free Software Foundation, Inc.
+
+   Author: Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>
+   Based on the work by Ulrich Drepper  <drepper@cygnus.com>, 1997.
+
+   Thanks to Daniel Appich who covered the relevant performance work
+   in his diploma thesis.
+
+   This is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   This is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <dlfcn.h>
+#include <stdint.h>
+#include <unistd.h>
+#include <dl-procinfo.h>
+#include <gconv.h>
+
+/* UTF-32 big endian byte order mark.  */
+#define BOM	                0x0000feffu
+
+#define DEFINE_INIT		0
+#define DEFINE_FINI		0
+/* These definitions apply to the UTF-8 to UTF-32 direction.  The
+   software implementation for UTF-8 still supports multibyte
+   characters up to 6 bytes whereas the hardware variant does not.  */
+#define MIN_NEEDED_FROM		1
+#define MAX_NEEDED_FROM		6
+#define MIN_NEEDED_TO		4
+#define FROM_LOOP		from_utf8_loop
+#define TO_LOOP			to_utf8_loop
+#define FROM_DIRECTION		(dir == from_utf8)
+#define PREPARE_LOOP							\
+  enum direction dir = ((struct utf8_data *) step->__data)->dir;	\
+  int emit_bom = ((struct utf8_data *) step->__data)->emit_bom;		\
+									\
+  if (emit_bom && !data->__internal_use					\
+      && data->__invocation_counter == 0)				\
+    {									\
+      /* Emit the Byte Order Mark.  */					\
+      if (__builtin_expect (outbuf + 4 > outend, 0))			\
+	return __GCONV_FULL_OUTPUT;					\
+									\
+      put32u (outbuf, BOM);						\
+      outbuf += 4;							\
+    }
+
+/* Direction of the transformation.  */
+enum direction
+{
+  illegal_dir,
+  to_utf8,
+  from_utf8
+};
+
+struct utf8_data
+{
+  enum direction dir;
+  int emit_bom;
+};
+
+
+extern int gconv_init (struct __gconv_step *step);
+int
+gconv_init (struct __gconv_step *step)
+{
+  /* Determine which direction.  */
+  struct utf8_data *new_data;
+  enum direction dir = illegal_dir;
+  int emit_bom;
+  int result;
+
+  emit_bom = (__strcasecmp (step->__to_name, "UTF-32//") == 0);
+
+  if (__strcasecmp (step->__from_name, "ISO-10646/UTF8/") == 0
+      && (__strcasecmp (step->__to_name, "UTF-32//") == 0
+	  || __strcasecmp (step->__to_name, "UTF-32BE//") == 0
+	  || __strcasecmp (step->__to_name, "INTERNAL") == 0))
+    {
+      dir = from_utf8;
+    }
+  else if (__strcasecmp (step->__to_name, "ISO-10646/UTF8/") == 0
+	   && (__strcasecmp (step->__from_name, "UTF-32BE//") == 0
+	       || __strcasecmp (step->__from_name, "INTERNAL") == 0))
+    {
+      dir = to_utf8;
+    }
+
+  result = __GCONV_NOCONV;
+  if (dir != illegal_dir)
+    {
+      new_data = (struct utf8_data *) malloc (sizeof (struct utf8_data));
+
+      result = __GCONV_NOMEM;
+      if (new_data != NULL)
+	{
+	  new_data->dir = dir;
+	  new_data->emit_bom = emit_bom;
+	  step->__data = new_data;
+
+	  if (dir == from_utf8)
+	    {
+	      step->__min_needed_from = MIN_NEEDED_FROM;
+	      step->__max_needed_from = MIN_NEEDED_FROM;
+	      step->__min_needed_to = MIN_NEEDED_TO;
+	      step->__max_needed_to = MIN_NEEDED_TO;
+	    }
+	  else
+	    {
+	      step->__min_needed_from = MIN_NEEDED_TO;
+	      step->__max_needed_from = MIN_NEEDED_TO;
+	      step->__min_needed_to = MIN_NEEDED_FROM;
+	      step->__max_needed_to = MIN_NEEDED_FROM;
+	    }
+
+	  step->__stateful = 0;
+
+	  result = __GCONV_OK;
+	}
+    }
+
+  return result;
+}
+
+
+extern void gconv_end (struct __gconv_step *data);
+void
+gconv_end (struct __gconv_step *data)
+{
+  free (data->__data);
+}
+
+/* The macro for the hardware loop.  This is used for both
+   directions.  */
+#define HARDWARE_CONVERT(INSTRUCTION)					\
+  {									\
+    register const unsigned char* pInput asm ("8") = inptr;		\
+    register unsigned long long inlen asm ("9") = inend - inptr;	\
+    register unsigned char* pOutput asm ("10") = outptr;		\
+    register unsigned long long outlen asm("11") = outend - outptr;	\
+    uint64_t cc = 0;							\
+									\
+    asm volatile ("0: " INSTRUCTION "  \n\t"				\
+                  "   jo     0b        \n\t"				\
+		  "   ipm    %2        \n"				\
+		  : "+a" (pOutput), "+a" (pInput), "+d" (cc),		\
+		    "+d" (outlen), "+d" (inlen)				\
+		  :							\
+		  : "cc", "memory");					\
+									\
+    inptr = pInput;							\
+    outptr = pOutput;							\
+    cc >>= 28;								\
+									\
+    if (cc == 1)							\
+      {									\
+	result = __GCONV_FULL_OUTPUT;					\
+	break;								\
+      }									\
+    else if (cc == 2)							\
+      {									\
+	result = __GCONV_ILLEGAL_INPUT;					\
+	break;								\
+      }									\
+  }
+
+/* Conversion function from UTF-8 to UTF-32 internal/BE.  */
+
+#define MIN_NEEDED_INPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_INPUT	MAX_NEEDED_FROM
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_TO
+#define LOOPFCT			FROM_LOOP
+/* The software routine is copied from gconv_simple.c.  */
+#define BODY								\
+  {									\
+    if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH)				\
+      {									\
+	HARDWARE_CONVERT ("cu14 %0, %1, 1");				\
+									\
+	if (inptr != inend)						\
+	  {								\
+	    int i;							\
+	    for (i = 1; inptr + i < inend; ++i)				\
+	      if ((inptr[i] & 0xc0) != 0x80)				\
+		break;							\
+									\
+	    if (__builtin_expect (inptr + i == inend, 1))		\
+	      {								\
+		result = __GCONV_INCOMPLETE_INPUT;			\
+		break;							\
+	      }								\
+	    STANDARD_FROM_LOOP_ERR_HANDLER (i);				\
+	  }								\
+	continue;							\
+      }									\
+									\
+    /* Next input byte.  */						\
+    uint32_t ch = *inptr;						\
+									\
+    if (__builtin_expect (ch < 0x80, 1))				\
+      {									\
+	/* One byte sequence.  */					\
+	++inptr;							\
+      }									\
+    else								\
+      {									\
+	uint_fast32_t cnt;						\
+	uint_fast32_t i;						\
+									\
+	if (ch >= 0xc2 && ch < 0xe0)					\
+	  {								\
+	    /* We expect two bytes.  The first byte cannot be 0xc0 or	\
+	       0xc1, otherwise the wide character could have been	\
+	       represented using a single byte.  */			\
+	    cnt = 2;							\
+	    ch &= 0x1f;							\
+	  }								\
+        else if (__builtin_expect ((ch & 0xf0) == 0xe0, 1))		\
+	  {								\
+	    /* We expect three bytes.  */				\
+	    cnt = 3;							\
+	    ch &= 0x0f;							\
+	  }								\
+	else if (__builtin_expect ((ch & 0xf8) == 0xf0, 1))		\
+	  {								\
+	    /* We expect four bytes.  */				\
+	    cnt = 4;							\
+	    ch &= 0x07;							\
+	  }								\
+	else if (__builtin_expect ((ch & 0xfc) == 0xf8, 1))		\
+	  {								\
+	    /* We expect five bytes.  */				\
+	    cnt = 5;							\
+	    ch &= 0x03;							\
+	  }								\
+	else if (__builtin_expect ((ch & 0xfe) == 0xfc, 1))		\
+	  {								\
+	    /* We expect six bytes.  */					\
+	    cnt = 6;							\
+	    ch &= 0x01;							\
+	  }								\
+	else								\
+	  {								\
+	    /* Search the end of this ill-formed UTF-8 character.  This	\
+	       is the next byte with (x & 0xc0) != 0x80.  */		\
+	    i = 0;							\
+	    do								\
+	      ++i;							\
+	    while (inptr + i < inend					\
+		   && (*(inptr + i) & 0xc0) == 0x80			\
+		   && i < 5);						\
+									\
+	  errout:							\
+	    STANDARD_FROM_LOOP_ERR_HANDLER (i);				\
+	  }								\
+									\
+	if (__builtin_expect (inptr + cnt > inend, 0))			\
+	  {								\
+	    /* We don't have enough input.  But before we report	\
+	       that check that all the bytes are correct.  */		\
+	    for (i = 1; inptr + i < inend; ++i)				\
+	      if ((inptr[i] & 0xc0) != 0x80)				\
+		break;							\
+									\
+	    if (__builtin_expect (inptr + i == inend, 1))		\
+	      {								\
+		result = __GCONV_INCOMPLETE_INPUT;			\
+		break;							\
+	      }								\
+									\
+	    goto errout;						\
+	  }								\
+									\
+	/* Read the possible remaining bytes.  */			\
+	for (i = 1; i < cnt; ++i)					\
+	  {								\
+	    uint32_t byte = inptr[i];					\
+									\
+	    if ((byte & 0xc0) != 0x80)					\
+	      /* This is an illegal encoding.  */			\
+	      break;							\
+									\
+	    ch <<= 6;							\
+	    ch |= byte & 0x3f;						\
+	  }								\
+									\
+	/* If i < cnt, some trail byte was not >= 0x80, < 0xc0.		\
+	   If cnt > 2 and ch < 2^(5*cnt-4), the wide character ch could	\
+	   have been represented with fewer than cnt bytes.  */		\
+	if (i < cnt || (cnt > 2 && (ch >> (5 * cnt - 4)) == 0))		\
+	  {								\
+	    /* This is an illegal encoding.  */				\
+	    goto errout;						\
+	  }								\
+									\
+	inptr += cnt;							\
+      }									\
+									\
+    /* Now adjust the pointers and store the result.  */		\
+    *((uint32_t *) outptr) = ch;					\
+    outptr += sizeof (uint32_t);					\
+  }
+#define LOOP_NEED_FLAGS
+
+#define STORE_REST							\
+  {									      \
+    /* We store the remaining bytes while converting them into the UCS4	      \
+       format.  We can assume that the first byte in the buffer is	      \
+       correct and that it requires a larger number of bytes than there	      \
+       are in the input buffer.  */					      \
+    wint_t ch = **inptrp;						      \
+    size_t cnt, r;							      \
+									      \
+    state->__count = inend - *inptrp;					      \
+									      \
+    if (ch >= 0xc2 && ch < 0xe0)					      \
+      {									      \
+	/* We expect two bytes.  The first byte cannot be 0xc0 or	      \
+	   0xc1, otherwise the wide character could have been		      \
+	   represented using a single byte.  */				      \
+	cnt = 2;							      \
+	ch &= 0x1f;							      \
+      }									      \
+    else if (__builtin_expect ((ch & 0xf0) == 0xe0, 1))			      \
+      {									      \
+	/* We expect three bytes.  */					      \
+	cnt = 3;							      \
+	ch &= 0x0f;							      \
+      }									      \
+    else if (__builtin_expect ((ch & 0xf8) == 0xf0, 1))			      \
+      {									      \
+	/* We expect four bytes.  */					      \
+	cnt = 4;							      \
+	ch &= 0x07;							      \
+      }									      \
+    else if (__builtin_expect ((ch & 0xfc) == 0xf8, 1))			      \
+      {									      \
+	/* We expect five bytes.  */					      \
+	cnt = 5;							      \
+	ch &= 0x03;							      \
+      }									      \
+    else								      \
+      {									      \
+	/* We expect six bytes.  */					      \
+	cnt = 6;							      \
+	ch &= 0x01;							      \
+      }									      \
+									      \
+    /* The first byte is already consumed.  */				      \
+    r = cnt - 1;							      \
+    while (++(*inptrp) < inend)						      \
+      {									      \
+	ch <<= 6;							      \
+	ch |= **inptrp & 0x3f;						      \
+	--r;								      \
+      }									      \
+									      \
+    /* Shift for the so far missing bytes.  */				      \
+    ch <<= r * 6;							      \
+									      \
+    /* Store the number of bytes expected for the entire sequence.  */	      \
+    state->__count |= cnt << 8;						      \
+									      \
+    /* Store the value.  */						      \
+    state->__value.__wch = ch;						      \
+  }
+
+#define UNPACK_BYTES \
+  {									      \
+    static const unsigned char inmask[5] = { 0xc0, 0xe0, 0xf0, 0xf8, 0xfc };  \
+    wint_t wch = state->__value.__wch;					      \
+    size_t ntotal = state->__count >> 8;				      \
+									      \
+    inlen = state->__count & 255;					      \
+									      \
+    bytebuf[0] = inmask[ntotal - 2];					      \
+									      \
+    do									      \
+      {									      \
+	if (--ntotal < inlen)						      \
+	  bytebuf[ntotal] = 0x80 | (wch & 0x3f);			      \
+	wch >>= 6;							      \
+      }									      \
+    while (ntotal > 1);							      \
+									      \
+    bytebuf[0] |= wch;							      \
+  }
+
+#define CLEAR_STATE \
+  state->__count = 0
+
+#include <iconv/loop.c>
+
+/* Conversion from UTF-32 internal/BE to UTF-8.  */
+
+#define MIN_NEEDED_INPUT	MIN_NEEDED_TO
+#define MIN_NEEDED_OUTPUT	MIN_NEEDED_FROM
+#define MAX_NEEDED_OUTPUT	MAX_NEEDED_FROM
+#define LOOPFCT			TO_LOOP
+/* The software routine mimics the S/390 cu41 instruction.  */
+#define BODY							\
+  {								\
+    if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH)			\
+      {								\
+	HARDWARE_CONVERT ("cu41 %0, %1");			\
+								\
+	if (inptr != inend)					\
+	  {							\
+	    result = __GCONV_INCOMPLETE_INPUT;			\
+	    break;						\
+	  }							\
+	continue;						\
+      }								\
+								\
+    uint32_t wc = *((const uint32_t *) inptr);			\
+								\
+    if (__builtin_expect (wc <= 0x7f, 1))			\
+      {								\
+        /* Single UTF-8 char.  */				\
+        *outptr = (uint8_t)wc;					\
+	outptr++;						\
+      }								\
+    else if (wc <= 0x7ff)					\
+      {								\
+        /* Two UTF-8 chars.  */					\
+        if (__builtin_expect (outptr + 2 > outend, 0))		\
+	  {							\
+	    /* Overflow in the output buffer.  */		\
+	    result = __GCONV_FULL_OUTPUT;			\
+	    break;						\
+	  }							\
+								\
+        outptr[0] = 0xc0;					\
+	outptr[0] |= wc >> 6;					\
+								\
+	outptr[1] = 0x80;					\
+	outptr[1] |= wc & 0x3f;					\
+								\
+	outptr += 2;						\
+      }								\
+    else if (wc <= 0xffff)					\
+      {								\
+	/* Three UTF-8 chars.  */				\
+	if (__builtin_expect (outptr + 3 > outend, 0))		\
+	  {							\
+	    /* Overflow in the output buffer.  */		\
+	    result = __GCONV_FULL_OUTPUT;			\
+	    break;						\
+	  }							\
+	outptr[0] = 0xe0;					\
+	outptr[0] |= wc >> 12;					\
+								\
+	outptr[1] = 0x80;					\
+	outptr[1] |= (wc >> 6) & 0x3f;				\
+								\
+	outptr[2] = 0x80;					\
+	outptr[2] |= wc & 0x3f;					\
+								\
+	outptr += 3;						\
+      }								\
+      else if (wc <= 0x10ffff)					\
+	{							\
+	  /* Four UTF-8 chars.  */				\
+	  if (__builtin_expect (outptr + 4 > outend, 0))	\
+	    {							\
+	      /* Overflow in the output buffer.  */		\
+	      result = __GCONV_FULL_OUTPUT;			\
+	      break;						\
+	    }							\
+	  outptr[0] = 0xf0;					\
+	  outptr[0] |= wc >> 18;				\
+								\
+	  outptr[1] = 0x80;					\
+	  outptr[1] |= (wc >> 12) & 0x3f;			\
+								\
+	  outptr[2] = 0x80;					\
+	  outptr[2] |= (wc >> 6) & 0x3f;			\
+								\
+	  outptr[3] = 0x80;					\
+	  outptr[3] |= wc & 0x3f;				\
+								\
+	  outptr += 4;						\
+	}							\
+      else							\
+	{							\
+	  STANDARD_TO_LOOP_ERR_HANDLER (4);			\
+	}							\
+    inptr += 4;							\
+  }
+#define LOOP_NEED_FLAGS
+#include <iconv/loop.c>
+
+#include <iconv/skeleton.c>
diff --git a/sysdeps/unix/sysv/linux/configure b/sysdeps/unix/sysv/linux/configure
index 253e9c57ff..199457a3ac 100644
--- a/sysdeps/unix/sysv/linux/configure
+++ b/sysdeps/unix/sysv/linux/configure
@@ -1,17 +1,6 @@
 # This file is generated from configure.in by Autoconf.  DO NOT EDIT!
  # Local configure fragment for sysdeps/unix/sysv/linux.
 
-# The Linux kernel headers can be found in
-#   /lib/modules/$(uname -r)/build/include
-# Check whether this directory is available.
-if test -z "$sysheaders" &&
-   test "x$cross_compiling" = xno &&
-   test -d /lib/modules/`uname -r`/build/include; then
-  sysheaders="/lib/modules/`uname -r`/build/include"
-  ccheaders=`$CC -print-file-name=include`
-      SYSINCLUDES="-I $sysheaders"
-fi
-
 # Don't bother trying to generate any glue code to be compatible with the
 # existing system library, because we are the only system library.
 inhibit_glue=yes
diff --git a/sysdeps/unix/sysv/linux/configure.in b/sysdeps/unix/sysv/linux/configure.in
index 5330e98c2d..8f00407a8b 100644
--- a/sysdeps/unix/sysv/linux/configure.in
+++ b/sysdeps/unix/sysv/linux/configure.in
@@ -1,19 +1,6 @@
 GLIBC_PROVIDES dnl See aclocal.m4 in the top level source directory.
 # Local configure fragment for sysdeps/unix/sysv/linux.
 
-# The Linux kernel headers can be found in
-#   /lib/modules/$(uname -r)/build/include
-# Check whether this directory is available.
-if test -z "$sysheaders" &&
-   test "x$cross_compiling" = xno &&
-   test -d /lib/modules/`uname -r`/build/include; then
-  sysheaders="/lib/modules/`uname -r`/build/include"
-  ccheaders=`$CC -print-file-name=include`
-  dnl We don't have to use -nostdinc.  We just want one more directory
-  dnl to be used.
-  SYSINCLUDES="-I $sysheaders"
-fi
-
 # Don't bother trying to generate any glue code to be compatible with the
 # existing system library, because we are the only system library.
 inhibit_glue=yes
diff --git a/sysdeps/unix/sysv/linux/eventfd.c b/sysdeps/unix/sysv/linux/eventfd.c
index 4cd557983e..7f69ecdb8c 100644
--- a/sysdeps/unix/sysv/linux/eventfd.c
+++ b/sysdeps/unix/sysv/linux/eventfd.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2007, 2008 Free Software Foundation, Inc.
+/* Copyright (C) 2007, 2008, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -19,14 +19,21 @@
 #include <errno.h>
 #include <sys/eventfd.h>
 #include <sysdep.h>
+#include <kernel-features.h>
 
 
 int
 eventfd (int count, int flags)
 {
 #ifdef __NR_eventfd2
-  return INLINE_SYSCALL (eventfd2, 2, count, flags);
-#else
+  int res = INLINE_SYSCALL (eventfd2, 2, count, flags);
+# ifndef __ASSUME_EVENTFD2
+  if (res != -1 || errno != ENOSYS)
+# endif
+    return res;
+#endif
+
+#ifndef __ASSUME_EVENTFD2
   /* The old system call has no flag parameter which is bad.  So we have
      to wait until we have to support to pass additional values to the
      kernel (sys_indirect) before implementing setting flags like
@@ -43,5 +50,7 @@ eventfd (int count, int flags)
   __set_errno (ENOSYS);
   return -1;
 # endif
+#elif !defined __NR_eventfd2
+# error "__ASSUME_EVENTFD2 defined but not __NR_eventfd2"
 #endif
 }
diff --git a/sysdeps/unix/sysv/linux/i386/sysconf.c b/sysdeps/unix/sysv/linux/i386/sysconf.c
index efe1a639cd..ff3cf9f7c7 100644
--- a/sysdeps/unix/sysv/linux/i386/sysconf.c
+++ b/sysdeps/unix/sysv/linux/i386/sysconf.c
@@ -138,6 +138,9 @@ static const struct intel_02_cache_info
     { 0xe3, 16, 64, M(_SC_LEVEL3_CACHE_SIZE),  2097152 },
     { 0xe3, 16, 64, M(_SC_LEVEL3_CACHE_SIZE),  4194304 },
     { 0xe4, 16, 64, M(_SC_LEVEL3_CACHE_SIZE),  8388608 },
+    { 0xea, 24, 64, M(_SC_LEVEL3_CACHE_SIZE), 12582912 },
+    { 0xeb, 24, 64, M(_SC_LEVEL3_CACHE_SIZE), 18874368 },
+    { 0xec, 24, 64, M(_SC_LEVEL3_CACHE_SIZE), 25165824 },
   };
 #define nintel_02_known (sizeof (intel_02_known) / sizeof (intel_02_known[0]))
 
diff --git a/sysdeps/unix/sysv/linux/kernel-features.h b/sysdeps/unix/sysv/linux/kernel-features.h
index 4562515790..ff065effb5 100644
--- a/sysdeps/unix/sysv/linux/kernel-features.h
+++ b/sysdeps/unix/sysv/linux/kernel-features.h
@@ -516,6 +516,8 @@
 # define __ASSUME_SOCK_CLOEXEC	1
 # define __ASSUME_IN_NONBLOCK	1
 # define __ASSUME_PIPE2		1
+# define __ASSUME_EVENTFD2	1
+# define __ASSUME_SIGNALFD4	1
 #endif
 
 /* Support for the accept4 syscall was added in 2.6.28.  */
diff --git a/sysdeps/unix/sysv/linux/signalfd.c b/sysdeps/unix/sysv/linux/signalfd.c
index 9898f29231..c2d974a45d 100644
--- a/sysdeps/unix/sysv/linux/signalfd.c
+++ b/sysdeps/unix/sysv/linux/signalfd.c
@@ -1,4 +1,4 @@
-/* Copyright (C) 2007, 2008 Free Software Foundation, Inc.
+/* Copyright (C) 2007, 2008, 2009 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -20,14 +20,21 @@
 #include <signal.h>
 #include <sys/signalfd.h>
 #include <sysdep.h>
+#include <kernel-features.h>
 
 
 int
 signalfd (int fd, const sigset_t *mask, int flags)
 {
 #ifdef __NR_signalfd4
-  return INLINE_SYSCALL (signalfd4, 4, fd, mask, _NSIG / 8, flags);
-#else
+  int res = INLINE_SYSCALL (signalfd4, 4, fd, mask, _NSIG / 8, flags);
+# ifndef __ASSUME_SIGNALFD4
+  if (res != -1 || errno != ENOSYS)
+# endif
+    return res;
+#endif
+
+#ifndef __ASSUME_SIGNALFD4
   /* The old system call has no flag parameter which is bad.  So we have
      to wait until we have to support to pass additional values to the
      kernel (sys_indirect) before implementing setting flags like
@@ -44,5 +51,7 @@ signalfd (int fd, const sigset_t *mask, int flags)
   __set_errno (ENOSYS);
   return -1;
 # endif
+#elif !defined __NR_signalfd4
+# error "__ASSUME_SIGNALFD4 defined but not __NR_signalfd4"
 #endif
 }
diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile
index 78fdb04fcb..57cd88432a 100644
--- a/sysdeps/x86_64/Makefile
+++ b/sysdeps/x86_64/Makefile
@@ -19,6 +19,10 @@ ifeq ($(subdir),elf)
 sysdep-dl-routines += tlsdesc dl-tlsdesc
 sysdep_routines += tlsdesc dl-tlsdesc
 sysdep-rtld-routines += tlsdesc dl-tlsdesc
+
+tests: $(objpfx)tst-xmmymm.out
+$(objpfx)tst-xmmymm.out: ../sysdeps/x86_64/tst-xmmymm.sh $(objpfx)ld.so
+	$(SHELL) -e $< $(objpfx) > $@
 endif
 
 ifeq ($(subdir),csu)
diff --git a/sysdeps/x86_64/cacheinfo.c b/sysdeps/x86_64/cacheinfo.c
index 362687c181..75b81958dd 100644
--- a/sysdeps/x86_64/cacheinfo.c
+++ b/sysdeps/x86_64/cacheinfo.c
@@ -25,6 +25,17 @@
 
 #ifdef USE_MULTIARCH
 # include "multiarch/init-arch.h"
+
+# define is_intel __cpu_features.kind == arch_kind_intel
+# define is_amd __cpu_features.kind == arch_kind_amd
+# define max_cpuid __cpu_features.max_cpuid
+#else
+  /* This spells out "GenuineIntel".  */
+# define is_intel \
+  ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69
+  /* This spells out "AuthenticAMD".  */
+# define is_amd \
+  ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65
 #endif
 
 static const struct intel_02_cache_info
@@ -100,6 +111,9 @@ static const struct intel_02_cache_info
     { 0xe3, 16, 64, M(_SC_LEVEL3_CACHE_SIZE),  2097152 },
     { 0xe3, 16, 64, M(_SC_LEVEL3_CACHE_SIZE),  4194304 },
     { 0xe4, 16, 64, M(_SC_LEVEL3_CACHE_SIZE),  8388608 },
+    { 0xea, 24, 64, M(_SC_LEVEL3_CACHE_SIZE), 12582912 },
+    { 0xeb, 24, 64, M(_SC_LEVEL3_CACHE_SIZE), 18874368 },
+    { 0xec, 24, 64, M(_SC_LEVEL3_CACHE_SIZE), 25165824 },
   };
 
 #define nintel_02_known (sizeof (intel_02_known) / sizeof (intel_02_known [0]))
@@ -152,6 +166,12 @@ intel_check_word (int name, unsigned int value, bool *has_level_2,
 	      /* Intel reused this value.  For family 15, model 6 it
 		 specifies the 3rd level cache.  Otherwise the 2nd
 		 level cache.  */
+	      unsigned int family;
+	      unsigned int model;
+#ifdef USE_MULTIARCH
+	      family = __cpu_features.family;
+	      model = __cpu_features.model;
+#else
 	      unsigned int eax;
 	      unsigned int ebx;
 	      unsigned int ecx;
@@ -160,9 +180,10 @@ intel_check_word (int name, unsigned int value, bool *has_level_2,
 			    : "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
 			    : "0" (1));
 
-	      unsigned int family = ((eax >> 20) & 0xff) + ((eax >> 8) & 0xf);
-	      unsigned int model = ((((eax >>16) & 0xf) << 4)
-				    + ((eax >> 4) & 0xf));
+	      family = ((eax >> 20) & 0xff) + ((eax >> 8) & 0xf);
+	      model = (((eax >>16) & 0xf) << 4) + ((eax >> 4) & 0xf);
+#endif
+
 	      if (family == 15 && model == 6)
 		{
 		  /* The level 3 cache is encoded for this model like
@@ -394,21 +415,24 @@ long int
 attribute_hidden
 __cache_sysconf (int name)
 {
+#ifdef USE_MULTIARCH
+  if (__cpu_features.kind == arch_kind_unknown)
+    __init_cpu_features ();
+#else
   /* Find out what brand of processor.  */
-  unsigned int eax;
+  unsigned int max_cpuid;
   unsigned int ebx;
   unsigned int ecx;
   unsigned int edx;
   asm volatile ("cpuid"
-		: "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
+		: "=a" (max_cpuid), "=b" (ebx), "=c" (ecx), "=d" (edx)
 		: "0" (0));
+#endif
 
-  /* This spells out "GenuineIntel".  */
-  if (ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69)
-    return handle_intel (name, eax);
+  if (is_intel)
+    return handle_intel (name, max_cpuid);
 
-  /* This spells out "AuthenticAMD".  */
-  if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
+  if (is_amd)
     return handle_amd (name);
 
   // XXX Fill in more vendors.
@@ -457,20 +481,11 @@ init_cacheinfo (void)
 #ifdef USE_MULTIARCH
   if (__cpu_features.kind == arch_kind_unknown)
     __init_cpu_features ();
-# define is_intel __cpu_features.kind == arch_kind_intel
-# define is_amd __cpu_features.kind == arch_kind_amd
-# define max_cpuid __cpu_features.max_cpuid
 #else
   int max_cpuid;
   asm volatile ("cpuid"
 		: "=a" (max_cpuid), "=b" (ebx), "=c" (ecx), "=d" (edx)
 		: "0" (0));
-  /* This spells out "GenuineIntel".  */
-# define is_intel \
-  ebx == 0x756e6547 && ecx == 0x6c65746e && edx == 0x49656e69
-  /* This spells out "AuthenticAMD".  */
-# define is_amd \
-  ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65
 #endif
 
   if (is_intel)
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index 5ce14aad8d..b066402204 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -4,7 +4,7 @@ gen-as-const-headers += ifunc-defines.sym
 endif
 
 ifeq ($(subdir),string)
-sysdep_routines += stpncpy-c strncpy-c strncmp-c
+sysdep_routines += stpncpy-c strncpy-c
 ifeq (yes,$(config-cflags-sse4))
 sysdep_routines += strcspn-c strpbrk-c strspn-c strstr-c strcasestr-c
 CFLAGS-strcspn-c.c += -msse4
diff --git a/sysdeps/x86_64/multiarch/init-arch.c b/sysdeps/x86_64/multiarch/init-arch.c
index 29e687344d..35fd19af0e 100644
--- a/sysdeps/x86_64/multiarch/init-arch.c
+++ b/sysdeps/x86_64/multiarch/init-arch.c
@@ -68,7 +68,13 @@ __init_cpu_features (void)
 	  __cpu_features.model += extended_model;
 	}
       else if (__cpu_features.family == 0x06)
-	__cpu_features.model += extended_model;
+	{
+	  __cpu_features.model += extended_model;
+
+	  if (__cpu_features.model == 0x1c)
+	    /* Avoid SSSE3 on Atom since it is slow.  */
+	    __cpu_features.cpuid[COMMON_CPUID_INDEX_1].ecx &= ~(1 << 9);
+	}
     }
   /* This spells out "AuthenticAMD".  */
   else if (ebx == 0x68747541 && ecx == 0x444d4163 && edx == 0x69746e65)
diff --git a/sysdeps/x86_64/multiarch/rtld-rawmemchr.c b/sysdeps/x86_64/multiarch/rtld-rawmemchr.c
new file mode 100644
index 0000000000..53a90675ab
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/rtld-rawmemchr.c
@@ -0,0 +1 @@
+#include "../rtld-rawmemchr.c"
diff --git a/sysdeps/x86_64/multiarch/rtld-strlen.S b/sysdeps/x86_64/multiarch/rtld-strlen.S
new file mode 100644
index 0000000000..596e0549ea
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/rtld-strlen.S
@@ -0,0 +1 @@
+#include "../rtld-strlen.S"
diff --git a/sysdeps/x86_64/multiarch/strcmp.S b/sysdeps/x86_64/multiarch/strcmp.S
index 37985036aa..1a315737af 100644
--- a/sysdeps/x86_64/multiarch/strcmp.S
+++ b/sysdeps/x86_64/multiarch/strcmp.S
@@ -28,9 +28,9 @@
 	/* calculate left number to compare */		\
 	lea	-16(%rcx, %r11), %r9;			\
 	cmp	%r9, %r11;				\
-	jb	LABEL(strcmp_exitz);			\
+	jb	LABEL(strcmp_exitz_sse4_2);		\
 	test	%r9, %r9;				\
-	je	LABEL(strcmp_exitz);			\
+	je	LABEL(strcmp_exitz_sse4_2);		\
 	mov	%r9, %r11
 
 #define STRCMP_SSE42	__strncmp_sse42
@@ -106,9 +106,9 @@ STRCMP_SSE42:
  */
 #ifdef USE_AS_STRNCMP
 	test	%rdx, %rdx
-	je	LABEL(strcmp_exitz)
+	je	LABEL(strcmp_exitz_sse4_2)
 	cmp	$1, %rdx
-	je	LABEL(Byte0)
+	je	LABEL(Byte0_sse4_2)
 	mov	%rdx, %r11
 #endif
 	mov	%esi, %ecx
@@ -117,23 +117,21 @@ STRCMP_SSE42:
 	and	$0x3f, %rcx		/* rsi alignment in cache line */
 	and	$0x3f, %rax		/* rdi alignment in cache line */
 	cmp	$0x30, %ecx
-	ja	LABEL(crosscache)	/* rsi: 16-byte load will cross cache line */
+	ja	LABEL(crosscache_sse4_2)/* rsi: 16-byte load will cross cache line */
 	cmp	$0x30, %eax
-	ja	LABEL(crosscache)	/* rdi: 16-byte load will cross cache line */
-	movlpd	(%rdi), %xmm1
-	movlpd	(%rsi), %xmm2
-	movhpd	8(%rdi), %xmm1
-	movhpd	8(%rsi), %xmm2
+	ja	LABEL(crosscache_sse4_2)/* rdi: 16-byte load will cross cache line */
+	movdqu	(%rdi), %xmm1
+	movdqu	(%rsi), %xmm2
 	pxor	%xmm0, %xmm0		/* clear %xmm0 for null char checks */
 	pcmpeqb	%xmm1, %xmm0		/* Any null chars? */
 	pcmpeqb	%xmm2, %xmm1		/* compare first 16 bytes for equality */
 	psubb	%xmm0, %xmm1		/* packed sub of comparison results*/
 	pmovmskb %xmm1, %edx
 	sub	$0xffff, %edx		/* if first 16 bytes are same, edx == 0xffff */
-	jnz	LABEL(less16bytes)	/* If not, find different value or null char */
+	jnz	LABEL(less16bytes_sse4_2)/* If not, find different value or null char */
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)	/* finish comparision */
+	jbe	LABEL(strcmp_exitz_sse4_2)/* finish comparision */
 #endif
 	add	$16, %rsi		/* prepare to search next 16 bytes */
 	add	$16, %rdi		/* prepare to search next 16 bytes */
@@ -144,7 +142,7 @@ STRCMP_SSE42:
 	 * below to use.
 	 */
 	.p2align 4
-LABEL(crosscache):
+LABEL(crosscache_sse4_2):
 	and	$0xfffffffffffffff0, %rsi	/* force %rsi is 16 byte aligned */
 	and	$0xfffffffffffffff0, %rdi	/* force %rdi is 16 byte aligned */
 	mov	$0xffff, %edx			/* for equivalent offset */
@@ -152,15 +150,15 @@ LABEL(crosscache):
 	and	$0xf, %ecx			/* offset of rsi */
 	and	$0xf, %eax			/* offset of rdi */
 	cmp	%eax, %ecx
-	je	LABEL(ashr_0)			/* rsi and rdi relative offset same */
-	ja	LABEL(bigger)
+	je	LABEL(ashr_0_sse4_2)		/* rsi and rdi relative offset same */
+	ja	LABEL(bigger_sse4_2)
 	mov	%edx, %r8d			/* r8d is offset flag for exit tail */
 	xchg	%ecx, %eax
 	xchg	%rsi, %rdi
-LABEL(bigger):
+LABEL(bigger_sse4_2):
 	lea	15(%rax), %r9
 	sub	%rcx, %r9
-	lea	LABEL(unaligned_table)(%rip), %r10
+	lea	LABEL(unaligned_table_sse4_2)(%rip), %r10
 	movslq	(%r10, %r9,4), %r9
 	lea	(%r10, %r9), %r10
 	jmp	*%r10				/* jump to corresponding case */
@@ -171,7 +169,7 @@ LABEL(bigger):
  *        n(0~15)            n(0~15)           15(15+ n-n)         ashr_0
  */
 	.p2align 4
-LABEL(ashr_0):
+LABEL(ashr_0_sse4_2):
 
 	movdqa	(%rsi), %xmm1
 	pxor	%xmm0, %xmm0			/* clear %xmm0 for null char check */
@@ -186,7 +184,7 @@ LABEL(ashr_0):
 	 * edx must be the same with r9d if in left byte (16-rcx) is equal to
 	 * the start from (16-rax) and no null char was seen.
 	 */
-	jne	LABEL(less32bytes)		/* mismatch or null char */
+	jne	LABEL(less32bytes_sse4_2)	/* mismatch or null char */
 	UPDATE_STRNCMP_COUNTER
 	mov	$16, %rcx
 	mov	$16, %r9
@@ -205,7 +203,7 @@ LABEL(ashr_0_use_sse4_2):
 	jbe	LABEL(ashr_0_use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	movdqa	(%rdi,%rdx), %xmm0
@@ -214,17 +212,17 @@ LABEL(ashr_0_use_sse4_2):
 	jbe	LABEL(ashr_0_use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	jmp	LABEL(ashr_0_use_sse4_2)
 
 
 	.p2align 4
 LABEL(ashr_0_use_sse4_2_exit):
-	jnc	LABEL(strcmp_exitz)
+	jnc	LABEL(strcmp_exitz_sse4_2)
 #ifdef USE_AS_STRNCMP
 	sub	%rcx, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	lea	-16(%rdx, %rcx), %rcx
 	movzbl	(%rdi, %rcx), %eax
@@ -241,7 +239,7 @@ LABEL(ashr_0_use_sse4_2_exit):
  *        n(15)            n -15            0(15 +(n-15) - n)         ashr_1
  */
 	.p2align 4
-LABEL(ashr_1):
+LABEL(ashr_1_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -253,7 +251,7 @@ LABEL(ashr_1):
 	shr	%cl, %edx		/* adjust 0xffff for offset */
 	shr	%cl, %r9d		/* adjust for 16-byte offset */
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)	/* mismatch or null char seen */
+	jnz	LABEL(less32bytes_sse4_2)/* mismatch or null char seen */
 	movdqa	(%rdi), %xmm3
 	UPDATE_STRNCMP_COUNTER
 
@@ -281,7 +279,7 @@ LABEL(loop_ashr_1_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -294,7 +292,7 @@ LABEL(loop_ashr_1_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_1_use_sse4_2)
@@ -320,7 +318,7 @@ LABEL(nibble_ashr_1_use_sse4_2):
  *        n(14~15)            n -14         1(15 +(n-14) - n)         ashr_2
  */
 	.p2align 4
-LABEL(ashr_2):
+LABEL(ashr_2_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -332,7 +330,7 @@ LABEL(ashr_2):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 	UPDATE_STRNCMP_COUNTER
 
@@ -360,7 +358,7 @@ LABEL(loop_ashr_2_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -373,7 +371,7 @@ LABEL(loop_ashr_2_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_2_use_sse4_2)
@@ -399,7 +397,7 @@ LABEL(nibble_ashr_2_use_sse4_2):
  *        n(13~15)            n -13         2(15 +(n-13) - n)         ashr_3
  */
 	.p2align 4
-LABEL(ashr_3):
+LABEL(ashr_3_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -411,7 +409,7 @@ LABEL(ashr_3):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -439,7 +437,7 @@ LABEL(loop_ashr_3_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -452,7 +450,7 @@ LABEL(loop_ashr_3_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_3_use_sse4_2)
@@ -478,7 +476,7 @@ LABEL(nibble_ashr_3_use_sse4_2):
  *        n(12~15)            n -12         3(15 +(n-12) - n)         ashr_4
  */
 	.p2align 4
-LABEL(ashr_4):
+LABEL(ashr_4_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -490,7 +488,7 @@ LABEL(ashr_4):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -519,7 +517,7 @@ LABEL(loop_ashr_4_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -532,7 +530,7 @@ LABEL(loop_ashr_4_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_4_use_sse4_2)
@@ -558,7 +556,7 @@ LABEL(nibble_ashr_4_use_sse4_2):
  *        n(11~15)          n - 11      	  4(15 +(n-11) - n)         ashr_5
  */
 	.p2align 4
-LABEL(ashr_5):
+LABEL(ashr_5_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -570,7 +568,7 @@ LABEL(ashr_5):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -599,7 +597,7 @@ LABEL(loop_ashr_5_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -613,7 +611,7 @@ LABEL(loop_ashr_5_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_5_use_sse4_2)
@@ -639,7 +637,7 @@ LABEL(nibble_ashr_5_use_sse4_2):
  *        n(10~15)          n - 10      	  5(15 +(n-10) - n)         ashr_6
  */
 	.p2align 4
-LABEL(ashr_6):
+LABEL(ashr_6_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -651,7 +649,7 @@ LABEL(ashr_6):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -680,7 +678,7 @@ LABEL(loop_ashr_6_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -693,7 +691,7 @@ LABEL(loop_ashr_6_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_6_use_sse4_2)
@@ -719,7 +717,7 @@ LABEL(nibble_ashr_6_use_sse4_2):
  *        n(9~15)          n - 9      	        6(15 +(n - 9) - n)         ashr_7
  */
 	.p2align 4
-LABEL(ashr_7):
+LABEL(ashr_7_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -731,7 +729,7 @@ LABEL(ashr_7):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -760,7 +758,7 @@ LABEL(loop_ashr_7_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -773,7 +771,7 @@ LABEL(loop_ashr_7_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_7_use_sse4_2)
@@ -799,7 +797,7 @@ LABEL(nibble_ashr_7_use_sse4_2):
  *        n(8~15)          n - 8      	        7(15 +(n - 8) - n)         ashr_8
  */
 	.p2align 4
-LABEL(ashr_8):
+LABEL(ashr_8_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -811,7 +809,7 @@ LABEL(ashr_8):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -840,7 +838,7 @@ LABEL(loop_ashr_8_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -853,7 +851,7 @@ LABEL(loop_ashr_8_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_8_use_sse4_2)
@@ -879,7 +877,7 @@ LABEL(nibble_ashr_8_use_sse4_2):
  *        n(7~15)          n - 7      	        8(15 +(n - 7) - n)         ashr_9
  */
 	.p2align 4
-LABEL(ashr_9):
+LABEL(ashr_9_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -891,7 +889,7 @@ LABEL(ashr_9):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -921,7 +919,7 @@ LABEL(loop_ashr_9_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -934,7 +932,7 @@ LABEL(loop_ashr_9_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_9_use_sse4_2)
@@ -960,7 +958,7 @@ LABEL(nibble_ashr_9_use_sse4_2):
  *        n(6~15)          n - 6      	        9(15 +(n - 6) - n)         ashr_10
  */
 	.p2align 4
-LABEL(ashr_10):
+LABEL(ashr_10_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -972,7 +970,7 @@ LABEL(ashr_10):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1001,7 +999,7 @@ LABEL(loop_ashr_10_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1014,7 +1012,7 @@ LABEL(loop_ashr_10_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_10_use_sse4_2)
@@ -1040,7 +1038,7 @@ LABEL(nibble_ashr_10_use_sse4_2):
  *        n(5~15)          n - 5      	        10(15 +(n - 5) - n)         ashr_11
  */
 	.p2align 4
-LABEL(ashr_11):
+LABEL(ashr_11_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1052,7 +1050,7 @@ LABEL(ashr_11):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1081,7 +1079,7 @@ LABEL(loop_ashr_11_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1094,7 +1092,7 @@ LABEL(loop_ashr_11_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_11_use_sse4_2)
@@ -1120,7 +1118,7 @@ LABEL(nibble_ashr_11_use_sse4_2):
  *        n(4~15)          n - 4      	        11(15 +(n - 4) - n)         ashr_12
  */
 	.p2align 4
-LABEL(ashr_12):
+LABEL(ashr_12_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1132,7 +1130,7 @@ LABEL(ashr_12):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1161,7 +1159,7 @@ LABEL(loop_ashr_12_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1174,7 +1172,7 @@ LABEL(loop_ashr_12_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_12_use_sse4_2)
@@ -1200,7 +1198,7 @@ LABEL(nibble_ashr_12_use_sse4_2):
  *        n(3~15)          n - 3      	        12(15 +(n - 3) - n)         ashr_13
  */
 	.p2align 4
-LABEL(ashr_13):
+LABEL(ashr_13_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1212,7 +1210,7 @@ LABEL(ashr_13):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1242,7 +1240,7 @@ LABEL(loop_ashr_13_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1255,7 +1253,7 @@ LABEL(loop_ashr_13_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_13_use_sse4_2)
@@ -1281,7 +1279,7 @@ LABEL(nibble_ashr_13_use_sse4_2):
  *        n(2~15)          n - 2      	        13(15 +(n - 2) - n)         ashr_14
  */
 	.p2align 4
-LABEL(ashr_14):
+LABEL(ashr_14_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1293,7 +1291,7 @@ LABEL(ashr_14):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 	movdqa	(%rdi), %xmm3
 
 	UPDATE_STRNCMP_COUNTER
@@ -1323,7 +1321,7 @@ LABEL(loop_ashr_14_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1336,7 +1334,7 @@ LABEL(loop_ashr_14_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_14_use_sse4_2)
@@ -1362,7 +1360,7 @@ LABEL(nibble_ashr_14_use_sse4_2):
  *        n(1~15)          n - 1      	        14(15 +(n - 1) - n)         ashr_15
  */
 	.p2align 4
-LABEL(ashr_15):
+LABEL(ashr_15_sse4_2):
 	pxor	%xmm0, %xmm0
 	movdqa	(%rdi), %xmm2
 	movdqa	(%rsi), %xmm1
@@ -1374,7 +1372,7 @@ LABEL(ashr_15):
 	shr	%cl, %edx
 	shr	%cl, %r9d
 	sub	%r9d, %edx
-	jnz	LABEL(less32bytes)
+	jnz	LABEL(less32bytes_sse4_2)
 
 	movdqa	(%rdi), %xmm3
 
@@ -1406,7 +1404,7 @@ LABEL(loop_ashr_15_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 
 	add	$16, %rdx
@@ -1419,7 +1417,7 @@ LABEL(loop_ashr_15_use_sse4_2):
 	jbe	LABEL(use_sse4_2_exit)
 #ifdef USE_AS_STRNCMP
 	sub	$16, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	$16, %rdx
 	jmp	LABEL(loop_ashr_15_use_sse4_2)
@@ -1441,219 +1439,78 @@ LABEL(nibble_ashr_use_sse4_2_exit):
 	pcmpistri      $0x1a,(%rsi,%rdx), %xmm0
 	.p2align 4
 LABEL(use_sse4_2_exit):
-	jnc	LABEL(strcmp_exitz)
+	jnc	LABEL(strcmp_exitz_sse4_2)
 #ifdef USE_AS_STRNCMP
 	sub	%rcx, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
 	add	%rcx, %rdx
 	lea	-16(%rdi, %r9), %rdi
 	movzbl	(%rdi, %rdx), %eax
 	movzbl	(%rsi, %rdx), %edx
 	test	%r8d, %r8d
-	jz	LABEL(use_sse4_2_ret)
+	jz	LABEL(use_sse4_2_ret_sse4_2)
 	xchg	%eax, %edx
-LABEL(use_sse4_2_ret):
+LABEL(use_sse4_2_ret_sse4_2):
 	sub	%edx, %eax
 	ret
 
-	.p2align 4
-LABEL(aftertail):
-	pcmpeqb	%xmm3, %xmm1
-	psubb	%xmm0, %xmm1
-	pmovmskb %xmm1, %edx
-	not	%edx
-
-	.p2align 4
-LABEL(exit):
-	lea	-16(%r9, %rcx), %rax	/* locate the exact offset for rdi */
-LABEL(less32bytes):
+LABEL(less32bytes_sse4_2):
 	lea	(%rdi, %rax), %rdi	/* locate the exact address for first operand(rdi) */
 	lea	(%rsi, %rcx), %rsi	/* locate the exact address for second operand(rsi) */
 	test	%r8d, %r8d
-	jz	LABEL(ret)
+	jz	LABEL(ret_sse4_2)
 	xchg	%rsi, %rdi		/* recover original order according to flag(%r8d) */
 
 	.p2align 4
-LABEL(ret):
-LABEL(less16bytes):
-	/*
-	 * Check to see if BSF is fast on this processor. If not, use a different
-	 * exit tail.
-	 */
+LABEL(ret_sse4_2):
+LABEL(less16bytes_sse4_2):
 	bsf	%rdx, %rdx		/* find and store bit index in %rdx */
 
 #ifdef USE_AS_STRNCMP
 	sub	%rdx, %r11
-	jbe	LABEL(strcmp_exitz)
+	jbe	LABEL(strcmp_exitz_sse4_2)
 #endif
-	xor	%ecx, %ecx		/* clear %ecx */
-	xor	%eax, %eax		/* clear %eax */
-
-	movb	(%rsi, %rdx), %cl
-	movb	(%rdi, %rdx), %al
+	movzbl	(%rsi, %rdx), %ecx
+	movzbl	(%rdi, %rdx), %eax
 
 	sub	%ecx, %eax
 	ret
 
-LABEL(strcmp_exitz):
+LABEL(strcmp_exitz_sse4_2):
 	xor	%eax, %eax
 	ret
 
 	.p2align 4
-LABEL(Byte0):
-	/*
-	 * never need to handle byte 0 for strncmpy
-#ifdef USE_AS_STRNCMP
-	sub	$0, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	*/
+LABEL(Byte0_sse4_2):
 	movzx	(%rsi), %ecx
 	movzx	(%rdi), %eax
 
 	sub	%ecx, %eax
 	ret
-
-	.p2align 4
-LABEL(Byte1):
-
-#ifdef USE_AS_STRNCMP
-	sub	$1, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	1(%rsi), %ecx
-	movzx	1(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte2):
-
-#ifdef USE_AS_STRNCMP
-	sub	$2, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	2(%rsi), %ecx
-	movzx	2(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte3):
-
-#ifdef USE_AS_STRNCMP
-	sub	$3, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	3(%rsi), %ecx
-	movzx	3(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte4):
-
-#ifdef USE_AS_STRNCMP
-	sub	$4, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	4(%rsi), %ecx
-	movzx	4(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte5):
-
-#ifdef USE_AS_STRNCMP
-	sub	$5, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	5(%rsi), %ecx
-	movzx	5(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(Byte6):
-
-#ifdef USE_AS_STRNCMP
-	sub	$6, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	6(%rsi), %ecx
-	movzx	6(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
-
-	.p2align 4
-LABEL(next_8_bytes):
-	add	$8, %rdi
-	add	$8, %rsi
-#ifdef USE_AS_STRNCMP
-	sub	$8, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	test	$0x01, %dh
-	jnz	LABEL(Byte0)
-
-	test	$0x02, %dh
-	jnz	LABEL(Byte1)
-
-	test	$0x04, %dh
-	jnz	LABEL(Byte2)
-
-	test	$0x08, %dh
-	jnz	LABEL(Byte3)
-
-	test	$0x10, %dh
-	jnz	LABEL(Byte4)
-
-	test	$0x20, %dh
-	jnz	LABEL(Byte5)
-
-	test	$0x40, %dh
-	jnz	LABEL(Byte6)
-
-#ifdef USE_AS_STRNCMP
-	sub	$7, %r11
-	jbe	LABEL(strcmp_exitz)
-#endif
-	movzx	7(%rsi), %ecx
-	movzx	7(%rdi), %eax
-
-	sub	%ecx, %eax
-	ret
 	cfi_endproc
 	.size	STRCMP_SSE42, .-STRCMP_SSE42
 
 	/* Put all SSE 4.2 functions together.  */
 	.section .rodata.sse4.2,"a",@progbits
-	.p2align 4
-LABEL(unaligned_table):
-	.int	LABEL(ashr_1) - LABEL(unaligned_table)
-	.int	LABEL(ashr_2) - LABEL(unaligned_table)
-	.int	LABEL(ashr_3) - LABEL(unaligned_table)
-	.int	LABEL(ashr_4) - LABEL(unaligned_table)
-	.int	LABEL(ashr_5) - LABEL(unaligned_table)
-	.int	LABEL(ashr_6) - LABEL(unaligned_table)
-	.int	LABEL(ashr_7) - LABEL(unaligned_table)
-	.int	LABEL(ashr_8) - LABEL(unaligned_table)
-	.int	LABEL(ashr_9) - LABEL(unaligned_table)
-	.int	LABEL(ashr_10) - LABEL(unaligned_table)
-	.int	LABEL(ashr_11) - LABEL(unaligned_table)
-	.int	LABEL(ashr_12) - LABEL(unaligned_table)
-	.int	LABEL(ashr_13) - LABEL(unaligned_table)
-	.int	LABEL(ashr_14) - LABEL(unaligned_table)
-	.int	LABEL(ashr_15) - LABEL(unaligned_table)
-	.int	LABEL(ashr_0) - LABEL(unaligned_table)
+	.p2align 3
+LABEL(unaligned_table_sse4_2):
+	.int	LABEL(ashr_1_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_2_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_3_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_4_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_5_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_6_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_7_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_8_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_9_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_10_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_11_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_12_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_13_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_14_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_15_sse4_2) - LABEL(unaligned_table_sse4_2)
+	.int	LABEL(ashr_0_sse4_2) - LABEL(unaligned_table_sse4_2)
 
 
 # undef ENTRY
@@ -1673,6 +1530,4 @@ LABEL(unaligned_table):
 	.globl __GI_STRCMP; __GI_STRCMP = STRCMP_SSE2
 #endif
 
-#ifndef USE_AS_STRNCMP
 #include "../strcmp.S"
-#endif
diff --git a/sysdeps/x86_64/multiarch/strcpy.S b/sysdeps/x86_64/multiarch/strcpy.S
index 25cd01307d..7e400a9140 100644
--- a/sysdeps/x86_64/multiarch/strcpy.S
+++ b/sysdeps/x86_64/multiarch/strcpy.S
@@ -64,16 +64,9 @@ ENTRY(STRCPY)
 	call	__init_cpu_features
 1:	leaq	STRCPY_SSE2(%rip), %rax
 	testl	$(1<<9), __cpu_features+CPUID_OFFSET+COMMON_CPUID_INDEX_1*CPUID_SIZE+CPUID_ECX_OFFSET(%rip)
-	jz	3f
-/* Avoid SSSE3 strcpy on Atom since it is slow.  */
-	cmpl	$1, __cpu_features+KIND_OFFSET(%rip)
-	jne	2f
-	cmpl	$6, __cpu_features+FAMILY_OFFSET(%rip)
-	jne	2f
-	cmpl	$28, __cpu_features+MODEL_OFFSET(%rip)
-	jz	3f
-2:	leaq	STRCPY_SSSE3(%rip), %rax
-3:	ret
+	jz	2f
+	leaq	STRCPY_SSSE3(%rip), %rax
+2:	ret
 END(STRCPY)
 
 	.section .text.ssse3,"ax",@progbits
diff --git a/sysdeps/x86_64/multiarch/strncmp-c.c b/sysdeps/x86_64/multiarch/strncmp-c.c
deleted file mode 100644
index d4f74a418d..0000000000
--- a/sysdeps/x86_64/multiarch/strncmp-c.c
+++ /dev/null
@@ -1,8 +0,0 @@
-#ifdef SHARED
-#define STRNCMP __strncmp_sse2
-#undef libc_hidden_builtin_def
-#define libc_hidden_builtin_def(name) \
-  __hidden_ver1 (__strncmp_sse2, __GI_strncmp, __strncmp_sse2);
-#endif
-
-#include "strncmp.c"
diff --git a/sysdeps/x86_64/rtld-memchr.c b/sysdeps/x86_64/rtld-memchr.c
new file mode 100644
index 0000000000..f63fefbcec
--- /dev/null
+++ b/sysdeps/x86_64/rtld-memchr.c
@@ -0,0 +1 @@
+#include <string/memchr.c>
diff --git a/sysdeps/x86_64/rtld-memcmp.c b/sysdeps/x86_64/rtld-memcmp.c
new file mode 100644
index 0000000000..2ee40328b8
--- /dev/null
+++ b/sysdeps/x86_64/rtld-memcmp.c
@@ -0,0 +1 @@
+#include <string/memcmp.c>
diff --git a/sysdeps/x86_64/rtld-rawmemchr.c b/sysdeps/x86_64/rtld-rawmemchr.c
new file mode 100644
index 0000000000..2b9189393c
--- /dev/null
+++ b/sysdeps/x86_64/rtld-rawmemchr.c
@@ -0,0 +1 @@
+#include <string/rawmemchr.c>
diff --git a/sysdeps/x86_64/rtld-strchr.S b/sysdeps/x86_64/rtld-strchr.S
new file mode 100644
index 0000000000..8934697972
--- /dev/null
+++ b/sysdeps/x86_64/rtld-strchr.S
@@ -0,0 +1,291 @@
+/* strchr (str, ch) -- Return pointer to first occurrence of CH in STR.
+   For AMD x86-64.
+   Copyright (C) 2002, 2005 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include "asm-syntax.h"
+#include "bp-sym.h"
+#include "bp-asm.h"
+
+
+	.text
+ENTRY (BP_SYM (strchr))
+
+	/* Before we start with the main loop we process single bytes
+	   until the source pointer is aligned.  This has two reasons:
+	   1. aligned 64-bit memory access is faster
+	   and (more important)
+	   2. we process in the main loop 64 bit in one step although
+	      we don't know the end of the string.  But accessing at
+	      8-byte alignment guarantees that we never access illegal
+	      memory if this would not also be done by the trivial
+	      implementation (this is because all processor inherent
+	      boundaries are multiples of 8).  */
+
+	movq	%rdi, %rdx
+	andl	$7, %edx	/* Mask alignment bits  */
+	movq	%rdi, %rax	/* duplicate destination.  */
+	jz	1f		/* aligned => start loop */
+	neg	%edx
+	addl	$8, %edx	/* Align to 8 bytes.  */
+
+	/* Search the first bytes directly.  */
+0:	movb	(%rax), %cl	/* load byte  */
+	cmpb	%cl,%sil	/* compare byte.  */
+	je	6f		/* target found */
+	testb	%cl,%cl		/* is byte NUL? */
+	je	7f		/* yes => return NULL */
+	incq	%rax		/* increment pointer */
+	decl	%edx
+	jnz	0b
+
+
+1:
+	/* At the moment %rsi contains C.  What we need for the
+	   algorithm is C in all bytes of the register.  Avoid
+	   operations on 16 bit words because these require an
+	   prefix byte (and one more cycle).  */
+	/* Populate 8 bit data to full 64-bit.  */
+	movabs	$0x0101010101010101,%r9
+	movzbl	%sil,%edx
+	imul	%rdx,%r9
+
+	movq $0xfefefefefefefeff, %r8 /* Save magic.  */
+
+      /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to
+	 change any of the hole bits of LONGWORD.
+
+	 1) Is this safe?  Will it catch all the zero bytes?
+	 Suppose there is a byte with all zeros.  Any carry bits
+	 propagating from its left will fall into the hole at its
+	 least significant bit and stop.  Since there will be no
+	 carry from its most significant bit, the LSB of the
+	 byte to the left will be unchanged, and the zero will be
+	 detected.
+
+	 2) Is this worthwhile?  Will it ignore everything except
+	 zero bytes?  Suppose every byte of QUARDWORD has a bit set
+	 somewhere.  There will be a carry into bit 8.	If bit 8
+	 is set, this will carry into bit 16.  If bit 8 is clear,
+	 one of bits 9-15 must be set, so there will be a carry
+	 into bit 16.  Similarly, there will be a carry into bit
+	 24 tec..  If one of bits 54-63 is set, there will be a carry
+	 into bit 64 (=carry flag), so all of the hole bits will
+	 be changed.
+
+	 3) But wait!  Aren't we looking for C, not zero?
+	 Good point.  So what we do is XOR LONGWORD with a longword,
+	 each of whose bytes is C.  This turns each byte that is C
+	 into a zero.  */
+
+	.p2align 4
+4:
+	/* Main Loop is unrolled 4 times.  */
+	/* First unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
+				   are now 0 */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found c => return pointer */
+
+	/* The quadword we looked at does not contain the value we're looking
+	   for.  Let's search now whether we have reached the end of the
+	   string.  */
+	xorq %r9, %rcx		/* restore original dword without reload */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 7f			/* highest byte is NUL => return NULL */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 7f			/* found NUL => return NULL */
+
+	/* Second unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
+				   are now 0 */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found c => return pointer */
+
+	/* The quadword we looked at does not contain the value we're looking
+	   for.  Let's search now whether we have reached the end of the
+	   string.  */
+	xorq %r9, %rcx		/* restore original dword without reload */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 7f			/* highest byte is NUL => return NULL */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 7f			/* found NUL => return NULL */
+	/* Third unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
+				   are now 0 */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found c => return pointer */
+
+	/* The quadword we looked at does not contain the value we're looking
+	   for.  Let's search now whether we have reached the end of the
+	   string.  */
+	xorq %r9, %rcx		/* restore original dword without reload */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 7f			/* highest byte is NUL => return NULL */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 7f			/* found NUL => return NULL */
+	/* Fourth unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	xorq %r9, %rcx		/* XOR with qword c|...|c => bytes of str == c
+				   are now 0 */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found c => return pointer */
+
+	/* The quadword we looked at does not contain the value we're looking
+	   for.  Let's search now whether we have reached the end of the
+	   string.  */
+	xorq %r9, %rcx		/* restore original dword without reload */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 7f			/* highest byte is NUL => return NULL */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jz 4b			/* no NUL found => restart loop */
+
+
+7:	/* Return NULL.  */
+	xorl %eax, %eax
+	retq
+
+
+	/* We now scan for the byte in which the character was matched.
+	   But we have to take care of the case that a NUL char is
+	   found before this in the dword.  Note that we XORed %rcx
+	   with the byte we're looking for, therefore the tests below look
+	   reversed.  */
+
+
+	.p2align 4		/* Align, it's a jump target.  */
+3:	movq	%r9,%rdx	/* move to %rdx so that we can access bytes */
+	subq	$8,%rax		/* correct pointer increment.  */
+	testb %cl, %cl		/* is first byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %cl		/* is first byte NUL? */
+	je 7b			/* yes => return NULL */
+	incq %rax		/* increment pointer */
+
+	testb %ch, %ch		/* is second byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %ch		/* is second byte NUL? */
+	je 7b			/* yes => return NULL? */
+	incq %rax		/* increment pointer */
+
+	shrq $16, %rcx		/* make upper bytes accessible */
+	testb %cl, %cl		/* is third byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %cl		/* is third byte NUL? */
+	je 7b			/* yes => return NULL */
+	incq %rax		/* increment pointer */
+
+	testb %ch, %ch		/* is fourth byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %ch		/* is fourth byte NUL? */
+	je 7b			/* yes => return NULL? */
+	incq %rax		/* increment pointer */
+
+	shrq $16, %rcx		/* make upper bytes accessible */
+	testb %cl, %cl		/* is fifth byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %cl		/* is fifth byte NUL? */
+	je 7b			/* yes => return NULL */
+	incq %rax		/* increment pointer */
+
+	testb %ch, %ch		/* is sixth byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %ch		/* is sixth byte NUL? */
+	je 7b			/* yes => return NULL? */
+	incq %rax		/* increment pointer */
+
+	shrq $16, %rcx		/* make upper bytes accessible */
+	testb %cl, %cl		/* is seventh byte C? */
+	jz 6f			/* yes => return pointer */
+	cmpb %dl, %cl		/* is seventh byte NUL? */
+	je 7b			/* yes => return NULL */
+
+	/* It must be in the eigth byte and it cannot be NUL.  */
+	incq %rax
+
+6:
+	nop
+	retq
+END (BP_SYM (strchr))
+
+weak_alias (BP_SYM (strchr), BP_SYM (index))
+libc_hidden_builtin_def (strchr)
diff --git a/sysdeps/x86_64/rtld-strcmp.S b/sysdeps/x86_64/rtld-strcmp.S
new file mode 100644
index 0000000000..a25535c161
--- /dev/null
+++ b/sysdeps/x86_64/rtld-strcmp.S
@@ -0,0 +1,28 @@
+#include <sysdep.h>
+#include "asm-syntax.h"
+#include "bp-sym.h"
+#include "bp-asm.h"
+
+#ifndef LABEL
+#define LABEL(l) L(l)
+#endif
+
+	.text
+ENTRY (BP_SYM (STRCMP))
+/* Simple version since we can't use SSE registers in ld.so.  */
+L(oop):	movb	(%rdi), %al
+	cmpb	(%rsi), %al
+	jne	L(neq)
+	incq	%rdi
+	incq	%rsi
+	testb	%al, %al
+	jnz	L(oop)
+
+	xorl	%eax, %eax
+	ret
+
+L(neq):	movl	$1, %eax
+	movl	$-1, %ecx
+	cmovbl	%ecx, %eax
+	ret
+END (BP_SYM (STRCMP))
diff --git a/sysdeps/x86_64/rtld-strlen.S b/sysdeps/x86_64/rtld-strlen.S
new file mode 100644
index 0000000000..fd950edaaa
--- /dev/null
+++ b/sysdeps/x86_64/rtld-strlen.S
@@ -0,0 +1,139 @@
+/* strlen(str) -- determine the length of the string STR.
+   Copyright (C) 2002, 2003 Free Software Foundation, Inc.
+   Based on i486 version contributed by Ulrich Drepper <drepper@redhat.com>.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#include <sysdep.h>
+#include "asm-syntax.h"
+#include "bp-sym.h"
+#include "bp-asm.h"
+
+
+	.text
+ENTRY (strlen)
+	movq %rdi, %rcx		/* Duplicate source pointer. */
+	andl $7, %ecx		/* mask alignment bits */
+	movq %rdi, %rax		/* duplicate destination.  */
+	jz 1f			/* aligned => start loop */
+
+	neg %ecx		/* We need to align to 8 bytes.  */
+	addl $8,%ecx
+	/* Search the first bytes directly.  */
+0:	cmpb $0x0,(%rax)	/* is byte NUL? */
+	je 2f			/* yes => return */
+	incq %rax		/* increment pointer */
+	decl %ecx
+	jnz 0b
+
+1:	movq $0xfefefefefefefeff,%r8 /* Save magic.  */
+
+	.p2align 4		/* Align loop.  */
+4:	/* Main Loop is unrolled 4 times.  */
+	/* First unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found NUL => return pointer */
+
+	/* Second unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found NUL => return pointer */
+
+	/* Third unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jnz 3f			/* found NUL => return pointer */
+
+	/* Fourth unroll.  */
+	movq (%rax), %rcx	/* get double word (= 8 bytes) in question */
+	addq $8,%rax		/* adjust pointer for next word */
+	movq %r8, %rdx		/* magic value */
+	addq %rcx, %rdx		/* add the magic value to the word.  We get
+				   carry bits reported for each byte which
+				   is *not* 0 */
+	jnc 3f			/* highest byte is NUL => return pointer */
+	xorq %rcx, %rdx		/* (word+magic)^word */
+	orq %r8, %rdx		/* set all non-carry bits */
+	incq %rdx		/* add 1: if one carry bit was *not* set
+				   the addition will not result in 0.  */
+	jz 4b			/* no NUL found => continue loop */
+
+	.p2align 4		/* Align, it's a jump target.  */
+3:	subq $8,%rax		/* correct pointer increment.  */
+
+	testb %cl, %cl		/* is first byte NUL? */
+	jz 2f			/* yes => return */
+	incq %rax		/* increment pointer */
+
+	testb %ch, %ch		/* is second byte NUL? */
+	jz 2f			/* yes => return */
+	incq %rax		/* increment pointer */
+
+	testl $0x00ff0000, %ecx /* is third byte NUL? */
+	jz 2f			/* yes => return pointer */
+	incq %rax		/* increment pointer */
+
+	testl $0xff000000, %ecx /* is fourth byte NUL? */
+	jz 2f			/* yes => return pointer */
+	incq %rax		/* increment pointer */
+
+	shrq $32, %rcx		/* look at other half.  */
+
+	testb %cl, %cl		/* is first byte NUL? */
+	jz 2f			/* yes => return */
+	incq %rax		/* increment pointer */
+
+	testb %ch, %ch		/* is second byte NUL? */
+	jz 2f			/* yes => return */
+	incq %rax		/* increment pointer */
+
+	testl $0xff0000, %ecx	/* is third byte NUL? */
+	jz 2f			/* yes => return pointer */
+	incq %rax		/* increment pointer */
+2:
+	subq %rdi, %rax		/* compute difference to string start */
+	ret
+END (strlen)
+libc_hidden_builtin_def (strlen)
diff --git a/sysdeps/x86_64/strcmp.S b/sysdeps/x86_64/strcmp.S
index 119b88e40b..340a64ba35 100644
--- a/sysdeps/x86_64/strcmp.S
+++ b/sysdeps/x86_64/strcmp.S
@@ -1,8 +1,10 @@
 /* Highly optimized version for x86-64.
-   Copyright (C) 1999, 2000, 2002, 2003, 2005 Free Software Foundation, Inc.
+   Copyright (C) 1999, 2000, 2002, 2003, 2005, 2009
+   Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Based on i686 version contributed by Ulrich Drepper
    <drepper@cygnus.com>, 1999.
+   Updated with SSE2 support contributed by Intel Corporation.
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
@@ -24,8 +26,35 @@
 #include "bp-sym.h"
 #include "bp-asm.h"
 
-        .text
-ENTRY (BP_SYM (strcmp))
+#undef UPDATE_STRNCMP_COUNTER
+
+#ifndef LABEL
+#define LABEL(l) L(l)
+#endif
+
+#ifdef USE_AS_STRNCMP
+/* Since the counter, %r11, is unsigned, we branch to strcmp_exitz
+   if the new counter > the old one or is 0.  */
+# define UPDATE_STRNCMP_COUNTER				\
+	/* calculate left number to compare */		\
+	lea	-16(%rcx, %r11), %r9;			\
+	cmp	%r9, %r11;				\
+	jb	LABEL(strcmp_exitz);			\
+	test	%r9, %r9;				\
+	je	LABEL(strcmp_exitz);			\
+	mov	%r9, %r11
+
+#else
+# define UPDATE_STRNCMP_COUNTER
+# ifndef STRCMP
+#  define STRCMP strcmp
+# endif
+#endif
+
+	.text
+ENTRY (BP_SYM (STRCMP))
+#ifdef NOT_IN_libc
+/* Simple version since we can't use SSE registers in ld.so.  */
 L(oop):	movb	(%rdi), %al
 	cmpb	(%rsi), %al
 	jne	L(neq)
@@ -41,5 +70,1914 @@ L(neq):	movl	$1, %eax
 	movl	$-1, %ecx
 	cmovbl	%ecx, %eax
 	ret
-END (BP_SYM (strcmp))
-libc_hidden_builtin_def (strcmp)
+END (BP_SYM (STRCMP))
+#else	/* NOT_IN_libc */
+/*
+ * This implementation uses SSE to compare up to 16 bytes at a time.
+ */
+#ifdef USE_AS_STRNCMP
+	test	%rdx, %rdx
+	je	LABEL(strcmp_exitz)
+	cmp	$1, %rdx
+	je	LABEL(Byte0)
+	mov	%rdx, %r11
+#endif
+	mov	%esi, %ecx
+	mov	%edi, %eax
+/* Use 64bit AND here to avoid long NOP padding.  */
+	and	$0x3f, %rcx		/* rsi alignment in cache line */
+	and	$0x3f, %rax		/* rdi alignment in cache line */
+	cmp	$0x30, %ecx
+	ja	LABEL(crosscache)	/* rsi: 16-byte load will cross cache line */
+	cmp	$0x30, %eax
+	ja	LABEL(crosscache)	/* rdi: 16-byte load will cross cache line */
+	movlpd	(%rdi), %xmm1
+	movlpd	(%rsi), %xmm2
+	movhpd	8(%rdi), %xmm1
+	movhpd	8(%rsi), %xmm2
+	pxor	%xmm0, %xmm0		/* clear %xmm0 for null char checks */
+	pcmpeqb	%xmm1, %xmm0		/* Any null chars? */
+	pcmpeqb	%xmm2, %xmm1		/* compare first 16 bytes for equality */
+	psubb	%xmm0, %xmm1		/* packed sub of comparison results*/
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx		/* if first 16 bytes are same, edx == 0xffff */
+	jnz	LABEL(less16bytes)	/* If not, find different value or null char */
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)	/* finish comparision */
+#endif
+	add	$16, %rsi		/* prepare to search next 16 bytes */
+	add	$16, %rdi		/* prepare to search next 16 bytes */
+
+	/*
+	 * Determine source and destination string offsets from 16-byte alignment.
+	 * Use relative offset difference between the two to determine which case
+	 * below to use.
+	 */
+	.p2align 4
+LABEL(crosscache):
+	and	$0xfffffffffffffff0, %rsi	/* force %rsi is 16 byte aligned */
+	and	$0xfffffffffffffff0, %rdi	/* force %rdi is 16 byte aligned */
+	mov	$0xffff, %edx			/* for equivalent offset */
+	xor	%r8d, %r8d
+	and	$0xf, %ecx			/* offset of rsi */
+	and	$0xf, %eax			/* offset of rdi */
+	cmp	%eax, %ecx
+	je	LABEL(ashr_0)			/* rsi and rdi relative offset same */
+	ja	LABEL(bigger)
+	mov	%edx, %r8d			/* r8d is offset flag for exit tail */
+	xchg	%ecx, %eax
+	xchg	%rsi, %rdi
+LABEL(bigger):
+	lea	15(%rax), %r9
+	sub	%rcx, %r9
+	lea	LABEL(unaligned_table)(%rip), %r10
+	movslq	(%r10, %r9,4), %r9
+	lea	(%r10, %r9), %r10
+	jmp	*%r10				/* jump to corresponding case */
+
+/*
+ * The following cases will be handled by ashr_0
+ *  rcx(offset of rsi)  rax(offset of rdi)  relative offset  corresponding case
+ *        n(0~15)            n(0~15)           15(15+ n-n)         ashr_0
+ */
+	.p2align 4
+LABEL(ashr_0):
+
+	movdqa	(%rsi), %xmm1
+	pxor	%xmm0, %xmm0			/* clear %xmm0 for null char check */
+	pcmpeqb	%xmm1, %xmm0			/* Any null chars? */
+	pcmpeqb	(%rdi), %xmm1			/* compare 16 bytes for equality */
+	psubb	%xmm0, %xmm1			/* packed sub of comparison results*/
+	pmovmskb %xmm1, %r9d
+	shr	%cl, %edx			/* adjust 0xffff for offset */
+	shr	%cl, %r9d			/* adjust for 16-byte offset */
+	sub	%r9d, %edx
+	/*
+	 * edx must be the same with r9d if in left byte (16-rcx) is equal to
+	 * the start from (16-rax) and no null char was seen.
+	 */
+	jne	LABEL(less32bytes)		/* mismatch or null char */
+	UPDATE_STRNCMP_COUNTER
+	mov	$16, %rcx
+	mov	$16, %r9
+	pxor	%xmm0, %xmm0			/* clear xmm0, may have changed above */
+
+	/*
+	 * Now both strings are aligned at 16-byte boundary. Loop over strings
+	 * checking 32-bytes per iteration.
+	 */
+	.p2align 4
+LABEL(loop_ashr_0):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)		/* mismatch or null char seen */
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+	add	$16, %rcx
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+	add	$16, %rcx
+	jmp	LABEL(loop_ashr_0)
+
+/*
+ * The following cases will be handled by ashr_1
+ * rcx(offset of rsi)  rax(offset of rdi)   relative offset   	corresponding case
+ *        n(15)            n -15            0(15 +(n-15) - n)         ashr_1
+ */
+	.p2align 4
+LABEL(ashr_1):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0		/* Any null chars? */
+	pslldq	$15, %xmm2		/* shift first string to align with second */
+	pcmpeqb	%xmm1, %xmm2		/* compare 16 bytes for equality */
+	psubb	%xmm0, %xmm2		/* packed sub of comparison results*/
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx		/* adjust 0xffff for offset */
+	shr	%cl, %r9d		/* adjust for 16-byte offset */
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)	/* mismatch or null char seen */
+	movdqa	(%rdi), %xmm3
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx		/* index for loads*/
+	mov	$1, %r9d		/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	1(%rdi), %r10
+	and	$0xfff, %r10		/* offset into 4K page */
+	sub	$0x1000, %r10		/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_1):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_1)	/* cross page boundary */
+
+LABEL(gobble_ashr_1):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4		 /* store for next cycle */
+
+	psrldq	$1, %xmm3
+	pslldq	$15, %xmm2
+	por	%xmm3, %xmm2		/* merge into one 16byte value */
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_1)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4		/* store for next cycle */
+
+	psrldq	$1, %xmm3
+	pslldq 	$15, %xmm2
+	por	%xmm3, %xmm2		/* merge into one 16byte value */
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_1)
+
+	/*
+	 * Nibble avoids loads across page boundary. This is to avoid a potential
+	 * access into unmapped memory.
+	 */
+	.p2align 4
+LABEL(nibble_ashr_1):
+	pcmpeqb	%xmm3, %xmm0		 /* check nibble for null char*/
+	pmovmskb %xmm0, %edx
+	test	$0xfffe, %edx
+	jnz	LABEL(ashr_1_exittail)	/* find null char*/
+
+#ifdef USE_AS_STRNCMP
+	cmp	$14, %r11
+	jbe	LABEL(ashr_1_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10		/* substract 4K from %r10 */
+	jmp	LABEL(gobble_ashr_1)
+
+	/*
+	 * Once find null char, determine if there is a string mismatch
+	 * before the null char.
+	 */
+	.p2align 4
+LABEL(ashr_1_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$1, %xmm0
+	psrldq	$1, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ * The following cases will be handled by ashr_2
+ * rcx(offset of rsi)  rax(offset of rdi)   relative offset   	corresponding case
+ *        n(14~15)            n -14         1(15 +(n-14) - n)         ashr_2
+ */
+	.p2align 4
+LABEL(ashr_2):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$14, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$2, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	2(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_2):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_2)
+
+LABEL(gobble_ashr_2):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$2, %xmm3
+	pslldq	$14, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_2)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$2, %xmm3
+	pslldq 	$14, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_2)
+
+	.p2align 4
+LABEL(nibble_ashr_2):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfffc, %edx
+	jnz	LABEL(ashr_2_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$13, %r11
+	jbe	LABEL(ashr_2_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_2)
+
+	.p2align 4
+LABEL(ashr_2_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$2, %xmm0
+	psrldq	$2, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ * The following cases will be handled by ashr_3
+ *  rcx(offset of rsi)  rax(offset of rdi)  relative offset	 corresponding case
+ *        n(13~15)            n -13         2(15 +(n-13) - n)         ashr_3
+ */
+	.p2align 4
+LABEL(ashr_3):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$13, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$3, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	3(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_3):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_3)
+
+LABEL(gobble_ashr_3):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$3, %xmm3
+	pslldq	$13, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_3)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$3, %xmm3
+	pslldq 	$13, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_3)
+
+	.p2align 4
+LABEL(nibble_ashr_3):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfff8, %edx
+	jnz	LABEL(ashr_3_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$12, %r11
+	jbe	LABEL(ashr_3_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_3)
+
+	.p2align 4
+LABEL(ashr_3_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$3, %xmm0
+	psrldq	$3, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ * The following cases will be handled by ashr_4
+ *  rcx(offset of rsi)  rax(offset of rdi)  relative offset	 corresponding case
+ *        n(12~15)            n -12         3(15 +(n-12) - n)         ashr_4
+ */
+	.p2align 4
+LABEL(ashr_4):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$12, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$4, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	4(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_4):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_4)
+
+LABEL(gobble_ashr_4):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$4, %xmm3
+	pslldq	$12, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_4)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$4, %xmm3
+	pslldq 	$12, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_4)
+
+	.p2align 4
+LABEL(nibble_ashr_4):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfff0, %edx
+	jnz	LABEL(ashr_4_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$11, %r11
+	jbe	LABEL(ashr_4_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_4)
+
+	.p2align 4
+LABEL(ashr_4_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$4, %xmm0
+	psrldq	$4, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ * The following cases will be handled by ashr_5
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset      corresponding case
+ *        n(11~15)          n - 11      	  4(15 +(n-11) - n)         ashr_5
+ */
+	.p2align 4
+LABEL(ashr_5):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$11, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$5, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	5(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_5):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_5)
+
+LABEL(gobble_ashr_5):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$5, %xmm3
+	pslldq	$11, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_5)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$5, %xmm3
+	pslldq 	$11, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_5)
+
+	.p2align 4
+LABEL(nibble_ashr_5):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xffe0, %edx
+	jnz	LABEL(ashr_5_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$10, %r11
+	jbe	LABEL(ashr_5_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_5)
+
+	.p2align 4
+LABEL(ashr_5_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$5, %xmm0
+	psrldq	$5, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ * The following cases will be handled by ashr_6
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset      corresponding case
+ *        n(10~15)          n - 10      	  5(15 +(n-10) - n)         ashr_6
+ */
+	.p2align 4
+LABEL(ashr_6):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$10, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$6, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	6(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_6):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_6)
+
+LABEL(gobble_ashr_6):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$6, %xmm3
+	pslldq	$10, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_6)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$6, %xmm3
+	pslldq 	$10, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_6)
+
+	.p2align 4
+LABEL(nibble_ashr_6):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xffc0, %edx
+	jnz	LABEL(ashr_6_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$9, %r11
+	jbe	LABEL(ashr_6_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_6)
+
+	.p2align 4
+LABEL(ashr_6_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$6, %xmm0
+	psrldq	$6, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ * The following cases will be handled by ashr_7
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset      corresponding case
+ *        n(9~15)          n - 9      	        6(15 +(n - 9) - n)         ashr_7
+ */
+	.p2align 4
+LABEL(ashr_7):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$9, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$7, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	7(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_7):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_7)
+
+LABEL(gobble_ashr_7):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$7, %xmm3
+	pslldq	$9, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_7)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$7, %xmm3
+	pslldq 	$9, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_7)
+
+	.p2align 4
+LABEL(nibble_ashr_7):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xff80, %edx
+	jnz	LABEL(ashr_7_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$8, %r11
+	jbe	LABEL(ashr_7_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_7)
+
+	.p2align 4
+LABEL(ashr_7_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$7, %xmm0
+	psrldq	$7, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_8
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(8~15)          n - 8      	        7(15 +(n - 8) - n)         ashr_8
+ */
+	.p2align 4
+LABEL(ashr_8):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$8, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$8, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	8(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_8):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_8)
+
+LABEL(gobble_ashr_8):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$8, %xmm3
+	pslldq	$8, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_8)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$8, %xmm3
+	pslldq 	$8, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_8)
+
+	.p2align 4
+LABEL(nibble_ashr_8):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xff00, %edx
+	jnz	LABEL(ashr_8_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$7, %r11
+	jbe	LABEL(ashr_8_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_8)
+
+	.p2align 4
+LABEL(ashr_8_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$8, %xmm0
+	psrldq	$8, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_9
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(7~15)          n - 7      	        8(15 +(n - 7) - n)         ashr_9
+ */
+	.p2align 4
+LABEL(ashr_9):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$7, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$9, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	9(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_9):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_9)
+
+LABEL(gobble_ashr_9):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$9, %xmm3
+	pslldq	$7, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_9)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$9, %xmm3
+	pslldq 	$7, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3		/* store for next cycle */
+	jmp	LABEL(loop_ashr_9)
+
+	.p2align 4
+LABEL(nibble_ashr_9):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfe00, %edx
+	jnz	LABEL(ashr_9_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$6, %r11
+	jbe	LABEL(ashr_9_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_9)
+
+	.p2align 4
+LABEL(ashr_9_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$9, %xmm0
+	psrldq	$9, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_10
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(6~15)          n - 6      	        9(15 +(n - 6) - n)         ashr_10
+ */
+	.p2align 4
+LABEL(ashr_10):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$6, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$10, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	10(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_10):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_10)
+
+LABEL(gobble_ashr_10):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$10, %xmm3
+	pslldq	$6, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_10)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$10, %xmm3
+	pslldq 	$6, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_10)
+
+	.p2align 4
+LABEL(nibble_ashr_10):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xfc00, %edx
+	jnz	LABEL(ashr_10_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$5, %r11
+	jbe	LABEL(ashr_10_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_10)
+
+	.p2align 4
+LABEL(ashr_10_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$10, %xmm0
+	psrldq	$10, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_11
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(5~15)          n - 5      	        10(15 +(n - 5) - n)         ashr_11
+ */
+	.p2align 4
+LABEL(ashr_11):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$5, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$11, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	11(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_11):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_11)
+
+LABEL(gobble_ashr_11):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$11, %xmm3
+	pslldq	$5, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_11)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$11, %xmm3
+	pslldq 	$5, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_11)
+
+	.p2align 4
+LABEL(nibble_ashr_11):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xf800, %edx
+	jnz	LABEL(ashr_11_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$4, %r11
+	jbe	LABEL(ashr_11_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_11)
+
+	.p2align 4
+LABEL(ashr_11_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$11, %xmm0
+	psrldq	$11, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_12
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(4~15)          n - 4      	        11(15 +(n - 4) - n)         ashr_12
+ */
+	.p2align 4
+LABEL(ashr_12):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$4, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$12, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	12(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_12):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_12)
+
+LABEL(gobble_ashr_12):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$12, %xmm3
+	pslldq	$4, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_12)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$12, %xmm3
+	pslldq 	$4, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_12)
+
+	.p2align 4
+LABEL(nibble_ashr_12):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xf000, %edx
+	jnz	LABEL(ashr_12_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$3, %r11
+	jbe	LABEL(ashr_12_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_12)
+
+	.p2align 4
+LABEL(ashr_12_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$12, %xmm0
+	psrldq	$12, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_13
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(3~15)          n - 3      	        12(15 +(n - 3) - n)         ashr_13
+ */
+	.p2align 4
+LABEL(ashr_13):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$3, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$13, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	13(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_13):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_13)
+
+LABEL(gobble_ashr_13):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$13, %xmm3
+	pslldq	$3, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_13)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$13, %xmm3
+	pslldq 	$3, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_13)
+
+	.p2align 4
+LABEL(nibble_ashr_13):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xe000, %edx
+	jnz	LABEL(ashr_13_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$2, %r11
+	jbe	LABEL(ashr_13_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_13)
+
+	.p2align 4
+LABEL(ashr_13_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq  $13, %xmm0
+	psrldq  $13, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_14
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(2~15)          n - 2      	        13(15 +(n - 2) - n)         ashr_14
+ */
+	.p2align 4
+LABEL(ashr_14):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq  $2, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$14, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	14(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_14):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_14)
+
+LABEL(gobble_ashr_14):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$14, %xmm3
+	pslldq	$2, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_14)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$14, %xmm3
+	pslldq 	$2, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_14)
+
+	.p2align 4
+LABEL(nibble_ashr_14):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0xc000, %edx
+	jnz	LABEL(ashr_14_exittail)
+
+#ifdef USE_AS_STRNCMP
+	cmp	$1, %r11
+	jbe	LABEL(ashr_14_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_14)
+
+	.p2align 4
+LABEL(ashr_14_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$14, %xmm0
+	psrldq	$14, %xmm3
+	jmp	LABEL(aftertail)
+
+/*
+ *  The following cases will be handled by ashr_15
+ *  rcx(offset of rsi)  rax(offset of rdi)        relative offset	 corresponding case
+ *        n(1~15)          n - 1      	        14(15 +(n - 1) - n)         ashr_15
+ */
+	.p2align 4
+LABEL(ashr_15):
+	pxor	%xmm0, %xmm0
+	movdqa	(%rdi), %xmm2
+	movdqa	(%rsi), %xmm1
+	pcmpeqb	%xmm1, %xmm0
+	pslldq	$1, %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	psubb	%xmm0, %xmm2
+	pmovmskb %xmm2, %r9d
+	shr	%cl, %edx
+	shr	%cl, %r9d
+	sub	%r9d, %edx
+	jnz	LABEL(less32bytes)
+
+	movdqa	(%rdi), %xmm3
+
+	UPDATE_STRNCMP_COUNTER
+
+	pxor	%xmm0, %xmm0
+	mov	$16, %rcx	/* index for loads */
+	mov	$15, %r9d	/* byte position left over from less32bytes case */
+	/*
+	 * Setup %r10 value allows us to detect crossing a page boundary.
+	 * When %r10 goes positive we have crossed a page boundary and
+	 * need to do a nibble.
+	 */
+	lea	15(%rdi), %r10
+	and	$0xfff, %r10	/* offset into 4K page */
+
+	sub	$0x1000, %r10	/* subtract 4K pagesize */
+
+	.p2align 4
+LABEL(loop_ashr_15):
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_15)
+
+LABEL(gobble_ashr_15):
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$15, %xmm3
+	pslldq	$1, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+
+	add	$16, %r10
+	jg	LABEL(nibble_ashr_15)	/* cross page boundary */
+
+	movdqa	(%rsi, %rcx), %xmm1
+	movdqa	(%rdi, %rcx), %xmm2
+	movdqa	%xmm2, %xmm4
+
+	psrldq	$15, %xmm3
+	pslldq 	$1, %xmm2
+	por	%xmm3, %xmm2
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm2, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	sub	$0xffff, %edx
+	jnz	LABEL(exit)
+
+#ifdef USE_AS_STRNCMP
+	sub	$16, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+
+	add	$16, %rcx
+	movdqa	%xmm4, %xmm3
+	jmp	LABEL(loop_ashr_15)
+
+	.p2align 4
+LABEL(nibble_ashr_15):
+	pcmpeqb	%xmm3, %xmm0		/* check nibble for null char */
+	pmovmskb %xmm0, %edx
+	test	$0x8000, %edx
+	jnz	LABEL(ashr_15_exittail)
+
+#ifdef USE_AS_STRNCMP
+	test	%r11, %r11
+	je	LABEL(ashr_15_exittail)
+#endif
+
+	pxor	%xmm0, %xmm0
+	sub	$0x1000, %r10
+	jmp	LABEL(gobble_ashr_15)
+
+	.p2align 4
+LABEL(ashr_15_exittail):
+	movdqa	(%rsi, %rcx), %xmm1
+	psrldq	$15, %xmm3
+	psrldq	$15, %xmm0
+
+	.p2align 4
+LABEL(aftertail):
+	pcmpeqb	%xmm3, %xmm1
+	psubb	%xmm0, %xmm1
+	pmovmskb %xmm1, %edx
+	not	%edx
+
+	.p2align 4
+LABEL(exit):
+	lea	-16(%r9, %rcx), %rax	/* locate the exact offset for rdi */
+LABEL(less32bytes):
+	lea	(%rdi, %rax), %rdi	/* locate the exact address for first operand(rdi) */
+	lea	(%rsi, %rcx), %rsi	/* locate the exact address for second operand(rsi) */
+	test	%r8d, %r8d
+	jz	LABEL(ret)
+	xchg	%rsi, %rdi		/* recover original order according to flag(%r8d) */
+
+	.p2align 4
+LABEL(ret):
+LABEL(less16bytes):
+	bsf	%rdx, %rdx		/* find and store bit index in %rdx */
+
+#ifdef USE_AS_STRNCMP
+	sub	%rdx, %r11
+	jbe	LABEL(strcmp_exitz)
+#endif
+	movzbl	(%rsi, %rdx), %ecx
+	movzbl	(%rdi, %rdx), %eax
+
+	sub	%ecx, %eax
+	ret
+
+LABEL(strcmp_exitz):
+	xor	%eax, %eax
+	ret
+
+	.p2align 4
+LABEL(Byte0):
+	movzx	(%rsi), %ecx
+	movzx	(%rdi), %eax
+
+	sub	%ecx, %eax
+	ret
+END (BP_SYM (STRCMP))
+
+	.section .rodata,"a",@progbits
+	.p2align 3
+LABEL(unaligned_table):
+	.int	LABEL(ashr_1) - LABEL(unaligned_table)
+	.int	LABEL(ashr_2) - LABEL(unaligned_table)
+	.int	LABEL(ashr_3) - LABEL(unaligned_table)
+	.int	LABEL(ashr_4) - LABEL(unaligned_table)
+	.int	LABEL(ashr_5) - LABEL(unaligned_table)
+	.int	LABEL(ashr_6) - LABEL(unaligned_table)
+	.int	LABEL(ashr_7) - LABEL(unaligned_table)
+	.int	LABEL(ashr_8) - LABEL(unaligned_table)
+	.int	LABEL(ashr_9) - LABEL(unaligned_table)
+	.int	LABEL(ashr_10) - LABEL(unaligned_table)
+	.int	LABEL(ashr_11) - LABEL(unaligned_table)
+	.int	LABEL(ashr_12) - LABEL(unaligned_table)
+	.int	LABEL(ashr_13) - LABEL(unaligned_table)
+	.int	LABEL(ashr_14) - LABEL(unaligned_table)
+	.int	LABEL(ashr_15) - LABEL(unaligned_table)
+	.int	LABEL(ashr_0) - LABEL(unaligned_table)
+#endif /* NOT_IN_libc */
+libc_hidden_builtin_def (STRCMP)
diff --git a/sysdeps/x86_64/strncmp.S b/sysdeps/x86_64/strncmp.S
new file mode 100644
index 0000000000..0af34e7f15
--- /dev/null
+++ b/sysdeps/x86_64/strncmp.S
@@ -0,0 +1,3 @@
+#define STRCMP strncmp
+#define USE_AS_STRNCMP
+#include "strcmp.S"
diff --git a/sysdeps/x86_64/tst-xmmymm.sh b/sysdeps/x86_64/tst-xmmymm.sh
new file mode 100755
index 0000000000..0735276e6d
--- /dev/null
+++ b/sysdeps/x86_64/tst-xmmymm.sh
@@ -0,0 +1,17 @@
+#! /bin/sh
+objpfx="$1"
+
+tmp=$(mktemp ${objpfx}tst-xmmymm.XXXXXX)
+trap 'rm -f "$tmp"' 1 2 3 15
+
+objdump -d "${objpfx}ld.so" |
+awk 'BEGIN { last="" } /^[[:xdigit:]]* <[_[:alnum:]]*>:$/ { fct=substr($2, 2, length($2)-3) } /,%[xy]mm[[:digit:]]*$/ { if (last != fct) { print fct; last=fct} }' |
+tee "$tmp"
+
+echo "Functions which incorrectly modify xmm/ymm registers:"
+err=1
+egrep -vs '^_dl_runtime_profile$' "$tmp" || err=0
+if test $err -eq 0; then echo "None"; fi
+
+rm "$tmp"
+exit $err