Merge "[MIPS] __dso_handle.S and __dso_handle_so.S not needed."

Restoring DEFAULT_MMAP_THRESHOLD to 64k, the way it was before
999089181e.

This forces allocations in the 64k-256k range to be mmaped.

Change-Id: Iace55ed638edd272b3e94fa6cd2ddd349042be84
Signed-off-by: Rom Lemarchand <romlem@google.com>

libc/Android.mk

@@ -6,67 +6,33 @@ include $(LOCAL_PATH)/arch-$(TARGET_ARCH)/syscalls.mk
 # =========================================================
 libc_common_src_files := \
 	$(syscall_src) \
-	unistd/abort.c \
 	unistd/alarm.c \
 	unistd/exec.c \
 	unistd/fnmatch.c \
-	unistd/getopt_long.c \
 	unistd/syslog.c \
 	unistd/system.c \
 	unistd/time.c \
 	stdio/asprintf.c \
-	stdio/clrerr.c \
-	stdio/fclose.c \
-	stdio/fdopen.c \
-	stdio/feof.c \
-	stdio/ferror.c \
 	stdio/fflush.c \
 	stdio/fgetc.c \
-	stdio/fgetln.c \
-	stdio/fgetpos.c \
-	stdio/fgets.c \
-	stdio/fileno.c \
 	stdio/findfp.c \
-	stdio/flags.c \
-	stdio/fopen.c \
 	stdio/fprintf.c \
-	stdio/fpurge.c \
 	stdio/fputc.c \
-	stdio/fputs.c \
 	stdio/fread.c \
 	stdio/freopen.c \
 	stdio/fscanf.c \
 	stdio/fseek.c \
-	stdio/fsetpos.c \
 	stdio/ftell.c \
-	stdio/funopen.c \
 	stdio/fvwrite.c \
-	stdio/fwalk.c \
-	stdio/fwrite.c \
-	stdio/getc.c \
-	stdio/getchar.c \
 	stdio/gets.c \
-	stdio/makebuf.c \
-	stdio/mktemp.c \
 	stdio/printf.c \
-	stdio/putc.c \
-	stdio/putchar.c \
-	stdio/puts.c \
-	stdio/putw.c \
 	stdio/refill.c \
-	stdio/remove.c \
 	stdio/rewind.c \
-	stdio/rget.c \
 	stdio/scanf.c \
-	stdio/setbuf.c \
-	stdio/setbuffer.c \
-	stdio/setvbuf.c \
 	stdio/snprintf.c\
 	stdio/sprintf.c \
 	stdio/sscanf.c \
 	stdio/stdio.c \
-	stdio/tempnam.c \
-	stdio/tmpnam.c \
 	stdio/ungetc.c \
 	stdio/vasprintf.c \
 	stdio/vfprintf.c \
@@ -77,13 +43,11 @@ libc_common_src_files := \
 	stdio/vscanf.c \
 	stdio/vsscanf.c \
 	stdio/wbuf.c \
-	stdio/wsetup.c \
 	stdlib/atexit.c \
 	stdlib/ctype_.c \
 	stdlib/exit.c \
 	stdlib/getenv.c \
 	stdlib/putenv.c \
-	stdlib/qsort.c \
 	stdlib/setenv.c \
 	stdlib/strtod.c \
 	stdlib/strtoimax.c \
@@ -94,29 +58,17 @@ libc_common_src_files := \
 	stdlib/strtoumax.c \
 	stdlib/tolower_.c \
 	stdlib/toupper_.c \
-	string/index.c \
 	string/strcasecmp.c \
-	string/strcat.c \
-	string/strchr.c \
 	string/strcspn.c \
 	string/strdup.c \
-	string/strlcat.c \
-	string/strlcpy.c \
-	string/strncat.c \
-	string/strncpy.c \
 	string/strpbrk.c \
-	string/strrchr.c \
+	string/__strrchr_chk.c \
 	string/strsep.c \
 	string/strspn.c \
 	string/strstr.c \
 	string/strtok.c \
 	wchar/wcswidth.c \
 	wchar/wcsxfrm.c \
-	tzcode/asctime.c \
-	tzcode/difftime.c \
-	tzcode/localtime.c \
-	tzcode/strftime.c \
-	tzcode/strptime.c \
 	bionic/arc4random.c \
 	bionic/atoi.c \
 	bionic/atol.c \
@@ -151,11 +103,8 @@ libc_common_src_files := \
 	bionic/ldexp.c \
 	bionic/lseek64.c \
 	bionic/md5.c \
-	bionic/memchr.c \
 	bionic/memmem.c \
-	bionic/memrchr.c \
 	bionic/memswap.c \
-	bionic/mmap.c \
 	bionic/openat.c \
 	bionic/open.c \
 	bionic/pathconf.c \
@@ -170,7 +119,6 @@ libc_common_src_files := \
 	bionic/recv.c \
 	bionic/sched_cpualloc.c \
 	bionic/sched_cpucount.c \
-	bionic/sched_getaffinity.c \
 	bionic/sched_getcpu.c \
 	bionic/semaphore.c \
 	bionic/send.c \
@@ -190,11 +138,11 @@ libc_common_src_files := \
 	bionic/sleep.c \
 	bionic/statfs.c \
 	bionic/strndup.c \
-	bionic/strnlen.c \
 	bionic/strntoimax.c \
 	bionic/strntoumax.c \
 	bionic/strtotimeval.c \
 	bionic/system_properties.c \
+	bionic/system_properties_compat.c \
 	bionic/tcgetpgrp.c \
 	bionic/tcsetpgrp.c \
 	bionic/thread_atexit.c \
@@ -234,6 +182,7 @@ libc_common_src_files := \
 	netbsd/nameser/ns_samedomain.c \
 
 libc_bionic_src_files := \
+    bionic/abort.cpp \
     bionic/assert.cpp \
     bionic/brk.cpp \
     bionic/dirent.cpp \
@@ -249,6 +198,7 @@ libc_bionic_src_files := \
     bionic/__memcpy_chk.cpp \
     bionic/__memmove_chk.cpp \
     bionic/__memset_chk.cpp \
+    bionic/mmap.cpp \
     bionic/pthread_attr.cpp \
     bionic/pthread_detach.cpp \
     bionic/pthread_equal.cpp \
@@ -264,11 +214,14 @@ libc_bionic_src_files := \
     bionic/raise.cpp \
     bionic/sbrk.cpp \
     bionic/scandir.cpp \
+    bionic/sched_getaffinity.cpp \
     bionic/__set_errno.cpp \
     bionic/setlocale.cpp \
     bionic/signalfd.cpp \
     bionic/sigwait.cpp \
+    bionic/statvfs.cpp \
     bionic/__strcat_chk.cpp \
+    bionic/__strchr_chk.cpp \
     bionic/__strcpy_chk.cpp \
     bionic/strerror.cpp \
     bionic/strerror_r.cpp \
@@ -288,32 +241,67 @@ libc_bionic_src_files := \
     bionic/wait.cpp \
     bionic/wchar.cpp \
 
+libc_tzcode_src_files := \
+    tzcode/asctime.c \
+    tzcode/difftime.c \
+    tzcode/localtime.c \
+    tzcode/strftime.c \
+    tzcode/strptime.c \
+
 libc_upstream_freebsd_src_files := \
+    upstream-freebsd/lib/libc/stdio/clrerr.c \
+    upstream-freebsd/lib/libc/stdio/fclose.c \
+    upstream-freebsd/lib/libc/stdio/fdopen.c \
+    upstream-freebsd/lib/libc/stdio/feof.c \
+    upstream-freebsd/lib/libc/stdio/ferror.c \
+    upstream-freebsd/lib/libc/stdio/fgetln.c \
+    upstream-freebsd/lib/libc/stdio/fgetpos.c \
+    upstream-freebsd/lib/libc/stdio/fgets.c \
+    upstream-freebsd/lib/libc/stdio/fileno.c \
+    upstream-freebsd/lib/libc/stdio/flags.c \
+    upstream-freebsd/lib/libc/stdio/fopen.c \
+    upstream-freebsd/lib/libc/stdio/fpurge.c \
+    upstream-freebsd/lib/libc/stdio/fputs.c \
+    upstream-freebsd/lib/libc/stdio/fsetpos.c \
+    upstream-freebsd/lib/libc/stdio/funopen.c \
+    upstream-freebsd/lib/libc/stdio/fwalk.c \
+    upstream-freebsd/lib/libc/stdio/fwrite.c \
+    upstream-freebsd/lib/libc/stdio/getc.c \
+    upstream-freebsd/lib/libc/stdio/getchar.c \
+    upstream-freebsd/lib/libc/stdio/makebuf.c \
+    upstream-freebsd/lib/libc/stdio/mktemp.c \
+    upstream-freebsd/lib/libc/stdio/putc.c \
+    upstream-freebsd/lib/libc/stdio/putchar.c \
+    upstream-freebsd/lib/libc/stdio/puts.c \
+    upstream-freebsd/lib/libc/stdio/putw.c \
+    upstream-freebsd/lib/libc/stdio/remove.c \
+    upstream-freebsd/lib/libc/stdio/rget.c \
+    upstream-freebsd/lib/libc/stdio/setbuf.c \
+    upstream-freebsd/lib/libc/stdio/setbuffer.c \
+    upstream-freebsd/lib/libc/stdio/setvbuf.c \
+    upstream-freebsd/lib/libc/stdio/tempnam.c \
+    upstream-freebsd/lib/libc/stdio/tmpnam.c \
+    upstream-freebsd/lib/libc/stdio/wsetup.c \
+    upstream-freebsd/lib/libc/stdlib/getopt_long.c \
+    upstream-freebsd/lib/libc/stdlib/qsort.c \
     upstream-freebsd/lib/libc/stdlib/realpath.c \
     upstream-freebsd/lib/libc/string/wcpcpy.c \
     upstream-freebsd/lib/libc/string/wcpncpy.c \
     upstream-freebsd/lib/libc/string/wcscasecmp.c \
-    upstream-freebsd/lib/libc/string/wcscat.c \
-    upstream-freebsd/lib/libc/string/wcschr.c \
-    upstream-freebsd/lib/libc/string/wcscmp.c \
-    upstream-freebsd/lib/libc/string/wcscpy.c \
     upstream-freebsd/lib/libc/string/wcscspn.c \
     upstream-freebsd/lib/libc/string/wcsdup.c \
     upstream-freebsd/lib/libc/string/wcslcat.c \
     upstream-freebsd/lib/libc/string/wcslcpy.c \
-    upstream-freebsd/lib/libc/string/wcslen.c \
     upstream-freebsd/lib/libc/string/wcsncasecmp.c \
     upstream-freebsd/lib/libc/string/wcsncat.c \
     upstream-freebsd/lib/libc/string/wcsncmp.c \
     upstream-freebsd/lib/libc/string/wcsncpy.c \
     upstream-freebsd/lib/libc/string/wcsnlen.c \
     upstream-freebsd/lib/libc/string/wcspbrk.c \
-    upstream-freebsd/lib/libc/string/wcsrchr.c \
     upstream-freebsd/lib/libc/string/wcsspn.c \
     upstream-freebsd/lib/libc/string/wcsstr.c \
     upstream-freebsd/lib/libc/string/wcstok.c \
     upstream-freebsd/lib/libc/string/wmemchr.c \
-    upstream-freebsd/lib/libc/string/wmemcmp.c \
     upstream-freebsd/lib/libc/string/wmemcpy.c \
     upstream-freebsd/lib/libc/string/wmemmove.c \
     upstream-freebsd/lib/libc/string/wmemset.c \
@@ -369,6 +357,24 @@ libc_common_src_files += \
 	bionic/memmove.c.arm \
 	string/bcopy.c \
 	string/strncmp.c \
+	string/strcat.c \
+	string/strncat.c \
+	string/strncpy.c \
+	bionic/strchr.cpp \
+	string/strrchr.c \
+	bionic/memchr.c \
+	bionic/memrchr.c \
+	string/index.c \
+	bionic/strnlen.c \
+	string/strlcat.c \
+	string/strlcpy.c \
+	upstream-freebsd/lib/libc/string/wcschr.c \
+	upstream-freebsd/lib/libc/string/wcsrchr.c \
+	upstream-freebsd/lib/libc/string/wcscmp.c \
+	upstream-freebsd/lib/libc/string/wcscpy.c \
+	upstream-freebsd/lib/libc/string/wmemcmp.c \
+	upstream-freebsd/lib/libc/string/wcslen.c \
+	upstream-freebsd/lib/libc/string/wcscat.c
 
 # These files need to be arm so that gdbserver
 # can set breakpoints in them without messing
@@ -392,7 +398,6 @@ libc_common_src_files += \
     bionic/pthread-rwlocks.c \
     bionic/pthread-timers.c \
     bionic/ptrace.c \
-    string/strcpy.c \
 
 libc_static_common_src_files += \
     bionic/pthread.c \
@@ -407,7 +412,25 @@ libc_common_src_files += \
 	string/bcopy.c \
 	string/strcmp.c \
 	string/strcpy.c \
-	string/strncmp.c
+	string/strncmp.c \
+	string/strcat.c \
+	string/strncat.c \
+	string/strncpy.c \
+	bionic/strchr.cpp \
+	string/strrchr.c \
+	bionic/memchr.c \
+	bionic/memrchr.c \
+	string/index.c \
+	bionic/strnlen.c \
+	string/strlcat.c \
+	string/strlcpy.c \
+	upstream-freebsd/lib/libc/string/wcschr.c \
+	upstream-freebsd/lib/libc/string/wcsrchr.c \
+	upstream-freebsd/lib/libc/string/wcscmp.c \
+	upstream-freebsd/lib/libc/string/wcscpy.c \
+	upstream-freebsd/lib/libc/string/wmemcmp.c \
+	upstream-freebsd/lib/libc/string/wcslen.c \
+	upstream-freebsd/lib/libc/string/wcscat.c
 
 libc_common_src_files += \
 	bionic/pthread-atfork.c \
@@ -469,14 +492,6 @@ libc_common_cflags := \
     -DLOG_ON_HEAP_ERROR \
     -Wall -Wextra
 
-# these macro definitions are required to implement the
-# 'timezone' and 'daylight' global variables, as well as
-# properly update the 'tm_gmtoff' field in 'struct tm'.
-#
-libc_common_cflags += \
-    -DTM_GMTOFF=tm_gmtoff \
-    -DUSG_COMPAT=1
-
 ifeq ($(strip $(DEBUG_BIONIC_LIBC)),true)
   libc_common_cflags += -DDEBUG
 endif
@@ -554,23 +569,17 @@ libc_crt_target_cflags += \
 # static C++ destructors are properly called on dlclose().
 #
 ifeq ($(TARGET_ARCH),arm)
-    libc_crtbegin_extension := c
     libc_crt_target_so_cflags :=
 endif
 ifeq ($(TARGET_ARCH),mips)
-    libc_crtbegin_extension := S
     libc_crt_target_so_cflags := -fPIC
 endif
 ifeq ($(TARGET_ARCH),x86)
-    libc_crtbegin_extension := c
     libc_crt_target_so_cflags := -fPIC
 endif
-ifeq ($(libc_crtbegin_extension),)
-    $(error $(TARGET_ARCH) not supported)
-endif
 libc_crt_target_so_cflags += $(libc_crt_target_cflags)
-libc_crt_target_crtbegin_file := $(LOCAL_PATH)/arch-$(TARGET_ARCH)/bionic/crtbegin.$(libc_crtbegin_extension)
+libc_crt_target_crtbegin_file := $(LOCAL_PATH)/arch-$(TARGET_ARCH)/bionic/crtbegin.c
-libc_crt_target_crtbegin_so_file := $(LOCAL_PATH)/arch-$(TARGET_ARCH)/bionic/crtbegin_so.$(libc_crtbegin_extension)
+libc_crt_target_crtbegin_so_file := $(LOCAL_PATH)/arch-$(TARGET_ARCH)/bionic/crtbegin_so.c
 
 # See the comment in crtbrand.c for the reason why we need to generate
 # crtbrand.s before generating crtbrand.o.
@@ -686,6 +695,28 @@ LOCAL_SYSTEM_SHARED_LIBRARIES :=
 include $(BUILD_STATIC_LIBRARY)
 
 
+# ========================================================
+# libc_tzcode.a - upstream 'tzcode' code
+# ========================================================
+
+include $(CLEAR_VARS)
+
+LOCAL_SRC_FILES := $(libc_tzcode_src_files)
+LOCAL_CFLAGS := \
+    $(libc_common_cflags) \
+    -std=gnu99 \
+    -DSTD_INSPIRED=1 \
+    -DTZDIR=\"/system/usr/share/zoneinfo\" \
+    -DTM_GMTOFF=tm_gmtoff \
+    -DUSG_COMPAT=1
+LOCAL_C_INCLUDES := $(libc_common_c_includes)
+LOCAL_MODULE := libc_tzcode
+LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
+LOCAL_SYSTEM_SHARED_LIBRARIES :=
+
+include $(BUILD_STATIC_LIBRARY)
+
+
 # ========================================================
 # libc_freebsd.a - upstream FreeBSD C library code
 # ========================================================
@@ -761,12 +792,14 @@ LOCAL_CFLAGS := $(libc_common_cflags) \
 LOCAL_C_INCLUDES := $(libc_common_c_includes)
 LOCAL_MODULE := libc_common
 LOCAL_ADDITIONAL_DEPENDENCIES := $(LOCAL_PATH)/Android.mk
-LOCAL_WHOLE_STATIC_LIBRARIES := libbionic_ssp libc_bionic libc_freebsd libc_netbsd
+LOCAL_WHOLE_STATIC_LIBRARIES := \
+    libbionic_ssp \
+    libc_bionic \
+    libc_freebsd \
+    libc_netbsd \
+    libc_tzcode
 LOCAL_SYSTEM_SHARED_LIBRARIES :=
 
-# TODO: split out the asflags.
-LOCAL_ASFLAGS := $(LOCAL_CFLAGS)
-
 include $(BUILD_STATIC_LIBRARY)
 
 

libc/NOTICE

@@ -1673,38 +1673,6 @@ SUCH DAMAGE.
 
 -------------------------------------------------------------------
 
-Copyright (c) 1990, 1993
-   The Regents of the University of California.  All rights reserved.
-
-This code is derived from software contributed to Berkeley by
-Chris Torek.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions
-are met:
-1. Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-2. Redistributions in binary form must reproduce the above copyright
-   notice, this list of conditions and the following disclaimer in the
-   documentation and/or other materials provided with the distribution.
-4. Neither the name of the University nor the names of its contributors
-   may be used to endorse or promote products derived from this software
-   without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
-ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
-IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
-ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
-FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
-DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
-OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
-HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
-LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
-OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
-SUCH DAMAGE.
-
--------------------------------------------------------------------
-
 Copyright (c) 1990, 1993
    The Regents of the University of California.  All rights reserved.
 
@@ -2852,7 +2820,7 @@ are met:
 2. Redistributions in binary form must reproduce the above copyright
    notices, this list of conditions and the following disclaimer in the
    documentation and/or other materials provided with the distribution.
-4. Neither the name of the University nor the names of its contributors
+3. Neither the name of the University nor the names of its contributors
    may be used to endorse or promote products derived from this software
    without specific prior written permission.
 
@@ -3092,13 +3060,6 @@ are met:
 2. Redistributions in binary form must reproduce the above copyright
    notice, this list of conditions and the following disclaimer in the
    documentation and/or other materials provided with the distribution.
-3. All advertising materials mentioning features or use of this software
-   must display the following acknowledgement:
-       This product includes software developed by the NetBSD
-       Foundation, Inc. and its contributors.
-4. Neither the name of The NetBSD Foundation nor the names of its
-   contributors may be used to endorse or promote products derived
-   from this software without specific prior written permission.
 
 THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
@@ -3916,6 +3877,36 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 -------------------------------------------------------------------
 
+Copyright (c) 2010, 2011 Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+   Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+
+   Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+   Neither the name of Intel Corporation nor the names of its contributors
+   may be used to endorse or promote products derived from this software
+   without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+-------------------------------------------------------------------
+
 Copyright (c) 2010, Intel Corporation
 All rights reserved.
 
@@ -3946,6 +3937,36 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 -------------------------------------------------------------------
 
+Copyright (c) 2011 Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+   Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+
+   Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+   Neither the name of Intel Corporation nor the names of its contributors
+   may be used to endorse or promote products derived from this software
+   without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+-------------------------------------------------------------------
+
 Copyright (c) 2011 The Android Open Source Project
 Copyright (c) 2008 ARM Ltd
 All rights reserved.
@@ -3975,6 +3996,36 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 -------------------------------------------------------------------
 
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+   Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+
+   Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+   Neither the name of Intel Corporation nor the names of its contributors
+   may be used to endorse or promote products derived from this software
+   without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+-------------------------------------------------------------------
+
 Copyright (c) 2011, VMware, Inc.
 All rights reserved.
 
@@ -4058,6 +4109,38 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 -------------------------------------------------------------------
 
+Copyright (c) 2013, Linaro Limited
+   All rights reserved.
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+     Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+     Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+
+     Neither the name of Linaro Limited nor the names of its
+      contributors may be used to endorse or promote products derived
+      from this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+-------------------------------------------------------------------
+
 Copyright (c)1999 Citrus Project,
 All rights reserved.
 

libc/SYSCALLS.TXT

@@ -201,6 +201,8 @@ int     removexattr(const char *, const char *) 1
 int     lremovexattr(const char *, const char *) 1
 int     __statfs64:statfs64(const char *, size_t, struct statfs *)  1
 long    unshare(unsigned long)  1
+int     swapon(const char *, int) 1
+int     swapoff(const char *) 1
 
 # time
 int           pause ()                       1
@@ -221,6 +223,9 @@ int           __timer_getoverrun:timer_getoverrun(timer_t)
 int           __timer_delete:timer_delete(timer_t)                                                      1
 int           utimes(const char*, const struct timeval tvp[2])                          1
 int           utimensat(int, const char *, const struct timespec times[2], int)         1
+int           timerfd_create(clockid_t, int)   1
+int           timerfd_settime(int, int, const struct itimerspec *, struct itimerspec *)   1
+int           timerfd_gettime(int, struct itimerspec *)   1
 
 # signals
 int     sigaction(int, const struct sigaction *, struct sigaction *)  1

libc/arch-arm/arm.mk

@@ -15,7 +15,6 @@ _LIBC_ARCH_COMMON_SRC_FILES := \
     arch-arm/bionic/setjmp.S \
     arch-arm/bionic/sigsetjmp.S \
     arch-arm/bionic/strcpy.S \
-    arch-arm/bionic/strlen.c.arm \
     arch-arm/bionic/syscall.S \
     arch-arm/bionic/tgkill.S \
     arch-arm/bionic/tkill.S \
@@ -29,7 +28,7 @@ _LIBC_ARCH_DYNAMIC_SRC_FILES := \
     arch-arm/bionic/exidx_dynamic.c
 
 ifeq ($(strip $(wildcard bionic/libc/arch-arm/$(TARGET_CPU_VARIANT)/$(TARGET_CPU_VARIANT).mk)),)
-$(error "TARGET_CPU_VARIANT not set or set to an unknown value. Possible values are cortex-a9, cortex-a15, krait. Use generic for devices that do not have a CPU similar to any of the supported cpu variants.")
+$(error "TARGET_CPU_VARIANT not set or set to an unknown value. Possible values are cortex-a7, cortex-a8, cortex-a9, cortex-a15, krait. Use generic for devices that do not have a CPU similar to any of the supported cpu variants.")
 endif
 
 include bionic/libc/arch-arm/$(TARGET_CPU_VARIANT)/$(TARGET_CPU_VARIANT).mk

libc/arch-arm/bionic/abort_arm.S

@@ -38,5 +38,5 @@
 ENTRY(abort)
     .save   {r3, r14}
     stmfd   sp!, {r3, r14}
-    blx     PIC_SYM(_C_LABEL(__libc_android_abort), PLT)
+    bl      PIC_SYM(_C_LABEL(__libc_android_abort), PLT)
 END(abort)

libc/arch-arm/bionic/clone.S

@@ -32,33 +32,33 @@
 
 // int  __pthread_clone(void* (*fn)(void*), void* child_stack, int flags, void* arg);
 ENTRY(__pthread_clone)
-    # Copy the args onto the new stack.
+    # Push 'fn' and 'arg' onto 'child_stack'.
     stmdb   r1!, {r0, r3}
 
     # The sys_clone system call only takes two arguments: 'flags' and 'child_stack'.
     # 'child_stack' is already in r1, but we need to move 'flags' into position.
     mov     r0, r2
-    stmfd   sp!, {r4, r7}
 
     # System call.
+    mov     ip, r7
     ldr     r7, =__NR_clone
     swi     #0
+
+    # Child?
     movs    r0, r0
     beq     1f
 
-    # In parent, reload saved registers then either return or set errno.
+    # Parent.
-    ldmfd   sp!, {r4, r7}
+    mov     r7, ip
     cmn     r0, #(MAX_ERRNO + 1)
     bxls    lr
     neg     r0, r0
     b       __set_errno
 
-1:  # The child.
+1:  # Child.
-    # pick the function arg and call address off the stack and jump
+    # Pop 'fn' and 'arg' back off the stack and call __thread_entry.
-    # to the C __thread_entry function which does some setup and then
-    # calls the thread's start function
     pop     {r0, r1}
-    # __thread_entry needs the TLS pointer
+    # __thread_entry also needs our stack pointer.
     mov     r2, sp
     b       __thread_entry
 END(__pthread_clone)

libc/arch-arm/bionic/futex_arm.S

@@ -34,11 +34,10 @@
 
 // __futex_syscall3(*ftx, op, val)
 ENTRY(__futex_syscall3)
-    stmdb   sp!, {r4, r7}
+    mov     ip, r7
-    .save   {r4, r7}
     ldr     r7, =__NR_futex
     swi     #0
-    ldmia   sp!, {r4, r7}
+    mov     r7, ip
     bx      lr
 END(__futex_syscall3)
 
@@ -49,25 +48,23 @@ END(__futex_syscall4)
 
 // __futex_wait(*ftx, val, *timespec)
 ENTRY(__futex_wait)
-    stmdb   sp!, {r4, r7}
+    mov     ip, r7
-    .save   {r4, r7}
     mov     r3, r2
     mov     r2, r1
     mov     r1, #FUTEX_WAIT
     ldr     r7, =__NR_futex
     swi     #0
-    ldmia   sp!, {r4, r7}
+    mov     r7, ip
     bx      lr
 END(__futex_wait)
 
 // __futex_wake(*ftx, counter)
 ENTRY(__futex_wake)
-    .save   {r4, r7}
+    mov     ip, r7
-    stmdb   sp!, {r4, r7}
     mov     r2, r1
     mov     r1, #FUTEX_WAKE
     ldr     r7, =__NR_futex
     swi     #0
-    ldmia   sp!, {r4, r7}
+    mov     r7, ip
     bx      lr
 END(__futex_wake)

libc/arch-arm/bionic/memcpy.a9.S

@@ -0,0 +1,614 @@
+/* Copyright (c) 2013, Linaro Limited
+   All rights reserved.
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions
+   are met:
+
+      * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+
+      * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+
+      * Neither the name of Linaro Limited nor the names of its
+      contributors may be used to endorse or promote products derived
+      from this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ */
+
+/*
+   This memcpy routine is optimised for Cortex-A15 cores and takes advantage
+   of VFP or NEON when built with the appropriate flags.
+
+   Assumptions:
+
+    ARMv6 (ARMv7-a if using Neon)
+    ARM state
+    Unaligned accesses
+    LDRD/STRD support unaligned word accesses
+
+ */
+
+#include <machine/cpu-features.h>
+#include <machine/asm.h>
+
+	.syntax unified
+	/* This implementation requires ARM state.  */
+	.arm
+
+#ifdef __ARM_NEON__
+
+	.fpu	neon
+	.arch	armv7-a
+# define FRAME_SIZE	4
+# define USE_VFP
+# define USE_NEON
+
+#elif !defined (__SOFTFP__)
+
+	.arch	armv6
+	.fpu	vfpv2
+# define FRAME_SIZE	32
+# define USE_VFP
+
+#else
+	.arch	armv6
+# define FRAME_SIZE    32
+
+#endif
+
+/* Old versions of GAS incorrectly implement the NEON align semantics.  */
+#ifdef BROKEN_ASM_NEON_ALIGN
+#define ALIGN(addr, align) addr,:align
+#else
+#define ALIGN(addr, align) addr:align
+#endif
+
+#define PC_OFFSET	8	/* PC pipeline compensation.  */
+#define INSN_SIZE	4
+
+/* Call parameters.  */
+#define dstin	r0
+#define src	r1
+#define count	r2
+
+/* Locals.  */
+#define tmp1	r3
+#define dst	ip
+#define tmp2	r10
+
+#ifndef USE_NEON
+/* For bulk copies using GP registers.  */
+#define	A_l	r2		/* Call-clobbered.  */
+#define	A_h	r3		/* Call-clobbered.  */
+#define	B_l	r4
+#define	B_h	r5
+#define	C_l	r6
+#define	C_h	r7
+#define	D_l	r8
+#define	D_h	r9
+#endif
+
+/* Number of lines ahead to pre-fetch data.  If you change this the code
+   below will need adjustment to compensate.  */
+
+#define prefetch_lines	5
+
+#ifdef USE_VFP
+	.macro	cpy_line_vfp vreg, base
+	vstr	\vreg, [dst, #\base]
+	vldr	\vreg, [src, #\base]
+	vstr	d0, [dst, #\base + 8]
+	vldr	d0, [src, #\base + 8]
+	vstr	d1, [dst, #\base + 16]
+	vldr	d1, [src, #\base + 16]
+	vstr	d2, [dst, #\base + 24]
+	vldr	d2, [src, #\base + 24]
+	vstr	\vreg, [dst, #\base + 32]
+	vldr	\vreg, [src, #\base + prefetch_lines * 64 - 32]
+	vstr	d0, [dst, #\base + 40]
+	vldr	d0, [src, #\base + 40]
+	vstr	d1, [dst, #\base + 48]
+	vldr	d1, [src, #\base + 48]
+	vstr	d2, [dst, #\base + 56]
+	vldr	d2, [src, #\base + 56]
+	.endm
+
+	.macro	cpy_tail_vfp vreg, base
+	vstr	\vreg, [dst, #\base]
+	vldr	\vreg, [src, #\base]
+	vstr	d0, [dst, #\base + 8]
+	vldr	d0, [src, #\base + 8]
+	vstr	d1, [dst, #\base + 16]
+	vldr	d1, [src, #\base + 16]
+	vstr	d2, [dst, #\base + 24]
+	vldr	d2, [src, #\base + 24]
+	vstr	\vreg, [dst, #\base + 32]
+	vstr	d0, [dst, #\base + 40]
+	vldr	d0, [src, #\base + 40]
+	vstr	d1, [dst, #\base + 48]
+	vldr	d1, [src, #\base + 48]
+	vstr	d2, [dst, #\base + 56]
+	vldr	d2, [src, #\base + 56]
+	.endm
+#endif
+
+	.p2align 6
+ENTRY(memcpy)
+
+	mov	dst, dstin	/* Preserve dstin, we need to return it.  */
+	cmp	count, #64
+	bge	.Lcpy_not_short
+	/* Deal with small copies quickly by dropping straight into the
+	   exit block.  */
+
+.Ltail63unaligned:
+#ifdef USE_NEON
+	and	tmp1, count, #0x38
+	rsb	tmp1, tmp1, #(56 - PC_OFFSET + INSN_SIZE)
+	add	pc, pc, tmp1
+	vld1.8	{d0}, [src]!	/* 14 words to go.  */
+	vst1.8	{d0}, [dst]!
+	vld1.8	{d0}, [src]!	/* 12 words to go.  */
+	vst1.8	{d0}, [dst]!
+	vld1.8	{d0}, [src]!	/* 10 words to go.  */
+	vst1.8	{d0}, [dst]!
+	vld1.8	{d0}, [src]!	/* 8 words to go.  */
+	vst1.8	{d0}, [dst]!
+	vld1.8	{d0}, [src]!	/* 6 words to go.  */
+	vst1.8	{d0}, [dst]!
+	vld1.8	{d0}, [src]!	/* 4 words to go.  */
+	vst1.8	{d0}, [dst]!
+	vld1.8	{d0}, [src]!	/* 2 words to go.  */
+	vst1.8	{d0}, [dst]!
+
+	tst	count, #4
+	ldrne	tmp1, [src], #4
+	strne	tmp1, [dst], #4
+#else
+	/* Copy up to 15 full words of data.  May not be aligned.  */
+	/* Cannot use VFP for unaligned data.  */
+	and	tmp1, count, #0x3c
+	add	dst, dst, tmp1
+	add	src, src, tmp1
+	rsb	tmp1, tmp1, #(60 - PC_OFFSET/2 + INSN_SIZE/2)
+	/* Jump directly into the sequence below at the correct offset.  */
+	add	pc, pc, tmp1, lsl #1
+
+	ldr	tmp1, [src, #-60]	/* 15 words to go.  */
+	str	tmp1, [dst, #-60]
+
+	ldr	tmp1, [src, #-56]	/* 14 words to go.  */
+	str	tmp1, [dst, #-56]
+	ldr	tmp1, [src, #-52]
+	str	tmp1, [dst, #-52]
+
+	ldr	tmp1, [src, #-48]	/* 12 words to go.  */
+	str	tmp1, [dst, #-48]
+	ldr	tmp1, [src, #-44]
+	str	tmp1, [dst, #-44]
+
+	ldr	tmp1, [src, #-40]	/* 10 words to go.  */
+	str	tmp1, [dst, #-40]
+	ldr	tmp1, [src, #-36]
+	str	tmp1, [dst, #-36]
+
+	ldr	tmp1, [src, #-32]	/* 8 words to go.  */
+	str	tmp1, [dst, #-32]
+	ldr	tmp1, [src, #-28]
+	str	tmp1, [dst, #-28]
+
+	ldr	tmp1, [src, #-24]	/* 6 words to go.  */
+	str	tmp1, [dst, #-24]
+	ldr	tmp1, [src, #-20]
+	str	tmp1, [dst, #-20]
+
+	ldr	tmp1, [src, #-16]	/* 4 words to go.  */
+	str	tmp1, [dst, #-16]
+	ldr	tmp1, [src, #-12]
+	str	tmp1, [dst, #-12]
+
+	ldr	tmp1, [src, #-8]	/* 2 words to go.  */
+	str	tmp1, [dst, #-8]
+	ldr	tmp1, [src, #-4]
+	str	tmp1, [dst, #-4]
+#endif
+
+	lsls	count, count, #31
+	ldrhcs	tmp1, [src], #2
+	ldrbne	src, [src]		/* Src is dead, use as a scratch.  */
+	strhcs	tmp1, [dst], #2
+	strbne	src, [dst]
+	bx	lr
+
+.Lcpy_not_short:
+	/* At least 64 bytes to copy, but don't know the alignment yet.  */
+	str	tmp2, [sp, #-FRAME_SIZE]!
+	and	tmp2, src, #7
+	and	tmp1, dst, #7
+	cmp	tmp1, tmp2
+	bne	.Lcpy_notaligned
+
+#ifdef USE_VFP
+	/* Magic dust alert!  Force VFP on Cortex-A9.  Experiments show
+	   that the FP pipeline is much better at streaming loads and
+	   stores.  This is outside the critical loop.  */
+	vmov.f32	s0, s0
+#endif
+
+	/* SRC and DST have the same mutual 32-bit alignment, but we may
+	   still need to pre-copy some bytes to get to natural alignment.
+	   We bring DST into full 64-bit alignment.  */
+	lsls	tmp2, dst, #29
+	beq	1f
+	rsbs	tmp2, tmp2, #0
+	sub	count, count, tmp2, lsr #29
+	ldrmi	tmp1, [src], #4
+	strmi	tmp1, [dst], #4
+	lsls	tmp2, tmp2, #2
+	ldrhcs	tmp1, [src], #2
+	ldrbne	tmp2, [src], #1
+	strhcs	tmp1, [dst], #2
+	strbne	tmp2, [dst], #1
+
+1:
+	subs	tmp2, count, #64	/* Use tmp2 for count.  */
+	blt	.Ltail63aligned
+
+	cmp	tmp2, #512
+	bge	.Lcpy_body_long
+
+.Lcpy_body_medium:			/* Count in tmp2.  */
+#ifdef USE_VFP
+1:
+	vldr	d0, [src, #0]
+	subs	tmp2, tmp2, #64
+	vldr	d1, [src, #8]
+	vstr	d0, [dst, #0]
+	vldr	d0, [src, #16]
+	vstr	d1, [dst, #8]
+	vldr	d1, [src, #24]
+	vstr	d0, [dst, #16]
+	vldr	d0, [src, #32]
+	vstr	d1, [dst, #24]
+	vldr	d1, [src, #40]
+	vstr	d0, [dst, #32]
+	vldr	d0, [src, #48]
+	vstr	d1, [dst, #40]
+	vldr	d1, [src, #56]
+	vstr	d0, [dst, #48]
+	add	src, src, #64
+	vstr	d1, [dst, #56]
+	add	dst, dst, #64
+	bge	1b
+	tst	tmp2, #0x3f
+	beq	.Ldone
+
+.Ltail63aligned:			/* Count in tmp2.  */
+	and	tmp1, tmp2, #0x38
+	add	dst, dst, tmp1
+	add	src, src, tmp1
+	rsb	tmp1, tmp1, #(56 - PC_OFFSET + INSN_SIZE)
+	add	pc, pc, tmp1
+
+	vldr	d0, [src, #-56]	/* 14 words to go.  */
+	vstr	d0, [dst, #-56]
+	vldr	d0, [src, #-48]	/* 12 words to go.  */
+	vstr	d0, [dst, #-48]
+	vldr	d0, [src, #-40]	/* 10 words to go.  */
+	vstr	d0, [dst, #-40]
+	vldr	d0, [src, #-32]	/* 8 words to go.  */
+	vstr	d0, [dst, #-32]
+	vldr	d0, [src, #-24]	/* 6 words to go.  */
+	vstr	d0, [dst, #-24]
+	vldr	d0, [src, #-16]	/* 4 words to go.  */
+	vstr	d0, [dst, #-16]
+	vldr	d0, [src, #-8]	/* 2 words to go.  */
+	vstr	d0, [dst, #-8]
+#else
+	sub	src, src, #8
+	sub	dst, dst, #8
+1:
+	ldrd	A_l, A_h, [src, #8]
+	strd	A_l, A_h, [dst, #8]
+	ldrd	A_l, A_h, [src, #16]
+	strd	A_l, A_h, [dst, #16]
+	ldrd	A_l, A_h, [src, #24]
+	strd	A_l, A_h, [dst, #24]
+	ldrd	A_l, A_h, [src, #32]
+	strd	A_l, A_h, [dst, #32]
+	ldrd	A_l, A_h, [src, #40]
+	strd	A_l, A_h, [dst, #40]
+	ldrd	A_l, A_h, [src, #48]
+	strd	A_l, A_h, [dst, #48]
+	ldrd	A_l, A_h, [src, #56]
+	strd	A_l, A_h, [dst, #56]
+	ldrd	A_l, A_h, [src, #64]!
+	strd	A_l, A_h, [dst, #64]!
+	subs	tmp2, tmp2, #64
+	bge	1b
+	tst	tmp2, #0x3f
+	bne	1f
+	ldr	tmp2,[sp], #FRAME_SIZE
+	bx	lr
+1:
+	add	src, src, #8
+	add	dst, dst, #8
+
+.Ltail63aligned:			/* Count in tmp2.  */
+	/* Copy up to 7 d-words of data.  Similar to Ltail63unaligned, but
+	   we know that the src and dest are 32-bit aligned so we can use
+	   LDRD/STRD to improve efficiency.  */
+	/* TMP2 is now negative, but we don't care about that.  The bottom
+	   six bits still tell us how many bytes are left to copy.  */
+
+	and	tmp1, tmp2, #0x38
+	add	dst, dst, tmp1
+	add	src, src, tmp1
+	rsb	tmp1, tmp1, #(56 - PC_OFFSET + INSN_SIZE)
+	add	pc, pc, tmp1
+	ldrd	A_l, A_h, [src, #-56]	/* 14 words to go.  */
+	strd	A_l, A_h, [dst, #-56]
+	ldrd	A_l, A_h, [src, #-48]	/* 12 words to go.  */
+	strd	A_l, A_h, [dst, #-48]
+	ldrd	A_l, A_h, [src, #-40]	/* 10 words to go.  */
+	strd	A_l, A_h, [dst, #-40]
+	ldrd	A_l, A_h, [src, #-32]	/* 8 words to go.  */
+	strd	A_l, A_h, [dst, #-32]
+	ldrd	A_l, A_h, [src, #-24]	/* 6 words to go.  */
+	strd	A_l, A_h, [dst, #-24]
+	ldrd	A_l, A_h, [src, #-16]	/* 4 words to go.  */
+	strd	A_l, A_h, [dst, #-16]
+	ldrd	A_l, A_h, [src, #-8]	/* 2 words to go.  */
+	strd	A_l, A_h, [dst, #-8]
+
+#endif
+	tst	tmp2, #4
+	ldrne	tmp1, [src], #4
+	strne	tmp1, [dst], #4
+	lsls	tmp2, tmp2, #31		/* Count (tmp2) now dead. */
+	ldrhcs	tmp1, [src], #2
+	ldrbne	tmp2, [src]
+	strhcs	tmp1, [dst], #2
+	strbne	tmp2, [dst]
+
+.Ldone:
+	ldr	tmp2, [sp], #FRAME_SIZE
+	bx	lr
+
+.Lcpy_body_long:			/* Count in tmp2.  */
+
+	/* Long copy.  We know that there's at least (prefetch_lines * 64)
+	   bytes to go.  */
+#ifdef USE_VFP
+	/* Don't use PLD.  Instead, read some data in advance of the current
+	   copy position into a register.  This should act like a PLD
+	   operation but we won't have to repeat the transfer.  */
+
+	vldr	d3, [src, #0]
+	vldr	d4, [src, #64]
+	vldr	d5, [src, #128]
+	vldr	d6, [src, #192]
+	vldr	d7, [src, #256]
+
+	vldr	d0, [src, #8]
+	vldr	d1, [src, #16]
+	vldr	d2, [src, #24]
+	add	src, src, #32
+
+	subs	tmp2, tmp2, #prefetch_lines * 64 * 2
+	blt	2f
+1:
+	cpy_line_vfp	d3, 0
+	cpy_line_vfp	d4, 64
+	cpy_line_vfp	d5, 128
+	add	dst, dst, #3 * 64
+	add	src, src, #3 * 64
+	cpy_line_vfp	d6, 0
+	cpy_line_vfp	d7, 64
+	add	dst, dst, #2 * 64
+	add	src, src, #2 * 64
+	subs	tmp2, tmp2, #prefetch_lines * 64
+	bge	1b
+
+2:
+	cpy_tail_vfp	d3, 0
+	cpy_tail_vfp	d4, 64
+	cpy_tail_vfp	d5, 128
+	add	src, src, #3 * 64
+	add	dst, dst, #3 * 64
+	cpy_tail_vfp	d6, 0
+	vstr	d7, [dst, #64]
+	vldr	d7, [src, #64]
+	vstr	d0, [dst, #64 + 8]
+	vldr	d0, [src, #64 + 8]
+	vstr	d1, [dst, #64 + 16]
+	vldr	d1, [src, #64 + 16]
+	vstr	d2, [dst, #64 + 24]
+	vldr	d2, [src, #64 + 24]
+	vstr	d7, [dst, #64 + 32]
+	add	src, src, #96
+	vstr	d0, [dst, #64 + 40]
+	vstr	d1, [dst, #64 + 48]
+	vstr	d2, [dst, #64 + 56]
+	add	dst, dst, #128
+	add	tmp2, tmp2, #prefetch_lines * 64
+	b	.Lcpy_body_medium
+#else
+	/* Long copy.  Use an SMS style loop to maximize the I/O
+	   bandwidth of the core.  We don't have enough spare registers
+	   to synthesise prefetching, so use PLD operations.  */
+	/* Pre-bias src and dst.  */
+	sub	src, src, #8
+	sub	dst, dst, #8
+	pld	[src, #8]
+	pld	[src, #72]
+	subs	tmp2, tmp2, #64
+	pld	[src, #136]
+	ldrd	A_l, A_h, [src, #8]
+	strd	B_l, B_h, [sp, #8]
+	ldrd	B_l, B_h, [src, #16]
+	strd	C_l, C_h, [sp, #16]
+	ldrd	C_l, C_h, [src, #24]
+	strd	D_l, D_h, [sp, #24]
+	pld	[src, #200]
+	ldrd	D_l, D_h, [src, #32]!
+	b	1f
+	.p2align	6
+2:
+	pld	[src, #232]
+	strd	A_l, A_h, [dst, #40]
+	ldrd	A_l, A_h, [src, #40]
+	strd	B_l, B_h, [dst, #48]
+	ldrd	B_l, B_h, [src, #48]
+	strd	C_l, C_h, [dst, #56]
+	ldrd	C_l, C_h, [src, #56]
+	strd	D_l, D_h, [dst, #64]!
+	ldrd	D_l, D_h, [src, #64]!
+	subs	tmp2, tmp2, #64
+1:
+	strd	A_l, A_h, [dst, #8]
+	ldrd	A_l, A_h, [src, #8]
+	strd	B_l, B_h, [dst, #16]
+	ldrd	B_l, B_h, [src, #16]
+	strd	C_l, C_h, [dst, #24]
+	ldrd	C_l, C_h, [src, #24]
+	strd	D_l, D_h, [dst, #32]
+	ldrd	D_l, D_h, [src, #32]
+	bcs	2b
+	/* Save the remaining bytes and restore the callee-saved regs.  */
+	strd	A_l, A_h, [dst, #40]
+	add	src, src, #40
+	strd	B_l, B_h, [dst, #48]
+	ldrd	B_l, B_h, [sp, #8]
+	strd	C_l, C_h, [dst, #56]
+	ldrd	C_l, C_h, [sp, #16]
+	strd	D_l, D_h, [dst, #64]
+	ldrd	D_l, D_h, [sp, #24]
+	add	dst, dst, #72
+	tst	tmp2, #0x3f
+	bne	.Ltail63aligned
+	ldr	tmp2, [sp], #FRAME_SIZE
+	bx	lr
+#endif
+
+.Lcpy_notaligned:
+	pld	[src]
+	pld	[src, #64]
+	/* There's at least 64 bytes to copy, but there is no mutual
+	   alignment.  */
+	/* Bring DST to 64-bit alignment.  */
+	lsls	tmp2, dst, #29
+	pld	[src, #(2 * 64)]
+	beq	1f
+	rsbs	tmp2, tmp2, #0
+	sub	count, count, tmp2, lsr #29
+	ldrmi	tmp1, [src], #4
+	strmi	tmp1, [dst], #4
+	lsls	tmp2, tmp2, #2
+	ldrbne	tmp1, [src], #1
+	ldrhcs	tmp2, [src], #2
+	strbne	tmp1, [dst], #1
+	strhcs	tmp2, [dst], #2
+1:
+	pld	[src, #(3 * 64)]
+	subs	count, count, #64
+	ldrmi	tmp2, [sp], #FRAME_SIZE
+	bmi	.Ltail63unaligned
+	pld	[src, #(4 * 64)]
+
+#ifdef USE_NEON
+	vld1.8	{d0-d3}, [src]!
+	vld1.8	{d4-d7}, [src]!
+	subs	count, count, #64
+	bmi	2f
+1:
+	pld	[src, #(4 * 64)]
+	vst1.8	{d0-d3}, [ALIGN (dst, 64)]!
+	vld1.8	{d0-d3}, [src]!
+	vst1.8	{d4-d7}, [ALIGN (dst, 64)]!
+	vld1.8	{d4-d7}, [src]!
+	subs	count, count, #64
+	bpl	1b
+2:
+	vst1.8	{d0-d3}, [ALIGN (dst, 64)]!
+	vst1.8	{d4-d7}, [ALIGN (dst, 64)]!
+	ands	count, count, #0x3f
+#else
+	/* Use an SMS style loop to maximize the I/O bandwidth.  */
+	sub	src, src, #4
+	sub	dst, dst, #8
+	subs	tmp2, count, #64	/* Use tmp2 for count.  */
+	ldr	A_l, [src, #4]
+	ldr	A_h, [src, #8]
+	strd	B_l, B_h, [sp, #8]
+	ldr	B_l, [src, #12]
+	ldr	B_h, [src, #16]
+	strd	C_l, C_h, [sp, #16]
+	ldr	C_l, [src, #20]
+	ldr	C_h, [src, #24]
+	strd	D_l, D_h, [sp, #24]
+	ldr	D_l, [src, #28]
+	ldr	D_h, [src, #32]!
+	b	1f
+	.p2align	6
+2:
+	pld	[src, #(5 * 64) - (32 - 4)]
+	strd	A_l, A_h, [dst, #40]
+	ldr	A_l, [src, #36]
+	ldr	A_h, [src, #40]
+	strd	B_l, B_h, [dst, #48]
+	ldr	B_l, [src, #44]
+	ldr	B_h, [src, #48]
+	strd	C_l, C_h, [dst, #56]
+	ldr	C_l, [src, #52]
+	ldr	C_h, [src, #56]
+	strd	D_l, D_h, [dst, #64]!
+	ldr	D_l, [src, #60]
+	ldr	D_h, [src, #64]!
+	subs	tmp2, tmp2, #64
+1:
+	strd	A_l, A_h, [dst, #8]
+	ldr	A_l, [src, #4]
+	ldr	A_h, [src, #8]
+	strd	B_l, B_h, [dst, #16]
+	ldr	B_l, [src, #12]
+	ldr	B_h, [src, #16]
+	strd	C_l, C_h, [dst, #24]
+	ldr	C_l, [src, #20]
+	ldr	C_h, [src, #24]
+	strd	D_l, D_h, [dst, #32]
+	ldr	D_l, [src, #28]
+	ldr	D_h, [src, #32]
+	bcs	2b
+
+	/* Save the remaining bytes and restore the callee-saved regs.  */
+	strd	A_l, A_h, [dst, #40]
+	add	src, src, #36
+	strd	B_l, B_h, [dst, #48]
+	ldrd	B_l, B_h, [sp, #8]
+	strd	C_l, C_h, [dst, #56]
+	ldrd	C_l, C_h, [sp, #16]
+	strd	D_l, D_h, [dst, #64]
+	ldrd	D_l, D_h, [sp, #24]
+	add	dst, dst, #72
+	ands	count, tmp2, #0x3f
+#endif
+	ldr	tmp2, [sp], #FRAME_SIZE
+	bne	.Ltail63unaligned
+	bx	lr
+END(memcpy)

libc/arch-arm/bionic/tgkill.S

@@ -39,6 +39,7 @@
 */
 
 ENTRY(tgkill)
+    .save   {r4-r7, ip, lr}
     stmfd   sp!, {r4-r7, ip, lr}
     ldr     r7, =__NR_tgkill
     swi     #0

libc/arch-arm/cortex-a15/bionic/memcpy.S

@@ -24,81 +24,110 @@
  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-/* Assumes neon instructions and a cache line size of 64 bytes. */
+    /* Prototype: void *memcpy (void *dst, const void *src, size_t count).  */
+
+        // This version is tuned for the Cortex-A15 processor.
 
 #include <machine/cpu-features.h>
 #include <machine/asm.h>
 
-/*
- * This code assumes it is running on a processor that supports all arm v7
- * instructions, that supports neon instructions, and that has a 64 byte
- * cache line.
- */
-
         .text
+        .syntax unified
         .fpu    neon
 
 #define CACHE_LINE_SIZE 64
 
 ENTRY(memcpy)
+        // Assumes that n >= 0, and dst, src are valid pointers.
+        // For any sizes less than 832 use the neon code that doesn't
+        // care about the src alignment. This avoids any checks
+        // for src alignment, and offers the best improvement since
+        // smaller sized copies are dominated by the overhead of
+        // the pre and post main loop.
+        // For larger copies, if src and dst cannot both be aligned to
+        // word boundaries, use the neon code.
+        // For all other copies, align dst to a double word boundary
+        // and copy using LDRD/STRD instructions.
+
+        // Save registers (r0 holds the return value):
+        // optimized push {r0, lr}.
         .save   {r0, lr}
-        /* start preloading as early as possible */
+        pld     [r1, #(CACHE_LINE_SIZE*16)]
-        pld         [r1, #(CACHE_LINE_SIZE*0)]
+        push    {r0, lr}
-        stmfd       sp!, {r0, lr}
-        pld         [r1, #(CACHE_LINE_SIZE*1)]
 
-        /* do we have at least 16-bytes to copy (needed for alignment below) */
         cmp     r2, #16
-        blo         5f
+        blo     copy_less_than_16_unknown_align
 
-        /* align destination to cache-line for the write-buffer */
+        cmp     r2, #832
+        bge     check_alignment
+
+copy_unknown_alignment:
+        // Unknown alignment of src and dst.
+        // Assumes that the first few bytes have already been prefetched.
+
+        // Align destination to 128 bits. The mainloop store instructions
+        // require this alignment or they will throw an exception.
         rsb         r3, r0, #0
         ands        r3, r3, #0xF
-        beq         0f
+        beq         2f
 
-        /* copy up to 15-bytes (count in r3) */
+        // Copy up to 15 bytes (count in r3).
         sub         r2, r2, r3
         movs        ip, r3, lsl #31
-        ldrmib      lr, [r1], #1
+
-        strmib      lr, [r0], #1
+        itt         mi
-        ldrcsb      ip, [r1], #1
+        ldrbmi      lr, [r1], #1
-        ldrcsb      lr, [r1], #1
+        strbmi      lr, [r0], #1
-        strcsb      ip, [r0], #1
+        itttt       cs
-        strcsb      lr, [r0], #1
+        ldrbcs      ip, [r1], #1
+        ldrbcs      lr, [r1], #1
+        strbcs      ip, [r0], #1
+        strbcs      lr, [r0], #1
+
         movs        ip, r3, lsl #29
         bge         1f
-        // copies 4 bytes, destination 32-bits aligned
+        // Copies 4 bytes, dst 32 bits aligned before, at least 64 bits after.
         vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]!
         vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0, :32]!
 1:      bcc         2f
-        // copies 8 bytes, destination 64-bits aligned
+        // Copies 8 bytes, dst 64 bits aligned before, at least 128 bits after.
         vld1.8      {d0}, [r1]!
         vst1.8      {d0}, [r0, :64]!
-2:
 
-0:      /* preload immediately the next cache line, which we may need */
+2:      // Make sure we have at least 64 bytes to copy.
-        pld         [r1, #(CACHE_LINE_SIZE*0)]
-        pld         [r1, #(CACHE_LINE_SIZE*1)]
-
-        /* make sure we have at least 64 bytes to copy */
         subs        r2, r2, #64
         blo         2f
 
-        /* Preload all the cache lines we need.
+1:      // The main loop copies 64 bytes at a time.
-         * NOTE: The number of pld below depends on CACHE_LINE_SIZE,
-         * ideally we would increase the distance in the main loop to
-         * avoid the goofy code below. In practice this doesn't seem to make
-         * a big difference.
-         * NOTE: The value CACHE_LINE_SIZE * 4 was chosen through
-         * experimentation.
-         */
-        pld         [r1, #(CACHE_LINE_SIZE*2)]
-        pld         [r1, #(CACHE_LINE_SIZE*3)]
-        pld         [r1, #(CACHE_LINE_SIZE*4)]
-
-1:      /* The main loop copies 64 bytes at a time */
         vld1.8      {d0  - d3},   [r1]!
         vld1.8      {d4  - d7},   [r1]!
         pld         [r1, #(CACHE_LINE_SIZE*4)]
@@ -107,25 +136,24 @@ ENTRY(memcpy)
         vst1.8      {d4  - d7},   [r0, :128]!
         bhs         1b
 
-2:      /* fix-up the remaining count and make sure we have >= 32 bytes left */
+2:      // Fix-up the remaining count and make sure we have >= 32 bytes left.
-        add         r2, r2, #64
+        adds        r2, r2, #32
-        subs        r2, r2, #32
+        blo         3f
-        blo         4f
 
-3:      /* 32 bytes at a time. These cache lines were already preloaded */
+        // 32 bytes. These cache lines were already preloaded.
         vld1.8      {d0 - d3},  [r1]!
-        subs        r2, r2, #32
+        sub         r2, r2, #32
         vst1.8      {d0 - d3},  [r0, :128]!
-        bhs         3b
+3:      // Less than 32 left.
-4:      /* less than 32 left */
         add         r2, r2, #32
         tst         r2, #0x10
-        beq         5f
+        beq         copy_less_than_16_unknown_align
-        // copies 16 bytes, 128-bits aligned
+        // Copies 16 bytes, destination 128 bits aligned.
         vld1.8      {d0, d1}, [r1]!
         vst1.8      {d0, d1}, [r0, :128]!
 
-5:      /* copy up to 15-bytes (count in r2) */
+copy_less_than_16_unknown_align:
+        // Copy up to 15 bytes (count in r2).
         movs        ip, r2, lsl #29
         bcc         1f
         vld1.8      {d0}, [r1]!
@@ -133,14 +161,164 @@ ENTRY(memcpy)
 1:      bge         2f
         vld4.8      {d0[0], d1[0], d2[0], d3[0]}, [r1]!
         vst4.8      {d0[0], d1[0], d2[0], d3[0]}, [r0]!
-2:      movs        ip, r2, lsl #31
-        ldrmib      r3, [r1], #1
-        ldrcsb      ip, [r1], #1
-        ldrcsb      lr, [r1], #1
-        strmib      r3, [r0], #1
-        strcsb      ip, [r0], #1
-        strcsb      lr, [r0], #1
 
-        ldmfd       sp!, {r0, lr}
+2:      // Copy 0 to 4 bytes.
-        bx          lr
+        lsls        r2, r2, #31
+        itt         ne
+        ldrbne      lr, [r1], #1
+        strbne      lr, [r0], #1
+        itttt       cs
+        ldrbcs      ip, [r1], #1
+        ldrbcs      lr, [r1]
+        strbcs      ip, [r0], #1
+        strbcs      lr, [r0]
+
+        pop         {r0, pc}
+
+check_alignment:
+        // If src and dst cannot both be aligned to a word boundary,
+        // use the unaligned copy version.
+        eor     r3, r0, r1
+        ands    r3, r3, #0x3
+        bne     copy_unknown_alignment
+
+        // To try and improve performance, stack layout changed,
+        // i.e., not keeping the stack looking like users expect
+        // (highest numbered register at highest address).
+        // TODO: Add debug frame directives.
+        // We don't need exception unwind directives, because the code below
+        // does not throw any exceptions and does not call any other functions.
+        // Generally, newlib functions like this lack debug information for
+        // assembler source.
+        .save   {r4, r5}
+        strd    r4, r5, [sp, #-8]!
+        .save   {r6, r7}
+        strd    r6, r7, [sp, #-8]!
+        .save   {r8, r9}
+        strd    r8, r9, [sp, #-8]!
+
+        // Optimized for already aligned dst code.
+        ands    ip, r0, #3
+        bne     dst_not_word_aligned
+
+word_aligned:
+        // Align the destination buffer to 8 bytes, to make sure double
+        // loads and stores don't cross a cache line boundary,
+        // as they are then more expensive even if the data is in the cache
+        // (require two load/store issue cycles instead of one).
+        // If only one of the buffers is not 8 bytes aligned,
+        // then it's more important to align dst than src,
+        // because there is more penalty for stores
+        // than loads that cross a cacheline boundary.
+        // This check and realignment are only done if there is >= 832
+        // bytes to copy.
+
+        // Dst is word aligned, but check if it is already double word aligned.
+        ands    r3, r0, #4
+        beq     1f
+        ldr     r3, [r1], #4
+        str     r3, [r0], #4
+        sub     r2, #4
+
+1:      // Can only get here if > 64 bytes to copy, so don't do check r2.
+        sub     r2, #64
+
+2:      // Every loop iteration copies 64 bytes.
+        .irp    offset, #0, #8, #16, #24, #32
+        ldrd    r4, r5, [r1, \offset]
+        strd    r4, r5, [r0, \offset]
+        .endr
+
+        ldrd    r4, r5, [r1, #40]
+        ldrd    r6, r7, [r1, #48]
+        ldrd    r8, r9, [r1, #56]
+
+        // Keep the pld as far from the next load as possible.
+        // The amount to prefetch was determined experimentally using
+        // large sizes, and verifying the prefetch size does not affect
+        // the smaller copies too much.
+        // WARNING: If the ldrd and strd instructions get too far away
+        //          from each other, performance suffers. Three loads
+        //          in a row is the best tradeoff.
+        pld     [r1, #(CACHE_LINE_SIZE*16)]
+        strd    r4, r5, [r0, #40]
+        strd    r6, r7, [r0, #48]
+        strd    r8, r9, [r0, #56]
+
+        add     r0, r0, #64
+        add     r1, r1, #64
+        subs    r2, r2, #64
+        bge     2b
+
+        // Fix-up the remaining count and make sure we have >= 32 bytes left.
+        adds    r2, r2, #32
+        blo     4f
+
+        // Copy 32 bytes. These cache lines were already preloaded.
+        .irp    offset, #0, #8, #16, #24
+        ldrd    r4, r5, [r1, \offset]
+        strd    r4, r5, [r0, \offset]
+        .endr
+        add     r1, r1, #32
+        add     r0, r0, #32
+        sub     r2, r2, #32
+4:      // Less than 32 left.
+        add     r2, r2, #32
+        tst     r2, #0x10
+        beq     5f
+        // Copy 16 bytes.
+        .irp    offset, #0, #8
+        ldrd    r4, r5, [r1, \offset]
+        strd    r4, r5, [r0, \offset]
+        .endr
+        add     r1, r1, #16
+        add     r0, r0, #16
+
+5:      // Copy up to 15 bytes (count in r2).
+        movs    ip, r2, lsl #29
+        bcc     1f
+        // Copy 8 bytes.
+        ldrd    r4, r5, [r1], #8
+        strd    r4, r5, [r0], #8
+1:      bge         2f
+        // Copy 4 bytes.
+        ldr     r4, [r1], #4
+        str     r4, [r0], #4
+2:      // Copy 0 to 4 bytes.
+        lsls    r2, r2, #31
+        itt     ne
+        ldrbne  lr, [r1], #1
+        strbne  lr, [r0], #1
+        itttt   cs
+        ldrbcs  ip, [r1], #1
+        ldrbcs  lr, [r1]
+        strbcs  ip, [r0], #1
+        strbcs  lr, [r0]
+
+        // Restore registers: optimized pop {r0, pc}
+        ldrd    r8, r9, [sp], #8
+        ldrd    r6, r7, [sp], #8
+        ldrd    r4, r5, [sp], #8
+        pop     {r0, pc}
+
+dst_not_word_aligned:
+        // Align dst to word.
+        rsb     ip, ip, #4
+        cmp     ip, #2
+
+        itt     gt
+        ldrbgt  lr, [r1], #1
+        strbgt  lr, [r0], #1
+
+        itt     ge
+        ldrbge  lr, [r1], #1
+        strbge  lr, [r0], #1
+
+        ldrb    lr, [r1], #1
+        strb    lr, [r0], #1
+
+        sub     r2, r2, ip
+
+        // Src is guaranteed to be at least word aligned by this point.
+        b       word_aligned
 END(memcpy)

libc/arch-arm/cortex-a15/bionic/memset.S

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008 The Android Open Source Project
+ * Copyright (C) 2013 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -36,6 +36,7 @@
 		 */
 
         .fpu        neon
+        .syntax     unified
 
 ENTRY(bzero)
         mov         r2, r1
@@ -47,60 +48,117 @@ ENTRY(memset)
         .save       {r0}
         stmfd       sp!, {r0}
 
-        vdup.8      q0, r1
+        // The new algorithm is slower for copies < 16 so use the old
-
+        // neon code in that case.
-        /* do we have at least 16-bytes to write (needed for alignment below) */
         cmp         r2, #16
-        blo         3f
+        blo         set_less_than_16_unknown_align
 
-        /* align destination to 16 bytes for the write-buffer */
+        // Use strd which requires an even and odd register so move the
-        rsb         r3, r0, #0
+        // values so that:
-        ands        r3, r3, #0xF
+        //   r0 and r1 contain the memset value
-        beq         2f
+        //   r2 is the number of bytes to set
+        //   r3 is the destination pointer
+        mov         r3, r0
 
-        /* write up to 15-bytes (count in r3) */
+        // Copy the byte value in every byte of r1.
-        sub         r2, r2, r3
+        mov         r1, r1, lsl #24
-        movs        ip, r3, lsl #31
+        orr         r1, r1, r1, lsr #8
-        strmib      r1, [r0], #1
+        orr         r1, r1, r1, lsr #16
-        strcsb      r1, [r0], #1
-        strcsb      r1, [r0], #1
-        movs        ip, r3, lsl #29
-        bge         1f
 
-        // writes 4 bytes, 32-bits aligned
+check_alignment:
-        vst1.32     {d0[0]}, [r0, :32]!
+        // Align destination to a double word to avoid the strd crossing
-1:      bcc         2f
+        // a cache line boundary.
+        ands        ip, r3, #7
+        bne         do_double_word_align
 
-        // writes 8 bytes, 64-bits aligned
+double_word_aligned:
-        vst1.8      {d0}, [r0, :64]!
+        mov         r0, r1
-2:
-        /* make sure we have at least 32 bytes to write */
-        subs        r2, r2, #32
-        blo         2f
-        vmov        q1, q0
 
-1:      /* The main loop writes 32 bytes at a time */
+        subs        r2, #64
-        subs        r2, r2, #32
+        blo         set_less_than_64
-        vst1.8      {d0 - d3}, [r0, :128]!
-        bhs         1b
 
-2:      /* less than 32 left */
+1:      // Main loop sets 64 bytes at a time.
-        add         r2, r2, #32
+        .irp        offset, #0, #8, #16, #24, #32, #40, #48, #56
-        tst         r2, #0x10
+        strd        r0, r1, [r3, \offset]
-        beq         3f
+        .endr
 
-        // writes 16 bytes, 128-bits aligned
+        add         r3, #64
-        vst1.8      {d0, d1}, [r0, :128]!
+        subs        r2, #64
-3:      /* write up to 15-bytes (count in r2) */
+        bge         1b
+
+set_less_than_64:
+        // Restore r2 to the count of bytes left to set.
+        add         r2, #64
+        lsls        ip, r2, #27
+        bcc         set_less_than_32
+        // Set 32 bytes.
+        .irp        offset, #0, #8, #16, #24
+        strd        r0, r1, [r3, \offset]
+        .endr
+        add         r3, #32
+
+set_less_than_32:
+        bpl         set_less_than_16
+        // Set 16 bytes.
+        .irp        offset, #0, #8
+        strd        r0, r1, [r3, \offset]
+        .endr
+        add         r3, #16
+
+set_less_than_16:
+        // Less than 16 bytes to set.
+        lsls        ip, r2, #29
+        bcc         set_less_than_8
+
+        // Set 8 bytes.
+        strd        r0, r1, [r3], #8
+
+set_less_than_8:
+        bpl         set_less_than_4
+        // Set 4 bytes
+        str         r1, [r3], #4
+
+set_less_than_4:
+        lsls        ip, r2, #31
+        it          ne
+        strbne      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3]
+
+        ldmfd       sp!, {r0}
+        bx          lr
+
+do_double_word_align:
+        rsb         ip, ip, #8
+        sub         r2, r2, ip
+        movs        r0, ip, lsl #31
+        it          mi
+        strbmi      r1, [r3], #1
+        itt         cs
+        strbcs      r1, [r3], #1
+        strbcs      r1, [r3], #1
+
+        // Dst is at least word aligned by this point.
+        cmp         ip, #4
+        blo         double_word_aligned
+        str         r1, [r3], #4
+        b           double_word_aligned
+
+set_less_than_16_unknown_align:
+        // Set up to 15 bytes.
+        vdup.8      d0, r1
         movs        ip, r2, lsl #29
         bcc         1f
         vst1.8      {d0}, [r0]!
 1:      bge         2f
         vst1.32     {d0[0]}, [r0]!
 2:      movs        ip, r2, lsl #31
-        strmib      r1, [r0], #1
+        it          mi
-        strcsb      r1, [r0], #1
+        strbmi      r1, [r0], #1
-        strcsb      r1, [r0], #1
+        itt         cs
+        strbcs      r1, [r0], #1
+        strbcs      r1, [r0], #1
         ldmfd       sp!, {r0}
         bx          lr
 END(memset)

libc/arch-arm/cortex-a15/bionic/strlen.S

@@ -0,0 +1,151 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+/*
+ * Copyright (c) 2013 ARM Ltd
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the company may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+ * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <machine/asm.h>
+
+    .syntax unified
+
+    .thumb
+    .thumb_func
+
+ENTRY(strlen)
+    pld [r1, #128]
+    mov r1, r0
+
+    rsb     r3, r0, #0
+    ands    r3, r3, #7
+    beq     mainloop
+
+    // Align to a double word (64 bits).
+    ands    ip, r3, #1
+    beq     align_to_32
+
+    ldrb    r2, [r1], #1
+    cmp     r2, #0
+    beq     update_count_and_return
+
+align_to_32:
+    ands    ip, r3, #2
+    beq     align_to_64
+
+    ldrb    r2, [r1], #1
+    cmp     r2, #0
+    beq     update_count_and_return
+    ldrb    r2, [r1], #1
+    cmp     r2, #0
+    beq     update_count_and_return
+
+align_to_64:
+    ands    ip, r3, #4
+    beq     mainloop
+    ldr     r3, [r1], #4
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     zero_in_second_register
+
+mainloop:
+    ldrd    r2, r3, [r1], #8
+
+    pld     [r1, #64]
+
+    sub     ip, r2, #0x01010101
+    bic     ip, ip, r2
+    ands    ip, ip, #0x80808080
+    bne     zero_in_first_register
+
+    sub     ip, r3, #0x01010101
+    bic     ip, ip, r3
+    ands    ip, ip, #0x80808080
+    bne     zero_in_second_register
+    b       mainloop
+
+zero_in_first_register:
+    sub     r1, r1, #4
+
+zero_in_second_register:
+    sub     r0, r1, r0
+
+    // Check for zero in byte 0.
+    ands    r1, ip, #0x80
+    beq     check_byte1
+
+    sub     r0, r0, #4
+    bx      lr
+
+check_byte1:
+    // Check for zero in byte 1.
+    ands    r1, ip, #0x8000
+    beq     check_byte2
+
+    sub     r0, r0, #3
+    bx      lr
+
+check_byte2:
+    // Check for zero in byte 2.
+    ands    r1, ip, #0x800000
+    beq     return
+
+    sub     r0, r0, #2
+    bx      lr
+
+update_count_and_return:
+    sub     r0, r1, r0
+
+return:
+    sub     r0, r0, #1
+    bx      lr
+END(strlen)

libc/arch-arm/cortex-a15/cortex-a15.mk

@@ -1,5 +1,6 @@
 $(call libc-add-cpu-variant-src,MEMCPY,arch-arm/cortex-a15/bionic/memcpy.S)
 $(call libc-add-cpu-variant-src,MEMSET,arch-arm/cortex-a15/bionic/memset.S)
 $(call libc-add-cpu-variant-src,STRCMP,arch-arm/cortex-a15/bionic/strcmp.S)
+$(call libc-add-cpu-variant-src,STRLEN,arch-arm/cortex-a15/bionic/strlen.S)
 
 include bionic/libc/arch-arm/generic/generic.mk

libc/arch-arm/cortex-a9/cortex-a9.mk

@@ -1,5 +1,7 @@
 $(call libc-add-cpu-variant-src,MEMCPY,arch-arm/cortex-a9/bionic/memcpy.S)
 $(call libc-add-cpu-variant-src,MEMSET,arch-arm/cortex-a9/bionic/memset.S)
 $(call libc-add-cpu-variant-src,STRCMP,arch-arm/cortex-a9/bionic/strcmp.S)
+# Use cortex-a15 version of strlen.
+$(call libc-add-cpu-variant-src,STRLEN,arch-arm/cortex-a15/bionic/strlen.S)
 
 include bionic/libc/arch-arm/generic/generic.mk

libc/arch-arm/generic/bionic/strlen.c

@@ -60,52 +60,60 @@ size_t strlen(const char *s)
     // We need to process 32 bytes per loop to schedule PLD properly
     // and achieve the maximum bus speed.
     asm(
-        "ldr     %[v], [ %[s] ], #4         \n"
+        "ldr     %[v], [%[s]], #4           \n"
         "sub     %[l], %[l], %[s]           \n"
         "0:                                 \n"
 #if __ARM_HAVE_PLD
-        "pld     [ %[s], #64 ]              \n"
+        "pld     [%[s], #64]                \n"
 #endif
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
 #if !defined(__OPTIMIZE_SIZE__)
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
         "bne     1f                         \n"
         "sub     %[t], %[v], %[mask], lsr #7\n"
         "and     %[t], %[t], %[mask]        \n"
         "bics    %[t], %[t], %[v]           \n"
-        "ldreq   %[v], [ %[s] ], #4         \n"
+        "it      eq                         \n"
+        "ldreq   %[v], [%[s]], #4           \n"
 #endif
         "beq     0b                         \n"
         "1:                                 \n"
@@ -117,6 +125,7 @@ size_t strlen(const char *s)
         "beq     2f                         \n"
         "add     %[l], %[l], #1             \n"
         "tst     %[v], #0xFF0000            \n"
+        "it      ne                         \n"
         "addne   %[l], %[l], #1             \n"
         "2:                                 \n"
         : [l]"=&r"(l), [v]"=&r"(v), [t]"=&r"(t), [s]"=&r"(u.b)

libc/arch-arm/generic/generic.mk

@@ -1,3 +1,4 @@
 $(call libc-add-cpu-variant-src,MEMCPY,arch-arm/generic/bionic/memcpy.S)
 $(call libc-add-cpu-variant-src,MEMSET,arch-arm/generic/bionic/memset.S)
 $(call libc-add-cpu-variant-src,STRCMP,arch-arm/generic/bionic/strcmp.S)
+$(call libc-add-cpu-variant-src,STRLEN,arch-arm/generic/bionic/strlen.c)

libc/arch-arm/krait/krait.mk

@@ -1,5 +1,7 @@
 $(call libc-add-cpu-variant-src,MEMCPY,arch-arm/krait/bionic/memcpy.S)
 $(call libc-add-cpu-variant-src,MEMSET,arch-arm/krait/bionic/memset.S)
 $(call libc-add-cpu-variant-src,STRCMP,arch-arm/krait/bionic/strcmp.S)
+# Use cortex-a15 version of strlen.
+$(call libc-add-cpu-variant-src,STRLEN,arch-arm/cortex-a15/bionic/strlen.S)
 
 include bionic/libc/arch-arm/generic/generic.mk

libc/arch-arm/syscalls.mk

@@ -126,6 +126,8 @@ syscall_src += arch-arm/syscalls/removexattr.S
 syscall_src += arch-arm/syscalls/lremovexattr.S
 syscall_src += arch-arm/syscalls/__statfs64.S
 syscall_src += arch-arm/syscalls/unshare.S
+syscall_src += arch-arm/syscalls/swapon.S
+syscall_src += arch-arm/syscalls/swapoff.S
 syscall_src += arch-arm/syscalls/pause.S
 syscall_src += arch-arm/syscalls/gettimeofday.S
 syscall_src += arch-arm/syscalls/settimeofday.S
@@ -144,6 +146,9 @@ syscall_src += arch-arm/syscalls/__timer_getoverrun.S
 syscall_src += arch-arm/syscalls/__timer_delete.S
 syscall_src += arch-arm/syscalls/utimes.S
 syscall_src += arch-arm/syscalls/utimensat.S
+syscall_src += arch-arm/syscalls/timerfd_create.S
+syscall_src += arch-arm/syscalls/timerfd_settime.S
+syscall_src += arch-arm/syscalls/timerfd_gettime.S
 syscall_src += arch-arm/syscalls/sigaction.S
 syscall_src += arch-arm/syscalls/sigprocmask.S
 syscall_src += arch-arm/syscalls/__sigsuspend.S

libc/arch-arm/syscalls/swapoff.S

@@ -0,0 +1,15 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+#include <linux/err.h>
+#include <machine/asm.h>
+
+ENTRY(swapoff)
+    mov     ip, r7
+    ldr     r7, =__NR_swapoff
+    swi     #0
+    mov     r7, ip
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno
+END(swapoff)

libc/arch-arm/syscalls/swapon.S

@@ -0,0 +1,15 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+#include <linux/err.h>
+#include <machine/asm.h>
+
+ENTRY(swapon)
+    mov     ip, r7
+    ldr     r7, =__NR_swapon
+    swi     #0
+    mov     r7, ip
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno
+END(swapon)

libc/arch-arm/syscalls/timerfd_create.S

@@ -0,0 +1,15 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+#include <linux/err.h>
+#include <machine/asm.h>
+
+ENTRY(timerfd_create)
+    mov     ip, r7
+    ldr     r7, =__NR_timerfd_create
+    swi     #0
+    mov     r7, ip
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno
+END(timerfd_create)

libc/arch-arm/syscalls/timerfd_gettime.S

@@ -0,0 +1,15 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+#include <linux/err.h>
+#include <machine/asm.h>
+
+ENTRY(timerfd_gettime)
+    mov     ip, r7
+    ldr     r7, =__NR_timerfd_gettime
+    swi     #0
+    mov     r7, ip
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno
+END(timerfd_gettime)

libc/arch-arm/syscalls/timerfd_settime.S

@@ -0,0 +1,15 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+#include <linux/err.h>
+#include <machine/asm.h>
+
+ENTRY(timerfd_settime)
+    mov     ip, r7
+    ldr     r7, =__NR_timerfd_settime
+    swi     #0
+    mov     r7, ip
+    cmn     r0, #(MAX_ERRNO + 1)
+    bxls    lr
+    neg     r0, r0
+    b       __set_errno
+END(timerfd_settime)

libc/arch-mips/bionic/atexit.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2010 The Android Open Source Project
+ * Copyright (C) 2013 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -26,13 +26,11 @@
  * SUCH DAMAGE.
  */
 
-# The __dso_handle global variable is used by static
+extern void *__dso_handle;
-# C++ constructors and destructors in the binary.
+extern int __cxa_atexit(void (*func)(void *), void *arg, void *dso);
-# See http://www.codesourcery.com/public/cxx-abi/abi.html#dso-dtor
+
-#
+__attribute__ ((visibility ("hidden")))
-	.data
+int atexit(void (*func)(void))
-        .align 4
+{
-	.hidden __dso_handle
+  return (__cxa_atexit((void (*)(void *))func, (void *)0, &__dso_handle));
-        .globl __dso_handle
+}
-__dso_handle:
-        .long __dso_handle

libc/arch-mips/bionic/crtbegin.S

@@ -1,146 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-	.text
-	.align 4
-	.type __start,@function
-	.globl __start
-	.globl _start
-
-# this is the small startup code that is first run when
-# any executable that is statically-linked with Bionic
-# runs.
-#
-# it's purpose is to call __libc_init with appropriate
-# arguments, which are:
-#
-#    - the address of the raw data block setup by the Linux
-#      kernel ELF loader
-#
-#    - address of an "onexit" function, not used on any
-#      platform supported by Bionic
-#
-#    - address of the "main" function of the program.
-#
-#    - address of the constructor list
-#
-
-	.ent	__start
-__start:	
-_start:
-	bal	1f
-1:
-	.set	noreorder
-	.cpload	$ra
-	.set	reorder
-
-	move	$a0, $sp
-	move	$a1, $0
-	la	$a2, main
-	la	$a3, 1f
-	subu	$sp, 32
-	la	$t9, __libc_init
-	j	$t9
-	.end	__start
-
-1:	.long	__PREINIT_ARRAY__
-	.long	__INIT_ARRAY__
-	.long	__FINI_ARRAY__
-
-	.section .preinit_array, "aw"
-	.type __PREINIT_ARRAY__, @object
-	.globl __PREINIT_ARRAY__
-__PREINIT_ARRAY__:
-	.long -1
-
-	.section .init_array, "aw"
-	.type __INIT_ARRAY__, @object
-	.globl __INIT_ARRAY__
-__INIT_ARRAY__:
-	.long -1
-
-	.section .fini_array, "aw"
-	.type __FINI_ARRAY__, @object
-	.globl __FINI_ARRAY__
-__FINI_ARRAY__:
-	.long -1
-	.long __do_global_dtors_aux
-
-	.abicalls
-	.text
-	.align	2
-	.set	nomips16
-	.ent	__do_global_dtors_aux
-	.type	__do_global_dtors_aux, @function
-__do_global_dtors_aux:
-	.frame	$sp,32,$31		# vars= 0, regs= 1/0, args= 16, gp= 8
-	.mask	0x80000000,-4
-	.fmask	0x00000000,0
-	.set	noreorder
-	.cpload	$25
-	.set	nomacro
-	addiu	$sp,$sp,-32
-	sw	$31,28($sp)
-	.cprestore	16
-	lw	$2,%got(completed.1269)($28)
-	lbu	$2,%lo(completed.1269)($2)
-	bne	$2,$0,$L8
-	nop
-
-$L4:
-	lw	$2,%got(__cxa_finalize)($28)
-	beq	$2,$0,$L6
-	nop
-
-	lw	$2,%got(__dso_handle)($28)
-	lw	$4,0($2)
-	lw	$25,%call16(__cxa_finalize)($28)
-	.reloc	1f,R_MIPS_JALR,__cxa_finalize
-1:	jalr	$25
-	nop
-
-	lw	$28,16($sp)
-$L6:
-	lw	$2,%got(completed.1269)($28)
-	li	$3,1			# 0x1
-	sb	$3,%lo(completed.1269)($2)
-$L8:
-	lw	$31,28($sp)
-	addiu	$sp,$sp,32
-	j	$31
-	nop
-
-	.set	macro
-	.set	reorder
-	.end	__do_global_dtors_aux
-	.size	__do_global_dtors_aux, .-__do_global_dtors_aux
-	.local	completed.1269
-	.comm	completed.1269,1,1
-	.weak	__cxa_finalize
-
-#include "__dso_handle.S"
-#include "atexit.S"

libc/arch-mips/bionic/crtbegin.c

@@ -0,0 +1,94 @@
+/*
+ * Copyright (C) 2013 The Android Open Source Project
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *  * Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in
+ *    the documentation and/or other materials provided with the
+ *    distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
+ * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
+ * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include "../../bionic/libc_init_common.h"
+#include <stddef.h>
+#include <stdint.h>
+
+__attribute__ ((section (".preinit_array")))
+void (*__PREINIT_ARRAY__)(void) = (void (*)(void)) -1;
+
+__attribute__ ((section (".init_array")))
+void (*__INIT_ARRAY__)(void) = (void (*)(void)) -1;
+
+__attribute__ ((section (".fini_array")))
+void (*__FINI_ARRAY__)(void) = (void (*)(void)) -1;
+
+
+__LIBC_HIDDEN__  void do_mips_start(void *raw_args) {
+  structors_array_t array;
+  array.preinit_array = &__PREINIT_ARRAY__;
+  array.init_array = &__INIT_ARRAY__;
+  array.fini_array = &__FINI_ARRAY__;
+
+  __libc_init(raw_args, NULL, &main, &array);
+}
+
+/*
+ * This function prepares the return address with a branch-and-link
+ * instruction (bal) and then uses a .cpload to compute the Global
+ * Offset Table (GOT) pointer ($gp). The $gp is then used to load
+ * the address of _do_start() into $t9 just before calling it.
+ * Terminating the stack with a NULL return address.
+ */
+__asm__ (
+"       .set push                   \n"
+"                                   \n"
+"       .text                       \n"
+"       .align  4                   \n"
+"       .type __start,@function     \n"
+"       .globl __start              \n"
+"       .globl  _start              \n"
+"                                   \n"
+"       .ent    __start             \n"
+"__start:                           \n"
+" _start:                           \n"
+"       .frame   $sp,32,$ra         \n"
+"       .mask   0x80000000,-4       \n"
+"                                   \n"
+"       .set noreorder              \n"
+"       bal     1f                  \n"
+"       nop                         \n"
+"1:                                 \n"
+"       .cpload $ra                 \n"
+"       .set reorder                \n"
+"                                   \n"
+"       move    $a0, $sp            \n"
+"       addiu   $sp, $sp, (-32)     \n"
+"       sw      $0, 28($sp)         \n"
+"       la      $t9, do_mips_start  \n"
+"       jalr    $t9                 \n"
+"                                   \n"
+"2:     b       2b                  \n"
+"       .end    __start             \n"
+"                                   \n"
+"       .set pop                    \n"
+);
+
+#include "__dso_handle.h"
+#include "atexit.h"

libc/arch-mips/bionic/crtbegin_so.S

@@ -1,94 +0,0 @@
-/*
- * Copyright (C) 2008 The Android Open Source Project
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- *  * Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- *  * Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in
- *    the documentation and/or other materials provided with the
- *    distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
- * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
- * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
- * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
- * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
- */
-	.section .init_array, "aw"
-	.type __INIT_ARRAY__, @object
-	.globl __INIT_ARRAY__
-__INIT_ARRAY__:
-	.long -1
-
-	.section .fini_array, "aw"
-	.type __FINI_ARRAY__, @object
-	.globl __FINI_ARRAY__
-__FINI_ARRAY__:
-	.long -1
-	.long __do_global_dtors_aux
-
-	.abicalls
-	.text
-	.align	2
-	.set	nomips16
-	.ent	__do_global_dtors_aux
-	.type	__do_global_dtors_aux, @function
-__do_global_dtors_aux:
-	.frame	$sp,32,$31		# vars= 0, regs= 1/0, args= 16, gp= 8
-	.mask	0x80000000,-4
-	.fmask	0x00000000,0
-	.set	noreorder
-	.cpload	$25
-	.set	nomacro
-	addiu	$sp,$sp,-32
-	sw	$31,28($sp)
-	.cprestore	16
-	lw	$2,%got(completed.1269)($28)
-	lbu	$2,%lo(completed.1269)($2)
-	bne	$2,$0,$L8
-	nop
-
-$L4:
-	lw	$2,%got(__cxa_finalize)($28)
-	beq	$2,$0,$L6
-	nop
-
-	lw	$2,%got(__dso_handle)($28)
-	lw	$4,0($2)
-	lw	$25,%call16(__cxa_finalize)($28)
-	.reloc	1f,R_MIPS_JALR,__cxa_finalize
-1:	jalr	$25
-	nop
-
-	lw	$28,16($sp)
-$L6:
-	lw	$2,%got(completed.1269)($28)
-	li	$3,1			# 0x1
-	sb	$3,%lo(completed.1269)($2)
-$L8:
-	lw	$31,28($sp)
-	addiu	$sp,$sp,32
-	j	$31
-	nop
-
-	.set	macro
-	.set	reorder
-	.end	__do_global_dtors_aux
-	.size	__do_global_dtors_aux, .-__do_global_dtors_aux
-	.local	completed.1269
-	.comm	completed.1269,1,1
-	.weak	__cxa_finalize
-
-#include "__dso_handle_so.S"
-#include "atexit.S"

libc/arch-mips/bionic/crtbegin_so.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008 The Android Open Source Project
+ * Copyright (C) 2013 The Android Open Source Project
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -25,19 +25,14 @@
  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
-	.text
+
-	.globl	atexit
+extern void __cxa_finalize(void *);
-	.hidden	atexit
+extern void *__dso_handle;
-	.type	atexit, @function
+
-	.align  4
+__attribute__((visibility("hidden"),destructor))
-	.ent	atexit
+void __on_dlclose() {
-atexit:
+  __cxa_finalize(&__dso_handle);
-	.set	noreorder
+}
-	.cpload	$t9
+
-	.set	reorder
+#include "__dso_handle_so.h"
-	la	$t9, __cxa_atexit
+#include "atexit.h"
-	move	$a1, $0
-	la      $a2, __dso_handle
-	j	$t9
-	.size	atexit, .-atexit
-	.end	atexit

libc/arch-mips/syscalls.mk

@@ -129,6 +129,8 @@ syscall_src += arch-mips/syscalls/removexattr.S
 syscall_src += arch-mips/syscalls/lremovexattr.S
 syscall_src += arch-mips/syscalls/__statfs64.S
 syscall_src += arch-mips/syscalls/unshare.S
+syscall_src += arch-mips/syscalls/swapon.S
+syscall_src += arch-mips/syscalls/swapoff.S
 syscall_src += arch-mips/syscalls/pause.S
 syscall_src += arch-mips/syscalls/gettimeofday.S
 syscall_src += arch-mips/syscalls/settimeofday.S
@@ -147,6 +149,9 @@ syscall_src += arch-mips/syscalls/__timer_getoverrun.S
 syscall_src += arch-mips/syscalls/__timer_delete.S
 syscall_src += arch-mips/syscalls/utimes.S
 syscall_src += arch-mips/syscalls/utimensat.S
+syscall_src += arch-mips/syscalls/timerfd_create.S
+syscall_src += arch-mips/syscalls/timerfd_settime.S
+syscall_src += arch-mips/syscalls/timerfd_gettime.S
 syscall_src += arch-mips/syscalls/sigaction.S
 syscall_src += arch-mips/syscalls/sigprocmask.S
 syscall_src += arch-mips/syscalls/__sigsuspend.S

libc/arch-mips/syscalls/swapoff.S

@@ -0,0 +1,22 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+    .text
+    .globl swapoff
+    .align 4
+    .ent swapoff
+
+swapoff:
+    .set noreorder
+    .cpload $t9
+    li $v0, __NR_swapoff
+    syscall
+    bnez $a3, 1f
+    move $a0, $v0
+    j $ra
+    nop
+1:
+    la $t9,__set_errno
+    j $t9
+    nop
+    .set reorder
+    .end swapoff

libc/arch-mips/syscalls/swapon.S

@@ -0,0 +1,22 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+    .text
+    .globl swapon
+    .align 4
+    .ent swapon
+
+swapon:
+    .set noreorder
+    .cpload $t9
+    li $v0, __NR_swapon
+    syscall
+    bnez $a3, 1f
+    move $a0, $v0
+    j $ra
+    nop
+1:
+    la $t9,__set_errno
+    j $t9
+    nop
+    .set reorder
+    .end swapon

libc/arch-mips/syscalls/timerfd_create.S

@@ -0,0 +1,22 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+    .text
+    .globl timerfd_create
+    .align 4
+    .ent timerfd_create
+
+timerfd_create:
+    .set noreorder
+    .cpload $t9
+    li $v0, __NR_timerfd_create
+    syscall
+    bnez $a3, 1f
+    move $a0, $v0
+    j $ra
+    nop
+1:
+    la $t9,__set_errno
+    j $t9
+    nop
+    .set reorder
+    .end timerfd_create

libc/arch-mips/syscalls/timerfd_gettime.S

@@ -0,0 +1,22 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+    .text
+    .globl timerfd_gettime
+    .align 4
+    .ent timerfd_gettime
+
+timerfd_gettime:
+    .set noreorder
+    .cpload $t9
+    li $v0, __NR_timerfd_gettime
+    syscall
+    bnez $a3, 1f
+    move $a0, $v0
+    j $ra
+    nop
+1:
+    la $t9,__set_errno
+    j $t9
+    nop
+    .set reorder
+    .end timerfd_gettime

libc/arch-mips/syscalls/timerfd_settime.S

@@ -0,0 +1,22 @@
+/* autogenerated by gensyscalls.py */
+#include <asm/unistd.h>
+    .text
+    .globl timerfd_settime
+    .align 4
+    .ent timerfd_settime
+
+timerfd_settime:
+    .set noreorder
+    .cpload $t9
+    li $v0, __NR_timerfd_settime
+    syscall
+    bnez $a3, 1f
+    move $a0, $v0
+    j $ra
+    nop
+1:
+    la $t9,__set_errno
+    j $t9
+    nop
+    .set reorder
+    .end timerfd_settime

libc/arch-x86/include/machine/endian.h

@@ -1,4 +1,4 @@
-/*	$OpenBSD: endian.h,v 1.14 2005/12/13 00:35:23 millert Exp $	*/
+/*	$OpenBSD: endian.h,v 1.17 2011/03/12 04:03:04 guenther Exp $	*/
 
 /*-
  * Copyright (c) 1997 Niklas Hallqvist.  All rights reserved.
@@ -24,38 +24,28 @@
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef _I386_ENDIAN_H_
+#ifndef _MACHINE_ENDIAN_H_
-#define _I386_ENDIAN_H_
+#define _MACHINE_ENDIAN_H_
 
 #ifdef __GNUC__
 
-#if defined(_KERNEL) && !defined(I386_CPU)
+#define	__swap32md(x) __statement({					\
-#define	__swap32md(x) ({						\
 	uint32_t __swap32md_x = (x);					\
 									\
-	__asm ("bswap %1" : "+r" (__swap32md_x));			\
+	__asm ("bswap %0" : "+r" (__swap32md_x));			\
 	__swap32md_x;							\
 })
-#else
-#define	__swap32md(x) ({						\
-	uint32_t __swap32md_x = (x);					\
-									\
-	__asm ("rorw $8, %w1; rorl $16, %1; rorw $8, %w1" :		\
-	    "+r" (__swap32md_x));					\
-	__swap32md_x;							\
-})
-#endif	/* _KERNEL && !I386_CPU */
 
-#define	__swap64md(x) ({						\
+#define	__swap64md(x) __statement({					\
 	uint64_t __swap64md_x = (x);					\
 									\
 	(uint64_t)__swap32md(__swap64md_x >> 32) |			\
 	    (uint64_t)__swap32md(__swap64md_x & 0xffffffff) << 32;	\
 })
-#define	__swap16md(x) ({						\
+#define	__swap16md(x) __statement({					\
 	uint16_t __swap16md_x = (x);					\
 									\
-	__asm ("rorw $8, %w1" : "+r" (__swap16md_x));			\
+	__asm ("rorw $8, %w0" : "+r" (__swap16md_x));			\
 	__swap16md_x;							\
 })
 
@@ -68,4 +58,4 @@
 #include <sys/types.h>
 #include <sys/endian.h>
 
-#endif /* _I386_ENDIAN_H_ */
+#endif /* _MACHINE_ENDIAN_H_ */

libc/arch-x86/string/bzero_wrapper.S

@@ -1,43 +0,0 @@
-/*
-Copyright (c) 2010, Intel Corporation
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-    * Redistributions of source code must retain the above copyright notice,
-    * this list of conditions and the following disclaimer.
-
-    * Redistributions in binary form must reproduce the above copyright notice,
-    * this list of conditions and the following disclaimer in the documentation
-    * and/or other materials provided with the distribution.
-
-    * Neither the name of Intel Corporation nor the names of its contributors
-    * may be used to endorse or promote products derived from this software
-    * without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
-ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
-ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
-ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#if defined(USE_SSE2)
-
-# include "cache_wrapper.S"
-# undef __i686
-# define USE_AS_BZERO
-# define sse2_memset5_atom bzero
-# include "sse2-memset5-atom.S"
-
-#else
-
-# include "bzero.S"
-
-#endif

libc/arch-x86/string/index.S

@@ -1,26 +0,0 @@
-/*	$OpenBSD: index.S,v 1.4 2005/08/07 11:30:38 espie Exp $ */
-/*
- * Written by J.T. Conklin <jtc@netbsd.org>.
- * Public domain.
- */
-
-#include <machine/asm.h>
-
-#ifdef STRCHR
-ENTRY(strchr)
-#else
-ENTRY(index)
-#endif
-	movl	4(%esp),%eax
-	movb	8(%esp),%cl
-	.align 2,0x90
-L1:
-	movb	(%eax),%dl
-	cmpb	%dl,%cl			/* found char??? */
-	je 	L2
-	incl	%eax
-	testb	%dl,%dl			/* null terminator??? */
-	jnz	L1
-	xorl	%eax,%eax
-L2:
-	ret

libc/arch-x86/string/memchr.S

@@ -1,27 +0,0 @@
-/*	$OpenBSD: memchr.S,v 1.3 2005/08/07 11:30:38 espie Exp $ */
-/*
- * Written by J.T. Conklin <jtc@netbsd.org>.
- * Public domain.
- */
-
-#include <machine/asm.h>
-
-ENTRY(memchr)
-	pushl	%edi
-	movl	8(%esp),%edi		/* string address */
-	movl	12(%esp),%eax		/* set character to search for */
-	movl	16(%esp),%ecx		/* set length of search */
-	testl	%ecx,%ecx		/* test for len == 0 */
-	jz	L1
-	cld				/* set search forward */
-	repne				/* search! */
-	scasb
-	jne	L1			/* scan failed, return null */
-	leal	-1(%edi),%eax		/* adjust result of scan */
-	popl	%edi
-	ret
-	.align 2,0x90
-L1:	xorl	%eax,%eax
-	popl	%edi
-	ret
-END(memchr)

libc/arch-x86/string/memcpy_wrapper.S

@@ -1,43 +0,0 @@
-/*
-Copyright (c) 2010, Intel Corporation
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-    * Redistributions of source code must retain the above copyright notice,
-    * this list of conditions and the following disclaimer.
-
-    * Redistributions in binary form must reproduce the above copyright notice,
-    * this list of conditions and the following disclaimer in the documentation
-    * and/or other materials provided with the distribution.
-
-    * Neither the name of Intel Corporation nor the names of its contributors
-    * may be used to endorse or promote products derived from this software
-    * without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
-ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
-ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
-ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#if defined(USE_SSSE3)
-
-# include "cache_wrapper.S"
-# undef __i686
-# define MEMCPY	memcpy
-# define USE_AS_MEMMOVE
-# include "ssse3-memcpy5.S"
-
-#else
-
-# include "memcpy.S"
-
-#endif

libc/arch-x86/string/memmove_wrapper.S

@@ -1,43 +0,0 @@
-/*
-Copyright (c) 2010, Intel Corporation
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-    * Redistributions of source code must retain the above copyright notice,
-    * this list of conditions and the following disclaimer.
-
-    * Redistributions in binary form must reproduce the above copyright notice,
-    * this list of conditions and the following disclaimer in the documentation
-    * and/or other materials provided with the distribution.
-
-    * Neither the name of Intel Corporation nor the names of its contributors
-    * may be used to endorse or promote products derived from this software
-    * without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
-ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
-ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
-ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#if defined(USE_SSSE3)
-
-# include "cache_wrapper.S"
-# undef __i686
-# define MEMCPY memmove
-# define USE_AS_MEMMOVE
-# include "ssse3-memcpy5.S"
-
-#else
-
-# include "memmove.S"
-
-#endif

libc/arch-x86/string/memset_wrapper.S

@@ -1,42 +0,0 @@
-/*
-Copyright (c) 2010, Intel Corporation
-All rights reserved.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions are met:
-
-    * Redistributions of source code must retain the above copyright notice,
-    * this list of conditions and the following disclaimer.
-
-    * Redistributions in binary form must reproduce the above copyright notice,
-    * this list of conditions and the following disclaimer in the documentation
-    * and/or other materials provided with the distribution.
-
-    * Neither the name of Intel Corporation nor the names of its contributors
-    * may be used to endorse or promote products derived from this software
-    * without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
-ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
-WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
-DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
-ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
-(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
-ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
-SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
-#if defined(USE_SSE2)
-
-# include "cache_wrapper.S"
-# undef __i686
-# define sse2_memset5_atom memset
-# include "sse2-memset5-atom.S"
-
-#else
-
-# include "memset.S"
-
-#endif

libc/arch-x86/string/sse2-bzero-atom.S

@@ -28,15 +28,6 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#if defined(USE_SSSE3)
+#define USE_AS_BZERO
-
+#define MEMSET  bzero
-# define USE_AS_STRNCMP
+#include "sse2-memset-atom.S"
-# define ssse3_strcmp_latest strncmp
-# include "ssse3-strcmp-latest.S"
-
-#else
-
-# include "strncmp.S"
-
-#endif
-

libc/arch-x86/string/sse2-index-atom.S

@@ -0,0 +1,32 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#define strchr  index
+#include "sse2-strchr-atom.S"

libc/arch-x86/string/sse2-memchr-atom.S

@@ -0,0 +1,556 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc	.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc	.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name,  @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name,	.-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG) pushl REG; CFI_PUSH (REG)
+#define POP(REG) popl REG; CFI_POP (REG)
+
+#define ENTRANCE PUSH (%edi);
+#define PARMS  8
+#define RETURN  POP (%edi); ret; CFI_PUSH (%edi);
+
+#define STR1  PARMS
+#define STR2  STR1+4
+#define LEN   STR2+4
+
+	.text
+ENTRY (memchr)
+	ENTRANCE
+	mov	STR1(%esp), %ecx
+	movd	STR2(%esp), %xmm1
+	mov	LEN(%esp), %edx
+	test	%edx, %edx
+	jz	L(return_null)
+
+	punpcklbw %xmm1, %xmm1
+	mov	%ecx, %edi
+	punpcklbw %xmm1, %xmm1
+
+	and	$63, %ecx
+	pshufd	$0, %xmm1, %xmm1
+	cmp	$48, %ecx
+	ja	L(crosscache)
+
+	movdqu	(%edi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(match_case2_prolog)
+
+	sub	$16, %edx
+	jbe	L(return_null)
+	lea	16(%edi), %edi
+	and	$15, %ecx
+	and	$-16, %edi
+	add	%ecx, %edx
+	sub	$64, %edx
+	jbe	L(exit_loop)
+	jmp	L(loop_prolog)
+
+	.p2align 4
+L(crosscache):
+	and	$15, %ecx
+	and	$-16, %edi
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	sar	%cl, %eax
+	test	%eax, %eax
+
+	jnz	L(match_case2_prolog1)
+	lea	-16(%edx), %edx
+	add	%ecx, %edx
+	jle	L(return_null)
+	lea	16(%edi), %edi
+	sub	$64, %edx
+	jbe	L(exit_loop)
+
+	.p2align 4
+L(loop_prolog):
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	xor	%ecx, %ecx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	16(%edi), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	32(%edi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	48(%edi), %xmm4
+	pcmpeqb	%xmm1, %xmm4
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm4, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	lea	64(%edi), %edi
+	sub	$64, %edx
+	jbe	L(exit_loop)
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	xor	%ecx, %ecx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	16(%edi), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	32(%edi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	48(%edi), %xmm4
+	pcmpeqb	%xmm1, %xmm4
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm4, %eax
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	lea	64(%edi), %edi
+	mov	%edi, %ecx
+	and	$-64, %edi
+	and	$63, %ecx
+	add	%ecx, %edx
+
+	.p2align 4
+L(align64_loop):
+	sub	$64, %edx
+	jbe	L(exit_loop)
+	movdqa	(%edi), %xmm0
+	movdqa	16(%edi), %xmm2
+	movdqa	32(%edi), %xmm3
+	movdqa	48(%edi), %xmm4
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm1, %xmm2
+	pcmpeqb	%xmm1, %xmm3
+	pcmpeqb	%xmm1, %xmm4
+
+	pmaxub	%xmm0, %xmm3
+	pmaxub	%xmm2, %xmm4
+	pmaxub	%xmm3, %xmm4
+	add	$64, %edi
+	pmovmskb %xmm4, %eax
+
+	test	%eax, %eax
+	jz	L(align64_loop)
+
+	sub	$64, %edi
+
+	pmovmskb %xmm0, %eax
+	xor	%ecx, %ecx
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	pmovmskb %xmm2, %eax
+	lea	16(%ecx), %ecx
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	movdqa	32(%edi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	lea	16(%ecx), %ecx
+	test	%eax, %eax
+	jnz	L(match_case1)
+
+	pcmpeqb	48(%edi), %xmm1
+	pmovmskb %xmm1, %eax
+	lea	16(%ecx), %ecx
+
+	.p2align 4
+L(match_case1):
+	add	%ecx, %edi
+	test	%al, %al
+	jz	L(match_case1_high)
+	mov	%al, %cl
+	and	$15, %cl
+	jz	L(match_case1_8)
+	test	$0x01, %al
+	jnz	L(exit_case1_1)
+	test	$0x02, %al
+	jnz	L(exit_case1_2)
+	test	$0x04, %al
+	jnz	L(exit_case1_3)
+	lea	3(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1_8):
+	test	$0x10, %al
+	jnz	L(exit_case1_5)
+	test	$0x20, %al
+	jnz	L(exit_case1_6)
+	test	$0x40, %al
+	jnz	L(exit_case1_7)
+	lea	7(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1_high):
+	mov	%ah, %ch
+	and	$15, %ch
+	jz	L(match_case1_high_8)
+	test	$0x01, %ah
+	jnz	L(exit_case1_9)
+	test	$0x02, %ah
+	jnz	L(exit_case1_10)
+	test	$0x04, %ah
+	jnz	L(exit_case1_11)
+	lea	11(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1_high_8):
+	test	$0x10, %ah
+	jnz	L(exit_case1_13)
+	test	$0x20, %ah
+	jnz	L(exit_case1_14)
+	test	$0x40, %ah
+	jnz	L(exit_case1_15)
+	lea	15(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_loop):
+	add	$64, %edx
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	xor	%ecx, %ecx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(match_case2)
+	cmp	$16, %edx
+	jbe	L(return_null)
+
+	movdqa	16(%edi), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(match_case2)
+	cmp	$32, %edx
+	jbe	L(return_null)
+
+	movdqa	32(%edi), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(match_case2)
+	cmp	$48, %edx
+	jbe	L(return_null)
+
+	pcmpeqb	48(%edi), %xmm1
+	lea	16(%ecx), %ecx
+	pmovmskb %xmm1, %eax
+	test	%eax, %eax
+	jnz	L(match_case2)
+
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_1):
+	mov	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_2):
+	lea	1(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_3):
+	lea	2(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_5):
+	lea	4(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_6):
+	lea	5(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_7):
+	lea	6(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_9):
+	lea	8(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_10):
+	lea	9(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_11):
+	lea	10(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_13):
+	lea	12(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_14):
+	lea	13(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case1_15):
+	lea	14(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case2):
+	sub	%ecx, %edx
+L(match_case2_prolog1):
+	add	%ecx, %edi
+L(match_case2_prolog):
+	test	%al, %al
+	jz	L(match_case2_high)
+	mov	%al, %cl
+	and	$15, %cl
+	jz	L(match_case2_8)
+	test	$0x01, %al
+	jnz	L(exit_case2_1)
+	test	$0x02, %al
+	jnz	L(exit_case2_2)
+	test	$0x04, %al
+	jnz	L(exit_case2_3)
+	sub	$4, %edx
+	jb	L(return_null)
+	lea	3(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case2_8):
+	test	$0x10, %al
+	jnz	L(exit_case2_5)
+	test	$0x20, %al
+	jnz	L(exit_case2_6)
+	test	$0x40, %al
+	jnz	L(exit_case2_7)
+	sub	$8, %edx
+	jb	L(return_null)
+	lea	7(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case2_high):
+	mov	%ah, %ch
+	and	$15, %ch
+	jz	L(match_case2_high_8)
+	test	$0x01, %ah
+	jnz	L(exit_case2_9)
+	test	$0x02, %ah
+	jnz	L(exit_case2_10)
+	test	$0x04, %ah
+	jnz	L(exit_case2_11)
+	sub	$12, %edx
+	jb	L(return_null)
+	lea	11(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case2_high_8):
+	test	$0x10, %ah
+	jnz	L(exit_case2_13)
+	test	$0x20, %ah
+	jnz	L(exit_case2_14)
+	test	$0x40, %ah
+	jnz	L(exit_case2_15)
+	sub	$16, %edx
+	jb	L(return_null)
+	lea	15(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_1):
+	mov	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_2):
+	sub	$2, %edx
+	jb	L(return_null)
+	lea	1(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_3):
+	sub	$3, %edx
+	jb	L(return_null)
+	lea	2(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_5):
+	sub	$5, %edx
+	jb	L(return_null)
+	lea	4(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_6):
+	sub	$6, %edx
+	jb	L(return_null)
+	lea	5(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_7):
+	sub	$7, %edx
+	jb	L(return_null)
+	lea	6(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_9):
+	sub	$9, %edx
+	jb	L(return_null)
+	lea	8(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_10):
+	sub	$10, %edx
+	jb	L(return_null)
+	lea	9(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_11):
+	sub	$11, %edx
+	jb	L(return_null)
+	lea	10(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_13):
+	sub	$13, %edx
+	jb	L(return_null)
+	lea	12(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_14):
+	sub	$14, %edx
+	jb	L(return_null)
+	lea	13(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(exit_case2_15):
+	sub	$15, %edx
+	jb	L(return_null)
+	lea	14(%edi), %eax
+	RETURN
+	.p2align 4
+L(return_null):
+	xor	%eax, %eax
+	RETURN
+END (memchr)

libc/arch-x86/string/sse2-memrchr-atom.S

@@ -0,0 +1,778 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc	.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc	.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name,  @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name,	.-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG) pushl REG; CFI_PUSH (REG)
+#define POP(REG) popl REG; CFI_POP (REG)
+
+#define PARMS  4
+#define STR1  PARMS
+#define STR2  STR1+4
+#define LEN   STR2+4
+
+	.text
+ENTRY (memrchr)
+	mov	STR1(%esp), %ecx
+	movd	STR2(%esp), %xmm1
+	mov	LEN(%esp), %edx
+
+	test	%edx, %edx
+	jz	L(return_null)
+	sub	$16, %edx
+	jbe	L(length_less16)
+
+	punpcklbw %xmm1, %xmm1
+	add	%edx, %ecx
+	punpcklbw %xmm1, %xmm1
+
+	movdqu	(%ecx), %xmm0
+	pshufd	$0, %xmm1, %xmm1
+	pcmpeqb	%xmm1, %xmm0
+
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(exit_dispatch)
+
+	sub	$64, %ecx
+	mov	%ecx, %eax
+	and	$15, %eax
+	jz	L(loop_prolog)
+
+	add	$16, %ecx
+	add	$16, %edx
+	and	$-16, %ecx
+	sub	%eax, %edx
+
+	.p2align 4
+/* Loop start on aligned string.  */
+L(loop_prolog):
+	sub	$64, %edx
+	jbe	L(exit_loop)
+
+	movdqa	48(%ecx), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches48)
+
+	movdqa	32(%ecx), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	movdqa	16(%ecx), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches16)
+
+	movdqa	(%ecx), %xmm4
+	pcmpeqb	%xmm1, %xmm4
+	pmovmskb %xmm4, %eax
+	test	%eax, %eax
+	jnz	L(exit_dispatch)
+
+	sub	$64, %ecx
+	sub	$64, %edx
+	jbe	L(exit_loop)
+
+	movdqa	48(%ecx), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches48)
+
+	movdqa	32(%ecx), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	movdqa	16(%ecx), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches16)
+
+	movdqa	(%ecx), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(exit_dispatch)
+
+	mov	%ecx, %eax
+	and	$63, %eax
+	test	%eax, %eax
+	jz	L(align64_loop)
+
+	add	$64, %ecx
+	add	$64, %edx
+	and	$-64, %ecx
+	sub	%eax, %edx
+
+	.p2align 4
+L(align64_loop):
+	sub	$64, %ecx
+	sub	$64, %edx
+	jbe	L(exit_loop)
+
+	movdqa	(%ecx), %xmm0
+	movdqa	16(%ecx), %xmm2
+	movdqa	32(%ecx), %xmm3
+	movdqa	48(%ecx), %xmm4
+
+	pcmpeqb	%xmm1, %xmm0
+	pcmpeqb	%xmm1, %xmm2
+	pcmpeqb	%xmm1, %xmm3
+	pcmpeqb	%xmm1, %xmm4
+
+	pmaxub	%xmm3, %xmm0
+	pmaxub	%xmm4, %xmm2
+	pmaxub	%xmm0, %xmm2
+	pmovmskb %xmm2, %eax
+
+	test	%eax, %eax
+	jz	L(align64_loop)
+
+	pmovmskb %xmm4, %eax
+	test	%eax, %eax
+	jnz	L(matches48)
+
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	movdqa	16(%ecx), %xmm2
+
+	pcmpeqb	%xmm1, %xmm2
+	pcmpeqb	(%ecx), %xmm1
+
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches16)
+
+	pmovmskb %xmm1, %eax
+	test	%ah, %ah
+	jnz	L(exit_dispatch_high)
+	mov	%al, %dl
+	and	$15 << 4, %dl
+	jnz	L(exit_dispatch_8)
+	test	$0x08, %al
+	jnz	L(exit_4)
+	test	$0x04, %al
+	jnz	L(exit_3)
+	test	$0x02, %al
+	jnz	L(exit_2)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(exit_loop):
+	add	$64, %edx
+	cmp	$32, %edx
+	jbe	L(exit_loop_32)
+
+	movdqa	48(%ecx), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches48)
+
+	movdqa	32(%ecx), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	pmovmskb %xmm2, %eax
+	test	%eax, %eax
+	jnz	L(matches32)
+
+	movdqa	16(%ecx), %xmm3
+	pcmpeqb	%xmm1, %xmm3
+	pmovmskb %xmm3, %eax
+	test	%eax, %eax
+	jnz	L(matches16_1)
+	cmp	$48, %edx
+	jbe	L(return_null)
+
+	pcmpeqb	(%ecx), %xmm1
+	pmovmskb %xmm1, %eax
+	test	%eax, %eax
+	jnz	L(matches0_1)
+	xor	%eax, %eax
+	ret
+
+	.p2align 4
+L(exit_loop_32):
+	movdqa	48(%ecx), %xmm0
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches48_1)
+	cmp	$16, %edx
+	jbe	L(return_null)
+
+	pcmpeqb	32(%ecx), %xmm1
+	pmovmskb %xmm1, %eax
+	test	%eax, %eax
+	jnz	L(matches32_1)
+	xor	%eax, %eax
+	ret
+
+	.p2align 4
+L(matches16):
+	lea	16(%ecx), %ecx
+	test	%ah, %ah
+	jnz	L(exit_dispatch_high)
+	mov	%al, %dl
+	and	$15 << 4, %dl
+	jnz	L(exit_dispatch_8)
+	test	$0x08, %al
+	jnz	L(exit_4)
+	test	$0x04, %al
+	jnz	L(exit_3)
+	test	$0x02, %al
+	jnz	L(exit_2)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(matches32):
+	lea	32(%ecx), %ecx
+	test	%ah, %ah
+	jnz	L(exit_dispatch_high)
+	mov	%al, %dl
+	and	$15 << 4, %dl
+	jnz	L(exit_dispatch_8)
+	test	$0x08, %al
+	jnz	L(exit_4)
+	test	$0x04, %al
+	jnz	L(exit_3)
+	test	$0x02, %al
+	jnz	L(exit_2)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(matches48):
+	lea	48(%ecx), %ecx
+
+	.p2align 4
+L(exit_dispatch):
+	test	%ah, %ah
+	jnz	L(exit_dispatch_high)
+	mov	%al, %dl
+	and	$15 << 4, %dl
+	jnz	L(exit_dispatch_8)
+	test	$0x08, %al
+	jnz	L(exit_4)
+	test	$0x04, %al
+	jnz	L(exit_3)
+	test	$0x02, %al
+	jnz	L(exit_2)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(exit_dispatch_8):
+	test	$0x80, %al
+	jnz	L(exit_8)
+	test	$0x40, %al
+	jnz	L(exit_7)
+	test	$0x20, %al
+	jnz	L(exit_6)
+	lea	4(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_dispatch_high):
+	mov	%ah, %dh
+	and	$15 << 4, %dh
+	jnz	L(exit_dispatch_high_8)
+	test	$0x08, %ah
+	jnz	L(exit_12)
+	test	$0x04, %ah
+	jnz	L(exit_11)
+	test	$0x02, %ah
+	jnz	L(exit_10)
+	lea	8(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_dispatch_high_8):
+	test	$0x80, %ah
+	jnz	L(exit_16)
+	test	$0x40, %ah
+	jnz	L(exit_15)
+	test	$0x20, %ah
+	jnz	L(exit_14)
+	lea	12(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_2):
+	lea	1(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_3):
+	lea	2(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_4):
+	lea	3(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_6):
+	lea	5(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_7):
+	lea	6(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_8):
+	lea	7(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_10):
+	lea	9(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_11):
+	lea	10(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_12):
+	lea	11(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_14):
+	lea	13(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_15):
+	lea	14(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_16):
+	lea	15(%ecx), %eax
+	ret
+
+	.p2align 4
+L(matches0_1):
+	lea	-64(%edx), %edx
+
+	test	%ah, %ah
+	jnz	L(exit_dispatch_1_high)
+	mov	%al, %ah
+	and	$15 << 4, %ah
+	jnz	L(exit_dispatch_1_8)
+	test	$0x08, %al
+	jnz	L(exit_1_4)
+	test	$0x04, %al
+	jnz	L(exit_1_3)
+	test	$0x02, %al
+	jnz	L(exit_1_2)
+
+	add	$0, %edx
+	jl	L(return_null)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(matches16_1):
+	lea	-48(%edx), %edx
+	lea	16(%ecx), %ecx
+
+	test	%ah, %ah
+	jnz	L(exit_dispatch_1_high)
+	mov	%al, %ah
+	and	$15 << 4, %ah
+	jnz	L(exit_dispatch_1_8)
+	test	$0x08, %al
+	jnz	L(exit_1_4)
+	test	$0x04, %al
+	jnz	L(exit_1_3)
+	test	$0x02, %al
+	jnz	L(exit_1_2)
+
+	add	$0, %edx
+	jl	L(return_null)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(matches32_1):
+	lea	-32(%edx), %edx
+	lea	32(%ecx), %ecx
+
+	test	%ah, %ah
+	jnz	L(exit_dispatch_1_high)
+	mov	%al, %ah
+	and	$15 << 4, %ah
+	jnz	L(exit_dispatch_1_8)
+	test	$0x08, %al
+	jnz	L(exit_1_4)
+	test	$0x04, %al
+	jnz	L(exit_1_3)
+	test	$0x02, %al
+	jnz	L(exit_1_2)
+
+	add	$0, %edx
+	jl	L(return_null)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(matches48_1):
+	lea	-16(%edx), %edx
+	lea	48(%ecx), %ecx
+
+	.p2align 4
+L(exit_dispatch_1):
+	test	%ah, %ah
+	jnz	L(exit_dispatch_1_high)
+	mov	%al, %ah
+	and	$15 << 4, %ah
+	jnz	L(exit_dispatch_1_8)
+	test	$0x08, %al
+	jnz	L(exit_1_4)
+	test	$0x04, %al
+	jnz	L(exit_1_3)
+	test	$0x02, %al
+	jnz	L(exit_1_2)
+
+	add	$0, %edx
+	jl	L(return_null)
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(exit_dispatch_1_8):
+	test	$0x80, %al
+	jnz	L(exit_1_8)
+	test	$0x40, %al
+	jnz	L(exit_1_7)
+	test	$0x20, %al
+	jnz	L(exit_1_6)
+
+	add	$4, %edx
+	jl	L(return_null)
+	lea	4(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_dispatch_1_high):
+	mov	%ah, %al
+	and	$15 << 4, %al
+	jnz	L(exit_dispatch_1_high_8)
+	test	$0x08, %ah
+	jnz	L(exit_1_12)
+	test	$0x04, %ah
+	jnz	L(exit_1_11)
+	test	$0x02, %ah
+	jnz	L(exit_1_10)
+
+	add	$8, %edx
+	jl	L(return_null)
+	lea	8(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_dispatch_1_high_8):
+	test	$0x80, %ah
+	jnz	L(exit_1_16)
+	test	$0x40, %ah
+	jnz	L(exit_1_15)
+	test	$0x20, %ah
+	jnz	L(exit_1_14)
+
+	add	$12, %edx
+	jl	L(return_null)
+	lea	12(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_2):
+	add	$1, %edx
+	jl	L(return_null)
+	lea	1(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_3):
+	add	$2, %edx
+	jl	L(return_null)
+	lea	2(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_4):
+	add	$3, %edx
+	jl	L(return_null)
+	lea	3(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_6):
+	add	$5, %edx
+	jl	L(return_null)
+	lea	5(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_7):
+	add	$6, %edx
+	jl	L(return_null)
+	lea	6(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_8):
+	add	$7, %edx
+	jl	L(return_null)
+	lea	7(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_10):
+	add	$9, %edx
+	jl	L(return_null)
+	lea	9(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_11):
+	add	$10, %edx
+	jl	L(return_null)
+	lea	10(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_12):
+	add	$11, %edx
+	jl	L(return_null)
+	lea	11(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_14):
+	add	$13, %edx
+	jl	L(return_null)
+	lea	13(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_15):
+	add	$14, %edx
+	jl	L(return_null)
+	lea	14(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit_1_16):
+	add	$15, %edx
+	jl	L(return_null)
+	lea	15(%ecx), %eax
+	ret
+
+	.p2align 4
+L(return_null):
+	xor	%eax, %eax
+	ret
+
+	.p2align 4
+L(length_less16_offset0):
+	mov	%dl, %cl
+	pcmpeqb	(%eax), %xmm1
+
+	mov	$1, %edx
+	sal	%cl, %edx
+	sub	$1, %edx
+
+	mov	%eax, %ecx
+	pmovmskb %xmm1, %eax
+
+	and	%edx, %eax
+	test	%eax, %eax
+	jnz	L(exit_dispatch)
+
+	xor	%eax, %eax
+	ret
+
+	.p2align 4
+L(length_less16):
+	punpcklbw %xmm1, %xmm1
+	add	$16, %edx
+	punpcklbw %xmm1, %xmm1
+
+	mov	%ecx, %eax
+	pshufd	$0, %xmm1, %xmm1
+
+	and	$15, %ecx
+	jz	L(length_less16_offset0)
+
+	PUSH	(%edi)
+
+	mov	%cl, %dh
+	add	%dl, %dh
+	and	$-16, %eax
+
+	sub	$16, %dh
+	ja	L(length_less16_part2)
+
+	pcmpeqb	(%eax), %xmm1
+	pmovmskb %xmm1, %edi
+
+	sar	%cl, %edi
+	add	%ecx, %eax
+	mov	%dl, %cl
+
+	mov	$1, %edx
+	sal	%cl, %edx
+	sub	$1, %edx
+
+	and	%edx, %edi
+	test	%edi, %edi
+	jz	L(ret_null)
+
+	bsr	%edi, %edi
+	add	%edi, %eax
+	POP	(%edi)
+	ret
+
+	CFI_PUSH     (%edi)
+
+	.p2align 4
+L(length_less16_part2):
+	movdqa	16(%eax), %xmm2
+	pcmpeqb	%xmm1, %xmm2
+	pmovmskb %xmm2, %edi
+
+	mov	%cl, %ch
+
+	mov	%dh, %cl
+	mov	$1, %edx
+	sal	%cl, %edx
+	sub	$1, %edx
+
+	and	%edx, %edi
+
+	test	%edi, %edi
+	jnz	L(length_less16_part2_return)
+
+	pcmpeqb	(%eax), %xmm1
+	pmovmskb %xmm1, %edi
+
+	mov	%ch, %cl
+	sar	%cl, %edi
+	test	%edi, %edi
+	jz	L(ret_null)
+
+	bsr	%edi, %edi
+	add	%edi, %eax
+	xor	%ch, %ch
+	add	%ecx, %eax
+	POP	(%edi)
+	ret
+
+	CFI_PUSH     (%edi)
+
+	.p2align 4
+L(length_less16_part2_return):
+	bsr	%edi, %edi
+	lea	16(%eax, %edi), %eax
+	POP	(%edi)
+	ret
+
+	CFI_PUSH     (%edi)
+
+	.p2align 4
+L(ret_null):
+	xor	%eax, %eax
+	POP	(%edi)
+	ret
+
+END (memrchr)

libc/arch-x86/string/sse2-memset-atom.S

@@ -28,6 +28,9 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
+#include "cache.h"
+#undef __i686
+
 #ifndef L
 # define L(label)	.L##label
 #endif
@@ -136,9 +139,13 @@ __i686.get_pc_thunk.bx:
     jmp		*TABLE(,%ecx,4)
 #endif
 
+#ifndef MEMSET
+# define MEMSET memset
+#endif
+
 	.section .text.sse2,"ax",@progbits
 	ALIGN (4)
-ENTRY (sse2_memset5_atom)
+ENTRY (MEMSET)
 	ENTRANCE
 
 	movl	LEN(%esp), %ecx
@@ -911,4 +918,4 @@ L(aligned_16_15bytes):
 	SETRTNVAL
 	RETURN_END
 
-END (sse2_memset5_atom)
+END (MEMSET)

libc/arch-x86/string/sse2-strchr-atom.S

@@ -0,0 +1,391 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc	.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc	.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name,  @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name,	.-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG;	CFI_PUSH (REG)
+#define POP(REG)	popl REG;	CFI_POP (REG)
+
+#define PARMS	8
+#define ENTRANCE	PUSH(%edi)
+#define RETURN	POP (%edi); ret; CFI_PUSH (%edi);
+
+
+#define STR1	PARMS
+#define STR2	STR1+4
+
+	.text
+ENTRY (strchr)
+
+	ENTRANCE
+	mov	STR1(%esp), %ecx
+	movd	STR2(%esp), %xmm1
+
+	pxor	%xmm2, %xmm2
+	mov	%ecx, %edi
+	punpcklbw %xmm1, %xmm1
+	punpcklbw %xmm1, %xmm1
+	/* ECX has OFFSET. */
+	and	$15, %ecx
+	pshufd	$0, %xmm1, %xmm1
+	je	L(loop)
+
+/* Handle unaligned string.  */
+	and	$-16, %edi
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	pcmpeqb	%xmm1, %xmm0
+	/* Find where NULL is.  */
+	pmovmskb %xmm2, %edx
+	/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+	/* Remove the leading bytes.  */
+	sarl	%cl, %edx
+	sarl	%cl, %eax
+	test	%eax, %eax
+	jz	L(unaligned_no_match)
+	add	%ecx, %edi
+	test	%edx, %edx
+	jz	L(match_case1)
+	jmp	L(match_case2)
+
+	.p2align 4
+L(unaligned_no_match):
+	test	%edx, %edx
+	jne	L(return_null)
+
+	pxor	%xmm2, %xmm2
+	add	$16, %edi
+
+	.p2align 4
+/* Loop start on aligned string.  */
+L(loop):
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+	test	%edx, %edx
+	jnz	L(return_null)
+	add	$16, %edi
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+	test	%edx, %edx
+	jnz	L(return_null)
+	add	$16, %edi
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+	test	%edx, %edx
+	jnz	L(return_null)
+	add	$16, %edi
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(matches)
+	test	%edx, %edx
+	jnz	L(return_null)
+	add	$16, %edi
+	jmp	L(loop)
+
+L(matches):
+	/* There is a match.  First find where NULL is.  */
+	test	%edx, %edx
+	jz	L(match_case1)
+
+	.p2align 4
+L(match_case2):
+	test	%al, %al
+	jz	L(match_higth_case2)
+
+	mov	%al, %cl
+	and	$15, %cl
+	jnz	L(match_case2_4)
+
+	mov	%dl, %ch
+	and	$15, %ch
+	jnz	L(return_null)
+
+	test	$0x10, %al
+	jnz	L(Exit5)
+	test	$0x10, %dl
+	jnz	L(return_null)
+	test	$0x20, %al
+	jnz	L(Exit6)
+	test	$0x20, %dl
+	jnz	L(return_null)
+	test	$0x40, %al
+	jnz	L(Exit7)
+	test	$0x40, %dl
+	jnz	L(return_null)
+	lea	7(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case2_4):
+	test	$0x01, %al
+	jnz	L(Exit1)
+	test	$0x01, %dl
+	jnz	L(return_null)
+	test	$0x02, %al
+	jnz	L(Exit2)
+	test	$0x02, %dl
+	jnz	L(return_null)
+	test	$0x04, %al
+	jnz	L(Exit3)
+	test	$0x04, %dl
+	jnz	L(return_null)
+	lea	3(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_higth_case2):
+	test	%dl, %dl
+	jnz	L(return_null)
+
+	mov	%ah, %cl
+	and	$15, %cl
+	jnz	L(match_case2_12)
+
+	mov	%dh, %ch
+	and	$15, %ch
+	jnz	L(return_null)
+
+	test	$0x10, %ah
+	jnz	L(Exit13)
+	test	$0x10, %dh
+	jnz	L(return_null)
+	test	$0x20, %ah
+	jnz	L(Exit14)
+	test	$0x20, %dh
+	jnz	L(return_null)
+	test	$0x40, %ah
+	jnz	L(Exit15)
+	test	$0x40, %dh
+	jnz	L(return_null)
+	lea	15(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case2_12):
+	test	$0x01, %ah
+	jnz	L(Exit9)
+	test	$0x01, %dh
+	jnz	L(return_null)
+	test	$0x02, %ah
+	jnz	L(Exit10)
+	test	$0x02, %dh
+	jnz	L(return_null)
+	test	$0x04, %ah
+	jnz	L(Exit11)
+	test	$0x04, %dh
+	jnz	L(return_null)
+	lea	11(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1):
+	test	%al, %al
+	jz	L(match_higth_case1)
+
+	test	$0x01, %al
+	jnz	L(Exit1)
+	test	$0x02, %al
+	jnz	L(Exit2)
+	test	$0x04, %al
+	jnz	L(Exit3)
+	test	$0x08, %al
+	jnz	L(Exit4)
+	test	$0x10, %al
+	jnz	L(Exit5)
+	test	$0x20, %al
+	jnz	L(Exit6)
+	test	$0x40, %al
+	jnz	L(Exit7)
+	lea	7(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_higth_case1):
+	test	$0x01, %ah
+	jnz	L(Exit9)
+	test	$0x02, %ah
+	jnz	L(Exit10)
+	test	$0x04, %ah
+	jnz	L(Exit11)
+	test	$0x08, %ah
+	jnz	L(Exit12)
+	test	$0x10, %ah
+	jnz	L(Exit13)
+	test	$0x20, %ah
+	jnz	L(Exit14)
+	test	$0x40, %ah
+	jnz	L(Exit15)
+	lea	15(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit1):
+	lea	(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit2):
+	lea	1(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit3):
+	lea	2(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit4):
+	lea	3(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit5):
+	lea	4(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit6):
+	lea	5(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit7):
+	lea	6(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit9):
+	lea	8(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit10):
+	lea	9(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit11):
+	lea	10(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit12):
+	lea	11(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit13):
+	lea	12(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit14):
+	lea	13(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit15):
+	lea	14(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(return_null):
+	xor	%eax, %eax
+	RETURN
+
+END (strchr)

libc/arch-x86/string/sse2-strlen-atom.S

@@ -1,71 +1,112 @@
-#define STRLEN sse2_strlen_atom
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
 
-#ifndef L
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef USE_AS_STRCAT
+
+# ifndef STRLEN
+#  define STRLEN strlen
+# endif
+
+# ifndef L
 #  define L(label)	.L##label
-#endif
+# endif
 
-#ifndef cfi_startproc
+# ifndef cfi_startproc
 #  define cfi_startproc	.cfi_startproc
-#endif
+# endif
 
-#ifndef cfi_endproc
+# ifndef cfi_endproc
 #  define cfi_endproc	.cfi_endproc
-#endif
+# endif
 
-#ifndef cfi_rel_offset
+/* calee safe register only for strnlen is required */
+
+# ifdef USE_AS_STRNLEN
+#  ifndef cfi_rel_offset
 #   define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
-#endif
+#  endif
 
-#ifndef cfi_restore
+#  ifndef cfi_restore
 #   define cfi_restore(reg)	.cfi_restore reg
-#endif
+#  endif
 
-#ifndef cfi_adjust_cfa_offset
+#  ifndef cfi_adjust_cfa_offset
 #   define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
-#endif
+#  endif
+# endif
 
-#ifndef cfi_remember_state
+# ifndef ENTRY
-# define cfi_remember_state		.cfi_remember_state
-#endif
-
-#ifndef cfi_restore_state
-# define cfi_restore_state		.cfi_restore_state
-#endif
-
-#ifndef ENTRY
 #  define ENTRY(name)	\
 	.type name,  @function;	\
 	.globl name;	\
 	.p2align 4;	\
 name:	\
 	cfi_startproc
-#endif
+# endif
 
-#ifndef END
+# ifndef END
 #  define END(name)	\
 	cfi_endproc;	\
 	.size name, .-name
-#endif
+# endif
 
-#define CFI_PUSH(REG)						\
+# define PARMS	4
+# define STR	PARMS
+# define RETURN	ret
+
+# ifdef USE_AS_STRNLEN
+#  define LEN	PARMS + 8
+#  define CFI_PUSH(REG)	\
 	cfi_adjust_cfa_offset (4);	\
 	cfi_rel_offset (REG, 0)
 
-#define CFI_POP(REG)						\
+#  define CFI_POP(REG)	\
 	cfi_adjust_cfa_offset (-4);	\
 	cfi_restore (REG)
 
-#define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+#  define PUSH(REG)	pushl	REG;	CFI_PUSH (REG)
-#define POP(REG)	popl REG; CFI_POP (REG)
+#  define POP(REG)	popl	REG;	CFI_POP (REG)
-#define PARMS		4
+#  undef RETURN
-#define	STR		PARMS
+#  define RETURN	POP (%edi); ret; CFI_PUSH(%edi);
-#define ENTRANCE
+# endif
-#define RETURN		ret
 
 	.text
 ENTRY (STRLEN)
-	ENTRANCE
 	mov	STR(%esp), %edx
+# ifdef USE_AS_STRNLEN
+	PUSH	(%edi)
+	movl	LEN(%esp), %edi
+	sub	$4, %edi
+	jbe	L(len_less4_prolog)
+# endif
+#endif
 	xor	%eax, %eax
 	cmpb	$0, (%edx)
 	jz	L(exit_tail0)
@@ -75,6 +116,12 @@ ENTRY (STRLEN)
 	jz	L(exit_tail2)
 	cmpb	$0, 3(%edx)
 	jz	L(exit_tail3)
+
+#ifdef USE_AS_STRNLEN
+	sub	$4, %edi
+	jbe	L(len_less8_prolog)
+#endif
+
 	cmpb	$0, 4(%edx)
 	jz	L(exit_tail4)
 	cmpb	$0, 5(%edx)
@@ -83,6 +130,12 @@ ENTRY (STRLEN)
 	jz	L(exit_tail6)
 	cmpb	$0, 7(%edx)
 	jz	L(exit_tail7)
+
+#ifdef USE_AS_STRNLEN
+	sub	$4, %edi
+	jbe	L(len_less12_prolog)
+#endif
+
 	cmpb	$0, 8(%edx)
 	jz	L(exit_tail8)
 	cmpb	$0, 9(%edx)
@@ -91,6 +144,12 @@ ENTRY (STRLEN)
 	jz	L(exit_tail10)
 	cmpb	$0, 11(%edx)
 	jz	L(exit_tail11)
+
+#ifdef USE_AS_STRNLEN
+	sub	$4, %edi
+	jbe	L(len_less16_prolog)
+#endif
+
 	cmpb	$0, 12(%edx)
 	jz	L(exit_tail12)
 	cmpb	$0, 13(%edx)
@@ -99,211 +158,531 @@ ENTRY (STRLEN)
 	jz	L(exit_tail14)
 	cmpb	$0, 15(%edx)
 	jz	L(exit_tail15)
+
 	pxor	%xmm0, %xmm0
-	mov	%edx, %eax
+	lea	16(%edx), %eax
-	mov	%edx, %ecx
+	mov	%eax, %ecx
 	and	$-16, %eax
-	add	$16, %ecx
+
-	add	$16, %eax
+#ifdef USE_AS_STRNLEN
+	and	$15, %edx
+	add	%edx, %edi
+	sub	$64, %edi
+	jbe	L(len_less64)
+#endif
 
 	pcmpeqb	(%eax), %xmm0
 	pmovmskb %xmm0, %edx
 	pxor	%xmm1, %xmm1
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm1
 	pmovmskb %xmm1, %edx
 	pxor	%xmm2, %xmm2
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
-
 	pcmpeqb	(%eax), %xmm2
 	pmovmskb %xmm2, %edx
 	pxor	%xmm3, %xmm3
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm3
 	pmovmskb %xmm3, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
+#ifdef USE_AS_STRNLEN
+	sub	$64, %edi
+	jbe	L(len_less64)
+#endif
+
 	pcmpeqb	(%eax), %xmm0
 	pmovmskb %xmm0, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm1
 	pmovmskb %xmm1, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm2
 	pmovmskb %xmm2, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm3
 	pmovmskb %xmm3, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
+#ifdef USE_AS_STRNLEN
+	sub	$64, %edi
+	jbe	L(len_less64)
+#endif
+
 	pcmpeqb	(%eax), %xmm0
 	pmovmskb %xmm0, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm1
 	pmovmskb %xmm1, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm2
 	pmovmskb %xmm2, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm3
 	pmovmskb %xmm3, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
+#ifdef USE_AS_STRNLEN
+	sub	$64, %edi
+	jbe	L(len_less64)
+#endif
+
 	pcmpeqb	(%eax), %xmm0
 	pmovmskb %xmm0, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm1
 	pmovmskb %xmm1, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm2
 	pmovmskb %xmm2, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
 	pcmpeqb	(%eax), %xmm3
 	pmovmskb %xmm3, %edx
-	test	%edx, %edx
 	lea	16(%eax), %eax
+	test	%edx, %edx
 	jnz	L(exit)
 
+#ifdef USE_AS_STRNLEN
+	mov	%eax, %edx
+	and	$63, %edx
+	add	%edx, %edi
+#endif
+
 	and	$-0x40, %eax
-	PUSH (%esi)
+
-	PUSH (%edi)
+	.p2align 4
-	PUSH (%ebx)
+L(aligned_64_loop):
-	PUSH (%ebp)
+#ifdef USE_AS_STRNLEN
-	xor	%ebp, %ebp
+	sub	$64, %edi
-L(aligned_64):
+	jbe	L(len_less64)
-	pcmpeqb	(%eax), %xmm0
+#endif
-	pcmpeqb	16(%eax), %xmm1
+	movaps	(%eax), %xmm0
-	pcmpeqb	32(%eax), %xmm2
+	movaps	16(%eax), %xmm1
-	pcmpeqb	48(%eax), %xmm3
+	movaps	32(%eax), %xmm2
-	pmovmskb %xmm0, %edx
+	movaps	48(%eax), %xmm6
-	pmovmskb %xmm1, %esi
+	pminub	%xmm1, %xmm0
-	pmovmskb %xmm2, %edi
+	pminub	%xmm6, %xmm2
-	pmovmskb %xmm3, %ebx
+	pminub	%xmm0, %xmm2
-	or	%edx, %ebp
+	pcmpeqb	%xmm3, %xmm2
-	or	%esi, %ebp
+	pmovmskb %xmm2, %edx
-	or	%edi, %ebp
-	or	%ebx, %ebp
 	lea	64(%eax), %eax
-	jz	L(aligned_64)
-L(48leave):
 	test	%edx, %edx
-	jnz	L(aligned_64_exit_16)
+	jz	L(aligned_64_loop)
-	test	%esi, %esi
+
-	jnz	L(aligned_64_exit_32)
+	pcmpeqb	-64(%eax), %xmm3
-	test	%edi, %edi
+	pmovmskb %xmm3, %edx
-	jnz	L(aligned_64_exit_48)
+	lea	48(%ecx), %ecx
-	mov	%ebx, %edx
+	test	%edx, %edx
-	lea	(%eax), %eax
+	jnz	L(exit)
-	jmp	L(aligned_64_exit)
+
-L(aligned_64_exit_48):
+	pcmpeqb	%xmm1, %xmm3
-	lea	-16(%eax), %eax
+	pmovmskb %xmm3, %edx
-	mov	%edi, %edx
+	lea	-16(%ecx), %ecx
-	jmp	L(aligned_64_exit)
+	test	%edx, %edx
-L(aligned_64_exit_32):
+	jnz	L(exit)
-	lea	-32(%eax), %eax
+
-	mov	%esi, %edx
+	pcmpeqb	-32(%eax), %xmm3
-	jmp	L(aligned_64_exit)
+	pmovmskb %xmm3, %edx
-L(aligned_64_exit_16):
+	lea	-16(%ecx), %ecx
-	lea	-48(%eax), %eax
+	test	%edx, %edx
-L(aligned_64_exit):
+	jnz	L(exit)
-	POP (%ebp)
+
-	POP (%ebx)
+	pcmpeqb	%xmm6, %xmm3
-	POP (%edi)
+	pmovmskb %xmm3, %edx
-	POP (%esi)
+	lea	-16(%ecx), %ecx
 L(exit):
 	sub	%ecx, %eax
 	test	%dl, %dl
 	jz	L(exit_high)
+
+	mov	%dl, %cl
+	and	$15, %cl
+	jz	L(exit_8)
 	test	$0x01, %dl
 	jnz	L(exit_tail0)
-
 	test	$0x02, %dl
 	jnz	L(exit_tail1)
-
 	test	$0x04, %dl
 	jnz	L(exit_tail2)
+	add	$3, %eax
+	RETURN
 
-	test	$0x08, %dl
+	.p2align 4
-	jnz	L(exit_tail3)
+L(exit_8):
-
 	test	$0x10, %dl
 	jnz	L(exit_tail4)
-
 	test	$0x20, %dl
 	jnz	L(exit_tail5)
-
 	test	$0x40, %dl
 	jnz	L(exit_tail6)
 	add	$7, %eax
+	RETURN
+
+	.p2align 4
+L(exit_high):
+	mov	%dh, %ch
+	and	$15, %ch
+	jz	L(exit_high_8)
+	test	$0x01, %dh
+	jnz	L(exit_tail8)
+	test	$0x02, %dh
+	jnz	L(exit_tail9)
+	test	$0x04, %dh
+	jnz	L(exit_tail10)
+	add	$11, %eax
+	RETURN
+
+	.p2align 4
+L(exit_high_8):
+	test	$0x10, %dh
+	jnz	L(exit_tail12)
+	test	$0x20, %dh
+	jnz	L(exit_tail13)
+	test	$0x40, %dh
+	jnz	L(exit_tail14)
+	add	$15, %eax
 L(exit_tail0):
 	RETURN
 
-L(exit_high):
+#ifdef USE_AS_STRNLEN
-	add	$8, %eax
-	test	$0x01, %dh
-	jnz	L(exit_tail0)
 
-	test	$0x02, %dh
+	.p2align 4
-	jnz	L(exit_tail1)
+L(len_less64):
+	pxor	%xmm0, %xmm0
+	add	$64, %edi
 
-	test	$0x04, %dh
+	pcmpeqb	(%eax), %xmm0
-	jnz	L(exit_tail2)
+	pmovmskb %xmm0, %edx
+	pxor	%xmm1, %xmm1
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(strnlen_exit)
 
-	test	$0x08, %dh
+	sub	$16, %edi
-	jnz	L(exit_tail3)
+	jbe	L(return_start_len)
 
-	test	$0x10, %dh
+	pcmpeqb	(%eax), %xmm1
-	jnz	L(exit_tail4)
+	pmovmskb %xmm1, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(strnlen_exit)
 
-	test	$0x20, %dh
+	sub	$16, %edi
-	jnz	L(exit_tail5)
+	jbe	L(return_start_len)
 
-	test	$0x40, %dh
+	pcmpeqb	(%eax), %xmm0
-	jnz	L(exit_tail6)
+	pmovmskb %xmm0, %edx
-	add	$7, %eax
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(strnlen_exit)
+
+	sub	$16, %edi
+	jbe	L(return_start_len)
+
+	pcmpeqb	(%eax), %xmm1
+	pmovmskb %xmm1, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(strnlen_exit)
+
+#ifndef USE_AS_STRLCAT
+	movl	LEN(%esp), %eax
 	RETURN
+#else
+	jmp	L(return_start_len)
+#endif
+
+	.p2align 4
+L(strnlen_exit):
+	sub	%ecx, %eax
+
+	test	%dl, %dl
+	jz	L(strnlen_exit_high)
+	mov	%dl, %cl
+	and	$15, %cl
+	jz	L(strnlen_exit_8)
+	test	$0x01, %dl
+	jnz	L(exit_tail0)
+	test	$0x02, %dl
+	jnz	L(strnlen_exit_tail1)
+	test	$0x04, %dl
+	jnz	L(strnlen_exit_tail2)
+	sub	$4, %edi
+	jb	L(return_start_len)
+	lea	3(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_8):
+	test	$0x10, %dl
+	jnz	L(strnlen_exit_tail4)
+	test	$0x20, %dl
+	jnz	L(strnlen_exit_tail5)
+	test	$0x40, %dl
+	jnz	L(strnlen_exit_tail6)
+	sub	$8, %edi
+	jb	L(return_start_len)
+	lea	7(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_high):
+	mov	%dh, %ch
+	and	$15, %ch
+	jz	L(strnlen_exit_high_8)
+	test	$0x01, %dh
+	jnz	L(strnlen_exit_tail8)
+	test	$0x02, %dh
+	jnz	L(strnlen_exit_tail9)
+	test	$0x04, %dh
+	jnz	L(strnlen_exit_tail10)
+	sub	$12, %edi
+	jb	L(return_start_len)
+	lea	11(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_high_8):
+	test	$0x10, %dh
+	jnz	L(strnlen_exit_tail12)
+	test	$0x20, %dh
+	jnz	L(strnlen_exit_tail13)
+	test	$0x40, %dh
+	jnz	L(strnlen_exit_tail14)
+	sub	$16, %edi
+	jb	L(return_start_len)
+	lea	15(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail1):
+	sub	$2, %edi
+	jb	L(return_start_len)
+	lea	1(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail2):
+	sub	$3, %edi
+	jb	L(return_start_len)
+	lea	2(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail4):
+	sub	$5, %edi
+	jb	L(return_start_len)
+	lea	4(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail5):
+	sub	$6, %edi
+	jb	L(return_start_len)
+	lea	5(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail6):
+	sub	$7, %edi
+	jb	L(return_start_len)
+	lea	6(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail8):
+	sub	$9, %edi
+	jb	L(return_start_len)
+	lea	8(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail9):
+	sub	$10, %edi
+	jb	L(return_start_len)
+	lea	9(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail10):
+	sub	$11, %edi
+	jb	L(return_start_len)
+	lea	10(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail12):
+	sub	$13, %edi
+	jb	L(return_start_len)
+	lea	12(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail13):
+	sub	$14, %edi
+	jb	L(return_start_len)
+	lea	13(%eax), %eax
+	RETURN
+
+	.p2align 4
+L(strnlen_exit_tail14):
+	sub	$15, %edi
+	jb	L(return_start_len)
+	lea	14(%eax), %eax
+	RETURN
+
+#ifndef USE_AS_STRLCAT
+	.p2align 4
+L(return_start_len):
+	movl	LEN(%esp), %eax
+	RETURN
+#endif
+
+/* for prolog only */
+
+	.p2align 4
+L(len_less4_prolog):
+	xor	%eax, %eax
+
+	add	$4, %edi
+	jz	L(exit_tail0)
+
+	cmpb	$0, (%edx)
+	jz	L(exit_tail0)
+	cmp	$1, %edi
+	je	L(exit_tail1)
+
+	cmpb	$0, 1(%edx)
+	jz	L(exit_tail1)
+	cmp	$2, %edi
+	je	L(exit_tail2)
+
+	cmpb	$0, 2(%edx)
+	jz	L(exit_tail2)
+	cmp	$3, %edi
+	je	L(exit_tail3)
+
+	cmpb	$0, 3(%edx)
+	jz	L(exit_tail3)
+	mov	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(len_less8_prolog):
+	add	$4, %edi
+
+	cmpb	$0, 4(%edx)
+	jz	L(exit_tail4)
+	cmp	$1, %edi
+	je	L(exit_tail5)
+
+	cmpb	$0, 5(%edx)
+	jz	L(exit_tail5)
+	cmp	$2, %edi
+	je	L(exit_tail6)
+
+	cmpb	$0, 6(%edx)
+	jz	L(exit_tail6)
+	cmp	$3, %edi
+	je	L(exit_tail7)
+
+	cmpb	$0, 7(%edx)
+	jz	L(exit_tail7)
+	mov	$8, %eax
+	RETURN
+
+
+	.p2align 4
+L(len_less12_prolog):
+	add	$4, %edi
+
+	cmpb	$0, 8(%edx)
+	jz	L(exit_tail8)
+	cmp	$1, %edi
+	je	L(exit_tail9)
+
+	cmpb	$0, 9(%edx)
+	jz	L(exit_tail9)
+	cmp	$2, %edi
+	je	L(exit_tail10)
+
+	cmpb	$0, 10(%edx)
+	jz	L(exit_tail10)
+	cmp	$3, %edi
+	je	L(exit_tail11)
+
+	cmpb	$0, 11(%edx)
+	jz	L(exit_tail11)
+	mov	$12, %eax
+	RETURN
+
+	.p2align 4
+L(len_less16_prolog):
+	add	$4, %edi
+
+	cmpb	$0, 12(%edx)
+	jz	L(exit_tail12)
+	cmp	$1, %edi
+	je	L(exit_tail13)
+
+	cmpb	$0, 13(%edx)
+	jz	L(exit_tail13)
+	cmp	$2, %edi
+	je	L(exit_tail14)
+
+	cmpb	$0, 14(%edx)
+	jz	L(exit_tail14)
+	cmp	$3, %edi
+	je	L(exit_tail15)
+
+	cmpb	$0, 15(%edx)
+	jz	L(exit_tail15)
+	mov	$16, %eax
+	RETURN
+#endif
 
 	.p2align 4
 L(exit_tail1):
@@ -364,6 +743,7 @@ L(exit_tail14):
 
 L(exit_tail15):
 	add	$15, %eax
-	ret
+#ifndef USE_AS_STRCAT
-
+	RETURN
 END (STRLEN)
+#endif

libc/arch-x86/string/sse2-strnlen-atom.S

@@ -0,0 +1,33 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#define USE_AS_STRNLEN 1
+#define STRLEN  strnlen
+#include "sse2-strlen-atom.S"

libc/arch-x86/string/sse2-strrchr-atom.S

@@ -0,0 +1,753 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc	.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc	.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name, @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name, .-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+#define POP(REG)	popl REG; CFI_POP (REG)
+
+#define PARMS	8
+#define ENTRANCE	PUSH(%edi);
+#define RETURN	POP (%edi); ret; CFI_PUSH (%edi);
+
+#define STR1  PARMS
+#define STR2  STR1+4
+
+	.text
+ENTRY (strrchr)
+
+	ENTRANCE
+	mov	STR1(%esp), %ecx
+	movd	STR2(%esp), %xmm1
+
+	pxor	%xmm2, %xmm2
+	mov	%ecx, %edi
+	punpcklbw %xmm1, %xmm1
+	punpcklbw %xmm1, %xmm1
+	/* ECX has OFFSET. */
+	and	$63, %ecx
+	pshufd	$0, %xmm1, %xmm1
+	cmp	$48, %ecx
+	ja	L(crosscache)
+
+/* unaligned string. */
+	movdqu	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	pcmpeqb	%xmm1, %xmm0
+	/* Find where NULL is.  */
+	pmovmskb %xmm2, %ecx
+	/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+	add	$16, %edi
+
+	test	%eax, %eax
+	jnz	L(unaligned_match1)
+
+	test	%ecx, %ecx
+	jnz	L(return_null)
+
+	and	$-16, %edi
+
+	PUSH	(%esi)
+	PUSH	(%ebx)
+
+	xor	%ebx, %ebx
+	jmp	L(loop)
+
+	CFI_POP    (%esi)
+	CFI_POP    (%ebx)
+
+	.p2align 4
+L(unaligned_match1):
+	test	%ecx, %ecx
+	jnz	L(prolog_find_zero_1)
+
+	PUSH	(%esi)
+	PUSH	(%ebx)
+
+	mov	%eax, %ebx
+	mov	%edi, %esi
+	and	$-16, %edi
+	jmp	L(loop)
+
+	CFI_POP    (%esi)
+	CFI_POP    (%ebx)
+
+	.p2align 4
+L(crosscache):
+/* Hancle unaligned string.  */
+	and	$15, %ecx
+	and	$-16, %edi
+	pxor	%xmm3, %xmm3
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm3
+	pcmpeqb	%xmm1, %xmm0
+	/* Find where NULL is.  */
+	pmovmskb %xmm3, %edx
+	/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+	/* Remove the leading bytes.  */
+	shr	%cl, %edx
+	shr	%cl, %eax
+	add	$16, %edi
+
+	test	%eax, %eax
+	jnz	L(unaligned_match)
+
+	test	%edx, %edx
+	jnz	L(return_null)
+
+	PUSH	(%esi)
+	PUSH	(%ebx)
+
+	xor	%ebx, %ebx
+	jmp	L(loop)
+
+	CFI_POP    (%esi)
+	CFI_POP    (%ebx)
+
+	.p2align 4
+L(unaligned_match):
+	test	%edx, %edx
+	jnz	L(prolog_find_zero)
+
+	PUSH	(%esi)
+	PUSH	(%ebx)
+
+	mov	%eax, %ebx
+	lea	(%edi, %ecx), %esi
+
+/* Loop start on aligned string.  */
+	.p2align 4
+L(loop):
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	add	$16, %edi
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm0, %eax
+	or	%eax, %ecx
+	jnz	L(matches)
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	add	$16, %edi
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm0, %eax
+	or	%eax, %ecx
+	jnz	L(matches)
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	add	$16, %edi
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm0, %eax
+	or	%eax, %ecx
+	jnz	L(matches)
+
+	movdqa	(%edi), %xmm0
+	pcmpeqb	%xmm0, %xmm2
+	add	$16, %edi
+	pcmpeqb	%xmm1, %xmm0
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm0, %eax
+	or	%eax, %ecx
+	jz	L(loop)
+
+L(matches):
+	test	%eax, %eax
+	jnz	L(match)
+L(return_value):
+	test	%ebx, %ebx
+	jz	L(return_null_1)
+	mov	%ebx, %eax
+	mov	%esi, %edi
+
+	POP	(%ebx)
+	POP	(%esi)
+
+	jmp	L(match_case1)
+
+	CFI_PUSH    (%ebx)
+	CFI_PUSH    (%esi)
+
+	.p2align 4
+L(return_null_1):
+	POP	(%ebx)
+	POP	(%esi)
+
+	xor	%eax, %eax
+	RETURN
+
+	CFI_PUSH    (%ebx)
+	CFI_PUSH    (%esi)
+
+	.p2align 4
+L(match):
+	pmovmskb %xmm2, %ecx
+	test	%ecx, %ecx
+	jnz	L(find_zero)
+	mov	%eax, %ebx
+	mov	%edi, %esi
+	jmp	L(loop)
+
+	.p2align 4
+L(find_zero):
+	test	%cl, %cl
+	jz	L(find_zero_high)
+	mov	%cl, %dl
+	and	$15, %dl
+	jz	L(find_zero_8)
+	test	$0x01, %cl
+	jnz	L(FindZeroExit1)
+	test	$0x02, %cl
+	jnz	L(FindZeroExit2)
+	test	$0x04, %cl
+	jnz	L(FindZeroExit3)
+	and	$1 << 4 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(find_zero_8):
+	test	$0x10, %cl
+	jnz	L(FindZeroExit5)
+	test	$0x20, %cl
+	jnz	L(FindZeroExit6)
+	test	$0x40, %cl
+	jnz	L(FindZeroExit7)
+	and	$1 << 8 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(find_zero_high):
+	mov	%ch, %dh
+	and	$15, %dh
+	jz	L(find_zero_high_8)
+	test	$0x01, %ch
+	jnz	L(FindZeroExit9)
+	test	$0x02, %ch
+	jnz	L(FindZeroExit10)
+	test	$0x04, %ch
+	jnz	L(FindZeroExit11)
+	and	$1 << 12 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(find_zero_high_8):
+	test	$0x10, %ch
+	jnz	L(FindZeroExit13)
+	test	$0x20, %ch
+	jnz	L(FindZeroExit14)
+	test	$0x40, %ch
+	jnz	L(FindZeroExit15)
+	and	$1 << 16 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit1):
+	and	$1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit2):
+	and	$1 << 2 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit3):
+	and	$1 << 3 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit5):
+	and	$1 << 5 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit6):
+	and	$1 << 6 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit7):
+	and	$1 << 7 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit9):
+	and	$1 << 9 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit10):
+	and	$1 << 10 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit11):
+	and	$1 << 11 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit13):
+	and	$1 << 13 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit14):
+	and	$1 << 14 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+	jmp     L(match_case1)
+
+	CFI_PUSH	(%ebx)
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(FindZeroExit15):
+	and	$1 << 15 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%ebx)
+	POP	(%esi)
+
+	.p2align 4
+L(match_case1):
+	test	%ah, %ah
+	jnz	L(match_case1_high)
+	mov	%al, %dl
+	and	$15 << 4, %dl
+	jnz	L(match_case1_8)
+	test	$0x08, %al
+	jnz	L(Exit4)
+	test	$0x04, %al
+	jnz	L(Exit3)
+	test	$0x02, %al
+	jnz	L(Exit2)
+	lea	-16(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1_8):
+	test	$0x80, %al
+	jnz	L(Exit8)
+	test	$0x40, %al
+	jnz	L(Exit7)
+	test	$0x20, %al
+	jnz	L(Exit6)
+	lea	-12(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1_high):
+	mov	%ah, %dh
+	and	$15 << 4, %dh
+	jnz	L(match_case1_high_8)
+	test	$0x08, %ah
+	jnz	L(Exit12)
+	test	$0x04, %ah
+	jnz	L(Exit11)
+	test	$0x02, %ah
+	jnz	L(Exit10)
+	lea	-8(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_case1_high_8):
+	test	$0x80, %ah
+	jnz	L(Exit16)
+	test	$0x40, %ah
+	jnz	L(Exit15)
+	test	$0x20, %ah
+	jnz	L(Exit14)
+	lea	-4(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit2):
+	lea	-15(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit3):
+	lea	-14(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit4):
+	lea	-13(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit6):
+	lea	-11(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit7):
+	lea	-10(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit8):
+	lea	-9(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit10):
+	lea	-7(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit11):
+	lea	-6(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit12):
+	lea	-5(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit14):
+	lea	-3(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit15):
+	lea	-2(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(Exit16):
+	lea	-1(%edi), %eax
+	RETURN
+
+/* Return NULL.  */
+	.p2align 4
+L(return_null):
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero):
+	add	%ecx, %edi
+	mov     %edx, %ecx
+L(prolog_find_zero_1):
+	test	%cl, %cl
+	jz	L(prolog_find_zero_high)
+	mov	%cl, %dl
+	and	$15, %dl
+	jz	L(prolog_find_zero_8)
+	test	$0x01, %cl
+	jnz	L(PrologFindZeroExit1)
+	test	$0x02, %cl
+	jnz	L(PrologFindZeroExit2)
+	test	$0x04, %cl
+	jnz	L(PrologFindZeroExit3)
+	and	$1 << 4 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero_8):
+	test	$0x10, %cl
+	jnz	L(PrologFindZeroExit5)
+	test	$0x20, %cl
+	jnz	L(PrologFindZeroExit6)
+	test	$0x40, %cl
+	jnz	L(PrologFindZeroExit7)
+	and	$1 << 8 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero_high):
+	mov	%ch, %dh
+	and	$15, %dh
+	jz	L(prolog_find_zero_high_8)
+	test	$0x01, %ch
+	jnz	L(PrologFindZeroExit9)
+	test	$0x02, %ch
+	jnz	L(PrologFindZeroExit10)
+	test	$0x04, %ch
+	jnz	L(PrologFindZeroExit11)
+	and	$1 << 12 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero_high_8):
+	test	$0x10, %ch
+	jnz	L(PrologFindZeroExit13)
+	test	$0x20, %ch
+	jnz	L(PrologFindZeroExit14)
+	test	$0x40, %ch
+	jnz	L(PrologFindZeroExit15)
+	and	$1 << 16 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit1):
+	and	$1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit2):
+	and	$1 << 2 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit3):
+	and	$1 << 3 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit5):
+	and	$1 << 5 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit6):
+	and	$1 << 6 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit7):
+	and	$1 << 7 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit9):
+	and	$1 << 9 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit10):
+	and	$1 << 10 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit11):
+	and	$1 << 11 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit13):
+	and	$1 << 13 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit14):
+	and	$1 << 14 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(PrologFindZeroExit15):
+	and	$1 << 15 - 1, %eax
+	jnz	L(match_case1)
+	xor	%eax, %eax
+	RETURN
+
+END (strrchr)

libc/arch-x86/string/sse2-wcschr-atom.S

@@ -0,0 +1,267 @@
+/*
+Copyright (c) 2011 Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc	.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc	.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name,  @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name,	.-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+#define POP(REG)	popl REG; CFI_POP (REG)
+
+#define PARMS	4
+
+
+#define STR1  PARMS
+#define STR2  STR1+4
+
+	.text
+ENTRY (wcschr)
+
+	mov	STR1(%esp), %ecx
+	movd	STR2(%esp), %xmm1
+
+	mov	%ecx, %eax
+	punpckldq %xmm1, %xmm1
+	pxor	%xmm2, %xmm2
+	punpckldq %xmm1, %xmm1
+
+	and	$63, %eax
+	cmp	$48, %eax
+	ja	L(cross_cache)
+
+	movdqu	(%ecx), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	or	%eax, %edx
+	jnz	L(matches)
+	and	$-16, %ecx
+	jmp	L(loop)
+
+	.p2align 4
+L(cross_cache):
+	PUSH	(%edi)
+	mov	%ecx, %edi
+	mov	%eax, %ecx
+	and	$-16, %edi
+	and	$15, %ecx
+	movdqa	(%edi), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+
+	sarl	%cl, %edx
+	sarl	%cl, %eax
+	test	%eax, %eax
+	jz	L(unaligned_no_match)
+
+	add	%edi, %ecx
+	POP	(%edi)
+
+	test	%edx, %edx
+	jz	L(match_case1)
+	test	%al, %al
+	jz	L(match_higth_case2)
+	test	$15, %al
+	jnz	L(match_case2_4)
+	test	$15, %dl
+	jnz	L(return_null)
+	lea	4(%ecx), %eax
+	ret
+
+	CFI_PUSH (%edi)
+
+	.p2align 4
+L(unaligned_no_match):
+	mov	%edi, %ecx
+	POP	(%edi)
+
+	test	%edx, %edx
+	jnz	L(return_null)
+
+	pxor	%xmm2, %xmm2
+
+/* Loop start on aligned string.  */
+	.p2align 4
+L(loop):
+	add	$16, %ecx
+	movdqa	(%ecx), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	or	%eax, %edx
+	jnz	L(matches)
+	add	$16, %ecx
+
+	movdqa	(%ecx), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	or	%eax, %edx
+	jnz	L(matches)
+	add	$16, %ecx
+
+	movdqa	(%ecx), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	or	%eax, %edx
+	jnz	L(matches)
+	add	$16, %ecx
+
+	movdqa	(%ecx), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %edx
+	pmovmskb %xmm0, %eax
+	or	%eax, %edx
+	jz	L(loop)
+
+	.p2align 4
+L(matches):
+	pmovmskb %xmm2, %edx
+	test	%eax, %eax
+	jz	L(return_null)
+	test	%edx, %edx
+	jz	L(match_case1)
+
+	.p2align 4
+L(match_case2):
+	test	%al, %al
+	jz	L(match_higth_case2)
+	test	$15, %al
+	jnz	L(match_case2_4)
+	test	$15, %dl
+	jnz	L(return_null)
+	lea	4(%ecx), %eax
+	ret
+
+	.p2align 4
+L(match_case2_4):
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(match_higth_case2):
+	test	%dl, %dl
+	jnz	L(return_null)
+	test	$15, %ah
+	jnz	L(match_case2_12)
+	test	$15, %dh
+	jnz	L(return_null)
+	lea	12(%ecx), %eax
+	ret
+
+	.p2align 4
+L(match_case2_12):
+	lea	8(%ecx), %eax
+	ret
+
+	.p2align 4
+L(match_case1):
+	test	%al, %al
+	jz	L(match_higth_case1)
+
+	test	$0x01, %al
+	jnz	L(exit0)
+	lea	4(%ecx), %eax
+	ret
+
+	.p2align 4
+L(match_higth_case1):
+	test	$0x01, %ah
+	jnz	L(exit3)
+	lea	12(%ecx), %eax
+	ret
+
+	.p2align 4
+L(exit0):
+	mov	%ecx, %eax
+	ret
+
+	.p2align 4
+L(exit3):
+	lea	8(%ecx), %eax
+	ret
+
+	.p2align 4
+L(return_null):
+	xor	%eax, %eax
+	ret
+
+END (wcschr)

libc/arch-x86/string/sse2-wcslen-atom.S

@@ -0,0 +1,306 @@
+/*
+Copyright (c) 2011 Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef USE_AS_WCSCAT
+
+# ifndef L
+#  define L(label)	.L##label
+# endif
+
+# ifndef cfi_startproc
+#  define cfi_startproc	.cfi_startproc
+# endif
+
+# ifndef cfi_endproc
+#  define cfi_endproc	.cfi_endproc
+# endif
+
+# ifndef ENTRY
+#  define ENTRY(name)	\
+	.type name,  @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+# endif
+
+# ifndef END
+#  define END(name)	\
+	cfi_endproc;	\
+	.size name, .-name
+# endif
+
+# define PARMS	4
+# define STR	PARMS
+# define RETURN ret
+
+	.text
+ENTRY (wcslen)
+	mov	STR(%esp), %edx
+#endif
+	cmp	$0, (%edx)
+	jz	L(exit_tail0)
+	cmp	$0, 4(%edx)
+	jz	L(exit_tail1)
+	cmp	$0, 8(%edx)
+	jz	L(exit_tail2)
+	cmp	$0, 12(%edx)
+	jz	L(exit_tail3)
+	cmp	$0, 16(%edx)
+	jz	L(exit_tail4)
+	cmp	$0, 20(%edx)
+	jz	L(exit_tail5)
+	cmp	$0, 24(%edx)
+	jz	L(exit_tail6)
+	cmp	$0, 28(%edx)
+	jz	L(exit_tail7)
+
+	pxor	%xmm0, %xmm0
+
+	lea	32(%edx), %eax
+	lea	-16(%eax), %ecx
+	and	$-16, %eax
+
+	pcmpeqd	(%eax), %xmm0
+	pmovmskb %xmm0, %edx
+	pxor	%xmm1, %xmm1
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm1
+	pmovmskb %xmm1, %edx
+	pxor	%xmm2, %xmm2
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm2
+	pmovmskb %xmm2, %edx
+	pxor	%xmm3, %xmm3
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm3
+	pmovmskb %xmm3, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm0
+	pmovmskb %xmm0, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm1
+	pmovmskb %xmm1, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm2
+	pmovmskb %xmm2, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm3
+	pmovmskb %xmm3, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm0
+	pmovmskb %xmm0, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm1
+	pmovmskb %xmm1, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm2
+	pmovmskb %xmm2, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm3
+	pmovmskb %xmm3, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm0
+	pmovmskb %xmm0, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm1
+	pmovmskb %xmm1, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm2
+	pmovmskb %xmm2, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	(%eax), %xmm3
+	pmovmskb %xmm3, %edx
+	lea	16(%eax), %eax
+	test	%edx, %edx
+	jnz	L(exit)
+
+	and	$-0x40, %eax
+
+	.p2align 4
+L(aligned_64_loop):
+	movaps	(%eax), %xmm0
+	movaps	16(%eax), %xmm1
+	movaps	32(%eax), %xmm2
+	movaps	48(%eax), %xmm6
+
+	pminub	%xmm1, %xmm0
+	pminub	%xmm6, %xmm2
+	pminub	%xmm0, %xmm2
+	pcmpeqd	%xmm3, %xmm2
+	pmovmskb %xmm2, %edx
+	lea	64(%eax), %eax
+	test	%edx, %edx
+	jz	L(aligned_64_loop)
+
+	pcmpeqd	-64(%eax), %xmm3
+	pmovmskb %xmm3, %edx
+	lea	48(%ecx), %ecx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	-16(%ecx), %ecx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	-32(%eax), %xmm3
+	pmovmskb %xmm3, %edx
+	lea	-16(%ecx), %ecx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pcmpeqd	%xmm6, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	-16(%ecx), %ecx
+	test	%edx, %edx
+	jnz	L(exit)
+
+	jmp	L(aligned_64_loop)
+
+	.p2align 4
+L(exit):
+	sub	%ecx, %eax
+	shr	$2, %eax
+	test	%dl, %dl
+	jz	L(exit_high)
+
+	mov	%dl, %cl
+	and	$15, %cl
+	jz	L(exit_1)
+	RETURN
+
+	.p2align 4
+L(exit_high):
+	mov	%dh, %ch
+	and	$15, %ch
+	jz	L(exit_3)
+	add	$2, %eax
+	RETURN
+
+	.p2align 4
+L(exit_1):
+	add	$1, %eax
+	RETURN
+
+	.p2align 4
+L(exit_3):
+	add	$3, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail0):
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail1):
+	mov	$1, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail2):
+	mov	$2, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail3):
+	mov	$3, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail4):
+	mov	$4, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail5):
+	mov	$5, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail6):
+	mov	$6, %eax
+	RETURN
+
+	.p2align 4
+L(exit_tail7):
+	mov	$7, %eax
+#ifndef USE_AS_WCSCAT
+	RETURN
+
+END (wcslen)
+#endif

libc/arch-x86/string/sse2-wcsrchr-atom.S

@@ -0,0 +1,402 @@
+/*
+Copyright (c) 2011 Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc	.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc	.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name, @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name, .-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG;	CFI_PUSH (REG)
+#define POP(REG)	popl REG;	CFI_POP (REG)
+
+#define PARMS  8
+#define ENTRANCE PUSH(%edi);
+#define RETURN  POP(%edi);	ret;	CFI_PUSH(%edi);
+
+#define STR1  PARMS
+#define STR2  STR1+4
+
+	.text
+ENTRY (wcsrchr)
+
+	ENTRANCE
+	mov	STR1(%esp), %ecx
+	movd	STR2(%esp), %xmm1
+
+	mov	%ecx, %edi
+	punpckldq %xmm1, %xmm1
+	pxor	%xmm2, %xmm2
+	punpckldq %xmm1, %xmm1
+
+/* ECX has OFFSET. */
+	and	$63, %ecx
+	cmp	$48, %ecx
+	ja	L(crosscache)
+
+/* unaligned string. */
+	movdqu	(%edi), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	pcmpeqd	%xmm1, %xmm0
+/* Find where NULL is.  */
+	pmovmskb %xmm2, %ecx
+/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+	add	$16, %edi
+
+	test	%eax, %eax
+	jnz	L(unaligned_match1)
+
+	test	%ecx, %ecx
+	jnz	L(return_null)
+
+	and	$-16, %edi
+
+	PUSH	(%esi)
+
+	xor	%edx, %edx
+	jmp	L(loop)
+
+	CFI_POP	(%esi)
+
+	.p2align 4
+L(unaligned_match1):
+	test	%ecx, %ecx
+	jnz	L(prolog_find_zero_1)
+
+	PUSH	(%esi)
+
+/* Save current match */
+	mov	%eax, %edx
+	mov	%edi, %esi
+	and	$-16, %edi
+	jmp	L(loop)
+
+	CFI_POP	(%esi)
+
+	.p2align 4
+L(crosscache):
+/* Hancle unaligned string.  */
+	and	$15, %ecx
+	and	$-16, %edi
+	pxor	%xmm3, %xmm3
+	movdqa	(%edi), %xmm0
+	pcmpeqd	%xmm0, %xmm3
+	pcmpeqd	%xmm1, %xmm0
+/* Find where NULL is.  */
+	pmovmskb %xmm3, %edx
+/* Check if there is a match.  */
+	pmovmskb %xmm0, %eax
+/* Remove the leading bytes.  */
+	shr	%cl, %edx
+	shr	%cl, %eax
+	add	$16, %edi
+
+	test	%eax, %eax
+	jnz	L(unaligned_match)
+
+	test	%edx, %edx
+	jnz	L(return_null)
+
+	PUSH	(%esi)
+
+	xor	%edx, %edx
+	jmp	L(loop)
+
+	CFI_POP	(%esi)
+
+	.p2align 4
+L(unaligned_match):
+	test	%edx, %edx
+	jnz	L(prolog_find_zero)
+
+	PUSH	(%esi)
+
+	mov	%eax, %edx
+	lea	(%edi, %ecx), %esi
+
+/* Loop start on aligned string.  */
+	.p2align 4
+L(loop):
+	movdqa	(%edi), %xmm0
+	pcmpeqd	%xmm0, %xmm2
+	add	$16, %edi
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm0, %eax
+	or	%eax, %ecx
+	jnz	L(matches)
+
+	movdqa	(%edi), %xmm3
+	pcmpeqd	%xmm3, %xmm2
+	add	$16, %edi
+	pcmpeqd	%xmm1, %xmm3
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm3, %eax
+	or	%eax, %ecx
+	jnz	L(matches)
+
+	movdqa	(%edi), %xmm4
+	pcmpeqd	%xmm4, %xmm2
+	add	$16, %edi
+	pcmpeqd	%xmm1, %xmm4
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm4, %eax
+	or	%eax, %ecx
+	jnz	L(matches)
+
+	movdqa	(%edi), %xmm5
+	pcmpeqd	%xmm5, %xmm2
+	add	$16, %edi
+	pcmpeqd	%xmm1, %xmm5
+	pmovmskb %xmm2, %ecx
+	pmovmskb %xmm5, %eax
+	or	%eax, %ecx
+	jz	L(loop)
+
+	.p2align 4
+L(matches):
+	test	%eax, %eax
+	jnz	L(match)
+L(return_value):
+	test	%edx, %edx
+	jz	L(return_null_1)
+	mov	%edx, %eax
+	mov	%esi, %edi
+
+	POP	(%esi)
+
+	test	%ah, %ah
+	jnz	L(match_third_or_fourth_wchar)
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(return_null_1):
+	POP	(%esi)
+
+	xor	%eax, %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(match):
+	pmovmskb %xmm2, %ecx
+	test	%ecx, %ecx
+	jnz	L(find_zero)
+/* save match info */
+	mov	%eax, %edx
+	mov	%edi, %esi
+	jmp	L(loop)
+
+	.p2align 4
+L(find_zero):
+	test	%cl, %cl
+	jz	L(find_zero_in_third_or_fourth_wchar)
+	test	$15, %cl
+	jz	L(find_zero_in_second_wchar)
+	and	$1, %eax
+	jz	L(return_value)
+
+	POP	(%esi)
+
+	lea	-16(%edi), %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(find_zero_in_second_wchar):
+	and	$1 << 5 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%esi)
+
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(find_zero_in_third_or_fourth_wchar):
+	test	$15, %ch
+	jz	L(find_zero_in_fourth_wchar)
+	and	$1 << 9 - 1, %eax
+	jz	L(return_value)
+
+	POP	(%esi)
+
+	test	%ah, %ah
+	jnz	L(match_third_wchar)
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(find_zero_in_fourth_wchar):
+
+	POP	(%esi)
+
+	test	%ah, %ah
+	jnz	L(match_third_or_fourth_wchar)
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(match_second_wchar):
+	lea	-12(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_third_or_fourth_wchar):
+	test	$15 << 4, %ah
+	jnz	L(match_fourth_wchar)
+	lea	-8(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_third_wchar):
+	lea	-8(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(match_fourth_wchar):
+	lea	-4(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(return_null):
+	xor	%eax, %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero):
+	add	%ecx, %edi
+	mov     %edx, %ecx
+L(prolog_find_zero_1):
+	test	%cl, %cl
+	jz	L(prolog_find_zero_in_third_or_fourth_wchar)
+	test	$15, %cl
+	jz	L(prolog_find_zero_in_second_wchar)
+	and	$1, %eax
+	jz	L(return_null)
+
+	lea	-16(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero_in_second_wchar):
+	and	$1 << 5 - 1, %eax
+	jz	L(return_null)
+
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero_in_third_or_fourth_wchar):
+	test	$15, %ch
+	jz	L(prolog_find_zero_in_fourth_wchar)
+	and	$1 << 9 - 1, %eax
+	jz	L(return_null)
+
+	test	%ah, %ah
+	jnz	L(match_third_wchar)
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+	.p2align 4
+L(prolog_find_zero_in_fourth_wchar):
+	test	%ah, %ah
+	jnz	L(match_third_or_fourth_wchar)
+	test	$15 << 4, %al
+	jnz	L(match_second_wchar)
+	lea	-16(%edi), %eax
+	RETURN
+
+END (wcsrchr)

libc/arch-x86/string/ssse3-bcopy-atom.S

@@ -28,13 +28,8 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#if defined(USE_SSE2)
 
-# define sse2_strlen_atom strlen
+#define MEMCPY	bcopy
-# include "sse2-strlen-atom.S"
+#define USE_AS_MEMMOVE
-
+#define USE_AS_BCOPY
-#else
+#include "ssse3-memcpy-atom.S"
-
-# include "strlen.S"
-
-#endif

libc/arch-x86/string/ssse3-memcmp-atom.S

@@ -1,5 +1,5 @@
 /*
-Copyright (c) 2010, Intel Corporation
+Copyright (c) 2010, 2011 Intel Corporation
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -28,18 +28,10 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#ifndef MEMCMP
-# define MEMCMP		ssse3_memcmp3_new
-#endif
-
 #ifndef L
 # define L(label)	.L##label
 #endif
 
-#ifndef ALIGN
-# define ALIGN(n)	.p2align n
-#endif
-
 #ifndef cfi_startproc
 # define cfi_startproc	.cfi_startproc
 #endif
@@ -83,6 +75,10 @@ name:					\
 	.size name, .-name
 #endif
 
+#ifndef MEMCMP
+# define MEMCMP	memcmp
+#endif
+
 #define CFI_PUSH(REG)	\
 	cfi_adjust_cfa_offset (4);	\
 	cfi_rel_offset (REG, 0)
@@ -101,22 +97,39 @@ name:					\
 #define RETURN_END	POP (%edi); POP (%esi); POP (%ebx); ret
 #define RETURN		RETURN_END; cfi_restore_state; cfi_remember_state
 
-	.section .text.ssse3,"ax",@progbits
+/* Warning!
+           wmemcmp has to use SIGNED comparison for elements.
+           memcmp has to use UNSIGNED comparison for elemnts.
+*/
+
+	.text
 ENTRY (MEMCMP)
 	movl	LEN(%esp), %ecx
+
+#ifdef USE_AS_WMEMCMP
+	shl	$2, %ecx
+	jz	L(zero)
+#endif
+
 	movl	BLK1(%esp), %eax
 	cmp	$48, %ecx
 	movl	BLK2(%esp), %edx
 	jae	L(48bytesormore)
+
+#ifndef USE_AS_WMEMCMP
 	cmp	$1, %ecx
 	jbe	L(less1bytes)
+#endif
+
 	PUSH	(%ebx)
 	add	%ecx, %edx
 	add	%ecx, %eax
 	jmp	L(less48bytes)
 
 	CFI_POP	(%ebx)
-	ALIGN (4)
+
+#ifndef USE_AS_WMEMCMP
+	.p2align 4
 L(less1bytes):
 	jb	L(zero)
 	movb	(%eax), %cl
@@ -127,13 +140,14 @@ L(less1bytes):
 	neg	%eax
 L(1bytesend):
 	ret
+#endif
 
-	ALIGN (4)
+	.p2align 4
 L(zero):
-	mov	$0, %eax
+	xor	%eax, %eax
 	ret
 
-	ALIGN (4)
+	.p2align 4
 L(48bytesormore):
 	PUSH	(%ebx)
 	PUSH	(%esi)
@@ -160,6 +174,7 @@ L(48bytesormore):
 	jz	L(shr_0)
 	xor	%edx, %esi
 
+#ifndef USE_AS_WMEMCMP
 	cmp	$8, %edx
 	jae	L(next_unaligned_table)
 	cmp	$0, %edx
@@ -178,7 +193,7 @@ L(48bytesormore):
 	je	L(shr_6)
 	jmp	L(shr_7)
 
-	ALIGN (4)
+	.p2align 2
 L(next_unaligned_table):
 	cmp	$8, %edx
 	je	L(shr_8)
@@ -195,8 +210,17 @@ L(next_unaligned_table):
 	cmp	$14, %edx
 	je	L(shr_14)
 	jmp	L(shr_15)
+#else
+	cmp	$0, %edx
+	je	L(shr_0)
+	cmp	$4, %edx
+	je	L(shr_4)
+	cmp	$8, %edx
+	je	L(shr_8)
+	jmp	L(shr_12)
+#endif
 
-	ALIGN (4)
+	.p2align 4
 L(shr_0):
 	cmp	$80, %ecx
 	jae	L(shr_0_gobble)
@@ -221,7 +245,7 @@ L(shr_0):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_0_gobble):
 	lea	-48(%ecx), %ecx
 	movdqa	(%esi), %xmm0
@@ -265,9 +289,10 @@ L(shr_0_gobble_loop_next):
 	POP	(%esi)
 	jmp	L(less48bytes)
 
+#ifndef USE_AS_WMEMCMP
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_1):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -297,7 +322,7 @@ L(shr_1):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_1_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -348,9 +373,10 @@ L(shr_1_gobble_next):
 	POP	(%esi)
 	jmp	L(less48bytes)
 
+
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_2):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -380,7 +406,7 @@ L(shr_2):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_2_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -433,7 +459,7 @@ L(shr_2_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_3):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -463,7 +489,7 @@ L(shr_3):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_3_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -513,10 +539,11 @@ L(shr_3_gobble_next):
 	POP	(%edi)
 	POP	(%esi)
 	jmp	L(less48bytes)
+#endif
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_4):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -546,7 +573,7 @@ L(shr_4):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_4_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -597,9 +624,10 @@ L(shr_4_gobble_next):
 	POP	(%esi)
 	jmp	L(less48bytes)
 
+#ifndef USE_AS_WMEMCMP
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_5):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -629,7 +657,7 @@ L(shr_5):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_5_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -682,7 +710,7 @@ L(shr_5_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_6):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -712,7 +740,7 @@ L(shr_6):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_6_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -765,7 +793,7 @@ L(shr_6_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_7):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -795,7 +823,7 @@ L(shr_7):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_7_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -845,10 +873,11 @@ L(shr_7_gobble_next):
 	POP	(%edi)
 	POP	(%esi)
 	jmp	L(less48bytes)
+#endif
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_8):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -878,7 +907,7 @@ L(shr_8):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_8_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -929,9 +958,10 @@ L(shr_8_gobble_next):
 	POP	(%esi)
 	jmp	L(less48bytes)
 
+#ifndef USE_AS_WMEMCMP
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_9):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -961,7 +991,7 @@ L(shr_9):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_9_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1014,7 +1044,7 @@ L(shr_9_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_10):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -1044,7 +1074,7 @@ L(shr_10):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_10_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1097,7 +1127,7 @@ L(shr_10_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_11):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -1127,7 +1157,7 @@ L(shr_11):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_11_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1177,10 +1207,11 @@ L(shr_11_gobble_next):
 	POP	(%edi)
 	POP	(%esi)
 	jmp	L(less48bytes)
+#endif
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_12):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -1210,7 +1241,7 @@ L(shr_12):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_12_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1261,9 +1292,10 @@ L(shr_12_gobble_next):
 	POP	(%esi)
 	jmp	L(less48bytes)
 
+#ifndef USE_AS_WMEMCMP
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_13):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -1293,7 +1325,7 @@ L(shr_13):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_13_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1346,7 +1378,7 @@ L(shr_13_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_14):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -1376,7 +1408,7 @@ L(shr_14):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_14_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1429,7 +1461,7 @@ L(shr_14_gobble_next):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_15):
 	cmp	$80, %ecx
 	lea	-48(%ecx), %ecx
@@ -1459,7 +1491,7 @@ L(shr_15):
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(shr_15_gobble):
 	sub	$32, %ecx
 	movdqa	16(%esi), %xmm0
@@ -1509,10 +1541,11 @@ L(shr_15_gobble_next):
 	POP	(%edi)
 	POP	(%esi)
 	jmp	L(less48bytes)
+#endif
 
 	cfi_restore_state
 	cfi_remember_state
-	ALIGN (4)
+	.p2align 4
 L(exit):
 	pmovmskb %xmm1, %ebx
 	sub	$0xffff, %ebx
@@ -1520,9 +1553,12 @@ L(exit):
 	lea	-16(%esi), %esi
 	lea	-16(%edi), %edi
 	mov	%ebx, %edx
+
 L(first16bytes):
 	add	%eax, %esi
 L(less16bytes):
+
+#ifndef USE_AS_WMEMCMP
 	test	%dl, %dl
 	jz	L(next_24_bytes)
 
@@ -1552,56 +1588,56 @@ L(Byte23):
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte16):
 	movzbl	-16(%edi), %eax
 	movzbl	-16(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte17):
 	movzbl	-15(%edi), %eax
 	movzbl	-15(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte18):
 	movzbl	-14(%edi), %eax
 	movzbl	-14(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte19):
 	movzbl	-13(%edi), %eax
 	movzbl	-13(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte20):
 	movzbl	-12(%edi), %eax
 	movzbl	-12(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte21):
 	movzbl	-11(%edi), %eax
 	movzbl	-11(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(Byte22):
 	movzbl	-10(%edi), %eax
 	movzbl	-10(%esi), %edx
 	sub	%edx, %eax
 	RETURN
 
-	ALIGN (4)
+	.p2align 4
 L(next_24_bytes):
 	lea	8(%edi), %edi
 	lea	8(%esi), %esi
@@ -1626,20 +1662,70 @@ L(next_24_bytes):
 	test	$0x40, %dh
 	jnz	L(Byte22)
 
-	ALIGN (4)
+	.p2align 4
 L(Byte31):
 	movzbl	-9(%edi), %eax
 	movzbl	-9(%esi), %edx
 	sub	%edx, %eax
 	RETURN_END
+#else
+
+/* special for wmemcmp */
+	test	%dl, %dl
+	jz	L(next_two_double_words)
+	and	$15, %dl
+	jz	L(second_double_word)
+	mov	-16(%edi), %ecx
+	cmp	-16(%esi), %ecx
+	mov	$1, %eax
+	jg	L(nequal_bigger)
+	neg	%eax
+	RETURN
+
+
+	.p2align 4
+L(second_double_word):
+	mov	-12(%edi), %ecx
+	cmp	-12(%esi), %ecx
+	mov	$1, %eax
+	jg	L(nequal_bigger)
+	neg	%eax
+	RETURN
+	
+	.p2align 4
+L(next_two_double_words):
+	and	$15, %dh
+	jz	L(fourth_double_word)
+	mov	-8(%edi), %ecx
+	cmp	-8(%esi), %ecx
+	mov	$1, %eax
+	jg	L(nequal_bigger)
+	neg	%eax
+	RETURN
+
+	.p2align 4
+L(fourth_double_word):
+	mov	-4(%edi), %ecx
+	cmp	-4(%esi), %ecx
+	mov	$1, %eax
+	jg	L(nequal_bigger)
+	neg	%eax
+	RETURN
+
+	.p2align 4
+L(nequal_bigger):
+	RETURN_END
+#endif
+
 	CFI_PUSH (%ebx)
 
-	ALIGN (4)
+	.p2align 4
 L(more8bytes):
 	cmp	$16, %ecx
 	jae	L(more16bytes)
 	cmp	$8, %ecx
 	je	L(8bytes)
+#ifndef USE_AS_WMEMCMP
 	cmp	$9, %ecx
 	je	L(9bytes)
 	cmp	$10, %ecx
@@ -1653,13 +1739,17 @@ L(more8bytes):
 	cmp	$14, %ecx
 	je	L(14bytes)
 	jmp	L(15bytes)
+#else
+	jmp	L(12bytes)
+#endif
 
-	ALIGN (4)
+	.p2align 4
 L(more16bytes):
 	cmp	$24, %ecx
 	jae	L(more24bytes)
 	cmp	$16, %ecx
 	je	L(16bytes)
+#ifndef USE_AS_WMEMCMP
 	cmp	$17, %ecx
 	je	L(17bytes)
 	cmp	$18, %ecx
@@ -1673,13 +1763,17 @@ L(more16bytes):
 	cmp	$22, %ecx
 	je	L(22bytes)
 	jmp	L(23bytes)
+#else
+	jmp	L(20bytes)
+#endif
 
-	ALIGN (4)
+	.p2align 4
 L(more24bytes):
 	cmp	$32, %ecx
 	jae	L(more32bytes)
 	cmp	$24, %ecx
 	je	L(24bytes)
+#ifndef USE_AS_WMEMCMP
 	cmp	$25, %ecx
 	je	L(25bytes)
 	cmp	$26, %ecx
@@ -1693,13 +1787,17 @@ L(more24bytes):
 	cmp	$30, %ecx
 	je	L(30bytes)
 	jmp	L(31bytes)
+#else
+	jmp	L(28bytes)
+#endif
 
-	ALIGN (4)
+	.p2align 4
 L(more32bytes):
 	cmp	$40, %ecx
 	jae	L(more40bytes)
 	cmp	$32, %ecx
 	je	L(32bytes)
+#ifndef USE_AS_WMEMCMP
 	cmp	$33, %ecx
 	je	L(33bytes)
 	cmp	$34, %ecx
@@ -1713,11 +1811,35 @@ L(more32bytes):
 	cmp	$38, %ecx
 	je	L(38bytes)
 	jmp	L(39bytes)
+#else
+	jmp	L(36bytes)
+#endif
 
-	ALIGN (4)
+	.p2align 4
+L(less48bytes):
+	cmp	$8, %ecx
+	jae	L(more8bytes)
+#ifndef USE_AS_WMEMCMP
+	cmp	$2, %ecx
+	je	L(2bytes)
+	cmp	$3, %ecx
+	je	L(3bytes)
+	cmp	$4, %ecx
+	je	L(4bytes)
+	cmp	$5, %ecx
+	je	L(5bytes)
+	cmp	$6, %ecx
+	je	L(6bytes)
+	jmp	L(7bytes)
+#else
+	jmp	L(4bytes)
+#endif
+
+	.p2align 4
 L(more40bytes):
 	cmp	$40, %ecx
 	je	L(40bytes)
+#ifndef USE_AS_WMEMCMP
 	cmp	$41, %ecx
 	je	L(41bytes)
 	cmp	$42, %ecx
@@ -1732,24 +1854,7 @@ L(more40bytes):
 	je	L(46bytes)
 	jmp	L(47bytes)
 
-	ALIGN (4)
+	.p2align 4
-L(less48bytes):
-	cmp	$8, %ecx
-	jae	L(more8bytes)
-	cmp	$2, %ecx
-	je	L(2bytes)
-	cmp	$3, %ecx
-	je	L(3bytes)
-	cmp	$4, %ecx
-	je	L(4bytes)
-	cmp	$5, %ecx
-	je	L(5bytes)
-	cmp	$6, %ecx
-	je	L(6bytes)
-	jmp	L(7bytes)
-
-
-	ALIGN (4)
 L(44bytes):
 	mov	-44(%eax), %ecx
 	mov	-44(%edx), %ebx
@@ -1809,8 +1914,61 @@ L(4bytes):
 	POP	(%ebx)
 	ret
 	CFI_PUSH (%ebx)
+#else
+	.p2align 4
+L(44bytes):
+	mov	-44(%eax), %ecx
+	cmp	-44(%edx), %ecx
+	jne	L(find_diff)
+L(40bytes):
+	mov	-40(%eax), %ecx
+	cmp	-40(%edx), %ecx
+	jne	L(find_diff)
+L(36bytes):
+	mov	-36(%eax), %ecx
+	cmp	-36(%edx), %ecx
+	jne	L(find_diff)
+L(32bytes):
+	mov	-32(%eax), %ecx
+	cmp	-32(%edx), %ecx
+	jne	L(find_diff)
+L(28bytes):
+	mov	-28(%eax), %ecx
+	cmp	-28(%edx), %ecx
+	jne	L(find_diff)
+L(24bytes):
+	mov	-24(%eax), %ecx
+	cmp	-24(%edx), %ecx
+	jne	L(find_diff)
+L(20bytes):
+	mov	-20(%eax), %ecx
+	cmp	-20(%edx), %ecx
+	jne	L(find_diff)
+L(16bytes):
+	mov	-16(%eax), %ecx
+	cmp	-16(%edx), %ecx
+	jne	L(find_diff)
+L(12bytes):
+	mov	-12(%eax), %ecx
+	cmp	-12(%edx), %ecx
+	jne	L(find_diff)
+L(8bytes):
+	mov	-8(%eax), %ecx
+	cmp	-8(%edx), %ecx
+	jne	L(find_diff)
+L(4bytes):
+	mov	-4(%eax), %ecx
+	xor	%eax, %eax
+	cmp	-4(%edx), %ecx
+	jne	L(find_diff)
+	POP	(%ebx)
+	ret
+	CFI_PUSH (%ebx)
+#endif
 
-	ALIGN (4)
+#ifndef USE_AS_WMEMCMP
+
+	.p2align 4
 L(45bytes):
 	mov	-45(%eax), %ecx
 	mov	-45(%edx), %ebx
@@ -1874,7 +2032,7 @@ L(5bytes):
 	ret
 	CFI_PUSH (%ebx)
 
-	ALIGN (4)
+	.p2align 4
 L(46bytes):
 	mov	-46(%eax), %ecx
 	mov	-46(%edx), %ebx
@@ -1942,7 +2100,7 @@ L(2bytes):
 	ret
 	CFI_PUSH (%ebx)
 
-	ALIGN (4)
+	.p2align 4
 L(47bytes):
 	movl	-47(%eax), %ecx
 	movl	-47(%edx), %ebx
@@ -2013,7 +2171,7 @@ L(3bytes):
 	ret
 	CFI_PUSH (%ebx)
 
-	ALIGN (4)
+	.p2align 4
 L(find_diff):
 	cmpb	%bl, %cl
 	jne	L(end)
@@ -2024,6 +2182,8 @@ L(find_diff):
 	cmp	%bl, %cl
 	jne	L(end)
 	cmp	%bx, %cx
+
+	.p2align 4
 L(end):
 	POP	(%ebx)
 	mov	$1, %eax
@@ -2031,5 +2191,20 @@ L(end):
 	neg	%eax
 L(bigger):
 	ret
+#else
 
+/* for wmemcmp */
+	.p2align 4
+L(find_diff):
+	POP	(%ebx)
+	mov	$1, %eax
+	jg	L(find_diff_bigger)
+	neg	%eax
+	ret
+
+	.p2align 4
+L(find_diff_bigger):
+	ret
+
+#endif
 END (MEMCMP)

libc/arch-x86/string/ssse3-memcpy-atom.S

@@ -28,8 +28,11 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
+#include "cache.h"
+#undef __i686
+
 #ifndef MEMCPY
-# define MEMCPY	ssse3_memcpy5
+# define MEMCPY	memcpy
 #endif
 
 #ifndef L

libc/arch-x86/string/ssse3-memmove-atom.S

@@ -28,13 +28,7 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#if defined(USE_SSSE3)
 
-# define ssse3_strcmp_latest strcmp
+#define MEMCPY memmove
-# include "ssse3-strcmp-latest.S"
+#define USE_AS_MEMMOVE
-
+#include "ssse3-memcpy-atom.S"
-#else
-
-# include "strcmp.S"
-
-#endif

libc/arch-x86/string/ssse3-strcat-atom.S

@@ -0,0 +1,620 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef L
+# define L(label)	.L##label
+#endif
+
+#ifndef cfi_startproc
+# define cfi_startproc			.cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc			.cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)		.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef cfi_remember_state
+# define cfi_remember_state		.cfi_remember_state
+#endif
+
+#ifndef cfi_restore_state
+# define cfi_restore_state		.cfi_restore_state
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)			\
+	.type name,  @function; 	\
+	.globl name;			\
+	.p2align 4;			\
+name:					\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)			\
+	cfi_endproc;			\
+	.size name, .-name
+#endif
+
+#define CFI_PUSH(REG)			\
+  cfi_adjust_cfa_offset (4);		\
+  cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)			\
+  cfi_adjust_cfa_offset (-4);		\
+  cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+#define POP(REG)	popl REG; CFI_POP (REG)
+
+#ifndef STRCAT
+# define STRCAT	strcat
+#endif
+
+#define PARMS	4
+#define STR1	PARMS+4
+#define STR2	STR1+4
+
+#ifdef USE_AS_STRNCAT
+# define LEN	STR2+8
+#endif
+
+#define USE_AS_STRCAT
+
+	.section .text.ssse3,"ax",@progbits
+ENTRY (STRCAT)
+	PUSH	(%edi)
+	mov	STR1(%esp), %edi
+	mov	%edi, %edx
+
+#define RETURN	jmp	L(StrcpyAtom)
+#include "sse2-strlen-atom.S"
+
+L(StrcpyAtom):
+	mov	STR2(%esp), %ecx
+	lea	(%edi, %eax), %edx
+#ifdef USE_AS_STRNCAT
+	PUSH	(%ebx)
+	mov	LEN(%esp), %ebx
+	test	%ebx, %ebx
+	jz	L(StrncatExit0)
+	cmp	$8, %ebx
+	jbe	L(StrncpyExit8Bytes)
+#endif
+	cmpb	$0, (%ecx)
+	jz	L(Exit1)
+	cmpb	$0, 1(%ecx)
+	jz	L(Exit2)
+	cmpb	$0, 2(%ecx)
+	jz	L(Exit3)
+	cmpb	$0, 3(%ecx)
+	jz	L(Exit4)
+	cmpb	$0, 4(%ecx)
+	jz	L(Exit5)
+	cmpb	$0, 5(%ecx)
+	jz	L(Exit6)
+	cmpb	$0, 6(%ecx)
+	jz	L(Exit7)
+	cmpb	$0, 7(%ecx)
+	jz	L(Exit8)
+	cmpb	$0, 8(%ecx)
+	jz	L(Exit9)
+#ifdef USE_AS_STRNCAT
+	cmp	$16, %ebx
+	jb	L(StrncpyExit15Bytes)
+#endif
+	cmpb	$0, 9(%ecx)
+	jz	L(Exit10)
+	cmpb	$0, 10(%ecx)
+	jz	L(Exit11)
+	cmpb	$0, 11(%ecx)
+	jz	L(Exit12)
+	cmpb	$0, 12(%ecx)
+	jz	L(Exit13)
+	cmpb	$0, 13(%ecx)
+	jz	L(Exit14)
+	cmpb	$0, 14(%ecx)
+	jz	L(Exit15)
+	cmpb	$0, 15(%ecx)
+	jz	L(Exit16)
+#ifdef USE_AS_STRNCAT
+	cmp	$16, %ebx
+	je	L(StrncatExit16)
+
+# define RETURN1	POP (%ebx); POP (%edi);	ret; \
+	CFI_PUSH (%ebx); CFI_PUSH (%edi)
+# define USE_AS_STRNCPY
+#else
+# define RETURN1	POP(%edi); ret; CFI_PUSH(%edi)
+#endif
+#include "ssse3-strcpy-atom.S"
+
+	.p2align 4
+L(CopyFrom1To16Bytes):
+	add	%esi, %edx
+	add	%esi, %ecx
+
+	POP	(%esi)
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit1)
+	test	$0x02, %al
+	jnz	L(Exit2)
+	test	$0x04, %al
+	jnz	L(Exit3)
+	test	$0x08, %al
+	jnz	L(Exit4)
+	test	$0x10, %al
+	jnz	L(Exit5)
+	test	$0x20, %al
+	jnz	L(Exit6)
+	test	$0x40, %al
+	jnz	L(Exit7)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(ExitHigh):
+	test	$0x01, %ah
+	jnz	L(Exit9)
+	test	$0x02, %ah
+	jnz	L(Exit10)
+	test	$0x04, %ah
+	jnz	L(Exit11)
+	test	$0x08, %ah
+	jnz	L(Exit12)
+	test	$0x10, %ah
+	jnz	L(Exit13)
+	test	$0x20, %ah
+	jnz	L(Exit14)
+	test	$0x40, %ah
+	jnz	L(Exit15)
+	movlpd	(%ecx), %xmm0
+	movlpd	8(%ecx), %xmm1
+	movlpd	%xmm0, (%edx)
+	movlpd	%xmm1, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit1):
+	movb	%bh, 1(%edx)
+L(Exit1):
+	movb	(%ecx), %al
+	movb	%al, (%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit2):
+	movb	%bh, 2(%edx)
+L(Exit2):
+	movw	(%ecx), %ax
+	movw	%ax, (%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit3):
+	movb	%bh, 3(%edx)
+L(Exit3):
+	movw	(%ecx), %ax
+	movw	%ax, (%edx)
+	movb	2(%ecx), %al
+	movb	%al, 2(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit4):
+	movb	%bh, 4(%edx)
+L(Exit4):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit5):
+	movb	%bh, 5(%edx)
+L(Exit5):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movb	4(%ecx), %al
+	movb	%al, 4(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit6):
+	movb	%bh, 6(%edx)
+L(Exit6):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movw	4(%ecx), %ax
+	movw	%ax, 4(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit7):
+	movb	%bh, 7(%edx)
+L(Exit7):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movl	3(%ecx), %eax
+	movl	%eax, 3(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit8):
+	movb	%bh, 8(%edx)
+L(Exit8):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit9):
+	movb	%bh, 9(%edx)
+L(Exit9):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movb	8(%ecx), %al
+	movb	%al, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit10):
+	movb	%bh, 10(%edx)
+L(Exit10):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movw	8(%ecx), %ax
+	movw	%ax, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit11):
+	movb	%bh, 11(%edx)
+L(Exit11):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	7(%ecx), %eax
+	movl	%eax, 7(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit12):
+	movb	%bh, 12(%edx)
+L(Exit12):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	8(%ecx), %eax
+	movl	%eax, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit13):
+	movb	%bh, 13(%edx)
+L(Exit13):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movlpd	5(%ecx), %xmm0
+	movlpd	%xmm0, 5(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit14):
+	movb	%bh, 14(%edx)
+L(Exit14):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movlpd	6(%ecx), %xmm0
+	movlpd	%xmm0, 6(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit15):
+	movb	%bh, 15(%edx)
+L(Exit15):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movlpd	7(%ecx), %xmm0
+	movlpd	%xmm0, 7(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit16):
+	movb	%bh, 16(%edx)
+L(Exit16):
+	movlpd	(%ecx), %xmm0
+	movlpd	8(%ecx), %xmm1
+	movlpd	%xmm0, (%edx)
+	movlpd	%xmm1, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+#ifdef USE_AS_STRNCPY
+
+	CFI_PUSH(%esi)
+
+	.p2align 4
+L(CopyFrom1To16BytesCase2):
+	add	$16, %ebx
+	add	%esi, %ecx
+	lea	(%esi, %edx), %esi
+	lea	-9(%ebx), %edx
+	and	$1<<7, %dh
+	or	%al, %dh
+	lea	(%esi), %edx
+	POP	(%esi)
+	jz	L(ExitHighCase2)
+
+	test	$0x01, %al
+	jnz	L(Exit1)
+	cmp	$1, %ebx
+	je	L(StrncatExit1)
+	test	$0x02, %al
+	jnz	L(Exit2)
+	cmp	$2, %ebx
+	je	L(StrncatExit2)
+	test	$0x04, %al
+	jnz	L(Exit3)
+	cmp	$3, %ebx
+	je	L(StrncatExit3)
+	test	$0x08, %al
+	jnz	L(Exit4)
+	cmp	$4, %ebx
+	je	L(StrncatExit4)
+	test	$0x10, %al
+	jnz	L(Exit5)
+	cmp	$5, %ebx
+	je	L(StrncatExit5)
+	test	$0x20, %al
+	jnz	L(Exit6)
+	cmp	$6, %ebx
+	je	L(StrncatExit6)
+	test	$0x40, %al
+	jnz	L(Exit7)
+	cmp	$7, %ebx
+	je	L(StrncatExit7)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	lea	7(%edx), %eax
+	cmpb	$1, (%eax)
+	sbb	$-1, %eax
+	xor	%cl, %cl
+	movb	%cl, (%eax)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(ExitHighCase2):
+	test	$0x01, %ah
+	jnz	L(Exit9)
+	cmp	$9, %ebx
+	je	L(StrncatExit9)
+	test	$0x02, %ah
+	jnz	L(Exit10)
+	cmp	$10, %ebx
+	je	L(StrncatExit10)
+	test	$0x04, %ah
+	jnz	L(Exit11)
+	cmp	$11, %ebx
+	je	L(StrncatExit11)
+	test	$0x8, %ah
+	jnz	L(Exit12)
+	cmp	$12, %ebx
+	je	L(StrncatExit12)
+	test	$0x10, %ah
+	jnz	L(Exit13)
+	cmp	$13, %ebx
+	je	L(StrncatExit13)
+	test	$0x20, %ah
+	jnz	L(Exit14)
+	cmp	$14, %ebx
+	je	L(StrncatExit14)
+	test	$0x40, %ah
+	jnz	L(Exit15)
+	cmp	$15, %ebx
+	je	L(StrncatExit15)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movlpd	8(%ecx), %xmm1
+	movlpd	%xmm1, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	CFI_PUSH(%esi)
+
+L(CopyFrom1To16BytesCase2OrCase3):
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16BytesCase2)
+
+	.p2align 4
+L(CopyFrom1To16BytesCase3):
+	add	$16, %ebx
+	add	%esi, %edx
+	add	%esi, %ecx
+
+	POP	(%esi)
+
+	cmp	$8, %ebx
+	ja	L(ExitHighCase3)
+	cmp	$1, %ebx
+	je	L(StrncatExit1)
+	cmp	$2, %ebx
+	je	L(StrncatExit2)
+	cmp	$3, %ebx
+	je	L(StrncatExit3)
+	cmp	$4, %ebx
+	je	L(StrncatExit4)
+	cmp	$5, %ebx
+	je	L(StrncatExit5)
+	cmp	$6, %ebx
+	je	L(StrncatExit6)
+	cmp	$7, %ebx
+	je	L(StrncatExit7)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movb	%bh, 8(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(ExitHighCase3):
+	cmp	$9, %ebx
+	je	L(StrncatExit9)
+	cmp	$10, %ebx
+	je	L(StrncatExit10)
+	cmp	$11, %ebx
+	je	L(StrncatExit11)
+	cmp	$12, %ebx
+	je	L(StrncatExit12)
+	cmp	$13, %ebx
+	je	L(StrncatExit13)
+	cmp	$14, %ebx
+	je	L(StrncatExit14)
+	cmp	$15, %ebx
+	je	L(StrncatExit15)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movlpd	8(%ecx), %xmm1
+	movlpd	%xmm1, 8(%edx)
+	movb	%bh, 16(%edx)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncatExit0):
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncpyExit15Bytes):
+	cmp	$9, %ebx
+	je	L(StrncatExit9)
+	cmpb	$0, 9(%ecx)
+	jz	L(Exit10)
+	cmp	$10, %ebx
+	je	L(StrncatExit10)
+	cmpb	$0, 10(%ecx)
+	jz	L(Exit11)
+	cmp	$11, %ebx
+	je	L(StrncatExit11)
+	cmpb	$0, 11(%ecx)
+	jz	L(Exit12)
+	cmp	$12, %ebx
+	je	L(StrncatExit12)
+	cmpb	$0, 12(%ecx)
+	jz	L(Exit13)
+	cmp	$13, %ebx
+	je	L(StrncatExit13)
+	cmpb	$0, 13(%ecx)
+	jz	L(Exit14)
+	cmp	$14, %ebx
+	je	L(StrncatExit14)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movlpd	7(%ecx), %xmm0
+	movlpd	%xmm0, 7(%edx)
+	lea	14(%edx), %eax
+	cmpb	$1, (%eax)
+	sbb	$-1, %eax
+	movb	%bh, (%eax)
+	movl	%edi, %eax
+	RETURN1
+
+	.p2align 4
+L(StrncpyExit8Bytes):
+	cmpb	$0, (%ecx)
+	jz	L(Exit1)
+	cmp	$1, %ebx
+	je	L(StrncatExit1)
+	cmpb	$0, 1(%ecx)
+	jz	L(Exit2)
+	cmp	$2, %ebx
+	je	L(StrncatExit2)
+	cmpb	$0, 2(%ecx)
+	jz	L(Exit3)
+	cmp	$3, %ebx
+	je	L(StrncatExit3)
+	cmpb	$0, 3(%ecx)
+	jz	L(Exit4)
+	cmp	$4, %ebx
+	je	L(StrncatExit4)
+	cmpb	$0, 4(%ecx)
+	jz	L(Exit5)
+	cmp	$5, %ebx
+	je	L(StrncatExit5)
+	cmpb	$0, 5(%ecx)
+	jz	L(Exit6)
+	cmp	$6, %ebx
+	je	L(StrncatExit6)
+	cmpb	$0, 6(%ecx)
+	jz	L(Exit7)
+	cmp	$7, %ebx
+	je	L(StrncatExit7)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	lea	7(%edx), %eax
+	cmpb	$1, (%eax)
+	sbb	$-1, %eax
+	movb	%bh, (%eax)
+	movl	%edi, %eax
+	RETURN1
+
+#endif
+END (STRCAT)

libc/arch-x86/string/ssse3-strcmp-atom.S

@@ -107,8 +107,12 @@ name:					\
 	sub	%esi, %ebp
 #endif
 
+#ifndef STRCMP
+# define STRCMP strcmp
+#endif
+
 	.section .text.ssse3,"ax",@progbits
-ENTRY (ssse3_strcmp_latest)
+ENTRY (STRCMP)
 #ifdef USE_AS_STRNCMP
 	PUSH	(%ebp)
 #endif
@@ -2271,4 +2275,4 @@ L(less16bytes_sncmp):
 	ret
 #endif
 
-END (ssse3_strcmp_latest)
+END (STRCMP)

libc/arch-x86/string/ssse3-strncat-atom.S

@@ -0,0 +1,34 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#define STRCAT  strncat
+#define USE_AS_STRNCAT
+
+#include "ssse3-strcat-atom.S"

libc/arch-x86/string/ssse3-strncmp-atom.S

@@ -28,13 +28,8 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-#if defined(USE_SSSE3)
 
-# define MEMCMP memcmp
+#define USE_AS_STRNCMP
-# include "ssse3-memcmp3-new.S"
+#define STRCMP  strncmp
+#include "ssse3-strcmp-atom.S"
 
-#else
-
-# include "memcmp.S"
-
-#endif

libc/arch-x86/string/ssse3-strncpy-atom.S

@@ -0,0 +1,33 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#define USE_AS_STRNCPY
+#define STRCPY strncpy
+#include "ssse3-strcpy-atom.S"

libc/arch-x86/string/ssse3-wcscat-atom.S

@@ -1,5 +1,5 @@
 /*
-Copyright (c) 2010, Intel Corporation
+Copyright (c) 2011 Intel Corporation
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
@@ -28,18 +28,87 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
 
-
+#ifndef L
-#if defined(USE_SSSE3)
+# define L(label)	.L##label
-
-# include "cache_wrapper.S"
-# undef __i686
-# define MEMCPY	bcopy
-# define USE_AS_MEMMOVE
-# define USE_AS_BCOPY
-# include "ssse3-memcpy5.S"
-
-#else
-
-# include "bcopy.S"
-
 #endif
+
+#ifndef cfi_startproc
+# define cfi_startproc                  .cfi_startproc
+#endif
+
+#ifndef cfi_endproc
+# define cfi_endproc                    .cfi_endproc
+#endif
+
+#ifndef cfi_rel_offset
+# define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+#endif
+
+#ifndef cfi_restore
+# define cfi_restore(reg)	.cfi_restore reg
+#endif
+
+#ifndef cfi_adjust_cfa_offset
+# define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+#endif
+
+#ifndef ENTRY
+# define ENTRY(name)	\
+	.type name,  @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+#endif
+
+#ifndef END
+# define END(name)	\
+	cfi_endproc;	\
+	.size name, .-name
+#endif
+
+#define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG;	CFI_PUSH (REG)
+#define POP(REG)	popl REG;	CFI_POP (REG)
+
+#define PARMS  4
+#define STR1  PARMS+4
+#define STR2  STR1+4
+
+#define USE_AS_WCSCAT
+
+.text
+ENTRY (wcscat)
+	PUSH    (%edi)
+	mov	STR1(%esp), %edi
+	mov	%edi, %edx
+
+#define RETURN  jmp L(WcscpyAtom)
+#include "sse2-wcslen-atom.S"
+
+L(WcscpyAtom):
+	shl	$2, %eax
+	mov	STR2(%esp), %ecx
+	lea	(%edi, %eax), %edx
+
+	cmp	$0, (%ecx)
+	jz	L(Exit4)
+	cmp	$0, 4(%ecx)
+	jz	L(Exit8)
+	cmp	$0, 8(%ecx)
+	jz	L(Exit12)
+	cmp	$0, 12(%ecx)
+	jz	L(Exit16)
+
+#undef RETURN
+#define RETURN  POP(%edi);	ret;	CFI_PUSH(%edi)
+#include "ssse3-wcscpy-atom.S"
+
+END (wcscat)

libc/arch-x86/string/ssse3-wcscpy-atom.S

@@ -0,0 +1,652 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef USE_AS_WCSCAT
+
+# ifndef L
+#  define L(label)	.L##label
+# endif
+
+# ifndef cfi_startproc
+#  define cfi_startproc	.cfi_startproc
+# endif
+
+# ifndef cfi_endproc
+#  define cfi_endproc	.cfi_endproc
+# endif
+
+# ifndef cfi_rel_offset
+#  define cfi_rel_offset(reg, off)	.cfi_rel_offset reg, off
+# endif
+
+# ifndef cfi_restore
+#  define cfi_restore(reg)	.cfi_restore reg
+# endif
+
+# ifndef cfi_adjust_cfa_offset
+#  define cfi_adjust_cfa_offset(off)	.cfi_adjust_cfa_offset off
+# endif
+
+# ifndef ENTRY
+#  define ENTRY(name)	\
+	.type name, @function;	\
+	.globl name;	\
+	.p2align 4;	\
+name:	\
+	cfi_startproc
+# endif
+
+# ifndef END
+#  define END(name)	\
+	cfi_endproc;	\
+	.size name, .-name
+# endif
+
+# define CFI_PUSH(REG)	\
+	cfi_adjust_cfa_offset (4);	\
+	cfi_rel_offset (REG, 0)
+
+# define CFI_POP(REG)	\
+	cfi_adjust_cfa_offset (-4);	\
+	cfi_restore (REG)
+
+# define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+# define POP(REG)	popl REG; CFI_POP (REG)
+
+# define PARMS	4
+# define RETURN	POP (%edi); ret; CFI_PUSH (%edi)
+
+# define STR1	PARMS
+# define STR2	STR1+4
+# define LEN	STR2+4
+
+.text
+ENTRY (wcscpy)
+	mov	STR1(%esp), %edx
+	mov	STR2(%esp), %ecx
+
+	cmp	$0, (%ecx)
+	jz	L(ExitTail4)
+	cmp	$0, 4(%ecx)
+	jz	L(ExitTail8)
+	cmp	$0, 8(%ecx)
+	jz	L(ExitTail12)
+	cmp	$0, 12(%ecx)
+	jz	L(ExitTail16)
+
+	PUSH	(%edi)
+	mov	%edx, %edi
+#endif
+	PUSH	(%esi)
+	lea	16(%ecx), %esi
+
+	and	$-16, %esi
+
+	pxor	%xmm0, %xmm0
+	pcmpeqd	(%esi), %xmm0
+	movdqu	(%ecx), %xmm1
+	movdqu	%xmm1, (%edx)
+
+	pmovmskb %xmm0, %eax
+	sub	%ecx, %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	mov	%edx, %eax
+	lea	16(%edx), %edx
+	and	$-16, %edx
+	sub	%edx, %eax
+
+	sub	%eax, %ecx
+	mov	%ecx, %eax
+	and	$0xf, %eax
+	mov	$0, %esi
+
+	jz	L(Align16Both)
+	cmp	$4, %eax
+	je	L(Shl4)
+	cmp	$8, %eax
+	je	L(Shl8)
+	jmp	L(Shl12)
+
+L(Align16Both):
+	movaps	(%ecx), %xmm1
+	movaps	16(%ecx), %xmm2
+	movaps	%xmm1, (%edx)
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm3
+	movaps	%xmm2, (%edx, %esi)
+	pcmpeqd	%xmm3, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm4
+	movaps	%xmm3, (%edx, %esi)
+	pcmpeqd	%xmm4, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm1
+	movaps	%xmm4, (%edx, %esi)
+	pcmpeqd	%xmm1, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm2
+	movaps	%xmm1, (%edx, %esi)
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	16(%ecx, %esi), %xmm3
+	movaps	%xmm2, (%edx, %esi)
+	pcmpeqd	%xmm3, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	%xmm3, (%edx, %esi)
+	mov	%ecx, %eax
+	lea	16(%ecx, %esi), %ecx
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	sub	%eax, %edx
+
+	mov	$-0x40, %esi
+
+L(Aligned64Loop):
+	movaps	(%ecx), %xmm2
+	movaps	32(%ecx), %xmm3
+	movaps	%xmm2, %xmm4
+	movaps	16(%ecx), %xmm5
+	movaps	%xmm3, %xmm6
+	movaps	48(%ecx), %xmm7
+	pminub	%xmm5, %xmm2
+	pminub	%xmm7, %xmm3
+	pminub	%xmm2, %xmm3
+	lea	64(%edx), %edx
+	pcmpeqd	%xmm0, %xmm3
+	lea	64(%ecx), %ecx
+	pmovmskb %xmm3, %eax
+
+	test	%eax, %eax
+	jnz	L(Aligned64Leave)
+	movaps	%xmm4, -64(%edx)
+	movaps	%xmm5, -48(%edx)
+	movaps	%xmm6, -32(%edx)
+	movaps	%xmm7, -16(%edx)
+	jmp	L(Aligned64Loop)
+
+L(Aligned64Leave):
+	pcmpeqd	%xmm4, %xmm0
+	pmovmskb %xmm0, %eax
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	pcmpeqd	%xmm5, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm4, -64(%edx)
+	lea	16(%esi), %esi
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	pcmpeqd	%xmm6, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm5, -48(%edx)
+	lea	16(%esi), %esi
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	movaps	%xmm6, -32(%edx)
+	pcmpeqd	%xmm7, %xmm0
+	pmovmskb %xmm0, %eax
+	lea	16(%esi), %esi
+	test	%eax, %eax
+	jnz	L(CopyFrom1To16Bytes)
+
+	mov	$-0x40, %esi
+	movaps	%xmm7, -16(%edx)
+	jmp	L(Aligned64Loop)
+
+	.p2align 4
+L(Shl4):
+	movaps	-4(%ecx), %xmm1
+	movaps	12(%ecx), %xmm2
+L(Shl4Start):
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	28(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm1
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm3, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	28(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	28(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+
+	test	%eax, %eax
+	jnz	L(Shl4LoopExit)
+
+	palignr	$4, %xmm3, %xmm2
+	movaps	%xmm2, (%edx)
+	lea	28(%ecx), %ecx
+	lea	16(%edx), %edx
+
+	mov	%ecx, %eax
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	lea	-12(%ecx), %ecx
+	sub	%eax, %edx
+
+	movaps	-4(%ecx), %xmm1
+
+L(Shl4LoopStart):
+	movaps	12(%ecx), %xmm2
+	movaps	28(%ecx), %xmm3
+	movaps	%xmm3, %xmm6
+	movaps	44(%ecx), %xmm4
+	movaps	%xmm4, %xmm7
+	movaps	60(%ecx), %xmm5
+	pminub	%xmm2, %xmm6
+	pminub	%xmm5, %xmm7
+	pminub	%xmm6, %xmm7
+	pcmpeqd	%xmm0, %xmm7
+	pmovmskb %xmm7, %eax
+	movaps	%xmm5, %xmm7
+	palignr	$4, %xmm4, %xmm5
+	palignr	$4, %xmm3, %xmm4
+	test	%eax, %eax
+	jnz	L(Shl4Start)
+
+	palignr	$4, %xmm2, %xmm3
+	lea	64(%ecx), %ecx
+	palignr	$4, %xmm1, %xmm2
+	movaps	%xmm7, %xmm1
+	movaps	%xmm5, 48(%edx)
+	movaps	%xmm4, 32(%edx)
+	movaps	%xmm3, 16(%edx)
+	movaps	%xmm2, (%edx)
+	lea	64(%edx), %edx
+	jmp	L(Shl4LoopStart)
+
+L(Shl4LoopExit):
+	movlpd	(%ecx), %xmm0
+	movl	8(%ecx), %esi
+	movlpd	%xmm0, (%edx)
+	movl	%esi, 8(%edx)
+	POP	(%esi)
+	add	$12, %edx
+	add	$12, %ecx
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit4)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(Shl8):
+	movaps	-8(%ecx), %xmm1
+	movaps	8(%ecx), %xmm2
+L(Shl8Start):
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	24(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm1
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm3, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	24(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	24(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+
+	test	%eax, %eax
+	jnz	L(Shl8LoopExit)
+
+	palignr	$8, %xmm3, %xmm2
+	movaps	%xmm2, (%edx)
+	lea	24(%ecx), %ecx
+	lea	16(%edx), %edx
+
+	mov	%ecx, %eax
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	lea	-8(%ecx), %ecx
+	sub	%eax, %edx
+
+	movaps	-8(%ecx), %xmm1
+
+L(Shl8LoopStart):
+	movaps	8(%ecx), %xmm2
+	movaps	24(%ecx), %xmm3
+	movaps	%xmm3, %xmm6
+	movaps	40(%ecx), %xmm4
+	movaps	%xmm4, %xmm7
+	movaps	56(%ecx), %xmm5
+	pminub	%xmm2, %xmm6
+	pminub	%xmm5, %xmm7
+	pminub	%xmm6, %xmm7
+	pcmpeqd	%xmm0, %xmm7
+	pmovmskb %xmm7, %eax
+	movaps	%xmm5, %xmm7
+	palignr	$8, %xmm4, %xmm5
+	palignr	$8, %xmm3, %xmm4
+	test	%eax, %eax
+	jnz	L(Shl8Start)
+
+	palignr	$8, %xmm2, %xmm3
+	lea	64(%ecx), %ecx
+	palignr	$8, %xmm1, %xmm2
+	movaps	%xmm7, %xmm1
+	movaps	%xmm5, 48(%edx)
+	movaps	%xmm4, 32(%edx)
+	movaps	%xmm3, 16(%edx)
+	movaps	%xmm2, (%edx)
+	lea	64(%edx), %edx
+	jmp	L(Shl8LoopStart)
+
+L(Shl8LoopExit):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	POP	(%esi)
+	add	$8, %edx
+	add	$8, %ecx
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit4)
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	CFI_PUSH	(%esi)
+
+	.p2align 4
+L(Shl12):
+	movaps	-12(%ecx), %xmm1
+	movaps	4(%ecx), %xmm2
+L(Shl12Start):
+	pcmpeqd	%xmm2, %xmm0
+	pmovmskb %xmm0, %eax
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	20(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm1
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm3, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	20(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+	movaps	%xmm2, %xmm3
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm2, (%edx)
+	movaps	20(%ecx), %xmm2
+
+	pcmpeqd	%xmm2, %xmm0
+	lea	16(%edx), %edx
+	pmovmskb %xmm0, %eax
+	lea	16(%ecx), %ecx
+
+	test	%eax, %eax
+	jnz	L(Shl12LoopExit)
+
+	palignr	$12, %xmm3, %xmm2
+	movaps	%xmm2, (%edx)
+	lea	20(%ecx), %ecx
+	lea	16(%edx), %edx
+
+	mov	%ecx, %eax
+	and	$-0x40, %ecx
+	sub	%ecx, %eax
+	lea	-4(%ecx), %ecx
+	sub	%eax, %edx
+
+	movaps	-12(%ecx), %xmm1
+
+L(Shl12LoopStart):
+	movaps	4(%ecx), %xmm2
+	movaps	20(%ecx), %xmm3
+	movaps	%xmm3, %xmm6
+	movaps	36(%ecx), %xmm4
+	movaps	%xmm4, %xmm7
+	movaps	52(%ecx), %xmm5
+	pminub	%xmm2, %xmm6
+	pminub	%xmm5, %xmm7
+	pminub	%xmm6, %xmm7
+	pcmpeqd	%xmm0, %xmm7
+	pmovmskb %xmm7, %eax
+	movaps	%xmm5, %xmm7
+	palignr	$12, %xmm4, %xmm5
+	palignr	$12, %xmm3, %xmm4
+	test	%eax, %eax
+	jnz	L(Shl12Start)
+
+	palignr	$12, %xmm2, %xmm3
+	lea	64(%ecx), %ecx
+	palignr	$12, %xmm1, %xmm2
+	movaps	%xmm7, %xmm1
+	movaps	%xmm5, 48(%edx)
+	movaps	%xmm4, 32(%edx)
+	movaps	%xmm3, 16(%edx)
+	movaps	%xmm2, (%edx)
+	lea	64(%edx), %edx
+	jmp	L(Shl12LoopStart)
+
+L(Shl12LoopExit):
+	movl	(%ecx), %esi
+	movl	%esi, (%edx)
+	mov	$4, %esi
+
+	.p2align 4
+L(CopyFrom1To16Bytes):
+	add	%esi, %edx
+	add	%esi, %ecx
+
+	POP	(%esi)
+	test	%al, %al
+	jz	L(ExitHigh)
+	test	$0x01, %al
+	jnz	L(Exit4)
+L(Exit8):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(ExitHigh):
+	test	$0x01, %ah
+	jnz	L(Exit12)
+L(Exit16):
+	movdqu	(%ecx), %xmm0
+	movdqu	%xmm0, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(Exit4):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movl	%edi, %eax
+	RETURN
+
+	.p2align 4
+L(Exit12):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	8(%ecx), %eax
+	movl	%eax, 8(%edx)
+	movl	%edi, %eax
+	RETURN
+
+CFI_POP	(%edi)
+
+	.p2align 4
+L(ExitTail4):
+	movl	(%ecx), %eax
+	movl	%eax, (%edx)
+	movl	%edx, %eax
+	ret
+
+	.p2align 4
+L(ExitTail8):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	%edx, %eax
+	ret
+
+	.p2align 4
+L(ExitTail12):
+	movlpd	(%ecx), %xmm0
+	movlpd	%xmm0, (%edx)
+	movl	8(%ecx), %eax
+	movl	%eax, 8(%edx)
+	movl	%edx, %eax
+	ret
+
+	.p2align 4
+L(ExitTail16):
+	movdqu	(%ecx), %xmm0
+	movdqu	%xmm0, (%edx)
+	movl	%edx, %eax
+	ret
+
+#ifndef USE_AS_WCSCAT
+END (wcscpy)
+#endif

libc/arch-x86/string/ssse3-wmemcmp-atom.S

@@ -0,0 +1,33 @@
+/*
+Copyright (c) 2011, Intel Corporation
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+    * this list of conditions and the following disclaimer.
+
+    * Redistributions in binary form must reproduce the above copyright notice,
+    * this list of conditions and the following disclaimer in the documentation
+    * and/or other materials provided with the distribution.
+
+    * Neither the name of Intel Corporation nor the names of its contributors
+    * may be used to endorse or promote products derived from this software
+    * without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#define MEMCMP  wmemcmp
+#define USE_AS_WMEMCMP 1
+#include "ssse3-memcmp-atom.S"

libc/arch-x86/string/strchr.S

@@ -1,3 +0,0 @@
-/*	$OpenBSD: strchr.S,v 1.3 2005/08/07 11:30:38 espie Exp $ */
-#define STRCHR
-#include "index.S"

libc/arch-x86/string/strcpy.S

@@ -1,64 +0,0 @@
-/*	$OpenBSD: strcpy.S,v 1.8 2005/08/07 11:30:38 espie Exp $ */
-/*
- * Written by J.T. Conklin <jtc@netbsd.org>.
- * Public domain.
- */
-
-#include <machine/asm.h>
-
-#if defined(APIWARN)
-#APP
-	.section .gnu.warning.strcpy
-	.ascii "warning: strcpy() is almost always misused, please use strlcpy()"
-#NO_APP
-#endif
-
-/*
- * NOTE: I've unrolled the loop eight times: large enough to make a
- * significant difference, and small enough not to totally trash the
- * cache.
- */
-
-ENTRY(strcpy)
-	movl	4(%esp),%ecx		/* dst address */
-	movl	8(%esp),%edx		/* src address */
-	pushl	%ecx			/* push dst address */
-
-	.align 2,0x90
-L1:	movb	(%edx),%al		/* unroll loop, but not too much */
-	movb	%al,(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	1(%edx),%al
-	movb	%al,1(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	2(%edx),%al
-	movb	%al,2(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	3(%edx),%al
-	movb	%al,3(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	4(%edx),%al
-	movb	%al,4(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	5(%edx),%al
-	movb	%al,5(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	6(%edx),%al
-	movb	%al,6(%ecx)
-	testb	%al,%al
-	jz	L2
-	movb	7(%edx),%al
-	movb	%al,7(%ecx)
-	addl	$8,%edx
-	addl	$8,%ecx
-	testb	%al,%al
-	jnz	L1
-L2:	popl	%eax			/* pop dst address */
-	ret
-END(strcpy)

libc/arch-x86/syscalls.mk

@@ -130,6 +130,8 @@ syscall_src += arch-x86/syscalls/removexattr.S
 syscall_src += arch-x86/syscalls/lremovexattr.S
 syscall_src += arch-x86/syscalls/__statfs64.S
 syscall_src += arch-x86/syscalls/unshare.S
+syscall_src += arch-x86/syscalls/swapon.S
+syscall_src += arch-x86/syscalls/swapoff.S
 syscall_src += arch-x86/syscalls/pause.S
 syscall_src += arch-x86/syscalls/gettimeofday.S
 syscall_src += arch-x86/syscalls/settimeofday.S
@@ -148,6 +150,9 @@ syscall_src += arch-x86/syscalls/__timer_getoverrun.S
 syscall_src += arch-x86/syscalls/__timer_delete.S
 syscall_src += arch-x86/syscalls/utimes.S
 syscall_src += arch-x86/syscalls/utimensat.S
+syscall_src += arch-x86/syscalls/timerfd_create.S
+syscall_src += arch-x86/syscalls/timerfd_settime.S
+syscall_src += arch-x86/syscalls/timerfd_gettime.S
 syscall_src += arch-x86/syscalls/sigaction.S
 syscall_src += arch-x86/syscalls/sigprocmask.S
 syscall_src += arch-x86/syscalls/__sigsuspend.S

libc/arch-x86/syscalls/swapoff.S

@@ -0,0 +1,21 @@
+/* autogenerated by gensyscalls.py */
+#include <linux/err.h>
+#include <machine/asm.h>
+#include <asm/unistd.h>
+
+ENTRY(swapoff)
+    pushl   %ebx
+    mov     8(%esp), %ebx
+    movl    $__NR_swapoff, %eax
+    int     $0x80
+    cmpl    $-MAX_ERRNO, %eax
+    jb      1f
+    negl    %eax
+    pushl   %eax
+    call    __set_errno
+    addl    $4, %esp
+    orl     $-1, %eax
+1:
+    popl    %ebx
+    ret
+END(swapoff)

libc/arch-x86/syscalls/swapon.S

@@ -0,0 +1,24 @@
+/* autogenerated by gensyscalls.py */
+#include <linux/err.h>
+#include <machine/asm.h>
+#include <asm/unistd.h>
+
+ENTRY(swapon)
+    pushl   %ebx
+    pushl   %ecx
+    mov     12(%esp), %ebx
+    mov     16(%esp), %ecx
+    movl    $__NR_swapon, %eax
+    int     $0x80
+    cmpl    $-MAX_ERRNO, %eax
+    jb      1f
+    negl    %eax
+    pushl   %eax
+    call    __set_errno
+    addl    $4, %esp
+    orl     $-1, %eax
+1:
+    popl    %ecx
+    popl    %ebx
+    ret
+END(swapon)

libc/arch-x86/syscalls/timerfd_create.S

@@ -0,0 +1,24 @@
+/* autogenerated by gensyscalls.py */
+#include <linux/err.h>
+#include <machine/asm.h>
+#include <asm/unistd.h>
+
+ENTRY(timerfd_create)
+    pushl   %ebx
+    pushl   %ecx
+    mov     12(%esp), %ebx
+    mov     16(%esp), %ecx
+    movl    $__NR_timerfd_create, %eax
+    int     $0x80
+    cmpl    $-MAX_ERRNO, %eax
+    jb      1f
+    negl    %eax
+    pushl   %eax
+    call    __set_errno
+    addl    $4, %esp
+    orl     $-1, %eax
+1:
+    popl    %ecx
+    popl    %ebx
+    ret
+END(timerfd_create)

libc/arch-x86/syscalls/timerfd_gettime.S

@@ -0,0 +1,24 @@
+/* autogenerated by gensyscalls.py */
+#include <linux/err.h>
+#include <machine/asm.h>
+#include <asm/unistd.h>
+
+ENTRY(timerfd_gettime)
+    pushl   %ebx
+    pushl   %ecx
+    mov     12(%esp), %ebx
+    mov     16(%esp), %ecx
+    movl    $__NR_timerfd_gettime, %eax
+    int     $0x80
+    cmpl    $-MAX_ERRNO, %eax
+    jb      1f
+    negl    %eax
+    pushl   %eax
+    call    __set_errno
+    addl    $4, %esp
+    orl     $-1, %eax
+1:
+    popl    %ecx
+    popl    %ebx
+    ret
+END(timerfd_gettime)

libc/arch-x86/syscalls/timerfd_settime.S

@@ -0,0 +1,30 @@
+/* autogenerated by gensyscalls.py */
+#include <linux/err.h>
+#include <machine/asm.h>
+#include <asm/unistd.h>
+
+ENTRY(timerfd_settime)
+    pushl   %ebx
+    pushl   %ecx
+    pushl   %edx
+    pushl   %esi
+    mov     20(%esp), %ebx
+    mov     24(%esp), %ecx
+    mov     28(%esp), %edx
+    mov     32(%esp), %esi
+    movl    $__NR_timerfd_settime, %eax
+    int     $0x80
+    cmpl    $-MAX_ERRNO, %eax
+    jb      1f
+    negl    %eax
+    pushl   %eax
+    call    __set_errno
+    addl    $4, %esp
+    orl     $-1, %eax
+1:
+    popl    %esi
+    popl    %edx
+    popl    %ecx
+    popl    %ebx
+    ret
+END(timerfd_settime)

libc/arch-x86/x86.mk

@@ -10,16 +10,75 @@ _LIBC_ARCH_COMMON_SRC_FILES := \
     arch-x86/bionic/sigsetjmp.S \
     arch-x86/bionic/syscall.S \
     arch-x86/bionic/vfork.S \
-    arch-x86/string/bcopy_wrapper.S \
+    arch-x86/string/ffs.S
-    arch-x86/string/bzero_wrapper.S \
+
-    arch-x86/string/ffs.S \
+ifeq ($(ARCH_X86_HAVE_SSSE3),true)
-    arch-x86/string/memcmp_wrapper.S \
+_LIBC_ARCH_COMMON_SRC_FILES += \
-    arch-x86/string/memcpy_wrapper.S \
+	arch-x86/string/ssse3-memcpy-atom.S \
-    arch-x86/string/memmove_wrapper.S \
+	arch-x86/string/ssse3-memmove-atom.S \
-    arch-x86/string/memset_wrapper.S \
+	arch-x86/string/ssse3-bcopy-atom.S \
-    arch-x86/string/strcmp_wrapper.S \
+	arch-x86/string/ssse3-strncat-atom.S \
-    arch-x86/string/strlen_wrapper.S \
+	arch-x86/string/ssse3-strncpy-atom.S \
-    arch-x86/string/strncmp_wrapper.S \
+	arch-x86/string/ssse3-strlcat-atom.S \
+	arch-x86/string/ssse3-strlcpy-atom.S \
+	arch-x86/string/ssse3-strcmp-atom.S \
+	arch-x86/string/ssse3-strncmp-atom.S \
+	arch-x86/string/ssse3-strcat-atom.S \
+	arch-x86/string/ssse3-strcpy-atom.S \
+	arch-x86/string/ssse3-memcmp-atom.S \
+	arch-x86/string/ssse3-wmemcmp-atom.S \
+	arch-x86/string/ssse3-wcscat-atom.S \
+	arch-x86/string/ssse3-wcscpy-atom.S
+else
+_LIBC_ARCH_COMMON_SRC_FILES += \
+	arch-x86/string/memcpy.S \
+	arch-x86/string/memmove.S \
+	arch-x86/string/bcopy.S \
+	arch-x86/string/strcmp.S \
+	arch-x86/string/strncmp.S \
+	arch-x86/string/strcat.S \
+	arch-x86/string/memcmp.S \
+	string/strcpy.c \
+	string/strncat.c \
+	string/strncpy.c \
+	string/strlcat.c \
+	string/strlcpy.c \
+	upstream-freebsd/lib/libc/string/wcscpy.c \
+	upstream-freebsd/lib/libc/string/wcscat.c \
+	upstream-freebsd/lib/libc/string/wmemcmp.c
+endif
+
+ifeq ($(ARCH_X86_HAVE_SSE2),true)
+_LIBC_ARCH_COMMON_SRC_FILES += \
+	arch-x86/string/sse2-memset-atom.S \
+	arch-x86/string/sse2-bzero-atom.S \
+	arch-x86/string/sse2-memchr-atom.S \
+	arch-x86/string/sse2-memrchr-atom.S \
+	arch-x86/string/sse2-strchr-atom.S \
+	arch-x86/string/sse2-strrchr-atom.S \
+	arch-x86/string/sse2-index-atom.S \
+	arch-x86/string/sse2-strlen-atom.S \
+	arch-x86/string/sse2-strnlen-atom.S \
+	arch-x86/string/sse2-wcschr-atom.S \
+	arch-x86/string/sse2-wcsrchr-atom.S \
+	arch-x86/string/sse2-wcslen-atom.S \
+	arch-x86/string/sse2-wcscmp-atom.S
+else
+_LIBC_ARCH_COMMON_SRC_FILES += \
+	arch-x86/string/memset.S \
+	arch-x86/string/strlen.S \
+	arch-x86/string/bzero.S \
+	bionic/memrchr.c \
+	bionic/memchr.c \
+	string/strchr.cpp \
+	string/strrchr.c \
+	string/index.c \
+	bionic/strnlen.c \
+	upstream-freebsd/lib/libc/string/wcschr.c \
+	upstream-freebsd/lib/libc/string/wcsrchr.c \
+	upstream-freebsd/lib/libc/string/wcslen.c \
+	upstream-freebsd/lib/libc/string/wcscmp.c
+endif
 
 _LIBC_ARCH_STATIC_SRC_FILES := \
     bionic/dl_iterate_phdr_static.c \

libc/bionic/__memcpy_chk.cpp

@@ -45,7 +45,7 @@
 extern "C" void *__memcpy_chk(void *dest, const void *src,
               size_t copy_amount, size_t dest_len)
 {
-    if (__builtin_expect(copy_amount > dest_len, 0)) {
+    if (__predict_false(copy_amount > dest_len)) {
         __fortify_chk_fail("memcpy buffer overflow",
                              BIONIC_EVENT_MEMCPY_BUFFER_OVERFLOW);
     }

libc/bionic/__memmove_chk.cpp

@@ -44,7 +44,7 @@
 extern "C" void *__memmove_chk (void *dest, const void *src,
               size_t len, size_t dest_len)
 {
-    if (len > dest_len) {
+    if (__predict_false(len > dest_len)) {
         __fortify_chk_fail("memmove buffer overflow",
                              BIONIC_EVENT_MEMMOVE_BUFFER_OVERFLOW);
     }

libc/bionic/__memset_chk.cpp

@@ -42,7 +42,7 @@
  * greater than 0.
  */
 extern "C" void *__memset_chk (void *dest, int c, size_t n, size_t dest_len) {
-    if (n > dest_len) {
+    if (__predict_false(n > dest_len)) {
         __fortify_chk_fail("memset buffer overflow",
                              BIONIC_EVENT_MEMSET_BUFFER_OVERFLOW);
     }

libc/bionic/__strcat_chk.cpp

@@ -29,7 +29,6 @@
 #include <string.h>
 #include <stdlib.h>
 #include "libc_logging.h"
-#include <safe_iop.h>
 
 /*
  * Runtime implementation of __builtin____strcat_chk.
@@ -42,22 +41,24 @@
  * This strcat check is called if _FORTIFY_SOURCE is defined and
  * greater than 0.
  */
-extern "C" char *__strcat_chk (char *dest, const char *src, size_t dest_buf_size) {
+extern "C" char* __strcat_chk(
-    // TODO: optimize so we don't scan src/dest twice.
+        char* __restrict dest,
-    size_t src_len  = strlen(src);
+        const char* __restrict src,
-    size_t dest_len = strlen(dest);
+        size_t dest_buf_size)
-    size_t sum;
+{
+    char* save = dest;
+    size_t dest_len = __strlen_chk(dest, dest_buf_size);
 
-    // sum = src_len + dest_len + 1 (with overflow protection)
+    dest += dest_len;
-    if (!safe_add3(&sum, src_len, dest_len, 1U)) {
+    dest_buf_size -= dest_len;
-        __fortify_chk_fail("strcat integer overflow",
-                             BIONIC_EVENT_STRCAT_INTEGER_OVERFLOW);
-    }
 
-    if (sum > dest_buf_size) {
+    while ((*dest++ = *src++) != '\0') {
+        dest_buf_size--;
+        if (__predict_false(dest_buf_size == 0)) {
             __fortify_chk_fail("strcat buffer overflow",
                                BIONIC_EVENT_STRCAT_BUFFER_OVERFLOW);
         }
+    }
 
-    return strcat(dest, src);
+    return save;
 }

libc/bionic/__strchr_chk.cpp

@@ -0,0 +1,46 @@
+/*-
+ * Copyright (c) 1990 The Regents of the University of California.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <string.h>
+#include "libc_logging.h"
+
+extern "C" char* __strchr_chk(const char* p, int ch, size_t s_len) {
+  for (;; ++p, s_len--) {
+    if (__predict_false(s_len == 0)) {
+      __fortify_chk_fail("read beyond buffer", 0);
+    }
+    if (*p == static_cast<char>(ch)) {
+      return const_cast<char*>(p);
+    }
+    if (*p == '\0') {
+      return NULL;
+    }
+  }
+  /* NOTREACHED */
+}

libc/bionic/__strcpy_chk.cpp

@@ -44,7 +44,7 @@
 extern "C" char *__strcpy_chk (char *dest, const char *src, size_t dest_len) {
     // TODO: optimize so we don't scan src twice.
     size_t src_len = strlen(src) + 1;
-    if (src_len > dest_len) {
+    if (__predict_false(src_len > dest_len)) {
         __fortify_chk_fail("strcpy buffer overflow",
                              BIONIC_EVENT_STRCPY_BUFFER_OVERFLOW);
     }

libc/bionic/__strlcat_chk.cpp

@@ -45,7 +45,7 @@
 extern "C" size_t __strlcat_chk(char *dest, const char *src,
               size_t supplied_size, size_t dest_len_from_compiler)
 {
-    if (supplied_size > dest_len_from_compiler) {
+    if (__predict_false(supplied_size > dest_len_from_compiler)) {
         __fortify_chk_fail("strlcat buffer overflow", 0);
     }
 

libc/bionic/__strlcpy_chk.cpp

@@ -45,7 +45,7 @@
 extern "C" size_t __strlcpy_chk(char *dest, const char *src,
               size_t supplied_size, size_t dest_len_from_compiler)
 {
-    if (supplied_size > dest_len_from_compiler) {
+    if (__predict_false(supplied_size > dest_len_from_compiler)) {
         __fortify_chk_fail("strlcpy buffer overflow", 0);
     }
 

libc/bionic/__strlen_chk.cpp

@@ -56,7 +56,7 @@
 extern "C" size_t __strlen_chk(const char *s, size_t s_len) {
     size_t ret = strlen(s);
 
-    if (__builtin_expect(ret >= s_len, 0)) {
+    if (__predict_false(ret >= s_len)) {
         __fortify_chk_fail("strlen read overflow", 0);
     }
 

libc/bionic/__strncat_chk.cpp

@@ -29,7 +29,6 @@
 #include <string.h>
 #include <stdlib.h>
 #include "libc_logging.h"
-#include <safe_iop.h>
 
 /*
  * Runtime implementation of __builtin____strncat_chk.
@@ -42,27 +41,33 @@
  * This strncat check is called if _FORTIFY_SOURCE is defined and
  * greater than 0.
  */
-extern "C" char *__strncat_chk (char *dest, const char *src,
+extern "C" char *__strncat_chk(
+        char* __restrict dest,
+        const char* __restrict src,
         size_t len, size_t dest_buf_size)
 {
-    // TODO: optimize so we don't scan src/dest twice.
+    if (len == 0) {
-    size_t dest_len = strlen(dest);
+        return dest;
-    size_t src_len = strlen(src);
-    if (src_len > len) {
-        src_len = len;
     }
 
-    size_t sum;
+    size_t dest_len = __strlen_chk(dest, dest_buf_size);
-    // sum = src_len + dest_len + 1 (with overflow protection)
+    char *d = dest + dest_len;
-    if (!safe_add3(&sum, src_len, dest_len, 1U)) {
+    dest_buf_size -= dest_len;
-        __fortify_chk_fail("strncat integer overflow",
-                             BIONIC_EVENT_STRNCAT_INTEGER_OVERFLOW);
-    }
 
-    if (sum > dest_buf_size) {
+    while (*src != '\0') {
+        *d++ = *src++;
+        len--; dest_buf_size--;
+
+        if (__predict_false(dest_buf_size == 0)) {
             __fortify_chk_fail("strncat buffer overflow",
                                BIONIC_EVENT_STRNCAT_BUFFER_OVERFLOW);
         }
 
-    return strncat(dest, src, len);
+        if (len == 0) {
+            break;
+        }
+    }
+
+    *d = '\0';
+    return dest;
 }

libc/bionic/__strncpy_chk.cpp

@@ -44,7 +44,7 @@
 extern "C" char *__strncpy_chk (char *dest, const char *src,
               size_t len, size_t dest_len)
 {
-    if (len > dest_len) {
+    if (__predict_false(len > dest_len)) {
         __fortify_chk_fail("strncpy buffer overflow",
                              BIONIC_EVENT_STRNCPY_BUFFER_OVERFLOW);
     }

libc/bionic/__umask_chk.cpp

@@ -42,7 +42,7 @@
  * greater than 0.
  */
 extern "C" mode_t __umask_chk(mode_t mode) {
-    if ((mode & 0777) != mode) {
+    if (__predict_false((mode & 0777) != mode)) {
         __fortify_chk_fail("umask called with invalid mask", 0);
     }
 

libc/bionic/__vsnprintf_chk.cpp

@@ -50,7 +50,7 @@ extern "C" int __vsnprintf_chk(
         const char *format,
         va_list va)
 {
-    if (supplied_size > dest_len_from_compiler) {
+    if (__predict_false(supplied_size > dest_len_from_compiler)) {
         __fortify_chk_fail("vsnprintf buffer overflow", 0);
     }
 

libc/bionic/abort.cpp

@@ -1,4 +1,3 @@
-/*	$OpenBSD: abort.c,v 1.14 2005/08/08 08:05:36 espie Exp $ */
 /*
  * Copyright (c) 1985 Regents of the University of California.
  * All rights reserved.
@@ -31,74 +30,38 @@
 #include <signal.h>
 #include <stdlib.h>
 #include <unistd.h>
-#include "thread_private.h"
 #include "atexit.h"
 
+__LIBC_HIDDEN__ void (*__cleanup)();
+
 #ifdef __arm__
-__LIBC_HIDDEN__ void
+extern "C" __LIBC_HIDDEN__ void __libc_android_abort()
-__libc_android_abort(void)
 #else
-void
+void abort()
-abort(void)
 #endif
 {
-	struct atexit *p = __atexit;
+  // Don't block SIGABRT to give any signal handler a chance; we ignore
-	static int cleanup_called = 0;
+  // any errors -- X311J doesn't allow abort to return anyway.
   sigset_t mask;
-
-
   sigfillset(&mask);
-	/*
-	 * don't block SIGABRT to give any handler a chance; we ignore
-	 * any errors -- X311J doesn't allow abort to return anyway.
-	 */
   sigdelset(&mask, SIGABRT);
-    /* temporary, so deliberate seg fault can be caught by debuggerd */
+  sigprocmask(SIG_SETMASK, &mask, NULL);
-	sigdelset(&mask, SIGSEGV);
-    /* -- */
-	(void)sigprocmask(SIG_SETMASK, &mask, (sigset_t *)NULL);
 
-	/*
+  // POSIX requires we flush stdio buffers on abort.
-	 * POSIX requires we flush stdio buffers on abort
+  if (__cleanup) {
-	 */
+    (*__cleanup)();
-	if (cleanup_called == 0) {
-		while (p != NULL && p->next != NULL)
-			p = p->next;
-		/* the check for fn_dso == NULL is mostly paranoia */
-		if (p != NULL && p->fns[0].fn_dso == NULL &&
-		    p->fns[0].fn_ptr.std_func != NULL) {
-			cleanup_called = 1;
-			(*p->fns[0].fn_ptr.std_func)();
-		}
   }
 
-    /* temporary, for bug hunting */
+  raise(SIGABRT);
-    /* seg fault seems to produce better debuggerd results than SIGABRT */
-#ifdef __mips__
-    /* An access that will generate SIGSEGV rather than SIGBUS. */
-    *((char*)0xdeadc0c0) = 39;
-#else
-    *((char*)0xdeadbaad) = 39;
-#endif
-    /* -- */
 
-	(void)kill(getpid(), SIGABRT);
+  // If SIGABRT ignored, or caught and the handler returns,
-
+  // remove the SIGABRT signal handler and raise SIGABRT again.
-	/*
-	 * if SIGABRT ignored, or caught and the handler returns, do
-	 * it again, only harder.
-	 */
-        {
   struct sigaction sa;
-
   sa.sa_handler = SIG_DFL;
   sa.sa_flags   = SA_RESTART;
   sigemptyset(&sa.sa_mask);
-
+  sigaction(SIGABRT, &sa, &sa);
-            (void)sigaction( SIGABRT, &sa, &sa );
+  sigprocmask(SIG_SETMASK, &mask, NULL);
-        }
+  raise(SIGABRT);
-
-	(void)sigprocmask(SIG_SETMASK, &mask, (sigset_t *)NULL);
-	(void)kill(getpid(), SIGABRT);
   _exit(1);
 }

libc/bionic/dl_iterate_phdr_static.c

@@ -27,39 +27,56 @@
  */
 
 #include <elf.h>
+#include <sys/auxv.h>
 #include <sys/types.h>
 #include <link.h>
 
-/* Dynamic binaries get this from the dynamic linker (system/linker), which
- * we don't pull in for static bins. We also don't have a list of so's to
- * iterate over, since there's really only a single monolithic blob of
- * code/data.
- *
- * All we need to do is to find where the executable is in memory, and grab the
- * phdr and phnum from there.
- */
-
 /* ld provides this to us in the default link script */
-extern void *__executable_start;
+extern void* __executable_start;
 
-int
+int dl_iterate_phdr(int (*cb)(struct dl_phdr_info* info, size_t size, void* data), void* data) {
-dl_iterate_phdr(int (*cb)(struct dl_phdr_info *info, size_t size, void *data),
+    Elf32_Ehdr* ehdr = (Elf32_Ehdr*) &__executable_start;
-                void *data)
-{
-    struct dl_phdr_info dl_info;
-    Elf32_Ehdr *ehdr = (Elf32_Ehdr *) &__executable_start;
-    Elf32_Phdr *phdr = (Elf32_Phdr *)((unsigned long)ehdr + ehdr->e_phoff);
 
-    /* TODO: again, copied from linker.c. Find a better home for this
+    // TODO: again, copied from linker.c. Find a better home for this later.
-     * later. */
     if (ehdr->e_ident[EI_MAG0] != ELFMAG0) return -1;
     if (ehdr->e_ident[EI_MAG1] != ELFMAG1) return -1;
     if (ehdr->e_ident[EI_MAG2] != ELFMAG2) return -1;
     if (ehdr->e_ident[EI_MAG3] != ELFMAG3) return -1;
 
-    dl_info.dlpi_addr = 0;
+    // Dynamic binaries get their dl_iterate_phdr from the dynamic linker, but
-    dl_info.dlpi_name = NULL;
+    // static binaries get this. We don't have a list of shared objects to
-    dl_info.dlpi_phdr = phdr;
+    // iterate over, since there's really only a single monolithic blob of
-    dl_info.dlpi_phnum = ehdr->e_phnum;
+    // code/data, plus optionally a VDSO.
-    return cb(&dl_info, sizeof (struct dl_phdr_info), data);
+
+    struct dl_phdr_info exe_info;
+    exe_info.dlpi_addr = 0;
+    exe_info.dlpi_name = NULL;
+    exe_info.dlpi_phdr = (Elf32_Phdr*) ((unsigned long) ehdr + ehdr->e_phoff);
+    exe_info.dlpi_phnum = ehdr->e_phnum;
+
+#ifdef AT_SYSINFO_EHDR
+    // Try the executable first.
+    int rc = cb(&exe_info, sizeof(exe_info), data);
+    if (rc != 0) {
+        return rc;
+    }
+
+    // Try the VDSO if that didn't work.
+    Elf32_Ehdr* ehdr_vdso = (Elf32_Ehdr*) getauxval(AT_SYSINFO_EHDR);
+    struct dl_phdr_info vdso_info;
+    vdso_info.dlpi_addr = 0;
+    vdso_info.dlpi_name = NULL;
+    vdso_info.dlpi_phdr = (Elf32_Phdr*) ((char*) ehdr_vdso + ehdr_vdso->e_phoff);
+    vdso_info.dlpi_phnum = ehdr_vdso->e_phnum;
+    for (size_t i = 0; i < vdso_info.dlpi_phnum; ++i) {
+        if (vdso_info.dlpi_phdr[i].p_type == PT_LOAD) {
+            vdso_info.dlpi_addr = (Elf32_Addr) ehdr_vdso - vdso_info.dlpi_phdr[i].p_vaddr;
+            break;
+        }
+    }
+    return cb(&vdso_info, sizeof(vdso_info), data);
+#else
+    // There's only the executable to try.
+    return cb(&exe_info, sizeof(exe_info), data);
+#endif
 }

libc/bionic/dlmalloc.c

@@ -28,13 +28,17 @@ static void __bionic_heap_usage_error(const char* function, void* address);
 // Ugly inclusion of C file so that bionic specific #defines configure dlmalloc.
 #include "../upstream-dlmalloc/malloc.c"
 
+extern void (*__cleanup)();
+
 static void __bionic_heap_corruption_error(const char* function) {
-  __libc_fatal("@@@ ABORTING: heap corruption detected by %s", function);
+  __cleanup = NULL; // The heap is corrupt. We can forget trying to shut down stdio.
+  __libc_fatal("heap corruption detected by %s", function);
 }
 
 static void __bionic_heap_usage_error(const char* function, void* address) {
-  __libc_fatal("@@@ ABORTING: invalid address or address of corrupt block %p passed to %s",
+  __libc_fatal_no_abort("invalid address or address of corrupt block %p passed to %s",
                address, function);
-  // So that we can get a memory dump around the specific address.
+  // So that debuggerd gives us a memory dump around the specific address.
+  // TODO: improve the debuggerd protocol so we can tell it to dump an address when we abort.
   *((int**) 0xdeadbaad) = (int*) address;
 }

libc/bionic/dlmalloc.h

@@ -27,6 +27,7 @@
 #define LOCK_AT_FORK 1
 #define USE_RECURSIVE_LOCK 0
 #define USE_SPIN_LOCKS 0
+#define DEFAULT_MMAP_THRESHOLD (64U * 1024U)
 
 /* Include the proper definitions. */
 #include "../upstream-dlmalloc/malloc.h"

libc/bionic/libc_logging.cpp

@@ -42,7 +42,6 @@
 #include <unistd.h>
 
 static pthread_mutex_t gAbortMsgLock = PTHREAD_MUTEX_INITIALIZER;
-static pthread_mutex_t gLogInitializationLock = PTHREAD_MUTEX_INITIALIZER;
 
 __LIBC_HIDDEN__ abort_msg_t** __abort_message_ptr; // Accessible to __libc_init_common.
 
@@ -348,7 +347,7 @@ static void out_vformat(Out& o, const char* format, va_list args) {
             buffer[0] = '0';
             buffer[1] = 'x';
             format_integer(buffer + 2, sizeof(buffer) - 2, value, 'x');
-        } else if (c == 'd' || c == 'i' || c == 'o' || c == 'x' || c == 'X') {
+        } else if (c == 'd' || c == 'i' || c == 'o' || c == 'u' || c == 'x' || c == 'X') {
             /* integers - first read value from stack */
             uint64_t value;
             int is_signed = (c == 'd' || c == 'i' || c == 'o');
@@ -421,14 +420,10 @@ int __libc_format_fd(int fd, const char* format, ...) {
 }
 
 static int __libc_write_log(int priority, const char* tag, const char* msg) {
-  static int main_log_fd = -1;
+  int main_log_fd = TEMP_FAILURE_RETRY(open("/dev/log/main", O_CLOEXEC | O_WRONLY));
-  if (main_log_fd == -1) {
-    ScopedPthreadMutexLocker locker(&gLogInitializationLock);
-    main_log_fd = TEMP_FAILURE_RETRY(open("/dev/log/main", O_CLOEXEC | O_WRONLY));
   if (main_log_fd == -1) {
     return -1;
   }
-  }
 
   iovec vec[3];
   vec[0].iov_base = &priority;
@@ -438,7 +433,9 @@ static int __libc_write_log(int priority, const char* tag, const char* msg) {
   vec[2].iov_base = const_cast<char*>(msg);
   vec[2].iov_len = strlen(msg) + 1;
 
-  return TEMP_FAILURE_RETRY(writev(main_log_fd, vec, 3));
+  int result = TEMP_FAILURE_RETRY(writev(main_log_fd, vec, 3));
+  close(main_log_fd);
+  return result;
 }
 
 int __libc_format_log_va_list(int priority, const char* tag, const char* format, va_list args) {
@@ -465,12 +462,13 @@ static int __libc_android_log_event(int32_t tag, char type, const void* payload,
   vec[2].iov_base = const_cast<void*>(payload);
   vec[2].iov_len = len;
 
-  static int event_log_fd = -1;
+  int event_log_fd = TEMP_FAILURE_RETRY(open("/dev/log/events", O_CLOEXEC | O_WRONLY));
   if (event_log_fd == -1) {
-    ScopedPthreadMutexLocker locker(&gLogInitializationLock);
+    return -1;
-    event_log_fd = TEMP_FAILURE_RETRY(open("/dev/log/events", O_CLOEXEC | O_WRONLY));
   }
-  return TEMP_FAILURE_RETRY(writev(event_log_fd, vec, 3));
+  int result = TEMP_FAILURE_RETRY(writev(event_log_fd, vec, 3));
+  close(event_log_fd);
+  return result;
 }
 
 void __libc_android_log_event_int(int32_t tag, int value) {
@@ -488,13 +486,10 @@ void __fortify_chk_fail(const char *msg, uint32_t tag) {
   __libc_fatal("FORTIFY_SOURCE: %s. Calling abort().", msg);
 }
 
-void __libc_fatal(const char* format, ...) {
+static void __libc_fatal(const char* format, va_list args) {
   char msg[1024];
   BufferOutputStream os(msg, sizeof(msg));
-  va_list args;
-  va_start(args, format);
   out_vformat(os, format, args);
-  va_end(args);
 
   // TODO: log to stderr for the benefit of "adb shell" users.
 
@@ -502,7 +497,20 @@ void __libc_fatal(const char* format, ...) {
   __libc_write_log(ANDROID_LOG_FATAL, "libc", msg);
 
   __libc_set_abort_message(msg);
+}
 
+void __libc_fatal_no_abort(const char* format, ...) {
+  va_list args;
+  va_start(args, format);
+  __libc_fatal(format, args);
+  va_end(args);
+}
+
+void __libc_fatal(const char* format, ...) {
+  va_list args;
+  va_start(args, format);
+  __libc_fatal(format, args);
+  va_end(args);
   abort();
 }
 

libc/bionic/malloc_debug_check.cpp

@@ -74,6 +74,10 @@ static void log_message(const char* format, ...) {
 
 struct hdr_t {
     uint32_t tag;
+    void* base;  // Always points to the memory allocated using dlmalloc.
+                 // For memory allocated in chk_memalign, this value will
+                 // not be the same as the location of the start of this
+                 // structure.
     hdr_t* prev;
     hdr_t* next;
     uintptr_t bt[MAX_BACKTRACE_DEPTH];
@@ -82,7 +86,7 @@ struct hdr_t {
     int freed_bt_depth;
     size_t size;
     char front_guard[FRONT_GUARD_LEN];
-} __attribute__((packed));
+} __attribute__((packed, aligned(MALLOC_ALIGNMENT)));
 
 struct ftr_t {
     char rear_guard[REAR_GUARD_LEN];
@@ -100,21 +104,26 @@ static inline hdr_t* meta(void* user) {
     return reinterpret_cast<hdr_t*>(user) - 1;
 }
 
+static inline const hdr_t* const_meta(const void* user) {
+    return reinterpret_cast<const hdr_t*>(user) - 1;
+}
+
+
 static unsigned gAllocatedBlockCount;
-static hdr_t *tail;
+static hdr_t* tail;
-static hdr_t *head;
+static hdr_t* head;
 static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;
 
 static unsigned backlog_num;
-static hdr_t *backlog_tail;
+static hdr_t* backlog_tail;
-static hdr_t *backlog_head;
+static hdr_t* backlog_head;
 static pthread_mutex_t backlog_lock = PTHREAD_MUTEX_INITIALIZER;
 
-static inline void init_front_guard(hdr_t *hdr) {
+static inline void init_front_guard(hdr_t* hdr) {
     memset(hdr->front_guard, FRONT_GUARD, FRONT_GUARD_LEN);
 }
 
-static inline bool is_front_guard_valid(hdr_t *hdr) {
+static inline bool is_front_guard_valid(hdr_t* hdr) {
     for (size_t i = 0; i < FRONT_GUARD_LEN; i++) {
         if (hdr->front_guard[i] != FRONT_GUARD) {
             return 0;
@@ -123,12 +132,12 @@ static inline bool is_front_guard_valid(hdr_t *hdr) {
     return 1;
 }
 
-static inline void init_rear_guard(hdr_t *hdr) {
+static inline void init_rear_guard(hdr_t* hdr) {
     ftr_t* ftr = to_ftr(hdr);
     memset(ftr->rear_guard, REAR_GUARD, REAR_GUARD_LEN);
 }
 
-static inline bool is_rear_guard_valid(hdr_t *hdr) {
+static inline bool is_rear_guard_valid(hdr_t* hdr) {
     unsigned i;
     int valid = 1;
     int first_mismatch = -1;
@@ -149,7 +158,7 @@ static inline bool is_rear_guard_valid(hdr_t *hdr) {
     return valid;
 }
 
-static inline void add_locked(hdr_t *hdr, hdr_t **tail, hdr_t **head) {
+static inline void add_locked(hdr_t* hdr, hdr_t** tail, hdr_t** head) {
     hdr->prev = NULL;
     hdr->next = *head;
     if (*head)
@@ -159,7 +168,7 @@ static inline void add_locked(hdr_t *hdr, hdr_t **tail, hdr_t **head) {
     *head = hdr;
 }
 
-static inline int del_locked(hdr_t *hdr, hdr_t **tail, hdr_t **head) {
+static inline int del_locked(hdr_t* hdr, hdr_t** tail, hdr_t** head) {
     if (hdr->prev) {
         hdr->prev->next = hdr->next;
     } else {
@@ -173,7 +182,7 @@ static inline int del_locked(hdr_t *hdr, hdr_t **tail, hdr_t **head) {
     return 0;
 }
 
-static inline void add(hdr_t *hdr, size_t size) {
+static inline void add(hdr_t* hdr, size_t size) {
     ScopedPthreadMutexLocker locker(&lock);
     hdr->tag = ALLOCATION_TAG;
     hdr->size = size;
@@ -183,7 +192,7 @@ static inline void add(hdr_t *hdr, size_t size) {
     add_locked(hdr, &tail, &head);
 }
 
-static inline int del(hdr_t *hdr) {
+static inline int del(hdr_t* hdr) {
     if (hdr->tag != ALLOCATION_TAG) {
         return -1;
     }
@@ -194,13 +203,13 @@ static inline int del(hdr_t *hdr) {
     return 0;
 }
 
-static inline void poison(hdr_t *hdr) {
+static inline void poison(hdr_t* hdr) {
     memset(user(hdr), FREE_POISON, hdr->size);
 }
 
-static int was_used_after_free(hdr_t *hdr) {
+static int was_used_after_free(hdr_t* hdr) {
     unsigned i;
-    const char *data = (const char *)user(hdr);
+    const char* data = reinterpret_cast<const char *>(user(hdr));
     for (i = 0; i < hdr->size; i++)
         if (data[i] != FREE_POISON)
             return 1;
@@ -208,7 +217,7 @@ static int was_used_after_free(hdr_t *hdr) {
 }
 
 /* returns 1 if valid, *safe == 1 if safe to dump stack */
-static inline int check_guards(hdr_t *hdr, int *safe) {
+static inline int check_guards(hdr_t* hdr, int* safe) {
     *safe = 1;
     if (!is_front_guard_valid(hdr)) {
         if (hdr->front_guard[0] == FRONT_GUARD) {
@@ -233,7 +242,7 @@ static inline int check_guards(hdr_t *hdr, int *safe) {
 }
 
 /* returns 1 if valid, *safe == 1 if safe to dump stack */
-static inline int check_allocation_locked(hdr_t *hdr, int *safe) {
+static inline int check_allocation_locked(hdr_t* hdr, int* safe) {
     int valid = 1;
     *safe = 1;
 
@@ -270,9 +279,9 @@ static inline int check_allocation_locked(hdr_t *hdr, int *safe) {
     return valid;
 }
 
-static inline int del_and_check_locked(hdr_t *hdr,
+static inline int del_and_check_locked(hdr_t* hdr,
-                                       hdr_t **tail, hdr_t **head, unsigned *cnt,
+                                       hdr_t** tail, hdr_t** head, unsigned* cnt,
-                                       int *safe) {
+                                       int* safe) {
     int valid = check_allocation_locked(hdr, safe);
     if (safe) {
         (*cnt)--;
@@ -281,7 +290,7 @@ static inline int del_and_check_locked(hdr_t *hdr,
     return valid;
 }
 
-static inline void del_from_backlog_locked(hdr_t *hdr) {
+static inline void del_from_backlog_locked(hdr_t* hdr) {
     int safe;
     del_and_check_locked(hdr,
                          &backlog_tail, &backlog_head, &backlog_num,
@@ -289,17 +298,17 @@ static inline void del_from_backlog_locked(hdr_t *hdr) {
     hdr->tag = 0; /* clear the tag */
 }
 
-static inline void del_from_backlog(hdr_t *hdr) {
+static inline void del_from_backlog(hdr_t* hdr) {
     ScopedPthreadMutexLocker locker(&backlog_lock);
     del_from_backlog_locked(hdr);
 }
 
-static inline int del_leak(hdr_t *hdr, int *safe) {
+static inline int del_leak(hdr_t* hdr, int* safe) {
     ScopedPthreadMutexLocker locker(&lock);
     return del_and_check_locked(hdr, &tail, &head, &gAllocatedBlockCount, safe);
 }
 
-static inline void add_to_backlog(hdr_t *hdr) {
+static inline void add_to_backlog(hdr_t* hdr) {
     ScopedPthreadMutexLocker locker(&backlog_lock);
     hdr->tag = BACKLOG_TAG;
     backlog_num++;
@@ -307,9 +316,9 @@ static inline void add_to_backlog(hdr_t *hdr) {
     poison(hdr);
     /* If we've exceeded the maximum backlog, clear it up */
     while (backlog_num > gMallocDebugBacklog) {
-        hdr_t *gone = backlog_tail;
+        hdr_t* gone = backlog_tail;
         del_from_backlog_locked(gone);
-        dlfree(gone);
+        dlfree(gone->base);
     }
 }
 
@@ -318,6 +327,7 @@ extern "C" void* chk_malloc(size_t size) {
 
     hdr_t* hdr = static_cast<hdr_t*>(dlmalloc(sizeof(hdr_t) + size + sizeof(ftr_t)));
     if (hdr) {
+        hdr->base = hdr;
         hdr->bt_depth = get_backtrace(hdr->bt, MAX_BACKTRACE_DEPTH);
         add(hdr, size);
         return user(hdr);
@@ -325,13 +335,44 @@ extern "C" void* chk_malloc(size_t size) {
     return NULL;
 }
 
-extern "C" void* chk_memalign(size_t, size_t bytes) {
+extern "C" void* chk_memalign(size_t alignment, size_t bytes) {
-//  log_message("%s: %s\n", __FILE__, __FUNCTION__);
+    if (alignment <= MALLOC_ALIGNMENT) {
-    // XXX: it's better to use malloc, than being wrong
         return chk_malloc(bytes);
+    }
+
+    // Make the alignment a power of two.
+    if (alignment & (alignment-1)) {
+        alignment = 1L << (31 - __builtin_clz(alignment));
+    }
+
+    // here, alignment is at least MALLOC_ALIGNMENT<<1 bytes
+    // we will align by at least MALLOC_ALIGNMENT bytes
+    // and at most alignment-MALLOC_ALIGNMENT bytes
+    size_t size = (alignment-MALLOC_ALIGNMENT) + bytes;
+    if (size < bytes) { // Overflow.
+        return NULL;
+    }
+
+    void* base = dlmalloc(sizeof(hdr_t) + size + sizeof(ftr_t));
+    if (base != NULL) {
+        // Check that the actual pointer that will be returned is aligned
+        // properly.
+        uintptr_t ptr = reinterpret_cast<uintptr_t>(user(reinterpret_cast<hdr_t*>(base)));
+        if ((ptr % alignment) != 0) {
+            // Align the pointer.
+            ptr += ((-ptr) % alignment);
+        }
+
+        hdr_t* hdr = meta(reinterpret_cast<void*>(ptr));
+        hdr->base = base;
+        hdr->bt_depth = get_backtrace(hdr->bt, MAX_BACKTRACE_DEPTH);
+        add(hdr, bytes);
+        return user(hdr);
+    }
+    return base;
 }
 
-extern "C" void chk_free(void *ptr) {
+extern "C" void chk_free(void* ptr) {
 //  log_message("%s: %s\n", __FILE__, __FUNCTION__);
 
     if (!ptr) /* ignore free(NULL) */
@@ -366,7 +407,7 @@ extern "C" void chk_free(void *ptr) {
     }
 }
 
-extern "C" void *chk_realloc(void *ptr, size_t size) {
+extern "C" void* chk_realloc(void* ptr, size_t size) {
 //  log_message("%s: %s\n", __FILE__, __FUNCTION__);
 
     if (!ptr) {
@@ -414,8 +455,23 @@ extern "C" void *chk_realloc(void *ptr, size_t size) {
         }
     }
 
+    if (hdr->base != hdr) {
+        // An allocation from memalign, so create another allocation and
+        // copy the data out.
+        void* newMem = dlmalloc(sizeof(hdr_t) + size + sizeof(ftr_t));
+        if (newMem) {
+            memcpy(newMem, hdr, sizeof(hdr_t) + hdr->size);
+            dlfree(hdr->base);
+            hdr = static_cast<hdr_t*>(newMem);
+        } else {
+            dlfree(hdr->base);
+            hdr = NULL;
+        }
+    } else {
         hdr = static_cast<hdr_t*>(dlrealloc(hdr, sizeof(hdr_t) + size + sizeof(ftr_t)));
+    }
     if (hdr) {
+        hdr->base = hdr;
         hdr->bt_depth = get_backtrace(hdr->bt, MAX_BACKTRACE_DEPTH);
         add(hdr, size);
         return user(hdr);
@@ -424,11 +480,12 @@ extern "C" void *chk_realloc(void *ptr, size_t size) {
     return NULL;
 }
 
-extern "C" void *chk_calloc(int nmemb, size_t size) {
+extern "C" void* chk_calloc(int nmemb, size_t size) {
 //  log_message("%s: %s\n", __FILE__, __FUNCTION__);
     size_t total_size = nmemb * size;
     hdr_t* hdr = static_cast<hdr_t*>(dlcalloc(1, sizeof(hdr_t) + total_size + sizeof(ftr_t)));
     if (hdr) {
+        hdr->base = hdr;
         hdr->bt_depth = get_backtrace(hdr->bt, MAX_BACKTRACE_DEPTH);
         add(hdr, total_size);
         return user(hdr);
@@ -436,6 +493,18 @@ extern "C" void *chk_calloc(int nmemb, size_t size) {
     return NULL;
 }
 
+extern "C" size_t chk_malloc_usable_size(const void* ptr) {
+    // dlmalloc_usable_size returns 0 for NULL and unknown blocks.
+    if (ptr == NULL)
+        return 0;
+
+    const hdr_t* hdr = const_meta(ptr);
+
+    // The sentinel tail is written just after the request block bytes
+    // so there is no extra room we can report here.
+    return hdr->size;
+}
+
 static void ReportMemoryLeaks() {
   // We only track leaks at level 10.
   if (gMallocDebugLevel != 10) {

libc/bionic/malloc_debug_common.cpp

@@ -190,10 +190,6 @@ extern "C" struct mallinfo mallinfo() {
     return dlmallinfo();
 }
 
-extern "C" size_t malloc_usable_size(void* mem) {
-    return dlmalloc_usable_size(mem);
-}
-
 extern "C" void* valloc(size_t bytes) {
     return dlvalloc(bytes);
 }
@@ -215,8 +211,9 @@ extern "C" int posix_memalign(void** memptr, size_t alignment, size_t size) {
 
 /* Table for dispatching malloc calls, initialized with default dispatchers. */
 extern const MallocDebug __libc_malloc_default_dispatch;
-const MallocDebug __libc_malloc_default_dispatch __attribute__((aligned(32))) = {
+const MallocDebug __libc_malloc_default_dispatch __attribute__((aligned(32))) =
-    dlmalloc, dlfree, dlcalloc, dlrealloc, dlmemalign
+{
+    dlmalloc, dlfree, dlcalloc, dlrealloc, dlmemalign, dlmalloc_usable_size,
 };
 
 /* Selector of dispatch table to use for dispatching malloc calls. */
@@ -242,6 +239,10 @@ extern "C" void* memalign(size_t alignment, size_t bytes) {
     return __libc_malloc_dispatch->memalign(alignment, bytes);
 }
 
+extern "C" size_t malloc_usable_size(const void* mem) {
+    return __libc_malloc_dispatch->malloc_usable_size(mem);
+}
+
 /* We implement malloc debugging only in libc.so, so code below
  * must be excluded if we compile this file for static libc.a
  */
@@ -253,7 +254,7 @@ extern "C" void* memalign(size_t alignment, size_t bytes) {
 
 /* Table for dispatching malloc calls, depending on environment. */
 static MallocDebug gMallocUse __attribute__((aligned(32))) = {
-    dlmalloc, dlfree, dlcalloc, dlrealloc, dlmemalign
+    dlmalloc, dlfree, dlcalloc, dlrealloc, dlmemalign, dlmalloc_usable_size
 };
 
 extern const char* __progname;
@@ -280,11 +281,6 @@ extern const char* __progname;
  */
 static void* libc_malloc_impl_handle = NULL;
 
-// This must match the alignment used by dlmalloc.
-#ifndef MALLOC_ALIGNMENT
-#define MALLOC_ALIGNMENT ((size_t)(2 * sizeof(void *)))
-#endif
-
 /* This variable is set to the value of property libc.debug.malloc.backlog,
  * when the value of libc.debug.malloc = 10.  It determines the size of the
  * backlog we use to detect multiple frees.  If the property is not set, the
@@ -296,41 +292,26 @@ unsigned int gMallocDebugBacklog;
 /* The value of libc.debug.malloc. */
 int gMallocDebugLevel;
 
-static void InitMalloc(MallocDebug* table, const char* prefix) {
+template<typename FunctionType>
+void InitMallocFunction(void* malloc_impl_handler, FunctionType* func, const char* prefix, const char* suffix) {
+    char symbol[128];
+    snprintf(symbol, sizeof(symbol), "%s_%s", prefix, suffix);
+    *func = reinterpret_cast<FunctionType>(dlsym(malloc_impl_handler, symbol));
+    if (*func == NULL) {
+        error_log("%s: dlsym(\"%s\") failed", __progname, symbol);
+    }
+}
+
+static void InitMalloc(void* malloc_impl_handler, MallocDebug* table, const char* prefix) {
     __libc_format_log(ANDROID_LOG_INFO, "libc", "%s: using libc.debug.malloc %d (%s)\n",
                       __progname, gMallocDebugLevel, prefix);
 
-  char symbol[128];
+    InitMallocFunction<MallocDebugMalloc>(malloc_impl_handler, &table->malloc, prefix, "malloc");
-
+    InitMallocFunction<MallocDebugFree>(malloc_impl_handler, &table->free, prefix, "free");
-  snprintf(symbol, sizeof(symbol), "%s_malloc", prefix);
+    InitMallocFunction<MallocDebugCalloc>(malloc_impl_handler, &table->calloc, prefix, "calloc");
-  table->malloc = reinterpret_cast<MallocDebugMalloc>(dlsym(libc_malloc_impl_handle, symbol));
+    InitMallocFunction<MallocDebugRealloc>(malloc_impl_handler, &table->realloc, prefix, "realloc");
-  if (table->malloc == NULL) {
+    InitMallocFunction<MallocDebugMemalign>(malloc_impl_handler, &table->memalign, prefix, "memalign");
-      error_log("%s: dlsym(\"%s\") failed", __progname, symbol);
+    InitMallocFunction<MallocDebugMallocUsableSize>(malloc_impl_handler, &table->malloc_usable_size, prefix, "malloc_usable_size");
-  }
-
-  snprintf(symbol, sizeof(symbol), "%s_free", prefix);
-  table->free = reinterpret_cast<MallocDebugFree>(dlsym(libc_malloc_impl_handle, symbol));
-  if (table->free == NULL) {
-      error_log("%s: dlsym(\"%s\") failed", __progname, symbol);
-  }
-
-  snprintf(symbol, sizeof(symbol), "%s_calloc", prefix);
-  table->calloc = reinterpret_cast<MallocDebugCalloc>(dlsym(libc_malloc_impl_handle, symbol));
-  if (table->calloc == NULL) {
-      error_log("%s: dlsym(\"%s\") failed", __progname, symbol);
-  }
-
-  snprintf(symbol, sizeof(symbol), "%s_realloc", prefix);
-  table->realloc = reinterpret_cast<MallocDebugRealloc>(dlsym(libc_malloc_impl_handle, symbol));
-  if (table->realloc == NULL) {
-      error_log("%s: dlsym(\"%s\") failed", __progname, symbol);
-  }
-
-  snprintf(symbol, sizeof(symbol), "%s_memalign", prefix);
-  table->memalign = reinterpret_cast<MallocDebugMemalign>(dlsym(libc_malloc_impl_handle, symbol));
-  if (table->memalign == NULL) {
-      error_log("%s: dlsym(\"%s\") failed", __progname, symbol);
-  }
 }
 
 /* Initializes memory allocation framework once per process. */
@@ -422,24 +403,24 @@ static void malloc_init_impl() {
     }
 
     // Load .so that implements the required malloc debugging functionality.
-    libc_malloc_impl_handle = dlopen(so_name, RTLD_LAZY);
+    void* malloc_impl_handle = dlopen(so_name, RTLD_LAZY);
-    if (libc_malloc_impl_handle == NULL) {
+    if (malloc_impl_handle == NULL) {
         error_log("%s: Missing module %s required for malloc debug level %d: %s",
                   __progname, so_name, gMallocDebugLevel, dlerror());
         return;
     }
 
     // Initialize malloc debugging in the loaded module.
-    malloc_debug_initialize = reinterpret_cast<MallocDebugInit>(dlsym(libc_malloc_impl_handle,
+    malloc_debug_initialize = reinterpret_cast<MallocDebugInit>(dlsym(malloc_impl_handle,
                                                                       "malloc_debug_initialize"));
     if (malloc_debug_initialize == NULL) {
         error_log("%s: Initialization routine is not found in %s\n",
                   __progname, so_name);
-        dlclose(libc_malloc_impl_handle);
+        dlclose(malloc_impl_handle);
         return;
     }
     if (malloc_debug_initialize() == -1) {
-        dlclose(libc_malloc_impl_handle);
+        dlclose(malloc_impl_handle);
         return;
     }
 
@@ -447,34 +428,35 @@ static void malloc_init_impl() {
         // For memory checker we need to do extra initialization.
         typedef int (*MemCheckInit)(int, const char*);
         MemCheckInit memcheck_initialize =
-            reinterpret_cast<MemCheckInit>(dlsym(libc_malloc_impl_handle,
+            reinterpret_cast<MemCheckInit>(dlsym(malloc_impl_handle,
                                                  "memcheck_initialize"));
         if (memcheck_initialize == NULL) {
             error_log("%s: memcheck_initialize routine is not found in %s\n",
                       __progname, so_name);
-            dlclose(libc_malloc_impl_handle);
+            dlclose(malloc_impl_handle);
             return;
         }
 
         if (memcheck_initialize(MALLOC_ALIGNMENT, memcheck_tracing)) {
-            dlclose(libc_malloc_impl_handle);
+            dlclose(malloc_impl_handle);
             return;
         }
     }
 
+
     // Initialize malloc dispatch table with appropriate routines.
     switch (gMallocDebugLevel) {
         case 1:
-            InitMalloc(&gMallocUse, "leak");
+            InitMalloc(malloc_impl_handle, &gMallocUse, "leak");
             break;
         case 5:
-            InitMalloc(&gMallocUse, "fill");
+            InitMalloc(malloc_impl_handle, &gMallocUse, "fill");
             break;
         case 10:
-            InitMalloc(&gMallocUse, "chk");
+            InitMalloc(malloc_impl_handle, &gMallocUse, "chk");
             break;
         case 20:
-            InitMalloc(&gMallocUse, "qemu_instrumented");
+            InitMalloc(malloc_impl_handle, &gMallocUse, "qemu_instrumented");
             break;
         default:
             break;
@@ -485,13 +467,14 @@ static void malloc_init_impl() {
         (gMallocUse.free == NULL) ||
         (gMallocUse.calloc == NULL) ||
         (gMallocUse.realloc == NULL) ||
-        (gMallocUse.memalign == NULL)) {
+        (gMallocUse.memalign == NULL) ||
+        (gMallocUse.malloc_usable_size == NULL)) {
         error_log("%s: some symbols for libc.debug.malloc level %d were not found (see above)",
                   __progname, gMallocDebugLevel);
-        dlclose(libc_malloc_impl_handle);
+        dlclose(malloc_impl_handle);
-        libc_malloc_impl_handle = NULL;
     } else {
         __libc_malloc_dispatch = &gMallocUse;
+        libc_malloc_impl_handle = malloc_impl_handle;
     }
 }
 
@@ -524,7 +507,7 @@ static pthread_once_t  malloc_fini_once_ctl = PTHREAD_ONCE_INIT;
  * This routine is called from __libc_init routines implemented
  * in libc_init_static.c and libc_init_dynamic.c files.
  */
-extern "C" void malloc_debug_init() {
+extern "C" __LIBC_HIDDEN__ void malloc_debug_init() {
     /* We need to initialize malloc iff we implement here custom
      * malloc routines (i.e. USE_DL_PREFIX is defined) for libc.so */
 #if defined(USE_DL_PREFIX) && !defined(LIBC_STATIC)
@@ -534,7 +517,7 @@ extern "C" void malloc_debug_init() {
 #endif  // USE_DL_PREFIX && !LIBC_STATIC
 }
 
-extern "C" void malloc_debug_fini() {
+extern "C" __LIBC_HIDDEN__ void malloc_debug_fini() {
     /* We need to finalize malloc iff we implement here custom
      * malloc routines (i.e. USE_DL_PREFIX is defined) for libc.so */
 #if defined(USE_DL_PREFIX) && !defined(LIBC_STATIC)