|
 |
00db10 |
Backport of the following commit:
|
|
|
00db10 |
|
|
|
00db10 |
commit 96d6fd6c4060d739abb1822e7ad633af749532b2
|
|
|
00db10 |
Author: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Tue Dec 23 05:59:44 2014 -0600
|
|
|
00db10 |
|
|
|
00db10 |
powerpc: Optimized st{r,p}cpy for POWER8/PPC64
|
|
|
00db10 |
|
|
|
00db10 |
This patch adds an optimized POWER8 strcpy using unaligned accesses.
|
|
|
00db10 |
For strings up to 16 bytes the implementation first calculate the
|
|
|
00db10 |
string size, like strlen, and issues a memcpy. For larger strings,
|
|
|
00db10 |
source is first aligned to 16 bytes and then tested over a loop that
|
|
|
00db10 |
reads 16 bytes am combine the cmpb results for speedup. Special case is
|
|
|
00db10 |
added for page cross reads.
|
|
|
00db10 |
|
|
|
00db10 |
It shows 30%-60% improvement over the optimized POWER7 one that uses
|
|
|
00db10 |
only aligned accesses.
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
2015-01-13 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/Makefile [sysdep_routines]: Add
|
|
|
00db10 |
strcpy-power8 and stpcpy-power8 objects.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
(__libc_ifunc_impl_list): Add __strcpy_power8 and __stpcpy_power8
|
|
|
00db10 |
implementations.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-power8.S: New file:
|
|
|
00db10 |
multiarch stpcpy implementation for POWER8.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/strcpy-power8.S: New file;
|
|
|
00db10 |
multiarch strcpy implementation for POWER8.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/strcpy.c (strcpy): Add
|
|
|
00db10 |
__strcpy_power8 function.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/power8/stpcpy.S: New file: optimized
|
|
|
00db10 |
stpcpy for POWER8.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/power8/strcpy.S: New file: optimized
|
|
|
00db10 |
strcpy for POWER8.
|
|
|
00db10 |
* NEWS: Update.
|
|
|
00db10 |
|
|
|
00db10 |
and the commits it depends on:
|
|
|
00db10 |
|
|
|
00db10 |
commit a52374e82b90a6039c720f7b9b7dfa9db24ff4f0
|
|
|
00db10 |
Author: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Fri Dec 13 14:55:22 2013 -0500
|
|
|
00db10 |
|
|
|
00db10 |
PowerPC: multiarch stpcpy for PowerPC64
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
2013-12-13 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/Makefile: Add stpcpy
|
|
|
00db10 |
multiarch implementations.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
(__libc_ifunc_impl_list): Likewise.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.c: New file.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.c: New file.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/stpcpy.c : New file:
|
|
|
00db10 |
multiarch stpcpy for PPC64.
|
|
|
00db10 |
|
|
|
00db10 |
commit 7f5ec11336e46d0449a6b5a8e5a0604c3c78ecbf
|
|
|
00db10 |
Author: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Fri Dec 13 14:54:41 2013 -0500
|
|
|
00db10 |
|
|
|
00db10 |
PowerPC: multiarch strcpy for PowerPC64
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
2013-12-13 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/Makefile: Add strcpy
|
|
|
00db10 |
multiarch implementations.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
(__libc_ifunc_impl_list): Likewise.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.c: New file.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.c: New file.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/strcpy.c : New file:
|
|
|
00db10 |
multiarch strcpy for PPC64.
|
|
|
00db10 |
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/Makefile b/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
00db10 |
index 8dceb09..1cdd5d6 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
00db10 |
@@ -5,6 +5,8 @@ sysdep_routines += memcpy-power7 memcpy-a2 memcpy-power6 memcpy-cell \
|
|
|
00db10 |
memset-ppc64 bzero-power4 bzero-power6 bzero-power7 \
|
|
|
00db10 |
mempcpy-power7 mempcpy-ppc64 memchr-power7 memchr-ppc64 \
|
|
|
00db10 |
memrchr-power7 memrchr-ppc64 rawmemchr-power7 \
|
|
|
00db10 |
+ stpcpy-power8 stpcpy-power7 stpcpy-ppc64 \
|
|
|
00db10 |
+ strcpy-power8 strcpy-power7 strcpy-ppc64 \
|
|
|
00db10 |
rawmemchr-ppc64 strlen-power7 strlen-ppc64 strnlen-power7 \
|
|
|
00db10 |
strnlen-ppc64 strcasecmp-power7 strcasecmp_l-power7 \
|
|
|
00db10 |
strncase-power7 strncase_l-power7 strncmp-power7 \
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
index 2d21ce1..e89fd3e 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
@@ -34,6 +34,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
00db10 |
size_t i = 0;
|
|
|
00db10 |
|
|
|
00db10 |
unsigned long int hwcap = GLRO(dl_hwcap);
|
|
|
00db10 |
+ unsigned long int hwcap2 = GLRO(dl_hwcap2);
|
|
|
00db10 |
+
|
|
|
00db10 |
/* hwcap contains only the latest supported ISA, the code checks which is
|
|
|
00db10 |
and fills the previous supported ones. */
|
|
|
00db10 |
if (hwcap & PPC_FEATURE_ARCH_2_06)
|
|
|
00db10 |
@@ -71,6 +73,24 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
00db10 |
__memset_power4)
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, memset, 1, __memset_ppc))
|
|
|
00db10 |
|
|
|
00db10 |
+ /* Support sysdeps/powerpc/powerpc64/multiarch/strcpy.c. */
|
|
|
00db10 |
+ IFUNC_IMPL (i, name, strcpy,
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, strcpy, hwcap2 & PPC_FEATURE2_ARCH_2_07,
|
|
|
00db10 |
+ __strcpy_power8)
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, strcpy, hwcap & PPC_FEATURE_HAS_VSX,
|
|
|
00db10 |
+ __strcpy_power7)
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, strcpy, 1,
|
|
|
00db10 |
+ __strcpy_ppc))
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Support sysdeps/powerpc/powerpc64/multiarch/stpcpy.c. */
|
|
|
00db10 |
+ IFUNC_IMPL (i, name, stpcpy,
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, stpcpy, hwcap2 & PPC_FEATURE2_ARCH_2_07,
|
|
|
00db10 |
+ __stpcpy_power8)
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, stpcpy, hwcap & PPC_FEATURE_HAS_VSX,
|
|
|
00db10 |
+ __stpcpy_power7)
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, stpcpy, 1,
|
|
|
00db10 |
+ __stpcpy_ppc))
|
|
|
00db10 |
+
|
|
|
00db10 |
/* Support sysdeps/powerpc/powerpc64/multiarch/strlen.c. */
|
|
|
00db10 |
IFUNC_IMPL (i, name, strlen,
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, strlen, hwcap & PPC_FEATURE_HAS_VSX,
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..0943611
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power7.S
|
|
|
00db10 |
@@ -0,0 +1,40 @@
|
|
|
00db10 |
+/* Optimized stpcpy implementation for POWER7.
|
|
|
00db10 |
+ Copyright (C) 2013-2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef EALIGN
|
|
|
00db10 |
+#define EALIGN(name, alignt, words) \
|
|
|
00db10 |
+ .section ".text"; \
|
|
|
00db10 |
+ ENTRY_2(__stpcpy_power7) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__stpcpy_power7): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__stpcpy_power7)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef END
|
|
|
00db10 |
+#define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__stpcpy_power7) \
|
|
|
00db10 |
+ END_2(__stpcpy_power7)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef libc_hidden_builtin_def
|
|
|
00db10 |
+#define libc_hidden_builtin_def(name)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power7/stpcpy.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power8.S b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power8.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..66e6f70
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-power8.S
|
|
|
00db10 |
@@ -0,0 +1,40 @@
|
|
|
00db10 |
+/* Optimized stpcpy implementation for POWER8/PPC64.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef EALIGN
|
|
|
00db10 |
+#define EALIGN(name, alignt, words) \
|
|
|
00db10 |
+ .section ".text"; \
|
|
|
00db10 |
+ ENTRY_2(__stpcpy_power8) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__stpcpy_power8): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__stpcpy_power8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef END
|
|
|
00db10 |
+#define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__stpcpy_power8) \
|
|
|
00db10 |
+ END_2(__stpcpy_power8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef libc_hidden_builtin_def
|
|
|
00db10 |
+#define libc_hidden_builtin_def(name)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power8/stpcpy.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..ac70c1b
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy-ppc64.S
|
|
|
00db10 |
@@ -0,0 +1,48 @@
|
|
|
00db10 |
+/* Default stpcpy implementation for PowerPC64.
|
|
|
00db10 |
+ Copyright (C) 2013-2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined SHARED && !defined NOT_IN_libc
|
|
|
00db10 |
+# undef EALIGN
|
|
|
00db10 |
+# define EALIGN(name, alignt, words) \
|
|
|
00db10 |
+ .section ".text"; \
|
|
|
00db10 |
+ ENTRY_2(__stpcpy_ppc) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__stpcpy_ppc): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__stpcpy_ppc)
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef END
|
|
|
00db10 |
+# define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__stpcpy_ppc) \
|
|
|
00db10 |
+ END_2(__stpcpy_ppc)
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef weak_alias
|
|
|
00db10 |
+# define weak_alias(name, alias)
|
|
|
00db10 |
+# undef libc_hidden_def
|
|
|
00db10 |
+# define libc_hidden_def(name)
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef libc_hidden_builtin_def
|
|
|
00db10 |
+# define libc_hidden_builtin_def(name) \
|
|
|
00db10 |
+ .globl __GI___stpcpy; __GI___stpcpy = __stpcpy_ppc
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/stpcpy.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c b/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..2ab62bf
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
|
|
|
00db10 |
@@ -0,0 +1,35 @@
|
|
|
00db10 |
+/* Multiple versions of stpcpy. PowerPC64 version.
|
|
|
00db10 |
+ Copyright (C) 2013-2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined SHARED && !defined NOT_IN_libc
|
|
|
00db10 |
+# define NO_MEMPCPY_STPCPY_REDIRECT
|
|
|
00db10 |
+# include <string.h>
|
|
|
00db10 |
+# include <shlib-compat.h>
|
|
|
00db10 |
+# include "init-arch.h"
|
|
|
00db10 |
+
|
|
|
00db10 |
+extern __typeof (__stpcpy) __stpcpy_ppc attribute_hidden;
|
|
|
00db10 |
+extern __typeof (__stpcpy) __stpcpy_power7 attribute_hidden;
|
|
|
00db10 |
+
|
|
|
00db10 |
+libc_ifunc (__stpcpy,
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_HAS_VSX)
|
|
|
00db10 |
+ ? __stpcpy_power7
|
|
|
00db10 |
+ : __stpcpy_ppc);
|
|
|
00db10 |
+
|
|
|
00db10 |
+weak_alias (__stpcpy, stpcpy)
|
|
|
00db10 |
+libc_hidden_def (stpcpy)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S b/sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..69851bb
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy-power7.S
|
|
|
00db10 |
@@ -0,0 +1,40 @@
|
|
|
00db10 |
+/* Optimized strcpy implementation for POWER7.
|
|
|
00db10 |
+ Copyright (C) 2013-2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef EALIGN
|
|
|
00db10 |
+#define EALIGN(name, alignt, words) \
|
|
|
00db10 |
+ .section ".text"; \
|
|
|
00db10 |
+ ENTRY_2(__strcpy_power7) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__strcpy_power7): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__strcpy_power7)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef END
|
|
|
00db10 |
+#define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__strcpy_power7) \
|
|
|
00db10 |
+ END_2(__strcpy_power7)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef libc_hidden_builtin_def
|
|
|
00db10 |
+#define libc_hidden_builtin_def(name)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power7/strcpy.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy-power8.S b/sysdeps/powerpc/powerpc64/multiarch/strcpy-power8.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..64cbc16
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy-power8.S
|
|
|
00db10 |
@@ -0,0 +1,40 @@
|
|
|
00db10 |
+/* Optimized strcpy implementation for POWER8/PPC64.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef EALIGN
|
|
|
00db10 |
+#define EALIGN(name, alignt, words) \
|
|
|
00db10 |
+ .section ".text"; \
|
|
|
00db10 |
+ ENTRY_2(__strcpy_power8) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__strcpy_power8): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__strcpy_power8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef END
|
|
|
00db10 |
+#define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__strcpy_power8) \
|
|
|
00db10 |
+ END_2(__strcpy_power8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef libc_hidden_builtin_def
|
|
|
00db10 |
+#define libc_hidden_builtin_def(name)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power8/strcpy.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S b/sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..e5452b1
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy-ppc64.S
|
|
|
00db10 |
@@ -0,0 +1,43 @@
|
|
|
00db10 |
+/* Default strcpy implementation for PowerPC64.
|
|
|
00db10 |
+ Copyright (C) 2013-2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined SHARED && !defined NOT_IN_libc
|
|
|
00db10 |
+# undef EALIGN
|
|
|
00db10 |
+# define EALIGN(name, alignt, words) \
|
|
|
00db10 |
+ .section ".text"; \
|
|
|
00db10 |
+ ENTRY_2(__strcpy_ppc) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__strcpy_ppc): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__strcpy_ppc)
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef END
|
|
|
00db10 |
+# define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__strcpy_ppc) \
|
|
|
00db10 |
+ END_2(__strcpy_ppc)
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef libc_hidden_builtin_def
|
|
|
00db10 |
+# define libc_hidden_builtin_def(name) \
|
|
|
00db10 |
+ .globl __GI_strcpy; __GI_strcpy = __strcpy_ppc
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/strcpy.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy.c b/sysdeps/powerpc/powerpc64/multiarch/strcpy.c
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..5a86f26
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy.c
|
|
|
00db10 |
@@ -0,0 +1,34 @@
|
|
|
00db10 |
+/* Multiple versions of strcpy. PowerPC64 version.
|
|
|
00db10 |
+ Copyright (C) 2013-2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined SHARED && !defined NOT_IN_libc
|
|
|
00db10 |
+# include <string.h>
|
|
|
00db10 |
+# include <shlib-compat.h>
|
|
|
00db10 |
+# include "init-arch.h"
|
|
|
00db10 |
+
|
|
|
00db10 |
+extern __typeof (strcpy) __strcpy_ppc attribute_hidden;
|
|
|
00db10 |
+extern __typeof (strcpy) __strcpy_power7 attribute_hidden;
|
|
|
00db10 |
+extern __typeof (strcpy) __strcpy_power8 attribute_hidden;
|
|
|
00db10 |
+
|
|
|
00db10 |
+libc_ifunc (strcpy,
|
|
|
00db10 |
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07)
|
|
|
00db10 |
+ ? __strcpy_power8 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_HAS_VSX)
|
|
|
00db10 |
+ ? __strcpy_power7
|
|
|
00db10 |
+ : __strcpy_ppc);
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/power8/stpcpy.S b/sysdeps/powerpc/powerpc64/power8/stpcpy.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..bf72065
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/power8/stpcpy.S
|
|
|
00db10 |
@@ -0,0 +1,24 @@
|
|
|
00db10 |
+/* Optimized stpcpy implementation for PowerPC64/POWER8.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#define USE_AS_STPCPY
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power8/strcpy.S>
|
|
|
00db10 |
+
|
|
|
00db10 |
+weak_alias (__stpcpy, stpcpy)
|
|
|
00db10 |
+libc_hidden_def (__stpcpy)
|
|
|
00db10 |
+libc_hidden_builtin_def (stpcpy)
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/power8/strcpy.S b/sysdeps/powerpc/powerpc64/power8/strcpy.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..d3e9a10
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/power8/strcpy.S
|
|
|
00db10 |
@@ -0,0 +1,262 @@
|
|
|
00db10 |
+/* Optimized strcpy/stpcpy implementation for PowerPC64/POWER8.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#ifdef USE_AS_STPCPY
|
|
|
00db10 |
+# define FUNC_NAME __stpcpy
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+# define FUNC_NAME strcpy
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Implements the function
|
|
|
00db10 |
+
|
|
|
00db10 |
+ char * [r3] strcpy (char *dest [r3], const char *src [r4])
|
|
|
00db10 |
+
|
|
|
00db10 |
+ or
|
|
|
00db10 |
+
|
|
|
00db10 |
+ char * [r3] stpcpy (char *dest [r3], const char *src [r4])
|
|
|
00db10 |
+
|
|
|
00db10 |
+ if USE_AS_STPCPY is defined.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The implementation uses unaligned doubleword access to avoid specialized
|
|
|
00db10 |
+ code paths depending of data alignment. Although recent powerpc64 uses
|
|
|
00db10 |
+ 64K as default, the page cross handling assumes minimum page size of
|
|
|
00db10 |
+ 4k. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .machine power7
|
|
|
00db10 |
+EALIGN (FUNC_NAME, 4, 0)
|
|
|
00db10 |
+ li r0,0 /* Doubleword with null chars to use
|
|
|
00db10 |
+ with cmpb. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Check if the [src]+15 will cross a 4K page by checking if the bit
|
|
|
00db10 |
+ indicating the page size changes. Basically:
|
|
|
00db10 |
+
|
|
|
00db10 |
+ uint64_t srcin = (uint64_t)src;
|
|
|
00db10 |
+ uint64_t ob = srcin & 4096UL;
|
|
|
00db10 |
+ uint64_t nb = (srcin+15UL) & 4096UL;
|
|
|
00db10 |
+ if (ob ^ nb)
|
|
|
00db10 |
+ goto pagecross; */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ addi r9,r4,15
|
|
|
00db10 |
+ xor r9,r9,r4
|
|
|
00db10 |
+ rlwinm. r9,r9,0,19,19
|
|
|
00db10 |
+ bne L(pagecross)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* For short string (less than 16 bytes), just calculate its size as
|
|
|
00db10 |
+ strlen and issues a memcpy if null is found. */
|
|
|
00db10 |
+ mr r7,r4
|
|
|
00db10 |
+ ld r12,0(r7) /* Load doubleword from memory. */
|
|
|
00db10 |
+ cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */
|
|
|
00db10 |
+ cmpdi cr7,r10,0 /* If r10 == 0, no null's have been found. */
|
|
|
00db10 |
+ bne cr7,L(done)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ ldu r8,8(r7)
|
|
|
00db10 |
+ cmpb r10,r8,r0
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ bne cr7,L(done)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ b L(loop_before)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(pagecross):
|
|
|
00db10 |
+ clrrdi r7,r4,3 /* Align the address to doubleword boundary. */
|
|
|
00db10 |
+ rlwinm r6,r4,3,26,28 /* Calculate padding. */
|
|
|
00db10 |
+ li r5,-1 /* MASK = 0xffffffffffffffff. */
|
|
|
00db10 |
+ ld r12,0(r7) /* Load doubleword from memory. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ sld r5,r5,r6
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ srd r5,r5,r6 /* MASK = MASK >> padding. */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ orc r9,r12,r5 /* Mask bits that are not part of the string. */
|
|
|
00db10 |
+ cmpb r10,r9,r0 /* Check for null bytes in DWORD1. */
|
|
|
00db10 |
+ cmpdi cr7,r10,0 /* If r10 == 0, no null's have been found. */
|
|
|
00db10 |
+ bne cr7,L(done)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ ldu r6,8(r7)
|
|
|
00db10 |
+ cmpb r10,r6,r0
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ bne cr7,L(done)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ ld r12,0(r7)
|
|
|
00db10 |
+ cmpb r10,r12,r0
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ bne cr7,L(done)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ ldu r6,8(r7)
|
|
|
00db10 |
+ cmpb r10,r6,r0
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ bne cr7,L(done)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* We checked for 24 - x bytes, with x being the source alignment
|
|
|
00db10 |
+ (0 <= x <= 16), and no zero has been found. Start the loop
|
|
|
00db10 |
+ copy with doubleword aligned address. */
|
|
|
00db10 |
+ mr r7,r4
|
|
|
00db10 |
+ ld r12, 0(r7)
|
|
|
00db10 |
+ ldu r8, 8(r7)
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(loop_before):
|
|
|
00db10 |
+ /* Save the two doublewords readed from source and align the source
|
|
|
00db10 |
+ to 16 bytes for the loop. */
|
|
|
00db10 |
+ mr r11,r3
|
|
|
00db10 |
+ std r12,0(r11)
|
|
|
00db10 |
+ std r8,8(r11)
|
|
|
00db10 |
+ addi r11,r11,16
|
|
|
00db10 |
+ rldicl r9,r4,0,60
|
|
|
00db10 |
+ subf r7,r9,r7
|
|
|
00db10 |
+ subf r11,r9,r11
|
|
|
00db10 |
+ b L(loop_start)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 5
|
|
|
00db10 |
+L(loop):
|
|
|
00db10 |
+ std r12, 0(r11)
|
|
|
00db10 |
+ std r6, 8(r11)
|
|
|
00db10 |
+ addi r11,r11,16
|
|
|
00db10 |
+L(loop_start):
|
|
|
00db10 |
+ /* Load two doublewords, compare and merge in a
|
|
|
00db10 |
+ single register for speed. This is an attempt
|
|
|
00db10 |
+ to speed up the null-checking process for bigger strings. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ ld r12, 8(r7)
|
|
|
00db10 |
+ ldu r6, 16(r7)
|
|
|
00db10 |
+ cmpb r10,r12,r0
|
|
|
00db10 |
+ cmpb r9,r6,r0
|
|
|
00db10 |
+ or r8,r9,r10 /* Merge everything in one doubleword. */
|
|
|
00db10 |
+ cmpdi cr7,r8,0
|
|
|
00db10 |
+ beq cr7,L(loop)
|
|
|
00db10 |
+
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* OK, one (or both) of the doublewords contains a null byte. Check
|
|
|
00db10 |
+ the first doubleword and decrement the address in case the first
|
|
|
00db10 |
+ doubleword really contains a null byte. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ addi r4,r7,-8
|
|
|
00db10 |
+ cmpdi cr6,r10,0
|
|
|
00db10 |
+ addi r7,r7,-8
|
|
|
00db10 |
+ bne cr6,L(done2)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* The null byte must be in the second doubleword. Adjust the address
|
|
|
00db10 |
+ again and move the result of cmpb to r10 so we can calculate the
|
|
|
00db10 |
+ length. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ mr r10,r9
|
|
|
00db10 |
+ addi r7,r7,8
|
|
|
00db10 |
+ b L(done2)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* r10 has the output of the cmpb instruction, that is, it contains
|
|
|
00db10 |
+ 0xff in the same position as the null byte in the original
|
|
|
00db10 |
+ doubleword from the string. Use that to calculate the length. */
|
|
|
00db10 |
+L(done):
|
|
|
00db10 |
+ mr r11,r3
|
|
|
00db10 |
+L(done2):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ addi r9, r10, -1 /* Form a mask from trailing zeros. */
|
|
|
00db10 |
+ andc r9, r9, r10
|
|
|
00db10 |
+ popcntd r6, r9 /* Count the bits in the mask. */
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ cntlzd r6,r10 /* Count leading zeros before the match. */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ subf r5,r4,r7
|
|
|
00db10 |
+ srdi r6,r6,3 /* Convert leading/trailing zeros to bytes. */
|
|
|
00db10 |
+ add r8,r5,r6 /* Compute final length. */
|
|
|
00db10 |
+#ifdef USE_AS_STPCPY
|
|
|
00db10 |
+ /* stpcpy returns the dest address plus the size not counting the
|
|
|
00db10 |
+ final '\0'. */
|
|
|
00db10 |
+ add r3,r11,r8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ addi r8,r8,1 /* Final '/0'. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ cmpldi cr6,r8,8
|
|
|
00db10 |
+ mtocrf 0x01,r8
|
|
|
00db10 |
+ ble cr6,L(copy_LE_8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ cmpldi cr1,r8,16
|
|
|
00db10 |
+ blt cr1,8f
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Handle copies of 0~31 bytes. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(copy_LT_32):
|
|
|
00db10 |
+ /* At least 6 bytes to go. */
|
|
|
00db10 |
+ blt cr1,8f
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Copy 16 bytes. */
|
|
|
00db10 |
+ ld r6,0(r4)
|
|
|
00db10 |
+ ld r8,8(r4)
|
|
|
00db10 |
+ addi r4,r4,16
|
|
|
00db10 |
+ std r6,0(r11)
|
|
|
00db10 |
+ std r8,8(r11)
|
|
|
00db10 |
+ addi r11,r11,16
|
|
|
00db10 |
+8: /* Copy 8 bytes. */
|
|
|
00db10 |
+ bf 28,L(tail4)
|
|
|
00db10 |
+ ld r6,0(r4)
|
|
|
00db10 |
+ addi r4,r4,8
|
|
|
00db10 |
+ std r6,0(r11)
|
|
|
00db10 |
+ addi r11,r11,8
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+/* Copies 4~7 bytes. */
|
|
|
00db10 |
+L(tail4):
|
|
|
00db10 |
+ bf 29,L(tail2)
|
|
|
00db10 |
+ lwz r6,0(r4)
|
|
|
00db10 |
+ stw r6,0(r11)
|
|
|
00db10 |
+ bf 30,L(tail5)
|
|
|
00db10 |
+ lhz r7,4(r4)
|
|
|
00db10 |
+ sth r7,4(r11)
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz r8,6(r4)
|
|
|
00db10 |
+ stb r8,6(r11)
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+/* Copies 2~3 bytes. */
|
|
|
00db10 |
+L(tail2):
|
|
|
00db10 |
+ bf 30,1f
|
|
|
00db10 |
+ lhz r6,0(r4)
|
|
|
00db10 |
+ sth r6,0(r11)
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz r7,2(r4)
|
|
|
00db10 |
+ stb r7,2(r11)
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(tail5):
|
|
|
00db10 |
+ bf 31,1f
|
|
|
00db10 |
+ lbz r6,4(r4)
|
|
|
00db10 |
+ stb r6,4(r11)
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+1:
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz r6,0(r4)
|
|
|
00db10 |
+ stb r6,0(r11)
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Handles copies of 0~8 bytes. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(copy_LE_8):
|
|
|
00db10 |
+ bne cr6,L(tail4)
|
|
|
00db10 |
+ ld r6,0(r4)
|
|
|
00db10 |
+ std r6,0(r11)
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+END (FUNC_NAME)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#ifndef USE_AS_STPCPY
|
|
|
00db10 |
+libc_hidden_builtin_def (strcpy)
|
|
|
00db10 |
+#endif
|