|
 |
a35d80 |
From 7beee7b39adeda657f45989b0635033dae25a1fd Mon Sep 17 00:00:00 2001
|
|
|
a35d80 |
From: Raphael M Zinsly <rzinsly@linux.ibm.com>
|
|
|
a35d80 |
Date: Thu, 12 Nov 2020 13:12:24 -0300
|
|
|
a35d80 |
Subject: powerpc: Add optimized stpncpy for POWER9
|
|
|
a35d80 |
|
|
|
a35d80 |
Add stpncpy support into the POWER9 strncpy.
|
|
|
a35d80 |
|
|
|
a35d80 |
Reviewed-by: Matheus Castanho <msc@linux.ibm.com>
|
|
|
a35d80 |
Reviewed-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>
|
|
|
a35d80 |
|
|
|
a35d80 |
diff --git a/sysdeps/powerpc/powerpc64/le/power9/stpncpy.S b/sysdeps/powerpc/powerpc64/le/power9/stpncpy.S
|
|
|
a35d80 |
new file mode 100644
|
|
|
a35d80 |
index 0000000000..81d9673d8b
|
|
|
a35d80 |
--- /dev/null
|
|
|
a35d80 |
+++ b/sysdeps/powerpc/powerpc64/le/power9/stpncpy.S
|
|
|
a35d80 |
@@ -0,0 +1,24 @@
|
|
|
a35d80 |
+/* Optimized stpncpy implementation for POWER9 LE.
|
|
|
a35d80 |
+ Copyright (C) 2020 Free Software Foundation, Inc.
|
|
|
a35d80 |
+ This file is part of the GNU C Library.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
a35d80 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
a35d80 |
+ License as published by the Free Software Foundation; either
|
|
|
a35d80 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
a35d80 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
a35d80 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
a35d80 |
+ Lesser General Public License for more details.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
a35d80 |
+ License along with the GNU C Library; if not, see
|
|
|
a35d80 |
+ <https://www.gnu.org/licenses/>. */
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+#define USE_AS_STPNCPY
|
|
|
a35d80 |
+#include <sysdeps/powerpc/powerpc64/le/power9/strncpy.S>
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+weak_alias (__stpncpy, stpncpy)
|
|
|
a35d80 |
+libc_hidden_def (__stpncpy)
|
|
|
a35d80 |
+libc_hidden_builtin_def (stpncpy)
|
|
|
a35d80 |
diff --git a/sysdeps/powerpc/powerpc64/le/power9/strncpy.S b/sysdeps/powerpc/powerpc64/le/power9/strncpy.S
|
|
|
a35d80 |
index cbfc37bda3..b4ba428662 100644
|
|
|
a35d80 |
--- a/sysdeps/powerpc/powerpc64/le/power9/strncpy.S
|
|
|
a35d80 |
+++ b/sysdeps/powerpc/powerpc64/le/power9/strncpy.S
|
|
|
a35d80 |
@@ -18,11 +18,19 @@
|
|
|
a35d80 |
|
|
|
a35d80 |
#include <sysdep.h>
|
|
|
a35d80 |
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+# ifndef STPNCPY
|
|
|
a35d80 |
+# define FUNC_NAME __stpncpy
|
|
|
a35d80 |
+# else
|
|
|
a35d80 |
+# define FUNC_NAME STPNCPY
|
|
|
a35d80 |
+# endif
|
|
|
a35d80 |
+#else
|
|
|
a35d80 |
# ifndef STRNCPY
|
|
|
a35d80 |
# define FUNC_NAME strncpy
|
|
|
a35d80 |
# else
|
|
|
a35d80 |
# define FUNC_NAME STRNCPY
|
|
|
a35d80 |
# endif
|
|
|
a35d80 |
+#endif /* !USE_AS_STPNCPY */
|
|
|
a35d80 |
|
|
|
a35d80 |
#ifndef MEMSET
|
|
|
a35d80 |
/* For builds without IFUNC support, local calls should be made to internal
|
|
|
a35d80 |
@@ -41,6 +49,12 @@
|
|
|
a35d80 |
|
|
|
a35d80 |
char * [r3] strncpy (char *dest [r3], const char *src [r4], size_t n [r5])
|
|
|
a35d80 |
|
|
|
a35d80 |
+ or
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ char * [r3] stpncpy (char *dest [r3], const char *src [r4], size_t n [r5])
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ if USE_AS_STPNCPY is defined.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
The implementation can load bytes past a null terminator, but only
|
|
|
a35d80 |
up to the next 16-byte aligned address, so it never crosses a page. */
|
|
|
a35d80 |
|
|
|
a35d80 |
@@ -66,7 +80,15 @@ ENTRY (FUNC_NAME, 4)
|
|
|
a35d80 |
|
|
|
a35d80 |
/* Empty/1-byte string optimization */
|
|
|
a35d80 |
cmpdi r5,0
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ bgt L(cont)
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ addi r3,r3,1
|
|
|
a35d80 |
+ blr
|
|
|
a35d80 |
+L(cont):
|
|
|
a35d80 |
+#else
|
|
|
a35d80 |
beqlr
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
|
|
|
a35d80 |
addi r4,r4,1
|
|
|
a35d80 |
neg r7,r4
|
|
|
a35d80 |
@@ -96,12 +118,20 @@ ENTRY (FUNC_NAME, 4)
|
|
|
a35d80 |
sldi r10,r5,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
stxvl 32+v0,r11,r10 /* Partial store */
|
|
|
a35d80 |
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r5
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
blr
|
|
|
a35d80 |
|
|
|
a35d80 |
L(null):
|
|
|
a35d80 |
sldi r10,r8,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
stxvl 32+v0,r11,r10 /* Partial store */
|
|
|
a35d80 |
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r7
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
add r11,r11,r8
|
|
|
a35d80 |
sub r5,r5,r8
|
|
|
a35d80 |
b L(zero_padding)
|
|
|
a35d80 |
@@ -185,6 +215,10 @@ L(n_tail4):
|
|
|
a35d80 |
sldi r10,r5,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
addi r11,r11,48 /* Offset */
|
|
|
a35d80 |
stxvl 32+v3,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r5
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
blr
|
|
|
a35d80 |
|
|
|
a35d80 |
L(prep_n_tail1):
|
|
|
a35d80 |
@@ -196,6 +230,10 @@ L(prep_n_tail1):
|
|
|
a35d80 |
L(n_tail1):
|
|
|
a35d80 |
sldi r10,r5,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
stxvl 32+v0,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r5
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
blr
|
|
|
a35d80 |
|
|
|
a35d80 |
L(prep_n_tail2):
|
|
|
a35d80 |
@@ -209,6 +247,10 @@ L(n_tail2):
|
|
|
a35d80 |
sldi r10,r5,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
addi r11,r11,16 /* offset */
|
|
|
a35d80 |
stxvl 32+v1,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r5
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
blr
|
|
|
a35d80 |
|
|
|
a35d80 |
L(prep_n_tail3):
|
|
|
a35d80 |
@@ -223,6 +265,10 @@ L(n_tail3):
|
|
|
a35d80 |
sldi r10,r5,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
addi r11,r11,32 /* Offset */
|
|
|
a35d80 |
stxvl 32+v2,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r5
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
blr
|
|
|
a35d80 |
|
|
|
a35d80 |
L(prep_tail1):
|
|
|
a35d80 |
@@ -232,6 +278,10 @@ L(tail1):
|
|
|
a35d80 |
addi r9,r8,1 /* Add null terminator */
|
|
|
a35d80 |
sldi r10,r9,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
stxvl 32+v0,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r8
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
add r11,r11,r9
|
|
|
a35d80 |
sub r5,r5,r9
|
|
|
a35d80 |
b L(zero_padding)
|
|
|
a35d80 |
@@ -246,6 +296,10 @@ L(tail2):
|
|
|
a35d80 |
sldi r10,r9,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
addi r11,r11,16 /* offset */
|
|
|
a35d80 |
stxvl 32+v1,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r8
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
add r11,r11,r9
|
|
|
a35d80 |
sub r5,r5,r9
|
|
|
a35d80 |
b L(zero_padding)
|
|
|
a35d80 |
@@ -261,6 +315,10 @@ L(tail3):
|
|
|
a35d80 |
sldi r10,r9,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
addi r11,r11,32 /* offset */
|
|
|
a35d80 |
stxvl 32+v2,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r8
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
add r11,r11,r9
|
|
|
a35d80 |
sub r5,r5,r9
|
|
|
a35d80 |
b L(zero_padding)
|
|
|
a35d80 |
@@ -276,6 +334,10 @@ L(tail4):
|
|
|
a35d80 |
sldi r10,r9,56 /* stxvl wants size in top 8 bits */
|
|
|
a35d80 |
addi r11,r11,48 /* offset */
|
|
|
a35d80 |
stxvl 32+v3,r11,r10 /* Partial store */
|
|
|
a35d80 |
+#ifdef USE_AS_STPNCPY
|
|
|
a35d80 |
+ /* Compute pointer to last byte copied into dest. */
|
|
|
a35d80 |
+ add r3,r11,r8
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
add r11,r11,r9
|
|
|
a35d80 |
sub r5,r5,r9
|
|
|
a35d80 |
|
|
|
a35d80 |
@@ -331,7 +393,8 @@ L(zero_padding_memset):
|
|
|
a35d80 |
ld r0,FRAMESIZE+16(r1)
|
|
|
a35d80 |
|
|
|
a35d80 |
mr r3,r30 /* Restore the return value of strncpy, i.e.:
|
|
|
a35d80 |
- dest. */
|
|
|
a35d80 |
+ dest. For stpncpy, the return value is the
|
|
|
a35d80 |
+ same as return value of memset. */
|
|
|
a35d80 |
ld r30,FRAMESIZE-8(r1) /* Restore r30. */
|
|
|
a35d80 |
/* Restore the stack frame. */
|
|
|
a35d80 |
addi r1,r1,FRAMESIZE
|
|
|
a35d80 |
@@ -342,3 +405,6 @@ L(zero_padding_memset):
|
|
|
a35d80 |
blr
|
|
|
a35d80 |
|
|
|
a35d80 |
END (FUNC_NAME)
|
|
|
a35d80 |
+#ifndef USE_AS_STPNCPY
|
|
|
a35d80 |
+libc_hidden_builtin_def (strncpy)
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/Makefile b/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
a35d80 |
index cd2b47b403..f46bf50732 100644
|
|
|
a35d80 |
--- a/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
a35d80 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
a35d80 |
@@ -33,7 +33,7 @@ sysdep_routines += memcpy-power8-cached memcpy-power7 memcpy-a2 memcpy-power6 \
|
|
|
a35d80 |
|
|
|
a35d80 |
ifneq (,$(filter %le,$(config-machine)))
|
|
|
a35d80 |
sysdep_routines += strcmp-power9 strncmp-power9 strcpy-power9 stpcpy-power9 \
|
|
|
a35d80 |
- rawmemchr-power9 strlen-power9 strncpy-power9
|
|
|
a35d80 |
+ rawmemchr-power9 strlen-power9 strncpy-power9 stpncpy-power9
|
|
|
a35d80 |
endif
|
|
|
a35d80 |
CFLAGS-strncase-power7.c += -mcpu=power7 -funroll-loops
|
|
|
a35d80 |
CFLAGS-strncase_l-power7.c += -mcpu=power7 -funroll-loops
|
|
|
a35d80 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
a35d80 |
index 135326c97a..8e19ebbf09 100644
|
|
|
a35d80 |
--- a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
a35d80 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
a35d80 |
@@ -318,6 +318,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
a35d80 |
|
|
|
a35d80 |
/* Support sysdeps/powerpc/powerpc64/multiarch/stpncpy.c. */
|
|
|
a35d80 |
IFUNC_IMPL (i, name, stpncpy,
|
|
|
a35d80 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
a35d80 |
+ IFUNC_IMPL_ADD (array, i, stpncpy,
|
|
|
a35d80 |
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00)
|
|
|
a35d80 |
+ && (hwcap & PPC_FEATURE_HAS_VSX),
|
|
|
a35d80 |
+ __stpncpy_power9)
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
IFUNC_IMPL_ADD (array, i, stpncpy,
|
|
|
a35d80 |
hwcap2 & PPC_FEATURE2_ARCH_2_07,
|
|
|
a35d80 |
__stpncpy_power8)
|
|
|
a35d80 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpncpy-power9.S b/sysdeps/powerpc/powerpc64/multiarch/stpncpy-power9.S
|
|
|
a35d80 |
new file mode 100644
|
|
|
a35d80 |
index 0000000000..1188bd0894
|
|
|
a35d80 |
--- /dev/null
|
|
|
a35d80 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpncpy-power9.S
|
|
|
a35d80 |
@@ -0,0 +1,29 @@
|
|
|
a35d80 |
+/* Optimized stpncpy implementation for POWER9 LE.
|
|
|
a35d80 |
+ Copyright (C) 2020 Free Software Foundation, Inc.
|
|
|
a35d80 |
+ This file is part of the GNU C Library.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
a35d80 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
a35d80 |
+ License as published by the Free Software Foundation; either
|
|
|
a35d80 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
a35d80 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
a35d80 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
a35d80 |
+ Lesser General Public License for more details.
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
a35d80 |
+ License along with the GNU C Library; if not, see
|
|
|
a35d80 |
+ <https://www.gnu.org/licenses/>. */
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+#define STPNCPY __stpncpy_power9
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+#undef libc_hidden_builtin_def
|
|
|
a35d80 |
+#define libc_hidden_builtin_def(name)
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+#define MEMSET __memset_power8
|
|
|
a35d80 |
+#ifdef SHARED
|
|
|
a35d80 |
+# define MEMSET_is_local
|
|
|
a35d80 |
+#endif
|
|
|
a35d80 |
+
|
|
|
a35d80 |
+#include <sysdeps/powerpc/powerpc64/le/power9/stpncpy.S>
|
|
|
a35d80 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c b/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
|
|
|
a35d80 |
index 17df886431..3758f29ad1 100644
|
|
|
a35d80 |
--- a/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
|
|
|
a35d80 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
|
|
|
a35d80 |
@@ -26,10 +26,18 @@
|
|
|
a35d80 |
extern __typeof (__stpncpy) __stpncpy_ppc attribute_hidden;
|
|
|
a35d80 |
extern __typeof (__stpncpy) __stpncpy_power7 attribute_hidden;
|
|
|
a35d80 |
extern __typeof (__stpncpy) __stpncpy_power8 attribute_hidden;
|
|
|
a35d80 |
+# ifdef __LITTLE_ENDIAN__
|
|
|
a35d80 |
+extern __typeof (__stpncpy) __stpncpy_power9 attribute_hidden;
|
|
|
a35d80 |
+# endif
|
|
|
a35d80 |
# undef stpncpy
|
|
|
a35d80 |
# undef __stpncpy
|
|
|
a35d80 |
|
|
|
a35d80 |
libc_ifunc_redirected (__redirect___stpncpy, __stpncpy,
|
|
|
a35d80 |
+# ifdef __LITTLE_ENDIAN__
|
|
|
a35d80 |
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00) &&
|
|
|
a35d80 |
+ (hwcap & PPC_FEATURE_HAS_VSX)
|
|
|
a35d80 |
+ ? __stpncpy_power9 :
|
|
|
a35d80 |
+# endif
|
|
|
a35d80 |
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
|
|
|
a35d80 |
? __stpncpy_power8
|
|
|
a35d80 |
: (hwcap & PPC_FEATURE_HAS_VSX)
|