|
 |
00db10 |
Backport of
|
|
|
00db10 |
commit d3b00f468bec441596877a685a19f43dee88657f
|
|
|
00db10 |
Author: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Fri Jan 9 16:04:26 2015 -0500
|
|
|
00db10 |
|
|
|
00db10 |
powerpc: Optimized strncmp for POWER8/PPC64
|
|
|
00db10 |
|
|
|
00db10 |
This patch adds an optimized POWER8 strncmp. The implementation focus
|
|
|
00db10 |
on speeding up unaligned cases follwing the ideas of power8 strcmp.
|
|
|
00db10 |
|
|
|
00db10 |
The algorithm first check the initial 16 bytes, then align the first
|
|
|
00db10 |
function source and uses unaligned loads on second argument only.
|
|
|
00db10 |
Aditional checks for page boundaries are done for unaligned cases
|
|
|
00db10 |
(where sources alignment are different).
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
2015-01-13 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/strncmp-power8.S: New file.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/power8/strncmp.S: New file.
|
|
|
00db10 |
* sysdeps/powerpc/powerpc64/multiarch/Makefile [sysdep_routines]: Add
|
|
|
00db10 |
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/Makefile b/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
00db10 |
index 27c8b65..677d8ce 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/Makefile
|
|
|
00db10 |
@@ -14,8 +14,9 @@ sysdep_routines += memcpy-power7 memcpy-a2 memcpy-power6 memcpy-cell \
|
|
|
00db10 |
strncat-power7 \
|
|
|
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 |
- strncmp-power4 strncmp-ppc64 strchr-power7 strchr-ppc64 \
|
|
|
00db10 |
+ strncase-power7 strncase_l-power7 \
|
|
|
00db10 |
+ strncmp-power8 strncmp-power7 strncmp-power4 strncmp-ppc64 \
|
|
|
00db10 |
+ strchr-power7 strchr-ppc64 \
|
|
|
00db10 |
strchrnul-power7 strchrnul-ppc64 wcschr-power7 \
|
|
|
00db10 |
wcschr-power6 wcschr-ppc64 wcsrchr-power7 wcsrchr-power6 \
|
|
|
00db10 |
wcsrchr-ppc64 wcscpy-power7 wcscpy-power6 wcscpy-ppc64 \
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
index 2b38c71..a540abf 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
@@ -100,6 +100,8 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
00db10 |
|
|
|
00db10 |
/* Support sysdeps/powerpc/powerpc64/multiarch/strncmp.c. */
|
|
|
00db10 |
IFUNC_IMPL (i, name, strncmp,
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, strncmp, hwcap2 & PPC_FEATURE2_ARCH_2_07,
|
|
|
00db10 |
+ __strncmp_power8)
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, strncmp, hwcap & PPC_FEATURE_HAS_VSX,
|
|
|
00db10 |
__strncmp_power7)
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, strncmp, hwcap & PPC_FEATURE_POWER4,
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp-power8.S b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power8.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..8d7223d
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp-power8.S
|
|
|
00db10 |
@@ -0,0 +1,40 @@
|
|
|
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(__strncmp_power8) \
|
|
|
00db10 |
+ .align ALIGNARG(alignt); \
|
|
|
00db10 |
+ EALIGN_W_##words; \
|
|
|
00db10 |
+ BODY_LABEL(__strncmp_power8): \
|
|
|
00db10 |
+ cfi_startproc; \
|
|
|
00db10 |
+ LOCALENTRY(__strncmp_power8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef END
|
|
|
00db10 |
+#define END(name) \
|
|
|
00db10 |
+ cfi_endproc; \
|
|
|
00db10 |
+ TRACEBACK(__strncmp_power8) \
|
|
|
00db10 |
+ END_2(__strncmp_power8)
|
|
|
00db10 |
+
|
|
|
00db10 |
+
|
|
|
00db10 |
+#undef libc_hidden_builtin_def
|
|
|
00db10 |
+#define libc_hidden_builtin_def(name)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdeps/powerpc/powerpc64/power8/strncmp.S>
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
|
|
|
00db10 |
index 9829d69..5e76783 100644
|
|
|
00db10 |
--- a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
|
|
|
00db10 |
@@ -25,13 +25,16 @@
|
|
|
00db10 |
extern __typeof (strncmp) __strncmp_ppc attribute_hidden;
|
|
|
00db10 |
extern __typeof (strncmp) __strncmp_power4 attribute_hidden;
|
|
|
00db10 |
extern __typeof (strncmp) __strncmp_power7 attribute_hidden;
|
|
|
00db10 |
+extern __typeof (strncmp) __strncmp_power8 attribute_hidden;
|
|
|
00db10 |
|
|
|
00db10 |
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
|
|
|
00db10 |
ifunc symbol properly. */
|
|
|
00db10 |
libc_ifunc (strncmp,
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_HAS_VSX)
|
|
|
00db10 |
- ? __strncmp_power7 :
|
|
|
00db10 |
- (hwcap & PPC_FEATURE_POWER4)
|
|
|
00db10 |
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07)
|
|
|
00db10 |
+ ? __strncmp_power8 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_HAS_VSX)
|
|
|
00db10 |
+ ? __strncmp_power7 :
|
|
|
00db10 |
+ (hwcap & PPC_FEATURE_POWER4)
|
|
|
00db10 |
? __strncmp_power4
|
|
|
00db10 |
: __strncmp_ppc);
|
|
|
00db10 |
#endif
|
|
|
00db10 |
diff --git a/sysdeps/powerpc/powerpc64/power8/strncmp.S b/sysdeps/powerpc/powerpc64/power8/strncmp.S
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..56c814b
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/powerpc/powerpc64/power8/strncmp.S
|
|
|
00db10 |
@@ -0,0 +1,323 @@
|
|
|
00db10 |
+/* Optimized strncmp 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 |
+/* Implements the function
|
|
|
00db10 |
+
|
|
|
00db10 |
+ int [r3] strncmp (const char *s1 [r3], const char *s2 [r4], size_t [r5] n)
|
|
|
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 (strncmp, 4, 0)
|
|
|
00db10 |
+ /* Check if size is 0. */
|
|
|
00db10 |
+ mr. r10,r5
|
|
|
00db10 |
+ beq cr0,L(ret0)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Check if [s1]+16 or [s2]+16 will cross a 4K page boundary using
|
|
|
00db10 |
+ the code:
|
|
|
00db10 |
+
|
|
|
00db10 |
+ (((size_t) s1) % PAGE_SIZE > (PAGE_SIZE - ITER_SIZE))
|
|
|
00db10 |
+
|
|
|
00db10 |
+ with PAGE_SIZE being 4096 and ITER_SIZE begin 16. */
|
|
|
00db10 |
+ rldicl r8,r3,0,52
|
|
|
00db10 |
+ cmpldi cr7,r8,4096-16
|
|
|
00db10 |
+ bgt cr7,L(pagecross)
|
|
|
00db10 |
+ rldicl r9,r4,0,52
|
|
|
00db10 |
+ cmpldi cr7,r9,4096-16
|
|
|
00db10 |
+ bgt cr7,L(pagecross)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* For short string up to 16 bytes, load both s1 and s2 using
|
|
|
00db10 |
+ unaligned dwords and compare. */
|
|
|
00db10 |
+ ld r7,0(r3)
|
|
|
00db10 |
+ ld r9,0(r4)
|
|
|
00db10 |
+ li r8,0
|
|
|
00db10 |
+ cmpb r8,r7,r8
|
|
|
00db10 |
+ cmpb r6,r7,r9
|
|
|
00db10 |
+ orc. r8,r8,r6
|
|
|
00db10 |
+ bne cr0,L(different1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* If the string compared are equal, but size is less or equal
|
|
|
00db10 |
+ to 8, return 0. */
|
|
|
00db10 |
+ cmpldi cr7,r10,8
|
|
|
00db10 |
+ li r9,0
|
|
|
00db10 |
+ ble cr7,L(ret1)
|
|
|
00db10 |
+ addi r5,r10,-8
|
|
|
00db10 |
+
|
|
|
00db10 |
+ ld r7,8(r3)
|
|
|
00db10 |
+ ld r9,8(r4)
|
|
|
00db10 |
+ cmpb r8,r7,r8
|
|
|
00db10 |
+ cmpb r6,r7,r9
|
|
|
00db10 |
+ orc. r8,r8,r6
|
|
|
00db10 |
+ bne cr0,L(different0)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ cmpldi cr7,r5,8
|
|
|
00db10 |
+ mr r9,r8
|
|
|
00db10 |
+ ble cr7,L(ret1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Update pointers and size. */
|
|
|
00db10 |
+ addi r10,r10,-16
|
|
|
00db10 |
+ addi r3,r3,16
|
|
|
00db10 |
+ addi r4,r4,16
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Now it has checked for first 16 bytes, align source1 to doubleword
|
|
|
00db10 |
+ and adjust source2 address. */
|
|
|
00db10 |
+L(align_8b):
|
|
|
00db10 |
+ rldicl r5,r3,0,61
|
|
|
00db10 |
+ rldicr r3,r3,0,60
|
|
|
00db10 |
+ subf r4,r5,r4
|
|
|
00db10 |
+ add r10,r10,r5
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* At this point, source1 alignment is 0 and source2 alignment is
|
|
|
00db10 |
+ between 0 and 7. Check is source2 alignment is 0, meaning both
|
|
|
00db10 |
+ sources have the same alignment. */
|
|
|
00db10 |
+ andi. r8,r4,0x7
|
|
|
00db10 |
+ beq cr0,L(loop_eq_align_0)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ li r5,0
|
|
|
00db10 |
+ b L(loop_ne_align_1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* If source2 is unaligned to doubleword, the code needs to check
|
|
|
00db10 |
+ on each interation if the unaligned doubleword access will cross
|
|
|
00db10 |
+ a 4k page boundary. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(loop_ne_align_0):
|
|
|
00db10 |
+ ld r7,0(r3)
|
|
|
00db10 |
+ ld r9,0(r4)
|
|
|
00db10 |
+ cmpb r8,r7,r5
|
|
|
00db10 |
+ cmpb r6,r7,r9
|
|
|
00db10 |
+ orc. r8,r8,r6
|
|
|
00db10 |
+ bne cr0,L(different1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ cmpldi cr7,r10,8
|
|
|
00db10 |
+ ble cr7,L(ret0)
|
|
|
00db10 |
+ addi r10,r10,-8
|
|
|
00db10 |
+ addi r3,r3,8
|
|
|
00db10 |
+ addi r4,r4,8
|
|
|
00db10 |
+L(loop_ne_align_1):
|
|
|
00db10 |
+ rldicl r9,r4,0,52
|
|
|
00db10 |
+ cmpldi r7,r9,4088
|
|
|
00db10 |
+ ble cr7,L(loop_ne_align_0)
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ beq cr7,L(ret0)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ lbz r9,0(r3)
|
|
|
00db10 |
+ lbz r8,0(r4)
|
|
|
00db10 |
+ cmplw cr7,r9,r8
|
|
|
00db10 |
+ bne cr7,L(byte_ne_4)
|
|
|
00db10 |
+ cmpdi cr7,r9,0
|
|
|
00db10 |
+ beq cr7,L(size_reached_0)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ li r9,r7
|
|
|
00db10 |
+ addi r8,r3,1
|
|
|
00db10 |
+ mtctr r9
|
|
|
00db10 |
+ addi r4,r4,1
|
|
|
00db10 |
+ addi r10,r10,-1
|
|
|
00db10 |
+ addi r3,r3,8
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* The unaligned read of source2 will cross a 4K page boundary,
|
|
|
00db10 |
+ and the different byte or NULL maybe be in the remaining page
|
|
|
00db10 |
+ bytes. Since it can not use the unaligned load the algorithm
|
|
|
00db10 |
+ reads and compares 8 bytes to keep source1 doubleword aligned. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(loop_ne_align_byte):
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ addi r10,r10,-1
|
|
|
00db10 |
+ beq cr7,L(ret0)
|
|
|
00db10 |
+ lbz r9,0(r8)
|
|
|
00db10 |
+ lbz r7,0(r4)
|
|
|
00db10 |
+ addi r8,r8,1
|
|
|
00db10 |
+ addi r4,r4,1
|
|
|
00db10 |
+ cmplw cr7,r9,r7
|
|
|
00db10 |
+ cmpdi cr5,r9,0
|
|
|
00db10 |
+ bne cr7,L(size_reached_2)
|
|
|
00db10 |
+ beq cr5,L(size_reached_0)
|
|
|
00db10 |
+ bdnz L(loop_ne_align_byte)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ bne+ cr7,L(loop_ne_align_0)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(ret0):
|
|
|
00db10 |
+ li r9,0
|
|
|
00db10 |
+L(ret1):
|
|
|
00db10 |
+ mr r3,r9
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* The code now check if r8 and r10 are different by issuing a
|
|
|
00db10 |
+ cmpb and shift the result based on its output:
|
|
|
00db10 |
+
|
|
|
00db10 |
+ #ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ leadzero = (__builtin_ffsl (z1) - 1);
|
|
|
00db10 |
+ leadzero = leadzero > (n-1)*8 ? (n-1)*8 : leadzero;
|
|
|
00db10 |
+ r1 = (r1 >> leadzero) & 0xFFUL;
|
|
|
00db10 |
+ r2 = (r2 >> leadzero) & 0xFFUL;
|
|
|
00db10 |
+ #else
|
|
|
00db10 |
+ leadzero = __builtin_clzl (z1);
|
|
|
00db10 |
+ leadzero = leadzero > (n-1)*8 ? (n-1)*8 : leadzero;
|
|
|
00db10 |
+ r1 = (r1 >> (56 - leadzero)) & 0xFFUL;
|
|
|
00db10 |
+ r2 = (r2 >> (56 - leadzero)) & 0xFFUL;
|
|
|
00db10 |
+ #endif
|
|
|
00db10 |
+ return r1 - r2; */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(different0):
|
|
|
00db10 |
+ mr r10,r5
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+L(different1):
|
|
|
00db10 |
+ neg r11,r8
|
|
|
00db10 |
+ sldi r10,r10,3
|
|
|
00db10 |
+ and r8,r11,r8
|
|
|
00db10 |
+ addi r10,r10,-8
|
|
|
00db10 |
+ cntlzd r8,r8
|
|
|
00db10 |
+ subfic r8,r8,63
|
|
|
00db10 |
+ extsw r8,r8
|
|
|
00db10 |
+ cmpld cr7,r8,r10
|
|
|
00db10 |
+ ble cr7,L(different2)
|
|
|
00db10 |
+ mr r8,r10
|
|
|
00db10 |
+L(different2):
|
|
|
00db10 |
+ extsw r8,r8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+L(different1):
|
|
|
00db10 |
+ addi r10,r10,-1
|
|
|
00db10 |
+ cntlzd r8,r8
|
|
|
00db10 |
+ sldi r10,r10,3
|
|
|
00db10 |
+ cmpld cr7,r8,r10
|
|
|
00db10 |
+ blt cr7,L(different2)
|
|
|
00db10 |
+ mr r8,r10
|
|
|
00db10 |
+L(different2):
|
|
|
00db10 |
+ subfic r8,r8,56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ srd r7,r7,r8
|
|
|
00db10 |
+ srd r9,r9,r8
|
|
|
00db10 |
+ rldicl r3,r7,0,56
|
|
|
00db10 |
+ rldicl r9,r9,0,56
|
|
|
00db10 |
+ subf r9,r9,3
|
|
|
00db10 |
+ extsw r9,r9
|
|
|
00db10 |
+ mr r3,r9
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* If unaligned 16 bytes reads across a 4K page boundary, it uses
|
|
|
00db10 |
+ a simple byte a byte comparison until the page alignment for s1
|
|
|
00db10 |
+ is reached. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(pagecross):
|
|
|
00db10 |
+ lbz r7,0(r3)
|
|
|
00db10 |
+ lbz r9,0(r4)
|
|
|
00db10 |
+ subfic r8,r8,4095
|
|
|
00db10 |
+ cmplw cr7,r9,r7
|
|
|
00db10 |
+ bne cr7,L(byte_ne_3)
|
|
|
00db10 |
+ cmpdi cr7,r9,0
|
|
|
00db10 |
+ beq cr7,L(byte_ne_0)
|
|
|
00db10 |
+ addi r10,r10,-1
|
|
|
00db10 |
+ subf r7,r8,r10
|
|
|
00db10 |
+ subf r9,r7,r10
|
|
|
00db10 |
+ addi r9,r9,1
|
|
|
00db10 |
+ mtctr r9
|
|
|
00db10 |
+ b L(pagecross_loop1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(pagecross_loop0):
|
|
|
00db10 |
+ beq cr7,L(ret0)
|
|
|
00db10 |
+ lbz r9,0(r3)
|
|
|
00db10 |
+ lbz r8,0(r4)
|
|
|
00db10 |
+ addi r10,r10,-1
|
|
|
00db10 |
+ cmplw cr7,r9,r8
|
|
|
00db10 |
+ cmpdi cr5,r9,0
|
|
|
00db10 |
+ bne r7,L(byte_ne_2)
|
|
|
00db10 |
+ beq r5,L(byte_ne_0)
|
|
|
00db10 |
+L(pagecross_loop1):
|
|
|
00db10 |
+ cmpdi cr7,r10,0
|
|
|
00db10 |
+ addi r3,r3,1
|
|
|
00db10 |
+ addi r4,r4,1
|
|
|
00db10 |
+ bdnz L(pagecross_loop0)
|
|
|
00db10 |
+ cmpdi cr7,r7,0
|
|
|
00db10 |
+ li r9,0
|
|
|
00db10 |
+ bne+ cr7,L(align_8b)
|
|
|
00db10 |
+ b L(ret1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* If both source1 and source2 are doubleword aligned, there is no
|
|
|
00db10 |
+ need for page boundary cross checks. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(loop_eq_align_0):
|
|
|
00db10 |
+ ld r7,0(r3)
|
|
|
00db10 |
+ ld r9,0(r4)
|
|
|
00db10 |
+ cmpb r8,r7,r8
|
|
|
00db10 |
+ cmpb r6,r7,r9
|
|
|
00db10 |
+ orc. r8,r8,r6
|
|
|
00db10 |
+ bne cr0,L(different1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ cmpldi cr7,r10,8
|
|
|
00db10 |
+ ble cr7,L(ret0)
|
|
|
00db10 |
+ addi r9,r10,-9
|
|
|
00db10 |
+
|
|
|
00db10 |
+ li r5,0
|
|
|
00db10 |
+ srdi r9,r9,3
|
|
|
00db10 |
+ addi r9,r9,1
|
|
|
00db10 |
+ mtctr r9
|
|
|
00db10 |
+ b L(loop_eq_align_2)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(loop_eq_align_1):
|
|
|
00db10 |
+ bdz L(ret0)
|
|
|
00db10 |
+L(loop_eq_align_2):
|
|
|
00db10 |
+ ldu r7,8(r3)
|
|
|
00db10 |
+ addi r10,r10,-8
|
|
|
00db10 |
+ ldu r9,8(r4)
|
|
|
00db10 |
+ cmpb r8,r7,r5
|
|
|
00db10 |
+ cmpb r6,r7,r9
|
|
|
00db10 |
+ orc. r8,r8,r6
|
|
|
00db10 |
+ beq cr0,L(loop_eq_align_1)
|
|
|
00db10 |
+ b L(different1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(byte_ne_0):
|
|
|
00db10 |
+ li r7,0
|
|
|
00db10 |
+L(byte_ne_1):
|
|
|
00db10 |
+ subf r9,r9,r7
|
|
|
00db10 |
+ extsw r9,r9
|
|
|
00db10 |
+ b L(ret1)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(byte_ne_2):
|
|
|
00db10 |
+ extsw r7,r9
|
|
|
00db10 |
+ mr r9,r8
|
|
|
00db10 |
+ b L(byte_ne_1)
|
|
|
00db10 |
+L(size_reached_0):
|
|
|
00db10 |
+ li r10,0
|
|
|
00db10 |
+L(size_reached_1):
|
|
|
00db10 |
+ subf r9,r9,r10
|
|
|
00db10 |
+ extsw r9,r9
|
|
|
00db10 |
+ b L(ret1)
|
|
|
00db10 |
+L(size_reached_2):
|
|
|
00db10 |
+ extsw r10,r9
|
|
|
00db10 |
+ mr r9,r7
|
|
|
00db10 |
+ b L(size_reached_1)
|
|
|
00db10 |
+L(byte_ne_3):
|
|
|
00db10 |
+ extsw r7,r7
|
|
|
00db10 |
+ b L(byte_ne_1)
|
|
|
00db10 |
+L(byte_ne_4):
|
|
|
00db10 |
+ extsw r10,r9
|
|
|
00db10 |
+ mr r9,r8
|
|
|
00db10 |
+ b L(size_reached_1)
|
|
|
00db10 |
+END(strncmp)
|
|
|
00db10 |
+libc_hidden_builtin_def(strncmp)
|