mirror of
https://sourceware.org/git/glibc.git
synced 2024-11-27 15:30:07 +00:00
33ee81de05
http://sourceware.org/ml/libc-alpha/2013-08/msg00098.html The existing strnlen code has a number of defects, so this patch is more than just adding little-endian support. The changes here are similar to those for memchr. * sysdeps/powerpc/powerpc64/power7/strnlen.S (strnlen): Add little-endian support. Remove unnecessary "are we done" tests. Handle "s" wrapping around zero and extremely large "size". Correct main loop count. Handle single left-over word from main loop inline rather than by using small_loop. Correct comments. Delete "zero" tail, use "end_max" instead. * sysdeps/powerpc/powerpc32/power7/strnlen.S: Likewise.
178 lines
4.7 KiB
ArmAsm
178 lines
4.7 KiB
ArmAsm
/* Optimized strnlen implementation for PowerPC64/POWER7 using cmpb insn.
|
|
Copyright (C) 2010-2013 Free Software Foundation, Inc.
|
|
Contributed by Luis Machado <luisgpm@br.ibm.com>.
|
|
This file is part of the GNU C Library.
|
|
|
|
The GNU C Library is free software; you can redistribute it and/or
|
|
modify it under the terms of the GNU Lesser General Public
|
|
License as published by the Free Software Foundation; either
|
|
version 2.1 of the License, or (at your option) any later version.
|
|
|
|
The GNU C Library is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Lesser General Public License for more details.
|
|
|
|
You should have received a copy of the GNU Lesser General Public
|
|
License along with the GNU C Library; if not, see
|
|
<http://www.gnu.org/licenses/>. */
|
|
|
|
#include <sysdep.h>
|
|
|
|
/* int [r3] strnlen (char *s [r3], int size [r4]) */
|
|
.machine power7
|
|
ENTRY (__strnlen)
|
|
CALL_MCOUNT 2
|
|
dcbt 0,r3
|
|
clrrdi r8,r3,3
|
|
add r7,r3,r4 /* Calculate the last acceptable address. */
|
|
cmpldi r4,32
|
|
li r0,0 /* Doubleword with null chars. */
|
|
addi r7,r7,-1
|
|
|
|
/* If we have less than 33 bytes to search, skip to a faster code. */
|
|
ble L(small_range)
|
|
|
|
rlwinm r6,r3,3,26,28 /* Calculate padding. */
|
|
ld r12,0(r8) /* Load doubleword from memory. */
|
|
cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */
|
|
#ifdef __LITTLE_ENDIAN__
|
|
srd r10,r10,r6
|
|
sld r10,r10,r6
|
|
#else
|
|
sld r10,r10,r6
|
|
srd r10,r10,r6
|
|
#endif
|
|
cmpldi cr7,r10,0 /* If r10 == 0, no null's have been found. */
|
|
bne cr7,L(done)
|
|
|
|
clrrdi r7,r7,3 /* Address of last doubleword. */
|
|
mtcrf 0x01,r8
|
|
/* Are we now aligned to a quadword boundary? If so, skip to
|
|
the main loop. Otherwise, go through the alignment code. */
|
|
|
|
bt 28,L(loop_setup)
|
|
|
|
/* Handle DWORD2 of pair. */
|
|
ldu r12,8(r8)
|
|
cmpb r10,r12,r0
|
|
cmpldi cr7,r10,0
|
|
bne cr7,L(done)
|
|
|
|
L(loop_setup):
|
|
/* The last dword we want to read in the loop below is the one
|
|
containing the last byte of the string, ie. the dword at
|
|
(s + size - 1) & ~7, or r7. The first dword read is at
|
|
r8 + 8, we read 2 * cnt dwords, so the last dword read will
|
|
be at r8 + 8 + 16 * cnt - 8. Solving for cnt gives
|
|
cnt = (r7 - r8) / 16 */
|
|
sub r5,r7,r8
|
|
srdi r6,r5,4 /* Number of loop iterations. */
|
|
mtctr r6 /* Setup the counter. */
|
|
|
|
/* Main loop to look for the null byte in the string. Since
|
|
it's a small loop (< 8 instructions), align it to 32-bytes. */
|
|
.p2align 5
|
|
L(loop):
|
|
/* Load two doublewords, compare and merge in a
|
|
single register for speed. This is an attempt
|
|
to speed up the null-checking process for bigger strings. */
|
|
|
|
ld r12,8(r8)
|
|
ldu r11,16(r8)
|
|
cmpb r10,r12,r0
|
|
cmpb r9,r11,r0
|
|
or r5,r9,r10 /* Merge everything in one doubleword. */
|
|
cmpldi cr7,r5,0
|
|
bne cr7,L(found)
|
|
bdnz L(loop)
|
|
|
|
/* We may have one more dword to read. */
|
|
cmpld cr6,r8,r7
|
|
beq cr6,L(end_max)
|
|
|
|
ldu r12,8(r8)
|
|
cmpb r10,r12,r0
|
|
cmpldi cr6,r10,0
|
|
bne cr6,L(done)
|
|
|
|
L(end_max):
|
|
mr r3,r4
|
|
blr
|
|
|
|
/* OK, one (or both) of the doublewords contains a null byte. Check
|
|
the first doubleword and decrement the address in case the first
|
|
doubleword really contains a null byte. */
|
|
.align 4
|
|
L(found):
|
|
cmpldi cr6,r10,0
|
|
addi r8,r8,-8
|
|
bne cr6,L(done)
|
|
|
|
/* The null byte must be in the second doubleword. Adjust the address
|
|
again and move the result of cmpb to r10 so we can calculate the
|
|
length. */
|
|
|
|
mr r10,r9
|
|
addi r8,r8,8
|
|
|
|
/* r10 has the output of the cmpb instruction, that is, it contains
|
|
0xff in the same position as the null byte in the original
|
|
doubleword from the string. Use that to calculate the length.
|
|
We need to make sure the null char is *before* the end of the
|
|
range. */
|
|
L(done):
|
|
#ifdef __LITTLE_ENDIAN__
|
|
addi r0,r10,-1
|
|
andc r0,r0,r10
|
|
popcntd r0,r0
|
|
#else
|
|
cntlzd r0,r10 /* Count leading zeros before the match. */
|
|
#endif
|
|
sub r3,r8,r3
|
|
srdi r0,r0,3 /* Convert leading/trailing zeros to bytes. */
|
|
add r3,r3,r0 /* Length until the match. */
|
|
cmpld r3,r4
|
|
blelr
|
|
mr r3,r4
|
|
blr
|
|
|
|
/* Deals with size <= 32. */
|
|
.align 4
|
|
L(small_range):
|
|
cmpldi r4,0
|
|
beq L(end_max)
|
|
|
|
clrrdi r7,r7,3 /* Address of last doubleword. */
|
|
|
|
rlwinm r6,r3,3,26,28 /* Calculate padding. */
|
|
ld r12,0(r8) /* Load doubleword from memory. */
|
|
cmpb r10,r12,r0 /* Check for null bytes in DWORD1. */
|
|
#ifdef __LITTLE_ENDIAN__
|
|
srd r10,r10,r6
|
|
sld r10,r10,r6
|
|
#else
|
|
sld r10,r10,r6
|
|
srd r10,r10,r6
|
|
#endif
|
|
cmpldi cr7,r10,0
|
|
bne cr7,L(done)
|
|
|
|
cmpld r8,r7
|
|
beq L(end_max)
|
|
|
|
.p2align 5
|
|
L(loop_small):
|
|
ldu r12,8(r8)
|
|
cmpb r10,r12,r0
|
|
cmpldi cr6,r10,0
|
|
bne cr6,L(done)
|
|
cmpld r8,r7
|
|
bne L(loop_small)
|
|
mr r3,r4
|
|
blr
|
|
|
|
END (__strnlen)
|
|
weak_alias (__strnlen, strnlen)
|
|
libc_hidden_builtin_def (strnlen)
|