|
Packit |
6c4009 |
/* strchrnul - find a character or nul in a string
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
Copyright (C) 2014-2018 Free Software Foundation, Inc.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
This file is part of the GNU C Library.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The GNU C Library is free software; you can redistribute it and/or
|
|
Packit |
6c4009 |
modify it under the terms of the GNU Lesser General Public
|
|
Packit |
6c4009 |
License as published by the Free Software Foundation; either
|
|
Packit |
6c4009 |
version 2.1 of the License, or (at your option) any later version.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
The GNU C Library is distributed in the hope that it will be useful,
|
|
Packit |
6c4009 |
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Packit |
6c4009 |
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
Packit |
6c4009 |
Lesser General Public License for more details.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
You should have received a copy of the GNU Lesser General Public
|
|
Packit |
6c4009 |
License along with the GNU C Library. If not, see
|
|
Packit |
6c4009 |
<http://www.gnu.org/licenses/>. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#include <sysdep.h>
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Assumptions:
|
|
Packit |
6c4009 |
*
|
|
Packit |
6c4009 |
* ARMv8-a, AArch64
|
|
Packit |
6c4009 |
* Neon Available.
|
|
Packit |
6c4009 |
*/
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Arguments and results. */
|
|
Packit |
6c4009 |
#define srcin x0
|
|
Packit |
6c4009 |
#define chrin w1
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#define result x0
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Locals and temporaries. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#define src x2
|
|
Packit |
6c4009 |
#define tmp1 x3
|
|
Packit |
6c4009 |
#define wtmp2 w4
|
|
Packit |
6c4009 |
#define tmp3 x5
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
#define vrepchr v0
|
|
Packit |
6c4009 |
#define vdata1 v1
|
|
Packit |
6c4009 |
#define vdata2 v2
|
|
Packit |
6c4009 |
#define vhas_nul1 v3
|
|
Packit |
6c4009 |
#define vhas_nul2 v4
|
|
Packit |
6c4009 |
#define vhas_chr1 v5
|
|
Packit |
6c4009 |
#define vhas_chr2 v6
|
|
Packit |
6c4009 |
#define vrepmask v7
|
|
Packit |
6c4009 |
#define vend1 v16
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Core algorithm.
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
For each 32-byte hunk we calculate a 64-bit syndrome value, with
|
|
Packit |
6c4009 |
two bits per byte (LSB is always in bits 0 and 1, for both big
|
|
Packit |
6c4009 |
and little-endian systems). For each tuple, bit 0 is set iff
|
|
Packit |
6c4009 |
the relevant byte matched the requested character or nul. Since the
|
|
Packit |
6c4009 |
bits in the syndrome reflect exactly the order in which things occur
|
|
Packit |
6c4009 |
in the original string a count_trailing_zeros() operation will
|
|
Packit |
6c4009 |
identify exactly which byte is causing the termination. */
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
ENTRY (__strchrnul)
|
|
Packit |
6c4009 |
DELOUSE (0)
|
|
Packit |
6c4009 |
/* Magic constant 0x40100401 to allow us to identify which lane
|
|
Packit |
6c4009 |
matches the termination condition. */
|
|
Packit |
6c4009 |
mov wtmp2, #0x0401
|
|
Packit |
6c4009 |
movk wtmp2, #0x4010, lsl #16
|
|
Packit |
6c4009 |
dup vrepchr.16b, chrin
|
|
Packit |
6c4009 |
bic src, srcin, #31 /* Work with aligned 32-byte hunks. */
|
|
Packit |
6c4009 |
dup vrepmask.4s, wtmp2
|
|
Packit |
6c4009 |
ands tmp1, srcin, #31
|
|
Packit |
6c4009 |
b.eq L(loop)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Input string is not 32-byte aligned. Rather than forcing
|
|
Packit |
6c4009 |
the padding bytes to a safe value, we calculate the syndrome
|
|
Packit |
6c4009 |
for all the bytes, but then mask off those bits of the
|
|
Packit |
6c4009 |
syndrome that are related to the padding. */
|
|
Packit |
6c4009 |
ld1 {vdata1.16b, vdata2.16b}, [src], #32
|
|
Packit |
6c4009 |
neg tmp1, tmp1
|
|
Packit |
6c4009 |
cmeq vhas_nul1.16b, vdata1.16b, #0
|
|
Packit |
6c4009 |
cmeq vhas_chr1.16b, vdata1.16b, vrepchr.16b
|
|
Packit |
6c4009 |
cmeq vhas_nul2.16b, vdata2.16b, #0
|
|
Packit |
6c4009 |
cmeq vhas_chr2.16b, vdata2.16b, vrepchr.16b
|
|
Packit |
6c4009 |
orr vhas_chr1.16b, vhas_chr1.16b, vhas_nul1.16b
|
|
Packit |
6c4009 |
orr vhas_chr2.16b, vhas_chr2.16b, vhas_nul2.16b
|
|
Packit |
6c4009 |
and vhas_chr1.16b, vhas_chr1.16b, vrepmask.16b
|
|
Packit |
6c4009 |
and vhas_chr2.16b, vhas_chr2.16b, vrepmask.16b
|
|
Packit |
6c4009 |
lsl tmp1, tmp1, #1
|
|
Packit |
6c4009 |
addp vend1.16b, vhas_chr1.16b, vhas_chr2.16b // 256->128
|
|
Packit |
6c4009 |
mov tmp3, #~0
|
|
Packit |
6c4009 |
addp vend1.16b, vend1.16b, vend1.16b // 128->64
|
|
Packit |
6c4009 |
lsr tmp1, tmp3, tmp1
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
mov tmp3, vend1.2d[0]
|
|
Packit |
6c4009 |
bic tmp1, tmp3, tmp1 // Mask padding bits.
|
|
Packit |
6c4009 |
cbnz tmp1, L(tail)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
L(loop):
|
|
Packit |
6c4009 |
ld1 {vdata1.16b, vdata2.16b}, [src], #32
|
|
Packit |
6c4009 |
cmeq vhas_nul1.16b, vdata1.16b, #0
|
|
Packit |
6c4009 |
cmeq vhas_chr1.16b, vdata1.16b, vrepchr.16b
|
|
Packit |
6c4009 |
cmeq vhas_nul2.16b, vdata2.16b, #0
|
|
Packit |
6c4009 |
cmeq vhas_chr2.16b, vdata2.16b, vrepchr.16b
|
|
Packit |
6c4009 |
/* Use a fast check for the termination condition. */
|
|
Packit |
6c4009 |
orr vhas_chr1.16b, vhas_nul1.16b, vhas_chr1.16b
|
|
Packit |
6c4009 |
orr vhas_chr2.16b, vhas_nul2.16b, vhas_chr2.16b
|
|
Packit |
6c4009 |
orr vend1.16b, vhas_chr1.16b, vhas_chr2.16b
|
|
Packit |
6c4009 |
addp vend1.2d, vend1.2d, vend1.2d
|
|
Packit |
6c4009 |
mov tmp1, vend1.2d[0]
|
|
Packit |
6c4009 |
cbz tmp1, L(loop)
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
/* Termination condition found. Now need to establish exactly why
|
|
Packit |
6c4009 |
we terminated. */
|
|
Packit |
6c4009 |
and vhas_chr1.16b, vhas_chr1.16b, vrepmask.16b
|
|
Packit |
6c4009 |
and vhas_chr2.16b, vhas_chr2.16b, vrepmask.16b
|
|
Packit |
6c4009 |
addp vend1.16b, vhas_chr1.16b, vhas_chr2.16b // 256->128
|
|
Packit |
6c4009 |
addp vend1.16b, vend1.16b, vend1.16b // 128->64
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
mov tmp1, vend1.2d[0]
|
|
Packit |
6c4009 |
L(tail):
|
|
Packit |
6c4009 |
/* Count the trailing zeros, by bit reversing... */
|
|
Packit |
6c4009 |
rbit tmp1, tmp1
|
|
Packit |
6c4009 |
/* Re-bias source. */
|
|
Packit |
6c4009 |
sub src, src, #32
|
|
Packit |
6c4009 |
clz tmp1, tmp1 /* ... and counting the leading zeros. */
|
|
Packit |
6c4009 |
/* tmp1 is twice the offset into the fragment. */
|
|
Packit |
6c4009 |
add result, src, tmp1, lsr #1
|
|
Packit |
6c4009 |
ret
|
|
Packit |
6c4009 |
|
|
Packit |
6c4009 |
END(__strchrnul)
|
|
Packit |
6c4009 |
weak_alias (__strchrnul, strchrnul)
|