glibc/sysdeps/aarch64/strlen.S
Wilco Dijkstra 3dc426b642 AArch64: Improve generic strlen
Improve performance by handling another 16 bytes before entering the loop.
Use ADDHN in the loop to avoid SHRN+FMOV when it terminates.  Change final
size computation to avoid increasing latency.  On Neoverse V1 performance
of the random strlen benchmark improves by 4.6%.

Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
2024-08-07 14:58:46 +01:00

109 lines
2.5 KiB
ArmAsm

/* Generic optimized strlen using SIMD.
Copyright (C) 2012-2024 Free Software Foundation, Inc.
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
<https://www.gnu.org/licenses/>. */
#include <sysdep.h>
/* Assumptions:
*
* ARMv8-a, AArch64, Advanced SIMD.
* MTE compatible.
*/
#ifndef STRLEN
# define STRLEN __strlen
#endif
#define srcin x0
#define result x0
#define src x1
#define synd x2
#define tmp x3
#define shift x4
#define data q0
#define vdata v0
#define vhas_nul v1
#define vend v2
#define dend d2
/* Core algorithm:
Process the string in 16-byte aligned chunks. Compute a 64-bit mask with
four bits per byte using the shrn instruction. A count trailing zeros then
identifies the first zero byte. */
ENTRY (STRLEN)
PTR_ARG (0)
bic src, srcin, 15
ld1 {vdata.16b}, [src]
cmeq vhas_nul.16b, vdata.16b, 0
lsl shift, srcin, 2
shrn vend.8b, vhas_nul.8h, 4 /* 128->64 */
fmov synd, dend
lsr synd, synd, shift
cbz synd, L(next16)
rbit synd, synd
clz result, synd
lsr result, result, 2
ret
L(next16):
ldr data, [src, 16]
cmeq vhas_nul.16b, vdata.16b, 0
shrn vend.8b, vhas_nul.8h, 4 /* 128->64 */
fmov synd, dend
cbz synd, L(loop)
add src, src, 16
#ifndef __AARCH64EB__
rbit synd, synd
#endif
sub result, src, srcin
clz tmp, synd
add result, result, tmp, lsr 2
ret
.p2align 5
L(loop):
ldr data, [src, 32]!
cmeq vhas_nul.16b, vdata.16b, 0
addhn vend.8b, vhas_nul.8h, vhas_nul.8h
fmov synd, dend
cbnz synd, L(loop_end)
ldr data, [src, 16]
cmeq vhas_nul.16b, vdata.16b, 0
addhn vend.8b, vhas_nul.8h, vhas_nul.8h
fmov synd, dend
cbz synd, L(loop)
add src, src, 16
L(loop_end):
sub result, shift, src, lsl 2 /* (srcin - src) << 2. */
#ifndef __AARCH64EB__
rbit synd, synd
sub result, result, 3
#endif
clz tmp, synd
sub result, tmp, result
lsr result, result, 2
ret
END (STRLEN)
weak_alias (STRLEN, strlen)
libc_hidden_builtin_def (strlen)