glibc/sysdeps/x86_64/wcslen.S
Noah Goldstein 4ff6ae069b x86: Small improvements for wcslen
Just a few QOL changes.
    1. Prefer `add` > `lea` as it has high execution units it can run
       on.
    2. Don't break macro-fusion between `test` and `jcc`
    3. Reduce code size by removing gratuitous padding bytes (-90
       bytes).

geometric_mean(N=20) of all benchmarks New / Original: 0.959

All string/memory tests pass.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>

(cherry picked from commit 244b415d38)
2022-05-16 18:55:09 -07:00

235 lines
3.8 KiB
ArmAsm

/* Optimized wcslen for x86-64 with SSE2.
Copyright (C) 2011-2021 Free Software Foundation, Inc.
Contributed by Intel Corporation.
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>
.text
ENTRY (__wcslen)
cmpl $0, (%rdi)
jz L(exit_tail0)
cmpl $0, 4(%rdi)
jz L(exit_tail1)
cmpl $0, 8(%rdi)
jz L(exit_tail2)
cmpl $0, 12(%rdi)
jz L(exit_tail3)
cmpl $0, 16(%rdi)
jz L(exit_tail4)
cmpl $0, 20(%rdi)
jz L(exit_tail5)
cmpl $0, 24(%rdi)
jz L(exit_tail6)
cmpl $0, 28(%rdi)
jz L(exit_tail7)
pxor %xmm0, %xmm0
lea 32(%rdi), %rax
addq $16, %rdi
and $-16, %rax
pcmpeqd (%rax), %xmm0
pmovmskb %xmm0, %edx
pxor %xmm1, %xmm1
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm1
pmovmskb %xmm1, %edx
pxor %xmm2, %xmm2
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm2
pmovmskb %xmm2, %edx
pxor %xmm3, %xmm3
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm3
pmovmskb %xmm3, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm0
pmovmskb %xmm0, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm1
pmovmskb %xmm1, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm2
pmovmskb %xmm2, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm3
pmovmskb %xmm3, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm0
pmovmskb %xmm0, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm1
pmovmskb %xmm1, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm2
pmovmskb %xmm2, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
pcmpeqd (%rax), %xmm3
pmovmskb %xmm3, %edx
addq $16, %rax
test %edx, %edx
jnz L(exit)
and $-0x40, %rax
.p2align 4
L(aligned_64_loop):
movaps (%rax), %xmm0
movaps 16(%rax), %xmm1
movaps 32(%rax), %xmm2
movaps 48(%rax), %xmm6
pminub %xmm1, %xmm0
pminub %xmm6, %xmm2
pminub %xmm0, %xmm2
pcmpeqd %xmm3, %xmm2
pmovmskb %xmm2, %edx
addq $64, %rax
test %edx, %edx
jz L(aligned_64_loop)
pcmpeqd -64(%rax), %xmm3
pmovmskb %xmm3, %edx
addq $48, %rdi
test %edx, %edx
jnz L(exit)
pcmpeqd %xmm1, %xmm3
pmovmskb %xmm3, %edx
addq $-16, %rdi
test %edx, %edx
jnz L(exit)
pcmpeqd -32(%rax), %xmm3
pmovmskb %xmm3, %edx
addq $-16, %rdi
test %edx, %edx
jnz L(exit)
pcmpeqd %xmm6, %xmm3
pmovmskb %xmm3, %edx
addq $-16, %rdi
test %edx, %edx
jz L(aligned_64_loop)
.p2align 4
L(exit):
sub %rdi, %rax
shr $2, %rax
test %dl, %dl
jz L(exit_high)
andl $15, %edx
jz L(exit_1)
ret
/* No align here. Naturally aligned % 16 == 1. */
L(exit_high):
andl $(15 << 8), %edx
jz L(exit_3)
add $2, %rax
ret
.p2align 3
L(exit_1):
add $1, %rax
ret
.p2align 3
L(exit_3):
add $3, %rax
ret
.p2align 3
L(exit_tail0):
xorl %eax, %eax
ret
.p2align 3
L(exit_tail1):
movl $1, %eax
ret
.p2align 3
L(exit_tail2):
movl $2, %eax
ret
.p2align 3
L(exit_tail3):
movl $3, %eax
ret
.p2align 3
L(exit_tail4):
movl $4, %eax
ret
.p2align 3
L(exit_tail5):
movl $5, %eax
ret
.p2align 3
L(exit_tail6):
movl $6, %eax
ret
.p2align 3
L(exit_tail7):
movl $7, %eax
ret
END (__wcslen)
weak_alias(__wcslen, wcslen)