summary refs log tree commit diff
path: root/sysdeps/x86_64/multiarch/memset-avx512-no-vzeroupper.S
blob: c5be8f57ef41c6b8a296ee04026e3520dd66b025 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
/* memset optimized with AVX512 for KNL hardware.
   Copyright (C) 2015-2022 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>

#if IS_IN (libc)

#include "asm-syntax.h"
#ifndef MEMSET
# define MEMSET __memset_avx512_no_vzeroupper
# define MEMSET_CHK __memset_chk_avx512_no_vzeroupper
#endif

	.section .text.avx512,"ax",@progbits
#if defined PIC
ENTRY (MEMSET_CHK)
	cmp	%RDX_LP, %RCX_LP
	jb	HIDDEN_JUMPTARGET (__chk_fail)
END (MEMSET_CHK)
#endif

ENTRY (MEMSET)
# ifdef __ILP32__
	/* Clear the upper 32 bits.  */
	mov	%edx, %edx
# endif
	vpxor	%xmm0, %xmm0, %xmm0
	vmovd	%esi, %xmm1
	lea	(%rdi, %rdx), %rsi
	mov	%rdi, %rax
	vpshufb	%xmm0, %xmm1, %xmm0
	cmp	$16, %rdx
	jb	L(less_16bytes)
	cmp	$512, %rdx
	vbroadcastss	%xmm0, %zmm2
	ja	L(512bytesormore)
	cmp	$256, %rdx
	jb	L(less_256bytes)
	vmovups	%zmm2, (%rdi)
	vmovups %zmm2, 0x40(%rdi)
	vmovups %zmm2, 0x80(%rdi)
	vmovups %zmm2, 0xC0(%rdi)
	vmovups %zmm2, -0x100(%rsi)
	vmovups %zmm2, -0xC0(%rsi)
	vmovups %zmm2, -0x80(%rsi)
	vmovups %zmm2, -0x40(%rsi)
	ret

L(less_256bytes):
	cmp	$128, %dl
	jb	L(less_128bytes)
	vmovups	%zmm2, (%rdi)
	vmovups %zmm2, 0x40(%rdi)
	vmovups %zmm2, -0x80(%rsi)
	vmovups %zmm2, -0x40(%rsi)
	ret

L(less_128bytes):
	cmp	$64, %dl
	jb	L(less_64bytes)
	vmovups	%zmm2, (%rdi)
	vmovups	%zmm2, -0x40(%rsi)
	ret

L(less_64bytes):
	cmp	$32, %dl
	jb	L(less_32bytes)
	vmovdqu	%ymm2, (%rdi)
	vmovdqu %ymm2, -0x20(%rsi)
	ret

L(less_32bytes):
	vmovdqu %xmm0, (%rdi)
	vmovdqu %xmm0, -0x10(%rsi)
	ret

L(less_16bytes):
	cmp	$8, %dl
	jb	L(less_8bytes)
	vmovq	%xmm0, (%rdi)
	vmovq	%xmm0, -0x08(%rsi)
	ret

L(less_8bytes):
	vmovd	%xmm0, %ecx
	cmp	$4, %dl
	jb	L(less_4bytes)
	mov	%ecx, (%rdi)
	mov	%ecx, -0x04(%rsi)
	ret

L(less_4bytes):
	cmp	$2, %dl
	jb	L(less_2bytes)
	mov	%cx, (%rdi)
	mov	%cx, -0x02(%rsi)
	ret

L(less_2bytes):
	cmp	$1, %dl
	jb	L(less_1bytes)
	mov	%cl, (%rdi)
L(less_1bytes):
	ret

L(512bytesormore):
	mov	__x86_shared_cache_size_half(%rip), %rcx
	cmp	%rcx, %rdx
	ja	L(preloop_large)
	cmp	$1024, %rdx
	ja	L(1024bytesormore)

	vmovups	%zmm2, (%rdi)
	vmovups	%zmm2, 0x40(%rdi)
	vmovups	%zmm2, 0x80(%rdi)
	vmovups	%zmm2, 0xC0(%rdi)
	vmovups	%zmm2, 0x100(%rdi)
	vmovups	%zmm2, 0x140(%rdi)
	vmovups	%zmm2, 0x180(%rdi)
	vmovups	%zmm2, 0x1C0(%rdi)
	vmovups %zmm2, -0x200(%rsi)
	vmovups %zmm2, -0x1C0(%rsi)
	vmovups %zmm2, -0x180(%rsi)
	vmovups %zmm2, -0x140(%rsi)
	vmovups %zmm2, -0x100(%rsi)
	vmovups %zmm2, -0xC0(%rsi)
	vmovups %zmm2, -0x80(%rsi)
	vmovups %zmm2, -0x40(%rsi)
	ret

/* Align on 64 and loop with aligned stores.  */
L(1024bytesormore):
	sub	$0x100, %rsi
	vmovups	%zmm2, (%rax)
	and	$-0x40, %rdi
	add	$0x40, %rdi

L(gobble_256bytes_loop):
	vmovaps	%zmm2, (%rdi)
	vmovaps	%zmm2, 0x40(%rdi)
	vmovaps	%zmm2, 0x80(%rdi)
	vmovaps	%zmm2, 0xC0(%rdi)
	add	$0x100, %rdi
	cmp	%rsi, %rdi
	jb	L(gobble_256bytes_loop)
	vmovups %zmm2, (%rsi)
	vmovups %zmm2, 0x40(%rsi)
	vmovups %zmm2, 0x80(%rsi)
	vmovups %zmm2, 0xC0(%rsi)
	ret

/* Align on 128 and loop with non-temporal stores.  */
L(preloop_large):
	and	$-0x80, %rdi
	add	$0x80, %rdi
	vmovups	%zmm2, (%rax)
	vmovups	%zmm2, 0x40(%rax)
	sub	$0x200, %rsi

L(gobble_512bytes_nt_loop):
	vmovntdq %zmm2, (%rdi)
	vmovntdq %zmm2, 0x40(%rdi)
	vmovntdq %zmm2, 0x80(%rdi)
	vmovntdq %zmm2, 0xC0(%rdi)
	vmovntdq %zmm2, 0x100(%rdi)
	vmovntdq %zmm2, 0x140(%rdi)
	vmovntdq %zmm2, 0x180(%rdi)
	vmovntdq %zmm2, 0x1C0(%rdi)
	add	$0x200, %rdi
	cmp	%rsi, %rdi
	jb	L(gobble_512bytes_nt_loop)
	sfence
	vmovups %zmm2, (%rsi)
	vmovups %zmm2, 0x40(%rsi)
	vmovups %zmm2, 0x80(%rsi)
	vmovups %zmm2, 0xC0(%rsi)
	vmovups	%zmm2, 0x100(%rsi)
	vmovups	%zmm2, 0x140(%rsi)
	vmovups	%zmm2, 0x180(%rsi)
	vmovups	%zmm2, 0x1C0(%rsi)
	ret
END (MEMSET)
#endif