1/* memset optimized with AVX512 for KNL hardware.
2 Copyright (C) 2015-2020 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <https://www.gnu.org/licenses/>. */
18
19#include <sysdep.h>
20
21#if IS_IN (libc)
22
23#include "asm-syntax.h"
24#ifndef MEMSET
25# define MEMSET __memset_avx512_no_vzeroupper
26# define MEMSET_CHK __memset_chk_avx512_no_vzeroupper
27#endif
28
29 .section .text.avx512,"ax",@progbits
30#if defined PIC
31ENTRY (MEMSET_CHK)
32 cmp %RDX_LP, %RCX_LP
33 jb HIDDEN_JUMPTARGET (__chk_fail)
34END (MEMSET_CHK)
35#endif
36
37ENTRY (MEMSET)
38# ifdef __ILP32__
39 /* Clear the upper 32 bits. */
40 mov %edx, %edx
41# endif
42 vpxor %xmm0, %xmm0, %xmm0
43 vmovd %esi, %xmm1
44 lea (%rdi, %rdx), %rsi
45 mov %rdi, %rax
46 vpshufb %xmm0, %xmm1, %xmm0
47 cmp $16, %rdx
48 jb L(less_16bytes)
49 cmp $512, %rdx
50 vbroadcastss %xmm0, %zmm2
51 ja L(512bytesormore)
52 cmp $256, %rdx
53 jb L(less_256bytes)
54 vmovups %zmm2, (%rdi)
55 vmovups %zmm2, 0x40(%rdi)
56 vmovups %zmm2, 0x80(%rdi)
57 vmovups %zmm2, 0xC0(%rdi)
58 vmovups %zmm2, -0x100(%rsi)
59 vmovups %zmm2, -0xC0(%rsi)
60 vmovups %zmm2, -0x80(%rsi)
61 vmovups %zmm2, -0x40(%rsi)
62 ret
63
64L(less_256bytes):
65 cmp $128, %dl
66 jb L(less_128bytes)
67 vmovups %zmm2, (%rdi)
68 vmovups %zmm2, 0x40(%rdi)
69 vmovups %zmm2, -0x80(%rsi)
70 vmovups %zmm2, -0x40(%rsi)
71 ret
72
73L(less_128bytes):
74 cmp $64, %dl
75 jb L(less_64bytes)
76 vmovups %zmm2, (%rdi)
77 vmovups %zmm2, -0x40(%rsi)
78 ret
79
80L(less_64bytes):
81 cmp $32, %dl
82 jb L(less_32bytes)
83 vmovdqu %ymm2, (%rdi)
84 vmovdqu %ymm2, -0x20(%rsi)
85 ret
86
87L(less_32bytes):
88 vmovdqu %xmm0, (%rdi)
89 vmovdqu %xmm0, -0x10(%rsi)
90 ret
91
92L(less_16bytes):
93 cmp $8, %dl
94 jb L(less_8bytes)
95 vmovq %xmm0, (%rdi)
96 vmovq %xmm0, -0x08(%rsi)
97 ret
98
99L(less_8bytes):
100 vmovd %xmm0, %ecx
101 cmp $4, %dl
102 jb L(less_4bytes)
103 mov %ecx, (%rdi)
104 mov %ecx, -0x04(%rsi)
105 ret
106
107L(less_4bytes):
108 cmp $2, %dl
109 jb L(less_2bytes)
110 mov %cx, (%rdi)
111 mov %cx, -0x02(%rsi)
112 ret
113
114L(less_2bytes):
115 cmp $1, %dl
116 jb L(less_1bytes)
117 mov %cl, (%rdi)
118L(less_1bytes):
119 ret
120
121L(512bytesormore):
122 mov __x86_shared_cache_size_half(%rip), %rcx
123 cmp %rcx, %rdx
124 ja L(preloop_large)
125 cmp $1024, %rdx
126 ja L(1024bytesormore)
127
128 vmovups %zmm2, (%rdi)
129 vmovups %zmm2, 0x40(%rdi)
130 vmovups %zmm2, 0x80(%rdi)
131 vmovups %zmm2, 0xC0(%rdi)
132 vmovups %zmm2, 0x100(%rdi)
133 vmovups %zmm2, 0x140(%rdi)
134 vmovups %zmm2, 0x180(%rdi)
135 vmovups %zmm2, 0x1C0(%rdi)
136 vmovups %zmm2, -0x200(%rsi)
137 vmovups %zmm2, -0x1C0(%rsi)
138 vmovups %zmm2, -0x180(%rsi)
139 vmovups %zmm2, -0x140(%rsi)
140 vmovups %zmm2, -0x100(%rsi)
141 vmovups %zmm2, -0xC0(%rsi)
142 vmovups %zmm2, -0x80(%rsi)
143 vmovups %zmm2, -0x40(%rsi)
144 ret
145
146/* Align on 64 and loop with aligned stores. */
147L(1024bytesormore):
148 sub $0x100, %rsi
149 vmovups %zmm2, (%rax)
150 and $-0x40, %rdi
151 add $0x40, %rdi
152
153L(gobble_256bytes_loop):
154 vmovaps %zmm2, (%rdi)
155 vmovaps %zmm2, 0x40(%rdi)
156 vmovaps %zmm2, 0x80(%rdi)
157 vmovaps %zmm2, 0xC0(%rdi)
158 add $0x100, %rdi
159 cmp %rsi, %rdi
160 jb L(gobble_256bytes_loop)
161 vmovups %zmm2, (%rsi)
162 vmovups %zmm2, 0x40(%rsi)
163 vmovups %zmm2, 0x80(%rsi)
164 vmovups %zmm2, 0xC0(%rsi)
165 ret
166
167/* Align on 128 and loop with non-temporal stores. */
168L(preloop_large):
169 and $-0x80, %rdi
170 add $0x80, %rdi
171 vmovups %zmm2, (%rax)
172 vmovups %zmm2, 0x40(%rax)
173 sub $0x200, %rsi
174
175L(gobble_512bytes_nt_loop):
176 vmovntdq %zmm2, (%rdi)
177 vmovntdq %zmm2, 0x40(%rdi)
178 vmovntdq %zmm2, 0x80(%rdi)
179 vmovntdq %zmm2, 0xC0(%rdi)
180 vmovntdq %zmm2, 0x100(%rdi)
181 vmovntdq %zmm2, 0x140(%rdi)
182 vmovntdq %zmm2, 0x180(%rdi)
183 vmovntdq %zmm2, 0x1C0(%rdi)
184 add $0x200, %rdi
185 cmp %rsi, %rdi
186 jb L(gobble_512bytes_nt_loop)
187 sfence
188 vmovups %zmm2, (%rsi)
189 vmovups %zmm2, 0x40(%rsi)
190 vmovups %zmm2, 0x80(%rsi)
191 vmovups %zmm2, 0xC0(%rsi)
192 vmovups %zmm2, 0x100(%rsi)
193 vmovups %zmm2, 0x140(%rsi)
194 vmovups %zmm2, 0x180(%rsi)
195 vmovups %zmm2, 0x1C0(%rsi)
196 ret
197END (MEMSET)
198#endif
199