1/* rawmemchr optimized with SSE2.
2 Copyright (C) 2017-2023 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 <isa-level.h>
20#include <sysdep.h>
21
22/* MINIMUM_X86_ISA_LEVEL <= 2 because there is no V2 implementation
23 so we need this to build for ISA V2 builds. */
24#if ISA_SHOULD_BUILD (2)
25
26# ifndef RAWMEMCHR
27# define RAWMEMCHR __rawmemchr_sse2
28# endif
29
30 .text
31ENTRY (RAWMEMCHR)
32 movd %rsi, %xmm1
33 mov %rdi, %rcx
34
35 punpcklbw %xmm1, %xmm1
36 punpcklbw %xmm1, %xmm1
37
38 and $63, %rcx
39 pshufd $0, %xmm1, %xmm1
40
41 cmp $48, %rcx
42 ja L(crosscache)
43
44 movdqu (%rdi), %xmm0
45 pcmpeqb %xmm1, %xmm0
46/* Check if there is a match. */
47 pmovmskb %xmm0, %eax
48 test %eax, %eax
49
50 jnz L(matches)
51 add $16, %rdi
52 and $-16, %rdi
53 jmp L(loop_prolog)
54
55 .p2align 4
56L(crosscache):
57 and $15, %rcx
58 and $-16, %rdi
59 movdqa (%rdi), %xmm0
60
61 pcmpeqb %xmm1, %xmm0
62/* Check if there is a match. */
63 pmovmskb %xmm0, %eax
64/* Remove the leading bytes. */
65 sar %cl, %eax
66 test %eax, %eax
67 je L(unaligned_no_match)
68/* Check which byte is a match. */
69 bsf %eax, %eax
70
71 add %rdi, %rax
72 add %rcx, %rax
73 ret
74
75 .p2align 4
76L(unaligned_no_match):
77 add $16, %rdi
78
79 .p2align 4
80L(loop_prolog):
81 movdqa (%rdi), %xmm0
82 pcmpeqb %xmm1, %xmm0
83 pmovmskb %xmm0, %eax
84 test %eax, %eax
85 jnz L(matches)
86
87 movdqa 16(%rdi), %xmm2
88 pcmpeqb %xmm1, %xmm2
89 pmovmskb %xmm2, %eax
90 test %eax, %eax
91 jnz L(matches16)
92
93 movdqa 32(%rdi), %xmm3
94 pcmpeqb %xmm1, %xmm3
95 pmovmskb %xmm3, %eax
96 test %eax, %eax
97 jnz L(matches32)
98
99 movdqa 48(%rdi), %xmm4
100 pcmpeqb %xmm1, %xmm4
101 add $64, %rdi
102 pmovmskb %xmm4, %eax
103 test %eax, %eax
104 jnz L(matches0)
105
106 test $0x3f, %rdi
107 jz L(align64_loop)
108
109 movdqa (%rdi), %xmm0
110 pcmpeqb %xmm1, %xmm0
111 pmovmskb %xmm0, %eax
112 test %eax, %eax
113 jnz L(matches)
114
115 movdqa 16(%rdi), %xmm2
116 pcmpeqb %xmm1, %xmm2
117 pmovmskb %xmm2, %eax
118 test %eax, %eax
119 jnz L(matches16)
120
121 movdqa 32(%rdi), %xmm3
122 pcmpeqb %xmm1, %xmm3
123 pmovmskb %xmm3, %eax
124 test %eax, %eax
125 jnz L(matches32)
126
127 movdqa 48(%rdi), %xmm3
128 pcmpeqb %xmm1, %xmm3
129 pmovmskb %xmm3, %eax
130
131 add $64, %rdi
132 test %eax, %eax
133 jnz L(matches0)
134
135 and $-64, %rdi
136
137 .p2align 4
138L(align64_loop):
139 movdqa (%rdi), %xmm0
140 movdqa 16(%rdi), %xmm2
141 movdqa 32(%rdi), %xmm3
142 movdqa 48(%rdi), %xmm4
143
144 pcmpeqb %xmm1, %xmm0
145 pcmpeqb %xmm1, %xmm2
146 pcmpeqb %xmm1, %xmm3
147 pcmpeqb %xmm1, %xmm4
148
149 pmaxub %xmm0, %xmm3
150 pmaxub %xmm2, %xmm4
151 pmaxub %xmm3, %xmm4
152 pmovmskb %xmm4, %eax
153
154 add $64, %rdi
155
156 test %eax, %eax
157 jz L(align64_loop)
158
159 sub $64, %rdi
160
161 pmovmskb %xmm0, %eax
162 test %eax, %eax
163 jnz L(matches)
164
165 pmovmskb %xmm2, %eax
166 test %eax, %eax
167 jnz L(matches16)
168
169 movdqa 32(%rdi), %xmm3
170 pcmpeqb %xmm1, %xmm3
171
172 pcmpeqb 48(%rdi), %xmm1
173 pmovmskb %xmm3, %eax
174 test %eax, %eax
175 jnz L(matches32)
176
177 pmovmskb %xmm1, %eax
178 bsf %eax, %eax
179 lea 48(%rdi, %rax), %rax
180 ret
181
182 .p2align 4
183L(matches0):
184 bsf %eax, %eax
185 lea -16(%rax, %rdi), %rax
186 ret
187
188 .p2align 4
189L(matches):
190 bsf %eax, %eax
191 add %rdi, %rax
192 ret
193
194 .p2align 4
195L(matches16):
196 bsf %eax, %eax
197 lea 16(%rax, %rdi), %rax
198 ret
199
200 .p2align 4
201L(matches32):
202 bsf %eax, %eax
203 lea 32(%rax, %rdi), %rax
204 ret
205
206END (RAWMEMCHR)
207#endif
208