1 | /* strrchr (str, ch) -- Return pointer to last occurrence of CH in STR. |
2 | Copyright (C) 2013-2022 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 | |
20 | #include <sysdep.h> |
21 | |
22 | .text |
23 | ENTRY (strrchr) |
24 | movd %esi, %xmm1 |
25 | movq %rdi, %rax |
26 | andl $4095, %eax |
27 | punpcklbw %xmm1, %xmm1 |
28 | cmpq $4032, %rax |
29 | punpcklwd %xmm1, %xmm1 |
30 | pshufd $0, %xmm1, %xmm1 |
31 | ja L(cross_page) |
32 | movdqu (%rdi), %xmm0 |
33 | pxor %xmm2, %xmm2 |
34 | movdqa %xmm0, %xmm3 |
35 | pcmpeqb %xmm1, %xmm0 |
36 | pcmpeqb %xmm2, %xmm3 |
37 | pmovmskb %xmm0, %ecx |
38 | pmovmskb %xmm3, %edx |
39 | testq %rdx, %rdx |
40 | je L(next_48_bytes) |
41 | leaq -1(%rdx), %rax |
42 | xorq %rdx, %rax |
43 | andq %rcx, %rax |
44 | je L(exit) |
45 | bsrq %rax, %rax |
46 | addq %rdi, %rax |
47 | ret |
48 | |
49 | .p2align 4 |
50 | L(next_48_bytes): |
51 | movdqu 16(%rdi), %xmm4 |
52 | movdqa %xmm4, %xmm5 |
53 | movdqu 32(%rdi), %xmm3 |
54 | pcmpeqb %xmm1, %xmm4 |
55 | pcmpeqb %xmm2, %xmm5 |
56 | movdqu 48(%rdi), %xmm0 |
57 | pmovmskb %xmm5, %edx |
58 | movdqa %xmm3, %xmm5 |
59 | pcmpeqb %xmm1, %xmm3 |
60 | pcmpeqb %xmm2, %xmm5 |
61 | pcmpeqb %xmm0, %xmm2 |
62 | salq $16, %rdx |
63 | pmovmskb %xmm3, %r8d |
64 | pmovmskb %xmm5, %eax |
65 | pmovmskb %xmm2, %esi |
66 | salq $32, %r8 |
67 | salq $32, %rax |
68 | pcmpeqb %xmm1, %xmm0 |
69 | orq %rdx, %rax |
70 | movq %rsi, %rdx |
71 | pmovmskb %xmm4, %esi |
72 | salq $48, %rdx |
73 | salq $16, %rsi |
74 | orq %r8, %rsi |
75 | orq %rcx, %rsi |
76 | pmovmskb %xmm0, %ecx |
77 | salq $48, %rcx |
78 | orq %rcx, %rsi |
79 | orq %rdx, %rax |
80 | je L(loop_header2) |
81 | leaq -1(%rax), %rcx |
82 | xorq %rax, %rcx |
83 | andq %rcx, %rsi |
84 | je L(exit) |
85 | bsrq %rsi, %rsi |
86 | leaq (%rdi,%rsi), %rax |
87 | ret |
88 | |
89 | .p2align 4 |
90 | L(loop_header2): |
91 | testq %rsi, %rsi |
92 | movq %rdi, %rcx |
93 | je L(no_c_found) |
94 | L(loop_header): |
95 | addq $64, %rdi |
96 | pxor %xmm7, %xmm7 |
97 | andq $-64, %rdi |
98 | jmp L(loop_entry) |
99 | |
100 | .p2align 4 |
101 | L(loop64): |
102 | testq %rdx, %rdx |
103 | cmovne %rdx, %rsi |
104 | cmovne %rdi, %rcx |
105 | addq $64, %rdi |
106 | L(loop_entry): |
107 | movdqa 32(%rdi), %xmm3 |
108 | pxor %xmm6, %xmm6 |
109 | movdqa 48(%rdi), %xmm2 |
110 | movdqa %xmm3, %xmm0 |
111 | movdqa 16(%rdi), %xmm4 |
112 | pminub %xmm2, %xmm0 |
113 | movdqa (%rdi), %xmm5 |
114 | pminub %xmm4, %xmm0 |
115 | pminub %xmm5, %xmm0 |
116 | pcmpeqb %xmm7, %xmm0 |
117 | pmovmskb %xmm0, %eax |
118 | movdqa %xmm5, %xmm0 |
119 | pcmpeqb %xmm1, %xmm0 |
120 | pmovmskb %xmm0, %r9d |
121 | movdqa %xmm4, %xmm0 |
122 | pcmpeqb %xmm1, %xmm0 |
123 | pmovmskb %xmm0, %edx |
124 | movdqa %xmm3, %xmm0 |
125 | pcmpeqb %xmm1, %xmm0 |
126 | salq $16, %rdx |
127 | pmovmskb %xmm0, %r10d |
128 | movdqa %xmm2, %xmm0 |
129 | pcmpeqb %xmm1, %xmm0 |
130 | salq $32, %r10 |
131 | orq %r10, %rdx |
132 | pmovmskb %xmm0, %r8d |
133 | orq %r9, %rdx |
134 | salq $48, %r8 |
135 | orq %r8, %rdx |
136 | testl %eax, %eax |
137 | je L(loop64) |
138 | pcmpeqb %xmm6, %xmm4 |
139 | pcmpeqb %xmm6, %xmm3 |
140 | pcmpeqb %xmm6, %xmm5 |
141 | pmovmskb %xmm4, %eax |
142 | pmovmskb %xmm3, %r10d |
143 | pcmpeqb %xmm6, %xmm2 |
144 | pmovmskb %xmm5, %r9d |
145 | salq $32, %r10 |
146 | salq $16, %rax |
147 | pmovmskb %xmm2, %r8d |
148 | orq %r10, %rax |
149 | orq %r9, %rax |
150 | salq $48, %r8 |
151 | orq %r8, %rax |
152 | leaq -1(%rax), %r8 |
153 | xorq %rax, %r8 |
154 | andq %r8, %rdx |
155 | cmovne %rdi, %rcx |
156 | cmovne %rdx, %rsi |
157 | bsrq %rsi, %rsi |
158 | leaq (%rcx,%rsi), %rax |
159 | ret |
160 | |
161 | .p2align 4 |
162 | L(no_c_found): |
163 | movl $1, %esi |
164 | xorl %ecx, %ecx |
165 | jmp L(loop_header) |
166 | |
167 | .p2align 4 |
168 | L(exit): |
169 | xorl %eax, %eax |
170 | ret |
171 | |
172 | .p2align 4 |
173 | L(cross_page): |
174 | movq %rdi, %rax |
175 | pxor %xmm0, %xmm0 |
176 | andq $-64, %rax |
177 | movdqu (%rax), %xmm5 |
178 | movdqa %xmm5, %xmm6 |
179 | movdqu 16(%rax), %xmm4 |
180 | pcmpeqb %xmm1, %xmm5 |
181 | pcmpeqb %xmm0, %xmm6 |
182 | movdqu 32(%rax), %xmm3 |
183 | pmovmskb %xmm6, %esi |
184 | movdqa %xmm4, %xmm6 |
185 | movdqu 48(%rax), %xmm2 |
186 | pcmpeqb %xmm1, %xmm4 |
187 | pcmpeqb %xmm0, %xmm6 |
188 | pmovmskb %xmm6, %edx |
189 | movdqa %xmm3, %xmm6 |
190 | pcmpeqb %xmm1, %xmm3 |
191 | pcmpeqb %xmm0, %xmm6 |
192 | pcmpeqb %xmm2, %xmm0 |
193 | salq $16, %rdx |
194 | pmovmskb %xmm3, %r9d |
195 | pmovmskb %xmm6, %r8d |
196 | pmovmskb %xmm0, %ecx |
197 | salq $32, %r9 |
198 | salq $32, %r8 |
199 | pcmpeqb %xmm1, %xmm2 |
200 | orq %r8, %rdx |
201 | salq $48, %rcx |
202 | pmovmskb %xmm5, %r8d |
203 | orq %rsi, %rdx |
204 | pmovmskb %xmm4, %esi |
205 | orq %rcx, %rdx |
206 | pmovmskb %xmm2, %ecx |
207 | salq $16, %rsi |
208 | salq $48, %rcx |
209 | orq %r9, %rsi |
210 | orq %r8, %rsi |
211 | orq %rcx, %rsi |
212 | movl %edi, %ecx |
213 | subl %eax, %ecx |
214 | shrq %cl, %rdx |
215 | shrq %cl, %rsi |
216 | testq %rdx, %rdx |
217 | je L(loop_header2) |
218 | leaq -1(%rdx), %rax |
219 | xorq %rdx, %rax |
220 | andq %rax, %rsi |
221 | je L(exit) |
222 | bsrq %rsi, %rax |
223 | addq %rdi, %rax |
224 | ret |
225 | END (strrchr) |
226 | |
227 | weak_alias (strrchr, rindex) |
228 | libc_hidden_builtin_def (strrchr) |
229 | |