1/* strcat(dest, src) -- Append SRC on the end of DEST.
2 Optimized for x86-64.
3 Copyright (C) 2002-2022 Free Software Foundation, Inc.
4 This file is part of the GNU C Library.
5
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
10
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
15
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, see
18 <https://www.gnu.org/licenses/>. */
19
20#include <sysdep.h>
21#include "asm-syntax.h"
22
23/* Will be removed when new strcpy implementation gets merged. */
24
25 .text
26ENTRY (strcat)
27 movq %rdi, %rcx /* Dest. register. */
28 andl $7, %ecx /* mask alignment bits */
29 movq %rdi, %rax /* Duplicate destination pointer. */
30 movq $0xfefefefefefefeff,%r8
31
32 /* First step: Find end of destination. */
33 jz 4f /* aligned => start loop */
34
35 neg %ecx /* We need to align to 8 bytes. */
36 addl $8,%ecx
37 /* Search the first bytes directly. */
380: cmpb $0x0,(%rax) /* is byte NUL? */
39 je 2f /* yes => start copy */
40 incq %rax /* increment pointer */
41 decl %ecx
42 jnz 0b
43
44
45
46 /* Now the source is aligned. Scan for NUL byte. */
47 .p2align 4
484:
49 /* First unroll. */
50 movq (%rax), %rcx /* get double word (= 8 bytes) in question */
51 addq $8,%rax /* adjust pointer for next word */
52 movq %r8, %rdx /* magic value */
53 addq %rcx, %rdx /* add the magic value to the word. We get
54 carry bits reported for each byte which
55 is *not* 0 */
56 jnc 3f /* highest byte is NUL => return pointer */
57 xorq %rcx, %rdx /* (word+magic)^word */
58 orq %r8, %rdx /* set all non-carry bits */
59 incq %rdx /* add 1: if one carry bit was *not* set
60 the addition will not result in 0. */
61 jnz 3f /* found NUL => return pointer */
62
63 /* Second unroll. */
64 movq (%rax), %rcx /* get double word (= 8 bytes) in question */
65 addq $8,%rax /* adjust pointer for next word */
66 movq %r8, %rdx /* magic value */
67 addq %rcx, %rdx /* add the magic value to the word. We get
68 carry bits reported for each byte which
69 is *not* 0 */
70 jnc 3f /* highest byte is NUL => return pointer */
71 xorq %rcx, %rdx /* (word+magic)^word */
72 orq %r8, %rdx /* set all non-carry bits */
73 incq %rdx /* add 1: if one carry bit was *not* set
74 the addition will not result in 0. */
75 jnz 3f /* found NUL => return pointer */
76
77 /* Third unroll. */
78 movq (%rax), %rcx /* get double word (= 8 bytes) in question */
79 addq $8,%rax /* adjust pointer for next word */
80 movq %r8, %rdx /* magic value */
81 addq %rcx, %rdx /* add the magic value to the word. We get
82 carry bits reported for each byte which
83 is *not* 0 */
84 jnc 3f /* highest byte is NUL => return pointer */
85 xorq %rcx, %rdx /* (word+magic)^word */
86 orq %r8, %rdx /* set all non-carry bits */
87 incq %rdx /* add 1: if one carry bit was *not* set
88 the addition will not result in 0. */
89 jnz 3f /* found NUL => return pointer */
90
91 /* Fourth unroll. */
92 movq (%rax), %rcx /* get double word (= 8 bytes) in question */
93 addq $8,%rax /* adjust pointer for next word */
94 movq %r8, %rdx /* magic value */
95 addq %rcx, %rdx /* add the magic value to the word. We get
96 carry bits reported for each byte which
97 is *not* 0 */
98 jnc 3f /* highest byte is NUL => return pointer */
99 xorq %rcx, %rdx /* (word+magic)^word */
100 orq %r8, %rdx /* set all non-carry bits */
101 incq %rdx /* add 1: if one carry bit was *not* set
102 the addition will not result in 0. */
103 jz 4b /* no NUL found => continue loop */
104
105 .p2align 4 /* Align, it's a jump target. */
1063: subq $8,%rax /* correct pointer increment. */
107
108 testb %cl, %cl /* is first byte NUL? */
109 jz 2f /* yes => return */
110 incq %rax /* increment pointer */
111
112 testb %ch, %ch /* is second byte NUL? */
113 jz 2f /* yes => return */
114 incq %rax /* increment pointer */
115
116 testl $0x00ff0000, %ecx /* is third byte NUL? */
117 jz 2f /* yes => return pointer */
118 incq %rax /* increment pointer */
119
120 testl $0xff000000, %ecx /* is fourth byte NUL? */
121 jz 2f /* yes => return pointer */
122 incq %rax /* increment pointer */
123
124 shrq $32, %rcx /* look at other half. */
125
126 testb %cl, %cl /* is first byte NUL? */
127 jz 2f /* yes => return */
128 incq %rax /* increment pointer */
129
130 testb %ch, %ch /* is second byte NUL? */
131 jz 2f /* yes => return */
132 incq %rax /* increment pointer */
133
134 testl $0xff0000, %ecx /* is third byte NUL? */
135 jz 2f /* yes => return pointer */
136 incq %rax /* increment pointer */
137
1382:
139 /* Second step: Copy source to destination. */
140
141 movq %rsi, %rcx /* duplicate */
142 andl $7,%ecx /* mask alignment bits */
143 movq %rax, %rdx /* move around */
144 jz 22f /* aligned => start loop */
145
146 neg %ecx /* align to 8 bytes. */
147 addl $8, %ecx
148 /* Align the source pointer. */
14921:
150 movb (%rsi), %al /* Fetch a byte */
151 testb %al, %al /* Is it NUL? */
152 movb %al, (%rdx) /* Store it */
153 jz 24f /* If it was NUL, done! */
154 incq %rsi
155 incq %rdx
156 decl %ecx
157 jnz 21b
158
159 /* Now the sources is aligned. Unfortunatly we cannot force
160 to have both source and destination aligned, so ignore the
161 alignment of the destination. */
162 .p2align 4
16322:
164 /* 1st unroll. */
165 movq (%rsi), %rax /* Read double word (8 bytes). */
166 addq $8, %rsi /* Adjust pointer for next word. */
167 movq %rax, %r9 /* Save a copy for NUL finding. */
168 addq %r8, %r9 /* add the magic value to the word. We get
169 carry bits reported for each byte which
170 is *not* 0 */
171 jnc 23f /* highest byte is NUL => return pointer */
172 xorq %rax, %r9 /* (word+magic)^word */
173 orq %r8, %r9 /* set all non-carry bits */
174 incq %r9 /* add 1: if one carry bit was *not* set
175 the addition will not result in 0. */
176
177 jnz 23f /* found NUL => return pointer */
178
179 movq %rax, (%rdx) /* Write value to destination. */
180 addq $8, %rdx /* Adjust pointer. */
181
182 /* 2nd unroll. */
183 movq (%rsi), %rax /* Read double word (8 bytes). */
184 addq $8, %rsi /* Adjust pointer for next word. */
185 movq %rax, %r9 /* Save a copy for NUL finding. */
186 addq %r8, %r9 /* add the magic value to the word. We get
187 carry bits reported for each byte which
188 is *not* 0 */
189 jnc 23f /* highest byte is NUL => return pointer */
190 xorq %rax, %r9 /* (word+magic)^word */
191 orq %r8, %r9 /* set all non-carry bits */
192 incq %r9 /* add 1: if one carry bit was *not* set
193 the addition will not result in 0. */
194
195 jnz 23f /* found NUL => return pointer */
196
197 movq %rax, (%rdx) /* Write value to destination. */
198 addq $8, %rdx /* Adjust pointer. */
199
200 /* 3rd unroll. */
201 movq (%rsi), %rax /* Read double word (8 bytes). */
202 addq $8, %rsi /* Adjust pointer for next word. */
203 movq %rax, %r9 /* Save a copy for NUL finding. */
204 addq %r8, %r9 /* add the magic value to the word. We get
205 carry bits reported for each byte which
206 is *not* 0 */
207 jnc 23f /* highest byte is NUL => return pointer */
208 xorq %rax, %r9 /* (word+magic)^word */
209 orq %r8, %r9 /* set all non-carry bits */
210 incq %r9 /* add 1: if one carry bit was *not* set
211 the addition will not result in 0. */
212
213 jnz 23f /* found NUL => return pointer */
214
215 movq %rax, (%rdx) /* Write value to destination. */
216 addq $8, %rdx /* Adjust pointer. */
217
218 /* 4th unroll. */
219 movq (%rsi), %rax /* Read double word (8 bytes). */
220 addq $8, %rsi /* Adjust pointer for next word. */
221 movq %rax, %r9 /* Save a copy for NUL finding. */
222 addq %r8, %r9 /* add the magic value to the word. We get
223 carry bits reported for each byte which
224 is *not* 0 */
225 jnc 23f /* highest byte is NUL => return pointer */
226 xorq %rax, %r9 /* (word+magic)^word */
227 orq %r8, %r9 /* set all non-carry bits */
228 incq %r9 /* add 1: if one carry bit was *not* set
229 the addition will not result in 0. */
230
231 jnz 23f /* found NUL => return pointer */
232
233 movq %rax, (%rdx) /* Write value to destination. */
234 addq $8, %rdx /* Adjust pointer. */
235 jmp 22b /* Next iteration. */
236
237 /* Do the last few bytes. %rax contains the value to write.
238 The loop is unrolled twice. */
239 .p2align 4
24023:
241 movb %al, (%rdx) /* 1st byte. */
242 testb %al, %al /* Is it NUL. */
243 jz 24f /* yes, finish. */
244 incq %rdx /* Increment destination. */
245 movb %ah, (%rdx) /* 2nd byte. */
246 testb %ah, %ah /* Is it NUL?. */
247 jz 24f /* yes, finish. */
248 incq %rdx /* Increment destination. */
249 shrq $16, %rax /* Shift... */
250 jmp 23b /* and look at next two bytes in %rax. */
251
252
25324:
254 movq %rdi, %rax /* Source is return value. */
255 retq
256END (strcat)
257libc_hidden_builtin_def (strcat)
258