]>
Commit | Line | Data |
---|---|---|
3f02f778 | 1 | /* strchrnul (str, chr) -- Return pointer to first occurrence of CHR in STR |
c4563d2d UD |
2 | or the final NUL byte. |
3 | For Intel 80x86, x>=3. | |
d4697bc9 | 4 | Copyright (C) 1994-2014 Free Software Foundation, Inc. |
c4563d2d UD |
5 | This file is part of the GNU C Library. |
6 | Contributed by Ulrich Drepper <drepper@gnu.org> | |
7 | Some optimisations by Alan Modra <Alan@SPRI.Levels.UniSA.Edu.Au> | |
8 | ||
9 | The GNU C Library is free software; you can redistribute it and/or | |
41bdb6e2 AJ |
10 | modify it under the terms of the GNU Lesser General Public |
11 | License as published by the Free Software Foundation; either | |
12 | version 2.1 of the License, or (at your option) any later version. | |
c4563d2d UD |
13 | |
14 | The GNU C Library is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
41bdb6e2 | 17 | Lesser General Public License for more details. |
c4563d2d | 18 | |
41bdb6e2 | 19 | You should have received a copy of the GNU Lesser General Public |
59ba27a6 PE |
20 | License along with the GNU C Library; if not, see |
21 | <http://www.gnu.org/licenses/>. */ | |
c4563d2d UD |
22 | |
23 | #include <sysdep.h> | |
24 | #include "asm-syntax.h" | |
25 | ||
2366713d | 26 | #define PARMS 4+4 /* space for 1 saved reg */ |
3f02f778 | 27 | #define RTN PARMS |
2366713d JM |
28 | #define STR RTN |
29 | #define CHR STR+4 | |
c4563d2d UD |
30 | |
31 | .text | |
2366713d | 32 | ENTRY (__strchrnul) |
3f02f778 | 33 | |
c4563d2d | 34 | pushl %edi /* Save callee-safe registers used here. */ |
1ad9da69 UD |
35 | cfi_adjust_cfa_offset (4) |
36 | cfi_rel_offset (edi, 0) | |
c4563d2d | 37 | |
3f02f778 GM |
38 | movl STR(%esp), %eax |
39 | movl CHR(%esp), %edx | |
c4563d2d | 40 | |
3f02f778 GM |
41 | /* At the moment %edx contains CHR. What we need for the |
42 | algorithm is CHR in all bytes of the dword. Avoid | |
c4563d2d UD |
43 | operations on 16 bit words because these require an |
44 | prefix byte (and one more cycle). */ | |
45 | movb %dl, %dh /* now it is 0|0|c|c */ | |
46 | movl %edx, %ecx | |
47 | shll $16, %edx /* now it is c|c|0|0 */ | |
48 | movw %cx, %dx /* and finally c|c|c|c */ | |
49 | ||
50 | /* Before we start with the main loop we process single bytes | |
51 | until the source pointer is aligned. This has two reasons: | |
52 | 1. aligned 32-bit memory access is faster | |
53 | and (more important) | |
54 | 2. we process in the main loop 32 bit in one step although | |
55 | we don't know the end of the string. But accessing at | |
56 | 4-byte alignment guarantees that we never access illegal | |
57 | memory if this would not also be done by the trivial | |
58 | implementation (this is because all processor inherent | |
59 | boundaries are multiples of 4. */ | |
60 | ||
a7123f0e | 61 | testb $3, %al /* correctly aligned ? */ |
c4563d2d UD |
62 | jz L(11) /* yes => begin loop */ |
63 | movb (%eax), %cl /* load byte in question (we need it twice) */ | |
64 | cmpb %cl, %dl /* compare byte */ | |
65 | je L(6) /* target found => return */ | |
66 | testb %cl, %cl /* is NUL? */ | |
67 | jz L(6) /* yes => return NULL */ | |
68 | incl %eax /* increment pointer */ | |
69 | ||
a7123f0e | 70 | testb $3, %al /* correctly aligned ? */ |
c4563d2d UD |
71 | jz L(11) /* yes => begin loop */ |
72 | movb (%eax), %cl /* load byte in question (we need it twice) */ | |
73 | cmpb %cl, %dl /* compare byte */ | |
74 | je L(6) /* target found => return */ | |
75 | testb %cl, %cl /* is NUL? */ | |
76 | jz L(6) /* yes => return NULL */ | |
77 | incl %eax /* increment pointer */ | |
78 | ||
a7123f0e | 79 | testb $3, %al /* correctly aligned ? */ |
c4563d2d UD |
80 | jz L(11) /* yes => begin loop */ |
81 | movb (%eax), %cl /* load byte in question (we need it twice) */ | |
82 | cmpb %cl, %dl /* compare byte */ | |
83 | je L(6) /* target found => return */ | |
84 | testb %cl, %cl /* is NUL? */ | |
85 | jz L(6) /* yes => return NULL */ | |
86 | incl %eax /* increment pointer */ | |
87 | ||
88 | /* No we have reached alignment. */ | |
89 | jmp L(11) /* begin loop */ | |
90 | ||
91 | /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to | |
92 | change any of the hole bits of LONGWORD. | |
93 | ||
94 | 1) Is this safe? Will it catch all the zero bytes? | |
95 | Suppose there is a byte with all zeros. Any carry bits | |
96 | propagating from its left will fall into the hole at its | |
97 | least significant bit and stop. Since there will be no | |
98 | carry from its most significant bit, the LSB of the | |
99 | byte to the left will be unchanged, and the zero will be | |
100 | detected. | |
101 | ||
102 | 2) Is this worthwhile? Will it ignore everything except | |
103 | zero bytes? Suppose every byte of LONGWORD has a bit set | |
104 | somewhere. There will be a carry into bit 8. If bit 8 | |
105 | is set, this will carry into bit 16. If bit 8 is clear, | |
106 | one of bits 9-15 must be set, so there will be a carry | |
107 | into bit 16. Similarly, there will be a carry into bit | |
108 | 24. If one of bits 24-31 is set, there will be a carry | |
109 | into bit 32 (=carry flag), so all of the hole bits will | |
110 | be changed. | |
111 | ||
3f02f778 | 112 | 3) But wait! Aren't we looking for CHR, not zero? |
c4563d2d | 113 | Good point. So what we do is XOR LONGWORD with a longword, |
3f02f778 | 114 | each of whose bytes is CHR. This turns each byte that is CHR |
c4563d2d UD |
115 | into a zero. */ |
116 | ||
117 | /* Each round the main loop processes 16 bytes. */ | |
118 | ||
119 | ALIGN(4) | |
120 | ||
121 | L(1): addl $16, %eax /* adjust pointer for whole round */ | |
122 | ||
123 | L(11): movl (%eax), %ecx /* get word (= 4 bytes) in question */ | |
124 | xorl %edx, %ecx /* XOR with word c|c|c|c => bytes of str == c | |
125 | are now 0 */ | |
126 | movl $0xfefefeff, %edi /* magic value */ | |
127 | addl %ecx, %edi /* add the magic value to the word. We get | |
128 | carry bits reported for each byte which | |
3f02f778 | 129 | is *not* CHR */ |
c4563d2d UD |
130 | |
131 | /* According to the algorithm we had to reverse the effect of the | |
132 | XOR first and then test the overflow bits. But because the | |
133 | following XOR would destroy the carry flag and it would (in a | |
134 | representation with more than 32 bits) not alter then last | |
135 | overflow, we can now test this condition. If no carry is signaled | |
136 | no overflow must have occurred in the last byte => it was 0. */ | |
137 | jnc L(7) | |
138 | ||
139 | /* We are only interested in carry bits that change due to the | |
140 | previous add, so remove original bits */ | |
141 | xorl %ecx, %edi /* ((word^charmask)+magic)^(word^charmask) */ | |
142 | ||
143 | /* Now test for the other three overflow bits. */ | |
144 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
145 | incl %edi /* add 1: if one carry bit was *not* set | |
146 | the addition will not result in 0. */ | |
147 | ||
3f02f778 | 148 | /* If at least one byte of the word is CHR we don't get 0 in %edi. */ |
c4563d2d UD |
149 | jnz L(7) /* found it => return pointer */ |
150 | ||
151 | /* Now we made sure the dword does not contain the character we are | |
152 | looking for. But because we deal with strings we have to check | |
153 | for the end of string before testing the next dword. */ | |
154 | ||
155 | xorl %edx, %ecx /* restore original dword without reload */ | |
156 | movl $0xfefefeff, %edi /* magic value */ | |
157 | addl %ecx, %edi /* add the magic value to the word. We get | |
158 | carry bits reported for each byte which | |
159 | is *not* 0 */ | |
d8b29eea | 160 | jnc L(7) /* highest byte is NUL => return NULL */ |
c4563d2d UD |
161 | xorl %ecx, %edi /* (word+magic)^word */ |
162 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
163 | incl %edi /* add 1: if one carry bit was *not* set | |
164 | the addition will not result in 0. */ | |
d8b29eea | 165 | jnz L(7) /* found NUL => return NULL */ |
c4563d2d UD |
166 | |
167 | movl 4(%eax), %ecx /* get word (= 4 bytes) in question */ | |
168 | xorl %edx, %ecx /* XOR with word c|c|c|c => bytes of str == c | |
169 | are now 0 */ | |
170 | movl $0xfefefeff, %edi /* magic value */ | |
171 | addl %ecx, %edi /* add the magic value to the word. We get | |
172 | carry bits reported for each byte which | |
3f02f778 GM |
173 | is *not* CHR */ |
174 | jnc L(71) /* highest byte is CHR => return pointer */ | |
c4563d2d UD |
175 | xorl %ecx, %edi /* ((word^charmask)+magic)^(word^charmask) */ |
176 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
177 | incl %edi /* add 1: if one carry bit was *not* set | |
178 | the addition will not result in 0. */ | |
179 | jnz L(71) /* found it => return pointer */ | |
180 | xorl %edx, %ecx /* restore original dword without reload */ | |
181 | movl $0xfefefeff, %edi /* magic value */ | |
182 | addl %ecx, %edi /* add the magic value to the word. We get | |
183 | carry bits reported for each byte which | |
184 | is *not* 0 */ | |
185 | jnc L(71) /* highest byte is NUL => return NULL */ | |
186 | xorl %ecx, %edi /* (word+magic)^word */ | |
187 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
188 | incl %edi /* add 1: if one carry bit was *not* set | |
189 | the addition will not result in 0. */ | |
190 | jnz L(71) /* found NUL => return NULL */ | |
191 | ||
192 | movl 8(%eax), %ecx /* get word (= 4 bytes) in question */ | |
193 | xorl %edx, %ecx /* XOR with word c|c|c|c => bytes of str == c | |
194 | are now 0 */ | |
195 | movl $0xfefefeff, %edi /* magic value */ | |
196 | addl %ecx, %edi /* add the magic value to the word. We get | |
197 | carry bits reported for each byte which | |
3f02f778 GM |
198 | is *not* CHR */ |
199 | jnc L(72) /* highest byte is CHR => return pointer */ | |
c4563d2d UD |
200 | xorl %ecx, %edi /* ((word^charmask)+magic)^(word^charmask) */ |
201 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
202 | incl %edi /* add 1: if one carry bit was *not* set | |
203 | the addition will not result in 0. */ | |
204 | jnz L(72) /* found it => return pointer */ | |
205 | xorl %edx, %ecx /* restore original dword without reload */ | |
206 | movl $0xfefefeff, %edi /* magic value */ | |
207 | addl %ecx, %edi /* add the magic value to the word. We get | |
208 | carry bits reported for each byte which | |
209 | is *not* 0 */ | |
210 | jnc L(72) /* highest byte is NUL => return NULL */ | |
211 | xorl %ecx, %edi /* (word+magic)^word */ | |
212 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
213 | incl %edi /* add 1: if one carry bit was *not* set | |
214 | the addition will not result in 0. */ | |
215 | jnz L(72) /* found NUL => return NULL */ | |
216 | ||
217 | movl 12(%eax), %ecx /* get word (= 4 bytes) in question */ | |
218 | xorl %edx, %ecx /* XOR with word c|c|c|c => bytes of str == c | |
219 | are now 0 */ | |
220 | movl $0xfefefeff, %edi /* magic value */ | |
221 | addl %ecx, %edi /* add the magic value to the word. We get | |
222 | carry bits reported for each byte which | |
3f02f778 GM |
223 | is *not* CHR */ |
224 | jnc L(73) /* highest byte is CHR => return pointer */ | |
c4563d2d UD |
225 | xorl %ecx, %edi /* ((word^charmask)+magic)^(word^charmask) */ |
226 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
227 | incl %edi /* add 1: if one carry bit was *not* set | |
228 | the addition will not result in 0. */ | |
229 | jnz L(73) /* found it => return pointer */ | |
230 | xorl %edx, %ecx /* restore original dword without reload */ | |
231 | movl $0xfefefeff, %edi /* magic value */ | |
232 | addl %ecx, %edi /* add the magic value to the word. We get | |
233 | carry bits reported for each byte which | |
234 | is *not* 0 */ | |
235 | jnc L(73) /* highest byte is NUL => return NULL */ | |
236 | xorl %ecx, %edi /* (word+magic)^word */ | |
237 | orl $0xfefefeff, %edi /* set all non-carry bits */ | |
238 | incl %edi /* add 1: if one carry bit was *not* set | |
239 | the addition will not result in 0. */ | |
240 | jz L(1) /* no NUL found => restart loop */ | |
241 | ||
242 | L(73): addl $4, %eax /* adjust pointer */ | |
243 | L(72): addl $4, %eax | |
244 | L(71): addl $4, %eax | |
245 | ||
246 | /* We now scan for the byte in which the character was matched. | |
247 | But we have to take care of the case that a NUL char is | |
248 | found before this in the dword. */ | |
249 | ||
3f02f778 | 250 | L(7): testb %cl, %cl /* is first byte CHR? */ |
c4563d2d UD |
251 | jz L(6) /* yes => return pointer */ |
252 | cmpb %dl, %cl /* is first byte NUL? */ | |
253 | je L(6) /* yes => return NULL */ | |
254 | incl %eax /* it's not in the first byte */ | |
255 | ||
3f02f778 | 256 | testb %ch, %ch /* is second byte CHR? */ |
c4563d2d UD |
257 | jz L(6) /* yes => return pointer */ |
258 | cmpb %dl, %ch /* is second byte NUL? */ | |
259 | je L(6) /* yes => return NULL? */ | |
260 | incl %eax /* it's not in the second byte */ | |
261 | ||
262 | shrl $16, %ecx /* make upper byte accessible */ | |
3f02f778 | 263 | testb %cl, %cl /* is third byte CHR? */ |
c4563d2d UD |
264 | jz L(6) /* yes => return pointer */ |
265 | cmpb %dl, %cl /* is third byte NUL? */ | |
266 | je L(6) /* yes => return NULL */ | |
267 | ||
268 | /* It must be in the fourth byte and it cannot be NUL. */ | |
269 | incl %eax | |
270 | ||
92945b52 | 271 | L(6): popl %edi /* restore saved register content */ |
1ad9da69 UD |
272 | cfi_adjust_cfa_offset (-4) |
273 | cfi_restore (edi) | |
c4563d2d | 274 | |
2366713d JM |
275 | ret |
276 | END (__strchrnul) | |
c4563d2d | 277 | |
2366713d | 278 | weak_alias (__strchrnul, strchrnul) |