about summary refs log tree commit diff
path: root/sysdeps/i386/i586/strchr.S
blob: 37c75260430179e1a4702d690d790b2ab2ef65f7 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
/* Find character CH in a NUL terminated string.
   Highly optimized version for ix85, x>=5.
   Copyright (C) 1995-2021 Free Software Foundation, Inc.
   This file is part of the GNU C Library.

   The GNU C Library is free software; you can redistribute it and/or
   modify it under the terms of the GNU Lesser General Public
   License as published by the Free Software Foundation; either
   version 2.1 of the License, or (at your option) any later version.

   The GNU C Library is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
   Lesser General Public License for more details.

   You should have received a copy of the GNU Lesser General Public
   License along with the GNU C Library; if not, see
   <https://www.gnu.org/licenses/>.  */

#include <sysdep.h>
#include "asm-syntax.h"

/* This version is especially optimized for the i586 (and following?)
   processors.  This is mainly done by using the two pipelines.  The
   version optimized for i486 is weak in this aspect because to get
   as much parallelism we have to execute some *more* instructions.

   The code below is structured to reflect the pairing of the instructions
   as *I think* it is.  I have no processor data book to verify this.
   If you find something you think is incorrect let me know.  */


/* The magic value which is used throughout in the whole code.  */
#define magic 0xfefefeff

#define PARMS	4+16	/* space for 4 saved regs */
#define RTN	PARMS
#define STR	RTN
#define CHR	STR+4

	.text
ENTRY (strchr)

	pushl %edi		/* Save callee-safe registers.  */
	cfi_adjust_cfa_offset (-4)
	pushl %esi
	cfi_adjust_cfa_offset (-4)

	pushl %ebx
	cfi_adjust_cfa_offset (-4)
	pushl %ebp
	cfi_adjust_cfa_offset (-4)

	movl STR(%esp), %eax
	movl CHR(%esp), %edx

	movl %eax, %edi		/* duplicate string pointer for later */
	cfi_rel_offset (edi, 12)
	xorl %ecx, %ecx		/* clear %ecx */

	/* At the moment %edx contains C.  What we need for the
	   algorithm is C in all bytes of the dword.  Avoid
	   operations on 16 bit words because these require an
	   prefix byte (and one more cycle).  */
	movb %dl, %dh		/* now it is 0|0|c|c */
	movb %dl, %cl		/* we construct the lower half in %ecx */

	shll $16, %edx		/* now %edx is c|c|0|0 */
	movb %cl, %ch		/* now %ecx is 0|0|c|c */

	orl %ecx, %edx		/* and finally c|c|c|c */
	andl $3, %edi		/* mask alignment bits */

	jz L(11)		/* alignment is 0 => start loop */

	movb %dl, %cl		/* 0 is needed below */
	jp L(0)			/* exactly two bits set */

	xorb (%eax), %cl	/* is byte the one we are looking for? */
	jz L(out)		/* yes => return pointer */

	xorb %dl, %cl		/* load single byte and test for NUL */
	je L(3)			/* yes => return NULL */

	movb 1(%eax), %cl	/* load single byte */
	incl %eax

	cmpb %cl, %dl		/* is byte == C? */
	je L(out)		/* aligned => return pointer */

	cmpb $0, %cl		/* is byte NUL? */
	je L(3)			/* yes => return NULL */

	incl %eax
	decl %edi

	jne L(11)

L(0):	movb (%eax), %cl	/* load single byte */

	cmpb %cl, %dl		/* is byte == C? */
	je L(out)		/* aligned => return pointer */

	cmpb $0, %cl		/* is byte NUL? */
	je L(3)			/* yes => return NULL */

	incl %eax		/* increment pointer */

	cfi_rel_offset (esi, 8)
	cfi_rel_offset (ebx, 4)
	cfi_rel_offset (ebp, 0)

	/* The following code is the preparation for the loop.  The
	   four instruction up to `L1' will not be executed in the loop
	   because the same code is found at the end of the loop, but
	   there it is executed in parallel with other instructions.  */
L(11):	movl (%eax), %ecx
	movl $magic, %ebp

	movl $magic, %edi
	addl %ecx, %ebp

	/* The main loop: it looks complex and indeed it is.  I would
	   love to say `it was hard to write, so it should he hard to
	   read' but I will give some more hints.  To fully understand
	   this code you should first take a look at the i486 version.
	   The basic algorithm is the same, but here the code organized
	   in a way which permits to use both pipelines all the time.

	   I tried to make it a bit more understandable by indenting
	   the code according to stage in the algorithm.  It goes as
	   follows:
		check for 0 in 1st word
			check for C in 1st word
					check for 0 in 2nd word
						check for C in 2nd word
		check for 0 in 3rd word
			check for C in 3rd word
					check for 0 in 4th word
						check for C in 4th word

	   Please note that doing the test for NUL before the test for
	   C allows us to overlap the test for 0 in the next word with
	   the test for C.  */

L(1):	xorl %ecx, %ebp			/* (word^magic) */
	addl %ecx, %edi			/* add magic word */

	leal 4(%eax), %eax		/* increment pointer */
	jnc L(4)			/* previous addl caused overflow? */

		movl %ecx, %ebx		/* duplicate original word */
	orl $magic, %ebp		/* (word^magic)|magic */

	addl $1, %ebp			/* (word^magic)|magic == 0xffffffff? */
	jne L(4)				/* yes => we found word with NUL */

		movl $magic, %esi	/* load magic value */
		xorl %edx, %ebx		/* clear words which are C */

					movl (%eax), %ecx
		addl %ebx, %esi		/* (word+magic) */

					movl $magic, %edi
		jnc L(5)		/* previous addl caused overflow? */

					movl %edi, %ebp
		xorl %ebx, %esi		/* (word+magic)^word */

					addl %ecx, %ebp
		orl $magic, %esi	/* ((word+magic)^word)|magic */

		addl $1, %esi		/* ((word+magic)^word)|magic==0xf..f?*/
		jne L(5)		/* yes => we found word with C */

					xorl %ecx, %ebp
					addl %ecx, %edi

					leal 4(%eax), %eax
					jnc L(4)

						movl %ecx, %ebx
					orl $magic, %ebp

					addl $1, %ebp
					jne L(4)

						movl $magic, %esi
						xorl %edx, %ebx

	movl (%eax), %ecx
						addl %ebx, %esi

	movl $magic, %edi
						jnc L(5)

	movl %edi, %ebp
						xorl %ebx, %esi

	addl %ecx, %ebp
						orl $magic, %esi

						addl $1, %esi
						jne L(5)

	xorl %ecx, %ebp
	addl %ecx, %edi

	leal 4(%eax), %eax
	jnc L(4)

		movl %ecx, %ebx
	orl $magic, %ebp

	addl $1, %ebp
	jne L(4)

		movl $magic, %esi
		xorl %edx, %ebx

					movl (%eax), %ecx
		addl %ebx, %esi

					movl $magic, %edi
		jnc L(5)

					movl %edi, %ebp
		xorl %ebx, %esi

					addl %ecx, %ebp
		orl $magic, %esi

		addl $1, %esi
		jne L(5)

					xorl %ecx, %ebp
					addl %ecx, %edi

					leal 4(%eax), %eax
					jnc L(4)

						movl %ecx, %ebx
					orl $magic, %ebp

					addl $1, %ebp
					jne L(4)

						movl $magic, %esi
						xorl %edx, %ebx

	movl (%eax), %ecx
						addl %ebx, %esi

	movl $magic, %edi
						jnc L(5)

	movl %edi, %ebp
						xorl %ebx, %esi

	addl %ecx, %ebp
						orl $magic, %esi

						addl $1, %esi

						je L(1)

	/* We know there is no NUL byte but a C byte in the word.
	   %ebx contains NUL in this particular byte.  */
L(5):	subl $4, %eax		/* adjust pointer */
	testb %bl, %bl		/* first byte == C? */

	jz L(out)		/* yes => return pointer */

	incl %eax		/* increment pointer */
	testb %bh, %bh		/* second byte == C? */

	jz L(out)		/* yes => return pointer */

	shrl $16, %ebx		/* make upper bytes accessible */
	incl %eax		/* increment pointer */

	cmp $0, %bl		/* third byte == C */
	je L(out)		/* yes => return pointer */

	incl %eax		/* increment pointer */

L(out):	popl %ebp		/* restore saved registers */
	cfi_adjust_cfa_offset (-4)
	cfi_restore (ebp)
	popl %ebx
	cfi_adjust_cfa_offset (-4)
	cfi_restore (ebx)

	popl %esi
	cfi_adjust_cfa_offset (-4)
	cfi_restore (esi)
	popl %edi
	cfi_adjust_cfa_offset (-4)
	cfi_restore (edi)

	ret

	cfi_adjust_cfa_offset (16)
	cfi_rel_offset (edi, 12)
	cfi_rel_offset (esi, 8)
	cfi_rel_offset (ebx, 4)
	cfi_rel_offset (ebp, 0)
	/* We know there is a NUL byte in the word.  But we have to test
	   whether there is an C byte before it in the word.  */
L(4):	subl $4, %eax		/* adjust pointer */
	cmpb %dl, %cl		/* first byte == C? */

	je L(out)		/* yes => return pointer */

	cmpb $0, %cl		/* first byte == NUL? */
	je L(3)			/* yes => return NULL */

	incl %eax		/* increment pointer */

	cmpb %dl, %ch		/* second byte == C? */
	je L(out)		/* yes => return pointer */

	cmpb $0, %ch		/* second byte == NUL? */
	je L(3)			/* yes => return NULL */

	shrl $16, %ecx		/* make upper bytes accessible */
	incl %eax		/* increment pointer */

	cmpb %dl, %cl		/* third byte == C? */
	je L(out)		/* yes => return pointer */

	cmpb $0, %cl		/* third byte == NUL? */
	je L(3)			/* yes => return NULL */

	incl %eax		/* increment pointer */

	/* The test four the fourth byte is necessary!  */
	cmpb %dl, %ch		/* fourth byte == C? */
	je L(out)		/* yes => return pointer */

L(3):	xorl %eax, %eax
	jmp L(out)
END (strchr)

#undef index
weak_alias (strchr, index)
libc_hidden_builtin_def (strchr)