about summary refs log tree commit diff
path: root/sysdeps/i386/rawmemchr.S
blob: 6df65e69afc5a4c7cecf8fe85b18fc94461b1bb9 (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
/* rawmemchr (str, ch) -- Return pointer to first occurrence of CH in STR.
   For Intel 80x86, x>=3.
   Copyright (C) 1994-2013 Free Software Foundation, Inc.
   This file is part of the GNU C Library.
   Contributed by Ulrich Drepper <drepper@gnu.ai.mit.edu>
   Optimised a little by Alan Modra <Alan@SPRI.Levels.UniSA.Edu.Au>
   This version is developed using the same algorithm as the fast C
   version which carries the following introduction:
   Based on strlen implementation by Torbjorn Granlund (tege@sics.se),
   with help from Dan Sahlin (dan@sics.se) and
   commentary by Jim Blandy (jimb@ai.mit.edu);
   adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu),
   and implemented by Roland McGrath (roland@ai.mit.edu).

   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
   <http://www.gnu.org/licenses/>.  */

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

#define PARMS	LINKAGE+4	/* space for 1 saved reg */
#define RTN	PARMS
#define STR	RTN+RTN_SIZE
#define CHR	STR+PTR_SIZE

	.text
ENTRY (BP_SYM (__rawmemchr))

	/* Save callee-safe register used in this function.  */
	pushl %edi
	cfi_adjust_cfa_offset (4)
	cfi_rel_offset (edi, 0)

	/* Load parameters into registers.  */
	movl STR(%esp), %eax
	movl CHR(%esp), %edx

	/* 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 */
	movl %edx, %ecx
	shll $16, %edx		/* Now c|c|0|0 */
	movw %cx, %dx		/* And finally c|c|c|c */

	/* Better performance can be achieved if the word (32
	   bit) memory access is aligned on a four-byte-boundary.
	   So process first bytes one by one until boundary is
	   reached. Don't use a loop for better performance.  */

	testb $3, %al		/* correctly aligned ? */
	je L(1)			/* yes => begin loop */
	cmpb %dl, (%eax)	/* compare byte */
	je L(9)			/* target found => return */
	incl %eax		/* increment source pointer */

	testb $3, %al		/* correctly aligned ? */
	je L(1)			/* yes => begin loop */
	cmpb %dl, (%eax)	/* compare byte */
	je L(9)			/* target found => return */
	incl %eax		/* increment source pointer */

	testb $3, %al		/* correctly aligned ? */
	je L(1)			/* yes => begin loop */
	cmpb %dl, (%eax)	/* compare byte */
	je L(9)			/* target found => return */
	incl %eax		/* increment source pointer */

      /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to
	 change any of the hole bits of LONGWORD.

	 1) Is this safe?  Will it catch all the zero bytes?
	 Suppose there is a byte with all zeros.  Any carry bits
	 propagating from its left will fall into the hole at its
	 least significant bit and stop.  Since there will be no
	 carry from its most significant bit, the LSB of the
	 byte to the left will be unchanged, and the zero will be
	 detected.

	 2) Is this worthwhile?  Will it ignore everything except
	 zero bytes?  Suppose every byte of LONGWORD has a bit set
	 somewhere.  There will be a carry into bit 8.	If bit 8
	 is set, this will carry into bit 16.  If bit 8 is clear,
	 one of bits 9-15 must be set, so there will be a carry
	 into bit 16.  Similarly, there will be a carry into bit
	 24.  If one of bits 24-31 is set, there will be a carry
	 into bit 32 (=carry flag), so all of the hole bits will
	 be changed.

	 3) But wait!  Aren't we looking for C, not zero?
	 Good point.  So what we do is XOR LONGWORD with a longword,
	 each of whose bytes is C.  This turns each byte that is C
	 into a zero.  */


	/* Each round the main loop processes 16 bytes.  */
	ALIGN (4)

L(1):	movl (%eax), %ecx	/* get word (= 4 bytes) in question */
	movl $0xfefefeff, %edi	/* magic value */
	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
				   are now 0 */
	addl %ecx, %edi		/* add the magic value to the word.  We get
				   carry bits reported for each byte which
				   is *not* 0 */

	/* According to the algorithm we had to reverse the effect of the
	   XOR first and then test the overflow bits.  But because the
	   following XOR would destroy the carry flag and it would (in a
	   representation with more than 32 bits) not alter then last
	   overflow, we can now test this condition.  If no carry is signaled
	   no overflow must have occurred in the last byte => it was 0.	*/
	jnc L(8)

	/* We are only interested in carry bits that change due to the
	   previous add, so remove original bits */
	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */

	/* Now test for the other three overflow bits.  */
	orl $0xfefefeff, %edi	/* set all non-carry bits */
	incl %edi		/* add 1: if one carry bit was *not* set
				   the addition will not result in 0.  */

	/* If at least one byte of the word is C we don't get 0 in %edi.  */
	jnz L(8)		/* found it => return pointer */

	/* This process is unfolded four times for better performance.
	   we don't increment the source pointer each time.  Instead we
	   use offsets and increment by 16 in each run of the loop.  But
	   before probing for the matching byte we need some extra code
	   (following LL(13) below).  Even the len can be compared with
	   constants instead of decrementing each time.  */

	movl 4(%eax), %ecx	/* get word (= 4 bytes) in question */
	movl $0xfefefeff, %edi	/* magic value */
	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
				   are now 0 */
	addl %ecx, %edi		/* add the magic value to the word.  We get
				   carry bits reported for each byte which
				   is *not* 0 */
	jnc L(7)		/* highest byte is C => return pointer */
	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
	orl $0xfefefeff, %edi	/* set all non-carry bits */
	incl %edi		/* add 1: if one carry bit was *not* set
				   the addition will not result in 0.  */
	jnz L(7)		/* found it => return pointer */

	movl 8(%eax), %ecx	/* get word (= 4 bytes) in question */
	movl $0xfefefeff, %edi	/* magic value */
	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
				   are now 0 */
	addl %ecx, %edi		/* add the magic value to the word.  We get
				   carry bits reported for each byte which
				   is *not* 0 */
	jnc L(6)		/* highest byte is C => return pointer */
	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
	orl $0xfefefeff, %edi	/* set all non-carry bits */
	incl %edi		/* add 1: if one carry bit was *not* set
				   the addition will not result in 0.  */
	jnz L(6)		/* found it => return pointer */

	movl 12(%eax), %ecx	/* get word (= 4 bytes) in question */
	movl $0xfefefeff, %edi	/* magic value */
	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
				   are now 0 */
	addl %ecx, %edi		/* add the magic value to the word.  We get
				   carry bits reported for each byte which
				   is *not* 0 */
	jnc L(5)		/* highest byte is C => return pointer */
	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
	orl $0xfefefeff, %edi	/* set all non-carry bits */
	incl %edi		/* add 1: if one carry bit was *not* set
				   the addition will not result in 0.  */
	jnz L(5)		/* found it => return pointer */

	/* Adjust both counters for a full round, i.e. 16 bytes.  */
	addl $16, %eax
	jmp L(1)
	/* add missing source pointer increments */
L(5):	addl $4, %eax
L(6):	addl $4, %eax
L(7):	addl $4, %eax

	/* Test for the matching byte in the word.  %ecx contains a NUL
	   char in the byte which originally was the byte we are looking
	   at.  */
L(8):	testb %cl, %cl		/* test first byte in dword */
	jz L(9)			/* if zero => return pointer */
	incl %eax		/* increment source pointer */

	testb %ch, %ch		/* test second byte in dword */
	jz L(9)			/* if zero => return pointer */
	incl %eax		/* increment source pointer */

	testl $0xff0000, %ecx	/* test third byte in dword */
	jz L(9)			/* if zero => return pointer */
	incl %eax		/* increment source pointer */

	/* No further test needed we we know it is one of the four bytes.  */

L(9):
	popl %edi		/* pop saved register */
	cfi_adjust_cfa_offset (-4)
	cfi_restore (edi)

	RET_PTR
END (BP_SYM (__rawmemchr))

libc_hidden_def (BP_SYM (__rawmemchr))
weak_alias (BP_SYM (__rawmemchr), BP_SYM (rawmemchr))