« back to all changes in this revision
Viewing changes to reference/bionic/memcmp.S
- browse files at revision 103
- compare with another revision
- download diff
- download tarball
- view history from revision 103
- Committer: Will Newton
- Date: 2013-04-30 14:31:08 UTC
- Revision ID: will.newton@linaro.org-20130430143108-ww31c741wek8dnus
Split bionic reference code into A15 and A9 versions.
- files added:
- reference/bionic-a15
- reference/bionic-a9
- files removed:
- reference/bionic
- files modified:
- Makefile.am
added
removed
reference/bionic/memcmp.S
1
/*
2
* Copyright (C) 2008 The Android Open Source Project
3
* All rights reserved.
4
*
5
* Redistribution and use in source and binary forms, with or without
6
* modification, are permitted provided that the following conditions
7
* are met:
8
* * Redistributions of source code must retain the above copyright
9
* notice, this list of conditions and the following disclaimer.
10
* * Redistributions in binary form must reproduce the above copyright
11
* notice, this list of conditions and the following disclaimer in
12
* the documentation and/or other materials provided with the
13
* distribution.
14
*
15
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22
* OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23
* AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25
* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26
* SUCH DAMAGE.
27
*/
28
29
.global memcmp
30
.type memcmp, %function
31
.text
32
33
/*
34
* Optimized memcmp() for ARM9.
35
* This would not be optimal on XScale or ARM11, where more prefetching
36
* and use of PLD will be needed.
37
* The 2 major optimzations here are
38
* (1) The main loop compares 16 bytes at a time
39
* (2) The loads are scheduled in a way they won't stall
40
*/
41
42
memcmp:
43
.fnstart
44
PLD [r0, #0]
45
PLD [r1, #0]
46
47
/* take of the case where length is 0 or the buffers are the same */
48
cmp r0, r1
49
cmpne r2, #0
50
moveq r0, #0
51
bxeq lr
52
53
.save {r4, lr}
54
/* save registers */
55
stmfd sp!, {r4, lr}
56
57
PLD [r0, #32]
58
PLD [r1, #32]
59
60
/* since r0 hold the result, move the first source
61
* pointer somewhere else
62
*/
63
64
mov r4, r0
65
66
/* make sure we have at least 8+4 bytes, this simplify things below
67
* and avoid some overhead for small blocks
68
*/
69
cmp r2, #(8+4)
70
bmi 8f
71
72
/* align first pointer to word boundary
73
* offset = -src & 3
74
*/
75
rsb r3, r4, #0
76
ands r3, r3, #3
77
beq 0f
78
79
/* align first pointer */
80
sub r2, r2, r3
81
1: ldrb r0, [r4], #1
82
ldrb ip, [r1], #1
83
subs r0, r0, ip
84
bne 9f
85
subs r3, r3, #1
86
bne 1b
87
88
89
0: /* here the first pointer is aligned, and we have at least 4 bytes
90
* to process.
91
*/
92
93
/* see if the pointers are congruent */
94
eor r0, r4, r1
95
ands r0, r0, #3
96
bne 5f
97
98
/* congruent case, 32 bytes per iteration
99
* We need to make sure there are at least 32+4 bytes left
100
* because we effectively read ahead one word, and we could
101
* read past the buffer (and segfault) if we're not careful.
102
*/
103
104
ldr ip, [r1]
105
subs r2, r2, #(32 + 4)
106
bmi 1f
107
108
0: PLD [r4, #64]
109
PLD [r1, #64]
110
ldr r0, [r4], #4
111
ldr lr, [r1, #4]!
112
eors r0, r0, ip
113
ldreq r0, [r4], #4
114
ldreq ip, [r1, #4]!
115
eoreqs r0, r0, lr
116
ldreq r0, [r4], #4
117
ldreq lr, [r1, #4]!
118
eoreqs r0, r0, ip
119
ldreq r0, [r4], #4
120
ldreq ip, [r1, #4]!
121
eoreqs r0, r0, lr
122
ldreq r0, [r4], #4
123
ldreq lr, [r1, #4]!
124
eoreqs r0, r0, ip
125
ldreq r0, [r4], #4
126
ldreq ip, [r1, #4]!
127
eoreqs r0, r0, lr
128
ldreq r0, [r4], #4
129
ldreq lr, [r1, #4]!
130
eoreqs r0, r0, ip
131
ldreq r0, [r4], #4
132
ldreq ip, [r1, #4]!
133
eoreqs r0, r0, lr
134
bne 2f
135
subs r2, r2, #32
136
bhs 0b
137
138
/* do we have at least 4 bytes left? */
139
1: adds r2, r2, #(32 - 4 + 4)
140
bmi 4f
141
142
/* finish off 4 bytes at a time */
143
3: ldr r0, [r4], #4
144
ldr ip, [r1], #4
145
eors r0, r0, ip
146
bne 2f
147
subs r2, r2, #4
148
bhs 3b
149
150
/* are we done? */
151
4: adds r2, r2, #4
152
moveq r0, #0
153
beq 9f
154
155
/* finish off the remaining bytes */
156
b 8f
157
158
2: /* the last 4 bytes are different, restart them */
159
sub r4, r4, #4
160
sub r1, r1, #4
161
mov r2, #4
162
163
/* process the last few bytes */
164
8: ldrb r0, [r4], #1
165
ldrb ip, [r1], #1
166
// stall
167
subs r0, r0, ip
168
bne 9f
169
subs r2, r2, #1
170
bne 8b
171
172
9: /* restore registers and return */
173
ldmfd sp!, {r4, lr}
174
bx lr
175
.fnend
176
177
178
179
180
181
5: /*************** non-congruent case ***************/
182
and r0, r1, #3
183
cmp r0, #2
184
bne 4f
185
186
/* here, offset is 2 (16-bits aligned, special cased) */
187
188
/* make sure we have at least 16 bytes to process */
189
subs r2, r2, #16
190
addmi r2, r2, #16
191
bmi 8b
192
193
/* align the unaligned pointer */
194
bic r1, r1, #3
195
ldr lr, [r1], #4
196
197
6: PLD [r1, #64]
198
PLD [r4, #64]
199
mov ip, lr, lsr #16
200
ldr lr, [r1], #4
201
ldr r0, [r4], #4
202
orr ip, ip, lr, lsl #16
203
eors r0, r0, ip
204
moveq ip, lr, lsr #16
205
ldreq lr, [r1], #4
206
ldreq r0, [r4], #4
207
orreq ip, ip, lr, lsl #16
208
eoreqs r0, r0, ip
209
moveq ip, lr, lsr #16
210
ldreq lr, [r1], #4
211
ldreq r0, [r4], #4
212
orreq ip, ip, lr, lsl #16
213
eoreqs r0, r0, ip
214
moveq ip, lr, lsr #16
215
ldreq lr, [r1], #4
216
ldreq r0, [r4], #4
217
orreq ip, ip, lr, lsl #16
218
eoreqs r0, r0, ip
219
bne 7f
220
subs r2, r2, #16
221
bhs 6b
222
sub r1, r1, #2
223
/* are we done? */
224
adds r2, r2, #16
225
moveq r0, #0
226
beq 9b
227
/* finish off the remaining bytes */
228
b 8b
229
230
7: /* fix up the 2 pointers and fallthrough... */
231
sub r1, r1, #(4+2)
232
sub r4, r4, #4
233
mov r2, #4
234
b 8b
235
236
237
4: /*************** offset is 1 or 3 (less optimized) ***************/
238
239
stmfd sp!, {r5, r6, r7}
240
241
// r5 = rhs
242
// r6 = lhs
243
// r7 = scratch
244
245
mov r5, r0, lsl #3 /* r5 = right shift */
246
rsb r6, r5, #32 /* r6 = left shift */
247
248
/* align the unaligned pointer */
249
bic r1, r1, #3
250
ldr r7, [r1], #4
251
sub r2, r2, #8
252
253
6: mov ip, r7, lsr r5
254
ldr r7, [r1], #4
255
ldr r0, [r4], #4
256
orr ip, ip, r7, lsl r6
257
eors r0, r0, ip
258
moveq ip, r7, lsr r5
259
ldreq r7, [r1], #4
260
ldreq r0, [r4], #4
261
orreq ip, ip, r7, lsl r6
262
eoreqs r0, r0, ip
263
bne 7f
264
subs r2, r2, #8
265
bhs 6b
266
267
sub r1, r1, r6, lsr #3
268
ldmfd sp!, {r5, r6, r7}
269
270
/* are we done? */
271
adds r2, r2, #8
272
moveq r0, #0
273
beq 9b
274
275
/* finish off the remaining bytes */
276
b 8b
277
278
7: /* fix up the 2 pointers and fallthrough... */
279
sub r1, r1, #4
280
sub r1, r1, r6, lsr #3
281
sub r4, r4, #4
282
mov r2, #4
283
ldmfd sp!, {r5, r6, r7}
284
b 8b
Loggerhead is a web-based interface for Breezy
Version: 2.0.1