« back to all changes in this revision
Viewing changes to src/reference/glibc/memcpy.S
- browse files at revision 1
- compare with another revision
- compare with revision 1
- download diff
- download tarball
- view history from revision 1
- Committer: Michael Hope
- Date: 2010-08-26 22:19:29 UTC
- Revision ID: michael.hope@linaro.org-20100826221929-ppeg01mnpx34aqrp
Pulled in the initial versions
- files added:
- src
- src/reference
- src/reference/bionic
- src/reference/glibc
- src/reference/helpers
- src/reference/newlib
- src/reference/plain
added
removed
src/reference/glibc/memcpy.S
1
/* Copyright (C) 2006, 2009 Free Software Foundation, Inc.
2
This file is part of the GNU C Library.
3
4
Contributed by MontaVista Software, Inc. (written by Nicolas Pitre)
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, write to the Free
18
Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
19
02111-1307 USA. */
20
21
#include <sysdep.h>
22
23
/*
24
* Data preload for architectures that support it (ARM V5TE and above)
25
*/
26
#if (!defined (__ARM_ARCH_2__) && !defined (__ARM_ARCH_3__) \
27
&& !defined (__ARM_ARCH_3M__) && !defined (__ARM_ARCH_4__) \
28
&& !defined (__ARM_ARCH_4T__) && !defined (__ARM_ARCH_5__) \
29
&& !defined (__ARM_ARCH_5T__))
30
#define PLD(code...) code
31
#else
32
#define PLD(code...)
33
#endif
34
35
/*
36
* This can be used to enable code to cacheline align the source pointer.
37
* Experiments on tested architectures (StrongARM and XScale) didn't show
38
* this a worthwhile thing to do. That might be different in the future.
39
*/
40
//#define CALGN(code...) code
41
#define CALGN(code...)
42
43
/*
44
* Endian independent macros for shifting bytes within registers.
45
*/
46
#ifndef __ARMEB__
47
#define pull lsr
48
#define push lsl
49
#else
50
#define pull lsl
51
#define push lsr
52
#endif
53
54
.text
55
56
/* Prototype: void *memcpy(void *dest, const void *src, size_t n); */
57
58
ENTRY(memcpy)
59
60
stmfd sp!, {r0, r4, lr}
61
62
subs r2, r2, #4
63
blt 8f
64
ands ip, r0, #3
65
PLD( pld [r1, #0] )
66
bne 9f
67
ands ip, r1, #3
68
bne 10f
69
70
1: subs r2, r2, #(28)
71
stmfd sp!, {r5 - r8}
72
blt 5f
73
74
CALGN( ands ip, r1, #31 )
75
CALGN( rsb r3, ip, #32 )
76
CALGN( sbcnes r4, r3, r2 ) @ C is always set here
77
CALGN( bcs 2f )
78
CALGN( adr r4, 6f )
79
CALGN( subs r2, r2, r3 ) @ C gets set
80
CALGN( add pc, r4, ip )
81
82
PLD( pld [r1, #0] )
83
2: PLD( subs r2, r2, #96 )
84
PLD( pld [r1, #28] )
85
PLD( blt 4f )
86
PLD( pld [r1, #60] )
87
PLD( pld [r1, #92] )
88
89
3: PLD( pld [r1, #124] )
90
4: ldmia r1!, {r3, r4, r5, r6, r7, r8, ip, lr}
91
subs r2, r2, #32
92
stmia r0!, {r3, r4, r5, r6, r7, r8, ip, lr}
93
bge 3b
94
PLD( cmn r2, #96 )
95
PLD( bge 4b )
96
97
5: ands ip, r2, #28
98
rsb ip, ip, #32
99
addne pc, pc, ip @ C is always clear here
100
b 7f
101
6: nop
102
ldr r3, [r1], #4
103
ldr r4, [r1], #4
104
ldr r5, [r1], #4
105
ldr r6, [r1], #4
106
ldr r7, [r1], #4
107
ldr r8, [r1], #4
108
ldr lr, [r1], #4
109
110
add pc, pc, ip
111
nop
112
nop
113
str r3, [r0], #4
114
str r4, [r0], #4
115
str r5, [r0], #4
116
str r6, [r0], #4
117
str r7, [r0], #4
118
str r8, [r0], #4
119
str lr, [r0], #4
120
121
CALGN( bcs 2b )
122
123
7: ldmfd sp!, {r5 - r8}
124
125
8: movs r2, r2, lsl #31
126
ldrneb r3, [r1], #1
127
ldrcsb r4, [r1], #1
128
ldrcsb ip, [r1]
129
strneb r3, [r0], #1
130
strcsb r4, [r0], #1
131
strcsb ip, [r0]
132
133
#if defined (__ARM_ARCH_4T__) && defined(__THUMB_INTERWORK__)
134
ldmfd sp!, {r0, r4, lr}
135
bx lr
136
#else
137
ldmfd sp!, {r0, r4, pc}
138
#endif
139
140
9: rsb ip, ip, #4
141
cmp ip, #2
142
ldrgtb r3, [r1], #1
143
ldrgeb r4, [r1], #1
144
ldrb lr, [r1], #1
145
strgtb r3, [r0], #1
146
strgeb r4, [r0], #1
147
subs r2, r2, ip
148
strb lr, [r0], #1
149
blt 8b
150
ands ip, r1, #3
151
beq 1b
152
153
10: bic r1, r1, #3
154
cmp ip, #2
155
ldr lr, [r1], #4
156
beq 17f
157
bgt 18f
158
159
160
.macro forward_copy_shift pull push
161
162
subs r2, r2, #28
163
blt 14f
164
165
CALGN( ands ip, r1, #31 )
166
CALGN( rsb ip, ip, #32 )
167
CALGN( sbcnes r4, ip, r2 ) @ C is always set here
168
CALGN( subcc r2, r2, ip )
169
CALGN( bcc 15f )
170
171
11: stmfd sp!, {r5 - r9}
172
173
PLD( pld [r1, #0] )
174
PLD( subs r2, r2, #96 )
175
PLD( pld [r1, #28] )
176
PLD( blt 13f )
177
PLD( pld [r1, #60] )
178
PLD( pld [r1, #92] )
179
180
12: PLD( pld [r1, #124] )
181
13: ldmia r1!, {r4, r5, r6, r7}
182
mov r3, lr, pull #\pull
183
subs r2, r2, #32
184
ldmia r1!, {r8, r9, ip, lr}
185
orr r3, r3, r4, push #\push
186
mov r4, r4, pull #\pull
187
orr r4, r4, r5, push #\push
188
mov r5, r5, pull #\pull
189
orr r5, r5, r6, push #\push
190
mov r6, r6, pull #\pull
191
orr r6, r6, r7, push #\push
192
mov r7, r7, pull #\pull
193
orr r7, r7, r8, push #\push
194
mov r8, r8, pull #\pull
195
orr r8, r8, r9, push #\push
196
mov r9, r9, pull #\pull
197
orr r9, r9, ip, push #\push
198
mov ip, ip, pull #\pull
199
orr ip, ip, lr, push #\push
200
stmia r0!, {r3, r4, r5, r6, r7, r8, r9, ip}
201
bge 12b
202
PLD( cmn r2, #96 )
203
PLD( bge 13b )
204
205
ldmfd sp!, {r5 - r9}
206
207
14: ands ip, r2, #28
208
beq 16f
209
210
15: mov r3, lr, pull #\pull
211
ldr lr, [r1], #4
212
subs ip, ip, #4
213
orr r3, r3, lr, push #\push
214
str r3, [r0], #4
215
bgt 15b
216
CALGN( cmp r2, #0 )
217
CALGN( bge 11b )
218
219
16: sub r1, r1, #(\push / 8)
220
b 8b
221
222
.endm
223
224
225
forward_copy_shift pull=8 push=24
226
227
17: forward_copy_shift pull=16 push=16
228
229
18: forward_copy_shift pull=24 push=8
230
231
END(memcpy)
232
libc_hidden_builtin_def (memcpy)
Loggerhead is a web-based interface for Breezy
Version: 2.0.1