1 /* $NetBSD: bcopy_page.S,v 1.7 2003/10/13 21:03:13 scw Exp $ */
2
3 /*-
4 * Copyright (c) 1995 Scott Stevens
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. All advertising materials mentioning features or use of this software
16 * must display the following acknowledgement:
17 * This product includes software developed by Scott Stevens.
18 * 4. The name of the author may not be used to endorse or promote products
19 * derived from this software without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 *
32 * RiscBSD kernel project
33 *
34 * bcopy_page.S
35 *
36 * page optimised bcopy and bzero routines
37 *
38 * Created : 08/04/95
39 */
40
41 #include <machine/asm.h>
42
43 __FBSDID("$FreeBSD: releng/10.0/sys/arm/arm/bcopy_page.S 248361 2013-03-16 02:48:49Z andrew $");
44
45 #include "assym.s"
46
47 #ifndef _ARM_ARCH_5E
48
49 /* #define BIG_LOOPS */
50
51 /*
52 * bcopy_page(src, dest)
53 *
54 * Optimised copy page routine.
55 *
56 * On entry:
57 * r0 - src address
58 * r1 - dest address
59 *
60 * Requires:
61 * number of bytes per page (PAGE_SIZE) is a multiple of 512 (BIG_LOOPS), 128
62 * otherwise.
63 */
64
65 #define CHUNK_SIZE 32
66
67 #define PREFETCH_FIRST_CHUNK /* nothing */
68 #define PREFETCH_NEXT_CHUNK /* nothing */
69
70 #ifndef COPY_CHUNK
71 #define COPY_CHUNK \
72 PREFETCH_NEXT_CHUNK ; \
73 ldmia r0!, {r3-r8,ip,lr} ; \
74 stmia r1!, {r3-r8,ip,lr}
75 #endif /* ! COPY_CHUNK */
76
77 #ifndef SAVE_REGS
78 #define SAVE_REGS stmfd sp!, {r4-r8, lr}
79 #define RESTORE_REGS ldmfd sp!, {r4-r8, pc}
80 #endif
81
82 ENTRY(bcopy_page)
83 PREFETCH_FIRST_CHUNK
84 SAVE_REGS
85 #ifdef BIG_LOOPS
86 mov r2, #(PAGE_SIZE >> 9)
87 #else
88 mov r2, #(PAGE_SIZE >> 7)
89 #endif
90
91 1:
92 COPY_CHUNK
93 COPY_CHUNK
94 COPY_CHUNK
95 COPY_CHUNK
96
97 #ifdef BIG_LOOPS
98 /* There is little point making the loop any larger; unless we are
99 running with the cache off, the load/store overheads will
100 completely dominate this loop. */
101 COPY_CHUNK
102 COPY_CHUNK
103 COPY_CHUNK
104 COPY_CHUNK
105
106 COPY_CHUNK
107 COPY_CHUNK
108 COPY_CHUNK
109 COPY_CHUNK
110
111 COPY_CHUNK
112 COPY_CHUNK
113 COPY_CHUNK
114 COPY_CHUNK
115 #endif
116 subs r2, r2, #1
117 bne 1b
118
119 RESTORE_REGS /* ...and return. */
120 END(bcopy_page)
121
122 /*
123 * bzero_page(dest)
124 *
125 * Optimised zero page routine.
126 *
127 * On entry:
128 * r0 - dest address
129 *
130 * Requires:
131 * number of bytes per page (PAGE_SIZE) is a multiple of 512 (BIG_LOOPS), 128
132 * otherwise
133 */
134
135 ENTRY(bzero_page)
136 stmfd sp!, {r4-r8, lr}
137 #ifdef BIG_LOOPS
138 mov r2, #(PAGE_SIZE >> 9)
139 #else
140 mov r2, #(PAGE_SIZE >> 7)
141 #endif
142 mov r3, #0
143 mov r4, #0
144 mov r5, #0
145 mov r6, #0
146 mov r7, #0
147 mov r8, #0
148 mov ip, #0
149 mov lr, #0
150
151 1:
152 stmia r0!, {r3-r8,ip,lr}
153 stmia r0!, {r3-r8,ip,lr}
154 stmia r0!, {r3-r8,ip,lr}
155 stmia r0!, {r3-r8,ip,lr}
156
157 #ifdef BIG_LOOPS
158 /* There is little point making the loop any larger; unless we are
159 running with the cache off, the load/store overheads will
160 completely dominate this loop. */
161 stmia r0!, {r3-r8,ip,lr}
162 stmia r0!, {r3-r8,ip,lr}
163 stmia r0!, {r3-r8,ip,lr}
164 stmia r0!, {r3-r8,ip,lr}
165
166 stmia r0!, {r3-r8,ip,lr}
167 stmia r0!, {r3-r8,ip,lr}
168 stmia r0!, {r3-r8,ip,lr}
169 stmia r0!, {r3-r8,ip,lr}
170
171 stmia r0!, {r3-r8,ip,lr}
172 stmia r0!, {r3-r8,ip,lr}
173 stmia r0!, {r3-r8,ip,lr}
174 stmia r0!, {r3-r8,ip,lr}
175
176 #endif
177
178 subs r2, r2, #1
179 bne 1b
180
181 ldmfd sp!, {r4-r8, pc}
182 END(bzero_page)
183
184 #else /* _ARM_ARCH_5E */
185
186 /*
187 * armv5e version of bcopy_page
188 */
189 ENTRY(bcopy_page)
190 pld [r0]
191 stmfd sp!, {r4, r5}
192 mov ip, #32
193 ldr r2, [r0], #0x04 /* 0x00 */
194 ldr r3, [r0], #0x04 /* 0x04 */
195 1: pld [r0, #0x18] /* Prefetch 0x20 */
196 ldr r4, [r0], #0x04 /* 0x08 */
197 ldr r5, [r0], #0x04 /* 0x0c */
198 strd r2, [r1], #0x08
199 ldr r2, [r0], #0x04 /* 0x10 */
200 ldr r3, [r0], #0x04 /* 0x14 */
201 strd r4, [r1], #0x08
202 ldr r4, [r0], #0x04 /* 0x18 */
203 ldr r5, [r0], #0x04 /* 0x1c */
204 strd r2, [r1], #0x08
205 ldr r2, [r0], #0x04 /* 0x20 */
206 ldr r3, [r0], #0x04 /* 0x24 */
207 pld [r0, #0x18] /* Prefetch 0x40 */
208 strd r4, [r1], #0x08
209 ldr r4, [r0], #0x04 /* 0x28 */
210 ldr r5, [r0], #0x04 /* 0x2c */
211 strd r2, [r1], #0x08
212 ldr r2, [r0], #0x04 /* 0x30 */
213 ldr r3, [r0], #0x04 /* 0x34 */
214 strd r4, [r1], #0x08
215 ldr r4, [r0], #0x04 /* 0x38 */
216 ldr r5, [r0], #0x04 /* 0x3c */
217 strd r2, [r1], #0x08
218 ldr r2, [r0], #0x04 /* 0x40 */
219 ldr r3, [r0], #0x04 /* 0x44 */
220 pld [r0, #0x18] /* Prefetch 0x60 */
221 strd r4, [r1], #0x08
222 ldr r4, [r0], #0x04 /* 0x48 */
223 ldr r5, [r0], #0x04 /* 0x4c */
224 strd r2, [r1], #0x08
225 ldr r2, [r0], #0x04 /* 0x50 */
226 ldr r3, [r0], #0x04 /* 0x54 */
227 strd r4, [r1], #0x08
228 ldr r4, [r0], #0x04 /* 0x58 */
229 ldr r5, [r0], #0x04 /* 0x5c */
230 strd r2, [r1], #0x08
231 ldr r2, [r0], #0x04 /* 0x60 */
232 ldr r3, [r0], #0x04 /* 0x64 */
233 pld [r0, #0x18] /* Prefetch 0x80 */
234 strd r4, [r1], #0x08
235 ldr r4, [r0], #0x04 /* 0x68 */
236 ldr r5, [r0], #0x04 /* 0x6c */
237 strd r2, [r1], #0x08
238 ldr r2, [r0], #0x04 /* 0x70 */
239 ldr r3, [r0], #0x04 /* 0x74 */
240 strd r4, [r1], #0x08
241 ldr r4, [r0], #0x04 /* 0x78 */
242 ldr r5, [r0], #0x04 /* 0x7c */
243 strd r2, [r1], #0x08
244 subs ip, ip, #0x01
245 ldrgt r2, [r0], #0x04 /* 0x80 */
246 ldrgt r3, [r0], #0x04 /* 0x84 */
247 strd r4, [r1], #0x08
248 bgt 1b
249 ldmfd sp!, {r4, r5}
250 RET
251 END(bcopy_page)
252
253 /*
254 * armv5e version of bzero_page
255 */
256 ENTRY(bzero_page)
257 mov r1, #PAGE_SIZE
258 mov r2, #0
259 mov r3, #0
260 1: strd r2, [r0], #8 /* 32 */
261 strd r2, [r0], #8
262 strd r2, [r0], #8
263 strd r2, [r0], #8
264 strd r2, [r0], #8 /* 64 */
265 strd r2, [r0], #8
266 strd r2, [r0], #8
267 strd r2, [r0], #8
268 strd r2, [r0], #8 /* 96 */
269 strd r2, [r0], #8
270 strd r2, [r0], #8
271 strd r2, [r0], #8
272 strd r2, [r0], #8 /* 128 */
273 strd r2, [r0], #8
274 strd r2, [r0], #8
275 strd r2, [r0], #8
276 subs r1, r1, #128
277 bne 1b
278 RET
279 END(bzero_page)
280 #endif /* _ARM_ARCH_5E */
Cache object: 49581b19356fbe1b0ad04eabf87bf9ef
|