blackfin architecture
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / blackfin / lib / memcpy.S
1 /*
2 * File: arch/blackfin/lib/memcpy.S
3 * Based on:
4 * Author:
5 *
6 * Created:
7 * Description: internal version of memcpy(), issued by the compiler
8 * to copy blocks of data around.
9 * This is really memmove() - it has to be able to deal with
10 * possible overlaps, because that ambiguity is when the compiler
11 * gives up and calls a function. We have our own, internal version
12 * so that we get something we trust, even if the user has redefined
13 * the normal symbol.
14 *
15 * Modified:
16 * Copyright 2004-2006 Analog Devices Inc.
17 *
18 * Bugs: Enter bugs at http://blackfin.uclinux.org/
19 *
20 * This program is free software; you can redistribute it and/or modify
21 * it under the terms of the GNU General Public License as published by
22 * the Free Software Foundation; either version 2 of the License, or
23 * (at your option) any later version.
24 *
25 * This program is distributed in the hope that it will be useful,
26 * but WITHOUT ANY WARRANTY; without even the implied warranty of
27 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
28 * GNU General Public License for more details.
29 *
30 * You should have received a copy of the GNU General Public License
31 * along with this program; if not, see the file COPYING, or write
32 * to the Free Software Foundation, Inc.,
33 * 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
34 */
35
36 #include <linux/linkage.h>
37
38 /* void *memcpy(void *dest, const void *src, size_t n);
39 * R0 = To Address (dest) (leave unchanged to form result)
40 * R1 = From Address (src)
41 * R2 = count
42 *
43 * Note: Favours word alignment
44 */
45
46 #ifdef CONFIG_MEMCPY_L1
47 .section .l1.text
48 #else
49 .text
50 #endif
51
52 .align 2
53
54 ENTRY(_memcpy)
55 CC = R2 <= 0; /* length not positive? */
56 IF CC JUMP .L_P1L2147483647; /* Nothing to do */
57
58 P0 = R0 ; /* dst*/
59 P1 = R1 ; /* src*/
60 P2 = R2 ; /* length */
61
62 /* check for overlapping data */
63 CC = R1 < R0; /* src < dst */
64 IF !CC JUMP .Lno_overlap;
65 R3 = R1 + R2;
66 CC = R0 < R3; /* and dst < src+len */
67 IF CC JUMP .Lhas_overlap;
68
69 .Lno_overlap:
70 /* Check for aligned data.*/
71
72 R3 = R1 | R0;
73 R0 = 0x3;
74 R3 = R3 & R0;
75 CC = R3; /* low bits set on either address? */
76 IF CC JUMP .Lnot_aligned;
77
78 /* Both addresses are word-aligned, so we can copy
79 at least part of the data using word copies.*/
80 P2 = P2 >> 2;
81 CC = P2 <= 2;
82 IF !CC JUMP .Lmore_than_seven;
83 /* less than eight bytes... */
84 P2 = R2;
85 LSETUP(.Lthree_start, .Lthree_end) LC0=P2;
86 R0 = R1; /* setup src address for return */
87 .Lthree_start:
88 R3 = B[P1++] (X);
89 .Lthree_end:
90 B[P0++] = R3;
91
92 RTS;
93
94 .Lmore_than_seven:
95 /* There's at least eight bytes to copy. */
96 P2 += -1; /* because we unroll one iteration */
97 LSETUP(.Lword_loop, .Lword_loop) LC0=P2;
98 R0 = R1;
99 I1 = P1;
100 R3 = [I1++];
101 .Lword_loop:
102 MNOP || [P0++] = R3 || R3 = [I1++];
103
104 [P0++] = R3;
105 /* Any remaining bytes to copy? */
106 R3 = 0x3;
107 R3 = R2 & R3;
108 CC = R3 == 0;
109 P1 = I1; /* in case there's something left, */
110 IF !CC JUMP .Lbytes_left;
111 RTS;
112 .Lbytes_left: P2 = R3;
113 .Lnot_aligned:
114 /* From here, we're copying byte-by-byte. */
115 LSETUP (.Lbyte_start, .Lbyte_end) LC0=P2;
116 R0 = R1; /* Save src address for return */
117 .Lbyte_start:
118 R1 = B[P1++] (X);
119 .Lbyte_end:
120 B[P0++] = R1;
121
122 .L_P1L2147483647:
123 RTS;
124
125 .Lhas_overlap:
126 /* Need to reverse the copying, because the
127 * dst would clobber the src.
128 * Don't bother to work out alignment for
129 * the reverse case.
130 */
131 R0 = R1; /* save src for later. */
132 P0 = P0 + P2;
133 P0 += -1;
134 P1 = P1 + P2;
135 P1 += -1;
136 LSETUP(.Lover_start, .Lover_end) LC0=P2;
137 .Lover_start:
138 R1 = B[P1--] (X);
139 .Lover_end:
140 B[P0--] = R1;
141
142 RTS;