1/* The assembly function for memcpy. C-SKY ABIV2 version.
2 Copyright (C) 2018-2024 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library. If not, see
17 <https://www.gnu.org/licenses/>. */
18
19#include <sysdep.h>
20
21ENTRY (memcpy)
22 /* Test if len less than 4 bytes. */
23 mov r12, r0
24 cmplti r2, 4
25 bt .L_copy_by_byte
26
27 andi r13, r0, 3
28 movi r19, 4
29 /* Test if dest is not 4 bytes aligned. */
30 bnez r13, .L_dest_not_aligned
31.L_dest_aligned:
32 /* If dest is aligned, then copy. */
33 zext r18, r2, 31, 4
34 /* Test if len less than 16 bytes. */
35 bez r18, .L_len_less_16bytes
36 movi r19, 0
37
38 LABLE_ALIGN
39.L_len_larger_16bytes:
40#if defined (__CSKY_VDSPV2__)
41 vldx.8 vr0, (r1), r19
42 PRE_BNEZAD (r18)
43 addi r1, 16
44 vstx.8 vr0, (r0), r19
45 addi r0, 16
46#elif defined (__csky_fpuv2__) && defined(__CK810__)
47 fldd fr4, (r1, 0)
48 PRE_BNEZAD (r18)
49 fstd fr4, (r0, 0)
50 fldd fr4, (r1, 8)
51 addi r1, 16
52 fstd fr4, (r0, 8)
53 addi r0, 16
54#elif defined (__CK860__)
55 ldw r3, (r1, 0)
56 stw r3, (r0, 0)
57 ldw r3, (r1, 4)
58 stw r3, (r0, 4)
59 ldw r3, (r1, 8)
60 stw r3, (r0, 8)
61 ldw r3, (r1, 12)
62 addi r1, 16
63 stw r3, (r0, 12)
64 addi r0, 16
65#else
66 ldw r20, (r1, 0)
67 ldw r21, (r1, 4)
68 ldw r22, (r1, 8)
69 ldw r23, (r1, 12)
70 stw r20, (r0, 0)
71 stw r21, (r0, 4)
72 stw r22, (r0, 8)
73 stw r23, (r0, 12)
74 PRE_BNEZAD (r18)
75 addi r1, 16
76 addi r0, 16
77#endif
78 BNEZAD (r18, .L_len_larger_16bytes)
79
80.L_len_less_16bytes:
81 zext r18, r2, 3, 2
82 bez r18, .L_copy_by_byte
83.L_len_less_16bytes_loop:
84 ldw r3, (r1, 0)
85 PRE_BNEZAD (r18)
86 addi r1, 4
87 stw r3, (r0, 0)
88 addi r0, 4
89 BNEZAD (r18, .L_len_less_16bytes_loop)
90
91 /* Test if len less than 4 bytes. */
92.L_copy_by_byte:
93 zext r18, r2, 1, 0
94 bez r18, .L_return
95.L_copy_by_byte_loop:
96 ldb r3, (r1, 0)
97 PRE_BNEZAD (r18)
98 addi r1, 1
99 stb r3, (r0, 0)
100 addi r0, 1
101 BNEZAD (r18, .L_copy_by_byte_loop)
102
103.L_return:
104 mov r0, r12
105 rts
106
107 /* If dest is not aligned, just copying some bytes makes the dest
108 align. */
109
110.L_dest_not_aligned:
111 sub r13, r19, r13
112 mov r19, r13
113.L_dest_not_aligned_loop:
114 /* Makes the dest align. */
115 ldb r3, (r1, 0)
116 PRE_BNEZAD (r13)
117 addi r1, 1
118 stb r3, (r0, 0)
119 addi r0, 1
120 BNEZAD (r13, .L_dest_not_aligned_loop)
121 sub r2, r19
122 cmplti r2, 4
123 bt .L_copy_by_byte
124 /* Check whether the src is aligned. */
125 br .L_dest_aligned
126END (memcpy)
127
128libc_hidden_builtin_def (memcpy)
129.weak memcpy
130
131
132ENTRY (memmove)
133 subu r3, r0, r1
134 cmphs r3, r2
135 bt memcpy
136
137 mov r12, r0
138 addu r0, r0, r2
139 addu r1, r1, r2
140
141 /* Test if len less than 4 bytes. */
142 cmplti r2, 4
143 bt .L_copy_by_byte_m
144
145 andi r13, r0, 3
146 /* Test if dest is not 4 bytes aligned. */
147 bnez r13, .L_dest_not_aligned_m
148.L_dest_aligned_m:
149 /* If dest is aligned, then copy. */
150 zext r18, r2, 31, 4
151 /* Test if len less than 16 bytes. */
152 bez r18, .L_len_less_16bytes_m
153 movi r19, 0
154
155 /* len > 16 bytes */
156 LABLE_ALIGN
157.L_len_larger_16bytes_m:
158 subi r1, 16
159 subi r0, 16
160#if defined (__CSKY_VDSPV2__)
161 vldx.8 vr0, (r1), r19
162 PRE_BNEZAD (r18)
163 vstx.8 vr0, (r0), r19
164#elif defined (__csky_fpuv2__) && defined(__CK810__)
165 fldd fr4, (r1, 8)
166 PRE_BNEZAD (r18)
167 fstd fr4, (r0, 8)
168 fldd fr4, (r1, 0)
169 fstd fr4, (r0, 0)
170#elif defined (__CK860__)
171 ldw r3, (r1, 12)
172 stw r3, (r0, 12)
173 ldw r3, (r1, 8)
174 stw r3, (r0, 8)
175 ldw r3, (r1, 4)
176 stw r3, (r0, 4)
177 ldw r3, (r1, 0)
178 stw r3, (r0, 0)
179#else
180 ldw r20, (r1, 0)
181 ldw r21, (r1, 4)
182 ldw r22, (r1, 8)
183 ldw r23, (r1, 12)
184 stw r20, (r0, 0)
185 stw r21, (r0, 4)
186 stw r22, (r0, 8)
187 stw r23, (r0, 12)
188 PRE_BNEZAD (r18)
189#endif
190 BNEZAD (r18, .L_len_larger_16bytes_m)
191
192.L_len_less_16bytes_m:
193 zext r18, r2, 3, 2
194 bez r18, .L_copy_by_byte_m
195.L_len_less_16bytes_loop_m:
196 subi r1, 4
197 subi r0, 4
198 ldw r3, (r1, 0)
199 PRE_BNEZAD (r18)
200 stw r3, (r0, 0)
201 BNEZAD (r18, .L_len_less_16bytes_loop_m)
202
203 /* Test if len less than 4 bytes. */
204.L_copy_by_byte_m:
205 zext r18, r2, 1, 0
206 bez r18, .L_return_m
207.L_copy_by_byte_loop_m:
208 subi r1, 1
209 subi r0, 1
210 ldb r3, (r1, 0)
211 PRE_BNEZAD (r18)
212 stb r3, (r0, 0)
213 BNEZAD (r18, .L_copy_by_byte_loop_m)
214
215.L_return_m:
216 mov r0, r12
217 rts
218
219 /* If dest is not aligned, just copying some bytes makes the dest
220 align. */
221.L_dest_not_aligned_m:
222 sub r2, r13
223.L_dest_not_aligned_loop_m:
224 subi r1, 1
225 subi r0, 1
226 /* Makes the dest align. */
227 ldb r3, (r1, 0)
228 PRE_BNEZAD (r13)
229 stb r3, (r0, 0)
230 BNEZAD (r13, .L_dest_not_aligned_loop_m)
231 cmplti r2, 4
232 bt .L_copy_by_byte_m
233 /* Check whether the src is aligned. */
234 br .L_dest_aligned_m
235END (memmove)
236
237libc_hidden_builtin_def (memmove)
238.weak memmove
239

source code of glibc/sysdeps/csky/abiv2/memcpy.S