yuezonghe | 824eb0c | 2024-06-27 02:32:26 -0700 | [diff] [blame] | 1 | /* Copy SRC to DEST returning DEST. |
| 2 | For SPARC v7. |
| 3 | Copyright (C) 1996, 1999, 2003 Free Software Foundation, Inc. |
| 4 | This file is part of the GNU C Library. |
| 5 | Contributed by Jakub Jelinek <jj@ultra.linux.cz>. |
| 6 | |
| 7 | The GNU C Library is free software; you can redistribute it and/or |
| 8 | modify it under the terms of the GNU Lesser General Public |
| 9 | License as published by the Free Software Foundation; either |
| 10 | version 2.1 of the License, or (at your option) any later version. |
| 11 | |
| 12 | The GNU C Library is distributed in the hope that it will be useful, |
| 13 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 15 | Lesser General Public License for more details. |
| 16 | |
| 17 | You should have received a copy of the GNU Lesser General Public |
| 18 | License along with the GNU C Library; if not, write to the Free |
| 19 | Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA |
| 20 | 02111-1307 USA. */ |
| 21 | |
| 22 | /* Normally, this uses ((xword - 0x01010101) & 0x80808080) test |
| 23 | to find out if any byte in xword could be zero. This is fast, but |
| 24 | also gives false alarm for any byte in range 0x81-0xff. It does |
| 25 | not matter for correctness, as if this test tells us there could |
| 26 | be some zero byte, we check it byte by byte, but if bytes with |
| 27 | high bits set are common in the strings, then this will give poor |
| 28 | performance. You can #define EIGHTBIT_NOT_RARE and the algorithm |
| 29 | will use one tick slower, but more precise test |
| 30 | ((xword - 0x01010101) & (~xword) & 0x80808080), |
| 31 | which does not give any false alarms (but if some bits are set, |
| 32 | one cannot assume from it which bytes are zero and which are not). |
| 33 | It is yet to be measured, what is the correct default for glibc |
| 34 | in these days for an average user. |
| 35 | */ |
| 36 | |
| 37 | .text |
| 38 | .align 4 |
| 39 | 1: ldub [%o1], %o5 |
| 40 | stb %o5, [%o0] |
| 41 | cmp %o5, 0 |
| 42 | add %o0, 1, %o0 |
| 43 | be 0f |
| 44 | add %o1, 1, %o1 |
| 45 | andcc %o1, 3, %g0 |
| 46 | be 4f |
| 47 | or %o4, %lo(0x80808080), %o3 |
| 48 | ldub [%o1], %o5 |
| 49 | stb %o5, [%o0] |
| 50 | cmp %o5, 0 |
| 51 | add %o0, 1, %o0 |
| 52 | be 0f |
| 53 | add %o1, 1, %o1 |
| 54 | andcc %o1, 3, %g0 |
| 55 | be 5f |
| 56 | sethi %hi(0x01010101), %o4 |
| 57 | ldub [%o1], %o5 |
| 58 | stb %o5, [%o0] |
| 59 | cmp %o5, 0 |
| 60 | add %o0, 1, %o0 |
| 61 | be 0f |
| 62 | add %o1, 1, %o1 |
| 63 | b 6f |
| 64 | andcc %o0, 3, %g3 |
| 65 | |
| 66 | ENTRY(strcpy) |
| 67 | mov %o0, %g2 |
| 68 | andcc %o1, 3, %g0 |
| 69 | bne 1b |
| 70 | sethi %hi(0x80808080), %o4 |
| 71 | or %o4, %lo(0x80808080), %o3 |
| 72 | 4: sethi %hi(0x01010101), %o4 |
| 73 | 5: andcc %o0, 3, %g3 |
| 74 | 6: bne 10f |
| 75 | or %o4, %lo(0x01010101), %o2 |
| 76 | 1: ld [%o1], %o5 |
| 77 | add %o1, 4, %o1 |
| 78 | sub %o5, %o2, %o4 |
| 79 | add %o0, 4, %o0 |
| 80 | andcc %o4, %o3, %g0 |
| 81 | be,a 1b |
| 82 | st %o5, [%o0 - 4] |
| 83 | |
| 84 | srl %o5, 24, %g5 |
| 85 | andcc %g5, 0xff, %g0 |
| 86 | be 1f |
| 87 | srl %o5, 16, %g5 |
| 88 | andcc %g5, 0xff, %g0 |
| 89 | be 2f |
| 90 | srl %o5, 8, %g5 |
| 91 | andcc %g5, 0xff, %g0 |
| 92 | be 3f |
| 93 | andcc %o5, 0xff, %g0 |
| 94 | bne 1b |
| 95 | st %o5, [%o0 - 4] |
| 96 | retl |
| 97 | mov %g2, %o0 |
| 98 | 3: srl %o5, 16, %o5 |
| 99 | sth %o5, [%o0 - 4] |
| 100 | stb %g0, [%o0 - 2] |
| 101 | retl |
| 102 | mov %g2, %o0 |
| 103 | 2: srl %o5, 16, %o5 |
| 104 | sth %o5, [%o0 - 4] |
| 105 | retl |
| 106 | mov %g2, %o0 |
| 107 | 1: stb %g0, [%o0 - 4] |
| 108 | retl |
| 109 | mov %g2, %o0 |
| 110 | |
| 111 | 10: ld [%o1], %o5 |
| 112 | add %o1, 4, %o1 |
| 113 | sub %o5, %o2, %o4 |
| 114 | cmp %g3, 2 |
| 115 | be 2f |
| 116 | cmp %g3, 3 |
| 117 | be 3f |
| 118 | andcc %o4, %o3, %g0 |
| 119 | bne 5f |
| 120 | srl %o5, 24, %g5 |
| 121 | stb %g5, [%o0] |
| 122 | sub %o0, 1, %o0 |
| 123 | srl %o5, 8, %g5 |
| 124 | sth %g5, [%o0 + 2] |
| 125 | 1: add %o0, 4, %o0 |
| 126 | 4: sll %o5, 24, %g6 |
| 127 | ld [%o1], %o5 |
| 128 | add %o1, 4, %o1 |
| 129 | srl %o5, 8, %g5 |
| 130 | sub %o5, %o2, %o4 |
| 131 | or %g5, %g6, %g5 |
| 132 | andcc %o4, %o3, %g0 |
| 133 | be,a 1b |
| 134 | st %g5, [%o0] |
| 135 | srl %o5, 24, %o4 |
| 136 | andcc %o4, 0xff, %g0 |
| 137 | be 6f |
| 138 | srl %o5, 16, %o4 |
| 139 | andcc %o4, 0xff, %g0 |
| 140 | be 7f |
| 141 | srl %o5, 8, %o4 |
| 142 | st %g5, [%o0] |
| 143 | andcc %o4, 0xff, %g0 |
| 144 | be 0f |
| 145 | andcc %o5, 0xff, %g0 |
| 146 | 1: bne 4b |
| 147 | add %o0, 4, %o0 |
| 148 | 9: stb %g0, [%o0] |
| 149 | 0: retl |
| 150 | mov %g2, %o0 |
| 151 | 6: srl %g5, 16, %g5 |
| 152 | sth %g5, [%o0] |
| 153 | retl |
| 154 | mov %g2, %o0 |
| 155 | 7: srl %g5, 16, %g5 |
| 156 | sth %g5, [%o0] |
| 157 | stb %g0, [%o0 + 2] |
| 158 | retl |
| 159 | mov %g2, %o0 |
| 160 | 5: andcc %g5, 0xff, %g4 |
| 161 | be 9b |
| 162 | srl %o5, 16, %g5 |
| 163 | andcc %g5, 0xff, %g0 |
| 164 | be 7f |
| 165 | srl %o5, 8, %g5 |
| 166 | andcc %g5, 0xff, %g0 |
| 167 | stb %g4, [%o0] |
| 168 | sth %g5, [%o0 + 1] |
| 169 | sub %o0, 1, %o0 |
| 170 | bne 1b |
| 171 | andcc %o5, 0xff, %g0 |
| 172 | retl |
| 173 | mov %g2, %o0 |
| 174 | 7: stb %g4, [%o0] |
| 175 | stb %g0, [%o0 + 1] |
| 176 | retl |
| 177 | mov %g2, %o0 |
| 178 | |
| 179 | 2: andcc %o4, %o3, %g0 |
| 180 | bne 5f |
| 181 | srl %o5, 16, %g5 |
| 182 | sth %g5, [%o0] |
| 183 | sub %o0, 2, %o0 |
| 184 | 1: add %o0, 4, %o0 |
| 185 | 4: sll %o5, 16, %g6 |
| 186 | ld [%o1], %o5 |
| 187 | add %o1, 4, %o1 |
| 188 | srl %o5, 16, %g5 |
| 189 | sub %o5, %o2, %o4 |
| 190 | or %g5, %g6, %g5 |
| 191 | andcc %o4, %o3, %g0 |
| 192 | be,a 1b |
| 193 | st %g5, [%o0] |
| 194 | srl %o5, 24, %o4 |
| 195 | andcc %o4, 0xff, %g0 |
| 196 | be 7f |
| 197 | srl %o5, 16, %o4 |
| 198 | st %g5, [%o0] |
| 199 | andcc %o4, 0xff, %g0 |
| 200 | be 0b |
| 201 | srl %o5, 8, %o4 |
| 202 | 1: andcc %o4, 0xff, %g0 |
| 203 | be 8f |
| 204 | andcc %o5, 0xff, %g0 |
| 205 | bne 4b |
| 206 | add %o0, 4, %o0 |
| 207 | sth %o5, [%o0] |
| 208 | retl |
| 209 | mov %g2, %o0 |
| 210 | 7: srl %g5, 16, %g5 |
| 211 | sth %g5, [%o0] |
| 212 | stb %g0, [%o0 + 2] |
| 213 | retl |
| 214 | mov %g2, %o0 |
| 215 | 8: stb %g0, [%o0 + 4] |
| 216 | retl |
| 217 | mov %g2, %o0 |
| 218 | 5: srl %o5, 24, %g5 |
| 219 | andcc %g5, 0xff, %g0 |
| 220 | be 9b |
| 221 | srl %o5, 16, %g5 |
| 222 | andcc %g5, 0xff, %g0 |
| 223 | sth %g5, [%o0] |
| 224 | sub %o0, 2, %o0 |
| 225 | bne 1b |
| 226 | srl %o5, 8, %o4 |
| 227 | retl |
| 228 | mov %g2, %o0 |
| 229 | |
| 230 | 3: bne 5f |
| 231 | srl %o5, 24, %g5 |
| 232 | stb %g5, [%o0] |
| 233 | sub %o0, 3, %o0 |
| 234 | 1: add %o0, 4, %o0 |
| 235 | 4: sll %o5, 8, %g6 |
| 236 | ld [%o1], %o5 |
| 237 | add %o1, 4, %o1 |
| 238 | srl %o5, 24, %g5 |
| 239 | sub %o5, %o2, %o4 |
| 240 | or %g5, %g6, %g5 |
| 241 | andcc %o4, %o3, %g0 |
| 242 | be 1b |
| 243 | st %g5, [%o0] |
| 244 | srl %o5, 24, %o4 |
| 245 | andcc %o4, 0xff, %g0 |
| 246 | be 0b |
| 247 | srl %o5, 16, %o4 |
| 248 | 1: andcc %o4, 0xff, %g0 |
| 249 | be 8b |
| 250 | srl %o5, 8, %o4 |
| 251 | andcc %o4, 0xff, %g0 |
| 252 | be 9f |
| 253 | andcc %o5, 0xff, %g0 |
| 254 | bne 4b |
| 255 | add %o0, 4, %o0 |
| 256 | srl %o5, 8, %o5 |
| 257 | sth %o5, [%o0] |
| 258 | stb %g0, [%o0 + 2] |
| 259 | retl |
| 260 | mov %g2, %o0 |
| 261 | 9: srl %o5, 8, %o5 |
| 262 | sth %o5, [%o0 + 4] |
| 263 | retl |
| 264 | mov %g2, %o0 |
| 265 | 5: andcc %g5, 0xff, %g0 |
| 266 | stb %g5, [%o0] |
| 267 | sub %o0, 3, %o0 |
| 268 | bne 1b |
| 269 | srl %o5, 16, %o4 |
| 270 | retl |
| 271 | mov %g2, %o0 |
| 272 | END(strcpy) |
| 273 | libc_hidden_def(strcpy) |