Christopher Ferris | 4e24dcc | 2013-07-15 12:49:26 -0700 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (C) 2013 The Android Open Source Project |
| 3 | * All rights reserved. |
| 4 | * |
| 5 | * Redistribution and use in source and binary forms, with or without |
| 6 | * modification, are permitted provided that the following conditions |
| 7 | * are met: |
| 8 | * * Redistributions of source code must retain the above copyright |
| 9 | * notice, this list of conditions and the following disclaimer. |
| 10 | * * Redistributions in binary form must reproduce the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer in |
| 12 | * the documentation and/or other materials provided with the |
| 13 | * distribution. |
| 14 | * |
| 15 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 16 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 17 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
| 18 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
| 19 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
| 20 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
| 21 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS |
| 22 | * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
| 23 | * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
| 24 | * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
| 25 | * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
| 26 | * SUCH DAMAGE. |
| 27 | */ |
| 28 | /* |
| 29 | * Copyright (c) 2013 ARM Ltd |
| 30 | * All rights reserved. |
| 31 | * |
| 32 | * Redistribution and use in source and binary forms, with or without |
| 33 | * modification, are permitted provided that the following conditions |
| 34 | * are met: |
| 35 | * 1. Redistributions of source code must retain the above copyright |
| 36 | * notice, this list of conditions and the following disclaimer. |
| 37 | * 2. Redistributions in binary form must reproduce the above copyright |
| 38 | * notice, this list of conditions and the following disclaimer in the |
| 39 | * documentation and/or other materials provided with the distribution. |
| 40 | * 3. The name of the company may not be used to endorse or promote |
| 41 | * products derived from this software without specific prior written |
| 42 | * permission. |
| 43 | * |
| 44 | * THIS SOFTWARE IS PROVIDED BY ARM LTD ``AS IS'' AND ANY EXPRESS OR IMPLIED |
| 45 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF |
| 46 | * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. |
| 47 | * IN NO EVENT SHALL ARM LTD BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 48 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED |
| 49 | * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
| 50 | * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF |
| 51 | * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING |
| 52 | * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
| 53 | * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 54 | */ |
| 55 | |
| 56 | #include <machine/asm.h> |
| 57 | |
| 58 | .syntax unified |
| 59 | |
| 60 | .thumb |
| 61 | .thumb_func |
| 62 | |
| 63 | .macro m_push |
| 64 | push {r0, r4, r5, lr} |
| 65 | .endm // m_push |
| 66 | |
| 67 | .macro m_ret inst |
| 68 | \inst {r0, r4, r5, pc} |
| 69 | .endm // m_ret |
| 70 | |
| 71 | .macro m_copy_byte reg, cmd, label |
| 72 | ldrb \reg, [r1], #1 |
| 73 | strb \reg, [r0], #1 |
| 74 | \cmd \reg, \label |
| 75 | .endm // m_copy_byte |
| 76 | |
| 77 | ENTRY(strcpy) |
| 78 | // Unroll the first 8 bytes that will be copied. |
| 79 | m_push |
| 80 | m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish |
| 81 | m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish |
| 82 | m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish |
| 83 | m_copy_byte reg=r5, cmd=cbz, label=strcpy_finish |
| 84 | m_copy_byte reg=r2, cmd=cbz, label=strcpy_finish |
| 85 | m_copy_byte reg=r3, cmd=cbz, label=strcpy_finish |
| 86 | m_copy_byte reg=r4, cmd=cbz, label=strcpy_finish |
| 87 | m_copy_byte reg=r5, cmd=cbnz, label=strcpy_continue |
| 88 | |
| 89 | strcpy_finish: |
| 90 | m_ret inst=pop |
| 91 | |
| 92 | strcpy_continue: |
| 93 | pld [r1, #0] |
| 94 | ands r3, r0, #7 |
| 95 | bne strcpy_align_dst |
| 96 | |
| 97 | strcpy_check_src_align: |
| 98 | // At this point dst is aligned to a double word, check if src |
| 99 | // is also aligned to a double word. |
| 100 | ands r3, r1, #7 |
| 101 | bne strcpy_unaligned_copy |
| 102 | |
| 103 | .p2align 2 |
| 104 | strcpy_mainloop: |
| 105 | ldmia r1!, {r2, r3} |
| 106 | |
| 107 | pld [r1, #64] |
| 108 | |
| 109 | sub ip, r2, #0x01010101 |
| 110 | bic ip, ip, r2 |
| 111 | ands ip, ip, #0x80808080 |
| 112 | bne strcpy_zero_in_first_register |
| 113 | |
| 114 | sub ip, r3, #0x01010101 |
| 115 | bic ip, ip, r3 |
| 116 | ands ip, ip, #0x80808080 |
| 117 | bne strcpy_zero_in_second_register |
| 118 | |
| 119 | stmia r0!, {r2, r3} |
| 120 | b strcpy_mainloop |
| 121 | |
| 122 | strcpy_zero_in_first_register: |
| 123 | lsls lr, ip, #17 |
| 124 | itt ne |
| 125 | strbne r2, [r0] |
| 126 | m_ret inst=popne |
| 127 | itt cs |
| 128 | strhcs r2, [r0] |
| 129 | m_ret inst=popcs |
| 130 | lsls ip, ip, #1 |
| 131 | itt eq |
| 132 | streq r2, [r0] |
| 133 | m_ret inst=popeq |
| 134 | strh r2, [r0], #2 |
| 135 | lsr r3, r2, #16 |
| 136 | strb r3, [r0] |
| 137 | m_ret inst=pop |
| 138 | |
| 139 | strcpy_zero_in_second_register: |
| 140 | lsls lr, ip, #17 |
| 141 | ittt ne |
| 142 | stmiane r0!, {r2} |
| 143 | strbne r3, [r0] |
| 144 | m_ret inst=popne |
| 145 | ittt cs |
| 146 | strcs r2, [r0], #4 |
| 147 | strhcs r3, [r0] |
| 148 | m_ret inst=popcs |
| 149 | lsls ip, ip, #1 |
| 150 | itt eq |
| 151 | stmiaeq r0, {r2, r3} |
| 152 | m_ret inst=popeq |
| 153 | stmia r0!, {r2} |
| 154 | strh r3, [r0], #2 |
| 155 | lsr r4, r3, #16 |
| 156 | strb r4, [r0] |
| 157 | m_ret inst=pop |
| 158 | |
| 159 | strcpy_align_dst: |
| 160 | // Align to a double word (64 bits). |
| 161 | rsb r3, r3, #8 |
| 162 | lsls ip, r3, #31 |
| 163 | beq strcpy_align_to_32 |
| 164 | |
| 165 | ldrb r2, [r1], #1 |
| 166 | strb r2, [r0], #1 |
| 167 | cbz r2, strcpy_complete |
| 168 | |
| 169 | strcpy_align_to_32: |
| 170 | bcc strcpy_align_to_64 |
| 171 | |
| 172 | ldrb r4, [r1], #1 |
| 173 | strb r4, [r0], #1 |
| 174 | cmp r4, #0 |
| 175 | it eq |
| 176 | m_ret inst=popeq |
| 177 | ldrb r5, [r1], #1 |
| 178 | strb r5, [r0], #1 |
| 179 | cmp r5, #0 |
| 180 | it eq |
| 181 | m_ret inst=popeq |
| 182 | |
| 183 | strcpy_align_to_64: |
| 184 | tst r3, #4 |
| 185 | beq strcpy_check_src_align |
| 186 | ldr r2, [r1], #4 |
| 187 | |
| 188 | sub ip, r2, #0x01010101 |
| 189 | bic ip, ip, r2 |
| 190 | ands ip, ip, #0x80808080 |
| 191 | bne strcpy_zero_in_first_register |
| 192 | stmia r0!, {r2} |
| 193 | b strcpy_check_src_align |
| 194 | |
| 195 | strcpy_complete: |
| 196 | m_ret inst=pop |
| 197 | |
| 198 | strcpy_unaligned_copy: |
| 199 | // Dst is aligned to a double word, while src is at an unknown alignment. |
| 200 | // There are 7 different versions of the unaligned copy code |
| 201 | // to prevent overreading the src. The mainloop of every single version |
| 202 | // will store 64 bits per loop. The difference is how much of src can |
| 203 | // be read without potentially crossing a page boundary. |
| 204 | tbb [pc, r3] |
| 205 | strcpy_unaligned_branchtable: |
| 206 | .byte 0 |
| 207 | .byte ((strcpy_unalign7 - strcpy_unaligned_branchtable)/2) |
| 208 | .byte ((strcpy_unalign6 - strcpy_unaligned_branchtable)/2) |
| 209 | .byte ((strcpy_unalign5 - strcpy_unaligned_branchtable)/2) |
| 210 | .byte ((strcpy_unalign4 - strcpy_unaligned_branchtable)/2) |
| 211 | .byte ((strcpy_unalign3 - strcpy_unaligned_branchtable)/2) |
| 212 | .byte ((strcpy_unalign2 - strcpy_unaligned_branchtable)/2) |
| 213 | .byte ((strcpy_unalign1 - strcpy_unaligned_branchtable)/2) |
| 214 | |
| 215 | .p2align 2 |
| 216 | // Can read 7 bytes before possibly crossing a page. |
| 217 | strcpy_unalign7: |
| 218 | ldr r2, [r1], #4 |
| 219 | |
| 220 | sub ip, r2, #0x01010101 |
| 221 | bic ip, ip, r2 |
| 222 | ands ip, ip, #0x80808080 |
| 223 | bne strcpy_zero_in_first_register |
| 224 | |
| 225 | ldrb r3, [r1] |
| 226 | cbz r3, strcpy_unalign7_copy5bytes |
| 227 | ldrb r4, [r1, #1] |
| 228 | cbz r4, strcpy_unalign7_copy6bytes |
| 229 | ldrb r5, [r1, #2] |
| 230 | cbz r5, strcpy_unalign7_copy7bytes |
| 231 | |
| 232 | ldr r3, [r1], #4 |
| 233 | pld [r1, #64] |
| 234 | |
| 235 | lsrs ip, r3, #24 |
| 236 | stmia r0!, {r2, r3} |
| 237 | beq strcpy_unalign_return |
| 238 | b strcpy_unalign7 |
| 239 | |
| 240 | strcpy_unalign7_copy5bytes: |
| 241 | stmia r0!, {r2} |
| 242 | strb r3, [r0] |
| 243 | strcpy_unalign_return: |
| 244 | m_ret inst=pop |
| 245 | |
| 246 | strcpy_unalign7_copy6bytes: |
| 247 | stmia r0!, {r2} |
| 248 | strb r3, [r0], #1 |
| 249 | strb r4, [r0], #1 |
| 250 | m_ret inst=pop |
| 251 | |
| 252 | strcpy_unalign7_copy7bytes: |
| 253 | stmia r0!, {r2} |
| 254 | strb r3, [r0], #1 |
| 255 | strb r4, [r0], #1 |
| 256 | strb r5, [r0], #1 |
| 257 | m_ret inst=pop |
| 258 | |
| 259 | .p2align 2 |
| 260 | // Can read 6 bytes before possibly crossing a page. |
| 261 | strcpy_unalign6: |
| 262 | ldr r2, [r1], #4 |
| 263 | |
| 264 | sub ip, r2, #0x01010101 |
| 265 | bic ip, ip, r2 |
| 266 | ands ip, ip, #0x80808080 |
| 267 | bne strcpy_zero_in_first_register |
| 268 | |
| 269 | ldrb r4, [r1] |
| 270 | cbz r4, strcpy_unalign_copy5bytes |
| 271 | ldrb r5, [r1, #1] |
| 272 | cbz r5, strcpy_unalign_copy6bytes |
| 273 | |
| 274 | ldr r3, [r1], #4 |
| 275 | pld [r1, #64] |
| 276 | |
| 277 | tst r3, #0xff0000 |
| 278 | beq strcpy_unalign6_copy7bytes |
| 279 | lsrs ip, r3, #24 |
| 280 | stmia r0!, {r2, r3} |
| 281 | beq strcpy_unalign_return |
| 282 | b strcpy_unalign6 |
| 283 | |
| 284 | strcpy_unalign6_copy7bytes: |
| 285 | stmia r0!, {r2} |
| 286 | strh r3, [r0], #2 |
| 287 | lsr r3, #16 |
| 288 | strb r3, [r0] |
| 289 | m_ret inst=pop |
| 290 | |
| 291 | .p2align 2 |
| 292 | // Can read 5 bytes before possibly crossing a page. |
| 293 | strcpy_unalign5: |
| 294 | ldr r2, [r1], #4 |
| 295 | |
| 296 | sub ip, r2, #0x01010101 |
| 297 | bic ip, ip, r2 |
| 298 | ands ip, ip, #0x80808080 |
| 299 | bne strcpy_zero_in_first_register |
| 300 | |
| 301 | ldrb r4, [r1] |
| 302 | cbz r4, strcpy_unalign_copy5bytes |
| 303 | |
| 304 | ldr r3, [r1], #4 |
| 305 | |
| 306 | pld [r1, #64] |
| 307 | |
| 308 | sub ip, r3, #0x01010101 |
| 309 | bic ip, ip, r3 |
| 310 | ands ip, ip, #0x80808080 |
| 311 | bne strcpy_zero_in_second_register |
| 312 | |
| 313 | stmia r0!, {r2, r3} |
| 314 | b strcpy_unalign5 |
| 315 | |
| 316 | strcpy_unalign_copy5bytes: |
| 317 | stmia r0!, {r2} |
| 318 | strb r4, [r0] |
| 319 | m_ret inst=pop |
| 320 | |
| 321 | strcpy_unalign_copy6bytes: |
| 322 | stmia r0!, {r2} |
| 323 | strb r4, [r0], #1 |
| 324 | strb r5, [r0] |
| 325 | m_ret inst=pop |
| 326 | |
| 327 | .p2align 2 |
| 328 | // Can read 4 bytes before possibly crossing a page. |
| 329 | strcpy_unalign4: |
| 330 | ldmia r1!, {r2} |
| 331 | |
| 332 | sub ip, r2, #0x01010101 |
| 333 | bic ip, ip, r2 |
| 334 | ands ip, ip, #0x80808080 |
| 335 | bne strcpy_zero_in_first_register |
| 336 | |
| 337 | ldmia r1!, {r3} |
| 338 | pld [r1, #64] |
| 339 | |
| 340 | sub ip, r3, #0x01010101 |
| 341 | bic ip, ip, r3 |
| 342 | ands ip, ip, #0x80808080 |
| 343 | bne strcpy_zero_in_second_register |
| 344 | |
| 345 | stmia r0!, {r2, r3} |
| 346 | b strcpy_unalign4 |
| 347 | |
| 348 | .p2align 2 |
| 349 | // Can read 3 bytes before possibly crossing a page. |
| 350 | strcpy_unalign3: |
| 351 | ldrb r2, [r1] |
| 352 | cbz r2, strcpy_unalign3_copy1byte |
| 353 | ldrb r3, [r1, #1] |
| 354 | cbz r3, strcpy_unalign3_copy2bytes |
| 355 | ldrb r4, [r1, #2] |
| 356 | cbz r4, strcpy_unalign3_copy3bytes |
| 357 | |
| 358 | ldr r2, [r1], #4 |
| 359 | ldr r3, [r1], #4 |
| 360 | |
| 361 | pld [r1, #64] |
| 362 | |
| 363 | lsrs lr, r2, #24 |
| 364 | beq strcpy_unalign_copy4bytes |
| 365 | |
| 366 | sub ip, r3, #0x01010101 |
| 367 | bic ip, ip, r3 |
| 368 | ands ip, ip, #0x80808080 |
| 369 | bne strcpy_zero_in_second_register |
| 370 | |
| 371 | stmia r0!, {r2, r3} |
| 372 | b strcpy_unalign3 |
| 373 | |
| 374 | strcpy_unalign3_copy1byte: |
| 375 | strb r2, [r0] |
| 376 | m_ret inst=pop |
| 377 | |
| 378 | strcpy_unalign3_copy2bytes: |
| 379 | strb r2, [r0], #1 |
| 380 | strb r3, [r0] |
| 381 | m_ret inst=pop |
| 382 | |
| 383 | strcpy_unalign3_copy3bytes: |
| 384 | strb r2, [r0], #1 |
| 385 | strb r3, [r0], #1 |
| 386 | strb r4, [r0] |
| 387 | m_ret inst=pop |
| 388 | |
| 389 | .p2align 2 |
| 390 | // Can read 2 bytes before possibly crossing a page. |
| 391 | strcpy_unalign2: |
| 392 | ldrb r2, [r1] |
| 393 | cbz r2, strcpy_unalign_copy1byte |
| 394 | ldrb r3, [r1, #1] |
| 395 | cbz r3, strcpy_unalign_copy2bytes |
| 396 | |
| 397 | ldr r2, [r1], #4 |
| 398 | ldr r3, [r1], #4 |
| 399 | pld [r1, #64] |
| 400 | |
| 401 | tst r2, #0xff0000 |
| 402 | beq strcpy_unalign_copy3bytes |
| 403 | lsrs ip, r2, #24 |
| 404 | beq strcpy_unalign_copy4bytes |
| 405 | |
| 406 | sub ip, r3, #0x01010101 |
| 407 | bic ip, ip, r3 |
| 408 | ands ip, ip, #0x80808080 |
| 409 | bne strcpy_zero_in_second_register |
| 410 | |
| 411 | stmia r0!, {r2, r3} |
| 412 | b strcpy_unalign2 |
| 413 | |
| 414 | .p2align 2 |
| 415 | // Can read 1 byte before possibly crossing a page. |
| 416 | strcpy_unalign1: |
| 417 | ldrb r2, [r1] |
| 418 | cbz r2, strcpy_unalign_copy1byte |
| 419 | |
| 420 | ldr r2, [r1], #4 |
| 421 | ldr r3, [r1], #4 |
| 422 | |
| 423 | pld [r1, #64] |
| 424 | |
| 425 | sub ip, r2, #0x01010101 |
| 426 | bic ip, ip, r2 |
| 427 | ands ip, ip, #0x80808080 |
| 428 | bne strcpy_zero_in_first_register |
| 429 | |
| 430 | sub ip, r3, #0x01010101 |
| 431 | bic ip, ip, r3 |
| 432 | ands ip, ip, #0x80808080 |
| 433 | bne strcpy_zero_in_second_register |
| 434 | |
| 435 | stmia r0!, {r2, r3} |
| 436 | b strcpy_unalign1 |
| 437 | |
| 438 | strcpy_unalign_copy1byte: |
| 439 | strb r2, [r0] |
| 440 | m_ret inst=pop |
| 441 | |
| 442 | strcpy_unalign_copy2bytes: |
| 443 | strb r2, [r0], #1 |
| 444 | strb r3, [r0] |
| 445 | m_ret inst=pop |
| 446 | |
| 447 | strcpy_unalign_copy3bytes: |
| 448 | strh r2, [r0], #2 |
| 449 | lsr r2, #16 |
| 450 | strb r2, [r0] |
| 451 | m_ret inst=pop |
| 452 | |
| 453 | strcpy_unalign_copy4bytes: |
| 454 | stmia r0, {r2} |
| 455 | m_ret inst=pop |
| 456 | END(strcpy) |