| 1 | /* strlen/strnlen/wcslen/wcsnlen optimized with 256-bit EVEX instructions. |
| 2 | Copyright (C) 2021-2022 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 <isa-level.h> |
| 20 | |
| 21 | #if ISA_SHOULD_BUILD (4) |
| 22 | |
| 23 | # include <sysdep.h> |
| 24 | |
| 25 | # ifndef STRLEN |
| 26 | # define STRLEN __strlen_evex |
| 27 | # endif |
| 28 | |
| 29 | # define VMOVA vmovdqa64 |
| 30 | |
| 31 | # ifdef USE_AS_WCSLEN |
| 32 | # define VPCMP vpcmpd |
| 33 | # define VPMINU vpminud |
| 34 | # define SHIFT_REG ecx |
| 35 | # define CHAR_SIZE 4 |
| 36 | # else |
| 37 | # define VPCMP vpcmpb |
| 38 | # define VPMINU vpminub |
| 39 | # define SHIFT_REG edx |
| 40 | # define CHAR_SIZE 1 |
| 41 | # endif |
| 42 | |
| 43 | # define XMMZERO xmm16 |
| 44 | # define YMMZERO ymm16 |
| 45 | # define YMM1 ymm17 |
| 46 | # define YMM2 ymm18 |
| 47 | # define YMM3 ymm19 |
| 48 | # define YMM4 ymm20 |
| 49 | # define YMM5 ymm21 |
| 50 | # define YMM6 ymm22 |
| 51 | |
| 52 | # define VEC_SIZE 32 |
| 53 | # define PAGE_SIZE 4096 |
| 54 | # define CHAR_PER_VEC (VEC_SIZE / CHAR_SIZE) |
| 55 | |
| 56 | .section .text.evex,"ax" ,@progbits |
| 57 | ENTRY (STRLEN) |
| 58 | # ifdef USE_AS_STRNLEN |
| 59 | /* Check zero length. */ |
| 60 | test %RSI_LP, %RSI_LP |
| 61 | jz L(zero) |
| 62 | # ifdef __ILP32__ |
| 63 | /* Clear the upper 32 bits. */ |
| 64 | movl %esi, %esi |
| 65 | # endif |
| 66 | mov %RSI_LP, %R8_LP |
| 67 | # endif |
| 68 | movl %edi, %eax |
| 69 | vpxorq %XMMZERO, %XMMZERO, %XMMZERO |
| 70 | /* Clear high bits from edi. Only keeping bits relevant to page |
| 71 | cross check. */ |
| 72 | andl $(PAGE_SIZE - 1), %eax |
| 73 | /* Check if we may cross page boundary with one vector load. */ |
| 74 | cmpl $(PAGE_SIZE - VEC_SIZE), %eax |
| 75 | ja L(cross_page_boundary) |
| 76 | |
| 77 | /* Check the first VEC_SIZE bytes. Each bit in K0 represents a |
| 78 | null byte. */ |
| 79 | VPCMP $0, (%rdi), %YMMZERO, %k0 |
| 80 | kmovd %k0, %eax |
| 81 | # ifdef USE_AS_STRNLEN |
| 82 | /* If length < CHAR_PER_VEC handle special. */ |
| 83 | cmpq $CHAR_PER_VEC, %rsi |
| 84 | jbe L(first_vec_x0) |
| 85 | # endif |
| 86 | testl %eax, %eax |
| 87 | jz L(aligned_more) |
| 88 | tzcntl %eax, %eax |
| 89 | ret |
| 90 | # ifdef USE_AS_STRNLEN |
| 91 | L(zero): |
| 92 | xorl %eax, %eax |
| 93 | ret |
| 94 | |
| 95 | .p2align 4 |
| 96 | L(first_vec_x0): |
| 97 | /* Set bit for max len so that tzcnt will return min of max len |
| 98 | and position of first match. */ |
| 99 | btsq %rsi, %rax |
| 100 | tzcntl %eax, %eax |
| 101 | ret |
| 102 | # endif |
| 103 | |
| 104 | .p2align 4 |
| 105 | L(first_vec_x1): |
| 106 | tzcntl %eax, %eax |
| 107 | /* Safe to use 32 bit instructions as these are only called for |
| 108 | size = [1, 159]. */ |
| 109 | # ifdef USE_AS_STRNLEN |
| 110 | /* Use ecx which was computed earlier to compute correct value. |
| 111 | */ |
| 112 | leal -(CHAR_PER_VEC * 4 + 1)(%rcx, %rax), %eax |
| 113 | # else |
| 114 | subl %edx, %edi |
| 115 | # ifdef USE_AS_WCSLEN |
| 116 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 117 | sarl $2, %edi |
| 118 | # endif |
| 119 | leal CHAR_PER_VEC(%rdi, %rax), %eax |
| 120 | # endif |
| 121 | ret |
| 122 | |
| 123 | .p2align 4 |
| 124 | L(first_vec_x2): |
| 125 | tzcntl %eax, %eax |
| 126 | /* Safe to use 32 bit instructions as these are only called for |
| 127 | size = [1, 159]. */ |
| 128 | # ifdef USE_AS_STRNLEN |
| 129 | /* Use ecx which was computed earlier to compute correct value. |
| 130 | */ |
| 131 | leal -(CHAR_PER_VEC * 3 + 1)(%rcx, %rax), %eax |
| 132 | # else |
| 133 | subl %edx, %edi |
| 134 | # ifdef USE_AS_WCSLEN |
| 135 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 136 | sarl $2, %edi |
| 137 | # endif |
| 138 | leal (CHAR_PER_VEC * 2)(%rdi, %rax), %eax |
| 139 | # endif |
| 140 | ret |
| 141 | |
| 142 | .p2align 4 |
| 143 | L(first_vec_x3): |
| 144 | tzcntl %eax, %eax |
| 145 | /* Safe to use 32 bit instructions as these are only called for |
| 146 | size = [1, 159]. */ |
| 147 | # ifdef USE_AS_STRNLEN |
| 148 | /* Use ecx which was computed earlier to compute correct value. |
| 149 | */ |
| 150 | leal -(CHAR_PER_VEC * 2 + 1)(%rcx, %rax), %eax |
| 151 | # else |
| 152 | subl %edx, %edi |
| 153 | # ifdef USE_AS_WCSLEN |
| 154 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 155 | sarl $2, %edi |
| 156 | # endif |
| 157 | leal (CHAR_PER_VEC * 3)(%rdi, %rax), %eax |
| 158 | # endif |
| 159 | ret |
| 160 | |
| 161 | .p2align 4 |
| 162 | L(first_vec_x4): |
| 163 | tzcntl %eax, %eax |
| 164 | /* Safe to use 32 bit instructions as these are only called for |
| 165 | size = [1, 159]. */ |
| 166 | # ifdef USE_AS_STRNLEN |
| 167 | /* Use ecx which was computed earlier to compute correct value. |
| 168 | */ |
| 169 | leal -(CHAR_PER_VEC + 1)(%rcx, %rax), %eax |
| 170 | # else |
| 171 | subl %edx, %edi |
| 172 | # ifdef USE_AS_WCSLEN |
| 173 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 174 | sarl $2, %edi |
| 175 | # endif |
| 176 | leal (CHAR_PER_VEC * 4)(%rdi, %rax), %eax |
| 177 | # endif |
| 178 | ret |
| 179 | |
| 180 | .p2align 5 |
| 181 | L(aligned_more): |
| 182 | movq %rdi, %rdx |
| 183 | /* Align data to VEC_SIZE. */ |
| 184 | andq $-(VEC_SIZE), %rdi |
| 185 | L(cross_page_continue): |
| 186 | /* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time |
| 187 | since data is only aligned to VEC_SIZE. */ |
| 188 | # ifdef USE_AS_STRNLEN |
| 189 | /* + CHAR_SIZE because it simplies the logic in |
| 190 | last_4x_vec_or_less. */ |
| 191 | leaq (VEC_SIZE * 5 + CHAR_SIZE)(%rdi), %rcx |
| 192 | subq %rdx, %rcx |
| 193 | # ifdef USE_AS_WCSLEN |
| 194 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 195 | sarl $2, %ecx |
| 196 | # endif |
| 197 | # endif |
| 198 | /* Load first VEC regardless. */ |
| 199 | VPCMP $0, VEC_SIZE(%rdi), %YMMZERO, %k0 |
| 200 | # ifdef USE_AS_STRNLEN |
| 201 | /* Adjust length. If near end handle specially. */ |
| 202 | subq %rcx, %rsi |
| 203 | jb L(last_4x_vec_or_less) |
| 204 | # endif |
| 205 | kmovd %k0, %eax |
| 206 | testl %eax, %eax |
| 207 | jnz L(first_vec_x1) |
| 208 | |
| 209 | VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMZERO, %k0 |
| 210 | kmovd %k0, %eax |
| 211 | test %eax, %eax |
| 212 | jnz L(first_vec_x2) |
| 213 | |
| 214 | VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMZERO, %k0 |
| 215 | kmovd %k0, %eax |
| 216 | testl %eax, %eax |
| 217 | jnz L(first_vec_x3) |
| 218 | |
| 219 | VPCMP $0, (VEC_SIZE * 4)(%rdi), %YMMZERO, %k0 |
| 220 | kmovd %k0, %eax |
| 221 | testl %eax, %eax |
| 222 | jnz L(first_vec_x4) |
| 223 | |
| 224 | addq $VEC_SIZE, %rdi |
| 225 | # ifdef USE_AS_STRNLEN |
| 226 | /* Check if at last VEC_SIZE * 4 length. */ |
| 227 | cmpq $(CHAR_PER_VEC * 4 - 1), %rsi |
| 228 | jbe L(last_4x_vec_or_less_load) |
| 229 | movl %edi, %ecx |
| 230 | andl $(VEC_SIZE * 4 - 1), %ecx |
| 231 | # ifdef USE_AS_WCSLEN |
| 232 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 233 | sarl $2, %ecx |
| 234 | # endif |
| 235 | /* Readjust length. */ |
| 236 | addq %rcx, %rsi |
| 237 | # endif |
| 238 | /* Align data to VEC_SIZE * 4. */ |
| 239 | andq $-(VEC_SIZE * 4), %rdi |
| 240 | |
| 241 | /* Compare 4 * VEC at a time forward. */ |
| 242 | .p2align 4 |
| 243 | L(loop_4x_vec): |
| 244 | /* Load first VEC regardless. */ |
| 245 | VMOVA (VEC_SIZE * 4)(%rdi), %YMM1 |
| 246 | # ifdef USE_AS_STRNLEN |
| 247 | /* Break if at end of length. */ |
| 248 | subq $(CHAR_PER_VEC * 4), %rsi |
| 249 | jb L(last_4x_vec_or_less_cmpeq) |
| 250 | # endif |
| 251 | /* Save some code size by microfusing VPMINU with the load. Since |
| 252 | the matches in ymm2/ymm4 can only be returned if there where no |
| 253 | matches in ymm1/ymm3 respectively there is no issue with overlap. |
| 254 | */ |
| 255 | VPMINU (VEC_SIZE * 5)(%rdi), %YMM1, %YMM2 |
| 256 | VMOVA (VEC_SIZE * 6)(%rdi), %YMM3 |
| 257 | VPMINU (VEC_SIZE * 7)(%rdi), %YMM3, %YMM4 |
| 258 | |
| 259 | VPCMP $0, %YMM2, %YMMZERO, %k0 |
| 260 | VPCMP $0, %YMM4, %YMMZERO, %k1 |
| 261 | subq $-(VEC_SIZE * 4), %rdi |
| 262 | kortestd %k0, %k1 |
| 263 | jz L(loop_4x_vec) |
| 264 | |
| 265 | /* Check if end was in first half. */ |
| 266 | kmovd %k0, %eax |
| 267 | subq %rdx, %rdi |
| 268 | # ifdef USE_AS_WCSLEN |
| 269 | shrq $2, %rdi |
| 270 | # endif |
| 271 | testl %eax, %eax |
| 272 | jz L(second_vec_return) |
| 273 | |
| 274 | VPCMP $0, %YMM1, %YMMZERO, %k2 |
| 275 | kmovd %k2, %edx |
| 276 | /* Combine VEC1 matches (edx) with VEC2 matches (eax). */ |
| 277 | # ifdef USE_AS_WCSLEN |
| 278 | sall $CHAR_PER_VEC, %eax |
| 279 | orl %edx, %eax |
| 280 | tzcntl %eax, %eax |
| 281 | # else |
| 282 | salq $CHAR_PER_VEC, %rax |
| 283 | orq %rdx, %rax |
| 284 | tzcntq %rax, %rax |
| 285 | # endif |
| 286 | addq %rdi, %rax |
| 287 | ret |
| 288 | |
| 289 | |
| 290 | # ifdef USE_AS_STRNLEN |
| 291 | |
| 292 | L(last_4x_vec_or_less_load): |
| 293 | /* Depending on entry adjust rdi / prepare first VEC in YMM1. */ |
| 294 | VMOVA (VEC_SIZE * 4)(%rdi), %YMM1 |
| 295 | L(last_4x_vec_or_less_cmpeq): |
| 296 | VPCMP $0, %YMM1, %YMMZERO, %k0 |
| 297 | addq $(VEC_SIZE * 3), %rdi |
| 298 | L(last_4x_vec_or_less): |
| 299 | kmovd %k0, %eax |
| 300 | /* If remaining length > VEC_SIZE * 2. This works if esi is off by |
| 301 | VEC_SIZE * 4. */ |
| 302 | testl $(CHAR_PER_VEC * 2), %esi |
| 303 | jnz L(last_4x_vec) |
| 304 | |
| 305 | /* length may have been negative or positive by an offset of |
| 306 | CHAR_PER_VEC * 4 depending on where this was called from. This |
| 307 | fixes that. */ |
| 308 | andl $(CHAR_PER_VEC * 4 - 1), %esi |
| 309 | testl %eax, %eax |
| 310 | jnz L(last_vec_x1_check) |
| 311 | |
| 312 | /* Check the end of data. */ |
| 313 | subl $CHAR_PER_VEC, %esi |
| 314 | jb L(max) |
| 315 | |
| 316 | VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMZERO, %k0 |
| 317 | kmovd %k0, %eax |
| 318 | tzcntl %eax, %eax |
| 319 | /* Check the end of data. */ |
| 320 | cmpl %eax, %esi |
| 321 | jb L(max) |
| 322 | |
| 323 | subq %rdx, %rdi |
| 324 | # ifdef USE_AS_WCSLEN |
| 325 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 326 | sarq $2, %rdi |
| 327 | # endif |
| 328 | leaq (CHAR_PER_VEC * 2)(%rdi, %rax), %rax |
| 329 | ret |
| 330 | L(max): |
| 331 | movq %r8, %rax |
| 332 | ret |
| 333 | # endif |
| 334 | |
| 335 | /* Placed here in strnlen so that the jcc L(last_4x_vec_or_less) |
| 336 | in the 4x VEC loop can use 2 byte encoding. */ |
| 337 | .p2align 4 |
| 338 | L(second_vec_return): |
| 339 | VPCMP $0, %YMM3, %YMMZERO, %k0 |
| 340 | /* Combine YMM3 matches (k0) with YMM4 matches (k1). */ |
| 341 | # ifdef USE_AS_WCSLEN |
| 342 | kunpckbw %k0, %k1, %k0 |
| 343 | kmovd %k0, %eax |
| 344 | tzcntl %eax, %eax |
| 345 | # else |
| 346 | kunpckdq %k0, %k1, %k0 |
| 347 | kmovq %k0, %rax |
| 348 | tzcntq %rax, %rax |
| 349 | # endif |
| 350 | leaq (CHAR_PER_VEC * 2)(%rdi, %rax), %rax |
| 351 | ret |
| 352 | |
| 353 | |
| 354 | # ifdef USE_AS_STRNLEN |
| 355 | L(last_vec_x1_check): |
| 356 | tzcntl %eax, %eax |
| 357 | /* Check the end of data. */ |
| 358 | cmpl %eax, %esi |
| 359 | jb L(max) |
| 360 | subq %rdx, %rdi |
| 361 | # ifdef USE_AS_WCSLEN |
| 362 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 363 | sarq $2, %rdi |
| 364 | # endif |
| 365 | leaq (CHAR_PER_VEC)(%rdi, %rax), %rax |
| 366 | ret |
| 367 | |
| 368 | .p2align 4 |
| 369 | L(last_4x_vec): |
| 370 | /* Test first 2x VEC normally. */ |
| 371 | testl %eax, %eax |
| 372 | jnz L(last_vec_x1) |
| 373 | |
| 374 | VPCMP $0, (VEC_SIZE * 2)(%rdi), %YMMZERO, %k0 |
| 375 | kmovd %k0, %eax |
| 376 | testl %eax, %eax |
| 377 | jnz L(last_vec_x2) |
| 378 | |
| 379 | /* Normalize length. */ |
| 380 | andl $(CHAR_PER_VEC * 4 - 1), %esi |
| 381 | VPCMP $0, (VEC_SIZE * 3)(%rdi), %YMMZERO, %k0 |
| 382 | kmovd %k0, %eax |
| 383 | testl %eax, %eax |
| 384 | jnz L(last_vec_x3) |
| 385 | |
| 386 | /* Check the end of data. */ |
| 387 | subl $(CHAR_PER_VEC * 3), %esi |
| 388 | jb L(max) |
| 389 | |
| 390 | VPCMP $0, (VEC_SIZE * 4)(%rdi), %YMMZERO, %k0 |
| 391 | kmovd %k0, %eax |
| 392 | tzcntl %eax, %eax |
| 393 | /* Check the end of data. */ |
| 394 | cmpl %eax, %esi |
| 395 | jb L(max_end) |
| 396 | |
| 397 | subq %rdx, %rdi |
| 398 | # ifdef USE_AS_WCSLEN |
| 399 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 400 | sarq $2, %rdi |
| 401 | # endif |
| 402 | leaq (CHAR_PER_VEC * 4)(%rdi, %rax), %rax |
| 403 | ret |
| 404 | |
| 405 | .p2align 4 |
| 406 | L(last_vec_x1): |
| 407 | tzcntl %eax, %eax |
| 408 | subq %rdx, %rdi |
| 409 | # ifdef USE_AS_WCSLEN |
| 410 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 411 | sarq $2, %rdi |
| 412 | # endif |
| 413 | leaq (CHAR_PER_VEC)(%rdi, %rax), %rax |
| 414 | ret |
| 415 | |
| 416 | .p2align 4 |
| 417 | L(last_vec_x2): |
| 418 | tzcntl %eax, %eax |
| 419 | subq %rdx, %rdi |
| 420 | # ifdef USE_AS_WCSLEN |
| 421 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 422 | sarq $2, %rdi |
| 423 | # endif |
| 424 | leaq (CHAR_PER_VEC * 2)(%rdi, %rax), %rax |
| 425 | ret |
| 426 | |
| 427 | .p2align 4 |
| 428 | L(last_vec_x3): |
| 429 | tzcntl %eax, %eax |
| 430 | subl $(CHAR_PER_VEC * 2), %esi |
| 431 | /* Check the end of data. */ |
| 432 | cmpl %eax, %esi |
| 433 | jb L(max_end) |
| 434 | subq %rdx, %rdi |
| 435 | # ifdef USE_AS_WCSLEN |
| 436 | /* NB: Divide bytes by 4 to get the wchar_t count. */ |
| 437 | sarq $2, %rdi |
| 438 | # endif |
| 439 | leaq (CHAR_PER_VEC * 3)(%rdi, %rax), %rax |
| 440 | ret |
| 441 | L(max_end): |
| 442 | movq %r8, %rax |
| 443 | ret |
| 444 | # endif |
| 445 | |
| 446 | /* Cold case for crossing page with first load. */ |
| 447 | .p2align 4 |
| 448 | L(cross_page_boundary): |
| 449 | movq %rdi, %rdx |
| 450 | /* Align data to VEC_SIZE. */ |
| 451 | andq $-VEC_SIZE, %rdi |
| 452 | VPCMP $0, (%rdi), %YMMZERO, %k0 |
| 453 | kmovd %k0, %eax |
| 454 | /* Remove the leading bytes. */ |
| 455 | # ifdef USE_AS_WCSLEN |
| 456 | /* NB: Divide shift count by 4 since each bit in K0 represent 4 |
| 457 | bytes. */ |
| 458 | movl %edx, %ecx |
| 459 | shrl $2, %ecx |
| 460 | andl $(CHAR_PER_VEC - 1), %ecx |
| 461 | # endif |
| 462 | /* SHIFT_REG is ecx for USE_AS_WCSLEN and edx otherwise. */ |
| 463 | sarxl %SHIFT_REG, %eax, %eax |
| 464 | testl %eax, %eax |
| 465 | # ifndef USE_AS_STRNLEN |
| 466 | jz L(cross_page_continue) |
| 467 | tzcntl %eax, %eax |
| 468 | ret |
| 469 | # else |
| 470 | jnz L(cross_page_less_vec) |
| 471 | # ifndef USE_AS_WCSLEN |
| 472 | movl %edx, %ecx |
| 473 | andl $(CHAR_PER_VEC - 1), %ecx |
| 474 | # endif |
| 475 | movl $CHAR_PER_VEC, %eax |
| 476 | subl %ecx, %eax |
| 477 | /* Check the end of data. */ |
| 478 | cmpq %rax, %rsi |
| 479 | ja L(cross_page_continue) |
| 480 | movl %esi, %eax |
| 481 | ret |
| 482 | L(cross_page_less_vec): |
| 483 | tzcntl %eax, %eax |
| 484 | /* Select min of length and position of first null. */ |
| 485 | cmpq %rax, %rsi |
| 486 | cmovb %esi, %eax |
| 487 | ret |
| 488 | # endif |
| 489 | |
| 490 | END (STRLEN) |
| 491 | #endif |
| 492 | |