cpu_info.cc 18.5 KB
Newer Older
T
tensor-tang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
// Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "paddle/fluid/lite/core/cpu_info.h"
#include <cstdarg>

namespace paddle {
namespace lite {

#ifdef LITE_WITH_ARM

void DeviceInfo::InitInternal(DeviceInfo* dev) {
  set_default_cache(dev);
  dev->compute_core_num_ = arm_get_cpucount();
  dev->max_memory_ = arm_get_meminfo();

// get max freq
#ifdef LITE_WITH_LINUX
  std::vector<int> max_freq(dev->compute_core_num_);
  for (int i = 0; i < dev->compute_core_num_; ++i) {
    max_freq[i] = get_max_freq_khz(i) / 1000;
  }
  std::string cpu_name = arm_get_cpu_name();
  if (get_cpu_info_from_name(dev, cpu_name) != true) {
    arm_sort_cpuid_by_max_frequency(dev->compute_core_num_, &dev->core_ids_,
                                    max_freq, &dev->cluster_ids_);
    dev->big_core_ids_.clear();
    dev->little_core_ids_.clear();
    for (int i = 0; i < dev->cluster_ids_.size(); ++i) {
      if (dev->cluster_ids_[i] == 0) {
        dev->big_core_ids_.push_back(dev->core_ids_[i]);
      } else {
        dev->little_core_ids_.push_back(dev->core_ids_[i]);
      }
    }
    arm_get_cpu_arch(&dev->archs_);
  }

  LOG(INFO) << "ARM multiprocessors number: " << dev->compute_core_num_;
  for (int i = 0; i < dev->compute_core_num_; ++i) {
    LOG(INFO) << "ARM multiprocessors ID: " << dev->core_ids_[i]
              << ", frequence: " << max_freq[i]
              << ", cluster ID: " << dev->cluster_ids_[dev->core_ids_[i]]
              << ", CPU ARCH: A" << dev->archs_[i];
  }
T
tensor-tang 已提交
57
  VLOG(1) << "L1 DataCache size is: ";
T
tensor-tang 已提交
58
  for (int i = 0; i < dev->compute_core_num_; ++i) {
T
tensor-tang 已提交
59
    VLOG(1) << dev->L1_cache_[i] / 1024 << " KB";
T
tensor-tang 已提交
60
  }
T
tensor-tang 已提交
61
  VLOG(1) << "L2 Cache size is: ";
T
tensor-tang 已提交
62
  for (int i = 0; i < dev->compute_core_num_; ++i) {
T
tensor-tang 已提交
63
    VLOG(1) << dev->L2_cache_[i] / 1024 << " KB";
T
tensor-tang 已提交
64
  }
T
tensor-tang 已提交
65
  VLOG(1) << "Total memory: " << dev->max_memory_ << "KB";
T
tensor-tang 已提交
66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629

  dev->max_freq_ = max_freq[0];
  for (int j = 1; j < dev->compute_core_num_; ++j) {
    if (dev->max_freq_ < max_freq[j]) {
      dev->max_freq_ = max_freq[j];
    }
  }
#elif defined(TARGET_IOS)
  arm_get_cpu_arch(&dev->archs_);
#endif
}

// cache_id : 0 -> L1, 1 -> L2, 2 -> L3
void set_cache_info(DeviceInfo* cpu_info, int cache_id, int argc, ...) {
  va_list arg_ptr;
  va_start(arg_ptr, argc);
  std::vector<int>* cache;
  switch (cache_id) {
    case 0:
      cache = &cpu_info->L1_cache_;
      break;
    case 1:
      cache = &cpu_info->L2_cache_;
      break;
    case 2:
      cache = &cpu_info->L3_cache_;
      break;
    default:
      break;
  }
  int core_num = cpu_info->compute_core_num_;
  cache->resize(core_num);
  if (argc == 1) {
    int cache_size = va_arg(arg_ptr, int);
    for (int i = 0; i < core_num; ++i) {
      (*cache)[i] = cache_size;
    }
  } else {
    int big_core_num = cpu_info->big_core_ids_.size();
    int little_core_num = cpu_info->little_core_ids_.size();
    int big_core_cache_size = va_arg(arg_ptr, int);
    int little_core_cache_size = va_arg(arg_ptr, int);
    for (int i = 0; i < big_core_num; ++i) {
      (*cache)[cpu_info->big_core_ids_[i]] = big_core_cache_size;
    }
    for (int i = 0; i < little_core_num; ++i) {
      (*cache)[cpu_info->little_core_ids_[i]] = little_core_cache_size;
    }
  }
  va_end(arg_ptr);
}

void set_arch_info(DeviceInfo* cpu_info, int argc, ...) {
  va_list arg_ptr;
  va_start(arg_ptr, argc);
  int core_num = cpu_info->compute_core_num_;
  cpu_info->archs_.resize(core_num);
  if (argc == 1) {
    ARMArch arch = (ARMArch)va_arg(arg_ptr, int);
    for (int i = 0; i < core_num; ++i) {
      cpu_info->archs_[i] = arch;
    }
  } else {
    ARMArch big_core_arch = (ARMArch)va_arg(arg_ptr, int);
    ARMArch little_core_arch = (ARMArch)va_arg(arg_ptr, int);
    int big_core_num = cpu_info->big_core_ids_.size();
    int little_core_num = cpu_info->little_core_ids_.size();
    for (int i = 0; i < big_core_num; ++i) {
      cpu_info->archs_[cpu_info->big_core_ids_[i]] = big_core_arch;
    }
    for (int i = 0; i < little_core_num; ++i) {
      cpu_info->archs_[cpu_info->little_core_ids_[i]] = little_core_arch;
    }
  }
  va_end(arg_ptr);
}

bool get_cpu_info_from_name(DeviceInfo* cpu_info, std::string hardware_name) {
  /* Snapdragon */
  if (hardware_name.find("SDM845") != std::string::npos) {  // 845
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {4, 5, 6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 0, 0, 0, 0};
    set_arch_info(cpu_info, 2, kA75, kA55);
    set_cache_info(cpu_info, 0, 1, 32 * 1024);
    set_cache_info(cpu_info, 1, 2, 256 * 1024, 128 * 1024);
    set_cache_info(cpu_info, 2, 1, 2048 * 1024);
    return true;

  } else if (hardware_name.find("SDM710") != std::string::npos) {  // 710
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3, 4, 5};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 1, 1, 0, 0};
    set_arch_info(cpu_info, 2, kA75, kA55);
    return true;
  } else if (hardware_name.find("MSM8998") != std::string::npos) {  // 835
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {4, 5, 6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 0, 0, 0, 0};
    set_arch_info(cpu_info, 2, kA73, kA53);
    set_cache_info(cpu_info, 0, 2, 64 * 1024);
    set_cache_info(cpu_info, 1, 2, 1024 * 1024,
                   /*real cache size is 2M, while that will get bad performace
                      on conv3x3s1 or gemm, set to 1M or 512K*/
                   1024 * 1024);
    return true;

  } else if (hardware_name.find("MSM8996") != std::string::npos) {  // 820
    cpu_info->compute_core_num_ = 4;
    cpu_info->core_ids_ = {0, 1, 2, 3};
    cpu_info->big_core_ids_ = {2, 3};
    cpu_info->little_core_ids_ = {0, 1};
    cpu_info->cluster_ids_ = {1, 1, 0, 0};
    set_arch_info(cpu_info, 1, kA72);
    set_cache_info(cpu_info, 0, 1, 24 * 1024);
    set_cache_info(cpu_info, 1, 2, 1024 * 1024, 512 * 1024);
    return true;

  } else if (hardware_name.find("SDM660") != std::string::npos ||
             hardware_name.find("SDM636") != std::string::npos) {  // 660, 636
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {4, 5, 6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 0, 0, 0, 0};
    set_arch_info(cpu_info, 1, kA73);
    set_cache_info(cpu_info, 0, 2, 64 * 1024, 32 * 1024);
    set_cache_info(cpu_info, 1, 1, 1024 * 1024);
    return true;

  } else if (hardware_name.find("MSM8976") != std::string::npos) {  // 652,653
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {4, 5, 6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 0, 0, 0, 0};
    set_arch_info(cpu_info, 2, kA72, kA53);
    set_cache_info(cpu_info, 0, 1, 32 * 1024);
    set_cache_info(cpu_info, 1, 2, 1024 * 1024, 512 * 1024);
    return true;

  } else if (hardware_name.find("MSM8953") != std::string::npos) {  // 625
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->little_core_ids_ = {};
    cpu_info->cluster_ids_ = {0, 0, 0, 0, 0, 0, 0, 0};
    set_arch_info(cpu_info, 1, kA53);
    set_cache_info(cpu_info, 0, 1, 32 * 1024);
    set_cache_info(cpu_info, 1, 1, 1024 * 1024);
    return true;

  } else if (hardware_name.find("MSM8939") != std::string::npos) {  // 615
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {0, 1, 2, 3};
    cpu_info->little_core_ids_ = {4, 5, 6, 7};
    cpu_info->cluster_ids_ = {0, 0, 0, 0, 1, 1, 1, 1};
    set_arch_info(cpu_info, 1, kA53);
    set_cache_info(cpu_info, 0, 1, 32 * 1024);
    set_cache_info(cpu_info, 1, 2, 512 * 1024, 256 * 1024);
    return true;

    /* MediaTek */

  } else if (hardware_name.find("MT6797") !=
             std::string::npos) {  // X20/X23/X25/X27
    cpu_info->compute_core_num_ = 10;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
    cpu_info->big_core_ids_ = {8, 9};
    cpu_info->little_core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 1, 1, 1, 1, 0, 0};
    set_arch_info(cpu_info, 2, kA72, kA53);
    set_cache_info(cpu_info, 0, 1, 32 * 1024);
    set_cache_info(cpu_info, 1, 2, 1024 * 1024, 512 * 1024);
    return true;

  } else if (hardware_name.find("MT6799") != std::string::npos) {  // X30
    cpu_info->compute_core_num_ = 10;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
    cpu_info->big_core_ids_ = {8, 9};
    cpu_info->little_core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 1, 1, 1, 1, 0, 0};
    set_arch_info(cpu_info, 2, kA73, kA53);
    return true;

  } else if (hardware_name.find("MT6795") != std::string::npos ||
             hardware_name.find("MT6762") != std::string::npos ||
             hardware_name.find("MT6755T") != std::string::npos ||
             hardware_name.find("MT6755S") != std::string::npos ||
             hardware_name.find("MT6753") != std::string::npos ||
             hardware_name.find("MT6752") != std::string::npos ||
             hardware_name.find("MT6750") != std::string::npos) {
    // X10, P22, P15/P18, MT6753, MT6752/MT6752M, MT6750
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->little_core_ids_ = {};
    cpu_info->cluster_ids_ = {0, 0, 0, 0, 0, 0, 0, 0};
    set_arch_info(cpu_info, 1, kA53);
    return true;

  } else if (hardware_name.find("MT6758") != std::string::npos ||
             hardware_name.find("MT6757") != std::string::npos ||
             hardware_name.find("MT6763") != std::string::npos ||
             hardware_name.find("MT6755M") != std::string::npos ||
             hardware_name.find("MT6755") !=
                 std::string::npos) {  // P30, P20/P25, P23, P10
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {4, 5, 6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 0, 0, 0, 0};
    set_arch_info(cpu_info, 1, kA53);
    return true;

  } else if (hardware_name.find("MT6771") != std::string::npos) {  // P60
    cpu_info->compute_core_num_ = 8;
    cpu_info->core_ids_ = {0, 1, 2, 3, 4, 5, 6, 7};
    cpu_info->big_core_ids_ = {4, 5, 6, 7};
    cpu_info->little_core_ids_ = {0, 1, 2, 3};
    cpu_info->cluster_ids_ = {1, 1, 1, 1, 0, 0, 0, 0};
    set_arch_info(cpu_info, 2, kA73, kA53);
    return true;

  } else if (hardware_name.find("MT6765") != std::string::npos ||
             hardware_name.find("MT6739") != std::string::npos ||
             hardware_name.find("MT6738") != std::string::npos ||
             hardware_name.find("MT6737") !=
                 std::string::npos) {  // A22, MT6739, MT6738, MT6767
    cpu_info->compute_core_num_ = 4;
    cpu_info->core_ids_ = {0, 1, 2, 3};
    cpu_info->big_core_ids_ = {0, 0, 0, 0};
    cpu_info->little_core_ids_ = {};
    cpu_info->cluster_ids_ = {0, 0, 0, 0};
    set_arch_info(cpu_info, 1, kA53);
    return true;
  }
  return false;
}

size_t arm_get_meminfo() {
#ifdef LITE_WITH_LINUX
  // get cpu count from /proc/cpuinfo
  FILE* fp = fopen("/proc/meminfo", "rb");
  if (!fp) {
    return 1;
  }

  size_t memsize = 0;
  char line[1024];
  while (!feof(fp)) {
    char* s = fgets(line, 1024, fp);
    if (!s) {
      break;
    }
    sscanf(s, "MemTotal:        %d kB", &memsize);
  }

  fclose(fp);

  return memsize;
#elif defined(TARGET_IOS)
  // to be implemented
  printf("not implemented\n");
  return 0;
#endif
}

int arm_get_cpucount() {
#ifdef LITE_WITH_LINUX
  // get cpu count from /sys/devices/system/cpu/cpunum/uevent
  int max_cpu_count = 20;
  int count = 0;
  for (int i = 0; i < max_cpu_count; ++i) {
    char path[256];
    snprintf(path, sizeof(path), "/sys/devices/system/cpu/cpu%d/uevent", i);
    FILE* fp = fopen(path, "rb");
    if (!fp) {
      break;
    }
    count++;
    fclose(fp);
  }
  if (count < 1) {
    count = 1;
  }
  return count;
#elif defined(TARGET_IOS)
  int count = 0;
  size_t len = sizeof(count);
  sysctlbyname("hw.ncpu", &count, &len, NULL, 0);
  if (count < 1) {
    count = 1;
  }
  return count;
#else
  return 1;
#endif
}

void arm_get_cpu_arch(std::vector<ARMArch>* archs) {
#ifdef LITE_WITH_LINUX
  archs->clear();
  //! get CPU ARCH
  FILE* fp = fopen("/proc/cpuinfo", "rb");
  if (!fp) {
    return;
  }
  char line[1024];
  while (!feof(fp)) {
    char* s = fgets(line, 1024, fp);
    if (!s) {
      break;
    }
    if (strstr(line, "part") != NULL) {
      int arch_id = 0;
      sscanf(s, "CPU part\t: %x", &arch_id);
      switch (arch_id) {
        case 0xd03:
          archs->push_back(kA53);
          break;
        case 0xd05:
          archs->push_back(kA55);
          break;
        case 0xd07:
          archs->push_back(kA57);
          break;
        case 0xd08:
          archs->push_back(kA72);
          break;
        case 0xd09:
          archs->push_back(kA73);
          break;
        case 0xd0a:
          archs->push_back(kA75);
          break;
        case 0x800:
          // 835
          archs->push_back(kA73);
          break;
        case 0x205:
          // 820
          archs->push_back(kA72);
          break;
        default:
          LOG(ERROR) << "unknow type";
          archs->push_back(kARMArch_UNKOWN);
      }
    }
  }
  fclose(fp);
  int cpu_count = arm_get_cpucount();
  if (archs->size() < cpu_count) {
    for (int i = archs->size(); i < cpu_count; ++i) {
      archs->push_back(archs->at(i - 1));
    }
  }
#endif
#ifdef TARGET_IOS
  int cpu_count = arm_get_cpucount();
  for (int i = 0; i < cpu_count; ++i) {
    archs->push_back(APPLE);
  }
#endif
}

#ifdef LITE_WITH_LINUX

void set_default_cache(DeviceInfo* dev) {
  int cpu_count = arm_get_cpucount();
  dev->L1_cache_.resize(cpu_count);
  dev->L2_cache_.resize(cpu_count);
  dev->L3_cache_.resize(cpu_count);
#ifdef TARGET_IOS
  for (int i = 0; i < cpu_count; ++i) {
    dev->L1_cache_[i] = 64 * 1024;
    dev->L2_cache_[i] = 2048 * 1024;
    dev->L3_cache_[i] = 0;
  }
#else
  for (int i = 0; i < cpu_count; ++i) {
    dev->L1_cache_[i] = 32 * 1024;
    dev->L2_cache_[i] = 512 * 1024;
    dev->L3_cache_[i] = 0;
  }
#endif
}
std::string arm_get_cpu_name() {
  FILE* fp = fopen("/proc/cpuinfo", "rb");
  if (!fp) {
    return "";
  }
  char line[1024];
  while (!feof(fp)) {
    char* s = fgets(line, 1024, fp);
    if (!s) {
      break;
    }
    if (strstr(line, "Hardware") != NULL) {
      fclose(fp);
      return std::string(line);
    }
  }
  fclose(fp);
  return "";
}

int get_max_freq_khz(int cpuid) {
  // first try, for all possible cpu
  char path[256];
  snprintf(path, sizeof(path),
           "/sys/devices/system/cpu/cpufreq/stats/cpu%d/time_in_state", cpuid);

  FILE* fp = fopen(path, "rb");

  if (!fp) {
    // second try, for online cpu
    snprintf(path, sizeof(path),
             "/sys/devices/system/cpu/cpu%d/cpufreq/stats/time_in_state",
             cpuid);
    fp = fopen(path, "rb");

    if (!fp) {
      // third try, for online cpu
      snprintf(path, sizeof(path),
               "/sys/devices/system/cpu/cpu%d/cpufreq/cpuinfo_max_freq", cpuid);
      fp = fopen(path, "rb");

      if (!fp) {
        return -1;
      }

      int max_freq_khz = -1;
      fscanf(fp, "%d", &max_freq_khz);

      fclose(fp);

      return max_freq_khz;
    }
  }

  int max_freq_khz = 0;
  while (!feof(fp)) {
    int freq_khz = 0;
    int nscan = fscanf(fp, "%d %*d", &freq_khz);
    if (nscan != 1) {
      break;
    }

    if (freq_khz > max_freq_khz) {
      max_freq_khz = freq_khz;
    }
  }

  fclose(fp);

  return max_freq_khz;
}

int arm_sort_cpuid_by_max_frequency(int cpu_count, std::vector<int>* cpuids,
                                    const std::vector<int>& cpu_freq,
                                    std::vector<int>* cluster_ids) {
  if (cpu_count == 0) {
    return 0;
  }

  cpuids->resize(cpu_count);
  cluster_ids->resize(cpu_count);

  for (int i = 0; i < cpu_count; i++) {
    cpuids->at(i) = i;
  }

  // sort cpuid as big core first
  // simple bubble sort

  for (int i = 0; i < cpu_count; i++) {
    for (int j = i + 1; j < cpu_count; j++) {
      if (cpu_freq[i] < cpu_freq[j]) {
        // swap
        int tmp = cpuids->at(i);
        cpuids->at(i) = cpuids->at(j);
        cpuids->at(j) = tmp;
      }
    }
  }
  // SMP
  int mid_max_freq_khz =
      (cpu_freq[cpuids->at(0)] + cpu_freq[cpuids->at(cpu_count - 1)]) / 2;

  for (int i = 0; i < cpu_count; i++) {
    cpuids->at(i) = i;
    if (cpu_freq[i] >= mid_max_freq_khz) {
      cluster_ids->at(i) = 0;
    } else {
      cluster_ids->at(i) = 1;
    }
  }
  return 0;
}

int check_online(const std::vector<int>& core_ids) {
  if (core_ids.size() == 0) {
    return 0;
  }
  char path[256];
  int online = 1;
  for (int i = 0; i < core_ids.size(); ++i) {
    snprintf(path, sizeof(path), "/sys/devices/system/cpu/cpu%d/online",
             core_ids[i]);
    FILE* fp = fopen(path, "rb");
    if (!fp) {
      return 0;
    }
    int cur_online = 0;
    fscanf(fp, "%d", &cur_online);
    online &= cur_online;
    fclose(fp);
  }
  return online;
}

int set_sched_affinity(const std::vector<int>& cpuids) {
// #define CPU_SETSIZE 1024
// #define __NCPUBITS  (8 * sizeof (unsigned long))
// typedef struct
// {
//    unsigned long __bits[CPU_SETSIZE / __NCPUBITS];
// } cpu_set_t;

// set affinity for thread
#ifdef __GLIBC__
  pid_t pid = syscall(SYS_gettid);
#else
  pid_t pid = gettid();
#endif
  cpu_set_t mask;
  CPU_ZERO(&mask);
  for (int i = 0; i < cpuids.size(); i++) {
    CPU_SET(cpuids[i], &mask);
  }

  int syscallret = syscall(__NR_sched_setaffinity, pid, sizeof(mask), &mask);
  if (syscallret) {
    LOG(ERROR) << "syscall error " << syscallret;
    return -1;
  }

  return 0;
}

#endif  // LITE_WITH_LINUX

#endif  // LITE_WITH_ARM

}  // namespace lite
}  // namespace paddle