You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

buddy.cc 12 kB

4 years ago
4 years ago
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415
  1. /**
  2. * Copyright 2019 Huawei Technologies Co., Ltd
  3. *
  4. * Licensed under the Apache License, Version 2.0 (the "License");
  5. * you may not use this file except in compliance with the License.
  6. * You may obtain a copy of the License at
  7. *
  8. * http://www.apache.org/licenses/LICENSE-2.0
  9. *
  10. * Unless required by applicable law or agreed to in writing, software
  11. * distributed under the License is distributed on an "AS IS" BASIS,
  12. * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13. * See the License for the specific language governing permissions and
  14. * limitations under the License.
  15. */
  16. #include "minddata/dataset/util/buddy.h"
  17. #include <iomanip>
  18. #include <stdexcept>
  19. #include "minddata/dataset/util/memory_pool.h"
  20. #include "minddata/dataset/util/log_adapter.h"
  21. inline uint64_t BitLeftShift(uint64_t v, uint64_t n) { return (v << n); }
  22. inline uint64_t BitRightShift(uint64_t v, uint64_t n) { return (v >> n); }
  23. inline uint64_t BitOr(uint64_t rhs, uint64_t lhs) { return rhs | lhs; }
  24. inline uint64_t BitEx(uint64_t rhs, uint64_t lhs) { return rhs ^ lhs; }
  25. inline uint64_t BitAnd(uint64_t rhs, uint64_t lhs) { return rhs & lhs; }
  26. namespace mindspore {
  27. namespace dataset {
  28. Status BuddySpace::Init() {
  29. const uint64_t kBitOffset = 3;
  30. const int kLvlMin = 3;
  31. const int kLvlMax = 18;
  32. if (log_min_ < 0) {
  33. return Status(StatusCode::kMDUnexpectedError, __LINE__, __FILE__,
  34. "log_min must be positive : " + std::to_string(log_min_));
  35. }
  36. if (num_lvl_ < kLvlMin || num_lvl_ > kLvlMax) {
  37. return Status(StatusCode::kMDUnexpectedError, __LINE__, __FILE__,
  38. "num_lvl must be between 3 and 18 : " + std::to_string(num_lvl_));
  39. }
  40. min_ = BitLeftShift(1, log_min_);
  41. max_ = BitLeftShift(1, log_min_ + num_lvl_ - 1);
  42. size_t offset_1 = sizeof(rel_addr_t) * num_lvl_;
  43. size_t offset_2 = sizeof(int) * num_lvl_ + offset_1;
  44. size_t offset_3 = sizeof(char) * BitLeftShift(1, static_cast<uint64_t>(num_lvl_ - kBitOffset)) + offset_2;
  45. try {
  46. mem_ = std::make_unique<uint8_t[]>(offset_3);
  47. } catch (const std::bad_alloc &e) {
  48. return Status(StatusCode::kMDOutOfMemory);
  49. }
  50. (void)memset_s(mem_.get(), offset_3, 0, offset_3);
  51. auto ptr = mem_.get();
  52. hint_ = reinterpret_cast<rel_addr_t *>(ptr);
  53. count_ = reinterpret_cast<int *>((reinterpret_cast<char *>(ptr) + offset_1));
  54. map_ = reinterpret_cast<char *>(ptr) + offset_2;
  55. count_[num_lvl_ - 1] = 1;
  56. map_[0] = BitOr(MORE_BIT, static_cast<uint64_t>(num_lvl_ - kBitOffset));
  57. return Status::OK();
  58. }
  59. Status BuddySpace::Alloc(const uint64_t sz, BSpaceDescriptor *desc, addr_t *p) noexcept {
  60. RETURN_UNEXPECTED_IF_NULL(desc);
  61. RETURN_UNEXPECTED_IF_NULL(p);
  62. std::lock_guard<std::mutex> lock(mutex_);
  63. addr_t addr = AllocNoLock(sz, desc);
  64. if (addr != NOSPACE) {
  65. *p = addr;
  66. return Status::OK();
  67. } else {
  68. return Status(StatusCode::kMDBuddySpaceFull, "BuddySpace full. Not an error. Please ignore.");
  69. }
  70. }
  71. addr_t BuddySpace::AllocNoLock(const uint64_t sz, BSpaceDescriptor *desc) noexcept {
  72. MS_ASSERT(sz <= max_);
  73. uint32_t reqSize = SizeToBlock(sz);
  74. rel_addr_t rel_addr = AllocBuddySeg(reqSize);
  75. if (rel_addr != static_cast<rel_addr_t>(NOSPACE)) {
  76. (void)memset_s(desc, sizeof(BSpaceDescriptor), 0, sizeof(BSpaceDescriptor));
  77. desc->sig = static_cast<int>(0xDEADBEEF);
  78. desc->addr = rel_addr;
  79. desc->req_size = reqSize;
  80. desc->blk_size = NextPowerOf2(reqSize);
  81. return static_cast<addr_t>(rel_addr * min_);
  82. } else {
  83. return NOSPACE;
  84. }
  85. }
  86. void BuddySpace::FreeNoLock(const BSpaceDescriptor *desc) {
  87. MS_ASSERT(desc->sig == 0XDEADBEEF);
  88. rel_addr_t rel_addr = desc->addr;
  89. size_t blk_size = desc->blk_size;
  90. size_t req_size = desc->req_size;
  91. FreeBuddySeg(rel_addr, blk_size, req_size);
  92. }
  93. void BuddySpace::Free(const BSpaceDescriptor *desc) {
  94. if (desc == nullptr) {
  95. MS_LOG(ERROR) << "The pointer[desc] is null.";
  96. return;
  97. }
  98. std::lock_guard<std::mutex> lock(mutex_);
  99. return FreeNoLock(desc);
  100. }
  101. std::ostream &operator<<(std::ostream &os, const BuddySpace &s) {
  102. const int32_t kLvlOffset = 4;
  103. os << "1 unit = " << s.GetMinSize() << "\n"
  104. << "Size of buddy space = " << s.GetMaxSize() << "\n"
  105. << "Number of levels = " << s.num_lvl_ << "\n\n"
  106. << "Percent free = " << s.PercentFree() << "\n"
  107. << "Dumping count array : "
  108. << "\n";
  109. for (int i = 0; i < s.num_lvl_; i++) {
  110. os << "[" << i << "] = " << s.count_[i] << " ";
  111. if (((i + 1) % kLvlOffset) == 0) {
  112. os << "\n";
  113. }
  114. }
  115. os << "\n";
  116. os << "Dumping allocation info:"
  117. << "\n";
  118. auto max_addr = static_cast<rel_addr_t>(BitLeftShift(1, s.num_lvl_ - 1));
  119. rel_addr_t addr = 0;
  120. while (addr < max_addr) {
  121. size_t sz = 0;
  122. BuddySpace::STATE st;
  123. s.GetBuddySegState(addr, &sz, &st);
  124. os << "Address : " << std::left << std::setw(8) << addr << " Size : " << std::setw(8) << sz << " State : "
  125. << ((st == BuddySpace::STATE::kAlloc) ? "ALLOC" : ((st == BuddySpace::STATE::kFree) ? "FREE" : "Unknown"))
  126. << "\n";
  127. addr += sz;
  128. }
  129. return os;
  130. }
  131. uint32_t BuddySpace::SizeToBlock(const uint64_t sz) const {
  132. if (min_ == 0) {
  133. MS_LOG(ERROR) << "min_ can not be zero.";
  134. return 0;
  135. }
  136. uint32_t reqSize = (sz / min_);
  137. if (sz % min_) {
  138. reqSize++;
  139. }
  140. return reqSize;
  141. }
  142. void BuddySpace::GetBuddySegState(const rel_addr_t rel_addr, size_t *rel_sz, STATE *st) const {
  143. const int32_t kAddrOffset = 4;
  144. const int32_t kShiftOffset = 2;
  145. const uint64_t kBitLeftShift = 2;
  146. const uint64_t kPosOffset = 2;
  147. char byte;
  148. int pos;
  149. int offset;
  150. uint64_t val = 0;
  151. int shift;
  152. pos = BitRightShift(rel_addr, kPosOffset);
  153. offset = rel_addr % kAddrOffset;
  154. shift = offset * kShiftOffset;
  155. byte = map_[pos];
  156. switch (offset) {
  157. case 0:
  158. val = byte;
  159. break;
  160. case 1:
  161. case 2:
  162. val = BitLeftShift(BitAnd(byte, 0x0F), shift);
  163. break;
  164. case 3:
  165. if (offset == 1) {
  166. val = BitLeftShift(BitAnd(byte, 0x30), shift);
  167. } else {
  168. val = BitLeftShift(BitAnd(byte, 0x03), shift);
  169. }
  170. break;
  171. }
  172. if (BitAnd(val, ONE_BIT)) {
  173. *rel_sz = 1;
  174. } else if (BitAnd(val, TWO_BIT)) {
  175. *rel_sz = 2;
  176. } else if (BitAnd(val, MORE_BIT)) {
  177. log_t lg = BitAnd(val, 0x0F);
  178. *rel_sz = BitLeftShift(1, static_cast<uint64_t>(lg + kBitLeftShift));
  179. } else {
  180. *st = STATE::kEmpty;
  181. return;
  182. }
  183. *st = BitAnd(val, ALLOC_BIT) ? STATE::kAlloc : STATE::kFree;
  184. }
  185. void BuddySpace::SetBuddySegState(rel_addr_t rel_addr, size_t rel_sz, STATE st) {
  186. const int32_t kAddrOffset = 4;
  187. const int32_t kShiftOffset = 2;
  188. const uint64_t kBitOffset = 2;
  189. const uint64_t kPosOffset = 2;
  190. int clr;
  191. int mask;
  192. int pos;
  193. int offset;
  194. int val = 0;
  195. int shift;
  196. auto log_sz = static_cast<log_t>(Log2(rel_sz));
  197. pos = BitRightShift(rel_addr, kPosOffset);
  198. offset = rel_addr % kAddrOffset;
  199. shift = offset * kShiftOffset;
  200. if (rel_sz == 1) {
  201. val = ONE_BIT;
  202. mask = 0xC0;
  203. } else if (rel_sz == 2) {
  204. val = TWO_BIT;
  205. mask = 0xF0;
  206. } else {
  207. val = BitOr(static_cast<uint64_t>(log_sz - kBitOffset), MORE_BIT);
  208. mask = 0xFF;
  209. }
  210. if (st == STATE::kAlloc) {
  211. val = BitOr(val, ALLOC_BIT);
  212. } else if (st == STATE::kFree) {
  213. val = BitAnd(val, ~(static_cast<uint64_t>(ALLOC_BIT)));
  214. } else if (st == STATE::kEmpty) {
  215. val = 0;
  216. }
  217. clr = static_cast<int>(~(BitRightShift(mask, shift)));
  218. map_[pos] = static_cast<char>(BitAnd(map_[pos], clr));
  219. map_[pos] = static_cast<char>(BitOr(map_[pos], BitRightShift(val, shift)));
  220. if (st == STATE::kAlloc) {
  221. count_[log_sz]--;
  222. } else if (st == STATE::kFree) {
  223. count_[log_sz]++;
  224. if (rel_addr < hint_[log_sz]) {
  225. hint_[log_sz] = rel_addr;
  226. }
  227. }
  228. }
  229. void BuddySpace::JoinBuddySeg(rel_addr_t addr, size_t blk_sz) {
  230. const int32_t kLogszStep = 2;
  231. while (blk_sz < BitLeftShift(1, num_lvl_)) {
  232. rel_addr_t buddy = BitEx(addr, blk_sz);
  233. size_t sz = 0;
  234. STATE st;
  235. GetBuddySegState(buddy, &sz, &st);
  236. if (st == STATE::kFree && sz == blk_sz) {
  237. auto log_sz = static_cast<log_t>(Log2(blk_sz));
  238. rel_addr_t left = (buddy < addr) ? buddy : addr;
  239. rel_addr_t right = left + blk_sz;
  240. MS_ASSERT(count_[log_sz] >= 2);
  241. count_[log_sz] -= kLogszStep;
  242. SetBuddySegState(right, blk_sz, STATE::kEmpty);
  243. SetBuddySegState(left, BitLeftShift(blk_sz, 1), STATE::kFree);
  244. for (int i = 0; i < log_sz; i++) {
  245. if (hint_[i] == right) {
  246. hint_[i] = left;
  247. }
  248. }
  249. addr = left;
  250. blk_sz <<= 1u;
  251. } else {
  252. break;
  253. }
  254. }
  255. }
  256. void BuddySpace::TrimBuddySeg(rel_addr_t addr, size_t blk_sz, size_t ask_sz) {
  257. MS_ASSERT(ask_sz < blk_sz);
  258. uint32_t inx = Log2(blk_sz);
  259. size_t remaining_sz = ask_sz;
  260. for (int i = inx; i > 0; i--) {
  261. size_t b_size = BitLeftShift(1, i);
  262. size_t half_sz = BitRightShift(b_size, 1);
  263. count_[i]--;
  264. SetBuddySegState(addr, half_sz, STATE::kFree);
  265. SetBuddySegState(addr + half_sz, half_sz, STATE::kFree);
  266. if (remaining_sz >= half_sz) {
  267. SetBuddySegState(addr, half_sz, STATE::kAlloc);
  268. remaining_sz -= half_sz;
  269. if (remaining_sz == 0) {
  270. break;
  271. }
  272. addr += half_sz;
  273. }
  274. }
  275. }
  276. void BuddySpace::UnTrimBuddySeg(rel_addr_t addr, size_t blk_sz, size_t ask_sz) {
  277. MS_ASSERT(ask_sz < blk_sz);
  278. uint32_t inx = Log2(blk_sz);
  279. size_t remaining_sz = ask_sz;
  280. for (int i = inx; i > 0; i--) {
  281. size_t b_size = BitLeftShift(1, i);
  282. size_t half_sz = BitRightShift(b_size, 1);
  283. if (remaining_sz >= half_sz) {
  284. #ifdef DEBUG
  285. {
  286. size_t sz = 0;
  287. STATE st;
  288. GetBuddySegState(addr, &sz, &st);
  289. MS_ASSERT(sz == half_sz && st == STATE::kAlloc);
  290. }
  291. #endif
  292. SetBuddySegState(addr, half_sz, STATE::kFree);
  293. remaining_sz -= half_sz;
  294. if (remaining_sz == 0) {
  295. JoinBuddySeg(addr, half_sz);
  296. break;
  297. }
  298. addr += half_sz;
  299. }
  300. }
  301. }
  302. rel_addr_t BuddySpace::AllocBuddySeg(uint32_t req_size) noexcept {
  303. uint32_t blk_size = NextPowerOf2(req_size);
  304. int start_inx = static_cast<int>(Log2(blk_size));
  305. bool found = false;
  306. rel_addr_t ask_addr = 0;
  307. auto max_addr = static_cast<rel_addr_t>(BitLeftShift(1, num_lvl_ - 1));
  308. STATE st;
  309. size_t sz = 0;
  310. for (int i = start_inx; !found && i < num_lvl_; i++) {
  311. MS_ASSERT(count_[i] >= 0);
  312. if (count_[i] == 0) {
  313. continue;
  314. }
  315. auto blk_sz = static_cast<size_t>(BitLeftShift(1, i));
  316. ask_addr = hint_[i];
  317. while (ask_addr < max_addr && !found) {
  318. GetBuddySegState(ask_addr, &sz, &st);
  319. if (st == STATE::kFree && sz == blk_sz) {
  320. found = true;
  321. } else {
  322. MS_ASSERT(st != STATE::kEmpty);
  323. ask_addr += ((sz > blk_sz) ? sz : blk_sz);
  324. }
  325. }
  326. }
  327. if (found) {
  328. if (sz > req_size) {
  329. TrimBuddySeg(ask_addr, sz, req_size);
  330. } else {
  331. SetBuddySegState(ask_addr, sz, STATE::kAlloc);
  332. hint_[start_inx] = ask_addr;
  333. }
  334. return ask_addr;
  335. } else {
  336. return static_cast<rel_addr_t>(NOSPACE);
  337. }
  338. }
  339. void BuddySpace::FreeBuddySeg(rel_addr_t addr, size_t blk_size, size_t req_size) {
  340. if (req_size == blk_size) {
  341. #ifdef DEBUG
  342. {
  343. size_t sz = 0;
  344. STATE st;
  345. GetBuddySegState(addr, &sz, &st);
  346. }
  347. #endif
  348. SetBuddySegState(addr, blk_size, STATE::kFree);
  349. JoinBuddySeg(addr, blk_size);
  350. } else {
  351. UnTrimBuddySeg(addr, blk_size, req_size);
  352. }
  353. }
  354. int BuddySpace::PercentFree() const {
  355. const int32_t kFloatToPercent = 100;
  356. uint64_t total_free_sz = 0;
  357. uint64_t max_sz_in_unit = BitLeftShift(1, num_lvl_ - 1);
  358. // Go through the count array without lock
  359. for (int i = 0; i < num_lvl_; i++) {
  360. int cnt = count_[i];
  361. if (cnt == 0) {
  362. continue;
  363. }
  364. uint64_t blk_sz = BitLeftShift(1, i);
  365. total_free_sz += (blk_sz * cnt);
  366. }
  367. return static_cast<int>(static_cast<float>(total_free_sz) / static_cast<float>(max_sz_in_unit) * kFloatToPercent);
  368. }
  369. BuddySpace::BuddySpace(int log_min, int num_lvl)
  370. : hint_(nullptr), count_(nullptr), map_(nullptr), log_min_(log_min), num_lvl_(num_lvl), min_(0), max_(0) {}
  371. BuddySpace::~BuddySpace() {
  372. hint_ = nullptr;
  373. count_ = nullptr;
  374. map_ = nullptr;
  375. }
  376. Status BuddySpace::CreateBuddySpace(std::unique_ptr<BuddySpace> *out_bs, int log_min, int num_lvl) {
  377. Status rc;
  378. auto bs = new (std::nothrow) BuddySpace(log_min, num_lvl);
  379. if (bs == nullptr) {
  380. return Status(StatusCode::kMDOutOfMemory);
  381. }
  382. rc = bs->Init();
  383. if (rc.IsOk()) {
  384. (*out_bs).reset(bs);
  385. } else {
  386. delete bs;
  387. }
  388. return rc;
  389. }
  390. } // namespace dataset
  391. } // namespace mindspore