| 1 | // Copyright (c) 2014, the Dart project authors. Please see the AUTHORS file |
| 2 | // for details. All rights reserved. Use of this source code is governed by a |
| 3 | // BSD-style license that can be found in the LICENSE file. |
| 4 | |
| 5 | #ifndef RUNTIME_VM_HASH_TABLE_H_ |
| 6 | #define RUNTIME_VM_HASH_TABLE_H_ |
| 7 | |
| 8 | #include "platform/assert.h" |
| 9 | #include "vm/object.h" |
| 10 | |
| 11 | namespace dart { |
| 12 | |
| 13 | // OVERVIEW: |
| 14 | // |
| 15 | // Hash maps and hash sets all use RawArray as backing storage. At the lowest |
| 16 | // level is a generic open-addressing table that supports deletion. |
| 17 | // - HashTable |
| 18 | // The next layer provides ordering and iteration functionality: |
| 19 | // - UnorderedHashTable |
| 20 | // - LinkedListHashTable (TODO(koda): Implement.) |
| 21 | // The utility class HashTables handles growth and conversion. |
| 22 | // The next layer fixes the payload size and provides a natural interface: |
| 23 | // - HashMap |
| 24 | // - HashSet |
| 25 | // Combining either of these with an iteration strategy, we get the templates |
| 26 | // intended for use outside this file: |
| 27 | // - UnorderedHashMap |
| 28 | // - LinkedListHashMap |
| 29 | // - UnorderedHashSet |
| 30 | // - LinkedListHashSet |
| 31 | // Each of these can be finally specialized with KeyTraits to support any set of |
| 32 | // lookup key types (e.g., look up a char* in a set of String objects), and |
| 33 | // any equality and hash code computation. |
| 34 | // |
| 35 | // The classes all wrap an Array handle, and methods like HashSet::Insert can |
| 36 | // trigger growth into a new RawArray, updating the handle. Debug mode asserts |
| 37 | // that 'Release' was called once to access the final array before destruction. |
| 38 | // NOTE: The handle returned by 'Release' is cleared by ~HashTable. |
| 39 | // |
| 40 | // Example use: |
| 41 | // typedef UnorderedHashMap<FooTraits> FooMap; |
| 42 | // ... |
| 43 | // FooMap cache(get_foo_cache()); |
| 44 | // cache.UpdateOrInsert(name0, obj0); |
| 45 | // cache.UpdateOrInsert(name1, obj1); |
| 46 | // ... |
| 47 | // set_foo_cache(cache.Release()); |
| 48 | // |
| 49 | // If you *know* that no mutating operations were called, you can optimize: |
| 50 | // ... |
| 51 | // obj ^= cache.GetOrNull(name); |
| 52 | // ASSERT(cache.Release().raw() == get_foo_cache()); |
| 53 | // |
| 54 | // TODO(koda): When exposing these to Dart code, document and assert that |
| 55 | // KeyTraits methods must not run Dart code (since the C++ code doesn't check |
| 56 | // for concurrent modification). |
| 57 | |
| 58 | // Open-addressing hash table template using a RawArray as backing storage. |
| 59 | // |
| 60 | // The elements of the array are partitioned into entries: |
| 61 | // [ header | metadata | entry0 | entry1 | ... | entryN ] |
| 62 | // Each entry contains a key, followed by zero or more payload components, |
| 63 | // and has 3 possible states: unused, occupied, or deleted. |
| 64 | // The header tracks the number of entries in each state. |
| 65 | // Any object except the backing storage array and Object::transition_sentinel() |
| 66 | // may be stored as a key. Any object may be stored in a payload. |
| 67 | // |
| 68 | // Parameters |
| 69 | // KeyTraits: defines static methods |
| 70 | // bool IsMatch(const Key& key, const Object& obj) and |
| 71 | // uword Hash(const Key& key) for any number of desired lookup key types. |
| 72 | // kPayloadSize: number of components of the payload in each entry. |
| 73 | // kMetaDataSize: number of elements reserved (e.g., for iteration order data). |
| 74 | template <typename KeyTraits, intptr_t kPayloadSize, intptr_t kMetaDataSize> |
| 75 | class HashTable : public ValueObject { |
| 76 | public: |
| 77 | typedef KeyTraits Traits; |
| 78 | // Uses the passed in handles for all handle operations. |
| 79 | // 'Release' must be called at the end to obtain the final table |
| 80 | // after potential growth/shrinkage. |
| 81 | HashTable(Object* key, Smi* index, Array* data) |
| 82 | : key_handle_(key), |
| 83 | smi_handle_(index), |
| 84 | data_(data), |
| 85 | released_data_(NULL) {} |
| 86 | // Uses 'zone' for handle allocation. 'Release' must be called at the end |
| 87 | // to obtain the final table after potential growth/shrinkage. |
| 88 | HashTable(Zone* zone, ArrayPtr data) |
| 89 | : key_handle_(&Object::Handle(zone)), |
| 90 | smi_handle_(&Smi::Handle(zone)), |
| 91 | data_(&Array::Handle(zone, data)), |
| 92 | released_data_(NULL) {} |
| 93 | |
| 94 | // Returns the final table. The handle is cleared when this HashTable is |
| 95 | // destroyed. |
| 96 | Array& Release() { |
| 97 | ASSERT(data_ != NULL); |
| 98 | ASSERT(released_data_ == NULL); |
| 99 | // Ensure that no methods are called after 'Release'. |
| 100 | released_data_ = data_; |
| 101 | data_ = NULL; |
| 102 | return *released_data_; |
| 103 | } |
| 104 | |
| 105 | ~HashTable() { |
| 106 | // In DEBUG mode, calling 'Release' is mandatory. |
| 107 | ASSERT(data_ == NULL); |
| 108 | if (released_data_ != NULL) { |
| 109 | *released_data_ = Array::null(); |
| 110 | } |
| 111 | } |
| 112 | |
| 113 | // Returns a backing storage size such that 'num_occupied' distinct keys can |
| 114 | // be inserted into the table. |
| 115 | static intptr_t ArrayLengthForNumOccupied(intptr_t num_occupied) { |
| 116 | // Because we use quadratic (actually triangle number) probing it is |
| 117 | // important that the size is a power of two (otherwise we could fail to |
| 118 | // find an empty slot). This is described in Knuth's The Art of Computer |
| 119 | // Programming Volume 2, Chapter 6.4, exercise 20 (solution in the |
| 120 | // appendix, 2nd edition). |
| 121 | intptr_t num_entries = Utils::RoundUpToPowerOfTwo(num_occupied + 1); |
| 122 | return kFirstKeyIndex + (kEntrySize * num_entries); |
| 123 | } |
| 124 | |
| 125 | // Initializes an empty table. |
| 126 | void Initialize() const { |
| 127 | ASSERT(data_->Length() >= ArrayLengthForNumOccupied(0)); |
| 128 | *smi_handle_ = Smi::New(0); |
| 129 | data_->SetAt(kOccupiedEntriesIndex, *smi_handle_); |
| 130 | data_->SetAt(kDeletedEntriesIndex, *smi_handle_); |
| 131 | |
| 132 | #if !defined(PRODUCT) |
| 133 | data_->SetAt(kNumGrowsIndex, *smi_handle_); |
| 134 | data_->SetAt(kNumLT5LookupsIndex, *smi_handle_); |
| 135 | data_->SetAt(kNumLT25LookupsIndex, *smi_handle_); |
| 136 | data_->SetAt(kNumGT25LookupsIndex, *smi_handle_); |
| 137 | data_->SetAt(kNumProbesIndex, *smi_handle_); |
| 138 | #endif // !defined(PRODUCT) |
| 139 | |
| 140 | for (intptr_t i = kHeaderSize; i < data_->Length(); ++i) { |
| 141 | data_->SetAt(i, UnusedMarker()); |
| 142 | } |
| 143 | } |
| 144 | |
| 145 | // Returns whether 'key' matches any key in the table. |
| 146 | template <typename Key> |
| 147 | bool ContainsKey(const Key& key) const { |
| 148 | return FindKey(key) != -1; |
| 149 | } |
| 150 | |
| 151 | // Returns the entry that matches 'key', or -1 if none exists. |
| 152 | template <typename Key> |
| 153 | intptr_t FindKey(const Key& key) const { |
| 154 | const intptr_t num_entries = NumEntries(); |
| 155 | ASSERT(NumOccupied() < num_entries); |
| 156 | // TODO(koda): Add salt. |
| 157 | NOT_IN_PRODUCT(intptr_t collisions = 0;) |
| 158 | uword hash = KeyTraits::Hash(key); |
| 159 | ASSERT(Utils::IsPowerOfTwo(num_entries)); |
| 160 | intptr_t probe = hash & (num_entries - 1); |
| 161 | int probe_distance = 1; |
| 162 | while (true) { |
| 163 | if (IsUnused(probe)) { |
| 164 | NOT_IN_PRODUCT(UpdateCollisions(collisions);) |
| 165 | return -1; |
| 166 | } else if (!IsDeleted(probe)) { |
| 167 | *key_handle_ = GetKey(probe); |
| 168 | if (KeyTraits::IsMatch(key, *key_handle_)) { |
| 169 | NOT_IN_PRODUCT(UpdateCollisions(collisions);) |
| 170 | return probe; |
| 171 | } |
| 172 | NOT_IN_PRODUCT(collisions += 1;) |
| 173 | } |
| 174 | // Advance probe. See ArrayLengthForNumOccupied comment for |
| 175 | // explanation of how we know this hits all slots. |
| 176 | probe = (probe + probe_distance) & (num_entries - 1); |
| 177 | probe_distance++; |
| 178 | } |
| 179 | UNREACHABLE(); |
| 180 | return -1; |
| 181 | } |
| 182 | |
| 183 | // Sets *entry to either: |
| 184 | // - an occupied entry matching 'key', and returns true, or |
| 185 | // - an unused/deleted entry where a matching key may be inserted, |
| 186 | // and returns false. |
| 187 | template <typename Key> |
| 188 | bool FindKeyOrDeletedOrUnused(const Key& key, intptr_t* entry) const { |
| 189 | const intptr_t num_entries = NumEntries(); |
| 190 | ASSERT(entry != NULL); |
| 191 | ASSERT(NumOccupied() < num_entries); |
| 192 | NOT_IN_PRODUCT(intptr_t collisions = 0;) |
| 193 | uword hash = KeyTraits::Hash(key); |
| 194 | ASSERT(Utils::IsPowerOfTwo(num_entries)); |
| 195 | intptr_t probe = hash & (num_entries - 1); |
| 196 | int probe_distance = 1; |
| 197 | intptr_t deleted = -1; |
| 198 | while (true) { |
| 199 | if (IsUnused(probe)) { |
| 200 | *entry = (deleted != -1) ? deleted : probe; |
| 201 | NOT_IN_PRODUCT(UpdateCollisions(collisions);) |
| 202 | return false; |
| 203 | } else if (IsDeleted(probe)) { |
| 204 | if (deleted == -1) { |
| 205 | deleted = probe; |
| 206 | } |
| 207 | } else { |
| 208 | *key_handle_ = GetKey(probe); |
| 209 | if (KeyTraits::IsMatch(key, *key_handle_)) { |
| 210 | *entry = probe; |
| 211 | NOT_IN_PRODUCT(UpdateCollisions(collisions);) |
| 212 | return true; |
| 213 | } |
| 214 | NOT_IN_PRODUCT(collisions += 1;) |
| 215 | } |
| 216 | // Advance probe. See ArrayLengthForNumOccupied comment for |
| 217 | // explanation of how we know this hits all slots. |
| 218 | probe = (probe + probe_distance) & (num_entries - 1); |
| 219 | probe_distance++; |
| 220 | } |
| 221 | UNREACHABLE(); |
| 222 | return false; |
| 223 | } |
| 224 | |
| 225 | // Sets the key of a previously unoccupied entry. This must not be the last |
| 226 | // unoccupied entry. |
| 227 | void InsertKey(intptr_t entry, const Object& key) const { |
| 228 | ASSERT(key.raw() != UnusedMarker().raw()); |
| 229 | ASSERT(key.raw() != DeletedMarker().raw()); |
| 230 | ASSERT(!IsOccupied(entry)); |
| 231 | AdjustSmiValueAt(kOccupiedEntriesIndex, 1); |
| 232 | if (IsDeleted(entry)) { |
| 233 | AdjustSmiValueAt(kDeletedEntriesIndex, -1); |
| 234 | } else { |
| 235 | ASSERT(IsUnused(entry)); |
| 236 | } |
| 237 | InternalSetKey(entry, key); |
| 238 | ASSERT(IsOccupied(entry)); |
| 239 | ASSERT(NumOccupied() < NumEntries()); |
| 240 | } |
| 241 | |
| 242 | const Object& UnusedMarker() const { return Object::transition_sentinel(); } |
| 243 | const Object& DeletedMarker() const { return *data_; } |
| 244 | |
| 245 | bool IsUnused(intptr_t entry) const { |
| 246 | return InternalGetKey(entry) == UnusedMarker().raw(); |
| 247 | } |
| 248 | bool IsOccupied(intptr_t entry) const { |
| 249 | return !IsUnused(entry) && !IsDeleted(entry); |
| 250 | } |
| 251 | bool IsDeleted(intptr_t entry) const { |
| 252 | return InternalGetKey(entry) == DeletedMarker().raw(); |
| 253 | } |
| 254 | |
| 255 | ObjectPtr GetKey(intptr_t entry) const { |
| 256 | ASSERT(IsOccupied(entry)); |
| 257 | return InternalGetKey(entry); |
| 258 | } |
| 259 | ObjectPtr GetPayload(intptr_t entry, intptr_t component) const { |
| 260 | ASSERT(IsOccupied(entry)); |
| 261 | return data_->At(PayloadIndex(entry, component)); |
| 262 | } |
| 263 | void UpdatePayload(intptr_t entry, |
| 264 | intptr_t component, |
| 265 | const Object& value) const { |
| 266 | ASSERT(IsOccupied(entry)); |
| 267 | ASSERT(0 <= component && component < kPayloadSize); |
| 268 | data_->SetAt(PayloadIndex(entry, component), value); |
| 269 | } |
| 270 | // Deletes both the key and payload of the specified entry. |
| 271 | void DeleteEntry(intptr_t entry) const { |
| 272 | ASSERT(IsOccupied(entry)); |
| 273 | for (intptr_t i = 0; i < kPayloadSize; ++i) { |
| 274 | UpdatePayload(entry, i, DeletedMarker()); |
| 275 | } |
| 276 | InternalSetKey(entry, DeletedMarker()); |
| 277 | AdjustSmiValueAt(kOccupiedEntriesIndex, -1); |
| 278 | AdjustSmiValueAt(kDeletedEntriesIndex, 1); |
| 279 | } |
| 280 | intptr_t NumEntries() const { |
| 281 | return (data_->Length() - kFirstKeyIndex) / kEntrySize; |
| 282 | } |
| 283 | intptr_t NumUnused() const { |
| 284 | return NumEntries() - NumOccupied() - NumDeleted(); |
| 285 | } |
| 286 | intptr_t NumOccupied() const { return GetSmiValueAt(kOccupiedEntriesIndex); } |
| 287 | intptr_t NumDeleted() const { return GetSmiValueAt(kDeletedEntriesIndex); } |
| 288 | Object& KeyHandle() const { return *key_handle_; } |
| 289 | Smi& SmiHandle() const { return *smi_handle_; } |
| 290 | |
| 291 | #if !defined(PRODUCT) |
| 292 | intptr_t NumGrows() const { return GetSmiValueAt(kNumGrowsIndex); } |
| 293 | intptr_t NumLT5Collisions() const { |
| 294 | return GetSmiValueAt(kNumLT5LookupsIndex); |
| 295 | } |
| 296 | intptr_t NumLT25Collisions() const { |
| 297 | return GetSmiValueAt(kNumLT25LookupsIndex); |
| 298 | } |
| 299 | intptr_t NumGT25Collisions() const { |
| 300 | return GetSmiValueAt(kNumGT25LookupsIndex); |
| 301 | } |
| 302 | intptr_t NumProbes() const { return GetSmiValueAt(kNumProbesIndex); } |
| 303 | void UpdateGrowth() const { |
| 304 | if (KeyTraits::ReportStats()) { |
| 305 | AdjustSmiValueAt(kNumGrowsIndex, 1); |
| 306 | } |
| 307 | } |
| 308 | void UpdateCollisions(intptr_t collisions) const { |
| 309 | if (KeyTraits::ReportStats()) { |
| 310 | if (data_->raw()->ptr()->InVMIsolateHeap()) { |
| 311 | return; |
| 312 | } |
| 313 | AdjustSmiValueAt(kNumProbesIndex, collisions + 1); |
| 314 | if (collisions < 5) { |
| 315 | AdjustSmiValueAt(kNumLT5LookupsIndex, 1); |
| 316 | } else if (collisions < 25) { |
| 317 | AdjustSmiValueAt(kNumLT25LookupsIndex, 1); |
| 318 | } else { |
| 319 | AdjustSmiValueAt(kNumGT25LookupsIndex, 1); |
| 320 | } |
| 321 | } |
| 322 | } |
| 323 | void PrintStats() const { |
| 324 | if (!KeyTraits::ReportStats()) { |
| 325 | return; |
| 326 | } |
| 327 | const intptr_t num5 = NumLT5Collisions(); |
| 328 | const intptr_t num25 = NumLT25Collisions(); |
| 329 | const intptr_t num_more = NumGT25Collisions(); |
| 330 | // clang-format off |
| 331 | OS::PrintErr("Stats for %s table :\n" |
| 332 | " Size of table = %" Pd ",Number of Occupied entries = %" Pd "\n" |
| 333 | " Number of Grows = %" Pd "\n" |
| 334 | " Number of lookups with < 5 collisions = %" Pd "\n" |
| 335 | " Number of lookups with < 25 collisions = %" Pd "\n" |
| 336 | " Number of lookups with > 25 collisions = %" Pd "\n" |
| 337 | " Average number of probes = %g\n" , |
| 338 | KeyTraits::Name(), |
| 339 | NumEntries(), NumOccupied(), NumGrows(), |
| 340 | num5, num25, num_more, |
| 341 | static_cast<double>(NumProbes()) / (num5 + num25 + num_more)); |
| 342 | // clang-format on |
| 343 | } |
| 344 | #endif // !PRODUCT |
| 345 | |
| 346 | protected: |
| 347 | static const intptr_t kOccupiedEntriesIndex = 0; |
| 348 | static const intptr_t kDeletedEntriesIndex = 1; |
| 349 | #if defined(PRODUCT) |
| 350 | static const intptr_t kHeaderSize = kDeletedEntriesIndex + 1; |
| 351 | #else |
| 352 | static const intptr_t kNumGrowsIndex = 2; |
| 353 | static const intptr_t kNumLT5LookupsIndex = 3; |
| 354 | static const intptr_t kNumLT25LookupsIndex = 4; |
| 355 | static const intptr_t kNumGT25LookupsIndex = 5; |
| 356 | static const intptr_t kNumProbesIndex = 6; |
| 357 | static const intptr_t = kNumProbesIndex + 1; |
| 358 | #endif |
| 359 | static const intptr_t kMetaDataIndex = kHeaderSize; |
| 360 | static const intptr_t kFirstKeyIndex = kHeaderSize + kMetaDataSize; |
| 361 | static const intptr_t kEntrySize = 1 + kPayloadSize; |
| 362 | |
| 363 | intptr_t KeyIndex(intptr_t entry) const { |
| 364 | ASSERT(0 <= entry && entry < NumEntries()); |
| 365 | return kFirstKeyIndex + (kEntrySize * entry); |
| 366 | } |
| 367 | |
| 368 | intptr_t PayloadIndex(intptr_t entry, intptr_t component) const { |
| 369 | ASSERT(0 <= component && component < kPayloadSize); |
| 370 | return KeyIndex(entry) + 1 + component; |
| 371 | } |
| 372 | |
| 373 | ObjectPtr InternalGetKey(intptr_t entry) const { |
| 374 | return data_->At(KeyIndex(entry)); |
| 375 | } |
| 376 | |
| 377 | void InternalSetKey(intptr_t entry, const Object& key) const { |
| 378 | data_->SetAt(KeyIndex(entry), key); |
| 379 | } |
| 380 | |
| 381 | intptr_t GetSmiValueAt(intptr_t index) const { |
| 382 | ASSERT(!data_->IsNull()); |
| 383 | ASSERT(!data_->At(index)->IsHeapObject()); |
| 384 | return Smi::Value(Smi::RawCast(data_->At(index))); |
| 385 | } |
| 386 | |
| 387 | void SetSmiValueAt(intptr_t index, intptr_t value) const { |
| 388 | *smi_handle_ = Smi::New(value); |
| 389 | data_->SetAt(index, *smi_handle_); |
| 390 | } |
| 391 | |
| 392 | void AdjustSmiValueAt(intptr_t index, intptr_t delta) const { |
| 393 | SetSmiValueAt(index, (GetSmiValueAt(index) + delta)); |
| 394 | } |
| 395 | |
| 396 | Object* key_handle_; |
| 397 | Smi* smi_handle_; |
| 398 | // Exactly one of these is non-NULL, depending on whether Release was called. |
| 399 | Array* data_; |
| 400 | Array* released_data_; |
| 401 | |
| 402 | friend class HashTables; |
| 403 | }; |
| 404 | |
| 405 | // Table with unspecified iteration order. No payload overhead or metadata. |
| 406 | template <typename KeyTraits, intptr_t kUserPayloadSize> |
| 407 | class UnorderedHashTable : public HashTable<KeyTraits, kUserPayloadSize, 0> { |
| 408 | public: |
| 409 | typedef HashTable<KeyTraits, kUserPayloadSize, 0> BaseTable; |
| 410 | static const intptr_t kPayloadSize = kUserPayloadSize; |
| 411 | explicit UnorderedHashTable(ArrayPtr data) |
| 412 | : BaseTable(Thread::Current()->zone(), data) {} |
| 413 | UnorderedHashTable(Zone* zone, ArrayPtr data) : BaseTable(zone, data) {} |
| 414 | UnorderedHashTable(Object* key, Smi* value, Array* data) |
| 415 | : BaseTable(key, value, data) {} |
| 416 | // Note: Does not check for concurrent modification. |
| 417 | class Iterator { |
| 418 | public: |
| 419 | explicit Iterator(const UnorderedHashTable* table) |
| 420 | : table_(table), entry_(-1) {} |
| 421 | bool MoveNext() { |
| 422 | while (entry_ < (table_->NumEntries() - 1)) { |
| 423 | ++entry_; |
| 424 | if (table_->IsOccupied(entry_)) { |
| 425 | return true; |
| 426 | } |
| 427 | } |
| 428 | return false; |
| 429 | } |
| 430 | intptr_t Current() { return entry_; } |
| 431 | |
| 432 | private: |
| 433 | const UnorderedHashTable* table_; |
| 434 | intptr_t entry_; |
| 435 | }; |
| 436 | |
| 437 | // No extra book-keeping needed for Initialize, InsertKey, DeleteEntry. |
| 438 | }; |
| 439 | |
| 440 | class HashTables : public AllStatic { |
| 441 | public: |
| 442 | // Allocates and initializes a table. |
| 443 | template <typename Table> |
| 444 | static ArrayPtr New(intptr_t initial_capacity, |
| 445 | Heap::Space space = Heap::kNew) { |
| 446 | Table table( |
| 447 | Thread::Current()->zone(), |
| 448 | Array::New(Table::ArrayLengthForNumOccupied(initial_capacity), space)); |
| 449 | table.Initialize(); |
| 450 | return table.Release().raw(); |
| 451 | } |
| 452 | |
| 453 | template <typename Table> |
| 454 | static ArrayPtr New(const Array& array) { |
| 455 | Table table(Thread::Current()->zone(), array.raw()); |
| 456 | table.Initialize(); |
| 457 | return table.Release().raw(); |
| 458 | } |
| 459 | |
| 460 | // Clears 'to' and inserts all elements from 'from', in iteration order. |
| 461 | // The tables must have the same user payload size. |
| 462 | template <typename From, typename To> |
| 463 | static void Copy(const From& from, const To& to) { |
| 464 | COMPILE_ASSERT(From::kPayloadSize == To::kPayloadSize); |
| 465 | to.Initialize(); |
| 466 | ASSERT(from.NumOccupied() < to.NumEntries()); |
| 467 | typename From::Iterator it(&from); |
| 468 | Object& obj = Object::Handle(); |
| 469 | while (it.MoveNext()) { |
| 470 | intptr_t from_entry = it.Current(); |
| 471 | obj = from.GetKey(from_entry); |
| 472 | intptr_t to_entry = -1; |
| 473 | const Object& key = obj; |
| 474 | bool present = to.FindKeyOrDeletedOrUnused(key, &to_entry); |
| 475 | ASSERT(!present); |
| 476 | to.InsertKey(to_entry, obj); |
| 477 | for (intptr_t i = 0; i < From::kPayloadSize; ++i) { |
| 478 | obj = from.GetPayload(from_entry, i); |
| 479 | to.UpdatePayload(to_entry, i, obj); |
| 480 | } |
| 481 | } |
| 482 | } |
| 483 | |
| 484 | template <typename Table> |
| 485 | static void EnsureLoadFactor(double high, const Table& table) { |
| 486 | // We count deleted elements because they take up space just |
| 487 | // like occupied slots in order to cause a rehashing. |
| 488 | const double current = (1 + table.NumOccupied() + table.NumDeleted()) / |
| 489 | static_cast<double>(table.NumEntries()); |
| 490 | const bool too_many_deleted = table.NumOccupied() <= table.NumDeleted(); |
| 491 | if (current < high && !too_many_deleted) { |
| 492 | return; |
| 493 | } |
| 494 | // Normally we double the size here, but if less than half are occupied |
| 495 | // then it won't grow (this would imply that there were quite a lot of |
| 496 | // deleted slots). We don't want to constantly rehash if we are adding |
| 497 | // and deleting entries at just under the load factor limit, so we may |
| 498 | // double the size even though the number of occupied slots would not |
| 499 | // necessarily justify it. For example if the max load factor is 71% and |
| 500 | // the table is 70% full we will double the size to avoid a rehash every |
| 501 | // time 1% has been added and deleted. |
| 502 | const intptr_t new_capacity = table.NumOccupied() * 2 + 1; |
| 503 | ASSERT(table.NumOccupied() == 0 || |
| 504 | ((1.0 + table.NumOccupied()) / |
| 505 | Utils::RoundUpToPowerOfTwo(new_capacity)) <= high); |
| 506 | Table new_table(New<Table>(new_capacity, // Is rounded up to power of 2. |
| 507 | table.data_->IsOld() ? Heap::kOld : Heap::kNew)); |
| 508 | Copy(table, new_table); |
| 509 | *table.data_ = new_table.Release().raw(); |
| 510 | NOT_IN_PRODUCT(table.UpdateGrowth(); table.PrintStats();) |
| 511 | } |
| 512 | |
| 513 | // Serializes a table by concatenating its entries as an array. |
| 514 | template <typename Table> |
| 515 | static ArrayPtr ToArray(const Table& table, bool include_payload) { |
| 516 | const intptr_t entry_size = include_payload ? (1 + Table::kPayloadSize) : 1; |
| 517 | Array& result = Array::Handle(Array::New(table.NumOccupied() * entry_size)); |
| 518 | typename Table::Iterator it(&table); |
| 519 | Object& obj = Object::Handle(); |
| 520 | intptr_t result_index = 0; |
| 521 | while (it.MoveNext()) { |
| 522 | intptr_t entry = it.Current(); |
| 523 | obj = table.GetKey(entry); |
| 524 | result.SetAt(result_index++, obj); |
| 525 | if (include_payload) { |
| 526 | for (intptr_t i = 0; i < Table::kPayloadSize; ++i) { |
| 527 | obj = table.GetPayload(entry, i); |
| 528 | result.SetAt(result_index++, obj); |
| 529 | } |
| 530 | } |
| 531 | } |
| 532 | return result.raw(); |
| 533 | } |
| 534 | }; |
| 535 | |
| 536 | template <typename BaseIterTable> |
| 537 | class HashMap : public BaseIterTable { |
| 538 | public: |
| 539 | explicit HashMap(ArrayPtr data) |
| 540 | : BaseIterTable(Thread::Current()->zone(), data) {} |
| 541 | HashMap(Zone* zone, ArrayPtr data) : BaseIterTable(zone, data) {} |
| 542 | HashMap(Object* key, Smi* value, Array* data) |
| 543 | : BaseIterTable(key, value, data) {} |
| 544 | template <typename Key> |
| 545 | ObjectPtr GetOrNull(const Key& key, bool* present = NULL) const { |
| 546 | intptr_t entry = BaseIterTable::FindKey(key); |
| 547 | if (present != NULL) { |
| 548 | *present = (entry != -1); |
| 549 | } |
| 550 | return (entry == -1) ? Object::null() : BaseIterTable::GetPayload(entry, 0); |
| 551 | } |
| 552 | template <typename Key> |
| 553 | ObjectPtr GetOrDie(const Key& key) const { |
| 554 | intptr_t entry = BaseIterTable::FindKey(key); |
| 555 | if (entry == -1) UNREACHABLE(); |
| 556 | return BaseIterTable::GetPayload(entry, 0); |
| 557 | } |
| 558 | bool UpdateOrInsert(const Object& key, const Object& value) const { |
| 559 | EnsureCapacity(); |
| 560 | intptr_t entry = -1; |
| 561 | bool present = BaseIterTable::FindKeyOrDeletedOrUnused(key, &entry); |
| 562 | if (!present) { |
| 563 | BaseIterTable::InsertKey(entry, key); |
| 564 | } |
| 565 | BaseIterTable::UpdatePayload(entry, 0, value); |
| 566 | return present; |
| 567 | } |
| 568 | // Update the value of an existing key. Note that 'key' need not be an Object. |
| 569 | template <typename Key> |
| 570 | void UpdateValue(const Key& key, const Object& value) const { |
| 571 | intptr_t entry = BaseIterTable::FindKey(key); |
| 572 | ASSERT(entry != -1); |
| 573 | BaseIterTable::UpdatePayload(entry, 0, value); |
| 574 | } |
| 575 | // If 'key' is not present, maps it to 'value_if_absent'. Returns the final |
| 576 | // value in the map. |
| 577 | ObjectPtr InsertOrGetValue(const Object& key, |
| 578 | const Object& value_if_absent) const { |
| 579 | EnsureCapacity(); |
| 580 | intptr_t entry = -1; |
| 581 | if (!BaseIterTable::FindKeyOrDeletedOrUnused(key, &entry)) { |
| 582 | BaseIterTable::InsertKey(entry, key); |
| 583 | BaseIterTable::UpdatePayload(entry, 0, value_if_absent); |
| 584 | return value_if_absent.raw(); |
| 585 | } else { |
| 586 | return BaseIterTable::GetPayload(entry, 0); |
| 587 | } |
| 588 | } |
| 589 | // Like InsertOrGetValue, but calls NewKey to allocate a key object if needed. |
| 590 | template <typename Key> |
| 591 | ObjectPtr InsertNewOrGetValue(const Key& key, |
| 592 | const Object& value_if_absent) const { |
| 593 | EnsureCapacity(); |
| 594 | intptr_t entry = -1; |
| 595 | if (!BaseIterTable::FindKeyOrDeletedOrUnused(key, &entry)) { |
| 596 | BaseIterTable::KeyHandle() = |
| 597 | BaseIterTable::BaseTable::Traits::NewKey(key); |
| 598 | BaseIterTable::InsertKey(entry, BaseIterTable::KeyHandle()); |
| 599 | BaseIterTable::UpdatePayload(entry, 0, value_if_absent); |
| 600 | return value_if_absent.raw(); |
| 601 | } else { |
| 602 | return BaseIterTable::GetPayload(entry, 0); |
| 603 | } |
| 604 | } |
| 605 | |
| 606 | template <typename Key> |
| 607 | bool Remove(const Key& key) const { |
| 608 | intptr_t entry = BaseIterTable::FindKey(key); |
| 609 | if (entry == -1) { |
| 610 | return false; |
| 611 | } else { |
| 612 | BaseIterTable::DeleteEntry(entry); |
| 613 | return true; |
| 614 | } |
| 615 | } |
| 616 | |
| 617 | void Clear() const { BaseIterTable::Initialize(); } |
| 618 | |
| 619 | protected: |
| 620 | void EnsureCapacity() const { |
| 621 | static const double kMaxLoadFactor = 0.71; |
| 622 | HashTables::EnsureLoadFactor(kMaxLoadFactor, *this); |
| 623 | } |
| 624 | }; |
| 625 | |
| 626 | template <typename KeyTraits> |
| 627 | class UnorderedHashMap : public HashMap<UnorderedHashTable<KeyTraits, 1> > { |
| 628 | public: |
| 629 | typedef HashMap<UnorderedHashTable<KeyTraits, 1> > BaseMap; |
| 630 | explicit UnorderedHashMap(ArrayPtr data) |
| 631 | : BaseMap(Thread::Current()->zone(), data) {} |
| 632 | UnorderedHashMap(Zone* zone, ArrayPtr data) : BaseMap(zone, data) {} |
| 633 | UnorderedHashMap(Object* key, Smi* value, Array* data) |
| 634 | : BaseMap(key, value, data) {} |
| 635 | }; |
| 636 | |
| 637 | template <typename BaseIterTable> |
| 638 | class HashSet : public BaseIterTable { |
| 639 | public: |
| 640 | explicit HashSet(ArrayPtr data) |
| 641 | : BaseIterTable(Thread::Current()->zone(), data) {} |
| 642 | HashSet(Zone* zone, ArrayPtr data) : BaseIterTable(zone, data) {} |
| 643 | HashSet(Object* key, Smi* value, Array* data) |
| 644 | : BaseIterTable(key, value, data) {} |
| 645 | bool Insert(const Object& key) { |
| 646 | EnsureCapacity(); |
| 647 | intptr_t entry = -1; |
| 648 | bool present = BaseIterTable::FindKeyOrDeletedOrUnused(key, &entry); |
| 649 | if (!present) { |
| 650 | BaseIterTable::InsertKey(entry, key); |
| 651 | } |
| 652 | return present; |
| 653 | } |
| 654 | |
| 655 | // If 'key' is not present, insert and return it. Else, return the existing |
| 656 | // key in the set (useful for canonicalization). |
| 657 | ObjectPtr InsertOrGet(const Object& key) const { |
| 658 | EnsureCapacity(); |
| 659 | intptr_t entry = -1; |
| 660 | if (!BaseIterTable::FindKeyOrDeletedOrUnused(key, &entry)) { |
| 661 | BaseIterTable::InsertKey(entry, key); |
| 662 | return key.raw(); |
| 663 | } else { |
| 664 | return BaseIterTable::GetKey(entry); |
| 665 | } |
| 666 | } |
| 667 | |
| 668 | // Like InsertOrGet, but calls NewKey to allocate a key object if needed. |
| 669 | template <typename Key> |
| 670 | ObjectPtr InsertNewOrGet(const Key& key) const { |
| 671 | EnsureCapacity(); |
| 672 | intptr_t entry = -1; |
| 673 | if (!BaseIterTable::FindKeyOrDeletedOrUnused(key, &entry)) { |
| 674 | BaseIterTable::KeyHandle() = |
| 675 | BaseIterTable::BaseTable::Traits::NewKey(key); |
| 676 | BaseIterTable::InsertKey(entry, BaseIterTable::KeyHandle()); |
| 677 | return BaseIterTable::KeyHandle().raw(); |
| 678 | } else { |
| 679 | return BaseIterTable::GetKey(entry); |
| 680 | } |
| 681 | } |
| 682 | |
| 683 | template <typename Key> |
| 684 | ObjectPtr GetOrNull(const Key& key, bool* present = NULL) const { |
| 685 | intptr_t entry = BaseIterTable::FindKey(key); |
| 686 | if (present != NULL) { |
| 687 | *present = (entry != -1); |
| 688 | } |
| 689 | return (entry == -1) ? Object::null() : BaseIterTable::GetKey(entry); |
| 690 | } |
| 691 | |
| 692 | template <typename Key> |
| 693 | bool Remove(const Key& key) const { |
| 694 | intptr_t entry = BaseIterTable::FindKey(key); |
| 695 | if (entry == -1) { |
| 696 | return false; |
| 697 | } else { |
| 698 | BaseIterTable::DeleteEntry(entry); |
| 699 | return true; |
| 700 | } |
| 701 | } |
| 702 | |
| 703 | void Clear() const { BaseIterTable::Initialize(); } |
| 704 | |
| 705 | protected: |
| 706 | void EnsureCapacity() const { |
| 707 | static const double kMaxLoadFactor = 0.71; |
| 708 | HashTables::EnsureLoadFactor(kMaxLoadFactor, *this); |
| 709 | } |
| 710 | }; |
| 711 | |
| 712 | template <typename KeyTraits> |
| 713 | class UnorderedHashSet : public HashSet<UnorderedHashTable<KeyTraits, 0> > { |
| 714 | public: |
| 715 | typedef HashSet<UnorderedHashTable<KeyTraits, 0> > BaseSet; |
| 716 | explicit UnorderedHashSet(ArrayPtr data) |
| 717 | : BaseSet(Thread::Current()->zone(), data) { |
| 718 | ASSERT(data != Array::null()); |
| 719 | } |
| 720 | UnorderedHashSet(Zone* zone, ArrayPtr data) : BaseSet(zone, data) {} |
| 721 | UnorderedHashSet(Object* key, Smi* value, Array* data) |
| 722 | : BaseSet(key, value, data) {} |
| 723 | |
| 724 | void Dump() const { |
| 725 | Object& entry = Object::Handle(); |
| 726 | for (intptr_t i = 0; i < this->data_->Length(); i++) { |
| 727 | entry = this->data_->At(i); |
| 728 | if (entry.raw() == BaseSet::UnusedMarker().raw() || |
| 729 | entry.raw() == BaseSet::DeletedMarker().raw() || entry.IsSmi()) { |
| 730 | // empty, deleted, num_used/num_deleted |
| 731 | OS::PrintErr("%" Pd ": %s\n" , i, entry.ToCString()); |
| 732 | } else { |
| 733 | intptr_t hash = KeyTraits::Hash(entry); |
| 734 | OS::PrintErr("%" Pd ": %" Pd ", %s\n" , i, hash, entry.ToCString()); |
| 735 | } |
| 736 | } |
| 737 | } |
| 738 | }; |
| 739 | |
| 740 | } // namespace dart |
| 741 | |
| 742 | #endif // RUNTIME_VM_HASH_TABLE_H_ |
| 743 | |