| 1 | //! This crate implements a hash table that can be used as is in its binary, on-disk format. |
| 2 | //! The goal is to provide a high performance data structure that can be used without any significant up-front decoding. |
| 3 | //! The implementation makes no assumptions about alignment or endianess of the underlying data, |
| 4 | //! so a table encoded on one platform can be used on any other platform and |
| 5 | //! the binary data can be mapped into memory at arbitrary addresses. |
| 6 | //! |
| 7 | //! |
| 8 | //! ## Usage |
| 9 | //! |
| 10 | //! In order to use the hash table one needs to implement the `Config` trait. |
| 11 | //! This trait defines how the table is encoded and what hash function is used. |
| 12 | //! With a `Config` in place the `HashTableOwned` type can be used to build and serialize a hash table. |
| 13 | //! The `HashTable` type can then be used to create an almost zero-cost view of the serialized hash table. |
| 14 | //! |
| 15 | //! ```rust |
| 16 | //! |
| 17 | //! use odht::{HashTable, HashTableOwned, Config, FxHashFn}; |
| 18 | //! |
| 19 | //! struct MyConfig; |
| 20 | //! |
| 21 | //! impl Config for MyConfig { |
| 22 | //! |
| 23 | //! type Key = u64; |
| 24 | //! type Value = u32; |
| 25 | //! |
| 26 | //! type EncodedKey = [u8; 8]; |
| 27 | //! type EncodedValue = [u8; 4]; |
| 28 | //! |
| 29 | //! type H = FxHashFn; |
| 30 | //! |
| 31 | //! #[inline ] fn encode_key(k: &Self::Key) -> Self::EncodedKey { k.to_le_bytes() } |
| 32 | //! #[inline ] fn encode_value(v: &Self::Value) -> Self::EncodedValue { v.to_le_bytes() } |
| 33 | //! #[inline ] fn decode_key(k: &Self::EncodedKey) -> Self::Key { u64::from_le_bytes(*k) } |
| 34 | //! #[inline ] fn decode_value(v: &Self::EncodedValue) -> Self::Value { u32::from_le_bytes(*v)} |
| 35 | //! } |
| 36 | //! |
| 37 | //! fn main() { |
| 38 | //! let mut builder = HashTableOwned::<MyConfig>::with_capacity(3, 95); |
| 39 | //! |
| 40 | //! builder.insert(&1, &2); |
| 41 | //! builder.insert(&3, &4); |
| 42 | //! builder.insert(&5, &6); |
| 43 | //! |
| 44 | //! let serialized = builder.raw_bytes().to_owned(); |
| 45 | //! |
| 46 | //! let table = HashTable::<MyConfig, &[u8]>::from_raw_bytes( |
| 47 | //! &serialized[..] |
| 48 | //! ).unwrap(); |
| 49 | //! |
| 50 | //! assert_eq!(table.get(&1), Some(2)); |
| 51 | //! assert_eq!(table.get(&3), Some(4)); |
| 52 | //! assert_eq!(table.get(&5), Some(6)); |
| 53 | //! } |
| 54 | //! ``` |
| 55 | |
| 56 | #![cfg_attr (feature = "nightly" , feature(core_intrinsics))] |
| 57 | |
| 58 | #[cfg (test)] |
| 59 | extern crate quickcheck; |
| 60 | |
| 61 | #[cfg (feature = "nightly" )] |
| 62 | macro_rules! likely { |
| 63 | ($x:expr) => { |
| 64 | std::intrinsics::likely($x) |
| 65 | }; |
| 66 | } |
| 67 | |
| 68 | #[cfg (not(feature = "nightly" ))] |
| 69 | macro_rules! likely { |
| 70 | ($x:expr) => { |
| 71 | $x |
| 72 | }; |
| 73 | } |
| 74 | |
| 75 | #[cfg (feature = "nightly" )] |
| 76 | macro_rules! unlikely { |
| 77 | ($x:expr) => { |
| 78 | std::intrinsics::unlikely($x) |
| 79 | }; |
| 80 | } |
| 81 | |
| 82 | #[cfg (not(feature = "nightly" ))] |
| 83 | macro_rules! unlikely { |
| 84 | ($x:expr) => { |
| 85 | $x |
| 86 | }; |
| 87 | } |
| 88 | |
| 89 | mod error; |
| 90 | mod fxhash; |
| 91 | mod memory_layout; |
| 92 | mod raw_table; |
| 93 | mod swisstable_group_query; |
| 94 | mod unhash; |
| 95 | |
| 96 | use error::Error; |
| 97 | use memory_layout::Header; |
| 98 | use std::borrow::{Borrow, BorrowMut}; |
| 99 | use swisstable_group_query::REFERENCE_GROUP_SIZE; |
| 100 | |
| 101 | pub use crate::fxhash::FxHashFn; |
| 102 | pub use crate::unhash::UnHashFn; |
| 103 | |
| 104 | use crate::raw_table::{ByteArray, RawIter, RawTable, RawTableMut}; |
| 105 | |
| 106 | /// This trait provides a complete "configuration" for a hash table, i.e. it |
| 107 | /// defines the key and value types, how these are encoded and what hash |
| 108 | /// function is being used. |
| 109 | /// |
| 110 | /// Implementations of the `encode_key` and `encode_value` methods must encode |
| 111 | /// the given key/value into a fixed size array. The encoding must be |
| 112 | /// deterministic (i.e. no random padding bytes) and must be independent of |
| 113 | /// platform endianess. It is always highly recommended to mark these methods |
| 114 | /// as `#[inline]`. |
| 115 | pub trait Config { |
| 116 | type Key; |
| 117 | type Value; |
| 118 | |
| 119 | // The EncodedKey and EncodedValue types must always be a fixed size array of bytes, |
| 120 | // e.g. [u8; 4]. |
| 121 | type EncodedKey: ByteArray; |
| 122 | type EncodedValue: ByteArray; |
| 123 | |
| 124 | type H: HashFn; |
| 125 | |
| 126 | /// Implementations of the `encode_key` and `encode_value` methods must encode |
| 127 | /// the given key/value into a fixed size array. See above for requirements. |
| 128 | fn encode_key(k: &Self::Key) -> Self::EncodedKey; |
| 129 | |
| 130 | /// Implementations of the `encode_key` and `encode_value` methods must encode |
| 131 | /// the given key/value into a fixed size array. See above for requirements. |
| 132 | fn encode_value(v: &Self::Value) -> Self::EncodedValue; |
| 133 | |
| 134 | fn decode_key(k: &Self::EncodedKey) -> Self::Key; |
| 135 | fn decode_value(v: &Self::EncodedValue) -> Self::Value; |
| 136 | } |
| 137 | |
| 138 | /// This trait represents hash functions as used by HashTable and |
| 139 | /// HashTableOwned. |
| 140 | pub trait HashFn: Eq { |
| 141 | fn hash(bytes: &[u8]) -> u32; |
| 142 | } |
| 143 | |
| 144 | /// A [HashTableOwned] keeps the underlying data on the heap and |
| 145 | /// can resize itself on demand. |
| 146 | #[derive (Clone)] |
| 147 | pub struct HashTableOwned<C: Config> { |
| 148 | allocation: memory_layout::Allocation<C, Box<[u8]>>, |
| 149 | } |
| 150 | |
| 151 | impl<C: Config> Default for HashTableOwned<C> { |
| 152 | fn default() -> Self { |
| 153 | HashTableOwned::with_capacity(max_item_count:12, max_load_factor_percent:87) |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | impl<C: Config> HashTableOwned<C> { |
| 158 | /// Creates a new [HashTableOwned] that can hold at least `max_item_count` |
| 159 | /// items while maintaining the specified load factor. |
| 160 | pub fn with_capacity(max_item_count: usize, max_load_factor_percent: u8) -> HashTableOwned<C> { |
| 161 | assert!(max_load_factor_percent <= 100); |
| 162 | assert!(max_load_factor_percent > 0); |
| 163 | |
| 164 | Self::with_capacity_internal( |
| 165 | max_item_count, |
| 166 | Factor::from_percent(max_load_factor_percent), |
| 167 | ) |
| 168 | } |
| 169 | |
| 170 | fn with_capacity_internal(max_item_count: usize, max_load_factor: Factor) -> HashTableOwned<C> { |
| 171 | let slots_needed = slots_needed(max_item_count, max_load_factor); |
| 172 | assert!(slots_needed > 0); |
| 173 | |
| 174 | let allocation = memory_layout::allocate(slots_needed, 0, max_load_factor); |
| 175 | |
| 176 | HashTableOwned { allocation } |
| 177 | } |
| 178 | |
| 179 | /// Retrieves the value for the given key. Returns `None` if no entry is found. |
| 180 | #[inline ] |
| 181 | pub fn get(&self, key: &C::Key) -> Option<C::Value> { |
| 182 | let encoded_key = C::encode_key(key); |
| 183 | if let Some(encoded_value) = self.as_raw().find(&encoded_key) { |
| 184 | Some(C::decode_value(encoded_value)) |
| 185 | } else { |
| 186 | None |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | #[inline ] |
| 191 | pub fn contains_key(&self, key: &C::Key) -> bool { |
| 192 | let encoded_key = C::encode_key(key); |
| 193 | self.as_raw().find(&encoded_key).is_some() |
| 194 | } |
| 195 | |
| 196 | /// Inserts the given key-value pair into the table. |
| 197 | /// Grows the table if necessary. |
| 198 | #[inline ] |
| 199 | pub fn insert(&mut self, key: &C::Key, value: &C::Value) -> Option<C::Value> { |
| 200 | let (item_count, max_item_count) = { |
| 201 | let header = self.allocation.header(); |
| 202 | let max_item_count = max_item_count_for(header.slot_count(), header.max_load_factor()); |
| 203 | (header.item_count(), max_item_count) |
| 204 | }; |
| 205 | |
| 206 | if unlikely!(item_count == max_item_count) { |
| 207 | self.grow(); |
| 208 | } |
| 209 | |
| 210 | debug_assert!( |
| 211 | item_count |
| 212 | < max_item_count_for( |
| 213 | self.allocation.header().slot_count(), |
| 214 | self.allocation.header().max_load_factor() |
| 215 | ) |
| 216 | ); |
| 217 | |
| 218 | let encoded_key = C::encode_key(key); |
| 219 | let encoded_value = C::encode_value(value); |
| 220 | |
| 221 | with_raw_mut(&mut self.allocation, |header, mut raw_table| { |
| 222 | if let Some(old_value) = raw_table.insert(encoded_key, encoded_value) { |
| 223 | Some(C::decode_value(&old_value)) |
| 224 | } else { |
| 225 | header.set_item_count(item_count + 1); |
| 226 | None |
| 227 | } |
| 228 | }) |
| 229 | } |
| 230 | |
| 231 | #[inline ] |
| 232 | pub fn iter(&self) -> Iter<'_, C> { |
| 233 | let (entry_metadata, entry_data) = self.allocation.data_slices(); |
| 234 | Iter(RawIter::new(entry_metadata, entry_data)) |
| 235 | } |
| 236 | |
| 237 | pub fn from_iterator<I: IntoIterator<Item = (C::Key, C::Value)>>( |
| 238 | it: I, |
| 239 | max_load_factor_percent: u8, |
| 240 | ) -> Self { |
| 241 | let it = it.into_iter(); |
| 242 | |
| 243 | let known_size = match it.size_hint() { |
| 244 | (min, Some(max)) => { |
| 245 | if min == max { |
| 246 | Some(max) |
| 247 | } else { |
| 248 | None |
| 249 | } |
| 250 | } |
| 251 | _ => None, |
| 252 | }; |
| 253 | |
| 254 | if let Some(known_size) = known_size { |
| 255 | let mut table = HashTableOwned::with_capacity(known_size, max_load_factor_percent); |
| 256 | |
| 257 | let initial_slot_count = table.allocation.header().slot_count(); |
| 258 | |
| 259 | for (k, v) in it { |
| 260 | table.insert(&k, &v); |
| 261 | } |
| 262 | |
| 263 | // duplicates |
| 264 | assert!(table.len() <= known_size); |
| 265 | assert_eq!(table.allocation.header().slot_count(), initial_slot_count); |
| 266 | |
| 267 | table |
| 268 | } else { |
| 269 | let items: Vec<_> = it.collect(); |
| 270 | Self::from_iterator(items, max_load_factor_percent) |
| 271 | } |
| 272 | } |
| 273 | |
| 274 | /// Constructs a [HashTableOwned] from its raw byte representation. |
| 275 | /// The provided data must have the exact right number of bytes. |
| 276 | /// |
| 277 | /// This method has linear time complexity as it needs to make its own |
| 278 | /// copy of the given data. |
| 279 | /// |
| 280 | /// The method will verify the header of the given data and return an |
| 281 | /// error if the verification fails. |
| 282 | pub fn from_raw_bytes(data: &[u8]) -> Result<HashTableOwned<C>, Box<dyn std::error::Error>> { |
| 283 | let data = data.to_owned().into_boxed_slice(); |
| 284 | let allocation = memory_layout::Allocation::from_raw_bytes(data)?; |
| 285 | |
| 286 | Ok(HashTableOwned { allocation }) |
| 287 | } |
| 288 | |
| 289 | #[inline ] |
| 290 | pub unsafe fn from_raw_bytes_unchecked(data: &[u8]) -> HashTableOwned<C> { |
| 291 | let data = data.to_owned().into_boxed_slice(); |
| 292 | let allocation = memory_layout::Allocation::from_raw_bytes_unchecked(data); |
| 293 | |
| 294 | HashTableOwned { allocation } |
| 295 | } |
| 296 | |
| 297 | /// Returns the number of items stored in the hash table. |
| 298 | #[inline ] |
| 299 | pub fn len(&self) -> usize { |
| 300 | self.allocation.header().item_count() |
| 301 | } |
| 302 | |
| 303 | #[inline ] |
| 304 | pub fn raw_bytes(&self) -> &[u8] { |
| 305 | self.allocation.raw_bytes() |
| 306 | } |
| 307 | |
| 308 | #[inline ] |
| 309 | fn as_raw(&self) -> RawTable<'_, C::EncodedKey, C::EncodedValue, C::H> { |
| 310 | let (entry_metadata, entry_data) = self.allocation.data_slices(); |
| 311 | RawTable::new(entry_metadata, entry_data) |
| 312 | } |
| 313 | |
| 314 | #[inline (never)] |
| 315 | #[cold ] |
| 316 | fn grow(&mut self) { |
| 317 | let initial_slot_count = self.allocation.header().slot_count(); |
| 318 | let initial_item_count = self.allocation.header().item_count(); |
| 319 | let initial_max_load_factor = self.allocation.header().max_load_factor(); |
| 320 | |
| 321 | let mut new_table = |
| 322 | Self::with_capacity_internal(initial_item_count * 2, initial_max_load_factor); |
| 323 | |
| 324 | // Copy the entries over with the internal `insert_entry()` method, |
| 325 | // which allows us to do insertions without hashing everything again. |
| 326 | { |
| 327 | with_raw_mut(&mut new_table.allocation, |header, mut raw_table| { |
| 328 | for (_, entry_data) in self.as_raw().iter() { |
| 329 | raw_table.insert(entry_data.key, entry_data.value); |
| 330 | } |
| 331 | |
| 332 | header.set_item_count(initial_item_count); |
| 333 | }); |
| 334 | } |
| 335 | |
| 336 | *self = new_table; |
| 337 | |
| 338 | assert!( |
| 339 | self.allocation.header().slot_count() >= 2 * initial_slot_count, |
| 340 | "Allocation did not grow properly. Slot count is {} but was expected to be \ |
| 341 | at least {}" , |
| 342 | self.allocation.header().slot_count(), |
| 343 | 2 * initial_slot_count |
| 344 | ); |
| 345 | assert_eq!(self.allocation.header().item_count(), initial_item_count); |
| 346 | assert_eq!( |
| 347 | self.allocation.header().max_load_factor(), |
| 348 | initial_max_load_factor |
| 349 | ); |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | impl<C: Config> std::fmt::Debug for HashTableOwned<C> { |
| 354 | fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { |
| 355 | let header: &Header = self.allocation.header(); |
| 356 | |
| 357 | writeln!( |
| 358 | f, |
| 359 | "(item_count= {}, max_item_count= {}, max_load_factor= {}%)" , |
| 360 | header.item_count(), |
| 361 | max_item_count_for(header.slot_count(), header.max_load_factor()), |
| 362 | header.max_load_factor().to_percent(), |
| 363 | )?; |
| 364 | |
| 365 | writeln!(f, " {:?}" , self.as_raw()) |
| 366 | } |
| 367 | } |
| 368 | |
| 369 | /// The [HashTable] type provides a cheap way to construct a non-resizable view |
| 370 | /// of a persisted hash table. If the underlying data storage `D` implements |
| 371 | /// `BorrowMut<[u8]>` then the table can be modified in place. |
| 372 | #[derive (Clone, Copy)] |
| 373 | pub struct HashTable<C: Config, D: Borrow<[u8]>> { |
| 374 | allocation: memory_layout::Allocation<C, D>, |
| 375 | } |
| 376 | |
| 377 | impl<C: Config, D: Borrow<[u8]>> HashTable<C, D> { |
| 378 | /// Constructs a [HashTable] from its raw byte representation. |
| 379 | /// The provided data must have the exact right number of bytes. |
| 380 | /// |
| 381 | /// This method has constant time complexity and will only verify the header |
| 382 | /// data of the hash table. It will not copy any data. |
| 383 | pub fn from_raw_bytes(data: D) -> Result<HashTable<C, D>, Box<dyn std::error::Error>> { |
| 384 | let allocation = memory_layout::Allocation::from_raw_bytes(data)?; |
| 385 | Ok(HashTable { allocation }) |
| 386 | } |
| 387 | |
| 388 | /// Constructs a [HashTable] from its raw byte representation without doing |
| 389 | /// any verification of the underlying data. It is the user's responsibility |
| 390 | /// to make sure that the underlying data is actually a valid hash table. |
| 391 | /// |
| 392 | /// The [HashTable::from_raw_bytes] method provides a safe alternative to this |
| 393 | /// method. |
| 394 | #[inline ] |
| 395 | pub unsafe fn from_raw_bytes_unchecked(data: D) -> HashTable<C, D> { |
| 396 | HashTable { |
| 397 | allocation: memory_layout::Allocation::from_raw_bytes_unchecked(data), |
| 398 | } |
| 399 | } |
| 400 | |
| 401 | #[inline ] |
| 402 | pub fn get(&self, key: &C::Key) -> Option<C::Value> { |
| 403 | let encoded_key = C::encode_key(key); |
| 404 | self.as_raw().find(&encoded_key).map(C::decode_value) |
| 405 | } |
| 406 | |
| 407 | #[inline ] |
| 408 | pub fn contains_key(&self, key: &C::Key) -> bool { |
| 409 | let encoded_key = C::encode_key(key); |
| 410 | self.as_raw().find(&encoded_key).is_some() |
| 411 | } |
| 412 | |
| 413 | #[inline ] |
| 414 | pub fn iter(&self) -> Iter<'_, C> { |
| 415 | let (entry_metadata, entry_data) = self.allocation.data_slices(); |
| 416 | Iter(RawIter::new(entry_metadata, entry_data)) |
| 417 | } |
| 418 | |
| 419 | /// Returns the number of items stored in the hash table. |
| 420 | #[inline ] |
| 421 | pub fn len(&self) -> usize { |
| 422 | self.allocation.header().item_count() |
| 423 | } |
| 424 | |
| 425 | #[inline ] |
| 426 | pub fn raw_bytes(&self) -> &[u8] { |
| 427 | self.allocation.raw_bytes() |
| 428 | } |
| 429 | |
| 430 | #[inline ] |
| 431 | fn as_raw(&self) -> RawTable<'_, C::EncodedKey, C::EncodedValue, C::H> { |
| 432 | let (entry_metadata, entry_data) = self.allocation.data_slices(); |
| 433 | RawTable::new(entry_metadata, entry_data) |
| 434 | } |
| 435 | } |
| 436 | |
| 437 | impl<C: Config, D: Borrow<[u8]> + BorrowMut<[u8]>> HashTable<C, D> { |
| 438 | pub fn init_in_place( |
| 439 | mut data: D, |
| 440 | max_item_count: usize, |
| 441 | max_load_factor_percent: u8, |
| 442 | ) -> Result<HashTable<C, D>, Box<dyn std::error::Error>> { |
| 443 | let max_load_factor = Factor::from_percent(max_load_factor_percent); |
| 444 | let byte_count = bytes_needed_internal::<C>(max_item_count, max_load_factor); |
| 445 | if data.borrow_mut().len() != byte_count { |
| 446 | return Err(Error(format!( |
| 447 | "byte slice to initialize has wrong length ( {} instead of {})" , |
| 448 | data.borrow_mut().len(), |
| 449 | byte_count |
| 450 | )))?; |
| 451 | } |
| 452 | |
| 453 | let slot_count = slots_needed(max_item_count, max_load_factor); |
| 454 | let allocation = memory_layout::init_in_place::<C, _>(data, slot_count, 0, max_load_factor); |
| 455 | Ok(HashTable { allocation }) |
| 456 | } |
| 457 | |
| 458 | /// Inserts the given key-value pair into the table. |
| 459 | /// Unlike [HashTableOwned::insert] this method cannot grow the underlying table |
| 460 | /// if there is not enough space for the new item. Instead the call will panic. |
| 461 | #[inline ] |
| 462 | pub fn insert(&mut self, key: &C::Key, value: &C::Value) -> Option<C::Value> { |
| 463 | let item_count = self.allocation.header().item_count(); |
| 464 | let max_load_factor = self.allocation.header().max_load_factor(); |
| 465 | let slot_count = self.allocation.header().slot_count(); |
| 466 | // FIXME: This is actually a bit to conservative because it does not account for |
| 467 | // cases where an entry is overwritten and thus the item count does not |
| 468 | // change. |
| 469 | assert!(item_count < max_item_count_for(slot_count, max_load_factor)); |
| 470 | |
| 471 | let encoded_key = C::encode_key(key); |
| 472 | let encoded_value = C::encode_value(value); |
| 473 | |
| 474 | with_raw_mut(&mut self.allocation, |header, mut raw_table| { |
| 475 | if let Some(old_value) = raw_table.insert(encoded_key, encoded_value) { |
| 476 | Some(C::decode_value(&old_value)) |
| 477 | } else { |
| 478 | header.set_item_count(item_count + 1); |
| 479 | None |
| 480 | } |
| 481 | }) |
| 482 | } |
| 483 | } |
| 484 | |
| 485 | /// Computes the exact number of bytes needed for storing a HashTable with the |
| 486 | /// given max item count and load factor. The result can be used for allocating |
| 487 | /// storage to be passed into [HashTable::init_in_place]. |
| 488 | pub fn bytes_needed<C: Config>(max_item_count: usize, max_load_factor_percent: u8) -> usize { |
| 489 | let max_load_factor: Factor = Factor::from_percent(max_load_factor_percent); |
| 490 | bytes_needed_internal::<C>(max_item_count, max_load_factor) |
| 491 | } |
| 492 | |
| 493 | fn bytes_needed_internal<C: Config>(max_item_count: usize, max_load_factor: Factor) -> usize { |
| 494 | let slot_count: usize = slots_needed(max_item_count, max_load_factor); |
| 495 | memory_layout::bytes_needed::<C>(slot_count) |
| 496 | } |
| 497 | |
| 498 | pub struct Iter<'a, C: Config>(RawIter<'a, C::EncodedKey, C::EncodedValue>); |
| 499 | |
| 500 | impl<'a, C: Config> Iterator for Iter<'a, C> { |
| 501 | type Item = (C::Key, C::Value); |
| 502 | |
| 503 | fn next(&mut self) -> Option<Self::Item> { |
| 504 | self.0.next().map(|(_, entry: &'a Entry<::EncodedKey, …>)| { |
| 505 | let key: ::Key = C::decode_key(&entry.key); |
| 506 | let value: ::Value = C::decode_value(&entry.value); |
| 507 | |
| 508 | (key, value) |
| 509 | }) |
| 510 | } |
| 511 | } |
| 512 | |
| 513 | // We use integer math here as not to run into any issues with |
| 514 | // platform-specific floating point math implementation. |
| 515 | fn slots_needed(item_count: usize, max_load_factor: Factor) -> usize { |
| 516 | // Note: we round up here |
| 517 | let slots_needed: usize = max_load_factor.apply_inverse(item_count); |
| 518 | std::cmp::max( |
| 519 | v1:slots_needed.checked_next_power_of_two().unwrap(), |
| 520 | REFERENCE_GROUP_SIZE, |
| 521 | ) |
| 522 | } |
| 523 | |
| 524 | fn max_item_count_for(slot_count: usize, max_load_factor: Factor) -> usize { |
| 525 | // Note: we round down here |
| 526 | max_load_factor.apply(slot_count) |
| 527 | } |
| 528 | |
| 529 | #[inline ] |
| 530 | fn with_raw_mut<C, M, F, R>(allocation: &mut memory_layout::Allocation<C, M>, f: F) -> R |
| 531 | where |
| 532 | C: Config, |
| 533 | M: BorrowMut<[u8]>, |
| 534 | F: FnOnce(&mut Header, RawTableMut<'_, C::EncodedKey, C::EncodedValue, C::H>) -> R, |
| 535 | { |
| 536 | allocation.with_mut_parts(|header: &mut Header, entry_metadata: &mut [u8], entry_data: &mut [Entry<::EncodedKey, …>]| { |
| 537 | f(header, RawTableMut::new(entry_metadata, entry_data)) |
| 538 | }) |
| 539 | } |
| 540 | |
| 541 | /// This type is used for computing max item counts for a given load factor |
| 542 | /// efficiently. We use integer math here so that things are the same on |
| 543 | /// all platforms and with all compiler settings. |
| 544 | #[derive (Debug, Clone, Copy, PartialEq, Eq)] |
| 545 | struct Factor(pub u16); |
| 546 | |
| 547 | impl Factor { |
| 548 | const BASE: usize = u16::MAX as usize; |
| 549 | |
| 550 | #[inline ] |
| 551 | fn from_percent(percent: u8) -> Factor { |
| 552 | let percent = percent as usize; |
| 553 | Factor(((percent * Self::BASE) / 100) as u16) |
| 554 | } |
| 555 | |
| 556 | fn to_percent(self) -> usize { |
| 557 | (self.0 as usize * 100) / Self::BASE |
| 558 | } |
| 559 | |
| 560 | // Note: we round down here |
| 561 | #[inline ] |
| 562 | fn apply(self, x: usize) -> usize { |
| 563 | // Let's make sure there's no overflow during the |
| 564 | // calculation below by doing everything with 128 bits. |
| 565 | let x = x as u128; |
| 566 | let factor = self.0 as u128; |
| 567 | ((x * factor) >> 16) as usize |
| 568 | } |
| 569 | |
| 570 | // Note: we round up here |
| 571 | #[inline ] |
| 572 | fn apply_inverse(self, x: usize) -> usize { |
| 573 | // Let's make sure there's no overflow during the |
| 574 | // calculation below by doing everything with 128 bits. |
| 575 | let x = x as u128; |
| 576 | let factor = self.0 as u128; |
| 577 | let base = Self::BASE as u128; |
| 578 | ((base * x + factor - 1) / factor) as usize |
| 579 | } |
| 580 | } |
| 581 | |
| 582 | #[cfg (test)] |
| 583 | mod tests { |
| 584 | use super::*; |
| 585 | use std::convert::TryInto; |
| 586 | |
| 587 | enum TestConfig {} |
| 588 | |
| 589 | impl Config for TestConfig { |
| 590 | type EncodedKey = [u8; 4]; |
| 591 | type EncodedValue = [u8; 4]; |
| 592 | |
| 593 | type Key = u32; |
| 594 | type Value = u32; |
| 595 | |
| 596 | type H = FxHashFn; |
| 597 | |
| 598 | fn encode_key(k: &Self::Key) -> Self::EncodedKey { |
| 599 | k.to_le_bytes() |
| 600 | } |
| 601 | |
| 602 | fn encode_value(v: &Self::Value) -> Self::EncodedValue { |
| 603 | v.to_le_bytes() |
| 604 | } |
| 605 | |
| 606 | fn decode_key(k: &Self::EncodedKey) -> Self::Key { |
| 607 | u32::from_le_bytes(k[..].try_into().unwrap()) |
| 608 | } |
| 609 | |
| 610 | fn decode_value(v: &Self::EncodedValue) -> Self::Value { |
| 611 | u32::from_le_bytes(v[..].try_into().unwrap()) |
| 612 | } |
| 613 | } |
| 614 | |
| 615 | fn make_test_items(count: usize) -> Vec<(u32, u32)> { |
| 616 | if count == 0 { |
| 617 | return vec![]; |
| 618 | } |
| 619 | |
| 620 | let mut items = vec![]; |
| 621 | |
| 622 | if count > 1 { |
| 623 | let steps = (count - 1) as u32; |
| 624 | let step = u32::MAX / steps; |
| 625 | |
| 626 | for i in 0..steps { |
| 627 | let x = i * step; |
| 628 | items.push((x, u32::MAX - x)); |
| 629 | } |
| 630 | } |
| 631 | |
| 632 | items.push((u32::MAX, 0)); |
| 633 | |
| 634 | items.sort(); |
| 635 | items.dedup(); |
| 636 | assert_eq!(items.len(), count); |
| 637 | |
| 638 | items |
| 639 | } |
| 640 | |
| 641 | #[test ] |
| 642 | fn from_iterator() { |
| 643 | for count in 0..33 { |
| 644 | let items = make_test_items(count); |
| 645 | let table = HashTableOwned::<TestConfig>::from_iterator(items.clone(), 95); |
| 646 | assert_eq!(table.len(), items.len()); |
| 647 | |
| 648 | let mut actual_items: Vec<_> = table.iter().collect(); |
| 649 | actual_items.sort(); |
| 650 | |
| 651 | assert_eq!(items, actual_items); |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | #[test ] |
| 656 | fn init_in_place() { |
| 657 | for count in 0..33 { |
| 658 | let items = make_test_items(count); |
| 659 | let byte_count = bytes_needed::<TestConfig>(items.len(), 87); |
| 660 | let data = vec![0u8; byte_count]; |
| 661 | |
| 662 | let mut table = |
| 663 | HashTable::<TestConfig, _>::init_in_place(data, items.len(), 87).unwrap(); |
| 664 | |
| 665 | for (i, (k, v)) in items.iter().enumerate() { |
| 666 | assert_eq!(table.len(), i); |
| 667 | assert_eq!(table.insert(k, v), None); |
| 668 | assert_eq!(table.len(), i + 1); |
| 669 | |
| 670 | // Make sure we still can find all items previously inserted. |
| 671 | for (k, v) in items.iter().take(i) { |
| 672 | assert_eq!(table.get(k), Some(*v)); |
| 673 | } |
| 674 | } |
| 675 | |
| 676 | let mut actual_items: Vec<_> = table.iter().collect(); |
| 677 | actual_items.sort(); |
| 678 | |
| 679 | assert_eq!(items, actual_items); |
| 680 | } |
| 681 | } |
| 682 | |
| 683 | #[test ] |
| 684 | fn hash_table_at_different_alignments() { |
| 685 | let items = make_test_items(33); |
| 686 | |
| 687 | let mut serialized = { |
| 688 | let table: HashTableOwned<TestConfig> = |
| 689 | HashTableOwned::from_iterator(items.clone(), 95); |
| 690 | |
| 691 | assert_eq!(table.len(), items.len()); |
| 692 | |
| 693 | table.raw_bytes().to_owned() |
| 694 | }; |
| 695 | |
| 696 | for alignment_shift in 0..4 { |
| 697 | let data = &serialized[alignment_shift..]; |
| 698 | |
| 699 | let table = HashTable::<TestConfig, _>::from_raw_bytes(data).unwrap(); |
| 700 | |
| 701 | assert_eq!(table.len(), items.len()); |
| 702 | |
| 703 | for (key, value) in items.iter() { |
| 704 | assert_eq!(table.get(key), Some(*value)); |
| 705 | } |
| 706 | |
| 707 | serialized.insert(0, 0xFFu8); |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | #[test ] |
| 712 | fn load_factor_and_item_count() { |
| 713 | assert_eq!( |
| 714 | slots_needed(0, Factor::from_percent(100)), |
| 715 | REFERENCE_GROUP_SIZE |
| 716 | ); |
| 717 | assert_eq!(slots_needed(6, Factor::from_percent(60)), 16); |
| 718 | assert_eq!(slots_needed(5, Factor::from_percent(50)), 16); |
| 719 | assert_eq!(slots_needed(5, Factor::from_percent(49)), 16); |
| 720 | assert_eq!(slots_needed(1000, Factor::from_percent(100)), 1024); |
| 721 | |
| 722 | // Factor cannot never be a full 100% because of the rounding involved. |
| 723 | assert_eq!(max_item_count_for(10, Factor::from_percent(100)), 9); |
| 724 | assert_eq!(max_item_count_for(10, Factor::from_percent(50)), 4); |
| 725 | assert_eq!(max_item_count_for(11, Factor::from_percent(50)), 5); |
| 726 | assert_eq!(max_item_count_for(12, Factor::from_percent(50)), 5); |
| 727 | } |
| 728 | |
| 729 | #[test ] |
| 730 | fn grow() { |
| 731 | let items = make_test_items(100); |
| 732 | let mut table = HashTableOwned::<TestConfig>::with_capacity(10, 87); |
| 733 | |
| 734 | for (key, value) in items.iter() { |
| 735 | assert_eq!(table.insert(key, value), None); |
| 736 | } |
| 737 | } |
| 738 | |
| 739 | #[test ] |
| 740 | fn factor_from_percent() { |
| 741 | assert_eq!(Factor::from_percent(100), Factor(u16::MAX)); |
| 742 | assert_eq!(Factor::from_percent(0), Factor(0)); |
| 743 | assert_eq!(Factor::from_percent(50), Factor(u16::MAX / 2)); |
| 744 | } |
| 745 | |
| 746 | #[test ] |
| 747 | fn factor_apply() { |
| 748 | assert_eq!(Factor::from_percent(100).apply(12345), 12344); |
| 749 | assert_eq!(Factor::from_percent(0).apply(12345), 0); |
| 750 | assert_eq!(Factor::from_percent(50).apply(66), 32); |
| 751 | |
| 752 | // Make sure we can handle large numbers without overflow |
| 753 | assert_basically_equal(Factor::from_percent(100).apply(usize::MAX), usize::MAX); |
| 754 | } |
| 755 | |
| 756 | #[test ] |
| 757 | fn factor_apply_inverse() { |
| 758 | assert_eq!(Factor::from_percent(100).apply_inverse(12345), 12345); |
| 759 | assert_eq!(Factor::from_percent(10).apply_inverse(100), 1001); |
| 760 | assert_eq!(Factor::from_percent(50).apply_inverse(33), 67); |
| 761 | |
| 762 | // // Make sure we can handle large numbers without overflow |
| 763 | assert_basically_equal( |
| 764 | Factor::from_percent(100).apply_inverse(usize::MAX), |
| 765 | usize::MAX, |
| 766 | ); |
| 767 | } |
| 768 | |
| 769 | fn assert_basically_equal(x: usize, y: usize) { |
| 770 | let larger_number = std::cmp::max(x, y) as f64; |
| 771 | let abs_difference = (x as f64 - y as f64).abs(); |
| 772 | let difference_in_percent = (abs_difference / larger_number) * 100.0; |
| 773 | |
| 774 | const MAX_ALLOWED_DIFFERENCE_IN_PERCENT: f64 = 0.01; |
| 775 | |
| 776 | assert!( |
| 777 | difference_in_percent < MAX_ALLOWED_DIFFERENCE_IN_PERCENT, |
| 778 | "{} and {} differ by {:.4} percent but the maximally allowed difference \ |
| 779 | is {:.2} percent. Large differences might be caused by integer overflow." , |
| 780 | x, |
| 781 | y, |
| 782 | difference_in_percent, |
| 783 | MAX_ALLOWED_DIFFERENCE_IN_PERCENT |
| 784 | ); |
| 785 | } |
| 786 | |
| 787 | mod quickchecks { |
| 788 | use super::*; |
| 789 | use crate::raw_table::ByteArray; |
| 790 | use quickcheck::{Arbitrary, Gen}; |
| 791 | use rustc_hash::FxHashMap; |
| 792 | |
| 793 | #[derive (Copy, Clone, Hash, Eq, PartialEq, Debug)] |
| 794 | struct Bytes<const BYTE_COUNT: usize>([u8; BYTE_COUNT]); |
| 795 | |
| 796 | impl<const L: usize> Arbitrary for Bytes<L> { |
| 797 | fn arbitrary(gen: &mut Gen) -> Self { |
| 798 | let mut xs = [0; L]; |
| 799 | for x in xs.iter_mut() { |
| 800 | *x = u8::arbitrary(gen); |
| 801 | } |
| 802 | Bytes(xs) |
| 803 | } |
| 804 | } |
| 805 | |
| 806 | impl<const L: usize> Default for Bytes<L> { |
| 807 | fn default() -> Self { |
| 808 | Bytes([0; L]) |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | impl<const L: usize> ByteArray for Bytes<L> { |
| 813 | #[inline (always)] |
| 814 | fn zeroed() -> Self { |
| 815 | Bytes([0u8; L]) |
| 816 | } |
| 817 | |
| 818 | #[inline (always)] |
| 819 | fn as_slice(&self) -> &[u8] { |
| 820 | &self.0[..] |
| 821 | } |
| 822 | |
| 823 | #[inline (always)] |
| 824 | fn equals(&self, other: &Self) -> bool { |
| 825 | self.as_slice() == other.as_slice() |
| 826 | } |
| 827 | } |
| 828 | |
| 829 | macro_rules! mk_quick_tests { |
| 830 | ($name: ident, $key_len:expr, $value_len:expr) => { |
| 831 | mod $name { |
| 832 | use super::*; |
| 833 | use quickcheck::quickcheck; |
| 834 | |
| 835 | struct Cfg; |
| 836 | |
| 837 | type Key = Bytes<$key_len>; |
| 838 | type Value = Bytes<$value_len>; |
| 839 | |
| 840 | impl Config for Cfg { |
| 841 | type EncodedKey = Key; |
| 842 | type EncodedValue = Value; |
| 843 | |
| 844 | type Key = Key; |
| 845 | type Value = Value; |
| 846 | |
| 847 | type H = FxHashFn; |
| 848 | |
| 849 | fn encode_key(k: &Self::Key) -> Self::EncodedKey { |
| 850 | *k |
| 851 | } |
| 852 | |
| 853 | fn encode_value(v: &Self::Value) -> Self::EncodedValue { |
| 854 | *v |
| 855 | } |
| 856 | |
| 857 | fn decode_key(k: &Self::EncodedKey) -> Self::Key { |
| 858 | *k |
| 859 | } |
| 860 | |
| 861 | fn decode_value(v: &Self::EncodedValue) -> Self::Value { |
| 862 | *v |
| 863 | } |
| 864 | } |
| 865 | |
| 866 | fn from_std_hashmap(m: &FxHashMap<Key, Value>) -> HashTableOwned<Cfg> { |
| 867 | HashTableOwned::<Cfg>::from_iterator(m.iter().map(|(x, y)| (*x, *y)), 87) |
| 868 | } |
| 869 | |
| 870 | quickcheck! { |
| 871 | fn len(xs: FxHashMap<Key, Value>) -> bool { |
| 872 | let table = from_std_hashmap(&xs); |
| 873 | |
| 874 | xs.len() == table.len() |
| 875 | } |
| 876 | } |
| 877 | |
| 878 | quickcheck! { |
| 879 | fn lookup(xs: FxHashMap<Key, Value>) -> bool { |
| 880 | let table = from_std_hashmap(&xs); |
| 881 | xs.iter().all(|(k, v)| table.get(k) == Some(*v)) |
| 882 | } |
| 883 | } |
| 884 | |
| 885 | quickcheck! { |
| 886 | fn insert_with_duplicates(xs: Vec<(Key, Value)>) -> bool { |
| 887 | let mut reference = FxHashMap::default(); |
| 888 | let mut table = HashTableOwned::<Cfg>::default(); |
| 889 | |
| 890 | for (k, v) in xs { |
| 891 | let expected = reference.insert(k, v); |
| 892 | let actual = table.insert(&k, &v); |
| 893 | |
| 894 | if expected != actual { |
| 895 | return false; |
| 896 | } |
| 897 | } |
| 898 | |
| 899 | true |
| 900 | } |
| 901 | } |
| 902 | |
| 903 | quickcheck! { |
| 904 | fn bytes_deterministic(xs: FxHashMap<Key, Value>) -> bool { |
| 905 | // NOTE: We only guarantee this given the exact same |
| 906 | // insertion order. |
| 907 | let table0 = from_std_hashmap(&xs); |
| 908 | let table1 = from_std_hashmap(&xs); |
| 909 | |
| 910 | table0.raw_bytes() == table1.raw_bytes() |
| 911 | } |
| 912 | } |
| 913 | |
| 914 | quickcheck! { |
| 915 | fn from_iterator_vs_manual_insertion(xs: Vec<(Key, Value)>) -> bool { |
| 916 | let mut table0 = HashTableOwned::<Cfg>::with_capacity(xs.len(), 87); |
| 917 | |
| 918 | for (k, v) in xs.iter() { |
| 919 | table0.insert(k, v); |
| 920 | } |
| 921 | |
| 922 | let table1 = HashTableOwned::<Cfg>::from_iterator(xs.into_iter(), 87); |
| 923 | |
| 924 | // Requiring bit for bit equality might be a bit too much in this case, |
| 925 | // as long as it works ... |
| 926 | table0.raw_bytes() == table1.raw_bytes() |
| 927 | } |
| 928 | } |
| 929 | } |
| 930 | }; |
| 931 | } |
| 932 | |
| 933 | // Test zero sized key and values |
| 934 | mk_quick_tests!(k0_v0, 0, 0); |
| 935 | mk_quick_tests!(k1_v0, 1, 0); |
| 936 | mk_quick_tests!(k2_v0, 2, 0); |
| 937 | mk_quick_tests!(k3_v0, 3, 0); |
| 938 | mk_quick_tests!(k4_v0, 4, 0); |
| 939 | mk_quick_tests!(k8_v0, 8, 0); |
| 940 | mk_quick_tests!(k15_v0, 15, 0); |
| 941 | mk_quick_tests!(k16_v0, 16, 0); |
| 942 | mk_quick_tests!(k17_v0, 17, 0); |
| 943 | mk_quick_tests!(k63_v0, 63, 0); |
| 944 | mk_quick_tests!(k64_v0, 64, 0); |
| 945 | |
| 946 | // Test a few different key sizes |
| 947 | mk_quick_tests!(k2_v4, 2, 4); |
| 948 | mk_quick_tests!(k4_v4, 4, 4); |
| 949 | mk_quick_tests!(k8_v4, 8, 4); |
| 950 | mk_quick_tests!(k17_v4, 17, 4); |
| 951 | mk_quick_tests!(k20_v4, 20, 4); |
| 952 | mk_quick_tests!(k64_v4, 64, 4); |
| 953 | |
| 954 | // Test a few different value sizes |
| 955 | mk_quick_tests!(k16_v1, 16, 1); |
| 956 | mk_quick_tests!(k16_v2, 16, 2); |
| 957 | mk_quick_tests!(k16_v3, 16, 3); |
| 958 | mk_quick_tests!(k16_v4, 16, 4); |
| 959 | mk_quick_tests!(k16_v8, 16, 8); |
| 960 | mk_quick_tests!(k16_v16, 16, 16); |
| 961 | mk_quick_tests!(k16_v17, 16, 17); |
| 962 | } |
| 963 | } |
| 964 | |