mirror of
https://github.com/italicsjenga/agb.git
synced 2024-12-23 08:11:33 +11:00
Refactor and slightly improve hashmap (#429)
Started as an attempt to make it easier to develop the hashmap, ended up with me benchmarking everything :). - [x] no changelog update needed
This commit is contained in:
commit
458280b68f
|
@ -11,6 +11,7 @@ rustc-hash = { version = "1", default-features = false }
|
|||
|
||||
[dev-dependencies]
|
||||
rand = { version = "0.8", default-features = false, features = ["small_rng"] }
|
||||
lazy_static = "1.4"
|
||||
|
||||
[profile.dev]
|
||||
opt-level = 3
|
||||
|
|
318
agb-hashmap/benches/bench.rs
Normal file
318
agb-hashmap/benches/bench.rs
Normal file
|
@ -0,0 +1,318 @@
|
|||
// These benchmarks were taken from hashbrown. They are impossible to run
|
||||
// on the target GBA hardware, but hopefully running these on something like a
|
||||
// raspberry pi zero will give something comparable.
|
||||
|
||||
// This benchmark suite contains some benchmarks along a set of dimensions:
|
||||
// Int key distribution: low bit heavy, top bit heavy, and random.
|
||||
// Task: basic functionality: insert, insert_erase, lookup, lookup_fail, iter
|
||||
#![feature(test)]
|
||||
|
||||
extern crate test;
|
||||
|
||||
use test::{black_box, Bencher};
|
||||
|
||||
use agb_hashmap::HashMap;
|
||||
use std::sync::atomic::{self, AtomicUsize};
|
||||
|
||||
const SIZE: usize = 1000;
|
||||
|
||||
type StdHashMap<K, V> = std::collections::hash_map::HashMap<K, V>;
|
||||
|
||||
// A random key iterator.
|
||||
#[derive(Clone, Copy)]
|
||||
struct RandomKeys {
|
||||
state: usize,
|
||||
}
|
||||
|
||||
impl RandomKeys {
|
||||
fn new() -> Self {
|
||||
RandomKeys { state: 0 }
|
||||
}
|
||||
}
|
||||
|
||||
impl Iterator for RandomKeys {
|
||||
type Item = usize;
|
||||
fn next(&mut self) -> Option<usize> {
|
||||
// Add 1 then multiply by some 32 bit prime.
|
||||
self.state = self.state.wrapping_add(1).wrapping_mul(3_787_392_781);
|
||||
Some(self.state)
|
||||
}
|
||||
}
|
||||
|
||||
// Just an arbitrary side effect to make the maps not shortcircuit to the non-dropping path
|
||||
// when dropping maps/entries (most real world usages likely have drop in the key or value)
|
||||
lazy_static::lazy_static! {
|
||||
static ref SIDE_EFFECT: AtomicUsize = AtomicUsize::new(0);
|
||||
}
|
||||
|
||||
#[derive(Clone)]
|
||||
struct DropType(usize);
|
||||
impl Drop for DropType {
|
||||
fn drop(&mut self) {
|
||||
SIDE_EFFECT.fetch_add(self.0, atomic::Ordering::SeqCst);
|
||||
}
|
||||
}
|
||||
|
||||
macro_rules! bench_suite {
|
||||
($bench_macro:ident, $bench_agb_hashmap_serial:ident, $bench_std_serial:ident,
|
||||
$bench_agb_hashmap_highbits:ident, $bench_std_highbits:ident,
|
||||
$bench_agb_hashmap_random:ident, $bench_std_random:ident) => {
|
||||
$bench_macro!($bench_agb_hashmap_serial, HashMap, 0..);
|
||||
$bench_macro!($bench_std_serial, StdHashMap, 0..);
|
||||
$bench_macro!(
|
||||
$bench_agb_hashmap_highbits,
|
||||
HashMap,
|
||||
(0..).map(usize::swap_bytes)
|
||||
);
|
||||
$bench_macro!(
|
||||
$bench_std_highbits,
|
||||
StdHashMap,
|
||||
(0..).map(usize::swap_bytes)
|
||||
);
|
||||
$bench_macro!($bench_agb_hashmap_random, HashMap, RandomKeys::new());
|
||||
$bench_macro!($bench_std_random, StdHashMap, RandomKeys::new());
|
||||
};
|
||||
}
|
||||
|
||||
macro_rules! bench_insert {
|
||||
($name:ident, $maptype:ident, $keydist:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
let mut m = $maptype::with_capacity(SIZE);
|
||||
b.iter(|| {
|
||||
m.clear();
|
||||
for i in ($keydist).take(SIZE) {
|
||||
m.insert(i, (DropType(i), [i; 20]));
|
||||
}
|
||||
black_box(&mut m);
|
||||
});
|
||||
eprintln!("{}", SIDE_EFFECT.load(atomic::Ordering::SeqCst));
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
bench_suite!(
|
||||
bench_insert,
|
||||
agb_hashmap_insert_serial,
|
||||
std_hashmap_insert_serial,
|
||||
agb_hashmap_insert_highbits,
|
||||
std_hashmap_insert_highbits,
|
||||
agb_hashmap_insert_random,
|
||||
std_hashmap_insert_random
|
||||
);
|
||||
|
||||
macro_rules! bench_grow_insert {
|
||||
($name:ident, $maptype:ident, $keydist:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
b.iter(|| {
|
||||
let mut m = $maptype::default();
|
||||
for i in ($keydist).take(SIZE) {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
black_box(&mut m);
|
||||
})
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
bench_suite!(
|
||||
bench_grow_insert,
|
||||
agb_hashmap_grow_insert_serial,
|
||||
std_hashmap_grow_insert_serial,
|
||||
agb_hashmap_grow_insert_highbits,
|
||||
std_hashmap_grow_insert_highbits,
|
||||
agb_hashmap_grow_insert_random,
|
||||
std_hashmap_grow_insert_random
|
||||
);
|
||||
|
||||
macro_rules! bench_insert_erase {
|
||||
($name:ident, $maptype:ident, $keydist:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
let mut base = $maptype::default();
|
||||
for i in ($keydist).take(SIZE) {
|
||||
base.insert(i, DropType(i));
|
||||
}
|
||||
let skip = $keydist.skip(SIZE);
|
||||
b.iter(|| {
|
||||
let mut m = base.clone();
|
||||
let mut add_iter = skip.clone();
|
||||
let mut remove_iter = $keydist;
|
||||
// While keeping the size constant,
|
||||
// replace the first keydist with the second.
|
||||
for (add, remove) in (&mut add_iter).zip(&mut remove_iter).take(SIZE) {
|
||||
m.insert(add, DropType(add));
|
||||
black_box(m.remove(&remove));
|
||||
}
|
||||
black_box(m);
|
||||
});
|
||||
eprintln!("{}", SIDE_EFFECT.load(atomic::Ordering::SeqCst));
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
bench_suite!(
|
||||
bench_insert_erase,
|
||||
agb_hashmap_insert_erase_serial,
|
||||
std_hashmap_insert_erase_serial,
|
||||
agb_hashmap_insert_erase_highbits,
|
||||
std_hashmap_insert_erase_highbits,
|
||||
agb_hashmap_insert_erase_random,
|
||||
std_hashmap_insert_erase_random
|
||||
);
|
||||
|
||||
macro_rules! bench_lookup {
|
||||
($name:ident, $maptype:ident, $keydist:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
let mut m = $maptype::default();
|
||||
for i in $keydist.take(SIZE) {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
for i in $keydist.take(SIZE) {
|
||||
black_box(m.get(&i));
|
||||
}
|
||||
});
|
||||
eprintln!("{}", SIDE_EFFECT.load(atomic::Ordering::SeqCst));
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
bench_suite!(
|
||||
bench_lookup,
|
||||
agb_hashmap_lookup_serial,
|
||||
std_hashmap_lookup_serial,
|
||||
agb_hashmap_lookup_highbits,
|
||||
std_hashmap_lookup_highbits,
|
||||
agb_hashmap_lookup_random,
|
||||
std_hashmap_lookup_random
|
||||
);
|
||||
|
||||
macro_rules! bench_lookup_fail {
|
||||
($name:ident, $maptype:ident, $keydist:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
let mut m = $maptype::default();
|
||||
let mut iter = $keydist;
|
||||
for i in (&mut iter).take(SIZE) {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
for i in (&mut iter).take(SIZE) {
|
||||
black_box(m.get(&i));
|
||||
}
|
||||
})
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
bench_suite!(
|
||||
bench_lookup_fail,
|
||||
agb_hashmap_lookup_fail_serial,
|
||||
std_hashmap_lookup_fail_serial,
|
||||
agb_hashmap_lookup_fail_highbits,
|
||||
std_hashmap_lookup_fail_highbits,
|
||||
agb_hashmap_lookup_fail_random,
|
||||
std_hashmap_lookup_fail_random
|
||||
);
|
||||
|
||||
macro_rules! bench_iter {
|
||||
($name:ident, $maptype:ident, $keydist:expr) => {
|
||||
#[bench]
|
||||
fn $name(b: &mut Bencher) {
|
||||
let mut m = $maptype::default();
|
||||
for i in ($keydist).take(SIZE) {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
for i in &m {
|
||||
black_box(i);
|
||||
}
|
||||
})
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
bench_suite!(
|
||||
bench_iter,
|
||||
agb_hashmap_iter_serial,
|
||||
std_hashmap_iter_serial,
|
||||
agb_hashmap_iter_highbits,
|
||||
std_hashmap_iter_highbits,
|
||||
agb_hashmap_iter_random,
|
||||
std_hashmap_iter_random
|
||||
);
|
||||
|
||||
macro_rules! clone_bench {
|
||||
($maptype:ident) => {
|
||||
use super::DropType;
|
||||
use test::{black_box, Bencher};
|
||||
|
||||
#[bench]
|
||||
fn clone_small(b: &mut Bencher) {
|
||||
let mut m = $maptype::new();
|
||||
for i in 0..10 {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
black_box(m.clone());
|
||||
})
|
||||
}
|
||||
|
||||
#[bench]
|
||||
fn clone_from_small(b: &mut Bencher) {
|
||||
let mut m = $maptype::new();
|
||||
let mut m2 = $maptype::new();
|
||||
for i in 0..10 {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
m2.clone_from(&m);
|
||||
black_box(&mut m2);
|
||||
})
|
||||
}
|
||||
|
||||
#[bench]
|
||||
fn clone_large(b: &mut Bencher) {
|
||||
let mut m = $maptype::new();
|
||||
for i in 0..1000 {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
black_box(m.clone());
|
||||
})
|
||||
}
|
||||
|
||||
#[bench]
|
||||
fn clone_from_large(b: &mut Bencher) {
|
||||
let mut m = $maptype::new();
|
||||
let mut m2 = $maptype::new();
|
||||
for i in 0..1000 {
|
||||
m.insert(i, DropType(i));
|
||||
}
|
||||
|
||||
b.iter(|| {
|
||||
m2.clone_from(&m);
|
||||
black_box(&mut m2);
|
||||
})
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
mod agb_hashmap_clone_benches {
|
||||
use agb_hashmap::HashMap;
|
||||
clone_bench!(HashMap);
|
||||
}
|
||||
|
||||
mod std_hashmap_clone_benches {
|
||||
use std::collections::hash_map::HashMap;
|
||||
clone_bench!(HashMap);
|
||||
}
|
|
@ -14,6 +14,15 @@
|
|||
#![deny(rustdoc::broken_intra_doc_links)]
|
||||
#![deny(rustdoc::private_intra_doc_links)]
|
||||
#![deny(rustdoc::invalid_html_tags)]
|
||||
#![deny(unreachable_pub)]
|
||||
#![deny(clippy::missing_safety_doc)]
|
||||
#![deny(clippy::undocumented_unsafe_blocks)]
|
||||
#![deny(clippy::manual_assert)]
|
||||
#![deny(clippy::default_trait_access)]
|
||||
#![deny(clippy::missing_panics_doc)]
|
||||
#![deny(clippy::doc_markdown)]
|
||||
#![deny(clippy::return_self_not_must_use)]
|
||||
#![deny(clippy::cast_possible_truncation)]
|
||||
|
||||
extern crate alloc;
|
||||
|
||||
|
@ -21,16 +30,20 @@ use alloc::{alloc::Global, vec::Vec};
|
|||
use core::{
|
||||
alloc::Allocator,
|
||||
borrow::Borrow,
|
||||
fmt::Debug,
|
||||
hash::{BuildHasher, BuildHasherDefault, Hash, Hasher},
|
||||
iter::FromIterator,
|
||||
mem::{self, MaybeUninit},
|
||||
num::Wrapping,
|
||||
ops::Index,
|
||||
ptr,
|
||||
};
|
||||
|
||||
use rustc_hash::FxHasher;
|
||||
|
||||
type HashType = u32;
|
||||
mod node;
|
||||
mod node_storage;
|
||||
|
||||
use node::Node;
|
||||
use node_storage::NodeStorage;
|
||||
|
||||
// # Robin Hood Hash Tables
|
||||
//
|
||||
|
@ -96,7 +109,7 @@ type HashType = u32;
|
|||
///
|
||||
/// The API surface provided is incredibly similar to the
|
||||
/// [`std::collections::HashMap`](https://doc.rust-lang.org/std/collections/struct.HashMap.html)
|
||||
/// implementation with fewer guarantees, and better optimised for the GameBoy Advance.
|
||||
/// implementation with fewer guarantees, and better optimised for the `GameBoy Advance`.
|
||||
///
|
||||
/// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html
|
||||
/// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html
|
||||
|
@ -132,6 +145,7 @@ type HashType = u32;
|
|||
/// println!("{game}: \"{review}\"");
|
||||
/// }
|
||||
/// ```
|
||||
#[derive(Clone)]
|
||||
pub struct HashMap<K, V, ALLOCATOR: Allocator = Global> {
|
||||
nodes: NodeStorage<K, V, ALLOCATOR>,
|
||||
|
||||
|
@ -161,6 +175,12 @@ impl<K, V> HashMap<K, V> {
|
|||
pub fn with_capacity(capacity: usize) -> Self {
|
||||
Self::with_capacity_in(capacity, Global)
|
||||
}
|
||||
|
||||
#[doc(hidden)]
|
||||
#[must_use]
|
||||
pub fn distance_histogram(&self) -> (Vec<usize>, usize) {
|
||||
self.nodes.distance_histogram()
|
||||
}
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> HashMap<K, V, ALLOCATOR> {
|
||||
|
@ -170,7 +190,7 @@ impl<K, V, ALLOCATOR: ClonableAllocator> HashMap<K, V, ALLOCATOR> {
|
|||
pub fn with_size_in(size: usize, alloc: ALLOCATOR) -> Self {
|
||||
Self {
|
||||
nodes: NodeStorage::with_size_in(size, alloc),
|
||||
hasher: Default::default(),
|
||||
hasher: BuildHasherDefault::default(),
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -187,6 +207,10 @@ impl<K, V, ALLOCATOR: ClonableAllocator> HashMap<K, V, ALLOCATOR> {
|
|||
|
||||
/// Creates an empty `HashMap` which can hold at least `capacity` elements before resizing. The actual
|
||||
/// internal size may be larger as it must be a power of 2
|
||||
///
|
||||
/// # Panics
|
||||
///
|
||||
/// Panics if capacity is larger than 2^32 * .85
|
||||
#[must_use]
|
||||
pub fn with_capacity_in(capacity: usize, alloc: ALLOCATOR) -> Self {
|
||||
for i in 0..32 {
|
||||
|
@ -231,8 +255,7 @@ impl<K, V, ALLOCATOR: ClonableAllocator> HashMap<K, V, ALLOCATOR> {
|
|||
|
||||
/// Removes all elements from the map
|
||||
pub fn clear(&mut self) {
|
||||
self.nodes =
|
||||
NodeStorage::with_size_in(self.nodes.backing_vec_size(), self.allocator().clone());
|
||||
self.nodes.clear();
|
||||
}
|
||||
|
||||
/// An iterator visiting all key-value pairs in an arbitrary order
|
||||
|
@ -246,7 +269,7 @@ impl<K, V, ALLOCATOR: ClonableAllocator> HashMap<K, V, ALLOCATOR> {
|
|||
|
||||
/// An iterator visiting all key-value pairs in an arbitrary order, with mutable references to the values
|
||||
pub fn iter_mut(&mut self) -> impl Iterator<Item = (&'_ K, &'_ mut V)> {
|
||||
self.nodes.nodes.iter_mut().filter_map(Node::key_value_mut)
|
||||
self.nodes.iter_mut().filter_map(Node::key_value_mut)
|
||||
}
|
||||
|
||||
/// Retains only the elements specified by the predicate `f`.
|
||||
|
@ -282,11 +305,6 @@ impl<K, V> Default for HashMap<K, V> {
|
|||
}
|
||||
}
|
||||
|
||||
const fn fast_mod(len: usize, hash: HashType) -> usize {
|
||||
debug_assert!(len.is_power_of_two(), "Length must be a power of 2");
|
||||
(hash as usize) & (len - 1)
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> HashMap<K, V, ALLOCATOR>
|
||||
where
|
||||
K: Eq + Hash,
|
||||
|
@ -302,7 +320,13 @@ where
|
|||
let hash = self.hash(&key);
|
||||
|
||||
if let Some(location) = self.nodes.location(&key, hash) {
|
||||
Some(self.nodes.replace_at_location(location, key, value))
|
||||
Some(
|
||||
// SAFETY: location is valid due to the above
|
||||
unsafe {
|
||||
self.nodes
|
||||
.replace_at_location_unchecked(location, key, value)
|
||||
},
|
||||
)
|
||||
} else {
|
||||
if self.nodes.capacity() <= self.len() {
|
||||
self.resize(self.nodes.backing_vec_size() * 2);
|
||||
|
@ -318,7 +342,11 @@ where
|
|||
let hash = self.hash(&key);
|
||||
|
||||
let location = if let Some(location) = self.nodes.location(&key, hash) {
|
||||
self.nodes.replace_at_location(location, key, value);
|
||||
// SAFETY: location is valid due to the above
|
||||
unsafe {
|
||||
self.nodes
|
||||
.replace_at_location_unchecked(location, key, value);
|
||||
}
|
||||
location
|
||||
} else {
|
||||
if self.nodes.capacity() <= self.len() {
|
||||
|
@ -328,7 +356,12 @@ where
|
|||
self.nodes.insert_new(key, value, hash)
|
||||
};
|
||||
|
||||
self.nodes.nodes[location].value_mut().unwrap()
|
||||
// SAFETY: location is always valid
|
||||
unsafe {
|
||||
self.nodes
|
||||
.node_at_unchecked_mut(location)
|
||||
.value_mut_unchecked()
|
||||
}
|
||||
}
|
||||
|
||||
/// Returns `true` if the map contains a value for the specified key.
|
||||
|
@ -349,9 +382,15 @@ where
|
|||
{
|
||||
let hash = self.hash(key);
|
||||
|
||||
self.nodes
|
||||
.location(key, hash)
|
||||
.and_then(|location| self.nodes.nodes[location].key_value_ref())
|
||||
let location = self.nodes.location(key, hash)?;
|
||||
Some(
|
||||
// SAFETY: we know that a node exists and has a value from the location call above
|
||||
unsafe {
|
||||
self.nodes
|
||||
.node_at_unchecked(location)
|
||||
.key_value_ref_unchecked()
|
||||
},
|
||||
)
|
||||
}
|
||||
|
||||
/// Returns a reference to the value corresponding to the key. Returns [`None`] if there is
|
||||
|
@ -397,11 +436,15 @@ where
|
|||
{
|
||||
let hash = self.hash(key);
|
||||
|
||||
if let Some(location) = self.nodes.location(key, hash) {
|
||||
self.nodes.nodes[location].value_mut()
|
||||
} else {
|
||||
None
|
||||
}
|
||||
let location = self.nodes.location(key, hash)?;
|
||||
Some(
|
||||
// SAFETY: we know that a node exists and has a value from the location call above
|
||||
unsafe {
|
||||
self.nodes
|
||||
.node_at_unchecked_mut(location)
|
||||
.value_mut_unchecked()
|
||||
},
|
||||
)
|
||||
}
|
||||
|
||||
/// Removes the given key from the map. Returns the current value if it existed, or [`None`]
|
||||
|
@ -440,7 +483,12 @@ where
|
|||
{
|
||||
let mut hasher = self.hasher.build_hasher();
|
||||
key.hash(&mut hasher);
|
||||
hasher.finish() as HashType
|
||||
let result = hasher.finish();
|
||||
|
||||
// we want to allow truncation here since we're reducing 64 bits to 32
|
||||
#[allow(clippy::cast_possible_truncation)]
|
||||
let reduced = (result as u32) ^ ((result >> 32) as u32);
|
||||
HashType::bit_mix(reduced)
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -463,12 +511,12 @@ impl<'a, K, V, ALLOCATOR: ClonableAllocator> Iterator for Iter<'a, K, V, ALLOCAT
|
|||
return None;
|
||||
}
|
||||
|
||||
let node = &self.map.nodes.nodes[self.at];
|
||||
let node = &self.map.nodes.node_at(self.at);
|
||||
self.at += 1;
|
||||
|
||||
if node.has_value() {
|
||||
if let Some(key_value) = node.key_value_ref() {
|
||||
self.num_found += 1;
|
||||
return Some((node.key_ref().unwrap(), node.value_ref().unwrap()));
|
||||
return Some(key_value);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
@ -497,7 +545,7 @@ impl<'a, K, V, ALLOCATOR: ClonableAllocator> IntoIterator for &'a HashMap<K, V,
|
|||
/// An iterator over entries of a [`HashMap`]
|
||||
///
|
||||
/// This struct is created using the `into_iter()` method on [`HashMap`] as part of its implementation
|
||||
/// of the IntoIterator trait.
|
||||
/// of the `IntoIterator` trait.
|
||||
pub struct IterOwned<K, V, ALLOCATOR: Allocator = Global> {
|
||||
map: HashMap<K, V, ALLOCATOR>,
|
||||
at: usize,
|
||||
|
@ -513,7 +561,7 @@ impl<K, V, ALLOCATOR: ClonableAllocator> Iterator for IterOwned<K, V, ALLOCATOR>
|
|||
return None;
|
||||
}
|
||||
|
||||
let maybe_kv = self.map.nodes.nodes[self.at].take_key_value();
|
||||
let maybe_kv = self.map.nodes.node_at_mut(self.at).take_key_value();
|
||||
self.at += 1;
|
||||
|
||||
if let Some((k, v, _)) = maybe_kv {
|
||||
|
@ -534,7 +582,7 @@ impl<K, V, ALLOCATOR: ClonableAllocator> Iterator for IterOwned<K, V, ALLOCATOR>
|
|||
/// An iterator over entries of a [`HashMap`]
|
||||
///
|
||||
/// This struct is created using the `into_iter()` method on [`HashMap`] as part of its implementation
|
||||
/// of the IntoIterator trait.
|
||||
/// of the `IntoIterator` trait.
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> IntoIterator for HashMap<K, V, ALLOCATOR> {
|
||||
type Item = (K, V);
|
||||
type IntoIter = IterOwned<K, V, ALLOCATOR>;
|
||||
|
@ -548,86 +596,133 @@ impl<K, V, ALLOCATOR: ClonableAllocator> IntoIterator for HashMap<K, V, ALLOCATO
|
|||
}
|
||||
}
|
||||
|
||||
/// A view into an occupied entry in a `HashMap`. This is part of the [`Entry`] enum.
|
||||
pub struct OccupiedEntry<'a, K: 'a, V: 'a, ALLOCATOR: Allocator> {
|
||||
key: K,
|
||||
map: &'a mut HashMap<K, V, ALLOCATOR>,
|
||||
location: usize,
|
||||
}
|
||||
mod entries {
|
||||
use core::{alloc::Allocator, hash::Hash};
|
||||
|
||||
impl<'a, K: 'a, V: 'a, ALLOCATOR: ClonableAllocator> OccupiedEntry<'a, K, V, ALLOCATOR> {
|
||||
/// Gets a reference to the key in the entry.
|
||||
pub fn key(&self) -> &K {
|
||||
&self.key
|
||||
use super::{ClonableAllocator, HashMap};
|
||||
|
||||
/// A view into an occupied entry in a `HashMap`. This is part of the [`crate::Entry`] enum.
|
||||
pub struct OccupiedEntry<'a, K: 'a, V: 'a, ALLOCATOR: Allocator> {
|
||||
key: K,
|
||||
map: &'a mut HashMap<K, V, ALLOCATOR>,
|
||||
location: usize,
|
||||
}
|
||||
|
||||
/// Take the ownership of the key and value from the map.
|
||||
pub fn remove_entry(self) -> (K, V) {
|
||||
let old_value = self.map.nodes.remove_from_location(self.location);
|
||||
(self.key, old_value)
|
||||
impl<'a, K: 'a, V: 'a, ALLOCATOR: ClonableAllocator> OccupiedEntry<'a, K, V, ALLOCATOR> {
|
||||
/// # Safety
|
||||
///
|
||||
/// You must call this with a valid location (one where the entry is defined)
|
||||
pub(crate) unsafe fn new(
|
||||
key: K,
|
||||
map: &'a mut HashMap<K, V, ALLOCATOR>,
|
||||
location: usize,
|
||||
) -> Self {
|
||||
Self { key, map, location }
|
||||
}
|
||||
|
||||
/// Gets a reference to the key in the entry.
|
||||
pub fn key(&self) -> &K {
|
||||
&self.key
|
||||
}
|
||||
|
||||
/// Take the ownership of the key and value from the map.
|
||||
pub fn remove_entry(self) -> (K, V) {
|
||||
let old_value = self.map.nodes.remove_from_location(self.location);
|
||||
(self.key, old_value)
|
||||
}
|
||||
|
||||
/// Gets a reference to the value in the entry.
|
||||
pub fn get(&self) -> &V {
|
||||
// SAFETY: This can only be constructed with valid locations
|
||||
unsafe {
|
||||
self.map
|
||||
.nodes
|
||||
.node_at_unchecked(self.location)
|
||||
.value_ref_unchecked()
|
||||
}
|
||||
}
|
||||
|
||||
/// Gets a mutable reference to the value in the entry.
|
||||
///
|
||||
/// If you need a reference to the `OccupiedEntry` which may outlive the destruction
|
||||
/// of the `Entry` value, see [`into_mut`].
|
||||
///
|
||||
/// [`into_mut`]: Self::into_mut
|
||||
pub fn get_mut(&mut self) -> &mut V {
|
||||
// SAFETY: This can only be constructed with valid locations
|
||||
unsafe {
|
||||
self.map
|
||||
.nodes
|
||||
.node_at_unchecked_mut(self.location)
|
||||
.value_mut_unchecked()
|
||||
}
|
||||
}
|
||||
|
||||
/// Converts the `OccupiedEntry` into a mutable reference to the value in the entry with
|
||||
/// a lifetime bound to the map itself.
|
||||
///
|
||||
/// If you need multiple references to the `OccupiedEntry`, see [`get_mut`].
|
||||
///
|
||||
/// [`get_mut`]: Self::get_mut
|
||||
pub fn into_mut(self) -> &'a mut V {
|
||||
// SAFETY: This can only be constructed with valid locations
|
||||
unsafe {
|
||||
self.map
|
||||
.nodes
|
||||
.node_at_unchecked_mut(self.location)
|
||||
.value_mut_unchecked()
|
||||
}
|
||||
}
|
||||
|
||||
/// Sets the value of the entry and returns the entry's old value.
|
||||
pub fn insert(&mut self, value: V) -> V {
|
||||
// SAFETY: This can only be constructed with valid locations
|
||||
unsafe {
|
||||
self.map
|
||||
.nodes
|
||||
.node_at_unchecked_mut(self.location)
|
||||
.replace_value_unchecked(value)
|
||||
}
|
||||
}
|
||||
|
||||
/// Takes the value out of the entry and returns it.
|
||||
pub fn remove(self) -> V {
|
||||
self.map.nodes.remove_from_location(self.location)
|
||||
}
|
||||
}
|
||||
|
||||
/// Gets a reference to the value in the entry.
|
||||
pub fn get(&self) -> &V {
|
||||
self.map.nodes.nodes[self.location].value_ref().unwrap()
|
||||
/// A view into a vacant entry in a `HashMap`. It is part of the [`crate::Entry`] enum.
|
||||
pub struct VacantEntry<'a, K: 'a, V: 'a, ALLOCATOR: Allocator> {
|
||||
key: K,
|
||||
map: &'a mut HashMap<K, V, ALLOCATOR>,
|
||||
}
|
||||
|
||||
/// Gets a mutable reference to the value in the entry.
|
||||
///
|
||||
/// If you need a reference to the `OccupiedEntry` which may outlive the destruction
|
||||
/// of the `Entry` value, see [`into_mut`].
|
||||
///
|
||||
/// [`into_mut`]: Self::into_mut
|
||||
pub fn get_mut(&mut self) -> &mut V {
|
||||
self.map.nodes.nodes[self.location].value_mut().unwrap()
|
||||
}
|
||||
impl<'a, K: 'a, V: 'a, ALLOCATOR: ClonableAllocator> VacantEntry<'a, K, V, ALLOCATOR> {
|
||||
pub(crate) fn new(key: K, map: &'a mut HashMap<K, V, ALLOCATOR>) -> Self {
|
||||
Self { key, map }
|
||||
}
|
||||
|
||||
/// Converts the `OccupiedEntry` into a mutable reference to the value in the entry with
|
||||
/// a lifetime bound to the map itself.
|
||||
///
|
||||
/// If you need multiple references to the `OccupiedEntry`, see [`get_mut`].
|
||||
///
|
||||
/// [`get_mut`]: Self::get_mut
|
||||
pub fn into_mut(self) -> &'a mut V {
|
||||
self.map.nodes.nodes[self.location].value_mut().unwrap()
|
||||
}
|
||||
/// Gets a reference to the key that would be used when inserting a value through `VacantEntry`
|
||||
pub fn key(&self) -> &K {
|
||||
&self.key
|
||||
}
|
||||
|
||||
/// Sets the value of the entry and returns the entry's old value.
|
||||
pub fn insert(&mut self, value: V) -> V {
|
||||
self.map.nodes.nodes[self.location].replace_value(value)
|
||||
}
|
||||
/// Take ownership of the key
|
||||
pub fn into_key(self) -> K {
|
||||
self.key
|
||||
}
|
||||
|
||||
/// Takes the value out of the entry and returns it.
|
||||
pub fn remove(self) -> V {
|
||||
self.map.nodes.remove_from_location(self.location)
|
||||
/// Sets the value of the entry with the `VacantEntry`'s key and returns a mutable reference to it.
|
||||
pub fn insert(self, value: V) -> &'a mut V
|
||||
where
|
||||
K: Hash + Eq,
|
||||
{
|
||||
self.map.insert_and_get(self.key, value)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// A view into a vacant entry in a `HashMap`. It is part of the [`Entry`] enum.
|
||||
pub struct VacantEntry<'a, K: 'a, V: 'a, ALLOCATOR: Allocator> {
|
||||
key: K,
|
||||
map: &'a mut HashMap<K, V, ALLOCATOR>,
|
||||
}
|
||||
|
||||
impl<'a, K: 'a, V: 'a, ALLOCATOR: ClonableAllocator> VacantEntry<'a, K, V, ALLOCATOR> {
|
||||
/// Gets a reference to the key that would be used when inserting a value through `VacantEntry`
|
||||
pub fn key(&self) -> &K {
|
||||
&self.key
|
||||
}
|
||||
|
||||
/// Take ownership of the key
|
||||
pub fn into_key(self) -> K {
|
||||
self.key
|
||||
}
|
||||
|
||||
/// Sets the value of the entry with the `VacantEntry`'s key and returns a mutable reference to it.
|
||||
pub fn insert(self, value: V) -> &'a mut V
|
||||
where
|
||||
K: Hash + Eq,
|
||||
{
|
||||
self.map.insert_and_get(self.key, value)
|
||||
}
|
||||
}
|
||||
pub use entries::{OccupiedEntry, VacantEntry};
|
||||
|
||||
/// A view into a single entry in a map, which may be vacant or occupied.
|
||||
///
|
||||
|
@ -679,7 +774,7 @@ where
|
|||
match self {
|
||||
Entry::Occupied(e) => e.into_mut(),
|
||||
Entry::Vacant(e) => {
|
||||
let value = f(&e.key);
|
||||
let value = f(e.key());
|
||||
e.insert(value)
|
||||
}
|
||||
}
|
||||
|
@ -687,6 +782,7 @@ where
|
|||
|
||||
/// Provides in-place mutable access to an occupied entry before any potential inserts
|
||||
/// into the map.
|
||||
#[must_use]
|
||||
pub fn and_modify<F>(self, f: F) -> Self
|
||||
where
|
||||
F: FnOnce(&mut V),
|
||||
|
@ -715,8 +811,8 @@ where
|
|||
/// Returns a reference to this entry's key.
|
||||
pub fn key(&self) -> &K {
|
||||
match self {
|
||||
Entry::Occupied(e) => &e.key,
|
||||
Entry::Vacant(e) => &e.key,
|
||||
Entry::Occupied(e) => e.key(),
|
||||
Entry::Vacant(e) => e.key(),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
@ -731,13 +827,12 @@ where
|
|||
let location = self.nodes.location(&key, hash);
|
||||
|
||||
if let Some(location) = location {
|
||||
Entry::Occupied(OccupiedEntry {
|
||||
key,
|
||||
location,
|
||||
map: self,
|
||||
})
|
||||
Entry::Occupied(
|
||||
// SAFETY: location is valid by the call to location above
|
||||
unsafe { OccupiedEntry::new(key, self, location) },
|
||||
)
|
||||
} else {
|
||||
Entry::Vacant(VacantEntry { key, map: self })
|
||||
Entry::Vacant(VacantEntry::new(key, self))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
@ -776,313 +871,81 @@ where
|
|||
}
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> PartialEq for HashMap<K, V, ALLOCATOR>
|
||||
where
|
||||
K: Eq + Hash,
|
||||
V: PartialEq,
|
||||
{
|
||||
fn eq(&self, other: &HashMap<K, V, ALLOCATOR>) -> bool {
|
||||
if self.len() != other.len() {
|
||||
return false;
|
||||
}
|
||||
|
||||
self.iter()
|
||||
.all(|(key, value)| other.get(key).map_or(false, |v| *value == *v))
|
||||
}
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> Eq for HashMap<K, V, ALLOCATOR>
|
||||
where
|
||||
K: Eq + Hash,
|
||||
V: PartialEq,
|
||||
{
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> Debug for HashMap<K, V, ALLOCATOR>
|
||||
where
|
||||
K: Debug,
|
||||
V: Debug,
|
||||
{
|
||||
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
|
||||
f.debug_map().entries(self.iter()).finish()
|
||||
}
|
||||
}
|
||||
|
||||
const fn number_before_resize(capacity: usize) -> usize {
|
||||
capacity * 85 / 100
|
||||
capacity * 60 / 100
|
||||
}
|
||||
|
||||
struct NodeStorage<K, V, ALLOCATOR: Allocator = Global> {
|
||||
nodes: Vec<Node<K, V>, ALLOCATOR>,
|
||||
max_distance_to_initial_bucket: i32,
|
||||
#[derive(Clone, Copy, PartialEq, Eq)]
|
||||
pub(crate) struct HashType(u32);
|
||||
|
||||
number_of_items: usize,
|
||||
max_number_before_resize: usize,
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> NodeStorage<K, V, ALLOCATOR> {
|
||||
fn with_size_in(capacity: usize, alloc: ALLOCATOR) -> Self {
|
||||
assert!(capacity.is_power_of_two(), "Capacity must be a power of 2");
|
||||
|
||||
let mut nodes = Vec::with_capacity_in(capacity, alloc);
|
||||
for _ in 0..capacity {
|
||||
nodes.push(Default::default());
|
||||
}
|
||||
|
||||
Self {
|
||||
nodes,
|
||||
max_distance_to_initial_bucket: 0,
|
||||
number_of_items: 0,
|
||||
max_number_before_resize: number_before_resize(capacity),
|
||||
}
|
||||
}
|
||||
|
||||
fn allocator(&self) -> &ALLOCATOR {
|
||||
self.nodes.allocator()
|
||||
}
|
||||
|
||||
fn capacity(&self) -> usize {
|
||||
self.max_number_before_resize
|
||||
}
|
||||
|
||||
fn backing_vec_size(&self) -> usize {
|
||||
self.nodes.len()
|
||||
}
|
||||
|
||||
fn len(&self) -> usize {
|
||||
self.number_of_items
|
||||
}
|
||||
|
||||
fn insert_new(&mut self, key: K, value: V, hash: HashType) -> usize {
|
||||
debug_assert!(
|
||||
self.capacity() > self.len(),
|
||||
"Do not have space to insert into len {} with {}",
|
||||
self.backing_vec_size(),
|
||||
self.len()
|
||||
);
|
||||
|
||||
let mut new_node = Node::new_with(key, value, hash);
|
||||
let mut inserted_location = usize::MAX;
|
||||
|
||||
loop {
|
||||
let location = fast_mod(
|
||||
self.backing_vec_size(),
|
||||
new_node.hash + new_node.distance() as HashType,
|
||||
);
|
||||
let current_node = &mut self.nodes[location];
|
||||
|
||||
if current_node.has_value() {
|
||||
if current_node.distance() <= new_node.distance() {
|
||||
mem::swap(&mut new_node, current_node);
|
||||
|
||||
if inserted_location == usize::MAX {
|
||||
inserted_location = location;
|
||||
}
|
||||
}
|
||||
} else {
|
||||
self.nodes[location] = new_node;
|
||||
if inserted_location == usize::MAX {
|
||||
inserted_location = location;
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
new_node.increment_distance();
|
||||
self.max_distance_to_initial_bucket =
|
||||
new_node.distance().max(self.max_distance_to_initial_bucket);
|
||||
}
|
||||
|
||||
self.number_of_items += 1;
|
||||
inserted_location
|
||||
}
|
||||
|
||||
fn retain<F>(&mut self, mut f: F)
|
||||
where
|
||||
F: FnMut(&K, &mut V) -> bool,
|
||||
{
|
||||
let num_nodes = self.nodes.len();
|
||||
let mut i = 0;
|
||||
|
||||
while i < num_nodes {
|
||||
let node = &mut self.nodes[i];
|
||||
|
||||
if let Some((k, v)) = node.key_value_mut() {
|
||||
if !f(k, v) {
|
||||
self.remove_from_location(i);
|
||||
|
||||
// Need to continue before adding 1 to i because remove from location could
|
||||
// put the element which was next into the ith location in the nodes array,
|
||||
// so we need to check if that one needs removing too.
|
||||
continue;
|
||||
}
|
||||
}
|
||||
|
||||
i += 1;
|
||||
}
|
||||
}
|
||||
|
||||
fn remove_from_location(&mut self, location: usize) -> V {
|
||||
let mut current_location = location;
|
||||
self.number_of_items -= 1;
|
||||
|
||||
loop {
|
||||
let next_location =
|
||||
fast_mod(self.backing_vec_size(), (current_location + 1) as HashType);
|
||||
|
||||
// if the next node is empty, or the next location has 0 distance to initial bucket then
|
||||
// we can clear the current node
|
||||
if !self.nodes[next_location].has_value() || self.nodes[next_location].distance() == 0 {
|
||||
return self.nodes[current_location].take_key_value().unwrap().1;
|
||||
}
|
||||
|
||||
self.nodes.swap(current_location, next_location);
|
||||
self.nodes[current_location].decrement_distance();
|
||||
current_location = next_location;
|
||||
}
|
||||
}
|
||||
|
||||
fn location<Q>(&self, key: &Q, hash: HashType) -> Option<usize>
|
||||
where
|
||||
K: Borrow<Q>,
|
||||
Q: Eq + ?Sized,
|
||||
{
|
||||
for distance_to_initial_bucket in 0..(self.max_distance_to_initial_bucket + 1) {
|
||||
let location = fast_mod(
|
||||
self.nodes.len(),
|
||||
hash + distance_to_initial_bucket as HashType,
|
||||
);
|
||||
|
||||
let node = &self.nodes[location];
|
||||
if let Some(node_key_ref) = node.key_ref() {
|
||||
if node_key_ref.borrow() == key {
|
||||
return Some(location);
|
||||
}
|
||||
} else {
|
||||
return None;
|
||||
}
|
||||
}
|
||||
|
||||
None
|
||||
}
|
||||
|
||||
fn resized_to(&mut self, new_size: usize) -> Self {
|
||||
let mut new_node_storage = Self::with_size_in(new_size, self.allocator().clone());
|
||||
|
||||
for mut node in self.nodes.drain(..) {
|
||||
if let Some((key, value, hash)) = node.take_key_value() {
|
||||
new_node_storage.insert_new(key, value, hash);
|
||||
}
|
||||
}
|
||||
|
||||
new_node_storage
|
||||
}
|
||||
|
||||
fn replace_at_location(&mut self, location: usize, key: K, value: V) -> V {
|
||||
self.nodes[location].replace(key, value).1
|
||||
impl From<usize> for HashType {
|
||||
fn from(value: usize) -> Self {
|
||||
// we explicitly want to allow truncation
|
||||
#[allow(clippy::cast_possible_truncation)]
|
||||
Self(value as u32)
|
||||
}
|
||||
}
|
||||
|
||||
struct Node<K, V> {
|
||||
hash: HashType,
|
||||
|
||||
// distance_to_initial_bucket = -1 => key and value are uninit.
|
||||
// distance_to_initial_bucket >= 0 => key and value are init
|
||||
distance_to_initial_bucket: i32,
|
||||
key: MaybeUninit<K>,
|
||||
value: MaybeUninit<V>,
|
||||
}
|
||||
|
||||
impl<K, V> Node<K, V> {
|
||||
fn new() -> Self {
|
||||
Self {
|
||||
hash: 0,
|
||||
distance_to_initial_bucket: -1,
|
||||
key: MaybeUninit::uninit(),
|
||||
value: MaybeUninit::uninit(),
|
||||
}
|
||||
impl HashType {
|
||||
pub(crate) const fn new() -> Self {
|
||||
Self(0)
|
||||
}
|
||||
|
||||
fn new_with(key: K, value: V, hash: HashType) -> Self {
|
||||
Self {
|
||||
hash,
|
||||
distance_to_initial_bucket: 0,
|
||||
key: MaybeUninit::new(key),
|
||||
value: MaybeUninit::new(value),
|
||||
}
|
||||
// 32 bit mix function from here: https://github.com/skeeto/hash-prospector
|
||||
fn bit_mix(key: u32) -> Self {
|
||||
let mut key = Wrapping(key);
|
||||
key ^= key >> 16;
|
||||
key *= 0x7feb352d;
|
||||
key ^= key >> 15;
|
||||
key *= 0x846ca68b;
|
||||
key ^= key >> 16;
|
||||
|
||||
Self(key.0)
|
||||
}
|
||||
|
||||
fn value_ref(&self) -> Option<&V> {
|
||||
if self.has_value() {
|
||||
Some(unsafe { self.value.assume_init_ref() })
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
fn value_mut(&mut self) -> Option<&mut V> {
|
||||
if self.has_value() {
|
||||
Some(unsafe { self.value.assume_init_mut() })
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
fn key_ref(&self) -> Option<&K> {
|
||||
if self.distance_to_initial_bucket >= 0 {
|
||||
Some(unsafe { self.key.assume_init_ref() })
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
fn key_value_ref(&self) -> Option<(&K, &V)> {
|
||||
if self.has_value() {
|
||||
Some(unsafe { (self.key.assume_init_ref(), self.value.assume_init_ref()) })
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
fn key_value_mut(&mut self) -> Option<(&K, &mut V)> {
|
||||
if self.has_value() {
|
||||
Some(unsafe { (self.key.assume_init_ref(), self.value.assume_init_mut()) })
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
fn has_value(&self) -> bool {
|
||||
self.distance_to_initial_bucket >= 0
|
||||
}
|
||||
|
||||
fn take_key_value(&mut self) -> Option<(K, V, HashType)> {
|
||||
if self.has_value() {
|
||||
let key = mem::replace(&mut self.key, MaybeUninit::uninit());
|
||||
let value = mem::replace(&mut self.value, MaybeUninit::uninit());
|
||||
self.distance_to_initial_bucket = -1;
|
||||
|
||||
Some(unsafe { (key.assume_init(), value.assume_init(), self.hash) })
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
fn replace_value(&mut self, value: V) -> V {
|
||||
if self.has_value() {
|
||||
let old_value = mem::replace(&mut self.value, MaybeUninit::new(value));
|
||||
unsafe { old_value.assume_init() }
|
||||
} else {
|
||||
panic!("Cannot replace an uninitialised node");
|
||||
}
|
||||
}
|
||||
|
||||
fn replace(&mut self, key: K, value: V) -> (K, V) {
|
||||
if self.has_value() {
|
||||
let old_key = mem::replace(&mut self.key, MaybeUninit::new(key));
|
||||
let old_value = mem::replace(&mut self.value, MaybeUninit::new(value));
|
||||
|
||||
unsafe { (old_key.assume_init(), old_value.assume_init()) }
|
||||
} else {
|
||||
panic!("Cannot replace an uninitialised node");
|
||||
}
|
||||
}
|
||||
|
||||
fn increment_distance(&mut self) {
|
||||
self.distance_to_initial_bucket += 1;
|
||||
}
|
||||
|
||||
fn decrement_distance(&mut self) {
|
||||
self.distance_to_initial_bucket -= 1;
|
||||
if self.distance_to_initial_bucket < 0 {
|
||||
panic!("Cannot decrement distance to below 0");
|
||||
}
|
||||
}
|
||||
|
||||
fn distance(&self) -> i32 {
|
||||
self.distance_to_initial_bucket
|
||||
pub(crate) fn fast_mod(self, len: usize) -> usize {
|
||||
debug_assert!(len.is_power_of_two(), "Length must be a power of 2");
|
||||
(self.0 as usize) & (len - 1)
|
||||
}
|
||||
}
|
||||
|
||||
impl<K, V> Drop for Node<K, V> {
|
||||
fn drop(&mut self) {
|
||||
if self.has_value() {
|
||||
unsafe { ptr::drop_in_place(self.key.as_mut_ptr()) };
|
||||
unsafe { ptr::drop_in_place(self.value.as_mut_ptr()) };
|
||||
}
|
||||
}
|
||||
}
|
||||
impl core::ops::Add<i32> for HashType {
|
||||
type Output = HashType;
|
||||
|
||||
impl<K, V> Default for Node<K, V> {
|
||||
fn default() -> Self {
|
||||
Self::new()
|
||||
fn add(self, rhs: i32) -> Self::Output {
|
||||
Self(self.0.wrapping_add_signed(rhs))
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -1090,6 +953,8 @@ impl<K, V> Default for Node<K, V> {
|
|||
mod test {
|
||||
use core::cell::RefCell;
|
||||
|
||||
use alloc::vec::Vec;
|
||||
|
||||
use super::*;
|
||||
|
||||
#[test]
|
||||
|
@ -1155,7 +1020,7 @@ mod test {
|
|||
let mut max_found = -1;
|
||||
let mut num_found = 0;
|
||||
|
||||
for (_, value) in map.into_iter() {
|
||||
for (_, value) in map {
|
||||
max_found = max_found.max(value);
|
||||
num_found += 1;
|
||||
}
|
||||
|
@ -1205,9 +1070,7 @@ mod test {
|
|||
|
||||
impl Drop for NoisyDrop {
|
||||
fn drop(&mut self) {
|
||||
if self.dropped {
|
||||
panic!("NoisyDropped dropped twice");
|
||||
}
|
||||
assert!(!self.dropped, "NoisyDropped dropped twice");
|
||||
|
||||
self.dropped = true;
|
||||
}
|
||||
|
@ -1234,11 +1097,11 @@ mod test {
|
|||
let mut map = HashMap::new();
|
||||
let mut rng = rand::rngs::SmallRng::seed_from_u64(20);
|
||||
|
||||
let mut answers: [Option<i32>; 128] = [None; 128];
|
||||
let mut answers: [Option<i32>; 512] = [None; 512];
|
||||
|
||||
for _ in 0..5_000 {
|
||||
for _ in 0..15_000 {
|
||||
let command = rng.next_i32().rem_euclid(2);
|
||||
let key = rng.next_i32().rem_euclid(answers.len() as i32);
|
||||
let key = rng.next_i32().rem_euclid(answers.len().try_into().unwrap());
|
||||
let value = rng.next_i32();
|
||||
|
||||
match command {
|
||||
|
@ -1257,7 +1120,8 @@ mod test {
|
|||
|
||||
for (i, answer) in answers.iter().enumerate() {
|
||||
assert_eq!(
|
||||
map.get(&NoisyDrop::new(i as i32)).map(|nd| &nd.i),
|
||||
map.get(&NoisyDrop::new(i.try_into().unwrap()))
|
||||
.map(|nd| &nd.i),
|
||||
answer.as_ref()
|
||||
);
|
||||
}
|
||||
|
@ -1295,13 +1159,13 @@ mod test {
|
|||
}
|
||||
|
||||
impl DropRegistry {
|
||||
pub fn new() -> Self {
|
||||
fn new() -> Self {
|
||||
Self {
|
||||
are_dropped: Default::default(),
|
||||
are_dropped: RefCell::default(),
|
||||
}
|
||||
}
|
||||
|
||||
pub fn new_droppable(&self) -> Droppable<'_> {
|
||||
fn new_droppable(&self) -> Droppable<'_> {
|
||||
self.are_dropped.borrow_mut().push(0);
|
||||
Droppable {
|
||||
id: self.are_dropped.borrow().len() - 1,
|
||||
|
@ -1309,19 +1173,19 @@ mod test {
|
|||
}
|
||||
}
|
||||
|
||||
pub fn dropped(&self, id: usize) {
|
||||
fn dropped(&self, id: usize) {
|
||||
self.are_dropped.borrow_mut()[id] += 1;
|
||||
}
|
||||
|
||||
pub fn assert_dropped_once(&self, id: usize) {
|
||||
fn assert_dropped_once(&self, id: usize) {
|
||||
assert_eq!(self.are_dropped.borrow()[id], 1);
|
||||
}
|
||||
|
||||
pub fn assert_not_dropped(&self, id: usize) {
|
||||
fn assert_not_dropped(&self, id: usize) {
|
||||
assert_eq!(self.are_dropped.borrow()[id], 0);
|
||||
}
|
||||
|
||||
pub fn assert_dropped_n_times(&self, id: usize, num_drops: i32) {
|
||||
fn assert_dropped_n_times(&self, id: usize, num_drops: i32) {
|
||||
assert_eq!(self.are_dropped.borrow()[id], num_drops);
|
||||
}
|
||||
}
|
||||
|
@ -1450,6 +1314,8 @@ mod test {
|
|||
// Following test cases copied from the rust source
|
||||
// https://github.com/rust-lang/rust/blob/master/library/std/src/collections/hash/map/tests.rs
|
||||
mod rust_std_tests {
|
||||
use alloc::format;
|
||||
|
||||
use crate::{Entry::*, HashMap};
|
||||
|
||||
#[test]
|
||||
|
@ -1548,5 +1414,37 @@ mod test {
|
|||
|
||||
assert_eq!(map[&2], 1);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_eq() {
|
||||
let mut m1 = HashMap::new();
|
||||
m1.insert(1, 2);
|
||||
m1.insert(2, 3);
|
||||
m1.insert(3, 4);
|
||||
|
||||
let mut m2 = HashMap::new();
|
||||
m2.insert(1, 2);
|
||||
m2.insert(2, 3);
|
||||
|
||||
assert!(m1 != m2);
|
||||
|
||||
m2.insert(3, 4);
|
||||
|
||||
assert_eq!(m1, m2);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_show() {
|
||||
let mut map = HashMap::new();
|
||||
let empty: HashMap<i32, i32> = HashMap::new();
|
||||
|
||||
map.insert(1, 2);
|
||||
map.insert(3, 4);
|
||||
|
||||
let map_str = format!("{map:?}");
|
||||
|
||||
assert!(map_str == "{1: 2, 3: 4}" || map_str == "{3: 4, 1: 2}");
|
||||
assert_eq!(format!("{empty:?}"), "{}");
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
170
agb-hashmap/src/node.rs
Normal file
170
agb-hashmap/src/node.rs
Normal file
|
@ -0,0 +1,170 @@
|
|||
use core::{
|
||||
mem::{self, MaybeUninit},
|
||||
ptr,
|
||||
};
|
||||
|
||||
use crate::HashType;
|
||||
|
||||
pub(crate) struct Node<K, V> {
|
||||
hash: HashType,
|
||||
|
||||
// distance_to_initial_bucket = -1 => key and value are uninit.
|
||||
// distance_to_initial_bucket >= 0 => key and value are init
|
||||
distance_to_initial_bucket: i32,
|
||||
key: MaybeUninit<K>,
|
||||
value: MaybeUninit<V>,
|
||||
}
|
||||
|
||||
impl<K, V> Node<K, V> {
|
||||
pub(crate) const fn new() -> Self {
|
||||
Self {
|
||||
hash: HashType::new(),
|
||||
distance_to_initial_bucket: -1,
|
||||
key: MaybeUninit::uninit(),
|
||||
value: MaybeUninit::uninit(),
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn new_with(key: K, value: V, hash: HashType) -> Self {
|
||||
Self {
|
||||
hash,
|
||||
distance_to_initial_bucket: 0,
|
||||
key: MaybeUninit::new(key),
|
||||
value: MaybeUninit::new(value),
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn value_ref_unchecked(&self) -> &V {
|
||||
self.value.assume_init_ref()
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn value_mut_unchecked(&mut self) -> &mut V {
|
||||
self.value.assume_init_mut()
|
||||
}
|
||||
|
||||
pub(crate) fn key_ref(&self) -> Option<&K> {
|
||||
if self.distance_to_initial_bucket >= 0 {
|
||||
Some(
|
||||
// SAFETY: has a value
|
||||
unsafe { self.key.assume_init_ref() },
|
||||
)
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn key_value_ref(&self) -> Option<(&K, &V)> {
|
||||
if self.has_value() {
|
||||
Some(
|
||||
// SAFETY: has a value
|
||||
unsafe { self.key_value_ref_unchecked() },
|
||||
)
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn key_value_ref_unchecked(&self) -> (&K, &V) {
|
||||
(self.key.assume_init_ref(), self.value.assume_init_ref())
|
||||
}
|
||||
|
||||
pub(crate) fn key_value_mut(&mut self) -> Option<(&K, &mut V)> {
|
||||
if self.has_value() {
|
||||
Some(
|
||||
// SAFETY: has a value
|
||||
unsafe { (self.key.assume_init_ref(), self.value.assume_init_mut()) },
|
||||
)
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn has_value(&self) -> bool {
|
||||
self.distance_to_initial_bucket >= 0
|
||||
}
|
||||
|
||||
pub(crate) fn take_key_value(&mut self) -> Option<(K, V, HashType)> {
|
||||
if self.has_value() {
|
||||
let key = mem::replace(&mut self.key, MaybeUninit::uninit());
|
||||
let value = mem::replace(&mut self.value, MaybeUninit::uninit());
|
||||
self.distance_to_initial_bucket = -1;
|
||||
|
||||
Some(
|
||||
// SAFETY: has a value
|
||||
unsafe { (key.assume_init(), value.assume_init(), self.hash) },
|
||||
)
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn replace_value_unchecked(&mut self, value: V) -> V {
|
||||
let old_value = mem::replace(&mut self.value, MaybeUninit::new(value));
|
||||
old_value.assume_init()
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn replace_unchecked(&mut self, key: K, value: V) -> (K, V) {
|
||||
let old_key = mem::replace(&mut self.key, MaybeUninit::new(key));
|
||||
let old_value = mem::replace(&mut self.value, MaybeUninit::new(value));
|
||||
|
||||
(old_key.assume_init(), old_value.assume_init())
|
||||
}
|
||||
|
||||
pub(crate) fn increment_distance(&mut self) {
|
||||
self.distance_to_initial_bucket += 1;
|
||||
}
|
||||
|
||||
pub(crate) fn decrement_distance(&mut self) {
|
||||
self.distance_to_initial_bucket -= 1;
|
||||
|
||||
assert!(
|
||||
self.distance_to_initial_bucket >= 0,
|
||||
"Cannot decrement distance below 0"
|
||||
);
|
||||
}
|
||||
|
||||
pub(crate) fn distance(&self) -> i32 {
|
||||
self.distance_to_initial_bucket
|
||||
}
|
||||
|
||||
pub(crate) fn hash(&self) -> HashType {
|
||||
self.hash
|
||||
}
|
||||
}
|
||||
|
||||
impl<K, V> Drop for Node<K, V> {
|
||||
fn drop(&mut self) {
|
||||
if self.has_value() {
|
||||
// SAFETY: has a value
|
||||
unsafe {
|
||||
ptr::drop_in_place(self.key.as_mut_ptr());
|
||||
ptr::drop_in_place(self.value.as_mut_ptr());
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl<K, V> Clone for Node<K, V>
|
||||
where
|
||||
K: Clone,
|
||||
V: Clone,
|
||||
{
|
||||
fn clone(&self) -> Self {
|
||||
if let Some((k, v)) = self.key_value_ref() {
|
||||
Self {
|
||||
hash: self.hash,
|
||||
distance_to_initial_bucket: self.distance_to_initial_bucket,
|
||||
key: MaybeUninit::new(k.clone()),
|
||||
value: MaybeUninit::new(v.clone()),
|
||||
}
|
||||
} else {
|
||||
Self {
|
||||
hash: self.hash,
|
||||
|
||||
distance_to_initial_bucket: self.distance_to_initial_bucket,
|
||||
key: MaybeUninit::uninit(),
|
||||
value: MaybeUninit::uninit(),
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
227
agb-hashmap/src/node_storage.rs
Normal file
227
agb-hashmap/src/node_storage.rs
Normal file
|
@ -0,0 +1,227 @@
|
|||
use core::{alloc::Allocator, borrow::Borrow, mem};
|
||||
|
||||
use alloc::{alloc::Global, vec::Vec};
|
||||
|
||||
use crate::{node::Node, number_before_resize, ClonableAllocator, HashType};
|
||||
|
||||
#[derive(Clone)]
|
||||
pub(crate) struct NodeStorage<K, V, ALLOCATOR: Allocator = Global> {
|
||||
nodes: Vec<Node<K, V>, ALLOCATOR>,
|
||||
max_distance_to_initial_bucket: i32,
|
||||
|
||||
number_of_items: usize,
|
||||
max_number_before_resize: usize,
|
||||
}
|
||||
|
||||
impl<K, V, ALLOCATOR: ClonableAllocator> NodeStorage<K, V, ALLOCATOR> {
|
||||
pub(crate) fn with_size_in(capacity: usize, alloc: ALLOCATOR) -> Self {
|
||||
assert!(capacity.is_power_of_two(), "Capacity must be a power of 2");
|
||||
|
||||
let mut nodes = Vec::with_capacity_in(capacity, alloc);
|
||||
for _ in 0..capacity {
|
||||
nodes.push(Node::new());
|
||||
}
|
||||
|
||||
Self {
|
||||
nodes,
|
||||
max_distance_to_initial_bucket: 0,
|
||||
number_of_items: 0,
|
||||
max_number_before_resize: number_before_resize(capacity),
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn allocator(&self) -> &ALLOCATOR {
|
||||
self.nodes.allocator()
|
||||
}
|
||||
|
||||
pub(crate) fn capacity(&self) -> usize {
|
||||
self.max_number_before_resize
|
||||
}
|
||||
|
||||
pub(crate) fn backing_vec_size(&self) -> usize {
|
||||
self.nodes.len()
|
||||
}
|
||||
|
||||
pub(crate) fn len(&self) -> usize {
|
||||
self.number_of_items
|
||||
}
|
||||
|
||||
pub(crate) fn insert_new(&mut self, key: K, value: V, hash: HashType) -> usize {
|
||||
debug_assert!(
|
||||
self.capacity() > self.len(),
|
||||
"Do not have space to insert into len {} with {}",
|
||||
self.backing_vec_size(),
|
||||
self.len()
|
||||
);
|
||||
|
||||
let mut new_node = Node::new_with(key, value, hash);
|
||||
let mut inserted_location = usize::MAX;
|
||||
|
||||
loop {
|
||||
let location =
|
||||
(new_node.hash() + new_node.distance()).fast_mod(self.backing_vec_size());
|
||||
|
||||
let current_node = &mut self.nodes[location];
|
||||
|
||||
if current_node.has_value() {
|
||||
if current_node.distance() <= new_node.distance() {
|
||||
mem::swap(&mut new_node, current_node);
|
||||
|
||||
if inserted_location == usize::MAX {
|
||||
inserted_location = location;
|
||||
}
|
||||
}
|
||||
} else {
|
||||
self.nodes[location] = new_node;
|
||||
if inserted_location == usize::MAX {
|
||||
inserted_location = location;
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
new_node.increment_distance();
|
||||
self.max_distance_to_initial_bucket =
|
||||
new_node.distance().max(self.max_distance_to_initial_bucket);
|
||||
}
|
||||
|
||||
self.number_of_items += 1;
|
||||
inserted_location
|
||||
}
|
||||
|
||||
pub(crate) fn retain<F>(&mut self, mut f: F)
|
||||
where
|
||||
F: FnMut(&K, &mut V) -> bool,
|
||||
{
|
||||
let num_nodes = self.nodes.len();
|
||||
let mut i = 0;
|
||||
|
||||
while i < num_nodes {
|
||||
let node = &mut self.nodes[i];
|
||||
|
||||
if let Some((k, v)) = node.key_value_mut() {
|
||||
if !f(k, v) {
|
||||
self.remove_from_location(i);
|
||||
|
||||
// Need to continue before adding 1 to i because remove from location could
|
||||
// put the element which was next into the ith location in the nodes array,
|
||||
// so we need to check if that one needs removing too.
|
||||
continue;
|
||||
}
|
||||
}
|
||||
|
||||
i += 1;
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn remove_from_location(&mut self, location: usize) -> V {
|
||||
let mut current_location = location;
|
||||
self.number_of_items -= 1;
|
||||
|
||||
loop {
|
||||
let next_location =
|
||||
HashType::from(current_location + 1).fast_mod(self.backing_vec_size());
|
||||
|
||||
// if the next node is empty, or the next location has 0 distance to initial bucket then
|
||||
// we can clear the current node
|
||||
if !self.nodes[next_location].has_value() || self.nodes[next_location].distance() == 0 {
|
||||
return self.nodes[current_location].take_key_value().unwrap().1;
|
||||
}
|
||||
|
||||
self.nodes.swap(current_location, next_location);
|
||||
self.nodes[current_location].decrement_distance();
|
||||
current_location = next_location;
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn location<Q>(&self, key: &Q, hash: HashType) -> Option<usize>
|
||||
where
|
||||
K: Borrow<Q>,
|
||||
Q: Eq + ?Sized,
|
||||
{
|
||||
for distance_to_initial_bucket in 0..(self.max_distance_to_initial_bucket + 1) {
|
||||
let location = (hash + distance_to_initial_bucket).fast_mod(self.backing_vec_size());
|
||||
|
||||
let node = &self.nodes[location];
|
||||
|
||||
// if we've seen a node which is further from home than what we'd expect to find, then
|
||||
// our node cannot exist because it would've been inserted here.
|
||||
if node.distance() < distance_to_initial_bucket {
|
||||
return None;
|
||||
}
|
||||
|
||||
let node_key_ref = node.key_ref()?;
|
||||
|
||||
if node_key_ref.borrow() == key {
|
||||
return Some(location);
|
||||
}
|
||||
}
|
||||
|
||||
None
|
||||
}
|
||||
|
||||
pub(crate) fn resized_to(&mut self, new_size: usize) -> Self {
|
||||
let mut new_node_storage = Self::with_size_in(new_size, self.allocator().clone());
|
||||
|
||||
for mut node in self.nodes.drain(..) {
|
||||
if let Some((key, value, hash)) = node.take_key_value() {
|
||||
new_node_storage.insert_new(key, value, hash);
|
||||
}
|
||||
}
|
||||
|
||||
new_node_storage
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn replace_at_location_unchecked(
|
||||
&mut self,
|
||||
location: usize,
|
||||
key: K,
|
||||
value: V,
|
||||
) -> V {
|
||||
self.node_at_unchecked_mut(location)
|
||||
.replace_unchecked(key, value)
|
||||
.1
|
||||
}
|
||||
|
||||
pub(crate) fn iter_mut(&mut self) -> impl Iterator<Item = &mut Node<K, V>> {
|
||||
self.nodes.iter_mut()
|
||||
}
|
||||
|
||||
pub(crate) fn node_at(&self, at: usize) -> &Node<K, V> {
|
||||
&self.nodes[at]
|
||||
}
|
||||
|
||||
pub(crate) fn node_at_mut(&mut self, at: usize) -> &mut Node<K, V> {
|
||||
&mut self.nodes[at]
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn node_at_unchecked(&self, at: usize) -> &Node<K, V> {
|
||||
self.nodes.get_unchecked(at)
|
||||
}
|
||||
|
||||
pub(crate) unsafe fn node_at_unchecked_mut(&mut self, at: usize) -> &mut Node<K, V> {
|
||||
self.nodes.get_unchecked_mut(at)
|
||||
}
|
||||
|
||||
pub(crate) fn distance_histogram(&self) -> (Vec<usize>, usize) {
|
||||
let mut ret = Vec::new();
|
||||
|
||||
for node in self.nodes.iter() {
|
||||
let distance = node.distance();
|
||||
|
||||
if distance >= 0 {
|
||||
let distance = distance as usize;
|
||||
ret.resize(ret.len().max(distance + 1), 0);
|
||||
ret[distance] += 1;
|
||||
}
|
||||
}
|
||||
|
||||
(ret, self.max_distance_to_initial_bucket as usize)
|
||||
}
|
||||
|
||||
pub(crate) fn clear(&mut self) {
|
||||
self.max_distance_to_initial_bucket = 0;
|
||||
self.number_of_items = 0;
|
||||
|
||||
self.nodes.fill_with(Node::new);
|
||||
}
|
||||
}
|
Loading…
Reference in a new issue