ustr/stringcache.rs
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410
use super::bumpalloc::LeakyBumpAlloc;
// `StringCache` stores a `Vec` of pointers to the `StringCacheEntry` structs.
// The actual memory for the `StringCacheEntry` is stored in the LeakyBumpAlloc,
// and each `Alloc` is rotated out when it's full and a new one twice its size
// is allocated. The Allocator memory is never freed so our strings essentialy
// have a 'static lifetime.
//
// The actual memory representation is as follows. Each `StringCacheEntry` is
// aligned to 8 bytes on a 64-bit system. The 64-bit memoized hash of the string
// is stored first, then a usize length, then the u8 characters, followed by a
// null terminator (not included in len), then x<8 bytes of uninitialized memory
// as padding before the next aligned entry.
//
// hash len H e l l o , W o r l d !\0
// |. . . . . . . .|. . . . . . . .|. . . . . . . .|. . . . . . . .|
// 0 8 16 len
// ^ StringCacheEntry ^ u8 chars ^ null ^ Next entry
//
// Proper alignment is guaranteed when allocating each entry as the alignment
// is baked into the allocator. `StringCache` is responsible for monitoring the
// Allocator and creating a new one when it would overflow -- the `Alloc` itself
// will just `abort()` if it runs out of memory. Note that we abort() rather
// than panic because the behaviour of the spinlock in case of a panic while
// holding the lock is undefined.
//
// Thread safety is ensured because we can only access the `StringCache` through
// the spinlock in the `lazy_static` ref. The initial capacity of the cache is
// divided evenly among a number of 'bins' or shards each with their own lock,
// in order to reduce contention.
#[repr(align(128))]
pub(crate) struct StringCache {
pub(crate) alloc: LeakyBumpAlloc,
pub(crate) old_allocs: Vec<LeakyBumpAlloc>,
entries: Vec<*mut StringCacheEntry>,
num_entries: usize,
mask: usize,
total_allocated: usize,
// Padding and aligning to 128 bytes gives up to 20% performance
// improvement this actually aligns to 256 bytes because of the Mutex
// around it.
_pad: [u32; 3],
}
// TODO: make these configurable?
// Initial size of the StringCache table
pub(crate) const INITIAL_CAPACITY: usize = 1 << 20;
// Initial size of the allocator storage (in bytes)
pub(crate) const INITIAL_ALLOC: usize = 4 << 20;
// Number of bins (shards) for map
pub(crate) const BIN_SHIFT: usize = 6;
pub(crate) const NUM_BINS: usize = 1 << BIN_SHIFT;
// Shift for top bits to determine bin a hash falls into
pub(crate) const TOP_SHIFT: usize =
8 * std::mem::size_of::<usize>() - BIN_SHIFT;
impl StringCache {
/// Create a new StringCache with the given starting capacity
pub fn new() -> StringCache {
let capacity = INITIAL_CAPACITY / NUM_BINS;
let alloc = LeakyBumpAlloc::new(
INITIAL_ALLOC / NUM_BINS,
std::mem::align_of::<StringCacheEntry>(),
);
StringCache {
// Current allocator.
alloc,
// Old allocators we'll keep around for iteration purposes.
// 16 would mean we've allocated 128GB of string storage since we
// double each time.
old_allocs: Vec::with_capacity(16),
// Vector of pointers to the `StringCacheEntry` headers.
entries: vec![std::ptr::null_mut(); capacity],
num_entries: 0,
mask: capacity - 1,
total_allocated: capacity,
_pad: [0u32; 3],
}
}
pub(crate) fn get_existing(
&self,
string: &str,
hash: u64,
) -> Option<*const u8> {
let mut pos = self.mask & hash as usize;
let mut dist = 0;
loop {
let entry = unsafe { self.entries.get_unchecked(pos) };
if entry.is_null() {
return None;
}
// This is safe as long as entry points to a valid address and the
// layout described in the `StringCache` doc comment holds.
unsafe {
// entry is a `*StringCacheEntry` so offseting by 1 gives us a
// pointer to the end of the entry, aka the beginning of the
// chars.
// As long as the memory is valid and the layout is correct,
// we're safe to create a string slice from the chars since
// they were copied directly from a valid `str`.
let entry_chars = entry.add(1) as *const u8;
// if entry is non-null then it must point to a valid
// StringCacheEntry
let sce = &**entry;
if sce.hash == hash
&& sce.len == string.len()
&& std::str::from_utf8_unchecked(
std::slice::from_raw_parts(entry_chars, sce.len),
) == string
{
// found matching string in the cache already, return it
return Some(entry_chars);
}
}
// Keep looking.
dist += 1;
debug_assert!(dist <= self.mask);
pos = (pos + dist) & self.mask;
}
}
// Insert the given string with its given hash into the cache.
pub(crate) fn insert(&mut self, string: &str, hash: u64) -> *const u8 {
let mut pos = self.mask & hash as usize;
let mut dist = 0;
loop {
let entry = unsafe { self.entries.get_unchecked(pos) };
if entry.is_null() {
// found empty slot to insert
break;
}
// This is safe as long as entry points to a valid address and the
// layout described in the `StringCache` doc comment holds.
unsafe {
// entry is a `*StringCacheEntry` so offseting by 1 gives us a
// pointer to the end of the entry, aka the beginning of the
// chars.
// As long as the memory is valid and the layout is correct,
// we're safe to create a string slice from the chars since
// they were copied directly from a valid `str`.
let entry_chars = entry.add(1) as *const u8;
// If entry is non-null then it must point to a valid
// `StringCacheEntry`.
let sce = &**entry;
if sce.hash == hash
&& sce.len == string.len()
&& std::str::from_utf8_unchecked(
std::slice::from_raw_parts(entry_chars, sce.len),
) == string
{
// found matching string in the cache already, return it
return entry_chars;
}
}
// keep looking
dist += 1;
debug_assert!(dist <= self.mask);
pos = (pos + dist) & self.mask;
}
//
// Insert the new string.
//
// We know pos is in bounds as it's &ed with the mask above.
let entry_ptr = unsafe { self.entries.get_unchecked_mut(pos) };
// Ddd one to length for null byte.
// There's no way we could overflow here in practice since that would
// require having allocated a `u64::MAX`-length string, by which time
// we'll be using 128-bit pointers and we'll need to rewrite this
// crate anyway.
let byte_len = string.len() + 1;
let alloc_size = std::mem::size_of::<StringCacheEntry>() + byte_len;
// if our new allocation would spill over the allocator, make a new
// allocator and let the old one leak
let capacity = self.alloc.capacity();
let allocated = self.alloc.allocated();
if alloc_size
.checked_add(allocated)
.expect("overflowed alloc_size + allocated")
> capacity
{
let new_capacity = capacity
.checked_mul(2)
.expect("capacity * 2 overflowed")
.max(alloc_size);
let old_alloc = std::mem::replace(
&mut self.alloc,
LeakyBumpAlloc::new(
new_capacity,
std::mem::align_of::<StringCacheEntry>(),
),
);
self.old_allocs.push(old_alloc);
self.total_allocated += new_capacity;
}
// This is safe as long as:
// 1. `alloc_size` is calculated correctly.
// 2. there is enough space in the allocator (checked in the block
// above).
// 3. The `StringCacheEntry` layout descibed above holds and the memory
// returned by allocate() is prooperly aligned.
unsafe {
*entry_ptr =
self.alloc.allocate(alloc_size) as *mut StringCacheEntry;
// Write the header.
// `entry_ptr` is guaranteed to point to a valid `StringCacheEntry`,
// or `alloc.allocate()` would have aborted.
std::ptr::write(
*entry_ptr,
StringCacheEntry {
hash,
len: string.len(),
},
);
// Write the characters after the `StringCacheEntry`.
let char_ptr = entry_ptr.add(1) as *mut u8;
std::ptr::copy_nonoverlapping(
string.as_bytes().as_ptr(),
char_ptr,
string.len(),
);
// Write the trailing null.
let write_ptr = char_ptr.add(string.len());
std::ptr::write(write_ptr, 0u8);
self.num_entries += 1;
// We want to keep an 0.5 load factor for the map, so grow if we've
// exceeded that.
if self.num_entries * 2 > self.mask {
self.grow();
}
char_ptr
}
}
// Double the size of the map storage.
//
// This is safe as long as:
// - The in-memory layout of the `StringCacheEntry` is correct.
//
// If there's not enough memory for the new entry table, it will just abort
pub(crate) unsafe fn grow(&mut self) {
let new_mask = self.mask * 2 + 1;
let mut new_entries: std::vec::Vec<*mut StringCacheEntry> =
vec![std::ptr::null_mut(); new_mask + 1];
// copy the existing map into the new map
let mut to_copy = self.num_entries;
for e in self.entries.iter_mut() {
if e.is_null() {
continue;
}
// Start of the entry is the hash.
let hash = *(*e as *const u64);
let mut pos = (hash as usize) & new_mask;
let mut dist = 0;
loop {
if new_entries[pos].is_null() {
// Here's an empty slot to put the pointer in.
break;
}
dist += 1;
// This should be impossble as we've allocated twice as many
// slots as we have entries.
debug_assert!(dist <= new_mask, "Probing wrapped around");
pos = pos.wrapping_add(dist) & new_mask;
}
new_entries[pos] = *e;
to_copy -= 1;
if to_copy == 0 {
break;
}
}
self.entries = new_entries;
self.mask = new_mask;
}
// This is only called by `clear()` during tests to clear the cache between
// runs. **DO NOT CALL THIS**.
pub(crate) unsafe fn clear(&mut self) {
// just zero all the pointers that have already been set
std::ptr::write_bytes(self.entries.as_mut_ptr(), 0, self.mask + 1);
self.num_entries = 0;
self.total_allocated = 0;
for a in self.old_allocs.iter_mut() {
a.clear();
}
self.old_allocs = Vec::new();
self.alloc.clear();
self.alloc = LeakyBumpAlloc::new(
INITIAL_ALLOC / NUM_BINS,
std::mem::align_of::<StringCacheEntry>(),
);
}
pub(crate) fn total_allocated(&self) -> usize {
self.alloc.allocated()
+ self.old_allocs.iter().map(|a| a.allocated()).sum::<usize>()
}
pub(crate) fn total_capacity(&self) -> usize {
self.alloc.capacity()
+ self.old_allocs.iter().map(|a| a.capacity()).sum::<usize>()
}
pub(crate) fn num_entries(&self) -> usize {
self.num_entries
}
}
impl Default for StringCache {
fn default() -> StringCache {
StringCache::new()
}
}
// We are safe to be `Send` but not `Sync` (we get Sync by wrapping in a mutex).
unsafe impl Send for StringCache {}
#[doc(hidden)]
pub struct StringCacheIterator {
pub(crate) allocs: Vec<(*const u8, *const u8)>,
pub(crate) current_alloc: usize,
pub(crate) current_ptr: *const u8,
}
fn round_up_to(n: usize, align: usize) -> usize {
debug_assert!(align.is_power_of_two());
(n.checked_add(align).expect("round_up_to overflowed") - 1) & !(align - 1)
}
impl Iterator for StringCacheIterator {
type Item = &'static str;
fn next(&mut self) -> Option<Self::Item> {
// check that the cache is not empty before accessing
if self.allocs.is_empty() {
return None;
}
let (_, end) = self.allocs[self.current_alloc];
if self.current_ptr >= end {
// We've reached the end of the current alloc.
if self.current_alloc == self.allocs.len() - 1 {
// We've reached the end.
return None;
} else {
// Advance to the next alloc.
self.current_alloc += 1;
let (current_ptr, _) = self.allocs[self.current_alloc];
self.current_ptr = current_ptr;
}
}
// Cast the current ptr to a `StringCacheEntry` and create the next
// string from it.
unsafe {
let sce = &*(self.current_ptr as *const StringCacheEntry);
// The next entry will be the size of the number of bytes in the
// string, +1 for the null byte, rounded up to the alignment (8).
self.current_ptr = sce.next_entry();
// We know we're safe not to check here since we put valid UTF-8 in.
let s = std::str::from_utf8_unchecked(std::slice::from_raw_parts(
sce.char_ptr(),
sce.len,
));
Some(s)
}
}
}
#[repr(C)]
#[derive(Clone)]
pub(crate) struct StringCacheEntry {
pub(crate) hash: u64,
pub(crate) len: usize,
}
impl StringCacheEntry {
// Get the pointer to the characters.
pub(crate) fn char_ptr(&self) -> *const u8 {
// We know the chars are always directly after this struct in memory
// because that's the way they're laid out on initialization.
unsafe { (self as *const StringCacheEntry).add(1) as *const u8 }
}
// Calcualte the address of the next entry in the cache. This is a utility
// function to hide the pointer arithmetic in iterators.
pub(crate) unsafe fn next_entry(&self) -> *const u8 {
#[allow(clippy::ptr_offset_with_cast)]
self.char_ptr().add(round_up_to(
self.len + 1,
std::mem::align_of::<StringCacheEntry>(),
))
}
}