may_queue 0.1.23

May's internal queue library
Documentation
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
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
use crossbeam_utils::{Backoff, CachePadded};
use smallvec::SmallVec;

use crate::atomic::{AtomicPtr, AtomicUsize};

use std::cell::UnsafeCell;
use std::cmp;
use std::marker::PhantomData;
use std::mem::MaybeUninit;
use std::ptr;
use std::sync::atomic::Ordering;

// size for block_node
pub const BLOCK_SIZE: usize = 1 << BLOCK_SHIFT;
// block mask
const BLOCK_MASK: usize = BLOCK_SIZE - 1;
// block shift
const BLOCK_SHIFT: usize = 6;

/// A slot in a block.
struct Slot<T> {
    /// The value.
    value: UnsafeCell<MaybeUninit<T>>,
    ready: AtomicUsize,
}

impl<T> std::panic::RefUnwindSafe for Slot<T> {}

impl<T> Slot<T> {
    #[allow(clippy::declare_interior_mutable_const)]
    const UNINIT: Self = Self {
        value: UnsafeCell::new(MaybeUninit::uninit()),
        ready: AtomicUsize::new(0),
    };
}

/// a block node contains a bunch of items stored in a array
/// this could make the malloc/free not that frequent, also
/// the array could speed up list operations
#[repr(align(64))]
struct BlockNode<T> {
    data: [Slot<T>; BLOCK_SIZE],
    next: AtomicPtr<BlockNode<T>>,
    start: usize, // start index of the block
}

/// we don't implement the block node Drop trait
/// the queue is responsible to drop all the items
/// and would call its get() method for the dropping
impl<T> BlockNode<T> {
    /// create a new BlockNode with uninitialized data
    #[inline]
    fn new_box(index: usize) -> *mut BlockNode<T> {
        Box::into_raw(Box::new(BlockNode::new(index)))
    }

    /// create a new BlockNode with uninitialized data
    #[inline]
    fn new(index: usize) -> BlockNode<T> {
        BlockNode {
            next: AtomicPtr::new(ptr::null_mut()),
            data: [Slot::UNINIT; BLOCK_SIZE],
            start: index,
        }
    }

    /// write index with data
    #[inline]
    fn set(&self, id: usize, v: T) {
        debug_assert!(id < BLOCK_SIZE);
        unsafe {
            let data = self.data.get_unchecked(id);
            data.value.get().write(MaybeUninit::new(v));

            std::sync::atomic::fence(Ordering::Release);
            // mark the data ready
            data.ready.store(1, Ordering::Release);
        }
    }

    #[inline]
    fn try_get(&self, id: usize) -> Option<T> {
        debug_assert!(id < BLOCK_SIZE);
        let data = unsafe { self.data.get_unchecked(id) };
        if data.ready.load(Ordering::Acquire) != 0 {
            Some(unsafe { data.value.get().read().assume_init() })
        } else {
            None
        }
    }

    #[inline]
    fn get(&self, id: usize) -> T {
        debug_assert!(id < BLOCK_SIZE);
        let data = unsafe { self.data.get_unchecked(id) };
        while data.ready.load(Ordering::Acquire) == 0 {
            std::hint::spin_loop();
        }
        unsafe { data.value.get().read().assume_init() }
    }

    /// peek the indexed value
    /// not safe if pop out a value when hold the data ref
    #[inline]
    unsafe fn peek(&self, id: usize) -> &T {
        let data = unsafe { self.data.get_unchecked(id) };
        while data.ready.load(Ordering::Acquire) == 0 {
            std::hint::spin_loop();
        }
        (*data.value.get()).assume_init_ref()
    }

    #[inline]
    fn wait_next_block(&self) -> *mut BlockNode<T> {
        let mut next: *mut BlockNode<T> = self.next.load(Ordering::Acquire);
        while next.is_null() {
            std::hint::spin_loop();
            next = self.next.load(Ordering::Acquire);
        }
        next
    }

    #[inline]
    fn copy_to_bulk(&self, start: usize, end: usize) -> SmallVec<[T; BLOCK_SIZE]> {
        let len = end - start;
        let start = start & BLOCK_MASK;
        SmallVec::from_iter((start..start + len).map(|i| self.get(i)))
    }
}

/// return the bulk end with in the block
#[inline]
fn bulk_end(start: usize, end: usize) -> usize {
    let block_end = (start + BLOCK_SIZE) & !BLOCK_MASK;
    cmp::min(end, block_end)
}

/// A position in a queue.
#[derive(Debug)]
struct Position<T> {
    /// The index in the queue.
    index: AtomicUsize,

    /// The block in the linked list.
    block: AtomicPtr<BlockNode<T>>,
}

impl<T> Position<T> {
    fn new(block: *mut BlockNode<T>) -> Self {
        Position {
            index: AtomicUsize::new(0),
            block: AtomicPtr::new(block),
        }
    }
}

#[derive(Debug)]
struct BlockPtr<T>(AtomicPtr<BlockNode<T>>);

impl<T> BlockPtr<T> {
    #[inline]
    fn new(block: *mut BlockNode<T>) -> Self {
        BlockPtr(AtomicPtr::new(block))
    }

    #[inline]
    fn unpack(ptr: *mut BlockNode<T>) -> (*mut BlockNode<T>, usize) {
        let ptr = ptr as usize;
        let index = ptr & BLOCK_MASK;
        let ptr = (ptr & !BLOCK_MASK) as *mut BlockNode<T>;
        (ptr, index)
    }

    #[inline]
    fn pack(ptr: *const BlockNode<T>, index: usize) -> *mut BlockNode<T> {
        ((ptr as usize) | index) as *mut BlockNode<T>
    }
}

/// mpsc unbounded queue
#[derive(Debug)]
pub struct Queue<T> {
    // -----------------------------------------
    // use for push
    tail: CachePadded<BlockPtr<T>>,

    // ----------------------------------------
    // use for pop
    head: Position<T>,
    // used to delay the drop of the old block
    old_block: UnsafeCell<Option<Box<BlockNode<T>>>>,

    /// Indicates that dropping a `Queue<T>` may drop values of type `T`.
    _marker: PhantomData<T>,
}

unsafe impl<T: Send> Send for Queue<T> {}
unsafe impl<T: Send> Sync for Queue<T> {}

// the old_block prevent RefUnwindSafe
impl<T> std::panic::RefUnwindSafe for Queue<T> {}

impl<T> Queue<T> {
    /// create a spsc queue
    pub fn new() -> Self {
        let init_block = BlockNode::new_box(0);
        let next_block = BlockNode::new_box(BLOCK_SIZE);
        unsafe { &*init_block }
            .next
            .store(next_block, Ordering::Relaxed);
        Queue {
            head: Position::new(init_block),
            tail: BlockPtr::new(init_block).into(),
            old_block: UnsafeCell::new(None),
            _marker: PhantomData,
        }
    }

    /// push a value to the back of queue
    pub fn push(&self, v: T) {
        let backoff = Backoff::new();
        let mut tail = self.tail.0.load(Ordering::Acquire);

        loop {
            tail = (tail as usize & !(1 << 63)) as *mut BlockNode<T>;
            let (block, id) = BlockPtr::unpack(tail);
            let block = unsafe { &*block };

            let new_tail = if id < BLOCK_MASK {
                BlockPtr::pack(block, id + 1)
            } else {
                (tail as usize | (1 << 63)) as *mut BlockNode<T>
            };

            match self.tail.0.compare_exchange_weak(
                tail,
                new_tail,
                Ordering::AcqRel,
                Ordering::Acquire,
            ) {
                Ok(_) => {
                    // set the data
                    block.set(id, v);
                    // the block may be released here by the consumer,
                    // so we need to use old_block to delay the drop
                    if id == BLOCK_MASK {
                        let new_block = BlockNode::new_box(block.start + BLOCK_SIZE * 2);
                        let next_block = unsafe { &mut *block.wait_next_block() };
                        // install the next-next block
                        next_block.next.store(new_block, Ordering::Release);
                        self.tail.0.store(next_block, Ordering::Release);
                    }
                    return;
                }
                Err(old) => {
                    tail = old;
                    backoff.spin();
                }
            }
        }
    }

    #[inline]
    fn push_index(&self) -> usize {
        let tail = self.tail.0.load(Ordering::Acquire);
        let (tail_block, id) = BlockPtr::unpack(tail);
        let tail_block = (tail_block as usize & !(1 << 63)) as *mut BlockNode<T>;
        unsafe { &*tail_block }.start + id
    }

    /// pop from the queue, if it's empty return None
    pub fn pop(&self) -> Option<T> {
        let head = unsafe { &mut *self.head.block.unsync_load() };
        let pop_index = unsafe { self.head.index.unsync_load() };
        let id = pop_index & BLOCK_MASK;

        // get the data
        let data = match head.try_get(id) {
            Some(v) => v,
            None => {
                if pop_index >= self.push_index() {
                    return None;
                } else {
                    head.get(id)
                }
            }
        };

        self.head.index.store(pop_index + 1, Ordering::Relaxed);

        if id == BLOCK_MASK {
            // we need to delay the drop of the block to let the push's `wait_next_block` return
            let old_block = unsafe { &mut *(self.old_block.get()) };
            old_block.replace(unsafe { Box::from_raw(head) });

            let next_block = head.wait_next_block();
            self.head.block.store(next_block, Ordering::Relaxed);
        }

        Some(data)
    }

    /// fast pop from the queue, if it's empty return None, or else return `SmallVec<[T; BLOCK_SIZE]>`
    /// don't check the push index, but only the ready flag
    #[inline]
    fn fast_bulk_pop(&self, index: usize, head: &mut BlockNode<T>) -> SmallVec<[T; BLOCK_SIZE]> {
        // only pop within a block
        let block_end = (index + BLOCK_SIZE) & !BLOCK_MASK;
        let len = block_end - index;

        let mut value = SmallVec::new();
        let start = index & BLOCK_MASK;
        for i in start..start + len {
            match head.try_get(i) {
                Some(v) => value.push(v),
                None => break,
            }
        }

        if value.is_empty() {
            return value;
        }

        let new_index = index + value.len();
        self.head.index.store(new_index, Ordering::Relaxed);

        // free the old block node
        if new_index & BLOCK_MASK == 0 {
            let old_block = unsafe { &mut *(self.old_block.get()) };
            old_block.replace(unsafe { Box::from_raw(head) });

            let next_block = head.wait_next_block();
            self.head.block.store(next_block, Ordering::Relaxed);
        }

        value
    }

    /// pop from the queue, if it's empty return None, or else return `SmallVec<[T; BLOCK_SIZE]>`
    pub fn bulk_pop(&self) -> SmallVec<[T; BLOCK_SIZE]> {
        let pop_index = unsafe { self.head.index.unsync_load() };
        let head = unsafe { &mut *self.head.block.unsync_load() };
        let v = self.fast_bulk_pop(pop_index, head);
        if !v.is_empty() {
            return v;
        }

        let push_index = self.push_index();
        if pop_index >= push_index {
            return SmallVec::new();
        }

        // only pop within a block
        let end = bulk_end(pop_index, push_index);
        let value = head.copy_to_bulk(pop_index, end);

        let new_index = end;
        self.head.index.store(new_index, Ordering::Relaxed);

        // free the old block node
        if new_index & BLOCK_MASK == 0 {
            let old_block = unsafe { &mut *(self.old_block.get()) };
            old_block.replace(unsafe { Box::from_raw(head) });

            let next_block = head.wait_next_block();
            self.head.block.store(next_block, Ordering::Relaxed);
        }

        value
    }

    /// peek the head
    ///
    /// # Safety
    ///
    /// not safe if you pop out the head value when hold the data ref
    pub unsafe fn peek(&self) -> Option<&T> {
        let pop_index = unsafe { self.head.index.unsync_load() };
        let push_index = self.push_index();
        if pop_index >= push_index {
            return None;
        }

        let head = &mut *self.head.block.unsync_load();
        Some(head.peek(pop_index & BLOCK_MASK))
    }

    /// get the size of queue
    #[inline]
    pub fn len(&self) -> usize {
        let pop_index = self.head.index.load(Ordering::Acquire);
        let push_index = self.push_index();
        if pop_index >= push_index {
            return 0;
        }
        push_index - pop_index
    }

    /// if the queue is empty
    #[inline]
    pub fn is_empty(&self) -> bool {
        self.len() == 0
    }
}

impl<T> Default for Queue<T> {
    fn default() -> Self {
        Queue::new()
    }
}

impl<T> Drop for Queue<T> {
    fn drop(&mut self) {
        //  pop all the element to make sure the queue is empty
        while self.pop().is_some() {}
        let head = self.head.block.load(Ordering::Acquire);
        let tail = self.tail.0.load(Ordering::Acquire);
        let (block, _id) = BlockPtr::unpack(tail);
        assert_eq!(block, head);

        let next_block = unsafe { &*block }.next.load(Ordering::Acquire);
        assert!(!next_block.is_null());
        let _unused_block = unsafe { Box::from_raw(block) };
        let _unused_block = unsafe { Box::from_raw(next_block) };
    }
}

#[cfg(all(nightly, test))]
mod test {
    extern crate test;
    use self::test::Bencher;
    use super::*;

    use std::sync::Arc;
    use std::thread;

    use crate::test_queue::ScBlockPop;

    impl<T: Send> ScBlockPop<T> for super::Queue<T> {
        fn block_pop(&self) -> T {
            let backoff = Backoff::new();
            loop {
                match self.pop() {
                    Some(v) => return v,
                    None => backoff.snooze(),
                }
            }
        }
    }

    #[test]
    fn queue_sanity() {
        let q = Queue::<usize>::new();
        assert_eq!(q.len(), 0);
        for i in 0..100 {
            q.push(i);
        }
        assert_eq!(q.len(), 100);
        // println!("{q:?}");

        for i in 0..100 {
            assert_eq!(q.pop(), Some(i));
        }
        assert_eq!(q.pop(), None);
        assert_eq!(q.len(), 0);
    }

    #[bench]
    fn single_thread_test(b: &mut Bencher) {
        let q = Queue::new();
        let mut i = 0;
        b.iter(|| {
            q.push(i);
            assert_eq!(q.pop(), Some(i));
            i += 1;
        });
    }

    #[bench]
    fn multi_1p1c_test(b: &mut Bencher) {
        b.iter(|| {
            let q = Arc::new(Queue::new());
            let total_work: usize = 1_000_000;
            // create worker threads that generate mono increasing index
            let _q = q.clone();
            // in other thread the value should be still 100
            thread::spawn(move || {
                for i in 0..total_work {
                    _q.push(i);
                }
            });

            for i in 0..total_work {
                let v = q.block_pop();
                assert_eq!(i, v);
            }
        });
    }

    #[bench]
    fn multi_2p1c_test(b: &mut Bencher) {
        b.iter(|| {
            let q = Arc::new(Queue::new());
            let total_work: usize = 1_000_000;
            // create worker threads that generate mono increasing index
            // in other thread the value should be still 100
            let mut total = 0;

            thread::scope(|s| {
                let threads = 20;
                for i in 0..threads {
                    let q = q.clone();
                    s.spawn(move || {
                        let len = total_work / threads;
                        let start = i * len;
                        for v in start..start + len {
                            q.push(v);
                        }
                    });
                }
                s.spawn(|| {
                    for _ in 0..total_work {
                        total += q.block_pop();
                    }
                });
            });
            assert!(q.is_empty());
            assert_eq!(total, (0..total_work).sum::<usize>());
        });
    }

    #[bench]
    fn bulk_pop_1p1c_bench(b: &mut Bencher) {
        b.iter(|| {
            let q = Arc::new(Queue::new());
            let total_work: usize = 1_000_000;
            // create worker threads that generate mono increasing index
            let _q = q.clone();
            // in other thread the value should be still 100
            thread::spawn(move || {
                for i in 0..total_work {
                    _q.push(i);
                }
            });

            let mut size = 0;
            while size < total_work {
                let v = q.bulk_pop();
                for (start, i) in v.iter().enumerate() {
                    assert_eq!(*i, start + size);
                }
                size += v.len();
            }
        });
    }

    #[bench]
    fn bulk_2p1c_test(b: &mut Bencher) {
        b.iter(|| {
            let q = Arc::new(Queue::new());
            let total_work: usize = 1_000_000;
            // create worker threads that generate mono increasing index
            // in other thread the value should be still 100
            let mut total = 0;

            thread::scope(|s| {
                let threads = 20;
                for i in 0..threads {
                    let q = q.clone();
                    s.spawn(move || {
                        let len = total_work / threads;
                        let start = i * len;
                        for v in start..start + len {
                            q.push(v);
                        }
                    });
                }
                s.spawn(|| {
                    let mut size = 0;
                    while size < total_work {
                        let v = q.bulk_pop();
                        size += v.len();
                        for data in v {
                            total += data;
                        }
                    }
                });
            });
            assert!(q.is_empty());
            assert_eq!(total, (0..total_work).sum::<usize>());
        });
    }
}