tiny_lfu 0.13.0

tiny_lfu provides a cache admission control policy.
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

use std::sync::atomic::{AtomicU64, Ordering};

////////////////////////////////////////////// Vector //////////////////////////////////////////////

pub struct Vector {
    elem_bits: usize,
    num_elems: usize,
    backing: Vec<AtomicU64>,
}

impl Vector {
    pub fn new(elem_bits: usize, num_elems: usize) -> Self {
        assert!(elem_bits <= 63);
        let total_bits = num_elems
            .checked_mul(elem_bits)
            .expect("tried to create too large a vector");
        let num_u64s = total_bits / 64 + 2;
        let mut backing = Vec::with_capacity(num_u64s);
        backing.resize_with(num_u64s, AtomicU64::default);
        Self {
            elem_bits,
            num_elems,
            backing,
        }
    }

    pub fn len(&self) -> usize {
        self.num_elems
    }

    pub fn load(&self, idx: usize) -> u64 {
        let (index, lower_shift, lower_mask, upper_mask, upper_shift) = self.shifts_and_masks(idx);
        let lower = self.backing[index].load(Ordering::Relaxed);
        let upper = self.backing[index + 1].load(Ordering::Relaxed);
        ((lower >> lower_shift) & lower_mask) | ((upper & upper_mask) << upper_shift)
    }

    pub fn increment(&self, idx: usize) {
        let (index, lower_shift, _, _, _) = self.shifts_and_masks(idx);
        loop {
            let lower = self.backing[index].load(Ordering::Relaxed);
            let upper = self.backing[index + 1].load(Ordering::Relaxed);
            let new_lower = lower.wrapping_add(1u64 << lower_shift);
            let new_upper = if new_lower < lower {
                upper.wrapping_add(1)
            } else {
                upper
            };
            if !self.compare_and_swap(index, lower, new_lower) {
                continue;
            }
            if upper == new_upper {
                break;
            }
            if !self.compare_and_swap(index + 1, upper, new_upper) {
                // TODO(rescrv) click!("tiny_lfu.increment_collision");
            }
            break;
        }
    }

    pub fn divide_two(&self, idx: usize) {
        let (index, lower_shift, lower_mask, upper_mask, _) = self.shifts_and_masks(idx);
        let lower_shifted_mask = lower_mask << lower_shift;
        loop {
            let lower = self.backing[index].load(Ordering::Relaxed);
            let upper = self.backing[index + 1].load(Ordering::Relaxed);
            let new_lower = (((lower & lower_shifted_mask) >> 1) & lower_shifted_mask)
                | (lower & !lower_shifted_mask)
                | ((upper & 0x1 & upper_mask) << 63);
            let new_upper = ((upper & upper_mask) >> 1) | (upper & !upper_mask);
            if !self.compare_and_swap(index + 1, upper, new_upper) {
                continue;
            }
            if !self.compare_and_swap(index, lower, new_lower) {
                // TODO(rescrv) click!("tiny_lfu.divide_two_collision");
            }
            break;
        }
    }

    #[inline(always)]
    fn shifts_and_masks(&self, index: usize) -> (usize, u64, u64, u64, u64) {
        let bit_index = index * self.elem_bits;
        let u64_index = bit_index / 64;
        let lower_shift = (bit_index - u64_index * 64) as u64;
        let lower_bits = if 64 - lower_shift < self.elem_bits as u64 {
            64 - lower_shift
        } else {
            self.elem_bits as u64
        };
        let lower_mask = (1u64 << lower_bits) - 1;
        let upper_mask = (1u64 << (self.elem_bits as u64 - lower_bits)) - 1;
        let upper_shift = if lower_bits == self.elem_bits as u64 {
            0
        } else {
            lower_bits
        };
        (u64_index, lower_shift, lower_mask, upper_mask, upper_shift)
    }

    #[inline(always)]
    fn compare_and_swap(&self, index: usize, current: u64, new: u64) -> bool {
        let ordering = Ordering::Relaxed;
        match self.backing[index].compare_exchange(current, new, ordering, ordering) {
            Ok(x) => x == current,
            Err(x) => x == current,
        }
    }
}

/////////////////////////////////////////////// tests //////////////////////////////////////////////

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn new() {
        let v = Vector::new(24, 47);
        assert_eq!(24, v.elem_bits);
        assert_eq!(47, v.num_elems);
        assert!(v.backing.len() >= 18);
    }

    #[test]
    fn load() {
        let v = Vector::new(24, 5);
        v.backing[0].store(0xdeadbeefcafe1eafu64, Ordering::Relaxed);
        v.backing[1].store(0xc0ffeef00dc0ffeeu64, Ordering::Relaxed);
        assert_eq!(0xfe1eafu64, v.load(0));
        assert_eq!(0xbeefcau64, v.load(1));
        assert_eq!(0xeedeadu64, v.load(2));
        assert_eq!(0x0dc0ffu64, v.load(3));
        assert_eq!(0xffeef0u64, v.load(4));
    }

    #[test]
    fn increment() {
        let v = Vector::new(24, 5);
        v.backing[0].store(0xdeadbeefcafe1eafu64, Ordering::Relaxed);
        v.backing[1].store(0xc0ffeef00dc0ffeeu64, Ordering::Relaxed);
        v.increment(0);
        assert_eq!(0xdeadbeefcafe1eb0u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc0ffeeu64, v.backing[1].load(Ordering::Relaxed));
        v.increment(1);
        assert_eq!(0xdeadbeefcbfe1eb0u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc0ffeeu64, v.backing[1].load(Ordering::Relaxed));
        v.increment(2);
        assert_eq!(0xdeaebeefcbfe1eb0u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc0ffeeu64, v.backing[1].load(Ordering::Relaxed));
        v.increment(3);
        assert_eq!(0xdeaebeefcbfe1eb0u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc100eeu64, v.backing[1].load(Ordering::Relaxed));
        v.increment(4);
        assert_eq!(0xdeaebeefcbfe1eb0u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef10dc100eeu64, v.backing[1].load(Ordering::Relaxed));
    }

    #[test]
    fn divide_two() {
        let v = Vector::new(24, 5);
        v.backing[0].store(0xdeadbeefcafe1eafu64, Ordering::Relaxed);
        v.backing[1].store(0xc0ffeef00dc0ffeeu64, Ordering::Relaxed);
        v.divide_two(0);
        assert_eq!(0xdeadbeefca7f0f57u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc0ffeeu64, v.backing[1].load(Ordering::Relaxed));
        v.divide_two(1);
        assert_eq!(0xdead5f77e57f0f57u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc0ffeeu64, v.backing[1].load(Ordering::Relaxed));
        v.divide_two(2);
        assert_eq!(0x6f565f77e57f0f57u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef00dc0ff77u64, v.backing[1].load(Ordering::Relaxed));
        v.divide_two(3);
        assert_eq!(0x6f565f77e57f0f57u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc0ffeef006e07f77u64, v.backing[1].load(Ordering::Relaxed));
        v.divide_two(4);
        assert_eq!(0x6f565f77e57f0f57u64, v.backing[0].load(Ordering::Relaxed));
        assert_eq!(0xc07ff77806e07f77u64, v.backing[1].load(Ordering::Relaxed));
    }

    #[test]
    fn shifts_and_masks() {
        let v = Vector::new(24, 5);
        v.backing[0].store(0xdeadbeefcafe1eafu64, Ordering::Relaxed);
        v.backing[1].store(0xc0ffeef00dc0ffeeu64, Ordering::Relaxed);
        let (index, lower_shift, lower_mask, upper_mask, upper_shift) = v.shifts_and_masks(0);
        assert_eq!(0, index);
        assert_eq!(0, lower_shift);
        assert_eq!(0xffffffu64, lower_mask);
        assert_eq!(0, upper_mask);
        assert_eq!(0, upper_shift);
        let (index, lower_shift, lower_mask, upper_mask, upper_shift) = v.shifts_and_masks(1);
        assert_eq!(0, index);
        assert_eq!(24, lower_shift);
        assert_eq!(0xffffffu64, lower_mask);
        assert_eq!(0, upper_mask);
        assert_eq!(0, upper_shift);
        let (index, lower_shift, lower_mask, upper_mask, upper_shift) = v.shifts_and_masks(2);
        assert_eq!(0, index);
        assert_eq!(48, lower_shift);
        assert_eq!(0xffffu64, lower_mask);
        assert_eq!(0xffu64, upper_mask);
        assert_eq!(16, upper_shift);
        let (index, lower_shift, lower_mask, upper_mask, upper_shift) = v.shifts_and_masks(3);
        assert_eq!(1, index);
        assert_eq!(8, lower_shift);
        assert_eq!(0xffffffu64, lower_mask);
        assert_eq!(0, upper_mask);
        assert_eq!(0, upper_shift);
        let (index, lower_shift, lower_mask, upper_mask, upper_shift) = v.shifts_and_masks(4);
        assert_eq!(1, index);
        assert_eq!(32, lower_shift);
        assert_eq!(0xffffffu64, lower_mask);
        assert_eq!(0, upper_mask);
        assert_eq!(0, upper_shift);
    }
}