multitude 0.5.1

Fast and flexible arena allocator.
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

# Multitude Performance Report

Generated by `scripts/perf_report.rs`:
- `cargo bench --bench criterion_alloc` and `criterion_drop` — criterion wall-clock timings.
- `cargo bench --bench criterion_arena_vs_allocator` — criterion wall-clock timing of allocating a mixed working set and then releasing it, comparing the arena (bulk reset) against the system allocator (mimalloc, per-object free).
- `cargo bench --bench gungraun_alloc` and `gungraun_drop` — Callgrind instruction-precise counts.

**Workload:** N = 1000 operations per measurement; slice element count = 8.  
Criterion median is reported (default 30 samples, 1 s warm-up, 2 s measurement; override with `--samples` / `--measurement-time` / `--warm-up-time`).  
Memory accesses = L1 Hits + LL Hits + RAM Hits (Callgrind D-cache references).  
Bench names are aligned between criterion and gungraun via the `GROUPS` table in `scripts/perf_report.rs`.

## `arena_creation`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `multitude_new` | 36 ns | 299 | 8 | 427 |
| `bumpalo_new` | 1 ns | 16 | 1 | 26 |

## `alloc_u64`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `alloc` | 5.29 µs | 15,020 | 11 | 23,034 |
| `alloc_with` | 4.57 µs | 15,020 | 12 | 23,034 |
| `alloc_box` | 4.93 µs | 23,039 | 9 | 37,070 |
| `alloc_box_with` | 5.01 µs | 24,039 | 9 | 38,070 |
| `alloc_uninit_box` | 2.50 µs | 20,039 | 9 | 31,070 |
| `alloc_zeroed_box` | 4.67 µs | 21,039 | 9 | 33,070 |
| `alloc_arc` | 9.34 µs | 25,039 | 9 | 40,070 |
| `alloc_arc_with` | 9.40 µs | 26,039 | 8 | 41,070 |
| `alloc_uninit_arc` | 8.75 µs | 22,039 | 8 | 34,070 |
| `alloc_zeroed_arc` | 9.07 µs | 23,039 | 9 | 36,070 |
| `alloc_rc` | 9.34 µs | 25,039 | 9 | 40,070 |
| `alloc_rc_with` | 9.34 µs | 26,039 | 9 | 41,070 |
| `alloc_uninit_rc` | 8.79 µs | 22,039 | 9 | 34,070 |
| `alloc_zeroed_rc` | 9.07 µs | 23,039 | 9 | 36,070 |
| `bumpalo_alloc` | 3.94 µs | 19,022 | 4 | 27,037 |
| `bumpalo_alloc_with` | 4.13 µs | 19,020 | 4 | 27,034 |

## `alloc_str`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `alloc_str` | 6.97 µs | 52,049 | 11 | 78,090 |
| `alloc_str_box` | 12.19 µs | 59,049 | 12 | 85,090 |
| `alloc_str_arc` | 14.26 µs | 52,050 | 4 | 76,090 |
| `alloc_str_rc` | 14.10 µs | 50,050 | 3 | 74,090 |
| `bumpalo_alloc_str` | 7.04 µs | 50,048 | 13 | 75,088 |

## `alloc_slice`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `alloc_slice_copy` | 42.62 µs | 42,045 | 3 | 59,082 |
| `alloc_slice_clone` | 41.75 µs | 46,046 | 10 | 60,083 |
| `alloc_slice_fill_with` | 45.34 µs | 39,024 | 10 | 70,039 |
| `alloc_slice_fill_iter` | 45.17 µs | 39,025 | 12 | 70,040 |
| `alloc_slice_copy_box` | 52.19 µs | 55,170 | 19 | 83,267 |
| `alloc_slice_clone_box` | 50.94 µs | 68,170 | 26 | 92,266 |
| `alloc_slice_fill_with_box` | 54.23 µs | 48,153 | 23 | 86,235 |
| `alloc_slice_fill_iter_box` | 54.98 µs | 50,153 | 19 | 90,235 |
| `alloc_uninit_slice_box` | 49.76 µs | 23,153 | 22 | 36,235 |
| `alloc_zeroed_slice_box` | 50.12 µs | 27,153 | 23 | 43,235 |
| `alloc_slice_copy_arc` | 56.79 µs | 50,203 | 15 | 74,312 |
| `alloc_slice_clone_arc` | 55.92 µs | 56,253 | 28 | 77,380 |
| `alloc_slice_fill_with_arc` | 58.66 µs | 46,015 | 15 | 83,019 |
| `alloc_slice_fill_iter_arc` | 59.14 µs | 46,015 | 15 | 83,019 |
| `alloc_uninit_slice_arc` | 53.71 µs | 22,174 | 25 | 35,265 |
| `alloc_zeroed_slice_arc` | 54.50 µs | 25,168 | 25 | 40,256 |
| `alloc_slice_copy_rc` | 56.62 µs | 50,203 | 18 | 74,312 |
| `alloc_slice_clone_rc` | 55.10 µs | 56,253 | 26 | 77,380 |
| `alloc_slice_fill_with_rc` | 57.75 µs | 46,015 | 16 | 83,019 |
| `alloc_slice_fill_iter_rc` | 58.77 µs | 46,015 | 15 | 83,019 |
| `alloc_uninit_slice_rc` | 53.73 µs | 22,174 | 24 | 35,265 |
| `alloc_zeroed_slice_rc` | 54.67 µs | 25,168 | 25 | 40,256 |
| `bumpalo_alloc_slice_copy` | 38.68 µs | 38,042 | 5 | 55,076 |
| `bumpalo_alloc_slice_clone` | 37.53 µs | 60,046 | 10 | 74,083 |
| `bumpalo_alloc_slice_fill_with` | 40.50 µs | 40,020 | 5 | 70,033 |
| `bumpalo_alloc_slice_fill_iter` | 42.89 µs | 40,020 | 4 | 70,033 |

## `string_builder`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `alloc_string` | 6.44 µs | 36,855 | 34 | 51,209 |
| `alloc_string_with_capacity` | 6.46 µs | 37,222 | 19 | 52,343 |
| `bumpalo_string_new_in` | 10.07 µs | 35,843 | 79 | 50,867 |
| `bumpalo_string_with_capacity_in` | 9.49 µs | 34,708 | 27 | 49,159 |

## `vec_builder`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `alloc_vec` | 1.03 µs | 11,814 | 28 | 17,109 |
| `alloc_vec_with_capacity` | 1.11 µs | 12,149 | 11 | 18,235 |
| `bumpalo_vec_new_in` | 3.91 µs | 12,281 | 61 | 18,888 |
| `bumpalo_vec_with_capacity_in` | 3.71 µs | 11,069 | 7 | 17,116 |

## `arena_vs_allocator`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `arena` | 10.73 µs | — | — | — |
| `system` | 34.82 µs | — | — | — |

## `drop`

| Variant | Time (criterion) | Instructions | Branch misses | Mem accesses |
|---|---:|---:|---:|---:|
| `box_u64` | 4.38 µs | 9,560 | 29 | 12,826 |
| `rc_u64` | 5.59 µs | 12,554 | 28 | 16,817 |
| `arc_u64` | 8.25 µs | 11,554 | 28 | 16,817 |
| `box_droppy` | 18.11 µs | 185,290 | 58 | 271,252 |
| `rc_droppy` | 18.97 µs | 188,289 | 58 | 275,251 |
| `arc_droppy` | 19.61 µs | 187,289 | 64 | 275,251 |
| `str_box` | 4.63 µs | 9,560 | 29 | 12,826 |
| `str_rc` | 5.55 µs | 12,554 | 25 | 16,817 |
| `str_arc` | 8.11 µs | 11,554 | 28 | 16,817 |
| `slice_box_u64` | 10.15 µs | 9,736 | 45 | 13,067 |
| `slice_rc_u64` | 12.22 µs | 12,730 | 42 | 17,058 |
| `slice_arc_u64` | 14.52 µs | 11,730 | 43 | 17,058 |
| `slice_box_droppy` | 115.72 µs | 1,511,455 | 1,105 | 2,203,111 |
| `slice_rc_droppy` | 115.63 µs | 1,514,454 | 1,112 | 2,207,109 |
| `slice_arc_droppy` | 115.27 µs | 1,514,454 | 1,100 | 2,208,109 |
| `alloc` | 1.44 µs | 371 | 16 | 541 |

## Multitude vs Bumpalo Head-to-Head

Direct comparisons of multitude versus bumpalo on identical workloads (the multitude variant chosen is the closest semantic equivalent to bumpalo's plain bump-allocation).

| Workload | Multitude time | Bumpalo time | Δ time | Multitude instr | Bumpalo instr | Δ instr |
|---|---:|---:|---:|---:|---:|---:|
| `alloc` vs `bumpalo_alloc` | 5.29 µs | 3.94 µs | +34.5% | 15,020 | 19,022 | -21.0% |
| `alloc_str` vs `bumpalo_alloc_str` | 6.97 µs | 7.04 µs | -1.0% | 52,049 | 50,048 | +4.0% |
| `alloc_slice_copy` vs `bumpalo_alloc_slice_copy` | 42.62 µs | 38.68 µs | +10.2% | 42,045 | 38,042 | +10.5% |
| `alloc_slice_clone` vs `bumpalo_alloc_slice_clone` | 41.75 µs | 37.53 µs | +11.3% | 46,046 | 60,046 | -23.3% |
| `alloc_slice_fill_with` vs `bumpalo_alloc_slice_fill_with` | 45.34 µs | 40.50 µs | +11.9% | 39,024 | 40,020 | -2.5% |
| `alloc_slice_fill_iter` vs `bumpalo_alloc_slice_fill_iter` | 45.17 µs | 42.89 µs | +5.3% | 39,025 | 40,020 | -2.5% |
| `alloc_string` vs `bumpalo_string_new_in` | 6.44 µs | 10.07 µs | -36.0% | 36,855 | 35,843 | +2.8% |
| `alloc_string_with_capacity` vs `bumpalo_string_with_capacity_in` | 6.46 µs | 9.49 µs | -31.9% | 37,222 | 34,708 | +7.2% |
| `alloc_vec` vs `bumpalo_vec_new_in` | 1.03 µs | 3.91 µs | -73.6% | 11,814 | 12,281 | -3.8% |
| `alloc_vec_with_capacity` vs `bumpalo_vec_with_capacity_in` | 1.11 µs | 3.71 µs | -70.2% | 12,149 | 11,069 | +9.8% |