stumpalo
A fast, zero-dependency, memory-efficient bump allocator with chunk reuse and scoped stack support.
Features
Fast allocations
The fast path contains as little as six instructions, and one conditional branch
Chunk reuse
.clear() resets the arena for reuse without freeing memory, allocated chunks are recycled on subsequent allocations.
Stack allocation support
with_scope() creates a temporary sub-arena whose
allocations only live as long as the closure.
This lets you use an arena as a heterogeneous stack allocator.
Usage
Basic allocation
use Arena;
let arena = new;
let x = arena.alloc;
let y = arena.alloc_with;
Slices and arrays
use Arena;
let arena = new;
// Copy a slice in
let nums = arena.alloc_slice_copy;
assert_eq!;
// Allocate a fixed-size array with Default
let arr: &mut = arena.alloc_slice_fill_default;
assert_eq!;
// Allocate a slice of length 12, filled by a function which takes the index
let arr: &mut = arena.alloc_slice_fill_with;
assert_eq!;
// Allocate a string
let s: &mut str = arena.alloc_str;
assert_eq!;
Scoped arenas
use Arena;
let mut arena = new;
// This is only necessary if you need to keep references alive from
// the outer scope, after the inner scope returns.
let arena = arena.as_arena_ref_mut;
let a = arena.alloc;
arena.with_scope;
// Scope returned, inner allocations are gone
// 'a' is still accessible
assert_eq!;
Clear and reuse
use Arena;
let mut arena = new;
let x = arena.alloc;
arena.clear; // x is now invalid, trying to use it will yield a compile-time error.
let y = arena.alloc; // reuses the same chunk
assert_eq!;
Specialized allocation methods
If your slice has a length known at compile-time, use alloc_sized_slice_copy instead of alloc_slice_copy.
If you have a string literal, use alloc_str_lit instead of alloc_str.
These methods can sometimes use the extra compile-time information to reduce the size of the fast-path.
Benchmarks
stumpalo includes a comparison benchmark against bumpalo and blink-alloc.
Each library measurement runs in its own forked process to eliminate heap-fragmentation interference.
Running
Each cell shows the slowdown relative to the fastest library for that operation (1.00x = fastest).
✅ <1.02x 🟢 <1.10x 🟡 <1.25x 🟠 <1.50x 🔴 <2.00x 🟥 2.00x+
Results
Benchmark machine: AMD Ryzen 3900x, Arch Linux, kernel 7.0.3-arch1-2
Num allocs: 100000
Warmup: 10
Samples: 60
Discarded samples: 6
operation stumpalo blink-alloc bumpalo
alloc_u8 ✅ 1.00x 128.8 µs 🔴 2.14x 275.0 µs 🟠 1.54x 197.8 µs
alloc_u16 ✅ 1.00x 111.0 µs 🔴 2.46x 272.6 µs 🟥 2.54x 281.5 µs
alloc_u32 ✅ 1.00x 74.8 µs 🟥 3.36x 251.0 µs 🟥 3.34x 250.0 µs
alloc_u64 ✅ 1.00x 75.1 µs 🟥 3.35x 251.5 µs 🟥 3.34x 250.8 µs
alloc_u128 ✅ 1.00x 215.5 µs 🟡 1.19x 256.6 µs 🟡 1.18x 254.4 µs
alloc_multiple_u8 ✅ 1.00x 1032.5 µs 🔴 1.82x 1874.1 µs 🔴 1.85x 1907.0 µs
alloc_multiple_u16 ✅ 1.00x 886.7 µs 🔴 2.30x 2043.5 µs 🔴 2.34x 2071.3 µs
alloc_multiple_u32 ✅ 1.00x 654.4 µs 🟥 3.12x 2043.7 µs 🟥 3.14x 2056.1 µs
alloc_multiple_u64 ✅ 1.00x 633.0 µs 🟥 3.23x 2044.3 µs 🟥 3.25x 2055.3 µs
alloc_multiple_u128 ✅ 1.00x 758.9 µs 🟥 2.70x 2045.9 µs 🟥 2.61x 1978.3 µs
alloc_array_u8_8 ✅ 1.00x 99.4 µs 🔴 1.99x 197.7 µs 🔴 2.11x 210.0 µs
alloc_array_u8_32 ✅ 1.00x 179.1 µs 🟢 1.15x 205.4 µs 🟡 1.20x 215.1 µs
alloc_array_u8_64 ✅ 1.00x 135.2 µs 🟠 1.55x 209.0 µs 🟠 1.59x 214.9 µs
alloc_array_u8_128 ✅ 1.00x 187.7 µs 🟡 1.30x 244.7 µs 🟠 1.50x 280.8 µs
alloc_slice_u8_8 🟢 1.11x 248.0 µs 🟡 1.27x 283.2 µs ✅ 1.00x 222.5 µs
alloc_slice_u8_32 🟢 1.06x 274.9 µs ✅ 1.00x 259.6 µs 🟢 1.08x 279.4 µs
alloc_slice_u8_64 ✅ 1.05x 278.7 µs ✅ 1.00x 266.7 µs 🟢 1.09x 289.6 µs
alloc_slice_u8_128 ✅ 1.00x 333.3 µs 🟢 1.06x 353.0 µs ✅ 1.04x 347.4 µs
alloc_slice_u16_8 ✅ 1.00x 229.4 µs 🟡 1.33x 304.3 µs 🟡 1.16x 265.5 µs
alloc_slice_u16_32 ✅ 1.00x 302.4 µs 🟢 1.14x 345.3 µs 🟢 1.11x 334.2 µs
alloc_slice_u16_64 ✅ 1.00x 358.4 µs 🟢 1.14x 408.4 µs 🟢 1.10x 394.5 µs
alloc_slice_u16_128 ✅ 1.04x 3401.3 µs ✅ 1.00x 3270.6 µs ✅ 1.02x 3336.0 µs
alloc_slice_u32_8 ✅ 1.00x 300.0 µs 🟢 1.14x 341.6 µs 🟢 1.09x 327.6 µs
alloc_slice_u32_32 ✅ 1.00x 358.1 µs 🟢 1.14x 409.1 µs 🟢 1.10x 393.9 µs
alloc_slice_u32_64 ✅ 1.05x 3459.3 µs ✅ 1.00x 3309.5 µs 🟢 1.06x 3509.9 µs
alloc_slice_u32_128 🟢 1.09x 7676.0 µs ✅ 1.00x 7027.1 µs 🟢 1.13x 7912.7 µs
alloc_slice_u64_8 ✅ 1.00x 274.8 µs 🟡 1.25x 344.8 µs 🟢 1.11x 305.0 µs
alloc_slice_u64_32 ✅ 1.04x 3422.2 µs ✅ 1.00x 3297.0 µs ✅ 1.02x 3351.6 µs
alloc_slice_u64_64 🟢 1.08x 7600.2 µs ✅ 1.00x 7036.4 µs 🟢 1.10x 7751.8 µs
alloc_slice_u64_128 🟢 1.07x 14956.8 µs ✅ 1.00x 14003.1 µs 🟢 1.08x 15073.6 µs
alloc_slice_u128_8 ✅ 1.00x 357.5 µs 🟢 1.12x 401.9 µs 🟢 1.11x 395.1 µs
alloc_slice_u128_32 🟢 1.08x 7595.3 µs ✅ 1.00x 7032.0 µs 🟢 1.12x 7898.0 µs
alloc_slice_u128_64 🟢 1.07x 15015.9 µs ✅ 1.00x 14065.2 µs 🟢 1.08x 15194.2 µs
alloc_slice_u128_128 ✅ 1.03x 29365.9 µs ✅ 1.00x 28623.2 µs ✅ 1.04x 29681.1 µs
alloc_sized_slice_u8_8 ✅ 1.00x 157.6 µs 🟡 1.28x 201.6 µs 🟡 1.34x 211.1 µs
alloc_sized_slice_u8_32 ✅ 1.00x 102.8 µs 🔴 1.91x 196.4 µs 🔴 2.24x 230.5 µs
alloc_sized_slice_u8_64 ✅ 1.00x 146.1 µs 🟠 1.51x 221.3 µs 🟠 1.54x 224.3 µs
alloc_sized_slice_u8_128 ✅ 1.00x 247.2 µs 🟢 1.10x 271.6 µs 🟢 1.14x 282.4 µs
alloc_sized_slice_u16_8 ✅ 1.00x 110.1 µs 🔴 2.19x 240.7 µs 🔴 2.12x 233.3 µs
alloc_sized_slice_u16_32 ✅ 1.00x 154.3 µs 🟠 1.60x 246.8 µs 🟠 1.53x 235.4 µs
alloc_sized_slice_u16_64 ✅ 1.00x 255.7 µs 🟢 1.14x 291.8 µs 🟢 1.09x 279.1 µs
alloc_sized_slice_u16_128 ✅ 1.00x 3278.2 µs ✅ 1.01x 3299.8 µs ✅ 1.01x 3305.7 µs
alloc_sized_slice_u32_8 ✅ 1.00x 179.1 µs 🟠 1.41x 252.8 µs 🟠 1.37x 244.9 µs
alloc_sized_slice_u32_32 ✅ 1.00x 261.1 µs 🟡 1.16x 301.7 µs 🟡 1.24x 323.0 µs
alloc_sized_slice_u32_64 ✅ 1.02x 3405.4 µs ✅ 1.00x 3340.3 µs 🟢 1.11x 3695.3 µs
alloc_sized_slice_u32_128 ✅ 1.05x 7576.7 µs ✅ 1.00x 7245.1 µs 🟢 1.08x 7831.9 µs
alloc_sized_slice_u64_8 ✅ 1.00x 235.2 µs 🟢 1.07x 252.0 µs 🟢 1.10x 257.6 µs
alloc_sized_slice_u64_32 ✅ 1.00x 3300.0 µs ✅ 1.02x 3360.5 µs ✅ 1.02x 3351.7 µs
alloc_sized_slice_u64_64 🟢 1.07x 7595.2 µs ✅ 1.00x 7080.6 µs 🟢 1.11x 7832.8 µs
alloc_sized_slice_u64_128 🟢 1.05x 14940.6 µs ✅ 1.00x 14174.8 µs 🟢 1.07x 15175.5 µs
alloc_sized_slice_u128_8 ✅ 1.00x 220.1 µs 🟡 1.24x 271.9 µs 🟡 1.21x 265.5 µs
alloc_sized_slice_u128_32 🟢 1.07x 7576.7 µs ✅ 1.00x 7063.4 µs 🟢 1.11x 7819.6 µs
alloc_sized_slice_u128_64 🟢 1.07x 14981.6 µs ✅ 1.00x 14055.3 µs 🟢 1.08x 15198.5 µs
alloc_sized_slice_u128_128 ✅ 1.02x 29366.2 µs ✅ 1.00x 28659.9 µs ✅ 1.03x 29414.6 µs
alloc_struct_13 ✅ 1.00x 161.3 µs 🟠 1.55x 250.1 µs 🟠 1.39x 224.4 µs
alloc_struct_24 ✅ 1.00x 108.3 µs 🔴 1.94x 209.6 µs 🔴 1.97x 213.6 µs
alloc_struct_26 ✅ 1.00x 155.7 µs 🟠 1.56x 242.4 µs 🟠 1.52x 236.2 µs
alloc_struct_30 ✅ 1.00x 159.1 µs 🟠 1.54x 245.8 µs 🟠 1.45x 230.1 µs
alloc_struct_32 ✅ 1.00x 151.5 µs 🟠 1.35x 204.9 µs 🟠 1.40x 211.7 µs
alloc_struct_64 ✅ 1.00x 145.3 µs 🟠 1.44x 208.7 µs 🟠 1.48x 215.1 µs
alloc_struct_96 ✅ 1.00x 188.0 µs 🟢 1.13x 211.6 µs 🟡 1.18x 222.6 µs
alloc_struct_128 ✅ 1.00x 209.1 µs 🟡 1.33x 278.4 µs 🟡 1.17x 245.2 µs
alloc_struct_192 ✅ 1.02x 1401.8 µs ✅ 1.00x 1371.7 µs 🟢 1.09x 1492.7 µs
alloc_struct_256 ✅ 1.00x 3324.6 µs 🟡 1.16x 3845.6 µs ✅ 1.01x 3368.2 µs
alloc_struct_512 🟢 1.06x 7708.0 µs ✅ 1.00x 7269.0 µs ✅ 1.02x 7419.6 µs
alloc_struct_1k ✅ 1.00x 14883.4 µs 🟢 1.05x 15633.8 µs ✅ 1.01x 15084.8 µs
alloc_struct_half_chunk_minus_1 ✅ 1.00x 345.1 µs ✅ 1.04x 357.6 µs 🟢 1.11x 382.6 µs
alloc_struct_half_chunk ✅ 1.00x 217.0 µs 🟢 1.12x 243.7 µs 🟡 1.28x 276.7 µs
alloc_struct_half_chunk_plus_1 ✅ 1.00x 353.7 µs ✅ 1.00x 354.1 µs ✅ 1.04x 366.3 µs
alloc_struct_one_chunk_minus_1 ✅ 1.00x 2916.8 µs ✅ 1.02x 2975.2 µs ✅ 1.02x 2968.8 µs
alloc_struct_one_chunk ✅ 1.00x 2924.1 µs ✅ 1.01x 2953.9 µs ✅ 1.03x 3021.9 µs
alloc_struct_one_chunk_plus_1 ✅ 1.00x 2949.6 µs ✅ 1.03x 3027.1 µs ✅ 1.02x 3016.5 µs
alloc_struct_two_chunks ✅ 1.00x 6645.9 µs ✅ 1.01x 6693.1 µs ✅ 1.02x 6749.7 µs
alloc_str_8 🟢 1.11x 246.5 µs ✅ 1.05x 232.7 µs ✅ 1.00x 221.9 µs
alloc_str_16 🟢 1.07x 237.9 µs ✅ 1.02x 226.7 µs ✅ 1.00x 223.3 µs
alloc_str_32 ✅ 1.04x 257.5 µs ✅ 1.00x 247.0 µs 🟢 1.07x 265.4 µs
alloc_str_40 ✅ 1.00x 275.4 µs 🟢 1.08x 296.9 µs 🟢 1.06x 293.0 µs
alloc_str_48 ✅ 1.00x 256.1 µs ✅ 1.03x 263.8 µs 🟢 1.06x 272.2 µs
alloc_str_64 ✅ 1.00x 267.1 µs ✅ 1.04x 277.1 µs 🟢 1.06x 283.8 µs
alloc_str_72 ✅ 1.04x 326.2 µs ✅ 1.00x 313.5 µs 🟢 1.07x 334.4 µs
alloc_str_80 ✅ 1.03x 328.7 µs ✅ 1.00x 318.9 µs 🟢 1.07x 341.1 µs
alloc_str_128 ✅ 1.00x 334.8 µs 🟢 1.11x 371.7 µs 🟢 1.08x 363.2 µs
alloc_slice_lit_u8_8 ✅ 1.00x 99.6 µs 🔴 2.47x 246.5 µs 🔴 2.23x 222.1 µs
alloc_slice_lit_u8_32 ✅ 1.00x 163.0 µs 🔴 1.83x 298.2 µs 🟠 1.71x 279.4 µs
alloc_slice_lit_u8_64 ✅ 1.00x 204.3 µs 🟡 1.34x 274.1 µs 🟠 1.42x 289.5 µs
alloc_slice_lit_u8_128 ✅ 1.00x 266.7 µs 🟡 1.31x 349.0 µs 🟡 1.31x 349.3 µs
alloc_str_lit_8 ✅ 1.00x 122.3 µs 🔴 2.02x 246.5 µs 🔴 1.82x 222.3 µs
alloc_str_lit_16 ✅ 1.00x 139.5 µs 🔴 1.78x 248.5 µs 🟠 1.60x 223.0 µs
alloc_str_lit_32 ✅ 1.00x 197.1 µs 🟠 1.51x 297.6 µs 🟠 1.42x 279.1 µs
alloc_str_lit_40 ✅ 1.00x 148.0 µs 🔴 1.76x 260.4 µs 🔴 1.93x 285.4 µs
alloc_str_lit_48 ✅ 1.00x 156.7 µs 🟠 1.74x 272.3 µs 🔴 1.82x 285.9 µs
alloc_str_lit_64 ✅ 1.00x 171.3 µs 🔴 1.75x 300.7 µs 🟠 1.69x 289.4 µs
alloc_str_lit_72 ✅ 1.00x 208.5 µs 🟠 1.53x 318.5 µs 🟠 1.61x 336.3 µs
alloc_str_lit_80 ✅ 1.00x 204.7 µs 🟠 1.54x 315.1 µs 🟠 1.63x 334.1 µs
alloc_str_lit_128 ✅ 1.00x 258.4 µs 🟠 1.36x 350.8 µs 🟠 1.35x 349.8 µs
clear ✅ 1.00x 241.6 µs ✅ 1.04x 251.9 µs ✅ 1.04x 251.0 µs
clear_and_reuse ✅ 1.00x 75.1 µs 🟥 3.35x 251.6 µs 🟥 3.35x 251.8 µs
miri
This library uses unsafe internally unapologetically, but is miri-clean.
Run the integration test suite under miri to check for undefined behavior.