performance-comparison-allocators.md 14KB
Performance Comparison: Allocators
This file is auto-generated
All measurments were taken on 12th Gen Intel® Core™ i9-12900K with 24 logical cores.
Unity Editor version: 2022.2.8f1
To regenerate this file locally use: DOTS -> Unity.Collections -> Generate *** menu.
Table of Contents
Benchmark Results
The following benchmarks make 150 consecutive allocations per sample set.
Multithreaded benchmarks make the full 150 consecutive allocations *per worker thread* per sample set.
The Min of 50 sample sets is compared against the baseline on the far right side of the table.
5 extra sample sets are run as warmup.
Legend
(S)= Safety Enabled
(B)= Burst Compiled with Safety Disabled
(S+B)= Burst Compiled with Safety Enabled
(E)= Engine Provided
italicresults are for benchmarking comparison only; these are not included in standard Performance Framework tests
RewindableAllocator
| Functionality | RewindableAllocator (S) | RewindableAllocator (S+B) | RewindableAllocator (B) | TempJob (E) | Temp (E) | Persistent (E) |
|---|---|---|---|---|---|---|
FixedSize(1, 1024)³ |
11.4µs (2.2x) | 4.0µs (6.3x) | 3.8µs (6.6x) 🟢 | 17.0µs (1.5x) | 10.1µs (2.5x) | 25.1µs (1.0x) 🟠 |
FixedSize(2, 1024)²˒³ |
23.0µs (2.1x) | 23.1µs (2.0x) | 9.0µs (5.2x) 🟢 | 20.2µs (2.3x) | 11.2µs (4.2x) | 47.2µs (1.0x) 🟠 |
FixedSize(4, 1024)²˒³ |
66.4µs (1.9x) | 71.9µs (1.7x) | 80.8µs (1.5x) | 23.5µs (5.3x) | 11.5µs (10.7x) 🟢 | 123.5µs (1.0x) 🟠 |
FixedSize(8, 1024)²˒³ |
167.1µs (2.2x) | 169.2µs (2.2x) | 167.3µs (2.2x) | 45.6µs (8.0x) | 12.8µs (28.6x) 🟢 | 366.4µs (1.0x) 🟠 |
FixedSize(1, 1048576)³ |
11.9µs (16.3x) | 4.7µs (41.3x) | 4.4µs (44.1x) 🟢 | 17.1µs (11.4x) | 10.9µs (17.8x) | 194.1µs (1.0x) 🟠 |
FixedSize(2, 1048576)²˒³ |
26.0µs (10.0x) | 17.0µs (15.2x) | 14.1µs (18.4x) | 32.0µs (8.1x) | 11.7µs (22.1x) 🟢 | 258.9µs (1.0x) 🟠 |
FixedSize(4, 1048576)²˒³ |
70.1µs (11.6x) | 71.3µs (11.4x) | 75.3µs (10.8x) | 208.5µs (3.9x) | 12.5µs (65.0x) 🟢 | 812.2µs (1.0x) 🟠 |
FixedSize(8, 1048576)²˒³ |
139.7µs (14.6x) | 161.0µs (12.7x) | 179.8µs (11.3x) | 1317.1µs (1.5x) | 19.5µs (104.6x) 🟢 | 2039.9µs (1.0x) 🟠 |
IncSize(1, 4096)⁴ |
11.9µs (4.0x) | 4.6µs (10.3x) | 4.3µs (11.0x) 🟢 | 17.9µs (2.6x) | 10.3µs (4.6x) | 47.2µs (1.0x) 🟠 |
IncSize(2, 4096)²˒⁴ |
26.8µs (4.8x) | 10.9µs (11.7x) | 10.5µs (12.2x) 🟢 | 31.7µs (4.0x) | 10.9µs (11.7x) | 127.6µs (1.0x) 🟠 |
IncSize(4, 4096)²˒⁴ |
58.9µs (7.6x) | 67.6µs (6.6x) | 64.5µs (6.9x) | 71.9µs (6.2x) | 11.2µs (39.7x) 🟢 | 444.7µs (1.0x) 🟠 |
IncSize(8, 4096)²˒⁴ |
169.3µs (7.8x) | 159.0µs (8.3x) | 185.7µs (7.1x) | 350.8µs (3.8x) | 11.5µs (114.7x) 🟢 | 1319.0µs (1.0x) 🟠 |
IncSize(1, 65536)⁴ |
12.7µs (49.2x) | 5.0µs (125.1x) | 4.7µs (133.1x) 🟢 | 19.0µs (32.9x) | 11.0µs (56.9x) | 625.4µs (1.0x) 🟠 |
IncSize(2, 65536)²˒⁴ |
25.0µs (46.3x) | 15.8µs (73.3x) | 13.0µs (89.1x) | 578.1µs (2.0x) | 11.3µs (102.5x) 🟢 | 1157.7µs (1.0x) 🟠 |
IncSize(4, 65536)²˒⁴ |
73.3µs (34.7x) | 73.0µs (34.8x) | 70.5µs (36.1x) | 2098.0µs (1.2x) | 11.9µs (213.6x) 🟢 | 2542.2µs (1.0x) 🟠 |
IncSize(8, 65536)²˒⁴ |
141.3µs (40.5x) | 168.1µs (34.1x) | 162.6µs (35.2x) | 6036.0µs (0.9x) 🟠 | 12.7µs (450.8x) 🟢 | 5724.9µs (1.0x) |
DecSize(1, 4096)⁵ |
12.2µs (6.1x) | 4.6µs (16.1x) | 4.3µs (17.2x) 🟢 | 16.9µs (4.4x) | 9.8µs (7.5x) | 73.9µs (1.0x) 🟠 |
DecSize(2, 4096)²˒⁵ |
27.6µs (3.4x) | 12.5µs (7.6x) | 11.9µs (8.0x) | 37.3µs (2.5x) | 11.4µs (8.3x) 🟢 | 94.9µs (1.0x) 🟠 |
DecSize(4, 4096)²˒⁵ |
68.5µs (7.5x) | 74.7µs (6.8x) | 69.4µs (7.4x) | 79.3µs (6.5x) | 11.0µs (46.5x) 🟢 | 511.6µs (1.0x) 🟠 |
DecSize(8, 4096)²˒⁵ |
173.5µs (7.4x) | 173.4µs (7.4x) | 168.3µs (7.6x) | 313.4µs (4.1x) | 17.1µs (75.1x) 🟢 | 1284.6µs (1.0x) 🟠 |
DecSize(1, 65536)⁵ |
12.1µs (47.9x) | 4.6µs (126.0x) | 4.3µs (134.8x) 🟢 | 20.8µs (27.9x) | 11.7µs (49.6x) | 579.8µs (1.0x) 🟠 |
DecSize(2, 65536)²˒⁵ |
28.6µs (37.1x) | 17.7µs (60.0x) | 11.5µs (92.3x) 🟢 | 658.8µs (1.6x) | 12.5µs (84.9x) | 1061.4µs (1.0x) 🟠 |
DecSize(4, 65536)²˒⁵ |
67.3µs (38.8x) | 69.4µs (37.6x) | 73.1µs (35.7x) | 2386.4µs (1.1x) | 14.2µs (183.8x) 🟢 | 2609.3µs (1.0x) 🟠 |
DecSize(8, 65536)²˒⁵ |
154.4µs (37.8x) | 166.6µs (35.0x) | 155.9µs (37.4x) | 5938.8µs (1.0x) 🟠 | 28.6µs (203.9x) 🟢 | 5830.8µs (1.0x) |
² Benchmark run on parallel job workers - results may vary
³ FixedSize(workerThreads, allocSize)
⁴ IncSize(workerThreads, allocSize) -- Makes linearly increasing allocations [1⋅allocSize, 2⋅allocSize … N⋅allocSize]
⁵ DecSize(workerThreads, allocSize) -- Makes linearly decreasing allocations [N⋅allocSize … 2⋅allocSize, 1⋅allocSize]