GC Parallel Sweeping

[Hirevo]

The idea for this PR came to me after realising that, technically, deallocating values is a pure waste of time for the interpreter, and doesn't affect any of its computations.

So this PR adds a new dedicated thread, maintained by the GC heap, that is dedicated to being sent the dead objects identified by the GC and deallocate them there, instead than on the main interpreter thread, in the hope that it then allows the interpreter to resume its work sooner.

I am unsure at the present moment on whether this is a performance improvement or not.

Depends on #33.

[som-rs-benchmarker]

Here are the benchmark results for feat/parallel-sweeping (commit: 58d555f):

AST interpreter

+-----------------+----------------------------------------+-------------------------------+
| Benchmark       | master (base)                          | feat/parallel-sweeping (head) |
+-----------------+----------------------------------------+-------------------------------+
| Bounce          | 202.27 ms ± 10.28 (185.55..217.29)     | 0.86x ± 0.15 (0.68..1.04)     |
| BubbleSort      | 283.07 ms ± 36.41 (243.28..364.36)     | 1.03x ± 0.14 (0.96..1.11)     |
| DeltaBlue       | 168.45 ms ± 12.41 (148.39..186.87)     | 1.02x ± 0.17 (0.80..1.17)     |
| Dispatch        | 217.88 ms ± 43.39 (180.77..298.29)     | 1.06x ± 0.24 (0.93..1.25)     |
| Fannkuch        | 131.81 ms ± 11.62 (112.85..148.99)     | 0.99x ± 0.11 (0.90..1.09)     |
| Fibonacci       | 377.62 ms ± 20.36 (353.01..407.17)     | 0.97x ± 0.12 (0.81..1.07)     |
| FieldLoop       | 343.02 ms ± 40.38 (305.33..451.97)     | 1.00x ± 0.15 (0.82..1.14)     |
| GraphSearch     | 100.28 ms ± 20.33 (83.20..148.40)      | 1.21x ± 0.25 (1.15..1.29)     |
| IntegerLoop     | 369.28 ms ± 48.60 (315.29..454.61)     | 1.02x ± 0.16 (0.86..1.09)     |
| JsonSmall       | 223.71 ms ± 22.60 (191.88..255.13)     | 1.11x ± 0.13 (1.02..1.21)     |
| List            | 249.96 ms ± 10.54 (238.68..266.56)     | 0.96x ± 0.12 (0.75..1.08)     |
| Loop            | 432.63 ms ± 21.98 (403.24..467.37)     | 1.03x ± 0.11 (0.86..1.12)     |
| Mandelbrot      | 274.07 ms ± 18.59 (250.79..311.52)     | 1.04x ± 0.11 (0.87..1.15)     |
| NBody           | 230.91 ms ± 20.71 (201.71..256.89)     | 1.03x ± 0.11 (0.88..1.09)     |
| PageRank        | 335.60 ms ± 30.96 (292.55..379.90)     | 1.07x ± 0.13 (0.90..1.21)     |
| Permute         | 321.81 ms ± 29.21 (294.16..383.19)     | 0.99x ± 0.12 (0.83..1.07)     |
| Queens          | 259.25 ms ± 23.19 (221.09..299.67)     | 0.93x ± 0.14 (0.76..1.13)     |
| QuickSort       | 73.75 ms ± 2.75 (71.34..79.88)         | 0.91x ± 0.10 (0.76..1.02)     |
| Recurse         | 309.77 ms ± 38.69 (272.89..396.57)     | 1.04x ± 0.15 (0.92..1.12)     |
| Richards        | 4107.32 ms ± 102.28 (3941.51..4277.60) | 1.00x ± 0.04 (0.92..1.03)     |
| Sieve           | 427.53 ms ± 26.13 (396.18..476.53)     | 0.93x ± 0.12 (0.73..1.06)     |
| Storage         | 92.85 ms ± 8.05 (83.27..104.08)        | 1.04x ± 0.15 (0.81..1.17)     |
| Sum             | 170.19 ms ± 13.69 (156.92..202.00)     | 1.08x ± 0.11 (0.97..1.18)     |
| Towers          | 329.32 ms ± 25.30 (298.39..370.25)     | 1.02x ± 0.10 (0.92..1.12)     |
| TreeSort        | 166.31 ms ± 11.74 (151.23..187.77)     | 0.99x ± 0.11 (0.84..1.11)     |
| WhileLoop       | 389.74 ms ± 36.04 (345.53..439.52)     | 1.07x ± 0.11 (0.99..1.12)     |
|                 |                                        |                               |
| Average Speedup |               (baseline)               | 1.02x ± 0.03 (0.86..1.21)     |
+-----------------+----------------------------------------+-------------------------------+

The raw ReBench data files are available for download here: baseline and head

Bytecode interpreter

+-----------------+---------------------------------------+-------------------------------+
| Benchmark       | master (base)                         | feat/parallel-sweeping (head) |
+-----------------+---------------------------------------+-------------------------------+
| Bounce          | 73.49 ms ± 4.34 (67.67..81.48)        | 0.82x ± 0.13 (0.63..0.96)     |
| BubbleSort      | 110.80 ms ± 11.98 (98.67..137.42)     | 0.90x ± 0.13 (0.80..1.00)     |
| DeltaBlue       | 59.52 ms ± 4.19 (54.52..68.01)        | 0.70x ± 0.17 (0.51..0.87)     |
| Dispatch        | 91.81 ms ± 17.23 (73.43..122.87)      | 0.92x ± 0.21 (0.72..1.12)     |
| Fannkuch        | 48.19 ms ± 5.25 (43.91..57.86)        | 0.70x ± 0.21 (0.48..1.00)     |
| Fibonacci       | 144.44 ms ± 21.87 (129.45..196.34)    | 0.80x ± 0.14 (0.67..0.87)     |
| FieldLoop       | 162.46 ms ± 13.01 (149.47..187.32)    | 0.99x ± 0.17 (0.72..1.14)     |
| GraphSearch     | 33.76 ms ± 5.25 (30.31..48.26)        | 0.88x ± 0.20 (0.66..1.06)     |
| IntegerLoop     | 147.27 ms ± 9.56 (136.69..170.50)     | 0.83x ± 0.13 (0.66..0.97)     |
| JsonSmall       | 87.17 ms ± 5.96 (78.66..98.68)        | 0.91x ± 0.10 (0.76..1.04)     |
| List            | 111.58 ms ± 16.08 (96.90..138.33)     | 0.83x ± 0.18 (0.58..0.98)     |
| Loop            | 194.04 ms ± 14.88 (175.88..224.23)    | 0.92x ± 0.09 (0.83..1.02)     |
| Mandelbrot      | 115.69 ms ± 15.09 (101.48..150.91)    | 0.86x ± 0.17 (0.65..1.02)     |
| NBody           | 79.22 ms ± 5.52 (75.17..90.97)        | 0.79x ± 0.14 (0.58..0.97)     |
| PageRank        | 124.69 ms ± 5.97 (115.19..138.17)     | 0.79x ± 0.12 (0.64..1.02)     |
| Permute         | 116.94 ms ± 11.41 (104.57..144.25)    | 0.83x ± 0.18 (0.56..1.00)     |
| Queens          | 93.48 ms ± 6.70 (85.59..106.84)       | 0.90x ± 0.19 (0.60..1.05)     |
| QuickSort       | 33.56 ms ± 7.73 (27.08..44.85)        | 0.95x ± 0.24 (0.81..1.13)     |
| Recurse         | 122.43 ms ± 17.71 (108.85..171.40)    | 0.85x ± 0.18 (0.59..0.98)     |
| Richards        | 1473.83 ms ± 97.54 (1345.55..1691.55) | 0.75x ± 0.08 (0.65..0.85)     |
| Sieve           | 166.66 ms ± 14.13 (156.32..205.15)    | 0.83x ± 0.16 (0.58..0.96)     |
| Storage         | 32.92 ms ± 1.52 (31.30..36.09)        | 0.79x ± 0.14 (0.59..0.97)     |
| Sum             | 69.05 ms ± 3.58 (65.13..75.51)        | 0.83x ± 0.20 (0.59..1.02)     |
| Towers          | 130.08 ms ± 12.86 (119.07..162.76)    | 0.99x ± 0.16 (0.73..1.09)     |
| TreeSort        | 52.32 ms ± 7.40 (45.34..64.02)        | 0.94x ± 0.16 (0.83..1.09)     |
| WhileLoop       | 170.93 ms ± 9.25 (159.72..191.28)     | 0.86x ± 0.14 (0.62..0.99)     |
|                 |                                       |                               |
| Average Speedup |              (baseline)               | 0.85x ± 0.03 (0.70..0.99)     |
+-----------------+---------------------------------------+-------------------------------+

The raw ReBench data files are available for download here: baseline and head

The benchmarks were run using ReBench v1.2.0
The statistical analysis was done using rebench-tabler v0.1.0

The source code of this benchmark runner is available as a GitHub Gist for more details about the setup