RESOLVED FIXED245797
[JSC] Introduce new_array_with_species bytecode and optimize it through all tiers 
https://bugs.webkit.org/show_bug.cgi?id=245797
Summary [JSC] Introduce new_array_with_species bytecode and optimize it through all t...
Jarred Sumner
Reported 2022-09-28 14:12:17 PDT
Created attachment 462692 [details] array-map.mjs A lot of modern JS uses .map and .forEach. Many codebases use these functions more frequently than for loops. For small array literals with constant values, V8 is about 3x faster than what JSC is doing. I've attached a benchmark script Bun using a copy of JSC from September 17th, 2022: cpu: Apple M1 Max runtime: bun 0.1.14 (arm64-darwin) benchmark time (avg) (min … max) p75 p99 p995 ------------------------------------------------------------- ----------------------------- Array.map x 0 50.91 ns/iter (45.7 ns … 274.74 ns) 47.35 ns 113.15 ns 131.34 ns Array.map x 1 55.15 ns/iter (51.91 ns … 453.84 ns) 52.44 ns 101.3 ns 119.57 ns Array.map x 2 58.1 ns/iter (51.91 ns … 183.54 ns) 52.77 ns 116.05 ns 151.96 ns Array.map x 3 52.16 ns/iter (46.56 ns … 152.19 ns) 47.15 ns 105.3 ns 125.74 ns Array.map x 4 78.14 ns/iter (51.94 ns … 164.94 ns) 83.45 ns 129.8 ns 139.77 ns Array.map x 5 54.94 ns/iter (51.36 ns … 226.1 ns) 53.67 ns 101.87 ns 109.88 ns Array.map x 6 55.97 ns/iter (51.9 ns … 257.6 ns) 54.43 ns 105.08 ns 165.98 ns Array.map x 7 55.5 ns/iter (51.84 ns … 162.09 ns) 54.43 ns 102.56 ns 112.98 ns Array.map x 8 57.03 ns/iter (51.84 ns … 476.93 ns) 54.97 ns 121.98 ns 148.4 ns Array.map x 9 56.4 ns/iter (51.85 ns … 370.1 ns) 54.74 ns 116.15 ns 131.97 ns Array.map x 10 56.05 ns/iter (51.86 ns … 212.73 ns) 54.53 ns 103.42 ns 108.92 ns Array.map x 11 55.92 ns/iter (51.86 ns … 187.18 ns) 54.53 ns 103.99 ns 133.61 ns Array.map x 12 55.93 ns/iter (51.88 ns … 272.41 ns) 54.48 ns 107.03 ns 120.77 ns Array.map x 13 56.08 ns/iter (51.86 ns … 174.88 ns) 54.64 ns 115.71 ns 127.24 ns Array.map x 14 55.97 ns/iter (51.88 ns … 191.69 ns) 54.55 ns 110.09 ns 119.5 ns Array.map x 15 55.87 ns/iter (51.86 ns … 357.33 ns) 54.51 ns 102.73 ns 111.7 ns Array.map x 16 55.98 ns/iter (51.85 ns … 181.74 ns) 54.52 ns 106.4 ns 117.05 ns Array.map x 17 56.25 ns/iter (51.88 ns … 380.5 ns) 54.54 ns 106.85 ns 117.7 ns Array.map x 18 55.41 ns/iter (51.85 ns … 127.13 ns) 54.54 ns 97.5 ns 103.66 ns Array.map x 19 55.73 ns/iter (51.87 ns … 267.85 ns) 54.61 ns 100.46 ns 106.79 ns inline Array.map x 0 31.8 ns/iter (28.91 ns … 239.6 ns) 31.2 ns 73.13 ns 84.8 ns inline Array.map x 1 33.6 ns/iter (30.74 ns … 106.19 ns) 33.89 ns 77.63 ns 90.15 ns inline Array.map x 2 34.59 ns/iter (31.94 ns … 239.49 ns) 34.82 ns 75.1 ns 85.36 ns inline Array.map x 3 36.49 ns/iter (33.39 ns … 151.94 ns) 36.17 ns 79.87 ns 93.93 ns inline Array.map x 4 38.94 ns/iter (35.6 ns … 295.58 ns) 38.79 ns 87.57 ns 93.78 ns inline Array.map x 5 40.37 ns/iter (36.88 ns … 268.68 ns) 39.93 ns 92.47 ns 97.58 ns inline Array.map x 6 43.33 ns/iter (39.58 ns … 272.25 ns) 42.77 ns 93.85 ns 101.4 ns inline Array.map x 7 44.73 ns/iter (40.56 ns … 505.39 ns) 43.34 ns 99.42 ns 138.26 ns inline Array.map x 8 41.82 ns/iter (37.98 ns … 161.33 ns) 41.22 ns 89.9 ns 98.47 ns inline Array.map x 9 42.2 ns/iter (38.85 ns … 104.19 ns) 41.62 ns 87.07 ns 92.6 ns inline Array.map x 10 54.53 ns/iter (49.97 ns … 185.29 ns) 53.61 ns 105.6 ns 113.41 ns inline Array.map x 11 55.63 ns/iter (50.92 ns … 251.47 ns) 54.47 ns 103.83 ns 173.58 ns inline Array.map x 12 56.66 ns/iter (51.74 ns … 278.88 ns) 56.14 ns 105.58 ns 107.75 ns inline Array.map x 13 57.31 ns/iter (52.67 ns … 212.94 ns) 56.72 ns 103.72 ns 115.22 ns inline Array.map x 14 53.85 ns/iter (48.2 ns … 203.39 ns) 54 ns 107.08 ns 115.79 ns inline Array.map x 15 55.64 ns/iter (48.61 ns … 265.81 ns) 55.93 ns 107.91 ns 114.97 ns inline Array.map x 16 56.43 ns/iter (49.25 ns … 408.1 ns) 55.23 ns 111.18 ns 137.67 ns inline Array.map x 17 57.4 ns/iter (50.05 ns … 143.59 ns) 56.14 ns 110.2 ns 116.92 ns inline Array.map x 18 58.43 ns/iter (51.64 ns … 267.74 ns) 58.65 ns 112.74 ns 121.68 ns inline Array.map x 19 59.89 ns/iter (51.9 ns … 476.5 ns) 59.19 ns 109.14 ns 112.25 ns Node v18.9.1: benchmark time (avg) (min … max) p75 p99 p995 ------------------------------------------------------------- ----------------------------- Array.map x 0 34.04 ns/iter (29.44 ns … 467.84 ns) 33.85 ns 42.88 ns 43.56 ns Array.map x 1 40.11 ns/iter (30.09 ns … 244.13 ns) 41.4 ns 109.83 ns 130.5 ns Array.map x 2 35.94 ns/iter (25.49 ns … 533.06 ns) 39.04 ns 47.24 ns 47.97 ns Array.map x 3 34.42 ns/iter (28.05 ns … 42.55 ns) 35.67 ns 39.95 ns 40.68 ns Array.map x 4 32.94 ns/iter (28 ns … 45.35 ns) 33.4 ns 38.43 ns 40.02 ns Array.map x 5 36.87 ns/iter (28.02 ns … 51.1 ns) 38.55 ns 45.81 ns 47.34 ns Array.map x 6 34.6 ns/iter (28.05 ns … 47.17 ns) 36.35 ns 41.41 ns 42.43 ns Array.map x 7 32.59 ns/iter (28 ns … 40.6 ns) 32.58 ns 37.35 ns 38.05 ns Array.map x 8 32.49 ns/iter (28.05 ns … 45.3 ns) 32.5 ns 36.84 ns 37.49 ns Array.map x 9 32.61 ns/iter (28 ns … 49.33 ns) 32.5 ns 38.6 ns 42.77 ns Array.map x 10 32.45 ns/iter (28.01 ns … 39.56 ns) 32.42 ns 36.15 ns 36.83 ns Array.map x 11 34.8 ns/iter (28.02 ns … 404.71 ns) 34.34 ns 66.5 ns 84.86 ns Array.map x 12 32.45 ns/iter (28.02 ns … 46.71 ns) 32.46 ns 35.9 ns 37.15 ns Array.map x 13 32.37 ns/iter (27.99 ns … 41.37 ns) 32.42 ns 37.14 ns 38.39 ns Array.map x 14 32.52 ns/iter (27.99 ns … 42.66 ns) 32.52 ns 36.39 ns 38.6 ns Array.map x 15 31.69 ns/iter (28.01 ns … 55.96 ns) 32.58 ns 34.9 ns 39.68 ns Array.map x 16 30.77 ns/iter (28.13 ns … 40.82 ns) 33.07 ns 35.27 ns 36.27 ns Array.map x 17 30.73 ns/iter (28.13 ns … 40.67 ns) 33 ns 35.59 ns 36.96 ns Array.map x 18 30.77 ns/iter (28.13 ns … 39.08 ns) 33.2 ns 34.04 ns 34.35 ns Array.map x 19 31.03 ns/iter (28.11 ns … 46.27 ns) 33.08 ns 37.92 ns 42.6 ns inline Array.map x 0 7.59 ns/iter (6.05 ns … 415.02 ns) 10.49 ns 11.85 ns 13.33 ns inline Array.map x 1 6.01 ns/iter (5.25 ns … 20.8 ns) 5.45 ns 10.9 ns 11.81 ns inline Array.map x 2 7.41 ns/iter (6.56 ns … 15.85 ns) 6.75 ns 11.69 ns 11.84 ns inline Array.map x 3 9.02 ns/iter (8.1 ns … 22.63 ns) 8.31 ns 13.24 ns 13.82 ns inline Array.map x 4 11.25 ns/iter (10.13 ns … 25.14 ns) 10.54 ns 16.12 ns 17.26 ns inline Array.map x 5 12.33 ns/iter (10.62 ns … 22.12 ns) 11.62 ns 16.96 ns 17.82 ns inline Array.map x 6 13.37 ns/iter (11.57 ns … 27.57 ns) 12.72 ns 17.59 ns 18.2 ns inline Array.map x 7 14.45 ns/iter (13.08 ns … 29.82 ns) 17.48 ns 19.14 ns 20.86 ns inline Array.map x 8 15.47 ns/iter (14.03 ns … 25.82 ns) 18.61 ns 19.32 ns 20.18 ns inline Array.map x 9 16.58 ns/iter (15 ns … 33.03 ns) 19.65 ns 20.6 ns 21.1 ns inline Array.map x 10 17.6 ns/iter (15.88 ns … 28.53 ns) 20.45 ns 22.87 ns 23.76 ns inline Array.map x 11 18.26 ns/iter (16.79 ns … 51.02 ns) 21.28 ns 24.35 ns 25.29 ns inline Array.map x 12 18.79 ns/iter (17.72 ns … 35.87 ns) 17.79 ns 23.95 ns 24.26 ns inline Array.map x 13 19.9 ns/iter (18.65 ns … 33.07 ns) 19.28 ns 26.57 ns 29.35 ns inline Array.map x 14 20.79 ns/iter (19.58 ns … 31.04 ns) 19.68 ns 25.87 ns 26.63 ns inline Array.map x 15 21.73 ns/iter (20.52 ns … 40.35 ns) 20.59 ns 26.63 ns 27.52 ns inline Array.map x 16 22.82 ns/iter (21.45 ns … 35.67 ns) 22.2 ns 29.01 ns 30.38 ns inline Array.map x 17 23.75 ns/iter (22.39 ns … 34.28 ns) 22.57 ns 28.8 ns 29.43 ns inline Array.map x 18 24.75 ns/iter (23.32 ns … 35.58 ns) 23.55 ns 29.62 ns 30.4 ns inline Array.map x 19 25.69 ns/iter (24.25 ns … 36.99 ns) 25.53 ns 30.86 ns 31.65 ns
Attachments
array-map.mjs (1.90 KB, text/plain)
2022-09-28 14:12 PDT, Jarred Sumner 		no flags
Details
View All Add attachment proposed patch, testcase, etc.
Radar WebKit Bug Importer
Comment 1 2022-09-28 20:27:29 PDT
<rdar://problem/100541275>
Yusuke Suzuki
Comment 2 2022-10-01 03:39:33 PDT
WIP cpu: unknown runtime: unknown (unknown) benchmark time (avg) (min … max) p75 p99 p995 ------------------------------------------------------------- ----------------------------- Array.map x 0 28.49 ns/iter (26.84 ns … 148.08 ns) 27.63 ns 59.3 ns 76.17 ns Array.map x 1 33.21 ns/iter (31.63 ns … 97.94 ns) 32.53 ns 62.02 ns 83.18 ns Array.map x 2 32.17 ns/iter (27.21 ns … 130.01 ns) 31.69 ns 73.57 ns 78.73 ns Array.map x 3 45.48 ns/iter (27.2 ns … 127.62 ns) 69.28 ns 81.28 ns 98.59 ns Array.map x 4 28.96 ns/iter (27.14 ns … 94.2 ns) 28.05 ns 64.62 ns 76.08 ns Array.map x 5 28.55 ns/iter (27.17 ns … 91.72 ns) 28.03 ns 39 ns 64.9 ns Array.map x 6 28.46 ns/iter (27.17 ns … 80.81 ns) 28.04 ns 39.17 ns 57.07 ns Array.map x 7 28.88 ns/iter (27.17 ns … 93.39 ns) 28.1 ns 62.61 ns 72.78 ns Array.map x 8 29.12 ns/iter (27.18 ns … 161.94 ns) 28.07 ns 67.97 ns 82.15 ns Array.map x 9 28.59 ns/iter (27.18 ns … 85.14 ns) 28.05 ns 47 ns 64.41 ns Array.map x 10 28.52 ns/iter (27.19 ns … 78.88 ns) 28.09 ns 39.6 ns 58.16 ns Array.map x 11 29.13 ns/iter (27.17 ns … 110.19 ns) 28.13 ns 79.15 ns 84.6 ns Array.map x 12 29.03 ns/iter (27.15 ns … 91.61 ns) 28.22 ns 65.89 ns 80.31 ns Array.map x 13 29.14 ns/iter (27.18 ns … 93.42 ns) 28.17 ns 69.51 ns 82.66 ns Array.map x 14 29.16 ns/iter (27.16 ns … 99.3 ns) 28.18 ns 70.8 ns 82.09 ns Array.map x 15 29.07 ns/iter (27.18 ns … 94.77 ns) 28.15 ns 65.42 ns 81.82 ns Array.map x 16 29.27 ns/iter (27.17 ns … 97.29 ns) 28.2 ns 71.23 ns 79.35 ns Array.map x 17 29.23 ns/iter (27.19 ns … 97.42 ns) 28.23 ns 67.93 ns 76.22 ns Array.map x 18 29.18 ns/iter (27.17 ns … 91.88 ns) 28.22 ns 71.79 ns 80.78 ns Array.map x 19 29.26 ns/iter (27.2 ns … 94.75 ns) 28.17 ns 71.42 ns 81.85 ns inline Array.map x 0 32.57 ns/iter (31.2 ns … 84.23 ns) 32.45 ns 41.88 ns 47.78 ns inline Array.map x 1 32.66 ns/iter (31.48 ns … 101.39 ns) 32.6 ns 42.28 ns 43.43 ns inline Array.map x 2 33.92 ns/iter (32.6 ns … 91.01 ns) 33.83 ns 44.92 ns 58.12 ns inline Array.map x 3 34.96 ns/iter (33.48 ns … 103.55 ns) 34.86 ns 45.21 ns 54.17 ns inline Array.map x 4 37.74 ns/iter (35.86 ns … 95.65 ns) 37.72 ns 48.17 ns 59.15 ns inline Array.map x 5 38.56 ns/iter (36.45 ns … 100.56 ns) 38.18 ns 54.9 ns 74.98 ns inline Array.map x 6 29.37 ns/iter (28.24 ns … 51.83 ns) 29.28 ns 38.95 ns 39.67 ns inline Array.map x 7 30.77 ns/iter (29.32 ns … 56.58 ns) 30.72 ns 40.22 ns 40.66 ns inline Array.map x 8 27.2 ns/iter (25.83 ns … 82.72 ns) 27.17 ns 37.07 ns 37.49 ns inline Array.map x 9 27.97 ns/iter (26.78 ns … 52.23 ns) 27.91 ns 37.84 ns 38.38 ns inline Array.map x 10 34.38 ns/iter (32.41 ns … 68.93 ns) 34.26 ns 44.48 ns 44.83 ns inline Array.map x 11 36.91 ns/iter (34.84 ns … 79.72 ns) 36.84 ns 47.31 ns 47.94 ns inline Array.map x 12 36.76 ns/iter (34.66 ns … 111.31 ns) 36.54 ns 46.86 ns 47.39 ns inline Array.map x 13 32.41 ns/iter (30.93 ns … 69.17 ns) 32.26 ns 43.06 ns 43.51 ns inline Array.map x 14 33.86 ns/iter (32.17 ns … 76.42 ns) 33.71 ns 43.96 ns 44.37 ns inline Array.map x 15 34.65 ns/iter (33.05 ns … 68.21 ns) 34.45 ns 44.38 ns 44.65 ns inline Array.map x 16 35.84 ns/iter (34.27 ns … 58.6 ns) 35.66 ns 45.48 ns 45.77 ns inline Array.map x 17 36.43 ns/iter (34.9 ns … 61.27 ns) 36.23 ns 46.1 ns 46.3 ns inline Array.map x 18 37.62 ns/iter (36.07 ns … 93.94 ns) 37.4 ns 47.55 ns 47.75 ns inline Array.map x 19 38.46 ns/iter (36.96 ns … 63.12 ns) 38.29 ns 47.67 ns 48.23 ns
Yusuke Suzuki
Comment 3 2022-10-02 00:01:52 PDT
Probably this is good enough since non-inline one is more important use case than inline case. inline one can be improved if we introduce a bit more complicated MovHint analysis. benchmark time (avg) (min … max) p75 p99 p995 ------------------------------------------------------------- ----------------------------- Array.map x 0 24.54 ns/iter (22.94 ns … 131.96 ns) 23.7 ns 52.89 ns 72.85 ns Array.map x 1 29.48 ns/iter (28.38 ns … 91.36 ns) 29.17 ns 39.09 ns 40.02 ns Array.map x 2 27.35 ns/iter (23.62 ns … 108.33 ns) 24.51 ns 68.3 ns 70.45 ns Array.map x 3 39.27 ns/iter (23.61 ns … 136.4 ns) 64.96 ns 76.57 ns 78.4 ns Array.map x 4 24.36 ns/iter (23.29 ns … 49.6 ns) 24.05 ns 34.45 ns 34.77 ns Array.map x 5 24.56 ns/iter (23.29 ns … 76.57 ns) 24.07 ns 36.74 ns 50.33 ns Array.map x 6 24.4 ns/iter (23.31 ns … 35.53 ns) 24.1 ns 34.31 ns 34.61 ns Array.map x 7 24.61 ns/iter (23.29 ns … 84.06 ns) 24.1 ns 34.63 ns 49.02 ns Array.map x 8 24.47 ns/iter (23.29 ns … 65.72 ns) 24.13 ns 34.63 ns 36.55 ns Array.map x 9 24.39 ns/iter (23.27 ns … 35.66 ns) 24.11 ns 34.44 ns 34.56 ns Array.map x 10 24.46 ns/iter (23.28 ns … 68.06 ns) 24.13 ns 34.26 ns 34.59 ns Array.map x 11 24.45 ns/iter (23.28 ns … 87.8 ns) 24.17 ns 34.46 ns 34.82 ns Array.map x 12 24.5 ns/iter (23.28 ns … 70.95 ns) 24.16 ns 34.5 ns 35.24 ns Array.map x 13 24.47 ns/iter (23.28 ns … 75.69 ns) 24.12 ns 34.26 ns 34.57 ns Array.map x 14 24.54 ns/iter (23.29 ns … 89.96 ns) 24.14 ns 34.45 ns 35.22 ns Array.map x 15 24.47 ns/iter (23.28 ns … 57.91 ns) 24.19 ns 34.42 ns 34.66 ns Array.map x 16 24.62 ns/iter (23.29 ns … 69.3 ns) 24.22 ns 34.65 ns 49 ns Array.map x 17 24.6 ns/iter (23.27 ns … 73.38 ns) 24.18 ns 34.9 ns 43.3 ns Array.map x 18 24.65 ns/iter (23.29 ns … 90.89 ns) 24.2 ns 35.15 ns 42.95 ns Array.map x 19 24.58 ns/iter (23.28 ns … 80.27 ns) 24.21 ns 34.79 ns 37.04 ns inline Array.map x 0 19.7 ns/iter (18.96 ns … 33.95 ns) 19.6 ns 28.68 ns 29.1 ns inline Array.map x 1 23.67 ns/iter (22.77 ns … 53.31 ns) 23.57 ns 33.13 ns 33.54 ns inline Array.map x 2 24.75 ns/iter (23.66 ns … 66.57 ns) 24.56 ns 35.31 ns 35.83 ns inline Array.map x 3 26.4 ns/iter (24.76 ns … 113.46 ns) 25.85 ns 50.35 ns 55.89 ns inline Array.map x 4 26.9 ns/iter (25.72 ns … 61.44 ns) 26.7 ns 36.27 ns 36.78 ns inline Array.map x 5 26.71 ns/iter (25.53 ns … 77.83 ns) 26.63 ns 36.52 ns 37.21 ns inline Array.map x 6 28.69 ns/iter (27.15 ns … 61.93 ns) 28.81 ns 38.05 ns 38.55 ns inline Array.map x 7 32.51 ns/iter (30.23 ns … 77.6 ns) 33 ns 44.39 ns 44.99 ns inline Array.map x 8 26.38 ns/iter (25.42 ns … 94.17 ns) 26.2 ns 35.94 ns 36.53 ns inline Array.map x 9 26.95 ns/iter (26.03 ns … 48.96 ns) 26.85 ns 36.67 ns 37.02 ns inline Array.map x 10 29.16 ns/iter (28.09 ns … 54.53 ns) 28.98 ns 38.25 ns 38.58 ns inline Array.map x 11 30.03 ns/iter (29 ns … 52.77 ns) 29.88 ns 38.81 ns 39.3 ns inline Array.map x 12 31.15 ns/iter (30.15 ns … 69.39 ns) 30.95 ns 39.96 ns 40.58 ns inline Array.map x 13 31.98 ns/iter (30.96 ns … 54.88 ns) 31.78 ns 41.58 ns 41.92 ns inline Array.map x 14 33.01 ns/iter (31.87 ns … 49.52 ns) 32.68 ns 42.47 ns 42.64 ns inline Array.map x 15 33.81 ns/iter (32.65 ns … 58.02 ns) 33.47 ns 43.13 ns 43.36 ns inline Array.map x 16 35.03 ns/iter (33.81 ns … 62.9 ns) 34.71 ns 44.63 ns 45.16 ns inline Array.map x 17 35.81 ns/iter (34.46 ns … 118.63 ns) 35.45 ns 45.33 ns 45.87 ns inline Array.map x 18 36.77 ns/iter (35.35 ns … 58.17 ns) 36.57 ns 46.85 ns 47.06 ns inline Array.map x 19 37.95 ns/iter (36.56 ns … 65.8 ns) 37.68 ns 48.07 ns 48.39 ns
Yusuke Suzuki
Comment 4 2022-10-02 00:04:26 PDT
Pull request: https://github.com/WebKit/WebKit/pull/4909
EWS
Comment 5 2022-10-03 18:47:12 PDT
Committed 255105@main (3c9c31d0fd7c): <https://commits.webkit.org/255105@main> Reviewed commits have been landed. Closing PR #4909 and removing active labels.
Note You need to log in before you can comment on or make changes to this bug.