use `memset_patternN` in `Buffer.fill` #14599

dylan-conway · 2024-10-16T00:30:18Z

What does this PR do?

Before:

❯ bun /Users/dylan/code/bun/bench/snippets/buffer-fill.mjs
cpu: Apple M1 Max
runtime: bun 1.1.30 (arm64-darwin)

benchmark                                          time (avg)             (min … max)       p75       p99      p995
------------------------------------------------------------------------------------- -----------------------------
Buffer.fill 32 bytes with 4 byte value          25.51 ns/iter   (24.36 ns … 51.63 ns)   25.9 ns  29.52 ns  30.23 ns
Buffer.fill 32 bytes with 8 byte value          19.79 ns/iter   (18.05 ns … 29.24 ns)  19.56 ns   24.7 ns  26.19 ns
Buffer.fill 32 bytes with 16 byte value         15.44 ns/iter   (15.15 ns … 21.93 ns)  15.51 ns  17.53 ns  18.36 ns
Buffer.fill 32 bytes with 11 byte value         19.38 ns/iter   (18.62 ns … 29.44 ns)  19.56 ns  21.87 ns   22.9 ns
Buffer.fill 2048 bytes with 4 byte value        75.03 ns/iter   (66.63 ns … 86.33 ns)  75.71 ns  81.61 ns   84.2 ns
Buffer.fill 2048 bytes with 8 byte value         64.2 ns/iter  (59.37 ns … 448.25 ns)  60.92 ns 183.53 ns 257.21 ns
Buffer.fill 2048 bytes with 16 byte value       56.88 ns/iter   (55.89 ns … 65.98 ns)   57.1 ns  62.32 ns  63.14 ns
Buffer.fill 2048 bytes with 11 byte value       67.37 ns/iter   (65.98 ns … 77.59 ns)  68.06 ns  72.23 ns  73.41 ns
Buffer.fill 16384 bytes with 4 byte value       253.1 ns/iter (238.01 ns … 306.48 ns) 254.21 ns 277.29 ns 277.69 ns
Buffer.fill 16384 bytes with 8 byte value      245.86 ns/iter (232.36 ns … 269.86 ns) 248.37 ns 260.68 ns 263.39 ns
Buffer.fill 16384 bytes with 16 byte value     242.69 ns/iter (229.21 ns … 250.71 ns) 245.08 ns 249.44 ns 250.02 ns
Buffer.fill 16384 bytes with 11 byte value        250 ns/iter (243.79 ns … 264.95 ns) 251.09 ns 261.87 ns  263.7 ns
Buffer.fill 2097152 bytes with 4 byte value     36.78 µs/iter   (35.88 µs … 71.83 µs)  36.92 µs  40.33 µs  42.17 µs
Buffer.fill 2097152 bytes with 8 byte value      36.4 µs/iter      (35 µs … 93.96 µs)  36.42 µs  41.71 µs  42.83 µs
Buffer.fill 2097152 bytes with 16 byte value    35.99 µs/iter   (35.13 µs … 62.58 µs)     36 µs  38.83 µs  40.46 µs
Buffer.fill 2097152 bytes with 11 byte value    35.17 µs/iter   (34.29 µs … 68.75 µs)  35.04 µs  41.25 µs  42.92 µs
Buffer.fill 16777216 bytes with 4 byte value   459.35 µs/iter (410.13 µs … 558.08 µs) 475.38 µs 514.38 µs 526.79 µs
Buffer.fill 16777216 bytes with 8 byte value   471.42 µs/iter    (438.75 µs … 580 µs)  478.5 µs 530.96 µs 547.08 µs
Buffer.fill 16777216 bytes with 16 byte value  463.69 µs/iter (435.75 µs … 522.29 µs) 468.92 µs 497.17 µs 503.21 µs
Buffer.fill 16777216 bytes with 11 byte value  460.78 µs/iter     (414 µs … 654.5 µs)    468 µs 558.88 µs 578.25 µs

After:

❯ ./build/release/bun /Users/dylan/code/bun/bench/snippets/buffer-fill.mjs
cpu: Apple M1 Max
runtime: bun 1.1.31 (arm64-darwin)

benchmark                                          time (avg)             (min … max)       p75       p99      p995
------------------------------------------------------------------------------------- -----------------------------
Buffer.fill 32 bytes with 4 byte value          19.61 ns/iter    (18.9 ns … 24.89 ns)  19.57 ns  22.37 ns  23.15 ns
Buffer.fill 32 bytes with 8 byte value          17.81 ns/iter   (16.73 ns … 31.59 ns)   17.6 ns  21.53 ns  22.71 ns
Buffer.fill 32 bytes with 16 byte value         14.13 ns/iter   (13.88 ns … 24.08 ns)  14.07 ns  16.13 ns   16.9 ns
Buffer.fill 32 bytes with 11 byte value         19.03 ns/iter    (18.62 ns … 27.8 ns)  18.86 ns  22.32 ns  22.89 ns
Buffer.fill 2048 bytes with 4 byte value        36.91 ns/iter  (35.69 ns … 345.48 ns)  36.15 ns  42.69 ns  44.08 ns
Buffer.fill 2048 bytes with 8 byte value        35.23 ns/iter   (34.66 ns … 60.75 ns)  35.11 ns  39.97 ns     42 ns
Buffer.fill 2048 bytes with 16 byte value       34.75 ns/iter    (34.25 ns … 43.4 ns)  34.69 ns  38.99 ns  40.24 ns
Buffer.fill 2048 bytes with 11 byte value       73.45 ns/iter   (72.39 ns … 90.92 ns)  73.46 ns   78.9 ns  80.79 ns
Buffer.fill 16384 bytes with 4 byte value      181.37 ns/iter  (179.1 ns … 200.78 ns) 181.72 ns 194.29 ns 196.05 ns
Buffer.fill 16384 bytes with 8 byte value      178.02 ns/iter  (176.37 ns … 194.8 ns) 178.23 ns 190.41 ns 193.99 ns
Buffer.fill 16384 bytes with 16 byte value     177.86 ns/iter  (176.53 ns … 187.2 ns) 178.36 ns 183.72 ns 184.47 ns
Buffer.fill 16384 bytes with 11 byte value     260.66 ns/iter (235.24 ns … 812.54 ns) 254.61 ns 732.31 ns 799.64 ns
Buffer.fill 2097152 bytes with 4 byte value     34.17 µs/iter     (33 µs … 246.79 µs)  33.79 µs   41.5 µs  44.58 µs
Buffer.fill 2097152 bytes with 8 byte value     33.97 µs/iter  (33.08 µs … 117.75 µs)  33.75 µs  40.17 µs  42.08 µs
Buffer.fill 2097152 bytes with 16 byte value    33.79 µs/iter      (33 µs … 50.71 µs)  33.71 µs  36.96 µs  38.46 µs
Buffer.fill 2097152 bytes with 11 byte value     35.7 µs/iter   (34.88 µs … 55.33 µs)  35.71 µs     39 µs  41.42 µs
Buffer.fill 16777216 bytes with 4 byte value   272.61 µs/iter    (268.08 µs … 848 µs) 269.83 µs 319.63 µs 334.13 µs
Buffer.fill 16777216 bytes with 8 byte value   270.25 µs/iter (267.46 µs … 334.83 µs) 269.67 µs 292.08 µs 297.46 µs
Buffer.fill 16777216 bytes with 16 byte value  270.97 µs/iter (268.58 µs … 374.13 µs) 270.17 µs 297.96 µs 301.63 µs
Buffer.fill 16777216 bytes with 11 byte value  485.21 µs/iter   (438.13 µs … 2.95 ms) 469.88 µs   1.03 ms    1.6 ms

Byte length of 11 is also tested to show results when memset_patternN isn't used.