Skip to content

buffer: directly use ArrayBuffer as the pool

Rodrigo Muino Tomonari requested to merge github/fork/addaleax/buffer-arraybuffer-pool into master
Checklist
  • make -j4 test (UNIX), or vcbuild test nosign (Windows) passes
  • tests and/or benchmarks are included (benchmarks are already there \o/)
  • commit message follows commit guidelines
Affected core subsystem(s)

buffer

Description of change

Make the buffer pool an ArrayBuffer which is used directly, speeding up allocation noticeably in some cases. The only drawback happens when creating pool-based Buffer instances from strings whose byte lengths got overestimated by Buffer.byteLength, e.g. for base64-encoded strings containing whitespace, where two Buffer instances are being created.

This may also be useful when providing Buffer classes in the future.

Benchmark results for benchmark/buffers/buffer-creation.js:

                                 improvement significant      p.value
len=1024 type="buffer()"             47.11 %         *** 5.202555e-12
len=1024 type="fast-alloc"           -3.41 %             3.823226e-01
len=1024 type="fast-alloc-fill"       1.11 %             7.985624e-01
len=1024 type="fast-allocUnsafe"     24.37 %         *** 4.264084e-05
len=1024 type="slow"                  4.81 %             2.634609e-01
len=1024 type="slow-allocUnsafe"      1.28 %             7.864850e-01
len=10 type="buffer()"               59.42 %         *** 9.953552e-13
len=10 type="fast-alloc"             -6.43 %             1.450524e-01
len=10 type="fast-alloc-fill"        -2.96 %             4.873766e-01
len=10 type="fast-allocUnsafe"       33.89 %         *** 6.517268e-07
len=10 type="slow"                   -1.48 %             7.357711e-01
len=10 type="slow-allocUnsafe"        0.04 %             9.939576e-01
len=2048 type="buffer()"             36.34 %         *** 3.201045e-10
len=2048 type="fast-alloc"           -4.67 %             2.172900e-01
len=2048 type="fast-alloc-fill"      -0.15 %             9.732945e-01
len=2048 type="fast-allocUnsafe"     20.13 %         *** 2.372115e-04
len=2048 type="slow"                  4.35 %             2.831340e-01
len=2048 type="slow-allocUnsafe"      1.13 %             8.055388e-01
len=4096 type="buffer()"              4.90 %             2.495340e-01
len=4096 type="fast-alloc"           -2.11 %             5.417520e-01
len=4096 type="fast-alloc-fill"      -0.29 %             9.460378e-01
len=4096 type="fast-allocUnsafe"      3.11 %             5.001959e-01
len=4096 type="slow"                  0.95 %             8.145888e-01
len=4096 type="slow-allocUnsafe"      3.74 %             4.227627e-01
len=8192 type="buffer()"              5.08 %             2.263029e-01
len=8192 type="fast-alloc"           -1.16 %             7.300235e-01
len=8192 type="fast-alloc-fill"       0.40 %             9.179919e-01
len=8192 type="fast-allocUnsafe"      5.16 %             2.591544e-01
len=8192 type="slow"                  2.57 %             5.212449e-01
len=8192 type="slow-allocUnsafe"     -3.19 %             4.699138e-01

(raw data)

/cc @nodejs/buffer

Merge request reports