Skip to content

buffer: improve Buffer.from performance

Rodrigo Muino Tomonari requested to merge github/fork/apapirovski/patch-buffer-from-perf into master

Using == null in code paths that are expected to mostly receive objects, arrays or other more complex data types is not great because typecasting these types is very slow. Change to instead check === null || === undefined.

Here's the benchmark:

buffers/buffer-from.js n=2048 len=10 source="array"                     7.26 %        *** 4.653123e-04
buffers/buffer-from.js n=2048 len=10 source="arraybuffer-middle"        2.50 %            1.937961e-01
buffers/buffer-from.js n=2048 len=10 source="arraybuffer"               0.57 %            7.604131e-01
buffers/buffer-from.js n=2048 len=10 source="buffer"                    4.23 %            5.768482e-02
buffers/buffer-from.js n=2048 len=10 source="object"                   16.31 %        *** 1.128762e-09
buffers/buffer-from.js n=2048 len=10 source="string-base64"             0.08 %            9.655769e-01
buffers/buffer-from.js n=2048 len=10 source="string-utf8"              -1.90 %            2.904045e-01
buffers/buffer-from.js n=2048 len=10 source="string"                   -0.80 %            6.723524e-01
buffers/buffer-from.js n=2048 len=10 source="uint8array"                3.63 %            5.416784e-02
buffers/buffer-from.js n=2048 len=2048 source="array"                   1.42 %            4.821207e-01
buffers/buffer-from.js n=2048 len=2048 source="arraybuffer-middle"     -1.04 %            7.025216e-01
buffers/buffer-from.js n=2048 len=2048 source="arraybuffer"             0.36 %            8.475033e-01
buffers/buffer-from.js n=2048 len=2048 source="buffer"                  0.99 %            7.272932e-01
buffers/buffer-from.js n=2048 len=2048 source="object"                 18.59 %        *** 1.131309e-11
buffers/buffer-from.js n=2048 len=2048 source="string-base64"          -0.17 %            9.388870e-01
buffers/buffer-from.js n=2048 len=2048 source="string-utf8"            -0.35 %            8.910105e-01
buffers/buffer-from.js n=2048 len=2048 source="string"                  0.30 %            9.041955e-01
buffers/buffer-from.js n=2048 len=2048 source="uint8array"              0.41 %            8.854981e-01

and here's the old vs old, just to show that the string slowdown is probably just within margin of error:

buffers/buffer-from.js n=2048 len=10 source="array"                    -0.27 %            0.9122062
buffers/buffer-from.js n=2048 len=10 source="arraybuffer-middle"       -0.50 %            0.8118019
buffers/buffer-from.js n=2048 len=10 source="arraybuffer"               1.77 %            0.5688189
buffers/buffer-from.js n=2048 len=10 source="buffer"                    0.98 %            0.7024176
buffers/buffer-from.js n=2048 len=10 source="object"                    0.92 %            0.7081286
buffers/buffer-from.js n=2048 len=10 source="string-base64"            -0.92 %            0.7706629
buffers/buffer-from.js n=2048 len=10 source="string-utf8"              -3.24 %            0.1525157
buffers/buffer-from.js n=2048 len=10 source="string"                    1.52 %            0.4168377
buffers/buffer-from.js n=2048 len=10 source="uint8array"               -0.42 %            0.8494492
buffers/buffer-from.js n=2048 len=2048 source="array"                   0.71 %            0.7683624
buffers/buffer-from.js n=2048 len=2048 source="arraybuffer-middle"      1.05 %            0.5941559
buffers/buffer-from.js n=2048 len=2048 source="arraybuffer"             1.96 %            0.4494948
buffers/buffer-from.js n=2048 len=2048 source="buffer"                 -0.57 %            0.7376221
buffers/buffer-from.js n=2048 len=2048 source="object"                 -1.35 %            0.5914752
buffers/buffer-from.js n=2048 len=2048 source="string-base64"           0.93 %            0.7160677
buffers/buffer-from.js n=2048 len=2048 source="string-utf8"             0.32 %            0.8848240
buffers/buffer-from.js n=2048 len=2048 source="string"                  0.77 %            0.7390453
buffers/buffer-from.js n=2048 len=2048 source="uint8array"             -0.40 %            0.8145751

For reference, == vs === is about 10x slower: https://jsperf.com/triple-equals-vs-double-equals/3

Checklist
  • make -j4 test (UNIX), or vcbuild test (Windows) passes
  • commit message follows commit guidelines
Affected core subsystem(s)

buffer

Merge request reports