This PR increases the number of (internal) functions that are given an opportunity to be inlined during grapheme detection.
Benchmarking indicates that this change results in a significant performance improvement (see below).

This PR also replaces bencher with criterion to enable continued performance monitoring.

$ cargo criterion   Compiling unicode-segmentation v1.7.1 (/home/tim/Open/unicode-segmentation)    Finished bench [optimized] target(s) in 2.96sGnuplot not found, using plotters backendBenchmarking graphemes_arabic: Collecting 100 samples in estimated 6.2276 s                                                                            graphemes_arabic         time:   [300.59 us 303.65 us 306.84 us]                        change: [-28.120% -27.089% -26.073%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking graphemes_english: Collecting 100 samples in estimated 5.8875 s                                                                           graphemes_english        time:   [341.92 us 344.80 us 348.53 us]                        change: [-43.442% -42.572% -41.739%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking graphemes_hindi: Collecting 100 samples in estimated 5.4019 s (                                                                           graphemes_hindi          time:   [340.59 us 343.50 us 346.90 us]                        change: [-20.806% -19.540% -18.205%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking grapheme japanese: Collecting 100 samples in estimated 5.2540                                                                            graphemes_japanese       time:   [336.86 us 338.72 us 341.01 us]                        change: [-23.941% -22.756% -21.497%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking graphemes_korean: Collecting 100 samples in estimated 6.8800 s                                                                            graphemes_korean         time:   [648.18 us 653.10 us 659.09 us]                        change: [-18.605% -17.519% -16.332%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking graphemes_mandarin: Collecting 100 samples in estimated 6.0117                                                                            graphemes_mandarin       time:   [215.14 us 217.71 us 220.84 us]                        change: [-29.560% -28.320% -27.072%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking graphemes_russian: Collecting 100 samples in estimated 6.3045 s                                                                           graphemes_russian        time:   [291.04 us 296.19 us 301.67 us]                        change: [-28.801% -27.643% -26.482%] (p = 0.00 < 0.05)                        Performance has improved.Benchmarking graphemes_source_code: Collecting 100 samples in estimated 6.40                                                                           graphemes_source_code    time:   [378.38 us 383.62 us 389.18 us]                        change: [-36.523% -35.751% -35.060%] (p = 0.00 < 0.05)                        Performance has improved.

timClicks added3 commits

June 2, 2021 15:41

Replace bencher with criterion

a08c4e5

Enable performance improvements to be tracked over time moreeasily.

Increase inlining

86509ee

Make bench output same as function names

9310f0f

Lokathor reviewed

Jun 2, 2021

View reviewed changes

src/grapheme.rs

		@@ -482,6 +488,7 @@ impl GraphemeCursor {
		self.state = GraphemeState::Emoji;
		}

		#[inline]

Copy link

LokathorJun 2, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

this should probably go below the docs

Copy link

ContributorAuthor

timClicksJun 2, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

Thanks. I thought I saw this pattern somewhere else in the crate and decided to follow. Will submit a follow-up PR that moves all of them below the docs.