8000 Modify benchmarks to compare against stdlib functions · unicode-rs/unicode-segmentation@d6a6eb8 · GitHub
[go: up one dir, main page]
Skip to content

Commit d6a6eb8

Browse files
Jonas-HeinrichJonas-Heinrich
committed
Modify benchmarks to compare against stdlib functions
This commit refactors and expands the microbenchmarks in order to evaluate the performance hit of handling full unicode. It is expected that `unicode-segmentation`'s functions are slower since they consider graphemes, the question is just how much. - bump criterion dependency - rename benchmarks to remove unicode/grapheme relationship - move benchmarks into benchmark group - add scalar versions with stdlib "equivalents" (scalars)
1 parent baab923 commit d6a6eb8

File tree

6 files changed

+154
-184
lines changed

6 files changed

+154
-184
lines changed

‎Cargo.toml

Lines changed: 3 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -23,17 +23,16 @@ no_std = [] # This is a no-op, preserved for backward compatibility only.
2323

2424
[dev-dependencies]
2525
quickcheck = "0.7"
26-
criterion = "0.3"
26+
criterion = "0.5"
2727

2828
[[bench]]
29-
name = "graphemes"
29+
name = "chars"
3030
harness = false
3131

3232
[[bench]]
33-
name = "unicode_words"
33+
name = "words"
3434
harness = false
3535

3636
[[bench]]
3737
name = "word_bounds"
3838
harness = false
39-

‎benches/chars.rs

Lines changed: 60 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,60 @@
1+
//! Compares the performance of `UnicodeSegmentation::graphemes` with stdlib's UTF-8 scalar-based
2+
//! `std::str::chars`.
3+
//!
4+
//! It is expected that `std::str::chars` is faster than `UnicodeSegmentation::graphemes` since it
5+
//! does not consider the complexity of grapheme clusters. The question in this benchmark
6+
//! is how much slower full unicode handling is.
7+
8+
use criterion::{black_box, criterion_group, criterion_main, BenchmarkId, Criterion};
9+
use unicode_segmentation;
10+
11+
use std::fs;
12+
use unicode_segmentation::UnicodeSegmentation;
13+
14+
const FILES: &[&str] = &[
15+
"arabic",
16+
"english",
17+
"hindi",
18+
"japanese",
19+
"korean",
20+
"mandarin",
21+
"russian",
22+
"source_code",
23+
];
24+
25+
#[inline(always)]
26+
fn grapheme(text: &str) {
27+
for c in UnicodeSegmentation::graphemes(black_box(&*text), true) {
28+
black_box(c);
29+
}
30+
}
31+
32+
#[inline(always)]
33+
fn scalar(text: &str) {
34+
for c in black_box(&*text).chars() {
35+
black_box(c);
36+
}
37+
}
38+
39+
fn bench_all(c: &mut Criterion) {
40+
let mut group = c.benchmark_group("chars");
41+
42+
for file in FILES {
43+
group.bench_with_input(
44+
BenchmarkId::new("grapheme", file),
45+
&fs::read_to_string(format!("benches/texts/{}.txt", file)).unwrap(),
46+
|b, content| b.iter(|| grapheme(content)),
47+
);
48+
}
49+
50+
for file in FILES {
51+
group.bench_with_input(
52+
BenchmarkId::new("scalar", file),
53+
&fs::read_to_string(format!("benches/texts/{}.txt", file)).unwrap(),
54+
|b, content| b.iter(|| scalar(content)),
55+
);
56+
}
57+
}
58+
59+
criterion_group!(benches, bench_all);
60+
criterion_main!(benches);

‎benches/graphemes.rs

Lines changed: 0 additions & 63 deletions
This file was deleted.

‎benches/unicode_words.rs

Lines changed: 0 additions & 61 deletions
This file was deleted.

‎benches/word_bounds.rs

Lines changed: 32 additions & 56 deletions
Original file line numberDiff line numberDiff line change
@@ -1,61 +1,37 @@
1-
use criterion::{black_box, criterion_group, criterion_main, Criterion};
1+
use criterion::{black_box, criterion_group, criterion_main, BenchmarkId, Criterion};
22

33
use std::fs;
44
use unicode_segmentation::UnicodeSegmentation;
55

6-
fn word_bounds(c: &mut Criterion, lang: &str, path: &str) {
7-
let text = fs::read_to_string(path).unwrap();
8-
c.bench_function(&format!("word_bounds_{}", lang), |bench| {
9-
bench.iter(|| {
10-
for w in text.split_word_bounds() {
11-
black_box(w);
12-
}
13-
});
14-
});
15-
}
16-
17-
fn word_bounds_arabic(c: &mut Criterion) {
18-
word_bounds(c, "arabic", "benches/texts/arabic.txt");
19-
}
20-
21-
fn word_bounds_english(c: &mut Criterion) {
22-
word_bounds(c, "english", "benches/texts/english.txt");
23-
}
24-
25-
fn word_bounds_hindi(c: &mut Criterion) {
26-
word_bounds(c, "hindi", "benches/texts/hindi.txt");
27-
}
28-
29-
fn word_bounds_japanese(c: &mut Criterion) {
30-
word_bounds(c, "japanese", "benches/texts/japanese.txt");
31-
}
32-
33-
fn word_bounds_korean(c: &mut Criterion) {
34-
word_bounds(c, "korean", "benches/texts/korean.txt");
35-
}
36-
37-
fn word_bounds_mandarin(c: &mut Criterion) {
38-
word_bounds(c, "mandarin", "benches/texts/mandarin.txt");
39-
}
40-
41-
fn word_bounds_russian(c: &mut Criterion) {
42-
word_bounds(c, "russian", "benches/texts/russian.txt");
43-
}
44-
45-
fn word_bounds_source_code(c: &mut Criterion) {
46-
word_bounds(c, "source_code", "benches/texts/source_code.txt");
47-
}
48-
49-
criterion_group!(
50-
benches,
51-
word_bounds_arabic,
52-
word_bounds_english,
53-
word_bounds_hindi,
54-
word_bounds_japanese,
55-
word_bounds_korean,
56-
word_bounds_mandarin,
57-
word_bounds_russian,
58-
word_bounds_source_code,
59-
);
60-
6+
const FILES: &[&str] = &[
7+
"arabic",
8+
"english",
9+
"hindi",
10+
"japanese",
11+
"korean",
12+
"mandarin",
13+
"russian",
14+
"source_code",
15+
];
16+
17+
#[inline(always)]
18+
fn grapheme(text: &str) {
19+
for w in text.split_word_bounds() {
20+
black_box(w);
21+
}
22+
}
23+
24+
fn bench_all(c: &mut Criterion) {
25+
let mut group = c.benchmark_group("word_bounds");
26+
27+
for file in FILES {
28+
group.bench_with_input(
29+
BenchmarkId::new("grapheme", file),
30+
&fs::read_to_string(format!("benches/texts/{}.txt", file)).unwrap(),
31+
|b, content| b.iter(|| grapheme(content)),
32+
);
33+
}
34+
}
35+
36+
criterion_group!(benches, bench_all);
6137
criterion_main!(benches);

‎benches/words.rs

Lines changed: 59 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,59 @@
1+
//! Compares the performance of `UnicodeSegmentation::unicode_words` with stdlib's UTF-8
2+
//! scalar-based `std::str::split_whitespace`.
3+
//!
4+
//! It is expected that `std::str::split_whitespace` is faster than
5+
//! `UnicodeSegmentation::unicode_words` since it does not consider the complexity of grapheme
6+
//! clusters. The question in this benchmark is how much slower full unicode handling is.
7+
8+
use criterion::{black_box, criterion_group, criterion_main, BenchmarkId, Criterion};
9+
10+
use std::fs;
11+
use unicode_segmentation::UnicodeSegmentation;
12+
13+
const FILES: &[&str] = &[
14+
"arabic",
15+
"english",
16+
"hindi",
17+
"japanese",
18+
"korean",
19+
"mandarin",
20+
"russian",
21+
"source_code",
22+
];
23+
24+
#[inline(always)]
25+
fn grapheme(text: &str) {
26+
for w in text.unicode_words() {
27+
black_box(w);
28+
}
29+
}
30+
31+
#[inline(always)]
32+
fn scalar(text: &str) {
33+
for w in text.split_whitespace() {
34+
black_box(w);
35+
}
36+
}
37+
38+
fn bench_all(c: &mut Criterion) {
39+
let mut group = c.benchmark_group("words");
40+
41+
for file in FILES {
42+
group.bench_with_input(
43+
BenchmarkId::new("grapheme", file),
44+
&fs::read_to_string(format!("benches/texts/{}.txt", file)).unwrap(),
45+
|b, content| b.iter(|| grapheme(content)),
46+
);
47+
}
48+
49+
for file in FILES {
50+
group.bench_with_input(
51+
BenchmarkId::new("scalar", file),
52+
&fs::read_to_string(format!("benches/texts/{}.txt", file)).unwrap(),
53+
|b, content| b.iter(|| scalar(content)),
54+
);
55+
}
56+
}
57+
58+
criterion_group!(benches, bench_all);
59+
criterion_main!(benches);

0 commit comments

Comments
 (0)
0