Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Use memchr for string lexing #9888

Merged
merged 1 commit into from
Feb 8, 2024
Merged

Use memchr for string lexing #9888

merged 1 commit into from
Feb 8, 2024

Conversation

charliermarsh
Copy link
Member

Summary

On main, string lexing consists of walking through the string character-by-character to search for the closing quote (with some nuance: we also need to skip escaped characters, and error if we see newlines in non-triple-quoted strings). This PR rewrites lex_string to instead use memchr to search for the closing quote, which is significantly faster. On my machine, at least, the globals.py benchmark (which contains a lot of docstrings) gets 40% faster...

lexer/numpy/globals.py  time:   [3.6410 µs 3.6496 µs 3.6585 µs]
                        thrpt:  [806.53 MiB/s 808.49 MiB/s 810.41 MiB/s]
                 change:
                        time:   [-40.413% -40.185% -39.984%] (p = 0.00 < 0.05)
                        thrpt:  [+66.623% +67.181% +67.822%]
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild
lexer/unicode/pypinyin.py
                        time:   [12.422 µs 12.445 µs 12.467 µs]
                        thrpt:  [337.03 MiB/s 337.65 MiB/s 338.27 MiB/s]
                 change:
                        time:   [-9.4213% -9.1930% -8.9586%] (p = 0.00 < 0.05)
                        thrpt:  [+9.8401% +10.124% +10.401%]
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  1 (1.00%) high mild
  2 (2.00%) high severe
lexer/pydantic/types.py time:   [107.45 µs 107.50 µs 107.56 µs]
                        thrpt:  [237.11 MiB/s 237.24 MiB/s 237.35 MiB/s]
                 change:
                        time:   [-4.0108% -3.7005% -3.3787%] (p = 0.00 < 0.05)
                        thrpt:  [+3.4968% +3.8427% +4.1784%]
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  2 (2.00%) high mild
  5 (5.00%) high severe
lexer/numpy/ctypeslib.py
                        time:   [46.123 µs 46.165 µs 46.208 µs]
                        thrpt:  [360.36 MiB/s 360.69 MiB/s 361.01 MiB/s]
                 change:
                        time:   [-19.313% -18.996% -18.710%] (p = 0.00 < 0.05)
                        thrpt:  [+23.016% +23.451% +23.935%]
                        Performance has improved.
Found 8 outliers among 100 measurements (8.00%)
  3 (3.00%) low mild
  1 (1.00%) high mild
  4 (4.00%) high severe
lexer/large/dataset.py  time:   [231.07 µs 231.19 µs 231.33 µs]
                        thrpt:  [175.87 MiB/s 175.97 MiB/s 176.06 MiB/s]
                 change:
                        time:   [-2.0437% -1.7663% -1.4922%] (p = 0.00 < 0.05)
                        thrpt:  [+1.5148% +1.7981% +2.0864%]
                        Performance has improved.
Found 10 outliers among 100 measurements (10.00%)
  5 (5.00%) high mild
  5 (5.00%) high severe

@charliermarsh charliermarsh added the performance Potential performance improvement label Feb 8, 2024
Copy link

codspeed-hq bot commented Feb 8, 2024

CodSpeed Performance Report

Merging #9888 will improve performances by 40.35%

Comparing charlie/l (4b4b298) with main (ad313b9)

Summary

⚡ 4 improvements
✅ 26 untouched benchmarks

Benchmarks breakdown

Benchmark main charlie/l Change
lexer[unicode/pypinyin.py] 571.5 µs 543.4 µs +5.18%
lexer[numpy/ctypeslib.py] 1.8 ms 1.7 ms +11.64%
lexer[numpy/globals.py] 223.5 µs 159.3 µs +40.35%
parser[numpy/globals.py] 1.1 ms 1.1 ms +6.07%

Copy link
Contributor

github-actions bot commented Feb 8, 2024

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

Copy link
Member

@MichaReiser MichaReiser left a comment

Choose a reason for hiding this comment

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

Nice!

Copy link
Member

@MichaReiser MichaReiser left a comment

Choose a reason for hiding this comment

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

Nice!

Copy link
Member

@BurntSushi BurntSushi left a comment

Choose a reason for hiding this comment

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

Nice, I like it!

crates/ruff_python_parser/src/lexer.rs Outdated Show resolved Hide resolved
let Some(index) = memchr::memchr3(
quote as u8,
'\r' as u8,
'\n' as u8,
Copy link
Member

Choose a reason for hiding this comment

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

Same as above for quote (or just do let quote_byte = u8::try_from(quote).unwrap() at the top of the function once). But for '\r' as u8 and '\n' as u8, you can just write b'\r' and b'\n', respectively.

// For non-triple-quoted strings, scan until we find the closing quote, but end early
// if we encounter a newline or the end of the file.
loop {
let Some(index) = memchr::memchr3(
Copy link
Member

Choose a reason for hiding this comment

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

Ah I see now why you were asking about memchr3. :P

@charliermarsh charliermarsh enabled auto-merge (squash) February 8, 2024 17:17
@charliermarsh charliermarsh merged commit 6fffde7 into main Feb 8, 2024
16 checks passed
@charliermarsh charliermarsh deleted the charlie/l branch February 8, 2024 17:23
nkxxll pushed a commit to nkxxll/ruff that referenced this pull request Mar 10, 2024
## Summary

On `main`, string lexing consists of walking through the string
character-by-character to search for the closing quote (with some
nuance: we also need to skip escaped characters, and error if we see
newlines in non-triple-quoted strings). This PR rewrites `lex_string` to
instead use `memchr` to search for the closing quote, which is
significantly faster. On my machine, at least, the `globals.py`
benchmark (which contains a lot of docstrings) gets 40% faster...

```text
lexer/numpy/globals.py  time:   [3.6410 µs 3.6496 µs 3.6585 µs]
                        thrpt:  [806.53 MiB/s 808.49 MiB/s 810.41 MiB/s]
                 change:
                        time:   [-40.413% -40.185% -39.984%] (p = 0.00 < 0.05)
                        thrpt:  [+66.623% +67.181% +67.822%]
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild
lexer/unicode/pypinyin.py
                        time:   [12.422 µs 12.445 µs 12.467 µs]
                        thrpt:  [337.03 MiB/s 337.65 MiB/s 338.27 MiB/s]
                 change:
                        time:   [-9.4213% -9.1930% -8.9586%] (p = 0.00 < 0.05)
                        thrpt:  [+9.8401% +10.124% +10.401%]
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  1 (1.00%) high mild
  2 (2.00%) high severe
lexer/pydantic/types.py time:   [107.45 µs 107.50 µs 107.56 µs]
                        thrpt:  [237.11 MiB/s 237.24 MiB/s 237.35 MiB/s]
                 change:
                        time:   [-4.0108% -3.7005% -3.3787%] (p = 0.00 < 0.05)
                        thrpt:  [+3.4968% +3.8427% +4.1784%]
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  2 (2.00%) high mild
  5 (5.00%) high severe
lexer/numpy/ctypeslib.py
                        time:   [46.123 µs 46.165 µs 46.208 µs]
                        thrpt:  [360.36 MiB/s 360.69 MiB/s 361.01 MiB/s]
                 change:
                        time:   [-19.313% -18.996% -18.710%] (p = 0.00 < 0.05)
                        thrpt:  [+23.016% +23.451% +23.935%]
                        Performance has improved.
Found 8 outliers among 100 measurements (8.00%)
  3 (3.00%) low mild
  1 (1.00%) high mild
  4 (4.00%) high severe
lexer/large/dataset.py  time:   [231.07 µs 231.19 µs 231.33 µs]
                        thrpt:  [175.87 MiB/s 175.97 MiB/s 176.06 MiB/s]
                 change:
                        time:   [-2.0437% -1.7663% -1.4922%] (p = 0.00 < 0.05)
                        thrpt:  [+1.5148% +1.7981% +2.0864%]
                        Performance has improved.
Found 10 outliers among 100 measurements (10.00%)
  5 (5.00%) high mild
  5 (5.00%) high severe
```
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
performance Potential performance improvement
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants