Use memchr to speed up [u8]::contains 3x #46713

Manishearth · 2017-12-13T16:01:11Z

No description provided.

Manishearth · 2017-12-13T16:14:48Z

related: #46704

#![feature(test)]

extern crate test;
use test::Bencher;

#[bench]
fn find_byte(b: &mut Bencher) {
    let x = test::black_box("Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged. It was popularised in the 1960s with the release of Letraset sheets containing Lorem Ipsum passages, and more recently with desktop publishing software like Aldus PageMaker including versions of Lorem Ipsum.").as_bytes();
    b.iter(|| test::black_box(x.contains(&0xff)));
}

#[bench]
fn find_byte_old(b: &mut Bencher) {
    let x = test::black_box("Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum has been the industry's standard dummy text ever since the 1500s, when an unknown printer took a galley of type and scrambled it to make a type specimen book. It has survived not only five centuries, but also the leap into electronic typesetting, remaining essentially unchanged. It was popularised in the 1960s with the release of Letraset sheets containing Lorem Ipsum passages, and more recently with desktop publishing software like Aldus PageMaker including versions of Lorem Ipsum.").as_bytes();
    b.iter(|| test::black_box(x.iter().any(|x| *x == 0xff)));
}

running 2 tests
test find_byte  ... bench:          58 ns/iter (+/- 26)
test find_byte_old ... bench:         159 ns/iter (+/- 29)

test result: ok. 0 passed; 0 failed; 0 ignored; 2 measured; 0 filtered out

Apparently our memchr implementation is actually slower than glibc, but @mystor is looking into that.

I also get similar numbers if I implement find_byte_old as something using [U8]::contains where U8 is struct U8(u8); so that it forces the fallback contains()

Manishearth · 2017-12-13T16:15:00Z

I'll next be working to speed up #46693

Manishearth · 2017-12-13T16:18:13Z

r? @bluss
cc @alexcrichton

ghost · 2017-12-13T16:32:48Z

src/libcore/slice/memchr.rs

+    x.wrapping_sub(LO_USIZE) & !x & HI_USIZE != 0
+}
+
+#[cfg(target_pointer_width = "32")]


Is this going to work on 16-bit platforms as well? I'm asking because I already broke rustc once by neglecting those. :) #40832

Fixed. It looks like it should work. But idk.

Manishearth · 2017-12-14T21:58:11Z

This PR is contained in #46735 (but feel free to land separately, this is a simpler change)

nagisa · 2017-12-15T11:55:13Z

How implausible it would be to start depending on memchr crate rather than including it verbatim, now that we have ability to depend on crates?

hanna-kruppe · 2017-12-15T12:19:05Z

If I'm up to date, there are no third party crates pulled in by libstd. All the crates.io dependencies are in compiler crates and tools. Not sure if there's any reason for what or if it just never came up before.

Manishearth · 2017-12-15T14:17:03Z

libcore can't depend on other crates; libcore contains most of the machinery that lets you write a crate. Also it can't link to libc, which memchr does. We can use this to only depend on the Rust impl, but the Rust impl requires libcore to exist.

…

-Manish Goregaokar

On Fri, Dec 15, 2017 at 6:19 AM, Robin Kruppe ***@***.***> wrote: If I'm up to date, there are no third party crates pulled in by libstd. All the crates.io dependencies are in compiler crates and tools. Not sure if there's any reason for what or if it just never came up before. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#46713 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABivSIXhWuMRUPak2fVx-Zkd-vWswtJ7ks5tAmPugaJpZM4RAxfl> .

bluss · 2017-12-18T16:09:19Z

src/libstd/sys/windows/memchr.rs

@@ -12,4 +12,4 @@
 // Copyright 2015 Andrew Gallant, bluss and Nicolas Koch

 // Fallback memchr is fastest on windows


Side note, this statement (Fallback memchr is fastest on windows) can be revisited by someone with the right development environment.

bluss · 2017-12-18T16:11:38Z

src/libcore/slice/mod.rs

+    fn slice_contains(&self, x: &[Self]) -> bool {
+        memchr::memchr(*self, x).is_some()
+    }
+}


We could add for i8 here too. Reinterpreting &[i8] as &[u8] is problem free

bluss · 2017-12-18T16:19:14Z

I'm glad to hear that the loopiness of Rust's bootstrapping does have a limit :-)

This PR looks good to me. The drawback is to add an unstable API (memchr function) to core with no intention of stabilizing it, that's normally not so popular.

Manishearth · 2017-12-18T18:13:32Z

This PR looks good to me. The drawback is to add an unstable API (memchr function) to core with no intention of stabilizing it, that's normally not so popular.

We already have a bunch of these APIs, like char_internals, whose sole raison d'être is that core needs that code but std needs to use it too.

Can I land this, r=you? This PR was split out so that it's easy to land independently if we want to.

Manishearth · 2017-12-18T18:18:23Z

Oh, hold on, adding an i8 impl

Manishearth · 2017-12-18T18:23:12Z

Done. r?

@mystor

Use memchr for str::find(char) This is a 10x improvement for searching for characters. This also contains the patches from #46713 . Feel free to land both separately or together. cc @mystor @alexcrichton r? @bluss fixes #46693

bluss

Sorry for being so slow :/ The review comments are about that transmute is a very convenient interface, but we prefer everything but transmute.

bluss · 2017-12-31T14:32:02Z

src/libcore/slice/mod.rs

+
+impl SliceContains for i8 {
+    fn slice_contains(&self, x: &[Self]) -> bool {
+        let byte: u8 = unsafe { mem::transmute::<i8, u8>(*self) };


this should be a cast with as.

bluss · 2017-12-31T14:32:04Z

src/libcore/slice/mod.rs

+impl SliceContains for i8 {
+    fn slice_contains(&self, x: &[Self]) -> bool {
+        let byte: u8 = unsafe { mem::transmute::<i8, u8>(*self) };
+        let bytes: &[u8] = unsafe { mem::transmute::<&[i8], &[u8]>(x) };


This with slice::from_raw_parts

Manishearth · 2017-12-31T14:35:45Z

cast with as is wrong, that is a semantic cast not a bitwise cast, and we need a bitwise cast for the search to happen of the same bits.

hanna-kruppe · 2017-12-31T14:37:37Z

as casts don't do overflow checks, they wrap, so they effectively reinterpret the bits when casting between uN and iN.

Manishearth · 2017-12-31T14:39:57Z

That's also UB iirc.

…

On Dec 31, 2017 8:08 PM, "Robin Kruppe" ***@***.***> wrote: as casts don't do overflow checks so they simply reinterpret the bits when casting between uN and iN. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#46713 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABivSIZAbObu6l9-PdSgVSzKutZiHGzPks5tF5xKgaJpZM4RAxfl> .

hanna-kruppe · 2017-12-31T14:49:56Z

UB?!? No.

Manishearth · 2017-12-31T14:51:45Z

We've had problems with casts being UB which iirc aren't fixed yet. Idk if that's true for this specific cast.

…

On Dec 31, 2017 8:20 PM, "Robin Kruppe" ***@***.***> wrote: UB?!? No. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#46713 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABivSJvaVlsuSnGl0lbdgVlJoAU2NZNbks5tF58vgaJpZM4RAxfl> .

bluss · 2017-12-31T14:53:13Z

In Rust the u8 to i8 cast using as is equivalent to the transmute in the sense that the input and output are bitwise equivalent.

Manishearth · 2017-12-31T14:54:05Z

Cool. Will fix then.

…

On Dec 31, 2017 8:23 PM, "bluss" ***@***.***> wrote: In Rust the u8 to i8 cast using as is equivalent to the transmute in the sense that the input and output are bitwise equivalent. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#46713 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABivSHMr4LJw4Gtt6S-u3uvHhjxmnfDoks5tF5_ugaJpZM4RAxfl> .

nagisa · 2017-12-31T14:58:12Z

as casts may only be UB for float-to-integer and integer-to-float involving numbers of certain magnitudes. Casting integers of the same bit-width is a bitwise-reinterpretation. Casting to a wider integer will do either sign or zero extension depending on signedness of the source integer. Casting to a less wide integer will truncate the MSB bits.

Manishearth · 2017-12-31T15:05:58Z

Updated

bluss · 2017-12-31T15:09:12Z

@bors r+

bors · 2017-12-31T15:09:13Z

📌 Commit 4ef6847 has been approved by bluss

bors · 2017-12-31T16:38:16Z

⌛ Testing commit 4ef6847 with merge 8c59418...

Use memchr to speed up [u8]::contains 3x None

bors · 2017-12-31T19:24:56Z

☀️ Test successful - status-appveyor, status-travis
Approved by: bluss
Pushing 8c59418 to master...

@mystor

Use memchr for str::find(char) This is a 10x improvement for searching for characters. This also contains the patches from #46713 . Feel free to land both separately or together. cc @mystor @alexcrichton r? @bluss fixes #46693

avl · 2018-01-04T19:18:54Z

Is this, which was mentioned above: "Casting to a less wide integer will truncate the LSB bits." really true? Isn't it the MSB which are truncated? 257u32 as u8 is 1u8, right?

nagisa · 2018-01-04T19:34:22Z

Ah, indeed. Overworked me switches the terms. Happens all the time.

Edited the comment to reflect the reality for future prosperity.

Manishearth added 2 commits December 13, 2017 01:15

Move rust memchr impl to libcore

2bf0df7

Use memchr in [u8]::contains

f8f2888

rust-highfive assigned bluss Dec 13, 2017

Manishearth changed the title ~~Use memchr to speed up [u8]::contains 2x~~ Use memchr to speed up [u8]::contains 3x Dec 13, 2017

Manishearth mentioned this pull request Dec 13, 2017

str::find(char) is slower than it ought ot be #46693

Closed

ghost reviewed Dec 13, 2017

View reviewed changes

Support 16 bit platforms

1d818a4

kennytm added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Dec 13, 2017

Manishearth mentioned this pull request Dec 14, 2017

Use memchr for str::find(char) #46735

Merged

bluss reviewed Dec 18, 2017

View reviewed changes

bluss reviewed Dec 31, 2017

View reviewed changes

Use memchr for [i8]::contains as well

4ef6847

Manishearth force-pushed the memchr branch from b9200d6 to 4ef6847 Compare December 31, 2017 15:05

kennytm added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Dec 31, 2017

bors added a commit that referenced this pull request Dec 31, 2017

Auto merge of #46713 - Manishearth:memchr, r=bluss

8c59418

Use memchr to speed up [u8]::contains 3x None

bors mentioned this pull request Dec 31, 2017

Improved error messages for linking failure #47052

Merged

bors merged commit 4ef6847 into rust-lang:master Dec 31, 2017

Manishearth deleted the memchr branch January 1, 2018 04:34

varkor mentioned this pull request Jan 4, 2018

Generalise slice::contains over PartialEq #46934

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use memchr to speed up [u8]::contains 3x #46713

Use memchr to speed up [u8]::contains 3x #46713

Manishearth commented Dec 13, 2017

Manishearth commented Dec 13, 2017 •

edited

Loading

Manishearth commented Dec 13, 2017

Manishearth commented Dec 13, 2017

ghost Dec 13, 2017 •

edited by ghost

Loading

Manishearth Dec 13, 2017

Manishearth commented Dec 14, 2017

nagisa commented Dec 15, 2017

hanna-kruppe commented Dec 15, 2017

Manishearth commented Dec 15, 2017 via email

bluss Dec 18, 2017

bluss Dec 18, 2017

bluss commented Dec 18, 2017 •

edited

Loading

Manishearth commented Dec 18, 2017 •

edited

Loading

Manishearth commented Dec 18, 2017

Manishearth commented Dec 18, 2017

bluss left a comment

bluss Dec 31, 2017

bluss Dec 31, 2017

Manishearth commented Dec 31, 2017

hanna-kruppe commented Dec 31, 2017 •

edited

Loading

Manishearth commented Dec 31, 2017 via email

hanna-kruppe commented Dec 31, 2017

Manishearth commented Dec 31, 2017 via email

bluss commented Dec 31, 2017

Manishearth commented Dec 31, 2017 via email

nagisa commented Dec 31, 2017 •

edited

Loading

Manishearth commented Dec 31, 2017

bluss commented Dec 31, 2017

bors commented Dec 31, 2017

bors commented Dec 31, 2017

bors commented Dec 31, 2017

avl commented Jan 4, 2018

nagisa commented Jan 4, 2018

		@@ -12,4 +12,4 @@
		// Copyright 2015 Andrew Gallant, bluss and Nicolas Koch

		// Fallback memchr is fastest on windows

Use memchr to speed up [u8]::contains 3x #46713

Use memchr to speed up [u8]::contains 3x #46713

Conversation

Manishearth commented Dec 13, 2017

Manishearth commented Dec 13, 2017 • edited Loading

Manishearth commented Dec 13, 2017

Manishearth commented Dec 13, 2017

ghost Dec 13, 2017 • edited by ghost Loading

Choose a reason for hiding this comment

Manishearth Dec 13, 2017

Choose a reason for hiding this comment

Manishearth commented Dec 14, 2017

nagisa commented Dec 15, 2017

hanna-kruppe commented Dec 15, 2017

Manishearth commented Dec 15, 2017 via email

bluss Dec 18, 2017

Choose a reason for hiding this comment

bluss Dec 18, 2017

Choose a reason for hiding this comment

bluss commented Dec 18, 2017 • edited Loading

Manishearth commented Dec 18, 2017 • edited Loading

Manishearth commented Dec 18, 2017

Manishearth commented Dec 18, 2017

bluss left a comment

Choose a reason for hiding this comment

bluss Dec 31, 2017

Choose a reason for hiding this comment

bluss Dec 31, 2017

Choose a reason for hiding this comment

Manishearth commented Dec 31, 2017

hanna-kruppe commented Dec 31, 2017 • edited Loading

Manishearth commented Dec 31, 2017 via email

hanna-kruppe commented Dec 31, 2017

Manishearth commented Dec 31, 2017 via email

bluss commented Dec 31, 2017

Manishearth commented Dec 31, 2017 via email

nagisa commented Dec 31, 2017 • edited Loading

Manishearth commented Dec 31, 2017

bluss commented Dec 31, 2017

bors commented Dec 31, 2017

bors commented Dec 31, 2017

bors commented Dec 31, 2017

avl commented Jan 4, 2018

nagisa commented Jan 4, 2018

Manishearth commented Dec 13, 2017 •

edited

Loading

ghost Dec 13, 2017 •

edited by ghost

Loading

bluss commented Dec 18, 2017 •

edited

Loading

Manishearth commented Dec 18, 2017 •

edited

Loading

hanna-kruppe commented Dec 31, 2017 •

edited

Loading

nagisa commented Dec 31, 2017 •

edited

Loading