This function was updated in a recent PR (92911) to be called without the caller
information passed in, but the function signature itself was not altered with
cfg_attr at the time.
Stop manually SIMDing in `swap_nonoverlapping`
Like I previously did for `reverse` (#90821), this leaves it to LLVM to pick how to vectorize it, since it can know better the chunk size to use, compared to the "32 bytes always" approach we currently have.
A variety of codegen tests are included to confirm that the various cases are still being vectorized.
It does still need logic to type-erase in some cases, though, as while LLVM is now smart enough to vectorize over slices of things like `[u8; 4]`, it fails to do so over slices of `[u8; 3]`.
As a bonus, this change also means one no longer gets the spurious `memcpy`(s?) at the end up swapping a slice of `__m256`s: <https://rust.godbolt.org/z/joofr4v8Y>
<details>
<summary>ASM for this example</summary>
## Before (from godbolt)
note the `push`/`pop`s and `memcpy`
```x86
swap_m256_slice:
push r15
push r14
push r13
push r12
push rbx
sub rsp, 32
cmp rsi, rcx
jne .LBB0_6
mov r14, rsi
shl r14, 5
je .LBB0_6
mov r15, rdx
mov rbx, rdi
xor eax, eax
.LBB0_3:
mov rcx, rax
vmovaps ymm0, ymmword ptr [rbx + rax]
vmovaps ymm1, ymmword ptr [r15 + rax]
vmovaps ymmword ptr [rbx + rax], ymm1
vmovaps ymmword ptr [r15 + rax], ymm0
add rax, 32
add rcx, 64
cmp rcx, r14
jbe .LBB0_3
sub r14, rax
jbe .LBB0_6
add rbx, rax
add r15, rax
mov r12, rsp
mov r13, qword ptr [rip + memcpy@GOTPCREL]
mov rdi, r12
mov rsi, rbx
mov rdx, r14
vzeroupper
call r13
mov rdi, rbx
mov rsi, r15
mov rdx, r14
call r13
mov rdi, r15
mov rsi, r12
mov rdx, r14
call r13
.LBB0_6:
add rsp, 32
pop rbx
pop r12
pop r13
pop r14
pop r15
vzeroupper
ret
```
## After (from my machine)
Note no `rsp` manipulation, sorry for different ASM syntax
```x86
swap_m256_slice:
cmpq %r9, %rdx
jne .LBB1_6
testq %rdx, %rdx
je .LBB1_6
cmpq $1, %rdx
jne .LBB1_7
xorl %r10d, %r10d
jmp .LBB1_4
.LBB1_7:
movq %rdx, %r9
andq $-2, %r9
movl $32, %eax
xorl %r10d, %r10d
.p2align 4, 0x90
.LBB1_8:
vmovaps -32(%rcx,%rax), %ymm0
vmovaps -32(%r8,%rax), %ymm1
vmovaps %ymm1, -32(%rcx,%rax)
vmovaps %ymm0, -32(%r8,%rax)
vmovaps (%rcx,%rax), %ymm0
vmovaps (%r8,%rax), %ymm1
vmovaps %ymm1, (%rcx,%rax)
vmovaps %ymm0, (%r8,%rax)
addq $2, %r10
addq $64, %rax
cmpq %r10, %r9
jne .LBB1_8
.LBB1_4:
testb $1, %dl
je .LBB1_6
shlq $5, %r10
vmovaps (%rcx,%r10), %ymm0
vmovaps (%r8,%r10), %ymm1
vmovaps %ymm1, (%rcx,%r10)
vmovaps %ymm0, (%r8,%r10)
.LBB1_6:
vzeroupper
retq
```
</details>
This does all its copying operations as either the original type or as `MaybeUninit`s, so as far as I know there should be no potential abstract machine issues with reading padding bytes as integers.
<details>
<summary>Perf is essentially unchanged</summary>
Though perhaps with more target features this would help more, if it could pick bigger chunks
## Before
```
running 10 tests
test slice::swap_with_slice_4x_usize_30 ... bench: 894 ns/iter (+/- 11)
test slice::swap_with_slice_4x_usize_3000 ... bench: 99,476 ns/iter (+/- 2,784)
test slice::swap_with_slice_5x_usize_30 ... bench: 1,257 ns/iter (+/- 7)
test slice::swap_with_slice_5x_usize_3000 ... bench: 139,922 ns/iter (+/- 959)
test slice::swap_with_slice_rgb_30 ... bench: 328 ns/iter (+/- 27)
test slice::swap_with_slice_rgb_3000 ... bench: 16,215 ns/iter (+/- 176)
test slice::swap_with_slice_u8_30 ... bench: 312 ns/iter (+/- 9)
test slice::swap_with_slice_u8_3000 ... bench: 5,401 ns/iter (+/- 123)
test slice::swap_with_slice_usize_30 ... bench: 368 ns/iter (+/- 3)
test slice::swap_with_slice_usize_3000 ... bench: 28,472 ns/iter (+/- 3,913)
```
## After
```
running 10 tests
test slice::swap_with_slice_4x_usize_30 ... bench: 868 ns/iter (+/- 36)
test slice::swap_with_slice_4x_usize_3000 ... bench: 99,642 ns/iter (+/- 1,507)
test slice::swap_with_slice_5x_usize_30 ... bench: 1,194 ns/iter (+/- 11)
test slice::swap_with_slice_5x_usize_3000 ... bench: 139,761 ns/iter (+/- 5,018)
test slice::swap_with_slice_rgb_30 ... bench: 324 ns/iter (+/- 6)
test slice::swap_with_slice_rgb_3000 ... bench: 15,962 ns/iter (+/- 287)
test slice::swap_with_slice_u8_30 ... bench: 281 ns/iter (+/- 5)
test slice::swap_with_slice_u8_3000 ... bench: 5,324 ns/iter (+/- 40)
test slice::swap_with_slice_usize_30 ... bench: 275 ns/iter (+/- 5)
test slice::swap_with_slice_usize_3000 ... bench: 28,277 ns/iter (+/- 277)
```
</detail>
remove feature gate in control_flow examples
Stabilization was done in https://github.com/rust-lang/rust/pull/91091, but the two examples weren't updated accordingly.
Probably too late to put it into stable, but it should be in the next release :)
Some improvements to the async docs
The goal here is to make the docs overall a little bit more comprehensive and add more links between the things.
One thing that's not working yet is the links to the keywords. Somehow I couldn't get them to work.
r? ````@GuillaumeGomez```` do you know how I could get the keyword links to work?
Like I previously did for `reverse`, this leaves it to LLVM to pick how to vectorize it, since it can know better the chunk size to use, compared to the "32 bytes always" approach we currently have.
It does still need logic to type-erase where appropriate, though, as while LLVM is now smart enough to vectorize over slices of things like `[u8; 4]`, it fails to do so over slices of `[u8; 3]`.
As a bonus, this also means one no longer gets the spurious `memcpy`(s?) at the end up swapping a slice of `__m256`s: <https://rust.godbolt.org/z/joofr4v8Y>
core: Implement ASCII trim functions on byte slices
Hi ````````@rust-lang/libs!```````` This is a feature that I wished for when implementing serial protocols with microcontrollers. Often these protocols may contain leading or trailing whitespace, which needs to be removed. Because oftentimes drivers will operate on the byte level, decoding to unicode and checking for unicode whitespace is unnecessary overhead.
This PR adds three new methods to byte slices:
- `trim_ascii_start`
- `trim_ascii_end`
- `trim_ascii`
I did not find any pre-existing discussions about this, which surprises me a bit. Maybe I'm missing something, and this functionality is already possible through other means? There's https://github.com/rust-lang/rfcs/issues/2547 ("Trim methods on slices"), but that has a different purpose.
As per the [std dev guide](https://std-dev-guide.rust-lang.org/feature-lifecycle/new-unstable-features.html), this is a proposed implementation without any issue / RFC. If this is the wrong process, please let me know. However, I thought discussing code is easier than discussing a mere idea, and hacking on the stdlib was fun.
Tracking issue: https://github.com/rust-lang/rust/issues/94035
Guard against unwinding in cleanup code
Currently the only safe guard we have against double unwind is the panic count (which is local to Rust). When double unwinds indeed happen (e.g. C++ exception + Rust panic, or two C++ exceptions), then the second unwind actually goes through and the first unwind is leaked. This can cause UB. cc rust-lang/project-ffi-unwind#6
E.g. given the following C++ code:
```c++
extern "C" void foo() {
throw "A";
}
extern "C" void execute(void (*fn)()) {
try {
fn();
} catch(...) {
}
}
```
This program is well-defined to terminate:
```c++
struct dtor {
~dtor() noexcept(false) {
foo();
}
};
void a() {
dtor a;
dtor b;
}
int main() {
execute(a);
return 0;
}
```
But this Rust code doesn't catch the double unwind:
```rust
extern "C-unwind" {
fn foo();
fn execute(f: unsafe extern "C-unwind" fn());
}
struct Dtor;
impl Drop for Dtor {
fn drop(&mut self) {
unsafe { foo(); }
}
}
extern "C-unwind" fn a() {
let _a = Dtor;
let _b = Dtor;
}
fn main() {
unsafe { execute(a) };
}
```
To address this issue, this PR adds an unwind edge to an abort block, so that the Rust example aborts. This is similar to how clang guards against double unwind (except clang calls terminate per C++ spec and we abort).
The cost should be very small; it's an additional trap instruction (well, two for now, since we use TrapUnreachable, but that's a different issue) for each function with landing pads; if LLVM gains support to encode "abort/terminate" info directly in LSDA like GCC does, then it'll be free. It's an additional basic block though so compile time may be worse, so I'd like a perf run.
r? `@ghost`
`@rustbot` label: F-c_unwind
Add a `try_collect()` helper method to `Iterator`
Implement `Iterator::try_collect()` as a helper around `Iterator::collect()` as discussed [here](https://internals.rust-lang.org/t/idea-fallible-iterator-mapping-with-try-map/15715/5?u=a.lafrance).
First time contributor so definitely open to any feedback about my implementation! Specifically wondering if I should open a tracking issue for the unstable feature I introduced.
As the main participant in the internals discussion: r? `@scottmcm`
Add documentation to more `From::from` implementations.
For users looking at documentation through IDE popups, this gives them relevant information rather than the generic trait documentation wording “Performs the conversion”. For users reading the documentation for a specific type for any reason, this informs them when the conversion may allocate or copy significant memory versus when it is always a move or cheap copy.
Notes on specific cases:
* The new documentation for `From<T> for T` explains that it is not a conversion at all.
* Also documented `impl<T, U> Into<U> for T where U: From<T>`, the other central blanket implementation of conversion.
* The new documentation for construction of maps and sets from arrays of keys mentions the handling of duplicates. Future work could be to do this for *all* code paths that convert an iterable to a map or set.
* I did not add documentation to conversions of a specific error type to a more general error type.
* I did not add documentation to unstable code.
This change was prepared by searching for the text "From<... for" and so may have missed some cases that for whatever reason did not match. I also looked for `Into` impls but did not find any worth documenting by the above criteria.
Destabilize cfg(target_has_atomic_load_store = ...)
This was not intended to be stabilized yet.
This keeps the cfg_target_has_atomic feature gate name since compiler-builtins otherwise depends on it and I'd rather not try to manage a bump across a crates.io published repository given the time-sensitivity here (we need to land this quickly to avoid a beta backport).
Closes https://github.com/rust-lang/rust/issues/32976
r? `@Amanieu`
Make [u8]::cmp implementation branchless
The current implementation generates rather ugly assembly code, branching when the common parts are equal. By performing the comparison of the lengths upfront using a subtraction, the assembly gets much prettier: https://godbolt.org/z/4e5fnEKGd.
This will probably not impact speed too much, as the expensive part is in most cases the `memcmp`, but it sure looks better (I'm porting a sorting algorithm currently, and that branch just bothered me).
Since `decl_macro`s and/or `Span::def_site()` is deemed quite unstable,
no public-facing macro that relies on it can hope to be, itself, stabilized.
We circumvent the issue by no longer relying on field privacy for safety and,
instead, relying on an unstable feature-gate to act as the gate keeper for
non users of the macro (thanks to `allow_internal_unstable`).
This is technically not correct (since a `nightly` user could technically enable
the feature and cause unsoundness with it); or, in other words, this makes the
feature-gate used to gate the access to the field be (technically unsound, and
in practice) `unsafe`. Hence it having `unsafe` in its name.
Back to the macro, we go back to `macro_rules!` / `mixed_site()`-span rules thanks
to declaring the `decl_macro` as `semitransparent`, which is a hack to basically have
`pub macro_rules!`
Co-Authored-By: Mara Bos <m-ou.se@m-ou.se>
add diagnostic items for clippy's `trim_split_whitespace`
Adding the following diagnostic items:
* str_split_whitespace,
* str_trim,
* str_trim_start,
* str_trim_end
They are needed for https://github.com/rust-lang/rust-clippy/pull/8575
r? `@flip1995`
add module-level documentation for vec's in-place iteration
As requested in the last libs team meeting and during previous reviews.
Feel free to point out any gaps you encounter, after all non-obvious things may with hindsight seem obvious to me.
r? `@yaahc`
CC `@steffahn`
Rename `~const Drop` to `~const Destruct`
r? `@oli-obk`
Completely switching to `~const Destructible` would be rather complicated, so it seems best to add it for now and wait for it to be backported to beta in the next release.
The rationale is to prevent complications such as #92149 and #94803 by introducing an entirely new trait. And `~const Destructible` reads a bit better than `~const Drop`. Name Bikesheddable.
Add u16::is_utf16_surrogate
Right now, there are methods in the standard library for encoding and decoding UTF-16, but at least for the moment, there aren't any methods specifically for `u16` to help work with UTF-16 data. Since the full logic already exists, this wouldn't really add any code, just expose what's already there.
This method in particular is useful for working with the data returned by Windows `OsStrExt::encode_wide`. Initially, I was planning to also offer a `TryFrom<u16> for char`, but decided against it for now. There is plenty of code in rustc that could be rewritten to use this method, but I only checked within the standard library to replace them.
I think that offering more UTF-16-related methods to u16 would be useful, but I think this one is a good start. For example, one useful method might be `u16::is_pattern_whitespace`, which would check if something is the Unicode `Pattern_Whitespace` category. We can get away with this because all of the `Pattern_Whitespace` characters are in the basic multilingual plane, and hence we don't need to check for surrogates.
Provide more useful documentation of conversion methods
I thought that the documentation for these methods needed to be a bit more explanatory for new users. For advanced users, the comments are relatively unnecessary. I think it would be useful to explain precisely what the method does. As a new user, when you see the `into` method, where the type is inferred, if you are new you don't even know what you convert to, because it is implicit. I believe this can help new users understand.
I thought that the documentation for these methods needed to be a bit more explanatory for new users. For advanced users, the comments are relatively unnecessary. I think it would be useful to explain precisely what the method does. As a new user, when you see the `into` method, where the type is inferred, if you are new you don't even know what you convert to, because it is implicit. I believe this can help new users understand.
Document that `Option<extern "abi" fn>` discriminant elision applies for any ABI
The current phrasing was not very clear on that aspect.
r? `@RalfJung`
`@rustbot` modify labels: A-docs A-ffi
Derive Eq for std::cmp::Ordering, instead of using manual impl.
This allows consts of type Ordering to be used in patterns, and with feature(adt_const_params) allows using `Ordering` as a const generic parameter.
Currently, `std::cmp::Ordering` implements `Eq` using a manually written `impl Eq for Ordering {}`, instead of `derive(Eq)`. This means that it does not implement `StructuralEq`.
This commit removes the manually written impl, and adds `derive(Eq)` to `Ordering`, so that it will implement `StructuralEq`.
Let `try_collect` take advantage of `try_fold` overrides
No public API changes.
With this change, `try_collect` (#94047) is no longer going through the `impl Iterator for &mut impl Iterator`, and thus will be able to use `try_fold` overrides instead of being forced through `next` for every element.
Here's the test added, to see that it fails before this PR (once a new enough nightly is out): https://play.rust-lang.org/?version=nightly&mode=debug&edition=2021&gist=462f2896f2fed2c238ee63ca1a7e7c56
This might as well go to the same person as my last `try_process` PR (#93572), so
r? ``@yaahc``
Stabilize ADX target feature
This is a continuation of #60109, which noted that while the ADX intrinsics were stabilized, the corresponding target feature never was.
This PR follows the same general structure and stabilizes the ADX target feature.
See also https://github.com/rust-lang/rust/issues/44839 - tracking issue for target feature
Format core and std macro rules, removing needless surrounding blocks
Many of the asserting and printing macros in `core` and `std` are written with prehistoric-looking formatting, like this:
335ffbfa54/library/std/src/macros.rs (L96-L101)
In modern Rust style this would conventionally be written as follows instead, always using braces and a trailing semicolon on the macro arms:
af53809c87/library/std/src/macros.rs (L98-L105)
Getting rid of the unneeded braces inside the expansion reduces extraneous indentation in macro-expanded code. For example:
```rust
println!("repro {}", true);
```
```rust
// before:
{
::std::io::_print(
::core::fmt::Arguments::new_v1(
&["repro ", "\n"],
&[::core::fmt::ArgumentV1::new_display(&true)],
),
);
};
```
```rust
// after:
::std::io::_print(
::core::fmt::Arguments::new_v1(
&["repro ", "\n"],
&[::core::fmt::ArgumentV1::new_display(&true)],
),
);
```
This is a continuation of #60109, which noted that while the ADX
intrinsics were stabilized, the corresponding target feature never was.
This PR follows the same general structure and stabilizes the ADX target
feature.
Add `Atomic*::get_mut_slice`
This PR adds the inverse of `Atomic*::from_mut_slice` introduced in #94384 with the following API:
```rust
// core::sync::atomic
impl Atomic* {
fn get_mut_slice(this: &mut [Self]) -> &mut [*];
}
```
cc `@cuviper`
-----
For now I've used the same tracking issue as `Atomic*::from_mut_slice`, should I open a new one?
Enable conditional checking of values in the Rust codebase
This pull-request enable conditional checking of (well known) values in the Rust codebase.
Well known values were added in https://github.com/rust-lang/rust/pull/94362. All the `target_*` values are taken from all the built-in targets which is why some extra values were needed do be added as they are not (yet ?) defined in any built-in targets.
r? `@Mark-Simulacrum`
Make float parsing docs more comprehensive
I was working on some code with some specialized restrictions on float parsing. I noticed the doc comments for `f32::from_str` and `f64::from_str` were missing several cases of valid inputs that are otherwise difficult to discover without looking at source code.
I'm not sure if the doc comments were initially intended to contain a comprehensive description of valid inputs, but I figured it's useful to include these extra cases for reference.
Rename `IntoFuture::Future` to `IntoFuture::IntoFuture`
Ref: https://github.com/rust-lang/rust/issues/67644#issuecomment-1051401459
This renames `IntoFuture::Future` to `IntoFuture::IntoFuture`. This adds the `Into*` prefix to the associated type, similar to the [`IntoIterator::IntoIter`](https://doc.rust-lang.org/std/iter/trait.IntoIterator.html#associatedtype.IntoIter) associated type. It's my mistake we didn't do so in the first place. This fixes that and brings the two closer together. Thanks!
### References
__`IntoIterator` trait def__
```rust
pub trait IntoIterator {
type Item;
type IntoIter: Iterator<Item = Self::Item>;
fn into_iter(self) -> Self::IntoIter;
}
```
__`IntoFuture` trait def__
```rust
pub trait IntoFuture {
type Output;
type IntoFuture: Future<Output = Self::Output>; // Prior to this PR: `type Future:`
fn into_future(self) -> Self::IntoFuture;
}
```
cc/ `@eholk` `@rust-lang/wg-async`
Stabilise inherent_ascii_escape (FCP in #77174)
Implements #77174, which completed its FCP.
This does *not* deprecate any existing methods or structs, as that is tracked in #93887. That stated, people should prefer using `u8::escape_ascii` to `std::ascii::escape_default`.
More practical examples for `Option::and_then` & `Result::and_then`
To be blatantly honest, I think the current example given for `Option::and_then` is objectively terrible. (No offence to whoever wrote them initially.)
```rust
fn sq(x: u32) -> Option<u32> { Some(x * x) }
fn nope(_: u32) -> Option<u32> { None }
assert_eq!(Some(2).and_then(sq).and_then(sq), Some(16));
assert_eq!(Some(2).and_then(sq).and_then(nope), None);
assert_eq!(Some(2).and_then(nope).and_then(sq), None);
assert_eq!(None.and_then(sq).and_then(sq), None);
```
Current example:
- does not demonstrate that `and_then` converts `Option<T>` to `Option<U>`
- is far removed from any realistic code
- generally just causes more confusion than it helps
So I replaced them with two blocks:
- the first one shows basic usage (including the type conversion)
- the second one shows an example of typical usage
Same thing with `Result::and_then`.
Hopefully this helps with clarity.
Change `ResultShunt` to be generic over `Try`
Just a refactor (and rename) for now, so it's not `Result`-specific.
This could be used for a future `Iterator::try_collect`, or similar, but anything like that is left for a future PR.
Add {floor,ceil}_char_boundary methods to str
This is technically already used internally by the standard library in the form of `truncate_to_char_boundary`.
Essentially these are two building blocks to allow for approximate string truncation, where you want to cut off the string at "approximately" a given length in bytes but don't know exactly where the character boundaries lie. It's also a good candidate for the standard library as it can easily be done naively, but would be difficult to properly optimise. Although the existing code that's done in error messages is done naively, this code will explicitly only check a window of 4 bytes since we know that a boundary must lie in that range, and because it will make it possible to vectorise.
Although this method doesn't take into account graphemes or other properties, this would still be a required building block for splitting that takes those into account. For example, if you wanted to split at a grapheme boundary, you could take your approximate splitting point and then determine the graphemes immediately following and preceeding the split. If you then notice that these two graphemes could be merged, you can decide to either include the whole grapheme or exclude it depending on whether you decide splitting should shrink or expand the string.
This takes the most conservative approach and just offers the raw indices to the user, and they can decide how to use them. That way, the methods are as useful as possible despite having as few methods as possible.
(Note: I'll add some tests and a tracking issue if it's decided that this is worth including.)
Just a refactor (and rename) for now, so it's not `Result`-specific.
This could be used for a future `Iterator::try_collect`, or similar, but anything like that is left for a future PR.
Impl {Add,Sub,Mul,Div,Rem,BitXor,BitOr,BitAnd}Assign<$t> for Wrapping<$t> for rust 1.60.0
Tracking issue #93204
This is about adding basic integer operations to the `Wrapping` type:
```rust
let mut value = Wrapping(2u8);
value += 3u8;
value -= 1u8;
value *= 2u8;
value /= 2u8;
value %= 2u8;
value ^= 255u8;
value |= 123u8;
value &= 2u8;
```
Because this adds stable impls on a stable type, it runs into the following issue if an `#[unstable(...)]` attribute is used:
```
an `#[unstable]` annotation here has no effect
note: see issue #55436 <https://github.com/rust-lang/rust/issues/55436> for more information
```
This means - if I understood this correctly - the new impls have to be stabilized instantly.
Which in turn means, this PR has to kick of an FCP on the tracking issue as well?
This impl is analog to 1c0dc1810d#92356 for the `Saturating` type ``@dtolnay`` ``@Mark-Simulacrum``
Fix invalid special casing of the unreachable! macro
This pull-request fix an invalid special casing of the `unreachable!` macro in the same way the `panic!` macro was solved, by adding two new internal only macros `unreachable_2015` and `unreachable_2021` edition dependent and turn `unreachable!` into a built-in macro that do dispatching. This logic is stolen from the `panic!` macro.
~~This pull-request also adds an internal feature `format_args_capture_non_literal` that allows capturing arguments from formatted string that expanded from macros. The original RFC #2795 mentioned this as a future possibility. This feature is [required](https://github.com/rust-lang/rust/issues/92137#issuecomment-1018630522) because of concatenation that needs to be done inside the macro:~~
```rust
$crate::concat!("internal error: entered unreachable code: ", $fmt)
```
**In summary** the new behavior for the `unreachable!` macro with this pr is:
Edition 2021:
```rust
let x = 5;
unreachable!("x is {x}");
```
```
internal error: entered unreachable code: x is 5
```
Edition <= 2018:
```rust
let x = 5;
unreachable!("x is {x}");
```
```
internal error: entered unreachable code: x is {x}
```
Also note that the change in this PR are **insta-stable** and **breaking changes** but this a considered as being a [bug](https://github.com/rust-lang/rust/issues/92137#issuecomment-998441613).
If someone could start a perf run and then a crater run this would be appreciated.
Fixes https://github.com/rust-lang/rust/issues/92137
Rollup of 2 pull requests
Successful merges:
- #90998 (Require const stability attribute on all stable functions that are `const`)
- #93489 (Mark the panic_no_unwind lang item as nounwind)
Failed merges:
r? `@ghost`
`@rustbot` modify labels: rollup
Mark the panic_no_unwind lang item as nounwind
This has 2 effects:
- It helps LLVM when inlining since it doesn't need to generate landing pads for `panic_no_unwind`.
- It makes it sound for a panic handler to unwind even if `PanicInfo::can_unwind` returns true. This will simply cause another panic once the unwind tries to go past the `panic_no_unwind` lang item. Eventually this will cause a stack overflow, which is safe.
Require const stability attribute on all stable functions that are `const`
This PR requires all stable functions (of all kinds) that are `const fn` to have a `#[rustc_const_stable]` or `#[rustc_const_unstable]` attribute. Stability was previously implied if omitted; a follow-up PR is planned to change the fallback to be unstable.
Optimize `core::str::Chars::count`
I wrote this a while ago after seeing this function as a bottleneck in a profile, but never got around to contributing it. I saw it again, and so here it is. The implementation is fairly complex, but I tried to explain what's happening at both a high level (in the header comment for the file), and in line comments in the impl. Hopefully it's clear enough.
This implementation (`case00_cur_libcore` in the benchmarks below) is somewhat consistently around 4x to 5x faster than the old implementation (`case01_old_libcore` in the benchmarks below), for a wide variety of workloads, without regressing performance on any of the workload sizes I've tried.
I also improved the benchmarks for this code, so that they explicitly check text in different languages and of different sizes (err, the cross product of language x size). The results of the benchmarks are here:
<details>
<summary>Benchmark results</summary>
<pre>
test str::char_count::emoji_huge::case00_cur_libcore ... bench: 20,216 ns/iter (+/- 3,673) = 17931 MB/s
test str::char_count::emoji_huge::case01_old_libcore ... bench: 108,851 ns/iter (+/- 12,777) = 3330 MB/s
test str::char_count::emoji_huge::case02_iter_increment ... bench: 329,502 ns/iter (+/- 4,163) = 1100 MB/s
test str::char_count::emoji_huge::case03_manual_char_len ... bench: 223,333 ns/iter (+/- 14,167) = 1623 MB/s
test str::char_count::emoji_large::case00_cur_libcore ... bench: 293 ns/iter (+/- 6) = 19331 MB/s
test str::char_count::emoji_large::case01_old_libcore ... bench: 1,681 ns/iter (+/- 28) = 3369 MB/s
test str::char_count::emoji_large::case02_iter_increment ... bench: 5,166 ns/iter (+/- 85) = 1096 MB/s
test str::char_count::emoji_large::case03_manual_char_len ... bench: 3,476 ns/iter (+/- 62) = 1629 MB/s
test str::char_count::emoji_medium::case00_cur_libcore ... bench: 48 ns/iter (+/- 0) = 14750 MB/s
test str::char_count::emoji_medium::case01_old_libcore ... bench: 217 ns/iter (+/- 4) = 3262 MB/s
test str::char_count::emoji_medium::case02_iter_increment ... bench: 642 ns/iter (+/- 7) = 1102 MB/s
test str::char_count::emoji_medium::case03_manual_char_len ... bench: 445 ns/iter (+/- 3) = 1591 MB/s
test str::char_count::emoji_small::case00_cur_libcore ... bench: 18 ns/iter (+/- 0) = 3777 MB/s
test str::char_count::emoji_small::case01_old_libcore ... bench: 23 ns/iter (+/- 0) = 2956 MB/s
test str::char_count::emoji_small::case02_iter_increment ... bench: 66 ns/iter (+/- 2) = 1030 MB/s
test str::char_count::emoji_small::case03_manual_char_len ... bench: 29 ns/iter (+/- 1) = 2344 MB/s
test str::char_count::en_huge::case00_cur_libcore ... bench: 25,909 ns/iter (+/- 39,260) = 13299 MB/s
test str::char_count::en_huge::case01_old_libcore ... bench: 102,887 ns/iter (+/- 3,257) = 3349 MB/s
test str::char_count::en_huge::case02_iter_increment ... bench: 166,370 ns/iter (+/- 12,439) = 2071 MB/s
test str::char_count::en_huge::case03_manual_char_len ... bench: 166,332 ns/iter (+/- 4,262) = 2071 MB/s
test str::char_count::en_large::case00_cur_libcore ... bench: 281 ns/iter (+/- 6) = 19160 MB/s
test str::char_count::en_large::case01_old_libcore ... bench: 1,598 ns/iter (+/- 19) = 3369 MB/s
test str::char_count::en_large::case02_iter_increment ... bench: 2,598 ns/iter (+/- 167) = 2072 MB/s
test str::char_count::en_large::case03_manual_char_len ... bench: 2,578 ns/iter (+/- 55) = 2088 MB/s
test str::char_count::en_medium::case00_cur_libcore ... bench: 44 ns/iter (+/- 1) = 15295 MB/s
test str::char_count::en_medium::case01_old_libcore ... bench: 201 ns/iter (+/- 51) = 3348 MB/s
test str::char_count::en_medium::case02_iter_increment ... bench: 322 ns/iter (+/- 40) = 2090 MB/s
test str::char_count::en_medium::case03_manual_char_len ... bench: 319 ns/iter (+/- 5) = 2109 MB/s
test str::char_count::en_small::case00_cur_libcore ... bench: 15 ns/iter (+/- 0) = 2333 MB/s
test str::char_count::en_small::case01_old_libcore ... bench: 14 ns/iter (+/- 0) = 2500 MB/s
test str::char_count::en_small::case02_iter_increment ... bench: 30 ns/iter (+/- 1) = 1166 MB/s
test str::char_count::en_small::case03_manual_char_len ... bench: 30 ns/iter (+/- 1) = 1166 MB/s
test str::char_count::ru_huge::case00_cur_libcore ... bench: 16,439 ns/iter (+/- 3,105) = 19777 MB/s
test str::char_count::ru_huge::case01_old_libcore ... bench: 89,480 ns/iter (+/- 2,555) = 3633 MB/s
test str::char_count::ru_huge::case02_iter_increment ... bench: 217,703 ns/iter (+/- 22,185) = 1493 MB/s
test str::char_count::ru_huge::case03_manual_char_len ... bench: 157,330 ns/iter (+/- 19,188) = 2066 MB/s
test str::char_count::ru_large::case00_cur_libcore ... bench: 243 ns/iter (+/- 6) = 20905 MB/s
test str::char_count::ru_large::case01_old_libcore ... bench: 1,384 ns/iter (+/- 51) = 3670 MB/s
test str::char_count::ru_large::case02_iter_increment ... bench: 3,381 ns/iter (+/- 543) = 1502 MB/s
test str::char_count::ru_large::case03_manual_char_len ... bench: 2,423 ns/iter (+/- 429) = 2096 MB/s
test str::char_count::ru_medium::case00_cur_libcore ... bench: 42 ns/iter (+/- 1) = 15119 MB/s
test str::char_count::ru_medium::case01_old_libcore ... bench: 180 ns/iter (+/- 4) = 3527 MB/s
test str::char_count::ru_medium::case02_iter_increment ... bench: 402 ns/iter (+/- 45) = 1579 MB/s
test str::char_count::ru_medium::case03_manual_char_len ... bench: 280 ns/iter (+/- 29) = 2267 MB/s
test str::char_count::ru_small::case00_cur_libcore ... bench: 12 ns/iter (+/- 0) = 2666 MB/s
test str::char_count::ru_small::case01_old_libcore ... bench: 12 ns/iter (+/- 0) = 2666 MB/s
test str::char_count::ru_small::case02_iter_increment ... bench: 19 ns/iter (+/- 0) = 1684 MB/s
test str::char_count::ru_small::case03_manual_char_len ... bench: 14 ns/iter (+/- 1) = 2285 MB/s
test str::char_count::zh_huge::case00_cur_libcore ... bench: 15,053 ns/iter (+/- 2,640) = 20067 MB/s
test str::char_count::zh_huge::case01_old_libcore ... bench: 82,622 ns/iter (+/- 3,602) = 3656 MB/s
test str::char_count::zh_huge::case02_iter_increment ... bench: 230,456 ns/iter (+/- 7,246) = 1310 MB/s
test str::char_count::zh_huge::case03_manual_char_len ... bench: 220,595 ns/iter (+/- 11,624) = 1369 MB/s
test str::char_count::zh_large::case00_cur_libcore ... bench: 227 ns/iter (+/- 65) = 20792 MB/s
test str::char_count::zh_large::case01_old_libcore ... bench: 1,136 ns/iter (+/- 144) = 4154 MB/s
test str::char_count::zh_large::case02_iter_increment ... bench: 3,147 ns/iter (+/- 253) = 1499 MB/s
test str::char_count::zh_large::case03_manual_char_len ... bench: 2,993 ns/iter (+/- 400) = 1577 MB/s
test str::char_count::zh_medium::case00_cur_libcore ... bench: 36 ns/iter (+/- 5) = 16388 MB/s
test str::char_count::zh_medium::case01_old_libcore ... bench: 142 ns/iter (+/- 18) = 4154 MB/s
test str::char_count::zh_medium::case02_iter_increment ... bench: 379 ns/iter (+/- 37) = 1556 MB/s
test str::char_count::zh_medium::case03_manual_char_len ... bench: 364 ns/iter (+/- 51) = 1620 MB/s
test str::char_count::zh_small::case00_cur_libcore ... bench: 11 ns/iter (+/- 1) = 3000 MB/s
test str::char_count::zh_small::case01_old_libcore ... bench: 11 ns/iter (+/- 1) = 3000 MB/s
test str::char_count::zh_small::case02_iter_increment ... bench: 20 ns/iter (+/- 3) = 1650 MB/s
</pre>
</details>
I also added fairly thorough tests for different sizes and alignments. This completes on my machine in 0.02s, which is surprising given how thorough they are, but it seems to detect bugs in the implementation. (I haven't run the tests on a 32 bit machine yet since before I reworked the code a little though, so... hopefully I'm not about to embarrass myself).
This uses similar SWAR-style techniques to the `is_ascii` impl I contributed in https://github.com/rust-lang/rust/pull/74066, so I'm going to request review from the same person who reviewed that one. That said am not particularly picky, and might not have the correct syntax for requesting a review from someone (so it goes).
r? `@nagisa`
