Stabilize x86/x86_64 SIMD
This commit stabilizes the SIMD in Rust for the x86/x86_64 platforms. Notably
this commit is stabilizing:
* The `std::arch::{x86, x86_64}` modules and the intrinsics contained inside.
* The `is_x86_feature_detected!` macro in the standard library
* The `#[target_feature(enable = "...")]` attribute
* The `#[cfg(target_feature = "...")]` matcher
Stabilization of the module and intrinsics were primarily done in
rust-lang-nursery/stdsimd#414 and the two attribute stabilizations are done in
this commit. The standard library is also tweaked a bit with the new way that
stdsimd is integrated.
Note that other architectures like `std::arch::arm` are not stabilized as part
of this commit, they will likely stabilize in the future after they've been
implemented and fleshed out. Similarly the `std::simd` module is also not being
stabilized in this commit, only `std::arch`. Finally, nothing related to `__m64`
is stabilized in this commit either (MMX), only SSE and up types and intrinsics
are stabilized.
Closes#29717Closes#44839Closes#48556
std: Minimize size of panicking on wasm
This commit applies a few code size optimizations for the wasm target to
the standard library, namely around panics. We notably know that in most
configurations it's impossible for us to print anything in
wasm32-unknown-unknown so we can skip larger portions of panicking that
are otherwise simply informative. This allows us to get quite a nice
size reduction.
Finally we can also tweak where the allocation happens for the
`Box<Any>` that we panic with. By only allocating once unwinding starts
we can reduce the size of a panicking wasm module from 44k to 350 bytes.
Move Range*::contains to a single default impl on RangeBounds
Per the ongoing discussion in #32311.
This is my first PR to Rust (woo!), so I don't know if this requires an amendment to the original range_contains RFC, or not, or if we can just do a psuedo-RFC here. While this may no longer follow the explicit decision made in that RFC, I believe this better follows its spirit by adding the new contains method to all Ranges. It also allows users to be generic over all ranges and use this method without writing it themselves (my personal desired use case).
This also somewhat answers the unanswered question about Wrapping ranges in the above issue by instead just punting it to the question of what those types should return for start() & end(), or if they should implement RangeArgument at all. Those types could also implement their own contains method without implementing this trait, in which case the question remains the same.
This does add a new contains method to types that already implemented RangeArgument but not contains. These types are RangeFull, (Bound<T>, Bound<T>), (Bound<&'a T>, Bound<&'a T>). No tests have been added for these types yet. No inherent method has been added either.
r? @alexcrichton
This commit stabilizes the SIMD in Rust for the x86/x86_64 platforms. Notably
this commit is stabilizing:
* The `std::arch::{x86, x86_64}` modules and the intrinsics contained inside.
* The `is_x86_feature_detected!` macro in the standard library
* The `#[target_feature(enable = "...")]` attribute
* The `#[cfg(target_feature = "...")]` matcher
Stabilization of the module and intrinsics were primarily done in
rust-lang-nursery/stdsimd#414 and the two attribute stabilizations are done in
this commit. The standard library is also tweaked a bit with the new way that
stdsimd is integrated.
Note that other architectures like `std::arch::arm` are not stabilized as part
of this commit, they will likely stabilize in the future after they've been
implemented and fleshed out. Similarly the `std::simd` module is also not being
stabilized in this commit, only `std::arch`. Finally, nothing related to `__m64`
is stabilized in this commit either (MMX), only SSE and up types and intrinsics
are stabilized.
Closes#29717Closes#44839Closes#48556
core: Remove panics from some `Layout` methods
`Layout` is often used at the core of allocation APIs and is as a result pretty
sensitive to codegen in various circumstances. I was profiling `-C opt-level=z`
with a wasm project recently and noticed that the `unwrap()` wasn't removed
inside of `Layout`, causing the program to be much larger than it otherwise
would be. If inlining were more aggressive LLVM would have figured out that the
panic could be eliminated, but in general the methods here can't panic in the
first place!
As a result this commit makes the following tweaks:
* Removes `unwrap()` and replaces it with `unsafe` in `Layout::new` and
`Layout::for_value`. For posterity though a debug assertion was left behind.
* Removes an `unwrap()` in favor of `?` in the `repeat` method. The comment
indicating that the function call couldn't panic wasn't quite right in that if
`alloc_size` becomes too large and if `align` is high enough it could indeed
cause a panic.
This'll hopefully mean that panics never get introduced into code in the first
place, ensuring that `opt-level=z` is closer to `opt-level=s` in this regard.
improve Atomic*::fetch_update docs
This clarifies that fetch_update *always* returns the previous value, either as `Ok(_)` or `Err(_)`, depending on whether the supplied update function returned `Some(_)` or `None`.
[doc] note the special type inference handling for shift ops
This adds a note to the docs about the difference between the shift ops and the corresponding trait methods when it comes to type inference.
This commit removes allocation of the panic message in instances like
`panic!("foo: {}", "bar")` if we don't actually end up needing the message. We
don't need it in the case of wasm32 right now, and in general it's not needed
for panic=abort instances that use the default panic hook.
For now this commit only solves the wasm use case where with LTO the allocation
is entirely removed, but the panic=abort use case can be implemented at a later
date if needed.
The expression `&s[..i]` in general can panic if `i` is out of bounds or not on
a character boundary for a string, and this caused the codegen for
`Formatter::pad` to be a bit larger than it otherwise needed to be. This commit
replaces this with `s.get(..i).unwrap_or(&s)` which while having different
behavior if `i` is out of bounds has a much smaller code footprint and otherwise
avoids the need for `unsafe` code.
This commit applies a few code size optimizations for the wasm target to
the standard library, namely around panics. We notably know that in most
configurations it's impossible for us to print anything in
wasm32-unknown-unknown so we can skip larger portions of panicking that
are otherwise simply informative. This allows us to get quite a nice
size reduction.
Finally we can also tweak where the allocation happens for the
`Box<Any>` that we panic with. By only allocating once unwinding starts
we can reduce the size of a panicking wasm module from 44k to 350 bytes.
`Layout` is often used at the core of allocation APIs and is as a result pretty
sensitive to codegen in various circumstances. I was profiling `-C opt-level=z`
with a wasm project recently and noticed that the `unwrap()` wasn't removed
inside of `Layout`, causing the program to be much larger than it otherwise
would be. If inlining were more aggressive LLVM would have figured out that the
panic could be eliminated, but in general the methods here can't panic in the
first place!
As a result this commit makes the following tweaks:
* Removes `unwrap()` and replaces it with `unsafe` in `Layout::new` and
`Layout::for_value`. For posterity though a debug assertion was left behind.
* Removes an `unwrap()` in favor of `?` in the `repeat` method. The comment
indicating that the function call couldn't panic wasn't quite right in that if
`alloc_size` becomes too large and if `align` is high enough it could indeed
cause a panic.
This'll hopefully mean that panics never get introduced into code in the first
place, ensuring that `opt-level=z` is closer to `opt-level=s` in this regard.
Add GlobalAlloc trait + tweaks for initial stabilization
This is the outcome of discussion at the Rust All Hands in Berlin. The high-level goal is stabilizing sooner rather than later the ability to [change the global allocator](https://github.com/rust-lang/rust/issues/27389), as well as allocating memory without abusing `Vec::with_capacity` + `mem::forget`.
Since we’re not ready to settle every detail of the `Alloc` trait for the purpose of collections that are generic over the allocator type (for example the possibility of a separate trait for deallocation only, and what that would look like exactly), we propose introducing separately **a new `GlobalAlloc` trait**, for use with the `#[global_allocator]` attribute.
We also propose a number of changes to existing APIs. They are batched in this one PR in order to minimize disruption to Nightly users.
The plan for initial stabilization is detailed in the tracking issue https://github.com/rust-lang/rust/issues/49668.
CC @rust-lang/libs, @glandium
## Immediate breaking changes to unstable features
* For pointers to allocated memory, change the pointed type from `u8` to `Opaque`, a new public [extern type](https://github.com/rust-lang/rust/issues/43467). Since extern types are not `Sized`, `<*mut _>::offset` cannot be used without first casting to another pointer type. (We hope that extern types can also be stabilized soon.)
* In the `Alloc` trait, change these pointers to `ptr::NonNull` and change the `AllocErr` type to a zero-size struct. This makes return types `Result<ptr::NonNull<Opaque>, AllocErr>` be pointer-sized.
* Instead of a new `Layout`, `realloc` takes only a new size (in addition to the pointer and old `Layout`). Changing the alignment is not supported with `realloc`.
* Change the return type of `Layout::from_size_align` from `Option<Self>` to `Result<Self, LayoutErr>`, with `LayoutErr` a new opaque struct.
* A `static` item registered as the global allocator with the `#[global_allocator]` **must now implement the new `GlobalAlloc` trait** instead of `Alloc`.
## Eventually-breaking changes to unstable features, with a deprecation period
* Rename the respective `heap` modules to `alloc` in the `core`, `alloc`, and `std` crates. (Yes, this does mean that `::alloc::alloc::Alloc::alloc` is a valid path to a trait method if you have `exetrn crate alloc;`)
* Rename the the `Heap` type to `Global`, since it is the entry point for what’s registered with `#[global_allocator]`.
Old names remain available for now, as deprecated `pub use` reexports.
## Backward-compatible changes
* Add a new [extern type](https://github.com/rust-lang/rust/issues/43467) `Opaque`, for use in pointers to allocated memory.
* Add a new `GlobalAlloc` trait shown below. Unlike `Alloc`, it uses bare `*mut Opaque` without `NonNull` or `Result`. NULL in return values indicates an error (of unspecified nature). This is easier to implement on top of `malloc`-like APIs.
* Add impls of `GlobalAlloc` for both the `Global` and `System` types, in addition to existing impls of `Alloc`. This enables calling `GlobalAlloc` methods on the stable channel before `Alloc` is stable. Implementing two traits with identical method names can make some calls ambiguous, but most code is expected to have no more than one of the two traits in scope. Erroneous code like `use std::alloc::Global; #[global_allocator] static A: Global = Global;` (where `Global` is defined to call itself, causing infinite recursion) is not statically prevented by the type system, but we count on it being hard enough to do accidentally and easy enough to diagnose.
```rust
extern {
pub type Opaque;
}
pub unsafe trait GlobalAlloc {
unsafe fn alloc(&self, layout: Layout) -> *mut Opaque;
unsafe fn dealloc(&self, ptr: *mut Opaque, layout: Layout);
unsafe fn alloc_zeroed(&self, layout: Layout) -> *mut Opaque {
// Default impl: self.alloc() and ptr::write_bytes()
}
unsafe fn realloc(&self, ptr: *mut Opaque, old_layout: Layout, new_size: usize) -> *mut Opaque {
// Default impl: self.alloc() and ptr::copy_nonoverlapping() and self.dealloc()
}
fn oom(&self) -> ! {
// intrinsics::abort
}
// More methods with default impls may be added in the future
}
```
## Bikeshed
The tracking issue https://github.com/rust-lang/rust/issues/49668 lists some open questions. If consensus is reached before this PR is merged, changes can be integrated.
Deprecate offset_to; switch core&alloc to using offset_from instead
Bonus: might make code than uses `.len()` on slice iterators faster
cc https://github.com/rust-lang/rust/issues/41079
