Add SIMD funnel shift and round-to-even intrinsics
This PR adds 3 new SIMD intrinsics
- `simd_funnel_shl` - funnel shift left
- `simd_funnel_shr` - funnel shift right
- `simd_round_ties_even` (vector version of `round_ties_even_fN`)
TODO (future PR): implement `simd_fsh{l,r}` in miri, cg_gcc and cg_clif (it is surprisingly hard to implement without branches, the common tricks that rotate uses doesn't work because we have 2 elements now. e.g, the `-n&31` trick used by cg_gcc to implement rotate doesn't work with this because then `fshl(a, b, 0)` will be `a | b`)
[#t-compiler > More SIMD intrinsics](https://rust-lang.zulipchat.com/#narrow/channel/131828-t-compiler/topic/More.20SIMD.20intrinsics/with/522130286)
`@rustbot` label T-compiler T-libs A-intrinsics F-core_intrinsics
r? `@workingjubilee`
New const traits syntax
This PR only affects the AST and doesn't actually change anything semantically.
All occurrences of `~const` outside of libcore have been replaced by `[const]`. Within libcore we have to wait for rustfmt to be bumped in the bootstrap compiler. This will happen "automatically" (when rustfmt is run) during the bootstrap bump, as rustfmt converts `~const` into `[const]`. After this we can remove the `~const` support from the parser
Caveat discovered during impl: there is no legacy bare trait object recovery for `[const] Trait` as that snippet in type position goes down the slice /array parsing code and will error
r? ``@fee1-dead``
cc ``@nikomatsakis`` ``@traviscross`` ``@compiler-errors``
This centralizes the placeholder type error reporting in one location, but it also exposes the granularity at which we convert things from hir to ty more. E.g. previously infer types in where bounds were errored together with the function signature, but now they are independent.
Add note to `find_const_ty_from_env`
Add a note to `find_const_ty_from_env` to explain why it has an `unwrap` which "often" causes ICEs.
Also, uplift it into the new trait solver. This avoids needing to go through the interner to call this method which is otherwise an inherent method in the compiler. I can remove this part if desired.
r? `@boxyuwu`
Merge unboxed trait object error suggestion into regular dyn incompat error
Another hir-walker removed from the well-formed queries. This error was always a duplicate of another, but it was able to provide more information because it could invoke `is_dyn_compatible` without worrying about cycle errors. That's also the reason we can't put the error directly into hir_ty_lowering when lowering a `dyn Trait` within an associated item signature. So instead I packed it into the error handling of wf obligation checking.
Reduce uses of `hir_crate`.
I tried rebasing my old incremental-HIR branch. This is a by-product, which is required if we want to get rid of `hir_crate` entirely.
The second commit is a drive-by cleanup. It can be pulled into its own PR.
r? ````@oli-obk````
AsyncDrop trait without sync Drop generates an error
When type implements `AsyncDrop` trait, it must also implement sync `Drop` trait to be used in sync context and unwinds.
This PR adds error generation in such a case.
Fixes: rust-lang/rust#140696
Opting-out of `Sized` with `?Sized` is now equivalent to adding a
`MetaSized` bound, and adding a `MetaSized` or `PointeeSized` bound
is equivalent to removing the default `Sized` bound - this commit
implements this change in `rustc_hir_analysis::hir_ty_lowering`.
`MetaSized` is also added as a supertrait of all traits, as this is
necessary to preserve backwards compatibility.
Unfortunately, non-global where clauses being preferred over item bounds
(where `PointeeSized` bounds would be proven) - which can result in
errors when a `PointeeSized` supertrait/bound/predicate is added to some
items. Rather than `PointeeSized` being a bound on everything, it can
be the absence of a bound on everything, as `?Sized` was.
Expand the automatic implementation of `MetaSized` and `PointeeSized` so
that it is also implemented on non-`Sized` types, just not `ty::Foreign`
(extern type).
Temporary lifetime extension through tuple struct and tuple variant constructors
This makes temporary lifetime extension work for tuple struct and tuple variant constructors, such as `Some()`.
Before:
```rust
let a = &temp(); // Extended
let a = Some(&temp()); // Not extended :(
let a = Some { 0: &temp() }; // Extended
```
After:
```rust
let a = &temp(); // Extended
let a = Some(&temp()); // Extended
let a = Some { 0: &temp() }; // Extended
```
So, with this change, this works:
```rust
let a = Some(&String::from("hello")); // New: String lifetime now extended!
println!("{a:?}");
```
Until now, we did not extend through tuple struct/variant constructors (like `Some`), because they are function calls syntactically, and we do not want to extend the String lifetime in:
```rust
let a = some_function(&String::from("hello")); // String not extended!
```
However, it turns out to be very easy to distinguish between regular functions and constructors at the point where we do lifetime extension.
In practice, constructors nearly always use UpperCamelCase while regular functions use lower_snake_case, so it should still be easy to for a human programmer at the call site to see whether something qualifies for lifetime extension or not.
This needs a lang fcp.
---
More examples of what will work after this change:
```rust
let x = Person {
name: "Ferris",
job: Some(&Job { // `Job` now extended!
title: "Chief Rustacean",
organisation: "Acme Ltd.",
}),
};
dbg!(x);
```
```rust
let file = if use_stdout {
None
} else {
Some(&File::create("asdf")?) // `File` now extended!
};
set_logger(file);
```
```rust
use std::path::Component;
let c = Component::Normal(&OsString::from(format!("test-{num}"))); // OsString now extended!
assert_eq!(path.components.first().unwrap(), c);
```
Add (back) `unsupported_calling_conventions` lint to reject more invalid calling conventions
This adds back the `unsupported_calling_conventions` lint that was removed in https://github.com/rust-lang/rust/pull/129935, in order to start the process of dealing with https://github.com/rust-lang/rust/issues/137018. Specifically, we are going for the plan laid out [here](https://github.com/rust-lang/rust/issues/137018#issuecomment-2672118326):
- thiscall, stdcall, fastcall, cdecl should only be accepted on x86-32
- vectorcall should only be accepted on x86-32 and x86-64
The difference to the status quo is that:
- We stop accepting stdcall, fastcall on targets that are windows && non-x86-32 (we already don't accept these on targets that are non-windows && non-x86-32)
- We stop accepting cdecl on targets that are non-x86-32
- (There is no difference for thiscall, this was already a hard error on non-x86-32)
- We stop accepting vectorcall on targets that are windows && non-x86-*
Vectorcall is an unstable ABI so we can just make this a hard error immediately. The others are stable, so we emit the `unsupported_calling_conventions` forward-compat lint. I set up the lint to show up in dependencies via cargo's future-compat report immediately, but we could also make it show up just for the local crate first if that is preferred.
try-job: i686-msvc-1
try-job: x86_64-msvc-1
try-job: test-various
Remove CollectItemTypesVisitor
I always felt like we were very unnecessarily walking the HIR, let's see if perf agrees
There is lots to ~~improve~~ consolidate further here, as we still have 3 item wfchecks:
* check_item (matching on the hir::ItemKind)
* actually doing trait solver based checks (by using HIR spans)
* lower_item (matching on the hir::ItemKind after loading it again??)
* just ensure_ok-ing a bunch of queries
* check_item_type (matching on DefKind)
* some type based checks, mostly ensure_ok-ing a bunch of queries
fixesrust-lang/rust#121429
`slice.get(i)` should use a slice projection in MIR, like `slice[i]` does
`slice[i]` is built-in magic, so ends up being quite different from `slice.get(i)` in MIR, even though they're both doing nearly identical operations -- checking the length of the slice then getting a ref/ptr to the element if it's in-bounds.
This PR adds a `slice_get_unchecked` intrinsic for `impl SliceIndex for usize` to use to fix that, so it no longer needs to do a bunch of lines of pointer math and instead just gets the obvious single statement. (This is *not* used for the range versions, since `slice[i..]` and `slice[..k]` can't use the mir Slice projection as they're using fenceposts, not indices.)
I originally tried to do this with some kind of GVN pattern, but realized that I'm pretty sure it's not legal to optimize `BinOp::Offset` to `PlaceElem::Index` without an extremely complicated condition. Basically, the problem is that the `Index` projection on a dereferenced slice pointer *cares about the metadata*, since it's UB to `PlaceElem::Index` outside the range described by the metadata. But then you cast the fat pointer to a thin pointer then offset it, that *ignores* the slice length metadata, so it's possible to write things that are legal with `Offset` but would be UB if translated in the obvious way to `Index`. Checking (or even determining) the necessary conditions for that would be complicated and error-prone, whereas this intrinsic-based approach is quite straight-forward.
Zero backend changes, because it just lowers to MIR, so it's already supported naturally by CTFE/Miri/cg_llvm/cg_clif.
