| Age | Commit message (Collapse) | Author | Lines |
|---|
|
|
|
|
|
Make <*const/mut T>::offset_from `const fn`
This reenables offset_of cc @mjbshaw after https://github.com/rust-lang/rust/pull/63075 broke it
|
|
The proper attribute was added to `simd_shuffle*` in
rust-lang/stdarch#825. This caused `promote_consts` to double-count its
second argument when recording promotion candidates, which caused
the promotion candidate compatibility check to fail.
|
|
Including removing a bunch of unnecessary `.as_str()` calls, and a bunch
of unnecessary sigils.
|
|
Make `promote_consts` emit the errors when required promotion fails
A very minimal version of #65942.
This will cause a generic "argument X is required to be a constant" message for `simd_shuffle` LLVM intrinsics instead of the [custom one](https://github.com/rust-lang/rust/blob/caa1f8d7b3b021c86a70ff62d23a07d97acff4c4/src/librustc_mir/transform/qualify_consts.rs#L1616). It may be possible to remove this special-casing altogether after rust-lang/stdarch#825.
r? @eddyb
|
|
|
|
|
|
|
|
This commit adds a suggestion to add the
`#![feature(const_in_array_repeat_expression)]` attribute to the crate
when a promotable expression is used in a repeat expression.
Signed-off-by: David Wood <david@davidtw.co>
|
|
`std::panic::Location` is a lang_item, add `core::intrinsics::caller_location` (RFC 2091 3/N)
[Tracking issue](https://github.com/rust-lang/rust/issues/47809)
[RFC text](https://github.com/rust-lang/rfcs/blob/master/text/2091-inline-semantic.md)
@eddyb suggested doing this intrinsic implementation ahead of actually implementing the `#[track_caller]` attribute so that there's an easily tested intermediate step between adding the shim and wiring up the attribute.
|
|
|
|
This is a relic from earlier attempts at dataflow-based const validation
that attempted to do promotion at the same time. #63812 takes a
different approach: `IsNotPromotable` is no longer a `Qualif` and is
computed lazily instead of eagerly. As a result, there's no need for an
eager `TempPromotionResolver`, and we can use the only implementer of
`QualifResolver` directly instead of through a trait.
|
|
|
|
|
|
|
|
This was originally only needed for validation, which is never run on
non-const `fn`s. The new promotion pass wants to use it, however.
|
|
|
|
|
|
Instead coerce to `*const [T; N]` and then cast.
|
|
Intern place projection
This should sit on top of https://github.com/rust-lang/rust/pull/65197. After that one merged, I'm gonna rebase on top of it.
The important commits are the last three and there's a bunch of code repetition that I'm going to remove but for that I need to refactor some things that probably need to be added before this PR.
Anyway this work helps as is because we can run perf tests :).
r? @oli-obk /cc @nikomatsakis
|
|
Remove `InternedString`
This PR removes `InternedString` by converting all occurrences to `Symbol`. There are a handful of places that need to use the symbol chars instead of the symbol index, e.g. for stable sorting; local conversions `LocalInternedString` is used in those places.
r? @eddyb
|
|
|
|
|
|
This prepares the code base for when projection is interned. Place's
projection field is going to be `&List<PlaceElem<'tcx>>` so we won't be
able to pattern match against it.
|
|
The `ConstProp` can cause many locals to be initialized to a constant
value and then never read from. `ConstProp` can also evaluate ZSTs into
constant values. Previously, many of these would be removed by other
parts of the MIR optimization pipeline. However, evaluating ZSTs
(especially `()`) into constant values defeated those parts of the
optimizer and so in a2e3ed5c054b544df6ceeb9e612d39af819f4aae, I added a
hack to `ConstProp` that skips evaluating ZSTs to avoid that regression.
This commit changes `SimplifyLocals` so that it doesn't consider writes
of const values to a local to be a use of that local. In doing so,
`SimplifyLocals` is able to remove otherwise unused locals left behind
by other optimization passes (`ConstProp` in particular).
|
|
|
|
|
|
|
|
|
|
[const-prop] Handle remaining MIR Rvalue cases
r? @oli-obk
|
|
r=eddyb
Suppress ICE when validators disagree on `LiveDrop`s in presence of `&mut`
Resolves #65394.
This hack disables the validator mismatch ICE in cases where a `MutBorrow` error has been emitted by both validators, but they don't agree on the number of `LiveDrop` errors.
The new validator is more conservative about whether a value is moved from in the presence of mutable borrows. For example, the new validator will emit a `LiveDrop` error on the following code.
```rust
const _: Vec<i32> = {
let mut x = Vec::new();
let px = &mut x as *mut _;
let y = x;
unsafe { ptr::write(px, Vec::new()); }
y
};
```
This code is not UB AFAIK (it passes MIRI at least). The current validator does not emit a `LiveDrop` error for `x` upon exit from the initializer. `x` is not actually dropped, so I think this is correct? A proper fix for this would require a new `MaybeInitializedLocals` dataflow analysis or maybe a relaxation of the existing `IndirectlyMutableLocals` one.
r? @RalfJung
|
|
Per code review feedback
|
|
|
|
|
|
rustc: arena-allocate the slice in `ty::GenericsPredicate`, not the whole struct.
While rebasing #59789 I noticed we can do this now. However, it doesn't help much without changing `inferred_outlives_of` to the same type, which I might try next.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
struct.
|
|
|
|
|
