The results of most analyses end up in a `Results<'tcx, A>`, where `A`
is the analysis. It's then possible to traverse the results via a
`ResultsVisitor`, which relies on the `ResultsVisitable` trait. (That
trait ends up using the same `apply_*` methods that were used when
computing the analysis, albeit indirectly.)
This pattern of "compute analysis results, then visit them" is common.
But there is one exception. For borrow checking we compute three
separate analyses (`Borrows`, `MaybeUninitializedPlaces`, and
`EverInitializedPlaces`), combine them into a single `BorrowckResults`,
and then do a single visit of that `BorrowckResults` with
`MirBorrowckResults`. `BorrowckResults` is just different enough from
`Results` that it requires the existence of `ResultsVisitable`, which
abstracts over the traversal differences between `Results` and
`BorrowckResults`.
This commit changes things by introducing `Borrowck` and bundling the
three borrowck analysis results into a standard `Results<Borrowck>`
instead of the exceptional `BorrowckResults`. Once that's done, the
results can be visited like any other analysis results.
`BorrowckResults` is removed, as is `impl ResultsVisitable for
BorrowckResults`. (It's instructive to see how similar the added `impl
Analysis for Borrowck` is to the removed `impl ResultsVisitable for
BorrowckResults`. They're both doing exactly the same things.)
Overall this increases the number of lines of code and might not seem
like a win. But it enables the removal of `ResultsVisitable` in the next
commit, which results in many simplifications.
They represent a lot of abstraction and indirection, but they're only
used for `ConstAnalysis`, and apparently won't be used for any other
analyses in the future. This commit inlines and removes them, which
makes `ConstAnalysis` easier to read and understand.
Rename `rustc_abi::Abi` to `BackendRepr`
Remove the confabulation of `rustc_abi::Abi` with what "ABI" actually means by renaming it to `BackendRepr`, and rename `Abi::Aggregate` to `BackendRepr::Memory`. The type never actually represented how things are passed, as that has to have `PassMode` considered, at minimum, but rather it just is how we represented some things to the backend. This conflation arose because LLVM, the primary backend at the time, would lower certain IR forms using certain ABIs. Even that only somewhat was true, as it broke down when one ventured significantly afield of what is described by the System V AMD64 ABI either by using different architectures, ABI-modifying IR annotations, the same architecture **with different ISA extensions enabled**, or other... unexpected delights.
Unfortunately both names are still somewhat of a misnomer right now, as people have written code for years based on this misunderstanding. Still, their original names are even moreso, and for better or worse, this backend code hasn't received as much maintenance as the rest of the compiler, lately. Actually arriving at a correct end-state will simply require us to disentangle a lot of code in order to fix, much of it pointlessly repeated in several places. Thus this is not an "actual fix", just a way to deflect further misunderstandings.
`Formatter` currently has a `RefCell<Option<Results>>` field. This is so
the `Results` can be temporarily taken and put into a `ResultsCursor`
that is used by `BlockFormatter`, and then put back, which is messy.
This commit changes `Formatter` to have a `RefCell<ResultsCursor>` and
`BlockFormatter` to have a `&mut ResultsCursor`, which greatly
simplifies the code at the `Formatter`/`BlockFormatter` interaction
point in `Formatter::node_label`. It also means we construct a
`ResultsCursor` once per `Formatter`, instead of once per `node_label`
call.
The commit also:
- documents the reason for the `RefCell`;
- adds a `Formatter::body` method, replacing the `Formatter::body`
field.
It's no longer needed. `Engine::iterate_to_fixpoint` can be inlined into
`Analysis::iterate_to_fixpoint` and removed. The commit also renames
`engine.rs` as `results.rs`.
This is a standard pattern:
```
MyAnalysis.into_engine(tcx, body).iterate_to_fixpoint()
```
`into_engine` and `iterate_to_fixpoint` are always called in pairs, but
sometimes with a builder-style `pass_name` call between them. But a
builder-style interface is overkill here. This has been bugging me a for
a while.
This commit:
- Merges `Engine::new` and `Engine::iterate_to_fixpoint`. This removes
the need for `Engine` to have fields, leaving it as a trivial type
that the next commit will remove.
- Renames `Analysis::into_engine` as `Analysis::iterate_to_fixpoint`,
gives it an extra argument for the optional pass name, and makes it
call `Engine::iterate_to_fixpoint` instead of `Engine::new`.
This turns the pattern from above into this:
```
MyAnalysis.iterate_to_fixpoint(tcx, body, None)
```
which is shorter at every call site, and there's less plumbing required
to support it.
The initial naming of "Abi" was an awful mistake, conveying wrong ideas
about how psABIs worked and even more about what the enum meant.
It was only meant to represent the way the value would be described to
a codegen backend as it was lowered to that intermediate representation.
It was never meant to mean anything about the actual psABI handling!
The conflation is because LLVM typically will associate a certain form
with a certain ABI, but even that does not hold when the special cases
that actually exist arise, plus the IR annotations that modify the ABI.
Reframe `rustc_abi::Abi` as the `BackendRepr` of the type, and rename
`BackendRepr::Aggregate` as `BackendRepr::Memory`. Unfortunately, due to
the persistent misunderstandings, this too is now incorrect:
- Scattered ABI-relevant code is entangled with BackendRepr
- We do not always pre-compute a correct BackendRepr that reflects how
we "actually" want this value to be handled, so we leave the backend
interface to also inject various special-cases here
- In some cases `BackendRepr::Memory` is a "real" aggregate, but in
others it is in fact using memory, and in some cases it is a scalar!
Our rustc-to-backend lowering code handles this sort of thing right now.
That will eventually be addressed by lifting duplicated lowering code
to either rustc_codegen_ssa or rustc_target as appropriate.
Depend on rustc_abi in compiler crates that use it indirectly but have
not yet taken on that dependency, and are not entangled in my other PRs.
This leaves an "excise rustc_target" step after the dust settles.
`GenKillAnalysis` has very similar methods to `Analysis`, but the first
two have a notable difference: the second argument is `&mut impl
GenKill<Self::Idx>` instead of `&mut Self::Domain`. But thanks to the
previous commit, this difference is no longer necessary.
This is an alternative to `Engine::new_generic` for gen/kill analyses.
It's supposed to be an optimization, but it has negligible effect.
The commit merges `Engine::new_generic` into `Engine::new`.
This allows the removal of various other things: `GenKillSet`,
`gen_kill_statement_effects_in_block`, `is_cfg_cyclic`.
- fix for divergence
- fix error message
- fix another cranelift test
- fix some cranelift things
- don't set the NORETURN option for naked asm
- fix use of naked_asm! in doc comment
- fix use of naked_asm! in run-make test
- use `span_bug` in unreachable branch
- Replace non-standard names like 's, 'p, 'rg, 'ck, 'parent, 'this, and
'me with vanilla 'a. These are cases where the original name isn't
really any more informative than 'a.
- Replace names like 'cx, 'mir, and 'body with vanilla 'a when the lifetime
applies to multiple fields and so the original lifetime name isn't
really accurate.
- Put 'tcx last in lifetime lists, and 'a before 'b.
No analysis needs `Copy`, and `MaybeBorrowedLocals` is the only analysis
that needs `Clone`. In `locals_live_across_suspend_points` it gets
cloned so it can be used within a `MaybeRequiresStorage`.
It's a very thin wrapper that pairs `MoveDataBuilder` with a `Location`,
and it has four lifetime arguments. This commit removes it by just
adding a `Location` to `MoveDataBuilder`.
There are four related dataflow structs: `MaybeInitializedPlaces`,
`MaybeUninitializedPlaces`, and `EverInitializedPlaces`,
`DefinitelyInitializedPlaces`. They all have a `&Body` and a
`&MoveData<'tcx>` field. The first three use different lifetimes for the
two fields, but the last one uses the same lifetime for both.
This commit changes the first three to use the same lifetime, removing
the need for one of the lifetimes. Other structs that also lose a
lifetime as a result of this are `LivenessContext`, `LivenessResults`,
`InitializationData`.
It then does similar things in various other structs.
The actual implementation remains in `rustc_mir_dataflow`, but this
commit moves the `MirPass` impl to `rustc_mir_transform` and changes it
to a `MirLint` (fixing a `FIXME` comment).
(I originally tried moving the full implementation from
`rustc_mir_dataflow` but I had some trait problems with `HasMoveData`
and `RustcPeekAt` and `MaybeLiveLocals`. This commit was much smaller
and simpler, but still will allow some follow-up cleanups.)
Shrink `TyKind::FnPtr`.
By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and `FnHeader`, which can be packed more efficiently. This reduces the size of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms. This reduces peak memory usage by a few percent on some benchmarks. It also reduces cache misses and page faults similarly, though this doesn't translate to clear cycles or wall-time improvements on CI.
r? `@compiler-errors`
By splitting the `FnSig` within `TyKind::FnPtr` into `FnSigTys` and
`FnHeader`, which can be packed more efficiently. This reduces the size
of the hot `TyKind` type from 32 bytes to 24 bytes on 64-bit platforms.
This reduces peak memory usage by a few percent on some benchmarks. It
also reduces cache misses and page faults similarly, though this doesn't
translate to clear cycles or wall-time improvements on CI.
Clean up a few minor refs in `format!` macro, as it has a performance cost. Apparently the compiler is unable to inline `format!("{}", &variable)`, and does a run-time double-reference instead (format macro already does one level referencing). Inlining format args prevents accidental `&` misuse.
Make jump threading state sparse
Continuation of https://github.com/rust-lang/rust/pull/127024
Both dataflow const-prop and jump threading involve cloning the state vector a lot. This PR replaces the data structure by a sparse vector, considering:
- that jump threading state is typically very sparse (at most 1 or 2 set entries);
- that dataflow const-prop is disabled by default;
- that place/value map is very eager, and prone to creating an overly large state.
The first commit is shared with the previous PR to avoid needless conflicts.
r? `@oli-obk`
Automatically taint InferCtxt when errors are emitted
r? `@nnethercote`
Basically `InferCtxt::dcx` now returns a `DiagCtxt` that refers back to the `Cell<Option<ErrorGuaranteed>>` of the `InferCtxt` and thus when invoking `Diag::emit`, and the diagnostic is an error, we taint the `InferCtxt` directly.
That change on its own has no effect at all, because `InferCtxt` already tracks whether errors have been emitted by recording the global error count when it gets opened, and checking at the end whether the count changed. So I removed that error count check, which had a bit of fallout that I immediately fixed by invoking `InferCtxt::dcx` instead of `TyCtxt::dcx` in a bunch of places.
The remaining new errors are because an error was reported in another query, and never bubbled up. I think they are minor enough for this to be ok, and sometimes it actually improves diagnostics, by not silencing useful diagnostics anymore.
fixes#126485 (cc `@olafes)`
There are more improvements we can do (like tainting in hir ty lowering), but I would rather do that in follow up PRs, because it requires some refactorings.
These things don't need to be `Vec`s; boxed slices are enough.
The frequent one here is call arguments, but MIR building knows the number of arguments from the THIR, so the collect is always getting the allocation right in the first place, and thus this shouldn't ever add the shrink-in-place overhead.
We already do this for a number of crates, e.g. `rustc_middle`,
`rustc_span`, `rustc_metadata`, `rustc_span`, `rustc_errors`.
For the ones we don't, in many cases the attributes are a mess.
- There is no consistency about order of attribute kinds (e.g.
`allow`/`deny`/`feature`).
- Within attribute kind groups (e.g. the `feature` attributes),
sometimes the order is alphabetical, and sometimes there is no
particular order.
- Sometimes the attributes of a particular kind aren't even grouped
all together, e.g. there might be a `feature`, then an `allow`, then
another `feature`.
This commit extends the existing sorting to all compiler crates,
increasing consistency. If any new attribute line is added there is now
only one place it can go -- no need for arbitrary decisions.
Exceptions:
- `rustc_log`, `rustc_next_trait_solver` and `rustc_type_ir_macros`,
because they have no crate attributes.
- `rustc_codegen_gcc`, because it's quasi-external to rustc (e.g. it's
ignored in `rustfmt.toml`).
Remove many `#[macro_use] extern crate foo` items
This requires the addition of more `use` items, which often make the code more verbose. But they also make the code easier to read, because `#[macro_use]` obscures where macros are defined.
r? `@fee1-dead`
Remove optionality from MoveData::base_local
This is an artifact from when Places could be based on statics and not just locals. Now, all move paths either are locals or have parents, so this doesn't need to return Option anymore.
Add asm goto support to `asm!`
Tracking issue: #119364
This PR implements asm-goto support, using the syntax described in "future possibilities" section of [RFC2873](https://rust-lang.github.io/rfcs/2873-inline-asm.html#asm-goto).
Currently I have only implemented the `label` part, not the `fallthrough` part (i.e. fallthrough is implicit). This doesn't reduce the expressive though, since you can use label-break to get arbitrary control flow or simply set a value and rely on jump threading optimisation to get the desired control flow. I can add that later if deemed necessary.
r? ``@Amanieu``
cc ``@ojeda``
Implement intrinsics with fallback bodies
fixes#93145 (though we can port many more intrinsics)
cc #63585
The way this works is that the backend logic for generating custom code for intrinsics has been made fallible. The only failure path is "this intrinsic is unknown". The `Instance` (that was `InstanceDef::Intrinsic`) then gets converted to `InstanceDef::Item`, which represents the fallback body. A regular function call to that body is then codegenned. This is currently implemented for
* codegen_ssa (so llvm and gcc)
* codegen_cranelift
other backends will need to adjust, but they can just keep doing what they were doing if they prefer (though adding new intrinsics to the compiler will then require them to implement them, instead of getting the fallback body).
cc `@scottmcm` `@WaffleLapkin`
### todo
* [ ] miri support
* [x] default intrinsic name to name of function instead of requiring it to be specified in attribute
* [x] make sure that the bodies are always available (must be collected for metadata)
These crates all needed specialization for `newtype_index!`, which will no
longer be necessary when the current nightly eventually becomes the next
bootstrap compiler.
Invert diagnostic lints.
That is, change `diagnostic_outside_of_impl` and `untranslatable_diagnostic` from `allow` to `deny`, because more than half of the compiler has been converted to use translated diagnostics.
This commit removes more `deny` attributes than it adds `allow` attributes, which proves that this change is warranted.
r? ````@davidtwco````
That is, change `diagnostic_outside_of_impl` and
`untranslatable_diagnostic` from `allow` to `deny`, because more than
half of the compiler has be converted to use translated diagnostics.
This commit removes more `deny` attributes than it adds `allow`
attributes, which proves that this change is warranted.
Added back previously available exports:
* Forward/Backward: used when implementing `AnalysisDomain`
* Engine: used in user's code to solve the dataflow problem
* SwitchIntEdgeEffects: used when implementing functions of the `Analysis` trait
* graphviz: potentially useful for debugging purposes
These exports are used when implementing external tools based on MIR
dataflow framework.
Closes#120130
Save liveness results for DestinationPropagation
`DestinationPropagation` needs to verify that merge candidates do not conflict with each other. This is done by verifying that a local is not live when its counterpart is written to.
To get the liveness information, the pass runs `MaybeLiveLocals` dataflow analysis repeatedly, once for each propagation round. This is quite costly, and the main driver for the perf impact on `ucd` and `diesel`. (See https://github.com/rust-lang/rust/pull/115105#issuecomment-1689205908)
In order to mitigate this cost, this PR proposes to save the result of the analysis into a `SparseIntervalMatrix`, and mirror merges of locals into that matrix: `liveness(destination) := liveness(destination) union liveness(source)`.
<details>
<summary>Proof</summary>
We denote by `'` all the quantities of the transformed program. Let $\varphi$ be a mapping of locals, which maps `source` to `destination`, and is identity otherwise. The exact liveness set after a statement is $out'(statement)$, and the proposed liveness set is $\varphi(out(statement))$.
Consider a statement. Suppose that the output state verifies $out' \subset phi(out)$. We want to prove that $in' \subset \varphi(in)$ where $in = (out - kill) \cup gen$, and conclude by induction.
We have 2 cases: either that statement is kept with locals renumbered by $\varphi$, or it is a tautological assignment and it removed.
1. If the statement is kept: the gen-set and the kill-set of $statement' = \varphi(statement)$ are $gen' = \varphi(gen)$ and $kill' = \varphi(kill)$ exactly.
From soundness requirement 3, $\varphi(in)$ is disjoint from $\varphi(kill)$.
This implies that $\varphi(out - kill)$ is disjoint from $\varphi(kill)$, and so $\varphi(out - kill) = \varphi(out) - \varphi(kill)$. Then $\varphi(in) = (\varphi(out) - \varphi(kill)) \cup \varphi(gen) = (\varphi(out) - kill') \cup gen'$.
We can conclude that $out' \subset \varphi(out) \implies in' \subset \varphi(in)$.
2. If the statement is removed. As $\varphi(statement)$ is a tautological assignment, we know that $\varphi(gen) = \varphi(kill) = \\{ destination \\}$, while $gen' = kill' = \emptyset$. So $\varphi(in) = \varphi(out) \cup \\{ destination \\}$. Then $in' = out' \subset out \subset \varphi(in)$.
By recursion, we can conclude by that $in' \subset \varphi(in)$ everywhere.
</details>
This approximate liveness results is only suboptimal if there are locals that fully disappear from the CFG due to an assignment cycle. These cases are quite unlikely, so we do not bother with them.
This change allows to reduce the perf impact of DestinationPropagation by half on diesel and ucd (https://github.com/rust-lang/rust/pull/115105#issuecomment-1694701904).
cc ````@JakobDegen````
