DestinationPropagation: compute liveness as ranges instead of traveling bitsets
The current implementation of `save_as_liveness` is very slow, and consists in inserting a traveling bitset in an interval set.
As the `MaybeLiveLocals` has a gen-kill property, we can leverage it to make it faster. "Gen" is creating a new interval. "Kill" is ending this interval, ripe to save in the interval set.
Change `SwitchInt` handling in dataflow analysis.
We call `get_switch_int_data` once for the switch and then pass that data to `apply_switch_int_edge_effect` for each switch target.
The only case in practice is `MaybePlacesSwitchIntData` which does an awkward thing, maintaining an index into the discriminants and updating it on each call to `apply_switch_int_edge_effect`.
This commit changes things to do more work up front in `get_switch_int_data`, in order to then do less work in `apply_switch_int_edge_effect`. This avoids the need for the `variants` and `next_discr` methods and the discriminants index. Overall it's a little simpler.
r? @cjgillot
We call `get_switch_int_data` once for the switch and then pass that
data to `apply_switch_int_edge_effect` for each switch target.
The only case in practice is `MaybePlacesSwitchIntData` which does an
awkward thing, maintaining an index into the discriminants and updating
it on each call to `apply_switch_int_edge_effect`.
This commit changes things to do more work up front in
`get_switch_int_data`, in order to then do less work in
`apply_switch_int_edge_effect`. This avoids the need for the `variants`
and `next_discr` methods and the discriminants index. Overall it's a
little simpler.
Fix some comments about dataflow analysis.
Mostly in the examples in `initialized.rs`. In particular, the `EverInitializedPlaces` example currently doesn't cover how it's initialization sites that are tracked, rather than local variables (that's the `b_0`/`b_1` distinction in the example.)
r? @cjgillot
Mostly in the examples in `initialized.rs`. In particular, the
`EverInitializedPlaces` example currently doesn't cover how it's
initialization sites that are tracked, rather than local variables
(that's the `b_0`/`b_1` distinction in the example.)
The `state: A::Domain` value is the primary things that's modified when
performing an analysis. The `Analysis` impl is immutable in every case
but one (`MaybeRequiredStorage`) and it now uses interior mutability.
As well as changing many `&mut A` arguments to `&A`, this also:
- lets `CowMut` be replaced with the simpler `SimpleCow` in `cursor.rs`;
- removes the need for the `RefCell` in `Formatter`;
- removes the need for `MaybeBorrowedLocals` to impl `Clone`, because
it's a unit type and it's now clear that its constructor can be used
directly instead of being put into a local variable and cloned.
Improve formatting of doc code blocks
We don't currently apply automatic formatting to doc comment code blocks. As a
result, it has built up various idiosyncracies, which make such automatic
formatting difficult. Some of those idiosyncracies also make things harder for
human readers or other tools.
This PR makes a few improvements to doc code formatting, in the hopes of making
future automatic formatting easier, as well as in many cases providing net
readability improvements.
I would suggest reading each commit separately, as each commit contains one
class of changes.
These examples feature Rust code that's presented primarily to
illustrate how its compilation would be handled, and these examples are
formatted to highlight those aspects in ways that rustfmt wouldn't
preserve. Turn formatting off in those examples.
(Doc code isn't formatted yet, but this will make it easier to enable
doc code formatting in the future.)
`rustc_mir_dataflow/src/elaborate_drops.rs` contains some infrastructure
used by a few MIR passes: the `elaborate_drop` function, the
`DropElaborator` trait, etc.
`rustc_mir_transform/src/elaborate_drops.rs` (same file name, different
crate) contains the `ElaborateDrops` pass. It relies on a lot of the
infrastructure from `rustc_mir_dataflow/src/elaborate_drops.rs`.
It turns out that the drop infrastructure is only used in
`rustc_mir_transform`, so this commit moves it there. (The only
exception is the small `DropFlagState` type, which is moved to the
existing `rustc_mir_dataflow/src/drop_flag_effects.rs`.) The file is
renamed from `rustc_mir_dataflow/src/elaborate_drops.rs` to
`rustc_mir_transform/src/elaborate_drop.rs` (with no trailing `s`)
because (a) the `elaborate_drop` function is the most important export,
and (b) `rustc_mir_transform/src/elaborate_drops.rs` already exists.
All the infrastructure pieces that used to be `pub` are now
`pub(crate)`, because they are now only used within
`rustc_mir_transform`.
Current `SwitchInt` handling has complicated control flow.
- The dataflow engine calls `Analysis::apply_switch_int_edge_effects`,
passing in an "applier" that impls `SwitchIntEdgeEffects`.
- `apply_switch_int_edge_effects` possibly calls `apply` on the applier,
passing it a closure.
- The `apply` method calls the closure on each `SwitchInt` edge.
- The closure operates on the edge.
I.e. control flow goes from the engine, to the analysis, to the applier
(which came from the engine), to the closure (which came from the
analysis). It took me a while to work this out.
This commit changes to a simpler structure that maintains the important
characteristics.
- The dataflow engine calls `Analysis::get_switch_int_data`.
- `get_switch_int_data` returns an `Option<Self::SwitchIntData>` value.
- If that returned value was `Some`, the dataflow engine calls
`Analysis::apply_switch_int_edge_effect` on each edge, passing the
`Self::SwitchIntData` value.
- `Analysis::apply_switch_int_edge_effect` operates on the edge.
I.e. control flow goes from the engine, to the analysis, to the
engine, to the analysis.
Added:
- The `Analysis::SwitchIntData` assoc type and the
`Analysis::get_switch_int_data` method. Both only need to be
defined by analyses that look at `SwitchInt` terminators.
- The `MaybePlacesSwitchIntData` struct, which has three fields.
Changes:
- `Analysis::apply_switch_int_edge_effects` becomes
`Analysis::apply_switch_int_edge_effect`, which is a little simpler
because it's dealing with a single edge instead of all edges.
Removed:
- The `SwitchIntEdgeEffects` trait, and its two impls:
`BackwardSwitchIntEdgeEffectsApplier` (which has six fields) and
`ForwardSwitchIntEdgeEffectsApplier` structs (which has four fields).
- The closure.
The new structure is more concise and simpler.
The words "before" and "after" have an obvious temporal meaning, e.g.
`seek_before_primary_effect`,
`visit_statement_{before,after}_primary_effect`. But "before" is also
used to name the effect that occurs before the primary effect of a
statement/terminator; this is `Effect::Before`. This leads to the
confusing possibility of talking about things happening "before/after
the before event".
This commit removes this awkward overloading of "before" by renaming
`Effect::Before` as `Effect::Early`. It also renames some of the
`Analysis` and `ResultsVisitor` methods to be more consistent.
Here are the before and after names:
- `Effect::{Before,Primary}` -> `Effect::{Early,Primary}`
- `apply_before_statement_effect` -> `apply_early_statement_effect`
- `apply_statement_effect` -> `apply_primary_statement_effect`
- `visit_statement_before_primary_effect` -> `visit_after_early_statement_effect`
- `visit_statement_after_primary_effect` -> `visit_after_primary_statement_effect`
(And s/statement/terminator/ for all the terminator events.)
Currently they are called (most common) `state`, or `trans`, or (rare)
`on_entry`. I think `trans` is short for "transfer function", which
perhaps made more sense when `GenKillAnalysis` existed. Using `state`
everywhere now is more consistent.
They are only present because it's currently defined in terms of the
domains of `Borrows` and `MaybeUninitializedPlaces` and
`EverInitializedPlaces` via associated types. This commit introduces
typedefs for those domains, avoiding the lifetimes.
The part about zero-sized structures is totally wrong. The rest of
it has almost no explanatory value; there are better explanations in
comments elsewhere.
It's a performance win because `MixedBitSet` is faster and uses less
memory than `ChunkedBitSet`.
Also reflow some overlong comment lines in
`lint_tail_expr_drop_order.rs`.
Remove the `DefinitelyInitializedPlaces` analysis.
Its only use is in the `tests/ui/mir-dataflow/def_inits-1.rs` where it is tested via `rustc_peek_definite_init`.
Also, it's probably buggy. It's supposed to be the inverse of `MaybeUninitializedPlaces`, and it mostly is, except that `apply_terminator_effect` is a little different, and `apply_switch_int_edge_effects` is missing. Unlike `MaybeUninitializedPlaces`, which is used extensively in borrow checking, any bugs in `DefinitelyInitializedPlaces` are easy to overlook because it is only used in one small test.
This commit removes the analysis. It also removes
`rustc_peek_definite_init`, `Dual` and `MeetSemiLattice`, all of which are no longer needed.
r? ``@cjgillot``
In `MaybeRequiresStorage::apply_before_statement_effect`, call
`transfer_function` directly, as is already done in
`MaybeRequiresStorage::apply_before_terminator_effect`. This makes it clear
that the operation doesn't rely on the `MaybeBorrowedLocals` results.
bors