Rollup merge of #107411 - cjgillot:dataflow-discriminant, r=oli-obk

Handle discriminant in DataflowConstProp

cc ``@jachris``
r? ``@JakobDegen``

This PR attempts to extend the DataflowConstProp pass to handle propagation of discriminants. We handle this by adding 2 new variants to `TrackElem`: `TrackElem::Variant` for enum variants and `TrackElem::Discriminant` for the enum discriminant pseudo-place.

The difficulty is that the enum discriminant and enum variants may alias each another. This is the issue of the `Option<NonZeroUsize>` test, which is the equivalent of rust-lang/unsafe-code-guidelines#84 with a direct write.

To handle that, we generalize the flood process to flood all the potentially aliasing places. In particular:
- any write to `(PLACE as Variant)`, either direct or through a projection, floods `(PLACE as OtherVariant)` for all other variants and `discriminant(PLACE)`;
- `SetDiscriminant(PLACE)` floods `(PLACE as Variant)` for each variant.

This implies that flooding is not hierarchical any more, and that an assignment to a non-tracked place may need to flood a tracked place. This is handled by `for_each_aliasing_place` which generalizes `preorder_invoke`.

As we deaggregate enums by putting `SetDiscriminant` last, this allows to propagate the value of the discriminant.

This refactor will allow to make #107009 able to handle discriminants too.

compiler/rustc_middle/src/mir/mod.rs

@@ -1642,6 +1642,14 @@ impl<'tcx> PlaceRef<'tcx> {
         }
     }
 
+    /// Returns `true` if this `Place` contains a `Deref` projection.
+    ///
+    /// If `Place::is_indirect` returns false, the caller knows that the `Place` refers to the
+    /// same region of memory as its base.
+    pub fn is_indirect(&self) -> bool {
+        self.projection.iter().any(|elem| elem.is_indirect())
+    }
+
     /// If MirPhase >= Derefered and if projection contains Deref,
     /// It's guaranteed to be in the first place
     pub fn has_deref(&self) -> bool {

compiler/rustc_mir_dataflow/src/impls/borrowed_locals.rs

@@ -121,7 +121,9 @@ where
                 // for now. See discussion on [#61069].
                 //
                 // [#61069]: https://github.com/rust-lang/rust/pull/61069
-                self.trans.gen(dropped_place.local);
+                if !dropped_place.is_indirect() {
+                    self.trans.gen(dropped_place.local);
+                }
             }
 
             TerminatorKind::Abort

compiler/rustc_mir_dataflow/src/lib.rs

@@ -1,6 +1,7 @@
 #![feature(associated_type_defaults)]
 #![feature(box_patterns)]
 #![feature(exact_size_is_empty)]
+#![feature(let_chains)]
 #![feature(min_specialization)]
 #![feature(once_cell)]
 #![feature(stmt_expr_attributes)]

compiler/rustc_mir_transform/src/dataflow_const_prop.rs

@@ -13,6 +13,7 @@ use rustc_mir_dataflow::value_analysis::{Map, State, TrackElem, ValueAnalysis, V
 use rustc_mir_dataflow::{lattice::FlatSet, Analysis, ResultsVisitor, SwitchIntEdgeEffects};
 use rustc_span::DUMMY_SP;
 use rustc_target::abi::Align;
+use rustc_target::abi::VariantIdx;
 
 use crate::MirPass;
 
@@ -30,14 +31,12 @@ impl<'tcx> MirPass<'tcx> for DataflowConstProp {
 
     #[instrument(skip_all level = "debug")]
     fn run_pass(&self, tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>) {
+        debug!(def_id = ?body.source.def_id());
         if tcx.sess.mir_opt_level() < 4 && body.basic_blocks.len() > BLOCK_LIMIT {
             debug!("aborted dataflow const prop due too many basic blocks");
             return;
         }
 
-        // Decide which places to track during the analysis.
-        let map = Map::from_filter(tcx, body, Ty::is_scalar);
-
         // We want to have a somewhat linear runtime w.r.t. the number of statements/terminators.
         // Let's call this number `n`. Dataflow analysis has `O(h*n)` transfer function
         // applications, where `h` is the height of the lattice. Because the height of our lattice
@@ -46,10 +45,10 @@ impl<'tcx> MirPass<'tcx> for DataflowConstProp {
         // `O(num_nodes * tracked_places * n)` in terms of time complexity. Since the number of
         // map nodes is strongly correlated to the number of tracked places, this becomes more or
         // less `O(n)` if we place a constant limit on the number of tracked places.
-        if tcx.sess.mir_opt_level() < 4 && map.tracked_places() > PLACE_LIMIT {
-            debug!("aborted dataflow const prop due to too many tracked places");
-            return;
-        }
+        let place_limit = if tcx.sess.mir_opt_level() < 4 { Some(PLACE_LIMIT) } else { None };
+
+        // Decide which places to track during the analysis.
+        let map = Map::from_filter(tcx, body, Ty::is_scalar, place_limit);
 
         // Perform the actual dataflow analysis.
         let analysis = ConstAnalysis::new(tcx, body, map);
@@ -63,14 +62,31 @@ impl<'tcx> MirPass<'tcx> for DataflowConstProp {
     }
 }
 
-struct ConstAnalysis<'tcx> {
+struct ConstAnalysis<'a, 'tcx> {
     map: Map,
     tcx: TyCtxt<'tcx>,
+    local_decls: &'a LocalDecls<'tcx>,
     ecx: InterpCx<'tcx, 'tcx, DummyMachine>,
     param_env: ty::ParamEnv<'tcx>,
 }
 
-impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'tcx> {
+impl<'tcx> ConstAnalysis<'_, 'tcx> {
+    fn eval_discriminant(
+        &self,
+        enum_ty: Ty<'tcx>,
+        variant_index: VariantIdx,
+    ) -> Option<ScalarTy<'tcx>> {
+        if !enum_ty.is_enum() {
+            return None;
+        }
+        let discr = enum_ty.discriminant_for_variant(self.tcx, variant_index)?;
+        let discr_layout = self.tcx.layout_of(self.param_env.and(discr.ty)).ok()?;
+        let discr_value = Scalar::try_from_uint(discr.val, discr_layout.size)?;
+        Some(ScalarTy(discr_value, discr.ty))
+    }
+}
+
+impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'_, 'tcx> {
     type Value = FlatSet<ScalarTy<'tcx>>;
 
     const NAME: &'static str = "ConstAnalysis";
@@ -79,6 +95,25 @@ impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'tcx> {
         &self.map
     }
 
+    fn handle_statement(&self, statement: &Statement<'tcx>, state: &mut State<Self::Value>) {
+        match statement.kind {
+            StatementKind::SetDiscriminant { box ref place, variant_index } => {
+                state.flood_discr(place.as_ref(), &self.map);
+                if self.map.find_discr(place.as_ref()).is_some() {
+                    let enum_ty = place.ty(self.local_decls, self.tcx).ty;
+                    if let Some(discr) = self.eval_discriminant(enum_ty, variant_index) {
+                        state.assign_discr(
+                            place.as_ref(),
+                            ValueOrPlace::Value(FlatSet::Elem(discr)),
+                            &self.map,
+                        );
+                    }
+                }
+            }
+            _ => self.super_statement(statement, state),
+        }
+    }
+
     fn handle_assign(
         &self,
         target: Place<'tcx>,
@@ -87,36 +122,47 @@ impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'tcx> {
     ) {
         match rvalue {
             Rvalue::Aggregate(kind, operands) => {
-                let target = self.map().find(target.as_ref());
-                if let Some(target) = target {
-                    state.flood_idx_with(target, self.map(), FlatSet::Bottom);
-                    let field_based = match **kind {
-                        AggregateKind::Tuple | AggregateKind::Closure(..) => true,
-                        AggregateKind::Adt(def_id, ..) => {
-                            matches!(self.tcx.def_kind(def_id), DefKind::Struct)
+                state.flood_with(target.as_ref(), self.map(), FlatSet::Bottom);
+                if let Some(target_idx) = self.map().find(target.as_ref()) {
+                    let (variant_target, variant_index) = match **kind {
+                        AggregateKind::Tuple | AggregateKind::Closure(..) => {
+                            (Some(target_idx), None)
                         }
-                        _ => false,
+                        AggregateKind::Adt(def_id, variant_index, ..) => {
+                            match self.tcx.def_kind(def_id) {
+                                DefKind::Struct => (Some(target_idx), None),
+                                DefKind::Enum => (Some(target_idx), Some(variant_index)),
+                                _ => (None, None),
+                            }
+                        }
+                        _ => (None, None),
                     };
-                    if field_based {
+                    if let Some(target) = variant_target {
                         for (field_index, operand) in operands.iter().enumerate() {
                             if let Some(field) = self
                                 .map()
                                 .apply(target, TrackElem::Field(Field::from_usize(field_index)))
                             {
                                 let result = self.handle_operand(operand, state);
-                                state.assign_idx(field, result, self.map());
+                                state.insert_idx(field, result, self.map());
                             }
                         }
                     }
+                    if let Some(variant_index) = variant_index
+                        && let Some(discr_idx) = self.map().apply(target_idx, TrackElem::Discriminant)
+                    {
+                        let enum_ty = target.ty(self.local_decls, self.tcx).ty;
+                        if let Some(discr_val) = self.eval_discriminant(enum_ty, variant_index) {
+                            state.insert_value_idx(discr_idx, FlatSet::Elem(discr_val), &self.map);
+                        }
+                    }
                 }
             }
             Rvalue::CheckedBinaryOp(op, box (left, right)) => {
+                // Flood everything now, so we can use `insert_value_idx` directly later.
+                state.flood(target.as_ref(), self.map());
+
                 let target = self.map().find(target.as_ref());
-                if let Some(target) = target {
-                    // We should not track any projections other than
-                    // what is overwritten below, but just in case...
-                    state.flood_idx(target, self.map());
-                }
 
                 let value_target = target
                     .and_then(|target| self.map().apply(target, TrackElem::Field(0_u32.into())));
@@ -127,7 +173,8 @@ impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'tcx> {
                     let (val, overflow) = self.binary_op(state, *op, left, right);
 
                     if let Some(value_target) = value_target {
-                        state.assign_idx(value_target, ValueOrPlace::Value(val), self.map());
+                        // We have flooded `target` earlier.
+                        state.insert_value_idx(value_target, val, self.map());
                     }
                     if let Some(overflow_target) = overflow_target {
                         let overflow = match overflow {
@@ -142,11 +189,8 @@ impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'tcx> {
                             }
                             FlatSet::Bottom => FlatSet::Bottom,
                         };
-                        state.assign_idx(
-                            overflow_target,
-                            ValueOrPlace::Value(overflow),
-                            self.map(),
-                        );
+                        // We have flooded `target` earlier.
+                        state.insert_value_idx(overflow_target, overflow, self.map());
                     }
                 }
             }
@@ -195,6 +239,9 @@ impl<'tcx> ValueAnalysis<'tcx> for ConstAnalysis<'tcx> {
                 FlatSet::Bottom => ValueOrPlace::Value(FlatSet::Bottom),
                 FlatSet::Top => ValueOrPlace::Value(FlatSet::Top),
             },
+            Rvalue::Discriminant(place) => {
+                ValueOrPlace::Value(state.get_discr(place.as_ref(), self.map()))
+            }
             _ => self.super_rvalue(rvalue, state),
         }
     }
@@ -268,12 +315,13 @@ impl<'tcx> std::fmt::Debug for ScalarTy<'tcx> {
     }
 }
 
-impl<'tcx> ConstAnalysis<'tcx> {
-    pub fn new(tcx: TyCtxt<'tcx>, body: &Body<'tcx>, map: Map) -> Self {
+impl<'a, 'tcx> ConstAnalysis<'a, 'tcx> {
+    pub fn new(tcx: TyCtxt<'tcx>, body: &'a Body<'tcx>, map: Map) -> Self {
         let param_env = tcx.param_env(body.source.def_id());
         Self {
             map,
             tcx,
+            local_decls: &body.local_decls,
             ecx: InterpCx::new(tcx, DUMMY_SP, param_env, DummyMachine),
             param_env: param_env,
         }
@@ -466,6 +514,21 @@ impl<'tcx, 'map, 'a> Visitor<'tcx> for OperandCollector<'tcx, 'map, 'a> {
             _ => (),
         }
     }
+
+    fn visit_rvalue(&mut self, rvalue: &Rvalue<'tcx>, location: Location) {
+        match rvalue {
+            Rvalue::Discriminant(place) => {
+                match self.state.get_discr(place.as_ref(), self.visitor.map) {
+                    FlatSet::Top => (),
+                    FlatSet::Elem(value) => {
+                        self.visitor.before_effect.insert((location, *place), value);
+                    }
+                    FlatSet::Bottom => (),
+                }
+            }
+            _ => self.super_rvalue(rvalue, location),
+        }
+    }
 }
 
 struct DummyMachine;

compiler/rustc_mir_transform/src/sroa.rs

@@ -1,5 +1,5 @@
 use crate::MirPass;
-use rustc_index::bit_set::BitSet;
+use rustc_index::bit_set::{BitSet, GrowableBitSet};
 use rustc_index::vec::IndexVec;
 use rustc_middle::mir::patch::MirPatch;
 use rustc_middle::mir::visit::*;
@@ -26,10 +26,12 @@ impl<'tcx> MirPass<'tcx> for ScalarReplacementOfAggregates {
             debug!(?replacements);
             let all_dead_locals = replace_flattened_locals(tcx, body, replacements);
             if !all_dead_locals.is_empty() {
-                for local in excluded.indices() {
-                    excluded[local] |= all_dead_locals.contains(local);
-                }
-                excluded.raw.resize(body.local_decls.len(), false);
+                excluded.union(&all_dead_locals);
+                excluded = {
+                    let mut growable = GrowableBitSet::from(excluded);
+                    growable.ensure(body.local_decls.len());
+                    growable.into()
+                };
             } else {
                 break;
             }
@@ -44,11 +46,11 @@ impl<'tcx> MirPass<'tcx> for ScalarReplacementOfAggregates {
 /// - the locals is a union or an enum;
 /// - the local's address is taken, and thus the relative addresses of the fields are observable to
 ///   client code.
-fn escaping_locals(excluded: &IndexVec<Local, bool>, body: &Body<'_>) -> BitSet<Local> {
+fn escaping_locals(excluded: &BitSet<Local>, body: &Body<'_>) -> BitSet<Local> {
     let mut set = BitSet::new_empty(body.local_decls.len());
     set.insert_range(RETURN_PLACE..=Local::from_usize(body.arg_count));
     for (local, decl) in body.local_decls().iter_enumerated() {
-        if decl.ty.is_union() || decl.ty.is_enum() || excluded[local] {
+        if decl.ty.is_union() || decl.ty.is_enum() || excluded.contains(local) {
             set.insert(local);
         }
     }
@@ -172,7 +174,7 @@ fn replace_flattened_locals<'tcx>(
     body: &mut Body<'tcx>,
     replacements: ReplacementMap<'tcx>,
 ) -> BitSet<Local> {
-    let mut all_dead_locals = BitSet::new_empty(body.local_decls.len());
+    let mut all_dead_locals = BitSet::new_empty(replacements.fragments.len());
     for (local, replacements) in replacements.fragments.iter_enumerated() {
         if replacements.is_some() {
             all_dead_locals.insert(local);

tests/mir-opt/dataflow-const-prop/enum.mutate_discriminant.DataflowConstProp.diff

@@ -0,0 +1,26 @@
+- // MIR for `mutate_discriminant` before DataflowConstProp
++ // MIR for `mutate_discriminant` after DataflowConstProp
+
+  fn mutate_discriminant() -> u8 {
+      let mut _0: u8;                      // return place in scope 0 at $DIR/enum.rs:+0:29: +0:31
+      let mut _1: std::option::Option<NonZeroUsize>; // in scope 0 at $SRC_DIR/core/src/intrinsics/mir.rs:LL:COL
+      let mut _2: isize;                   // in scope 0 at $SRC_DIR/core/src/intrinsics/mir.rs:LL:COL
+
+      bb0: {
+          discriminant(_1) = 1;            // scope 0 at $DIR/enum.rs:+4:13: +4:34
+          (((_1 as variant#1).0: NonZeroUsize).0: usize) = const 0_usize; // scope 0 at $DIR/enum.rs:+6:13: +6:64
+          _2 = discriminant(_1);           // scope 0 at $SRC_DIR/core/src/intrinsics/mir.rs:LL:COL
+          switchInt(_2) -> [0: bb1, otherwise: bb2]; // scope 0 at $DIR/enum.rs:+9:13: +12:14
+      }
+
+      bb1: {
+          _0 = const 1_u8;                 // scope 0 at $DIR/enum.rs:+15:13: +15:20
+          return;                          // scope 0 at $DIR/enum.rs:+16:13: +16:21
+      }
+
+      bb2: {
+          _0 = const 2_u8;                 // scope 0 at $DIR/enum.rs:+19:13: +19:20
+          unreachable;                     // scope 0 at $DIR/enum.rs:+20:13: +20:26
+      }
+  }
+

tests/mir-opt/dataflow-const-prop/enum.rs

@@ -1,13 +1,52 @@
 // unit-test: DataflowConstProp
 
-// Not trackable, because variants could be aliased.
+#![feature(custom_mir, core_intrinsics, rustc_attrs)]
+
+use std::intrinsics::mir::*;
+
 enum E {
     V1(i32),
     V2(i32)
 }
 
-// EMIT_MIR enum.main.DataflowConstProp.diff
-fn main() {
+// EMIT_MIR enum.simple.DataflowConstProp.diff
+fn simple() {
     let e = E::V1(0);
     let x = match e { E::V1(x) => x, E::V2(x) => x };
 }
+
+#[rustc_layout_scalar_valid_range_start(1)]
+#[rustc_nonnull_optimization_guaranteed]
+struct NonZeroUsize(usize);
+
+// EMIT_MIR enum.mutate_discriminant.DataflowConstProp.diff
+#[custom_mir(dialect = "runtime", phase = "post-cleanup")]
+fn mutate_discriminant() -> u8 {
+    mir!(
+        let x: Option<NonZeroUsize>;
+        {
+            SetDiscriminant(x, 1);
+            // This assignment overwrites the niche in which the discriminant is stored.
+            place!(Field(Field(Variant(x, 1), 0), 0)) = 0_usize;
+            // So we cannot know the value of this discriminant.
+            let a = Discriminant(x);
+            match a {
+                0 => bb1,
+                _ => bad,
+            }
+        }
+        bb1 = {
+            RET = 1;
+            Return()
+        }
+        bad = {
+            RET = 2;
+            Unreachable()
+        }
+    )
+}
+
+fn main() {
+    simple();
+    mutate_discriminant();
+}