Auto merge of #147791 - cuviper:beta-next, r=cuviper

[beta] backports

- Change int-to-ptr transmute lowering back to inttoptr #147541
- rewrite outlives placeholder constraints to outlives static when handling opaque types #147566
- GVN: Invalidate derefs at loop headers #147607

r? cuviper

Author: bors

compiler/rustc_borrowck/src/handle_placeholders.rs

@@ -348,7 +348,7 @@ pub(crate) fn compute_sccs_applying_placeholder_outlives_constraints<'tcx>(
     }
 }
 
-fn rewrite_placeholder_outlives<'tcx>(
+pub(crate) fn rewrite_placeholder_outlives<'tcx>(
     sccs: &Sccs<RegionVid, ConstraintSccIndex>,
     annotations: &SccAnnotations<'_, '_, RegionTracker>,
     fr_static: RegionVid,

compiler/rustc_borrowck/src/region_infer/opaque_types/member_constraints.rs

@@ -39,7 +39,7 @@ pub(super) fn apply_member_constraints<'tcx>(
     debug!(?member_constraints);
     for scc_a in rcx.constraint_sccs.all_sccs() {
         debug!(?scc_a);
-        // Start by  adding the region values required by outlives constraints. This
+        // Start by adding the region values required by outlives constraints. This
         // matches how we compute the final region values in `fn compute_regions`.
         //
         // We need to do this here to get a lower bound when applying member constraints.
@@ -64,6 +64,7 @@ fn apply_member_constraint<'tcx>(
     // If the member region lives in a higher universe, we currently choose
     // the most conservative option by leaving it unchanged.
     if !rcx.max_placeholder_universe_reached(member).is_root() {
+        debug!("member region reached non root universe, bailing");
         return;
     }
 

compiler/rustc_borrowck/src/region_infer/opaque_types/mod.rs

@@ -254,6 +254,10 @@ fn collect_defining_uses<'tcx>(
                 }
             } else {
                 errors.push(DeferredOpaqueTypeError::InvalidOpaqueTypeArgs(err));
+                debug!(
+                    "collect_defining_uses: InvalidOpaqueTypeArgs for {:?} := {:?}",
+                    non_nll_opaque_type_key, hidden_type
+                );
             }
             continue;
         }
@@ -277,6 +281,7 @@ fn collect_defining_uses<'tcx>(
     defining_uses
 }
 
+#[instrument(level = "debug", skip(rcx, concrete_opaque_types, defining_uses, errors))]
 fn compute_concrete_types_from_defining_uses<'tcx>(
     rcx: &RegionCtxt<'_, 'tcx>,
     concrete_opaque_types: &mut ConcreteOpaqueTypes<'tcx>,
@@ -288,13 +293,15 @@ fn compute_concrete_types_from_defining_uses<'tcx>(
     let mut decls_modulo_regions: FxIndexMap<OpaqueTypeKey<'tcx>, (OpaqueTypeKey<'tcx>, Span)> =
         FxIndexMap::default();
     for &DefiningUse { opaque_type_key, ref arg_regions, hidden_type } in defining_uses {
+        debug!(?opaque_type_key, ?arg_regions, ?hidden_type);
         // After applying member constraints, we now map all regions in the hidden type
         // to the `arg_regions` of this defining use. In case a region in the hidden type
         // ended up not being equal to any such region, we error.
         let hidden_type =
             match hidden_type.try_fold_with(&mut ToArgRegionsFolder::new(rcx, arg_regions)) {
                 Ok(hidden_type) => hidden_type,
                 Err(r) => {
+                    debug!("UnexpectedHiddenRegion: {:?}", r);
                     errors.push(DeferredOpaqueTypeError::UnexpectedHiddenRegion {
                         hidden_type,
                         opaque_type_key,

compiler/rustc_borrowck/src/region_infer/opaque_types/region_ctxt.rs

@@ -11,7 +11,9 @@ use crate::constraints::ConstraintSccIndex;
 use crate::handle_placeholders::{SccAnnotations, region_definitions};
 use crate::region_infer::reverse_sccs::ReverseSccGraph;
 use crate::region_infer::values::RegionValues;
-use crate::region_infer::{ConstraintSccs, RegionDefinition, RegionTracker, Representative};
+use crate::region_infer::{
+    ConstraintSccs, OutlivesConstraintSet, RegionDefinition, RegionTracker, Representative,
+};
 use crate::type_check::MirTypeckRegionConstraints;
 use crate::type_check::free_region_relations::UniversalRegionRelations;
 use crate::universal_regions::UniversalRegions;
@@ -39,16 +41,36 @@ impl<'a, 'tcx> RegionCtxt<'a, 'tcx> {
         location_map: Rc<DenseLocationMap>,
         constraints: &MirTypeckRegionConstraints<'tcx>,
     ) -> RegionCtxt<'a, 'tcx> {
+        let mut outlives_constraints = constraints.outlives_constraints.clone();
         let universal_regions = &universal_region_relations.universal_regions;
         let (definitions, _has_placeholders) = region_definitions(infcx, universal_regions);
+
+        let compute_sccs =
+            |outlives_constraints: &OutlivesConstraintSet<'tcx>,
+             annotations: &mut SccAnnotations<'_, 'tcx, RegionTracker>| {
+                ConstraintSccs::new_with_annotation(
+                    &outlives_constraints
+                        .graph(definitions.len())
+                        .region_graph(outlives_constraints, universal_regions.fr_static),
+                    annotations,
+                )
+            };
+
         let mut scc_annotations = SccAnnotations::init(&definitions);
-        let constraint_sccs = ConstraintSccs::new_with_annotation(
-            &constraints
-                .outlives_constraints
-                .graph(definitions.len())
-                .region_graph(&constraints.outlives_constraints, universal_regions.fr_static),
-            &mut scc_annotations,
+        let mut constraint_sccs = compute_sccs(&outlives_constraints, &mut scc_annotations);
+
+        let added_constraints = crate::handle_placeholders::rewrite_placeholder_outlives(
+            &constraint_sccs,
+            &scc_annotations,
+            universal_regions.fr_static,
+            &mut outlives_constraints,
         );
+
+        if added_constraints {
+            scc_annotations = SccAnnotations::init(&definitions);
+            constraint_sccs = compute_sccs(&outlives_constraints, &mut scc_annotations);
+        }
+
         let scc_annotations = scc_annotations.scc_to_annotation;
 
         // Unlike the `RegionInferenceContext`, we only care about free regions

compiler/rustc_codegen_ssa/src/mir/rvalue.rs

@@ -1048,14 +1048,14 @@ pub(super) fn transmute_scalar<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
     imm = match (from_scalar.primitive(), to_scalar.primitive()) {
         (Int(..) | Float(_), Int(..) | Float(_)) => bx.bitcast(imm, to_backend_ty),
         (Pointer(..), Pointer(..)) => bx.pointercast(imm, to_backend_ty),
-        (Int(..), Pointer(..)) => bx.ptradd(bx.const_null(bx.type_ptr()), imm),
+        (Int(..), Pointer(..)) => bx.inttoptr(imm, to_backend_ty),
         (Pointer(..), Int(..)) => {
             // FIXME: this exposes the provenance, which shouldn't be necessary.
             bx.ptrtoint(imm, to_backend_ty)
         }
         (Float(_), Pointer(..)) => {
             let int_imm = bx.bitcast(imm, bx.cx().type_isize());
-            bx.ptradd(bx.const_null(bx.type_ptr()), int_imm)
+            bx.inttoptr(int_imm, to_backend_ty)
         }
         (Pointer(..), Float(_)) => {
             // FIXME: this exposes the provenance, which shouldn't be necessary.

compiler/rustc_middle/src/mir/loops.rs

@@ -0,0 +1,29 @@
+use rustc_index::bit_set::DenseBitSet;
+
+use super::*;
+
+/// Compute the set of loop headers in the given body. A loop header is usually defined as a block
+/// which dominates one of its predecessors. This definition is only correct for reducible CFGs.
+/// However, computing dominators is expensive, so we approximate according to the post-order
+/// traversal order. A loop header for us is a block which is visited after its predecessor in
+/// post-order. This is ok as we mostly need a heuristic.
+pub fn maybe_loop_headers(body: &Body<'_>) -> DenseBitSet<BasicBlock> {
+    let mut maybe_loop_headers = DenseBitSet::new_empty(body.basic_blocks.len());
+    let mut visited = DenseBitSet::new_empty(body.basic_blocks.len());
+    for (bb, bbdata) in traversal::postorder(body) {
+        // Post-order means we visit successors before the block for acyclic CFGs.
+        // If the successor is not visited yet, consider it a loop header.
+        for succ in bbdata.terminator().successors() {
+            if !visited.contains(succ) {
+                maybe_loop_headers.insert(succ);
+            }
+        }
+
+        // Only mark `bb` as visited after we checked the successors, in case we have a self-loop.
+        //     bb1: goto -> bb1;
+        let _new = visited.insert(bb);
+        debug_assert!(_new);
+    }
+
+    maybe_loop_headers
+}

compiler/rustc_middle/src/mir/mod.rs

@@ -51,6 +51,7 @@ mod statement;
 mod syntax;
 mod terminator;
 
+pub mod loops;
 pub mod traversal;
 pub mod visit;
 

compiler/rustc_mir_transform/src/gvn.rs

@@ -126,6 +126,7 @@ impl<'tcx> crate::MirPass<'tcx> for GVN {
         let ssa = SsaLocals::new(tcx, body, typing_env);
         // Clone dominators because we need them while mutating the body.
         let dominators = body.basic_blocks.dominators().clone();
+        let maybe_loop_headers = loops::maybe_loop_headers(body);
 
         let mut state = VnState::new(tcx, body, typing_env, &ssa, dominators, &body.local_decls);
 
@@ -136,6 +137,11 @@ impl<'tcx> crate::MirPass<'tcx> for GVN {
 
         let reverse_postorder = body.basic_blocks.reverse_postorder().to_vec();
         for bb in reverse_postorder {
+            // N.B. With loops, reverse postorder cannot produce a valid topological order.
+            // A statement or terminator from inside the loop, that is not processed yet, may have performed an indirect write.
+            if maybe_loop_headers.contains(bb) {
+                state.invalidate_derefs();
+            }
             let data = &mut body.basic_blocks.as_mut_preserves_cfg()[bb];
             state.visit_basic_block_data(bb, data);
         }

compiler/rustc_mir_transform/src/jump_threading.rs

@@ -84,7 +84,7 @@ impl<'tcx> crate::MirPass<'tcx> for JumpThreading {
             body,
             arena,
             map: Map::new(tcx, body, Some(MAX_PLACES)),
-            loop_headers: loop_headers(body),
+            loop_headers: loops::maybe_loop_headers(body),
             opportunities: Vec::new(),
         };
 
@@ -832,21 +832,3 @@ enum Update {
     Incr,
     Decr,
 }
-
-/// Compute the set of loop headers in the given body. We define a loop header as a block which has
-/// at least a predecessor which it dominates. This definition is only correct for reducible CFGs.
-/// But if the CFG is already irreducible, there is no point in trying much harder.
-/// is already irreducible.
-fn loop_headers(body: &Body<'_>) -> DenseBitSet<BasicBlock> {
-    let mut loop_headers = DenseBitSet::new_empty(body.basic_blocks.len());
-    let dominators = body.basic_blocks.dominators();
-    // Only visit reachable blocks.
-    for (bb, bbdata) in traversal::preorder(body) {
-        for succ in bbdata.terminator().successors() {
-            if dominators.dominates(succ, bb) {
-                loop_headers.insert(succ);
-            }
-        }
-    }
-    loop_headers
-}

tests/codegen-llvm/common_prim_int_ptr.rs

@@ -11,7 +11,7 @@
 #[no_mangle]
 pub fn insert_int(x: usize) -> Result<usize, Box<()>> {
     // CHECK: start:
-    // CHECK-NEXT: %[[WO_PROV:.+]] = getelementptr i8, ptr null, [[USIZE:i[0-9]+]] %x
+    // CHECK-NEXT: %[[WO_PROV:.+]] = inttoptr [[USIZE:i[0-9]+]] %x to ptr
     // CHECK-NEXT: %[[R:.+]] = insertvalue { [[USIZE]], ptr } { [[USIZE]] 0, ptr poison }, ptr %[[WO_PROV]], 1
     // CHECK-NEXT: ret { [[USIZE]], ptr } %[[R]]
     Ok(x)