[TOY] Run MIR opts before coroutine state transform

compiler/rustc_middle/src/mir/mod.rs

@@ -471,7 +471,7 @@ impl<'tcx> Body<'tcx> {
 
     /// Returns an iterator over all function arguments.
     #[inline]
-    pub fn args_iter(&self) -> impl Iterator<Item = Local> + ExactSizeIterator {
+    pub fn args_iter(&self) -> impl Iterator<Item = Local> + ExactSizeIterator + use<> {
         (1..self.arg_count + 1).map(Local::new)
     }
 

compiler/rustc_middle/src/mir/query.rs

@@ -17,7 +17,7 @@ use crate::ty::{self, CoroutineArgsExt, OpaqueHiddenType, Ty};
 rustc_index::newtype_index! {
     #[derive(HashStable)]
     #[encodable]
-    #[debug_format = "_{}"]
+    #[debug_format = "_s{}"]
     pub struct CoroutineSavedLocal {}
 }
 

compiler/rustc_mir_transform/src/coroutine/drop.rs

@@ -126,6 +126,7 @@ fn build_pin_fut<'tcx>(
 //     Ready() => ready_block
 //     Pending => yield_block
 //}
+#[tracing::instrument(level = "trace", skip(tcx, body), ret)]
 fn build_poll_switch<'tcx>(
     tcx: TyCtxt<'tcx>,
     body: &mut Body<'tcx>,
@@ -179,6 +180,7 @@ fn build_poll_switch<'tcx>(
 }
 
 // Gather blocks, reachable through 'drop' targets of Yield and Drop terminators (chained)
+#[tracing::instrument(level = "trace", skip(body), ret)]
 fn gather_dropline_blocks<'tcx>(body: &mut Body<'tcx>) -> DenseBitSet<BasicBlock> {
     let mut dropline: DenseBitSet<BasicBlock> = DenseBitSet::new_empty(body.basic_blocks.len());
     for (bb, data) in traversal::reverse_postorder(body) {
@@ -249,6 +251,7 @@ pub(super) fn has_expandable_async_drops<'tcx>(
 }
 
 /// Expand Drop terminator for async drops into mainline poll-switch and dropline poll-switch
+#[tracing::instrument(level = "trace", skip(tcx, body), ret)]
 pub(super) fn expand_async_drops<'tcx>(
     tcx: TyCtxt<'tcx>,
     body: &mut Body<'tcx>,
@@ -259,6 +262,7 @@ pub(super) fn expand_async_drops<'tcx>(
     let dropline = gather_dropline_blocks(body);
     // Clean drop and async_fut fields if potentially async drop is not expanded (stays sync)
     let remove_asyncness = |block: &mut BasicBlockData<'tcx>| {
+        tracing::trace!("remove_asyncness");
         if let TerminatorKind::Drop {
             place: _,
             target: _,
@@ -273,6 +277,7 @@ pub(super) fn expand_async_drops<'tcx>(
         }
     };
     for bb in START_BLOCK..body.basic_blocks.next_index() {
+        tracing::trace!(?bb);
         // Drops in unwind path (cleanup blocks) are not expanded to async drops, only sync drops in unwind path
         if body[bb].is_cleanup {
             remove_asyncness(&mut body[bb]);
@@ -461,6 +466,7 @@ pub(super) fn expand_async_drops<'tcx>(
     }
 }
 
+#[tracing::instrument(level = "trace", skip(tcx, body))]
 pub(super) fn elaborate_coroutine_drops<'tcx>(tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>) {
     use crate::elaborate_drop::{Unwind, elaborate_drop};
     use crate::patch::MirPatch;
@@ -519,6 +525,7 @@ pub(super) fn elaborate_coroutine_drops<'tcx>(tcx: TyCtxt<'tcx>, body: &mut Body
     elaborator.patch.apply(body);
 }
 
+#[tracing::instrument(level = "trace", skip(tcx, body), ret)]
 pub(super) fn insert_clean_drop<'tcx>(
     tcx: TyCtxt<'tcx>,
     body: &mut Body<'tcx>,
@@ -550,6 +557,7 @@ pub(super) fn insert_clean_drop<'tcx>(
         .push(BasicBlockData::new(Some(Terminator { source_info, kind: term }), false))
 }
 
+#[tracing::instrument(level = "trace", skip(tcx, transform, body))]
 pub(super) fn create_coroutine_drop_shim<'tcx>(
     tcx: TyCtxt<'tcx>,
     transform: &TransformVisitor<'tcx>,
@@ -621,6 +629,7 @@ pub(super) fn create_coroutine_drop_shim<'tcx>(
 }
 
 // Create async drop shim function to drop coroutine itself
+#[tracing::instrument(level = "trace", skip(tcx, transform, body))]
 pub(super) fn create_coroutine_drop_shim_async<'tcx>(
     tcx: TyCtxt<'tcx>,
     transform: &TransformVisitor<'tcx>,
@@ -676,12 +685,13 @@ pub(super) fn create_coroutine_drop_shim_async<'tcx>(
     let poll_enum = Ty::new_adt(tcx, poll_adt_ref, tcx.mk_args(&[tcx.types.unit.into()]));
     body.local_decls[RETURN_PLACE] = LocalDecl::with_source_info(poll_enum, source_info);
 
-    make_coroutine_state_argument_indirect(tcx, &mut body);
-
     match transform.coroutine_kind {
         // Iterator::next doesn't accept a pinned argument,
         // unlike for all other coroutine kinds.
-        CoroutineKind::Desugared(CoroutineDesugaring::Gen, _) => {}
+        CoroutineKind::Desugared(CoroutineDesugaring::Gen, _) => {
+            make_coroutine_state_argument_indirect(tcx, &mut body);
+        }
+
         _ => {
             make_coroutine_state_argument_pinned(tcx, &mut body);
         }

compiler/rustc_mir_transform/src/dataflow_const_prop.rs

@@ -46,6 +46,12 @@ impl<'tcx> crate::MirPass<'tcx> for DataflowConstProp {
             return;
         }
 
+        // Avoid computing layout inside coroutines, since their `optimized_mir` is used for layout
+        // computation, which can create a cycle.
+        if body.coroutine.is_some() {
+            return;
+        }
+
         // 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
@@ -241,9 +247,8 @@ impl<'a, 'tcx> ConstAnalysis<'a, 'tcx> {
             TerminatorKind::Drop { place, .. } => {
                 state.flood_with(place.as_ref(), &self.map, FlatSet::<Scalar>::BOTTOM);
             }
-            TerminatorKind::Yield { .. } => {
-                // They would have an effect, but are not allowed in this phase.
-                bug!("encountered disallowed terminator");
+            TerminatorKind::Yield { resume_arg, .. } => {
+                state.flood_with(resume_arg.as_ref(), &self.map, FlatSet::<Scalar>::BOTTOM);
             }
             TerminatorKind::SwitchInt { discr, targets } => {
                 return self.handle_switch_int(discr, targets, state);

compiler/rustc_mir_transform/src/gvn.rs

@@ -1730,14 +1730,18 @@ impl<'tcx> MutVisitor<'tcx> for VnState<'_, 'tcx> {
     }
 
     fn visit_terminator(&mut self, terminator: &mut Terminator<'tcx>, location: Location) {
-        if let Terminator { kind: TerminatorKind::Call { destination, .. }, .. } = terminator {
-            if let Some(local) = destination.as_local()
-                && self.ssa.is_ssa(local)
-            {
-                let ty = self.local_decls[local].ty;
-                let opaque = self.new_opaque(ty);
-                self.assign(local, opaque);
-            }
+        let destination = match terminator.kind {
+            TerminatorKind::Call { destination, .. } => Some(destination),
+            TerminatorKind::Yield { resume_arg, .. } => Some(resume_arg),
+            _ => None,
+        };
+        if let Some(destination) = destination
+            && let Some(local) = destination.as_local()
+            && self.ssa.is_ssa(local)
+        {
+            let ty = self.local_decls[local].ty;
+            let opaque = self.new_opaque(ty);
+            self.assign(local, opaque);
         }
         // Function calls and ASM may invalidate (nested) derefs. We must handle them carefully.
         // Currently, only preserving derefs for trivial terminators like SwitchInt and Goto.

compiler/rustc_mir_transform/src/jump_threading.rs

@@ -612,9 +612,9 @@ impl<'a, 'tcx> TOFinder<'a, 'tcx> {
             | TerminatorKind::Unreachable
             | TerminatorKind::CoroutineDrop => bug!("{term:?} has no terminators"),
             // Disallowed during optimizations.
-            TerminatorKind::FalseEdge { .. }
-            | TerminatorKind::FalseUnwind { .. }
-            | TerminatorKind::Yield { .. } => bug!("{term:?} invalid"),
+            TerminatorKind::FalseEdge { .. } | TerminatorKind::FalseUnwind { .. } => {
+                bug!("{term:?} invalid")
+            }
             // Cannot reason about inline asm.
             TerminatorKind::InlineAsm { .. } => return,
             // `SwitchInt` is handled specially.
@@ -623,6 +623,7 @@ impl<'a, 'tcx> TOFinder<'a, 'tcx> {
             TerminatorKind::Goto { .. } => None,
             // Flood the overwritten place, and progress through.
             TerminatorKind::Drop { place: destination, .. }
+            | TerminatorKind::Yield { resume_arg: destination, .. }
             | TerminatorKind::Call { destination, .. } => Some(destination),
             // Ignore, as this can be a no-op at codegen time.
             TerminatorKind::Assert { .. } => None,

compiler/rustc_mir_transform/src/lib.rs

@@ -80,7 +80,7 @@ mod ssa;
 macro_rules! declare_passes {
     (
         $(
-            $vis:vis mod $mod_name:ident : $($pass_name:ident $( { $($ident:ident),* } )?),+ $(,)?;
+            $vis:vis mod $mod_name:ident : $($pass_name:ident $( { $($ident:ident),* $(,)? } )?),+ $(,)?;
         )*
     ) => {
         $(
@@ -180,12 +180,14 @@ declare_passes! {
             PreOptimizations,
             Final,
             MakeShim,
-            AfterUnreachableEnumBranching
+            AfterUnreachableEnumBranching,
+            PostStateTransform,
         },
         SimplifyLocals {
             BeforeConstProp,
             AfterGVN,
-            Final
+            Final,
+            PostStateTransform,
         };
     mod simplify_branches : SimplifyConstCondition {
         AfterConstProp,
@@ -621,7 +623,6 @@ fn run_runtime_lowering_passes<'tcx>(tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>) {
         // Otherwise it should run fairly late, but before optimizations begin.
         &add_retag::AddRetag,
         &elaborate_box_derefs::ElaborateBoxDerefs,
-        &coroutine::StateTransform,
         &Lint(known_panics_lint::KnownPanicsLint),
     ];
     pm::run_passes_no_validate(tcx, body, passes, Some(MirPhase::Runtime(RuntimePhase::Initial)));
@@ -727,6 +728,7 @@ pub(crate) fn run_optimization_passes<'tcx>(tcx: TyCtxt<'tcx>, body: &mut Body<'
             &simplify::SimplifyLocals::Final,
             &multiple_return_terminators::MultipleReturnTerminators,
             &large_enums::EnumSizeOpt { discrepancy: 128 },
+            &coroutine::StateTransform,
             // Some cleanup necessary at least for LLVM and potentially other codegen backends.
             &add_call_guards::CriticalCallEdges,
             // Cleanup for human readability, off by default.

compiler/rustc_mir_transform/src/shim.rs

@@ -149,7 +149,6 @@ fn make_shim<'tcx>(tcx: TyCtxt<'tcx>, instance: ty::InstanceKind<'tcx>) -> Body<
                     tcx,
                     &mut body,
                     &[
-                        &mentioned_items::MentionedItems,
                         &abort_unwinding_calls::AbortUnwindingCalls,
                         &add_call_guards::CriticalCallEdges,
                     ],

compiler/rustc_mir_transform/src/shim/async_destructor_ctor.rs

@@ -216,6 +216,7 @@ fn build_adrop_for_coroutine_shim<'tcx>(
     body.source.instance = instance;
     body.phase = MirPhase::Runtime(RuntimePhase::Initial);
     body.var_debug_info.clear();
+    body.mentioned_items = None;
     let pin_adt_ref = tcx.adt_def(tcx.require_lang_item(LangItem::Pin, span));
     let args = tcx.mk_args(&[proxy_ref.into()]);
     let pin_proxy_ref = Ty::new_adt(tcx, pin_adt_ref, args);

compiler/rustc_mir_transform/src/simplify.rs

@@ -55,6 +55,8 @@ pub(super) enum SimplifyCfg {
     Final,
     MakeShim,
     AfterUnreachableEnumBranching,
+    /// Extra run introduced by `StateTransform`.
+    PostStateTransform,
 }
 
 impl SimplifyCfg {
@@ -70,6 +72,7 @@ impl SimplifyCfg {
             SimplifyCfg::AfterUnreachableEnumBranching => {
                 "SimplifyCfg-after-unreachable-enum-branching"
             }
+            SimplifyCfg::PostStateTransform => "SimplifyCfg-post-StateTransform",
         }
     }
 }
@@ -390,6 +393,8 @@ pub(super) enum SimplifyLocals {
     BeforeConstProp,
     AfterGVN,
     Final,
+    /// Extra run introduced by `StateTransform`.
+    PostStateTransform,
 }
 
 impl<'tcx> crate::MirPass<'tcx> for SimplifyLocals {
@@ -398,6 +403,7 @@ impl<'tcx> crate::MirPass<'tcx> for SimplifyLocals {
             SimplifyLocals::BeforeConstProp => "SimplifyLocals-before-const-prop",
             SimplifyLocals::AfterGVN => "SimplifyLocals-after-value-numbering",
             SimplifyLocals::Final => "SimplifyLocals-final",
+            SimplifyLocals::PostStateTransform => "SimplifyLocals-post-StateTransform",
         }
     }
 

compiler/rustc_mir_transform/src/validate.rs

@@ -450,7 +450,7 @@ impl<'a, 'tcx> Visitor<'tcx> for CfgChecker<'a, 'tcx> {
                 if self.body.coroutine.is_none() {
                     self.fail(location, "`Yield` cannot appear outside coroutine bodies");
                 }
-                if self.body.phase >= MirPhase::Runtime(RuntimePhase::Initial) {
+                if self.body.phase >= MirPhase::Runtime(RuntimePhase::Optimized) {
                     self.fail(location, "`Yield` should have been replaced by coroutine lowering");
                 }
                 self.check_edge(location, *resume, EdgeKind::Normal);
@@ -488,7 +488,7 @@ impl<'a, 'tcx> Visitor<'tcx> for CfgChecker<'a, 'tcx> {
                 if self.body.coroutine.is_none() {
                     self.fail(location, "`CoroutineDrop` cannot appear outside coroutine bodies");
                 }
-                if self.body.phase >= MirPhase::Runtime(RuntimePhase::Initial) {
+                if self.body.phase >= MirPhase::Runtime(RuntimePhase::Optimized) {
                     self.fail(
                         location,
                         "`CoroutineDrop` should have been replaced by coroutine lowering",