diff options
Diffstat (limited to 'compiler/rustc_monomorphize/src')
| -rw-r--r-- | compiler/rustc_monomorphize/src/collector.rs | 248 | ||||
| -rw-r--r-- | compiler/rustc_monomorphize/src/lib.rs | 5 | ||||
| -rw-r--r-- | compiler/rustc_monomorphize/src/partitioning.rs | 1274 | ||||
| -rw-r--r-- | compiler/rustc_monomorphize/src/partitioning/default.rs | 644 | ||||
| -rw-r--r-- | compiler/rustc_monomorphize/src/partitioning/mod.rs | 673 | ||||
| -rw-r--r-- | compiler/rustc_monomorphize/src/util.rs | 4 |
6 files changed, 1365 insertions, 1483 deletions
diff --git a/compiler/rustc_monomorphize/src/collector.rs b/compiler/rustc_monomorphize/src/collector.rs index 35b154b7b..242269e9d 100644 --- a/compiler/rustc_monomorphize/src/collector.rs +++ b/compiler/rustc_monomorphize/src/collector.rs @@ -35,15 +35,15 @@ //! //! - A "mono item" is something that results in a function or global in //! the LLVM IR of a codegen unit. Mono items do not stand on their -//! own, they can reference other mono items. For example, if function +//! own, they can use other mono items. For example, if function //! `foo()` calls function `bar()` then the mono item for `foo()` -//! references the mono item for function `bar()`. In general, the -//! definition for mono item A referencing a mono item B is that -//! the LLVM artifact produced for A references the LLVM artifact produced +//! uses the mono item for function `bar()`. In general, the +//! definition for mono item A using a mono item B is that +//! the LLVM artifact produced for A uses the LLVM artifact produced //! for B. //! -//! - Mono items and the references between them form a directed graph, -//! where the mono items are the nodes and references form the edges. +//! - Mono items and the uses between them form a directed graph, +//! where the mono items are the nodes and uses form the edges. //! Let's call this graph the "mono item graph". //! //! - The mono item graph for a program contains all mono items @@ -53,12 +53,11 @@ //! mono item graph for the current crate. It runs in two phases: //! //! 1. Discover the roots of the graph by traversing the HIR of the crate. -//! 2. Starting from the roots, find neighboring nodes by inspecting the MIR +//! 2. Starting from the roots, find uses by inspecting the MIR //! representation of the item corresponding to a given node, until no more //! new nodes are found. //! //! ### Discovering roots -//! //! The roots of the mono item graph correspond to the public non-generic //! syntactic items in the source code. We find them by walking the HIR of the //! crate, and whenever we hit upon a public function, method, or static item, @@ -69,25 +68,23 @@ //! specified. Functions marked `#[no_mangle]` and functions called by inlinable //! functions also always act as roots.) //! -//! ### Finding neighbor nodes -//! Given a mono item node, we can discover neighbors by inspecting its -//! MIR. We walk the MIR and any time we hit upon something that signifies a -//! reference to another mono item, we have found a neighbor. Since the -//! mono item we are currently at is always monomorphic, we also know the -//! concrete type arguments of its neighbors, and so all neighbors again will be -//! monomorphic. The specific forms a reference to a neighboring node can take -//! in MIR are quite diverse. Here is an overview: +//! ### Finding uses +//! Given a mono item node, we can discover uses by inspecting its MIR. We walk +//! the MIR to find other mono items used by each mono item. Since the mono +//! item we are currently at is always monomorphic, we also know the concrete +//! type arguments of its used mono items. The specific forms a use can take in +//! MIR are quite diverse. Here is an overview: //! //! #### Calling Functions/Methods -//! The most obvious form of one mono item referencing another is a +//! The most obvious way for one mono item to use another is a //! function or method call (represented by a CALL terminator in MIR). But -//! calls are not the only thing that might introduce a reference between two +//! calls are not the only thing that might introduce a use between two //! function mono items, and as we will see below, they are just a //! specialization of the form described next, and consequently will not get any //! special treatment in the algorithm. //! //! #### Taking a reference to a function or method -//! A function does not need to actually be called in order to be a neighbor of +//! A function does not need to actually be called in order to be used by //! another function. It suffices to just take a reference in order to introduce //! an edge. Consider the following example: //! @@ -109,29 +106,23 @@ //! The MIR of none of these functions will contain an explicit call to //! `print_val::<i32>`. Nonetheless, in order to mono this program, we need //! an instance of this function. Thus, whenever we encounter a function or -//! method in operand position, we treat it as a neighbor of the current +//! method in operand position, we treat it as a use of the current //! mono item. Calls are just a special case of that. //! //! #### Drop glue //! Drop glue mono items are introduced by MIR drop-statements. The -//! generated mono item will again have drop-glue item neighbors if the +//! generated mono item will have additional drop-glue item uses if the //! type to be dropped contains nested values that also need to be dropped. It -//! might also have a function item neighbor for the explicit `Drop::drop` +//! might also have a function item use for the explicit `Drop::drop` //! implementation of its type. //! //! #### Unsizing Casts -//! A subtle way of introducing neighbor edges is by casting to a trait object. +//! A subtle way of introducing use edges is by casting to a trait object. //! Since the resulting fat-pointer contains a reference to a vtable, we need to //! instantiate all object-safe methods of the trait, as we need to store //! pointers to these functions even if they never get called anywhere. This can //! be seen as a special case of taking a function reference. //! -//! #### Boxes -//! Since `Box` expression have special compiler support, no explicit calls to -//! `exchange_malloc()` and `box_free()` may show up in MIR, even if the -//! compiler will generate them. We have to observe `Rvalue::Box` expressions -//! and Box-typed drop-statements for that purpose. -//! //! //! Interaction with Cross-Crate Inlining //! ------------------------------------- @@ -151,7 +142,7 @@ //! Mono item collection can be performed in one of two modes: //! //! - Lazy mode means that items will only be instantiated when actually -//! referenced. The goal is to produce the least amount of machine code +//! used. The goal is to produce the least amount of machine code //! possible. //! //! - Eager mode is meant to be used in conjunction with incremental compilation @@ -179,14 +170,13 @@ use rustc_hir as hir; use rustc_hir::def::DefKind; use rustc_hir::def_id::{DefId, DefIdMap, LocalDefId}; use rustc_hir::lang_items::LangItem; -use rustc_index::bit_set::GrowableBitSet; use rustc_middle::mir::interpret::{AllocId, ConstValue}; use rustc_middle::mir::interpret::{ErrorHandled, GlobalAlloc, Scalar}; use rustc_middle::mir::mono::{InstantiationMode, MonoItem}; use rustc_middle::mir::visit::Visitor as MirVisitor; use rustc_middle::mir::{self, Local, Location}; use rustc_middle::query::TyCtxtAt; -use rustc_middle::ty::adjustment::{CustomCoerceUnsized, PointerCast}; +use rustc_middle::ty::adjustment::{CustomCoerceUnsized, PointerCoercion}; use rustc_middle::ty::print::with_no_trimmed_paths; use rustc_middle::ty::subst::{GenericArgKind, InternalSubsts}; use rustc_middle::ty::{ @@ -199,7 +189,6 @@ use rustc_session::lint::builtin::LARGE_ASSIGNMENTS; use rustc_session::Limit; use rustc_span::source_map::{dummy_spanned, respan, Span, Spanned, DUMMY_SP}; use rustc_target::abi::Size; -use std::ops::Range; use std::path::PathBuf; use crate::errors::{ @@ -212,114 +201,51 @@ pub enum MonoItemCollectionMode { Lazy, } -/// Maps every mono item to all mono items it references in its -/// body. -pub struct InliningMap<'tcx> { - // Maps a source mono item to the range of mono items - // accessed by it. - // The range selects elements within the `targets` vecs. - index: FxHashMap<MonoItem<'tcx>, Range<usize>>, - targets: Vec<MonoItem<'tcx>>, - - // Contains one bit per mono item in the `targets` field. That bit - // is true if that mono item needs to be inlined into every CGU. - inlines: GrowableBitSet<usize>, -} - -/// Struct to store mono items in each collecting and if they should -/// be inlined. We call `instantiation_mode` to get their inlining -/// status when inserting new elements, which avoids calling it in -/// `inlining_map.lock_mut()`. See the `collect_items_rec` implementation -/// below. -struct MonoItems<'tcx> { - // If this is false, we do not need to compute whether items - // will need to be inlined. - compute_inlining: bool, - - // The TyCtxt used to determine whether the a item should - // be inlined. - tcx: TyCtxt<'tcx>, +pub struct UsageMap<'tcx> { + // Maps every mono item to the mono items used by it. + used_map: FxHashMap<MonoItem<'tcx>, Vec<MonoItem<'tcx>>>, - // The collected mono items. The bool field in each element - // indicates whether this element should be inlined. - items: Vec<(Spanned<MonoItem<'tcx>>, bool /*inlined*/)>, + // Maps every mono item to the mono items that use it. + user_map: FxHashMap<MonoItem<'tcx>, Vec<MonoItem<'tcx>>>, } -impl<'tcx> MonoItems<'tcx> { - #[inline] - fn push(&mut self, item: Spanned<MonoItem<'tcx>>) { - self.extend([item]); - } - - #[inline] - fn extend<T: IntoIterator<Item = Spanned<MonoItem<'tcx>>>>(&mut self, iter: T) { - self.items.extend(iter.into_iter().map(|mono_item| { - let inlined = if !self.compute_inlining { - false - } else { - mono_item.node.instantiation_mode(self.tcx) == InstantiationMode::LocalCopy - }; - (mono_item, inlined) - })) - } -} +type MonoItems<'tcx> = Vec<Spanned<MonoItem<'tcx>>>; -impl<'tcx> InliningMap<'tcx> { - fn new() -> InliningMap<'tcx> { - InliningMap { - index: FxHashMap::default(), - targets: Vec::new(), - inlines: GrowableBitSet::with_capacity(1024), - } +impl<'tcx> UsageMap<'tcx> { + fn new() -> UsageMap<'tcx> { + UsageMap { used_map: FxHashMap::default(), user_map: FxHashMap::default() } } - fn record_accesses<'a>( + fn record_used<'a>( &mut self, - source: MonoItem<'tcx>, - new_targets: &'a [(Spanned<MonoItem<'tcx>>, bool)], + user_item: MonoItem<'tcx>, + used_items: &'a [Spanned<MonoItem<'tcx>>], ) where 'tcx: 'a, { - let start_index = self.targets.len(); - let new_items_count = new_targets.len(); - let new_items_count_total = new_items_count + self.targets.len(); - - self.targets.reserve(new_items_count); - self.inlines.ensure(new_items_count_total); - - for (i, (Spanned { node: mono_item, .. }, inlined)) in new_targets.into_iter().enumerate() { - self.targets.push(*mono_item); - if *inlined { - self.inlines.insert(i + start_index); - } + let used_items: Vec<_> = used_items.iter().map(|item| item.node).collect(); + for &used_item in used_items.iter() { + self.user_map.entry(used_item).or_default().push(user_item); } - let end_index = self.targets.len(); - assert!(self.index.insert(source, start_index..end_index).is_none()); + assert!(self.used_map.insert(user_item, used_items).is_none()); } - /// Internally iterate over all items referenced by `source` which will be - /// made available for inlining. - pub fn with_inlining_candidates<F>(&self, source: MonoItem<'tcx>, mut f: F) - where - F: FnMut(MonoItem<'tcx>), - { - if let Some(range) = self.index.get(&source) { - for (i, candidate) in self.targets[range.clone()].iter().enumerate() { - if self.inlines.contains(range.start + i) { - f(*candidate); - } - } - } + pub fn get_user_items(&self, item: MonoItem<'tcx>) -> &[MonoItem<'tcx>] { + self.user_map.get(&item).map(|items| items.as_slice()).unwrap_or(&[]) } - /// Internally iterate over all items and the things each accesses. - pub fn iter_accesses<F>(&self, mut f: F) + /// Internally iterate over all inlined items used by `item`. + pub fn for_each_inlined_used_item<F>(&self, tcx: TyCtxt<'tcx>, item: MonoItem<'tcx>, mut f: F) where - F: FnMut(MonoItem<'tcx>, &[MonoItem<'tcx>]), + F: FnMut(MonoItem<'tcx>), { - for (&accessor, range) in &self.index { - f(accessor, &self.targets[range.clone()]) + let used_items = self.used_map.get(&item).unwrap(); + for used_item in used_items.iter() { + let is_inlined = used_item.instantiation_mode(tcx) == InstantiationMode::LocalCopy; + if is_inlined { + f(*used_item); + } } } } @@ -328,7 +254,7 @@ impl<'tcx> InliningMap<'tcx> { pub fn collect_crate_mono_items( tcx: TyCtxt<'_>, mode: MonoItemCollectionMode, -) -> (FxHashSet<MonoItem<'_>>, InliningMap<'_>) { +) -> (FxHashSet<MonoItem<'_>>, UsageMap<'_>) { let _prof_timer = tcx.prof.generic_activity("monomorphization_collector"); let roots = @@ -337,12 +263,12 @@ pub fn collect_crate_mono_items( debug!("building mono item graph, beginning at roots"); let mut visited = MTLock::new(FxHashSet::default()); - let mut inlining_map = MTLock::new(InliningMap::new()); + let mut usage_map = MTLock::new(UsageMap::new()); let recursion_limit = tcx.recursion_limit(); { let visited: MTLockRef<'_, _> = &mut visited; - let inlining_map: MTLockRef<'_, _> = &mut inlining_map; + let usage_map: MTLockRef<'_, _> = &mut usage_map; tcx.sess.time("monomorphization_collector_graph_walk", || { par_for_each_in(roots, |root| { @@ -353,13 +279,13 @@ pub fn collect_crate_mono_items( visited, &mut recursion_depths, recursion_limit, - inlining_map, + usage_map, ); }); }); } - (visited.into_inner(), inlining_map.into_inner()) + (visited.into_inner(), usage_map.into_inner()) } // Find all non-generic items by walking the HIR. These items serve as roots to @@ -367,7 +293,7 @@ pub fn collect_crate_mono_items( #[instrument(skip(tcx, mode), level = "debug")] fn collect_roots(tcx: TyCtxt<'_>, mode: MonoItemCollectionMode) -> Vec<MonoItem<'_>> { debug!("collecting roots"); - let mut roots = MonoItems { compute_inlining: false, tcx, items: Vec::new() }; + let mut roots = Vec::new(); { let entry_fn = tcx.entry_fn(()); @@ -393,9 +319,8 @@ fn collect_roots(tcx: TyCtxt<'_>, mode: MonoItemCollectionMode) -> Vec<MonoItem< // whose predicates hold. Luckily, items that aren't instantiable // can't actually be used, so we can just skip codegenning them. roots - .items .into_iter() - .filter_map(|(Spanned { node: mono_item, .. }, _)| { + .filter_map(|Spanned { node: mono_item, .. }| { mono_item.is_instantiable(tcx).then_some(mono_item) }) .collect() @@ -403,24 +328,23 @@ fn collect_roots(tcx: TyCtxt<'_>, mode: MonoItemCollectionMode) -> Vec<MonoItem< /// Collect all monomorphized items reachable from `starting_point`, and emit a note diagnostic if a /// post-monomorphization error is encountered during a collection step. -#[instrument(skip(tcx, visited, recursion_depths, recursion_limit, inlining_map), level = "debug")] +#[instrument(skip(tcx, visited, recursion_depths, recursion_limit, usage_map), level = "debug")] fn collect_items_rec<'tcx>( tcx: TyCtxt<'tcx>, - starting_point: Spanned<MonoItem<'tcx>>, + starting_item: Spanned<MonoItem<'tcx>>, visited: MTLockRef<'_, FxHashSet<MonoItem<'tcx>>>, recursion_depths: &mut DefIdMap<usize>, recursion_limit: Limit, - inlining_map: MTLockRef<'_, InliningMap<'tcx>>, + usage_map: MTLockRef<'_, UsageMap<'tcx>>, ) { - if !visited.lock_mut().insert(starting_point.node) { + if !visited.lock_mut().insert(starting_item.node) { // We've been here already, no need to search again. return; } - let mut neighbors = MonoItems { compute_inlining: true, tcx, items: Vec::new() }; + let mut used_items = Vec::new(); let recursion_depth_reset; - // // Post-monomorphization errors MVP // // We can encounter errors while monomorphizing an item, but we don't have a good way of @@ -446,7 +370,7 @@ fn collect_items_rec<'tcx>( // FIXME: don't rely on global state, instead bubble up errors. Note: this is very hard to do. let error_count = tcx.sess.diagnostic().err_count(); - match starting_point.node { + match starting_item.node { MonoItem::Static(def_id) => { let instance = Instance::mono(tcx, def_id); @@ -454,19 +378,19 @@ fn collect_items_rec<'tcx>( debug_assert!(should_codegen_locally(tcx, &instance)); let ty = instance.ty(tcx, ty::ParamEnv::reveal_all()); - visit_drop_use(tcx, ty, true, starting_point.span, &mut neighbors); + visit_drop_use(tcx, ty, true, starting_item.span, &mut used_items); recursion_depth_reset = None; if let Ok(alloc) = tcx.eval_static_initializer(def_id) { for &id in alloc.inner().provenance().ptrs().values() { - collect_miri(tcx, id, &mut neighbors); + collect_miri(tcx, id, &mut used_items); } } if tcx.needs_thread_local_shim(def_id) { - neighbors.push(respan( - starting_point.span, + used_items.push(respan( + starting_item.span, MonoItem::Fn(Instance { def: InstanceDef::ThreadLocalShim(def_id), substs: InternalSubsts::empty(), @@ -482,14 +406,14 @@ fn collect_items_rec<'tcx>( recursion_depth_reset = Some(check_recursion_limit( tcx, instance, - starting_point.span, + starting_item.span, recursion_depths, recursion_limit, )); check_type_length_limit(tcx, instance); rustc_data_structures::stack::ensure_sufficient_stack(|| { - collect_neighbours(tcx, instance, &mut neighbors); + collect_used_items(tcx, instance, &mut used_items); }); } MonoItem::GlobalAsm(item_id) => { @@ -507,13 +431,13 @@ fn collect_items_rec<'tcx>( hir::InlineAsmOperand::SymFn { anon_const } => { let fn_ty = tcx.typeck_body(anon_const.body).node_type(anon_const.hir_id); - visit_fn_use(tcx, fn_ty, false, *op_sp, &mut neighbors); + visit_fn_use(tcx, fn_ty, false, *op_sp, &mut used_items); } hir::InlineAsmOperand::SymStatic { path: _, def_id } => { let instance = Instance::mono(tcx, *def_id); if should_codegen_locally(tcx, &instance) { trace!("collecting static {:?}", def_id); - neighbors.push(dummy_spanned(MonoItem::Static(*def_id))); + used_items.push(dummy_spanned(MonoItem::Static(*def_id))); } } hir::InlineAsmOperand::In { .. } @@ -533,19 +457,19 @@ fn collect_items_rec<'tcx>( // Check for PMEs and emit a diagnostic if one happened. To try to show relevant edges of the // mono item graph. if tcx.sess.diagnostic().err_count() > error_count - && starting_point.node.is_generic_fn() - && starting_point.node.is_user_defined() + && starting_item.node.is_generic_fn() + && starting_item.node.is_user_defined() { - let formatted_item = with_no_trimmed_paths!(starting_point.node.to_string()); + let formatted_item = with_no_trimmed_paths!(starting_item.node.to_string()); tcx.sess.emit_note(EncounteredErrorWhileInstantiating { - span: starting_point.span, + span: starting_item.span, formatted_item, }); } - inlining_map.lock_mut().record_accesses(starting_point.node, &neighbors.items); + usage_map.lock_mut().record_used(starting_item.node, &used_items); - for (neighbour, _) in neighbors.items { - collect_items_rec(tcx, neighbour, visited, recursion_depths, recursion_limit, inlining_map); + for used_item in used_items { + collect_items_rec(tcx, used_item, visited, recursion_depths, recursion_limit, usage_map); } if let Some((def_id, depth)) = recursion_depth_reset { @@ -661,14 +585,14 @@ fn check_type_length_limit<'tcx>(tcx: TyCtxt<'tcx>, instance: Instance<'tcx>) { } } -struct MirNeighborCollector<'a, 'tcx> { +struct MirUsedCollector<'a, 'tcx> { tcx: TyCtxt<'tcx>, body: &'a mir::Body<'tcx>, output: &'a mut MonoItems<'tcx>, instance: Instance<'tcx>, } -impl<'a, 'tcx> MirNeighborCollector<'a, 'tcx> { +impl<'a, 'tcx> MirUsedCollector<'a, 'tcx> { pub fn monomorphize<T>(&self, value: T) -> T where T: TypeFoldable<TyCtxt<'tcx>>, @@ -677,12 +601,12 @@ impl<'a, 'tcx> MirNeighborCollector<'a, 'tcx> { self.instance.subst_mir_and_normalize_erasing_regions( self.tcx, ty::ParamEnv::reveal_all(), - ty::EarlyBinder(value), + ty::EarlyBinder::bind(value), ) } } -impl<'a, 'tcx> MirVisitor<'tcx> for MirNeighborCollector<'a, 'tcx> { +impl<'a, 'tcx> MirVisitor<'tcx> for MirUsedCollector<'a, 'tcx> { fn visit_rvalue(&mut self, rvalue: &mir::Rvalue<'tcx>, location: Location) { debug!("visiting rvalue {:?}", *rvalue); @@ -693,7 +617,7 @@ impl<'a, 'tcx> MirVisitor<'tcx> for MirNeighborCollector<'a, 'tcx> { // have to instantiate all methods of the trait being cast to, so we // can build the appropriate vtable. mir::Rvalue::Cast( - mir::CastKind::Pointer(PointerCast::Unsize), + mir::CastKind::PointerCoercion(PointerCoercion::Unsize), ref operand, target_ty, ) @@ -719,7 +643,7 @@ impl<'a, 'tcx> MirVisitor<'tcx> for MirNeighborCollector<'a, 'tcx> { } } mir::Rvalue::Cast( - mir::CastKind::Pointer(PointerCast::ReifyFnPointer), + mir::CastKind::PointerCoercion(PointerCoercion::ReifyFnPointer), ref operand, _, ) => { @@ -728,7 +652,7 @@ impl<'a, 'tcx> MirVisitor<'tcx> for MirNeighborCollector<'a, 'tcx> { visit_fn_use(self.tcx, fn_ty, false, span, &mut self.output); } mir::Rvalue::Cast( - mir::CastKind::Pointer(PointerCast::ClosureFnPointer(_)), + mir::CastKind::PointerCoercion(PointerCoercion::ClosureFnPointer(_)), ref operand, _, ) => { @@ -1442,13 +1366,13 @@ fn collect_miri<'tcx>(tcx: TyCtxt<'tcx>, alloc_id: AllocId, output: &mut MonoIte /// Scans the MIR in order to find function calls, closures, and drop-glue. #[instrument(skip(tcx, output), level = "debug")] -fn collect_neighbours<'tcx>( +fn collect_used_items<'tcx>( tcx: TyCtxt<'tcx>, instance: Instance<'tcx>, output: &mut MonoItems<'tcx>, ) { let body = tcx.instance_mir(instance.def); - MirNeighborCollector { tcx, body: &body, output, instance }.visit_body(&body); + MirUsedCollector { tcx, body: &body, output, instance }.visit_body(&body); } #[instrument(skip(tcx, output), level = "debug")] diff --git a/compiler/rustc_monomorphize/src/lib.rs b/compiler/rustc_monomorphize/src/lib.rs index ecc50c3f6..5f05020ac 100644 --- a/compiler/rustc_monomorphize/src/lib.rs +++ b/compiler/rustc_monomorphize/src/lib.rs @@ -1,4 +1,5 @@ #![feature(array_windows)] +#![feature(is_sorted)] #![recursion_limit = "256"] #![allow(rustc::potential_query_instability)] #![deny(rustc::untranslatable_diagnostic)] @@ -30,12 +31,12 @@ fn custom_coerce_unsize_info<'tcx>( source_ty: Ty<'tcx>, target_ty: Ty<'tcx>, ) -> CustomCoerceUnsized { - let trait_ref = ty::Binder::dummy(ty::TraitRef::from_lang_item( + let trait_ref = ty::TraitRef::from_lang_item( tcx.tcx, LangItem::CoerceUnsized, tcx.span, [source_ty, target_ty], - )); + ); match tcx.codegen_select_candidate((ty::ParamEnv::reveal_all(), trait_ref)) { Ok(traits::ImplSource::UserDefined(traits::ImplSourceUserDefinedData { diff --git a/compiler/rustc_monomorphize/src/partitioning.rs b/compiler/rustc_monomorphize/src/partitioning.rs new file mode 100644 index 000000000..da76cf223 --- /dev/null +++ b/compiler/rustc_monomorphize/src/partitioning.rs @@ -0,0 +1,1274 @@ +//! Partitioning Codegen Units for Incremental Compilation +//! ====================================================== +//! +//! The task of this module is to take the complete set of monomorphizations of +//! a crate and produce a set of codegen units from it, where a codegen unit +//! is a named set of (mono-item, linkage) pairs. That is, this module +//! decides which monomorphization appears in which codegen units with which +//! linkage. The following paragraphs describe some of the background on the +//! partitioning scheme. +//! +//! The most important opportunity for saving on compilation time with +//! incremental compilation is to avoid re-codegenning and re-optimizing code. +//! Since the unit of codegen and optimization for LLVM is "modules" or, how +//! we call them "codegen units", the particulars of how much time can be saved +//! by incremental compilation are tightly linked to how the output program is +//! partitioned into these codegen units prior to passing it to LLVM -- +//! especially because we have to treat codegen units as opaque entities once +//! they are created: There is no way for us to incrementally update an existing +//! LLVM module and so we have to build any such module from scratch if it was +//! affected by some change in the source code. +//! +//! From that point of view it would make sense to maximize the number of +//! codegen units by, for example, putting each function into its own module. +//! That way only those modules would have to be re-compiled that were actually +//! affected by some change, minimizing the number of functions that could have +//! been re-used but just happened to be located in a module that is +//! re-compiled. +//! +//! However, since LLVM optimization does not work across module boundaries, +//! using such a highly granular partitioning would lead to very slow runtime +//! code since it would effectively prohibit inlining and other inter-procedure +//! optimizations. We want to avoid that as much as possible. +//! +//! Thus we end up with a trade-off: The bigger the codegen units, the better +//! LLVM's optimizer can do its work, but also the smaller the compilation time +//! reduction we get from incremental compilation. +//! +//! Ideally, we would create a partitioning such that there are few big codegen +//! units with few interdependencies between them. For now though, we use the +//! following heuristic to determine the partitioning: +//! +//! - There are two codegen units for every source-level module: +//! - One for "stable", that is non-generic, code +//! - One for more "volatile" code, i.e., monomorphized instances of functions +//! defined in that module +//! +//! In order to see why this heuristic makes sense, let's take a look at when a +//! codegen unit can get invalidated: +//! +//! 1. The most straightforward case is when the BODY of a function or global +//! changes. Then any codegen unit containing the code for that item has to be +//! re-compiled. Note that this includes all codegen units where the function +//! has been inlined. +//! +//! 2. The next case is when the SIGNATURE of a function or global changes. In +//! this case, all codegen units containing a REFERENCE to that item have to be +//! re-compiled. This is a superset of case 1. +//! +//! 3. The final and most subtle case is when a REFERENCE to a generic function +//! is added or removed somewhere. Even though the definition of the function +//! might be unchanged, a new REFERENCE might introduce a new monomorphized +//! instance of this function which has to be placed and compiled somewhere. +//! Conversely, when removing a REFERENCE, it might have been the last one with +//! that particular set of generic arguments and thus we have to remove it. +//! +//! From the above we see that just using one codegen unit per source-level +//! module is not such a good idea, since just adding a REFERENCE to some +//! generic item somewhere else would invalidate everything within the module +//! containing the generic item. The heuristic above reduces this detrimental +//! side-effect of references a little by at least not touching the non-generic +//! code of the module. +//! +//! A Note on Inlining +//! ------------------ +//! As briefly mentioned above, in order for LLVM to be able to inline a +//! function call, the body of the function has to be available in the LLVM +//! module where the call is made. This has a few consequences for partitioning: +//! +//! - The partitioning algorithm has to take care of placing functions into all +//! codegen units where they should be available for inlining. It also has to +//! decide on the correct linkage for these functions. +//! +//! - The partitioning algorithm has to know which functions are likely to get +//! inlined, so it can distribute function instantiations accordingly. Since +//! there is no way of knowing for sure which functions LLVM will decide to +//! inline in the end, we apply a heuristic here: Only functions marked with +//! `#[inline]` are considered for inlining by the partitioner. The current +//! implementation will not try to determine if a function is likely to be +//! inlined by looking at the functions definition. +//! +//! Note though that as a side-effect of creating a codegen units per +//! source-level module, functions from the same module will be available for +//! inlining, even when they are not marked `#[inline]`. + +use std::cmp; +use std::collections::hash_map::Entry; +use std::fs::{self, File}; +use std::io::{BufWriter, Write}; +use std::path::{Path, PathBuf}; + +use rustc_data_structures::fx::{FxHashMap, FxHashSet}; +use rustc_data_structures::sync; +use rustc_hir::def::DefKind; +use rustc_hir::def_id::{DefId, DefIdSet, LOCAL_CRATE}; +use rustc_hir::definitions::DefPathDataName; +use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrFlags; +use rustc_middle::middle::exported_symbols::{SymbolExportInfo, SymbolExportLevel}; +use rustc_middle::mir; +use rustc_middle::mir::mono::{ + CodegenUnit, CodegenUnitNameBuilder, InstantiationMode, Linkage, MonoItem, Visibility, +}; +use rustc_middle::query::Providers; +use rustc_middle::ty::print::{characteristic_def_id_of_type, with_no_trimmed_paths}; +use rustc_middle::ty::{self, visit::TypeVisitableExt, InstanceDef, TyCtxt}; +use rustc_session::config::{DumpMonoStatsFormat, SwitchWithOptPath}; +use rustc_session::CodegenUnits; +use rustc_span::symbol::Symbol; + +use crate::collector::UsageMap; +use crate::collector::{self, MonoItemCollectionMode}; +use crate::errors::{CouldntDumpMonoStats, SymbolAlreadyDefined, UnknownCguCollectionMode}; + +struct PartitioningCx<'a, 'tcx> { + tcx: TyCtxt<'tcx>, + usage_map: &'a UsageMap<'tcx>, +} + +struct PlacedMonoItems<'tcx> { + /// The codegen units, sorted by name to make things deterministic. + codegen_units: Vec<CodegenUnit<'tcx>>, + + internalization_candidates: FxHashSet<MonoItem<'tcx>>, + + /// These must be obtained when the iterator in `partition` runs. They + /// can't be obtained later because some inlined functions might not be + /// reachable. + unique_inlined_stats: (usize, usize), +} + +// The output CGUs are sorted by name. +fn partition<'tcx, I>( + tcx: TyCtxt<'tcx>, + mono_items: I, + usage_map: &UsageMap<'tcx>, +) -> Vec<CodegenUnit<'tcx>> +where + I: Iterator<Item = MonoItem<'tcx>>, +{ + let _prof_timer = tcx.prof.generic_activity("cgu_partitioning"); + + let cx = &PartitioningCx { tcx, usage_map }; + + // Place all mono items into a codegen unit. `place_mono_items` is + // responsible for initializing the CGU size estimates. + let PlacedMonoItems { mut codegen_units, internalization_candidates, unique_inlined_stats } = { + let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_place_items"); + let placed = place_mono_items(cx, mono_items); + + debug_dump(tcx, "PLACE", &placed.codegen_units, placed.unique_inlined_stats); + + placed + }; + + // Merge until we have at most `max_cgu_count` codegen units. + // `merge_codegen_units` is responsible for updating the CGU size + // estimates. + { + let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_merge_cgus"); + merge_codegen_units(cx, &mut codegen_units); + debug_dump(tcx, "MERGE", &codegen_units, unique_inlined_stats); + } + + // Make as many symbols "internal" as possible, so LLVM has more freedom to + // optimize. + if !tcx.sess.link_dead_code() { + let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_internalize_symbols"); + internalize_symbols(cx, &mut codegen_units, internalization_candidates); + + debug_dump(tcx, "INTERNALIZE", &codegen_units, unique_inlined_stats); + } + + // Mark one CGU for dead code, if necessary. + let instrument_dead_code = + tcx.sess.instrument_coverage() && !tcx.sess.instrument_coverage_except_unused_functions(); + if instrument_dead_code { + mark_code_coverage_dead_code_cgu(&mut codegen_units); + } + + // Ensure CGUs are sorted by name, so that we get deterministic results. + if !codegen_units.is_sorted_by(|a, b| Some(a.name().as_str().cmp(b.name().as_str()))) { + let mut names = String::new(); + for cgu in codegen_units.iter() { + names += &format!("- {}\n", cgu.name()); + } + bug!("unsorted CGUs:\n{names}"); + } + + codegen_units +} + +fn place_mono_items<'tcx, I>(cx: &PartitioningCx<'_, 'tcx>, mono_items: I) -> PlacedMonoItems<'tcx> +where + I: Iterator<Item = MonoItem<'tcx>>, +{ + let mut codegen_units = FxHashMap::default(); + let is_incremental_build = cx.tcx.sess.opts.incremental.is_some(); + let mut internalization_candidates = FxHashSet::default(); + + // Determine if monomorphizations instantiated in this crate will be made + // available to downstream crates. This depends on whether we are in + // share-generics mode and whether the current crate can even have + // downstream crates. + let export_generics = + cx.tcx.sess.opts.share_generics() && cx.tcx.local_crate_exports_generics(); + + let cgu_name_builder = &mut CodegenUnitNameBuilder::new(cx.tcx); + let cgu_name_cache = &mut FxHashMap::default(); + + let mut num_unique_inlined_items = 0; + let mut unique_inlined_items_size = 0; + for mono_item in mono_items { + // Handle only root items directly here. Inlined items are handled at + // the bottom of the loop based on reachability. + match mono_item.instantiation_mode(cx.tcx) { + InstantiationMode::GloballyShared { .. } => {} + InstantiationMode::LocalCopy => { + num_unique_inlined_items += 1; + unique_inlined_items_size += mono_item.size_estimate(cx.tcx); + continue; + } + } + + let characteristic_def_id = characteristic_def_id_of_mono_item(cx.tcx, mono_item); + let is_volatile = is_incremental_build && mono_item.is_generic_fn(); + + let cgu_name = match characteristic_def_id { + Some(def_id) => compute_codegen_unit_name( + cx.tcx, + cgu_name_builder, + def_id, + is_volatile, + cgu_name_cache, + ), + None => fallback_cgu_name(cgu_name_builder), + }; + + let cgu = codegen_units.entry(cgu_name).or_insert_with(|| CodegenUnit::new(cgu_name)); + + let mut can_be_internalized = true; + let (linkage, visibility) = mono_item_linkage_and_visibility( + cx.tcx, + &mono_item, + &mut can_be_internalized, + export_generics, + ); + if visibility == Visibility::Hidden && can_be_internalized { + internalization_candidates.insert(mono_item); + } + + cgu.items_mut().insert(mono_item, (linkage, visibility)); + + // Get all inlined items that are reachable from `mono_item` without + // going via another root item. This includes drop-glue, functions from + // external crates, and local functions the definition of which is + // marked with `#[inline]`. + let mut reachable_inlined_items = FxHashSet::default(); + get_reachable_inlined_items(cx.tcx, mono_item, cx.usage_map, &mut reachable_inlined_items); + + // Add those inlined items. It's possible an inlined item is reachable + // from multiple root items within a CGU, which is fine, it just means + // the `insert` will be a no-op. + for inlined_item in reachable_inlined_items { + // This is a CGU-private copy. + cgu.items_mut().insert(inlined_item, (Linkage::Internal, Visibility::Default)); + } + } + + // Always ensure we have at least one CGU; otherwise, if we have a + // crate with just types (for example), we could wind up with no CGU. + if codegen_units.is_empty() { + let cgu_name = fallback_cgu_name(cgu_name_builder); + codegen_units.insert(cgu_name, CodegenUnit::new(cgu_name)); + } + + let mut codegen_units: Vec<_> = codegen_units.into_values().collect(); + codegen_units.sort_by(|a, b| a.name().as_str().cmp(b.name().as_str())); + + for cgu in codegen_units.iter_mut() { + cgu.compute_size_estimate(cx.tcx); + } + + return PlacedMonoItems { + codegen_units, + internalization_candidates, + unique_inlined_stats: (num_unique_inlined_items, unique_inlined_items_size), + }; + + fn get_reachable_inlined_items<'tcx>( + tcx: TyCtxt<'tcx>, + item: MonoItem<'tcx>, + usage_map: &UsageMap<'tcx>, + visited: &mut FxHashSet<MonoItem<'tcx>>, + ) { + usage_map.for_each_inlined_used_item(tcx, item, |inlined_item| { + let is_new = visited.insert(inlined_item); + if is_new { + get_reachable_inlined_items(tcx, inlined_item, usage_map, visited); + } + }); + } +} + +// This function requires the CGUs to be sorted by name on input, and ensures +// they are sorted by name on return, for deterministic behaviour. +fn merge_codegen_units<'tcx>( + cx: &PartitioningCx<'_, 'tcx>, + codegen_units: &mut Vec<CodegenUnit<'tcx>>, +) { + assert!(cx.tcx.sess.codegen_units().as_usize() >= 1); + + // A sorted order here ensures merging is deterministic. + assert!(codegen_units.is_sorted_by(|a, b| Some(a.name().as_str().cmp(b.name().as_str())))); + + // This map keeps track of what got merged into what. + let mut cgu_contents: FxHashMap<Symbol, Vec<Symbol>> = + codegen_units.iter().map(|cgu| (cgu.name(), vec![cgu.name()])).collect(); + + // Having multiple CGUs can drastically speed up compilation. But for + // non-incremental builds, tiny CGUs slow down compilation *and* result in + // worse generated code. So we don't allow CGUs smaller than this (unless + // there is just one CGU, of course). Note that CGU sizes of 100,000+ are + // common in larger programs, so this isn't all that large. + const NON_INCR_MIN_CGU_SIZE: usize = 1800; + + // Repeatedly merge the two smallest codegen units as long as: + // - we have more CGUs than the upper limit, or + // - (Non-incremental builds only) the user didn't specify a CGU count, and + // there are multiple CGUs, and some are below the minimum size. + // + // The "didn't specify a CGU count" condition is because when an explicit + // count is requested we observe it as closely as possible. For example, + // the `compiler_builtins` crate sets `codegen-units = 10000` and it's + // critical they aren't merged. Also, some tests use explicit small values + // and likewise won't work if small CGUs are merged. + while codegen_units.len() > cx.tcx.sess.codegen_units().as_usize() + || (cx.tcx.sess.opts.incremental.is_none() + && matches!(cx.tcx.sess.codegen_units(), CodegenUnits::Default(_)) + && codegen_units.len() > 1 + && codegen_units.iter().any(|cgu| cgu.size_estimate() < NON_INCR_MIN_CGU_SIZE)) + { + // Sort small cgus to the back. + codegen_units.sort_by_cached_key(|cgu| cmp::Reverse(cgu.size_estimate())); + + let mut smallest = codegen_units.pop().unwrap(); + let second_smallest = codegen_units.last_mut().unwrap(); + + // Move the items from `smallest` to `second_smallest`. Some of them + // may be duplicate inlined items, in which case the destination CGU is + // unaffected. Recalculate size estimates afterwards. + second_smallest.items_mut().extend(smallest.items_mut().drain()); + second_smallest.compute_size_estimate(cx.tcx); + + // Record that `second_smallest` now contains all the stuff that was + // in `smallest` before. + let mut consumed_cgu_names = cgu_contents.remove(&smallest.name()).unwrap(); + cgu_contents.get_mut(&second_smallest.name()).unwrap().append(&mut consumed_cgu_names); + + debug!( + "CodegenUnit {} merged into CodegenUnit {}", + smallest.name(), + second_smallest.name() + ); + } + + let cgu_name_builder = &mut CodegenUnitNameBuilder::new(cx.tcx); + + // Rename the newly merged CGUs. + if cx.tcx.sess.opts.incremental.is_some() { + // If we are doing incremental compilation, we want CGU names to + // reflect the path of the source level module they correspond to. + // For CGUs that contain the code of multiple modules because of the + // merging done above, we use a concatenation of the names of all + // contained CGUs. + let new_cgu_names: FxHashMap<Symbol, String> = cgu_contents + .into_iter() + // This `filter` makes sure we only update the name of CGUs that + // were actually modified by merging. + .filter(|(_, cgu_contents)| cgu_contents.len() > 1) + .map(|(current_cgu_name, cgu_contents)| { + let mut cgu_contents: Vec<&str> = cgu_contents.iter().map(|s| s.as_str()).collect(); + + // Sort the names, so things are deterministic and easy to + // predict. We are sorting primitive `&str`s here so we can + // use unstable sort. + cgu_contents.sort_unstable(); + + (current_cgu_name, cgu_contents.join("--")) + }) + .collect(); + + for cgu in codegen_units.iter_mut() { + if let Some(new_cgu_name) = new_cgu_names.get(&cgu.name()) { + if cx.tcx.sess.opts.unstable_opts.human_readable_cgu_names { + cgu.set_name(Symbol::intern(&new_cgu_name)); + } else { + // If we don't require CGU names to be human-readable, + // we use a fixed length hash of the composite CGU name + // instead. + let new_cgu_name = CodegenUnit::mangle_name(&new_cgu_name); + cgu.set_name(Symbol::intern(&new_cgu_name)); + } + } + } + + // A sorted order here ensures what follows can be deterministic. + codegen_units.sort_by(|a, b| a.name().as_str().cmp(b.name().as_str())); + } else { + // When compiling non-incrementally, we rename the CGUS so they have + // identical names except for the numeric suffix, something like + // `regex.f10ba03eb5ec7975-cgu.N`, where `N` varies. + // + // It is useful for debugging and profiling purposes if the resulting + // CGUs are sorted by name *and* reverse sorted by size. (CGU 0 is the + // biggest, CGU 1 is the second biggest, etc.) + // + // So first we reverse sort by size. Then we generate the names with + // zero-padded suffixes, which means they are automatically sorted by + // names. The numeric suffix width depends on the number of CGUs, which + // is always greater than zero: + // - [1,9] CGUs: `0`, `1`, `2`, ... + // - [10,99] CGUs: `00`, `01`, `02`, ... + // - [100,999] CGUs: `000`, `001`, `002`, ... + // - etc. + // + // If we didn't zero-pad the sorted-by-name order would be `XYZ-cgu.0`, + // `XYZ-cgu.1`, `XYZ-cgu.10`, `XYZ-cgu.11`, ..., `XYZ-cgu.2`, etc. + codegen_units.sort_by_key(|cgu| cmp::Reverse(cgu.size_estimate())); + let num_digits = codegen_units.len().ilog10() as usize + 1; + for (index, cgu) in codegen_units.iter_mut().enumerate() { + // Note: `WorkItem::short_description` depends on this name ending + // with `-cgu.` followed by a numeric suffix. Please keep it in + // sync with this code. + let suffix = format!("{index:0num_digits$}"); + let numbered_codegen_unit_name = + cgu_name_builder.build_cgu_name_no_mangle(LOCAL_CRATE, &["cgu"], Some(suffix)); + cgu.set_name(numbered_codegen_unit_name); + } + } +} + +fn internalize_symbols<'tcx>( + cx: &PartitioningCx<'_, 'tcx>, + codegen_units: &mut [CodegenUnit<'tcx>], + internalization_candidates: FxHashSet<MonoItem<'tcx>>, +) { + /// For symbol internalization, we need to know whether a symbol/mono-item + /// is used from outside the codegen unit it is defined in. This type is + /// used to keep track of that. + #[derive(Clone, PartialEq, Eq, Debug)] + enum MonoItemPlacement { + SingleCgu(Symbol), + MultipleCgus, + } + + let mut mono_item_placements = FxHashMap::default(); + let single_codegen_unit = codegen_units.len() == 1; + + if !single_codegen_unit { + for cgu in codegen_units.iter() { + for item in cgu.items().keys() { + // If there is more than one codegen unit, we need to keep track + // in which codegen units each monomorphization is placed. + match mono_item_placements.entry(*item) { + Entry::Occupied(e) => { + let placement = e.into_mut(); + debug_assert!(match *placement { + MonoItemPlacement::SingleCgu(cgu_name) => cgu_name != cgu.name(), + MonoItemPlacement::MultipleCgus => true, + }); + *placement = MonoItemPlacement::MultipleCgus; + } + Entry::Vacant(e) => { + e.insert(MonoItemPlacement::SingleCgu(cgu.name())); + } + } + } + } + } + + // For each internalization candidates in each codegen unit, check if it is + // used from outside its defining codegen unit. + for cgu in codegen_units { + let home_cgu = MonoItemPlacement::SingleCgu(cgu.name()); + + for (item, linkage_and_visibility) in cgu.items_mut() { + if !internalization_candidates.contains(item) { + // This item is no candidate for internalizing, so skip it. + continue; + } + + if !single_codegen_unit { + debug_assert_eq!(mono_item_placements[item], home_cgu); + + if cx + .usage_map + .get_user_items(*item) + .iter() + .filter_map(|user_item| { + // Some user mono items might not have been + // instantiated. We can safely ignore those. + mono_item_placements.get(user_item) + }) + .any(|placement| *placement != home_cgu) + { + // Found a user from another CGU, so skip to the next item + // without marking this one as internal. + continue; + } + } + + // If we got here, we did not find any uses from other CGUs, so + // it's fine to make this monomorphization internal. + *linkage_and_visibility = (Linkage::Internal, Visibility::Default); + } + } +} + +fn mark_code_coverage_dead_code_cgu<'tcx>(codegen_units: &mut [CodegenUnit<'tcx>]) { + assert!(!codegen_units.is_empty()); + + // Find the smallest CGU that has exported symbols and put the dead + // function stubs in that CGU. We look for exported symbols to increase + // the likelihood the linker won't throw away the dead functions. + // FIXME(#92165): In order to truly resolve this, we need to make sure + // the object file (CGU) containing the dead function stubs is included + // in the final binary. This will probably require forcing these + // function symbols to be included via `-u` or `/include` linker args. + let dead_code_cgu = codegen_units + .iter_mut() + .filter(|cgu| cgu.items().iter().any(|(_, (linkage, _))| *linkage == Linkage::External)) + .min_by_key(|cgu| cgu.size_estimate()); + + // If there are no CGUs that have externally linked items, then we just + // pick the first CGU as a fallback. + let dead_code_cgu = if let Some(cgu) = dead_code_cgu { cgu } else { &mut codegen_units[0] }; + + dead_code_cgu.make_code_coverage_dead_code_cgu(); +} + +fn characteristic_def_id_of_mono_item<'tcx>( + tcx: TyCtxt<'tcx>, + mono_item: MonoItem<'tcx>, +) -> Option<DefId> { + match mono_item { + MonoItem::Fn(instance) => { + let def_id = match instance.def { + ty::InstanceDef::Item(def) => def, + ty::InstanceDef::VTableShim(..) + | ty::InstanceDef::ReifyShim(..) + | ty::InstanceDef::FnPtrShim(..) + | ty::InstanceDef::ClosureOnceShim { .. } + | ty::InstanceDef::Intrinsic(..) + | ty::InstanceDef::DropGlue(..) + | ty::InstanceDef::Virtual(..) + | ty::InstanceDef::CloneShim(..) + | ty::InstanceDef::ThreadLocalShim(..) + | ty::InstanceDef::FnPtrAddrShim(..) => return None, + }; + + // If this is a method, we want to put it into the same module as + // its self-type. If the self-type does not provide a characteristic + // DefId, we use the location of the impl after all. + + if tcx.trait_of_item(def_id).is_some() { + let self_ty = instance.substs.type_at(0); + // This is a default implementation of a trait method. + return characteristic_def_id_of_type(self_ty).or(Some(def_id)); + } + + if let Some(impl_def_id) = tcx.impl_of_method(def_id) { + if tcx.sess.opts.incremental.is_some() + && tcx.trait_id_of_impl(impl_def_id) == tcx.lang_items().drop_trait() + { + // Put `Drop::drop` into the same cgu as `drop_in_place` + // since `drop_in_place` is the only thing that can + // call it. + return None; + } + + // When polymorphization is enabled, methods which do not depend on their generic + // parameters, but the self-type of their impl block do will fail to normalize. + if !tcx.sess.opts.unstable_opts.polymorphize || !instance.has_param() { + // This is a method within an impl, find out what the self-type is: + let impl_self_ty = tcx.subst_and_normalize_erasing_regions( + instance.substs, + ty::ParamEnv::reveal_all(), + tcx.type_of(impl_def_id), + ); + if let Some(def_id) = characteristic_def_id_of_type(impl_self_ty) { + return Some(def_id); + } + } + } + + Some(def_id) + } + MonoItem::Static(def_id) => Some(def_id), + MonoItem::GlobalAsm(item_id) => Some(item_id.owner_id.to_def_id()), + } +} + +fn compute_codegen_unit_name( + tcx: TyCtxt<'_>, + name_builder: &mut CodegenUnitNameBuilder<'_>, + def_id: DefId, + volatile: bool, + cache: &mut CguNameCache, +) -> Symbol { + // Find the innermost module that is not nested within a function. + let mut current_def_id = def_id; + let mut cgu_def_id = None; + // Walk backwards from the item we want to find the module for. + loop { + if current_def_id.is_crate_root() { + if cgu_def_id.is_none() { + // If we have not found a module yet, take the crate root. + cgu_def_id = Some(def_id.krate.as_def_id()); + } + break; + } else if tcx.def_kind(current_def_id) == DefKind::Mod { + if cgu_def_id.is_none() { + cgu_def_id = Some(current_def_id); + } + } else { + // If we encounter something that is not a module, throw away + // any module that we've found so far because we now know that + // it is nested within something else. + cgu_def_id = None; + } + + current_def_id = tcx.parent(current_def_id); + } + + let cgu_def_id = cgu_def_id.unwrap(); + + *cache.entry((cgu_def_id, volatile)).or_insert_with(|| { + let def_path = tcx.def_path(cgu_def_id); + + let components = def_path.data.iter().map(|part| match part.data.name() { + DefPathDataName::Named(name) => name, + DefPathDataName::Anon { .. } => unreachable!(), + }); + + let volatile_suffix = volatile.then_some("volatile"); + + name_builder.build_cgu_name(def_path.krate, components, volatile_suffix) + }) +} + +// Anything we can't find a proper codegen unit for goes into this. +fn fallback_cgu_name(name_builder: &mut CodegenUnitNameBuilder<'_>) -> Symbol { + name_builder.build_cgu_name(LOCAL_CRATE, &["fallback"], Some("cgu")) +} + +fn mono_item_linkage_and_visibility<'tcx>( + tcx: TyCtxt<'tcx>, + mono_item: &MonoItem<'tcx>, + can_be_internalized: &mut bool, + export_generics: bool, +) -> (Linkage, Visibility) { + if let Some(explicit_linkage) = mono_item.explicit_linkage(tcx) { + return (explicit_linkage, Visibility::Default); + } + let vis = mono_item_visibility(tcx, mono_item, can_be_internalized, export_generics); + (Linkage::External, vis) +} + +type CguNameCache = FxHashMap<(DefId, bool), Symbol>; + +fn static_visibility<'tcx>( + tcx: TyCtxt<'tcx>, + can_be_internalized: &mut bool, + def_id: DefId, +) -> Visibility { + if tcx.is_reachable_non_generic(def_id) { + *can_be_internalized = false; + default_visibility(tcx, def_id, false) + } else { + Visibility::Hidden + } +} + +fn mono_item_visibility<'tcx>( + tcx: TyCtxt<'tcx>, + mono_item: &MonoItem<'tcx>, + can_be_internalized: &mut bool, + export_generics: bool, +) -> Visibility { + let instance = match mono_item { + // This is pretty complicated; see below. + MonoItem::Fn(instance) => instance, + + // Misc handling for generics and such, but otherwise: + MonoItem::Static(def_id) => return static_visibility(tcx, can_be_internalized, *def_id), + MonoItem::GlobalAsm(item_id) => { + return static_visibility(tcx, can_be_internalized, item_id.owner_id.to_def_id()); + } + }; + + let def_id = match instance.def { + InstanceDef::Item(def_id) | InstanceDef::DropGlue(def_id, Some(_)) => def_id, + + // We match the visibility of statics here + InstanceDef::ThreadLocalShim(def_id) => { + return static_visibility(tcx, can_be_internalized, def_id); + } + + // These are all compiler glue and such, never exported, always hidden. + InstanceDef::VTableShim(..) + | InstanceDef::ReifyShim(..) + | InstanceDef::FnPtrShim(..) + | InstanceDef::Virtual(..) + | InstanceDef::Intrinsic(..) + | InstanceDef::ClosureOnceShim { .. } + | InstanceDef::DropGlue(..) + | InstanceDef::CloneShim(..) + | InstanceDef::FnPtrAddrShim(..) => return Visibility::Hidden, + }; + + // The `start_fn` lang item is actually a monomorphized instance of a + // function in the standard library, used for the `main` function. We don't + // want to export it so we tag it with `Hidden` visibility but this symbol + // is only referenced from the actual `main` symbol which we unfortunately + // don't know anything about during partitioning/collection. As a result we + // forcibly keep this symbol out of the `internalization_candidates` set. + // + // FIXME: eventually we don't want to always force this symbol to have + // hidden visibility, it should indeed be a candidate for + // internalization, but we have to understand that it's referenced + // from the `main` symbol we'll generate later. + // + // This may be fixable with a new `InstanceDef` perhaps? Unsure! + if tcx.lang_items().start_fn() == Some(def_id) { + *can_be_internalized = false; + return Visibility::Hidden; + } + + let is_generic = instance.substs.non_erasable_generics().next().is_some(); + + // Upstream `DefId` instances get different handling than local ones. + let Some(def_id) = def_id.as_local() else { + return if export_generics && is_generic { + // If it is an upstream monomorphization and we export generics, we must make + // it available to downstream crates. + *can_be_internalized = false; + default_visibility(tcx, def_id, true) + } else { + Visibility::Hidden + }; + }; + + if is_generic { + if export_generics { + if tcx.is_unreachable_local_definition(def_id) { + // This instance cannot be used from another crate. + Visibility::Hidden + } else { + // This instance might be useful in a downstream crate. + *can_be_internalized = false; + default_visibility(tcx, def_id.to_def_id(), true) + } + } else { + // We are not exporting generics or the definition is not reachable + // for downstream crates, we can internalize its instantiations. + Visibility::Hidden + } + } else { + // If this isn't a generic function then we mark this a `Default` if + // this is a reachable item, meaning that it's a symbol other crates may + // use when they link to us. + if tcx.is_reachable_non_generic(def_id.to_def_id()) { + *can_be_internalized = false; + debug_assert!(!is_generic); + return default_visibility(tcx, def_id.to_def_id(), false); + } + + // If this isn't reachable then we're gonna tag this with `Hidden` + // visibility. In some situations though we'll want to prevent this + // symbol from being internalized. + // + // There's two categories of items here: + // + // * First is weak lang items. These are basically mechanisms for + // libcore to forward-reference symbols defined later in crates like + // the standard library or `#[panic_handler]` definitions. The + // definition of these weak lang items needs to be referencable by + // libcore, so we're no longer a candidate for internalization. + // Removal of these functions can't be done by LLVM but rather must be + // done by the linker as it's a non-local decision. + // + // * Second is "std internal symbols". Currently this is primarily used + // for allocator symbols. Allocators are a little weird in their + // implementation, but the idea is that the compiler, at the last + // minute, defines an allocator with an injected object file. The + // `alloc` crate references these symbols (`__rust_alloc`) and the + // definition doesn't get hooked up until a linked crate artifact is + // generated. + // + // The symbols synthesized by the compiler (`__rust_alloc`) are thin + // veneers around the actual implementation, some other symbol which + // implements the same ABI. These symbols (things like `__rg_alloc`, + // `__rdl_alloc`, `__rde_alloc`, etc), are all tagged with "std + // internal symbols". + // + // The std-internal symbols here **should not show up in a dll as an + // exported interface**, so they return `false` from + // `is_reachable_non_generic` above and we'll give them `Hidden` + // visibility below. Like the weak lang items, though, we can't let + // LLVM internalize them as this decision is left up to the linker to + // omit them, so prevent them from being internalized. + let attrs = tcx.codegen_fn_attrs(def_id); + if attrs.flags.contains(CodegenFnAttrFlags::RUSTC_STD_INTERNAL_SYMBOL) { + *can_be_internalized = false; + } + + Visibility::Hidden + } +} + +fn default_visibility(tcx: TyCtxt<'_>, id: DefId, is_generic: bool) -> Visibility { + if !tcx.sess.target.default_hidden_visibility { + return Visibility::Default; + } + + // Generic functions never have export-level C. + if is_generic { + return Visibility::Hidden; + } + + // Things with export level C don't get instantiated in + // downstream crates. + if !id.is_local() { + return Visibility::Hidden; + } + + // C-export level items remain at `Default`, all other internal + // items become `Hidden`. + match tcx.reachable_non_generics(id.krate).get(&id) { + Some(SymbolExportInfo { level: SymbolExportLevel::C, .. }) => Visibility::Default, + _ => Visibility::Hidden, + } +} + +fn debug_dump<'a, 'tcx: 'a>( + tcx: TyCtxt<'tcx>, + label: &str, + cgus: &[CodegenUnit<'tcx>], + (unique_inlined_items, unique_inlined_size): (usize, usize), +) { + let dump = move || { + use std::fmt::Write; + + let mut num_cgus = 0; + let mut all_cgu_sizes = Vec::new(); + + // Note: every unique root item is placed exactly once, so the number + // of unique root items always equals the number of placed root items. + + let mut root_items = 0; + // unique_inlined_items is passed in above. + let mut placed_inlined_items = 0; + + let mut root_size = 0; + // unique_inlined_size is passed in above. + let mut placed_inlined_size = 0; + + for cgu in cgus.iter() { + num_cgus += 1; + all_cgu_sizes.push(cgu.size_estimate()); + + for (item, _) in cgu.items() { + match item.instantiation_mode(tcx) { + InstantiationMode::GloballyShared { .. } => { + root_items += 1; + root_size += item.size_estimate(tcx); + } + InstantiationMode::LocalCopy => { + placed_inlined_items += 1; + placed_inlined_size += item.size_estimate(tcx); + } + } + } + } + + all_cgu_sizes.sort_unstable_by_key(|&n| cmp::Reverse(n)); + + let unique_items = root_items + unique_inlined_items; + let placed_items = root_items + placed_inlined_items; + let items_ratio = placed_items as f64 / unique_items as f64; + + let unique_size = root_size + unique_inlined_size; + let placed_size = root_size + placed_inlined_size; + let size_ratio = placed_size as f64 / unique_size as f64; + + let mean_cgu_size = placed_size as f64 / num_cgus as f64; + + assert_eq!(placed_size, all_cgu_sizes.iter().sum::<usize>()); + + let s = &mut String::new(); + let _ = writeln!(s, "{label}"); + let _ = writeln!( + s, + "- unique items: {unique_items} ({root_items} root + {unique_inlined_items} inlined), \ + unique size: {unique_size} ({root_size} root + {unique_inlined_size} inlined)\n\ + - placed items: {placed_items} ({root_items} root + {placed_inlined_items} inlined), \ + placed size: {placed_size} ({root_size} root + {placed_inlined_size} inlined)\n\ + - placed/unique items ratio: {items_ratio:.2}, \ + placed/unique size ratio: {size_ratio:.2}\n\ + - CGUs: {num_cgus}, mean size: {mean_cgu_size:.1}, sizes: {}", + list(&all_cgu_sizes), + ); + let _ = writeln!(s); + + for (i, cgu) in cgus.iter().enumerate() { + let name = cgu.name(); + let size = cgu.size_estimate(); + let num_items = cgu.items().len(); + let mean_size = size as f64 / num_items as f64; + + let mut placed_item_sizes: Vec<_> = + cgu.items().iter().map(|(item, _)| item.size_estimate(tcx)).collect(); + placed_item_sizes.sort_unstable_by_key(|&n| cmp::Reverse(n)); + let sizes = list(&placed_item_sizes); + + let _ = writeln!(s, "- CGU[{i}]"); + let _ = writeln!(s, " - {name}, size: {size}"); + let _ = + writeln!(s, " - items: {num_items}, mean size: {mean_size:.1}, sizes: {sizes}",); + + for (item, linkage) in cgu.items_in_deterministic_order(tcx) { + let symbol_name = item.symbol_name(tcx).name; + let symbol_hash_start = symbol_name.rfind('h'); + let symbol_hash = symbol_hash_start.map_or("<no hash>", |i| &symbol_name[i..]); + let size = item.size_estimate(tcx); + let kind = match item.instantiation_mode(tcx) { + InstantiationMode::GloballyShared { .. } => "root", + InstantiationMode::LocalCopy => "inlined", + }; + let _ = with_no_trimmed_paths!(writeln!( + s, + " - {item} [{linkage:?}] [{symbol_hash}] ({kind}, size: {size})" + )); + } + + let _ = writeln!(s); + } + + return std::mem::take(s); + + // Converts a slice to a string, capturing repetitions to save space. + // E.g. `[4, 4, 4, 3, 2, 1, 1, 1, 1, 1]` -> "[4 (x3), 3, 2, 1 (x5)]". + fn list(ns: &[usize]) -> String { + let mut v = Vec::new(); + if ns.is_empty() { + return "[]".to_string(); + } + + let mut elem = |curr, curr_count| { + if curr_count == 1 { + v.push(format!("{curr}")); + } else { + v.push(format!("{curr} (x{curr_count})")); + } + }; + + let mut curr = ns[0]; + let mut curr_count = 1; + + for &n in &ns[1..] { + if n != curr { + elem(curr, curr_count); + curr = n; + curr_count = 1; + } else { + curr_count += 1; + } + } + elem(curr, curr_count); + + let mut s = "[".to_string(); + s.push_str(&v.join(", ")); + s.push_str("]"); + s + } + }; + + debug!("{}", dump()); +} + +#[inline(never)] // give this a place in the profiler +fn assert_symbols_are_distinct<'a, 'tcx, I>(tcx: TyCtxt<'tcx>, mono_items: I) +where + I: Iterator<Item = &'a MonoItem<'tcx>>, + 'tcx: 'a, +{ + let _prof_timer = tcx.prof.generic_activity("assert_symbols_are_distinct"); + + let mut symbols: Vec<_> = + mono_items.map(|mono_item| (mono_item, mono_item.symbol_name(tcx))).collect(); + + symbols.sort_by_key(|sym| sym.1); + + for &[(mono_item1, ref sym1), (mono_item2, ref sym2)] in symbols.array_windows() { + if sym1 == sym2 { + let span1 = mono_item1.local_span(tcx); + let span2 = mono_item2.local_span(tcx); + + // Deterministically select one of the spans for error reporting + let span = match (span1, span2) { + (Some(span1), Some(span2)) => { + Some(if span1.lo().0 > span2.lo().0 { span1 } else { span2 }) + } + (span1, span2) => span1.or(span2), + }; + + tcx.sess.emit_fatal(SymbolAlreadyDefined { span, symbol: sym1.to_string() }); + } + } +} + +fn collect_and_partition_mono_items(tcx: TyCtxt<'_>, (): ()) -> (&DefIdSet, &[CodegenUnit<'_>]) { + let collection_mode = match tcx.sess.opts.unstable_opts.print_mono_items { + Some(ref s) => { + let mode = s.to_lowercase(); + let mode = mode.trim(); + if mode == "eager" { + MonoItemCollectionMode::Eager + } else { + if mode != "lazy" { + tcx.sess.emit_warning(UnknownCguCollectionMode { mode }); + } + + MonoItemCollectionMode::Lazy + } + } + None => { + if tcx.sess.link_dead_code() { + MonoItemCollectionMode::Eager + } else { + MonoItemCollectionMode::Lazy + } + } + }; + + let (items, usage_map) = collector::collect_crate_mono_items(tcx, collection_mode); + + tcx.sess.abort_if_errors(); + + let (codegen_units, _) = tcx.sess.time("partition_and_assert_distinct_symbols", || { + sync::join( + || { + let mut codegen_units = partition(tcx, items.iter().copied(), &usage_map); + codegen_units[0].make_primary(); + &*tcx.arena.alloc_from_iter(codegen_units) + }, + || assert_symbols_are_distinct(tcx, items.iter()), + ) + }); + + if tcx.prof.enabled() { + // Record CGU size estimates for self-profiling. + for cgu in codegen_units { + tcx.prof.artifact_size( + "codegen_unit_size_estimate", + cgu.name().as_str(), + cgu.size_estimate() as u64, + ); + } + } + + let mono_items: DefIdSet = items + .iter() + .filter_map(|mono_item| match *mono_item { + MonoItem::Fn(ref instance) => Some(instance.def_id()), + MonoItem::Static(def_id) => Some(def_id), + _ => None, + }) + .collect(); + + // Output monomorphization stats per def_id + if let SwitchWithOptPath::Enabled(ref path) = tcx.sess.opts.unstable_opts.dump_mono_stats { + if let Err(err) = + dump_mono_items_stats(tcx, &codegen_units, path, tcx.crate_name(LOCAL_CRATE)) + { + tcx.sess.emit_fatal(CouldntDumpMonoStats { error: err.to_string() }); + } + } + + if tcx.sess.opts.unstable_opts.print_mono_items.is_some() { + let mut item_to_cgus: FxHashMap<_, Vec<_>> = Default::default(); + + for cgu in codegen_units { + for (&mono_item, &linkage) in cgu.items() { + item_to_cgus.entry(mono_item).or_default().push((cgu.name(), linkage)); + } + } + + let mut item_keys: Vec<_> = items + .iter() + .map(|i| { + let mut output = with_no_trimmed_paths!(i.to_string()); + output.push_str(" @@"); + let mut empty = Vec::new(); + let cgus = item_to_cgus.get_mut(i).unwrap_or(&mut empty); + cgus.sort_by_key(|(name, _)| *name); + cgus.dedup(); + for &(ref cgu_name, (linkage, _)) in cgus.iter() { + output.push(' '); + output.push_str(cgu_name.as_str()); + + let linkage_abbrev = match linkage { + Linkage::External => "External", + Linkage::AvailableExternally => "Available", + Linkage::LinkOnceAny => "OnceAny", + Linkage::LinkOnceODR => "OnceODR", + Linkage::WeakAny => "WeakAny", + Linkage::WeakODR => "WeakODR", + Linkage::Appending => "Appending", + Linkage::Internal => "Internal", + Linkage::Private => "Private", + Linkage::ExternalWeak => "ExternalWeak", + Linkage::Common => "Common", + }; + + output.push('['); + output.push_str(linkage_abbrev); + output.push(']'); + } + output + }) + .collect(); + + item_keys.sort(); + + for item in item_keys { + println!("MONO_ITEM {item}"); + } + } + + (tcx.arena.alloc(mono_items), codegen_units) +} + +/// Outputs stats about instantiation counts and estimated size, per `MonoItem`'s +/// def, to a file in the given output directory. +fn dump_mono_items_stats<'tcx>( + tcx: TyCtxt<'tcx>, + codegen_units: &[CodegenUnit<'tcx>], + output_directory: &Option<PathBuf>, + crate_name: Symbol, +) -> Result<(), Box<dyn std::error::Error>> { + let output_directory = if let Some(ref directory) = output_directory { + fs::create_dir_all(directory)?; + directory + } else { + Path::new(".") + }; + + let format = tcx.sess.opts.unstable_opts.dump_mono_stats_format; + let ext = format.extension(); + let filename = format!("{crate_name}.mono_items.{ext}"); + let output_path = output_directory.join(&filename); + let file = File::create(&output_path)?; + let mut file = BufWriter::new(file); + + // Gather instantiated mono items grouped by def_id + let mut items_per_def_id: FxHashMap<_, Vec<_>> = Default::default(); + for cgu in codegen_units { + for (&mono_item, _) in cgu.items() { + // Avoid variable-sized compiler-generated shims + if mono_item.is_user_defined() { + items_per_def_id.entry(mono_item.def_id()).or_default().push(mono_item); + } + } + } + + #[derive(serde::Serialize)] + struct MonoItem { + name: String, + instantiation_count: usize, + size_estimate: usize, + total_estimate: usize, + } + + // Output stats sorted by total instantiated size, from heaviest to lightest + let mut stats: Vec<_> = items_per_def_id + .into_iter() + .map(|(def_id, items)| { + let name = with_no_trimmed_paths!(tcx.def_path_str(def_id)); + let instantiation_count = items.len(); + let size_estimate = items[0].size_estimate(tcx); + let total_estimate = instantiation_count * size_estimate; + MonoItem { name, instantiation_count, size_estimate, total_estimate } + }) + .collect(); + stats.sort_unstable_by_key(|item| cmp::Reverse(item.total_estimate)); + + if !stats.is_empty() { + match format { + DumpMonoStatsFormat::Json => serde_json::to_writer(file, &stats)?, + DumpMonoStatsFormat::Markdown => { + writeln!( + file, + "| Item | Instantiation count | Estimated Cost Per Instantiation | Total Estimated Cost |" + )?; + writeln!(file, "| --- | ---: | ---: | ---: |")?; + + for MonoItem { name, instantiation_count, size_estimate, total_estimate } in stats { + writeln!( + file, + "| `{name}` | {instantiation_count} | {size_estimate} | {total_estimate} |" + )?; + } + } + } + } + + Ok(()) +} + +fn codegened_and_inlined_items(tcx: TyCtxt<'_>, (): ()) -> &DefIdSet { + let (items, cgus) = tcx.collect_and_partition_mono_items(()); + let mut visited = DefIdSet::default(); + let mut result = items.clone(); + + for cgu in cgus { + for (item, _) in cgu.items() { + if let MonoItem::Fn(ref instance) = item { + let did = instance.def_id(); + if !visited.insert(did) { + continue; + } + let body = tcx.instance_mir(instance.def); + for block in body.basic_blocks.iter() { + for statement in &block.statements { + let mir::StatementKind::Coverage(_) = statement.kind else { continue }; + let scope = statement.source_info.scope; + if let Some(inlined) = scope.inlined_instance(&body.source_scopes) { + result.insert(inlined.def_id()); + } + } + } + } + } + } + + tcx.arena.alloc(result) +} + +pub fn provide(providers: &mut Providers) { + providers.collect_and_partition_mono_items = collect_and_partition_mono_items; + providers.codegened_and_inlined_items = codegened_and_inlined_items; + + providers.is_codegened_item = |tcx, def_id| { + let (all_mono_items, _) = tcx.collect_and_partition_mono_items(()); + all_mono_items.contains(&def_id) + }; + + providers.codegen_unit = |tcx, name| { + let (_, all) = tcx.collect_and_partition_mono_items(()); + all.iter() + .find(|cgu| cgu.name() == name) + .unwrap_or_else(|| panic!("failed to find cgu with name {name:?}")) + }; +} diff --git a/compiler/rustc_monomorphize/src/partitioning/default.rs b/compiler/rustc_monomorphize/src/partitioning/default.rs deleted file mode 100644 index 603b3ddc1..000000000 --- a/compiler/rustc_monomorphize/src/partitioning/default.rs +++ /dev/null @@ -1,644 +0,0 @@ -use std::cmp; -use std::collections::hash_map::Entry; - -use rustc_data_structures::fx::{FxHashMap, FxHashSet}; -use rustc_hir::def::DefKind; -use rustc_hir::def_id::{DefId, LOCAL_CRATE}; -use rustc_hir::definitions::DefPathDataName; -use rustc_middle::middle::codegen_fn_attrs::CodegenFnAttrFlags; -use rustc_middle::middle::exported_symbols::{SymbolExportInfo, SymbolExportLevel}; -use rustc_middle::mir::mono::{CodegenUnit, CodegenUnitNameBuilder, Linkage, Visibility}; -use rustc_middle::mir::mono::{InstantiationMode, MonoItem}; -use rustc_middle::ty::print::characteristic_def_id_of_type; -use rustc_middle::ty::{self, visit::TypeVisitableExt, InstanceDef, TyCtxt}; -use rustc_span::symbol::Symbol; - -use super::PartitioningCx; -use crate::collector::InliningMap; -use crate::partitioning::{MonoItemPlacement, Partition, PlacedRootMonoItems}; - -pub struct DefaultPartitioning; - -impl<'tcx> Partition<'tcx> for DefaultPartitioning { - fn place_root_mono_items<I>( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - mono_items: &mut I, - ) -> PlacedRootMonoItems<'tcx> - where - I: Iterator<Item = MonoItem<'tcx>>, - { - let mut roots = FxHashSet::default(); - let mut codegen_units = FxHashMap::default(); - let is_incremental_build = cx.tcx.sess.opts.incremental.is_some(); - let mut internalization_candidates = FxHashSet::default(); - - // Determine if monomorphizations instantiated in this crate will be made - // available to downstream crates. This depends on whether we are in - // share-generics mode and whether the current crate can even have - // downstream crates. - let export_generics = - cx.tcx.sess.opts.share_generics() && cx.tcx.local_crate_exports_generics(); - - let cgu_name_builder = &mut CodegenUnitNameBuilder::new(cx.tcx); - let cgu_name_cache = &mut FxHashMap::default(); - - for mono_item in mono_items { - match mono_item.instantiation_mode(cx.tcx) { - InstantiationMode::GloballyShared { .. } => {} - InstantiationMode::LocalCopy => continue, - } - - let characteristic_def_id = characteristic_def_id_of_mono_item(cx.tcx, mono_item); - let is_volatile = is_incremental_build && mono_item.is_generic_fn(); - - let codegen_unit_name = match characteristic_def_id { - Some(def_id) => compute_codegen_unit_name( - cx.tcx, - cgu_name_builder, - def_id, - is_volatile, - cgu_name_cache, - ), - None => fallback_cgu_name(cgu_name_builder), - }; - - let codegen_unit = codegen_units - .entry(codegen_unit_name) - .or_insert_with(|| CodegenUnit::new(codegen_unit_name)); - - let mut can_be_internalized = true; - let (linkage, visibility) = mono_item_linkage_and_visibility( - cx.tcx, - &mono_item, - &mut can_be_internalized, - export_generics, - ); - if visibility == Visibility::Hidden && can_be_internalized { - internalization_candidates.insert(mono_item); - } - - codegen_unit.items_mut().insert(mono_item, (linkage, visibility)); - roots.insert(mono_item); - } - - // Always ensure we have at least one CGU; otherwise, if we have a - // crate with just types (for example), we could wind up with no CGU. - if codegen_units.is_empty() { - let codegen_unit_name = fallback_cgu_name(cgu_name_builder); - codegen_units.insert(codegen_unit_name, CodegenUnit::new(codegen_unit_name)); - } - - let codegen_units = codegen_units.into_values().collect(); - PlacedRootMonoItems { codegen_units, roots, internalization_candidates } - } - - fn merge_codegen_units( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut Vec<CodegenUnit<'tcx>>, - ) { - assert!(cx.target_cgu_count >= 1); - - // Note that at this point in time the `codegen_units` here may not be - // in a deterministic order (but we know they're deterministically the - // same set). We want this merging to produce a deterministic ordering - // of codegen units from the input. - // - // Due to basically how we've implemented the merging below (merge the - // two smallest into each other) we're sure to start off with a - // deterministic order (sorted by name). This'll mean that if two cgus - // have the same size the stable sort below will keep everything nice - // and deterministic. - codegen_units.sort_by(|a, b| a.name().as_str().cmp(b.name().as_str())); - - // This map keeps track of what got merged into what. - let mut cgu_contents: FxHashMap<Symbol, Vec<Symbol>> = - codegen_units.iter().map(|cgu| (cgu.name(), vec![cgu.name()])).collect(); - - // Merge the two smallest codegen units until the target size is - // reached. - while codegen_units.len() > cx.target_cgu_count { - // Sort small cgus to the back - codegen_units.sort_by_cached_key(|cgu| cmp::Reverse(cgu.size_estimate())); - let mut smallest = codegen_units.pop().unwrap(); - let second_smallest = codegen_units.last_mut().unwrap(); - - // Move the mono-items from `smallest` to `second_smallest` - second_smallest.modify_size_estimate(smallest.size_estimate()); - for (k, v) in smallest.items_mut().drain() { - second_smallest.items_mut().insert(k, v); - } - - // Record that `second_smallest` now contains all the stuff that was - // in `smallest` before. - let mut consumed_cgu_names = cgu_contents.remove(&smallest.name()).unwrap(); - cgu_contents.get_mut(&second_smallest.name()).unwrap().append(&mut consumed_cgu_names); - - debug!( - "CodegenUnit {} merged into CodegenUnit {}", - smallest.name(), - second_smallest.name() - ); - } - - let cgu_name_builder = &mut CodegenUnitNameBuilder::new(cx.tcx); - - if cx.tcx.sess.opts.incremental.is_some() { - // If we are doing incremental compilation, we want CGU names to - // reflect the path of the source level module they correspond to. - // For CGUs that contain the code of multiple modules because of the - // merging done above, we use a concatenation of the names of all - // contained CGUs. - let new_cgu_names: FxHashMap<Symbol, String> = cgu_contents - .into_iter() - // This `filter` makes sure we only update the name of CGUs that - // were actually modified by merging. - .filter(|(_, cgu_contents)| cgu_contents.len() > 1) - .map(|(current_cgu_name, cgu_contents)| { - let mut cgu_contents: Vec<&str> = - cgu_contents.iter().map(|s| s.as_str()).collect(); - - // Sort the names, so things are deterministic and easy to - // predict. We are sorting primitive `&str`s here so we can - // use unstable sort. - cgu_contents.sort_unstable(); - - (current_cgu_name, cgu_contents.join("--")) - }) - .collect(); - - for cgu in codegen_units.iter_mut() { - if let Some(new_cgu_name) = new_cgu_names.get(&cgu.name()) { - if cx.tcx.sess.opts.unstable_opts.human_readable_cgu_names { - cgu.set_name(Symbol::intern(&new_cgu_name)); - } else { - // If we don't require CGU names to be human-readable, - // we use a fixed length hash of the composite CGU name - // instead. - let new_cgu_name = CodegenUnit::mangle_name(&new_cgu_name); - cgu.set_name(Symbol::intern(&new_cgu_name)); - } - } - } - } else { - // If we are compiling non-incrementally we just generate simple CGU - // names containing an index. - for (index, cgu) in codegen_units.iter_mut().enumerate() { - let numbered_codegen_unit_name = - cgu_name_builder.build_cgu_name_no_mangle(LOCAL_CRATE, &["cgu"], Some(index)); - cgu.set_name(numbered_codegen_unit_name); - } - } - } - - fn place_inlined_mono_items( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut [CodegenUnit<'tcx>], - roots: FxHashSet<MonoItem<'tcx>>, - ) -> FxHashMap<MonoItem<'tcx>, MonoItemPlacement> { - let mut mono_item_placements = FxHashMap::default(); - - let single_codegen_unit = codegen_units.len() == 1; - - for old_codegen_unit in codegen_units.iter_mut() { - // Collect all items that need to be available in this codegen unit. - let mut reachable = FxHashSet::default(); - for root in old_codegen_unit.items().keys() { - follow_inlining(*root, cx.inlining_map, &mut reachable); - } - - let mut new_codegen_unit = CodegenUnit::new(old_codegen_unit.name()); - - // Add all monomorphizations that are not already there. - for mono_item in reachable { - if let Some(linkage) = old_codegen_unit.items().get(&mono_item) { - // This is a root, just copy it over. - new_codegen_unit.items_mut().insert(mono_item, *linkage); - } else { - if roots.contains(&mono_item) { - bug!( - "GloballyShared mono-item inlined into other CGU: \ - {:?}", - mono_item - ); - } - - // This is a CGU-private copy. - new_codegen_unit - .items_mut() - .insert(mono_item, (Linkage::Internal, Visibility::Default)); - } - - if !single_codegen_unit { - // If there is more than one codegen unit, we need to keep track - // in which codegen units each monomorphization is placed. - match mono_item_placements.entry(mono_item) { - Entry::Occupied(e) => { - let placement = e.into_mut(); - debug_assert!(match *placement { - MonoItemPlacement::SingleCgu { cgu_name } => { - cgu_name != new_codegen_unit.name() - } - MonoItemPlacement::MultipleCgus => true, - }); - *placement = MonoItemPlacement::MultipleCgus; - } - Entry::Vacant(e) => { - e.insert(MonoItemPlacement::SingleCgu { - cgu_name: new_codegen_unit.name(), - }); - } - } - } - } - - *old_codegen_unit = new_codegen_unit; - } - - return mono_item_placements; - - fn follow_inlining<'tcx>( - mono_item: MonoItem<'tcx>, - inlining_map: &InliningMap<'tcx>, - visited: &mut FxHashSet<MonoItem<'tcx>>, - ) { - if !visited.insert(mono_item) { - return; - } - - inlining_map.with_inlining_candidates(mono_item, |target| { - follow_inlining(target, inlining_map, visited); - }); - } - } - - fn internalize_symbols( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut [CodegenUnit<'tcx>], - mono_item_placements: FxHashMap<MonoItem<'tcx>, MonoItemPlacement>, - internalization_candidates: FxHashSet<MonoItem<'tcx>>, - ) { - if codegen_units.len() == 1 { - // Fast path for when there is only one codegen unit. In this case we - // can internalize all candidates, since there is nowhere else they - // could be accessed from. - for cgu in codegen_units { - for candidate in &internalization_candidates { - cgu.items_mut().insert(*candidate, (Linkage::Internal, Visibility::Default)); - } - } - - return; - } - - // Build a map from every monomorphization to all the monomorphizations that - // reference it. - let mut accessor_map: FxHashMap<MonoItem<'tcx>, Vec<MonoItem<'tcx>>> = Default::default(); - cx.inlining_map.iter_accesses(|accessor, accessees| { - for accessee in accessees { - accessor_map.entry(*accessee).or_default().push(accessor); - } - }); - - // For each internalization candidates in each codegen unit, check if it is - // accessed from outside its defining codegen unit. - for cgu in codegen_units { - let home_cgu = MonoItemPlacement::SingleCgu { cgu_name: cgu.name() }; - - for (accessee, linkage_and_visibility) in cgu.items_mut() { - if !internalization_candidates.contains(accessee) { - // This item is no candidate for internalizing, so skip it. - continue; - } - debug_assert_eq!(mono_item_placements[accessee], home_cgu); - - if let Some(accessors) = accessor_map.get(accessee) { - if accessors - .iter() - .filter_map(|accessor| { - // Some accessors might not have been - // instantiated. We can safely ignore those. - mono_item_placements.get(accessor) - }) - .any(|placement| *placement != home_cgu) - { - // Found an accessor from another CGU, so skip to the next - // item without marking this one as internal. - continue; - } - } - - // If we got here, we did not find any accesses from other CGUs, - // so it's fine to make this monomorphization internal. - *linkage_and_visibility = (Linkage::Internal, Visibility::Default); - } - } - } -} - -fn characteristic_def_id_of_mono_item<'tcx>( - tcx: TyCtxt<'tcx>, - mono_item: MonoItem<'tcx>, -) -> Option<DefId> { - match mono_item { - MonoItem::Fn(instance) => { - let def_id = match instance.def { - ty::InstanceDef::Item(def) => def, - ty::InstanceDef::VTableShim(..) - | ty::InstanceDef::ReifyShim(..) - | ty::InstanceDef::FnPtrShim(..) - | ty::InstanceDef::ClosureOnceShim { .. } - | ty::InstanceDef::Intrinsic(..) - | ty::InstanceDef::DropGlue(..) - | ty::InstanceDef::Virtual(..) - | ty::InstanceDef::CloneShim(..) - | ty::InstanceDef::ThreadLocalShim(..) - | ty::InstanceDef::FnPtrAddrShim(..) => return None, - }; - - // If this is a method, we want to put it into the same module as - // its self-type. If the self-type does not provide a characteristic - // DefId, we use the location of the impl after all. - - if tcx.trait_of_item(def_id).is_some() { - let self_ty = instance.substs.type_at(0); - // This is a default implementation of a trait method. - return characteristic_def_id_of_type(self_ty).or(Some(def_id)); - } - - if let Some(impl_def_id) = tcx.impl_of_method(def_id) { - if tcx.sess.opts.incremental.is_some() - && tcx.trait_id_of_impl(impl_def_id) == tcx.lang_items().drop_trait() - { - // Put `Drop::drop` into the same cgu as `drop_in_place` - // since `drop_in_place` is the only thing that can - // call it. - return None; - } - - // When polymorphization is enabled, methods which do not depend on their generic - // parameters, but the self-type of their impl block do will fail to normalize. - if !tcx.sess.opts.unstable_opts.polymorphize || !instance.has_param() { - // This is a method within an impl, find out what the self-type is: - let impl_self_ty = tcx.subst_and_normalize_erasing_regions( - instance.substs, - ty::ParamEnv::reveal_all(), - tcx.type_of(impl_def_id), - ); - if let Some(def_id) = characteristic_def_id_of_type(impl_self_ty) { - return Some(def_id); - } - } - } - - Some(def_id) - } - MonoItem::Static(def_id) => Some(def_id), - MonoItem::GlobalAsm(item_id) => Some(item_id.owner_id.to_def_id()), - } -} - -fn compute_codegen_unit_name( - tcx: TyCtxt<'_>, - name_builder: &mut CodegenUnitNameBuilder<'_>, - def_id: DefId, - volatile: bool, - cache: &mut CguNameCache, -) -> Symbol { - // Find the innermost module that is not nested within a function. - let mut current_def_id = def_id; - let mut cgu_def_id = None; - // Walk backwards from the item we want to find the module for. - loop { - if current_def_id.is_crate_root() { - if cgu_def_id.is_none() { - // If we have not found a module yet, take the crate root. - cgu_def_id = Some(def_id.krate.as_def_id()); - } - break; - } else if tcx.def_kind(current_def_id) == DefKind::Mod { - if cgu_def_id.is_none() { - cgu_def_id = Some(current_def_id); - } - } else { - // If we encounter something that is not a module, throw away - // any module that we've found so far because we now know that - // it is nested within something else. - cgu_def_id = None; - } - - current_def_id = tcx.parent(current_def_id); - } - - let cgu_def_id = cgu_def_id.unwrap(); - - *cache.entry((cgu_def_id, volatile)).or_insert_with(|| { - let def_path = tcx.def_path(cgu_def_id); - - let components = def_path.data.iter().map(|part| match part.data.name() { - DefPathDataName::Named(name) => name, - DefPathDataName::Anon { .. } => unreachable!(), - }); - - let volatile_suffix = volatile.then_some("volatile"); - - name_builder.build_cgu_name(def_path.krate, components, volatile_suffix) - }) -} - -// Anything we can't find a proper codegen unit for goes into this. -fn fallback_cgu_name(name_builder: &mut CodegenUnitNameBuilder<'_>) -> Symbol { - name_builder.build_cgu_name(LOCAL_CRATE, &["fallback"], Some("cgu")) -} - -fn mono_item_linkage_and_visibility<'tcx>( - tcx: TyCtxt<'tcx>, - mono_item: &MonoItem<'tcx>, - can_be_internalized: &mut bool, - export_generics: bool, -) -> (Linkage, Visibility) { - if let Some(explicit_linkage) = mono_item.explicit_linkage(tcx) { - return (explicit_linkage, Visibility::Default); - } - let vis = mono_item_visibility(tcx, mono_item, can_be_internalized, export_generics); - (Linkage::External, vis) -} - -type CguNameCache = FxHashMap<(DefId, bool), Symbol>; - -fn static_visibility<'tcx>( - tcx: TyCtxt<'tcx>, - can_be_internalized: &mut bool, - def_id: DefId, -) -> Visibility { - if tcx.is_reachable_non_generic(def_id) { - *can_be_internalized = false; - default_visibility(tcx, def_id, false) - } else { - Visibility::Hidden - } -} - -fn mono_item_visibility<'tcx>( - tcx: TyCtxt<'tcx>, - mono_item: &MonoItem<'tcx>, - can_be_internalized: &mut bool, - export_generics: bool, -) -> Visibility { - let instance = match mono_item { - // This is pretty complicated; see below. - MonoItem::Fn(instance) => instance, - - // Misc handling for generics and such, but otherwise: - MonoItem::Static(def_id) => return static_visibility(tcx, can_be_internalized, *def_id), - MonoItem::GlobalAsm(item_id) => { - return static_visibility(tcx, can_be_internalized, item_id.owner_id.to_def_id()); - } - }; - - let def_id = match instance.def { - InstanceDef::Item(def_id) | InstanceDef::DropGlue(def_id, Some(_)) => def_id, - - // We match the visibility of statics here - InstanceDef::ThreadLocalShim(def_id) => { - return static_visibility(tcx, can_be_internalized, def_id); - } - - // These are all compiler glue and such, never exported, always hidden. - InstanceDef::VTableShim(..) - | InstanceDef::ReifyShim(..) - | InstanceDef::FnPtrShim(..) - | InstanceDef::Virtual(..) - | InstanceDef::Intrinsic(..) - | InstanceDef::ClosureOnceShim { .. } - | InstanceDef::DropGlue(..) - | InstanceDef::CloneShim(..) - | InstanceDef::FnPtrAddrShim(..) => return Visibility::Hidden, - }; - - // The `start_fn` lang item is actually a monomorphized instance of a - // function in the standard library, used for the `main` function. We don't - // want to export it so we tag it with `Hidden` visibility but this symbol - // is only referenced from the actual `main` symbol which we unfortunately - // don't know anything about during partitioning/collection. As a result we - // forcibly keep this symbol out of the `internalization_candidates` set. - // - // FIXME: eventually we don't want to always force this symbol to have - // hidden visibility, it should indeed be a candidate for - // internalization, but we have to understand that it's referenced - // from the `main` symbol we'll generate later. - // - // This may be fixable with a new `InstanceDef` perhaps? Unsure! - if tcx.lang_items().start_fn() == Some(def_id) { - *can_be_internalized = false; - return Visibility::Hidden; - } - - let is_generic = instance.substs.non_erasable_generics().next().is_some(); - - // Upstream `DefId` instances get different handling than local ones. - let Some(def_id) = def_id.as_local() else { - return if export_generics && is_generic { - // If it is an upstream monomorphization and we export generics, we must make - // it available to downstream crates. - *can_be_internalized = false; - default_visibility(tcx, def_id, true) - } else { - Visibility::Hidden - }; - }; - - if is_generic { - if export_generics { - if tcx.is_unreachable_local_definition(def_id) { - // This instance cannot be used from another crate. - Visibility::Hidden - } else { - // This instance might be useful in a downstream crate. - *can_be_internalized = false; - default_visibility(tcx, def_id.to_def_id(), true) - } - } else { - // We are not exporting generics or the definition is not reachable - // for downstream crates, we can internalize its instantiations. - Visibility::Hidden - } - } else { - // If this isn't a generic function then we mark this a `Default` if - // this is a reachable item, meaning that it's a symbol other crates may - // access when they link to us. - if tcx.is_reachable_non_generic(def_id.to_def_id()) { - *can_be_internalized = false; - debug_assert!(!is_generic); - return default_visibility(tcx, def_id.to_def_id(), false); - } - - // If this isn't reachable then we're gonna tag this with `Hidden` - // visibility. In some situations though we'll want to prevent this - // symbol from being internalized. - // - // There's two categories of items here: - // - // * First is weak lang items. These are basically mechanisms for - // libcore to forward-reference symbols defined later in crates like - // the standard library or `#[panic_handler]` definitions. The - // definition of these weak lang items needs to be referencable by - // libcore, so we're no longer a candidate for internalization. - // Removal of these functions can't be done by LLVM but rather must be - // done by the linker as it's a non-local decision. - // - // * Second is "std internal symbols". Currently this is primarily used - // for allocator symbols. Allocators are a little weird in their - // implementation, but the idea is that the compiler, at the last - // minute, defines an allocator with an injected object file. The - // `alloc` crate references these symbols (`__rust_alloc`) and the - // definition doesn't get hooked up until a linked crate artifact is - // generated. - // - // The symbols synthesized by the compiler (`__rust_alloc`) are thin - // veneers around the actual implementation, some other symbol which - // implements the same ABI. These symbols (things like `__rg_alloc`, - // `__rdl_alloc`, `__rde_alloc`, etc), are all tagged with "std - // internal symbols". - // - // The std-internal symbols here **should not show up in a dll as an - // exported interface**, so they return `false` from - // `is_reachable_non_generic` above and we'll give them `Hidden` - // visibility below. Like the weak lang items, though, we can't let - // LLVM internalize them as this decision is left up to the linker to - // omit them, so prevent them from being internalized. - let attrs = tcx.codegen_fn_attrs(def_id); - if attrs.flags.contains(CodegenFnAttrFlags::RUSTC_STD_INTERNAL_SYMBOL) { - *can_be_internalized = false; - } - - Visibility::Hidden - } -} - -fn default_visibility(tcx: TyCtxt<'_>, id: DefId, is_generic: bool) -> Visibility { - if !tcx.sess.target.default_hidden_visibility { - return Visibility::Default; - } - - // Generic functions never have export-level C. - if is_generic { - return Visibility::Hidden; - } - - // Things with export level C don't get instantiated in - // downstream crates. - if !id.is_local() { - return Visibility::Hidden; - } - - // C-export level items remain at `Default`, all other internal - // items become `Hidden`. - match tcx.reachable_non_generics(id.krate).get(&id) { - Some(SymbolExportInfo { level: SymbolExportLevel::C, .. }) => Visibility::Default, - _ => Visibility::Hidden, - } -} diff --git a/compiler/rustc_monomorphize/src/partitioning/mod.rs b/compiler/rustc_monomorphize/src/partitioning/mod.rs deleted file mode 100644 index d0b23ca9e..000000000 --- a/compiler/rustc_monomorphize/src/partitioning/mod.rs +++ /dev/null @@ -1,673 +0,0 @@ -//! Partitioning Codegen Units for Incremental Compilation -//! ====================================================== -//! -//! The task of this module is to take the complete set of monomorphizations of -//! a crate and produce a set of codegen units from it, where a codegen unit -//! is a named set of (mono-item, linkage) pairs. That is, this module -//! decides which monomorphization appears in which codegen units with which -//! linkage. The following paragraphs describe some of the background on the -//! partitioning scheme. -//! -//! The most important opportunity for saving on compilation time with -//! incremental compilation is to avoid re-codegenning and re-optimizing code. -//! Since the unit of codegen and optimization for LLVM is "modules" or, how -//! we call them "codegen units", the particulars of how much time can be saved -//! by incremental compilation are tightly linked to how the output program is -//! partitioned into these codegen units prior to passing it to LLVM -- -//! especially because we have to treat codegen units as opaque entities once -//! they are created: There is no way for us to incrementally update an existing -//! LLVM module and so we have to build any such module from scratch if it was -//! affected by some change in the source code. -//! -//! From that point of view it would make sense to maximize the number of -//! codegen units by, for example, putting each function into its own module. -//! That way only those modules would have to be re-compiled that were actually -//! affected by some change, minimizing the number of functions that could have -//! been re-used but just happened to be located in a module that is -//! re-compiled. -//! -//! However, since LLVM optimization does not work across module boundaries, -//! using such a highly granular partitioning would lead to very slow runtime -//! code since it would effectively prohibit inlining and other inter-procedure -//! optimizations. We want to avoid that as much as possible. -//! -//! Thus we end up with a trade-off: The bigger the codegen units, the better -//! LLVM's optimizer can do its work, but also the smaller the compilation time -//! reduction we get from incremental compilation. -//! -//! Ideally, we would create a partitioning such that there are few big codegen -//! units with few interdependencies between them. For now though, we use the -//! following heuristic to determine the partitioning: -//! -//! - There are two codegen units for every source-level module: -//! - One for "stable", that is non-generic, code -//! - One for more "volatile" code, i.e., monomorphized instances of functions -//! defined in that module -//! -//! In order to see why this heuristic makes sense, let's take a look at when a -//! codegen unit can get invalidated: -//! -//! 1. The most straightforward case is when the BODY of a function or global -//! changes. Then any codegen unit containing the code for that item has to be -//! re-compiled. Note that this includes all codegen units where the function -//! has been inlined. -//! -//! 2. The next case is when the SIGNATURE of a function or global changes. In -//! this case, all codegen units containing a REFERENCE to that item have to be -//! re-compiled. This is a superset of case 1. -//! -//! 3. The final and most subtle case is when a REFERENCE to a generic function -//! is added or removed somewhere. Even though the definition of the function -//! might be unchanged, a new REFERENCE might introduce a new monomorphized -//! instance of this function which has to be placed and compiled somewhere. -//! Conversely, when removing a REFERENCE, it might have been the last one with -//! that particular set of generic arguments and thus we have to remove it. -//! -//! From the above we see that just using one codegen unit per source-level -//! module is not such a good idea, since just adding a REFERENCE to some -//! generic item somewhere else would invalidate everything within the module -//! containing the generic item. The heuristic above reduces this detrimental -//! side-effect of references a little by at least not touching the non-generic -//! code of the module. -//! -//! A Note on Inlining -//! ------------------ -//! As briefly mentioned above, in order for LLVM to be able to inline a -//! function call, the body of the function has to be available in the LLVM -//! module where the call is made. This has a few consequences for partitioning: -//! -//! - The partitioning algorithm has to take care of placing functions into all -//! codegen units where they should be available for inlining. It also has to -//! decide on the correct linkage for these functions. -//! -//! - The partitioning algorithm has to know which functions are likely to get -//! inlined, so it can distribute function instantiations accordingly. Since -//! there is no way of knowing for sure which functions LLVM will decide to -//! inline in the end, we apply a heuristic here: Only functions marked with -//! `#[inline]` are considered for inlining by the partitioner. The current -//! implementation will not try to determine if a function is likely to be -//! inlined by looking at the functions definition. -//! -//! Note though that as a side-effect of creating a codegen units per -//! source-level module, functions from the same module will be available for -//! inlining, even when they are not marked `#[inline]`. - -mod default; - -use std::cmp; -use std::fs::{self, File}; -use std::io::{BufWriter, Write}; -use std::path::{Path, PathBuf}; - -use rustc_data_structures::fx::{FxHashMap, FxHashSet}; -use rustc_data_structures::sync; -use rustc_hir::def_id::{DefIdSet, LOCAL_CRATE}; -use rustc_middle::mir; -use rustc_middle::mir::mono::MonoItem; -use rustc_middle::mir::mono::{CodegenUnit, Linkage}; -use rustc_middle::query::Providers; -use rustc_middle::ty::print::with_no_trimmed_paths; -use rustc_middle::ty::TyCtxt; -use rustc_session::config::{DumpMonoStatsFormat, SwitchWithOptPath}; -use rustc_span::symbol::Symbol; - -use crate::collector::InliningMap; -use crate::collector::{self, MonoItemCollectionMode}; -use crate::errors::{ - CouldntDumpMonoStats, SymbolAlreadyDefined, UnknownCguCollectionMode, UnknownPartitionStrategy, -}; - -enum Partitioner { - Default(default::DefaultPartitioning), - // Other partitioning strategies can go here. - Unknown, -} - -impl<'tcx> Partition<'tcx> for Partitioner { - fn place_root_mono_items<I>( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - mono_items: &mut I, - ) -> PlacedRootMonoItems<'tcx> - where - I: Iterator<Item = MonoItem<'tcx>>, - { - match self { - Partitioner::Default(partitioner) => partitioner.place_root_mono_items(cx, mono_items), - Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy), - } - } - - fn merge_codegen_units( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut Vec<CodegenUnit<'tcx>>, - ) { - match self { - Partitioner::Default(partitioner) => partitioner.merge_codegen_units(cx, codegen_units), - Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy), - } - } - - fn place_inlined_mono_items( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut [CodegenUnit<'tcx>], - roots: FxHashSet<MonoItem<'tcx>>, - ) -> FxHashMap<MonoItem<'tcx>, MonoItemPlacement> { - match self { - Partitioner::Default(partitioner) => { - partitioner.place_inlined_mono_items(cx, codegen_units, roots) - } - Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy), - } - } - - fn internalize_symbols( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut [CodegenUnit<'tcx>], - mono_item_placements: FxHashMap<MonoItem<'tcx>, MonoItemPlacement>, - internalization_candidates: FxHashSet<MonoItem<'tcx>>, - ) { - match self { - Partitioner::Default(partitioner) => partitioner.internalize_symbols( - cx, - codegen_units, - mono_item_placements, - internalization_candidates, - ), - Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy), - } - } -} - -struct PartitioningCx<'a, 'tcx> { - tcx: TyCtxt<'tcx>, - target_cgu_count: usize, - inlining_map: &'a InliningMap<'tcx>, -} - -pub struct PlacedRootMonoItems<'tcx> { - codegen_units: Vec<CodegenUnit<'tcx>>, - roots: FxHashSet<MonoItem<'tcx>>, - internalization_candidates: FxHashSet<MonoItem<'tcx>>, -} - -trait Partition<'tcx> { - fn place_root_mono_items<I>( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - mono_items: &mut I, - ) -> PlacedRootMonoItems<'tcx> - where - I: Iterator<Item = MonoItem<'tcx>>; - - fn merge_codegen_units( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut Vec<CodegenUnit<'tcx>>, - ); - - fn place_inlined_mono_items( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut [CodegenUnit<'tcx>], - roots: FxHashSet<MonoItem<'tcx>>, - ) -> FxHashMap<MonoItem<'tcx>, MonoItemPlacement>; - - fn internalize_symbols( - &mut self, - cx: &PartitioningCx<'_, 'tcx>, - codegen_units: &mut [CodegenUnit<'tcx>], - mono_item_placements: FxHashMap<MonoItem<'tcx>, MonoItemPlacement>, - internalization_candidates: FxHashSet<MonoItem<'tcx>>, - ); -} - -fn get_partitioner(tcx: TyCtxt<'_>) -> Partitioner { - let strategy = match &tcx.sess.opts.unstable_opts.cgu_partitioning_strategy { - None => "default", - Some(s) => &s[..], - }; - - match strategy { - "default" => Partitioner::Default(default::DefaultPartitioning), - _ => Partitioner::Unknown, - } -} - -fn partition<'tcx, I>( - tcx: TyCtxt<'tcx>, - mono_items: &mut I, - max_cgu_count: usize, - inlining_map: &InliningMap<'tcx>, -) -> Vec<CodegenUnit<'tcx>> -where - I: Iterator<Item = MonoItem<'tcx>>, -{ - let _prof_timer = tcx.prof.generic_activity("cgu_partitioning"); - - let mut partitioner = get_partitioner(tcx); - let cx = &PartitioningCx { tcx, target_cgu_count: max_cgu_count, inlining_map }; - // In the first step, we place all regular monomorphizations into their - // respective 'home' codegen unit. Regular monomorphizations are all - // functions and statics defined in the local crate. - let PlacedRootMonoItems { mut codegen_units, roots, internalization_candidates } = { - let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_place_roots"); - partitioner.place_root_mono_items(cx, mono_items) - }; - - for cgu in &mut codegen_units { - cgu.create_size_estimate(tcx); - } - - debug_dump(tcx, "INITIAL PARTITIONING", &codegen_units); - - // Merge until we have at most `max_cgu_count` codegen units. - // `merge_codegen_units` is responsible for updating the CGU size - // estimates. - { - let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_merge_cgus"); - partitioner.merge_codegen_units(cx, &mut codegen_units); - debug_dump(tcx, "POST MERGING", &codegen_units); - } - - // In the next step, we use the inlining map to determine which additional - // monomorphizations have to go into each codegen unit. These additional - // monomorphizations can be drop-glue, functions from external crates, and - // local functions the definition of which is marked with `#[inline]`. - let mono_item_placements = { - let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_place_inline_items"); - partitioner.place_inlined_mono_items(cx, &mut codegen_units, roots) - }; - - for cgu in &mut codegen_units { - cgu.create_size_estimate(tcx); - } - - debug_dump(tcx, "POST INLINING", &codegen_units); - - // Next we try to make as many symbols "internal" as possible, so LLVM has - // more freedom to optimize. - if !tcx.sess.link_dead_code() { - let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_internalize_symbols"); - partitioner.internalize_symbols( - cx, - &mut codegen_units, - mono_item_placements, - internalization_candidates, - ); - } - - let instrument_dead_code = - tcx.sess.instrument_coverage() && !tcx.sess.instrument_coverage_except_unused_functions(); - - if instrument_dead_code { - assert!( - codegen_units.len() > 0, - "There must be at least one CGU that code coverage data can be generated in." - ); - - // Find the smallest CGU that has exported symbols and put the dead - // function stubs in that CGU. We look for exported symbols to increase - // the likelihood the linker won't throw away the dead functions. - // FIXME(#92165): In order to truly resolve this, we need to make sure - // the object file (CGU) containing the dead function stubs is included - // in the final binary. This will probably require forcing these - // function symbols to be included via `-u` or `/include` linker args. - let mut cgus: Vec<_> = codegen_units.iter_mut().collect(); - cgus.sort_by_key(|cgu| cgu.size_estimate()); - - let dead_code_cgu = - if let Some(cgu) = cgus.into_iter().rev().find(|cgu| { - cgu.items().iter().any(|(_, (linkage, _))| *linkage == Linkage::External) - }) { - cgu - } else { - // If there are no CGUs that have externally linked items, - // then we just pick the first CGU as a fallback. - &mut codegen_units[0] - }; - dead_code_cgu.make_code_coverage_dead_code_cgu(); - } - - // Finally, sort by codegen unit name, so that we get deterministic results. - codegen_units.sort_by(|a, b| a.name().as_str().cmp(b.name().as_str())); - - debug_dump(tcx, "FINAL", &codegen_units); - - codegen_units -} - -/// For symbol internalization, we need to know whether a symbol/mono-item is -/// accessed from outside the codegen unit it is defined in. This type is used -/// to keep track of that. -#[derive(Clone, PartialEq, Eq, Debug)] -enum MonoItemPlacement { - SingleCgu { cgu_name: Symbol }, - MultipleCgus, -} - -fn debug_dump<'a, 'tcx: 'a>(tcx: TyCtxt<'tcx>, label: &str, cgus: &[CodegenUnit<'tcx>]) { - let dump = move || { - use std::fmt::Write; - - let num_cgus = cgus.len(); - let max = cgus.iter().map(|cgu| cgu.size_estimate()).max().unwrap(); - let min = cgus.iter().map(|cgu| cgu.size_estimate()).min().unwrap(); - let ratio = max as f64 / min as f64; - - let s = &mut String::new(); - let _ = writeln!( - s, - "{label} ({num_cgus} CodegenUnits, max={max}, min={min}, max/min={ratio:.1}):" - ); - for cgu in cgus { - let _ = - writeln!(s, "CodegenUnit {} estimated size {}:", cgu.name(), cgu.size_estimate()); - - for (mono_item, linkage) in cgu.items() { - let symbol_name = mono_item.symbol_name(tcx).name; - let symbol_hash_start = symbol_name.rfind('h'); - let symbol_hash = symbol_hash_start.map_or("<no hash>", |i| &symbol_name[i..]); - - let _ = with_no_trimmed_paths!(writeln!( - s, - " - {} [{:?}] [{}] estimated size {}", - mono_item, - linkage, - symbol_hash, - mono_item.size_estimate(tcx) - )); - } - - let _ = writeln!(s); - } - - std::mem::take(s) - }; - - debug!("{}", dump()); -} - -#[inline(never)] // give this a place in the profiler -fn assert_symbols_are_distinct<'a, 'tcx, I>(tcx: TyCtxt<'tcx>, mono_items: I) -where - I: Iterator<Item = &'a MonoItem<'tcx>>, - 'tcx: 'a, -{ - let _prof_timer = tcx.prof.generic_activity("assert_symbols_are_distinct"); - - let mut symbols: Vec<_> = - mono_items.map(|mono_item| (mono_item, mono_item.symbol_name(tcx))).collect(); - - symbols.sort_by_key(|sym| sym.1); - - for &[(mono_item1, ref sym1), (mono_item2, ref sym2)] in symbols.array_windows() { - if sym1 == sym2 { - let span1 = mono_item1.local_span(tcx); - let span2 = mono_item2.local_span(tcx); - - // Deterministically select one of the spans for error reporting - let span = match (span1, span2) { - (Some(span1), Some(span2)) => { - Some(if span1.lo().0 > span2.lo().0 { span1 } else { span2 }) - } - (span1, span2) => span1.or(span2), - }; - - tcx.sess.emit_fatal(SymbolAlreadyDefined { span, symbol: sym1.to_string() }); - } - } -} - -fn collect_and_partition_mono_items(tcx: TyCtxt<'_>, (): ()) -> (&DefIdSet, &[CodegenUnit<'_>]) { - let collection_mode = match tcx.sess.opts.unstable_opts.print_mono_items { - Some(ref s) => { - let mode = s.to_lowercase(); - let mode = mode.trim(); - if mode == "eager" { - MonoItemCollectionMode::Eager - } else { - if mode != "lazy" { - tcx.sess.emit_warning(UnknownCguCollectionMode { mode }); - } - - MonoItemCollectionMode::Lazy - } - } - None => { - if tcx.sess.link_dead_code() { - MonoItemCollectionMode::Eager - } else { - MonoItemCollectionMode::Lazy - } - } - }; - - let (items, inlining_map) = collector::collect_crate_mono_items(tcx, collection_mode); - - tcx.sess.abort_if_errors(); - - let (codegen_units, _) = tcx.sess.time("partition_and_assert_distinct_symbols", || { - sync::join( - || { - let mut codegen_units = partition( - tcx, - &mut items.iter().copied(), - tcx.sess.codegen_units(), - &inlining_map, - ); - codegen_units[0].make_primary(); - &*tcx.arena.alloc_from_iter(codegen_units) - }, - || assert_symbols_are_distinct(tcx, items.iter()), - ) - }); - - if tcx.prof.enabled() { - // Record CGU size estimates for self-profiling. - for cgu in codegen_units { - tcx.prof.artifact_size( - "codegen_unit_size_estimate", - cgu.name().as_str(), - cgu.size_estimate() as u64, - ); - } - } - - let mono_items: DefIdSet = items - .iter() - .filter_map(|mono_item| match *mono_item { - MonoItem::Fn(ref instance) => Some(instance.def_id()), - MonoItem::Static(def_id) => Some(def_id), - _ => None, - }) - .collect(); - - // Output monomorphization stats per def_id - if let SwitchWithOptPath::Enabled(ref path) = tcx.sess.opts.unstable_opts.dump_mono_stats { - if let Err(err) = - dump_mono_items_stats(tcx, &codegen_units, path, tcx.crate_name(LOCAL_CRATE)) - { - tcx.sess.emit_fatal(CouldntDumpMonoStats { error: err.to_string() }); - } - } - - if tcx.sess.opts.unstable_opts.print_mono_items.is_some() { - let mut item_to_cgus: FxHashMap<_, Vec<_>> = Default::default(); - - for cgu in codegen_units { - for (&mono_item, &linkage) in cgu.items() { - item_to_cgus.entry(mono_item).or_default().push((cgu.name(), linkage)); - } - } - - let mut item_keys: Vec<_> = items - .iter() - .map(|i| { - let mut output = with_no_trimmed_paths!(i.to_string()); - output.push_str(" @@"); - let mut empty = Vec::new(); - let cgus = item_to_cgus.get_mut(i).unwrap_or(&mut empty); - cgus.sort_by_key(|(name, _)| *name); - cgus.dedup(); - for &(ref cgu_name, (linkage, _)) in cgus.iter() { - output.push(' '); - output.push_str(cgu_name.as_str()); - - let linkage_abbrev = match linkage { - Linkage::External => "External", - Linkage::AvailableExternally => "Available", - Linkage::LinkOnceAny => "OnceAny", - Linkage::LinkOnceODR => "OnceODR", - Linkage::WeakAny => "WeakAny", - Linkage::WeakODR => "WeakODR", - Linkage::Appending => "Appending", - Linkage::Internal => "Internal", - Linkage::Private => "Private", - Linkage::ExternalWeak => "ExternalWeak", - Linkage::Common => "Common", - }; - - output.push('['); - output.push_str(linkage_abbrev); - output.push(']'); - } - output - }) - .collect(); - - item_keys.sort(); - - for item in item_keys { - println!("MONO_ITEM {item}"); - } - } - - (tcx.arena.alloc(mono_items), codegen_units) -} - -/// Outputs stats about instantiation counts and estimated size, per `MonoItem`'s -/// def, to a file in the given output directory. -fn dump_mono_items_stats<'tcx>( - tcx: TyCtxt<'tcx>, - codegen_units: &[CodegenUnit<'tcx>], - output_directory: &Option<PathBuf>, - crate_name: Symbol, -) -> Result<(), Box<dyn std::error::Error>> { - let output_directory = if let Some(ref directory) = output_directory { - fs::create_dir_all(directory)?; - directory - } else { - Path::new(".") - }; - - let format = tcx.sess.opts.unstable_opts.dump_mono_stats_format; - let ext = format.extension(); - let filename = format!("{crate_name}.mono_items.{ext}"); - let output_path = output_directory.join(&filename); - let file = File::create(&output_path)?; - let mut file = BufWriter::new(file); - - // Gather instantiated mono items grouped by def_id - let mut items_per_def_id: FxHashMap<_, Vec<_>> = Default::default(); - for cgu in codegen_units { - for (&mono_item, _) in cgu.items() { - // Avoid variable-sized compiler-generated shims - if mono_item.is_user_defined() { - items_per_def_id.entry(mono_item.def_id()).or_default().push(mono_item); - } - } - } - - #[derive(serde::Serialize)] - struct MonoItem { - name: String, - instantiation_count: usize, - size_estimate: usize, - total_estimate: usize, - } - - // Output stats sorted by total instantiated size, from heaviest to lightest - let mut stats: Vec<_> = items_per_def_id - .into_iter() - .map(|(def_id, items)| { - let name = with_no_trimmed_paths!(tcx.def_path_str(def_id)); - let instantiation_count = items.len(); - let size_estimate = items[0].size_estimate(tcx); - let total_estimate = instantiation_count * size_estimate; - MonoItem { name, instantiation_count, size_estimate, total_estimate } - }) - .collect(); - stats.sort_unstable_by_key(|item| cmp::Reverse(item.total_estimate)); - - if !stats.is_empty() { - match format { - DumpMonoStatsFormat::Json => serde_json::to_writer(file, &stats)?, - DumpMonoStatsFormat::Markdown => { - writeln!( - file, - "| Item | Instantiation count | Estimated Cost Per Instantiation | Total Estimated Cost |" - )?; - writeln!(file, "| --- | ---: | ---: | ---: |")?; - - for MonoItem { name, instantiation_count, size_estimate, total_estimate } in stats { - writeln!( - file, - "| `{name}` | {instantiation_count} | {size_estimate} | {total_estimate} |" - )?; - } - } - } - } - - Ok(()) -} - -fn codegened_and_inlined_items(tcx: TyCtxt<'_>, (): ()) -> &DefIdSet { - let (items, cgus) = tcx.collect_and_partition_mono_items(()); - let mut visited = DefIdSet::default(); - let mut result = items.clone(); - - for cgu in cgus { - for (item, _) in cgu.items() { - if let MonoItem::Fn(ref instance) = item { - let did = instance.def_id(); - if !visited.insert(did) { - continue; - } - let body = tcx.instance_mir(instance.def); - for block in body.basic_blocks.iter() { - for statement in &block.statements { - let mir::StatementKind::Coverage(_) = statement.kind else { continue }; - let scope = statement.source_info.scope; - if let Some(inlined) = scope.inlined_instance(&body.source_scopes) { - result.insert(inlined.def_id()); - } - } - } - } - } - } - - tcx.arena.alloc(result) -} - -pub fn provide(providers: &mut Providers) { - providers.collect_and_partition_mono_items = collect_and_partition_mono_items; - providers.codegened_and_inlined_items = codegened_and_inlined_items; - - providers.is_codegened_item = |tcx, def_id| { - let (all_mono_items, _) = tcx.collect_and_partition_mono_items(()); - all_mono_items.contains(&def_id) - }; - - providers.codegen_unit = |tcx, name| { - let (_, all) = tcx.collect_and_partition_mono_items(()); - all.iter() - .find(|cgu| cgu.name() == name) - .unwrap_or_else(|| panic!("failed to find cgu with name {name:?}")) - }; -} diff --git a/compiler/rustc_monomorphize/src/util.rs b/compiler/rustc_monomorphize/src/util.rs index d12bfc6f6..f6a80b043 100644 --- a/compiler/rustc_monomorphize/src/util.rs +++ b/compiler/rustc_monomorphize/src/util.rs @@ -29,12 +29,12 @@ pub(crate) fn dump_closure_profile<'tcx>(tcx: TyCtxt<'tcx>, closure_instance: In let before_feature_tys = tcx.subst_and_normalize_erasing_regions( closure_instance.substs, param_env, - ty::EarlyBinder(before_feature_tys), + ty::EarlyBinder::bind(before_feature_tys), ); let after_feature_tys = tcx.subst_and_normalize_erasing_regions( closure_instance.substs, param_env, - ty::EarlyBinder(after_feature_tys), + ty::EarlyBinder::bind(after_feature_tys), ); let new_size = tcx |