1 // Copyright 2016 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
11 //! Partitioning Codegen Units for Incremental Compilation
12 //! ======================================================
14 //! The task of this module is to take the complete set of translation items of
15 //! a crate and produce a set of codegen units from it, where a codegen unit
16 //! is a named set of (translation-item, linkage) pairs. That is, this module
17 //! decides which translation item appears in which codegen units with which
18 //! linkage. The following paragraphs describe some of the background on the
19 //! partitioning scheme.
21 //! The most important opportunity for saving on compilation time with
22 //! incremental compilation is to avoid re-translating and re-optimizing code.
23 //! Since the unit of translation and optimization for LLVM is "modules" or, how
24 //! we call them "codegen units", the particulars of how much time can be saved
25 //! by incremental compilation are tightly linked to how the output program is
26 //! partitioned into these codegen units prior to passing it to LLVM --
27 //! especially because we have to treat codegen units as opaque entities once
28 //! they are created: There is no way for us to incrementally update an existing
29 //! LLVM module and so we have to build any such module from scratch if it was
30 //! affected by some change in the source code.
32 //! From that point of view it would make sense to maximize the number of
33 //! codegen units by, for example, putting each function into its own module.
34 //! That way only those modules would have to be re-compiled that were actually
35 //! affected by some change, minimizing the number of functions that could have
36 //! been re-used but just happened to be located in a module that is
39 //! However, since LLVM optimization does not work across module boundaries,
40 //! using such a highly granular partitioning would lead to very slow runtime
41 //! code since it would effectively prohibit inlining and other inter-procedure
42 //! optimizations. We want to avoid that as much as possible.
44 //! Thus we end up with a trade-off: The bigger the codegen units, the better
45 //! LLVM's optimizer can do its work, but also the smaller the compilation time
46 //! reduction we get from incremental compilation.
48 //! Ideally, we would create a partitioning such that there are few big codegen
49 //! units with few interdependencies between them. For now though, we use the
50 //! following heuristic to determine the partitioning:
52 //! - There are two codegen units for every source-level module:
53 //! - One for "stable", that is non-generic, code
54 //! - One for more "volatile" code, i.e. monomorphized instances of functions
55 //! defined in that module
57 //! In order to see why this heuristic makes sense, let's take a look at when a
58 //! codegen unit can get invalidated:
60 //! 1. The most straightforward case is when the BODY of a function or global
61 //! changes. Then any codegen unit containing the code for that item has to be
62 //! re-compiled. Note that this includes all codegen units where the function
65 //! 2. The next case is when the SIGNATURE of a function or global changes. In
66 //! this case, all codegen units containing a REFERENCE to that item have to be
67 //! re-compiled. This is a superset of case 1.
69 //! 3. The final and most subtle case is when a REFERENCE to a generic function
70 //! is added or removed somewhere. Even though the definition of the function
71 //! might be unchanged, a new REFERENCE might introduce a new monomorphized
72 //! instance of this function which has to be placed and compiled somewhere.
73 //! Conversely, when removing a REFERENCE, it might have been the last one with
74 //! that particular set of generic arguments and thus we have to remove it.
76 //! From the above we see that just using one codegen unit per source-level
77 //! module is not such a good idea, since just adding a REFERENCE to some
78 //! generic item somewhere else would invalidate everything within the module
79 //! containing the generic item. The heuristic above reduces this detrimental
80 //! side-effect of references a little by at least not touching the non-generic
81 //! code of the module.
83 //! A Note on Inlining
84 //! ------------------
85 //! As briefly mentioned above, in order for LLVM to be able to inline a
86 //! function call, the body of the function has to be available in the LLVM
87 //! module where the call is made. This has a few consequences for partitioning:
89 //! - The partitioning algorithm has to take care of placing functions into all
90 //! codegen units where they should be available for inlining. It also has to
91 //! decide on the correct linkage for these functions.
93 //! - The partitioning algorithm has to know which functions are likely to get
94 //! inlined, so it can distribute function instantiations accordingly. Since
95 //! there is no way of knowing for sure which functions LLVM will decide to
96 //! inline in the end, we apply a heuristic here: Only functions marked with
97 //! #[inline] are considered for inlining by the partitioner. The current
98 //! implementation will not try to determine if a function is likely to be
99 //! inlined by looking at the functions definition.
101 //! Note though that as a side-effect of creating a codegen units per
102 //! source-level module, functions from the same module will be available for
103 //! inlining, even when they are not marked #[inline].
105 use collector::InliningMap;
107 use context::SharedCrateContext;
109 use rustc::dep_graph::{DepNode, WorkProductId};
110 use rustc::hir::def_id::DefId;
111 use rustc::hir::map::DefPathData;
112 use rustc::session::config::NUMBERED_CODEGEN_UNIT_MARKER;
113 use rustc::ty::{self, TyCtxt};
114 use rustc::ty::item_path::characteristic_def_id_of_type;
115 use rustc_incremental::IchHasher;
116 use std::cmp::Ordering;
119 use symbol_cache::SymbolCache;
120 use syntax::ast::NodeId;
121 use syntax::symbol::{Symbol, InternedString};
122 use trans_item::{TransItem, InstantiationMode};
123 use util::nodemap::{FxHashMap, FxHashSet};
125 pub enum PartitioningStrategy {
126 /// Generate one codegen unit per source-level module.
129 /// Partition the whole crate into a fixed number of codegen units.
130 FixedUnitCount(usize)
133 pub struct CodegenUnit<'tcx> {
134 /// A name for this CGU. Incremental compilation requires that
135 /// name be unique amongst **all** crates. Therefore, it should
136 /// contain something unique to this crate (e.g., a module path)
137 /// as well as the crate name and disambiguator.
138 name: InternedString,
140 items: FxHashMap<TransItem<'tcx>, llvm::Linkage>,
143 impl<'tcx> CodegenUnit<'tcx> {
144 pub fn new(name: InternedString,
145 items: FxHashMap<TransItem<'tcx>, llvm::Linkage>)
153 pub fn empty(name: InternedString) -> Self {
154 Self::new(name, FxHashMap())
157 pub fn contains_item(&self, item: &TransItem<'tcx>) -> bool {
158 self.items.contains_key(item)
161 pub fn name(&self) -> &str {
165 pub fn items(&self) -> &FxHashMap<TransItem<'tcx>, llvm::Linkage> {
169 pub fn work_product_id(&self) -> Arc<WorkProductId> {
170 Arc::new(WorkProductId(self.name().to_string()))
173 pub fn work_product_dep_node(&self) -> DepNode<DefId> {
174 DepNode::WorkProduct(self.work_product_id())
177 pub fn compute_symbol_name_hash<'a>(&self,
178 scx: &SharedCrateContext<'a, 'tcx>,
179 symbol_cache: &SymbolCache<'a, 'tcx>)
181 let mut state = IchHasher::new();
182 let exported_symbols = scx.exported_symbols();
183 let all_items = self.items_in_deterministic_order(scx.tcx(), symbol_cache);
184 for (item, _) in all_items {
185 let symbol_name = symbol_cache.get(item);
186 symbol_name.len().hash(&mut state);
187 symbol_name.hash(&mut state);
188 let exported = match item {
189 TransItem::Fn(ref instance) => {
191 scx.tcx().hir.as_local_node_id(instance.def_id());
192 node_id.map(|node_id| exported_symbols.contains(&node_id))
195 TransItem::Static(node_id) => {
196 exported_symbols.contains(&node_id)
198 TransItem::GlobalAsm(..) => true,
200 exported.hash(&mut state);
202 state.finish().to_smaller_hash()
205 pub fn items_in_deterministic_order<'a>(&self,
207 symbol_cache: &SymbolCache<'a, 'tcx>)
208 -> Vec<(TransItem<'tcx>, llvm::Linkage)> {
209 let mut items: Vec<(TransItem<'tcx>, llvm::Linkage)> =
210 self.items.iter().map(|(item, linkage)| (*item, *linkage)).collect();
212 // The codegen tests rely on items being process in the same order as
213 // they appear in the file, so for local items, we sort by node_id first
214 items.sort_by(|&(trans_item1, _), &(trans_item2, _)| {
215 let node_id1 = local_node_id(tcx, trans_item1);
216 let node_id2 = local_node_id(tcx, trans_item2);
218 match (node_id1, node_id2) {
220 let symbol_name1 = symbol_cache.get(trans_item1);
221 let symbol_name2 = symbol_cache.get(trans_item2);
222 symbol_name1.cmp(&symbol_name2)
224 // In the following two cases we can avoid looking up the symbol
225 (None, Some(_)) => Ordering::Less,
226 (Some(_), None) => Ordering::Greater,
227 (Some(node_id1), Some(node_id2)) => {
228 let ordering = node_id1.cmp(&node_id2);
230 if ordering != Ordering::Equal {
234 let symbol_name1 = symbol_cache.get(trans_item1);
235 let symbol_name2 = symbol_cache.get(trans_item2);
236 symbol_name1.cmp(&symbol_name2)
243 fn local_node_id(tcx: TyCtxt, trans_item: TransItem) -> Option<NodeId> {
245 TransItem::Fn(instance) => {
246 tcx.hir.as_local_node_id(instance.def_id())
248 TransItem::Static(node_id) | TransItem::GlobalAsm(node_id) => {
257 // Anything we can't find a proper codegen unit for goes into this.
258 const FALLBACK_CODEGEN_UNIT: &'static str = "__rustc_fallback_codegen_unit";
260 pub fn partition<'a, 'tcx, I>(scx: &SharedCrateContext<'a, 'tcx>,
262 strategy: PartitioningStrategy,
263 inlining_map: &InliningMap<'tcx>)
264 -> Vec<CodegenUnit<'tcx>>
265 where I: Iterator<Item = TransItem<'tcx>>
269 // In the first step, we place all regular translation items into their
270 // respective 'home' codegen unit. Regular translation items are all
271 // functions and statics defined in the local crate.
272 let mut initial_partitioning = place_root_translation_items(scx,
275 debug_dump(tcx, "INITIAL PARTITONING:", initial_partitioning.codegen_units.iter());
277 // If the partitioning should produce a fixed count of codegen units, merge
278 // until that count is reached.
279 if let PartitioningStrategy::FixedUnitCount(count) = strategy {
280 merge_codegen_units(&mut initial_partitioning, count, &tcx.crate_name.as_str());
282 debug_dump(tcx, "POST MERGING:", initial_partitioning.codegen_units.iter());
285 // In the next step, we use the inlining map to determine which addtional
286 // translation items have to go into each codegen unit. These additional
287 // translation items can be drop-glue, functions from external crates, and
288 // local functions the definition of which is marked with #[inline].
289 let post_inlining = place_inlined_translation_items(initial_partitioning,
292 debug_dump(tcx, "POST INLINING:", post_inlining.0.iter());
294 // Finally, sort by codegen unit name, so that we get deterministic results
295 let mut result = post_inlining.0;
296 result.sort_by(|cgu1, cgu2| {
297 (&cgu1.name[..]).cmp(&cgu2.name[..])
303 struct PreInliningPartitioning<'tcx> {
304 codegen_units: Vec<CodegenUnit<'tcx>>,
305 roots: FxHashSet<TransItem<'tcx>>,
308 struct PostInliningPartitioning<'tcx>(Vec<CodegenUnit<'tcx>>);
310 fn place_root_translation_items<'a, 'tcx, I>(scx: &SharedCrateContext<'a, 'tcx>,
312 -> PreInliningPartitioning<'tcx>
313 where I: Iterator<Item = TransItem<'tcx>>
316 let mut roots = FxHashSet();
317 let mut codegen_units = FxHashMap();
318 let is_incremental_build = tcx.sess.opts.incremental.is_some();
320 for trans_item in trans_items {
321 let is_root = trans_item.instantiation_mode(tcx) == InstantiationMode::GloballyShared;
324 let characteristic_def_id = characteristic_def_id_of_trans_item(scx, trans_item);
325 let is_volatile = is_incremental_build &&
326 trans_item.is_generic_fn();
328 let codegen_unit_name = match characteristic_def_id {
329 Some(def_id) => compute_codegen_unit_name(tcx, def_id, is_volatile),
330 None => Symbol::intern(FALLBACK_CODEGEN_UNIT).as_str(),
333 let make_codegen_unit = || {
334 CodegenUnit::empty(codegen_unit_name.clone())
337 let mut codegen_unit = codegen_units.entry(codegen_unit_name.clone())
338 .or_insert_with(make_codegen_unit);
340 let linkage = match trans_item.explicit_linkage(tcx) {
341 Some(explicit_linkage) => explicit_linkage,
345 TransItem::Static(..) |
346 TransItem::GlobalAsm(..) => llvm::ExternalLinkage,
351 codegen_unit.items.insert(trans_item, linkage);
352 roots.insert(trans_item);
356 // always ensure we have at least one CGU; otherwise, if we have a
357 // crate with just types (for example), we could wind up with no CGU
358 if codegen_units.is_empty() {
359 let codegen_unit_name = Symbol::intern(FALLBACK_CODEGEN_UNIT).as_str();
360 codegen_units.entry(codegen_unit_name.clone())
361 .or_insert_with(|| CodegenUnit::empty(codegen_unit_name.clone()));
364 PreInliningPartitioning {
365 codegen_units: codegen_units.into_iter()
366 .map(|(_, codegen_unit)| codegen_unit)
372 fn merge_codegen_units<'tcx>(initial_partitioning: &mut PreInliningPartitioning<'tcx>,
373 target_cgu_count: usize,
375 assert!(target_cgu_count >= 1);
376 let codegen_units = &mut initial_partitioning.codegen_units;
378 // Merge the two smallest codegen units until the target size is reached.
379 // Note that "size" is estimated here rather inaccurately as the number of
380 // translation items in a given unit. This could be improved on.
381 while codegen_units.len() > target_cgu_count {
382 // Sort small cgus to the back
383 codegen_units.sort_by_key(|cgu| -(cgu.items.len() as i64));
384 let smallest = codegen_units.pop().unwrap();
385 let second_smallest = codegen_units.last_mut().unwrap();
387 for (k, v) in smallest.items.into_iter() {
388 second_smallest.items.insert(k, v);
392 for (index, cgu) in codegen_units.iter_mut().enumerate() {
393 cgu.name = numbered_codegen_unit_name(crate_name, index);
396 // If the initial partitioning contained less than target_cgu_count to begin
397 // with, we won't have enough codegen units here, so add a empty units until
398 // we reach the target count
399 while codegen_units.len() < target_cgu_count {
400 let index = codegen_units.len();
402 CodegenUnit::empty(numbered_codegen_unit_name(crate_name, index)));
406 fn place_inlined_translation_items<'tcx>(initial_partitioning: PreInliningPartitioning<'tcx>,
407 inlining_map: &InliningMap<'tcx>)
408 -> PostInliningPartitioning<'tcx> {
409 let mut new_partitioning = Vec::new();
411 for codegen_unit in &initial_partitioning.codegen_units[..] {
412 // Collect all items that need to be available in this codegen unit
413 let mut reachable = FxHashSet();
414 for root in codegen_unit.items.keys() {
415 follow_inlining(*root, inlining_map, &mut reachable);
418 let mut new_codegen_unit =
419 CodegenUnit::empty(codegen_unit.name.clone());
421 // Add all translation items that are not already there
422 for trans_item in reachable {
423 if let Some(linkage) = codegen_unit.items.get(&trans_item) {
424 // This is a root, just copy it over
425 new_codegen_unit.items.insert(trans_item, *linkage);
427 if initial_partitioning.roots.contains(&trans_item) {
428 bug!("GloballyShared trans-item inlined into other CGU: \
432 // This is a cgu-private copy
433 new_codegen_unit.items.insert(trans_item, llvm::InternalLinkage);
437 new_partitioning.push(new_codegen_unit);
440 return PostInliningPartitioning(new_partitioning);
442 fn follow_inlining<'tcx>(trans_item: TransItem<'tcx>,
443 inlining_map: &InliningMap<'tcx>,
444 visited: &mut FxHashSet<TransItem<'tcx>>) {
445 if !visited.insert(trans_item) {
449 inlining_map.with_inlining_candidates(trans_item, |target| {
450 follow_inlining(target, inlining_map, visited);
455 fn characteristic_def_id_of_trans_item<'a, 'tcx>(scx: &SharedCrateContext<'a, 'tcx>,
456 trans_item: TransItem<'tcx>)
460 TransItem::Fn(instance) => {
461 let def_id = match instance.def {
462 ty::InstanceDef::Item(def_id) => def_id,
463 ty::InstanceDef::FnPtrShim(..) |
464 ty::InstanceDef::ClosureOnceShim { .. } |
465 ty::InstanceDef::Intrinsic(..) |
466 ty::InstanceDef::DropGlue(..) |
467 ty::InstanceDef::Virtual(..) => return None
470 // If this is a method, we want to put it into the same module as
471 // its self-type. If the self-type does not provide a characteristic
472 // DefId, we use the location of the impl after all.
474 if tcx.trait_of_item(def_id).is_some() {
475 let self_ty = instance.substs.type_at(0);
476 // This is an implementation of a trait method.
477 return characteristic_def_id_of_type(self_ty).or(Some(def_id));
480 if let Some(impl_def_id) = tcx.impl_of_method(def_id) {
481 // This is a method within an inherent impl, find out what the
483 let impl_self_ty = common::def_ty(scx, impl_def_id, instance.substs);
484 if let Some(def_id) = characteristic_def_id_of_type(impl_self_ty) {
491 TransItem::Static(node_id) |
492 TransItem::GlobalAsm(node_id) => Some(tcx.hir.local_def_id(node_id)),
496 fn compute_codegen_unit_name<'a, 'tcx>(tcx: TyCtxt<'a, 'tcx, 'tcx>,
500 // Unfortunately we cannot just use the `ty::item_path` infrastructure here
501 // because we need paths to modules and the DefIds of those are not
502 // available anymore for external items.
503 let mut mod_path = String::with_capacity(64);
505 let def_path = tcx.def_path(def_id);
506 mod_path.push_str(&tcx.crate_name(def_path.krate).as_str());
508 for part in tcx.def_path(def_id)
513 DefPathData::Module(..) => true,
517 mod_path.push_str("-");
518 mod_path.push_str(&part.data.as_interned_str());
522 mod_path.push_str(".volatile");
525 return Symbol::intern(&mod_path[..]).as_str();
528 fn numbered_codegen_unit_name(crate_name: &str, index: usize) -> InternedString {
529 Symbol::intern(&format!("{}{}{}", crate_name, NUMBERED_CODEGEN_UNIT_MARKER, index)).as_str()
532 fn debug_dump<'a, 'b, 'tcx, I>(tcx: TyCtxt<'a, 'tcx, 'tcx>,
535 where I: Iterator<Item=&'b CodegenUnit<'tcx>>,
538 if cfg!(debug_assertions) {
540 let symbol_cache = SymbolCache::new(tcx);
542 debug!("CodegenUnit {}:", cgu.name);
544 for (trans_item, linkage) in &cgu.items {
545 let symbol_name = symbol_cache.get(*trans_item);
546 let symbol_hash_start = symbol_name.rfind('h');
547 let symbol_hash = symbol_hash_start.map(|i| &symbol_name[i ..])
548 .unwrap_or("<no hash>");
550 debug!(" - {} [{:?}] [{}]",
551 trans_item.to_string(tcx),