dnrops
/
rust
mirror of https://github.com/rust-lang/rust.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Rollup merge of #111899 - nnethercote:cgu-cleanups, r=wesleywiser 

CGU cleanups

Some code clarity improvements I found when reading this code closely.

r? ``@wesleywiser``
This commit is contained in:
Matthias Krüger 2023-05-26 08:24:07 +02:00 committed by GitHub
parent 42c7b8a7de e6b99a6521
commit 78cc117f7b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 169 additions and 206 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
compiler/rustc_middle/src/mir/mono.rs
View File

@ -334,10 +334,7 @@ impl<'tcx> CodegenUnit<'tcx> {
}
pub fn modify_size_estimate(&mut self, delta: usize) {
assert!(self.size_estimate.is_some());
if let Some(size_estimate) = self.size_estimate {
self.size_estimate = Some(size_estimate + delta);
}
*self.size_estimate.as_mut().unwrap() += delta;
}
pub fn contains_item(&self, item: &MonoItem<'tcx>) -> bool {

150
compiler/rustc_monomorphize/src/partitioning/default.rs
View File

@ -1,3 +1,4 @@
use std::cmp;
use std::collections::hash_map::Entry;
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
@ -14,10 +15,7 @@ use rustc_span::symbol::Symbol;
use super::PartitioningCx;
use crate::collector::InliningMap;
use crate::partitioning::merging;
use crate::partitioning::{
MonoItemPlacement, Partition, PostInliningPartitioning, PreInliningPartitioning,
};
use crate::partitioning::{MonoItemPlacement, Partition, PlacedRootMonoItems};
pub struct DefaultPartitioning;
@ -26,7 +24,7 @@ impl<'tcx> Partition<'tcx> for DefaultPartitioning {
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
mono_items: &mut I,
) -> PreInliningPartitioning<'tcx>
) -> PlacedRootMonoItems<'tcx>
where
I: Iterator<Item = MonoItem<'tcx>>,
{
@ -91,38 +89,120 @@ impl<'tcx> Partition<'tcx> for DefaultPartitioning {
codegen_units.insert(codegen_unit_name, CodegenUnit::new(codegen_unit_name));
}
PreInliningPartitioning {
codegen_units: codegen_units.into_values().collect(),
roots,
internalization_candidates,
}
let codegen_units = codegen_units.into_values().collect();
PlacedRootMonoItems { codegen_units, roots, internalization_candidates }
}
fn merge_codegen_units(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
initial_partitioning: &mut PreInliningPartitioning<'tcx>,
codegen_units: &mut Vec<CodegenUnit<'tcx>>,
) {
merging::merge_codegen_units(cx, initial_partitioning);
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>,
initial_partitioning: PreInliningPartitioning<'tcx>,
) -> PostInliningPartitioning<'tcx> {
let mut new_partitioning = Vec::new();
codegen_units: &mut [CodegenUnit<'tcx>],
roots: FxHashSet<MonoItem<'tcx>>,
) -> FxHashMap<MonoItem<'tcx>, MonoItemPlacement> {
let mut mono_item_placements = FxHashMap::default();
let PreInliningPartitioning {
codegen_units: initial_cgus,
roots,
internalization_candidates,
} = initial_partitioning;
let single_codegen_unit = codegen_units.len() == 1;
let single_codegen_unit = initial_cgus.len() == 1;
for old_codegen_unit in initial_cgus {
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() {
@ -174,14 +254,10 @@ impl<'tcx> Partition<'tcx> for DefaultPartitioning {
}
}
new_partitioning.push(new_codegen_unit);
*old_codegen_unit = new_codegen_unit;
}
return PostInliningPartitioning {
codegen_units: new_partitioning,
mono_item_placements,
internalization_candidates,
};
return mono_item_placements;
fn follow_inlining<'tcx>(
mono_item: MonoItem<'tcx>,
@ -201,14 +277,16 @@ impl<'tcx> Partition<'tcx> for DefaultPartitioning {
fn internalize_symbols(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
partitioning: &mut PostInliningPartitioning<'tcx>,
codegen_units: &mut [CodegenUnit<'tcx>],
mono_item_placements: FxHashMap<MonoItem<'tcx>, MonoItemPlacement>,
internalization_candidates: FxHashSet<MonoItem<'tcx>>,
) {
if partitioning.codegen_units.len() == 1 {
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 &mut partitioning.codegen_units {
for candidate in &partitioning.internalization_candidates {
for cgu in codegen_units {
for candidate in &internalization_candidates {
cgu.items_mut().insert(*candidate, (Linkage::Internal, Visibility::Default));
}
}
@ -225,15 +303,13 @@ impl<'tcx> Partition<'tcx> for DefaultPartitioning {
}
});
let mono_item_placements = &partitioning.mono_item_placements;
// For each internalization candidates in each codegen unit, check if it is
// accessed from outside its defining codegen unit.
for cgu in &mut partitioning.codegen_units {
for cgu in codegen_units {
let home_cgu = MonoItemPlacement::SingleCgu { cgu_name: cgu.name() };
for (accessee, linkage_and_visibility) in cgu.items_mut() {
if !partitioning.internalization_candidates.contains(accessee) {
if !internalization_candidates.contains(accessee) {
// This item is no candidate for internalizing, so skip it.
continue;
}

111
compiler/rustc_monomorphize/src/partitioning/merging.rs
View File

@ -1,111 +0,0 @@
use std::cmp;
use rustc_data_structures::fx::FxHashMap;
use rustc_hir::def_id::LOCAL_CRATE;
use rustc_middle::mir::mono::{CodegenUnit, CodegenUnitNameBuilder};
use rustc_span::symbol::Symbol;
use super::PartitioningCx;
use crate::partitioning::PreInliningPartitioning;
pub fn merge_codegen_units<'tcx>(
cx: &PartitioningCx<'_, 'tcx>,
initial_partitioning: &mut PreInliningPartitioning<'tcx>,
) {
assert!(cx.target_cgu_count >= 1);
let codegen_units = &mut initial_partitioning.codegen_units;
// 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 &strs 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() {
cgu.set_name(numbered_codegen_unit_name(cgu_name_builder, index));
}
}
}
fn numbered_codegen_unit_name(
name_builder: &mut CodegenUnitNameBuilder<'_>,
index: usize,
) -> Symbol {
name_builder.build_cgu_name_no_mangle(LOCAL_CRATE, &["cgu"], Some(index))
}

109
compiler/rustc_monomorphize/src/partitioning/mod.rs
View File

@ -93,7 +93,6 @@
//! inlining, even when they are not marked `#[inline]`.
mod default;
mod merging;
use std::cmp;
use std::fs::{self, File};
@ -129,7 +128,7 @@ impl<'tcx> Partition<'tcx> for Partitioner {
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
mono_items: &mut I,
) -> PreInliningPartitioning<'tcx>
) -> PlacedRootMonoItems<'tcx>
where
I: Iterator<Item = MonoItem<'tcx>>,
{
@ -142,12 +141,10 @@ impl<'tcx> Partition<'tcx> for Partitioner {
fn merge_codegen_units(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
initial_partitioning: &mut PreInliningPartitioning<'tcx>,
codegen_units: &mut Vec<CodegenUnit<'tcx>>,
) {
match self {
Partitioner::Default(partitioner) => {
partitioner.merge_codegen_units(cx, initial_partitioning)
}
Partitioner::Default(partitioner) => partitioner.merge_codegen_units(cx, codegen_units),
Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy),
}
}
@ -155,11 +152,12 @@ impl<'tcx> Partition<'tcx> for Partitioner {
fn place_inlined_mono_items(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
initial_partitioning: PreInliningPartitioning<'tcx>,
) -> PostInliningPartitioning<'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, initial_partitioning)
partitioner.place_inlined_mono_items(cx, codegen_units, roots)
}
Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy),
}
@ -168,48 +166,62 @@ impl<'tcx> Partition<'tcx> for Partitioner {
fn internalize_symbols(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
post_inlining_partitioning: &mut PostInliningPartitioning<'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, post_inlining_partitioning)
}
Partitioner::Default(partitioner) => partitioner.internalize_symbols(
cx,
codegen_units,
mono_item_placements,
internalization_candidates,
),
Partitioner::Unknown => cx.tcx.sess.emit_fatal(UnknownPartitionStrategy),
}
}
}
pub struct PartitioningCx<'a, 'tcx> {
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,
) -> PreInliningPartitioning<'tcx>
) -> PlacedRootMonoItems<'tcx>
where
I: Iterator<Item = MonoItem<'tcx>>;
fn merge_codegen_units(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
initial_partitioning: &mut PreInliningPartitioning<'tcx>,
codegen_units: &mut Vec<CodegenUnit<'tcx>>,
);
fn place_inlined_mono_items(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
initial_partitioning: PreInliningPartitioning<'tcx>,
) -> PostInliningPartitioning<'tcx>;
codegen_units: &mut [CodegenUnit<'tcx>],
roots: FxHashSet<MonoItem<'tcx>>,
) -> FxHashMap<MonoItem<'tcx>, MonoItemPlacement>;
fn internalize_symbols(
&mut self,
cx: &PartitioningCx<'_, 'tcx>,
partitioning: &mut PostInliningPartitioning<'tcx>,
codegen_units: &mut [CodegenUnit<'tcx>],
mono_item_placements: FxHashMap<MonoItem<'tcx>, MonoItemPlacement>,
internalization_candidates: FxHashSet<MonoItem<'tcx>>,
);
}
@ -225,7 +237,7 @@ fn get_partitioner(tcx: TyCtxt<'_>) -> Partitioner {
}
}
pub fn partition<'tcx, I>(
fn partition<'tcx, I>(
tcx: TyCtxt<'tcx>,
mono_items: &mut I,
max_cgu_count: usize,
@ -241,44 +253,51 @@ where
// 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 mut initial_partitioning = {
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 initial_partitioning.codegen_units {
for cgu in &mut codegen_units {
cgu.create_size_estimate(tcx);
}
debug_dump(tcx, "INITIAL PARTITIONING", &initial_partitioning.codegen_units);
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 initial_partitioning);
debug_dump(tcx, "POST MERGING", &initial_partitioning.codegen_units);
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 mut post_inlining = {
let mono_item_placements = {
let _prof_timer = tcx.prof.generic_activity("cgu_partitioning_place_inline_items");
partitioner.place_inlined_mono_items(cx, initial_partitioning)
partitioner.place_inlined_mono_items(cx, &mut codegen_units, roots)
};
for cgu in &mut post_inlining.codegen_units {
for cgu in &mut codegen_units {
cgu.create_size_estimate(tcx);
}
debug_dump(tcx, "POST INLINING", &post_inlining.codegen_units);
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 post_inlining);
partitioner.internalize_symbols(
cx,
&mut codegen_units,
mono_item_placements,
internalization_candidates,
);
}
let instrument_dead_code =
@ -286,7 +305,7 @@ where
if instrument_dead_code {
assert!(
post_inlining.codegen_units.len() > 0,
codegen_units.len() > 0,
"There must be at least one CGU that code coverage data can be generated in."
);
@ -297,7 +316,7 @@ where
// 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<_> = post_inlining.codegen_units.iter_mut().collect();
let mut cgus: Vec<_> = codegen_units.iter_mut().collect();
cgus.sort_by_key(|cgu| cgu.size_estimate());
let dead_code_cgu =
@ -308,29 +327,17 @@ where
} else {
// If there are no CGUs that have externally linked items,
// then we just pick the first CGU as a fallback.
&mut post_inlining.codegen_units[0]
&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.
let PostInliningPartitioning {
codegen_units: mut result,
mono_item_placements: _,
internalization_candidates: _,
} = post_inlining;
codegen_units.sort_by(|a, b| a.name().as_str().cmp(b.name().as_str()));
result.sort_by(|a, b| a.name().as_str().cmp(b.name().as_str()));
debug_dump(tcx, "FINAL", &codegen_units);
debug_dump(tcx, "FINAL", &result);
result
}
pub struct PreInliningPartitioning<'tcx> {
codegen_units: Vec<CodegenUnit<'tcx>>,
roots: FxHashSet<MonoItem<'tcx>>,
internalization_candidates: FxHashSet<MonoItem<'tcx>>,
codegen_units
}
/// For symbol internalization, we need to know whether a symbol/mono-item is
@ -342,12 +349,6 @@ enum MonoItemPlacement {
MultipleCgus,
}
struct PostInliningPartitioning<'tcx> {
codegen_units: Vec<CodegenUnit<'tcx>>,
mono_item_placements: FxHashMap<MonoItem<'tcx>, MonoItemPlacement>,
internalization_candidates: FxHashSet<MonoItem<'tcx>>,
}
fn debug_dump<'a, 'tcx: 'a>(tcx: TyCtxt<'tcx>, label: &str, cgus: &[CodegenUnit<'tcx>]) {
let dump = move || {
use std::fmt::Write;