Optimize lock_shards

Author: Zoxc

compiler/rustc_data_structures/src/sharded.rs

@@ -2,9 +2,12 @@ use crate::fx::{FxHashMap, FxHasher};
 #[cfg(parallel_compiler)]
 use crate::sync::{is_dyn_thread_safe, CacheAligned};
 use crate::sync::{Lock, LockGuard};
+#[cfg(parallel_compiler)]
+use itertools::Either;
 use std::borrow::Borrow;
 use std::collections::hash_map::RawEntryMut;
 use std::hash::{Hash, Hasher};
+use std::iter;
 use std::mem;
 
 // 32 shards is sufficient to reduce contention on an 8-core Ryzen 7 1700,
@@ -70,19 +73,27 @@ impl<T> Sharded<T> {
         }
     }
 
-    pub fn lock_shards(&self) -> Vec<LockGuard<'_, T>> {
+    #[inline]
+    pub fn lock_shards(&self) -> impl Iterator<Item = LockGuard<'_, T>> {
         match self {
-            Self::Single(single) => vec![single.lock()],
+            #[cfg(not(parallel_compiler))]
+            Self::Single(single) => iter::once(single.lock()),
+            #[cfg(parallel_compiler)]
+            Self::Single(single) => Either::Left(iter::once(single.lock())),
             #[cfg(parallel_compiler)]
-            Self::Shards(shards) => shards.iter().map(|shard| shard.0.lock()).collect(),
+            Self::Shards(shards) => Either::Right(shards.iter().map(|shard| shard.0.lock())),
         }
     }
 
-    pub fn try_lock_shards(&self) -> Option<Vec<LockGuard<'_, T>>> {
+    #[inline]
+    pub fn try_lock_shards(&self) -> impl Iterator<Item = Option<LockGuard<'_, T>>> {
         match self {
-            Self::Single(single) => Some(vec![single.try_lock()?]),
+            #[cfg(not(parallel_compiler))]
+            Self::Single(single) => iter::once(single.try_lock()),
+            #[cfg(parallel_compiler)]
+            Self::Single(single) => Either::Left(iter::once(single.try_lock())),
             #[cfg(parallel_compiler)]
-            Self::Shards(shards) => shards.iter().map(|shard| shard.0.try_lock()).collect(),
+            Self::Shards(shards) => Either::Right(shards.iter().map(|shard| shard.0.try_lock())),
         }
     }
 }
@@ -101,7 +112,7 @@ pub type ShardedHashMap<K, V> = Sharded<FxHashMap<K, V>>;
 
 impl<K: Eq, V> ShardedHashMap<K, V> {
     pub fn len(&self) -> usize {
-        self.lock_shards().iter().map(|shard| shard.len()).sum()
+        self.lock_shards().map(|shard| shard.len()).sum()
     }
 }
 

compiler/rustc_middle/src/ty/context.rs

@@ -1296,25 +1296,26 @@ macro_rules! sty_debug_print {
                 };
                 $(let mut $variant = total;)*
 
-                let shards = tcx.interners.type_.lock_shards();
-                let types = shards.iter().flat_map(|shard| shard.keys());
-                for &InternedInSet(t) in types {
-                    let variant = match t.internee {
-                        ty::Bool | ty::Char | ty::Int(..) | ty::Uint(..) |
-                            ty::Float(..) | ty::Str | ty::Never => continue,
-                        ty::Error(_) => /* unimportant */ continue,
-                        $(ty::$variant(..) => &mut $variant,)*
-                    };
-                    let lt = t.flags.intersects(ty::TypeFlags::HAS_RE_INFER);
-                    let ty = t.flags.intersects(ty::TypeFlags::HAS_TY_INFER);
-                    let ct = t.flags.intersects(ty::TypeFlags::HAS_CT_INFER);
-
-                    variant.total += 1;
-                    total.total += 1;
-                    if lt { total.lt_infer += 1; variant.lt_infer += 1 }
-                    if ty { total.ty_infer += 1; variant.ty_infer += 1 }
-                    if ct { total.ct_infer += 1; variant.ct_infer += 1 }
-                    if lt && ty && ct { total.all_infer += 1; variant.all_infer += 1 }
+                for shard in tcx.interners.type_.lock_shards() {
+                    let types = shard.keys();
+                    for &InternedInSet(t) in types {
+                        let variant = match t.internee {
+                            ty::Bool | ty::Char | ty::Int(..) | ty::Uint(..) |
+                                ty::Float(..) | ty::Str | ty::Never => continue,
+                            ty::Error(_) => /* unimportant */ continue,
+                            $(ty::$variant(..) => &mut $variant,)*
+                        };
+                        let lt = t.flags.intersects(ty::TypeFlags::HAS_RE_INFER);
+                        let ty = t.flags.intersects(ty::TypeFlags::HAS_TY_INFER);
+                        let ct = t.flags.intersects(ty::TypeFlags::HAS_CT_INFER);
+
+                        variant.total += 1;
+                        total.total += 1;
+                        if lt { total.lt_infer += 1; variant.lt_infer += 1 }
+                        if ty { total.ty_infer += 1; variant.ty_infer += 1 }
+                        if ct { total.ct_infer += 1; variant.ct_infer += 1 }
+                        if lt && ty && ct { total.all_infer += 1; variant.all_infer += 1 }
+                    }
                 }
                 writeln!(fmt, "Ty interner             total           ty lt ct all")?;
                 $(writeln!(fmt, "    {:18}: {uses:6} {usespc:4.1}%, \

compiler/rustc_query_system/src/query/caches.rs

@@ -70,8 +70,7 @@ where
     }
 
     fn iter(&self, f: &mut dyn FnMut(&Self::Key, &Self::Value, DepNodeIndex)) {
-        let shards = self.cache.lock_shards();
-        for shard in shards.iter() {
+        for shard in self.cache.lock_shards() {
             for (k, v) in shard.iter() {
                 f(k, &v.0, v.1);
             }
@@ -160,8 +159,7 @@ where
     }
 
     fn iter(&self, f: &mut dyn FnMut(&Self::Key, &Self::Value, DepNodeIndex)) {
-        let shards = self.cache.lock_shards();
-        for shard in shards.iter() {
+        for shard in self.cache.lock_shards() {
             for (k, v) in shard.iter_enumerated() {
                 if let Some(v) = v {
                     f(&k, &v.0, v.1);

compiler/rustc_query_system/src/query/plumbing.rs

@@ -50,8 +50,7 @@ where
     D: DepKind,
 {
     pub fn all_inactive(&self) -> bool {
-        let shards = self.active.lock_shards();
-        shards.iter().all(|shard| shard.is_empty())
+        self.active.lock_shards().all(|shard| shard.is_empty())
     }
 
     pub fn try_collect_active_jobs<Qcx: Copy>(
@@ -64,9 +63,8 @@ where
 
         // We use try_lock_shards here since we are called from the
         // deadlock handler, and this shouldn't be locked.
-        let shards = self.active.try_lock_shards()?;
-        for shard in shards.iter() {
-            for (k, v) in shard.iter() {
+        for shard in self.active.try_lock_shards() {
+            for (k, v) in shard?.iter() {
                 if let QueryResult::Started(ref job) = *v {
                     active.push((*k, job.clone()));
                 }