Auto merge of #84719 - Mark-Simulacrum:reduce-query-impl, r=davidtwco

Move iter_results to dyn FnMut rather than a generic

This means that we're no longer generating the iteration/locking code for each invocation site of iter_results, rather just once per query (roughly), which seems much better: this is a 15% win in instruction counts when compiling the rustc_query_impl crate. The code where this is used also is pretty cold, I suspect; the old solution didn't fully monomorphize either.
This commit is contained in:
bors 2021-04-30 22:21:07 +00:00
commit 8a9fa3682d
5 changed files with 53 additions and 47 deletions

View file

@ -1185,20 +1185,27 @@ where
assert!(Q::query_state(tcx).all_inactive());
let cache = Q::query_cache(tcx);
cache.iter_results(|results| {
for (key, value, dep_node) in results {
if Q::cache_on_disk(tcx, &key, Some(value)) {
let dep_node = SerializedDepNodeIndex::new(dep_node.index());
let mut res = Ok(());
cache.iter_results(&mut |key, value, dep_node| {
if res.is_err() {
return;
}
if Q::cache_on_disk(tcx, &key, Some(value)) {
let dep_node = SerializedDepNodeIndex::new(dep_node.index());
// Record position of the cache entry.
query_result_index
.push((dep_node, AbsoluteBytePos::new(encoder.encoder.position())));
// Record position of the cache entry.
query_result_index.push((dep_node, AbsoluteBytePos::new(encoder.encoder.position())));
// Encode the type check tables with the `SerializedDepNodeIndex`
// as tag.
encoder.encode_tagged(dep_node, value)?;
// Encode the type check tables with the `SerializedDepNodeIndex`
// as tag.
match encoder.encode_tagged(dep_node, value) {
Ok(()) => {}
Err(e) => {
res = Err(e);
}
}
}
Ok(())
})
});
res
}

View file

@ -250,8 +250,8 @@ fn alloc_self_profile_query_strings_for_query_cache<'tcx, C>(
// need to invoke queries itself, we cannot keep the query caches
// locked while doing so. Instead we copy out the
// `(query_key, dep_node_index)` pairs and release the lock again.
let query_keys_and_indices: Vec<_> = query_cache
.iter_results(|results| results.map(|(k, _, i)| (k.clone(), i)).collect());
let mut query_keys_and_indices = Vec::new();
query_cache.iter_results(&mut |k, _, i| query_keys_and_indices.push((k.clone(), i)));
// Now actually allocate the strings. If allocating the strings
// generates new entries in the query cache, we'll miss them but
@ -275,14 +275,15 @@ fn alloc_self_profile_query_strings_for_query_cache<'tcx, C>(
let query_name = profiler.get_or_alloc_cached_string(query_name);
let event_id = event_id_builder.from_label(query_name).to_string_id();
query_cache.iter_results(|results| {
let query_invocation_ids: Vec<_> = results.map(|v| v.2.into()).collect();
profiler.bulk_map_query_invocation_id_to_single_string(
query_invocation_ids.into_iter(),
event_id,
);
let mut query_invocation_ids = Vec::new();
query_cache.iter_results(&mut |_, _, i| {
query_invocation_ids.push(i.into());
});
profiler.bulk_map_query_invocation_id_to_single_string(
query_invocation_ids.into_iter(),
event_id,
);
}
});
}

View file

@ -50,13 +50,12 @@ where
key_type: type_name::<C::Key>(),
value_size: mem::size_of::<C::Value>(),
value_type: type_name::<C::Value>(),
entry_count: map.iter_results(|results| results.count()),
entry_count: 0,
local_def_id_keys: None,
};
map.iter_results(|results| {
for (key, _, _) in results {
key.key_stats(&mut stats)
}
map.iter_results(&mut |key, _, _| {
stats.entry_count += 1;
key.key_stats(&mut stats)
});
stats
}

View file

@ -49,13 +49,11 @@ pub trait QueryCache: QueryStorage {
index: DepNodeIndex,
) -> Self::Stored;
fn iter<R>(
fn iter(
&self,
shards: &Sharded<Self::Sharded>,
f: impl for<'a> FnOnce(
&'a mut dyn Iterator<Item = (&'a Self::Key, &'a Self::Value, DepNodeIndex)>,
) -> R,
) -> R;
f: &mut dyn FnMut(&Self::Key, &Self::Value, DepNodeIndex),
);
}
pub struct DefaultCacheSelector;
@ -124,14 +122,17 @@ where
value
}
fn iter<R>(
fn iter(
&self,
shards: &Sharded<Self::Sharded>,
f: impl for<'a> FnOnce(&'a mut dyn Iterator<Item = (&'a K, &'a V, DepNodeIndex)>) -> R,
) -> R {
f: &mut dyn FnMut(&Self::Key, &Self::Value, DepNodeIndex),
) {
let shards = shards.lock_shards();
let mut results = shards.iter().flat_map(|shard| shard.iter()).map(|(k, v)| (k, &v.0, v.1));
f(&mut results)
for shard in shards.iter() {
for (k, v) in shard.iter() {
f(k, &v.0, v.1);
}
}
}
}
@ -207,13 +208,16 @@ where
&value.0
}
fn iter<R>(
fn iter(
&self,
shards: &Sharded<Self::Sharded>,
f: impl for<'a> FnOnce(&'a mut dyn Iterator<Item = (&'a K, &'a V, DepNodeIndex)>) -> R,
) -> R {
f: &mut dyn FnMut(&Self::Key, &Self::Value, DepNodeIndex),
) {
let shards = shards.lock_shards();
let mut results = shards.iter().flat_map(|shard| shard.iter()).map(|(k, v)| (k, &v.0, v.1));
f(&mut results)
for shard in shards.iter() {
for (k, v) in shard.iter() {
f(k, &v.0, v.1);
}
}
}
}

View file

@ -73,12 +73,7 @@ impl<C: QueryCache> QueryCacheStore<C> {
(QueryLookup { key_hash, shard }, lock)
}
pub fn iter_results<R>(
&self,
f: impl for<'a> FnOnce(
&'a mut dyn Iterator<Item = (&'a C::Key, &'a C::Value, DepNodeIndex)>,
) -> R,
) -> R {
pub fn iter_results(&self, f: &mut dyn FnMut(&C::Key, &C::Value, DepNodeIndex)) {
self.cache.iter(&self.shards, f)
}
}