gecko-dev/js/public/GCHashTable.h

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

759 строки
24 KiB
C
Исходник Обычный вид История

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
* vim: set ts=8 sts=2 et sw=2 tw=80:
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef GCHashTable_h
#define GCHashTable_h
#include "mozilla/Maybe.h"
#include "js/GCPolicyAPI.h"
#include "js/HashTable.h"
#include "js/RootingAPI.h"
#include "js/SweepingAPI.h"
#include "js/TracingAPI.h"
namespace JS {
// Define a reasonable default GC policy for GC-aware Maps.
template <typename Key, typename Value>
struct DefaultMapSweepPolicy {
static bool needsSweep(Key* key, Value* value) {
return GCPolicy<Key>::needsSweep(key) || GCPolicy<Value>::needsSweep(value);
}
static bool traceWeak(JSTracer* trc, Key* key, Value* value) {
return GCPolicy<Key>::traceWeak(trc, key) &&
GCPolicy<Value>::traceWeak(trc, value);
}
};
// A GCHashMap is a GC-aware HashMap, meaning that it has additional trace and
// sweep methods that know how to visit all keys and values in the table.
// HashMaps that contain GC pointers will generally want to use this GCHashMap
// specialization instead of HashMap, because this conveniently supports tracing
// keys and values, and cleaning up weak entries.
//
// GCHashMap::trace applies GCPolicy<T>::trace to each entry's key and value.
// Most types of GC pointers already have appropriate specializations of
// GCPolicy, so they should just work as keys and values. Any struct type with a
// default constructor and trace and sweep functions should work as well. If you
// need to define your own GCPolicy specialization, generic helpers can be found
// in js/public/TracingAPI.h.
//
// The MapSweepPolicy template parameter controls how the table drops entries
// when swept. GCHashMap::sweep applies MapSweepPolicy::needsSweep to each table
// entry; if it returns true, the entry is dropped. The default MapSweepPolicy
// drops the entry if either the key or value is about to be finalized,
// according to its GCPolicy<T>::needsSweep method. (This default is almost
// always fine: it's hard to imagine keeping such an entry around anyway.)
//
// Note that this HashMap only knows *how* to trace and sweep, but it does not
// itself cause tracing or sweeping to be invoked. For tracing, it must be used
// with Rooted or PersistentRooted, or barriered and traced manually. For
// sweeping, currently it requires an explicit call to <map>.sweep().
template <typename Key, typename Value,
typename HashPolicy = js::DefaultHasher<Key>,
typename AllocPolicy = js::TempAllocPolicy,
typename MapSweepPolicy = DefaultMapSweepPolicy<Key, Value>>
class GCHashMap : public js::HashMap<Key, Value, HashPolicy, AllocPolicy> {
using Base = js::HashMap<Key, Value, HashPolicy, AllocPolicy>;
public:
explicit GCHashMap(AllocPolicy a = AllocPolicy()) : Base(std::move(a)) {}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
explicit GCHashMap(size_t length) : Base(length) {}
GCHashMap(AllocPolicy a, size_t length) : Base(std::move(a), length) {}
void trace(JSTracer* trc) {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
GCPolicy<Value>::trace(trc, &e.front().value(), "hashmap value");
GCPolicy<Key>::trace(trc, &e.front().mutableKey(), "hashmap key");
}
}
bool needsSweep() const { return !this->empty(); }
void sweep() {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
if (MapSweepPolicy::needsSweep(&e.front().mutableKey(),
&e.front().value())) {
e.removeFront();
}
}
}
void traceWeak(JSTracer* trc) {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
if (!MapSweepPolicy::traceWeak(trc, &e.front().mutableKey(),
&e.front().value())) {
e.removeFront();
}
}
}
// GCHashMap is movable
GCHashMap(GCHashMap&& rhs) : Base(std::move(rhs)) {}
void operator=(GCHashMap&& rhs) {
MOZ_ASSERT(this != &rhs, "self-move assignment is prohibited");
Base::operator=(std::move(rhs));
}
private:
// GCHashMap is not copyable or assignable
GCHashMap(const GCHashMap& hm) = delete;
GCHashMap& operator=(const GCHashMap& hm) = delete;
} MOZ_INHERIT_TYPE_ANNOTATIONS_FROM_TEMPLATE_ARGS;
} // namespace JS
namespace js {
// HashMap that supports rekeying.
//
// If your keys are pointers to something like JSObject that can be tenured or
// compacted, prefer to use GCHashMap with MovableCellHasher, which takes
// advantage of the Zone's stable id table to make rekeying unnecessary.
template <typename Key, typename Value,
typename HashPolicy = DefaultHasher<Key>,
typename AllocPolicy = TempAllocPolicy,
typename MapSweepPolicy = JS::DefaultMapSweepPolicy<Key, Value>>
class GCRekeyableHashMap : public JS::GCHashMap<Key, Value, HashPolicy,
AllocPolicy, MapSweepPolicy> {
using Base = JS::GCHashMap<Key, Value, HashPolicy, AllocPolicy>;
public:
explicit GCRekeyableHashMap(AllocPolicy a = AllocPolicy())
: Base(std::move(a)) {}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
explicit GCRekeyableHashMap(size_t length) : Base(length) {}
GCRekeyableHashMap(AllocPolicy a, size_t length)
: Base(std::move(a), length) {}
void sweep() {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
Key key(e.front().key());
if (MapSweepPolicy::needsSweep(&key, &e.front().value())) {
e.removeFront();
} else if (!HashPolicy::match(key, e.front().key())) {
e.rekeyFront(key);
}
}
}
void traceWeak(JSTracer* trc) {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
Key key(e.front().key());
if (!MapSweepPolicy::traceWeak(trc, &key, &e.front().value())) {
e.removeFront();
} else if (!HashPolicy::match(key, e.front().key())) {
e.rekeyFront(key);
}
}
}
// GCRekeyableHashMap is movable
GCRekeyableHashMap(GCRekeyableHashMap&& rhs) : Base(std::move(rhs)) {}
void operator=(GCRekeyableHashMap&& rhs) {
MOZ_ASSERT(this != &rhs, "self-move assignment is prohibited");
Base::operator=(std::move(rhs));
}
} MOZ_INHERIT_TYPE_ANNOTATIONS_FROM_TEMPLATE_ARGS;
template <typename Wrapper, typename... Args>
class WrappedPtrOperations<JS::GCHashMap<Args...>, Wrapper> {
using Map = JS::GCHashMap<Args...>;
using Lookup = typename Map::Lookup;
const Map& map() const { return static_cast<const Wrapper*>(this)->get(); }
public:
using AddPtr = typename Map::AddPtr;
using Ptr = typename Map::Ptr;
using Range = typename Map::Range;
Ptr lookup(const Lookup& l) const { return map().lookup(l); }
Range all() const { return map().all(); }
bool empty() const { return map().empty(); }
uint32_t count() const { return map().count(); }
size_t capacity() const { return map().capacity(); }
bool has(const Lookup& l) const { return map().lookup(l).found(); }
size_t sizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return map().sizeOfExcludingThis(mallocSizeOf);
}
size_t sizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return mallocSizeOf(this) + map().sizeOfExcludingThis(mallocSizeOf);
}
};
template <typename Wrapper, typename... Args>
class MutableWrappedPtrOperations<JS::GCHashMap<Args...>, Wrapper>
: public WrappedPtrOperations<JS::GCHashMap<Args...>, Wrapper> {
using Map = JS::GCHashMap<Args...>;
using Lookup = typename Map::Lookup;
Map& map() { return static_cast<Wrapper*>(this)->get(); }
public:
using AddPtr = typename Map::AddPtr;
struct Enum : public Map::Enum {
explicit Enum(Wrapper& o) : Map::Enum(o.map()) {}
};
using Ptr = typename Map::Ptr;
using Range = typename Map::Range;
void clear() { map().clear(); }
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
void clearAndCompact() { map().clearAndCompact(); }
void remove(Ptr p) { map().remove(p); }
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
AddPtr lookupForAdd(const Lookup& l) { return map().lookupForAdd(l); }
template <typename KeyInput, typename ValueInput>
bool add(AddPtr& p, KeyInput&& k, ValueInput&& v) {
return map().add(p, std::forward<KeyInput>(k), std::forward<ValueInput>(v));
}
template <typename KeyInput>
bool add(AddPtr& p, KeyInput&& k) {
return map().add(p, std::forward<KeyInput>(k), Map::Value());
}
template <typename KeyInput, typename ValueInput>
bool relookupOrAdd(AddPtr& p, KeyInput&& k, ValueInput&& v) {
return map().relookupOrAdd(p, k, std::forward<KeyInput>(k),
std::forward<ValueInput>(v));
}
template <typename KeyInput, typename ValueInput>
bool put(KeyInput&& k, ValueInput&& v) {
return map().put(std::forward<KeyInput>(k), std::forward<ValueInput>(v));
}
template <typename KeyInput, typename ValueInput>
bool putNew(KeyInput&& k, ValueInput&& v) {
return map().putNew(std::forward<KeyInput>(k), std::forward<ValueInput>(v));
}
};
} // namespace js
namespace JS {
// A GCHashSet is a HashSet with an additional trace method that knows
// be traced to be kept alive will generally want to use this GCHashSet
// specialization in lieu of HashSet.
//
// Most types of GC pointers can be traced with no extra infrastructure. For
// structs and non-gc-pointer members, ensure that there is a specialization of
// GCPolicy<T> with an appropriate trace method available to handle the custom
// type. Generic helpers can be found in js/public/TracingAPI.h.
//
// Note that although this HashSet's trace will deal correctly with moved
// elements, it does not itself know when to barrier or trace elements. To
// function properly it must either be used with Rooted or barriered and traced
// manually.
template <typename T, typename HashPolicy = js::DefaultHasher<T>,
typename AllocPolicy = js::TempAllocPolicy>
class GCHashSet : public js::HashSet<T, HashPolicy, AllocPolicy> {
using Base = js::HashSet<T, HashPolicy, AllocPolicy>;
public:
explicit GCHashSet(AllocPolicy a = AllocPolicy()) : Base(std::move(a)) {}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
explicit GCHashSet(size_t length) : Base(length) {}
GCHashSet(AllocPolicy a, size_t length) : Base(std::move(a), length) {}
void trace(JSTracer* trc) {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
GCPolicy<T>::trace(trc, &e.mutableFront(), "hashset element");
}
}
bool needsSweep() const { return !this->empty(); }
void sweep() {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
if (GCPolicy<T>::needsSweep(&e.mutableFront())) {
e.removeFront();
}
}
}
void traceWeak(JSTracer* trc) {
for (typename Base::Enum e(*this); !e.empty(); e.popFront()) {
if (!GCPolicy<T>::traceWeak(trc, &e.mutableFront())) {
e.removeFront();
}
}
}
// GCHashSet is movable
GCHashSet(GCHashSet&& rhs) : Base(std::move(rhs)) {}
void operator=(GCHashSet&& rhs) {
MOZ_ASSERT(this != &rhs, "self-move assignment is prohibited");
Base::operator=(std::move(rhs));
}
private:
// GCHashSet is not copyable or assignable
GCHashSet(const GCHashSet& hs) = delete;
GCHashSet& operator=(const GCHashSet& hs) = delete;
} MOZ_INHERIT_TYPE_ANNOTATIONS_FROM_TEMPLATE_ARGS;
} // namespace JS
namespace js {
template <typename Wrapper, typename... Args>
class WrappedPtrOperations<JS::GCHashSet<Args...>, Wrapper> {
using Set = JS::GCHashSet<Args...>;
const Set& set() const { return static_cast<const Wrapper*>(this)->get(); }
public:
using Lookup = typename Set::Lookup;
using AddPtr = typename Set::AddPtr;
using Entry = typename Set::Entry;
using Ptr = typename Set::Ptr;
using Range = typename Set::Range;
Ptr lookup(const Lookup& l) const { return set().lookup(l); }
Range all() const { return set().all(); }
bool empty() const { return set().empty(); }
uint32_t count() const { return set().count(); }
size_t capacity() const { return set().capacity(); }
bool has(const Lookup& l) const { return set().lookup(l).found(); }
size_t sizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return set().sizeOfExcludingThis(mallocSizeOf);
}
size_t sizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return mallocSizeOf(this) + set().sizeOfExcludingThis(mallocSizeOf);
}
};
template <typename Wrapper, typename... Args>
class MutableWrappedPtrOperations<JS::GCHashSet<Args...>, Wrapper>
: public WrappedPtrOperations<JS::GCHashSet<Args...>, Wrapper> {
using Set = JS::GCHashSet<Args...>;
using Lookup = typename Set::Lookup;
Set& set() { return static_cast<Wrapper*>(this)->get(); }
public:
using AddPtr = typename Set::AddPtr;
using Entry = typename Set::Entry;
struct Enum : public Set::Enum {
explicit Enum(Wrapper& o) : Set::Enum(o.set()) {}
};
using Ptr = typename Set::Ptr;
using Range = typename Set::Range;
void clear() { set().clear(); }
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
void clearAndCompact() { set().clearAndCompact(); }
MOZ_MUST_USE bool reserve(uint32_t len) { return set().reserve(len); }
void remove(Ptr p) { set().remove(p); }
void remove(const Lookup& l) { set().remove(l); }
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
AddPtr lookupForAdd(const Lookup& l) { return set().lookupForAdd(l); }
template <typename TInput>
bool add(AddPtr& p, TInput&& t) {
return set().add(p, std::forward<TInput>(t));
}
template <typename TInput>
bool relookupOrAdd(AddPtr& p, const Lookup& l, TInput&& t) {
return set().relookupOrAdd(p, l, std::forward<TInput>(t));
}
template <typename TInput>
bool put(TInput&& t) {
return set().put(std::forward<TInput>(t));
}
template <typename TInput>
bool putNew(TInput&& t) {
return set().putNew(std::forward<TInput>(t));
}
template <typename TInput>
bool putNew(const Lookup& l, TInput&& t) {
return set().putNew(l, std::forward<TInput>(t));
}
};
} /* namespace js */
namespace JS {
// Specialize WeakCache for GCHashMap to provide a barriered map that does not
// need to be swept immediately.
template <typename Key, typename Value, typename HashPolicy,
typename AllocPolicy, typename MapSweepPolicy>
class WeakCache<GCHashMap<Key, Value, HashPolicy, AllocPolicy, MapSweepPolicy>>
: protected detail::WeakCacheBase {
using Map = GCHashMap<Key, Value, HashPolicy, AllocPolicy, MapSweepPolicy>;
using Self = WeakCache<Map>;
Map map;
bool needsBarrier;
public:
template <typename... Args>
explicit WeakCache(Zone* zone, Args&&... args)
: WeakCacheBase(zone),
map(std::forward<Args>(args)...),
needsBarrier(false) {}
template <typename... Args>
explicit WeakCache(JSRuntime* rt, Args&&... args)
: WeakCacheBase(rt),
map(std::forward<Args>(args)...),
needsBarrier(false) {}
~WeakCache() { MOZ_ASSERT(!needsBarrier); }
bool needsSweep() override { return map.needsSweep(); }
size_t sweep() override {
size_t steps = map.count();
map.sweep();
return steps;
}
bool setNeedsIncrementalBarrier(bool needs) override {
MOZ_ASSERT(needsBarrier != needs);
needsBarrier = needs;
return true;
}
bool needsIncrementalBarrier() const override { return needsBarrier; }
private:
using Entry = typename Map::Entry;
static bool entryNeedsSweep(const Entry& prior) {
Key key(prior.key());
Value value(prior.value());
bool result = MapSweepPolicy::needsSweep(&key, &value);
MOZ_ASSERT(prior.key() == key); // We shouldn't update here.
MOZ_ASSERT(prior.value() == value); // We shouldn't update here.
return result;
}
public:
using Lookup = typename Map::Lookup;
using Ptr = typename Map::Ptr;
using AddPtr = typename Map::AddPtr;
struct Range {
explicit Range(const typename Map::Range& r) : range(r) { settle(); }
Range() {}
bool empty() const { return range.empty(); }
const Entry& front() const { return range.front(); }
void popFront() {
range.popFront();
settle();
}
private:
typename Map::Range range;
void settle() {
while (!empty() && entryNeedsSweep(front())) {
popFront();
}
}
};
struct Enum : public Map::Enum {
explicit Enum(Self& cache) : Map::Enum(cache.map) {
// This operation is not allowed while barriers are in place as we
// may also need to enumerate the set for sweeping.
MOZ_ASSERT(!cache.needsBarrier);
}
};
Ptr lookup(const Lookup& l) const {
Ptr ptr = map.lookup(l);
if (needsBarrier && ptr && entryNeedsSweep(*ptr)) {
const_cast<Map&>(map).remove(ptr);
return Ptr();
}
return ptr;
}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
AddPtr lookupForAdd(const Lookup& l) {
AddPtr ptr = map.lookupForAdd(l);
if (needsBarrier && ptr && entryNeedsSweep(*ptr)) {
const_cast<Map&>(map).remove(ptr);
return map.lookupForAdd(l);
}
return ptr;
}
Range all() const { return Range(map.all()); }
bool empty() const {
// This operation is not currently allowed while barriers are in place
// as it would require iterating the map and the caller expects a
// constant time operation.
MOZ_ASSERT(!needsBarrier);
return map.empty();
}
uint32_t count() const {
// This operation is not currently allowed while barriers are in place
// as it would require iterating the set and the caller expects a
// constant time operation.
MOZ_ASSERT(!needsBarrier);
return map.count();
}
size_t capacity() const { return map.capacity(); }
bool has(const Lookup& l) const { return lookup(l).found(); }
size_t sizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return map.sizeOfExcludingThis(mallocSizeOf);
}
size_t sizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return mallocSizeOf(this) + map.shallowSizeOfExcludingThis(mallocSizeOf);
}
void clear() {
// This operation is not currently allowed while barriers are in place
// since it doesn't make sense to clear a cache while it is being swept.
MOZ_ASSERT(!needsBarrier);
map.clear();
}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
void clearAndCompact() {
// This operation is not currently allowed while barriers are in place
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
// since it doesn't make sense to clear a cache while it is being swept.
MOZ_ASSERT(!needsBarrier);
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
map.clearAndCompact();
}
void remove(Ptr p) {
// This currently supports removing entries during incremental
// sweeping. If we allow these tables to be swept incrementally this may
// no longer be possible.
map.remove(p);
}
void remove(const Lookup& l) {
Ptr p = lookup(l);
if (p) {
remove(p);
}
}
template <typename KeyInput, typename ValueInput>
bool add(AddPtr& p, KeyInput&& k, ValueInput&& v) {
return map.add(p, std::forward<KeyInput>(k), std::forward<ValueInput>(v));
}
template <typename KeyInput, typename ValueInput>
bool relookupOrAdd(AddPtr& p, KeyInput&& k, ValueInput&& v) {
return map.relookupOrAdd(p, std::forward<KeyInput>(k),
std::forward<ValueInput>(v));
}
template <typename KeyInput, typename ValueInput>
bool put(KeyInput&& k, ValueInput&& v) {
return map.put(std::forward<KeyInput>(k), std::forward<ValueInput>(v));
}
template <typename KeyInput, typename ValueInput>
bool putNew(KeyInput&& k, ValueInput&& v) {
return map.putNew(std::forward<KeyInput>(k), std::forward<ValueInput>(v));
}
} JS_HAZ_NON_GC_POINTER;
// Specialize WeakCache for GCHashSet to provide a barriered set that does not
// need to be swept immediately.
template <typename T, typename HashPolicy, typename AllocPolicy>
class WeakCache<GCHashSet<T, HashPolicy, AllocPolicy>>
: protected detail::WeakCacheBase {
using Set = GCHashSet<T, HashPolicy, AllocPolicy>;
using Self = WeakCache<Set>;
Set set;
bool needsBarrier;
public:
using Entry = typename Set::Entry;
template <typename... Args>
explicit WeakCache(Zone* zone, Args&&... args)
: WeakCacheBase(zone),
set(std::forward<Args>(args)...),
needsBarrier(false) {}
template <typename... Args>
explicit WeakCache(JSRuntime* rt, Args&&... args)
: WeakCacheBase(rt),
set(std::forward<Args>(args)...),
needsBarrier(false) {}
size_t sweep() override {
size_t steps = set.count();
set.sweep();
return steps;
}
bool needsSweep() override { return set.needsSweep(); }
bool setNeedsIncrementalBarrier(bool needs) override {
MOZ_ASSERT(needsBarrier != needs);
needsBarrier = needs;
return true;
}
bool needsIncrementalBarrier() const override { return needsBarrier; }
private:
static bool entryNeedsSweep(const Entry& prior) {
Entry entry(prior);
bool result = GCPolicy<T>::needsSweep(&entry);
MOZ_ASSERT(prior == entry); // We shouldn't update here.
return result;
}
public:
using Lookup = typename Set::Lookup;
using Ptr = typename Set::Ptr;
using AddPtr = typename Set::AddPtr;
struct Range {
explicit Range(const typename Set::Range& r) : range(r) { settle(); }
Range() {}
bool empty() const { return range.empty(); }
const Entry& front() const { return range.front(); }
void popFront() {
range.popFront();
settle();
}
private:
typename Set::Range range;
void settle() {
while (!empty() && entryNeedsSweep(front())) {
popFront();
}
}
};
struct Enum : public Set::Enum {
explicit Enum(Self& cache) : Set::Enum(cache.set) {
// This operation is not allowed while barriers are in place as we
// may also need to enumerate the set for sweeping.
MOZ_ASSERT(!cache.needsBarrier);
}
};
Ptr lookup(const Lookup& l) const {
Ptr ptr = set.lookup(l);
if (needsBarrier && ptr && entryNeedsSweep(*ptr)) {
const_cast<Set&>(set).remove(ptr);
return Ptr();
}
return ptr;
}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
AddPtr lookupForAdd(const Lookup& l) {
AddPtr ptr = set.lookupForAdd(l);
if (needsBarrier && ptr && entryNeedsSweep(*ptr)) {
const_cast<Set&>(set).remove(ptr);
return set.lookupForAdd(l);
}
return ptr;
}
Range all() const { return Range(set.all()); }
bool empty() const {
// This operation is not currently allowed while barriers are in place
// as it would require iterating the set and the caller expects a
// constant time operation.
MOZ_ASSERT(!needsBarrier);
return set.empty();
}
uint32_t count() const {
// This operation is not currently allowed while barriers are in place
// as it would require iterating the set and the caller expects a
// constant time operation.
MOZ_ASSERT(!needsBarrier);
return set.count();
}
size_t capacity() const { return set.capacity(); }
bool has(const Lookup& l) const { return lookup(l).found(); }
size_t sizeOfExcludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return set.shallowSizeOfExcludingThis(mallocSizeOf);
}
size_t sizeOfIncludingThis(mozilla::MallocSizeOf mallocSizeOf) const {
return mallocSizeOf(this) + set.shallowSizeOfExcludingThis(mallocSizeOf);
}
void clear() {
// This operation is not currently allowed while barriers are in place
// since it doesn't make sense to clear a cache while it is being swept.
MOZ_ASSERT(!needsBarrier);
set.clear();
}
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
void clearAndCompact() {
// This operation is not currently allowed while barriers are in place
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
// since it doesn't make sense to clear a cache while it is being swept.
MOZ_ASSERT(!needsBarrier);
Bug 1481998 - Make mozilla::Hash{Map,Set}'s entry storage allocation lazy. r=luke,sfink Entry storage allocation now occurs on the first lookupForAdd()/put()/putNew(). This removes the need for init() and initialized(), and matches how PLDHashTable/nsTHashtable work. It also removes the need for init() functions in a lot of types that are built on top of mozilla::Hash{Map,Set}. Pros: - No need for init() calls and subsequent checks. - No memory allocated for empty tables, which are not that uncommon. Cons: - An extra branch in lookup() and lookupForAdd(), but not in put()/putNew(), because the existing checkOverloaded() can handle it. Specifics: - Construction now can take a length parameter. - init() is removed. Explicit length-setting, when necessary, now occurs in the constructors. - initialized() is removed. - capacity() now returns zero when the entry storage is absent. - lookupForAdd() is no longer `const`, because it can instantiate the storage, which requires modifications. - lookupForAdd() can now return an invalid AddPtr in two cases: - old: hashing failure (due to OOM in the hasher) - new: OOM while instantiating entry storage The existing failure handling paths for the old case work for the new case. - clear(), finish(), and clearAndShrink() are replaced by clear(), compact(), and reserve(). The old compactIfUnderloaded() is also removed. - Capacity computation code is now in its own functions, bestCapacity() and hashShift(). setTableSizeLog2() is removed. - uint32_t is used throughout for capacities, instead of size_t, for consistency with other similar values. - changeTableSize() now takes a capacity instead of a deltaLog2, and it can now handle !mTable. Measurements: - Total source code size is reduced by over 900 lines. Also, lots of existing lines got shorter (i.e. two checks were reduced to one). - Executable size barely changed, down by 2 KiB on Linux64. The extra branches are compensated for by the lack of init() calls. - Speed changed negligibly. The instruction count for Bench_Cpp_MozHash increased from 2.84 billion to 2.89 billion but any execution time change was well below noise.
2018-08-10 11:00:29 +03:00
set.clearAndCompact();
}
void remove(Ptr p) {
// This currently supports removing entries during incremental
// sweeping. If we allow these tables to be swept incrementally this may
// no longer be possible.
set.remove(p);
}
void remove(const Lookup& l) {
Ptr p = lookup(l);
if (p) {
remove(p);
}
}
template <typename TInput>
bool add(AddPtr& p, TInput&& t) {
return set.add(p, std::forward<TInput>(t));
}
template <typename TInput>
bool relookupOrAdd(AddPtr& p, const Lookup& l, TInput&& t) {
return set.relookupOrAdd(p, l, std::forward<TInput>(t));
}
template <typename TInput>
bool put(TInput&& t) {
return set.put(std::forward<TInput>(t));
}
template <typename TInput>
bool putNew(TInput&& t) {
return set.putNew(std::forward<TInput>(t));
}
template <typename TInput>
bool putNew(const Lookup& l, TInput&& t) {
return set.putNew(l, std::forward<TInput>(t));
}
} JS_HAZ_NON_GC_POINTER;
} // namespace JS
#endif /* GCHashTable_h */