/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- * vim: set ts=8 sw=4 et tw=99 ft=cpp: * * 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 js_HashTable_h__ #define js_HashTable_h__ #include "mozilla/Attributes.h" #include "mozilla/DebugOnly.h" #include "mozilla/Util.h" #include "js/TemplateLib.h" #include "js/Utility.h" namespace js { class TempAllocPolicy; template struct DefaultHasher; template class HashMapEntry; namespace detail { template class HashTableEntry; template class HashTable; } /*****************************************************************************/ // A JS-friendly, STL-like container providing a hash-based map from keys to // values. In particular, HashMap calls constructors and destructors of all // objects added so non-PODs may be used safely. // // Key/Value requirements: // - movable, destructible, assignable // HashPolicy requirements: // - see Hash Policy section below // AllocPolicy: // - see jsalloc.h // // Note: // - HashMap is not reentrant: Key/Value/HashPolicy/AllocPolicy members // called by HashMap must not call back into the same HashMap object. // - Due to the lack of exception handling, the user must call |init()|. template , class AllocPolicy = TempAllocPolicy> class HashMap { typedef HashMapEntry TableEntry; struct MapHashPolicy : HashPolicy { typedef Key KeyType; static const Key &getKey(TableEntry &e) { return e.key; } static void setKey(TableEntry &e, Key &k) { const_cast(e.key) = k; } }; typedef detail::HashTable Impl; Impl impl; public: typedef typename HashPolicy::Lookup Lookup; typedef TableEntry Entry; // HashMap construction is fallible (due to OOM); thus the user must call // init after constructing a HashMap and check the return value. HashMap(AllocPolicy a = AllocPolicy()) : impl(a) {} bool init(uint32_t len = 16) { return impl.init(len); } bool initialized() const { return impl.initialized(); } // Return whether the given lookup value is present in the map. E.g.: // // typedef HashMap HM; // HM h; // if (HM::Ptr p = h.lookup(3)) { // const HM::Entry &e = *p; // p acts like a pointer to Entry // assert(p->key == 3); // Entry contains the key // char val = p->value; // and value // } // // Also see the definition of Ptr in HashTable above (with T = Entry). typedef typename Impl::Ptr Ptr; Ptr lookup(const Lookup &l) const { return impl.lookup(l); } // Assuming |p.found()|, remove |*p|. void remove(Ptr p) { impl.remove(p); } // Like |lookup(l)|, but on miss, |p = lookupForAdd(l)| allows efficient // insertion of Key |k| (where |HashPolicy::match(k,l) == true|) using // |add(p,k,v)|. After |add(p,k,v)|, |p| points to the new Entry. E.g.: // // typedef HashMap HM; // HM h; // HM::AddPtr p = h.lookupForAdd(3); // if (!p) { // if (!h.add(p, 3, 'a')) // return false; // } // const HM::Entry &e = *p; // p acts like a pointer to Entry // assert(p->key == 3); // Entry contains the key // char val = p->value; // and value // // Also see the definition of AddPtr in HashTable above (with T = Entry). // // N.B. The caller must ensure that no mutating hash table operations // occur between a pair of |lookupForAdd| and |add| calls. To avoid // looking up the key a second time, the caller may use the more efficient // relookupOrAdd method. This method reuses part of the hashing computation // to more efficiently insert the key if it has not been added. For // example, a mutation-handling version of the previous example: // // HM::AddPtr p = h.lookupForAdd(3); // if (!p) { // call_that_may_mutate_h(); // if (!h.relookupOrAdd(p, 3, 'a')) // return false; // } // const HM::Entry &e = *p; // assert(p->key == 3); // char val = p->value; typedef typename Impl::AddPtr AddPtr; AddPtr lookupForAdd(const Lookup &l) const { return impl.lookupForAdd(l); } template bool add(AddPtr &p, const KeyInput &k, const ValueInput &v) { Entry e(k, v); return impl.add(p, Move(e)); } bool add(AddPtr &p, const Key &k) { Entry e(k, Value()); return impl.add(p, Move(e)); } template bool relookupOrAdd(AddPtr &p, const KeyInput &k, const ValueInput &v) { Entry e(k, v); return impl.relookupOrAdd(p, k, Move(e)); } // |all()| returns a Range containing |count()| elements. E.g.: // // typedef HashMap HM; // HM h; // for (HM::Range r = h.all(); !r.empty(); r.popFront()) // char c = r.front().value; // // Also see the definition of Range in HashTable above (with T = Entry). typedef typename Impl::Range Range; Range all() const { return impl.all(); } // Typedef for the enumeration class. An Enum may be used to examine and // remove table entries: // // typedef HashMap HM; // HM s; // for (HM::Enum e(s); !e.empty(); e.popFront()) // if (e.front().value == 'l') // e.removeFront(); // // Table resize may occur in Enum's destructor. Also see the definition of // Enum in HashTable above (with T = Entry). typedef typename Impl::Enum Enum; // Remove all entries. This does not shrink the table. For that consider // using the finish() method. void clear() { impl.clear(); } // Remove all the entries and release all internal buffers. The map must // be initialized again before any use. void finish() { impl.finish(); } // Does the table contain any entries? bool empty() const { return impl.empty(); } // Number of live elements in the map. uint32_t count() const { return impl.count(); } // Total number of allocation in the dynamic table. Note: resize will // happen well before count() == capacity(). size_t capacity() const { return impl.capacity(); } // Don't just call |impl.sizeOfExcludingThis()| because there's no // guarantee that |impl| is the first field in HashMap. size_t sizeOfExcludingThis(JSMallocSizeOfFun mallocSizeOf) const { return impl.sizeOfExcludingThis(mallocSizeOf); } size_t sizeOfIncludingThis(JSMallocSizeOfFun mallocSizeOf) const { return mallocSizeOf(this) + impl.sizeOfExcludingThis(mallocSizeOf); } // If |generation()| is the same before and after a HashMap operation, // pointers into the table remain valid. unsigned generation() const { return impl.generation(); } /************************************************** Shorthand operations */ bool has(const Lookup &l) const { return impl.lookup(l) != NULL; } // Overwrite existing value with v. Return false on oom. template bool put(const KeyInput &k, const ValueInput &v) { AddPtr p = lookupForAdd(k); if (p) { p->value = v; return true; } return add(p, k, v); } // Like put, but assert that the given key is not already present. template bool putNew(const KeyInput &k, const ValueInput &v) { Entry e(k, v); return impl.putNew(k, Move(e)); } // Add (k,defaultValue) if |k| is not found. Return a false-y Ptr on oom. Ptr lookupWithDefault(const Key &k, const Value &defaultValue) { AddPtr p = lookupForAdd(k); if (p) return p; (void)add(p, k, defaultValue); // p is left false-y on oom. return p; } // Remove if present. void remove(const Lookup &l) { if (Ptr p = lookup(l)) remove(p); } private: // Not implicitly copyable (expensive). May add explicit |clone| later. HashMap(const HashMap &hm) MOZ_DELETE; HashMap &operator=(const HashMap &hm) MOZ_DELETE; friend class Impl::Enum; typedef typename tl::StaticAssert::result>::result keyAssert; typedef typename tl::StaticAssert::result>::result valAssert; }; /*****************************************************************************/ // A JS-friendly, STL-like container providing a hash-based set of values. In // particular, HashSet calls constructors and destructors of all objects added // so non-PODs may be used safely. // // T requirements: // - movable, destructible, assignable // HashPolicy requirements: // - see Hash Policy section below // AllocPolicy: // - see jsalloc.h // // Note: // - HashSet is not reentrant: T/HashPolicy/AllocPolicy members called by // HashSet must not call back into the same HashSet object. // - Due to the lack of exception handling, the user must call |init()|. template , class AllocPolicy = TempAllocPolicy> class HashSet { struct SetOps : HashPolicy { typedef T KeyType; static const KeyType &getKey(const T &t) { return t; } static void setKey(T &t, KeyType &k) { t = k; } }; typedef detail::HashTable Impl; Impl impl; public: typedef typename HashPolicy::Lookup Lookup; typedef T Entry; // HashSet construction is fallible (due to OOM); thus the user must call // init after constructing a HashSet and check the return value. HashSet(AllocPolicy a = AllocPolicy()) : impl(a) {} bool init(uint32_t len = 16) { return impl.init(len); } bool initialized() const { return impl.initialized(); } // Return whether the given lookup value is present in the map. E.g.: // // typedef HashSet HS; // HS h; // if (HS::Ptr p = h.lookup(3)) { // assert(*p == 3); // p acts like a pointer to int // } // // Also see the definition of Ptr in HashTable above. typedef typename Impl::Ptr Ptr; Ptr lookup(const Lookup &l) const { return impl.lookup(l); } // Assuming |p.found()|, remove |*p|. void remove(Ptr p) { impl.remove(p); } // Like |lookup(l)|, but on miss, |p = lookupForAdd(l)| allows efficient // insertion of T value |t| (where |HashPolicy::match(t,l) == true|) using // |add(p,t)|. After |add(p,t)|, |p| points to the new element. E.g.: // // typedef HashSet HS; // HS h; // HS::AddPtr p = h.lookupForAdd(3); // if (!p) { // if (!h.add(p, 3)) // return false; // } // assert(*p == 3); // p acts like a pointer to int // // Also see the definition of AddPtr in HashTable above. // // N.B. The caller must ensure that no mutating hash table operations // occur between a pair of |lookupForAdd| and |add| calls. To avoid // looking up the key a second time, the caller may use the more efficient // relookupOrAdd method. This method reuses part of the hashing computation // to more efficiently insert the key if it has not been added. For // example, a mutation-handling version of the previous example: // // HS::AddPtr p = h.lookupForAdd(3); // if (!p) { // call_that_may_mutate_h(); // if (!h.relookupOrAdd(p, 3, 3)) // return false; // } // assert(*p == 3); // // Note that relookupOrAdd(p,l,t) performs Lookup using |l| and adds the // entry |t|, where the caller ensures match(l,t). typedef typename Impl::AddPtr AddPtr; AddPtr lookupForAdd(const Lookup &l) const { return impl.lookupForAdd(l); } bool add(AddPtr &p, const T &t) { return impl.add(p, t); } bool relookupOrAdd(AddPtr &p, const Lookup &l, const T &t) { return impl.relookupOrAdd(p, l, t); } // |all()| returns a Range containing |count()| elements: // // typedef HashSet HS; // HS h; // for (HS::Range r = h.all(); !r.empty(); r.popFront()) // int i = r.front(); // // Also see the definition of Range in HashTable above. typedef typename Impl::Range Range; Range all() const { return impl.all(); } // Typedef for the enumeration class. An Enum may be used to examine and // remove table entries: // // typedef HashSet HS; // HS s; // for (HS::Enum e(s); !e.empty(); e.popFront()) // if (e.front() == 42) // e.removeFront(); // // Table resize may occur in Enum's destructor. Also see the definition of // Enum in HashTable above. typedef typename Impl::Enum Enum; // Remove all entries. This does not shrink the table. For that consider // using the finish() method. void clear() { impl.clear(); } // Remove all the entries and release all internal buffers. The set must // be initialized again before any use. void finish() { impl.finish(); } // Does the table contain any entries? bool empty() const { return impl.empty(); } // Number of live elements in the map. uint32_t count() const { return impl.count(); } // Total number of allocation in the dynamic table. Note: resize will // happen well before count() == capacity(). size_t capacity() const { return impl.capacity(); } // Don't just call |impl.sizeOfExcludingThis()| because there's no // guarantee that |impl| is the first field in HashSet. size_t sizeOfExcludingThis(JSMallocSizeOfFun mallocSizeOf) const { return impl.sizeOfExcludingThis(mallocSizeOf); } size_t sizeOfIncludingThis(JSMallocSizeOfFun mallocSizeOf) const { return mallocSizeOf(this) + impl.sizeOfExcludingThis(mallocSizeOf); } // If |generation()| is the same before and after a HashSet operation, // pointers into the table remain valid. unsigned generation() const { return impl.generation(); } /************************************************** Shorthand operations */ bool has(const Lookup &l) const { return impl.lookup(l) != NULL; } // Overwrite existing value with v. Return false on oom. bool put(const T &t) { AddPtr p = lookupForAdd(t); return p ? true : add(p, t); } // Like put, but assert that the given key is not already present. bool putNew(const T &t) { return impl.putNew(t, t); } bool putNew(const Lookup &l, const T &t) { return impl.putNew(l, t); } void remove(const Lookup &l) { if (Ptr p = lookup(l)) remove(p); } private: // Not implicitly copyable (expensive). May add explicit |clone| later. HashSet(const HashSet &hs) MOZ_DELETE; HashSet &operator=(const HashSet &hs) MOZ_DELETE; friend class Impl::Enum; typedef typename tl::StaticAssert::result>::result _; }; /*****************************************************************************/ // Hash Policy // // A hash policy P for a hash table with key-type Key must provide: // - a type |P::Lookup| to use to lookup table entries; // - a static member function |P::hash| with signature // // static js::HashNumber hash(Lookup) // // to use to hash the lookup type; and // - a static member function |P::match| with signature // // static bool match(Key, Lookup) // // to use to test equality of key and lookup values. // // Normally, Lookup = Key. In general, though, different values and types of // values can be used to lookup and store. If a Lookup value |l| is != to the // added Key value |k|, the user must ensure that |P::match(k,l)|. E.g.: // // js::HashSet::AddPtr p = h.lookup(l); // if (!p) { // assert(P::match(k, l)); // must hold // h.add(p, k); // } // Pointer hashing policy that strips the lowest zeroBits when calculating the // hash to improve key distribution. template struct PointerHasher { typedef Key Lookup; static HashNumber hash(const Lookup &l) { JS_ASSERT(!js::IsPoisonedPtr(l)); size_t word = reinterpret_cast(l) >> zeroBits; JS_STATIC_ASSERT(sizeof(HashNumber) == 4); #if JS_BYTES_PER_WORD == 4 return HashNumber(word); #else JS_STATIC_ASSERT(sizeof word == 8); return HashNumber((word >> 32) ^ word); #endif } static bool match(const Key &k, const Lookup &l) { JS_ASSERT(!js::IsPoisonedPtr(k)); JS_ASSERT(!js::IsPoisonedPtr(l)); return k == l; } }; // Default hash policy: just use the 'lookup' value. This of course only // works if the lookup value is integral. HashTable applies ScrambleHashCode to // the result of the 'hash' which means that it is 'ok' if the lookup value is // not well distributed over the HashNumber domain. template struct DefaultHasher { typedef Key Lookup; static HashNumber hash(const Lookup &l) { // Hash if can implicitly cast to hash number type. return l; } static bool match(const Key &k, const Lookup &l) { // Use builtin or overloaded operator==. return k == l; } }; // Specialize hashing policy for pointer types. It assumes that the type is // at least word-aligned. For types with smaller size use PointerHasher. template struct DefaultHasher : PointerHasher::result> {}; /*****************************************************************************/ // Both HashMap and HashSet are implemented by a single HashTable that is even // more heavily parameterized than the other two. This leaves HashTable gnarly // and extremely coupled to HashMap and HashSet; thus code should not use // HashTable directly. template class HashMapEntry { template friend class detail::HashTable; template friend class detail::HashTableEntry; HashMapEntry(const HashMapEntry &) MOZ_DELETE; void operator=(const HashMapEntry &) MOZ_DELETE; public: template HashMapEntry(const KeyInput &k, const ValueInput &v) : key(k), value(v) {} HashMapEntry(MoveRef rhs) : key(Move(rhs->key)), value(Move(rhs->value)) { } const Key key; Value value; }; namespace tl { template struct IsPodType > { static const bool result = IsPodType::result; }; template struct IsPodType > { static const bool result = IsPodType::result && IsPodType::result; }; } // namespace tl namespace detail { template class HashTable; template class HashTableEntry { template friend class HashTable; typedef typename tl::StripConst::result NonConstT; HashNumber keyHash; mozilla::AlignedStorage2 mem; static const HashNumber sFreeKey = 0; static const HashNumber sRemovedKey = 1; static const HashNumber sCollisionBit = 1; // Assumed by calloc in createTable. JS_STATIC_ASSERT(sFreeKey == 0); static bool isLiveHash(HashNumber hash) { return hash > sRemovedKey; } HashTableEntry(const HashTableEntry &) MOZ_DELETE; void operator=(const HashTableEntry &) MOZ_DELETE; ~HashTableEntry() MOZ_DELETE; public: // NB: HashTableEntry is treated as a POD: no constructor or destructor calls. void destroyIfLive() { if (isLive()) mem.addr()->~T(); } void destroy() { JS_ASSERT(isLive()); mem.addr()->~T(); } void swap(HashTableEntry *other) { Swap(keyHash, other->keyHash); Swap(mem, other->mem); } T &get() { JS_ASSERT(isLive()); return *mem.addr(); } bool isFree() const { return keyHash == sFreeKey; } void clearLive() { JS_ASSERT(isLive()); keyHash = sFreeKey; mem.addr()->~T(); } void clear() { if (isLive()) mem.addr()->~T(); keyHash = sFreeKey; } bool isRemoved() const { return keyHash == sRemovedKey; } void removeLive() { JS_ASSERT(isLive()); keyHash = sRemovedKey; mem.addr()->~T(); } bool isLive() const { return isLiveHash(keyHash); } void setCollision() { JS_ASSERT(isLive()); keyHash |= sCollisionBit; } void setCollision(HashNumber bit) { JS_ASSERT(isLive()); keyHash |= bit; } void unsetCollision() { keyHash &= ~sCollisionBit; } bool hasCollision() const { return keyHash & sCollisionBit; } bool matchHash(HashNumber hn) { return (keyHash & ~sCollisionBit) == hn; } HashNumber getKeyHash() const { return keyHash & ~sCollisionBit; } template void setLive(HashNumber hn, const U &u) { JS_ASSERT(!isLive()); keyHash = hn; new(mem.addr()) T(u); JS_ASSERT(isLive()); } }; template class HashTable : private AllocPolicy { typedef typename tl::StripConst::result NonConstT; typedef typename HashPolicy::KeyType Key; typedef typename HashPolicy::Lookup Lookup; public: typedef HashTableEntry Entry; // A nullable pointer to a hash table element. A Ptr |p| can be tested // either explicitly |if (p.found()) p->...| or using boolean conversion // |if (p) p->...|. Ptr objects must not be used after any mutating hash // table operations unless |generation()| is tested. class Ptr { friend class HashTable; typedef void (Ptr::* ConvertibleToBool)(); void nonNull() {} Entry *entry; protected: Ptr(Entry &entry) : entry(&entry) {} public: // Leaves Ptr uninitialized. Ptr() { #ifdef DEBUG entry = (Entry *)0xbad; #endif } bool found() const { return entry->isLive(); } operator ConvertibleToBool() const { return found() ? &Ptr::nonNull : 0; } bool operator==(const Ptr &rhs) const { JS_ASSERT(found() && rhs.found()); return entry == rhs.entry; } bool operator!=(const Ptr &rhs) const { return !(*this == rhs); } T &operator*() const { return entry->get(); } T *operator->() const { return &entry->get(); } }; // A Ptr that can be used to add a key after a failed lookup. class AddPtr : public Ptr { friend class HashTable; HashNumber keyHash; mozilla::DebugOnly mutationCount; AddPtr(Entry &entry, HashNumber hn) : Ptr(entry), keyHash(hn) {} public: // Leaves AddPtr uninitialized. AddPtr() {} }; // A collection of hash table entries. The collection is enumerated by // calling |front()| followed by |popFront()| as long as |!empty()|. As // with Ptr/AddPtr, Range objects must not be used after any mutating hash // table operation unless the |generation()| is tested. class Range { protected: friend class HashTable; Range(Entry *c, Entry *e) : cur(c), end(e), validEntry(true) { while (cur < end && !cur->isLive()) ++cur; } Entry *cur, *end; mozilla::DebugOnly validEntry; public: Range() : cur(NULL), end(NULL), validEntry(false) {} bool empty() const { return cur == end; } T &front() const { JS_ASSERT(validEntry); JS_ASSERT(!empty()); return cur->get(); } void popFront() { JS_ASSERT(!empty()); while (++cur < end && !cur->isLive()) continue; validEntry = true; } }; // A Range whose lifetime delimits a mutating enumeration of a hash table. // Since rehashing when elements were removed during enumeration would be // bad, it is postponed until |endEnumeration()| is called. If // |endEnumeration()| is not called before an Enum's constructor, it will // be called automatically. Since |endEnumeration()| touches the hash // table, the user must ensure that the hash table is still alive when this // happens. class Enum : public Range { friend class HashTable; HashTable &table; bool rekeyed; bool removed; /* Not copyable. */ Enum(const Enum &); void operator=(const Enum &); public: template explicit Enum(Map &map) : Range(map.all()), table(map.impl), rekeyed(false), removed(false) {} // Removes the |front()| element from the table, leaving |front()| // invalid until the next call to |popFront()|. For example: // // HashSet s; // for (HashSet::Enum e(s); !e.empty(); e.popFront()) // if (e.front() == 42) // e.removeFront(); void removeFront() { table.remove(*this->cur); removed = true; this->validEntry = false; } // Removes the |front()| element and re-inserts it into the table with // a new key at the new Lookup position. |front()| is invalid after // this operation until the next call to |popFront()|. void rekeyFront(const Lookup &l, const Key &k) { typename HashTableEntry::NonConstT t(Move(this->cur->get())); HashPolicy::setKey(t, const_cast(k)); table.remove(*this->cur); table.putNewInfallible(l, Move(t)); rekeyed = true; this->validEntry = false; } void rekeyFront(const Key &k) { rekeyFront(k, k); } // Potentially rehashes the table. ~Enum() { if (rekeyed) table.checkOverRemoved(); if (removed) table.checkUnderloaded(); } }; private: uint32_t hashShift; // multiplicative hash shift uint32_t entryCount; // number of entries in table uint32_t gen; // entry storage generation number uint32_t removedCount; // removed entry sentinels in table Entry *table; // entry storage void setTableSizeLog2(unsigned sizeLog2) { hashShift = sHashBits - sizeLog2; } #ifdef DEBUG mutable struct Stats { uint32_t searches; // total number of table searches uint32_t steps; // hash chain links traversed uint32_t hits; // searches that found key uint32_t misses; // searches that didn't find key uint32_t addOverRemoved; // adds that recycled a removed entry uint32_t removes; // calls to remove uint32_t removeFrees; // calls to remove that freed the entry uint32_t grows; // table expansions uint32_t shrinks; // table contractions uint32_t compresses; // table compressions uint32_t rehashes; // tombstone decontaminations } stats; # define METER(x) x #else # define METER(x) #endif friend class js::ReentrancyGuard; mutable mozilla::DebugOnly entered; mozilla::DebugOnly mutationCount; // The default initial capacity is 16, but you can ask for as small as 4. static const unsigned sMinSizeLog2 = 2; static const unsigned sMinSize = 1 << sMinSizeLog2; static const unsigned sMaxInit = JS_BIT(23); static const unsigned sMaxCapacity = JS_BIT(24); static const unsigned sHashBits = tl::BitSize::result; static const uint8_t sMinAlphaFrac = 64; // (0x100 * .25) static const uint8_t sMaxAlphaFrac = 192; // (0x100 * .75) static const uint8_t sInvMaxAlpha = 171; // (ceil(0x100 / .75) >> 1) static const HashNumber sFreeKey = Entry::sFreeKey; static const HashNumber sRemovedKey = Entry::sRemovedKey; static const HashNumber sCollisionBit = Entry::sCollisionBit; static void staticAsserts() { // Rely on compiler "constant overflow warnings". JS_STATIC_ASSERT(((sMaxInit * sInvMaxAlpha) >> 7) < sMaxCapacity); JS_STATIC_ASSERT((sMaxCapacity * sInvMaxAlpha) <= UINT32_MAX); JS_STATIC_ASSERT((sMaxCapacity * sizeof(Entry)) <= UINT32_MAX); } static bool isLiveHash(HashNumber hash) { return Entry::isLiveHash(hash); } static HashNumber prepareHash(const Lookup& l) { HashNumber keyHash = ScrambleHashCode(HashPolicy::hash(l)); // Avoid reserved hash codes. if (!isLiveHash(keyHash)) keyHash -= (sRemovedKey + 1); return keyHash & ~sCollisionBit; } static Entry *createTable(AllocPolicy &alloc, uint32_t capacity) { // See JS_STATIC_ASSERT(sFreeKey == 0) in HashTableEntry. return (Entry *)alloc.calloc_(capacity * sizeof(Entry)); } static void destroyTable(AllocPolicy &alloc, Entry *oldTable, uint32_t capacity) { for (Entry *e = oldTable, *end = e + capacity; e < end; ++e) e->destroyIfLive(); alloc.free_(oldTable); } public: HashTable(AllocPolicy ap) : AllocPolicy(ap), hashShift(sHashBits), entryCount(0), gen(0), removedCount(0), table(NULL), entered(false), mutationCount(0) {} MOZ_WARN_UNUSED_RESULT bool init(uint32_t length) { JS_ASSERT(!initialized()); // Correct for sMaxAlphaFrac such that the table will not resize // when adding 'length' entries. if (length > sMaxInit) { this->reportAllocOverflow(); return false; } uint32_t capacity = (length * sInvMaxAlpha) >> 7; if (capacity < sMinSize) capacity = sMinSize; // FIXME: use JS_CEILING_LOG2 when PGO stops crashing (bug 543034). uint32_t roundUp = sMinSize, roundUpLog2 = sMinSizeLog2; while (roundUp < capacity) { roundUp <<= 1; ++roundUpLog2; } capacity = roundUp; JS_ASSERT(capacity <= sMaxCapacity); table = createTable(*this, capacity); if (!table) return false; setTableSizeLog2(roundUpLog2); METER(memset(&stats, 0, sizeof(stats))); return true; } bool initialized() const { return !!table; } ~HashTable() { if (table) destroyTable(*this, table, capacity()); } private: HashNumber hash1(HashNumber hash0) const { return hash0 >> hashShift; } struct DoubleHash { HashNumber h2; HashNumber sizeMask; }; DoubleHash hash2(HashNumber curKeyHash) const { unsigned sizeLog2 = sHashBits - hashShift; DoubleHash dh = { ((curKeyHash << sizeLog2) >> hashShift) | 1, (HashNumber(1) << sizeLog2) - 1 }; return dh; } static HashNumber applyDoubleHash(HashNumber h1, const DoubleHash &dh) { return (h1 - dh.h2) & dh.sizeMask; } bool overloaded() { return entryCount + removedCount >= ((sMaxAlphaFrac * capacity()) >> 8); } bool underloaded() { uint32_t tableCapacity = capacity(); return tableCapacity > sMinSize && entryCount <= ((sMinAlphaFrac * tableCapacity) >> 8); } static bool match(Entry &e, const Lookup &l) { return HashPolicy::match(HashPolicy::getKey(e.get()), l); } Entry &lookup(const Lookup &l, HashNumber keyHash, unsigned collisionBit) const { JS_ASSERT(isLiveHash(keyHash)); JS_ASSERT(!(keyHash & sCollisionBit)); JS_ASSERT(collisionBit == 0 || collisionBit == sCollisionBit); JS_ASSERT(table); METER(stats.searches++); // Compute the primary hash address. HashNumber h1 = hash1(keyHash); Entry *entry = &table[h1]; // Miss: return space for a new entry. if (entry->isFree()) { METER(stats.misses++); return *entry; } // Hit: return entry. if (entry->matchHash(keyHash) && match(*entry, l)) { METER(stats.hits++); return *entry; } // Collision: double hash. DoubleHash dh = hash2(keyHash); // Save the first removed entry pointer so we can recycle later. Entry *firstRemoved = NULL; while(true) { if (JS_UNLIKELY(entry->isRemoved())) { if (!firstRemoved) firstRemoved = entry; } else { entry->setCollision(collisionBit); } METER(stats.steps++); h1 = applyDoubleHash(h1, dh); entry = &table[h1]; if (entry->isFree()) { METER(stats.misses++); return firstRemoved ? *firstRemoved : *entry; } if (entry->matchHash(keyHash) && match(*entry, l)) { METER(stats.hits++); return *entry; } } } // This is a copy of lookup hardcoded to the assumptions: // 1. the lookup is a lookupForAdd // 2. the key, whose |keyHash| has been passed is not in the table, // 3. no entries have been removed from the table. // This specialized search avoids the need for recovering lookup values // from entries, which allows more flexible Lookup/Key types. Entry &findFreeEntry(HashNumber keyHash) { JS_ASSERT(!(keyHash & sCollisionBit)); JS_ASSERT(table); METER(stats.searches++); // We assume 'keyHash' has already been distributed. // Compute the primary hash address. HashNumber h1 = hash1(keyHash); Entry *entry = &table[h1]; // Miss: return space for a new entry. if (!entry->isLive()) { METER(stats.misses++); return *entry; } // Collision: double hash. DoubleHash dh = hash2(keyHash); while(true) { JS_ASSERT(!entry->isRemoved()); entry->setCollision(); METER(stats.steps++); h1 = applyDoubleHash(h1, dh); entry = &table[h1]; if (!entry->isLive()) { METER(stats.misses++); return *entry; } } } enum RebuildStatus { NotOverloaded, Rehashed, RehashFailed }; RebuildStatus changeTableSize(int deltaLog2) { // Look, but don't touch, until we succeed in getting new entry store. Entry *oldTable = table; uint32_t oldCap = capacity(); uint32_t newLog2 = sHashBits - hashShift + deltaLog2; uint32_t newCapacity = JS_BIT(newLog2); if (newCapacity > sMaxCapacity) { this->reportAllocOverflow(); return RehashFailed; } Entry *newTable = createTable(*this, newCapacity); if (!newTable) return RehashFailed; // We can't fail from here on, so update table parameters. setTableSizeLog2(newLog2); removedCount = 0; gen++; table = newTable; // Copy only live entries, leaving removed ones behind. for (Entry *src = oldTable, *end = src + oldCap; src < end; ++src) { if (src->isLive()) { HashNumber hn = src->getKeyHash(); findFreeEntry(hn).setLive(hn, Move(src->get())); src->destroy(); } } // All entries have been destroyed, no need to destroyTable. this->free_(oldTable); return Rehashed; } RebuildStatus checkOverloaded() { if (!overloaded()) return NotOverloaded; // Compress if a quarter or more of all entries are removed. int deltaLog2; if (removedCount >= (capacity() >> 2)) { METER(stats.compresses++); deltaLog2 = 0; } else { METER(stats.grows++); deltaLog2 = 1; } return changeTableSize(deltaLog2); } // Infallibly rehash the table if we are overloaded with removals. void checkOverRemoved() { if (overloaded()) { METER(stats.rehashes++); rehashTable(); JS_ASSERT(!overloaded()); } } void remove(Entry &e) { JS_ASSERT(table); METER(stats.removes++); if (e.hasCollision()) { e.removeLive(); removedCount++; } else { METER(stats.removeFrees++); e.clearLive(); } entryCount--; mutationCount++; } void checkUnderloaded() { if (underloaded()) { METER(stats.shrinks++); (void) changeTableSize(-1); } } // This is identical to changeTableSize(currentSize), but without requiring // a second table. We do this by recycling the collision bits to tell us if // the element is already inserted or still waiting to be inserted. Since // already-inserted elements win any conflicts, we get the same table as we // would have gotten through random insertion order. void rehashTable() { removedCount = 0; for (size_t i = 0; i < capacity(); ++i) table[i].unsetCollision(); for (size_t i = 0; i < capacity();) { Entry *src = &table[i]; if (!src->isLive() || src->hasCollision()) { ++i; continue; } HashNumber keyHash = src->getKeyHash(); HashNumber h1 = hash1(keyHash); DoubleHash dh = hash2(keyHash); Entry *tgt = &table[h1]; while (true) { if (!tgt->hasCollision()) { src->swap(tgt); tgt->setCollision(); break; } h1 = applyDoubleHash(h1, dh); tgt = &table[h1]; } } // TODO: this algorithm leaves collision bits on *all* elements, even if // they are on no collision path. We have the option of setting the // collision bits correctly on a subsequent pass or skipping the rehash // unless we are totally filled with tombstones: benchmark to find out // which approach is best. } public: void clear() { if (tl::IsPodType::result) { memset(table, 0, sizeof(*table) * capacity()); } else { uint32_t tableCapacity = capacity(); for (Entry *e = table, *end = table + tableCapacity; e < end; ++e) e->clear(); } removedCount = 0; entryCount = 0; mutationCount++; } void finish() { JS_ASSERT(!entered); if (!table) return; destroyTable(*this, table, capacity()); table = NULL; gen++; entryCount = 0; removedCount = 0; mutationCount++; } Range all() const { JS_ASSERT(table); return Range(table, table + capacity()); } bool empty() const { JS_ASSERT(table); return !entryCount; } uint32_t count() const { JS_ASSERT(table); return entryCount; } uint32_t capacity() const { JS_ASSERT(table); return JS_BIT(sHashBits - hashShift); } uint32_t generation() const { JS_ASSERT(table); return gen; } size_t sizeOfExcludingThis(JSMallocSizeOfFun mallocSizeOf) const { return mallocSizeOf(table); } size_t sizeOfIncludingThis(JSMallocSizeOfFun mallocSizeOf) const { return mallocSizeOf(this) + sizeOfExcludingThis(mallocSizeOf); } Ptr lookup(const Lookup &l) const { ReentrancyGuard g(*this); HashNumber keyHash = prepareHash(l); return Ptr(lookup(l, keyHash, 0)); } AddPtr lookupForAdd(const Lookup &l) const { ReentrancyGuard g(*this); HashNumber keyHash = prepareHash(l); Entry &entry = lookup(l, keyHash, sCollisionBit); AddPtr p(entry, keyHash); p.mutationCount = mutationCount; return p; } template bool add(AddPtr &p, const U &rhs) { ReentrancyGuard g(*this); JS_ASSERT(mutationCount == p.mutationCount); JS_ASSERT(table); JS_ASSERT(!p.found()); JS_ASSERT(!(p.keyHash & sCollisionBit)); // Changing an entry from removed to live does not affect whether we // are overloaded and can be handled separately. if (p.entry->isRemoved()) { METER(stats.addOverRemoved++); removedCount--; p.keyHash |= sCollisionBit; } else { // Preserve the validity of |p.entry|. RebuildStatus status = checkOverloaded(); if (status == RehashFailed) return false; if (status == Rehashed) p.entry = &findFreeEntry(p.keyHash); } p.entry->setLive(p.keyHash, rhs); entryCount++; mutationCount++; return true; } template void putNewInfallible(const Lookup &l, const U &u) { JS_ASSERT(table); HashNumber keyHash = prepareHash(l); Entry *entry = &findFreeEntry(keyHash); if (entry->isRemoved()) { METER(stats.addOverRemoved++); removedCount--; keyHash |= sCollisionBit; } entry->setLive(keyHash, u); entryCount++; mutationCount++; } template bool putNew(const Lookup &l, const U &u) { if (checkOverloaded() == RehashFailed) return false; putNewInfallible(l, u); return true; } template bool relookupOrAdd(AddPtr& p, const Lookup &l, const U &u) { p.mutationCount = mutationCount; { ReentrancyGuard g(*this); p.entry = &lookup(l, p.keyHash, sCollisionBit); } return p.found() || add(p, u); } void remove(Ptr p) { JS_ASSERT(table); ReentrancyGuard g(*this); JS_ASSERT(p.found()); remove(*p.entry); checkUnderloaded(); } #undef METER }; } // namespace detail } // namespace js #endif // js_HashTable_h__