//========= Copyright Valve Corporation, All rights reserved. ============// // // Purpose: a fast growable hashtable with stored hashes, L2-friendly behavior. // Useful as a string dictionary or a low-overhead set/map for small POD types. // // Usage notes: // - handles are NOT STABLE across element removal! use RemoveAndAdvance() // if you are removing elements while iterating through the hashtable. // Use CUtlStableHashtable if you need stable handles (less efficient). // - handles are also NOT STABLE across element insertion. The handle // resulting from the insertion of an element may not retreive the // same (or any!) element after further insertions. Again, use // CUtlStableHashtable if you need stable handles // - Insert() first searches for an existing match and returns it if found // - a value type of "empty_t" can be used to eliminate value storage and // switch Element() to return const Key references instead of values // - an extra user flag bit is accessible via Get/SetUserFlag() // - hash function pointer / functor is exposed via GetHashRef() // - comparison function pointer / functor is exposed via GetEqualRef() // - if your value type cannot be copy-constructed, use key-only Insert() // to default-initialize the value and then manipulate it afterwards. // - The reason that UtlHashtable permutes itself and invalidates // iterators is to make it faster in the case where you are not // tracking iterators. If you use it as a set or a map ("is this // value a member?") as opposed to a long-term container, then you // probably don't need stable iterators. Hashtable tries to place // newly inserted data in the primary hash slot, making an // assumption that if you inserted it recently, you're more likely // to access it than if you inserted something a long time // ago. It's effectively trying to minimize cache misses for hot // data if you add and remove a lot. // If you don't care too much about cache misses, UtlStableHashtable // is what you're looking for // // Implementation notes: // - overall hash table load is kept between .25 and .75 // - items which would map to the same ideal slot are chained together // - chained items are stored sequentially in adjacent free spaces // - "root" entries are prioritized over chained entries; if a // slot is not occupied by an item in its root position, the table // is guaranteed to contain no keys which would hash to that slot. // - new items go at the head of the chain (ie, in their root slot) // and evict / "bump" any chained entries which occupy that slot // - chain-following skips over unused holes and continues examining // table entries until a chain entry with FLAG_LAST is encountered // // CUtlHashtable< uint32 > setOfIntegers; // CUtlHashtable< const char* > setOfStringPointers; // CUtlHashtable< int, CUtlVector > mapFromIntsToArrays; // // $NoKeywords: $ // // A closed-form (open addressing) hashtable with linear sequential probing. //=============================================================================// #ifndef UTLHASHTABLE_H #define UTLHASHTABLE_H #pragma once #include "utlcommon.h" #include "utlmemory.h" #include "mathlib/mathlib.h" #include "utllinkedlist.h" //----------------------------------------------------------------------------- // Henry Goffin (henryg) was here. Questions? Bugs? Go slap him around a bit. //----------------------------------------------------------------------------- typedef unsigned int UtlHashHandle_t; #define FOR_EACH_HASHTABLE( table, iter ) \ for ( UtlHashHandle_t iter = (table).FirstHandle(); iter != (table).InvalidHandle(); iter = (table).NextHandle( iter ) ) // CUtlHashtableEntry selects between 16 and 32 bit storage backing // for flags_and_hash depending on the size of the stored types. template < typename KeyT, typename ValueT = empty_t > class CUtlHashtableEntry { public: typedef CUtlKeyValuePair< KeyT, ValueT > KVPair; enum { INT16_STORAGE = ( sizeof( KVPair ) <= 2 ) }; typedef typename CTypeSelect< INT16_STORAGE, int16, int32 >::type storage_t; enum { FLAG_FREE = INT16_STORAGE ? 0x8000 : 0x80000000, // must be high bit for IsValid and IdealIndex to work FLAG_LAST = INT16_STORAGE ? 0x4000 : 0x40000000, MASK_HASH = INT16_STORAGE ? 0x3FFF : 0x3FFFFFFF }; storage_t flags_and_hash; storage_t data[ ( sizeof(KVPair) + sizeof(storage_t) - 1 ) / sizeof(storage_t) ]; bool IsValid() const { return flags_and_hash >= 0; } void MarkInvalid() { int32 flag = FLAG_FREE; flags_and_hash = (storage_t)flag; } const KVPair *Raw() const { return reinterpret_cast< const KVPair * >( &data[0] ); } const KVPair *operator->() const { Assert( IsValid() ); return reinterpret_cast< const KVPair * >( &data[0] ); } KVPair *Raw() { return reinterpret_cast< KVPair * >( &data[0] ); } KVPair *operator->() { Assert( IsValid() ); return reinterpret_cast< KVPair * >( &data[0] ); } // Returns the ideal index of the data in this slot, or all bits set if invalid uint32 FORCEINLINE IdealIndex( uint32 slotmask ) const { return IdealIndex( flags_and_hash, slotmask ) | ( (int32)flags_and_hash >> 31 ); } // Use template tricks to fully define only one function that takes either 16 or 32 bits // and performs different logic without using "if ( INT16_STORAGE )", because GCC and MSVC // sometimes have trouble removing the constant branch, which is dumb... but whatever. // 16-bit hashes are simply too narrow for large hashtables; more mask bits than hash bits! // So we duplicate the hash bits. (Note: h *= MASK_HASH+2 is the same as h += h<::type uint32_if16BitStorage; typedef typename CTypeSelect< INT16_STORAGE, undefined_t, int32 >::type uint32_if32BitStorage; static FORCEINLINE uint32 IdealIndex( uint32_if16BitStorage h, uint32 m ) { h &= MASK_HASH; h *= MASK_HASH + 2; return h & m; } static FORCEINLINE uint32 IdealIndex( uint32_if32BitStorage h, uint32 m ) { return h & m; } // More efficient than memcpy for the small types that are stored in a hashtable void MoveDataFrom( CUtlHashtableEntry &src ) { storage_t * RESTRICT srcData = &src.data[0]; for ( int i = 0; i < ARRAYSIZE( data ); ++i ) { data[i] = srcData[i]; } } }; template , typename KeyIsEqualT = DefaultEqualFunctor, typename AlternateKeyT = typename ArgumentTypeInfo::Alt_t > class CUtlHashtable { public: typedef UtlHashHandle_t handle_t; protected: typedef CUtlKeyValuePair KVPair; typedef typename ArgumentTypeInfo::Arg_t KeyArg_t; typedef typename ArgumentTypeInfo::Arg_t ValueArg_t; typedef typename ArgumentTypeInfo::Arg_t KeyAlt_t; typedef CUtlHashtableEntry< KeyT, ValueT > entry_t; enum { FLAG_FREE = entry_t::FLAG_FREE }; enum { FLAG_LAST = entry_t::FLAG_LAST }; enum { MASK_HASH = entry_t::MASK_HASH }; CUtlMemory< entry_t > m_table; int m_nUsed; int m_nMinSize; bool m_bSizeLocked; KeyIsEqualT m_eq; KeyHashT m_hash; // Allocate an empty table and then re-insert all existing entries. void DoRealloc( int size ); // Move an existing entry to a free slot, leaving a hole behind void BumpEntry( unsigned int idx ); // Insert an unconstructed KVPair at the primary slot int DoInsertUnconstructed( unsigned int h, bool allowGrow ); // Implementation for Insert functions, constructs a KVPair // with either a default-construted or copy-constructed value template handle_t DoInsert( KeyParamT k, unsigned int h ); template handle_t DoInsert( KeyParamT k, typename ArgumentTypeInfo::Arg_t v, unsigned int h, bool* pDidInsert ); template handle_t DoInsertNoCheck( KeyParamT k, typename ArgumentTypeInfo::Arg_t v, unsigned int h ); // Key lookup. Can also return previous-in-chain if result is chained. template handle_t DoLookup( KeyParamT x, unsigned int h, handle_t *pPreviousInChain ) const; // Remove single element by key + hash. Returns the index of the new hole // that was created. Returns InvalidHandle() if element was not found. template int DoRemove( KeyParamT x, unsigned int h ); // Friend CUtlStableHashtable so that it can call our Do* functions directly template < typename K, typename V, typename S, typename H, typename E, typename A > friend class CUtlStableHashtable; public: explicit CUtlHashtable( int minimumSize = 32 ) : m_nUsed(0), m_nMinSize(MAX(8, minimumSize)), m_bSizeLocked(false), m_eq(), m_hash() { } CUtlHashtable( int minimumSize, const KeyHashT &hash, KeyIsEqualT const &eq = KeyIsEqualT() ) : m_nUsed(0), m_nMinSize(MAX(8, minimumSize)), m_bSizeLocked(false), m_eq(eq), m_hash(hash) { } CUtlHashtable( entry_t* pMemory, unsigned int nCount, const KeyHashT &hash = KeyHashT(), KeyIsEqualT const &eq = KeyIsEqualT() ) : m_nUsed(0), m_nMinSize(8), m_bSizeLocked(false), m_eq(eq), m_hash(hash) { SetExternalBuffer( pMemory, nCount ); } ~CUtlHashtable() { RemoveAll(); } CUtlHashtable &operator=( CUtlHashtable const &src ); // Set external memory void SetExternalBuffer( byte* pRawBuffer, unsigned int nBytes, bool bAssumeOwnership = false, bool bGrowable = false ); void SetExternalBuffer( entry_t* pBuffer, unsigned int nSize, bool bAssumeOwnership = false, bool bGrowable = false ); // Functor/function-pointer access KeyHashT& GetHashRef() { return m_hash; } KeyIsEqualT& GetEqualRef() { return m_eq; } KeyHashT const &GetHashRef() const { return m_hash; } KeyIsEqualT const &GetEqualRef() const { return m_eq; } // Handle validation bool IsValidHandle( handle_t idx ) const { return (unsigned)idx < (unsigned)m_table.Count() && m_table[idx].IsValid(); } static handle_t InvalidHandle() { return (handle_t) -1; } // Iteration functions handle_t FirstHandle() const { return NextHandle( (handle_t) -1 ); } handle_t NextHandle( handle_t start ) const; // Returns the number of unique keys in the table int Count() const { return m_nUsed; } // Key lookup, returns InvalidHandle() if not found handle_t Find( KeyArg_t k ) const { return DoLookup( k, m_hash(k), NULL ); } handle_t Find( KeyArg_t k, unsigned int hash) const { Assert( hash == m_hash(k) ); return DoLookup( k, hash, NULL ); } // Alternate-type key lookup, returns InvalidHandle() if not found handle_t Find( KeyAlt_t k ) const { return DoLookup( k, m_hash(k), NULL ); } handle_t Find( KeyAlt_t k, unsigned int hash) const { Assert( hash == m_hash(k) ); return DoLookup( k, hash, NULL ); } // True if the key is in the table bool HasElement( KeyArg_t k ) const { return InvalidHandle() != Find( k ); } bool HasElement( KeyAlt_t k ) const { return InvalidHandle() != Find( k ); } // Key insertion or lookup, always returns a valid handle handle_t Insert( KeyArg_t k ) { return DoInsert( k, m_hash(k) ); } handle_t Insert( KeyArg_t k, ValueArg_t v, bool *pDidInsert = NULL ) { return DoInsert( k, v, m_hash(k), pDidInsert ); } handle_t Insert( KeyArg_t k, ValueArg_t v, unsigned int hash, bool *pDidInsert = NULL ) { Assert( hash == m_hash(k) ); return DoInsert( k, v, hash, pDidInsert ); } // Alternate-type key insertion or lookup, always returns a valid handle handle_t Insert( KeyAlt_t k ) { return DoInsert( k, m_hash(k) ); } handle_t Insert( KeyAlt_t k, ValueArg_t v, bool *pDidInsert = NULL ) { return DoInsert( k, v, m_hash(k), pDidInsert ); } handle_t Insert( KeyAlt_t k, ValueArg_t v, unsigned int hash, bool *pDidInsert = NULL ) { Assert( hash == m_hash(k) ); return DoInsert( k, v, hash, pDidInsert ); } // Key removal, returns false if not found bool Remove( KeyArg_t k ) { return DoRemove( k, m_hash(k) ) >= 0; } bool Remove( KeyArg_t k, unsigned int hash ) { Assert( hash == m_hash(k) ); return DoRemove( k, hash ) >= 0; } // Alternate-type key removal, returns false if not found bool Remove( KeyAlt_t k ) { return DoRemove( k, m_hash(k) ) >= 0; } bool Remove( KeyAlt_t k, unsigned int hash ) { Assert( hash == m_hash(k) ); return DoRemove( k, hash ) >= 0; } // Remove while iterating, returns the next handle for forward iteration // Note: aside from this, ALL handles are invalid if an element is removed handle_t RemoveAndAdvance( handle_t idx ); // Remove by handle, convenient when you look up a handle and do something with it before removing the element void RemoveByHandle( handle_t idx ); // Nuke contents void RemoveAll(); // Nuke and release memory. void Purge() { RemoveAll(); m_table.Purge(); } // Reserve table capacity up front to avoid reallocation during insertions void Reserve( int expected ) { if ( expected > m_nUsed ) DoRealloc( expected * 4 / 3 ); } // Shrink to best-fit size, re-insert keys for optimal lookup void Compact( bool bMinimal ) { DoRealloc( bMinimal ? m_nUsed : ( m_nUsed * 4 / 3 ) ); } // Access functions. Note: if ValueT is empty_t, all functions return const keys. typedef typename KVPair::ValueReturn_t Element_t; KeyT const &Key( handle_t idx ) const { return m_table[idx]->m_key; } Element_t const &Element( handle_t idx ) const { return m_table[idx]->GetValue(); } Element_t &Element(handle_t idx) { return m_table[idx]->GetValue(); } Element_t const &operator[]( handle_t idx ) const { return m_table[idx]->GetValue(); } Element_t &operator[]( handle_t idx ) { return m_table[idx]->GetValue(); } void ReplaceKey( handle_t idx, KeyArg_t k ) { Assert( m_eq( m_table[idx]->m_key, k ) && m_hash( k ) == m_hash( m_table[idx]->m_key ) ); m_table[idx]->m_key = k; } void ReplaceKey( handle_t idx, KeyAlt_t k ) { Assert( m_eq( m_table[idx]->m_key, k ) && m_hash( k ) == m_hash( m_table[idx]->m_key ) ); m_table[idx]->m_key = k; } Element_t const &Get( KeyArg_t k, Element_t const &defaultValue ) const { handle_t h = Find( k ); if ( h != InvalidHandle() ) return Element( h ); return defaultValue; } Element_t const &Get( KeyAlt_t k, Element_t const &defaultValue ) const { handle_t h = Find( k ); if ( h != InvalidHandle() ) return Element( h ); return defaultValue; } Element_t const *GetPtr( KeyArg_t k ) const { handle_t h = Find(k); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } Element_t const *GetPtr( KeyAlt_t k ) const { handle_t h = Find(k); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } Element_t *GetPtr( KeyArg_t k ) { handle_t h = Find( k ); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } Element_t *GetPtr( KeyAlt_t k ) { handle_t h = Find( k ); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } // Swap memory and contents with another identical hashtable // (NOTE: if using function pointers or functors with state, // it is up to the caller to ensure that they are compatible!) void Swap( CUtlHashtable &other ) { m_table.Swap(other.m_table); ::V_swap(m_nUsed, other.m_nUsed); } // GetMemoryUsage returns all memory held by this class // and its held classes. It does not include sizeof(*this). size_t GetMemoryUsage() const { return m_table.AllocSize(); } size_t GetReserveCount( )const { return m_table.Count(); } #if _DEBUG // Validate the integrity of the hashtable void DbgCheckIntegrity() const; #endif private: CUtlHashtable(const CUtlHashtable& copyConstructorIsNotImplemented); }; // Set external memory (raw byte buffer, best-fit) template void CUtlHashtable::SetExternalBuffer( byte* pRawBuffer, unsigned int nBytes, bool bAssumeOwnership, bool bGrowable ) { Assert( ((uintptr_t)pRawBuffer % __alignof(int)) == 0 ); uint32 bestSize = LargestPowerOfTwoLessThanOrEqual( nBytes / sizeof(entry_t) ); Assert( bestSize != 0 && bestSize*sizeof(entry_t) <= nBytes ); return SetExternalBuffer( (entry_t*) pRawBuffer, bestSize, bAssumeOwnership, bGrowable ); } // Set external memory (typechecked, must be power of two) template void CUtlHashtable::SetExternalBuffer( entry_t* pBuffer, unsigned int nSize, bool bAssumeOwnership, bool bGrowable ) { Assert( IsPowerOfTwo(nSize) ); Assert( m_nUsed == 0 ); for ( uint i = 0; i < nSize; ++i ) pBuffer[i].MarkInvalid(); if ( bAssumeOwnership ) m_table.AssumeMemory( pBuffer, nSize ); else m_table.SetExternalBuffer( pBuffer, nSize ); m_bSizeLocked = !bGrowable; } // Allocate an empty table and then re-insert all existing entries. template void CUtlHashtable::DoRealloc( int size ) { Assert( !m_bSizeLocked ); size = SmallestPowerOfTwoGreaterOrEqual( MAX( m_nMinSize, size ) ); Assert( size > 0 && (uint)size <= entry_t::IdealIndex( ~0, 0x1FFFFFFF ) ); // reasonable power of 2 Assert( size > m_nUsed ); CUtlMemory oldTable; oldTable.Swap( m_table ); entry_t * RESTRICT const pOldBase = oldTable.Base(); m_table.EnsureCapacity( size ); entry_t * const pNewBase = m_table.Base(); for ( int i = 0; i < size; ++i ) pNewBase[i].MarkInvalid(); int nLeftToMove = m_nUsed; m_nUsed = 0; for ( int i = oldTable.Count() - 1; i >= 0; --i ) { if ( pOldBase[i].IsValid() ) { int newIdx = DoInsertUnconstructed( pOldBase[i].flags_and_hash, false ); pNewBase[newIdx].MoveDataFrom( pOldBase[i] ); if ( --nLeftToMove == 0 ) break; } } Assert( nLeftToMove == 0 ); } // Move an existing entry to a free slot, leaving a hole behind template void CUtlHashtable::BumpEntry( unsigned int idx ) { Assert( m_table[idx].IsValid() ); Assert( m_nUsed < m_table.Count() ); entry_t* table = m_table.Base(); unsigned int slotmask = m_table.Count()-1; unsigned int new_flags_and_hash = table[idx].flags_and_hash & (FLAG_LAST | MASK_HASH); unsigned int chainid = entry_t::IdealIndex( new_flags_and_hash, slotmask ); // Look for empty slots scanning forward, stripping FLAG_LAST as we go. // Note: this potentially strips FLAG_LAST from table[idx] if we pass it int newIdx = chainid; // start at ideal slot for ( ; ; newIdx = (newIdx + 1) & slotmask ) { if ( table[newIdx].IdealIndex( slotmask ) == chainid ) { if ( table[newIdx].flags_and_hash & FLAG_LAST ) { table[newIdx].flags_and_hash &= ~FLAG_LAST; new_flags_and_hash |= FLAG_LAST; } continue; } if ( table[newIdx].IsValid() ) { continue; } break; } // Did we pick something closer to the ideal slot, leaving behind a // FLAG_LAST bit on the current slot because we didn't scan past it? if ( table[idx].flags_and_hash & FLAG_LAST ) { #ifdef _DEBUG Assert( new_flags_and_hash & FLAG_LAST ); // Verify logic: we must have moved to an earlier slot, right? uint offset = ((uint)idx - chainid + slotmask + 1) & slotmask; uint newOffset = ((uint)newIdx - chainid + slotmask + 1) & slotmask; Assert( newOffset < offset ); #endif // Scan backwards from old to new location, depositing FLAG_LAST on // the first match we find. (+slotmask) is the same as (-1) without // having to make anyone think about two's complement shenanigans. int scan = (idx + slotmask) & slotmask; while ( scan != newIdx ) { if ( table[scan].IdealIndex( slotmask ) == chainid ) { table[scan].flags_and_hash |= FLAG_LAST; new_flags_and_hash &= ~FLAG_LAST; break; } scan = (scan + slotmask) & slotmask; } } // Move entry to the free slot we found, leaving a hole at idx table[newIdx].flags_and_hash = new_flags_and_hash; table[newIdx].MoveDataFrom( table[idx] ); table[idx].MarkInvalid(); } // Insert a value at the root position for that value's hash chain. template int CUtlHashtable::DoInsertUnconstructed( unsigned int h, bool allowGrow ) { if ( allowGrow && !m_bSizeLocked ) { // Keep the load factor between .25 and .75 int newSize = m_nUsed + 1; if ( ( newSize*4 < m_table.Count() && m_table.Count() > m_nMinSize*2 ) || newSize*4 > m_table.Count()*3 ) { DoRealloc( newSize * 4 / 3 ); } } Assert( m_nUsed < m_table.Count() ); ++m_nUsed; entry_t* table = m_table.Base(); unsigned int slotmask = m_table.Count()-1; unsigned int new_flags_and_hash = FLAG_LAST | (h & MASK_HASH); unsigned int idx = entry_t::IdealIndex( h, slotmask ); if ( table[idx].IdealIndex( slotmask ) == idx ) { // There is already an entry in this chain. new_flags_and_hash &= ~FLAG_LAST; BumpEntry(idx); } else if ( table[idx].IsValid() ) { // Somebody else is living in our ideal index but does not belong // to our entry chain; move it out of the way, start a new chain. BumpEntry(idx); } table[idx].flags_and_hash = new_flags_and_hash; return idx; } // Key lookup. Can also return previous-in-chain if result is a chained slot. template template UtlHashHandle_t CUtlHashtable::DoLookup( KeyParamT x, unsigned int h, handle_t *pPreviousInChain ) const { if ( m_nUsed == 0 ) { // Empty table. return (handle_t) -1; } const entry_t* table = m_table.Base(); unsigned int slotmask = m_table.Count()-1; Assert( m_table.Count() > 0 && (slotmask & m_table.Count()) == 0 ); unsigned int chainid = entry_t::IdealIndex( h, slotmask ); unsigned int idx = chainid; if ( table[idx].IdealIndex( slotmask ) != chainid ) { // Nothing in root position? No match. return (handle_t) -1; } // Linear scan until found or end of chain handle_t lastIdx = (handle_t) -1; while (1) { // Only examine this slot if it is valid and belongs to our hash chain if ( table[idx].IdealIndex( slotmask ) == chainid ) { // Test the full-width hash to avoid unnecessary calls to m_eq() if ( ((table[idx].flags_and_hash ^ h) & MASK_HASH) == 0 && m_eq( table[idx]->m_key, x ) ) { // Found match! if (pPreviousInChain) *pPreviousInChain = lastIdx; return (handle_t) idx; } if ( table[idx].flags_and_hash & FLAG_LAST ) { // End of chain. No match. return (handle_t) -1; } lastIdx = (handle_t) idx; } idx = (idx + 1) & slotmask; } } // Key insertion, or return index of existing key if found template template UtlHashHandle_t CUtlHashtable::DoInsert( KeyParamT k, unsigned int h ) { handle_t idx = DoLookup( k, h, NULL ); if ( idx == (handle_t) -1 ) { idx = (handle_t) DoInsertUnconstructed( h, true ); ConstructOneArg( m_table[ idx ].Raw(), k ); } return idx; } // Key insertion, or return index of existing key if found template template UtlHashHandle_t CUtlHashtable::DoInsert( KeyParamT k, typename ArgumentTypeInfo::Arg_t v, unsigned int h, bool *pDidInsert ) { handle_t idx = DoLookup( k, h, NULL ); if ( idx == (handle_t) -1 ) { idx = (handle_t) DoInsertUnconstructed( h, true ); ConstructTwoArg( m_table[ idx ].Raw(), k, v ); if ( pDidInsert ) *pDidInsert = true; } else { if ( pDidInsert ) *pDidInsert = false; } return idx; } // Key insertion template template UtlHashHandle_t CUtlHashtable::DoInsertNoCheck( KeyParamT k, typename ArgumentTypeInfo::Arg_t v, unsigned int h ) { Assert( DoLookup( k, h, NULL ) == (handle_t) -1 ); handle_t idx = (handle_t) DoInsertUnconstructed( h, true ); ConstructTwoArg( m_table[ idx ].Raw(), k, v ); return idx; } // Remove single element by key + hash. Returns the location of the new empty hole. template template int CUtlHashtable::DoRemove( KeyParamT x, unsigned int h ) { unsigned int slotmask = m_table.Count()-1; handle_t previous = (handle_t) -1; int idx = (int) DoLookup( x, h, &previous ); if (idx == -1) { return -1; } enum { FAKEFLAG_ROOT = 1 }; int nLastAndRootFlags = m_table[idx].flags_and_hash & FLAG_LAST; nLastAndRootFlags |= ( (uint)idx == m_table[idx].IdealIndex( slotmask ) ); // Remove from table m_table[idx].MarkInvalid(); Destruct( m_table[idx].Raw() ); --m_nUsed; if ( nLastAndRootFlags == FLAG_LAST ) // last only, not root { // This was the end of the chain - mark previous as last. // (This isn't the root, so there must be a previous.) Assert( previous != (handle_t) -1 ); m_table[previous].flags_and_hash |= FLAG_LAST; } if ( nLastAndRootFlags == FAKEFLAG_ROOT ) // root only, not last { // If we are removing the root and there is more to the chain, // scan to find the next chain entry and move it to the root. unsigned int chainid = entry_t::IdealIndex( h, slotmask ); unsigned int nextIdx = idx; while (1) { nextIdx = (nextIdx + 1) & slotmask; if ( m_table[nextIdx].IdealIndex( slotmask ) == chainid ) { break; } } Assert( !(m_table[nextIdx].flags_and_hash & FLAG_FREE) ); // Leave a hole where the next entry in the chain was. m_table[idx].flags_and_hash = m_table[nextIdx].flags_and_hash; m_table[idx].MoveDataFrom( m_table[nextIdx] ); m_table[nextIdx].MarkInvalid(); return nextIdx; } // The hole is still where the element used to be. return idx; } // Assignment operator. It's up to the user to make sure that the hash and equality functors match. template CUtlHashtable &CUtlHashtable::operator=( CUtlHashtable const &src ) { if ( &src != this ) { Assert( !m_bSizeLocked || m_table.Count() >= src.m_nUsed ); if ( !m_bSizeLocked ) { Purge(); Reserve(src.m_nUsed); } else { RemoveAll(); } const entry_t * srcTable = src.m_table.Base(); for ( int i = src.m_table.Count() - 1; i >= 0; --i ) { if ( srcTable[i].IsValid() ) { // If this assert trips, double-check that both hashtables // have the same hash function pointers or hash functor state! Assert( m_hash(srcTable[i]->m_key) == src.m_hash(srcTable[i]->m_key) ); int newIdx = DoInsertUnconstructed( srcTable[i].flags_and_hash , false ); CopyConstruct( m_table[newIdx].Raw(), *srcTable[i].Raw() ); // copy construct KVPair } } } return *this; } // Remove and return the next valid iterator for a forward iteration. template UtlHashHandle_t CUtlHashtable::RemoveAndAdvance( UtlHashHandle_t idx ) { Assert( IsValidHandle( idx ) ); // TODO optimize, implement DoRemoveAt that does not need to re-evaluate equality in DoLookup int hole = DoRemove< KeyArg_t >( m_table[idx]->m_key, m_table[idx].flags_and_hash & MASK_HASH ); // DoRemove returns the index of the element that it moved to fill the hole, if any. if ( hole <= (int) idx ) { // Didn't fill, or filled from a previously seen element. return NextHandle( idx ); } else { // Do not advance; slot has a new un-iterated value. Assert( IsValidHandle(idx) ); return idx; } } // Remove and return the next valid iterator for a forward iteration. template void CUtlHashtable::RemoveByHandle( UtlHashHandle_t idx ) { Assert( IsValidHandle( idx ) ); // Copied from RemoveAndAdvance(): TODO optimize, implement DoRemoveAt that does not need to re-evaluate equality in DoLookup DoRemove< KeyArg_t >( m_table[idx]->m_key, m_table[idx].flags_and_hash & MASK_HASH ); } // Burn it with fire. template void CUtlHashtable::RemoveAll() { int used = m_nUsed; if ( used != 0 ) { entry_t* table = m_table.Base(); for ( int i = m_table.Count() - 1; i >= 0; --i ) { if ( table[i].IsValid() ) { table[i].MarkInvalid(); Destruct( table[i].Raw() ); if ( --used == 0 ) break; } } m_nUsed = 0; } } template UtlHashHandle_t CUtlHashtable::NextHandle( handle_t start ) const { const entry_t *table = m_table.Base(); for ( int i = (int)start + 1; i < m_table.Count(); ++i ) { if ( table[i].IsValid() ) return (handle_t) i; } return (handle_t) -1; } #if _DEBUG template void CUtlHashtable::DbgCheckIntegrity() const { // Stress test the hash table as a test of both container functionality // and also the validity of the user's Hash and Equal function objects. // NOTE: will fail if function objects require any sort of state! CUtlHashtable clone; unsigned int bytes = sizeof(entry_t)*max(16,m_table.Count()); byte* tempbuf = (byte*) malloc(bytes); clone.SetExternalBuffer( tempbuf, bytes, false, false ); clone = *this; int count = 0, roots = 0, ends = 0; int slotmask = m_table.Count() - 1; for (int i = 0; i < m_table.Count(); ++i) { if (!(m_table[i].flags_and_hash & FLAG_FREE)) ++count; if (m_table[i].IdealIndex(slotmask) == (uint)i) ++roots; if (m_table[i].flags_and_hash & FLAG_LAST) ++ends; if (m_table[i].IsValid()) { Assert( Find(m_table[i]->m_key) == (handle_t)i ); Verify( clone.Remove(m_table[i]->m_key) ); } else { Assert( m_table[i].flags_and_hash == FLAG_FREE ); } } Assert( count == Count() && count >= roots && roots == ends ); Assert( clone.Count() == 0 ); clone.Purge(); free(tempbuf); } #endif //----------------------------------------------------------------------- // CUtlStableHashtable //----------------------------------------------------------------------- // Stable hashtables are less memory and cache efficient, but can be // iterated quickly and their element handles are completely stable. // Implemented as a hashtable which only stores indices, and a separate // CUtlLinkedList data table which contains key-value pairs; this may // change to a more efficient structure in the future if space becomes // critical. I have some ideas about that but not the time to implement // at the moment. -henryg // Note: RemoveAndAdvance is slower than in CUtlHashtable because the // key needs to be re-hashed under the current implementation. template , typename KeyIsEqualT = DefaultEqualFunctor, typename IndexStorageT = uint16, typename AlternateKeyT = typename ArgumentTypeInfo::Alt_t > class CUtlStableHashtable { public: typedef typename ArgumentTypeInfo::Arg_t KeyArg_t; typedef typename ArgumentTypeInfo::Arg_t ValueArg_t; typedef typename ArgumentTypeInfo::Arg_t KeyAlt_t; typedef typename CTypeSelect< sizeof( IndexStorageT ) == 2, uint16, uint32 >::type IndexStorage_t; protected: COMPILE_TIME_ASSERT( sizeof( IndexStorage_t ) == sizeof( IndexStorageT ) ); typedef CUtlKeyValuePair< KeyT, ValueT > KVPair; struct HashProxy; struct EqualProxy; struct IndirectIndex; typedef CUtlHashtable< IndirectIndex, empty_t, HashProxy, EqualProxy, AlternateKeyT > Hashtable_t; typedef CUtlLinkedList< KVPair, IndexStorage_t > LinkedList_t; template bool DoRemove( KeyArgumentT k ); template UtlHashHandle_t DoFind( KeyArgumentT k ) const; template UtlHashHandle_t DoInsert( KeyArgumentT k ); template UtlHashHandle_t DoInsert( KeyArgumentT k, ValueArgumentT v ); public: KeyHashT &GetHashRef() { return m_table.GetHashRef().m_hash; } KeyIsEqualT &GetEqualRef() { return m_table.GetEqualRef().m_eq; } KeyHashT const &GetHashRef() const { return m_table.GetHashRef().m_hash; } KeyIsEqualT const &GetEqualRef() const { return m_table.GetEqualRef().m_eq; } UtlHashHandle_t Insert( KeyArg_t k ) { return DoInsert( k ); } UtlHashHandle_t Insert( KeyAlt_t k ) { return DoInsert( k ); } UtlHashHandle_t Insert( KeyArg_t k, ValueArg_t v ) { return DoInsert( k, v ); } UtlHashHandle_t Insert( KeyAlt_t k, ValueArg_t v ) { return DoInsert( k, v ); } UtlHashHandle_t Find( KeyArg_t k ) const { return DoFind( k ); } UtlHashHandle_t Find( KeyAlt_t k ) const { return DoFind( k ); } bool Remove( KeyArg_t k ) { return DoRemove( k ); } bool Remove( KeyAlt_t k ) { return DoRemove( k ); } void RemoveAll() { m_table.RemoveAll(); m_data.RemoveAll(); } void Purge() { m_table.Purge(); m_data.Purge(); } int Count() const { return m_table.Count(); } typedef typename KVPair::ValueReturn_t Element_t; KeyT const &Key( UtlHashHandle_t idx ) const { return m_data[idx].m_key; } Element_t const &Element( UtlHashHandle_t idx ) const { return m_data[idx].GetValue(); } Element_t &Element( UtlHashHandle_t idx ) { return m_data[idx].GetValue(); } Element_t const &operator[]( UtlHashHandle_t idx ) const { return m_data[idx].GetValue(); } Element_t &operator[]( UtlHashHandle_t idx ) { return m_data[idx].GetValue(); } void ReplaceKey( UtlHashHandle_t idx, KeyArg_t k ) { Assert( GetEqualRef()( m_data[idx].m_key, k ) && GetHashRef()( k ) == GetHashRef()( m_data[idx].m_key ) ); m_data[idx].m_key = k; } void ReplaceKey( UtlHashHandle_t idx, KeyAlt_t k ) { Assert( GetEqualRef()( m_data[idx].m_key, k ) && GetHashRef()( k ) == GetHashRef()( m_data[idx].m_key ) ); m_data[idx].m_key = k; } Element_t const &Get( KeyArg_t k, Element_t const &defaultValue ) const { UtlHashHandle_t h = Find( k ); if ( h != InvalidHandle() ) return Element( h ); return defaultValue; } Element_t const &Get( KeyAlt_t k, Element_t const &defaultValue ) const { UtlHashHandle_t h = Find( k ); if ( h != InvalidHandle() ) return Element( h ); return defaultValue; } Element_t const *GetPtr( KeyArg_t k ) const { UtlHashHandle_t h = Find(k); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } Element_t const *GetPtr( KeyAlt_t k ) const { UtlHashHandle_t h = Find(k); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } Element_t *GetPtr( KeyArg_t k ) { UtlHashHandle_t h = Find( k ); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } Element_t *GetPtr( KeyAlt_t k ) { UtlHashHandle_t h = Find( k ); if ( h != InvalidHandle() ) return &Element( h ); return NULL; } UtlHashHandle_t FirstHandle() const { return ExtendInvalidHandle( m_data.Head() ); } UtlHashHandle_t NextHandle( UtlHashHandle_t h ) const { return ExtendInvalidHandle( m_data.Next( h ) ); } bool IsValidHandle( UtlHashHandle_t h ) const { return m_data.IsValidIndex( h ); } UtlHashHandle_t InvalidHandle() const { return (UtlHashHandle_t)-1; } UtlHashHandle_t RemoveAndAdvance( UtlHashHandle_t h ) { Assert( m_data.IsValidIndex( h ) ); m_table.Remove( IndirectIndex( h ) ); IndexStorage_t next = m_data.Next( h ); m_data.Remove( h ); return ExtendInvalidHandle(next); } void Compact( bool bMinimal ) { m_table.Compact( bMinimal ); /*m_data.Compact();*/ } void Swap( CUtlStableHashtable &other ) { m_table.Swap(other.m_table); // XXX swapping CUtlLinkedList by block memory swap, ugh char buf[ sizeof(m_data) ]; memcpy( buf, &m_data, sizeof(m_data) ); memcpy( &m_data, &other.m_data, sizeof(m_data) ); memcpy( &other.m_data, buf, sizeof(m_data) ); } protected: // Perform extension of 0xFFFF to 0xFFFFFFFF if necessary. Note: ( a < CONSTANT ) ? 0 : -1 is usually branchless static UtlHashHandle_t ExtendInvalidHandle( uint32 x ) { return x; } static UtlHashHandle_t ExtendInvalidHandle( uint16 x ) { uint32 a = x; return a | ( ( a < 0xFFFFu ) ? 0 : -1 ); } struct IndirectIndex { explicit IndirectIndex(IndexStorage_t i) : m_index(i) { } IndexStorage_t m_index; }; struct HashProxy { KeyHashT m_hash; unsigned int operator()( IndirectIndex idx ) const { const ptrdiff_t tableoffset = (uintptr_t)(&((Hashtable_t*)1024)->GetHashRef()) - 1024; const ptrdiff_t owneroffset = offsetof(CUtlStableHashtable, m_table) + tableoffset; CUtlStableHashtable* pOwner = (CUtlStableHashtable*)((uintptr_t)this - owneroffset); return m_hash( pOwner->m_data[ idx.m_index ].m_key ); } unsigned int operator()( KeyArg_t k ) const { return m_hash( k ); } unsigned int operator()( KeyAlt_t k ) const { return m_hash( k ); } }; struct EqualProxy { KeyIsEqualT m_eq; unsigned int operator()( IndirectIndex lhs, IndirectIndex rhs ) const { return lhs.m_index == rhs.m_index; } unsigned int operator()( IndirectIndex lhs, KeyArg_t rhs ) const { const ptrdiff_t tableoffset = (uintptr_t)(&((Hashtable_t*)1024)->GetEqualRef()) - 1024; const ptrdiff_t owneroffset = offsetof(CUtlStableHashtable, m_table) + tableoffset; CUtlStableHashtable* pOwner = (CUtlStableHashtable*)((uintptr_t)this - owneroffset); return m_eq( pOwner->m_data[ lhs.m_index ].m_key, rhs ); } unsigned int operator()( IndirectIndex lhs, KeyAlt_t rhs ) const { const ptrdiff_t tableoffset = (uintptr_t)(&((Hashtable_t*)1024)->GetEqualRef()) - 1024; const ptrdiff_t owneroffset = offsetof(CUtlStableHashtable, m_table) + tableoffset; CUtlStableHashtable* pOwner = (CUtlStableHashtable*)((uintptr_t)this - owneroffset); return m_eq( pOwner->m_data[ lhs.m_index ].m_key, rhs ); } }; class CCustomLinkedList : public LinkedList_t { public: int AddToTailUnconstructed() { IndexStorage_t newNode = this->AllocInternal(); if ( newNode != this->InvalidIndex() ) this->LinkToTail( newNode ); return newNode; } }; Hashtable_t m_table; CCustomLinkedList m_data; }; template template inline bool CUtlStableHashtable::DoRemove( KeyArgumentT k ) { unsigned int hash = m_table.GetHashRef()( k ); UtlHashHandle_t h = m_table.template DoLookup( k, hash, NULL ); if ( h == m_table.InvalidHandle() ) return false; int idx = m_table[ h ].m_index; m_table.template DoRemove( IndirectIndex( idx ), hash ); m_data.Remove( idx ); return true; } template template inline UtlHashHandle_t CUtlStableHashtable::DoFind( KeyArgumentT k ) const { unsigned int hash = m_table.GetHashRef()( k ); UtlHashHandle_t h = m_table.template DoLookup( k, hash, NULL ); if ( h != m_table.InvalidHandle() ) return m_table[ h ].m_index; return (UtlHashHandle_t) -1; } template template inline UtlHashHandle_t CUtlStableHashtable::DoInsert( KeyArgumentT k ) { unsigned int hash = m_table.GetHashRef()( k ); UtlHashHandle_t h = m_table.template DoLookup( k, hash, NULL ); if ( h != m_table.InvalidHandle() ) return m_table[ h ].m_index; int idx = m_data.AddToTailUnconstructed(); ConstructOneArg( &m_data[idx], k ); m_table.template DoInsertNoCheck( IndirectIndex( idx ), empty_t(), hash ); return idx; } template template inline UtlHashHandle_t CUtlStableHashtable::DoInsert( KeyArgumentT k, ValueArgumentT v ) { unsigned int hash = m_table.GetHashRef()( k ); UtlHashHandle_t h = m_table.template DoLookup( k, hash, NULL ); if ( h != m_table.InvalidHandle() ) return m_table[ h ].m_index; int idx = m_data.AddToTailUnconstructed(); ConstructTwoArg( &m_data[idx], k, v ); m_table.template DoInsertNoCheck( IndirectIndex( idx ), empty_t(), hash ); return idx; } #endif // UTLHASHTABLE_H