/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ /* 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 jsdhash_h___ #define jsdhash_h___ /* * Double hashing, a la Knuth 6. * * Try to keep this file in sync with xpcom/glue/pldhash.h. */ #include "jstypes.h" #include "jsutil.h" #if defined(__GNUC__) && defined(__i386__) && !defined(XP_OS2) #define JS_DHASH_FASTCALL __attribute__ ((regparm (3),stdcall)) #elif defined(XP_WIN) #define JS_DHASH_FASTCALL __fastcall #else #define JS_DHASH_FASTCALL #endif #ifdef DEBUG_XXXbrendan #define JS_DHASHMETER 1 #endif /* Table size limit, do not equal or exceed (see min&maxAlphaFrac, below). */ #undef JS_DHASH_SIZE_LIMIT #define JS_DHASH_SIZE_LIMIT JS_BIT(24) /* Minimum table size, or gross entry count (net is at most .75 loaded). */ #ifndef JS_DHASH_MIN_SIZE #define JS_DHASH_MIN_SIZE 16 #elif (JS_DHASH_MIN_SIZE & (JS_DHASH_MIN_SIZE - 1)) != 0 #error "JS_DHASH_MIN_SIZE must be a power of two!" #endif /* * Multiplicative hash uses an unsigned 32 bit integer and the golden ratio, * expressed as a fixed-point 32-bit fraction. */ #define JS_DHASH_BITS 32 #define JS_DHASH_GOLDEN_RATIO 0x9E3779B9U /* Primitive and forward-struct typedefs. */ typedef uint32_t JSDHashNumber; typedef struct JSDHashEntryHdr JSDHashEntryHdr; typedef struct JSDHashEntryStub JSDHashEntryStub; typedef struct JSDHashTable JSDHashTable; typedef struct JSDHashTableOps JSDHashTableOps; /* * Table entry header structure. * * In order to allow in-line allocation of key and value, we do not declare * either here. Instead, the API uses const void *key as a formal parameter. * The key need not be stored in the entry; it may be part of the value, but * need not be stored at all. * * Callback types are defined below and grouped into the JSDHashTableOps * structure, for single static initialization per hash table sub-type. * * Each hash table sub-type should nest the JSDHashEntryHdr structure at the * front of its particular entry type. The keyHash member contains the result * of multiplying the hash code returned from the hashKey callback (see below) * by JS_DHASH_GOLDEN_RATIO, then constraining the result to avoid the magic 0 * and 1 values. The stored keyHash value is table size invariant, and it is * maintained automatically by JS_DHashTableOperate -- users should never set * it, and its only uses should be via the entry macros below. * * The JS_DHASH_ENTRY_IS_LIVE macro tests whether entry is neither free nor * removed. An entry may be either busy or free; if busy, it may be live or * removed. Consumers of this API should not access members of entries that * are not live. * * However, use JS_DHASH_ENTRY_IS_BUSY for faster liveness testing of entries * returned by JS_DHashTableOperate, as JS_DHashTableOperate never returns a * non-live, busy (i.e., removed) entry pointer to its caller. See below for * more details on JS_DHashTableOperate's calling rules. */ struct JSDHashEntryHdr { JSDHashNumber keyHash; /* every entry must begin like this */ }; MOZ_ALWAYS_INLINE bool JS_DHASH_ENTRY_IS_FREE(JSDHashEntryHdr* entry) { return entry->keyHash == 0; } MOZ_ALWAYS_INLINE bool JS_DHASH_ENTRY_IS_BUSY(JSDHashEntryHdr* entry) { return !JS_DHASH_ENTRY_IS_FREE(entry); } MOZ_ALWAYS_INLINE bool JS_DHASH_ENTRY_IS_LIVE(JSDHashEntryHdr* entry) { return entry->keyHash >= 2; } /* * A JSDHashTable is currently 8 words (without the JS_DHASHMETER overhead) * on most architectures, and may be allocated on the stack or within another * structure or class (see below for the Init and Finish functions to use). * * To decide whether to use double hashing vs. chaining, we need to develop a * trade-off relation, as follows: * * Let alpha be the load factor, esize the entry size in words, count the * entry count, and pow2 the power-of-two table size in entries. * * (JSDHashTable overhead) > (JSHashTable overhead) * (unused table entry space) > (malloc and .next overhead per entry) + * (buckets overhead) * (1 - alpha) * esize * pow2 > 2 * count + pow2 * * Notice that alpha is by definition (count / pow2): * * (1 - alpha) * esize * pow2 > 2 * alpha * pow2 + pow2 * (1 - alpha) * esize > 2 * alpha + 1 * * esize > (1 + 2 * alpha) / (1 - alpha) * * This assumes both tables must keep keyHash, key, and value for each entry, * where key and value point to separately allocated strings or structures. * If key and value can be combined into one pointer, then the trade-off is: * * esize > (1 + 3 * alpha) / (1 - alpha) * * If the entry value can be a subtype of JSDHashEntryHdr, rather than a type * that must be allocated separately and referenced by an entry.value pointer * member, and provided key's allocation can be fused with its entry's, then * k (the words wasted per entry with chaining) is 4. * * To see these curves, feed gnuplot input like so: * * gnuplot> f(x,k) = (1 + k * x) / (1 - x) * gnuplot> plot [0:.75] f(x,2), f(x,3), f(x,4) * * For k of 2 and a well-loaded table (alpha > .5), esize must be more than 4 * words for chaining to be more space-efficient than double hashing. * * Solving for alpha helps us decide when to shrink an underloaded table: * * esize > (1 + k * alpha) / (1 - alpha) * esize - alpha * esize > 1 + k * alpha * esize - 1 > (k + esize) * alpha * (esize - 1) / (k + esize) > alpha * * alpha < (esize - 1) / (esize + k) * * Therefore double hashing should keep alpha >= (esize - 1) / (esize + k), * assuming esize is not too large (in which case, chaining should probably be * used for any alpha). For esize=2 and k=3, we want alpha >= .2; for esize=3 * and k=2, we want alpha >= .4. For k=4, esize could be 6, and alpha >= .5 * would still obtain. See the JS_DHASH_MIN_ALPHA macro further below. * * The current implementation uses a configurable lower bound on alpha, which * defaults to .25, when deciding to shrink the table (while still respecting * JS_DHASH_MIN_SIZE). * * Note a qualitative difference between chaining and double hashing: under * chaining, entry addresses are stable across table shrinks and grows. With * double hashing, you can't safely hold an entry pointer and use it after an * ADD or REMOVE operation, unless you sample table->generation before adding * or removing, and compare the sample after, dereferencing the entry pointer * only if table->generation has not changed. * * The moral of this story: there is no one-size-fits-all hash table scheme, * but for small table entry size, and assuming entry address stability is not * required, double hashing wins. */ struct JSDHashTable { const JSDHashTableOps *ops; /* virtual operations, see below */ void *data; /* ops- and instance-specific data */ int16_t hashShift; /* multiplicative hash shift */ uint8_t maxAlphaFrac; /* 8-bit fixed point max alpha */ uint8_t minAlphaFrac; /* 8-bit fixed point min alpha */ uint32_t entrySize; /* number of bytes in an entry */ uint32_t entryCount; /* number of entries in table */ uint32_t removedCount; /* removed entry sentinels in table */ uint32_t generation; /* entry storage generation number */ char *entryStore; /* entry storage */ #ifdef JS_DHASHMETER struct JSDHashStats { 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 lookups; /* number of JS_DHASH_LOOKUPs */ uint32_t addMisses; /* adds that miss, and do work */ uint32_t addOverRemoved; /* adds that recycled a removed entry */ uint32_t addHits; /* adds that hit an existing entry */ uint32_t addFailures; /* out-of-memory during add growth */ uint32_t removeHits; /* removes that hit, and do work */ uint32_t removeMisses; /* useless removes that miss */ uint32_t removeFrees; /* removes that freed entry directly */ uint32_t removeEnums; /* removes done by Enumerate */ uint32_t grows; /* table expansions */ uint32_t shrinks; /* table contractions */ uint32_t compresses; /* table compressions */ uint32_t enumShrinks; /* contractions after Enumerate */ } stats; #endif }; /* * Size in entries (gross, not net of free and removed sentinels) for table. * We store hashShift rather than sizeLog2 to optimize the collision-free case * in SearchTable. */ #define JS_DHASH_TABLE_SIZE(table) JS_BIT(JS_DHASH_BITS - (table)->hashShift) /* * Table space at entryStore is allocated and freed using these callbacks. * The allocator should return null on error only (not if called with nbytes * equal to 0; but note that jsdhash.c code will never call with 0 nbytes). */ typedef void * (* JSDHashAllocTable)(JSDHashTable *table, uint32_t nbytes); typedef void (* JSDHashFreeTable) (JSDHashTable *table, void *ptr); /* * Compute the hash code for a given key to be looked up, added, or removed * from table. A hash code may have any JSDHashNumber value. */ typedef JSDHashNumber (* JSDHashHashKey) (JSDHashTable *table, const void *key); /* * Compare the key identifying entry in table with the provided key parameter. * Return JS_TRUE if keys match, JS_FALSE otherwise. */ typedef JSBool (* JSDHashMatchEntry)(JSDHashTable *table, const JSDHashEntryHdr *entry, const void *key); /* * Copy the data starting at from to the new entry storage at to. Do not add * reference counts for any strong references in the entry, however, as this * is a "move" operation: the old entry storage at from will be freed without * any reference-decrementing callback shortly. */ typedef void (* JSDHashMoveEntry)(JSDHashTable *table, const JSDHashEntryHdr *from, JSDHashEntryHdr *to); /* * Clear the entry and drop any strong references it holds. This callback is * invoked during a JS_DHASH_REMOVE operation (see below for operation codes), * but only if the given key is found in the table. */ typedef void (* JSDHashClearEntry)(JSDHashTable *table, JSDHashEntryHdr *entry); /* * Called when a table (whether allocated dynamically by itself, or nested in * a larger structure, or allocated on the stack) is finished. This callback * allows table->ops-specific code to finalize table->data. */ typedef void (* JSDHashFinalize) (JSDHashTable *table); /* * Initialize a new entry, apart from keyHash. This function is called when * JS_DHashTableOperate's JS_DHASH_ADD case finds no existing entry for the * given key, and must add a new one. At that point, entry->keyHash is not * set yet, to avoid claiming the last free entry in a severely overloaded * table. */ typedef JSBool (* JSDHashInitEntry)(JSDHashTable *table, JSDHashEntryHdr *entry, const void *key); /* * Finally, the "vtable" structure for JSDHashTable. The first eight hooks * must be provided by implementations; they're called unconditionally by the * generic jsdhash.c code. Hooks after these may be null. * * Summary of allocation-related hook usage with C++ placement new emphasis: * allocTable Allocate raw bytes with malloc, no ctors run. * freeTable Free raw bytes with free, no dtors run. * initEntry Call placement new using default key-based ctor. * Return JS_TRUE on success, JS_FALSE on error. * moveEntry Call placement new using copy ctor, run dtor on old * entry storage. * clearEntry Run dtor on entry. * finalize Stub unless table->data was initialized and needs to * be finalized. * * Note the reason why initEntry is optional: the default hooks (stubs) clear * entry storage: On successful JS_DHashTableOperate(tbl, key, JS_DHASH_ADD), * the returned entry pointer addresses an entry struct whose keyHash member * has been set non-zero, but all other entry members are still clear (null). * JS_DHASH_ADD callers can test such members to see whether the entry was * newly created by the JS_DHASH_ADD call that just succeeded. If placement * new or similar initialization is required, define an initEntry hook. Of * course, the clearEntry hook must zero or null appropriately. * * XXX assumes 0 is null for pointer types. */ struct JSDHashTableOps { /* Mandatory hooks. All implementations must provide these. */ JSDHashAllocTable allocTable; JSDHashFreeTable freeTable; JSDHashHashKey hashKey; JSDHashMatchEntry matchEntry; JSDHashMoveEntry moveEntry; JSDHashClearEntry clearEntry; JSDHashFinalize finalize; /* Optional hooks start here. If null, these are not called. */ JSDHashInitEntry initEntry; }; /* * Default implementations for the above ops. */ extern JS_PUBLIC_API(void *) JS_DHashAllocTable(JSDHashTable *table, uint32_t nbytes); extern JS_PUBLIC_API(void) JS_DHashFreeTable(JSDHashTable *table, void *ptr); extern JS_PUBLIC_API(JSDHashNumber) JS_DHashStringKey(JSDHashTable *table, const void *key); /* A minimal entry contains a keyHash header and a void key pointer. */ struct JSDHashEntryStub { JSDHashEntryHdr hdr; const void *key; }; extern JS_PUBLIC_API(JSDHashNumber) JS_DHashVoidPtrKeyStub(JSDHashTable *table, const void *key); extern JS_PUBLIC_API(JSBool) JS_DHashMatchEntryStub(JSDHashTable *table, const JSDHashEntryHdr *entry, const void *key); extern JS_PUBLIC_API(JSBool) JS_DHashMatchStringKey(JSDHashTable *table, const JSDHashEntryHdr *entry, const void *key); extern JS_PUBLIC_API(void) JS_DHashMoveEntryStub(JSDHashTable *table, const JSDHashEntryHdr *from, JSDHashEntryHdr *to); extern JS_PUBLIC_API(void) JS_DHashClearEntryStub(JSDHashTable *table, JSDHashEntryHdr *entry); extern JS_PUBLIC_API(void) JS_DHashFreeStringKey(JSDHashTable *table, JSDHashEntryHdr *entry); extern JS_PUBLIC_API(void) JS_DHashFinalizeStub(JSDHashTable *table); /* * If you use JSDHashEntryStub or a subclass of it as your entry struct, and * if your entries move via memcpy and clear via memset(0), you can use these * stub operations. */ extern JS_PUBLIC_API(const JSDHashTableOps *) JS_DHashGetStubOps(void); /* * Dynamically allocate a new JSDHashTable using malloc, initialize it using * JS_DHashTableInit, and return its address. Return null on malloc failure. * Note that the entry storage at table->entryStore will be allocated using * the ops->allocTable callback. */ extern JS_PUBLIC_API(JSDHashTable *) JS_NewDHashTable(const JSDHashTableOps *ops, void *data, uint32_t entrySize, uint32_t capacity); /* * Finalize table's data, free its entry storage (via table->ops->freeTable), * and return the memory starting at table to the malloc heap. */ extern JS_PUBLIC_API(void) JS_DHashTableDestroy(JSDHashTable *table); /* * Initialize table with ops, data, entrySize, and capacity. Capacity is a * guess for the smallest table size at which the table will usually be less * than 75% loaded (the table will grow or shrink as needed; capacity serves * only to avoid inevitable early growth from JS_DHASH_MIN_SIZE). */ extern JS_PUBLIC_API(JSBool) JS_DHashTableInit(JSDHashTable *table, const JSDHashTableOps *ops, void *data, uint32_t entrySize, uint32_t capacity); /* * Set maximum and minimum alpha for table. The defaults are 0.75 and .25. * maxAlpha must be in [0.5, 0.9375] for the default JS_DHASH_MIN_SIZE; or if * MinSize=JS_DHASH_MIN_SIZE <= 256, in [0.5, (float)(MinSize-1)/MinSize]; or * else in [0.5, 255.0/256]. minAlpha must be in [0, maxAlpha / 2), so that * we don't shrink on the very next remove after growing a table upon adding * an entry that brings entryCount past maxAlpha * tableSize. */ extern JS_PUBLIC_API(void) JS_DHashTableSetAlphaBounds(JSDHashTable *table, float maxAlpha, float minAlpha); /* * Call this macro with k, the number of pointer-sized words wasted per entry * under chaining, to compute the minimum alpha at which double hashing still * beats chaining. */ #define JS_DHASH_MIN_ALPHA(table, k) \ ((float)((table)->entrySize / sizeof(void *) - 1) \ / ((table)->entrySize / sizeof(void *) + (k))) /* * Default max/min alpha, and macros to compute the value for the |capacity| * parameter to JS_NewDHashTable and JS_DHashTableInit, given default or any * max alpha, such that adding entryCount entries right after initializing the * table will not require a reallocation (so JS_DHASH_ADD can't fail for those * JS_DHashTableOperate calls). * * NB: JS_DHASH_CAP is a helper macro meant for use only in JS_DHASH_CAPACITY. * Don't use it directly! */ #define JS_DHASH_DEFAULT_MAX_ALPHA 0.75 #define JS_DHASH_DEFAULT_MIN_ALPHA 0.25 #define JS_DHASH_CAP(entryCount, maxAlpha) \ ((uint32_t)((double)(entryCount) / (maxAlpha))) #define JS_DHASH_CAPACITY(entryCount, maxAlpha) \ (JS_DHASH_CAP(entryCount, maxAlpha) + \ (((JS_DHASH_CAP(entryCount, maxAlpha) * (uint8_t)(0x100 * (maxAlpha))) \ >> 8) < (entryCount))) #define JS_DHASH_DEFAULT_CAPACITY(entryCount) \ JS_DHASH_CAPACITY(entryCount, JS_DHASH_DEFAULT_MAX_ALPHA) /* * Finalize table's data, free its entry storage using table->ops->freeTable, * and leave its members unchanged from their last live values (which leaves * pointers dangling). If you want to burn cycles clearing table, it's up to * your code to call memset. */ extern JS_PUBLIC_API(void) JS_DHashTableFinish(JSDHashTable *table); /* * To consolidate keyHash computation and table grow/shrink code, we use a * single entry point for lookup, add, and remove operations. The operation * codes are declared here, along with codes returned by JSDHashEnumerator * functions, which control JS_DHashTableEnumerate's behavior. */ typedef enum JSDHashOperator { JS_DHASH_LOOKUP = 0, /* lookup entry */ JS_DHASH_ADD = 1, /* add entry */ JS_DHASH_REMOVE = 2, /* remove entry, or enumerator says remove */ JS_DHASH_NEXT = 0, /* enumerator says continue */ JS_DHASH_STOP = 1 /* enumerator says stop */ } JSDHashOperator; /* * To lookup a key in table, call: * * entry = JS_DHashTableOperate(table, key, JS_DHASH_LOOKUP); * * If JS_DHASH_ENTRY_IS_BUSY(entry) is true, key was found and it identifies * entry. If JS_DHASH_ENTRY_IS_FREE(entry) is true, key was not found. * * To add an entry identified by key to table, call: * * entry = JS_DHashTableOperate(table, key, JS_DHASH_ADD); * * If entry is null upon return, then either the table is severely overloaded, * and memory can't be allocated for entry storage via table->ops->allocTable; * Or if table->ops->initEntry is non-null, the table->ops->initEntry op may * have returned false. * * Otherwise, entry->keyHash has been set so that JS_DHASH_ENTRY_IS_BUSY(entry) * is true, and it is up to the caller to initialize the key and value parts * of the entry sub-type, if they have not been set already (i.e. if entry was * not already in the table, and if the optional initEntry hook was not used). * * To remove an entry identified by key from table, call: * * (void) JS_DHashTableOperate(table, key, JS_DHASH_REMOVE); * * If key's entry is found, it is cleared (via table->ops->clearEntry) and * the entry is marked so that JS_DHASH_ENTRY_IS_FREE(entry). This operation * returns null unconditionally; you should ignore its return value. */ extern JS_PUBLIC_API(JSDHashEntryHdr *) JS_DHASH_FASTCALL JS_DHashTableOperate(JSDHashTable *table, const void *key, JSDHashOperator op); /* * Remove an entry already accessed via LOOKUP or ADD. * * NB: this is a "raw" or low-level routine, intended to be used only where * the inefficiency of a full JS_DHashTableOperate (which rehashes in order * to find the entry given its key) is not tolerable. This function does not * shrink the table if it is underloaded. It does not update stats #ifdef * JS_DHASHMETER, either. */ extern JS_PUBLIC_API(void) JS_DHashTableRawRemove(JSDHashTable *table, JSDHashEntryHdr *entry); /* * Enumerate entries in table using etor: * * count = JS_DHashTableEnumerate(table, etor, arg); * * JS_DHashTableEnumerate calls etor like so: * * op = etor(table, entry, number, arg); * * where number is a zero-based ordinal assigned to live entries according to * their order in table->entryStore. * * The return value, op, is treated as a set of flags. If op is JS_DHASH_NEXT, * then continue enumerating. If op contains JS_DHASH_REMOVE, then clear (via * table->ops->clearEntry) and free entry. Then we check whether op contains * JS_DHASH_STOP; if so, stop enumerating and return the number of live entries * that were enumerated so far. Return the total number of live entries when * enumeration completes normally. * * If etor calls JS_DHashTableOperate on table with op != JS_DHASH_LOOKUP, it * must return JS_DHASH_STOP; otherwise undefined behavior results. * * If any enumerator returns JS_DHASH_REMOVE, table->entryStore may be shrunk * or compressed after enumeration, but before JS_DHashTableEnumerate returns. * Such an enumerator therefore can't safely set aside entry pointers, but an * enumerator that never returns JS_DHASH_REMOVE can set pointers to entries * aside, e.g., to avoid copying live entries into an array of the entry type. * Copying entry pointers is cheaper, and safe so long as the caller of such a * "stable" Enumerate doesn't use the set-aside pointers after any call either * to PL_DHashTableOperate, or to an "unstable" form of Enumerate, which might * grow or shrink entryStore. * * If your enumerator wants to remove certain entries, but set aside pointers * to other entries that it retains, it can use JS_DHashTableRawRemove on the * entries to be removed, returning JS_DHASH_NEXT to skip them. Likewise, if * you want to remove entries, but for some reason you do not want entryStore * to be shrunk or compressed, you can call JS_DHashTableRawRemove safely on * the entry being enumerated, rather than returning JS_DHASH_REMOVE. */ typedef JSDHashOperator (* JSDHashEnumerator)(JSDHashTable *table, JSDHashEntryHdr *hdr, uint32_t number, void *arg); extern JS_PUBLIC_API(uint32_t) JS_DHashTableEnumerate(JSDHashTable *table, JSDHashEnumerator etor, void *arg); typedef size_t (* JSDHashSizeOfEntryExcludingThisFun)(JSDHashEntryHdr *hdr, JSMallocSizeOfFun mallocSizeOf, void *arg); /** * Measure the size of the table's entry storage, and if * |sizeOfEntryExcludingThis| is non-NULL, measure the size of things pointed * to by entries. Doesn't measure |ops| because it's often shared between * tables, nor |data| because it's opaque. */ extern JS_PUBLIC_API(size_t) JS_DHashTableSizeOfExcludingThis(const JSDHashTable *table, JSDHashSizeOfEntryExcludingThisFun sizeOfEntryExcludingThis, JSMallocSizeOfFun mallocSizeOf, void *arg = NULL); /** * Like JS_DHashTableSizeOfExcludingThis, but includes sizeof(*this). */ extern JS_PUBLIC_API(size_t) JS_DHashTableSizeOfIncludingThis(const JSDHashTable *table, JSDHashSizeOfEntryExcludingThisFun sizeOfEntryExcludingThis, JSMallocSizeOfFun mallocSizeOf, void *arg = NULL); #ifdef DEBUG /** * Mark a table as immutable for the remainder of its lifetime. This * changes the implementation from ASSERTing one set of invariants to * ASSERTing a different set. * * When a table is NOT marked as immutable, the table implementation * asserts that the table is not mutated from its own callbacks. It * assumes the caller protects the table from being accessed on multiple * threads simultaneously. * * When the table is marked as immutable, the re-entry assertions will * no longer trigger erroneously due to multi-threaded access. Instead, * mutations will cause assertions. */ extern JS_PUBLIC_API(void) JS_DHashMarkTableImmutable(JSDHashTable *table); #endif #ifdef JS_DHASHMETER #include extern JS_PUBLIC_API(void) JS_DHashTableDumpMeter(JSDHashTable *table, JSDHashEnumerator dump, FILE *fp); #endif #endif /* jsdhash_h___ */