dict.cpp revision 1472:c18cbe5936b8
1/* 2 * Copyright (c) 1997, 2009, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25// Dictionaries - An Abstract Data Type 26 27#include "incls/_precompiled.incl" 28#include "incls/_dict.cpp.incl" 29 30// %%%%% includes not needed with AVM framework - Ungar 31 32// #include "port.hpp" 33//IMPLEMENTATION 34// #include "dict.hpp" 35 36#include <assert.h> 37 38// The iostream is not needed and it gets confused for gcc by the 39// define of bool. 40// 41// #include <iostream.h> 42 43//------------------------------data----------------------------------------- 44// String hash tables 45#define MAXID 20 46static byte initflag = 0; // True after 1st initialization 47static const char shft[MAXID] = {1,2,3,4,5,6,7,1,2,3,4,5,6,7,1,2,3,4,5,6}; 48static short xsum[MAXID]; 49 50//------------------------------bucket--------------------------------------- 51class bucket : public ResourceObj { 52public: 53 uint _cnt, _max; // Size of bucket 54 void **_keyvals; // Array of keys and values 55}; 56 57//------------------------------Dict----------------------------------------- 58// The dictionary is kept has a hash table. The hash table is a even power 59// of two, for nice modulo operations. Each bucket in the hash table points 60// to a linear list of key-value pairs; each key & value is just a (void *). 61// The list starts with a count. A hash lookup finds the list head, then a 62// simple linear scan finds the key. If the table gets too full, it's 63// doubled in size; the total amount of EXTRA times all hash functions are 64// computed for the doubling is no more than the current size - thus the 65// doubling in size costs no more than a constant factor in speed. 66Dict::Dict(CmpKey initcmp, Hash inithash) : _hash(inithash), _cmp(initcmp), 67 _arena(Thread::current()->resource_area()) { 68 int i; 69 70 // Precompute table of null character hashes 71 if( !initflag ) { // Not initializated yet? 72 xsum[0] = (1<<shft[0])+1; // Initialize 73 for(i=1; i<MAXID; i++) { 74 xsum[i] = (1<<shft[i])+1+xsum[i-1]; 75 } 76 initflag = 1; // Never again 77 } 78 79 _size = 16; // Size is a power of 2 80 _cnt = 0; // Dictionary is empty 81 _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 82 memset(_bin,0,sizeof(bucket)*_size); 83} 84 85Dict::Dict(CmpKey initcmp, Hash inithash, Arena *arena, int size) 86: _hash(inithash), _cmp(initcmp), _arena(arena) { 87 int i; 88 89 // Precompute table of null character hashes 90 if( !initflag ) { // Not initializated yet? 91 xsum[0] = (1<<shft[0])+1; // Initialize 92 for(i=1; i<MAXID; i++) { 93 xsum[i] = (1<<shft[i])+1+xsum[i-1]; 94 } 95 initflag = 1; // Never again 96 } 97 98 i=16; 99 while( i < size ) i <<= 1; 100 _size = i; // Size is a power of 2 101 _cnt = 0; // Dictionary is empty 102 _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 103 memset(_bin,0,sizeof(bucket)*_size); 104} 105 106//------------------------------~Dict------------------------------------------ 107// Delete an existing dictionary. 108Dict::~Dict() { 109 /* 110 tty->print("~Dict %d/%d: ",_cnt,_size); 111 for( uint i=0; i < _size; i++) // For complete new table do 112 tty->print("%d ",_bin[i]._cnt); 113 tty->print("\n");*/ 114 /*for( uint i=0; i<_size; i++ ) { 115 FREE_FAST( _bin[i]._keyvals ); 116 } */ 117} 118 119//------------------------------Clear---------------------------------------- 120// Zap to empty; ready for re-use 121void Dict::Clear() { 122 _cnt = 0; // Empty contents 123 for( uint i=0; i<_size; i++ ) 124 _bin[i]._cnt = 0; // Empty buckets, but leave allocated 125 // Leave _size & _bin alone, under the assumption that dictionary will 126 // grow to this size again. 127} 128 129//------------------------------doubhash--------------------------------------- 130// Double hash table size. If can't do so, just suffer. If can, then run 131// thru old hash table, moving things to new table. Note that since hash 132// table doubled, exactly 1 new bit is exposed in the mask - so everything 133// in the old table ends up on 1 of two lists in the new table; a hi and a 134// lo list depending on the value of the bit. 135void Dict::doubhash(void) { 136 uint oldsize = _size; 137 _size <<= 1; // Double in size 138 _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*oldsize, sizeof(bucket)*_size ); 139 memset( &_bin[oldsize], 0, oldsize*sizeof(bucket) ); 140 // Rehash things to spread into new table 141 for( uint i=0; i < oldsize; i++) { // For complete OLD table do 142 bucket *b = &_bin[i]; // Handy shortcut for _bin[i] 143 if( !b->_keyvals ) continue; // Skip empties fast 144 145 bucket *nb = &_bin[i+oldsize]; // New bucket shortcut 146 uint j = b->_max; // Trim new bucket to nearest power of 2 147 while( j > b->_cnt ) j >>= 1; // above old bucket _cnt 148 if( !j ) j = 1; // Handle zero-sized buckets 149 nb->_max = j<<1; 150 // Allocate worst case space for key-value pairs 151 nb->_keyvals = (void**)_arena->Amalloc_4( sizeof(void *)*nb->_max*2 ); 152 uint nbcnt = 0; 153 154 for( j=0; j<b->_cnt; j++ ) { // Rehash all keys in this bucket 155 void *key = b->_keyvals[j+j]; 156 if( (_hash( key ) & (_size-1)) != i ) { // Moving to hi bucket? 157 nb->_keyvals[nbcnt+nbcnt] = key; 158 nb->_keyvals[nbcnt+nbcnt+1] = b->_keyvals[j+j+1]; 159 nb->_cnt = nbcnt = nbcnt+1; 160 b->_cnt--; // Remove key/value from lo bucket 161 b->_keyvals[j+j ] = b->_keyvals[b->_cnt+b->_cnt ]; 162 b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1]; 163 j--; // Hash compacted element also 164 } 165 } // End of for all key-value pairs in bucket 166 } // End of for all buckets 167 168 169} 170 171//------------------------------Dict----------------------------------------- 172// Deep copy a dictionary. 173Dict::Dict( const Dict &d ) : _size(d._size), _cnt(d._cnt), _hash(d._hash),_cmp(d._cmp), _arena(d._arena) { 174 _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 175 memcpy( _bin, d._bin, sizeof(bucket)*_size ); 176 for( uint i=0; i<_size; i++ ) { 177 if( !_bin[i]._keyvals ) continue; 178 _bin[i]._keyvals=(void**)_arena->Amalloc_4( sizeof(void *)*_bin[i]._max*2); 179 memcpy( _bin[i]._keyvals, d._bin[i]._keyvals,_bin[i]._cnt*2*sizeof(void*)); 180 } 181} 182 183//------------------------------Dict----------------------------------------- 184// Deep copy a dictionary. 185Dict &Dict::operator =( const Dict &d ) { 186 if( _size < d._size ) { // If must have more buckets 187 _arena = d._arena; 188 _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*_size, sizeof(bucket)*d._size ); 189 memset( &_bin[_size], 0, (d._size-_size)*sizeof(bucket) ); 190 _size = d._size; 191 } 192 uint i; 193 for( i=0; i<_size; i++ ) // All buckets are empty 194 _bin[i]._cnt = 0; // But leave bucket allocations alone 195 _cnt = d._cnt; 196 *(Hash*)(&_hash) = d._hash; 197 *(CmpKey*)(&_cmp) = d._cmp; 198 for( i=0; i<_size; i++ ) { 199 bucket *b = &d._bin[i]; // Shortcut to source bucket 200 for( uint j=0; j<b->_cnt; j++ ) 201 Insert( b->_keyvals[j+j], b->_keyvals[j+j+1] ); 202 } 203 return *this; 204} 205 206//------------------------------Insert---------------------------------------- 207// Insert or replace a key/value pair in the given dictionary. If the 208// dictionary is too full, it's size is doubled. The prior value being 209// replaced is returned (NULL if this is a 1st insertion of that key). If 210// an old value is found, it's swapped with the prior key-value pair on the 211// list. This moves a commonly searched-for value towards the list head. 212void *Dict::Insert(void *key, void *val, bool replace) { 213 uint hash = _hash( key ); // Get hash key 214 uint i = hash & (_size-1); // Get hash key, corrected for size 215 bucket *b = &_bin[i]; // Handy shortcut 216 for( uint j=0; j<b->_cnt; j++ ) { 217 if( !_cmp(key,b->_keyvals[j+j]) ) { 218 if (!replace) { 219 return b->_keyvals[j+j+1]; 220 } else { 221 void *prior = b->_keyvals[j+j+1]; 222 b->_keyvals[j+j ] = key; // Insert current key-value 223 b->_keyvals[j+j+1] = val; 224 return prior; // Return prior 225 } 226 } 227 } 228 if( ++_cnt > _size ) { // Hash table is full 229 doubhash(); // Grow whole table if too full 230 i = hash & (_size-1); // Rehash 231 b = &_bin[i]; // Handy shortcut 232 } 233 if( b->_cnt == b->_max ) { // Must grow bucket? 234 if( !b->_keyvals ) { 235 b->_max = 2; // Initial bucket size 236 b->_keyvals = (void**)_arena->Amalloc_4(sizeof(void*) * b->_max * 2); 237 } else { 238 b->_keyvals = (void**)_arena->Arealloc(b->_keyvals, sizeof(void*) * b->_max * 2, sizeof(void*) * b->_max * 4); 239 b->_max <<= 1; // Double bucket 240 } 241 } 242 b->_keyvals[b->_cnt+b->_cnt ] = key; 243 b->_keyvals[b->_cnt+b->_cnt+1] = val; 244 b->_cnt++; 245 return NULL; // Nothing found prior 246} 247 248//------------------------------Delete--------------------------------------- 249// Find & remove a value from dictionary. Return old value. 250void *Dict::Delete(void *key) { 251 uint i = _hash( key ) & (_size-1); // Get hash key, corrected for size 252 bucket *b = &_bin[i]; // Handy shortcut 253 for( uint j=0; j<b->_cnt; j++ ) 254 if( !_cmp(key,b->_keyvals[j+j]) ) { 255 void *prior = b->_keyvals[j+j+1]; 256 b->_cnt--; // Remove key/value from lo bucket 257 b->_keyvals[j+j ] = b->_keyvals[b->_cnt+b->_cnt ]; 258 b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1]; 259 _cnt--; // One less thing in table 260 return prior; 261 } 262 return NULL; 263} 264 265//------------------------------FindDict------------------------------------- 266// Find a key-value pair in the given dictionary. If not found, return NULL. 267// If found, move key-value pair towards head of list. 268void *Dict::operator [](const void *key) const { 269 uint i = _hash( key ) & (_size-1); // Get hash key, corrected for size 270 bucket *b = &_bin[i]; // Handy shortcut 271 for( uint j=0; j<b->_cnt; j++ ) 272 if( !_cmp(key,b->_keyvals[j+j]) ) 273 return b->_keyvals[j+j+1]; 274 return NULL; 275} 276 277//------------------------------CmpDict-------------------------------------- 278// CmpDict compares two dictionaries; they must have the same keys (their 279// keys must match using CmpKey) and they must have the same values (pointer 280// comparison). If so 1 is returned, if not 0 is returned. 281int32 Dict::operator ==(const Dict &d2) const { 282 if( _cnt != d2._cnt ) return 0; 283 if( _hash != d2._hash ) return 0; 284 if( _cmp != d2._cmp ) return 0; 285 for( uint i=0; i < _size; i++) { // For complete hash table do 286 bucket *b = &_bin[i]; // Handy shortcut 287 if( b->_cnt != d2._bin[i]._cnt ) return 0; 288 if( memcmp(b->_keyvals, d2._bin[i]._keyvals, b->_cnt*2*sizeof(void*) ) ) 289 return 0; // Key-value pairs must match 290 } 291 return 1; // All match, is OK 292} 293 294//------------------------------print------------------------------------------ 295// Handier print routine 296void Dict::print() { 297 DictI i(this); // Moved definition in iterator here because of g++. 298 tty->print("Dict@0x%lx[%d] = {", this, _cnt); 299 for( ; i.test(); ++i ) { 300 tty->print("(0x%lx,0x%lx),", i._key, i._value); 301 } 302 tty->print_cr("}"); 303} 304 305//------------------------------Hashing Functions---------------------------- 306// Convert string to hash key. This algorithm implements a universal hash 307// function with the multipliers frozen (ok, so it's not universal). The 308// multipliers (and allowable characters) are all odd, so the resultant sum 309// is odd - guaranteed not divisible by any power of two, so the hash tables 310// can be any power of two with good results. Also, I choose multipliers 311// that have only 2 bits set (the low is always set to be odd) so 312// multiplication requires only shifts and adds. Characters are required to 313// be in the range 0-127 (I double & add 1 to force oddness). Keys are 314// limited to MAXID characters in length. Experimental evidence on 150K of 315// C text shows excellent spreading of values for any size hash table. 316int hashstr(const void *t) { 317 register char c, k = 0; 318 register int32 sum = 0; 319 register const char *s = (const char *)t; 320 321 while( ((c = *s++) != '\0') && (k < MAXID-1) ) { // Get characters till null or MAXID-1 322 c = (c<<1)+1; // Characters are always odd! 323 sum += c + (c<<shft[k++]); // Universal hash function 324 } 325 return (int)((sum+xsum[k]) >> 1); // Hash key, un-modulo'd table size 326} 327 328//------------------------------hashptr-------------------------------------- 329// Slimey cheap hash function; no guaranteed performance. Better than the 330// default for pointers, especially on MS-DOS machines. 331int hashptr(const void *key) { 332#ifdef __TURBOC__ 333 return ((intptr_t)key >> 16); 334#else // __TURBOC__ 335 return ((intptr_t)key >> 2); 336#endif 337} 338 339// Slimey cheap hash function; no guaranteed performance. 340int hashkey(const void *key) { 341 return (intptr_t)key; 342} 343 344//------------------------------Key Comparator Functions--------------------- 345int32 cmpstr(const void *k1, const void *k2) { 346 return strcmp((const char *)k1,(const char *)k2); 347} 348 349// Cheap key comparator. 350int32 cmpkey(const void *key1, const void *key2) { 351 if (key1 == key2) return 0; 352 intptr_t delta = (intptr_t)key1 - (intptr_t)key2; 353 if (delta > 0) return 1; 354 return -1; 355} 356 357//============================================================================= 358//------------------------------reset------------------------------------------ 359// Create an iterator and initialize the first variables. 360void DictI::reset( const Dict *dict ) { 361 _d = dict; // The dictionary 362 _i = (uint)-1; // Before the first bin 363 _j = 0; // Nothing left in the current bin 364 ++(*this); // Step to first real value 365} 366 367//------------------------------next------------------------------------------- 368// Find the next key-value pair in the dictionary, or return a NULL key and 369// value. 370void DictI::operator ++(void) { 371 if( _j-- ) { // Still working in current bin? 372 _key = _d->_bin[_i]._keyvals[_j+_j]; 373 _value = _d->_bin[_i]._keyvals[_j+_j+1]; 374 return; 375 } 376 377 while( ++_i < _d->_size ) { // Else scan for non-zero bucket 378 _j = _d->_bin[_i]._cnt; 379 if( !_j ) continue; 380 _j--; 381 _key = _d->_bin[_i]._keyvals[_j+_j]; 382 _value = _d->_bin[_i]._keyvals[_j+_j+1]; 383 return; 384 } 385 _key = _value = NULL; 386} 387