1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/src/share/vm/adlc/dict2.cpp Sat Dec 01 00:00:00 2007 +0000 1.3 @@ -0,0 +1,353 @@ 1.4 +/* 1.5 + * Copyright 1998-2002 Sun Microsystems, Inc. All Rights Reserved. 1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 1.7 + * 1.8 + * This code is free software; you can redistribute it and/or modify it 1.9 + * under the terms of the GNU General Public License version 2 only, as 1.10 + * published by the Free Software Foundation. 1.11 + * 1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT 1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 1.15 + * version 2 for more details (a copy is included in the LICENSE file that 1.16 + * accompanied this code). 1.17 + * 1.18 + * You should have received a copy of the GNU General Public License version 1.19 + * 2 along with this work; if not, write to the Free Software Foundation, 1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 1.21 + * 1.22 + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 1.23 + * CA 95054 USA or visit www.sun.com if you need additional information or 1.24 + * have any questions. 1.25 + * 1.26 + */ 1.27 + 1.28 +// Dictionaries - An Abstract Data Type 1.29 + 1.30 +#include "adlc.hpp" 1.31 + 1.32 +// #include "dict.hpp" 1.33 + 1.34 + 1.35 +//------------------------------data----------------------------------------- 1.36 +// String hash tables 1.37 +#define MAXID 20 1.38 +static char initflag = 0; // True after 1st initialization 1.39 +static char shft[MAXID] = {1,2,3,4,5,6,7,1,2,3,4,5,6,7,1,2,3,4,5,6}; 1.40 +static short xsum[MAXID + 1]; 1.41 + 1.42 +//------------------------------bucket--------------------------------------- 1.43 +class bucket { 1.44 +public: 1.45 + int _cnt, _max; // Size of bucket 1.46 + const void **_keyvals; // Array of keys and values 1.47 +}; 1.48 + 1.49 +//------------------------------Dict----------------------------------------- 1.50 +// The dictionary is kept has a hash table. The hash table is a even power 1.51 +// of two, for nice modulo operations. Each bucket in the hash table points 1.52 +// to a linear list of key-value pairs; each key & value is just a (void *). 1.53 +// The list starts with a count. A hash lookup finds the list head, then a 1.54 +// simple linear scan finds the key. If the table gets too full, it's 1.55 +// doubled in size; the total amount of EXTRA times all hash functions are 1.56 +// computed for the doubling is no more than the current size - thus the 1.57 +// doubling in size costs no more than a constant factor in speed. 1.58 +Dict::Dict(CmpKey initcmp, Hash inithash) : _hash(inithash), _cmp(initcmp), _arena(NULL) { 1.59 + init(); 1.60 +} 1.61 + 1.62 +Dict::Dict(CmpKey initcmp, Hash inithash, Arena *arena) : _hash(inithash), _cmp(initcmp), _arena(arena) { 1.63 + init(); 1.64 +} 1.65 + 1.66 +void Dict::init() { 1.67 + int i; 1.68 + 1.69 + // Precompute table of null character hashes 1.70 + if( !initflag ) { // Not initializated yet? 1.71 + xsum[0] = (1<<shft[0])+1; // Initialize 1.72 + for( i = 1; i < MAXID + 1; i++) { 1.73 + xsum[i] = (1<<shft[i])+1+xsum[i-1]; 1.74 + } 1.75 + initflag = 1; // Never again 1.76 + } 1.77 + 1.78 + _size = 16; // Size is a power of 2 1.79 + _cnt = 0; // Dictionary is empty 1.80 + _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 1.81 + memset(_bin,0,sizeof(bucket)*_size); 1.82 +} 1.83 + 1.84 +//------------------------------~Dict------------------------------------------ 1.85 +// Delete an existing dictionary. 1.86 +Dict::~Dict() { 1.87 +} 1.88 + 1.89 +//------------------------------Clear---------------------------------------- 1.90 +// Zap to empty; ready for re-use 1.91 +void Dict::Clear() { 1.92 + _cnt = 0; // Empty contents 1.93 + for( int i=0; i<_size; i++ ) 1.94 + _bin[i]._cnt = 0; // Empty buckets, but leave allocated 1.95 + // Leave _size & _bin alone, under the assumption that dictionary will 1.96 + // grow to this size again. 1.97 +} 1.98 + 1.99 +//------------------------------doubhash--------------------------------------- 1.100 +// Double hash table size. If can't do so, just suffer. If can, then run 1.101 +// thru old hash table, moving things to new table. Note that since hash 1.102 +// table doubled, exactly 1 new bit is exposed in the mask - so everything 1.103 +// in the old table ends up on 1 of two lists in the new table; a hi and a 1.104 +// lo list depending on the value of the bit. 1.105 +void Dict::doubhash(void) { 1.106 + int oldsize = _size; 1.107 + _size <<= 1; // Double in size 1.108 + _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*oldsize, sizeof(bucket)*_size ); 1.109 + memset( &_bin[oldsize], 0, oldsize*sizeof(bucket) ); 1.110 + // Rehash things to spread into new table 1.111 + for( int i=0; i < oldsize; i++) { // For complete OLD table do 1.112 + bucket *b = &_bin[i]; // Handy shortcut for _bin[i] 1.113 + if( !b->_keyvals ) continue; // Skip empties fast 1.114 + 1.115 + bucket *nb = &_bin[i+oldsize]; // New bucket shortcut 1.116 + int j = b->_max; // Trim new bucket to nearest power of 2 1.117 + while( j > b->_cnt ) j >>= 1; // above old bucket _cnt 1.118 + if( !j ) j = 1; // Handle zero-sized buckets 1.119 + nb->_max = j<<1; 1.120 + // Allocate worst case space for key-value pairs 1.121 + nb->_keyvals = (const void**)_arena->Amalloc_4( sizeof(void *)*nb->_max*2 ); 1.122 + int nbcnt = 0; 1.123 + 1.124 + for( j=0; j<b->_cnt; j++ ) { // Rehash all keys in this bucket 1.125 + const void *key = b->_keyvals[j+j]; 1.126 + if( (_hash( key ) & (_size-1)) != i ) { // Moving to hi bucket? 1.127 + nb->_keyvals[nbcnt+nbcnt] = key; 1.128 + nb->_keyvals[nbcnt+nbcnt+1] = b->_keyvals[j+j+1]; 1.129 + nb->_cnt = nbcnt = nbcnt+1; 1.130 + b->_cnt--; // Remove key/value from lo bucket 1.131 + b->_keyvals[j+j ] = b->_keyvals[b->_cnt+b->_cnt ]; 1.132 + b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1]; 1.133 + j--; // Hash compacted element also 1.134 + } 1.135 + } // End of for all key-value pairs in bucket 1.136 + } // End of for all buckets 1.137 + 1.138 + 1.139 +} 1.140 + 1.141 +//------------------------------Dict----------------------------------------- 1.142 +// Deep copy a dictionary. 1.143 +Dict::Dict( const Dict &d ) : _size(d._size), _cnt(d._cnt), _hash(d._hash),_cmp(d._cmp), _arena(d._arena) { 1.144 + _bin = (bucket*)_arena->Amalloc_4(sizeof(bucket)*_size); 1.145 + memcpy( _bin, d._bin, sizeof(bucket)*_size ); 1.146 + for( int i=0; i<_size; i++ ) { 1.147 + if( !_bin[i]._keyvals ) continue; 1.148 + _bin[i]._keyvals=(const void**)_arena->Amalloc_4( sizeof(void *)*_bin[i]._max*2); 1.149 + memcpy( _bin[i]._keyvals, d._bin[i]._keyvals,_bin[i]._cnt*2*sizeof(void*)); 1.150 + } 1.151 +} 1.152 + 1.153 +//------------------------------Dict----------------------------------------- 1.154 +// Deep copy a dictionary. 1.155 +Dict &Dict::operator =( const Dict &d ) { 1.156 + if( _size < d._size ) { // If must have more buckets 1.157 + _arena = d._arena; 1.158 + _bin = (bucket*)_arena->Arealloc( _bin, sizeof(bucket)*_size, sizeof(bucket)*d._size ); 1.159 + memset( &_bin[_size], 0, (d._size-_size)*sizeof(bucket) ); 1.160 + _size = d._size; 1.161 + } 1.162 + for( int i=0; i<_size; i++ ) // All buckets are empty 1.163 + _bin[i]._cnt = 0; // But leave bucket allocations alone 1.164 + _cnt = d._cnt; 1.165 + *(Hash*)(&_hash) = d._hash; 1.166 + *(CmpKey*)(&_cmp) = d._cmp; 1.167 + for(int k=0; k<_size; k++ ) { 1.168 + bucket *b = &d._bin[k]; // Shortcut to source bucket 1.169 + for( int j=0; j<b->_cnt; j++ ) 1.170 + Insert( b->_keyvals[j+j], b->_keyvals[j+j+1] ); 1.171 + } 1.172 + return *this; 1.173 +} 1.174 + 1.175 +//------------------------------Insert--------------------------------------- 1.176 +// Insert or replace a key/value pair in the given dictionary. If the 1.177 +// dictionary is too full, it's size is doubled. The prior value being 1.178 +// replaced is returned (NULL if this is a 1st insertion of that key). If 1.179 +// an old value is found, it's swapped with the prior key-value pair on the 1.180 +// list. This moves a commonly searched-for value towards the list head. 1.181 +const void *Dict::Insert(const void *key, const void *val) { 1.182 + int hash = _hash( key ); // Get hash key 1.183 + int i = hash & (_size-1); // Get hash key, corrected for size 1.184 + bucket *b = &_bin[i]; // Handy shortcut 1.185 + for( int j=0; j<b->_cnt; j++ ) 1.186 + if( !_cmp(key,b->_keyvals[j+j]) ) { 1.187 + const void *prior = b->_keyvals[j+j+1]; 1.188 + b->_keyvals[j+j ] = key; // Insert current key-value 1.189 + b->_keyvals[j+j+1] = val; 1.190 + return prior; // Return prior 1.191 + } 1.192 + 1.193 + if( ++_cnt > _size ) { // Hash table is full 1.194 + doubhash(); // Grow whole table if too full 1.195 + i = hash & (_size-1); // Rehash 1.196 + b = &_bin[i]; // Handy shortcut 1.197 + } 1.198 + if( b->_cnt == b->_max ) { // Must grow bucket? 1.199 + if( !b->_keyvals ) { 1.200 + b->_max = 2; // Initial bucket size 1.201 + b->_keyvals = (const void**)_arena->Amalloc_4( sizeof(void *)*b->_max*2 ); 1.202 + } else { 1.203 + b->_keyvals = (const void**)_arena->Arealloc( b->_keyvals, sizeof(void *)*b->_max*2, sizeof(void *)*b->_max*4 ); 1.204 + b->_max <<= 1; // Double bucket 1.205 + } 1.206 + } 1.207 + b->_keyvals[b->_cnt+b->_cnt ] = key; 1.208 + b->_keyvals[b->_cnt+b->_cnt+1] = val; 1.209 + b->_cnt++; 1.210 + return NULL; // Nothing found prior 1.211 +} 1.212 + 1.213 +//------------------------------Delete--------------------------------------- 1.214 +// Find & remove a value from dictionary. Return old value. 1.215 +const void *Dict::Delete(void *key) { 1.216 + int i = _hash( key ) & (_size-1); // Get hash key, corrected for size 1.217 + bucket *b = &_bin[i]; // Handy shortcut 1.218 + for( int j=0; j<b->_cnt; j++ ) 1.219 + if( !_cmp(key,b->_keyvals[j+j]) ) { 1.220 + const void *prior = b->_keyvals[j+j+1]; 1.221 + b->_cnt--; // Remove key/value from lo bucket 1.222 + b->_keyvals[j+j ] = b->_keyvals[b->_cnt+b->_cnt ]; 1.223 + b->_keyvals[j+j+1] = b->_keyvals[b->_cnt+b->_cnt+1]; 1.224 + _cnt--; // One less thing in table 1.225 + return prior; 1.226 + } 1.227 + return NULL; 1.228 +} 1.229 + 1.230 +//------------------------------FindDict------------------------------------- 1.231 +// Find a key-value pair in the given dictionary. If not found, return NULL. 1.232 +// If found, move key-value pair towards head of list. 1.233 +const void *Dict::operator [](const void *key) const { 1.234 + int i = _hash( key ) & (_size-1); // Get hash key, corrected for size 1.235 + bucket *b = &_bin[i]; // Handy shortcut 1.236 + for( int j=0; j<b->_cnt; j++ ) 1.237 + if( !_cmp(key,b->_keyvals[j+j]) ) 1.238 + return b->_keyvals[j+j+1]; 1.239 + return NULL; 1.240 +} 1.241 + 1.242 +//------------------------------CmpDict-------------------------------------- 1.243 +// CmpDict compares two dictionaries; they must have the same keys (their 1.244 +// keys must match using CmpKey) and they must have the same values (pointer 1.245 +// comparison). If so 1 is returned, if not 0 is returned. 1.246 +int Dict::operator ==(const Dict &d2) const { 1.247 + if( _cnt != d2._cnt ) return 0; 1.248 + if( _hash != d2._hash ) return 0; 1.249 + if( _cmp != d2._cmp ) return 0; 1.250 + for( int i=0; i < _size; i++) { // For complete hash table do 1.251 + bucket *b = &_bin[i]; // Handy shortcut 1.252 + if( b->_cnt != d2._bin[i]._cnt ) return 0; 1.253 + if( memcmp(b->_keyvals, d2._bin[i]._keyvals, b->_cnt*2*sizeof(void*) ) ) 1.254 + return 0; // Key-value pairs must match 1.255 + } 1.256 + return 1; // All match, is OK 1.257 +} 1.258 + 1.259 + 1.260 +//------------------------------print---------------------------------------- 1.261 +static void printvoid(const void* x) { printf("%p", x); } 1.262 +void Dict::print() { 1.263 + print(printvoid, printvoid); 1.264 +} 1.265 +void Dict::print(PrintKeyOrValue print_key, PrintKeyOrValue print_value) { 1.266 + for( int i=0; i < _size; i++) { // For complete hash table do 1.267 + bucket *b = &_bin[i]; // Handy shortcut 1.268 + for( int j=0; j<b->_cnt; j++ ) { 1.269 + print_key( b->_keyvals[j+j ]); 1.270 + printf(" -> "); 1.271 + print_value(b->_keyvals[j+j+1]); 1.272 + printf("\n"); 1.273 + } 1.274 + } 1.275 +} 1.276 + 1.277 +//------------------------------Hashing Functions---------------------------- 1.278 +// Convert string to hash key. This algorithm implements a universal hash 1.279 +// function with the multipliers frozen (ok, so it's not universal). The 1.280 +// multipliers (and allowable characters) are all odd, so the resultant sum 1.281 +// is odd - guarenteed not divisible by any power of two, so the hash tables 1.282 +// can be any power of two with good results. Also, I choose multipliers 1.283 +// that have only 2 bits set (the low is always set to be odd) so 1.284 +// multiplication requires only shifts and adds. Characters are required to 1.285 +// be in the range 0-127 (I double & add 1 to force oddness). Keys are 1.286 +// limited to MAXID characters in length. Experimental evidence on 150K of 1.287 +// C text shows excellent spreading of values for any size hash table. 1.288 +int hashstr(const void *t) { 1.289 + register char c, k = 0; 1.290 + register int sum = 0; 1.291 + register const char *s = (const char *)t; 1.292 + 1.293 + while( ((c = s[k]) != '\0') && (k < MAXID-1) ) { // Get characters till nul 1.294 + c = (c<<1)+1; // Characters are always odd! 1.295 + sum += c + (c<<shft[k++]); // Universal hash function 1.296 + } 1.297 + assert( k < (MAXID + 1), "Exceeded maximum name length"); 1.298 + return (int)((sum+xsum[k]) >> 1); // Hash key, un-modulo'd table size 1.299 +} 1.300 + 1.301 +//------------------------------hashptr-------------------------------------- 1.302 +// Slimey cheap hash function; no guarenteed performance. Better than the 1.303 +// default for pointers, especially on MS-DOS machines. 1.304 +int hashptr(const void *key) { 1.305 +#ifdef __TURBOC__ 1.306 + return (int)((intptr_t)key >> 16); 1.307 +#else // __TURBOC__ 1.308 + return (int)((intptr_t)key >> 2); 1.309 +#endif 1.310 +} 1.311 + 1.312 +// Slimey cheap hash function; no guarenteed performance. 1.313 +int hashkey(const void *key) { 1.314 + return (int)((intptr_t)key); 1.315 +} 1.316 + 1.317 +//------------------------------Key Comparator Functions--------------------- 1.318 +int cmpstr(const void *k1, const void *k2) { 1.319 + return strcmp((const char *)k1,(const char *)k2); 1.320 +} 1.321 + 1.322 +// Slimey cheap key comparator. 1.323 +int cmpkey(const void *key1, const void *key2) { 1.324 + return (int)((intptr_t)key1 - (intptr_t)key2); 1.325 +} 1.326 + 1.327 +//============================================================================= 1.328 +//------------------------------reset------------------------------------------ 1.329 +// Create an iterator and initialize the first variables. 1.330 +void DictI::reset( const Dict *dict ) { 1.331 + _d = dict; // The dictionary 1.332 + _i = (int)-1; // Before the first bin 1.333 + _j = 0; // Nothing left in the current bin 1.334 + ++(*this); // Step to first real value 1.335 +} 1.336 + 1.337 +//------------------------------next------------------------------------------- 1.338 +// Find the next key-value pair in the dictionary, or return a NULL key and 1.339 +// value. 1.340 +void DictI::operator ++(void) { 1.341 + if( _j-- ) { // Still working in current bin? 1.342 + _key = _d->_bin[_i]._keyvals[_j+_j]; 1.343 + _value = _d->_bin[_i]._keyvals[_j+_j+1]; 1.344 + return; 1.345 + } 1.346 + 1.347 + while( ++_i < _d->_size ) { // Else scan for non-zero bucket 1.348 + _j = _d->_bin[_i]._cnt; 1.349 + if( !_j ) continue; 1.350 + _j--; 1.351 + _key = _d->_bin[_i]._keyvals[_j+_j]; 1.352 + _value = _d->_bin[_i]._keyvals[_j+_j+1]; 1.353 + return; 1.354 + } 1.355 + _key = _value = NULL; 1.356 +}