| 1 | /* Implement a cached obstack. |
| 2 | Written by Fred Fish <fnf@cygnus.com> |
| 3 | Rewritten by Jim Blandy <jimb@cygnus.com> |
| 4 | Copyright 1999 Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of GDB. |
| 7 | |
| 8 | This program is free software; you can redistribute it and/or modify |
| 9 | it under the terms of the GNU General Public License as published by |
| 10 | the Free Software Foundation; either version 2 of the License, or |
| 11 | (at your option) any later version. |
| 12 | |
| 13 | This program is distributed in the hope that it will be useful, |
| 14 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 15 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 16 | GNU General Public License for more details. |
| 17 | |
| 18 | You should have received a copy of the GNU General Public License |
| 19 | along with this program; if not, write to the Free Software |
| 20 | Foundation, Inc., 59 Temple Place - Suite 330, |
| 21 | Boston, MA 02111-1307, USA. */ |
| 22 | |
| 23 | #include <stddef.h> |
| 24 | #include <stdlib.h> |
| 25 | |
| 26 | #include "defs.h" |
| 27 | #include "obstack.h" |
| 28 | #include "bcache.h" |
| 29 | #include "gdb_string.h" /* For memcpy declaration */ |
| 30 | |
| 31 | /* The old hash function was stolen from SDBM. This is what DB 3.0 uses now, |
| 32 | * and is better than the old one. |
| 33 | */ |
| 34 | \f |
| 35 | unsigned long |
| 36 | hash(void *addr, int length) |
| 37 | { |
| 38 | const unsigned char *k, *e; |
| 39 | unsigned long h; |
| 40 | |
| 41 | k = (const unsigned char *)addr; |
| 42 | e = k+length; |
| 43 | for (h=0; k< e;++k) |
| 44 | { |
| 45 | h *=16777619; |
| 46 | h ^= *k; |
| 47 | } |
| 48 | return (h); |
| 49 | } |
| 50 | \f |
| 51 | /* Growing the bcache's hash table. */ |
| 52 | |
| 53 | /* If the average chain length grows beyond this, then we want to |
| 54 | resize our hash table. */ |
| 55 | #define CHAIN_LENGTH_THRESHOLD (5) |
| 56 | |
| 57 | static void |
| 58 | expand_hash_table (struct bcache *bcache) |
| 59 | { |
| 60 | /* A table of good hash table sizes. Whenever we grow, we pick the |
| 61 | next larger size from this table. sizes[i] is close to 1 << (i+10), |
| 62 | so we roughly double the table size each time. After we fall off |
| 63 | the end of this table, we just double. Don't laugh --- there have |
| 64 | been executables sighted with a gigabyte of debug info. */ |
| 65 | static unsigned long sizes[] = { |
| 66 | 1021, 2053, 4099, 8191, 16381, 32771, |
| 67 | 65537, 131071, 262144, 524287, 1048573, 2097143, |
| 68 | 4194301, 8388617, 16777213, 33554467, 67108859, 134217757, |
| 69 | 268435459, 536870923, 1073741827, 2147483659UL |
| 70 | }; |
| 71 | unsigned int new_num_buckets; |
| 72 | struct bstring **new_buckets; |
| 73 | unsigned int i; |
| 74 | |
| 75 | /* Find the next size. */ |
| 76 | new_num_buckets = bcache->num_buckets * 2; |
| 77 | for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++) |
| 78 | if (sizes[i] > bcache->num_buckets) |
| 79 | { |
| 80 | new_num_buckets = sizes[i]; |
| 81 | break; |
| 82 | } |
| 83 | |
| 84 | /* Allocate the new table. */ |
| 85 | { |
| 86 | size_t new_size = new_num_buckets * sizeof (new_buckets[0]); |
| 87 | new_buckets = (struct bstring **) xmalloc (new_size); |
| 88 | memset (new_buckets, 0, new_size); |
| 89 | |
| 90 | bcache->structure_size -= (bcache->num_buckets |
| 91 | * sizeof (bcache->bucket[0])); |
| 92 | bcache->structure_size += new_size; |
| 93 | } |
| 94 | |
| 95 | /* Rehash all existing strings. */ |
| 96 | for (i = 0; i < bcache->num_buckets; i++) |
| 97 | { |
| 98 | struct bstring *s, *next; |
| 99 | |
| 100 | for (s = bcache->bucket[i]; s; s = next) |
| 101 | { |
| 102 | struct bstring **new_bucket; |
| 103 | next = s->next; |
| 104 | |
| 105 | new_bucket = &new_buckets[(hash (&s->d.data, s->length) |
| 106 | % new_num_buckets)]; |
| 107 | s->next = *new_bucket; |
| 108 | *new_bucket = s; |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | /* Plug in the new table. */ |
| 113 | if (bcache->bucket) |
| 114 | xfree (bcache->bucket); |
| 115 | bcache->bucket = new_buckets; |
| 116 | bcache->num_buckets = new_num_buckets; |
| 117 | } |
| 118 | |
| 119 | \f |
| 120 | /* Looking up things in the bcache. */ |
| 121 | |
| 122 | /* The number of bytes needed to allocate a struct bstring whose data |
| 123 | is N bytes long. */ |
| 124 | #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n)) |
| 125 | |
| 126 | /* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has |
| 127 | never seen those bytes before, add a copy of them to BCACHE. In |
| 128 | either case, return a pointer to BCACHE's copy of that string. */ |
| 129 | void * |
| 130 | bcache (void *addr, int length, struct bcache *bcache) |
| 131 | { |
| 132 | int hash_index; |
| 133 | struct bstring *s; |
| 134 | |
| 135 | /* If our average chain length is too high, expand the hash table. */ |
| 136 | if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD) |
| 137 | expand_hash_table (bcache); |
| 138 | |
| 139 | bcache->total_count++; |
| 140 | bcache->total_size += length; |
| 141 | |
| 142 | hash_index = hash (addr, length) % bcache->num_buckets; |
| 143 | |
| 144 | /* Search the hash bucket for a string identical to the caller's. */ |
| 145 | for (s = bcache->bucket[hash_index]; s; s = s->next) |
| 146 | if (s->length == length |
| 147 | && ! memcmp (&s->d.data, addr, length)) |
| 148 | return &s->d.data; |
| 149 | |
| 150 | /* The user's string isn't in the list. Insert it after *ps. */ |
| 151 | { |
| 152 | struct bstring *new |
| 153 | = obstack_alloc (&bcache->cache, BSTRING_SIZE (length)); |
| 154 | memcpy (&new->d.data, addr, length); |
| 155 | new->length = length; |
| 156 | new->next = bcache->bucket[hash_index]; |
| 157 | bcache->bucket[hash_index] = new; |
| 158 | |
| 159 | bcache->unique_count++; |
| 160 | bcache->unique_size += length; |
| 161 | bcache->structure_size += BSTRING_SIZE (length); |
| 162 | |
| 163 | return &new->d.data; |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | \f |
| 168 | /* Freeing bcaches. */ |
| 169 | |
| 170 | /* Free all the storage associated with BCACHE. */ |
| 171 | void |
| 172 | free_bcache (struct bcache *bcache) |
| 173 | { |
| 174 | obstack_free (&bcache->cache, 0); |
| 175 | if (bcache->bucket) |
| 176 | xfree (bcache->bucket); |
| 177 | |
| 178 | /* This isn't necessary, but at least the bcache is always in a |
| 179 | consistent state. */ |
| 180 | memset (bcache, 0, sizeof (*bcache)); |
| 181 | } |
| 182 | |
| 183 | |
| 184 | \f |
| 185 | /* Printing statistics. */ |
| 186 | |
| 187 | static int |
| 188 | compare_ints (const void *ap, const void *bp) |
| 189 | { |
| 190 | /* Because we know we're comparing two ints which are positive, |
| 191 | there's no danger of overflow here. */ |
| 192 | return * (int *) ap - * (int *) bp; |
| 193 | } |
| 194 | |
| 195 | |
| 196 | static void |
| 197 | print_percentage (int portion, int total) |
| 198 | { |
| 199 | if (total == 0) |
| 200 | printf_filtered ("(not applicable)\n"); |
| 201 | else |
| 202 | printf_filtered ("%3d%%\n", portion * 100 / total); |
| 203 | } |
| 204 | |
| 205 | |
| 206 | /* Print statistics on BCACHE's memory usage and efficacity at |
| 207 | eliminating duplication. NAME should describe the kind of data |
| 208 | BCACHE holds. Statistics are printed using `printf_filtered' and |
| 209 | its ilk. */ |
| 210 | void |
| 211 | print_bcache_statistics (struct bcache *c, char *type) |
| 212 | { |
| 213 | int occupied_buckets; |
| 214 | int max_chain_length; |
| 215 | int median_chain_length; |
| 216 | |
| 217 | /* Count the number of occupied buckets, and measure chain lengths. */ |
| 218 | { |
| 219 | unsigned int b; |
| 220 | int *chain_length |
| 221 | = (int *) alloca (c->num_buckets * sizeof (*chain_length)); |
| 222 | |
| 223 | occupied_buckets = 0; |
| 224 | |
| 225 | for (b = 0; b < c->num_buckets; b++) |
| 226 | { |
| 227 | struct bstring *s = c->bucket[b]; |
| 228 | |
| 229 | chain_length[b] = 0; |
| 230 | |
| 231 | if (s) |
| 232 | { |
| 233 | occupied_buckets++; |
| 234 | |
| 235 | while (s) |
| 236 | { |
| 237 | chain_length[b]++; |
| 238 | s = s->next; |
| 239 | } |
| 240 | } |
| 241 | } |
| 242 | |
| 243 | /* To compute the median, we need the set of chain lengths sorted. */ |
| 244 | qsort (chain_length, c->num_buckets, sizeof (chain_length[0]), |
| 245 | compare_ints); |
| 246 | |
| 247 | if (c->num_buckets > 0) |
| 248 | { |
| 249 | max_chain_length = chain_length[c->num_buckets - 1]; |
| 250 | median_chain_length = chain_length[c->num_buckets / 2]; |
| 251 | } |
| 252 | else |
| 253 | { |
| 254 | max_chain_length = 0; |
| 255 | median_chain_length = 0; |
| 256 | } |
| 257 | } |
| 258 | |
| 259 | printf_filtered (" Cached '%s' statistics:\n", type); |
| 260 | printf_filtered (" Total object count: %ld\n", c->total_count); |
| 261 | printf_filtered (" Unique object count: %lu\n", c->unique_count); |
| 262 | printf_filtered (" Percentage of duplicates, by count: "); |
| 263 | print_percentage (c->total_count - c->unique_count, c->total_count); |
| 264 | printf_filtered ("\n"); |
| 265 | |
| 266 | printf_filtered (" Total object size: %ld\n", c->total_size); |
| 267 | printf_filtered (" Unique object size: %ld\n", c->unique_size); |
| 268 | printf_filtered (" Percentage of duplicates, by size: "); |
| 269 | print_percentage (c->total_size - c->unique_size, c->total_size); |
| 270 | printf_filtered ("\n"); |
| 271 | |
| 272 | printf_filtered (" Total memory used by bcache, including overhead: %ld\n", |
| 273 | c->structure_size); |
| 274 | printf_filtered (" Percentage memory overhead: "); |
| 275 | print_percentage (c->structure_size - c->unique_size, c->unique_size); |
| 276 | printf_filtered (" Net memory savings: "); |
| 277 | print_percentage (c->total_size - c->structure_size, c->total_size); |
| 278 | printf_filtered ("\n"); |
| 279 | |
| 280 | printf_filtered (" Hash table size: %3d\n", c->num_buckets); |
| 281 | printf_filtered (" Hash table population: "); |
| 282 | print_percentage (occupied_buckets, c->num_buckets); |
| 283 | printf_filtered (" Median hash chain length: %3d\n", |
| 284 | median_chain_length); |
| 285 | printf_filtered (" Average hash chain length: "); |
| 286 | if (c->num_buckets > 0) |
| 287 | printf_filtered ("%3lu\n", c->unique_count / c->num_buckets); |
| 288 | else |
| 289 | printf_filtered ("(not applicable)\n"); |
| 290 | printf_filtered (" Maximum hash chain length: %3d\n", max_chain_length); |
| 291 | printf_filtered ("\n"); |
| 292 | } |