gdb/bcache.c

   1 /* Implement a cached obstack.
   2    Written by Fred Fish <fnf@cygnus.com>
   3    Rewritten by Jim Blandy <jimb@cygnus.com>
   4
   5    Copyright (C) 1999, 2000, 2002, 2003, 2007, 2008
   6    Free Software Foundation, Inc.
   7
   8    This file is part of GDB.
   9
  10    This program is free software; you can redistribute it and/or modify
  11    it under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 3 of the License, or
  13    (at your option) any later version.
  14
  15    This program is distributed in the hope that it will be useful,
  16    but WITHOUT ANY WARRANTY; without even the implied warranty of
  17    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18    GNU General Public License for more details.
  19
  20    You should have received a copy of the GNU General Public License
  21    along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
  22
  23 #include "defs.h"
  24 #include "gdb_obstack.h"
  25 #include "bcache.h"
  26 #include "gdb_string.h"         /* For memcpy declaration */
  27 #include "gdb_assert.h"
  28
  29 #include <stddef.h>
  30 #include <stdlib.h>
  31
  32 /* The type used to hold a single bcache string.  The user data is
  33    stored in d.data.  Since it can be any type, it needs to have the
  34    same alignment as the most strict alignment of any type on the host
  35    machine.  I don't know of any really correct way to do this in
  36    stock ANSI C, so just do it the same way obstack.h does.  */
  37
  38 struct bstring
  39 {
  40   /* Hash chain.  */
  41   struct bstring *next;
  42   /* Assume the data length is no more than 64k.  */
  43   unsigned short length;
  44   /* The half hash hack.  This contains the upper 16 bits of the hash
  45      value and is used as a pre-check when comparing two strings and
  46      avoids the need to do length or memcmp calls.  It proves to be
  47      roughly 100% effective.  */
  48   unsigned short half_hash;
  49
  50   union
  51   {
  52     char data[1];
  53     double dummy;
  54   }
  55   d;
  56 };
  57
  58
  59 /* The structure for a bcache itself.  The bcache is initialized, in
  60    bcache_xmalloc(), by filling it with zeros and then setting the
  61    corresponding obstack's malloc() and free() methods.  */
  62
  63 struct bcache
  64 {
  65   /* All the bstrings are allocated here.  */
  66   struct obstack cache;
  67
  68   /* How many hash buckets we're using.  */
  69   unsigned int num_buckets;
  70
  71   /* Hash buckets.  This table is allocated using malloc, so when we
  72      grow the table we can return the old table to the system.  */
  73   struct bstring **bucket;
  74
  75   /* Statistics.  */
  76   unsigned long unique_count;   /* number of unique strings */
  77   long total_count;     /* total number of strings cached, including dups */
  78   long unique_size;     /* size of unique strings, in bytes */
  79   long total_size;      /* total number of bytes cached, including dups */
  80   long structure_size;  /* total size of bcache, including infrastructure */
  81   /* Number of times that the hash table is expanded and hence
  82      re-built, and the corresponding number of times that a string is
  83      [re]hashed as part of entering it into the expanded table.  The
  84      total number of hashes can be computed by adding TOTAL_COUNT to
  85      expand_hash_count.  */
  86   unsigned long expand_count;
  87   unsigned long expand_hash_count;
  88   /* Number of times that the half-hash compare hit (compare the upper
  89      16 bits of hash values) hit, but the corresponding combined
  90      length/data compare missed.  */
  91   unsigned long half_hash_miss_count;
  92 };
  93
  94 /* The old hash function was stolen from SDBM. This is what DB 3.0 uses now,
  95  * and is better than the old one.
  96  */
  97 \f
  98 unsigned long
  99 hash(const void *addr, int length)
 100 {
 101                 const unsigned char *k, *e;
 102                 unsigned long h;
 103
 104                 k = (const unsigned char *)addr;
 105                 e = k+length;
 106                 for (h=0; k< e;++k)
 107                 {
 108                                 h *=16777619;
 109                                 h ^= *k;
 110                 }
 111                 return (h);
 112 }
 113 \f
 114 /* Growing the bcache's hash table.  */
 115
 116 /* If the average chain length grows beyond this, then we want to
 117    resize our hash table.  */
 118 #define CHAIN_LENGTH_THRESHOLD (5)
 119
 120 static void
 121 expand_hash_table (struct bcache *bcache)
 122 {
 123   /* A table of good hash table sizes.  Whenever we grow, we pick the
 124      next larger size from this table.  sizes[i] is close to 1 << (i+10),
 125      so we roughly double the table size each time.  After we fall off
 126      the end of this table, we just double.  Don't laugh --- there have
 127      been executables sighted with a gigabyte of debug info.  */
 128   static unsigned long sizes[] = {
 129     1021, 2053, 4099, 8191, 16381, 32771,
 130     65537, 131071, 262144, 524287, 1048573, 2097143,
 131     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
 132     268435459, 536870923, 1073741827, 2147483659UL
 133   };
 134   unsigned int new_num_buckets;
 135   struct bstring **new_buckets;
 136   unsigned int i;
 137
 138   /* Count the stats.  Every unique item needs to be re-hashed and
 139      re-entered.  */
 140   bcache->expand_count++;
 141   bcache->expand_hash_count += bcache->unique_count;
 142
 143   /* Find the next size.  */
 144   new_num_buckets = bcache->num_buckets * 2;
 145   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 146     if (sizes[i] > bcache->num_buckets)
 147       {
 148         new_num_buckets = sizes[i];
 149         break;
 150       }
 151
 152   /* Allocate the new table.  */
 153   {
 154     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 155     new_buckets = (struct bstring **) xmalloc (new_size);
 156     memset (new_buckets, 0, new_size);
 157
 158     bcache->structure_size -= (bcache->num_buckets
 159                                * sizeof (bcache->bucket[0]));
 160     bcache->structure_size += new_size;
 161   }
 162
 163   /* Rehash all existing strings.  */
 164   for (i = 0; i < bcache->num_buckets; i++)
 165     {
 166       struct bstring *s, *next;
 167
 168       for (s = bcache->bucket[i]; s; s = next)
 169         {
 170           struct bstring **new_bucket;
 171           next = s->next;
 172
 173           new_bucket = &new_buckets[(hash (&s->d.data, s->length)
 174                                      % new_num_buckets)];
 175           s->next = *new_bucket;
 176           *new_bucket = s;
 177         }
 178     }
 179
 180   /* Plug in the new table.  */
 181   if (bcache->bucket)
 182     xfree (bcache->bucket);
 183   bcache->bucket = new_buckets;
 184   bcache->num_buckets = new_num_buckets;
 185 }
 186
 187 \f
 188 /* Looking up things in the bcache.  */
 189
 190 /* The number of bytes needed to allocate a struct bstring whose data
 191    is N bytes long.  */
 192 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 193
 194 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 195    never seen those bytes before, add a copy of them to BCACHE.  In
 196    either case, return a pointer to BCACHE's copy of that string.  */
 197 static void *
 198 bcache_data (const void *addr, int length, struct bcache *bcache)
 199 {
 200   return deprecated_bcache_added (addr, length, bcache, NULL);
 201 }
 202
 203
 204 void *
 205 deprecated_bcache (const void *addr, int length, struct bcache *bcache)
 206 {
 207   return bcache_data (addr, length, bcache);
 208 }
 209
 210 const void *
 211 bcache (const void *addr, int length, struct bcache *bcache)
 212 {
 213   return bcache_data (addr, length, bcache);
 214 }
 215
 216 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 217    never seen those bytes before, add a copy of them to BCACHE.  In
 218    either case, return a pointer to BCACHE's copy of that string.  If
 219    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 220    returning an old entry.  */
 221
 222 void *
 223 deprecated_bcache_added (const void *addr, int length, struct bcache *bcache,
 224                          int *added)
 225 {
 226   unsigned long full_hash;
 227   unsigned short half_hash;
 228   int hash_index;
 229   struct bstring *s;
 230
 231   if (added)
 232     *added = 0;
 233
 234   /* If our average chain length is too high, expand the hash table.  */
 235   if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
 236     expand_hash_table (bcache);
 237
 238   bcache->total_count++;
 239   bcache->total_size += length;
 240
 241   full_hash = hash (addr, length);
 242   half_hash = (full_hash >> 16);
 243   hash_index = full_hash % bcache->num_buckets;
 244
 245   /* Search the hash bucket for a string identical to the caller's.
 246      As a short-circuit first compare the upper part of each hash
 247      values.  */
 248   for (s = bcache->bucket[hash_index]; s; s = s->next)
 249     {
 250       if (s->half_hash == half_hash)
 251         {
 252           if (s->length == length
 253               && ! memcmp (&s->d.data, addr, length))
 254             return &s->d.data;
 255           else
 256             bcache->half_hash_miss_count++;
 257         }
 258     }
 259
 260   /* The user's string isn't in the list.  Insert it after *ps.  */
 261   {
 262     struct bstring *new
 263       = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
 264     memcpy (&new->d.data, addr, length);
 265     new->length = length;
 266     new->next = bcache->bucket[hash_index];
 267     new->half_hash = half_hash;
 268     bcache->bucket[hash_index] = new;
 269
 270     bcache->unique_count++;
 271     bcache->unique_size += length;
 272     bcache->structure_size += BSTRING_SIZE (length);
 273
 274     if (added)
 275       *added = 1;
 276
 277     return &new->d.data;
 278   }
 279 }
 280 \f
 281 /* Allocating and freeing bcaches.  */
 282
 283 struct bcache *
 284 bcache_xmalloc (void)
 285 {
 286   /* Allocate the bcache pre-zeroed.  */
 287   struct bcache *b = XCALLOC (1, struct bcache);
 288   /* We could use obstack_specify_allocation here instead, but
 289      gdb_obstack.h specifies the allocation/deallocation
 290      functions.  */
 291   obstack_init (&b->cache);
 292   return b;
 293 }
 294
 295 /* Free all the storage associated with BCACHE.  */
 296 void
 297 bcache_xfree (struct bcache *bcache)
 298 {
 299   if (bcache == NULL)
 300     return;
 301   obstack_free (&bcache->cache, 0);
 302   xfree (bcache->bucket);
 303   xfree (bcache);
 304 }
 305
 306
 307 \f
 308 /* Printing statistics.  */
 309
 310 static int
 311 compare_ints (const void *ap, const void *bp)
 312 {
 313   /* Because we know we're comparing two ints which are positive,
 314      there's no danger of overflow here.  */
 315   return * (int *) ap - * (int *) bp;
 316 }
 317
 318
 319 static void
 320 print_percentage (int portion, int total)
 321 {
 322   if (total == 0)
 323     /* i18n: Like "Percentage of duplicates, by count: (not applicable)" */
 324     printf_filtered (_("(not applicable)\n"));
 325   else
 326     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 327 }
 328
 329
 330 /* Print statistics on BCACHE's memory usage and efficacity at
 331    eliminating duplication.  NAME should describe the kind of data
 332    BCACHE holds.  Statistics are printed using `printf_filtered' and
 333    its ilk.  */
 334 void
 335 print_bcache_statistics (struct bcache *c, char *type)
 336 {
 337   int occupied_buckets;
 338   int max_chain_length;
 339   int median_chain_length;
 340   int max_entry_size;
 341   int median_entry_size;
 342
 343   /* Count the number of occupied buckets, tally the various string
 344      lengths, and measure chain lengths.  */
 345   {
 346     unsigned int b;
 347     int *chain_length = XCALLOC (c->num_buckets + 1, int);
 348     int *entry_size = XCALLOC (c->unique_count + 1, int);
 349     int stringi = 0;
 350
 351     occupied_buckets = 0;
 352
 353     for (b = 0; b < c->num_buckets; b++)
 354       {
 355         struct bstring *s = c->bucket[b];
 356
 357         chain_length[b] = 0;
 358
 359         if (s)
 360           {
 361             occupied_buckets++;
 362
 363             while (s)
 364               {
 365                 gdb_assert (b < c->num_buckets);
 366                 chain_length[b]++;
 367                 gdb_assert (stringi < c->unique_count);
 368                 entry_size[stringi++] = s->length;
 369                 s = s->next;
 370               }
 371           }
 372       }
 373
 374     /* To compute the median, we need the set of chain lengths sorted.  */
 375     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 376            compare_ints);
 377     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 378            compare_ints);
 379
 380     if (c->num_buckets > 0)
 381       {
 382         max_chain_length = chain_length[c->num_buckets - 1];
 383         median_chain_length = chain_length[c->num_buckets / 2];
 384       }
 385     else
 386       {
 387         max_chain_length = 0;
 388         median_chain_length = 0;
 389       }
 390     if (c->unique_count > 0)
 391       {
 392         max_entry_size = entry_size[c->unique_count - 1];
 393         median_entry_size = entry_size[c->unique_count / 2];
 394       }
 395     else
 396       {
 397         max_entry_size = 0;
 398         median_entry_size = 0;
 399       }
 400
 401     xfree (chain_length);
 402     xfree (entry_size);
 403   }
 404
 405   printf_filtered (_("  Cached '%s' statistics:\n"), type);
 406   printf_filtered (_("    Total object count:  %ld\n"), c->total_count);
 407   printf_filtered (_("    Unique object count: %lu\n"), c->unique_count);
 408   printf_filtered (_("    Percentage of duplicates, by count: "));
 409   print_percentage (c->total_count - c->unique_count, c->total_count);
 410   printf_filtered ("\n");
 411
 412   printf_filtered (_("    Total object size:   %ld\n"), c->total_size);
 413   printf_filtered (_("    Unique object size:  %ld\n"), c->unique_size);
 414   printf_filtered (_("    Percentage of duplicates, by size:  "));
 415   print_percentage (c->total_size - c->unique_size, c->total_size);
 416   printf_filtered ("\n");
 417
 418   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 419   printf_filtered (_("    Average entry size: "));
 420   if (c->unique_count > 0)
 421     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 422   else
 423     /* i18n: "Average entry size: (not applicable)" */
 424     printf_filtered (_("(not applicable)\n"));
 425   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 426   printf_filtered ("\n");
 427
 428   printf_filtered (_("    Total memory used by bcache, including overhead: %ld\n"),
 429                    c->structure_size);
 430   printf_filtered (_("    Percentage memory overhead: "));
 431   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 432   printf_filtered (_("    Net memory savings:         "));
 433   print_percentage (c->total_size - c->structure_size, c->total_size);
 434   printf_filtered ("\n");
 435
 436   printf_filtered (_("    Hash table size:           %3d\n"), c->num_buckets);
 437   printf_filtered (_("    Hash table expands:        %lu\n"),
 438                    c->expand_count);
 439   printf_filtered (_("    Hash table hashes:         %lu\n"),
 440                    c->total_count + c->expand_hash_count);
 441   printf_filtered (_("    Half hash misses:          %lu\n"),
 442                    c->half_hash_miss_count);
 443   printf_filtered (_("    Hash table population:     "));
 444   print_percentage (occupied_buckets, c->num_buckets);
 445   printf_filtered (_("    Median hash chain length:  %3d\n"),
 446                    median_chain_length);
 447   printf_filtered (_("    Average hash chain length: "));
 448   if (c->num_buckets > 0)
 449     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 450   else
 451     /* i18n: "Average hash chain length: (not applicable)" */
 452     printf_filtered (_("(not applicable)\n"));
 453   printf_filtered (_("    Maximum hash chain length: %3d\n"), max_chain_length);
 454   printf_filtered ("\n");
 455 }
 456
 457 int
 458 bcache_memory_used (struct bcache *bcache)
 459 {
 460   return obstack_memory_used (&bcache->cache);
 461 }