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