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, 2009, 2010
   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   /* Hash function to be used for this bcache object.  */
  94   unsigned long (*hash_function)(const void *addr, int length);
  95
  96   /* Compare function to be used for this bcache object.  */
  97   int (*compare_function)(const void *, const void *, int length);
  98 };
  99
 100 /* The old hash function was stolen from SDBM. This is what DB 3.0 uses now,
 101  * and is better than the old one.
 102  */
 103 \f
 104 unsigned long
 105 hash(const void *addr, int length)
 106 {
 107   return hash_continue (addr, length, 0);
 108 }
 109
 110 /* Continue the calculation of the hash H at the given address.  */
 111
 112 unsigned long
 113 hash_continue (const void *addr, int length, unsigned long h)
 114 {
 115   const unsigned char *k, *e;
 116
 117   k = (const unsigned char *)addr;
 118   e = k+length;
 119   for (; k< e;++k)
 120     {
 121       h *=16777619;
 122       h ^= *k;
 123     }
 124   return (h);
 125 }
 126 \f
 127 /* Growing the bcache's hash table.  */
 128
 129 /* If the average chain length grows beyond this, then we want to
 130    resize our hash table.  */
 131 #define CHAIN_LENGTH_THRESHOLD (5)
 132
 133 static void
 134 expand_hash_table (struct bcache *bcache)
 135 {
 136   /* A table of good hash table sizes.  Whenever we grow, we pick the
 137      next larger size from this table.  sizes[i] is close to 1 << (i+10),
 138      so we roughly double the table size each time.  After we fall off
 139      the end of this table, we just double.  Don't laugh --- there have
 140      been executables sighted with a gigabyte of debug info.  */
 141   static unsigned long sizes[] = {
 142     1021, 2053, 4099, 8191, 16381, 32771,
 143     65537, 131071, 262144, 524287, 1048573, 2097143,
 144     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
 145     268435459, 536870923, 1073741827, 2147483659UL
 146   };
 147   unsigned int new_num_buckets;
 148   struct bstring **new_buckets;
 149   unsigned int i;
 150
 151   /* Count the stats.  Every unique item needs to be re-hashed and
 152      re-entered.  */
 153   bcache->expand_count++;
 154   bcache->expand_hash_count += bcache->unique_count;
 155
 156   /* Find the next size.  */
 157   new_num_buckets = bcache->num_buckets * 2;
 158   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 159     if (sizes[i] > bcache->num_buckets)
 160       {
 161         new_num_buckets = sizes[i];
 162         break;
 163       }
 164
 165   /* Allocate the new table.  */
 166   {
 167     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 168
 169     new_buckets = (struct bstring **) xmalloc (new_size);
 170     memset (new_buckets, 0, new_size);
 171
 172     bcache->structure_size -= (bcache->num_buckets
 173                                * sizeof (bcache->bucket[0]));
 174     bcache->structure_size += new_size;
 175   }
 176
 177   /* Rehash all existing strings.  */
 178   for (i = 0; i < bcache->num_buckets; i++)
 179     {
 180       struct bstring *s, *next;
 181
 182       for (s = bcache->bucket[i]; s; s = next)
 183         {
 184           struct bstring **new_bucket;
 185           next = s->next;
 186
 187           new_bucket = &new_buckets[(bcache->hash_function (&s->d.data,
 188                                                             s->length)
 189                                      % new_num_buckets)];
 190           s->next = *new_bucket;
 191           *new_bucket = s;
 192         }
 193     }
 194
 195   /* Plug in the new table.  */
 196   if (bcache->bucket)
 197     xfree (bcache->bucket);
 198   bcache->bucket = new_buckets;
 199   bcache->num_buckets = new_num_buckets;
 200 }
 201
 202 \f
 203 /* Looking up things in the bcache.  */
 204
 205 /* The number of bytes needed to allocate a struct bstring whose data
 206    is N bytes long.  */
 207 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 208
 209 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 210    never seen those bytes before, add a copy of them to BCACHE.  In
 211    either case, return a pointer to BCACHE's copy of that string.  */
 212 const void *
 213 bcache (const void *addr, int length, struct bcache *bcache)
 214 {
 215   return bcache_full (addr, length, bcache, NULL);
 216 }
 217
 218 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 219    never seen those bytes before, add a copy of them to BCACHE.  In
 220    either case, return a pointer to BCACHE's copy of that string.  If
 221    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 222    returning an old entry.  */
 223
 224 const void *
 225 bcache_full (const void *addr, int length, struct bcache *bcache, int *added)
 226 {
 227   unsigned long full_hash;
 228   unsigned short half_hash;
 229   int hash_index;
 230   struct bstring *s;
 231
 232   if (added)
 233     *added = 0;
 234
 235   /* Lazily initialize the obstack.  This can save quite a bit of
 236      memory in some cases.  */
 237   if (bcache->total_count == 0)
 238     {
 239       /* We could use obstack_specify_allocation here instead, but
 240          gdb_obstack.h specifies the allocation/deallocation
 241          functions.  */
 242       obstack_init (&bcache->cache);
 243     }
 244
 245   /* If our average chain length is too high, expand the hash table.  */
 246   if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
 247     expand_hash_table (bcache);
 248
 249   bcache->total_count++;
 250   bcache->total_size += length;
 251
 252   full_hash = bcache->hash_function (addr, length);
 253
 254   half_hash = (full_hash >> 16);
 255   hash_index = full_hash % bcache->num_buckets;
 256
 257   /* Search the hash bucket for a string identical to the caller's.
 258      As a short-circuit first compare the upper part of each hash
 259      values.  */
 260   for (s = bcache->bucket[hash_index]; s; s = s->next)
 261     {
 262       if (s->half_hash == half_hash)
 263         {
 264           if (s->length == length
 265               && bcache->compare_function (&s->d.data, addr, length))
 266             return &s->d.data;
 267           else
 268             bcache->half_hash_miss_count++;
 269         }
 270     }
 271
 272   /* The user's string isn't in the list.  Insert it after *ps.  */
 273   {
 274     struct bstring *new
 275       = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
 276
 277     memcpy (&new->d.data, addr, length);
 278     new->length = length;
 279     new->next = bcache->bucket[hash_index];
 280     new->half_hash = half_hash;
 281     bcache->bucket[hash_index] = new;
 282
 283     bcache->unique_count++;
 284     bcache->unique_size += length;
 285     bcache->structure_size += BSTRING_SIZE (length);
 286
 287     if (added)
 288       *added = 1;
 289
 290     return &new->d.data;
 291   }
 292 }
 293 \f
 294
 295 /* Compare the byte string at ADDR1 of lenght LENGHT to the
 296    string at ADDR2.  Return 1 if they are equal.  */
 297
 298 static int
 299 bcache_compare (const void *addr1, const void *addr2, int length)
 300 {
 301   return memcmp (addr1, addr2, length) == 0;
 302 }
 303
 304 /* Allocating and freeing bcaches.  */
 305
 306 /* Allocated a bcache.  HASH_FUNCTION and COMPARE_FUNCTION can be used
 307    to pass in custom hash, and compare functions to be used by this
 308    bcache. If HASH_FUNCTION is NULL hash() is used and if COMPARE_FUNCTION
 309    is NULL memcmp() is used.  */
 310
 311 struct bcache *
 312 bcache_xmalloc (unsigned long (*hash_function)(const void *, int length),
 313                 int (*compare_function)(const void *, const void *, int length))
 314 {
 315   /* Allocate the bcache pre-zeroed.  */
 316   struct bcache *b = XCALLOC (1, struct bcache);
 317
 318   if (hash_function)
 319     b->hash_function = hash_function;
 320   else
 321     b->hash_function = hash;
 322
 323   if (compare_function)
 324     b->compare_function = compare_function;
 325   else
 326     b->compare_function = bcache_compare;
 327   return b;
 328 }
 329
 330 /* Free all the storage associated with BCACHE.  */
 331 void
 332 bcache_xfree (struct bcache *bcache)
 333 {
 334   if (bcache == NULL)
 335     return;
 336   /* Only free the obstack if we actually initialized it.  */
 337   if (bcache->total_count > 0)
 338     obstack_free (&bcache->cache, 0);
 339   xfree (bcache->bucket);
 340   xfree (bcache);
 341 }
 342
 343
 344 \f
 345 /* Printing statistics.  */
 346
 347 static void
 348 print_percentage (int portion, int total)
 349 {
 350   if (total == 0)
 351     /* i18n: Like "Percentage of duplicates, by count: (not applicable)" */
 352     printf_filtered (_("(not applicable)\n"));
 353   else
 354     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 355 }
 356
 357
 358 /* Print statistics on BCACHE's memory usage and efficacity at
 359    eliminating duplication.  NAME should describe the kind of data
 360    BCACHE holds.  Statistics are printed using `printf_filtered' and
 361    its ilk.  */
 362 void
 363 print_bcache_statistics (struct bcache *c, char *type)
 364 {
 365   int occupied_buckets;
 366   int max_chain_length;
 367   int median_chain_length;
 368   int max_entry_size;
 369   int median_entry_size;
 370
 371   /* Count the number of occupied buckets, tally the various string
 372      lengths, and measure chain lengths.  */
 373   {
 374     unsigned int b;
 375     int *chain_length = XCALLOC (c->num_buckets + 1, int);
 376     int *entry_size = XCALLOC (c->unique_count + 1, int);
 377     int stringi = 0;
 378
 379     occupied_buckets = 0;
 380
 381     for (b = 0; b < c->num_buckets; b++)
 382       {
 383         struct bstring *s = c->bucket[b];
 384
 385         chain_length[b] = 0;
 386
 387         if (s)
 388           {
 389             occupied_buckets++;
 390
 391             while (s)
 392               {
 393                 gdb_assert (b < c->num_buckets);
 394                 chain_length[b]++;
 395                 gdb_assert (stringi < c->unique_count);
 396                 entry_size[stringi++] = s->length;
 397                 s = s->next;
 398               }
 399           }
 400       }
 401
 402     /* To compute the median, we need the set of chain lengths sorted.  */
 403     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 404            compare_positive_ints);
 405     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 406            compare_positive_ints);
 407
 408     if (c->num_buckets > 0)
 409       {
 410         max_chain_length = chain_length[c->num_buckets - 1];
 411         median_chain_length = chain_length[c->num_buckets / 2];
 412       }
 413     else
 414       {
 415         max_chain_length = 0;
 416         median_chain_length = 0;
 417       }
 418     if (c->unique_count > 0)
 419       {
 420         max_entry_size = entry_size[c->unique_count - 1];
 421         median_entry_size = entry_size[c->unique_count / 2];
 422       }
 423     else
 424       {
 425         max_entry_size = 0;
 426         median_entry_size = 0;
 427       }
 428
 429     xfree (chain_length);
 430     xfree (entry_size);
 431   }
 432
 433   printf_filtered (_("  Cached '%s' statistics:\n"), type);
 434   printf_filtered (_("    Total object count:  %ld\n"), c->total_count);
 435   printf_filtered (_("    Unique object count: %lu\n"), c->unique_count);
 436   printf_filtered (_("    Percentage of duplicates, by count: "));
 437   print_percentage (c->total_count - c->unique_count, c->total_count);
 438   printf_filtered ("\n");
 439
 440   printf_filtered (_("    Total object size:   %ld\n"), c->total_size);
 441   printf_filtered (_("    Unique object size:  %ld\n"), c->unique_size);
 442   printf_filtered (_("    Percentage of duplicates, by size:  "));
 443   print_percentage (c->total_size - c->unique_size, c->total_size);
 444   printf_filtered ("\n");
 445
 446   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 447   printf_filtered (_("    Average entry size: "));
 448   if (c->unique_count > 0)
 449     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 450   else
 451     /* i18n: "Average entry size: (not applicable)" */
 452     printf_filtered (_("(not applicable)\n"));
 453   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 454   printf_filtered ("\n");
 455
 456   printf_filtered (_("    Total memory used by bcache, including overhead: %ld\n"),
 457                    c->structure_size);
 458   printf_filtered (_("    Percentage memory overhead: "));
 459   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 460   printf_filtered (_("    Net memory savings:         "));
 461   print_percentage (c->total_size - c->structure_size, c->total_size);
 462   printf_filtered ("\n");
 463
 464   printf_filtered (_("    Hash table size:           %3d\n"), c->num_buckets);
 465   printf_filtered (_("    Hash table expands:        %lu\n"),
 466                    c->expand_count);
 467   printf_filtered (_("    Hash table hashes:         %lu\n"),
 468                    c->total_count + c->expand_hash_count);
 469   printf_filtered (_("    Half hash misses:          %lu\n"),
 470                    c->half_hash_miss_count);
 471   printf_filtered (_("    Hash table population:     "));
 472   print_percentage (occupied_buckets, c->num_buckets);
 473   printf_filtered (_("    Median hash chain length:  %3d\n"),
 474                    median_chain_length);
 475   printf_filtered (_("    Average hash chain length: "));
 476   if (c->num_buckets > 0)
 477     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 478   else
 479     /* i18n: "Average hash chain length: (not applicable)" */
 480     printf_filtered (_("(not applicable)\n"));
 481   printf_filtered (_("    Maximum hash chain length: %3d\n"), max_chain_length);
 482   printf_filtered ("\n");
 483 }
 484
 485 int
 486 bcache_memory_used (struct bcache *bcache)
 487 {
 488   if (bcache->total_count == 0)
 489     return 0;
 490   return obstack_memory_used (&bcache->cache);
 491 }