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) 1999-2019 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
  26 /* The type used to hold a single bcache string.  The user data is
  27    stored in d.data.  Since it can be any type, it needs to have the
  28    same alignment as the most strict alignment of any type on the host
  29    machine.  I don't know of any really correct way to do this in
  30    stock ANSI C, so just do it the same way obstack.h does.  */
  31
  32 struct bstring
  33 {
  34   /* Hash chain.  */
  35   struct bstring *next;
  36   /* Assume the data length is no more than 64k.  */
  37   unsigned short length;
  38   /* The half hash hack.  This contains the upper 16 bits of the hash
  39      value and is used as a pre-check when comparing two strings and
  40      avoids the need to do length or memcmp calls.  It proves to be
  41      roughly 100% effective.  */
  42   unsigned short half_hash;
  43
  44   union
  45   {
  46     char data[1];
  47     double dummy;
  48   }
  49   d;
  50 };
  51
  52
  53 /* The structure for a bcache itself.  The bcache is initialized, in
  54    bcache_xmalloc(), by filling it with zeros and then setting the
  55    corresponding obstack's malloc() and free() methods.  */
  56
  57 struct bcache
  58 {
  59   /* All the bstrings are allocated here.  */
  60   struct obstack cache;
  61
  62   /* How many hash buckets we're using.  */
  63   unsigned int num_buckets;
  64
  65   /* Hash buckets.  This table is allocated using malloc, so when we
  66      grow the table we can return the old table to the system.  */
  67   struct bstring **bucket;
  68
  69   /* Statistics.  */
  70   unsigned long unique_count;   /* number of unique strings */
  71   long total_count;     /* total number of strings cached, including dups */
  72   long unique_size;     /* size of unique strings, in bytes */
  73   long total_size;      /* total number of bytes cached, including dups */
  74   long structure_size;  /* total size of bcache, including infrastructure */
  75   /* Number of times that the hash table is expanded and hence
  76      re-built, and the corresponding number of times that a string is
  77      [re]hashed as part of entering it into the expanded table.  The
  78      total number of hashes can be computed by adding TOTAL_COUNT to
  79      expand_hash_count.  */
  80   unsigned long expand_count;
  81   unsigned long expand_hash_count;
  82   /* Number of times that the half-hash compare hit (compare the upper
  83      16 bits of hash values) hit, but the corresponding combined
  84      length/data compare missed.  */
  85   unsigned long half_hash_miss_count;
  86
  87   /* Hash function to be used for this bcache object.  */
  88   unsigned long (*hash_function)(const void *addr, int length);
  89
  90   /* Compare function to be used for this bcache object.  */
  91   int (*compare_function)(const void *, const void *, int length);
  92 };
  93
  94 /* The old hash function was stolen from SDBM. This is what DB 3.0
  95    uses now, and is better than the old one.  */
  96 \f
  97 unsigned long
  98 hash(const void *addr, int length)
  99 {
 100   return hash_continue (addr, length, 0);
 101 }
 102
 103 /* Continue the calculation of the hash H at the given address.  */
 104
 105 unsigned long
 106 hash_continue (const void *addr, int length, unsigned long h)
 107 {
 108   const unsigned char *k, *e;
 109
 110   k = (const unsigned char *)addr;
 111   e = k+length;
 112   for (; k< e;++k)
 113     {
 114       h *=16777619;
 115       h ^= *k;
 116     }
 117   return (h);
 118 }
 119 \f
 120 /* Growing the bcache's hash table.  */
 121
 122 /* If the average chain length grows beyond this, then we want to
 123    resize our hash table.  */
 124 #define CHAIN_LENGTH_THRESHOLD (5)
 125
 126 static void
 127 expand_hash_table (struct bcache *bcache)
 128 {
 129   /* A table of good hash table sizes.  Whenever we grow, we pick the
 130      next larger size from this table.  sizes[i] is close to 1 << (i+10),
 131      so we roughly double the table size each time.  After we fall off
 132      the end of this table, we just double.  Don't laugh --- there have
 133      been executables sighted with a gigabyte of debug info.  */
 134   static unsigned long sizes[] = {
 135     1021, 2053, 4099, 8191, 16381, 32771,
 136     65537, 131071, 262144, 524287, 1048573, 2097143,
 137     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
 138     268435459, 536870923, 1073741827, 2147483659UL
 139   };
 140   unsigned int new_num_buckets;
 141   struct bstring **new_buckets;
 142   unsigned int i;
 143
 144   /* Count the stats.  Every unique item needs to be re-hashed and
 145      re-entered.  */
 146   bcache->expand_count++;
 147   bcache->expand_hash_count += bcache->unique_count;
 148
 149   /* Find the next size.  */
 150   new_num_buckets = bcache->num_buckets * 2;
 151   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 152     if (sizes[i] > bcache->num_buckets)
 153       {
 154         new_num_buckets = sizes[i];
 155         break;
 156       }
 157
 158   /* Allocate the new table.  */
 159   {
 160     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 161
 162     new_buckets = (struct bstring **) xmalloc (new_size);
 163     memset (new_buckets, 0, new_size);
 164
 165     bcache->structure_size -= (bcache->num_buckets
 166                                * sizeof (bcache->bucket[0]));
 167     bcache->structure_size += new_size;
 168   }
 169
 170   /* Rehash all existing strings.  */
 171   for (i = 0; i < bcache->num_buckets; i++)
 172     {
 173       struct bstring *s, *next;
 174
 175       for (s = bcache->bucket[i]; s; s = next)
 176         {
 177           struct bstring **new_bucket;
 178           next = s->next;
 179
 180           new_bucket = &new_buckets[(bcache->hash_function (&s->d.data,
 181                                                             s->length)
 182                                      % new_num_buckets)];
 183           s->next = *new_bucket;
 184           *new_bucket = s;
 185         }
 186     }
 187
 188   /* Plug in the new table.  */
 189   if (bcache->bucket)
 190     xfree (bcache->bucket);
 191   bcache->bucket = new_buckets;
 192   bcache->num_buckets = new_num_buckets;
 193 }
 194
 195 \f
 196 /* Looking up things in the bcache.  */
 197
 198 /* The number of bytes needed to allocate a struct bstring whose data
 199    is N bytes long.  */
 200 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 201
 202 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 203    never seen those bytes before, add a copy of them to BCACHE.  In
 204    either case, return a pointer to BCACHE's copy of that string.  */
 205 const void *
 206 bcache (const void *addr, int length, struct bcache *cache)
 207 {
 208   return bcache_full (addr, length, cache, NULL);
 209 }
 210
 211 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 212    never seen those bytes before, add a copy of them to BCACHE.  In
 213    either case, return a pointer to BCACHE's copy of that string.  If
 214    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 215    returning an old entry.  */
 216
 217 const void *
 218 bcache_full (const void *addr, int length, struct bcache *bcache, int *added)
 219 {
 220   unsigned long full_hash;
 221   unsigned short half_hash;
 222   int hash_index;
 223   struct bstring *s;
 224
 225   if (added)
 226     *added = 0;
 227
 228   /* Lazily initialize the obstack.  This can save quite a bit of
 229      memory in some cases.  */
 230   if (bcache->total_count == 0)
 231     {
 232       /* We could use obstack_specify_allocation here instead, but
 233          gdb_obstack.h specifies the allocation/deallocation
 234          functions.  */
 235       obstack_init (&bcache->cache);
 236     }
 237
 238   /* If our average chain length is too high, expand the hash table.  */
 239   if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
 240     expand_hash_table (bcache);
 241
 242   bcache->total_count++;
 243   bcache->total_size += length;
 244
 245   full_hash = bcache->hash_function (addr, length);
 246
 247   half_hash = (full_hash >> 16);
 248   hash_index = full_hash % bcache->num_buckets;
 249
 250   /* Search the hash bucket for a string identical to the caller's.
 251      As a short-circuit first compare the upper part of each hash
 252      values.  */
 253   for (s = bcache->bucket[hash_index]; s; s = s->next)
 254     {
 255       if (s->half_hash == half_hash)
 256         {
 257           if (s->length == length
 258               && bcache->compare_function (&s->d.data, addr, length))
 259             return &s->d.data;
 260           else
 261             bcache->half_hash_miss_count++;
 262         }
 263     }
 264
 265   /* The user's string isn't in the list.  Insert it after *ps.  */
 266   {
 267     struct bstring *newobj
 268       = (struct bstring *) obstack_alloc (&bcache->cache,
 269                                           BSTRING_SIZE (length));
 270
 271     memcpy (&newobj->d.data, addr, length);
 272     newobj->length = length;
 273     newobj->next = bcache->bucket[hash_index];
 274     newobj->half_hash = half_hash;
 275     bcache->bucket[hash_index] = newobj;
 276
 277     bcache->unique_count++;
 278     bcache->unique_size += length;
 279     bcache->structure_size += BSTRING_SIZE (length);
 280
 281     if (added)
 282       *added = 1;
 283
 284     return &newobj->d.data;
 285   }
 286 }
 287 \f
 288
 289 /* Compare the byte string at ADDR1 of lenght LENGHT to the
 290    string at ADDR2.  Return 1 if they are equal.  */
 291
 292 static int
 293 bcache_compare (const void *addr1, const void *addr2, int length)
 294 {
 295   return memcmp (addr1, addr2, length) == 0;
 296 }
 297
 298 /* Allocating and freeing bcaches.  */
 299
 300 /* Allocated a bcache.  HASH_FUNCTION and COMPARE_FUNCTION can be used
 301    to pass in custom hash, and compare functions to be used by this
 302    bcache.  If HASH_FUNCTION is NULL hash() is used and if
 303    COMPARE_FUNCTION is NULL memcmp() is used.  */
 304
 305 struct bcache *
 306 bcache_xmalloc (unsigned long (*hash_function)(const void *, int length),
 307                 int (*compare_function)(const void *,
 308                                         const void *,
 309                                         int length))
 310 {
 311   /* Allocate the bcache pre-zeroed.  */
 312   struct bcache *b = XCNEW (struct bcache);
 313
 314   if (hash_function)
 315     b->hash_function = hash_function;
 316   else
 317     b->hash_function = hash;
 318
 319   if (compare_function)
 320     b->compare_function = compare_function;
 321   else
 322     b->compare_function = bcache_compare;
 323   return b;
 324 }
 325
 326 /* Free all the storage associated with BCACHE.  */
 327 void
 328 bcache_xfree (struct bcache *bcache)
 329 {
 330   if (bcache == NULL)
 331     return;
 332   /* Only free the obstack if we actually initialized it.  */
 333   if (bcache->total_count > 0)
 334     obstack_free (&bcache->cache, 0);
 335   xfree (bcache->bucket);
 336   xfree (bcache);
 337 }
 338
 339
 340 \f
 341 /* Printing statistics.  */
 342
 343 static void
 344 print_percentage (int portion, int total)
 345 {
 346   if (total == 0)
 347     /* i18n: Like "Percentage of duplicates, by count: (not applicable)".  */
 348     printf_filtered (_("(not applicable)\n"));
 349   else
 350     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 351 }
 352
 353
 354 /* Print statistics on BCACHE's memory usage and efficacity at
 355    eliminating duplication.  NAME should describe the kind of data
 356    BCACHE holds.  Statistics are printed using `printf_filtered' and
 357    its ilk.  */
 358 void
 359 print_bcache_statistics (struct bcache *c, const char *type)
 360 {
 361   int occupied_buckets;
 362   int max_chain_length;
 363   int median_chain_length;
 364   int max_entry_size;
 365   int median_entry_size;
 366
 367   /* Count the number of occupied buckets, tally the various string
 368      lengths, and measure chain lengths.  */
 369   {
 370     unsigned int b;
 371     int *chain_length = XCNEWVEC (int, c->num_buckets + 1);
 372     int *entry_size = XCNEWVEC (int, c->unique_count + 1);
 373     int stringi = 0;
 374
 375     occupied_buckets = 0;
 376
 377     for (b = 0; b < c->num_buckets; b++)
 378       {
 379         struct bstring *s = c->bucket[b];
 380
 381         chain_length[b] = 0;
 382
 383         if (s)
 384           {
 385             occupied_buckets++;
 386
 387             while (s)
 388               {
 389                 gdb_assert (b < c->num_buckets);
 390                 chain_length[b]++;
 391                 gdb_assert (stringi < c->unique_count);
 392                 entry_size[stringi++] = s->length;
 393                 s = s->next;
 394               }
 395           }
 396       }
 397
 398     /* To compute the median, we need the set of chain lengths
 399        sorted.  */
 400     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 401            compare_positive_ints);
 402     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 403            compare_positive_ints);
 404
 405     if (c->num_buckets > 0)
 406       {
 407         max_chain_length = chain_length[c->num_buckets - 1];
 408         median_chain_length = chain_length[c->num_buckets / 2];
 409       }
 410     else
 411       {
 412         max_chain_length = 0;
 413         median_chain_length = 0;
 414       }
 415     if (c->unique_count > 0)
 416       {
 417         max_entry_size = entry_size[c->unique_count - 1];
 418         median_entry_size = entry_size[c->unique_count / 2];
 419       }
 420     else
 421       {
 422         max_entry_size = 0;
 423         median_entry_size = 0;
 424       }
 425
 426     xfree (chain_length);
 427     xfree (entry_size);
 428   }
 429
 430   printf_filtered (_("  Cached '%s' statistics:\n"), type);
 431   printf_filtered (_("    Total object count:  %ld\n"), c->total_count);
 432   printf_filtered (_("    Unique object count: %lu\n"), c->unique_count);
 433   printf_filtered (_("    Percentage of duplicates, by count: "));
 434   print_percentage (c->total_count - c->unique_count, c->total_count);
 435   printf_filtered ("\n");
 436
 437   printf_filtered (_("    Total object size:   %ld\n"), c->total_size);
 438   printf_filtered (_("    Unique object size:  %ld\n"), c->unique_size);
 439   printf_filtered (_("    Percentage of duplicates, by size:  "));
 440   print_percentage (c->total_size - c->unique_size, c->total_size);
 441   printf_filtered ("\n");
 442
 443   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 444   printf_filtered (_("    Average entry size: "));
 445   if (c->unique_count > 0)
 446     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 447   else
 448     /* i18n: "Average entry size: (not applicable)".  */
 449     printf_filtered (_("(not applicable)\n"));
 450   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 451   printf_filtered ("\n");
 452
 453   printf_filtered (_("    \
 454 Total memory used by bcache, including overhead: %ld\n"),
 455                    c->structure_size);
 456   printf_filtered (_("    Percentage memory overhead: "));
 457   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 458   printf_filtered (_("    Net memory savings:         "));
 459   print_percentage (c->total_size - c->structure_size, c->total_size);
 460   printf_filtered ("\n");
 461
 462   printf_filtered (_("    Hash table size:           %3d\n"),
 463                    c->num_buckets);
 464   printf_filtered (_("    Hash table expands:        %lu\n"),
 465                    c->expand_count);
 466   printf_filtered (_("    Hash table hashes:         %lu\n"),
 467                    c->total_count + c->expand_hash_count);
 468   printf_filtered (_("    Half hash misses:          %lu\n"),
 469                    c->half_hash_miss_count);
 470   printf_filtered (_("    Hash table population:     "));
 471   print_percentage (occupied_buckets, c->num_buckets);
 472   printf_filtered (_("    Median hash chain length:  %3d\n"),
 473                    median_chain_length);
 474   printf_filtered (_("    Average hash chain length: "));
 475   if (c->num_buckets > 0)
 476     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 477   else
 478     /* i18n: "Average hash chain length: (not applicable)".  */
 479     printf_filtered (_("(not applicable)\n"));
 480   printf_filtered (_("    Maximum hash chain length: %3d\n"),
 481                    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 }