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) 1999-2015 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 *new
 268       = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
 269
 270     memcpy (&new->d.data, addr, length);
 271     new->length = length;
 272     new->next = bcache->bucket[hash_index];
 273     new->half_hash = half_hash;
 274     bcache->bucket[hash_index] = new;
 275
 276     bcache->unique_count++;
 277     bcache->unique_size += length;
 278     bcache->structure_size += BSTRING_SIZE (length);
 279
 280     if (added)
 281       *added = 1;
 282
 283     return &new->d.data;
 284   }
 285 }
 286 \f
 287
 288 /* Compare the byte string at ADDR1 of lenght LENGHT to the
 289    string at ADDR2.  Return 1 if they are equal.  */
 290
 291 static int
 292 bcache_compare (const void *addr1, const void *addr2, int length)
 293 {
 294   return memcmp (addr1, addr2, length) == 0;
 295 }
 296
 297 /* Allocating and freeing bcaches.  */
 298
 299 /* Allocated a bcache.  HASH_FUNCTION and COMPARE_FUNCTION can be used
 300    to pass in custom hash, and compare functions to be used by this
 301    bcache.  If HASH_FUNCTION is NULL hash() is used and if
 302    COMPARE_FUNCTION is NULL memcmp() is used.  */
 303
 304 struct bcache *
 305 bcache_xmalloc (unsigned long (*hash_function)(const void *, int length),
 306                 int (*compare_function)(const void *,
 307                                         const void *,
 308                                         int length))
 309 {
 310   /* Allocate the bcache pre-zeroed.  */
 311   struct bcache *b = XCNEW (struct bcache);
 312
 313   if (hash_function)
 314     b->hash_function = hash_function;
 315   else
 316     b->hash_function = hash;
 317
 318   if (compare_function)
 319     b->compare_function = compare_function;
 320   else
 321     b->compare_function = bcache_compare;
 322   return b;
 323 }
 324
 325 /* Free all the storage associated with BCACHE.  */
 326 void
 327 bcache_xfree (struct bcache *bcache)
 328 {
 329   if (bcache == NULL)
 330     return;
 331   /* Only free the obstack if we actually initialized it.  */
 332   if (bcache->total_count > 0)
 333     obstack_free (&bcache->cache, 0);
 334   xfree (bcache->bucket);
 335   xfree (bcache);
 336 }
 337
 338
 339 \f
 340 /* Printing statistics.  */
 341
 342 static void
 343 print_percentage (int portion, int total)
 344 {
 345   if (total == 0)
 346     /* i18n: Like "Percentage of duplicates, by count: (not applicable)".  */
 347     printf_filtered (_("(not applicable)\n"));
 348   else
 349     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 350 }
 351
 352
 353 /* Print statistics on BCACHE's memory usage and efficacity at
 354    eliminating duplication.  NAME should describe the kind of data
 355    BCACHE holds.  Statistics are printed using `printf_filtered' and
 356    its ilk.  */
 357 void
 358 print_bcache_statistics (struct bcache *c, char *type)
 359 {
 360   int occupied_buckets;
 361   int max_chain_length;
 362   int median_chain_length;
 363   int max_entry_size;
 364   int median_entry_size;
 365
 366   /* Count the number of occupied buckets, tally the various string
 367      lengths, and measure chain lengths.  */
 368   {
 369     unsigned int b;
 370     int *chain_length = XCNEWVEC (int, c->num_buckets + 1);
 371     int *entry_size = XCNEWVEC (int, c->unique_count + 1);
 372     int stringi = 0;
 373
 374     occupied_buckets = 0;
 375
 376     for (b = 0; b < c->num_buckets; b++)
 377       {
 378         struct bstring *s = c->bucket[b];
 379
 380         chain_length[b] = 0;
 381
 382         if (s)
 383           {
 384             occupied_buckets++;
 385
 386             while (s)
 387               {
 388                 gdb_assert (b < c->num_buckets);
 389                 chain_length[b]++;
 390                 gdb_assert (stringi < c->unique_count);
 391                 entry_size[stringi++] = s->length;
 392                 s = s->next;
 393               }
 394           }
 395       }
 396
 397     /* To compute the median, we need the set of chain lengths
 398        sorted.  */
 399     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 400            compare_positive_ints);
 401     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 402            compare_positive_ints);
 403
 404     if (c->num_buckets > 0)
 405       {
 406         max_chain_length = chain_length[c->num_buckets - 1];
 407         median_chain_length = chain_length[c->num_buckets / 2];
 408       }
 409     else
 410       {
 411         max_chain_length = 0;
 412         median_chain_length = 0;
 413       }
 414     if (c->unique_count > 0)
 415       {
 416         max_entry_size = entry_size[c->unique_count - 1];
 417         median_entry_size = entry_size[c->unique_count / 2];
 418       }
 419     else
 420       {
 421         max_entry_size = 0;
 422         median_entry_size = 0;
 423       }
 424
 425     xfree (chain_length);
 426     xfree (entry_size);
 427   }
 428
 429   printf_filtered (_("  Cached '%s' statistics:\n"), type);
 430   printf_filtered (_("    Total object count:  %ld\n"), c->total_count);
 431   printf_filtered (_("    Unique object count: %lu\n"), c->unique_count);
 432   printf_filtered (_("    Percentage of duplicates, by count: "));
 433   print_percentage (c->total_count - c->unique_count, c->total_count);
 434   printf_filtered ("\n");
 435
 436   printf_filtered (_("    Total object size:   %ld\n"), c->total_size);
 437   printf_filtered (_("    Unique object size:  %ld\n"), c->unique_size);
 438   printf_filtered (_("    Percentage of duplicates, by size:  "));
 439   print_percentage (c->total_size - c->unique_size, c->total_size);
 440   printf_filtered ("\n");
 441
 442   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 443   printf_filtered (_("    Average entry size: "));
 444   if (c->unique_count > 0)
 445     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 446   else
 447     /* i18n: "Average entry size: (not applicable)".  */
 448     printf_filtered (_("(not applicable)\n"));
 449   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 450   printf_filtered ("\n");
 451
 452   printf_filtered (_("    \
 453 Total memory used by bcache, including overhead: %ld\n"),
 454                    c->structure_size);
 455   printf_filtered (_("    Percentage memory overhead: "));
 456   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 457   printf_filtered (_("    Net memory savings:         "));
 458   print_percentage (c->total_size - c->structure_size, c->total_size);
 459   printf_filtered ("\n");
 460
 461   printf_filtered (_("    Hash table size:           %3d\n"),
 462                    c->num_buckets);
 463   printf_filtered (_("    Hash table expands:        %lu\n"),
 464                    c->expand_count);
 465   printf_filtered (_("    Hash table hashes:         %lu\n"),
 466                    c->total_count + c->expand_hash_count);
 467   printf_filtered (_("    Half hash misses:          %lu\n"),
 468                    c->half_hash_miss_count);
 469   printf_filtered (_("    Hash table population:     "));
 470   print_percentage (occupied_buckets, c->num_buckets);
 471   printf_filtered (_("    Median hash chain length:  %3d\n"),
 472                    median_chain_length);
 473   printf_filtered (_("    Average hash chain length: "));
 474   if (c->num_buckets > 0)
 475     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 476   else
 477     /* i18n: "Average hash chain length: (not applicable)".  */
 478     printf_filtered (_("(not applicable)\n"));
 479   printf_filtered (_("    Maximum hash chain length: %3d\n"),
 480                    max_chain_length);
 481   printf_filtered ("\n");
 482 }
 483
 484 int
 485 bcache_memory_used (struct bcache *bcache)
 486 {
 487   if (bcache->total_count == 0)
 488     return 0;
 489   return obstack_memory_used (&bcache->cache);
 490 }