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) 1999-2014 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 <string.h>             /* For memcpy declaration */
  26 #include "gdb_assert.h"
  27
  28 /* The type used to hold a single bcache string.  The user data is
  29    stored in d.data.  Since it can be any type, it needs to have the
  30    same alignment as the most strict alignment of any type on the host
  31    machine.  I don't know of any really correct way to do this in
  32    stock ANSI C, so just do it the same way obstack.h does.  */
  33
  34 struct bstring
  35 {
  36   /* Hash chain.  */
  37   struct bstring *next;
  38   /* Assume the data length is no more than 64k.  */
  39   unsigned short length;
  40   /* The half hash hack.  This contains the upper 16 bits of the hash
  41      value and is used as a pre-check when comparing two strings and
  42      avoids the need to do length or memcmp calls.  It proves to be
  43      roughly 100% effective.  */
  44   unsigned short half_hash;
  45
  46   union
  47   {
  48     char data[1];
  49     double dummy;
  50   }
  51   d;
  52 };
  53
  54
  55 /* The structure for a bcache itself.  The bcache is initialized, in
  56    bcache_xmalloc(), by filling it with zeros and then setting the
  57    corresponding obstack's malloc() and free() methods.  */
  58
  59 struct bcache
  60 {
  61   /* All the bstrings are allocated here.  */
  62   struct obstack cache;
  63
  64   /* How many hash buckets we're using.  */
  65   unsigned int num_buckets;
  66
  67   /* Hash buckets.  This table is allocated using malloc, so when we
  68      grow the table we can return the old table to the system.  */
  69   struct bstring **bucket;
  70
  71   /* Statistics.  */
  72   unsigned long unique_count;   /* number of unique strings */
  73   long total_count;     /* total number of strings cached, including dups */
  74   long unique_size;     /* size of unique strings, in bytes */
  75   long total_size;      /* total number of bytes cached, including dups */
  76   long structure_size;  /* total size of bcache, including infrastructure */
  77   /* Number of times that the hash table is expanded and hence
  78      re-built, and the corresponding number of times that a string is
  79      [re]hashed as part of entering it into the expanded table.  The
  80      total number of hashes can be computed by adding TOTAL_COUNT to
  81      expand_hash_count.  */
  82   unsigned long expand_count;
  83   unsigned long expand_hash_count;
  84   /* Number of times that the half-hash compare hit (compare the upper
  85      16 bits of hash values) hit, but the corresponding combined
  86      length/data compare missed.  */
  87   unsigned long half_hash_miss_count;
  88
  89   /* Hash function to be used for this bcache object.  */
  90   unsigned long (*hash_function)(const void *addr, int length);
  91
  92   /* Compare function to be used for this bcache object.  */
  93   int (*compare_function)(const void *, const void *, int length);
  94 };
  95
  96 /* The old hash function was stolen from SDBM. This is what DB 3.0
  97    uses now, and is better than the old one.  */
  98 \f
  99 unsigned long
 100 hash(const void *addr, int length)
 101 {
 102   return hash_continue (addr, length, 0);
 103 }
 104
 105 /* Continue the calculation of the hash H at the given address.  */
 106
 107 unsigned long
 108 hash_continue (const void *addr, int length, unsigned long h)
 109 {
 110   const unsigned char *k, *e;
 111
 112   k = (const unsigned char *)addr;
 113   e = k+length;
 114   for (; k< e;++k)
 115     {
 116       h *=16777619;
 117       h ^= *k;
 118     }
 119   return (h);
 120 }
 121 \f
 122 /* Growing the bcache's hash table.  */
 123
 124 /* If the average chain length grows beyond this, then we want to
 125    resize our hash table.  */
 126 #define CHAIN_LENGTH_THRESHOLD (5)
 127
 128 static void
 129 expand_hash_table (struct bcache *bcache)
 130 {
 131   /* A table of good hash table sizes.  Whenever we grow, we pick the
 132      next larger size from this table.  sizes[i] is close to 1 << (i+10),
 133      so we roughly double the table size each time.  After we fall off
 134      the end of this table, we just double.  Don't laugh --- there have
 135      been executables sighted with a gigabyte of debug info.  */
 136   static unsigned long sizes[] = {
 137     1021, 2053, 4099, 8191, 16381, 32771,
 138     65537, 131071, 262144, 524287, 1048573, 2097143,
 139     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
 140     268435459, 536870923, 1073741827, 2147483659UL
 141   };
 142   unsigned int new_num_buckets;
 143   struct bstring **new_buckets;
 144   unsigned int i;
 145
 146   /* Count the stats.  Every unique item needs to be re-hashed and
 147      re-entered.  */
 148   bcache->expand_count++;
 149   bcache->expand_hash_count += bcache->unique_count;
 150
 151   /* Find the next size.  */
 152   new_num_buckets = bcache->num_buckets * 2;
 153   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 154     if (sizes[i] > bcache->num_buckets)
 155       {
 156         new_num_buckets = sizes[i];
 157         break;
 158       }
 159
 160   /* Allocate the new table.  */
 161   {
 162     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 163
 164     new_buckets = (struct bstring **) xmalloc (new_size);
 165     memset (new_buckets, 0, new_size);
 166
 167     bcache->structure_size -= (bcache->num_buckets
 168                                * sizeof (bcache->bucket[0]));
 169     bcache->structure_size += new_size;
 170   }
 171
 172   /* Rehash all existing strings.  */
 173   for (i = 0; i < bcache->num_buckets; i++)
 174     {
 175       struct bstring *s, *next;
 176
 177       for (s = bcache->bucket[i]; s; s = next)
 178         {
 179           struct bstring **new_bucket;
 180           next = s->next;
 181
 182           new_bucket = &new_buckets[(bcache->hash_function (&s->d.data,
 183                                                             s->length)
 184                                      % new_num_buckets)];
 185           s->next = *new_bucket;
 186           *new_bucket = s;
 187         }
 188     }
 189
 190   /* Plug in the new table.  */
 191   if (bcache->bucket)
 192     xfree (bcache->bucket);
 193   bcache->bucket = new_buckets;
 194   bcache->num_buckets = new_num_buckets;
 195 }
 196
 197 \f
 198 /* Looking up things in the bcache.  */
 199
 200 /* The number of bytes needed to allocate a struct bstring whose data
 201    is N bytes long.  */
 202 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 203
 204 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 205    never seen those bytes before, add a copy of them to BCACHE.  In
 206    either case, return a pointer to BCACHE's copy of that string.  */
 207 const void *
 208 bcache (const void *addr, int length, struct bcache *cache)
 209 {
 210   return bcache_full (addr, length, cache, NULL);
 211 }
 212
 213 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 214    never seen those bytes before, add a copy of them to BCACHE.  In
 215    either case, return a pointer to BCACHE's copy of that string.  If
 216    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 217    returning an old entry.  */
 218
 219 const void *
 220 bcache_full (const void *addr, int length, struct bcache *bcache, int *added)
 221 {
 222   unsigned long full_hash;
 223   unsigned short half_hash;
 224   int hash_index;
 225   struct bstring *s;
 226
 227   if (added)
 228     *added = 0;
 229
 230   /* Lazily initialize the obstack.  This can save quite a bit of
 231      memory in some cases.  */
 232   if (bcache->total_count == 0)
 233     {
 234       /* We could use obstack_specify_allocation here instead, but
 235          gdb_obstack.h specifies the allocation/deallocation
 236          functions.  */
 237       obstack_init (&bcache->cache);
 238     }
 239
 240   /* If our average chain length is too high, expand the hash table.  */
 241   if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
 242     expand_hash_table (bcache);
 243
 244   bcache->total_count++;
 245   bcache->total_size += length;
 246
 247   full_hash = bcache->hash_function (addr, length);
 248
 249   half_hash = (full_hash >> 16);
 250   hash_index = full_hash % bcache->num_buckets;
 251
 252   /* Search the hash bucket for a string identical to the caller's.
 253      As a short-circuit first compare the upper part of each hash
 254      values.  */
 255   for (s = bcache->bucket[hash_index]; s; s = s->next)
 256     {
 257       if (s->half_hash == half_hash)
 258         {
 259           if (s->length == length
 260               && bcache->compare_function (&s->d.data, addr, length))
 261             return &s->d.data;
 262           else
 263             bcache->half_hash_miss_count++;
 264         }
 265     }
 266
 267   /* The user's string isn't in the list.  Insert it after *ps.  */
 268   {
 269     struct bstring *new
 270       = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
 271
 272     memcpy (&new->d.data, addr, length);
 273     new->length = length;
 274     new->next = bcache->bucket[hash_index];
 275     new->half_hash = half_hash;
 276     bcache->bucket[hash_index] = new;
 277
 278     bcache->unique_count++;
 279     bcache->unique_size += length;
 280     bcache->structure_size += BSTRING_SIZE (length);
 281
 282     if (added)
 283       *added = 1;
 284
 285     return &new->d.data;
 286   }
 287 }
 288 \f
 289
 290 /* Compare the byte string at ADDR1 of lenght LENGHT to the
 291    string at ADDR2.  Return 1 if they are equal.  */
 292
 293 static int
 294 bcache_compare (const void *addr1, const void *addr2, int length)
 295 {
 296   return memcmp (addr1, addr2, length) == 0;
 297 }
 298
 299 /* Allocating and freeing bcaches.  */
 300
 301 /* Allocated a bcache.  HASH_FUNCTION and COMPARE_FUNCTION can be used
 302    to pass in custom hash, and compare functions to be used by this
 303    bcache.  If HASH_FUNCTION is NULL hash() is used and if
 304    COMPARE_FUNCTION is NULL memcmp() is used.  */
 305
 306 struct bcache *
 307 bcache_xmalloc (unsigned long (*hash_function)(const void *, int length),
 308                 int (*compare_function)(const void *,
 309                                         const void *,
 310                                         int length))
 311 {
 312   /* Allocate the bcache pre-zeroed.  */
 313   struct bcache *b = XCNEW (struct bcache);
 314
 315   if (hash_function)
 316     b->hash_function = hash_function;
 317   else
 318     b->hash_function = hash;
 319
 320   if (compare_function)
 321     b->compare_function = compare_function;
 322   else
 323     b->compare_function = bcache_compare;
 324   return b;
 325 }
 326
 327 /* Free all the storage associated with BCACHE.  */
 328 void
 329 bcache_xfree (struct bcache *bcache)
 330 {
 331   if (bcache == NULL)
 332     return;
 333   /* Only free the obstack if we actually initialized it.  */
 334   if (bcache->total_count > 0)
 335     obstack_free (&bcache->cache, 0);
 336   xfree (bcache->bucket);
 337   xfree (bcache);
 338 }
 339
 340
 341 \f
 342 /* Printing statistics.  */
 343
 344 static void
 345 print_percentage (int portion, int total)
 346 {
 347   if (total == 0)
 348     /* i18n: Like "Percentage of duplicates, by count: (not applicable)".  */
 349     printf_filtered (_("(not applicable)\n"));
 350   else
 351     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 352 }
 353
 354
 355 /* Print statistics on BCACHE's memory usage and efficacity at
 356    eliminating duplication.  NAME should describe the kind of data
 357    BCACHE holds.  Statistics are printed using `printf_filtered' and
 358    its ilk.  */
 359 void
 360 print_bcache_statistics (struct bcache *c, char *type)
 361 {
 362   int occupied_buckets;
 363   int max_chain_length;
 364   int median_chain_length;
 365   int max_entry_size;
 366   int median_entry_size;
 367
 368   /* Count the number of occupied buckets, tally the various string
 369      lengths, and measure chain lengths.  */
 370   {
 371     unsigned int b;
 372     int *chain_length = XCNEWVEC (int, c->num_buckets + 1);
 373     int *entry_size = XCNEWVEC (int, c->unique_count + 1);
 374     int stringi = 0;
 375
 376     occupied_buckets = 0;
 377
 378     for (b = 0; b < c->num_buckets; b++)
 379       {
 380         struct bstring *s = c->bucket[b];
 381
 382         chain_length[b] = 0;
 383
 384         if (s)
 385           {
 386             occupied_buckets++;
 387
 388             while (s)
 389               {
 390                 gdb_assert (b < c->num_buckets);
 391                 chain_length[b]++;
 392                 gdb_assert (stringi < c->unique_count);
 393                 entry_size[stringi++] = s->length;
 394                 s = s->next;
 395               }
 396           }
 397       }
 398
 399     /* To compute the median, we need the set of chain lengths
 400        sorted.  */
 401     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 402            compare_positive_ints);
 403     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 404            compare_positive_ints);
 405
 406     if (c->num_buckets > 0)
 407       {
 408         max_chain_length = chain_length[c->num_buckets - 1];
 409         median_chain_length = chain_length[c->num_buckets / 2];
 410       }
 411     else
 412       {
 413         max_chain_length = 0;
 414         median_chain_length = 0;
 415       }
 416     if (c->unique_count > 0)
 417       {
 418         max_entry_size = entry_size[c->unique_count - 1];
 419         median_entry_size = entry_size[c->unique_count / 2];
 420       }
 421     else
 422       {
 423         max_entry_size = 0;
 424         median_entry_size = 0;
 425       }
 426
 427     xfree (chain_length);
 428     xfree (entry_size);
 429   }
 430
 431   printf_filtered (_("  Cached '%s' statistics:\n"), type);
 432   printf_filtered (_("    Total object count:  %ld\n"), c->total_count);
 433   printf_filtered (_("    Unique object count: %lu\n"), c->unique_count);
 434   printf_filtered (_("    Percentage of duplicates, by count: "));
 435   print_percentage (c->total_count - c->unique_count, c->total_count);
 436   printf_filtered ("\n");
 437
 438   printf_filtered (_("    Total object size:   %ld\n"), c->total_size);
 439   printf_filtered (_("    Unique object size:  %ld\n"), c->unique_size);
 440   printf_filtered (_("    Percentage of duplicates, by size:  "));
 441   print_percentage (c->total_size - c->unique_size, c->total_size);
 442   printf_filtered ("\n");
 443
 444   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 445   printf_filtered (_("    Average entry size: "));
 446   if (c->unique_count > 0)
 447     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 448   else
 449     /* i18n: "Average entry size: (not applicable)".  */
 450     printf_filtered (_("(not applicable)\n"));
 451   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 452   printf_filtered ("\n");
 453
 454   printf_filtered (_("    \
 455 Total memory used by bcache, including overhead: %ld\n"),
 456                    c->structure_size);
 457   printf_filtered (_("    Percentage memory overhead: "));
 458   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 459   printf_filtered (_("    Net memory savings:         "));
 460   print_percentage (c->total_size - c->structure_size, c->total_size);
 461   printf_filtered ("\n");
 462
 463   printf_filtered (_("    Hash table size:           %3d\n"),
 464                    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"),
 482                    max_chain_length);
 483   printf_filtered ("\n");
 484 }
 485
 486 int
 487 bcache_memory_used (struct bcache *bcache)
 488 {
 489   if (bcache->total_count == 0)
 490     return 0;
 491   return obstack_memory_used (&bcache->cache);
 492 }