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, 2011
   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
 101    uses now, and is better than the old one.  */
 102 \f
 103 unsigned long
 104 hash(const void *addr, int length)
 105 {
 106   return hash_continue (addr, length, 0);
 107 }
 108
 109 /* Continue the calculation of the hash H at the given address.  */
 110
 111 unsigned long
 112 hash_continue (const void *addr, int length, unsigned long h)
 113 {
 114   const unsigned char *k, *e;
 115
 116   k = (const unsigned char *)addr;
 117   e = k+length;
 118   for (; k< e;++k)
 119     {
 120       h *=16777619;
 121       h ^= *k;
 122     }
 123   return (h);
 124 }
 125 \f
 126 /* Growing the bcache's hash table.  */
 127
 128 /* If the average chain length grows beyond this, then we want to
 129    resize our hash table.  */
 130 #define CHAIN_LENGTH_THRESHOLD (5)
 131
 132 static void
 133 expand_hash_table (struct bcache *bcache)
 134 {
 135   /* A table of good hash table sizes.  Whenever we grow, we pick the
 136      next larger size from this table.  sizes[i] is close to 1 << (i+10),
 137      so we roughly double the table size each time.  After we fall off
 138      the end of this table, we just double.  Don't laugh --- there have
 139      been executables sighted with a gigabyte of debug info.  */
 140   static unsigned long sizes[] = {
 141     1021, 2053, 4099, 8191, 16381, 32771,
 142     65537, 131071, 262144, 524287, 1048573, 2097143,
 143     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
 144     268435459, 536870923, 1073741827, 2147483659UL
 145   };
 146   unsigned int new_num_buckets;
 147   struct bstring **new_buckets;
 148   unsigned int i;
 149
 150   /* Count the stats.  Every unique item needs to be re-hashed and
 151      re-entered.  */
 152   bcache->expand_count++;
 153   bcache->expand_hash_count += bcache->unique_count;
 154
 155   /* Find the next size.  */
 156   new_num_buckets = bcache->num_buckets * 2;
 157   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 158     if (sizes[i] > bcache->num_buckets)
 159       {
 160         new_num_buckets = sizes[i];
 161         break;
 162       }
 163
 164   /* Allocate the new table.  */
 165   {
 166     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 167
 168     new_buckets = (struct bstring **) xmalloc (new_size);
 169     memset (new_buckets, 0, new_size);
 170
 171     bcache->structure_size -= (bcache->num_buckets
 172                                * sizeof (bcache->bucket[0]));
 173     bcache->structure_size += new_size;
 174   }
 175
 176   /* Rehash all existing strings.  */
 177   for (i = 0; i < bcache->num_buckets; i++)
 178     {
 179       struct bstring *s, *next;
 180
 181       for (s = bcache->bucket[i]; s; s = next)
 182         {
 183           struct bstring **new_bucket;
 184           next = s->next;
 185
 186           new_bucket = &new_buckets[(bcache->hash_function (&s->d.data,
 187                                                             s->length)
 188                                      % new_num_buckets)];
 189           s->next = *new_bucket;
 190           *new_bucket = s;
 191         }
 192     }
 193
 194   /* Plug in the new table.  */
 195   if (bcache->bucket)
 196     xfree (bcache->bucket);
 197   bcache->bucket = new_buckets;
 198   bcache->num_buckets = new_num_buckets;
 199 }
 200
 201 \f
 202 /* Looking up things in the bcache.  */
 203
 204 /* The number of bytes needed to allocate a struct bstring whose data
 205    is N bytes long.  */
 206 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 207
 208 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 209    never seen those bytes before, add a copy of them to BCACHE.  In
 210    either case, return a pointer to BCACHE's copy of that string.  */
 211 const void *
 212 bcache (const void *addr, int length, struct bcache *bcache)
 213 {
 214   return bcache_full (addr, length, bcache, NULL);
 215 }
 216
 217 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 218    never seen those bytes before, add a copy of them to BCACHE.  In
 219    either case, return a pointer to BCACHE's copy of that string.  If
 220    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 221    returning an old entry.  */
 222
 223 const void *
 224 bcache_full (const void *addr, int length, struct bcache *bcache, int *added)
 225 {
 226   unsigned long full_hash;
 227   unsigned short half_hash;
 228   int hash_index;
 229   struct bstring *s;
 230
 231   if (added)
 232     *added = 0;
 233
 234   /* Lazily initialize the obstack.  This can save quite a bit of
 235      memory in some cases.  */
 236   if (bcache->total_count == 0)
 237     {
 238       /* We could use obstack_specify_allocation here instead, but
 239          gdb_obstack.h specifies the allocation/deallocation
 240          functions.  */
 241       obstack_init (&bcache->cache);
 242     }
 243
 244   /* If our average chain length is too high, expand the hash table.  */
 245   if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
 246     expand_hash_table (bcache);
 247
 248   bcache->total_count++;
 249   bcache->total_size += length;
 250
 251   full_hash = bcache->hash_function (addr, length);
 252
 253   half_hash = (full_hash >> 16);
 254   hash_index = full_hash % bcache->num_buckets;
 255
 256   /* Search the hash bucket for a string identical to the caller's.
 257      As a short-circuit first compare the upper part of each hash
 258      values.  */
 259   for (s = bcache->bucket[hash_index]; s; s = s->next)
 260     {
 261       if (s->half_hash == half_hash)
 262         {
 263           if (s->length == length
 264               && bcache->compare_function (&s->d.data, addr, length))
 265             return &s->d.data;
 266           else
 267             bcache->half_hash_miss_count++;
 268         }
 269     }
 270
 271   /* The user's string isn't in the list.  Insert it after *ps.  */
 272   {
 273     struct bstring *new
 274       = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
 275
 276     memcpy (&new->d.data, addr, length);
 277     new->length = length;
 278     new->next = bcache->bucket[hash_index];
 279     new->half_hash = half_hash;
 280     bcache->bucket[hash_index] = new;
 281
 282     bcache->unique_count++;
 283     bcache->unique_size += length;
 284     bcache->structure_size += BSTRING_SIZE (length);
 285
 286     if (added)
 287       *added = 1;
 288
 289     return &new->d.data;
 290   }
 291 }
 292 \f
 293
 294 /* Compare the byte string at ADDR1 of lenght LENGHT to the
 295    string at ADDR2.  Return 1 if they are equal.  */
 296
 297 static int
 298 bcache_compare (const void *addr1, const void *addr2, int length)
 299 {
 300   return memcmp (addr1, addr2, length) == 0;
 301 }
 302
 303 /* Allocating and freeing bcaches.  */
 304
 305 /* Allocated a bcache.  HASH_FUNCTION and COMPARE_FUNCTION can be used
 306    to pass in custom hash, and compare functions to be used by this
 307    bcache.  If HASH_FUNCTION is NULL hash() is used and if
 308    COMPARE_FUNCTION is NULL memcmp() is used.  */
 309
 310 struct bcache *
 311 bcache_xmalloc (unsigned long (*hash_function)(const void *, int length),
 312                 int (*compare_function)(const void *,
 313                                         const void *,
 314                                         int length))
 315 {
 316   /* Allocate the bcache pre-zeroed.  */
 317   struct bcache *b = XCALLOC (1, struct bcache);
 318
 319   if (hash_function)
 320     b->hash_function = hash_function;
 321   else
 322     b->hash_function = hash;
 323
 324   if (compare_function)
 325     b->compare_function = compare_function;
 326   else
 327     b->compare_function = bcache_compare;
 328   return b;
 329 }
 330
 331 /* Free all the storage associated with BCACHE.  */
 332 void
 333 bcache_xfree (struct bcache *bcache)
 334 {
 335   if (bcache == NULL)
 336     return;
 337   /* Only free the obstack if we actually initialized it.  */
 338   if (bcache->total_count > 0)
 339     obstack_free (&bcache->cache, 0);
 340   xfree (bcache->bucket);
 341   xfree (bcache);
 342 }
 343
 344
 345 \f
 346 /* Printing statistics.  */
 347
 348 static void
 349 print_percentage (int portion, int total)
 350 {
 351   if (total == 0)
 352     /* i18n: Like "Percentage of duplicates, by count: (not applicable)".  */
 353     printf_filtered (_("(not applicable)\n"));
 354   else
 355     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 356 }
 357
 358
 359 /* Print statistics on BCACHE's memory usage and efficacity at
 360    eliminating duplication.  NAME should describe the kind of data
 361    BCACHE holds.  Statistics are printed using `printf_filtered' and
 362    its ilk.  */
 363 void
 364 print_bcache_statistics (struct bcache *c, char *type)
 365 {
 366   int occupied_buckets;
 367   int max_chain_length;
 368   int median_chain_length;
 369   int max_entry_size;
 370   int median_entry_size;
 371
 372   /* Count the number of occupied buckets, tally the various string
 373      lengths, and measure chain lengths.  */
 374   {
 375     unsigned int b;
 376     int *chain_length = XCALLOC (c->num_buckets + 1, int);
 377     int *entry_size = XCALLOC (c->unique_count + 1, int);
 378     int stringi = 0;
 379
 380     occupied_buckets = 0;
 381
 382     for (b = 0; b < c->num_buckets; b++)
 383       {
 384         struct bstring *s = c->bucket[b];
 385
 386         chain_length[b] = 0;
 387
 388         if (s)
 389           {
 390             occupied_buckets++;
 391
 392             while (s)
 393               {
 394                 gdb_assert (b < c->num_buckets);
 395                 chain_length[b]++;
 396                 gdb_assert (stringi < c->unique_count);
 397                 entry_size[stringi++] = s->length;
 398                 s = s->next;
 399               }
 400           }
 401       }
 402
 403     /* To compute the median, we need the set of chain lengths
 404        sorted.  */
 405     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 406            compare_positive_ints);
 407     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 408            compare_positive_ints);
 409
 410     if (c->num_buckets > 0)
 411       {
 412         max_chain_length = chain_length[c->num_buckets - 1];
 413         median_chain_length = chain_length[c->num_buckets / 2];
 414       }
 415     else
 416       {
 417         max_chain_length = 0;
 418         median_chain_length = 0;
 419       }
 420     if (c->unique_count > 0)
 421       {
 422         max_entry_size = entry_size[c->unique_count - 1];
 423         median_entry_size = entry_size[c->unique_count / 2];
 424       }
 425     else
 426       {
 427         max_entry_size = 0;
 428         median_entry_size = 0;
 429       }
 430
 431     xfree (chain_length);
 432     xfree (entry_size);
 433   }
 434
 435   printf_filtered (_("  Cached '%s' statistics:\n"), type);
 436   printf_filtered (_("    Total object count:  %ld\n"), c->total_count);
 437   printf_filtered (_("    Unique object count: %lu\n"), c->unique_count);
 438   printf_filtered (_("    Percentage of duplicates, by count: "));
 439   print_percentage (c->total_count - c->unique_count, c->total_count);
 440   printf_filtered ("\n");
 441
 442   printf_filtered (_("    Total object size:   %ld\n"), c->total_size);
 443   printf_filtered (_("    Unique object size:  %ld\n"), c->unique_size);
 444   printf_filtered (_("    Percentage of duplicates, by size:  "));
 445   print_percentage (c->total_size - c->unique_size, c->total_size);
 446   printf_filtered ("\n");
 447
 448   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 449   printf_filtered (_("    Average entry size: "));
 450   if (c->unique_count > 0)
 451     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 452   else
 453     /* i18n: "Average entry size: (not applicable)".  */
 454     printf_filtered (_("(not applicable)\n"));
 455   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 456   printf_filtered ("\n");
 457
 458   printf_filtered (_("    \
 459 Total memory used by bcache, including overhead: %ld\n"),
 460                    c->structure_size);
 461   printf_filtered (_("    Percentage memory overhead: "));
 462   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 463   printf_filtered (_("    Net memory savings:         "));
 464   print_percentage (c->total_size - c->structure_size, c->total_size);
 465   printf_filtered ("\n");
 466
 467   printf_filtered (_("    Hash table size:           %3d\n"),
 468                    c->num_buckets);
 469   printf_filtered (_("    Hash table expands:        %lu\n"),
 470                    c->expand_count);
 471   printf_filtered (_("    Hash table hashes:         %lu\n"),
 472                    c->total_count + c->expand_hash_count);
 473   printf_filtered (_("    Half hash misses:          %lu\n"),
 474                    c->half_hash_miss_count);
 475   printf_filtered (_("    Hash table population:     "));
 476   print_percentage (occupied_buckets, c->num_buckets);
 477   printf_filtered (_("    Median hash chain length:  %3d\n"),
 478                    median_chain_length);
 479   printf_filtered (_("    Average hash chain length: "));
 480   if (c->num_buckets > 0)
 481     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 482   else
 483     /* i18n: "Average hash chain length: (not applicable)".  */
 484     printf_filtered (_("(not applicable)\n"));
 485   printf_filtered (_("    Maximum hash chain length: %3d\n"),
 486                    max_chain_length);
 487   printf_filtered ("\n");
 488 }
 489
 490 int
 491 bcache_memory_used (struct bcache *bcache)
 492 {
 493   if (bcache->total_count == 0)
 494     return 0;
 495   return obstack_memory_used (&bcache->cache);
 496 }