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 1999, 2000, 2002, 2003 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 2 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, write to the Free Software
  21    Foundation, Inc., 59 Temple Place - Suite 330,
  22    Boston, MA 02111-1307, USA.  */
  23
  24 #include "defs.h"
  25 #include "gdb_obstack.h"
  26 #include "bcache.h"
  27 #include "gdb_string.h"         /* For memcpy declaration */
  28 #include "gdb_assert.h"
  29
  30 #include <stddef.h>
  31 #include <stdlib.h>
  32
  33 /* The type used to hold a single bcache string.  The user data is
  34    stored in d.data.  Since it can be any type, it needs to have the
  35    same alignment as the most strict alignment of any type on the host
  36    machine.  I don't know of any really correct way to do this in
  37    stock ANSI C, so just do it the same way obstack.h does.  */
  38
  39 struct bstring
  40 {
  41   /* Hash chain.  */
  42   struct bstring *next;
  43   /* Assume the data length is no more than 64k.  */
  44   unsigned short length;
  45   /* The half hash hack.  This contains the upper 16 bits of the hash
  46      value and is used as a pre-check when comparing two strings and
  47      avoids the need to do length or memcmp calls.  It proves to be
  48      roughly 100% effective.  */
  49   unsigned short half_hash;
  50
  51   union
  52   {
  53     char data[1];
  54     double dummy;
  55   }
  56   d;
  57 };
  58
  59
  60 /* The structure for a bcache itself.  The bcache is initialized, in
  61    bcache_xmalloc(), by filling it with zeros and then setting the
  62    corresponding obstack's malloc() and free() methods.  */
  63
  64 struct bcache
  65 {
  66   /* All the bstrings are allocated here.  */
  67   struct obstack cache;
  68
  69   /* How many hash buckets we're using.  */
  70   unsigned int num_buckets;
  71
  72   /* Hash buckets.  This table is allocated using malloc, so when we
  73      grow the table we can return the old table to the system.  */
  74   struct bstring **bucket;
  75
  76   /* Statistics.  */
  77   unsigned long unique_count;   /* number of unique strings */
  78   long total_count;     /* total number of strings cached, including dups */
  79   long unique_size;     /* size of unique strings, in bytes */
  80   long total_size;      /* total number of bytes cached, including dups */
  81   long structure_size;  /* total size of bcache, including infrastructure */
  82   /* Number of times that the hash table is expanded and hence
  83      re-built, and the corresponding number of times that a string is
  84      [re]hashed as part of entering it into the expanded table.  The
  85      total number of hashes can be computed by adding TOTAL_COUNT to
  86      expand_hash_count.  */
  87   unsigned long expand_count;
  88   unsigned long expand_hash_count;
  89   /* Number of times that the half-hash compare hit (compare the upper
  90      16 bits of hash values) hit, but the corresponding combined
  91      length/data compare missed.  */
  92   unsigned long half_hash_miss_count;
  93 };
  94
  95 /* The old hash function was stolen from SDBM. This is what DB 3.0 uses now,
  96  * and is better than the old one.
  97  */
  98 \f
  99 unsigned long
 100 hash(const void *addr, int length)
 101 {
 102                 const unsigned char *k, *e;
 103                 unsigned long h;
 104
 105                 k = (const unsigned char *)addr;
 106                 e = k+length;
 107                 for (h=0; k< e;++k)
 108                 {
 109                                 h *=16777619;
 110                                 h ^= *k;
 111                 }
 112                 return (h);
 113 }
 114 \f
 115 /* Growing the bcache's hash table.  */
 116
 117 /* If the average chain length grows beyond this, then we want to
 118    resize our hash table.  */
 119 #define CHAIN_LENGTH_THRESHOLD (5)
 120
 121 static void
 122 expand_hash_table (struct bcache *bcache)
 123 {
 124   /* A table of good hash table sizes.  Whenever we grow, we pick the
 125      next larger size from this table.  sizes[i] is close to 1 << (i+10),
 126      so we roughly double the table size each time.  After we fall off
 127      the end of this table, we just double.  Don't laugh --- there have
 128      been executables sighted with a gigabyte of debug info.  */
 129   static unsigned long sizes[] = {
 130     1021, 2053, 4099, 8191, 16381, 32771,
 131     65537, 131071, 262144, 524287, 1048573, 2097143,
 132     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
 133     268435459, 536870923, 1073741827, 2147483659UL
 134   };
 135   unsigned int new_num_buckets;
 136   struct bstring **new_buckets;
 137   unsigned int i;
 138
 139   /* Count the stats.  Every unique item needs to be re-hashed and
 140      re-entered.  */
 141   bcache->expand_count++;
 142   bcache->expand_hash_count += bcache->unique_count;
 143
 144   /* Find the next size.  */
 145   new_num_buckets = bcache->num_buckets * 2;
 146   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 147     if (sizes[i] > bcache->num_buckets)
 148       {
 149         new_num_buckets = sizes[i];
 150         break;
 151       }
 152
 153   /* Allocate the new table.  */
 154   {
 155     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 156     new_buckets = (struct bstring **) xmalloc (new_size);
 157     memset (new_buckets, 0, new_size);
 158
 159     bcache->structure_size -= (bcache->num_buckets
 160                                * sizeof (bcache->bucket[0]));
 161     bcache->structure_size += new_size;
 162   }
 163
 164   /* Rehash all existing strings.  */
 165   for (i = 0; i < bcache->num_buckets; i++)
 166     {
 167       struct bstring *s, *next;
 168
 169       for (s = bcache->bucket[i]; s; s = next)
 170         {
 171           struct bstring **new_bucket;
 172           next = s->next;
 173
 174           new_bucket = &new_buckets[(hash (&s->d.data, s->length)
 175                                      % new_num_buckets)];
 176           s->next = *new_bucket;
 177           *new_bucket = s;
 178         }
 179     }
 180
 181   /* Plug in the new table.  */
 182   if (bcache->bucket)
 183     xfree (bcache->bucket);
 184   bcache->bucket = new_buckets;
 185   bcache->num_buckets = new_num_buckets;
 186 }
 187
 188 \f
 189 /* Looking up things in the bcache.  */
 190
 191 /* The number of bytes needed to allocate a struct bstring whose data
 192    is N bytes long.  */
 193 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 194
 195 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 196    never seen those bytes before, add a copy of them to BCACHE.  In
 197    either case, return a pointer to BCACHE's copy of that string.  */
 198 void *
 199 bcache (const void *addr, int length, struct bcache *bcache)
 200 {
 201   unsigned long full_hash;
 202   unsigned short half_hash;
 203   int hash_index;
 204   struct bstring *s;
 205
 206   /* If our average chain length is too high, expand the hash table.  */
 207   if (bcache->unique_count >= bcache->num_buckets * CHAIN_LENGTH_THRESHOLD)
 208     expand_hash_table (bcache);
 209
 210   bcache->total_count++;
 211   bcache->total_size += length;
 212
 213   full_hash = hash (addr, length);
 214   half_hash = (full_hash >> 16);
 215   hash_index = full_hash % bcache->num_buckets;
 216
 217   /* Search the hash bucket for a string identical to the caller's.
 218      As a short-circuit first compare the upper part of each hash
 219      values.  */
 220   for (s = bcache->bucket[hash_index]; s; s = s->next)
 221     {
 222       if (s->half_hash == half_hash)
 223         {
 224           if (s->length == length
 225               && ! memcmp (&s->d.data, addr, length))
 226             return &s->d.data;
 227           else
 228             bcache->half_hash_miss_count++;
 229         }
 230     }
 231
 232   /* The user's string isn't in the list.  Insert it after *ps.  */
 233   {
 234     struct bstring *new
 235       = obstack_alloc (&bcache->cache, BSTRING_SIZE (length));
 236     memcpy (&new->d.data, addr, length);
 237     new->length = length;
 238     new->next = bcache->bucket[hash_index];
 239     new->half_hash = half_hash;
 240     bcache->bucket[hash_index] = new;
 241
 242     bcache->unique_count++;
 243     bcache->unique_size += length;
 244     bcache->structure_size += BSTRING_SIZE (length);
 245
 246     return &new->d.data;
 247   }
 248 }
 249
 250 \f
 251 /* Allocating and freeing bcaches.  */
 252
 253 struct bcache *
 254 bcache_xmalloc (void)
 255 {
 256   /* Allocate the bcache pre-zeroed.  */
 257   struct bcache *b = XCALLOC (1, struct bcache);
 258   obstack_specify_allocation (&b->cache, 0, 0, xmalloc, xfree);
 259   return b;
 260 }
 261
 262 /* Free all the storage associated with BCACHE.  */
 263 void
 264 bcache_xfree (struct bcache *bcache)
 265 {
 266   if (bcache == NULL)
 267     return;
 268   obstack_free (&bcache->cache, 0);
 269   xfree (bcache->bucket);
 270   xfree (bcache);
 271 }
 272
 273
 274 \f
 275 /* Printing statistics.  */
 276
 277 static int
 278 compare_ints (const void *ap, const void *bp)
 279 {
 280   /* Because we know we're comparing two ints which are positive,
 281      there's no danger of overflow here.  */
 282   return * (int *) ap - * (int *) bp;
 283 }
 284
 285
 286 static void
 287 print_percentage (int portion, int total)
 288 {
 289   if (total == 0)
 290     printf_filtered ("(not applicable)\n");
 291   else
 292     printf_filtered ("%3d%%\n", portion * 100 / total);
 293 }
 294
 295
 296 /* Print statistics on BCACHE's memory usage and efficacity at
 297    eliminating duplication.  NAME should describe the kind of data
 298    BCACHE holds.  Statistics are printed using `printf_filtered' and
 299    its ilk.  */
 300 void
 301 print_bcache_statistics (struct bcache *c, char *type)
 302 {
 303   int occupied_buckets;
 304   int max_chain_length;
 305   int median_chain_length;
 306   int max_entry_size;
 307   int median_entry_size;
 308
 309   /* Count the number of occupied buckets, tally the various string
 310      lengths, and measure chain lengths.  */
 311   {
 312     unsigned int b;
 313     int *chain_length = XCALLOC (c->num_buckets + 1, int);
 314     int *entry_size = XCALLOC (c->unique_count + 1, int);
 315     int stringi = 0;
 316
 317     occupied_buckets = 0;
 318
 319     for (b = 0; b < c->num_buckets; b++)
 320       {
 321         struct bstring *s = c->bucket[b];
 322
 323         chain_length[b] = 0;
 324
 325         if (s)
 326           {
 327             occupied_buckets++;
 328
 329             while (s)
 330               {
 331                 gdb_assert (b < c->num_buckets);
 332                 chain_length[b]++;
 333                 gdb_assert (stringi < c->unique_count);
 334                 entry_size[stringi++] = s->length;
 335                 s = s->next;
 336               }
 337           }
 338       }
 339
 340     /* To compute the median, we need the set of chain lengths sorted.  */
 341     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 342            compare_ints);
 343     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 344            compare_ints);
 345
 346     if (c->num_buckets > 0)
 347       {
 348         max_chain_length = chain_length[c->num_buckets - 1];
 349         median_chain_length = chain_length[c->num_buckets / 2];
 350       }
 351     else
 352       {
 353         max_chain_length = 0;
 354         median_chain_length = 0;
 355       }
 356     if (c->unique_count > 0)
 357       {
 358         max_entry_size = entry_size[c->unique_count - 1];
 359         median_entry_size = entry_size[c->unique_count / 2];
 360       }
 361     else
 362       {
 363         max_entry_size = 0;
 364         median_entry_size = 0;
 365       }
 366
 367     xfree (chain_length);
 368     xfree (entry_size);
 369   }
 370
 371   printf_filtered ("  Cached '%s' statistics:\n", type);
 372   printf_filtered ("    Total object count:  %ld\n", c->total_count);
 373   printf_filtered ("    Unique object count: %lu\n", c->unique_count);
 374   printf_filtered ("    Percentage of duplicates, by count: ");
 375   print_percentage (c->total_count - c->unique_count, c->total_count);
 376   printf_filtered ("\n");
 377
 378   printf_filtered ("    Total object size:   %ld\n", c->total_size);
 379   printf_filtered ("    Unique object size:  %ld\n", c->unique_size);
 380   printf_filtered ("    Percentage of duplicates, by size:  ");
 381   print_percentage (c->total_size - c->unique_size, c->total_size);
 382   printf_filtered ("\n");
 383
 384   printf_filtered ("    Max entry size:     %d\n", max_entry_size);
 385   printf_filtered ("    Average entry size: ");
 386   if (c->unique_count > 0)
 387     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 388   else
 389     printf_filtered ("(not applicable)\n");
 390   printf_filtered ("    Median entry size:  %d\n", median_entry_size);
 391   printf_filtered ("\n");
 392
 393   printf_filtered ("    Total memory used by bcache, including overhead: %ld\n",
 394                    c->structure_size);
 395   printf_filtered ("    Percentage memory overhead: ");
 396   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 397   printf_filtered ("    Net memory savings:         ");
 398   print_percentage (c->total_size - c->structure_size, c->total_size);
 399   printf_filtered ("\n");
 400
 401   printf_filtered ("    Hash table size:           %3d\n", c->num_buckets);
 402   printf_filtered ("    Hash table expands:        %lu\n",
 403                    c->expand_count);
 404   printf_filtered ("    Hash table hashes:         %lu\n",
 405                    c->total_count + c->expand_hash_count);
 406   printf_filtered ("    Half hash misses:          %lu\n",
 407                    c->half_hash_miss_count);
 408   printf_filtered ("    Hash table population:     ");
 409   print_percentage (occupied_buckets, c->num_buckets);
 410   printf_filtered ("    Median hash chain length:  %3d\n",
 411                    median_chain_length);
 412   printf_filtered ("    Average hash chain length: ");
 413   if (c->num_buckets > 0)
 414     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 415   else
 416     printf_filtered ("(not applicable)\n");
 417   printf_filtered ("    Maximum hash chain length: %3d\n", max_chain_length);
 418   printf_filtered ("\n");
 419 }
 420
 421 int
 422 bcache_memory_used (struct bcache *bcache)
 423 {
 424   return obstack_memory_used (&bcache->cache);
 425 }