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 static void *
 199 bcache_data (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 void *
 251 deprecated_bcache (const void *addr, int length, struct bcache *bcache)
 252 {
 253   return bcache_data (addr, length, bcache);
 254 }
 255
 256 const void *
 257 bcache (const void *addr, int length, struct bcache *bcache)
 258 {
 259   return bcache_data (addr, length, bcache);
 260 }
 261 \f
 262 /* Allocating and freeing bcaches.  */
 263
 264 struct bcache *
 265 bcache_xmalloc (void)
 266 {
 267   /* Allocate the bcache pre-zeroed.  */
 268   struct bcache *b = XCALLOC (1, struct bcache);
 269   /* We could use obstack_specify_allocation here instead, but
 270      gdb_obstack.h specifies the allocation/deallocation
 271      functions.  */
 272   obstack_init (&b->cache);
 273   return b;
 274 }
 275
 276 /* Free all the storage associated with BCACHE.  */
 277 void
 278 bcache_xfree (struct bcache *bcache)
 279 {
 280   if (bcache == NULL)
 281     return;
 282   obstack_free (&bcache->cache, 0);
 283   xfree (bcache->bucket);
 284   xfree (bcache);
 285 }
 286
 287
 288 \f
 289 /* Printing statistics.  */
 290
 291 static int
 292 compare_ints (const void *ap, const void *bp)
 293 {
 294   /* Because we know we're comparing two ints which are positive,
 295      there's no danger of overflow here.  */
 296   return * (int *) ap - * (int *) bp;
 297 }
 298
 299
 300 static void
 301 print_percentage (int portion, int total)
 302 {
 303   if (total == 0)
 304     printf_filtered ("(not applicable)\n");
 305   else
 306     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 307 }
 308
 309
 310 /* Print statistics on BCACHE's memory usage and efficacity at
 311    eliminating duplication.  NAME should describe the kind of data
 312    BCACHE holds.  Statistics are printed using `printf_filtered' and
 313    its ilk.  */
 314 void
 315 print_bcache_statistics (struct bcache *c, char *type)
 316 {
 317   int occupied_buckets;
 318   int max_chain_length;
 319   int median_chain_length;
 320   int max_entry_size;
 321   int median_entry_size;
 322
 323   /* Count the number of occupied buckets, tally the various string
 324      lengths, and measure chain lengths.  */
 325   {
 326     unsigned int b;
 327     int *chain_length = XCALLOC (c->num_buckets + 1, int);
 328     int *entry_size = XCALLOC (c->unique_count + 1, int);
 329     int stringi = 0;
 330
 331     occupied_buckets = 0;
 332
 333     for (b = 0; b < c->num_buckets; b++)
 334       {
 335         struct bstring *s = c->bucket[b];
 336
 337         chain_length[b] = 0;
 338
 339         if (s)
 340           {
 341             occupied_buckets++;
 342
 343             while (s)
 344               {
 345                 gdb_assert (b < c->num_buckets);
 346                 chain_length[b]++;
 347                 gdb_assert (stringi < c->unique_count);
 348                 entry_size[stringi++] = s->length;
 349                 s = s->next;
 350               }
 351           }
 352       }
 353
 354     /* To compute the median, we need the set of chain lengths sorted.  */
 355     qsort (chain_length, c->num_buckets, sizeof (chain_length[0]),
 356            compare_ints);
 357     qsort (entry_size, c->unique_count, sizeof (entry_size[0]),
 358            compare_ints);
 359
 360     if (c->num_buckets > 0)
 361       {
 362         max_chain_length = chain_length[c->num_buckets - 1];
 363         median_chain_length = chain_length[c->num_buckets / 2];
 364       }
 365     else
 366       {
 367         max_chain_length = 0;
 368         median_chain_length = 0;
 369       }
 370     if (c->unique_count > 0)
 371       {
 372         max_entry_size = entry_size[c->unique_count - 1];
 373         median_entry_size = entry_size[c->unique_count / 2];
 374       }
 375     else
 376       {
 377         max_entry_size = 0;
 378         median_entry_size = 0;
 379       }
 380
 381     xfree (chain_length);
 382     xfree (entry_size);
 383   }
 384
 385   printf_filtered ("  Cached '%s' statistics:\n", type);
 386   printf_filtered ("    Total object count:  %ld\n", c->total_count);
 387   printf_filtered ("    Unique object count: %lu\n", c->unique_count);
 388   printf_filtered ("    Percentage of duplicates, by count: ");
 389   print_percentage (c->total_count - c->unique_count, c->total_count);
 390   printf_filtered ("\n");
 391
 392   printf_filtered ("    Total object size:   %ld\n", c->total_size);
 393   printf_filtered ("    Unique object size:  %ld\n", c->unique_size);
 394   printf_filtered ("    Percentage of duplicates, by size:  ");
 395   print_percentage (c->total_size - c->unique_size, c->total_size);
 396   printf_filtered ("\n");
 397
 398   printf_filtered ("    Max entry size:     %d\n", max_entry_size);
 399   printf_filtered ("    Average entry size: ");
 400   if (c->unique_count > 0)
 401     printf_filtered ("%ld\n", c->unique_size / c->unique_count);
 402   else
 403     printf_filtered ("(not applicable)\n");
 404   printf_filtered ("    Median entry size:  %d\n", median_entry_size);
 405   printf_filtered ("\n");
 406
 407   printf_filtered ("    Total memory used by bcache, including overhead: %ld\n",
 408                    c->structure_size);
 409   printf_filtered ("    Percentage memory overhead: ");
 410   print_percentage (c->structure_size - c->unique_size, c->unique_size);
 411   printf_filtered ("    Net memory savings:         ");
 412   print_percentage (c->total_size - c->structure_size, c->total_size);
 413   printf_filtered ("\n");
 414
 415   printf_filtered ("    Hash table size:           %3d\n", c->num_buckets);
 416   printf_filtered ("    Hash table expands:        %lu\n",
 417                    c->expand_count);
 418   printf_filtered ("    Hash table hashes:         %lu\n",
 419                    c->total_count + c->expand_hash_count);
 420   printf_filtered ("    Half hash misses:          %lu\n",
 421                    c->half_hash_miss_count);
 422   printf_filtered ("    Hash table population:     ");
 423   print_percentage (occupied_buckets, c->num_buckets);
 424   printf_filtered ("    Median hash chain length:  %3d\n",
 425                    median_chain_length);
 426   printf_filtered ("    Average hash chain length: ");
 427   if (c->num_buckets > 0)
 428     printf_filtered ("%3lu\n", c->unique_count / c->num_buckets);
 429   else
 430     printf_filtered ("(not applicable)\n");
 431   printf_filtered ("    Maximum hash chain length: %3d\n", max_chain_length);
 432   printf_filtered ("\n");
 433 }
 434
 435 int
 436 bcache_memory_used (struct bcache *bcache)
 437 {
 438   return obstack_memory_used (&bcache->cache);
 439 }