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) 1999-2019 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 /* The old hash function was stolen from SDBM. This is what DB 3.0
  53    uses now, and is better than the old one.  */
  54 \f
  55 unsigned long
  56 hash(const void *addr, int length)
  57 {
  58   return hash_continue (addr, length, 0);
  59 }
  60
  61 /* Continue the calculation of the hash H at the given address.  */
  62
  63 unsigned long
  64 hash_continue (const void *addr, int length, unsigned long h)
  65 {
  66   const unsigned char *k, *e;
  67
  68   k = (const unsigned char *)addr;
  69   e = k+length;
  70   for (; k< e;++k)
  71     {
  72       h *=16777619;
  73       h ^= *k;
  74     }
  75   return (h);
  76 }
  77 \f
  78 /* Growing the bcache's hash table.  */
  79
  80 /* If the average chain length grows beyond this, then we want to
  81    resize our hash table.  */
  82 #define CHAIN_LENGTH_THRESHOLD (5)
  83
  84 void
  85 bcache::expand_hash_table ()
  86 {
  87   /* A table of good hash table sizes.  Whenever we grow, we pick the
  88      next larger size from this table.  sizes[i] is close to 1 << (i+10),
  89      so we roughly double the table size each time.  After we fall off
  90      the end of this table, we just double.  Don't laugh --- there have
  91      been executables sighted with a gigabyte of debug info.  */
  92   static unsigned long sizes[] = {
  93     1021, 2053, 4099, 8191, 16381, 32771,
  94     65537, 131071, 262144, 524287, 1048573, 2097143,
  95     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
  96     268435459, 536870923, 1073741827, 2147483659UL
  97   };
  98   unsigned int new_num_buckets;
  99   struct bstring **new_buckets;
 100   unsigned int i;
 101
 102   /* Count the stats.  Every unique item needs to be re-hashed and
 103      re-entered.  */
 104   m_expand_count++;
 105   m_expand_hash_count += m_unique_count;
 106
 107   /* Find the next size.  */
 108   new_num_buckets = m_num_buckets * 2;
 109   for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
 110     if (sizes[i] > m_num_buckets)
 111       {
 112         new_num_buckets = sizes[i];
 113         break;
 114       }
 115
 116   /* Allocate the new table.  */
 117   {
 118     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
 119
 120     new_buckets = (struct bstring **) xmalloc (new_size);
 121     memset (new_buckets, 0, new_size);
 122
 123     m_structure_size -= m_num_buckets * sizeof (m_bucket[0]);
 124     m_structure_size += new_size;
 125   }
 126
 127   /* Rehash all existing strings.  */
 128   for (i = 0; i < m_num_buckets; i++)
 129     {
 130       struct bstring *s, *next;
 131
 132       for (s = m_bucket[i]; s; s = next)
 133         {
 134           struct bstring **new_bucket;
 135           next = s->next;
 136
 137           new_bucket = &new_buckets[(m_hash_function (&s->d.data, s->length)
 138                                      % new_num_buckets)];
 139           s->next = *new_bucket;
 140           *new_bucket = s;
 141         }
 142     }
 143
 144   /* Plug in the new table.  */
 145   xfree (m_bucket);
 146   m_bucket = new_buckets;
 147   m_num_buckets = new_num_buckets;
 148 }
 149
 150 \f
 151 /* Looking up things in the bcache.  */
 152
 153 /* The number of bytes needed to allocate a struct bstring whose data
 154    is N bytes long.  */
 155 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 156
 157 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 158    never seen those bytes before, add a copy of them to BCACHE.  In
 159    either case, return a pointer to BCACHE's copy of that string.  If
 160    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 161    returning an old entry.  */
 162
 163 const void *
 164 bcache::insert (const void *addr, int length, int *added)
 165 {
 166   unsigned long full_hash;
 167   unsigned short half_hash;
 168   int hash_index;
 169   struct bstring *s;
 170
 171   if (added)
 172     *added = 0;
 173
 174   /* Lazily initialize the obstack.  This can save quite a bit of
 175      memory in some cases.  */
 176   if (m_total_count == 0)
 177     {
 178       /* We could use obstack_specify_allocation here instead, but
 179          gdb_obstack.h specifies the allocation/deallocation
 180          functions.  */
 181       obstack_init (&m_cache);
 182     }
 183
 184   /* If our average chain length is too high, expand the hash table.  */
 185   if (m_unique_count >= m_num_buckets * CHAIN_LENGTH_THRESHOLD)
 186     expand_hash_table ();
 187
 188   m_total_count++;
 189   m_total_size += length;
 190
 191   full_hash = m_hash_function (addr, length);
 192
 193   half_hash = (full_hash >> 16);
 194   hash_index = full_hash % m_num_buckets;
 195
 196   /* Search the hash m_bucket for a string identical to the caller's.
 197      As a short-circuit first compare the upper part of each hash
 198      values.  */
 199   for (s = m_bucket[hash_index]; s; s = s->next)
 200     {
 201       if (s->half_hash == half_hash)
 202         {
 203           if (s->length == length
 204               && m_compare_function (&s->d.data, addr, length))
 205             return &s->d.data;
 206           else
 207             m_half_hash_miss_count++;
 208         }
 209     }
 210
 211   /* The user's string isn't in the list.  Insert it after *ps.  */
 212   {
 213     struct bstring *newobj
 214       = (struct bstring *) obstack_alloc (&m_cache,
 215                                           BSTRING_SIZE (length));
 216
 217     memcpy (&newobj->d.data, addr, length);
 218     newobj->length = length;
 219     newobj->next = m_bucket[hash_index];
 220     newobj->half_hash = half_hash;
 221     m_bucket[hash_index] = newobj;
 222
 223     m_unique_count++;
 224     m_unique_size += length;
 225     m_structure_size += BSTRING_SIZE (length);
 226
 227     if (added)
 228       *added = 1;
 229
 230     return &newobj->d.data;
 231   }
 232 }
 233 \f
 234
 235 /* Compare the byte string at ADDR1 of lenght LENGHT to the
 236    string at ADDR2.  Return 1 if they are equal.  */
 237
 238 int
 239 bcache::compare (const void *addr1, const void *addr2, int length)
 240 {
 241   return memcmp (addr1, addr2, length) == 0;
 242 }
 243
 244 /* Free all the storage associated with BCACHE.  */
 245 bcache::~bcache ()
 246 {
 247   /* Only free the obstack if we actually initialized it.  */
 248   if (m_total_count > 0)
 249     obstack_free (&m_cache, 0);
 250   xfree (m_bucket);
 251 }
 252
 253
 254 \f
 255 /* Printing statistics.  */
 256
 257 static void
 258 print_percentage (int portion, int total)
 259 {
 260   if (total == 0)
 261     /* i18n: Like "Percentage of duplicates, by count: (not applicable)".  */
 262     printf_filtered (_("(not applicable)\n"));
 263   else
 264     printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
 265 }
 266
 267
 268 /* Print statistics on BCACHE's memory usage and efficacity at
 269    eliminating duplication.  NAME should describe the kind of data
 270    BCACHE holds.  Statistics are printed using `printf_filtered' and
 271    its ilk.  */
 272 void
 273 bcache::print_statistics (const char *type)
 274 {
 275   int occupied_buckets;
 276   int max_chain_length;
 277   int median_chain_length;
 278   int max_entry_size;
 279   int median_entry_size;
 280
 281   /* Count the number of occupied buckets, tally the various string
 282      lengths, and measure chain lengths.  */
 283   {
 284     unsigned int b;
 285     int *chain_length = XCNEWVEC (int, m_num_buckets + 1);
 286     int *entry_size = XCNEWVEC (int, m_unique_count + 1);
 287     int stringi = 0;
 288
 289     occupied_buckets = 0;
 290
 291     for (b = 0; b < m_num_buckets; b++)
 292       {
 293         struct bstring *s = m_bucket[b];
 294
 295         chain_length[b] = 0;
 296
 297         if (s)
 298           {
 299             occupied_buckets++;
 300
 301             while (s)
 302               {
 303                 gdb_assert (b < m_num_buckets);
 304                 chain_length[b]++;
 305                 gdb_assert (stringi < m_unique_count);
 306                 entry_size[stringi++] = s->length;
 307                 s = s->next;
 308               }
 309           }
 310       }
 311
 312     /* To compute the median, we need the set of chain lengths
 313        sorted.  */
 314     qsort (chain_length, m_num_buckets, sizeof (chain_length[0]),
 315            compare_positive_ints);
 316     qsort (entry_size, m_unique_count, sizeof (entry_size[0]),
 317            compare_positive_ints);
 318
 319     if (m_num_buckets > 0)
 320       {
 321         max_chain_length = chain_length[m_num_buckets - 1];
 322         median_chain_length = chain_length[m_num_buckets / 2];
 323       }
 324     else
 325       {
 326         max_chain_length = 0;
 327         median_chain_length = 0;
 328       }
 329     if (m_unique_count > 0)
 330       {
 331         max_entry_size = entry_size[m_unique_count - 1];
 332         median_entry_size = entry_size[m_unique_count / 2];
 333       }
 334     else
 335       {
 336         max_entry_size = 0;
 337         median_entry_size = 0;
 338       }
 339
 340     xfree (chain_length);
 341     xfree (entry_size);
 342   }
 343
 344   printf_filtered (_("  M_Cached '%s' statistics:\n"), type);
 345   printf_filtered (_("    Total object count:  %ld\n"), m_total_count);
 346   printf_filtered (_("    Unique object count: %lu\n"), m_unique_count);
 347   printf_filtered (_("    Percentage of duplicates, by count: "));
 348   print_percentage (m_total_count - m_unique_count, m_total_count);
 349   printf_filtered ("\n");
 350
 351   printf_filtered (_("    Total object size:   %ld\n"), m_total_size);
 352   printf_filtered (_("    Unique object size:  %ld\n"), m_unique_size);
 353   printf_filtered (_("    Percentage of duplicates, by size:  "));
 354   print_percentage (m_total_size - m_unique_size, m_total_size);
 355   printf_filtered ("\n");
 356
 357   printf_filtered (_("    Max entry size:     %d\n"), max_entry_size);
 358   printf_filtered (_("    Average entry size: "));
 359   if (m_unique_count > 0)
 360     printf_filtered ("%ld\n", m_unique_size / m_unique_count);
 361   else
 362     /* i18n: "Average entry size: (not applicable)".  */
 363     printf_filtered (_("(not applicable)\n"));
 364   printf_filtered (_("    Median entry size:  %d\n"), median_entry_size);
 365   printf_filtered ("\n");
 366
 367   printf_filtered (_("    \
 368 Total memory used by bcache, including overhead: %ld\n"),
 369                    m_structure_size);
 370   printf_filtered (_("    Percentage memory overhead: "));
 371   print_percentage (m_structure_size - m_unique_size, m_unique_size);
 372   printf_filtered (_("    Net memory savings:         "));
 373   print_percentage (m_total_size - m_structure_size, m_total_size);
 374   printf_filtered ("\n");
 375
 376   printf_filtered (_("    Hash table size:           %3d\n"),
 377                    m_num_buckets);
 378   printf_filtered (_("    Hash table expands:        %lu\n"),
 379                    m_expand_count);
 380   printf_filtered (_("    Hash table hashes:         %lu\n"),
 381                    m_total_count + m_expand_hash_count);
 382   printf_filtered (_("    Half hash misses:          %lu\n"),
 383                    m_half_hash_miss_count);
 384   printf_filtered (_("    Hash table population:     "));
 385   print_percentage (occupied_buckets, m_num_buckets);
 386   printf_filtered (_("    Median hash chain length:  %3d\n"),
 387                    median_chain_length);
 388   printf_filtered (_("    Average hash chain length: "));
 389   if (m_num_buckets > 0)
 390     printf_filtered ("%3lu\n", m_unique_count / m_num_buckets);
 391   else
 392     /* i18n: "Average hash chain length: (not applicable)".  */
 393     printf_filtered (_("(not applicable)\n"));
 394   printf_filtered (_("    Maximum hash chain length: %3d\n"),
 395                    max_chain_length);
 396   printf_filtered ("\n");
 397 }
 398
 399 int
 400 bcache::memory_used ()
 401 {
 402   if (m_total_count == 0)
 403     return 0;
 404   return obstack_memory_used (&m_cache);
 405 }