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