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) 1999-2024 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 "gdbsupport/gdb_obstack.h"
  24 #include "bcache.h"
  25
  26 #include <algorithm>
  27
  28 namespace gdb {
  29
  30 /* The type used to hold a single bcache string.  The user data is
  31    stored in d.data.  Since it can be any type, it needs to have the
  32    same alignment as the most strict alignment of any type on the host
  33    machine.  I don't know of any really correct way to do this in
  34    stock ANSI C, so just do it the same way obstack.h does.  */
  35
  36 struct bstring
  37 {
  38   /* Hash chain.  */
  39   struct bstring *next;
  40   /* Assume the data length is no more than 64k.  */
  41   unsigned short length;
  42   /* The half hash hack.  This contains the upper 16 bits of the hash
  43      value and is used as a pre-check when comparing two strings and
  44      avoids the need to do length or memcmp calls.  It proves to be
  45      roughly 100% effective.  */
  46   unsigned short half_hash;
  47
  48   union
  49   {
  50     char data[1];
  51     double dummy;
  52   }
  53   d;
  54 };
  55
  56 \f
  57 /* Growing the bcache's hash table.  */
  58
  59 /* If the average chain length grows beyond this, then we want to
  60    resize our hash table.  */
  61 #define CHAIN_LENGTH_THRESHOLD (5)
  62
  63 void
  64 bcache::expand_hash_table ()
  65 {
  66   /* A table of good hash table sizes.  Whenever we grow, we pick the
  67      next larger size from this table.  sizes[i] is close to 1 << (i+10),
  68      so we roughly double the table size each time.  After we fall off
  69      the end of this table, we just double.  Don't laugh --- there have
  70      been executables sighted with a gigabyte of debug info.  */
  71   static const unsigned long sizes[] = {
  72     1021, 2053, 4099, 8191, 16381, 32771,
  73     65537, 131071, 262144, 524287, 1048573, 2097143,
  74     4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
  75     268435459, 536870923, 1073741827, 2147483659UL
  76   };
  77   unsigned int new_num_buckets;
  78   struct bstring **new_buckets;
  79   unsigned int i;
  80
  81   /* Count the stats.  Every unique item needs to be re-hashed and
  82      re-entered.  */
  83   m_expand_count++;
  84   m_expand_hash_count += m_unique_count;
  85
  86   /* Find the next size.  */
  87   new_num_buckets = m_num_buckets * 2;
  88   for (unsigned long a_size : sizes)
  89     if (a_size > m_num_buckets)
  90       {
  91         new_num_buckets = a_size;
  92         break;
  93       }
  94
  95   /* Allocate the new table.  */
  96   {
  97     size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
  98
  99     new_buckets = (struct bstring **) xmalloc (new_size);
 100     memset (new_buckets, 0, new_size);
 101
 102     m_structure_size -= m_num_buckets * sizeof (m_bucket[0]);
 103     m_structure_size += new_size;
 104   }
 105
 106   /* Rehash all existing strings.  */
 107   for (i = 0; i < m_num_buckets; i++)
 108     {
 109       struct bstring *s, *next;
 110
 111       for (s = m_bucket[i]; s; s = next)
 112         {
 113           struct bstring **new_bucket;
 114           next = s->next;
 115
 116           new_bucket = &new_buckets[(this->hash (&s->d.data, s->length)
 117                                      % new_num_buckets)];
 118           s->next = *new_bucket;
 119           *new_bucket = s;
 120         }
 121     }
 122
 123   /* Plug in the new table.  */
 124   xfree (m_bucket);
 125   m_bucket = new_buckets;
 126   m_num_buckets = new_num_buckets;
 127 }
 128
 129 \f
 130 /* Looking up things in the bcache.  */
 131
 132 /* The number of bytes needed to allocate a struct bstring whose data
 133    is N bytes long.  */
 134 #define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
 135
 136 /* Find a copy of the LENGTH bytes at ADDR in BCACHE.  If BCACHE has
 137    never seen those bytes before, add a copy of them to BCACHE.  In
 138    either case, return a pointer to BCACHE's copy of that string.  If
 139    optional ADDED is not NULL, return 1 in case of new entry or 0 if
 140    returning an old entry.  */
 141
 142 const void *
 143 bcache::insert (const void *addr, int length, bool *added)
 144 {
 145   unsigned long full_hash;
 146   unsigned short half_hash;
 147   int hash_index;
 148   struct bstring *s;
 149
 150   if (added != nullptr)
 151     *added = false;
 152
 153   /* Lazily initialize the obstack.  This can save quite a bit of
 154      memory in some cases.  */
 155   if (m_total_count == 0)
 156     {
 157       /* We could use obstack_specify_allocation here instead, but
 158          gdb_obstack.h specifies the allocation/deallocation
 159          functions.  */
 160       obstack_init (&m_cache);
 161     }
 162
 163   /* If our average chain length is too high, expand the hash table.  */
 164   if (m_unique_count >= m_num_buckets * CHAIN_LENGTH_THRESHOLD)
 165     expand_hash_table ();
 166
 167   m_total_count++;
 168   m_total_size += length;
 169
 170   full_hash = this->hash (addr, length);
 171
 172   half_hash = (full_hash >> 16);
 173   hash_index = full_hash % m_num_buckets;
 174
 175   /* Search the hash m_bucket for a string identical to the caller's.
 176      As a short-circuit first compare the upper part of each hash
 177      values.  */
 178   for (s = m_bucket[hash_index]; s; s = s->next)
 179     {
 180       if (s->half_hash == half_hash)
 181         {
 182           if (s->length == length
 183               && this->compare (&s->d.data, addr, length))
 184             return &s->d.data;
 185           else
 186             m_half_hash_miss_count++;
 187         }
 188     }
 189
 190   /* The user's string isn't in the list.  Insert it after *ps.  */
 191   {
 192     struct bstring *newobj
 193       = (struct bstring *) obstack_alloc (&m_cache,
 194                                           BSTRING_SIZE (length));
 195
 196     memcpy (&newobj->d.data, addr, length);
 197     newobj->length = length;
 198     newobj->next = m_bucket[hash_index];
 199     newobj->half_hash = half_hash;
 200     m_bucket[hash_index] = newobj;
 201
 202     m_unique_count++;
 203     m_unique_size += length;
 204     m_structure_size += BSTRING_SIZE (length);
 205
 206     if (added != nullptr)
 207       *added = true;
 208
 209     return &newobj->d.data;
 210   }
 211 }
 212 \f
 213
 214 /* See bcache.h.  */
 215
 216 unsigned long
 217 bcache::hash (const void *addr, int length)
 218 {
 219   return fast_hash (addr, length, 0);
 220 }
 221
 222 /* See bcache.h.  */
 223
 224 int
 225 bcache::compare (const void *left, const void *right, int length)
 226 {
 227   return memcmp (left, right, length) == 0;
 228 }
 229
 230 /* Free all the storage associated with BCACHE.  */
 231 bcache::~bcache ()
 232 {
 233   /* Only free the obstack if we actually initialized it.  */
 234   if (m_total_count > 0)
 235     obstack_free (&m_cache, 0);
 236   xfree (m_bucket);
 237 }
 238
 239
 240 \f
 241 /* Printing statistics.  */
 242
 243 static void
 244 print_percentage (int portion, int total)
 245 {
 246   if (total == 0)
 247     /* i18n: Like "Percentage of duplicates, by count: (not applicable)".  */
 248     gdb_printf (_("(not applicable)\n"));
 249   else
 250     gdb_printf ("%3d%%\n", (int) (portion * 100.0 / total));
 251 }
 252
 253
 254 /* Print statistics on BCACHE's memory usage and efficacity at
 255    eliminating duplication.  NAME should describe the kind of data
 256    BCACHE holds.  Statistics are printed using `gdb_printf' and
 257    its ilk.  */
 258 void
 259 bcache::print_statistics (const char *type)
 260 {
 261   int occupied_buckets;
 262   int max_chain_length;
 263   int median_chain_length;
 264   int max_entry_size;
 265   int median_entry_size;
 266
 267   /* Count the number of occupied buckets, tally the various string
 268      lengths, and measure chain lengths.  */
 269   {
 270     unsigned int b;
 271     int *chain_length = XCNEWVEC (int, m_num_buckets + 1);
 272     int *entry_size = XCNEWVEC (int, m_unique_count + 1);
 273     int stringi = 0;
 274
 275     occupied_buckets = 0;
 276
 277     for (b = 0; b < m_num_buckets; b++)
 278       {
 279         struct bstring *s = m_bucket[b];
 280
 281         chain_length[b] = 0;
 282
 283         if (s)
 284           {
 285             occupied_buckets++;
 286
 287             while (s)
 288               {
 289                 gdb_assert (b < m_num_buckets);
 290                 chain_length[b]++;
 291                 gdb_assert (stringi < m_unique_count);
 292                 entry_size[stringi++] = s->length;
 293                 s = s->next;
 294               }
 295           }
 296       }
 297
 298     /* To compute the median, we need the set of chain lengths
 299        sorted.  */
 300     std::sort (chain_length, chain_length + m_num_buckets);
 301     std::sort (entry_size, entry_size + m_unique_count);
 302
 303     if (m_num_buckets > 0)
 304       {
 305         max_chain_length = chain_length[m_num_buckets - 1];
 306         median_chain_length = chain_length[m_num_buckets / 2];
 307       }
 308     else
 309       {
 310         max_chain_length = 0;
 311         median_chain_length = 0;
 312       }
 313     if (m_unique_count > 0)
 314       {
 315         max_entry_size = entry_size[m_unique_count - 1];
 316         median_entry_size = entry_size[m_unique_count / 2];
 317       }
 318     else
 319       {
 320         max_entry_size = 0;
 321         median_entry_size = 0;
 322       }
 323
 324     xfree (chain_length);
 325     xfree (entry_size);
 326   }
 327
 328   gdb_printf (_("  M_Cached '%s' statistics:\n"), type);
 329   gdb_printf (_("    Total object count:  %ld\n"), m_total_count);
 330   gdb_printf (_("    Unique object count: %lu\n"), m_unique_count);
 331   gdb_printf (_("    Percentage of duplicates, by count: "));
 332   print_percentage (m_total_count - m_unique_count, m_total_count);
 333   gdb_printf ("\n");
 334
 335   gdb_printf (_("    Total object size:   %ld\n"), m_total_size);
 336   gdb_printf (_("    Unique object size:  %ld\n"), m_unique_size);
 337   gdb_printf (_("    Percentage of duplicates, by size:  "));
 338   print_percentage (m_total_size - m_unique_size, m_total_size);
 339   gdb_printf ("\n");
 340
 341   gdb_printf (_("    Max entry size:     %d\n"), max_entry_size);
 342   gdb_printf (_("    Average entry size: "));
 343   if (m_unique_count > 0)
 344     gdb_printf ("%ld\n", m_unique_size / m_unique_count);
 345   else
 346     /* i18n: "Average entry size: (not applicable)".  */
 347     gdb_printf (_("(not applicable)\n"));
 348   gdb_printf (_("    Median entry size:  %d\n"), median_entry_size);
 349   gdb_printf ("\n");
 350
 351   gdb_printf (_("    \
 352 Total memory used by bcache, including overhead: %ld\n"),
 353               m_structure_size);
 354   gdb_printf (_("    Percentage memory overhead: "));
 355   print_percentage (m_structure_size - m_unique_size, m_unique_size);
 356   gdb_printf (_("    Net memory savings:         "));
 357   print_percentage (m_total_size - m_structure_size, m_total_size);
 358   gdb_printf ("\n");
 359
 360   gdb_printf (_("    Hash table size:           %3d\n"),
 361               m_num_buckets);
 362   gdb_printf (_("    Hash table expands:        %lu\n"),
 363               m_expand_count);
 364   gdb_printf (_("    Hash table hashes:         %lu\n"),
 365               m_total_count + m_expand_hash_count);
 366   gdb_printf (_("    Half hash misses:          %lu\n"),
 367               m_half_hash_miss_count);
 368   gdb_printf (_("    Hash table population:     "));
 369   print_percentage (occupied_buckets, m_num_buckets);
 370   gdb_printf (_("    Median hash chain length:  %3d\n"),
 371               median_chain_length);
 372   gdb_printf (_("    Average hash chain length: "));
 373   if (m_num_buckets > 0)
 374     gdb_printf ("%3lu\n", m_unique_count / m_num_buckets);
 375   else
 376     /* i18n: "Average hash chain length: (not applicable)".  */
 377     gdb_printf (_("(not applicable)\n"));
 378   gdb_printf (_("    Maximum hash chain length: %3d\n"),
 379               max_chain_length);
 380   gdb_printf ("\n");
 381 }
 382
 383 int
 384 bcache::memory_used ()
 385 {
 386   if (m_total_count == 0)
 387     return 0;
 388   return obstack_memory_used (&m_cache);
 389 }
 390
 391 } /* namespace gdb */