/* Implement a cached obstack.
- Written by Fred Fish (fnf@cygnus.com)
- Copyright 1995, 1998 Free Software Foundation, Inc.
+ Written by Fred Fish <fnf@cygnus.com>
+ Rewritten by Jim Blandy <jimb@cygnus.com>
-This file is part of GDB.
+ Copyright (C) 1999-2020 Free Software Foundation, Inc.
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2 of the License, or
-(at your option) any later version.
+ This file is part of GDB.
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-GNU General Public License for more details.
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
-You should have received a copy of the GNU General Public License
-along with this program; if not, write to the Free Software
-Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
#include "defs.h"
-#include "obstack.h"
+#include "gdb_obstack.h"
#include "bcache.h"
-#include "gdb_string.h" /* For memcpy declaration */
-
-/* Prototypes for local functions. */
-static unsigned int hash PARAMS ((void *, int));
+#include <algorithm>
-static void *lookup_cache PARAMS ((void *, int, int, struct bcache *));
+namespace gdb {
-/* FIXME: Incredibly simplistic hash generator. Probably way too expensive
- (consider long strings) and unlikely to have good distribution across hash
- values for typical input. */
+/* The type used to hold a single bcache string. The user data is
+ stored in d.data. Since it can be any type, it needs to have the
+ same alignment as the most strict alignment of any type on the host
+ machine. I don't know of any really correct way to do this in
+ stock ANSI C, so just do it the same way obstack.h does. */
-static unsigned int
-hash (bytes, count)
- void *bytes;
- int count;
+struct bstring
{
- unsigned int len;
- unsigned long hashval;
- unsigned int c;
- const unsigned char *data = bytes;
-
- hashval = 0;
- len = 0;
- while (count-- > 0)
- {
- c = *data++;
- hashval += c + (c << 17);
- hashval ^= hashval >> 2;
- ++len;
- }
- hashval += len + (len << 17);
- hashval ^= hashval >> 2;
- return (hashval % BCACHE_HASHSIZE);
-}
+ /* Hash chain. */
+ struct bstring *next;
+ /* Assume the data length is no more than 64k. */
+ unsigned short length;
+ /* The half hash hack. This contains the upper 16 bits of the hash
+ value and is used as a pre-check when comparing two strings and
+ avoids the need to do length or memcmp calls. It proves to be
+ roughly 100% effective. */
+ unsigned short half_hash;
+
+ union
+ {
+ char data[1];
+ double dummy;
+ }
+ d;
+};
-static void *
-lookup_cache (bytes, count, hashval, bcachep)
- void *bytes;
- int count;
- int hashval;
- struct bcache *bcachep;
+\f
+/* Growing the bcache's hash table. */
+
+/* If the average chain length grows beyond this, then we want to
+ resize our hash table. */
+#define CHAIN_LENGTH_THRESHOLD (5)
+
+void
+bcache::expand_hash_table ()
{
- void *location = NULL;
- struct hashlink **hashtablep;
- struct hashlink *linkp;
+ /* A table of good hash table sizes. Whenever we grow, we pick the
+ next larger size from this table. sizes[i] is close to 1 << (i+10),
+ so we roughly double the table size each time. After we fall off
+ the end of this table, we just double. Don't laugh --- there have
+ been executables sighted with a gigabyte of debug info. */
+ static unsigned long sizes[] = {
+ 1021, 2053, 4099, 8191, 16381, 32771,
+ 65537, 131071, 262144, 524287, 1048573, 2097143,
+ 4194301, 8388617, 16777213, 33554467, 67108859, 134217757,
+ 268435459, 536870923, 1073741827, 2147483659UL
+ };
+ unsigned int new_num_buckets;
+ struct bstring **new_buckets;
+ unsigned int i;
+
+ /* Count the stats. Every unique item needs to be re-hashed and
+ re-entered. */
+ m_expand_count++;
+ m_expand_hash_count += m_unique_count;
+
+ /* Find the next size. */
+ new_num_buckets = m_num_buckets * 2;
+ for (i = 0; i < (sizeof (sizes) / sizeof (sizes[0])); i++)
+ if (sizes[i] > m_num_buckets)
+ {
+ new_num_buckets = sizes[i];
+ break;
+ }
- hashtablep = bcachep -> indextable[count];
- if (hashtablep != NULL)
+ /* Allocate the new table. */
+ {
+ size_t new_size = new_num_buckets * sizeof (new_buckets[0]);
+
+ new_buckets = (struct bstring **) xmalloc (new_size);
+ memset (new_buckets, 0, new_size);
+
+ m_structure_size -= m_num_buckets * sizeof (m_bucket[0]);
+ m_structure_size += new_size;
+ }
+
+ /* Rehash all existing strings. */
+ for (i = 0; i < m_num_buckets; i++)
{
- linkp = hashtablep[hashval];
- while (linkp != NULL)
+ struct bstring *s, *next;
+
+ for (s = m_bucket[i]; s; s = next)
{
- if (memcmp (BCACHE_DATA (linkp), bytes, count) == 0)
- {
- location = BCACHE_DATA (linkp);
- break;
- }
- linkp = linkp -> next;
+ struct bstring **new_bucket;
+ next = s->next;
+
+ new_bucket = &new_buckets[(m_hash_function (&s->d.data, s->length)
+ % new_num_buckets)];
+ s->next = *new_bucket;
+ *new_bucket = s;
}
}
- return (location);
+
+ /* Plug in the new table. */
+ xfree (m_bucket);
+ m_bucket = new_buckets;
+ m_num_buckets = new_num_buckets;
}
-void *
-bcache (bytes, count, bcachep)
- void *bytes;
- int count;
- struct bcache *bcachep;
+\f
+/* Looking up things in the bcache. */
+
+/* The number of bytes needed to allocate a struct bstring whose data
+ is N bytes long. */
+#define BSTRING_SIZE(n) (offsetof (struct bstring, d.data) + (n))
+
+/* Find a copy of the LENGTH bytes at ADDR in BCACHE. If BCACHE has
+ never seen those bytes before, add a copy of them to BCACHE. In
+ either case, return a pointer to BCACHE's copy of that string. If
+ optional ADDED is not NULL, return 1 in case of new entry or 0 if
+ returning an old entry. */
+
+const void *
+bcache::insert (const void *addr, int length, int *added)
{
- int hashval;
- void *location;
- struct hashlink *newlink;
- struct hashlink **linkpp;
- struct hashlink ***hashtablepp;
+ unsigned long full_hash;
+ unsigned short half_hash;
+ int hash_index;
+ struct bstring *s;
+
+ if (added)
+ *added = 0;
- if (count >= BCACHE_MAXLENGTH)
+ /* Lazily initialize the obstack. This can save quite a bit of
+ memory in some cases. */
+ if (m_total_count == 0)
{
- /* Rare enough to just stash unique copies */
- location = (void *) obstack_alloc (&bcachep->cache, count);
- bcachep -> cache_bytes += count;
- memcpy (location, bytes, count);
- bcachep -> bcache_overflows++;
+ /* We could use obstack_specify_allocation here instead, but
+ gdb_obstack.h specifies the allocation/deallocation
+ functions. */
+ obstack_init (&m_cache);
}
- else
+
+ /* If our average chain length is too high, expand the hash table. */
+ if (m_unique_count >= m_num_buckets * CHAIN_LENGTH_THRESHOLD)
+ expand_hash_table ();
+
+ m_total_count++;
+ m_total_size += length;
+
+ full_hash = m_hash_function (addr, length);
+
+ half_hash = (full_hash >> 16);
+ hash_index = full_hash % m_num_buckets;
+
+ /* Search the hash m_bucket for a string identical to the caller's.
+ As a short-circuit first compare the upper part of each hash
+ values. */
+ for (s = m_bucket[hash_index]; s; s = s->next)
{
- hashval = hash (bytes, count);
- location = lookup_cache (bytes, count, hashval, bcachep);
- if (location != NULL)
- {
- bcachep -> cache_savings += count;
- bcachep -> cache_hits++;
- }
- else
+ if (s->half_hash == half_hash)
{
- bcachep -> cache_misses++;
- hashtablepp = &bcachep -> indextable[count];
- if (*hashtablepp == NULL)
- {
- *hashtablepp = (struct hashlink **)
- obstack_alloc (&bcachep->cache, BCACHE_HASHSIZE * sizeof (struct hashlink *));
- bcachep -> cache_bytes += BCACHE_HASHSIZE * sizeof (struct hashlink *);
- memset (*hashtablepp, 0, BCACHE_HASHSIZE * sizeof (struct hashlink *));
- }
- linkpp = &(*hashtablepp)[hashval];
- newlink = (struct hashlink *)
- obstack_alloc (&bcachep->cache, BCACHE_DATA_ALIGNMENT + count);
- bcachep -> cache_bytes += BCACHE_DATA_ALIGNMENT + count;
- memcpy (BCACHE_DATA (newlink), bytes, count);
- newlink -> next = *linkpp;
- *linkpp = newlink;
- location = BCACHE_DATA (newlink);
+ if (s->length == length
+ && m_compare_function (&s->d.data, addr, length))
+ return &s->d.data;
+ else
+ m_half_hash_miss_count++;
}
}
- return (location);
+
+ /* The user's string isn't in the list. Insert it after *ps. */
+ {
+ struct bstring *newobj
+ = (struct bstring *) obstack_alloc (&m_cache,
+ BSTRING_SIZE (length));
+
+ memcpy (&newobj->d.data, addr, length);
+ newobj->length = length;
+ newobj->next = m_bucket[hash_index];
+ newobj->half_hash = half_hash;
+ m_bucket[hash_index] = newobj;
+
+ m_unique_count++;
+ m_unique_size += length;
+ m_structure_size += BSTRING_SIZE (length);
+
+ if (added)
+ *added = 1;
+
+ return &newobj->d.data;
+ }
}
+\f
-void
-print_bcache_statistics (bcachep, id)
- struct bcache *bcachep;
- char *id;
+/* Compare the byte string at ADDR1 of lenght LENGHT to the
+ string at ADDR2. Return 1 if they are equal. */
+
+int
+bcache::compare (const void *addr1, const void *addr2, int length)
{
- struct hashlink **hashtablep;
- struct hashlink *linkp;
- int tidx, tcount, hidx, hcount, lcount, lmax, temp, lmaxt, lmaxh;
+ return memcmp (addr1, addr2, length) == 0;
+}
- for (lmax = lcount = tcount = hcount = tidx = 0; tidx < BCACHE_MAXLENGTH; tidx++)
- {
- hashtablep = bcachep -> indextable[tidx];
- if (hashtablep != NULL)
- {
- tcount++;
- for (hidx = 0; hidx < BCACHE_HASHSIZE; hidx++)
- {
- linkp = hashtablep[hidx];
- if (linkp != NULL)
- {
- hcount++;
- for (temp = 0; linkp != NULL; linkp = linkp -> next)
- {
- lcount++;
- temp++;
- }
- if (temp > lmax)
- {
- lmax = temp;
- lmaxt = tidx;
- lmaxh = hidx;
- }
- }
- }
- }
- }
- printf_filtered (" Cached '%s' statistics:\n", id);
- printf_filtered (" Cache hits: %d\n", bcachep -> cache_hits);
- printf_filtered (" Cache misses: %d\n", bcachep -> cache_misses);
- printf_filtered (" Cache hit ratio: ");
- if (bcachep -> cache_hits + bcachep -> cache_misses > 0)
- {
- printf_filtered ("%d%%\n", ((bcachep -> cache_hits) * 100) /
- (bcachep -> cache_hits + bcachep -> cache_misses));
- }
+/* Free all the storage associated with BCACHE. */
+bcache::~bcache ()
+{
+ /* Only free the obstack if we actually initialized it. */
+ if (m_total_count > 0)
+ obstack_free (&m_cache, 0);
+ xfree (m_bucket);
+}
+
+
+\f
+/* Printing statistics. */
+
+static void
+print_percentage (int portion, int total)
+{
+ if (total == 0)
+ /* i18n: Like "Percentage of duplicates, by count: (not applicable)". */
+ printf_filtered (_("(not applicable)\n"));
else
- {
- printf_filtered ("(not applicable)\n");
- }
- printf_filtered (" Space used for caching: %d\n", bcachep -> cache_bytes);
- printf_filtered (" Space saved by cache hits: %d\n", bcachep -> cache_savings);
- printf_filtered (" Number of bcache overflows: %d\n", bcachep -> bcache_overflows);
- printf_filtered (" Number of index buckets used: %d\n", tcount);
- printf_filtered (" Number of hash table buckets used: %d\n", hcount);
- printf_filtered (" Number of chained items: %d\n", lcount);
- printf_filtered (" Average hash table population: ");
- if (tcount > 0)
- {
- printf_filtered ("%d%%\n", (hcount * 100) / (tcount * BCACHE_HASHSIZE));
- }
+ printf_filtered ("%3d%%\n", (int) (portion * 100.0 / total));
+}
+
+
+/* Print statistics on BCACHE's memory usage and efficacity at
+ eliminating duplication. NAME should describe the kind of data
+ BCACHE holds. Statistics are printed using `printf_filtered' and
+ its ilk. */
+void
+bcache::print_statistics (const char *type)
+{
+ int occupied_buckets;
+ int max_chain_length;
+ int median_chain_length;
+ int max_entry_size;
+ int median_entry_size;
+
+ /* Count the number of occupied buckets, tally the various string
+ lengths, and measure chain lengths. */
+ {
+ unsigned int b;
+ int *chain_length = XCNEWVEC (int, m_num_buckets + 1);
+ int *entry_size = XCNEWVEC (int, m_unique_count + 1);
+ int stringi = 0;
+
+ occupied_buckets = 0;
+
+ for (b = 0; b < m_num_buckets; b++)
+ {
+ struct bstring *s = m_bucket[b];
+
+ chain_length[b] = 0;
+
+ if (s)
+ {
+ occupied_buckets++;
+
+ while (s)
+ {
+ gdb_assert (b < m_num_buckets);
+ chain_length[b]++;
+ gdb_assert (stringi < m_unique_count);
+ entry_size[stringi++] = s->length;
+ s = s->next;
+ }
+ }
+ }
+
+ /* To compute the median, we need the set of chain lengths
+ sorted. */
+ std::sort (chain_length, chain_length + m_num_buckets);
+ std::sort (entry_size, entry_size + m_unique_count);
+
+ if (m_num_buckets > 0)
+ {
+ max_chain_length = chain_length[m_num_buckets - 1];
+ median_chain_length = chain_length[m_num_buckets / 2];
+ }
+ else
+ {
+ max_chain_length = 0;
+ median_chain_length = 0;
+ }
+ if (m_unique_count > 0)
+ {
+ max_entry_size = entry_size[m_unique_count - 1];
+ median_entry_size = entry_size[m_unique_count / 2];
+ }
+ else
+ {
+ max_entry_size = 0;
+ median_entry_size = 0;
+ }
+
+ xfree (chain_length);
+ xfree (entry_size);
+ }
+
+ printf_filtered (_(" M_Cached '%s' statistics:\n"), type);
+ printf_filtered (_(" Total object count: %ld\n"), m_total_count);
+ printf_filtered (_(" Unique object count: %lu\n"), m_unique_count);
+ printf_filtered (_(" Percentage of duplicates, by count: "));
+ print_percentage (m_total_count - m_unique_count, m_total_count);
+ printf_filtered ("\n");
+
+ printf_filtered (_(" Total object size: %ld\n"), m_total_size);
+ printf_filtered (_(" Unique object size: %ld\n"), m_unique_size);
+ printf_filtered (_(" Percentage of duplicates, by size: "));
+ print_percentage (m_total_size - m_unique_size, m_total_size);
+ printf_filtered ("\n");
+
+ printf_filtered (_(" Max entry size: %d\n"), max_entry_size);
+ printf_filtered (_(" Average entry size: "));
+ if (m_unique_count > 0)
+ printf_filtered ("%ld\n", m_unique_size / m_unique_count);
else
- {
- printf_filtered ("(not applicable)\n");
- }
- printf_filtered (" Average chain length ");
- if (hcount > 0)
- {
- printf_filtered ("%d\n", lcount / hcount);
- }
+ /* i18n: "Average entry size: (not applicable)". */
+ printf_filtered (_("(not applicable)\n"));
+ printf_filtered (_(" Median entry size: %d\n"), median_entry_size);
+ printf_filtered ("\n");
+
+ printf_filtered (_(" \
+Total memory used by bcache, including overhead: %ld\n"),
+ m_structure_size);
+ printf_filtered (_(" Percentage memory overhead: "));
+ print_percentage (m_structure_size - m_unique_size, m_unique_size);
+ printf_filtered (_(" Net memory savings: "));
+ print_percentage (m_total_size - m_structure_size, m_total_size);
+ printf_filtered ("\n");
+
+ printf_filtered (_(" Hash table size: %3d\n"),
+ m_num_buckets);
+ printf_filtered (_(" Hash table expands: %lu\n"),
+ m_expand_count);
+ printf_filtered (_(" Hash table hashes: %lu\n"),
+ m_total_count + m_expand_hash_count);
+ printf_filtered (_(" Half hash misses: %lu\n"),
+ m_half_hash_miss_count);
+ printf_filtered (_(" Hash table population: "));
+ print_percentage (occupied_buckets, m_num_buckets);
+ printf_filtered (_(" Median hash chain length: %3d\n"),
+ median_chain_length);
+ printf_filtered (_(" Average hash chain length: "));
+ if (m_num_buckets > 0)
+ printf_filtered ("%3lu\n", m_unique_count / m_num_buckets);
else
- {
- printf_filtered ("(not applicable)\n");
- }
- printf_filtered (" Maximum chain length %d at %d:%d\n", lmax, lmaxt, lmaxh);
+ /* i18n: "Average hash chain length: (not applicable)". */
+ printf_filtered (_("(not applicable)\n"));
+ printf_filtered (_(" Maximum hash chain length: %3d\n"),
+ max_chain_length);
+ printf_filtered ("\n");
}
+
+int
+bcache::memory_used ()
+{
+ if (m_total_count == 0)
+ return 0;
+ return obstack_memory_used (&m_cache);
+}
+
+} /* namespace gdb */
projects / deliverable / binutils-gdb.git / blobdiff