Sweden-Number/dlls/dbghelp/storage.c

420 lines
12 KiB
C
Raw Normal View History

2004-04-06 00:21:27 +02:00
/*
* Various storage structures (pool allocation, vector, hash table)
*
* Copyright (C) 1993, Eric Youngdale.
* 2004, Eric Pouech
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
2004-04-06 00:21:27 +02:00
*/
#include "config.h"
#include <assert.h>
#include <stdlib.h>
#include "wine/debug.h"
#include "dbghelp_private.h"
#ifdef USE_STATS
#include <math.h>
#endif
WINE_DEFAULT_DEBUG_CHANNEL(dbghelp);
struct pool_arena
{
struct pool_arena* next;
char* current;
};
void pool_init(struct pool* a, unsigned arena_size)
{
a->arena_size = arena_size;
a->first = NULL;
}
void pool_destroy(struct pool* pool)
{
struct pool_arena* arena;
struct pool_arena* next;
#ifdef USE_STATS
unsigned alloc, used, num;
2007-05-15 02:05:21 +02:00
alloc = used = num = 0;
arena = pool->first;
while (arena)
2004-04-06 00:21:27 +02:00
{
alloc += pool->arena_size;
used += arena->current - (char*)arena;
num++;
2007-05-15 02:05:21 +02:00
arena = arena->next;
2004-04-06 00:21:27 +02:00
}
2007-05-15 02:05:21 +02:00
if (alloc == 0) alloc = 1; /* avoid division by zero */
2004-04-06 00:21:27 +02:00
FIXME("STATS: pool %p has allocated %u kbytes, used %u kbytes in %u arenas,\n"
"\t\t\t\tnon-allocation ratio: %.2f%%\n",
pool, alloc >> 10, used >> 10, num, 100.0 - (float)used / (float)alloc * 100.0);
#endif
2007-05-15 02:05:21 +02:00
arena = pool->first;
while (arena)
2004-04-06 00:21:27 +02:00
{
next = arena->next;
HeapFree(GetProcessHeap(), 0, arena);
2007-05-15 02:05:21 +02:00
arena = next;
2004-04-06 00:21:27 +02:00
}
pool_init(pool, 0);
}
void* pool_alloc(struct pool* pool, unsigned len)
{
struct pool_arena* arena;
void* ret;
len = (len + 3) & ~3; /* round up size on DWORD boundary */
assert(sizeof(struct pool_arena) + len <= pool->arena_size && len);
for (arena = pool->first; arena; arena = arena->next)
2004-04-06 00:21:27 +02:00
{
if ((char*)arena + pool->arena_size - arena->current >= len)
2004-04-06 00:21:27 +02:00
{
ret = arena->current;
arena->current += len;
2004-04-06 00:21:27 +02:00
return ret;
}
}
2004-04-06 00:21:27 +02:00
arena = HeapAlloc(GetProcessHeap(), 0, pool->arena_size);
2008-11-23 22:07:36 +01:00
if (!arena) {ERR("OOM\n");return NULL;}
2004-04-06 00:21:27 +02:00
ret = (char*)arena + sizeof(*arena);
arena->next = pool->first;
pool->first = arena;
2004-04-06 00:21:27 +02:00
arena->current = (char*)ret + len;
return ret;
}
char* pool_strdup(struct pool* pool, const char* str)
{
char* ret;
if ((ret = pool_alloc(pool, strlen(str) + 1))) strcpy(ret, str);
return ret;
}
void vector_init(struct vector* v, unsigned esz, unsigned bucket_sz)
{
v->buckets = NULL;
/* align size on DWORD boundaries */
v->elt_size = (esz + 3) & ~3;
switch (bucket_sz)
{
case 2: v->shift = 1; break;
case 4: v->shift = 2; break;
case 8: v->shift = 3; break;
case 16: v->shift = 4; break;
case 32: v->shift = 5; break;
case 64: v->shift = 6; break;
case 128: v->shift = 7; break;
case 256: v->shift = 8; break;
case 512: v->shift = 9; break;
case 1024: v->shift = 10; break;
default: assert(0);
}
v->num_buckets = 0;
v->buckets_allocated = 0;
2004-04-06 00:21:27 +02:00
v->num_elts = 0;
}
unsigned vector_length(const struct vector* v)
{
return v->num_elts;
}
void* vector_at(const struct vector* v, unsigned pos)
{
unsigned o;
if (pos >= v->num_elts) return NULL;
o = pos & ((1 << v->shift) - 1);
return (char*)v->buckets[pos >> v->shift] + o * v->elt_size;
}
void* vector_add(struct vector* v, struct pool* pool)
{
unsigned ncurr = v->num_elts++;
/* check that we don't wrap around */
assert(v->num_elts > ncurr);
if (ncurr == (v->num_buckets << v->shift))
{
if(v->num_buckets == v->buckets_allocated)
{
/* Double the bucket cache, so it scales well with big vectors.*/
unsigned new_reserved;
void* new;
new_reserved = 2*v->buckets_allocated;
if(new_reserved == 0) new_reserved = 1;
/* Don't even try to resize memory.
Pool datastructure is very inefficient with reallocs. */
new = pool_alloc(pool, new_reserved * sizeof(void*));
memcpy(new, v->buckets, v->buckets_allocated * sizeof(void*));
v->buckets = new;
v->buckets_allocated = new_reserved;
}
2004-04-06 00:21:27 +02:00
v->buckets[v->num_buckets] = pool_alloc(pool, v->elt_size << v->shift);
return v->buckets[v->num_buckets++];
}
return vector_at(v, ncurr);
}
/* We construct the sparse array as two vectors (of equal size)
* The first vector (key2index) is the lookup table between the key and
* an index in the second vector (elements)
* When inserting an element, it's always appended in second vector (and
* never moved in memory later on), only the first vector is reordered
*/
struct key2index
{
unsigned long key;
unsigned index;
};
void sparse_array_init(struct sparse_array* sa, unsigned elt_sz, unsigned bucket_sz)
{
vector_init(&sa->key2index, sizeof(struct key2index), bucket_sz);
vector_init(&sa->elements, elt_sz, bucket_sz);
}
/******************************************************************
2006-12-09 14:30:56 +01:00
* sparse_array_lookup
*
* Returns the first index which key is >= at passed key
*/
2006-12-09 14:30:56 +01:00
static struct key2index* sparse_array_lookup(const struct sparse_array* sa,
unsigned long key, unsigned* idx)
{
struct key2index* pk2i;
2006-12-09 14:30:56 +01:00
unsigned low, high;
2006-12-09 14:30:56 +01:00
if (!sa->elements.num_elts)
{
2006-12-09 14:30:56 +01:00
*idx = 0;
return NULL;
}
high = sa->elements.num_elts;
pk2i = vector_at(&sa->key2index, high - 1);
if (pk2i->key < key)
{
*idx = high;
return NULL;
}
if (pk2i->key == key)
{
*idx = high - 1;
return pk2i;
}
low = 0;
pk2i = vector_at(&sa->key2index, low);
if (pk2i->key >= key)
{
*idx = 0;
return pk2i;
}
/* now we have: sa(lowest key) < key < sa(highest key) */
while (low < high)
{
*idx = (low + high) / 2;
pk2i = vector_at(&sa->key2index, *idx);
2006-12-09 14:30:56 +01:00
if (pk2i->key > key) high = *idx;
else if (pk2i->key < key) low = *idx + 1;
else return pk2i;
}
2006-12-09 14:30:56 +01:00
/* binary search could return exact item, we search for highest one
* below the key
*/
if (pk2i->key < key)
pk2i = vector_at(&sa->key2index, ++(*idx));
return pk2i;
}
void* sparse_array_find(const struct sparse_array* sa, unsigned long key)
{
unsigned idx;
struct key2index* pk2i;
2006-12-09 14:30:56 +01:00
if ((pk2i = sparse_array_lookup(sa, key, &idx)) && pk2i->key == key)
return vector_at(&sa->elements, pk2i->index);
return NULL;
}
void* sparse_array_add(struct sparse_array* sa, unsigned long key,
struct pool* pool)
{
unsigned idx, i;
struct key2index* pk2i;
struct key2index* to;
2006-12-09 14:30:56 +01:00
pk2i = sparse_array_lookup(sa, key, &idx);
if (pk2i && pk2i->key == key)
{
FIXME("re adding an existing key\n");
return NULL;
}
to = vector_add(&sa->key2index, pool);
if (pk2i)
{
/* we need to shift vector's content... */
/* let's do it brute force... (FIXME) */
assert(sa->key2index.num_elts >= 2);
for (i = sa->key2index.num_elts - 1; i > idx; i--)
{
pk2i = vector_at(&sa->key2index, i - 1);
*to = *pk2i;
to = pk2i;
}
}
to->key = key;
to->index = sa->elements.num_elts;
return vector_add(&sa->elements, pool);
}
unsigned sparse_array_length(const struct sparse_array* sa)
{
return sa->elements.num_elts;
}
2004-04-06 00:21:27 +02:00
unsigned hash_table_hash(const char* name, unsigned num_buckets)
{
unsigned hash = 0;
while (*name)
{
hash += *name++;
hash += (hash << 10);
hash ^= (hash >> 6);
}
hash += (hash << 3);
hash ^= (hash >> 11);
hash += (hash << 15);
return hash % num_buckets;
}
void hash_table_init(struct pool* pool, struct hash_table* ht, unsigned num_buckets)
{
ht->num_elts = 0;
2004-04-06 00:21:27 +02:00
ht->num_buckets = num_buckets;
2007-05-15 02:05:21 +02:00
ht->pool = pool;
ht->buckets = NULL;
2004-04-06 00:21:27 +02:00
}
void hash_table_destroy(struct hash_table* ht)
{
#if defined(USE_STATS)
int i;
unsigned len;
unsigned min = 0xffffffff, max = 0, sq = 0;
2004-04-06 00:21:27 +02:00
struct hash_table_elt* elt;
double mean, variance;
for (i = 0; i < ht->num_buckets; i++)
{
for (len = 0, elt = ht->buckets[i]; elt; elt = elt->next) len++;
if (len < min) min = len;
if (len > max) max = len;
sq += len * len;
}
mean = (double)ht->num_elts / ht->num_buckets;
2004-04-06 00:21:27 +02:00
variance = (double)sq / ht->num_buckets - mean * mean;
FIXME("STATS: elts[num:%-4u size:%u mean:%f] buckets[min:%-4u variance:%+f max:%-4u]\n",
ht->num_elts, ht->num_buckets, mean, min, variance, max);
2004-04-06 00:21:27 +02:00
#if 1
for (i = 0; i < ht->num_buckets; i++)
{
for (len = 0, elt = ht->buckets[i]; elt; elt = elt->next) len++;
if (len == max)
{
FIXME("Longuest bucket:\n");
for (elt = ht->buckets[i]; elt; elt = elt->next)
FIXME("\t%s\n", elt->name);
break;
}
}
#endif
#endif
}
void hash_table_add(struct hash_table* ht, struct hash_table_elt* elt)
{
unsigned hash = hash_table_hash(elt->name, ht->num_buckets);
struct hash_table_elt** p;
2007-05-15 02:05:21 +02:00
if (!ht->buckets)
{
ht->buckets = pool_alloc(ht->pool, ht->num_buckets * sizeof(struct hash_table_elt*));
assert(ht->buckets);
memset(ht->buckets, 0, ht->num_buckets * sizeof(struct hash_table_elt*));
}
/* in some cases, we need to get back the symbols of same name in the order
* in which they've been inserted. So insert new elements at the end of the list.
*/
for (p = &ht->buckets[hash]; *p; p = &((*p)->next));
*p = elt;
elt->next = NULL;
ht->num_elts++;
2004-04-06 00:21:27 +02:00
}
void* hash_table_find(const struct hash_table* ht, const char* name)
{
unsigned hash = hash_table_hash(name, ht->num_buckets);
struct hash_table_elt* elt;
2007-05-15 02:05:21 +02:00
if(!ht->buckets) return NULL;
2004-04-06 00:21:27 +02:00
for (elt = ht->buckets[hash]; elt; elt = elt->next)
if (!strcmp(name, elt->name)) return elt;
return NULL;
}
void hash_table_iter_init(const struct hash_table* ht,
struct hash_table_iter* hti, const char* name)
{
hti->ht = ht;
if (name)
{
hti->last = hash_table_hash(name, ht->num_buckets);
hti->index = hti->last - 1;
}
else
{
hti->last = ht->num_buckets - 1;
hti->index = -1;
}
hti->element = NULL;
}
void* hash_table_iter_up(struct hash_table_iter* hti)
{
2007-05-15 02:05:21 +02:00
if(!hti->ht->buckets) return NULL;
2004-04-06 00:21:27 +02:00
if (hti->element) hti->element = hti->element->next;
while (!hti->element && hti->index < hti->last)
hti->element = hti->ht->buckets[++hti->index];
return hti->element;
}