raze/source/build/src/hash.cpp
hendricks266 1cc9d13ccf The great repository rearrangement of 2017.
Files moved but not modified. Changes to follow in a subsequent commit.

You down with CPP?

git-svn-id: https://svn.eduke32.com/eduke32@6055 1a8010ca-5511-0410-912e-c29ae57300e0
2017-02-01 10:01:11 +00:00

367 lines
7.4 KiB
C++

#include "compat.h"
#include "hash.h"
#include "baselayer.h"
void hash_init(hashtable_t *t)
{
hash_free(t);
t->items=(hashitem_t **) Xcalloc(1, t->size * sizeof(hashitem_t));
}
void hash_loop(hashtable_t *t, void(*func)(const char *, intptr_t))
{
if (t->items == NULL)
return;
for (bssize_t i=0; i < t->size; i++)
for (hashitem_t *item=t->items[i]; item != NULL; item = item->next)
func(item->string, item->key);
}
void hash_free(hashtable_t *t)
{
if (t->items == NULL)
return;
int remaining = t->size - 1;
do
{
hashitem_t *cur = t->items[remaining];
int num = 0;
while (cur)
{
hashitem_t * const tmp = cur;
cur = cur->next;
Bfree(tmp->string);
Bfree(tmp);
num++;
}
} while (--remaining >= 0);
DO_FREE_AND_NULL(t->items);
}
// djb3 algorithm
static inline uint32_t hash_getcode(const char *s)
{
uint32_t h = 5381;
int32_t ch;
while ((ch = *s++) != '\0')
h = ((h << 5) + h) ^ ch;
return h;
}
void hash_add(hashtable_t *t, const char *s, intptr_t key, int32_t replace)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (EDUKE32_PREDICT_FALSE(t->items == NULL))
{
initprintf("hash_add(): table not initialized!\n");
return;
}
#endif
uint32_t code = hash_getcode(s) % t->size;
hashitem_t *cur = t->items[code];
if (!cur)
{
cur = (hashitem_t *) Xcalloc(1, sizeof(hashitem_t));
cur->string = Xstrdup(s);
cur->key = key;
cur->next = NULL;
t->items[code] = cur;
return;
}
hashitem_t *prev = NULL;
do
{
if (Bstrcmp(s, cur->string) == 0)
{
if (replace) cur->key = key;
return;
}
prev = cur;
} while ((cur = cur->next));
cur = (hashitem_t *) Xcalloc(1, sizeof(hashitem_t));
cur->string = Xstrdup(s);
cur->key = key;
cur->next = NULL;
prev->next = cur;
}
// delete at most once
void hash_delete(hashtable_t *t, const char *s)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (t->items == NULL)
{
initprintf("hash_delete(): table not initialized!\n");
return;
}
#endif
uint32_t code = hash_getcode(s) % t->size;
hashitem_t *cur = t->items[code];
if (!cur)
return;
hashitem_t *prev = NULL;
do
{
if (Bstrcmp(s, cur->string) == 0)
{
Bfree(cur->string);
if (!prev)
t->items[code] = cur->next;
else
prev->next = cur->next;
Bfree(cur);
break;
}
prev = cur;
} while ((cur = cur->next));
}
intptr_t hash_find(const hashtable_t * const t, char const * const s)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (t->items == NULL)
{
initprintf("hash_find(): table not initialized!\n");
return -1;
}
#endif
hashitem_t *cur = t->items[hash_getcode(s) % t->size];
if (!cur)
return -1;
do
if (Bstrcmp(s, cur->string) == 0)
return cur->key;
while ((cur = cur->next));
return -1;
}
intptr_t hash_findcase(const hashtable_t * const t, char const * const s)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (t->items == NULL)
{
initprintf("hash_findcase(): table not initialized!\n");
return -1;
}
#endif
hashitem_t *cur = t->items[hash_getcode(s) % t->size];
if (!cur)
return -1;
do
if (Bstrcasecmp(s, cur->string) == 0)
return cur->key;
while ((cur = cur->next));
return -1;
}
void inthash_init(inthashtable_t *t)
{
if (EDUKE32_PREDICT_FALSE(!t->count))
{
initputs("inthash_add(): count is zero!\n");
return;
}
inthash_free(t);
t->items = (inthashitem_t *) Xcalloc(t->count, sizeof(inthashitem_t));
}
void inthash_loop(inthashtable_t const *t, void(*func)(intptr_t, intptr_t))
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (EDUKE32_PREDICT_FALSE(t->items == NULL))
{
initputs("inthash_loop(): table not initialized!\n");
return;
}
#endif
for (inthashitem_t const * item = t->items, *const items_end = t->items + t->count; item < items_end; ++item)
func(item->key, item->value);
}
void inthash_free(inthashtable_t *t)
{
DO_FREE_AND_NULL(t->items);
}
// djb3 algorithm
static inline uint32_t inthash_getcode(intptr_t key)
{
uint32_t h = 5381;
for (uint8_t const * keybuf = (uint8_t *) &key, *const keybuf_end = keybuf + sizeof(intptr_t); keybuf < keybuf_end; ++keybuf)
h = ((h << 5) + h) ^ (uint32_t) *keybuf;
return h;
}
void inthash_add(inthashtable_t *t, intptr_t key, intptr_t value, int32_t replace)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (EDUKE32_PREDICT_FALSE(t->items == NULL))
{
initputs("inthash_add(): table not initialized!\n");
return;
}
#endif
inthashitem_t * seeker = t->items + inthash_getcode(key) % t->count;
if (seeker->collision == NULL)
{
seeker->key = key;
seeker->value = value;
seeker->collision = seeker;
return;
}
if (seeker->key == key)
{
if (replace)
seeker->value = value;
return;
}
while (seeker != seeker->collision)
{
seeker = seeker->collision;
if (seeker->key == key)
{
if (replace)
seeker->value = value;
return;
}
}
inthashitem_t *tail = seeker;
do
tail = t->items + (tail - t->items + 1) % t->count;
while (tail->collision != NULL && tail != seeker);
if (EDUKE32_PREDICT_FALSE(tail == seeker))
{
initputs("inthash_add(): table full!\n");
return;
}
tail->key = key;
tail->value = value;
tail->collision = seeker->collision = tail;
}
// delete at most once
void inthash_delete(inthashtable_t *t, intptr_t key)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (EDUKE32_PREDICT_FALSE(t->items == NULL))
{
initputs("inthash_delete(): table not initialized!\n");
return;
}
#endif
inthashitem_t * seeker = t->items + inthash_getcode(key) % t->count;
if (seeker->collision == NULL)
return;
if (seeker->key == key)
{
seeker->collision = NULL;
return;
}
while (seeker != seeker->collision)
{
inthashitem_t * const prev = seeker;
seeker = seeker->collision;
if (seeker->key == key)
{
prev->collision = seeker == seeker->collision ? prev : seeker->collision;
seeker->collision = NULL;
return;
}
}
}
intptr_t inthash_find(inthashtable_t const *t, intptr_t key)
{
#ifdef DEBUGGINGAIDS
Bassert(t->items != NULL);
#else
if (EDUKE32_PREDICT_FALSE(t->items == NULL))
{
initputs("inthash_find(): table not initialized!\n");
return -1;
}
#endif
inthashitem_t const * seeker = t->items + inthash_getcode(key) % t->count;
if (seeker->collision == NULL)
return -1;
if (seeker->key == key)
return seeker->value;
while (seeker != seeker->collision)
{
seeker = seeker->collision;
if (seeker->key == key)
return seeker->value;
}
return -1;
}