void fch_destroy(cmph_t *mphf)
{
fch_data_t *data = (fch_data_t *)mphf->data;
free(data->g);
hash_state_destroy(data->h1);
hash_state_destroy(data->h2);
free(data);
free(mphf);
}
void bmz8_destroy(cmph_t *mphf)
{
bmz8_data_t *data = (bmz8_data_t *)mphf->data;
free(data->g);
hash_state_destroy(data->hashes[0]);
hash_state_destroy(data->hashes[1]);
free(data->hashes);
free(data);
free(mphf);
}
static fch_buckets_t * mapping(cmph_config_t *mph)
{
cmph_uint32 i = 0;
fch_buckets_t *buckets = NULL;
fch_config_data_t *fch = (fch_config_data_t *)mph->data;
if (fch->h1) hash_state_destroy(fch->h1);
fch->h1 = hash_state_new(fch->hashfuncs[0], fch->m);
fch->b = fch_calc_b(fch->c, fch->m);
fch->p1 = fch_calc_p1(fch->m);
fch->p2 = fch_calc_p2(fch->b);
//DEBUGP("b:%u p1:%f p2:%f\n", fch->b, fch->p1, fch->p2);
buckets = fch_buckets_new(fch->b);
mph->key_source->rewind(mph->key_source->data);
for(i = 0; i < fch->m; i++)
{
cmph_uint32 h1, keylen;
char *key = NULL;
mph->key_source->read(mph->key_source->data, &key, &keylen);
h1 = hash(fch->h1, key, keylen) % fch->m;
h1 = mixh10h11h12 (fch->b, fch->p1, fch->p2, h1);
fch_buckets_insert(buckets, h1, key, keylen);
key = NULL; // transger memory ownership
}
return buckets;
}
void chd_ph_destroy(cmph_t *mphf)
{
chd_ph_data_t *data = (chd_ph_data_t *)mphf->data;
compressed_seq_destroy(data->cs);
free(data->cs);
hash_state_destroy(data->hl);
free(data);
free(mphf);
}
void brz_destroy(cmph_t *mphf)
{
cmph_uint32 i;
brz_data_t *data = (brz_data_t *)mphf->data;
if(data->g)
{
for(i = 0; i < data->k; i++)
{
free(data->g[i]);
hash_state_destroy(data->h1[i]);
hash_state_destroy(data->h2[i]);
}
free(data->g);
free(data->h1);
free(data->h2);
}
hash_state_destroy(data->h0);
free(data->size);
free(data->offset);
free(data);
free(mphf);
}
cmph_t *bmz8_new(cmph_config_t *mph, double c)
{
cmph_t *mphf = NULL;
bmz8_data_t *bmz8f = NULL;
cmph_uint8 i;
cmph_uint8 iterations;
cmph_uint8 iterations_map = 20;
cmph_uint8 *used_edges = NULL;
cmph_uint8 restart_mapping = 0;
cmph_uint8 * visited = NULL;
bmz8_config_data_t *bmz8 = (bmz8_config_data_t *)mph->data;
if (mph->key_source->nkeys >= 256)
{
if (mph->verbosity) fprintf(stderr, "The number of keys in BMZ8 must be lower than 256.\n");
return NULL;
}
if (c == 0) c = 1.15; // validating restrictions over parameter c.
DEBUGP("c: %f\n", c);
bmz8->m = (cmph_uint8) mph->key_source->nkeys;
bmz8->n = (cmph_uint8) ceil(c * mph->key_source->nkeys);
DEBUGP("m (edges): %u n (vertices): %u c: %f\n", bmz8->m, bmz8->n, c);
bmz8->graph = graph_new(bmz8->n, bmz8->m);
DEBUGP("Created graph\n");
bmz8->hashes = (hash_state_t **)malloc(sizeof(hash_state_t *)*3);
for(i = 0; i < 3; ++i) bmz8->hashes[i] = NULL;
do
{
// Mapping step
cmph_uint8 biggest_g_value = 0;
cmph_uint8 biggest_edge_value = 1;
iterations = 100;
if (mph->verbosity)
{
fprintf(stderr, "Entering mapping step for mph creation of %u keys with graph sized %u\n", bmz8->m, bmz8->n);
}
while(1)
{
int ok;
DEBUGP("hash function 1\n");
bmz8->hashes[0] = hash_state_new(bmz8->hashfuncs[0], bmz8->n);
DEBUGP("hash function 2\n");
bmz8->hashes[1] = hash_state_new(bmz8->hashfuncs[1], bmz8->n);
DEBUGP("Generating edges\n");
ok = bmz8_gen_edges(mph);
if (!ok)
{
--iterations;
hash_state_destroy(bmz8->hashes[0]);
bmz8->hashes[0] = NULL;
hash_state_destroy(bmz8->hashes[1]);
bmz8->hashes[1] = NULL;
DEBUGP("%u iterations remaining\n", iterations);
if (mph->verbosity)
{
fprintf(stderr, "simple graph creation failure - %u iterations remaining\n", iterations);
}
if (iterations == 0) break;
}
else break;
}
if (iterations == 0)
{
graph_destroy(bmz8->graph);
return NULL;
}
// Ordering step
if (mph->verbosity)
{
fprintf(stderr, "Starting ordering step\n");
}
graph_obtain_critical_nodes(bmz8->graph);
// Searching step
if (mph->verbosity)
{
fprintf(stderr, "Starting Searching step.\n");
fprintf(stderr, "\tTraversing critical vertices.\n");
}
DEBUGP("Searching step\n");
visited = (cmph_uint8 *)malloc((size_t)bmz8->n/8 + 1);
memset(visited, 0, (size_t)bmz8->n/8 + 1);
used_edges = (cmph_uint8 *)malloc((size_t)bmz8->m/8 + 1);
memset(used_edges, 0, (size_t)bmz8->m/8 + 1);
free(bmz8->g);
bmz8->g = (cmph_uint8 *)calloc((size_t)bmz8->n, sizeof(cmph_uint8));
assert(bmz8->g);
for (i = 0; i < bmz8->n; ++i) // critical nodes
{
if (graph_node_is_critical(bmz8->graph, i) && (!GETBIT(visited,i)))
{
if(c > 1.14) restart_mapping = bmz8_traverse_critical_nodes(bmz8, i, &biggest_g_value, &biggest_edge_value, used_edges, visited);
else restart_mapping = bmz8_traverse_critical_nodes_heuristic(bmz8, i, &biggest_g_value, &biggest_edge_value, used_edges, visited);
if(restart_mapping) break;
}
}
if(!restart_mapping)
{
if (mph->verbosity)
{
fprintf(stderr, "\tTraversing non critical vertices.\n");
}
bmz8_traverse_non_critical_nodes(bmz8, used_edges, visited); // non_critical_nodes
}
else
{
iterations_map--;
if (mph->verbosity) fprintf(stderr, "Restarting mapping step. %u iterations remaining.\n", iterations_map);
}
free(used_edges);
free(visited);
}while(restart_mapping && iterations_map > 0);
graph_destroy(bmz8->graph);
bmz8->graph = NULL;
if (iterations_map == 0)
{
return NULL;
}
mphf = (cmph_t *)malloc(sizeof(cmph_t));
mphf->algo = mph->algo;
bmz8f = (bmz8_data_t *)malloc(sizeof(bmz8_data_t));
bmz8f->g = bmz8->g;
bmz8->g = NULL; //transfer memory ownership
bmz8f->hashes = bmz8->hashes;
bmz8->hashes = NULL; //transfer memory ownership
bmz8f->n = bmz8->n;
bmz8f->m = bmz8->m;
mphf->data = bmz8f;
mphf->size = bmz8->m;
DEBUGP("Successfully generated minimal perfect hash\n");
if (mph->verbosity)
{
fprintf(stderr, "Successfully generated minimal perfect hash function\n");
}
return mphf;
}
static cmph_uint8 searching(fch_config_data_t *fch, fch_buckets_t *buckets, cmph_uint32 *sorted_indexes)
{
cmph_uint32 * random_table = (cmph_uint32 *) calloc((size_t)fch->m, sizeof(cmph_uint32));
cmph_uint32 * map_table = (cmph_uint32 *) calloc((size_t)fch->m, sizeof(cmph_uint32));
cmph_uint32 iteration_to_generate_h2 = 0;
cmph_uint32 searching_iterations = 0;
cmph_uint8 restart = 0;
cmph_uint32 nbuckets = fch_buckets_get_nbuckets(buckets);
cmph_uint32 i, j, z, counter = 0, filled_count = 0;
if (fch->g) free (fch->g);
fch->g = (cmph_uint32 *) calloc((size_t)fch->b, sizeof(cmph_uint32));
//DEBUGP("max bucket size: %u\n", fch_buckets_get_max_size(buckets));
for(i = 0; i < fch->m; i++)
{
random_table[i] = i;
}
permut(random_table, fch->m);
for(i = 0; i < fch->m; i++)
{
map_table[random_table[i]] = i;
}
do {
if (fch->h2) hash_state_destroy(fch->h2);
fch->h2 = hash_state_new(fch->hashfuncs[1], fch->m);
restart = check_for_collisions_h2(fch, buckets, sorted_indexes);
filled_count = 0;
if (!restart)
{
searching_iterations++; iteration_to_generate_h2 = 0;
//DEBUGP("searching_iterations: %u\n", searching_iterations);
}
else {
iteration_to_generate_h2++;
//DEBUGP("iteration_to_generate_h2: %u\n", iteration_to_generate_h2);
}
for(i = 0; (i < nbuckets) && !restart; i++) {
cmph_uint32 bucketsize = fch_buckets_get_size(buckets, sorted_indexes[i]);
if (bucketsize == 0)
{
restart = 0; // false
break;
}
else restart = 1; // true
for(z = 0; (z < (fch->m - filled_count)) && restart; z++) {
char * key = fch_buckets_get_key(buckets, sorted_indexes[i], INDEX);
cmph_uint32 keylen = fch_buckets_get_keylength(buckets, sorted_indexes[i], INDEX);
cmph_uint32 h2 = hash(fch->h2, key, keylen) % fch->m;
counter = 0;
restart = 0; // false
fch->g[sorted_indexes[i]] = (fch->m + random_table[filled_count + z] - h2) % fch->m;
//DEBUGP("g[%u]: %u\n", sorted_indexes[i], fch->g[sorted_indexes[i]]);
j = INDEX;
do {
cmph_uint32 index = 0;
key = fch_buckets_get_key(buckets, sorted_indexes[i], j);
keylen = fch_buckets_get_keylength(buckets, sorted_indexes[i], j);
h2 = hash(fch->h2, key, keylen) % fch->m;
index = (h2 + fch->g[sorted_indexes[i]]) % fch->m;
//DEBUGP("key:%s keylen:%u index: %u h2:%u bucketsize:%u\n", key, keylen, index, h2, bucketsize);
if (map_table[index] >= filled_count) {
cmph_uint32 y = map_table[index];
cmph_uint32 ry = random_table[y];
random_table[y] = random_table[filled_count];
random_table[filled_count] = ry;
map_table[random_table[y]] = y;
map_table[random_table[filled_count]] = filled_count;
filled_count++;
counter ++;
}
else {
restart = 1; // true
filled_count = filled_count - counter;
counter = 0;
break;
}
j = (j + 1) % bucketsize;
} while(j % bucketsize != INDEX);
}
//getchar();
}
//} while(restart && (searching_iterations < 10));
} while(restart && (searching_iterations < 10) && (iteration_to_generate_h2 < 1000)); // DC
free(map_table);
free(random_table);
return restart;
}
cmph_t *chd_ph_new(cmph_config_t *mph, double c)
{
cmph_t *mphf = NULL;
chd_ph_data_t *chd_phf = NULL;
chd_ph_config_data_t *chd_ph = (chd_ph_config_data_t *)mph->data;
register double load_factor = c;
register cmph_uint8 searching_success = 0;
register cmph_uint32 max_probes = 1 << 20; // default value for max_probes
register cmph_uint32 iterations = 100;
chd_ph_bucket_t * buckets = NULL;
chd_ph_item_t * items = NULL;
register cmph_uint8 failure = 0;
cmph_uint32 max_bucket_size = 0;
chd_ph_sorted_list_t * sorted_lists = NULL;
cmph_uint32 * disp_table = NULL;
register double space_lower_bound = 0;
#ifdef CMPH_TIMING
double construction_time_begin = 0.0;
double construction_time = 0.0;
ELAPSED_TIME_IN_SECONDS(&construction_time_begin);
#endif
chd_ph->m = mph->key_source->nkeys;
DEBUGP("m = %u\n", chd_ph->m);
chd_ph->nbuckets = (cmph_uint32)(chd_ph->m/chd_ph->keys_per_bucket) + 1;
DEBUGP("nbuckets = %u\n", chd_ph->nbuckets);
if(load_factor < 0.5 )
{
load_factor = 0.5;
}
if(load_factor >= 0.99)
{
load_factor = 0.99;
}
DEBUGP("load_factor = %.3f\n", load_factor);
chd_ph->n = (cmph_uint32)(chd_ph->m/(chd_ph->keys_per_bin * load_factor)) + 1;
//Round the number of bins to the prime immediately above
if(chd_ph->n % 2 == 0) chd_ph->n++;
for(;;)
{
if(check_primality(chd_ph->n) == 1)
break;
chd_ph->n += 2; // just odd numbers can be primes for n > 2
};
DEBUGP("n = %u \n", chd_ph->n);
if(chd_ph->keys_per_bin == 1)
{
space_lower_bound = chd_ph_space_lower_bound(chd_ph->m, chd_ph->n);
}
if(mph->verbosity)
{
fprintf(stderr, "space lower bound is %.3f bits per key\n", space_lower_bound);
}
// We allocate the working tables
buckets = chd_ph_bucket_new(chd_ph->nbuckets);
items = (chd_ph_item_t *) calloc(chd_ph->m, sizeof(chd_ph_item_t));
max_probes = (cmph_uint32)(((log(chd_ph->m)/log(2))/20) * max_probes);
if(chd_ph->keys_per_bin == 1)
chd_ph->occup_table = (cmph_uint8 *) calloc(((chd_ph->n + 31)/32), sizeof(cmph_uint32));
else
chd_ph->occup_table = (cmph_uint8 *) calloc(chd_ph->n, sizeof(cmph_uint8));
disp_table = (cmph_uint32 *) calloc(chd_ph->nbuckets, sizeof(cmph_uint32));
//
// init_genrand(time(0));
while(1)
{
iterations --;
if (mph->verbosity)
{
fprintf(stderr, "Starting mapping step for mph creation of %u keys with %u bins\n", chd_ph->m, chd_ph->n);
}
if(!chd_ph_mapping(mph, buckets, items, &max_bucket_size))
{
if (mph->verbosity)
{
fprintf(stderr, "Failure in mapping step\n");
}
failure = 1;
goto cleanup;
}
if (mph->verbosity)
{
fprintf(stderr, "Starting ordering step\n");
}
if(sorted_lists)
{
free(sorted_lists);
}
sorted_lists = chd_ph_ordering(&buckets, &items, chd_ph->nbuckets, chd_ph->m, max_bucket_size);
if (mph->verbosity)
{
fprintf(stderr, "Starting searching step\n");
}
searching_success = chd_ph_searching(chd_ph, buckets, items, max_bucket_size, sorted_lists, max_probes, disp_table);
if(searching_success) break;
// reset occup_table
if(chd_ph->keys_per_bin > 1)
memset(chd_ph->occup_table, 0, chd_ph->n);
else
memset(chd_ph->occup_table, 0, ((chd_ph->n + 31)/32) * sizeof(cmph_uint32));
if(iterations == 0)
{
// Cleanup memory
if (mph->verbosity)
{
fprintf(stderr, "Failure because the max trials was exceeded\n");
}
failure = 1;
goto cleanup;
};
}
#ifdef DEBUG
{
if(!chd_ph_check_bin_hashing(chd_ph, buckets, items, disp_table,sorted_lists,max_bucket_size))
{
DEBUGP("Error for bin packing generation");
failure = 1;
goto cleanup;
}
}
#endif
if (mph->verbosity)
{
fprintf(stderr, "Starting compressing step\n");
}
if(chd_ph->cs)
{
free(chd_ph->cs);
}
chd_ph->cs = (compressed_seq_t *) calloc(1, sizeof(compressed_seq_t));
compressed_seq_init(chd_ph->cs);
compressed_seq_generate(chd_ph->cs, disp_table, chd_ph->nbuckets);
#ifdef CMPH_TIMING
ELAPSED_TIME_IN_SECONDS(&construction_time);
register double entropy = chd_ph_get_entropy(disp_table, chd_ph->nbuckets, max_probes);
DEBUGP("Entropy = %.4f\n", entropy/chd_ph->m);
#endif
cleanup:
chd_ph_bucket_destroy(buckets);
free(items);
free(sorted_lists);
free(disp_table);
if(failure)
{
if(chd_ph->hl)
{
hash_state_destroy(chd_ph->hl);
}
chd_ph->hl = NULL;
return NULL;
}
mphf = (cmph_t *)malloc(sizeof(cmph_t));
mphf->algo = mph->algo;
chd_phf = (chd_ph_data_t *)malloc(sizeof(chd_ph_data_t));
chd_phf->cs = chd_ph->cs;
chd_ph->cs = NULL; //transfer memory ownership
chd_phf->hl = chd_ph->hl;
chd_ph->hl = NULL; //transfer memory ownership
chd_phf->n = chd_ph->n;
chd_phf->nbuckets = chd_ph->nbuckets;
mphf->data = chd_phf;
mphf->size = chd_ph->n;
DEBUGP("Successfully generated minimal perfect hash\n");
if (mph->verbosity)
{
fprintf(stderr, "Successfully generated minimal perfect hash function\n");
}
#ifdef CMPH_TIMING
register cmph_uint32 space_usage = chd_ph_packed_size(mphf)*8;
construction_time = construction_time - construction_time_begin;
fprintf(stdout, "%u\t%.2f\t%u\t%.4f\t%.4f\t%.4f\t%.4f\n", chd_ph->m, load_factor, chd_ph->keys_per_bucket, construction_time, space_usage/(double)chd_ph->m, space_lower_bound, entropy/chd_ph->m);
#endif
return mphf;
}
cmph_uint8 chd_ph_mapping(cmph_config_t *mph, chd_ph_bucket_t * buckets, chd_ph_item_t * items, cmph_uint32 *max_bucket_size)
{
register cmph_uint32 i = 0, g = 0;
cmph_uint32 hl[3];
chd_ph_config_data_t *chd_ph = (chd_ph_config_data_t *)mph->data;
char * key = NULL;
cmph_uint32 keylen = 0;
chd_ph_map_item_t * map_item;
chd_ph_map_item_t * map_items = malloc(chd_ph->m*sizeof(chd_ph_map_item_t));
register cmph_uint32 mapping_iterations = 1000;
*max_bucket_size = 0;
while(1)
{
mapping_iterations--;
if (chd_ph->hl) hash_state_destroy(chd_ph->hl);
chd_ph->hl = hash_state_new(chd_ph->hashfunc, chd_ph->m);
chd_ph_bucket_clean(buckets, chd_ph->nbuckets);
mph->key_source->rewind(mph->key_source->data);
for(i = 0; i < chd_ph->m; i++)
{
mph->key_source->read(mph->key_source->data, &key, &keylen);
hash_vector(chd_ph->hl, key, keylen, hl);
map_item = (map_items + i);
g = hl[0] % chd_ph->nbuckets;
map_item->f = hl[1] % chd_ph->n;
map_item->h = hl[2] % (chd_ph->n - 1) + 1;
map_item->bucket_num=g;
mph->key_source->dispose(mph->key_source->data, key, keylen);
// if(buckets[g].size == (chd_ph->keys_per_bucket << 2))
// {
// DEBUGP("BUCKET = %u -- SIZE = %u -- MAXIMUM SIZE = %u\n", g, buckets[g].size, (chd_ph->keys_per_bucket << 2));
// goto error;
// }
buckets[g].size++;
if(buckets[g].size > *max_bucket_size)
{
*max_bucket_size = buckets[g].size;
}
}
buckets[0].items_list = 0;
for(i = 1; i < chd_ph->nbuckets; i++)
{
buckets[i].items_list = buckets[i-1].items_list + buckets[i - 1].size;
buckets[i - 1].size = 0;
};
buckets[i - 1].size = 0;
for(i = 0; i < chd_ph->m; i++)
{
map_item = (map_items + i);
if(!chd_ph_bucket_insert(buckets, map_items, items, chd_ph->nbuckets, i))
break;
}
if(i == chd_ph->m)
{
free(map_items);
return 1; // SUCCESS
}
if(mapping_iterations == 0)
{
goto error;
}
}
error:
free(map_items);
hash_state_destroy(chd_ph->hl);
chd_ph->hl = NULL;
return 0; // FAILURE
}
cmph_t *chm_new(cmph_config_t *mph, double c)
{
cmph_t *mphf = NULL;
chm_data_t *chmf = NULL;
cmph_uint32 i;
cmph_uint32 iterations = 20;
cmph_uint8 *visited = NULL;
chm_config_data_t *chm = (chm_config_data_t *)mph->data;
chm->m = mph->key_source->nkeys;
if (c == 0) c = 2.09;
chm->n = (cmph_uint32)ceil(c * mph->key_source->nkeys);
DEBUGP("m (edges): %u n (vertices): %u c: %f\n", chm->m, chm->n, c);
chm->graph = graph_new(chm->n, chm->m);
DEBUGP("Created graph\n");
chm->hashes = (hash_state_t **)malloc(sizeof(hash_state_t *)*3);
for(i = 0; i < 3; ++i) chm->hashes[i] = NULL;
//Mapping step
if (mph->verbosity)
{
fprintf(stderr, "Entering mapping step for mph creation of %u keys with graph sized %u\n", chm->m, chm->n);
}
while(1)
{
int ok;
chm->hashes[0] = hash_state_new(chm->hashfuncs[0], chm->n);
chm->hashes[1] = hash_state_new(chm->hashfuncs[1], chm->n);
ok = chm_gen_edges(mph);
if (!ok)
{
--iterations;
hash_state_destroy(chm->hashes[0]);
chm->hashes[0] = NULL;
hash_state_destroy(chm->hashes[1]);
chm->hashes[1] = NULL;
DEBUGP("%u iterations remaining\n", iterations);
if (mph->verbosity)
{
fprintf(stderr, "Acyclic graph creation failure - %u iterations remaining\n", iterations);
}
if (iterations == 0) break;
}
else break;
}
if (iterations == 0)
{
graph_destroy(chm->graph);
return NULL;
}
//Assignment step
if (mph->verbosity)
{
fprintf(stderr, "Starting assignment step\n");
}
DEBUGP("Assignment step\n");
visited = (cmph_uint8 *)malloc((size_t)(chm->n/8 + 1));
memset(visited, 0, (size_t)(chm->n/8 + 1));
free(chm->g);
chm->g = (cmph_uint32 *)malloc(chm->n * sizeof(cmph_uint32));
assert(chm->g);
for (i = 0; i < chm->n; ++i)
{
if (!GETBIT(visited,i))
{
chm->g[i] = 0;
chm_traverse(chm, visited, i);
}
}
graph_destroy(chm->graph);
free(visited);
chm->graph = NULL;
mphf = (cmph_t *)malloc(sizeof(cmph_t));
mphf->algo = mph->algo;
chmf = (chm_data_t *)malloc(sizeof(chm_data_t));
chmf->g = chm->g;
chm->g = NULL; //transfer memory ownership
chmf->hashes = chm->hashes;
chm->hashes = NULL; //transfer memory ownership
chmf->n = chm->n;
chmf->m = chm->m;
mphf->data = chmf;
mphf->size = chm->m;
DEBUGP("Successfully generated minimal perfect hash\n");
if (mph->verbosity)
{
fprintf(stderr, "Successfully generated minimal perfect hash function\n");
}
return mphf;
}
cmph_t *bdz_new(cmph_config_t *mph, double c)
{
cmph_t *mphf = NULL;
bdz_data_t *bdzf = NULL;
cmph_uint32 iterations;
bdz_queue_t edges;
bdz_graph3_t graph3;
bdz_config_data_t *bdz = (bdz_config_data_t *)mph->data;
#ifdef CMPH_TIMING
double construction_time_begin = 0.0;
double construction_time = 0.0;
ELAPSED_TIME_IN_SECONDS(&construction_time_begin);
#endif
if (c == 0) c = 1.23; // validating restrictions over parameter c.
DEBUGP("c: %f\n", c);
bdz->m = mph->key_source->nkeys;
bdz->r = (cmph_uint32)ceil((c * mph->key_source->nkeys)/3);
if ((bdz->r % 2) == 0) bdz->r+=1;
bdz->n = 3*bdz->r;
bdz->k = (1U << bdz->b);
DEBUGP("b: %u -- k: %u\n", bdz->b, bdz->k);
bdz->ranktablesize = (cmph_uint32)ceil(bdz->n/(double)bdz->k);
DEBUGP("ranktablesize: %u\n", bdz->ranktablesize);
bdz_alloc_graph3(&graph3, bdz->m, bdz->n);
bdz_alloc_queue(&edges,bdz->m);
DEBUGP("Created hypergraph\n");
DEBUGP("m (edges): %u n (vertices): %u r: %u c: %f \n", bdz->m, bdz->n, bdz->r, c);
// Mapping step
iterations = 1000;
if (mph->verbosity)
{
fprintf(stderr, "Entering mapping step for mph creation of %u keys with graph sized %u\n", bdz->m, bdz->n);
}
while(1)
{
int ok;
DEBUGP("linear hash function \n");
bdz->hl = hash_state_new(bdz->hashfunc, 15);
ok = bdz_mapping(mph, &graph3, edges);
//ok = 0;
if (!ok)
{
--iterations;
hash_state_destroy(bdz->hl);
bdz->hl = NULL;
DEBUGP("%u iterations remaining\n", iterations);
if (mph->verbosity)
{
fprintf(stderr, "acyclic graph creation failure - %u iterations remaining\n", iterations);
}
if (iterations == 0) break;
}
else break;
}
if (iterations == 0)
{
bdz_free_queue(&edges);
bdz_free_graph3(&graph3);
return NULL;
}
bdz_partial_free_graph3(&graph3);
// Assigning step
if (mph->verbosity)
{
fprintf(stderr, "Entering assigning step for mph creation of %u keys with graph sized %u\n", bdz->m, bdz->n);
}
assigning(bdz, &graph3, edges);
bdz_free_queue(&edges);
bdz_free_graph3(&graph3);
if (mph->verbosity)
{
fprintf(stderr, "Entering ranking step for mph creation of %u keys with graph sized %u\n", bdz->m, bdz->n);
}
ranking(bdz);
#ifdef CMPH_TIMING
ELAPSED_TIME_IN_SECONDS(&construction_time);
#endif
mphf = (cmph_t *)malloc(sizeof(cmph_t));
mphf->algo = mph->algo;
bdzf = (bdz_data_t *)malloc(sizeof(bdz_data_t));
bdzf->g = bdz->g;
bdz->g = NULL; //transfer memory ownership
bdzf->hl = bdz->hl;
bdz->hl = NULL; //transfer memory ownership
bdzf->ranktable = bdz->ranktable;
bdz->ranktable = NULL; //transfer memory ownership
bdzf->ranktablesize = bdz->ranktablesize;
bdzf->k = bdz->k;
bdzf->b = bdz->b;
bdzf->n = bdz->n;
bdzf->m = bdz->m;
bdzf->r = bdz->r;
mphf->data = bdzf;
mphf->size = bdz->m;
DEBUGP("Successfully generated minimal perfect hash\n");
if (mph->verbosity)
{
fprintf(stderr, "Successfully generated minimal perfect hash function\n");
}
#ifdef CMPH_TIMING
register cmph_uint32 space_usage = bdz_packed_size(mphf)*8;
register cmph_uint32 keys_per_bucket = 1;
construction_time = construction_time - construction_time_begin;
fprintf(stdout, "%u\t%.2f\t%u\t%.4f\t%.4f\n", bdz->m, bdz->m/(double)bdz->n, keys_per_bucket, construction_time, space_usage/(double)bdz->m);
#endif
return mphf;
}
cmph_t *brz_new(cmph_config_t *mph, double c)
{
cmph_t *mphf = NULL;
brz_data_t *brzf = NULL;
cmph_uint32 i;
cmph_uint32 iterations = 20;
DEBUGP("c: %f\n", c);
brz_config_data_t *brz = (brz_config_data_t *)mph->data;
switch(brz->algo) // validating restrictions over parameter c.
{
case CMPH_BMZ8:
if (c == 0 || c >= 2.0) c = 1;
break;
case CMPH_FCH:
if (c <= 2.0) c = 2.6;
break;
default:
assert(0);
}
brz->c = c;
brz->m = mph->key_source->nkeys;
DEBUGP("m: %u\n", brz->m);
brz->k = (cmph_uint32)ceil(brz->m/((double)brz->b));
DEBUGP("k: %u\n", brz->k);
brz->size = (cmph_uint8 *) calloc((size_t)brz->k, sizeof(cmph_uint8));
// Clustering the keys by graph id.
if (mph->verbosity)
{
fprintf(stderr, "Partioning the set of keys.\n");
}
while(1)
{
int ok;
DEBUGP("hash function 3\n");
brz->h0 = hash_state_new(brz->hashfuncs[2], brz->k);
DEBUGP("Generating graphs\n");
ok = brz_gen_mphf(mph);
if (!ok)
{
--iterations;
hash_state_destroy(brz->h0);
brz->h0 = NULL;
DEBUGP("%u iterations remaining to create the graphs in a external file\n", iterations);
if (mph->verbosity)
{
fprintf(stderr, "Failure: A graph with more than 255 keys was created - %u iterations remaining\n", iterations);
}
if (iterations == 0) break;
}
else break;
}
if (iterations == 0)
{
DEBUGP("Graphs with more than 255 keys were created in all 20 iterations\n");
free(brz->size);
return NULL;
}
DEBUGP("Graphs generated\n");
brz->offset = (cmph_uint32 *)calloc((size_t)brz->k, sizeof(cmph_uint32));
for (i = 1; i < brz->k; ++i)
{
brz->offset[i] = brz->size[i-1] + brz->offset[i-1];
}
// Generating a mphf
mphf = (cmph_t *)malloc(sizeof(cmph_t));
mphf->algo = mph->algo;
brzf = (brz_data_t *)malloc(sizeof(brz_data_t));
brzf->g = brz->g;
brz->g = NULL; //transfer memory ownership
brzf->h1 = brz->h1;
brz->h1 = NULL; //transfer memory ownership
brzf->h2 = brz->h2;
brz->h2 = NULL; //transfer memory ownership
brzf->h0 = brz->h0;
brz->h0 = NULL; //transfer memory ownership
brzf->size = brz->size;
brz->size = NULL; //transfer memory ownership
brzf->offset = brz->offset;
brz->offset = NULL; //transfer memory ownership
brzf->k = brz->k;
brzf->c = brz->c;
brzf->m = brz->m;
brzf->algo = brz->algo;
mphf->data = brzf;
mphf->size = brz->m;
DEBUGP("Successfully generated minimal perfect hash\n");
if (mph->verbosity)
{
fprintf(stderr, "Successfully generated minimal perfect hash function\n");
}
return mphf;
}
