/** \fn cmph_uint32 cmph_packed_size(cmph_t *mphf);
* \brief Return the amount of space needed to pack mphf.
* \param mphf pointer to a mphf
* \return the size of the packed function or zero for failures
*/
cmph_uint32 cmph_packed_size(cmph_t *mphf)
{
switch(mphf->algo)
{
case CMPH_CHM:
return chm_packed_size(mphf);
case CMPH_BMZ: /* included -- Fabiano */
return bmz_packed_size(mphf);
case CMPH_BMZ8: /* included -- Fabiano */
return bmz8_packed_size(mphf);
case CMPH_BRZ: /* included -- Fabiano */
return brz_packed_size(mphf);
case CMPH_FCH: /* included -- Fabiano */
return fch_packed_size(mphf);
case CMPH_BDZ: /* included -- Fabiano */
return bdz_packed_size(mphf);
case CMPH_BDZ_PH: /* included -- Fabiano */
return bdz_ph_packed_size(mphf);
case CMPH_CHD_PH: /* included -- Fabiano */
return chd_ph_packed_size(mphf);
case CMPH_CHD: /* included -- Fabiano */
return chd_packed_size(mphf);
default:
assert(0);
}
return 0; // FAILURE
}
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;
}
