/** cmph_uint32 fch_search(void *packed_mphf, const char *key, cmph_uint32 keylen);
* \brief Use the packed mphf to do a search.
* \param packed_mphf pointer to the packed mphf
* \param key key to be hashed
* \param keylen key legth in bytes
* \return The mphf value
*/
cmph_uint32 fch_search_packed(void *packed_mphf, const char *key, cmph_uint32 keylen)
{
register cmph_uint8 *h1_ptr = packed_mphf;
register CMPH_HASH h1_type = *((cmph_uint32 *)h1_ptr);
h1_ptr += 4;
register cmph_uint8 *h2_ptr = h1_ptr + hash_state_packed_size(h1_type);
register CMPH_HASH h2_type = *((cmph_uint32 *)h2_ptr);
h2_ptr += 4;
register cmph_uint32 *g_ptr = (cmph_uint32 *)(h2_ptr + hash_state_packed_size(h2_type));
register cmph_uint32 m = *g_ptr++;
register cmph_uint32 b = *g_ptr++;
register double p1 = (double)(*((cmph_uint64 *)g_ptr));
g_ptr += 2;
register double p2 = (double)(*((cmph_uint64 *)g_ptr));
g_ptr += 2;
register cmph_uint32 h1 = hash_packed(h1_ptr, h1_type, key, keylen) % m;
register cmph_uint32 h2 = hash_packed(h2_ptr, h2_type, key, keylen) % m;
h1 = mixh10h11h12 (b, p1, p2, h1);
return (h2 + g_ptr[h1]) % m;
}
/** \fn void bmz8_pack(cmph_t *mphf, void *packed_mphf);
* \brief Support the ability to pack a perfect hash function into a preallocated contiguous memory space pointed by packed_mphf.
* \param mphf pointer to the resulting mphf
* \param packed_mphf pointer to the contiguous memory area used to store the resulting mphf. The size of packed_mphf must be at least cmph_packed_size()
*/
void bmz8_pack(cmph_t *mphf, void *packed_mphf)
{
bmz8_data_t *data = (bmz8_data_t *)mphf->data;
cmph_uint8 * ptr = (cmph_uint8 *)packed_mphf;
// packing h1 type
CMPH_HASH h1_type = hash_get_type(data->hashes[0]);
*((cmph_uint32 *) ptr) = h1_type;
ptr += sizeof(cmph_uint32);
// packing h1
hash_state_pack(data->hashes[0], ptr);
ptr += hash_state_packed_size(h1_type);
// packing h2 type
CMPH_HASH h2_type = hash_get_type(data->hashes[1]);
*((cmph_uint32 *) ptr) = h2_type;
ptr += sizeof(cmph_uint32);
// packing h2
hash_state_pack(data->hashes[1], ptr);
ptr += hash_state_packed_size(h2_type);
// packing n
*ptr++ = data->n;
// packing g
memcpy(ptr, data->g, sizeof(cmph_uint8)*data->n);
}
/** cmph_uint32 bmz_search(void *packed_mphf, const char *key, cmph_uint32 keylen);
* \brief Use the packed mphf to do a search.
* \param packed_mphf pointer to the packed mphf
* \param key key to be hashed
* \param keylen key legth in bytes
* \return The mphf value
*/
cmph_uint32 bmz_search_packed(void *packed_mphf, const char *key, cmph_uint32 keylen)
{
register cmph_uint8 *h1_ptr = packed_mphf;
register CMPH_HASH h1_type = *((cmph_uint32 *)h1_ptr);
register cmph_uint8 *h2_ptr;
register CMPH_HASH h2_type;
register cmph_uint32 *g_ptr;
register cmph_uint32 n;
register cmph_uint32 h1;
register cmph_uint32 h2;
h1_ptr += 4;
h2_ptr = h1_ptr + hash_state_packed_size(h1_type);
h2_type = *((cmph_uint32 *)h2_ptr);
h2_ptr += 4;
g_ptr = (cmph_uint32 *)(h2_ptr + hash_state_packed_size(h2_type));
n = *g_ptr++;
h1 = hash_packed(h1_ptr, h1_type, key, keylen) % n;
h2 = hash_packed(h2_ptr, h2_type, key, keylen) % n;
if (h1 == h2 && ++h2 > n) h2 = 0;
return (g_ptr[h1] + g_ptr[h2]);
}
/** \fn cmph_uint32 bmz8_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 bmz8_packed_size(cmph_t *mphf)
{
bmz8_data_t *data = (bmz8_data_t *)mphf->data;
CMPH_HASH h1_type = hash_get_type(data->hashes[0]);
CMPH_HASH h2_type = hash_get_type(data->hashes[1]);
return (cmph_uint32)(sizeof(CMPH_ALGO) + hash_state_packed_size(h1_type) + hash_state_packed_size(h2_type) +
2*sizeof(cmph_uint32) + sizeof(cmph_uint8) + sizeof(cmph_uint8)*data->n);
}
/** \fn cmph_uint32 fch_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 fch_packed_size(cmph_t *mphf)
{
fch_data_t *data = (fch_data_t *)mphf->data;
CMPH_HASH h1_type = hash_get_type(data->h1);
CMPH_HASH h2_type = hash_get_type(data->h2);
return (sizeof(CMPH_ALGO) + hash_state_packed_size(h1_type) + hash_state_packed_size(h2_type) +
4*sizeof(cmph_uint32) + 2*sizeof(double) + sizeof(cmph_uint32)*(data->b));
}
static cmph_uint32 brz_fch_search_packed(cmph_uint32 *packed_mphf, const char *key, cmph_uint32 keylen, cmph_uint32 * fingerprint)
{
register CMPH_HASH h0_type = *packed_mphf++;
register cmph_uint32 *h0_ptr = packed_mphf;
packed_mphf = (cmph_uint32 *)(((cmph_uint8 *)packed_mphf) + hash_state_packed_size(h0_type));
register cmph_uint32 k = *packed_mphf++;
register double c = (double)(*((cmph_uint64*)packed_mphf));
packed_mphf += 2;
register CMPH_HASH h1_type = *packed_mphf++;
register CMPH_HASH h2_type = *packed_mphf++;
register cmph_uint8 * size = (cmph_uint8 *) packed_mphf;
packed_mphf = (cmph_uint32 *)(size + k);
register cmph_uint32 * offset = packed_mphf;
packed_mphf += k;
register cmph_uint32 h0;
hash_vector_packed(h0_ptr, h0_type, key, keylen, fingerprint);
h0 = fingerprint[2] % k;
register cmph_uint32 m = size[h0];
register cmph_uint32 b = fch_calc_b(c, m);
register double p1 = fch_calc_p1(m);
register double p2 = fch_calc_p2(b);
#if defined (__ia64) || defined (__x86_64__)
register cmph_uint64 * g_is_ptr = (cmph_uint64 *)packed_mphf;
#else
register cmph_uint32 * g_is_ptr = packed_mphf;
#endif
register cmph_uint8 * h1_ptr = (cmph_uint8 *) g_is_ptr[h0];
register cmph_uint8 * h2_ptr = h1_ptr + hash_state_packed_size(h1_type);
register cmph_uint8 * g = h2_ptr + hash_state_packed_size(h2_type);
register cmph_uint32 h1 = hash_packed(h1_ptr, h1_type, key, keylen) % m;
register cmph_uint32 h2 = hash_packed(h2_ptr, h2_type, key, keylen) % m;
register cmph_uint8 mphf_bucket = 0;
h1 = mixh10h11h12(b, p1, p2, h1);
mphf_bucket = (cmph_uint8)((h2 + g[h1]) % m);
return (mphf_bucket + offset[h0]);
}
static cmph_uint32 brz_bmz8_search_packed(cmph_uint32 *packed_mphf, const char *key, cmph_uint32 keylen, cmph_uint32 * fingerprint)
{
register CMPH_HASH h0_type = *packed_mphf++;
register cmph_uint32 *h0_ptr = packed_mphf;
packed_mphf = (cmph_uint32 *)(((cmph_uint8 *)packed_mphf) + hash_state_packed_size(h0_type));
register cmph_uint32 k = *packed_mphf++;
register double c = (double)(*((cmph_uint64*)packed_mphf));
packed_mphf += 2;
register CMPH_HASH h1_type = *packed_mphf++;
register CMPH_HASH h2_type = *packed_mphf++;
register cmph_uint8 * size = (cmph_uint8 *) packed_mphf;
packed_mphf = (cmph_uint32 *)(size + k);
register cmph_uint32 * offset = packed_mphf;
packed_mphf += k;
register cmph_uint32 h0;
hash_vector_packed(h0_ptr, h0_type, key, keylen, fingerprint);
h0 = fingerprint[2] % k;
register cmph_uint32 m = size[h0];
register cmph_uint32 n = (cmph_uint32)ceil(c * m);
#if defined (__ia64) || defined (__x86_64__)
register cmph_uint64 * g_is_ptr = (cmph_uint64 *)packed_mphf;
#else
register cmph_uint32 * g_is_ptr = packed_mphf;
#endif
register cmph_uint8 * h1_ptr = (cmph_uint8 *) g_is_ptr[h0];
register cmph_uint8 * h2_ptr = h1_ptr + hash_state_packed_size(h1_type);
register cmph_uint8 * g = h2_ptr + hash_state_packed_size(h2_type);
register cmph_uint32 h1 = hash_packed(h1_ptr, h1_type, key, keylen) % n;
register cmph_uint32 h2 = hash_packed(h2_ptr, h2_type, key, keylen) % n;
register cmph_uint8 mphf_bucket;
if (h1 == h2 && ++h2 >= n) h2 = 0;
mphf_bucket = (cmph_uint8)(g[h1] + g[h2]);
DEBUGP("key: %s h1: %u h2: %u h0: %u\n", key, h1, h2, h0);
DEBUGP("Address: %u\n", mphf_bucket + offset[h0]);
return (mphf_bucket + offset[h0]);
}
cmph_uint32 chd_ph_search_packed(void *packed_mphf, const char *key, cmph_uint32 keylen)
{
register CMPH_HASH hl_type = *(cmph_uint32 *)packed_mphf;
register cmph_uint8 *hl_ptr = (cmph_uint8 *)(packed_mphf) + 4;
register cmph_uint32 * ptr = (cmph_uint32 *)(hl_ptr + hash_state_packed_size(hl_type));
register cmph_uint32 n = *ptr++;
register cmph_uint32 nbuckets = *ptr++;
cmph_uint32 hl[3];
register cmph_uint32 disp,position;
register cmph_uint32 probe0_num,probe1_num;
register cmph_uint32 f,g,h;
hash_vector_packed(hl_ptr, hl_type, key, keylen, hl);
g = hl[0] % nbuckets;
f = hl[1] % n;
h = hl[2] % (n-1) + 1;
disp = compressed_seq_query_packed(ptr, g);
probe0_num = disp % n;
probe1_num = disp/n;
position = (cmph_uint32)((f + ((cmph_uint64 )h)*probe0_num + probe1_num) % n);
return position;
}
void chd_ph_pack(cmph_t *mphf, void *packed_mphf)
{
chd_ph_data_t *data = (chd_ph_data_t *)mphf->data;
cmph_uint8 * ptr = packed_mphf;
// packing hl type
CMPH_HASH hl_type = hash_get_type(data->hl);
*((cmph_uint32 *) ptr) = hl_type;
ptr += sizeof(cmph_uint32);
// packing hl
hash_state_pack(data->hl, ptr);
ptr += hash_state_packed_size(hl_type);
// packing n
*((cmph_uint32 *) ptr) = data->n;
ptr += sizeof(data->n);
// packing nbuckets
*((cmph_uint32 *) ptr) = data->nbuckets;
ptr += sizeof(data->nbuckets);
// packing cs
compressed_seq_pack(data->cs, ptr);
//ptr += compressed_seq_packed_size(data->cs);
}
cmph_uint32 chd_ph_packed_size(cmph_t *mphf)
{
register chd_ph_data_t *data = (chd_ph_data_t *)mphf->data;
register CMPH_HASH hl_type = hash_get_type(data->hl);
register cmph_uint32 hash_state_pack_size = hash_state_packed_size(hl_type);
register cmph_uint32 cs_pack_size = compressed_seq_packed_size(data->cs);
return (cmph_uint32)(sizeof(CMPH_ALGO) + hash_state_pack_size + cs_pack_size + 3*sizeof(cmph_uint32));
}
/** \fn void fch_pack(cmph_t *mphf, void *packed_mphf);
* \brief Support the ability to pack a perfect hash function into a preallocated contiguous memory space pointed by packed_mphf.
* \param mphf pointer to the resulting mphf
* \param packed_mphf pointer to the contiguous memory area used to store the resulting mphf. The size of packed_mphf must be at least cmph_packed_size()
*/
void fch_pack(cmph_t *mphf, void *packed_mphf)
{
fch_data_t *data = (fch_data_t *)mphf->data;
cmph_uint8 * ptr = packed_mphf;
// packing h1 type
CMPH_HASH h1_type = hash_get_type(data->h1);
CMPH_HASH h2_type;
*((cmph_uint32 *) ptr) = h1_type;
ptr += sizeof(cmph_uint32);
// packing h1
hash_state_pack(data->h1, ptr);
ptr += hash_state_packed_size(h1_type);
// packing h2 type
h2_type = hash_get_type(data->h2);
*((cmph_uint32 *) ptr) = h2_type;
ptr += sizeof(cmph_uint32);
// packing h2
hash_state_pack(data->h2, ptr);
ptr += hash_state_packed_size(h2_type);
// packing m
*((cmph_uint32 *) ptr) = data->m;
ptr += sizeof(data->m);
// packing b
*((cmph_uint32 *) ptr) = data->b;
ptr += sizeof(data->b);
// packing p1
*((cmph_uint64 *)ptr) = (cmph_uint64)data->p1;
ptr += sizeof(data->p1);
// packing p2
*((cmph_uint64 *)ptr) = (cmph_uint64)data->p2;
ptr += sizeof(data->p2);
// packing g
memcpy(ptr, data->g, sizeof(cmph_uint32)*(data->b));
}
/** cmph_uint8 bmz8_search(void *packed_mphf, const char *key, cmph_uint32 keylen);
* \brief Use the packed mphf to do a search.
* \param packed_mphf pointer to the packed mphf
* \param key key to be hashed
* \param keylen key legth in bytes
* \return The mphf value
*/
cmph_uint8 bmz8_search_packed(void *packed_mphf, const char *key, cmph_uint32 keylen)
{
register cmph_uint8 *h1_ptr = (cmph_uint8 *)packed_mphf;
register CMPH_HASH h1_type = (CMPH_HASH)(*((cmph_uint32 *)h1_ptr));
h1_ptr += 4;
register cmph_uint8 *h2_ptr = h1_ptr + hash_state_packed_size(h1_type);
register CMPH_HASH h2_type = (CMPH_HASH)(*((cmph_uint32 *)h2_ptr));
h2_ptr += 4;
register cmph_uint8 *g_ptr = h2_ptr + hash_state_packed_size(h2_type);
register cmph_uint8 n = *g_ptr++;
register cmph_uint8 h1 = (cmph_uint8)(hash_packed(h1_ptr, h1_type, key, keylen) % n);
register cmph_uint8 h2 = (cmph_uint8)(hash_packed(h2_ptr, h2_type, key, keylen) % n);
DEBUGP("key: %s h1: %u h2: %u\n", key, h1, h2);
if (h1 == h2 && ++h2 > n) h2 = 0;
return (cmph_uint8)(g_ptr[h1] + g_ptr[h2]);
}
/** cmph_uint32 chm_search(void *packed_mphf, const char *key, cmph_uint32 keylen);
* \brief Use the packed mphf to do a search.
* \param packed_mphf pointer to the packed mphf
* \param key key to be hashed
* \param keylen key legth in bytes
* \return The mphf value
*/
cmph_uint32 chm_search_packed(void *packed_mphf, const char *key, cmph_uint32 keylen)
{
register cmph_uint8 *h1_ptr = packed_mphf;
register CMPH_HASH h1_type = *((cmph_uint32 *)h1_ptr);
h1_ptr += 4;
register cmph_uint8 *h2_ptr = h1_ptr + hash_state_packed_size(h1_type);
register CMPH_HASH h2_type = *((cmph_uint32 *)h2_ptr);
h2_ptr += 4;
register cmph_uint32 *g_ptr = (cmph_uint32 *)(h2_ptr + hash_state_packed_size(h2_type));
register cmph_uint32 n = *g_ptr++;
register cmph_uint32 m = *g_ptr++;
register cmph_uint32 h1 = hash_packed(h1_ptr, h1_type, key, keylen) % n;
register cmph_uint32 h2 = hash_packed(h2_ptr, h2_type, key, keylen) % n;
DEBUGP("key: %s h1: %u h2: %u\n", key, h1, h2);
if (h1 == h2 && ++h2 >= n) h2 = 0;
DEBUGP("key: %s g[h1]: %u g[h2]: %u edges: %u\n", key, g_ptr[h1], g_ptr[h2], m);
return (g_ptr[h1] + g_ptr[h2]) % m;
}
/** \fn cmph_uint32 brz_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 brz_packed_size(cmph_t *mphf)
{
cmph_uint32 i;
cmph_uint32 size = 0;
brz_data_t *data = (brz_data_t *)mphf->data;
CMPH_HASH h0_type = hash_get_type(data->h0);
CMPH_HASH h1_type = hash_get_type(data->h1[0]);
CMPH_HASH h2_type = hash_get_type(data->h2[0]);
size = (cmph_uint32)(2*sizeof(CMPH_ALGO) + 3*sizeof(CMPH_HASH) + hash_state_packed_size(h0_type) + sizeof(cmph_uint32) +
sizeof(double) + sizeof(cmph_uint8)*data->k + sizeof(cmph_uint32)*data->k);
// pointers to g_is
#if defined (__ia64) || defined (__x86_64__)
size += (cmph_uint32) sizeof(cmph_uint64)*data->k;
#else
size += (cmph_uint32) sizeof(cmph_uint32)*data->k;
#endif
size += hash_state_packed_size(h1_type) * data->k;
size += hash_state_packed_size(h2_type) * data->k;
cmph_uint32 n = 0;
for(i = 0; i < data->k; i++)
{
switch(data->algo)
{
case CMPH_FCH:
n = fch_calc_b(data->c, data->size[i]);
break;
case CMPH_BMZ8:
n = (cmph_uint32)ceil(data->c * data->size[i]);
break;
default: assert(0);
}
size += n;
}
return size;
}
/** \fn void brz_pack(cmph_t *mphf, void *packed_mphf);
* \brief Support the ability to pack a perfect hash function into a preallocated contiguous memory space pointed by packed_mphf.
* \param mphf pointer to the resulting mphf
* \param packed_mphf pointer to the contiguous memory area used to store the resulting mphf. The size of packed_mphf must be at least cmph_packed_size()
*/
void brz_pack(cmph_t *mphf, void *packed_mphf)
{
brz_data_t *data = (brz_data_t *)mphf->data;
cmph_uint8 * ptr = packed_mphf;
cmph_uint32 i,n;
// packing internal algo type
memcpy(ptr, &(data->algo), sizeof(data->algo));
ptr += sizeof(data->algo);
// packing h0 type
CMPH_HASH h0_type = hash_get_type(data->h0);
memcpy(ptr, &h0_type, sizeof(h0_type));
ptr += sizeof(h0_type);
// packing h0
hash_state_pack(data->h0, ptr);
ptr += hash_state_packed_size(h0_type);
// packing k
memcpy(ptr, &(data->k), sizeof(data->k));
ptr += sizeof(data->k);
// packing c
*((cmph_uint64 *)ptr) = (cmph_uint64)data->c;
ptr += sizeof(data->c);
// packing h1 type
CMPH_HASH h1_type = hash_get_type(data->h1[0]);
memcpy(ptr, &h1_type, sizeof(h1_type));
ptr += sizeof(h1_type);
// packing h2 type
CMPH_HASH h2_type = hash_get_type(data->h2[0]);
memcpy(ptr, &h2_type, sizeof(h2_type));
ptr += sizeof(h2_type);
// packing size
memcpy(ptr, data->size, sizeof(cmph_uint8)*data->k);
ptr += data->k;
// packing offset
memcpy(ptr, data->offset, sizeof(cmph_uint32)*data->k);
ptr += sizeof(cmph_uint32)*data->k;
#if defined (__ia64) || defined (__x86_64__)
cmph_uint64 * g_is_ptr = (cmph_uint64 *)ptr;
#else
cmph_uint32 * g_is_ptr = (cmph_uint32 *)ptr;
#endif
cmph_uint8 * g_i = (cmph_uint8 *) (g_is_ptr + data->k);
for(i = 0; i < data->k; i++)
{
#if defined (__ia64) || defined (__x86_64__)
*g_is_ptr++ = (cmph_uint64)g_i;
#else
*g_is_ptr++ = (cmph_uint32)g_i;
#endif
// packing h1[i]
hash_state_pack(data->h1[i], g_i);
g_i += hash_state_packed_size(h1_type);
// packing h2[i]
hash_state_pack(data->h2[i], g_i);
g_i += hash_state_packed_size(h2_type);
// packing g_i
switch(data->algo)
{
case CMPH_FCH:
n = fch_calc_b(data->c, data->size[i]);
break;
case CMPH_BMZ8:
n = (cmph_uint32)ceil(data->c * data->size[i]);
break;
default: assert(0);
}
memcpy(g_i, data->g[i], sizeof(cmph_uint8)*n);
g_i += n;
}
}