BitHandle
BitField::get_handle()
{
// find first free bit
// first find first free word
unsigned int word;
for (word = 0 ; word < f_num_words ; word++ )
if ( word_has_zero_bit(f_bits_in_use[word]) )
break;
if (word == f_num_words)
{
check_allocation (word);
#if 0
// have to allocate more
unsigned long *new_words = new unsigned long [f_num_words + 1] ;
unsigned long *new_use = new unsigned long [f_num_words + 1] ;
for (int i = 0 ; i < f_num_words ; i++ ){
// copy old info
new_words[i] = f_words[i] ;
new_use[i] = f_bits_in_use[i] ;
}
// zero out newest memory
new_words[f_num_words] = 0 ;
new_use[f_num_words] = 0 ;
delete f_words ;
f_words = new_words ;
delete f_bits_in_use ;
f_bits_in_use = new_use ;
f_num_words++ ;
#endif
return get_handle(); // NOTE: recursive and function exit
}
// now find first free byte in word
int byte;
for (byte = 0 ; byte < bytes_per_word ; byte++ )
if (byte_has_zero_bit(get_byte(byte, f_bits_in_use[word])))
break;
unsigned char the_byte = (unsigned char)
(f_bits_in_use[word] >> (byte * bits_per_byte)) & 0xFF ;
// now check byte for zero bit
int bit;
for (bit = 0 ; bit < bits_per_byte ; bit++ )
if ((the_byte & mask[bit]) == 0)
break ;
// mark the bit as in use
f_bits_in_use[word] |= 1 << ((byte * bits_per_byte) + bit) ;
return (word * bits_per_word) + (byte * bits_per_byte) + bit ;
}
int allocate_send_lists(int myrank, int *nghb_cnt, int** nghb_to_rank, int** send_cnt,
int*** send_lst, int **recv_cnt) {
MPI_Status status;
int nghb_idx=0;
// Allocate send_cnt
*send_cnt = (int*) calloc(sizeof(int), *nghb_cnt);
// Send size of receive list for that processor, it will be equal to the size of send list of
// received processor
for (nghb_idx=0; nghb_idx<(*nghb_cnt); ++nghb_idx) {
// Send sizes
MPI_Send(&(*recv_cnt)[nghb_idx], 1, MPI_INT, (*nghb_to_rank)[nghb_idx], myrank, MPI_COMM_WORLD);
// Receive sizes
MPI_Recv(&(*send_cnt)[nghb_idx],1 , MPI_INT, (*nghb_to_rank)[nghb_idx],
(*nghb_to_rank)[nghb_idx], MPI_COMM_WORLD, &status);
}
// Allocate send_lst with given sizes
*send_lst = (int**) malloc((*nghb_cnt)*sizeof(int*));
check_allocation(myrank, send_lst, "malloc of send_lst failed");
for (nghb_idx=0; nghb_idx<(*nghb_cnt); ++nghb_idx) {
(*send_lst)[nghb_idx] = (int *) malloc((*send_cnt)[nghb_idx]*sizeof(int));
}
return POSL_OK;
}
struct lsh_iterator*
lsh_initialize(int len1, int len2, int(*penalty_)(int, int),
int bandwidth, int *result)
{
struct lsh_iterator* ret;
assert (!busy);
busy = 1;
n1 = len1 + 1, n2 = len2 + 1;
/* this is ugly comunication between lsh_initialize_mcmp and lsh_initialize
* to check if we have mpenalty
*/
if (bandwidth < 0) {
bandwidth *= -1;
} else {
mpenalty = NULL;
mmpenalty = NULL;
}
penalty = penalty_;
K = bandwidth;
adjust_bandwidth();
double_K_plus_1 = 2*K + 1;
check_allocation();
edit_distance();
output();
if (result != NULL) {
*result = dp_read[n2 - n1 + K];
}
ret = (struct lsh_iterator*) malloc(sizeof(struct lsh_iterator));
ret->_row = n1 - 1;
ret->_col = n2 - n1 + K;
return ret;
}
void initiate_inode_list() {
int arr[512];
memset(arr,0 ,sizeof(arr));
int i;
uint max_inode_block = 1 + curr_superblock.isize;
for(i = 2; i <= max_inode_block; i++)
write_block(i, arr, sizeof(arr));
//int brr[512];
//read_block(20,brr,512);
uint inode_number = curr_superblock.isize * INODES_PER_BLOCK;
for(i = 2; i <= inode_number; i++) {
if(curr_superblock.ninode == MAX_SIZE)
break;
struct inode ino;
read_inode(i, &ino);
if(check_allocation(&ino) != 1) {
free_inode(i);
}
}
if(curr_superblock.ninode == 0){
fprintf(stderr,"Error: the inode blocks are full, allocation falure!");
exit(-1);
}
}
int allocate_lcc_elems_points(int read_key, int myrank, int nprocs, int *nintci, int *nintcf, int *nextci,
int ***lcc, int* points_count, int*** points, int** elems, int **local_global_index,
int points_count_g, int *int_cells_per_proc) {
int i=0;
MPI_Status status;
//starting index is same for all processors
*nintci = 0;
if (read_key == POSL_INIT_ONE_READ) {
if (myrank == 0) {
*points_count = points_count_g;
for (i=1; i<nprocs; i++) {
MPI_Send(&int_cells_per_proc[i], 1, MPI_INT, i, POSL_MPI_TAG_NINTCF, MPI_COMM_WORLD);
MPI_Send(&points_count_g, 1, MPI_INT, i, POSL_MPI_TAG_POINTS_COUNT, MPI_COMM_WORLD);
}
} else {
MPI_Recv(nintcf, 1, MPI_INT, 0, POSL_MPI_TAG_NINTCF, MPI_COMM_WORLD, &status);
MPI_Recv(points_count, 1, MPI_INT, 0, POSL_MPI_TAG_POINTS_COUNT, MPI_COMM_WORLD, &status);
//decrement the value, since our indexing starts from zero
*nextci = *nintcf;
--(*nintcf);
}
} else {
*points_count = points_count_g;
}
*lcc = (int**) malloc(((*nintcf)+1)*sizeof(int*));
check_allocation(myrank, lcc, "Failed to allocate first dimension of LCC");
for (i=0; i<(*nintcf)+1; i++) {
(*lcc)[i] = (int *) malloc(6*sizeof(int));
check_allocation(myrank, lcc+i, "malloc failed to allocate second dimension of lcc");
}
*elems = (int *) malloc(((*nintcf)+1)*8*sizeof(int));
check_allocation(myrank, elems, "Failed to allocate memory for elems");
*points = (int **) calloc(*points_count, sizeof(int*));
check_allocation(myrank, elems, "Failed to allocate memory for points 1st dimention");
for (i=0; i<*points_count; i++) {
(*points)[i] = (int *) calloc(3, sizeof(int));
check_allocation(myrank, points+i, "malloc() POINTS 2nd dim. failed");
}
return POSL_OK;
}
int fill_l2g(int read_key, int myrank, int nproc, int nintcf, int** local_global_index,
int ***local_global_index_g, int *partitioning_map, int nelems_g, int *int_cells_per_proc) {
int i=0, local_idx=0, current_proc=0;
*local_global_index = (int *) malloc(((nintcf)+1)*sizeof(int));
check_allocation(myrank, local_global_index, "Malloc for local_global_index failed");
for (i=0; i<nelems_g; ++i) {
if (partitioning_map[i] == myrank) {
(*local_global_index)[local_idx] = i;
++local_idx;
}
}
if (read_key == POSL_INIT_ONE_READ && myrank == 0) {
*local_global_index_g = (int**) malloc(nproc*sizeof(int*));
check_allocation(myrank, local_global_index_g, "malloc failed to allocate first dimension of l2g_g");
for (i=0; i<nproc; i++) {
(*local_global_index_g)[i] = (int*) malloc(int_cells_per_proc[i]*sizeof(int));
check_allocation(myrank, (local_global_index_g+i), "malloc failed to allocate 2nd dimension of l2g_g");
}
//initialize array of local indexes
int last_local_idx[nproc];
for (i=0; i<nproc; ++i) {
last_local_idx[i] = 0;
}
for (i=0; i<nelems_g; ++i) {
current_proc = partitioning_map[i];
local_idx = last_local_idx[current_proc];
if (local_idx > int_cells_per_proc[current_proc]) {
log_err("Wrong local index for processor #%d", current_proc);
return POSL_ERROR;
}
(*local_global_index_g)[current_proc][local_idx] = i;
++(last_local_idx[current_proc]);
}
}
return POSL_OK;
}
bool
BitField::i_is_set(BitHandle handle)
{
assert(handle);
unsigned long word = handle / bits_per_word ;
check_allocation (word);
unsigned long mask = 1 << (handle % bits_per_word) ;
return (f_words[word] & mask) != 0 ;
}
bool
BitField::i_unset(BitHandle handle)
{
assert(handle);
unsigned long word = handle / bits_per_word ;
check_allocation (word);
unsigned long mask = 1 << (handle % bits_per_word) ;
f_words[word] &= ~0L ^ mask ;
return 0 ;
}
bool
BitField::i_set(BitHandle handle)
{
assert(handle);
unsigned long word = handle / bits_per_word ;
check_allocation (word);
unsigned long mask = 1 << (handle % bits_per_word) ;
ON_DEBUG (printf ("BitField: setting bit %ld (in word %ld)\n", handle, word));
f_words[word] |= mask ;
return 1 ;
}
int partition(int part_key, int read_key, int myrank, int nprocs, int nintci_g,
int nintcf_g, int nextci_g, int nextcf_g, int *nintci, int *nintcf, int *nextci, int *nextcf,
int **lcc_g, int points_count_g, int **points_g, int *elems_g, int *int_cells_per_proc,
int *extcell_per_proc, int **local_global_index_g, int **local_global_index, int **partitioning_map) {
int i=0;
idx_t nelems, nnodes, ncommon=4, nparts, objval;
idx_t *elem_ptr, *elem_idx, *elem_part, *node_part;
nelems = nintcf_g-nintci_g+1;
*partitioning_map = (int *) calloc(sizeof(int), (nintcf_g-nintci_g+1));
check_allocation(myrank, partitioning_map, "Partitioning map allocation failed");
if (((read_key == POSL_INIT_ONE_READ) && (myrank == 0)) || (read_key == POSL_INIT_ALL_READ)) {
*nintci = 0; *nintcf = 0;
if (part_key == POSL_PARTITIONING_CLASSIC) {
//the last processor always gets different number of cells
int elem_per_proc = (nelems+(nprocs-1))/nprocs;
*nextci = (myrank == nprocs-1) ? nelems-(nprocs-1)*elem_per_proc : elem_per_proc;
*nintcf = *nextci-1;
//build global cell allocation
if (read_key == POSL_INIT_ONE_READ) {
for (i=0; i<(nprocs-1); ++i) {
int_cells_per_proc[i] = elem_per_proc;
}
int_cells_per_proc[nprocs-1] = nelems-(nprocs-1)*elem_per_proc;
}
for (i=0; i<nelems; ++i) {
(*partitioning_map)[i] = i/elem_per_proc;
}
} else {
//initialize variables for metis
nnodes = points_count_g;
nparts = nprocs;
elem_ptr = (idx_t *) calloc(nelems+1, sizeof(idx_t));
elem_idx = (idx_t *) calloc(nelems*8, sizeof(idx_t));
elem_part = (idx_t *) calloc(nelems, sizeof(idx_t));
node_part = (idx_t *) calloc(nnodes, sizeof(idx_t));
//assign arrays that store metis graph mesh
for (i=0; i<(nelems+1); i++) {
elem_ptr[i] = 8*i;
}
for (i=0; i<(nelems*8); i++) {
elem_idx[i] = elems_g[i];
}
//perform metis partitioning
if (part_key == POSL_PARTITIONING_DUAL) {
METIS_PartMeshDual(&nelems, &nnodes, elem_ptr, elem_idx, NULL, NULL, &ncommon,
&nparts, NULL, NULL, &objval, elem_part, node_part);
} else {
METIS_PartMeshNodal(&nelems, &nnodes, elem_ptr, elem_idx, NULL, NULL, &nparts,
NULL, NULL, &objval, elem_part, node_part);
}
for (i=0; i<nelems; i++) {
(*partitioning_map)[i] = (int) elem_part[i];
}
//initialize global cell counters
if (read_key == POSL_INIT_ONE_READ) {
for (i=0; i<nprocs; ++i) {
int_cells_per_proc[i] = 0;
}
}
//TODO: consider performance gains when if statement is outside of the loop
//compute position of last internal cell
for (i=0; i<nelems; i++) {
if (read_key == POSL_INIT_ONE_READ) {
int_cells_per_proc[(*partitioning_map)[i]] += 1;
} else {
if (myrank == (*partitioning_map)[i]) {
(*nintcf) += 1;
}
}
}
//assign local internal cell ending idx
if (read_key == POSL_INIT_ONE_READ) {
*nintcf = int_cells_per_proc[myrank];
}
*nextci = (*nintcf)--;
}
}
return POSL_OK;
}
int build_lists_g2l_next(int nprocs, int myrank, int *partitioning_map, int nintcf_g, int nextcf_g,
int* nintcf, int* nextcf, int*** lcc, int** local_global_index, int** global_local_index,
int *nghb_cnt, int** nghb_to_rank, int **recv_cnt, int*** recv_lst) {
/*********** Initialize and allocate g2l and tmp variables ************************************/
int proc=0, i=0, j=0, is_ghost_cell=0, idx_g=-1, neighbor_rank=-1;
int n_ghost_cells[nprocs], start_idx_per_proc[nprocs];
int nghb_idx=0;
memset(n_ghost_cells, 0, nprocs*sizeof(int));
memset(start_idx_per_proc, 0, nprocs*sizeof(int));
*nextcf = 0;
// Used to check if we already saved the cel index
int is_saved[nextcf_g+1];
/** Lists for neighboring information */
/// number of cells to be received from each neighbor (size: nprocs)
int tmp_recv_cnt[nprocs];
memset(tmp_recv_cnt, 0, nprocs*sizeof(int));
// TODO: try not to use tmp_recv_lst and save all data immediately in revc_lst
// lists of cells to be received from each neighbor (size: nprocs x recv_cnt[*])
int *tmp_recv_lst[nprocs];
*global_local_index = (int *) malloc((nextcf_g+1)*sizeof(int));
check_allocation(myrank, global_local_index, "malloc(global_local_index) failed");
// We will fill whole g2l by -1 for easier debugging
for (i=0; i<=nextcf_g; ++i) {
(*global_local_index)[i] = -1;
}
/*********** End initialize and allocate g2l and tmp variables ********************************/
// TODO: We can do all work in one big loop but then we need to use the buffer before we allocate
// memory for tmp_recv_lst
/************************ Start processing lcc and generating data ****************************/
memset(is_saved, 0, (nextcf_g+1)*sizeof(int));
// Count number of external cells and fill g2l with external cells, and fill tmp_recv_cnt
for (i=0; i<=(*nintcf); ++i) {
for (j=0; j<6; ++j) {
idx_g = (*lcc)[i][j];
if(idx_g<=nintcf_g) {
neighbor_rank = partitioning_map[idx_g];
is_ghost_cell = neighbor_rank != myrank;
}
if (!is_saved[idx_g]) {
if((*lcc)[i][j] > nintcf_g) {
++(*nextcf);
is_saved[idx_g] = 1;
(*global_local_index)[idx_g] = (*nintcf) + (*nextcf);
} else if (is_ghost_cell) {
++tmp_recv_cnt[neighbor_rank];
is_saved[idx_g] = 1;
}
}
}
// Fill g2l with internal cell
(*global_local_index)[(*local_global_index)[i]] = i;
}
/// After the loop nextcf shows amount of elements
*nextcf = *nintcf + *nextcf;
// Allocate tmp_recv_lst
for (proc=0; proc<nprocs; ++proc) {
if(tmp_recv_cnt[proc] == 0) {
tmp_recv_lst[proc] = NULL;
} else {
tmp_recv_lst[proc] = (int *) calloc(sizeof(int), tmp_recv_cnt[proc]);
}
}
// Fill tmp_recv_list and g2l with ghost cells
start_idx_per_proc[0]=0;
for(proc=1; proc<nprocs; ++proc) {
start_idx_per_proc[proc] = start_idx_per_proc[proc-1] + tmp_recv_cnt[proc-1];
}
memset(is_saved, 0, (nextcf_g+1)*sizeof(int));
for (i=0; i<=(*nintcf); ++i) {
for (j=0; j<6; ++j) {
idx_g = (*lcc)[i][j];
if(idx_g<=nintcf_g) {
neighbor_rank = partitioning_map[idx_g];
is_ghost_cell = neighbor_rank != myrank;
} else {
is_ghost_cell = 0;
}
if (is_ghost_cell && !is_saved[idx_g]) {
is_saved[idx_g] = 1;
tmp_recv_lst[neighbor_rank][n_ghost_cells[neighbor_rank]] = idx_g;
++n_ghost_cells[neighbor_rank];
(*global_local_index)[idx_g] = (*nextcf) +
start_idx_per_proc[neighbor_rank] + n_ghost_cells[neighbor_rank];
}
}
}
// Count neighbors it is already set to zero in gccg.c
for (proc=0; proc<nprocs; ++proc) {
if (tmp_recv_cnt[proc] !=0) ++(*nghb_cnt);
}
// Allocate and fill nhb_to_rank
*nghb_to_rank = (int *) malloc((*nghb_cnt)*sizeof(int));
check_allocation(myrank, nghb_to_rank, "malloc(nghb_to_rank) failed");
for (proc=0;proc<nprocs; ++proc) {
if(tmp_recv_cnt[proc] !=0) {
(*nghb_to_rank)[nghb_idx] = proc;
++nghb_idx;
}
}
// Allocate and fill recv_cnt and recv_lst with global lcc indexing for further use
*recv_cnt = (int*) calloc(sizeof(int), *nghb_cnt);
for (nghb_idx=0; nghb_idx<(*nghb_cnt); ++nghb_idx) {
(*recv_cnt)[nghb_idx] = tmp_recv_cnt[ (*nghb_to_rank)[nghb_idx] ];
}
*recv_lst = (int**) malloc( (*nghb_cnt)*sizeof(int*) );
for (nghb_idx=0; nghb_idx<(*nghb_cnt); ++nghb_idx) {
(*recv_lst)[nghb_idx] = (int *) malloc( (*recv_cnt)[nghb_idx] * sizeof(int) );
}
for (nghb_idx=0; nghb_idx<(*nghb_cnt); ++nghb_idx) {
memcpy( (*recv_lst)[nghb_idx], tmp_recv_lst[ (*nghb_to_rank)[nghb_idx] ],
(*recv_cnt)[nghb_idx]*sizeof(int) );
}
// Total number of external cells
for (nghb_idx=0; nghb_idx<(*nghb_cnt); ++nghb_idx) {
*nextcf += (*recv_cnt)[nghb_idx];
}
/************************ End processing lcc and generating data ******************************/
// Free memory
for (proc=0; proc<nprocs; ++proc) {
free(tmp_recv_lst[proc]);
}
return POSL_OK;
}