/**
* Shortest Path solution implemented using the Dijkstra algorithm
*/
int dijkstra(Graph *graph, int source, pri_queue priq,
Energy *distance, int previous[], Energy shortest_distance)
{
int vertex, adjacent, v;
Energy alt;
int *visited = calloc(graph->list_size, sizeof(int));
for(v = 0; v < graph->list_size; distance[v++] = ENERGY_MAX);
previous[source] = NONE;
distance[source] = graph->vertexes[source].pixel->energy;
visited[source] = 1;
priq_purge(priq);
priq_push(priq, source, distance[source]);
int dest = NONE;
while(priq_size(priq))
{
vertex = priq_pop(priq);
v = 0;
adjacent = graph->vertexes[vertex].adj[v++];
if(adjacent == NONE)
{
dest = vertex;
break;
}
#ifdef OPT_GRAPH_SHORTEST_PATH_BREAK
else
if (distance[vertex] >= shortest_distance)
{
dest = BREAK;
break;
}
#endif
do
{
if(adjacent >= 0 && !visited[adjacent])
{
alt = distance[vertex] +
graph->vertexes[adjacent].pixel->energy;
if(alt < distance[adjacent])
{
distance[adjacent] = alt;
previous[adjacent] = vertex;
priq_push(priq, adjacent, alt);
}
}
adjacent = graph->vertexes[vertex].adj[v++];
} while(adjacent != NONE);
visited[vertex] = 1;
}
free(visited);
return dest;
}
/* this is O(n log n), but probably not the best */
void priq_combine(pri_queue q, pri_queue q2)
{
int i;
q_elem_t *e = q2->buf + 1;
for (i = q2->n - 1; i >= 1; i--, e++)
priq_push(q, e->data, e->pri);
priq_purge(q2);
}
int main()
{
int i, p;
const char *c, *tasks[] ={
"Clear drains", "Feed cat", "Make tea", "Solve RC tasks", "Tax return" };
int pri[] = { 3, 4, 5, 1, 2 };
/* make two queues */
pri_queue q = priq_new(0), q2 = priq_new(0);
/* push all 5 tasks into q */
for (i = 0; i < 5; i++)
priq_push(q, tasks[i], pri[i]);
/* pop them and print one by one */
while ((c = priq_pop(q, &p)))
printf("%d: %s\n", p, c);
/* put a million random tasks in each queue */
for (i = 0; i < 1 << 20; i++) {
p = rand() / ( RAND_MAX / 5 );
priq_push(q, tasks[p], pri[p]);
p = rand() / ( RAND_MAX / 5 );
priq_push(q2, tasks[p], pri[p]);
}
printf("\nq has %d items, q2 has %d items\n", priq_size(q), priq_size(q2));
/* merge q2 into q; q2 is empty */
priq_combine(q, q2);
printf("After merge, q has %d items, q2 has %d items\n",
priq_size(q), priq_size(q2));
/* pop q until it's empty */
for (i = 0; (c = priq_pop(q, 0)); i++);
printf("Popped %d items out of q\n", i);
return 0;
}
int main(int argc, char **argv)
{
if (argc < 3) {
fprintf(stderr,"%s <bcf file> <num vars>\n", argv[0]);
return 1;
}
char *fname = argv[1];
uint32_t num_vars = atoi(argv[2]);
htsFile *fp = hts_open(fname,"rb");
bcf_hdr_t *hdr = bcf_hdr_read(fp);
bcf1_t *line = bcf_init1();
int32_t *gt_p = NULL;
uint32_t num_inds = bcf_hdr_nsamples(hdr);
int32_t i, j, k, ntmp = 0, int_i = 0, two_bit_i = 0, sum, t_sum = 0;
uint32_t num_ind_ints = 1 + ((num_inds - 1) / 16);
pri_queue q = priq_new(0);
priority p;
uint32_t *packed_ints = (uint32_t *) calloc(num_ind_ints,
sizeof(uint32_t));
FILE *gt_of = fopen("gt.tmp.packed","wb");
FILE *md_of = fopen("md.tmp.packed","w");
uint32_t *md_index = (uint32_t *) malloc(num_vars * sizeof(uint32_t));
uint32_t md_i = 0;
unsigned long t_bcf_read = 0,
t_bcf_dup = 0,
t_bcf_unpack = 0,
t_bcf_get_genotypes = 0,
t_bcf_hdr_nsamples = 0,
t_q = 0,
t_write = 0,
t_get_md = 0,
t_md_write = 0,
t_pack = 0;
for (i = 0; i < num_vars; ++i) {
sum = 0;
int_i = 0;
two_bit_i = 0;
int r = bcf_read(fp, hdr, line);
// Copy
bcf1_t *t_line = bcf_dup(line);
// Unpack
bcf_unpack(t_line, BCF_UN_ALL);
// Get metadata
size_t len = strlen(bcf_hdr_id2name(hdr, t_line->rid)) +
10 + // max length of pos
strlen(t_line->d.id) +
strlen(t_line->d.allele[0]) +
strlen(t_line->d.allele[1]) +
4; //tabs
char *md = (char *) malloc(len * sizeof(char));
sprintf(md, "%s\t%d\t%s\t%s\t%s",
bcf_hdr_id2name(hdr, t_line->rid),
t_line->pos + 1,
t_line->d.id,
t_line->d.allele[0],
t_line->d.allele[1]);
// Write metadata
md_i += strlen(md);
md_index[i] = md_i;
fprintf(md_of, "%s", md);
// Get gentotypes
uint32_t num_gts_per_sample = bcf_get_genotypes(hdr,
t_line,
>_p,
&ntmp);
num_gts_per_sample /= num_inds;
int32_t *gt_i = gt_p;
// Pack genotypes
for (j = 0; j < num_inds; ++j) {
uint32_t gt = 0;
for (k = 0; k < num_gts_per_sample; ++k) {
gt += bcf_gt_allele(gt_i[k]);
}
packed_ints[int_i] += gt << (30 - 2*two_bit_i);
two_bit_i += 1;
if (two_bit_i == 16) {
two_bit_i = 0;
int_i += 1;
}
sum += gt;
gt_i += num_gts_per_sample;
}
// Get a priority for the variant based on the sum and number of
// leading zeros
p.sum = sum;
uint32_t prefix_len = 0;
j = 0;
while ((j < num_ind_ints) && (packed_ints[j] == 0)){
prefix_len += 32;
j += 1;
}
if (j < num_ind_ints)
prefix_len += nlz1(packed_ints[j]);
// Push it into the q
p.len = prefix_len;
int *j = (int *) malloc (sizeof(int));
j[0] = i;
priq_push(q, j, p);
// Write to file
fwrite(packed_ints, sizeof(uint32_t), num_ind_ints,gt_of);
memset(packed_ints, 0, num_ind_ints*sizeof(uint32_t));
t_sum += sum;
bcf_destroy(t_line);
free(md);
}
fclose(gt_of);
fclose(md_of);
md_of = fopen("md.tmp.packed","r");
FILE *md_out = fopen("md.bim","w");
gt_of = fopen("gt.tmp.packed","rb");
FILE *s_gt_of = fopen("s.gt.tmp.packed","wb");
// Get variants in order and rewrite a variant-major sorted matrix
while ( priq_top(q, &p) != NULL ) {
int *d = priq_pop(q, &p);
uint32_t start = 0;
if (*d != 0)
start = md_index[*d - 1];
uint32_t len = md_index[*d] - start;
fseek(md_of, start*sizeof(char), SEEK_SET);
char buf[len+1];
fread(buf, sizeof(char), len, md_of);
buf[len] = '\0';
fseek(gt_of, (*d)*num_ind_ints*sizeof(uint32_t), SEEK_SET);
fread(packed_ints, sizeof(uint32_t), num_ind_ints, gt_of);
fwrite(packed_ints, sizeof(uint32_t), num_ind_ints,s_gt_of);
fprintf(md_out, "%s\n", buf);
}
fclose(md_out);
fclose(md_of);
fclose(gt_of);
fclose(s_gt_of);
/*
* In a packed-int variant-major matrix there will be a num_vars
* number of rows, and a num_inds number of values packed into
* num_inds_ints number of intergers. For examples, 16 rows of 16 values
* will be 16 ints, where each int encodes 16 values.
*
*/
uint32_t num_var_ints = 1 + ((num_vars - 1) / 16);
uint32_t *I_data = (uint32_t *)calloc(num_var_ints*16,sizeof(uint32_t));
uint32_t **I = (uint32_t **)malloc(16*sizeof(uint32_t*));
for (i = 0; i < 16; ++i)
I[i] = I_data + i*num_var_ints;
uint32_t I_i = 0, I_int_i = 0;
uint32_t v;
s_gt_of = fopen("s.gt.tmp.packed","rb");
FILE *rs_gt_of = fopen("r.s.gt.tmp.packed","wb");
// Write these to values to that this is a well-formed uncompressed
// packed int binary file (ubin) file
fwrite(&num_vars, sizeof(uint32_t), 1, rs_gt_of);
fwrite(&num_inds, sizeof(uint32_t), 1, rs_gt_of);
/*
* we need to loop over the columns in the v-major file.
* There are num_vars rows, and num_ind_ints 16-ind packed columns
*
* In this loop :
* i: cols in var-major form, rows in ind-major form
* j: rows in var-major form, cols in ind-major form
*/
uint32_t num_inds_to_write = num_inds;
for (i = 0; i < num_ind_ints; ++i) { // loop over each int col
for (j = 0; j < num_vars; ++j) { // loop over head row in that col
// skip to the value at the row/col
fseek(s_gt_of,
j*num_ind_ints*sizeof(uint32_t) + //row
i*sizeof(uint32_t), //col
SEEK_SET);
fread(&v, sizeof(uint32_t), 1, s_gt_of);
// one int corresponds to a col of 16 two-bit values
// two_bit_i will move across the cols
for (two_bit_i = 0; two_bit_i < 16; ++two_bit_i) {
I[two_bit_i][I_i] += ((v >> (30 - 2*two_bit_i)) & 3) <<
(30 - 2*I_int_i);
}
I_int_i += 1;
if (I_int_i == 16) {
I_i += 1;
I_int_i = 0;
}
}
// When we are at the end of the file, and the number of lines
// is not a factor of 16, only write out the lines that contain values
if (num_inds_to_write >= 16) {
fwrite(I_data,
sizeof(uint32_t),
num_var_ints*16,
rs_gt_of);
num_inds_to_write -= 16;
} else {
fwrite(I_data,
sizeof(uint32_t),
num_var_ints*num_inds_to_write,
rs_gt_of);
}
memset(I_data, 0, num_var_ints*16*sizeof(uint32_t));
I_int_i = 0;
I_i = 0;
}
fclose(s_gt_of);
fclose(rs_gt_of);
free(md_index);
free(packed_ints);
}
