static int
thread_loop(ESL_THREADS *obj, ESL_WORK_QUEUE *queue, P7_HMMFILE *hfp)
{
int status = eslOK;
int sstatus = eslOK;
int eofCount = 0;
P7_OM_BLOCK *block;
ESL_ALPHABET *abc = NULL;
void *newBlock;
esl_workqueue_Reset(queue);
esl_threads_WaitForStart(obj);
status = esl_workqueue_ReaderUpdate(queue, NULL, &newBlock);
if (status != eslOK) esl_fatal("Work queue reader failed");
/* Main loop: */
while (sstatus == eslOK)
{
block = (P7_OM_BLOCK *) newBlock;
sstatus = p7_oprofile_ReadBlockMSV(hfp, &abc, block);
if (sstatus == eslEOF)
{
if (eofCount < esl_threads_GetWorkerCount(obj)) sstatus = eslOK;
++eofCount;
}
if (sstatus == eslOK)
{
status = esl_workqueue_ReaderUpdate(queue, block, &newBlock);
if (status != eslOK) esl_fatal("Work queue reader failed");
}
}
status = esl_workqueue_ReaderUpdate(queue, block, NULL);
if (status != eslOK) esl_fatal("Work queue reader failed");
if (sstatus == eslEOF)
{
/* wait for all the threads to complete */
esl_threads_WaitForFinish(obj);
esl_workqueue_Complete(queue);
}
esl_alphabet_Destroy(abc);
return sstatus;
}
static void
process_SearchCmd(HMMD_COMMAND *cmd, WORKER_ENV *env)
{
int i;
int cnt;
int limit;
int status;
int blk_size;
WORKER_INFO *info = NULL;
ESL_ALPHABET *abc;
ESL_STOPWATCH *w;
ESL_THREADS *threadObj = NULL;
pthread_mutex_t inx_mutex;
int current_index;
QUEUE_DATA *query = NULL;
time_t date;
char timestamp[32];
w = esl_stopwatch_Create();
abc = esl_alphabet_Create(eslAMINO);
if (pthread_mutex_init(&inx_mutex, NULL) != 0) p7_Fail("mutex init failed");
ESL_ALLOC(info, sizeof(*info) * env->ncpus);
/* Log the current time (at search start) */
date = time(NULL);
ctime_r(&date, timestamp);
printf("\n%s", timestamp); /* note that ctime_r() leaves \n on end of timestamp */
/* initialize thread data */
query = process_QueryCmd(cmd, env);
esl_stopwatch_Start(w);
info->range_list = NULL;
if (esl_opt_IsUsed(query->opts, "--seqdb_ranges")) {
ESL_ALLOC(info->range_list, sizeof(RANGE_LIST));
hmmpgmd_GetRanges(info->range_list, esl_opt_GetString(query->opts, "--seqdb_ranges"));
}
if (query->cmd_type == HMMD_CMD_SEARCH) threadObj = esl_threads_Create(&search_thread);
else threadObj = esl_threads_Create(&scan_thread);
if (query->query_type == HMMD_SEQUENCE) {
fprintf(stdout, "Search seq %s [L=%ld]", query->seq->name, (long) query->seq->n);
} else {
fprintf(stdout, "Search hmm %s [M=%d]", query->hmm->name, query->hmm->M);
}
fprintf(stdout, " vs %s DB %d [%d - %d]",
(query->cmd_type == HMMD_CMD_SEARCH) ? "SEQ" : "HMM",
query->dbx, query->inx, query->inx + query->cnt - 1);
if (info->range_list)
fprintf(stdout, " in range(s) %s", esl_opt_GetString(query->opts, "--seqdb_ranges"));
fprintf(stdout, "\n");
/* Create processing pipeline and hit list */
for (i = 0; i < env->ncpus; ++i) {
info[i].abc = query->abc;
info[i].hmm = query->hmm;
info[i].seq = query->seq;
info[i].opts = query->opts;
info[i].range_list = info[0].range_list;
info[i].th = NULL;
info[i].pli = NULL;
info[i].inx_mutex = &inx_mutex;
info[i].inx = ¤t_index;/* this is confusing trickery - to share a single variable across all threads */
info[i].blk_size = &blk_size; /* ditto */
info[i].limit = &limit; /* ditto. TODO: come back and clean this up. */
if (query->cmd_type == HMMD_CMD_SEARCH) {
HMMER_SEQ **list = env->seq_db->db[query->dbx].list;
info[i].sq_list = &list[query->inx];
info[i].sq_cnt = query->cnt;
info[i].db_Z = env->seq_db->db[query->dbx].K;
info[i].om_list = NULL;
info[i].om_cnt = 0;
} else {
info[i].sq_list = NULL;
info[i].sq_cnt = 0;
info[i].db_Z = 0;
info[i].om_list = &env->hmm_db->list[query->inx];
info[i].om_cnt = query->cnt;
}
esl_threads_AddThread(threadObj, &info[i]);
}
/* try block size of 5000. we will need enough sequences for four
* blocks per thread or better.
*/
blk_size = 5000;
cnt = query->cnt / env->ncpus / blk_size;
limit = query->cnt * 2 / 3;
if (cnt < 4) {
/* try block size of 1000 */
blk_size /= 5;
cnt = query->cnt / env->ncpus / blk_size;
if (cnt < 4) {
/* still not enough. just divide it up into one block per thread */
blk_size = query->cnt / env->ncpus + 1;
limit = query->cnt * 2;
}
}
current_index = 0;
esl_threads_WaitForStart(threadObj);
esl_threads_WaitForFinish(threadObj);
esl_stopwatch_Stop(w);
#if 1
fprintf (stdout, " Sequences Residues Elapsed\n");
for (i = 0; i < env->ncpus; ++i) {
print_timings(i, info[i].elapsed, info[i].pli);
}
#endif
/* merge the results of the search results */
for (i = 1; i < env->ncpus; ++i) {
p7_tophits_Merge(info[0].th, info[i].th);
p7_pipeline_Merge(info[0].pli, info[i].pli);
p7_pipeline_Destroy(info[i].pli);
p7_tophits_Destroy(info[i].th);
}
print_timings(99, w->elapsed, info[0].pli);
send_results(env->fd, w, info);
/* free the last of the pipeline data */
p7_pipeline_Destroy(info->pli);
p7_tophits_Destroy(info->th);
free_QueueData(query);
esl_threads_Destroy(threadObj);
pthread_mutex_destroy(&inx_mutex);
if (info->range_list) {
if (info->range_list->starts) free(info->range_list->starts);
if (info->range_list->ends) free(info->range_list->ends);
free (info->range_list);
}
free(info);
esl_stopwatch_Destroy(w);
esl_alphabet_Destroy(abc);
return;
ERROR:
LOG_FATAL_MSG("malloc", errno);
}
static int
thread_loop(ESL_THREADS *obj, ESL_WORK_QUEUE *queue, struct cfg_s *cfg)
{
int status = eslOK;
int sstatus = eslOK;
int processed = 0;
WORK_ITEM *item;
void *newItem;
int next = 1;
PENDING_ITEM *top = NULL;
PENDING_ITEM *empty = NULL;
PENDING_ITEM *tmp = NULL;
char errmsg[eslERRBUFSIZE];
esl_workqueue_Reset(queue);
esl_threads_WaitForStart(obj);
status = esl_workqueue_ReaderUpdate(queue, NULL, &newItem);
if (status != eslOK) esl_fatal("Work queue reader failed");
/* Main loop: */
item = (WORK_ITEM *) newItem;
while (sstatus == eslOK) {
sstatus = esl_msa_Read(cfg->afp, &item->msa);
if (sstatus == eslOK) {
item->nali = ++cfg->nali;
if (set_msa_name(cfg, errmsg, item->msa) != eslOK) p7_Fail("%s\n", errmsg);
}
if (sstatus == eslEOF && processed < cfg->nali) sstatus = eslOK;
if (sstatus == eslOK) {
status = esl_workqueue_ReaderUpdate(queue, item, &newItem);
if (status != eslOK) esl_fatal("Work queue reader failed");
/* process any results */
item = (WORK_ITEM *) newItem;
if (item->processed == TRUE) {
++processed;
/* try to keep the input output order the same */
if (item->nali == next) {
sstatus = output_result(cfg, errmsg, item->nali, item->msa, item->hmm, item->postmsa, item->entropy);
if (sstatus != eslOK) p7_Fail(errmsg);
p7_hmm_Destroy(item->hmm);
esl_msa_Destroy(item->msa);
esl_msa_Destroy(item->postmsa);
++next;
/* output any pending msa as long as the order
* remains the same as read in.
*/
while (top != NULL && top->nali == next) {
sstatus = output_result(cfg, errmsg, top->nali, top->msa, top->hmm, top->postmsa, top->entropy);
if (sstatus != eslOK) p7_Fail(errmsg);
p7_hmm_Destroy(top->hmm);
esl_msa_Destroy(top->msa);
esl_msa_Destroy(top->postmsa);
tmp = top;
top = tmp->next;
tmp->next = empty;
empty = tmp;
++next;
}
} else {
/* queue up the msa so the sequence order is the same in
* the .sto and .hmm
*/
if (empty != NULL) {
tmp = empty;
empty = tmp->next;
} else {
ESL_ALLOC(tmp, sizeof(PENDING_ITEM));
}
tmp->nali = item->nali;
tmp->hmm = item->hmm;
tmp->msa = item->msa;
tmp->postmsa = item->postmsa;
tmp->entropy = item->entropy;
/* add the msa to the pending list */
if (top == NULL || tmp->nali < top->nali) {
tmp->next = top;
top = tmp;
} else {
PENDING_ITEM *ptr = top;
while (ptr->next != NULL && tmp->nali > ptr->next->nali) {
ptr = ptr->next;
}
tmp->next = ptr->next;
ptr->next = tmp;
}
}
item->nali = 0;
item->processed = FALSE;
item->hmm = NULL;
item->msa = NULL;
item->postmsa = NULL;
item->entropy = 0.0;
}
}
}
if (top != NULL) esl_fatal("Top is not empty\n");
while (empty != NULL) {
tmp = empty;
empty = tmp->next;
free(tmp);
}
status = esl_workqueue_ReaderUpdate(queue, item, NULL);
if (status != eslOK) esl_fatal("Work queue reader failed");
if (sstatus == eslEOF)
{
/* wait for all the threads to complete */
esl_threads_WaitForFinish(obj);
esl_workqueue_Complete(queue);
}
return sstatus;
ERROR:
return eslEMEM;
}
int
main(void)
{
int i;
int ncpu = 4;
int iter = 25;
WORK_INFO *worker = NULL;
ESL_THREADS *thr = NULL;
ESL_WORK_QUEUE *queue = NULL;
int *objs = NULL;
int *obj;
objs = malloc(sizeof(int) * ncpu * 2);
worker = malloc(sizeof(WORK_INFO) * ncpu);
thr = esl_threads_Create(&worker_thread);
/* Create a work queue that is able to hold two items per thread.
* The idea is that while one object is being processed by a
* worker thread, another item is being readied. So, when the
* worker thread has completed processing its current object,
* its next object to processes is hopefully waiting.
*/
queue = esl_workqueue_Create(ncpu * 2);
for (i = 0; i < ncpu * 2; i++)
{
objs[i] = 0;
esl_workqueue_Init(queue, &objs[i]);
}
for (i = 0; i < ncpu; i++)
{
worker[i].id = 'A' + i;
worker[i].queue = queue;
esl_threads_AddThread(thr, (void *) &worker[i]);
}
esl_threads_WaitForStart (thr);
/* For N number of iterations, get an object that has been
* processed, i.e. on the readers input queue and place it
* on the ready queue.
*/
esl_workqueue_ReaderUpdate(queue, NULL, (void **) &obj);
for (i = 1; i <= iter; ++i)
{
*obj = i;
printf("Item %d is ready to be processed\n", *obj);
esl_workqueue_ReaderUpdate(queue, obj, (void **) &obj);
}
/* put zeros on the queues to signal the worker that we are done */
for (i = 0; i < ncpu; ++i)
{
*obj = 0;
esl_workqueue_ReaderUpdate(queue, obj, (void **) &obj);
}
/* The worker threads now run their work. */
esl_threads_WaitForFinish(thr);
esl_threads_Destroy(thr);
esl_workqueue_Destroy(queue);
free(worker);
free(objs);
return eslOK;
}
