/* Add a new entry to the registry */
int registry_add(char *name,int object_type,void *data)
{
registry_entry_t *entry;
if (!name)
return(-1);
REGISTRY_LOCK();
/* check if we have already a reference for this name */
if ((entry = registry_find_entry(name,object_type))) {
REGISTRY_UNLOCK();
return(-1);
}
/* create a new entry */
if (!(entry = mp_alloc(®istry->mp,sizeof(*entry)))) {
REGISTRY_UNLOCK();
return(-1);
}
entry->name = name;
entry->data = data;
entry->object_type = object_type;
entry->ref_count = 1; /* consider object is referenced by the caller */
registry_insert_entry(entry);
REGISTRY_UNLOCK();
return(0);
}
int get_chunk_body_buffer(struct worker_context *ctx, struct connection *conn, char **p_start, int *p_len)
{
int flag = 0;
struct body_buffer *buffer;
if ( DLIST_EMPTY(&conn->_client._body) )
{
flag = 1;
}
else
{
__dlist_t *cur = DLIST_PRE(&conn->_client._body);
buffer = GET_OWNER(cur, struct body_buffer, _list);
if ( buffer->_data_len - buffer->_curpos == 0 )
{
flag = 1;
}
}
if ( flag )
{
buffer = (struct body_buffer *)mp_alloc(ctx->_body_pool);
if ( NULL == buffer )
{
WARNING("failed to alloc mem for body buffer.");
return -1;
}
DLIST_INIT(&buffer->_list);
buffer->_curpos = 0;
buffer->_data_len = ctx->_data_len;
DLIST_INSERT_B(&buffer->_list, &conn->_client._body);
}
*p_start = buffer->_data + buffer->_curpos;
*p_len = buffer->_data_len - buffer->_curpos;
return 0;
}
void func_worker_1(void* arg) {
// Cast and load the parameter
int worker_1_param = *((int*)arg);
// Create the queuing ports
spd_t * chan = mp_create_sport(MP_CHAN_1_ID, SOURCE, MP_CHAN_1_MSG_SIZE);
if (chan == NULL) {
DEBUGF(chan);
abort();
}
volatile unsigned long long _SPM * time_sample = mp_alloc(MP_CHAN_1_MSG_SIZE);
// Initialize the communication channels
int retval = mp_init_ports();
// TODO: check on retval
slave= 1;
for (unsigned long long start = get_cpu_usecs(); start + RUNTIME > get_cpu_usecs(); ) {
*time_sample = get_cpu_usecs();
mp_write(chan,time_sample);
}
int ret = 0;
corethread_exit(&ret);
return;
}
int bam_plp_push(bam_plp_t iter, const bam1_t *b)
{
if (iter->error) return -1;
if (b) {
if (b->core.tid < 0) return 0;
if (b->core.flag & iter->flag_mask) return 0;
if (iter->tid == b->core.tid && iter->pos == b->core.pos && iter->mp->cnt > iter->maxcnt) return 0;
bam_copy1(&iter->tail->b, b);
iter->tail->beg = b->core.pos; iter->tail->end = bam_calend(&b->core, bam1_cigar(b));
iter->tail->s = g_cstate_null; iter->tail->s.end = iter->tail->end - 1; // initialize cstate_t
if (b->core.tid < iter->max_tid) {
fprintf(stderr, "[bam_pileup_core] the input is not sorted (chromosomes out of order)\n");
iter->error = 1;
return -1;
}
if ((b->core.tid == iter->max_tid) && (iter->tail->beg < iter->max_pos)) {
fprintf(stderr, "[bam_pileup_core] the input is not sorted (reads out of order)\n");
iter->error = 1;
return -1;
}
iter->max_tid = b->core.tid; iter->max_pos = iter->tail->beg;
if (iter->tail->end > iter->pos || iter->tail->b.core.tid > iter->tid) {
iter->tail->next = mp_alloc(iter->mp);
iter->tail = iter->tail->next;
}
} else iter->is_eof = 1;
return 0;
}
bam_plp_t bam_plp_init(bam_plp_auto_f func, void *data)
{
bam_plp_t iter;
iter = calloc(1, sizeof(struct __bam_plp_t));
iter->mp = mp_init();
iter->head = iter->tail = mp_alloc(iter->mp);
iter->dummy = mp_alloc(iter->mp);
iter->max_tid = iter->max_pos = -1;
iter->flag_mask = BAM_DEF_MASK;
iter->maxcnt = 8000;
if (func) {
iter->func = func;
iter->data = data;
iter->b = bam_init1();
}
return iter;
}
LOCK_T * initialize_lock(unsigned remote) {
LOCK_T * lock = (LOCK_T *)mp_alloc(sizeof(LOCK_T));
lock->remote_entering = 0;
lock->remote_number = 0;
lock->local_entering = 0;
lock->local_number = 0;
lock->remote_cpuid = remote;
return lock;
}
struct sound_params *sg_create_sound_params(int type, void *params)
{
if( !sound_params_mem )
sound_params_mem = mp_create(sizeof(struct sound_params), 2);
struct sound_params * sc = mp_alloc(sound_params_mem);
sg_set_params(sc, type, params);
return sc;
}
void func_worker_1(void* arg) {
mp_init();
spd_t * sport2 = mp_create_sport(CHAN_ID_TWO,SINK,SAMPLE_SIZE*sizeof(short));
spd_t * sport1 = mp_create_sport(CHAN_ID_ONE,SOURCE,SAMPLE_SIZE*sizeof(short));
if (sport1 == NULL || sport2 == NULL) {
//exit(1);
}
volatile short _SPM * sample = mp_alloc(SAMPLE_SIZE*sizeof(short));
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[i] = i;
}
mp_init_ports();
while(done == 0);
for (int i = 0; i < ITERATIONS; ++i) {
int ret = mp_read(sport2,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
if(sample[i] != i) {
break;
}
}
for (int i = 0; i < 100000; ++i)
{
asm volatile (""::);
}
}
//for (int i = 0; i < ITERATIONS/2; ++i) {
for (int i = 0; i < ITERATIONS*20; ++i) {
mp_write(sport1,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[i] = i;
}
//for (int i = 0; i < 100000; ++i) {
// asm volatile (""::);
//}
mp_write(sport1,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[SAMPLE_SIZE-1-i] = i;
}
//for (int i = 0; i < 100000; ++i) {
// asm volatile (""::);
//}
}
int ret = 0;
corethread_exit(&ret);
return;
}
struct odes *
obj_new(struct mempool *pool)
{
struct odes *o = mp_alloc(pool, sizeof(struct odes));
o->pool = pool;
o->attrs = NULL;
o->cached_attr = NULL;
o->parent = NULL;
return o;
}
static struct oattr *
oa_new_ref(struct odes *o, uns x, byte *v)
{
struct oattr *a = mp_alloc(o->pool, sizeof(struct oattr));
a->next = a->same = NULL;
a->attr = x;
a->val = v;
return a;
}
/* Initialize registry */
int registry_init(void)
{
registry_entry_t *p;
pthread_mutexattr_t attr;
size_t len;
int i;
registry = malloc(sizeof(*registry));
assert(registry != NULL);
pthread_mutexattr_init(&attr);
pthread_mutexattr_settype(&attr,PTHREAD_MUTEX_RECURSIVE);
pthread_mutex_init(®istry->lock,&attr);
/* initialize registry memory pool */
mp_create_fixed_pool(®istry->mp,"registry");
registry->ht_name_entries = REGISTRY_HT_NAME_ENTRIES;
registry->ht_type_entries = REGISTRY_MAX_TYPES;
/* initialize hash table for names, with sentinels */
len = registry->ht_name_entries * sizeof(registry_entry_t);
registry->ht_names = mp_alloc(®istry->mp,len);
assert(registry->ht_names != NULL);
for(i=0;i<registry->ht_name_entries;i++) {
p = ®istry->ht_names[i];
p->hname_next = p->hname_prev = p;
}
/* initialize hash table for types, with sentinels */
len = registry->ht_type_entries * sizeof(registry_entry_t);
registry->ht_types = mp_alloc(®istry->mp,len);
assert(registry->ht_types != NULL);
for(i=0;i<registry->ht_type_entries;i++) {
p = ®istry->ht_types[i];
p->htype_next = p->htype_prev = p;
}
return(0);
}
static struct oattr *
oa_new_son(struct odes *o, uns x, struct odes *son)
{
struct oattr *a = mp_alloc(o->pool, sizeof(struct oattr));
a->next = a->same = NULL;
a->attr = x;
a->son = son;
son->parent = o;
return a;
}
void
mp_basket_add (mem_pool_t * mp, basket_t * bsk, void *token)
{
basket_t *newn = (basket_t *) mp_alloc (mp, sizeof (basket_t));
if (bsk->bsk_count == 0)
basket_init (bsk);
newn->bsk_pointer = token;
LISTPUTBEFORE (bsk, newn, bsk_next, bsk_prev);
bsk->bsk_count++;
}
bam_lplbuf_t *bam_lplbuf_init(bam_pileup_f func, void *data)
{
bam_lplbuf_t *tv;
tv = (bam_lplbuf_t*)calloc(1, sizeof(bam_lplbuf_t));
tv->mp = mp_init();
tv->head = tv->tail = mp_alloc(tv->mp);
tv->func = func;
tv->user_data = data;
tv->plbuf = bam_plbuf_init(tview_func, tv);
return (bam_lplbuf_t*)tv;
}
static struct oattr *
oa_new(struct odes *o, uns x, byte *v)
{
uns l = strlen(v)+1;
struct oattr *a = mp_alloc(o->pool, sizeof(struct oattr) + l);
a->next = a->same = NULL;
a->attr = x;
a->val = (byte*) (a+1);
memcpy(a->val, v, l);
return a;
}
inline void *
mp_calloc(size_t number, size_t size)
{
void *ptr;
size_t num_size;
num_size = number * size;
ptr = mp_alloc(num_size);
if (ptr != NULL)
memset(ptr, 0, num_size);
return ptr;
}
int main(int argc, char *argv[])
{
int *h_mem = NULL;
int cnt = 0;
struct mp_mem_head_t *p_mem_head = NULL;
struct mp_node_t *pp_node_array[NEW_NODE_NUM];
struct mp_node_t *p_node = NULL;
p_mem_head = mp_alloc(&h_mem, ALLOC_NODE_NUM);
for (cnt = 0; cnt < NEW_NODE_NUM; cnt++) {
pp_node_array[cnt] = mp_new_node(h_mem);
if (pp_node_array[cnt] == NULL) {
mp_clean(h_mem);
return -1;
}
}
for (cnt = 0; cnt < NEW_NODE_NUM; cnt += 3) {
mp_del_node(h_mem, pp_node_array[cnt]);
}
mp_del_node(h_mem, pp_node_array[NEW_NODE_NUM - 1]);
for (cnt = 0; cnt < NEW_NODE_NUM; cnt++) {
pp_node_array[cnt] = mp_new_node(h_mem);
}
// pp_node_array[3] = mp_new_node(h_mem);
// pp_node_array[6] = mp_new_node(h_mem);
// pp_node_array[9] = mp_new_node(h_mem);
// pp_node_array[12] = mp_new_node(h_mem);
// pp_node_array[15] = mp_new_node(h_mem);
p_node = (struct mp_node_t *)(p_mem_head + 1);
for (cnt = 0; cnt < ALLOC_NODE_NUM; cnt++, p_node++) {
printf("node: %p\tprev: %p\tnext: %p\tused: %d\n",
p_node, p_node->p_prev, p_node->p_next, p_node->used);
}
#if 0
for (cnt = 0; cnt < NEW_NODE_NUM; cnt++) {
if (pp_node_array[cnt] != NULL) {
printf("p_node: %p, p_node prev: %p, p_node next: %p\n",
pp_node_array[cnt],
pp_node_array[cnt]->p_prev,
pp_node_array[cnt]->p_next);
}
}
#endif
printf("------------- memory pool info ------------\n");
printf("first %p, tail %p, used: %d\n",
p_mem_head->p_node_first,
p_mem_head->p_node_tail,
p_mem_head->used_node_num);
mp_clean(h_mem);
return 0;
}
struct triangle_parameters *sg_create_triangle_parameters(float frequency,
float sample_rate)
{
if( !triangle_parameter_mp )
triangle_parameter_mp = mp_create(sizeof(struct triangle_parameters), 16);
struct triangle_parameters *p = mp_alloc(triangle_parameter_mp);
p->frequency = frequency;
p->ticks_per_rev = sample_rate/frequency;
p->counter = 0;
p->sample_rate = sample_rate;
p->current_value = 0;
p->slope = 1 / (p->ticks_per_rev/4.0);
return p;
}
struct sine_parameters *sg_create_sine_parameters(float frequency,
float sample_rate)
{
if( !sine_parameter_mp )
sine_parameter_mp = mp_create(sizeof(struct sine_parameters), 16);
struct sine_parameters *p = mp_alloc(sine_parameter_mp);
p->frequency = frequency;
p->ticks_per_rev = sample_rate/frequency;
p->rev = (sample_rate/frequency) / (2*PI);
p->counter = 0;
p->sample_rate = sample_rate;
p->current_value = 0;
return p;
}
struct sawtooth_parameters *sg_create_sawtooth_parameters(float frequency,
float sample_rate)
{
if( !sawtooth_parameter_mp )
sawtooth_parameter_mp = mp_create(sizeof(struct sawtooth_parameters), 16);
struct sawtooth_parameters *p = mp_alloc(sawtooth_parameter_mp);
p->frequency = frequency;
p->ticks_per_rev = sample_rate/frequency;
p->counter = 0;
p->sample_rate = sample_rate;
p->current_value = -1;
p->slope = 2.0 / p->ticks_per_rev;
return p;
}
static void
load(const char *filename, uint flags, void *img_data, uint img_size, uintptr_t group, TextureLoaded sync_cb, void *cb_data)
{
assert(sync_cb != NULL);
SDL_mutexP(storage_mutex);
{
/* See if task for this filename already exists. */
Task *task;
HASH_FIND_STR(task_hash, filename, task);
if (task == NULL) {
task = mp_alloc(&mp_tasks);
/* Set task filename and add to hash. */
assert(*filename != '\0' && strlen(filename) < sizeof(task->filename));
strcpy(task->filename, filename);
HASH_ADD_STR(task_hash, filename, task);
/* Set source buffer and size. */
task->img_data = img_data;
task->img_size = img_size;
if (texture_is_loaded(filename, flags)) {
/* Already loaded: add to finished_tasks. */
DL_PREPEND(finished_tasks, task);
num_finished++;
} else {
/*
* Insert into active list and signal task
* processing thread.
*/
task->active = 1;
DL_PREPEND(active_tasks, task);
SDL_CondSignal(checktask_cond);
}
} else if (task->active) {
/* Move to the front of active task queue. */
DL_DELETE(active_tasks, task);
DL_PREPEND(active_tasks, task);
}
/* Set/change group, flags, callback and its data pointer. */
task->group = (group != 0) ? group : (uintptr_t)sync_cb;
task->sync_cb = sync_cb;
task->cb_data = cb_data;
task->flags = flags;
}
SDL_mutexV(storage_mutex);
}
struct square_wave_parameters *sg_create_square_wave_parameters(float frequency,
float duty_cycle,
float sample_rate)
{
if( !square_wave_parameter_mem )
square_wave_parameter_mem = mp_create(sizeof(struct square_wave_parameters), 16);
struct square_wave_parameters *p = mp_alloc(square_wave_parameter_mem);
p->frequency = frequency;
p->ticks_per_rev = sample_rate/frequency;
p->counter = 0;
p->sample_rate = sample_rate;
p->current_value = 0;
p->duty_cycle = duty_cycle;
return p;
}
/*
* @brief initialize system monitor
* @param none
* @return status
* @note only monitor daemon can call this function
*/
HRESULT monitor_init(VOID)
{
HRESULT ret = E_FAIL;
struct mp_mem_head_t *p_mem_head = NULL;
M_DAEMON_PRINT(LOG_DEBUG, "enter\n");
p_mem_head = (struct mp_mem_head_t *)h_mem;
cnt_init++;
if (cnt_init != 1) {
M_DAEMON_PRINT(LOG_INFO,
"warning: daemon has already been initialized\n");
return S_OK;
}
#ifndef __BIONIC__
shmid = mp_shm_alloc(&h_mem, PUBLIC_SHM_KEY, MODULE_NUMBER_MAX);
if (h_mem == NULL)
return E_FAIL;
mp_shm_init(h_mem, MODULE_NUMBER_MAX);
M_DAEMON_PRINT(LOG_WARN, "shmid: %d, h_mem: %p\n", shmid, h_mem);
semid = mp_sem_create(PUBLIC_SEM_KEY);
M_DAEMON_PRINT(LOG_WARN, "semid: %d\n", semid);
M_DAEMON_ASSERT(semid != -1);
#else
mp_alloc(&h_mem, MODULE_NUMBER_MAX);
#endif
ret = M_OSAL_Init();
M_DAEMON_ASSERT(ret == S_OK);
ret = M_OSAL_Task_Create_Advanced(&h_daemon, daemon_thread, NULL,
M_OSAL_TASK_PRIORITY_DEFAULT, MSG_NUMBER_MAX);
M_DAEMON_ASSERT(ret == S_OK);
M_DAEMON_PRINT(LOG_DEBUG, "leave, shmid: %d, h_mem: %p\n",
shmid, h_mem);
return ret;
}
void *mp_memdup(struct mempool *mp, const void *buf, unsigned len) {
void *b = mp_alloc(mp, len, 0);
if (b)
memcpy(b, buf, len);
return b;
}
int
main(int argc, char **argv)
/*
* Initial main driver for GOMA. Derived from a (1/93) release of
* the rf_salsa program by
*
* Original Authors: John Shadid (1421)
* Scott Hutchinson (1421)
* Harry Moffat (1421)
*
* Date: 12/3/92
*
*
* Updates and Changes by:
* Randy Schunk (9111)
* P. A. Sackinger (9111)
* R. R. Rao (9111)
* R. A. Cairncross (Univ. of Delaware)
* Dates: 2/93 - 6/96
*
* Modified for continuation
* Ian Gates
* Dates: 2/98 - 10/98
* Dates: 7/99 - 8/99
*
* Last modified: Wed June 26 14:21:35 MST 1994 [email protected]
* Hello.
*
* Note: Many modifications from an early 2/93 pre-release
* version of rf_salsa were made by various persons
* in order to test ideas about moving/deforming meshes...
*/
{
/* Local Declarations */
double time_start, total_time; /* timing variables */
#ifndef PARALLEL
/* struct tm *tm_ptr; additional serial timing variables */
time_t now;
#endif
int error;
int i;
int j;
char **ptmp;
char *yo;
struct Command_line_command **clc=NULL; /* point to command line structure */
int nclc = 0; /* number of command line commands */
/********************** BEGIN EXECUTION ***************************************/
#ifdef FP_EXCEPT
feenableexcept ((FE_OVERFLOW | FE_DIVBYZERO | FE_INVALID));
#endif
/* assume number of commands is less than or equal to the number of
* arguments in the command line minus 1 (1st is program name) */
/*
* Get the name of the executable, yo
*/
yo = argv[0];
#ifdef PARALLEL
MPI_Init(&argc, &argv);
time_start = MPI_Wtime();
#endif /* PARALLEL */
#ifndef PARALLEL
(void)time(&now);
time_start = (double)now;
#endif /* PARALLEL */
time_goma_started = time_start;
Argv = argv;
Argc = argc;
#ifdef PARALLEL
/*
* Determine the parallel processing status, if any. We need to know
* pretty early if we're "one of many" or the only process.
*/
error = MPI_Comm_size(MPI_COMM_WORLD, &Num_Proc);
error = MPI_Comm_rank(MPI_COMM_WORLD, &ProcID);
/*
* Setup a default Proc_config so we can use utility routines
* from Aztec
*/
AZ_set_proc_config(Proc_Config, MPI_COMM_WORLD);
/* set the output limit flag if need be */
if( Num_Proc > DP_PROC_PRINT_LIMIT ) Unlimited_Output = FALSE;
#ifdef HAVE_MPE_H
error = MPE_Init_log();
#endif /* HAVE_MPE_H */
Dim = 0; /* for any hypercube legacy code... */
#endif /* PARALLEL */
#ifndef PARALLEL
Dim = 0;
ProcID = 0;
Num_Proc = 1;
#endif /* PARALLEL */
/*
* HKM - Change the ieee exception handling based on the machine and
* the level of debugging/speed desired. This call currently causes
* core dumps for floating point exceptions.
*/
handle_ieee();
log_msg("--------------");
log_msg("GOMA begins...");
/*
* Some initial stuff that only the master process does.
*/
if ( ProcID == 0 )
{
if (argc > 1)
{
log_msg("Preprocessing command line options.");
clc = (struct Command_line_command **)
smalloc( argc * sizeof(struct Command_line_command *));
for (i=0; i<argc; i++)
{
clc[i] = (struct Command_line_command *)
smalloc(sizeof(struct Command_line_command));
clc[i]->type = 0; /* initialize command line structure */
clc[i]->i_val = 0;
clc[i]->r_val = 0.;
clc[i]->string = (char *)
smalloc(MAX_COMMAND_LINE_LENGTH*sizeof(char));
for ( j=0; j<MAX_COMMAND_LINE_LENGTH; j++)
{
clc[i]->string[j] = '\0';
}
#ifdef DEBUG
fprintf(stderr, "clc[%d]->string is at 0x%x\n", i, clc[i]->string);
fprintf(stderr, "clc[%d] is at 0x%x\n", i, clc[i]);
#endif
}
}
strcpy(Input_File, "input");
strcpy(Echo_Input_File , "echo_input");
if (argc > 1) translate_command_line(argc, argv, clc, &nclc);
ECHO("OPEN", Echo_Input_File);
echo_command_line( argc, argv, Echo_Input_File );
print_code_version();
ptmp = legal_notice;
while ( strcmp(*ptmp, LAST_LEGAL_STRING) != 0 )
{
fprintf(stderr, "%s", *ptmp++);
}
}
/*
* Allocate the uniform problem description structure and
* the problem description structures on all processors
*/
error = pd_alloc();
EH(error, "pd_alloc problem");
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier after pd_alloc\n", ProcID);
#ifdef PARALLEL
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
#endif
log_msg("Allocating mp, gn, ...");
error = mp_alloc();
EH(error, "mp_alloc problem");
error = gn_alloc();
EH(error, "gn_alloc problem");
error = ve_alloc();
EH(error, "ve_alloc problem");
error = elc_alloc();
EH(error, "elc_alloc problem");
error = elc_rs_alloc();
EH(error, "elc_alloc problem");
error = cr_alloc();
EH(error, "cr_alloc problem");
error = evp_alloc();
EH(error, "evp_alloc problem");
error = tran_alloc();
EH(error, "tran_alloc problem");
error = eigen_alloc();
EH(error, "eigen_alloc problem");
error = cont_alloc();
EH(error, "cont_alloc problem");
error = loca_alloc();
EH(error, "loca_alloc problem");
error = efv_alloc();
EH(error, "efv_alloc problem");
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before read_input_file()\n", ProcID);
#ifdef PARALLEL
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
#endif
/*
* Read ASCII input file, data files, related exodusII FEM databases.
*/
if ( ProcID == 0 )
{
log_msg("Reading input file ...");
read_input_file(clc, nclc); /* Read ascii input file get file names */
/* update inputed data to account for command line arguments that
* might override the input deck...
*/
log_msg("Overriding any input file specs w/ any command line specs...");
if (argc > 1) apply_command_line(clc, nclc);
#ifdef DEBUG
DPRINTF(stderr, "apply_command_line() is done.\n");
#endif
}
/*
* The user-defined material properties, etc. available to goma users
* mean that some dynamically allocated data needs to be communicated.
*
* To handle this, sizing information from the input file scan is
* broadcast in stages so that the other processors can allocate space
* accordingly to hold the data.
*
* Note: instead of handpacking a data structure, use MPI derived datatypes
* to gather and scatter. Pray this is done efficiently. Certainly it costs
* less from a memory standpoint.
*/
#ifdef PARALLEL
/*
* Make sure the input file was successully processed before moving on
*/
check_parallel_error("Input file error");
/*
* This is some sizing information that helps fit a little bit more
* onto the ark later on.
*/
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before noahs_raven()\n", ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
noahs_raven();
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before MPI_Bcast of Noahs_Raven\n", ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
MPI_Bcast(MPI_BOTTOM, 1, Noahs_Raven->new_type, 0, MPI_COMM_WORLD);
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier after Bcast/before raven_landing()\n",
ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
/*
* Get the other processors ready to handle ark data.
*/
raven_landing();
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before noahs_ark()\n", ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
/*
* This is the main body of communicated information, including some
* whose sizes were determined because of advanced legwork by the raven.
*/
noahs_ark();
MPI_Bcast(MPI_BOTTOM, 1, Noahs_Ark->new_type, 0, MPI_COMM_WORLD);
/*
* Chemkin was initialized on processor zero during the input file
* process. Now, distribute it to all processors
*/
#ifdef USE_CHEMKIN
if (Chemkin_Needed) {
chemkin_initialize_mp();
}
#endif
/*
* Once the ark has landed, there are additional things that will need to
* be sent by dove. Example: BC_Types[]->u-BC arrays.
*
*/
ark_landing();
noahs_dove();
MPI_Bcast(MPI_BOTTOM, 1, Noahs_Dove->new_type, 0, MPI_COMM_WORLD);
#endif /* End of ifdef PARALLEL */
/*
* We sent the packed line to all processors that contained geometry
* creation commands. Now we need to step through it and create
* geometry as we go (including possibly reading an ACIS .sat file).
*
*/
/* Check to see if BRK File option exists and if so check if file exits */
if (Brk_Flag == 1) {
check_for_brkfile(Brk_File);
}
check_parallel_error("Error encountered in check for brkfile");
/* Now break the exodus files */
if (Num_Proc > 1 && ProcID == 0 && Brk_Flag == 1) {
call_brk();
}
check_parallel_error("Error in brking exodus files");
MPI_Barrier(MPI_COMM_WORLD);
/*
* For parallel execution, assume the following variables will be changed
* to reflect the multiple file aspect of the problem.
*
* FEM file = file.exoII --> file_3of15.exoII
*
* Output EXODUS II file = out.exoII --> out_3of15.exoII
*
*/
/*
* Allocate space for structures holding the EXODUS II finite element
* database information and for the Distributed Processing information.
*
* These are mostly skeletons with pointers that get allocated in the
* rd_exoII and rd_dpi routines. Remember to free up those arrays first
* before freeing the major pointers.
*/
EXO_ptr = alloc_struct_1(Exo_DB, 1);
init_exo_struct(EXO_ptr);
DPI_ptr = alloc_struct_1(Dpi, 1);
init_dpi_struct(DPI_ptr);
log_msg("Reading mesh from EXODUS II file...");
error = read_mesh_exoII(EXO_ptr, DPI_ptr);
/*
* Missing files on any processor are detected at a lower level
* forcing a return to the higher level
* rd_exo --> rd_mesh --> main
* Shutdown now, if any of the exodus files weren't found
*/
if (error < 0) {
#ifdef PARALLEL
MPI_Finalize();
#endif
return(-1);
}
/*
* All of the MPI_Type_commit() calls called behind the scenes that build
* the dove, ark and raven really allocated memory. Let's free it up now that
* the initial information has been communicated.
*/
#ifdef PARALLEL
MPI_Type_free(&(Noahs_Raven->new_type));
MPI_Type_free(&(Noahs_Ark->new_type));
MPI_Type_free(&(Noahs_Dove->new_type));
#endif
/*
* Setup the rest of the Problem Description structure that depends on
* the mesh that was read in from the EXODUS II file...
*
* Note that memory allocation and some setup has already been performed
* in mm_input()...
*/
error = setup_pd();
EH( error, "Problem setting up Problem_Description.");
/*
* Let's check to see if we need the large elasto-plastic global tensors
* and allocate them if so
*/
error = evp_tensor_alloc(EXO_ptr);
EH( error, "Problems setting up evp tensors");
/*
* Now that we know about what kind of problem we're solving and the
* mesh information, let's allocate space for elemental assembly structures
*
*/
#ifdef DEBUG
DPRINTF(stderr, "About to assembly_alloc()...\n");
#endif
log_msg("Assembly allocation...");
error = assembly_alloc(EXO_ptr);
EH( error, "Problem from assembly_alloc");
if (Debug_Flag) {
DPRINTF(stderr, "%s: setting up EXODUS II output files...\n", yo);
}
/*
* These are not critical - just niceties. Also, they should not overburden
* your db with too much of this - they're capped verbiage compliant routines.
*/
add_qa_stamp(EXO_ptr);
add_info_stamp(EXO_ptr);
#ifdef DEBUG
fprintf(stderr, "added qa and info stamps\n");
#endif
/*
* If the output EXODUS II database file is different from the input
* file, then we'll need to replicate all the basic mesh information.
* But, remember that if we're parallel, that the output file names must
* be multiplexed first...
*/
if ( Num_Proc > 1 )
{
multiname(ExoFileOut, ProcID, Num_Proc);
multiname(Init_GuessFile, ProcID, Num_Proc);
if ( strcmp( Soln_OutFile, "" ) != 0 )
{
multiname(Soln_OutFile, ProcID, Num_Proc);
}
if( strcmp( ExoAuxFile, "" ) != 0 )
{
multiname(ExoAuxFile, ProcID, Num_Proc);
}
if( efv->Num_external_field != 0 )
{
for( i=0; i<efv->Num_external_field; i++ )
{
multiname(efv->file_nm[i], ProcID, Num_Proc);
}
}
}
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* Preprocess the exodus mesh
* -> Allocate pointers to structures containing element
* side bc info, First_Elem_Side_BC_Array, and
* element edge info, First_Elem_Edge_BC_Array.
* -> Determine Unique_Element_Types[] array
*/
#ifdef DEBUG
fprintf(stderr, "pre_process()...\n");
#endif
log_msg("Pre processing of mesh...");
#ifdef PARALLEL
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
pre_process(EXO_ptr);
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* Load up a few key indeces in the bfd prototype basis function structures
* and make sure that each active eqn/vbl has a bf[v] that points to the
* right bfd[]...needs pre_process to find out the number of unique
* element types in the problem.
*/
#ifdef DEBUG
fprintf(stderr, "bf_init()...\n");
#endif
log_msg("Basis function initialization...");
error = bf_init(EXO_ptr);
EH( error, "Problem from bf_init");
/*
* check for parallel errors before continuing
*/
check_parallel_error("Error encountered in problem setup");
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* Allocate space for each communication exchange description.
*/
#ifdef PARALLEL
#ifdef DEBUG
fprintf(stderr, "P_%d: Parallel cx allocation\n", ProcID);
#endif
if (DPI_ptr->num_neighbors > 0) {
cx = alloc_struct_1(Comm_Ex, DPI_ptr->num_neighbors);
Request = alloc_struct_1(MPI_Request,
Num_Requests * DPI_ptr->num_neighbors);
Status = alloc_struct_1(MPI_Status,
Num_Requests * DPI_ptr->num_neighbors);
}
#endif
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* SET UP THE PROBLEM
*
* Setup node-based structures
* Finalise how boundary conditions are to be handled
* Determine what unknowns are at each owned node and then tell
* neighboring processors about your nodes
* Set up communications pattern for fast unknown updates between
* processors.
*/
(void) setup_problem(EXO_ptr, DPI_ptr);
/*
* check for parallel errors before continuing
*/
check_parallel_error("Error encountered in problem setup");
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* CREATE BRK_FILE IF ONE DOES NOT EXIST
*
* If no Brk_File exists but the option was configured in the input or
* optional command we create one now and exit from goma.
*/
if ( Brk_Flag == 2 ) {
write_brk_file(Brk_File, EXO_ptr);
exit(0);
}
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* WRITE OUT INITIAL INFO TO EXODUS FILE
*/
/*
* Only have to initialize the exodus file if we are using different
* files for the output versus the input mesh
*/
if (strcmp(ExoFile, ExoFileOut)) {
/*
* Temporarily we'll need to renumber the nodes and elements in the
* mesh to be 1-based. After writing, return to the 0 based indexing
* that is more convenient in C.
*/
#ifdef DEBUG
fprintf(stderr, "1-base; wr_mesh; 0-base\n");
#endif
one_base(EXO_ptr);
wr_mesh_exo(EXO_ptr, ExoFileOut, 0);
zero_base(EXO_ptr);
/*
* If running on a distributed computer, augment the plain finite
* element information of EXODUS with the description of how this
* piece fits into the global problem.
*/
if (Num_Proc > 1) {
#ifdef PARALLEL
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before wr_dpi()\n", ProcID);
fprintf(stderr, "P_%d ExoFileOut = \"%s\"\n", ProcID, ExoFileOut);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
#endif
wr_dpi(DPI_ptr, ExoFileOut, 0);
}
}
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* SOLVE THE PROBLEM
*/
if (Debug_Flag) {
switch (Continuation) {
case ALC_ZEROTH:
P0PRINTF("%s: continue_problem (zeroth order) ...\n", yo);
break;
case ALC_FIRST:
P0PRINTF("%s: continue_problem (first order) ...\n", yo);
break;
case HUN_ZEROTH:
P0PRINTF("%s: hunt_problem (zeroth order) ...\n", yo);
break;
case HUN_FIRST:
P0PRINTF("%s: hunt_problem (first order) ...\n", yo);
break;
case LOCA:
P0PRINTF("%s: do_loca ...\n", yo);
break;
default:
P0PRINTF("%s: solve_problem...\n", yo);
break;
}
}
#ifdef DEBUG
switch (Continuation) {
case ALC_ZEROTH:
DPRINTF(stderr, "%s: continue_problem (zeroth order) ...\n", yo);
break;
case ALC_FIRST:
DPRINTF(stderr, "%s: continue_problem (first order) ...\n", yo);
break;
case HUN_ZEROTH:
DPRINTF(stderr, "%s: hunt_problem (zeroth order) ...\n", yo);
break;
case HUN_FIRST:
DPRINTF(stderr, "%s: hunt_problem (first order) ...\n", yo);
break;
case LOCA:
DPRINTF(stderr, "%s: do_loca ...\n", yo);
break;
default:
DPRINTF(stderr, "%s: solve_problem...\n", yo);
break;
}
#endif
if( TimeIntegration == TRANSIENT)
{
Continuation = ALC_NONE;
if (Debug_Flag) {
P0PRINTF("%s: solve_problem...TRANSIENT superceded Continuation...\n", yo);
}
#ifdef DEBUG
DPRINTF(stderr, "%s: solve_problem...TRANSIENT superceded Continuation...\n", yo);
#endif
solve_problem(EXO_ptr, DPI_ptr, NULL);
}
switch (Continuation) {
case ALC_ZEROTH:
case ALC_FIRST:
log_msg("Solving continuation problem");
continue_problem(cx, EXO_ptr, DPI_ptr);
break;
case HUN_ZEROTH:
case HUN_FIRST:
log_msg("Solving hunt problem");
hunt_problem(cx, EXO_ptr, DPI_ptr);
break;
case LOCA:
log_msg("Solving continuation problem with LOCA");
error = do_loca(cx, EXO_ptr, DPI_ptr);
break;
default:
log_msg("Solving problem");
if (loca_in->Cont_Alg == LOCA_LSA_ONLY)
{
error = do_loca(cx, EXO_ptr, DPI_ptr);
}
else if(TimeIntegration != TRANSIENT)
{
solve_problem(EXO_ptr, DPI_ptr, NULL);
}
break;
}
#ifdef PARALLEL
MPI_Barrier(MPI_COMM_WORLD);
#endif
if (ProcID == 0 && Brk_Flag == 1 && Num_Proc > 1) {
fix_output();
}
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* PRINT A MESSAGE TO STDOUT SAYING WE ARE DONE
*/
P0PRINTF("\n-done\n\n");
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* FREE MEMORY ALLOCATED BY THE PROGRAM
*/
/*
* free the element block / element based structures
*/
free_element_blocks(EXO_ptr);
/*
* free nodal based structures
*/
free_nodes();
#ifdef FREE_PROBLEM
free_problem ( EXO_ptr, DPI_ptr );
#endif
/*
* Free command line stuff
*/
if ( ProcID == 0 )
{
if ( argc > 1 )
{
for (i=0; i<argc; i++)
{
#ifdef DEBUG
fprintf(stderr, "clc[%d]->string &= 0x%x\n", i, clc[i]->string);
fprintf(stderr, "clc[%d] &= 0x%x\n", i, clc[i]);
#endif
safer_free((void **) &(clc[i]->string));
safer_free((void **) (clc + i));
}
safer_free((void **) &clc);
}
}
/*
* Free exodus database structures
*/
free_exo(EXO_ptr);
safer_free((void **) &EXO_ptr);
if ( Num_Proc > 1 )
{
free_dpi(DPI_ptr);
}
else
{
free_dpi_uni(DPI_ptr);
}
safer_free((void **) &DPI_ptr);
/*
* Remove front scratch file [/tmp/lu.'pid'.0]
*/
if (Linear_Solver == FRONT)
{
unlerr = unlink(front_scratch_directory);
WH(unlerr, "Unlink problem with front scratch file");
}
#ifdef PARALLEL
total_time = ( MPI_Wtime() - time_start )/ 60. ;
DPRINTF(stderr, "\nProc 0 runtime: %10.2f Minutes.\n\n",total_time);
MPI_Finalize();
#endif
#ifndef PARALLEL
(void)time(&now);
total_time = (double)(now) - time_start;
fprintf(stderr, "\nProc 0 runtime: %10.2f Minutes.\n\n",total_time/60);
#endif
fflush(stdout);
fflush(stderr);
log_msg("GOMA ends normally.");
return (0);
}
int
goma_init_(dbl *time1, int *nnodes, int *nelems,
int *nnv_in, int *nev_in, int *i_soln, int *i_post)
/*
* Initial main driver for GOMA. Derived from a (1/93) release of
* the rf_salsa program by
*
* Original Authors: John Shadid (1421)
* Scott Hutchinson (1421)
* Harry Moffat (1421)
*
* Date: 12/3/92
*
*
* Updates and Changes by:
* Randy Schunk (9111)
* P. A. Sackinger (9111)
* R. R. Rao (9111)
* R. A. Cairncross (Univ. of Delaware)
* Dates: 2/93 - 6/96
*
* Modified for continuation
* Ian Gates
* Dates: 2/98 - 10/98
* Dates: 7/99 - 8/99
*
* Last modified: Wed June 26 14:21:35 MST 1994 [email protected]
* Hello.
*
* Note: Many modifications from an early 2/93 pre-release
* version of rf_salsa were made by various persons
* in order to test ideas about moving/deforming meshes...
*/
{
/* Local Declarations */
double time_start, total_time; /* timing variables */
#ifndef PARALLEL
struct tm *tm_ptr; /* additional serial timing variables */
time_t the_time;
#endif
int error;
int i;
int j;
static int first_goma_call=TRUE;
char **ptmp;
static const char *yo="goma_init";
struct Command_line_command **clc=NULL; /* point to command line structure */
int nclc = 0; /* number of command line commands */
/********************** BEGIN EXECUTION ***************************************/
/* assume number of commands is less than or equal to the number of
* arguments in the command line minus 1 (1st is program name) */
/*
* Get the name of the executable, yo
*/
#ifdef PARALLEL
if( first_goma_call ) {
Argc = 1;
Argv = (char **) smalloc( Argc*sizeof(char *) );
Argv[0] = (char *) yo;
MPI_Init(&Argc, &Argv); /*PRS will have to fix this. Too late TAB already did. */
}
time_start = MPI_Wtime();
#else /* PARALLEL */
(void) time(&the_time);
tm_ptr = gmtime(&the_time);
time_start = (double) ( tm_ptr->tm_sec
+ 60. * ( 60. * ( tm_ptr->tm_yday * 24. + tm_ptr->tm_hour )
+ tm_ptr->tm_min )
);
#endif /* PARALLEL */
*time1 = time_start;
/* Argv = argv; */
/* Argc = argc; */
time_goma_started = time_start;
#ifdef PARALLEL
/*
* Determine the parallel processing status, if any. We need to know
* pretty early if we're "one of many" or the only process.
*/
error = MPI_Comm_size(MPI_COMM_WORLD, &Num_Proc);
error = MPI_Comm_rank(MPI_COMM_WORLD, &ProcID);
/*
* Setup a default Proc_config so we can use utility routines
* from Aztec
*/
AZ_set_proc_config(Proc_Config, MPI_COMM_WORLD);
/* set the output limit flag if need be */
if( Num_Proc > DP_PROC_PRINT_LIMIT ) Unlimited_Output = FALSE;
#ifdef HAVE_MPE_H
error = MPE_Init_log();
#endif /* HAVE_MPE_H */
Dim = 0; /* for any hypercube legacy code... */
#endif /* PARALLEL */
#ifndef PARALLEL
Dim = 0;
ProcID = 0;
Num_Proc = 1;
#endif /* PARALLEL */
/*
* HKM - Change the ieee exception handling based on the machine and
* the level of debugging/speed desired. This call currently causes
* core dumps for floating point exceptions.
*/
handle_ieee();
log_msg("--------------");
log_msg("GOMA begins...");
#ifdef USE_CGM
cgm_initialize();
#endif
/*
* Some initial stuff that only the master process does.
*/
/*PRS: Disable this command line stuff for the jas coupled version */
/*-----------------------------------------------------------------*/
/* if ( ProcID == 0 ) */
/* { */
/* if (argc > 1) */
/* { */
/* log_msg("Preprocessing command line options."); */
/* clc = (struct Command_line_command **) */
/* smalloc( argc * sizeof(struct Command_line_command *)); */
/* for (i=0; i<argc; i++) */
/* { */
/* clc[i] = (struct Command_line_command *) */
/* smalloc(sizeof(struct Command_line_command)); */
/* clc[i]->type = 0; /\* initialize command line structure *\/ */
/* clc[i]->i_val = 0; */
/* clc[i]->r_val = 0.; */
/* clc[i]->string = (char *) */
/* smalloc(MAX_COMMAND_LINE_LENGTH*sizeof(char)); */
/* for ( j=0; j<MAX_COMMAND_LINE_LENGTH; j++) */
/* { */
/* clc[i]->string[j] = '\0'; */
/* } */
/* #ifdef DEBUG */
/* fprintf(stderr, "clc[%d]->string is at 0x%x\n", i, clc[i]->string); */
/* fprintf(stderr, "clc[%d] is at 0x%x\n", i, clc[i]); */
/* #endif */
/* } */
/* } */
/* PRS For the JAS version we will use the default input file name "input" */
strcpy(Input_File, "input");
/* if (argc > 1) translate_command_line(argc, argv, clc, &nclc); */
/* print_code_version(); */
/* ptmp = legal_notice; */
/* while ( strcmp(*ptmp, LAST_LEGAL_STRING) != 0 ) */
/* { */
/* fprintf(stderr, "%s", *ptmp++); */
/* } */
/* } */
/*
* Allocate the uniform problem description structure and
* the problem description structures on all processors
*/
error = pd_alloc();
EH(error, "pd_alloc problem");
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier after pd_alloc\n", ProcID);
#ifdef PARALLEL
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
#endif
log_msg("Allocating mp, gn, ...");
error = mp_alloc();
EH(error, "mp_alloc problem");
error = gn_alloc();
EH(error, "gn_alloc problem");
error = ve_alloc();
EH(error, "ve_alloc problem");
error = elc_alloc();
EH(error, "elc_alloc problem");
error = elc_rs_alloc();
EH(error, "elc_alloc problem");
error = cr_alloc();
EH(error, "cr_alloc problem");
error = evp_alloc();
EH(error, "evp_alloc problem");
error = tran_alloc();
EH(error, "tran_alloc problem");
error = libio_alloc();
EH(error, "libio_alloc problem");
error = eigen_alloc();
EH(error, "eigen_alloc problem");
error = cont_alloc();
EH(error, "cont_alloc problem");
error = loca_alloc();
EH(error, "loca_alloc problem");
error = efv_alloc();
EH(error, "efv_alloc problem");
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before read_input_file()\n", ProcID);
#ifdef PARALLEL
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
#endif
/*PRS AGAIN, NO COMMAND LINE OVERRIDES IN THIS JAS3D VERSION */
/*
* Read ASCII input file, data files, related exodusII FEM databases.
*/
if ( ProcID == 0 )
{
log_msg("Reading input file ...");
read_input_file(clc, nclc);
}
/*
* The user-defined material properties, etc. available to goma users
* mean that some dynamically allocated data needs to be communicated.
*
* To handle this, sizing information from the input file scan is
* broadcast in stages so that the other processors can allocate space
* accordingly to hold the data.
*
* Note: instead of handpacking a data structure, use MPI derived datatypes
* to gather and scatter. Pray this is done efficiently. Certainly it costs
* less from a memory standpoint.
*/
#ifdef PARALLEL
/*
* Make sure the input file was successully processed before moving on
*/
check_parallel_error("Input file error");
/*
* This is some sizing information that helps fit a little bit more
* onto the ark later on.
*/
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before noahs_raven()\n", ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
noahs_raven();
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before MPI_Bcast of Noahs_Raven\n", ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
MPI_Bcast(MPI_BOTTOM, 1, Noahs_Raven->new_type, 0, MPI_COMM_WORLD);
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier after Bcast/before raven_landing()\n",
ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
/*
* Get the other processors ready to handle ark data.
*/
raven_landing();
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before noahs_ark()\n", ProcID);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
/*
* This is the main body of communicated information, including some
* whose sizes were determined because of advanced legwork by the raven.
*/
noahs_ark();
MPI_Bcast(MPI_BOTTOM, 1, Noahs_Ark->new_type, 0, MPI_COMM_WORLD);
/*
* Chemkin was initialized on processor zero during the input file
* process. Now, distribute it to all processors
*/
#ifdef USE_CHEMKIN
if (Chemkin_Needed) {
chemkin_initialize_mp();
}
#endif
/*
* Once the ark has landed, there are additional things that will need to
* be sent by dove. Example: BC_Types[]->u-BC arrays.
*
*/
ark_landing();
noahs_dove();
MPI_Bcast(MPI_BOTTOM, 1, Noahs_Dove->new_type, 0, MPI_COMM_WORLD);
#endif /* End of ifdef PARALLEL */
/*
* We sent the packed line to all processors that contained geometry
* creation commands. Now we need to step through it and create
* geometry as we go (including possibly reading an ACIS .sat file).
*
*/
#ifdef USE_CGM
create_cgm_geometry();
#endif
/*
* For parallel execution, assume the following variables will be changed
* to reflect the multiple file aspect of the problem.
*
* FEM file = file.exoII --> file_3of15.exoII
*
* Output EXODUS II file = out.exoII --> out_3of15.exoII
*
*/
/*
* Allocate space for structures holding the EXODUS II finite element
* database information and for the Distributed Processing information.
*
* These are mostly skeletons with pointers that get allocated in the
* rd_exoII and rd_dpi routines. Remember to free up those arrays first
* before freeing the major pointers.
*/
EXO_ptr = alloc_struct_1(Exo_DB, 1);
init_exo_struct(EXO_ptr);
DPI_ptr = alloc_struct_1(Dpi, 1);
init_dpi_struct(DPI_ptr);
log_msg("Reading mesh from EXODUS II file...");
error = read_mesh_exoII(EXO_ptr, DPI_ptr);
/*
* Missing files on any processor are detected at a lower level
* forcing a return to the higher level
* rd_exo --> rd_mesh --> main
* Shutdown now, if any of the exodus files weren't found
*/
if (error < 0) {
#ifdef PARALLEL
MPI_Finalize();
#endif
return(-1);
}
/*
* All of the MPI_Type_commit() calls called behind the scenes that build
* the dove, ark and raven really allocated memory. Let's free it up now that
* the initial information has been communicated.
*/
#ifdef PARALLEL
MPI_Type_free(&(Noahs_Raven->new_type));
MPI_Type_free(&(Noahs_Ark->new_type));
MPI_Type_free(&(Noahs_Dove->new_type));
#endif
/*
* Setup the rest of the Problem Description structure that depends on
* the mesh that was read in from the EXODUS II file...
*
* Note that memory allocation and some setup has already been performed
* in mm_input()...
*/
error = setup_pd();
EH( error, "Problem setting up Problem_Description.");
/*
* Let's check to see if we need the large elasto-plastic global tensors
* and allocate them if so
*/
error = evp_tensor_alloc(EXO_ptr);
EH( error, "Problems setting up evp tensors");
/*
* Now that we know about what kind of problem we're solving and the
* mesh information, let's allocate space for elemental assembly structures
*
*/
#ifdef DEBUG
DPRINTF(stderr, "About to assembly_alloc()...\n");
#endif
log_msg("Assembly allocation...");
error = assembly_alloc(EXO_ptr);
EH( error, "Problem from assembly_alloc");
if (Debug_Flag) {
DPRINTF(stderr, "%s: setting up EXODUS II output files...\n", yo);
}
/*
* These are not critical - just niceties. Also, they should not overburden
* your db with too much of this - they're capped verbiage compliant routines.
*/
add_qa_stamp(EXO_ptr);
add_info_stamp(EXO_ptr);
#ifdef DEBUG
fprintf(stderr, "added qa and info stamps\n");
#endif
/*
* If the output EXODUS II database file is different from the input
* file, then we'll need to replicate all the basic mesh information.
* But, remember that if we're parallel, that the output file names must
* be multiplexed first...
*/
if ( Num_Proc > 1 )
{
multiname(ExoFileOut, ProcID, Num_Proc);
multiname(Init_GuessFile, ProcID, Num_Proc);
if ( strcmp( Soln_OutFile, "" ) != 0 )
{
multiname(Soln_OutFile, ProcID, Num_Proc);
}
if( strcmp( ExoAuxFile, "" ) != 0 )
{
multiname(ExoAuxFile, ProcID, Num_Proc);
}
if( efv->Num_external_field != 0 )
{
for( i=0; i<efv->Num_external_field; i++ )
{
multiname(efv->file_nm[i], ProcID, Num_Proc);
}
}
}
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* Preprocess the exodus mesh
* -> Allocate pointers to structures containing element
* side bc info, First_Elem_Side_BC_Array, and
* element edge info, First_Elem_Edge_BC_Array.
* -> Determine Unique_Element_Types[] array
*/
#ifdef DEBUG
fprintf(stderr, "pre_process()...\n");
#endif
log_msg("Pre processing of mesh...");
#ifdef PARALLEL
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
pre_process(EXO_ptr);
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* Load up a few key indeces in the bfd prototype basis function structures
* and make sure that each active eqn/vbl has a bf[v] that points to the
* right bfd[]...needs pre_process to find out the number of unique
* element types in the problem.
*/
#ifdef DEBUG
fprintf(stderr, "bf_init()...\n");
#endif
log_msg("Basis function initialization...");
error = bf_init(EXO_ptr);
EH( error, "Problem from bf_init");
/*
* check for parallel errors before continuing
*/
check_parallel_error("Error encountered in problem setup");
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* Allocate space for each communication exchange description.
*/
#ifdef PARALLEL
#ifdef DEBUG
fprintf(stderr, "P_%d: Parallel cx allocation\n", ProcID);
#endif
if (DPI_ptr->num_neighbors > 0) {
cx = alloc_struct_1(Comm_Ex, DPI_ptr->num_neighbors);
Request = alloc_struct_1(MPI_Request,
Num_Requests * DPI_ptr->num_neighbors);
Status = alloc_struct_1(MPI_Status,
Num_Requests * DPI_ptr->num_neighbors);
}
#endif
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* SET UP THE PROBLEM
*
* Setup node-based structures
* Finalise how boundary conditions are to be handled
* Determine what unknowns are at each owned node and then tell
* neighboring processors about your nodes
* Set up communications pattern for fast unknown updates between
* processors.
*/
(void) setup_problem(EXO_ptr, DPI_ptr);
/*
* check for parallel errors before continuing
*/
check_parallel_error("Error encountered in problem setup");
/***********************************************************************/
/***********************************************************************/
/***********************************************************************/
/*
* WRITE OUT INITIAL INFO TO EXODUS FILE
*/
/*
* Only have to initialize the exodus file if we are using different
* files for the output versus the input mesh
*/
if (strcmp(ExoFile, ExoFileOut)) {
/*
* Temporarily we'll need to renumber the nodes and elements in the
* mesh to be 1-based. After writing, return to the 0 based indexing
* that is more convenient in C.
*/
#ifdef DEBUG
fprintf(stderr, "1-base; wr_mesh; 0-base\n");
#endif
one_base(EXO_ptr);
wr_mesh_exo(EXO_ptr, ExoFileOut, 0);
zero_base(EXO_ptr);
/*
* If running on a distributed computer, augment the plain finite
* element information of EXODUS with the description of how this
* piece fits into the global problem.
*/
if (Num_Proc > 1) {
#ifdef PARALLEL
#ifdef DEBUG
fprintf(stderr, "P_%d at barrier before wr_dpi()\n", ProcID);
fprintf(stderr, "P_%d ExoFileOut = \"%s\"\n", ProcID, ExoFileOut);
error = MPI_Barrier(MPI_COMM_WORLD);
#endif
#endif
wr_dpi(DPI_ptr, ExoFileOut, 0);
}
}
if (Num_Import_NV > 0 || Num_Import_EV > 0) printf
(" Goma will import %d nodal and %d element variables.\n",
Num_Import_NV, Num_Import_EV);
if (Num_Export_XS > 0 || Num_Export_XP > 0) printf
(" Goma will export %d solution and %d post-processing variables.\n",
Num_Export_XS, Num_Export_XP);
/* Return counts to calling program */
*nnodes = EXO_ptr->num_nodes;
*nelems = EXO_ptr->num_elems;
*nnv_in = Num_Import_NV;
*nev_in = Num_Import_EV;
*i_soln = Num_Export_XS;
*i_post = Num_Export_XP;
return (0); /* Back to animas*/
}
static int tview_func(uint32_t tid, uint32_t pos, int n, const bam_pileup1_t *pl, void *data)
{
bam_lplbuf_t *tv = (bam_lplbuf_t*)data;
freenode_t *p;
int i, l, max_level;
// allocate memory if necessary
if (tv->max < n) { // enlarge
tv->max = n;
kroundup32(tv->max);
tv->cur_level = (int*)realloc(tv->cur_level, sizeof(int) * tv->max);
tv->pre_level = (int*)realloc(tv->pre_level, sizeof(int) * tv->max);
}
tv->n_cur = n;
// update cnt
for (p = tv->head; p->next; p = p->next)
if (p->cnt > 0) --p->cnt;
// calculate cur_level[]
max_level = 0;
for (i = l = 0; i < n; ++i) {
const bam_pileup1_t *p = pl + i;
if (p->is_head) {
if (tv->head->next && tv->head->cnt == 0) { // then take a free slot
freenode_t *p = tv->head->next;
tv->cur_level[i] = tv->head->level;
mp_free(tv->mp, tv->head);
tv->head = p;
--tv->n_nodes;
} else tv->cur_level[i] = ++tv->max_level;
} else {
tv->cur_level[i] = tv->pre_level[l++];
if (p->is_tail) { // then return a free slot
tv->tail->level = tv->cur_level[i];
tv->tail->next = mp_alloc(tv->mp);
tv->tail = tv->tail->next;
++tv->n_nodes;
}
}
if (tv->cur_level[i] > max_level) max_level = tv->cur_level[i];
((bam_pileup1_t*)p)->level = tv->cur_level[i];
}
assert(l == tv->n_pre);
tv->func(tid, pos, n, pl, tv->user_data);
// sort the linked list
if (tv->n_nodes) {
freenode_t *q;
if (tv->n_nodes + 1 > tv->m_aux) { // enlarge
tv->m_aux = tv->n_nodes + 1;
kroundup32(tv->m_aux);
tv->aux = (freenode_t**)realloc(tv->aux, sizeof(void*) * tv->m_aux);
}
for (p = tv->head, i = l = 0; p->next;) {
if (p->level > max_level) { // then discard this entry
q = p->next;
mp_free(tv->mp, p);
p = q;
} else {
tv->aux[i++] = p;
p = p->next;
}
}
tv->aux[i] = tv->tail; // add a proper tail for the loop below
tv->n_nodes = i;
if (tv->n_nodes) {
ks_introsort(node, tv->n_nodes, tv->aux);
for (i = 0; i < tv->n_nodes; ++i) tv->aux[i]->next = tv->aux[i+1];
tv->head = tv->aux[0];
} else tv->head = tv->tail;
}
// clean up
tv->max_level = max_level;
memcpy(tv->pre_level, tv->cur_level, tv->n_cur * 4);
// squeeze out terminated levels
for (i = l = 0; i < n; ++i) {
const bam_pileup1_t *p = pl + i;
if (!p->is_tail)
tv->pre_level[l++] = tv->pre_level[i];
}
tv->n_pre = l;
/*
fprintf(stderr, "%d\t", pos+1);
for (i = 0; i < n; ++i) {
const bam_pileup1_t *p = pl + i;
if (p->is_head) fprintf(stderr, "^");
if (p->is_tail) fprintf(stderr, "$");
fprintf(stderr, "%d,", p->level);
}
fprintf(stderr, "\n");
*/
return 0;
}
int main() {
puts("Master");
corethread_t worker_1 = 1; // For now the core ID
int worker_1_param = 1;
corethread_create(&worker_1,&func_worker_1,(void*)&worker_1_param);
// Create the queuing ports
spd_t * chan = mp_create_sport(MP_CHAN_1_ID, SINK, MP_CHAN_1_MSG_SIZE);
volatile unsigned long long _SPM * time_sample = mp_alloc(MP_CHAN_1_MSG_SIZE);
if (chan == NULL || time_sample == NULL) {
DEBUGF(chan);
abort();
}
// Initialize the communication channels
int retval = mp_init_ports();
// TODO: check on retval
puts("Initialized ports");
while(slave != 1) {
;
}
puts("Slave is ready");
unsigned long long min_time_diff = -1;
unsigned long long max_time_diff = 0;
unsigned long long accum_time_diff = 0;
unsigned long long cnt_time_diff = 0;
unsigned long long percent = 0;
int done = 0;
unsigned long long start = get_cpu_usecs();
while(!done) {
int success = mp_read(chan,time_sample);
unsigned long long time_diff = get_cpu_usecs() - (*time_sample);
if (success == 0) {
printf("No sample received\n");
} else if ((*time_sample) == 0) {
printf("Received empty sample, newest: %u, sample size: %u\n",chan->newest,chan->sample_size);
} else {
if (time_diff > 2000 ) {
// Time difference is larger than a micro second
printf("Time sample: %llu\tdiff: %llu\n",*time_sample,time_diff);
}
cnt_time_diff++;
if (time_diff < min_time_diff) {
min_time_diff = time_diff;
}
if (time_diff > max_time_diff) {
max_time_diff = time_diff;
}
accum_time_diff += time_diff;
}
if (start + percent < get_cpu_usecs()) {
percent += RUNTIME/10;
printf("+");
fflush(stdout);
}
if ( start + RUNTIME < get_cpu_usecs()) {
done = 1;
}
}
printf("\n");
printf("Status:\n\tMin time diff: %llu\n\tMax time diff: %llu\n\tAvg time diff: %llu\n", min_time_diff,max_time_diff,accum_time_diff/cnt_time_diff);
int* res;
corethread_join(worker_1,&res);
return *res;
}
int main() {
corethread_t worker_1 = SLAVE_CORE; // For now the core ID
corethread_create(&worker_1,&func_worker_1,(void*)&worker_1);
puts("Corethread created");
unsigned short int local_phase = 0;
min_time = ULONG_MAX;
max_time = 0;
accum_time = 0;
cnt_time = 0;
unsigned long long int start = 0;
unsigned long long int stop = 0;
spd_t * sport1 = mp_create_sport(CHAN_ID_ONE,SINK,SAMPLE_SIZE*sizeof(short));
spd_t * sport2 = mp_create_sport(CHAN_ID_TWO,SOURCE,SAMPLE_SIZE*sizeof(short));
if (sport1 == NULL || sport2 == NULL) {
//exit(1);
}
volatile short _SPM * sample = mp_alloc(SAMPLE_SIZE*sizeof(short));
mp_init_ports();
done = 1;
int balance = 0;
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[i] = i;
}
for (int i = 0; i < ITERATIONS/2; ++i) {
mp_write(sport2,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[i] = i;
}
}
for (int i = 0; i < ITERATIONS/2; ++i) {
mp_write(sport2,sample);
for (int i = 0; i < SAMPLE_SIZE; ++i) {
sample[SAMPLE_SIZE-1-i] = i;
}
}
for (int i = 0; i < ITERATIONS; ++i) {
start = get_cpu_usecs();
int ret = mp_read(sport1,sample);
stop = get_cpu_usecs();
if (ret == 0)
{
puts("No value written yet.");
} else {
unsigned long long int exe_time = stop - start;
min_time = (exe_time < min_time) ? exe_time : min_time;
max_time = (exe_time > max_time) ? exe_time : max_time;
accum_time += exe_time;
cnt_time++;
if (sample[0] == 0) {
balance++;
for (int i = 0; i < SAMPLE_SIZE; ++i) {
if(sample[i] != i) {
printf("Error: sample[%i] = %i\n",i,sample[i]);
break;
}
}
} else if (sample[0] == SAMPLE_SIZE-1) {
balance--;
for (int i = 0; i < SAMPLE_SIZE; ++i) {
if(sample[SAMPLE_SIZE-1-i] != i) {
printf("Error: sample[%i] = %i\n",i,sample[i]);
break;
}
}
} else {
printf("Wrong sample values sample[0] = %i\n",sample[0]);
}
}
}
printf("Local phase: %d\n",local_phase);
inval_dcache();
int* res;
corethread_join(worker_1,&res);
printf("Balance: %i\n",balance);
printf("Min time: %llu\tMax time: %llu\tAccumulated time: %llu\nCount time: %llu\tAverage time: %llu\n", min_time,max_time,accum_time,cnt_time,accum_time/cnt_time);
puts("Corethread joined");
return *res;
}
inline void *
mp_malloc(size_t size)
{
return mp_alloc(size);
}
