void compute_step_factor(int nelr, double* variables, double* areas, double* step_factors)
{
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for default(shared) schedule(static)
for(int i = 0; i < nelr; i++)
{
double density = variables[NVAR*i + VAR_DENSITY];
cfd_double3 momentum;
momentum.x = variables[NVAR*i + (VAR_MOMENTUM+0)];
momentum.y = variables[NVAR*i + (VAR_MOMENTUM+1)];
momentum.z = variables[NVAR*i + (VAR_MOMENTUM+2)];
double density_energy = variables[NVAR*i + VAR_DENSITY_ENERGY];
cfd_double3 velocity; compute_velocity(density, momentum, velocity);
double speed_sqd = compute_speed_sqd(velocity);
double pressure = compute_pressure(density, density_energy, speed_sqd);
double speed_of_sound = compute_speed_of_sound(density, pressure);
// dt = double(0.5) * std::sqrt(areas[i]) / (||v|| + c).... but when we do time stepping, this later would need to be divided by the area, so we just do it all at once
step_factors[i] = double(0.5) / (std::sqrt(areas[i]) * (std::sqrt(speed_sqd) + speed_of_sound));
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma155_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
}
/*
* Verifies the correctness of the sort.
* Ensures all keys are within a PE's bucket boundaries.
* Ensures the final number of keys is equal to the initial.
*/
static int verify_results(int const * const my_local_key_counts,
KEY_TYPE const * const my_local_keys)
{
shmem_barrier_all();
int error = 0;
const int my_rank = shmem_my_pe();
const int my_min_key = my_rank * BUCKET_WIDTH;
const int my_max_key = (my_rank+1) * BUCKET_WIDTH - 1;
#ifdef ISX_PROFILING
unsigned long long start = current_time_ns();
#endif
// Verify all keys are within bucket boundaries
for(long long int i = 0; i < my_bucket_size; ++i){
const int key = my_local_keys[i];
if((key < my_min_key) || (key > my_max_key)){
printf("Rank %d Failed Verification!\n",my_rank);
printf("Key: %d is outside of bounds [%d, %d]\n", key, my_min_key, my_max_key);
error = 1;
}
}
#ifdef ISX_PROFILING
unsigned long long end = current_time_ns();
if (shmem_my_pe() == 0)
printf("Verifying took %llu ns\n", end - start);
#endif
// Verify the sum of the key population equals the expected bucket size
long long int bucket_size_test = 0;
for(uint64_t i = 0; i < BUCKET_WIDTH; ++i){
bucket_size_test += my_local_key_counts[i];
}
if(bucket_size_test != my_bucket_size){
printf("Rank %d Failed Verification!\n",my_rank);
printf("Actual Bucket Size: %lld Should be %lld\n", bucket_size_test, my_bucket_size);
error = 1;
}
// Verify the final number of keys equals the initial number of keys
static long long int total_num_keys = 0;
shmem_longlong_sum_to_all(&total_num_keys, &my_bucket_size, 1, 0, 0, NUM_PES, llWrk, pSync);
shmem_barrier_all();
if(total_num_keys != (long long int)(NUM_KEYS_PER_PE * NUM_PES)){
if(my_rank == ROOT_PE){
printf("Verification Failed!\n");
printf("Actual total number of keys: %lld Expected %" PRIu64 "\n", total_num_keys, NUM_KEYS_PER_PE * NUM_PES );
error = 1;
}
}
return error;
}
void copy(double *dst, double *src, int N)
{
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for default(shared) schedule(static)
for(int i = 0; i < N; i++)
{
dst[i] = src[i];
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma53_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
}
void initialize_variables(int nelr, double* variables)
{
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for default(shared) schedule(static)
for(int i = 0; i < nelr; i++)
{
for(int j = 0; j < NVAR; j++) variables[i*NVAR + j] = ff_variable[j];
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma102_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
}
/*
* Counts the occurence of each key in my bucket.
* Key indices into the count array are the key's value minus my bucket's
* minimum key value to allow indexing from 0.
* my_bucket_keys: All keys in my bucket unsorted [my_rank * BUCKET_WIDTH, (my_rank+1)*BUCKET_WIDTH)
*/
static int * count_local_keys(KEY_TYPE const * const my_bucket_keys)
{
int * const my_local_key_counts = malloc(BUCKET_WIDTH * sizeof(int));
assert(my_local_key_counts);
memset(my_local_key_counts, 0, BUCKET_WIDTH * sizeof(int));
timer_start(&timers[TIMER_SORT]);
const int my_rank = shmem_my_pe();
const int my_min_key = my_rank * BUCKET_WIDTH;
#ifdef ISX_PROFILING
unsigned long long start = current_time_ns();
#endif
// Count the occurences of each key in my bucket
for(long long int i = 0; i < my_bucket_size; ++i){
const unsigned int key_index = my_bucket_keys[i] - my_min_key;
assert(my_bucket_keys[i] >= my_min_key);
assert(key_index < BUCKET_WIDTH);
my_local_key_counts[key_index]++;
}
#ifdef ISX_PROFILING
unsigned long long end = current_time_ns();
if (shmem_my_pe() == 0)
printf("Counting local took %llu ns, my_bucket_size = %u, BUCKET_WIDTH = "
"%llu\n", end - start, my_bucket_size, BUCKET_WIDTH);
#endif
timer_stop(&timers[TIMER_SORT]);
#ifdef DEBUG
wait_my_turn();
char msg[4096];
sprintf(msg,"Rank %d: Bucket Size %lld | Local Key Counts:", my_rank, my_bucket_size);
for(uint64_t i = 0; i < BUCKET_WIDTH; ++i){
if(i < PRINT_MAX)
sprintf(msg + strlen(msg),"%d ", my_local_key_counts[i]);
}
sprintf(msg + strlen(msg),"\n");
printf("%s",msg);
fflush(stdout);
my_turn_complete();
#endif
return my_local_key_counts;
}
/*
* Places local keys into their corresponding local bucket.
* The contents of each bucket are not sorted.
*/
static KEY_TYPE * bucketize_local_keys(KEY_TYPE const * const my_keys,
int * const local_bucket_offsets)
{
KEY_TYPE * const my_local_bucketed_keys = malloc(NUM_KEYS_PER_PE * sizeof(KEY_TYPE));
assert(my_local_bucketed_keys);
timer_start(&timers[TIMER_BUCKETIZE]);
#ifdef ISX_PROFILING
unsigned long long start = current_time_ns();
#endif
for(uint64_t i = 0; i < NUM_KEYS_PER_PE; ++i){
const KEY_TYPE key = my_keys[i];
const uint32_t bucket_index = key / BUCKET_WIDTH;
uint32_t index;
assert(local_bucket_offsets[bucket_index] >= 0);
index = local_bucket_offsets[bucket_index]++;
assert(index < NUM_KEYS_PER_PE);
my_local_bucketed_keys[index] = key;
}
#ifdef ISX_PROFILING
unsigned long long end = current_time_ns();
if (shmem_my_pe() == 0)
printf("Bucketizing took %llu ns\n", end - start);
#endif
timer_stop(&timers[TIMER_BUCKETIZE]);
#ifdef DEBUG
wait_my_turn();
char msg[1024];
const int my_rank = shmem_my_pe();
sprintf(msg,"Rank %d: local bucketed keys: ", my_rank);
for(uint64_t i = 0; i < NUM_KEYS_PER_PE; ++i){
if(i < PRINT_MAX)
sprintf(msg + strlen(msg),"%d ", my_local_bucketed_keys[i]);
}
sprintf(msg + strlen(msg),"\n");
printf("%s",msg);
fflush(stdout);
my_turn_complete();
#endif
return my_local_bucketed_keys;
}
void time_step(int j, int nelr, double* old_variables, double* variables, double* step_factors, double* fluxes)
{
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for default(shared) schedule(static)
for(int i = 0; i < nelr; i++)
{
double factor = step_factors[i]/double(RK+1-j);
variables[NVAR*i + VAR_DENSITY] = old_variables[NVAR*i + VAR_DENSITY] + factor*fluxes[NVAR*i + VAR_DENSITY];
variables[NVAR*i + VAR_DENSITY_ENERGY] = old_variables[NVAR*i + VAR_DENSITY_ENERGY] + factor*fluxes[NVAR*i + VAR_DENSITY_ENERGY];
variables[NVAR*i + (VAR_MOMENTUM+0)] = old_variables[NVAR*i + (VAR_MOMENTUM+0)] + factor*fluxes[NVAR*i + (VAR_MOMENTUM+0)];
variables[NVAR*i + (VAR_MOMENTUM+1)] = old_variables[NVAR*i + (VAR_MOMENTUM+1)] + factor*fluxes[NVAR*i + (VAR_MOMENTUM+1)];
variables[NVAR*i + (VAR_MOMENTUM+2)] = old_variables[NVAR*i + (VAR_MOMENTUM+2)] + factor*fluxes[NVAR*i + (VAR_MOMENTUM+2)];
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma317_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
}
/*
* Computes the size of each bucket by iterating all keys and incrementing
* their corresponding bucket's size
*/
static int * count_local_bucket_sizes(KEY_TYPE const * const my_keys)
{
int * const local_bucket_sizes = malloc(NUM_BUCKETS * sizeof(int));
assert(local_bucket_sizes);
timer_start(&timers[TIMER_BCOUNT]);
init_array(local_bucket_sizes, NUM_BUCKETS);
#ifdef ISX_PROFILING
unsigned long long start = current_time_ns();
#endif
for(uint64_t i = 0; i < NUM_KEYS_PER_PE; ++i){
const uint32_t bucket_index = my_keys[i]/BUCKET_WIDTH;
local_bucket_sizes[bucket_index]++;
}
#ifdef ISX_PROFILING
unsigned long long end = current_time_ns();
if (shmem_my_pe() == 0)
printf("Counting local bucket sizes took %llu ns\n", end - start);
#endif
timer_stop(&timers[TIMER_BCOUNT]);
#ifdef DEBUG
wait_my_turn();
char msg[1024];
const int my_rank = shmem_my_pe();
sprintf(msg,"Rank %d: local bucket sizes: ", my_rank);
for(uint64_t i = 0; i < NUM_BUCKETS; ++i){
if(i < PRINT_MAX)
sprintf(msg + strlen(msg),"%d ", local_bucket_sizes[i]);
}
sprintf(msg + strlen(msg),"\n");
printf("%s",msg);
fflush(stdout);
my_turn_complete();
#endif
return local_bucket_sizes;
}
void hclib_launch(generic_frame_ptr fct_ptr, void *arg, const char **deps,
int ndeps) {
unsigned long long start_time = 0;
unsigned long long end_time;
const int instrument = (getenv("HCLIB_INSTRUMENT") != NULL);
hclib_init(deps, ndeps, instrument);
if (profile_launch_body) {
start_time = current_time_ns();
}
hclib_async(fct_ptr, arg, NULL, 0, hclib_get_closest_locale());
hclib_finalize(instrument);
if (profile_launch_body) {
end_time = current_time_ns();
printf("\nHCLIB TIME %llu ns\n", end_time - start_time);
}
}
void sim_village_main_par(struct Village *top)
{
long i;
const unsigned long long full_program_start = current_time_ns();
{
#pragma omp parallel
{
#pragma omp single
{
#pragma omp task untied
{
for (i = 0; i < sim_time; i++) sim_village_par(top);
}
}
}
} ;
const unsigned long long full_program_end = current_time_ns();
printf("full_program %llu ns\n", full_program_end - full_program_start);
}
/*
* Generates uniformly random keys [0, MAX_KEY_VAL] on each rank using the time and rank
* number as a seed
*/
static KEY_TYPE * make_input(void)
{
timer_start(&timers[TIMER_INPUT]);
KEY_TYPE * const my_keys = malloc(NUM_KEYS_PER_PE * sizeof(KEY_TYPE));
assert(my_keys);
pcg32_random_t rng = seed_my_rank();
#ifdef ISX_PROFILING
unsigned long long start = current_time_ns();
#endif
for(uint64_t i = 0; i < NUM_KEYS_PER_PE; ++i) {
my_keys[i] = pcg32_boundedrand_r(&rng, MAX_KEY_VAL);
}
#ifdef ISX_PROFILING
unsigned long long end = current_time_ns();
if (shmem_my_pe() == 0)
printf("Making input took %llu ns\n", end - start);
#endif
timer_stop(&timers[TIMER_INPUT]);
#ifdef DEBUG
wait_my_turn();
char msg[1024];
const int my_rank = shmem_my_pe();
sprintf(msg,"Rank %d: Initial Keys: ", my_rank);
for(uint64_t i = 0; i < NUM_KEYS_PER_PE; ++i){
if(i < PRINT_MAX)
sprintf(msg + strlen(msg),"%d ", my_keys[i]);
}
sprintf(msg + strlen(msg),"\n");
printf("%s",msg);
fflush(stdout);
my_turn_complete();
#endif
return my_keys;
}
/*---< main() >-------------------------------------------------------------*/
int main(int argc, char **argv) {
int opt;
extern char *optarg;
extern int optind;
int nclusters=5;
char *filename = 0;
float *buf;
float *attributes;
float *cluster_centres=NULL;
int i, j;
int numAttributes;
int numObjects;
char line[1024];
int isBinaryFile = 0;
int nloops = 1;
float threshold = 0.001;
double timing;
while ( (opt=getopt(argc,argv,"i:k:t:b:n:?"))!= EOF) {
switch (opt) {
case 'i': filename=optarg;
break;
case 'b': isBinaryFile = 1;
break;
case 't': threshold=atof(optarg);
break;
case 'k': nclusters = atoi(optarg);
break;
case '?': usage(argv[0]);
break;
default: usage(argv[0]);
break;
}
}
if (filename == 0) usage(argv[0]);
numAttributes = numObjects = 0;
/* from the input file, get the numAttributes and numObjects ------------*/
if (isBinaryFile) {
int infile;
if ((infile = open(filename, O_RDONLY, "0600")) == -1) {
fprintf(stderr, "Error: no such file (%s)\n", filename);
exit(1);
}
read(infile, &numObjects, sizeof(int));
read(infile, &numAttributes, sizeof(int));
/* allocate space for attributes[] and read attributes of all objects */
attributes = (float*) malloc(numObjects * numAttributes * sizeof(float));
read(infile, attributes, numObjects*numAttributes*sizeof(float));
close(infile);
}
else {
FILE *infile;
if ((infile = fopen(filename, "r")) == NULL) {
fprintf(stderr, "Error: no such file (%s)\n", filename);
exit(1);
}
while (fgets(line, 1024, infile) != NULL)
if (strtok(line, " \t\n") != 0)
numObjects++;
rewind(infile);
while (fgets(line, 1024, infile) != NULL) {
if (strtok(line, " \t\n") != 0) {
/* ignore the id (first attribute): numAttributes = 1; */
while (strtok(NULL, " ,\t\n") != NULL) numAttributes++;
break;
}
}
/* allocate space for attributes[] and read attributes of all objects */
attributes = (float*) malloc(numObjects*numAttributes*sizeof(float));
rewind(infile);
i = 0;
while (fgets(line, 1024, infile) != NULL) {
if (strtok(line, " \t\n") == NULL) continue;
for (j=0; j<numAttributes; j++) {
attributes[i] = atof(strtok(NULL, " ,\t\n"));
i++;
}
}
fclose(infile);
}
printf("I/O completed\n");
const unsigned long long full_program_start = current_time_ns();
for (i=0; i<nloops; i++) {
cluster_centres = NULL;
cluster(numObjects,
numAttributes,
attributes, /* [numObjects][numAttributes] */
nclusters,
threshold,
&cluster_centres
);
} ;
const unsigned long long full_program_end = current_time_ns();
printf("full_program %llu ns\n", full_program_end - full_program_start);
printf("number of Clusters %d\n",nclusters);
printf("number of Attributes %d\n\n",numAttributes);
/* printf("Cluster Centers Output\n");
printf("The first number is cluster number and the following data is arribute value\n");
printf("=============================================================================\n\n");
for (i=0; i< nclusters; i++) {
printf("%d: ", i);
for (j=0; j<numAttributes; j++)
printf("%.2f ", cluster_centres[i][j]);
printf("\n\n");
}
*/
free(attributes);
return(0);
}
int main()
{
int l;
uint64_t start, end;
long int diff = 0;
int i, j, k;
long double h, t1, t2, dppi, ans = 5.795776322412856L;
long double s1;
// variables for logging/checking
long double log[INPUTS];
long double threshold = 0.0;
long double epsilon = -4.0;
// 0. read input from the file final_inputs
int finputs[INPUTS];
FILE* infile = fopen("final_inputs", "r");
if (!infile)
{
printf("Could not open final_inputs\n");
}
char *s = malloc(10);
for (i = 0; i < INPUTS; i++)
{
if (!feof(infile))
{
fscanf(infile, "%s", s);
finputs[i] = (int)cov_deserialize(s, 10);
}
}
// dummy calls
sqrtf(0);
acosf(0);
sinf(0);
start = current_time_ns();
for (l = 0; l < INPUTS; l++)
{
int n = finputs[l];
t1 = -1.0;
dppi = acos(t1);
s1 = 0.0;
t1 = 0.0;
h = dppi / n;
for( i = 1; i <= n; i++ )
{
t2 = fun (i * h);
s1 = s1 + sqrt (h*h + (t2 - t1)*(t2 - t1));
t1 = t2;
}
// 1. compute threshold and record result
log[l] = (long double) s1;
if (s1*pow(10, epsilon) > threshold)
{
threshold = s1*pow(10, epsilon);
}
}
end = current_time_ns();
diff = (end-start);
// 2. create spec, or checking results
cov_arr_spec_log("spec.cov", threshold, INPUTS, log);
cov_arr_log(log, INPUTS, "result", "log.cov");
cov_check("log.cov", "spec.cov", INPUTS);
// 3. print score (diff) to a file
FILE* file;
file = fopen("score.cov", "w");
fprintf(file, "%ld\n", diff);
fclose(file);
return 0;
}
/*
* Main function
*/
int main(int argc, char** argv)
{
if (argc < 2)
{
std::cout << "specify data file name" << std::endl;
return 0;
}
const char* data_file_name = argv[1];
const unsigned long long full_program_start = current_time_ns();
{
// set far field conditions
{
const double angle_of_attack = double(3.1415926535897931 / 180.0) * double(deg_angle_of_attack);
ff_variable[VAR_DENSITY] = double(1.4);
double ff_pressure = double(1.0);
double ff_speed_of_sound = sqrt(GAMMA*ff_pressure / ff_variable[VAR_DENSITY]);
double ff_speed = double(ff_mach)*ff_speed_of_sound;
cfd_double3 ff_velocity;
ff_velocity.x = ff_speed*double(cos((double)angle_of_attack));
ff_velocity.y = ff_speed*double(sin((double)angle_of_attack));
ff_velocity.z = 0.0;
ff_variable[VAR_MOMENTUM+0] = ff_variable[VAR_DENSITY] * ff_velocity.x;
ff_variable[VAR_MOMENTUM+1] = ff_variable[VAR_DENSITY] * ff_velocity.y;
ff_variable[VAR_MOMENTUM+2] = ff_variable[VAR_DENSITY] * ff_velocity.z;
ff_variable[VAR_DENSITY_ENERGY] = ff_variable[VAR_DENSITY]*(double(0.5)*(ff_speed*ff_speed)) + (ff_pressure / double(GAMMA-1.0));
cfd_double3 ff_momentum;
ff_momentum.x = *(ff_variable+VAR_MOMENTUM+0);
ff_momentum.y = *(ff_variable+VAR_MOMENTUM+1);
ff_momentum.z = *(ff_variable+VAR_MOMENTUM+2);
compute_flux_contribution(ff_variable[VAR_DENSITY], ff_momentum, ff_variable[VAR_DENSITY_ENERGY], ff_pressure, ff_velocity, ff_flux_contribution_momentum_x, ff_flux_contribution_momentum_y, ff_flux_contribution_momentum_z, ff_flux_contribution_density_energy);
}
int nel;
int nelr;
// read in domain geometry
double* areas;
int* elements_surrounding_elements;
double* normals;
{
std::ifstream file(data_file_name);
file >> nel;
nelr = block_length*((nel / block_length )+ std::min(1, nel % block_length));
areas = new double[nelr];
elements_surrounding_elements = new int[nelr*NNB];
normals = new double[NDIM*NNB*nelr];
// read in data
for(int i = 0; i < nel; i++)
{
file >> areas[i];
for(int j = 0; j < NNB; j++)
{
file >> elements_surrounding_elements[i*NNB + j];
if(elements_surrounding_elements[i*NNB+j] < 0) elements_surrounding_elements[i*NNB+j] = -1;
elements_surrounding_elements[i*NNB + j]--; //it's coming in with Fortran numbering
for(int k = 0; k < NDIM; k++)
{
file >> normals[(i*NNB + j)*NDIM + k];
normals[(i*NNB + j)*NDIM + k] = -normals[(i*NNB + j)*NDIM + k];
}
}
}
// fill in remaining data
int last = nel-1;
for(int i = nel; i < nelr; i++)
{
areas[i] = areas[last];
for(int j = 0; j < NNB; j++)
{
// duplicate the last element
elements_surrounding_elements[i*NNB + j] = elements_surrounding_elements[last*NNB + j];
for(int k = 0; k < NDIM; k++) normals[(i*NNB + j)*NDIM + k] = normals[(last*NNB + j)*NDIM + k];
}
}
}
// Create arrays and set initial conditions
double* variables = alloc<double>(nelr*NVAR);
initialize_variables(nelr, variables);
double* old_variables = alloc<double>(nelr*NVAR);
double* fluxes = alloc<double>(nelr*NVAR);
double* step_factors = alloc<double>(nelr);
// these need to be computed the first time in order to compute time step
std::cout << "Starting..." << std::endl;
// Begin iterations
for(int i = 0; i < iterations; i++)
{
copy(old_variables, variables, nelr*NVAR);
// for the first iteration we compute the time step
compute_step_factor(nelr, variables, areas, step_factors);
for(int j = 0; j < RK; j++)
{
compute_flux(nelr, elements_surrounding_elements, normals, variables, fluxes);
time_step(j, nelr, old_variables, variables, step_factors, fluxes);
}
}
std::cout << "Saving solution..." << std::endl;
dump(variables, nel, nelr);
std::cout << "Saved solution..." << std::endl;
std::cout << "Cleaning up..." << std::endl;
dealloc<double>(areas);
dealloc<int>(elements_surrounding_elements);
dealloc<double>(normals);
dealloc<double>(variables);
dealloc<double>(old_variables);
dealloc<double>(fluxes);
dealloc<double>(step_factors);
} ;
const unsigned long long full_program_end = current_time_ns();
printf("full_program %llu ns\n", full_program_end - full_program_start);
std::cout << "Done..." << std::endl;
return 0;
}
int main (int argc, char *argv[]) {
/**** Initialising ****/
const unsigned long long full_program_start = current_time_ns();
{
shmem_init ();
/* Variable Declarations */
int Numprocs,MyRank, Root = 0;
int i,j,k, NoofElements, NoofElements_Bloc,
NoElementsToSort;
int count, temp;
TYPE *Input, *InputData;
TYPE *Splitter, *AllSplitter;
TYPE *Buckets, *BucketBuffer, *LocalBucket;
TYPE *OutputBuffer, *Output;
MyRank = shmem_my_pe ();
Numprocs = shmem_n_pes ();
NoofElements = SIZE;
if(( NoofElements % Numprocs) != 0){
if(MyRank == Root)
printf("Number of Elements are not divisible by Numprocs \n");
shmem_finalize ();
exit(0);
}
/**** Reading Input ****/
Input = (TYPE *) shmem_malloc (NoofElements*sizeof(*Input));
if(Input == NULL) {
printf("Error : Can not allocate memory \n");
}
if (MyRank == Root){
/* Initialise random number generator */
printf ("Generating input Array for Sorting %d uint64_t numbers\n",SIZE);
srand48((TYPE)NoofElements);
for(i=0; i< NoofElements; i++) {
Input[i] = rand();
}
}
/**** Sending Data ****/
NoofElements_Bloc = NoofElements / Numprocs;
InputData = (TYPE *) shmem_malloc (NoofElements_Bloc * sizeof (*InputData));
if(InputData == NULL) {
printf("Error : Can not allocate memory \n");
}
//MPI_Scatter(Input, NoofElements_Bloc, TYPE_MPI, InputData,
// NoofElements_Bloc, TYPE_MPI, Root, MPI_COMM_WORLD);
shmem_barrier_all();
if(MyRank == Root) {
for(i=0; i<Numprocs; i++) {
TYPE* start = &Input[i * NoofElements_Bloc];
shmem_put64(InputData, start, NoofElements_Bloc, i);
}
}
shmem_barrier_all();
/**** Sorting Locally ****/
sorting(InputData, NoofElements_Bloc);
/**** Choosing Local Splitters ****/
Splitter = (TYPE *) shmem_malloc (sizeof (TYPE) * (Numprocs-1));
if(Splitter == NULL) {
printf("Error : Can not allocate memory \n");
}
for (i=0; i< (Numprocs-1); i++){
Splitter[i] = InputData[NoofElements/(Numprocs*Numprocs) * (i+1)];
}
/**** Gathering Local Splitters at Root ****/
AllSplitter = (TYPE *) shmem_malloc (sizeof (TYPE) * Numprocs * (Numprocs-1));
if(AllSplitter == NULL) {
printf("Error : Can not allocate memory \n");
}
//MPI_Gather (Splitter, Numprocs-1, TYPE_MPI, AllSplitter, Numprocs-1,
// TYPE_MPI, Root, MPI_COMM_WORLD);
shmem_barrier_all();
TYPE* target_index = &AllSplitter[MyRank * (Numprocs-1)];
shmem_put64(target_index, Splitter, Numprocs-1, Root);
shmem_barrier_all();
/**** Choosing Global Splitters ****/
if (MyRank == Root){
sorting (AllSplitter, Numprocs*(Numprocs-1));
for (i=0; i<Numprocs-1; i++)
Splitter[i] = AllSplitter[(Numprocs-1)*(i+1)];
}
/**** Broadcasting Global Splitters ****/
//MPI_Bcast (Splitter, Numprocs-1, TYPE_MPI, 0, MPI_COMM_WORLD);
{ int _i; for(_i=0; _i<_SHMEM_BCAST_SYNC_SIZE; _i++) { pSync[_i] = _SHMEM_SYNC_VALUE; } shmem_barrier_all(); }
shmem_broadcast64(Splitter, Splitter, Numprocs-1, 0, 0, 0, Numprocs, pSync);
shmem_barrier_all();
/**** Creating Numprocs Buckets locally ****/
Buckets = (TYPE *) shmem_malloc (sizeof (TYPE) * (NoofElements + Numprocs));
if(Buckets == NULL) {
printf("Error : Can not allocate memory \n");
}
j = 0;
k = 1;
for (i=0; i<NoofElements_Bloc; i++){
if(j < (Numprocs-1)){
if (InputData[i] < Splitter[j])
Buckets[((NoofElements_Bloc + 1) * j) + k++] = InputData[i];
else{
Buckets[(NoofElements_Bloc + 1) * j] = k-1;
k=1;
j++;
i--;
}
}
else
Buckets[((NoofElements_Bloc + 1) * j) + k++] = InputData[i];
}
Buckets[(NoofElements_Bloc + 1) * j] = k - 1;
shmem_free(Splitter);
shmem_free(AllSplitter);
/**** Sending buckets to respective processors ****/
BucketBuffer = (TYPE *) shmem_malloc (sizeof (TYPE) * (NoofElements + Numprocs));
if(BucketBuffer == NULL) {
printf("Error : Can not allocate memory \n");
}
//MPI_Alltoall (Buckets, NoofElements_Bloc + 1, TYPE_MPI, BucketBuffer,
// NoofElements_Bloc + 1, TYPE_MPI, MPI_COMM_WORLD);
shmem_barrier_all();
for(i=0; i<Numprocs; i++) {
shmem_put64(&BucketBuffer[MyRank*(NoofElements_Bloc + 1)], &Buckets[i*(NoofElements_Bloc + 1)], NoofElements_Bloc + 1, i);
}
shmem_barrier_all();
/**** Rearranging BucketBuffer ****/
LocalBucket = (TYPE *) shmem_malloc (sizeof (TYPE) * 2 * NoofElements / Numprocs);
if(LocalBucket == NULL) {
printf("Error : Can not allocate memory \n");
}
count = 1;
for (j=0; j<Numprocs; j++) {
k = 1;
for (i=0; i<BucketBuffer[(NoofElements/Numprocs + 1) * j]; i++)
LocalBucket[count++] = BucketBuffer[(NoofElements/Numprocs + 1) * j + k++];
}
LocalBucket[0] = count-1;
/**** Sorting Local Buckets using Bubble Sort ****/
/*sorting (InputData, NoofElements_Bloc, sizeof(int), intcompare); */
NoElementsToSort = LocalBucket[0];
sorting (&LocalBucket[1], NoElementsToSort);
/**** Gathering sorted sub blocks at root ****/
OutputBuffer = (TYPE *) shmem_malloc (sizeof(TYPE) * 2 * NoofElements);
if(OutputBuffer == NULL) {
printf("Error : Can not allocate memory \n");
}
//MPI_Gather (LocalBucket, 2*NoofElements_Bloc, TYPE_MPI, OutputBuffer,
// 2*NoofElements_Bloc, TYPE_MPI, Root, MPI_COMM_WORLD);
shmem_barrier_all();
target_index = &OutputBuffer[MyRank * (2*NoofElements_Bloc)];
shmem_put64(target_index, LocalBucket, 2*NoofElements_Bloc, Root);
shmem_barrier_all();
/**** Rearranging output buffer ****/
if (MyRank == Root){
Output = (TYPE *) malloc (sizeof (TYPE) * NoofElements);
count = 0;
for(j=0; j<Numprocs; j++){
k = 1;
for(i=0; i<OutputBuffer[(2 * NoofElements/Numprocs) * j]; i++)
Output[count++] = OutputBuffer[(2*NoofElements/Numprocs) * j + k++];
}
printf ( "Number of Elements to be sorted : %d \n", NoofElements);
TYPE prev = 0;
int fail = 0;
for (i=0; i<NoofElements; i++){
if(Output[i] < prev) { printf("Failed at index %d\n",i); fail = 1; }
prev = Output[i];
}
if(fail) printf("Sorting FAILED\n");
else printf("Sorting PASSED\n");
free(Output);
}/* MyRank==0*/
shmem_free(Input);
shmem_free(OutputBuffer);
shmem_free(InputData);
shmem_free(Buckets);
shmem_free(BucketBuffer);
shmem_free(LocalBucket);
/**** Finalize ****/
shmem_finalize();
} ;
const unsigned long long full_program_end = current_time_ns();
printf("full_program %llu ns\n", full_program_end - full_program_start);
}
void compute_flux(int nelr, int* elements_surrounding_elements, double* normals, double* variables, double* fluxes)
{
double smoothing_coefficient = double(0.2f);
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for default(shared) schedule(static)
for(int i = 0; i < nelr; i++)
{
int j, nb;
cfd_double3 normal; double normal_len;
double factor;
double density_i = variables[NVAR*i + VAR_DENSITY];
cfd_double3 momentum_i;
momentum_i.x = variables[NVAR*i + (VAR_MOMENTUM+0)];
momentum_i.y = variables[NVAR*i + (VAR_MOMENTUM+1)];
momentum_i.z = variables[NVAR*i + (VAR_MOMENTUM+2)];
double density_energy_i = variables[NVAR*i + VAR_DENSITY_ENERGY];
cfd_double3 velocity_i; compute_velocity(density_i, momentum_i, velocity_i);
double speed_sqd_i = compute_speed_sqd(velocity_i);
double speed_i = std::sqrt(speed_sqd_i);
double pressure_i = compute_pressure(density_i, density_energy_i, speed_sqd_i);
double speed_of_sound_i = compute_speed_of_sound(density_i, pressure_i);
cfd_double3 flux_contribution_i_momentum_x, flux_contribution_i_momentum_y, flux_contribution_i_momentum_z;
cfd_double3 flux_contribution_i_density_energy;
compute_flux_contribution(density_i, momentum_i, density_energy_i, pressure_i, velocity_i, flux_contribution_i_momentum_x, flux_contribution_i_momentum_y, flux_contribution_i_momentum_z, flux_contribution_i_density_energy);
double flux_i_density = double(0.0);
cfd_double3 flux_i_momentum;
flux_i_momentum.x = double(0.0);
flux_i_momentum.y = double(0.0);
flux_i_momentum.z = double(0.0);
double flux_i_density_energy = double(0.0);
cfd_double3 velocity_nb;
double density_nb, density_energy_nb;
cfd_double3 momentum_nb;
cfd_double3 flux_contribution_nb_momentum_x, flux_contribution_nb_momentum_y, flux_contribution_nb_momentum_z;
cfd_double3 flux_contribution_nb_density_energy;
double speed_sqd_nb, speed_of_sound_nb, pressure_nb;
for(j = 0; j < NNB; j++)
{
nb = elements_surrounding_elements[i*NNB + j];
normal.x = normals[(i*NNB + j)*NDIM + 0];
normal.y = normals[(i*NNB + j)*NDIM + 1];
normal.z = normals[(i*NNB + j)*NDIM + 2];
normal_len = std::sqrt(normal.x*normal.x + normal.y*normal.y + normal.z*normal.z);
if(nb >= 0) // a legitimate neighbor
{
density_nb = variables[nb*NVAR + VAR_DENSITY];
momentum_nb.x = variables[nb*NVAR + (VAR_MOMENTUM+0)];
momentum_nb.y = variables[nb*NVAR + (VAR_MOMENTUM+1)];
momentum_nb.z = variables[nb*NVAR + (VAR_MOMENTUM+2)];
density_energy_nb = variables[nb*NVAR + VAR_DENSITY_ENERGY];
compute_velocity(density_nb, momentum_nb, velocity_nb);
speed_sqd_nb = compute_speed_sqd(velocity_nb);
pressure_nb = compute_pressure(density_nb, density_energy_nb, speed_sqd_nb);
speed_of_sound_nb = compute_speed_of_sound(density_nb, pressure_nb);
compute_flux_contribution(density_nb, momentum_nb, density_energy_nb, pressure_nb, velocity_nb, flux_contribution_nb_momentum_x, flux_contribution_nb_momentum_y, flux_contribution_nb_momentum_z, flux_contribution_nb_density_energy);
// artificial viscosity
factor = -normal_len*smoothing_coefficient*double(0.5)*(speed_i + std::sqrt(speed_sqd_nb) + speed_of_sound_i + speed_of_sound_nb);
flux_i_density += factor*(density_i-density_nb);
flux_i_density_energy += factor*(density_energy_i-density_energy_nb);
flux_i_momentum.x += factor*(momentum_i.x-momentum_nb.x);
flux_i_momentum.y += factor*(momentum_i.y-momentum_nb.y);
flux_i_momentum.z += factor*(momentum_i.z-momentum_nb.z);
// accumulate cell-centered fluxes
factor = double(0.5)*normal.x;
flux_i_density += factor*(momentum_nb.x+momentum_i.x);
flux_i_density_energy += factor*(flux_contribution_nb_density_energy.x+flux_contribution_i_density_energy.x);
flux_i_momentum.x += factor*(flux_contribution_nb_momentum_x.x+flux_contribution_i_momentum_x.x);
flux_i_momentum.y += factor*(flux_contribution_nb_momentum_y.x+flux_contribution_i_momentum_y.x);
flux_i_momentum.z += factor*(flux_contribution_nb_momentum_z.x+flux_contribution_i_momentum_z.x);
factor = double(0.5)*normal.y;
flux_i_density += factor*(momentum_nb.y+momentum_i.y);
flux_i_density_energy += factor*(flux_contribution_nb_density_energy.y+flux_contribution_i_density_energy.y);
flux_i_momentum.x += factor*(flux_contribution_nb_momentum_x.y+flux_contribution_i_momentum_x.y);
flux_i_momentum.y += factor*(flux_contribution_nb_momentum_y.y+flux_contribution_i_momentum_y.y);
flux_i_momentum.z += factor*(flux_contribution_nb_momentum_z.y+flux_contribution_i_momentum_z.y);
factor = double(0.5)*normal.z;
flux_i_density += factor*(momentum_nb.z+momentum_i.z);
flux_i_density_energy += factor*(flux_contribution_nb_density_energy.z+flux_contribution_i_density_energy.z);
flux_i_momentum.x += factor*(flux_contribution_nb_momentum_x.z+flux_contribution_i_momentum_x.z);
flux_i_momentum.y += factor*(flux_contribution_nb_momentum_y.z+flux_contribution_i_momentum_y.z);
flux_i_momentum.z += factor*(flux_contribution_nb_momentum_z.z+flux_contribution_i_momentum_z.z);
}
else if(nb == -1) // a wing boundary
{
flux_i_momentum.x += normal.x*pressure_i;
flux_i_momentum.y += normal.y*pressure_i;
flux_i_momentum.z += normal.z*pressure_i;
}
else if(nb == -2) // a far field boundary
{
factor = double(0.5)*normal.x;
flux_i_density += factor*(ff_variable[VAR_MOMENTUM+0]+momentum_i.x);
flux_i_density_energy += factor*(ff_flux_contribution_density_energy.x+flux_contribution_i_density_energy.x);
flux_i_momentum.x += factor*(ff_flux_contribution_momentum_x.x + flux_contribution_i_momentum_x.x);
flux_i_momentum.y += factor*(ff_flux_contribution_momentum_y.x + flux_contribution_i_momentum_y.x);
flux_i_momentum.z += factor*(ff_flux_contribution_momentum_z.x + flux_contribution_i_momentum_z.x);
factor = double(0.5)*normal.y;
flux_i_density += factor*(ff_variable[VAR_MOMENTUM+1]+momentum_i.y);
flux_i_density_energy += factor*(ff_flux_contribution_density_energy.y+flux_contribution_i_density_energy.y);
flux_i_momentum.x += factor*(ff_flux_contribution_momentum_x.y + flux_contribution_i_momentum_x.y);
flux_i_momentum.y += factor*(ff_flux_contribution_momentum_y.y + flux_contribution_i_momentum_y.y);
flux_i_momentum.z += factor*(ff_flux_contribution_momentum_z.y + flux_contribution_i_momentum_z.y);
factor = double(0.5)*normal.z;
flux_i_density += factor*(ff_variable[VAR_MOMENTUM+2]+momentum_i.z);
flux_i_density_energy += factor*(ff_flux_contribution_density_energy.z+flux_contribution_i_density_energy.z);
flux_i_momentum.x += factor*(ff_flux_contribution_momentum_x.z + flux_contribution_i_momentum_x.z);
flux_i_momentum.y += factor*(ff_flux_contribution_momentum_y.z + flux_contribution_i_momentum_y.z);
flux_i_momentum.z += factor*(ff_flux_contribution_momentum_z.z + flux_contribution_i_momentum_z.z);
}
}
fluxes[i*NVAR + VAR_DENSITY] = flux_i_density;
fluxes[i*NVAR + (VAR_MOMENTUM+0)] = flux_i_momentum.x;
fluxes[i*NVAR + (VAR_MOMENTUM+1)] = flux_i_momentum.y;
fluxes[i*NVAR + (VAR_MOMENTUM+2)] = flux_i_momentum.z;
fluxes[i*NVAR + VAR_DENSITY_ENERGY] = flux_i_density_energy;
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma186_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
}
unsigned long long hclib_current_time_ns() {
return current_time_ns();
}
/**
* The implementation of the particle filter using OpenMP for many frames
* @see http://openmp.org/wp/
* @note This function is designed to work with a video of several frames. In addition, it references a provided MATLAB function which takes the video, the objxy matrix and the x and y arrays as arguments and returns the likelihoods
* @param I The video to be run
* @param IszX The x dimension of the video
* @param IszY The y dimension of the video
* @param Nfr The number of frames
* @param seed The seed array used for random number generation
* @param Nparticles The number of particles to be used
*/
void particleFilter(int * I, int IszX, int IszY, int Nfr, int * seed, int Nparticles){
int max_size = IszX*IszY*Nfr;
long long start = get_time();
//original particle centroid
double xe = roundDouble(IszY/2.0);
double ye = roundDouble(IszX/2.0);
//expected object locations, compared to center
int radius = 5;
int diameter = radius*2 - 1;
int * disk = (int *)malloc(diameter*diameter*sizeof(int));
strelDisk(disk, radius);
int countOnes = 0;
int x, y;
for(x = 0; x < diameter; x++){
for(y = 0; y < diameter; y++){
if(disk[x*diameter + y] == 1)
countOnes++;
}
}
double * objxy = (double *)malloc(countOnes*2*sizeof(double));
getneighbors(disk, countOnes, objxy, radius);
long long get_neighbors = get_time();
printf("TIME TO GET NEIGHBORS TOOK: %f\n", elapsed_time(start, get_neighbors));
//initial weights are all equal (1/Nparticles)
double * weights = (double *)malloc(sizeof(double)*Nparticles);
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(weights, Nparticles) private(x)
for(x = 0; x < Nparticles; x++){
weights[x] = 1/((double)(Nparticles));
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma373_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long get_weights = get_time();
printf("TIME TO GET WEIGHTSTOOK: %f\n", elapsed_time(get_neighbors, get_weights));
//initial likelihood to 0.0
double * likelihood = (double *)malloc(sizeof(double)*Nparticles);
double * arrayX = (double *)malloc(sizeof(double)*Nparticles);
double * arrayY = (double *)malloc(sizeof(double)*Nparticles);
double * xj = (double *)malloc(sizeof(double)*Nparticles);
double * yj = (double *)malloc(sizeof(double)*Nparticles);
double * CDF = (double *)malloc(sizeof(double)*Nparticles);
double * u = (double *)malloc(sizeof(double)*Nparticles);
int * ind = (int*)malloc(sizeof(int)*countOnes*Nparticles);
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(arrayX, arrayY, xe, ye) private(x)
for(x = 0; x < Nparticles; x++){
arrayX[x] = xe;
arrayY[x] = ye;
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma388_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
int k;
printf("TIME TO SET ARRAYS TOOK: %f\n", elapsed_time(get_weights, get_time()));
int indX, indY;
for(k = 1; k < Nfr; k++){
long long set_arrays = get_time();
//apply motion model
//draws sample from motion model (random walk). The only prior information
//is that the object moves 2x as fast as in the y direction
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(arrayX, arrayY, Nparticles, seed) private(x)
for(x = 0; x < Nparticles; x++){
arrayX[x] += 1 + 5*randn(seed, x);
arrayY[x] += -2 + 2*randn(seed, x);
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma402_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long error = get_time();
printf("TIME TO SET ERROR TOOK: %f\n", elapsed_time(set_arrays, error));
//particle filter likelihood
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(likelihood, I, arrayX, arrayY, objxy, ind) private(x, y, indX, indY)
for(x = 0; x < Nparticles; x++){
//compute the likelihood: remember our assumption is that you know
// foreground and the background image intensity distribution.
// Notice that we consider here a likelihood ratio, instead of
// p(z|x). It is possible in this case. why? a hometask for you.
//calc ind
for(y = 0; y < countOnes; y++){
indX = roundDouble(arrayX[x]) + objxy[y*2 + 1];
indY = roundDouble(arrayY[x]) + objxy[y*2];
ind[x*countOnes + y] = fabs((double)(indX*IszY*Nfr + indY*Nfr + k));
if(ind[x*countOnes + y] >= max_size)
ind[x*countOnes + y] = 0;
}
likelihood[x] = 0;
for(y = 0; y < countOnes; y++)
likelihood[x] += (pow((I[ind[x*countOnes + y]] - 100),2) - pow((I[ind[x*countOnes + y]]-228),2))/50.0;
likelihood[x] = likelihood[x]/((double) countOnes);
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma410_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long likelihood_time = get_time();
printf("TIME TO GET LIKELIHOODS TOOK: %f\n", elapsed_time(error, likelihood_time));
// update & normalize weights
// using equation (63) of Arulampalam Tutorial
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(Nparticles, weights, likelihood) private(x)
for(x = 0; x < Nparticles; x++){
weights[x] = weights[x] * exp(likelihood[x]);
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma433_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long exponential = get_time();
printf("TIME TO GET EXP TOOK: %f\n", elapsed_time(likelihood_time, exponential));
double sumWeights = 0;
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for private(x) reduction(+:sumWeights)
for(x = 0; x < Nparticles; x++){
sumWeights += weights[x];
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma440_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long sum_time = get_time();
printf("TIME TO SUM WEIGHTS TOOK: %f\n", elapsed_time(exponential, sum_time));
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(sumWeights, weights) private(x)
for(x = 0; x < Nparticles; x++){
weights[x] = weights[x]/sumWeights;
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma446_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long normalize = get_time();
printf("TIME TO NORMALIZE WEIGHTS TOOK: %f\n", elapsed_time(sum_time, normalize));
xe = 0;
ye = 0;
// estimate the object location by expected values
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for private(x) reduction(+:xe, ye)
for(x = 0; x < Nparticles; x++){
xe += arrayX[x] * weights[x];
ye += arrayY[x] * weights[x];
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma455_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long move_time = get_time();
printf("TIME TO MOVE OBJECT TOOK: %f\n", elapsed_time(normalize, move_time));
printf("XE: %lf\n", xe);
printf("YE: %lf\n", ye);
double distance = sqrt( pow((double)(xe-(int)roundDouble(IszY/2.0)),2) + pow((double)(ye-(int)roundDouble(IszX/2.0)),2) );
printf("%lf\n", distance);
//display(hold off for now)
//pause(hold off for now)
//resampling
CDF[0] = weights[0];
for(x = 1; x < Nparticles; x++){
CDF[x] = weights[x] + CDF[x-1];
}
long long cum_sum = get_time();
printf("TIME TO CALC CUM SUM TOOK: %f\n", elapsed_time(move_time, cum_sum));
double u1 = (1/((double)(Nparticles)))*randu(seed, 0);
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(u, u1, Nparticles) private(x)
for(x = 0; x < Nparticles; x++){
u[x] = u1 + x/((double)(Nparticles));
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma480_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long u_time = get_time();
printf("TIME TO CALC U TOOK: %f\n", elapsed_time(cum_sum, u_time));
int j, i;
{ const unsigned long long parallel_for_start = current_time_ns();
#pragma omp parallel for shared(CDF, Nparticles, xj, yj, u, arrayX, arrayY) private(i, j)
for(j = 0; j < Nparticles; j++){
i = findIndex(CDF, Nparticles, u[j]);
if(i == -1)
i = Nparticles-1;
xj[j] = arrayX[i];
yj[j] = arrayY[i];
} ;
const unsigned long long parallel_for_end = current_time_ns();
printf("pragma488_omp_parallel %llu ns\n", parallel_for_end - parallel_for_start); }
long long xyj_time = get_time();
printf("TIME TO CALC NEW ARRAY X AND Y TOOK: %f\n", elapsed_time(u_time, xyj_time));
//#pragma omp parallel for shared(weights, Nparticles) private(x)
for(x = 0; x < Nparticles; x++){
//reassign arrayX and arrayY
arrayX[x] = xj[x];
arrayY[x] = yj[x];
weights[x] = 1/((double)(Nparticles));
}
long long reset = get_time();
printf("TIME TO RESET WEIGHTS TOOK: %f\n", elapsed_time(xyj_time, reset));
}
free(disk);
free(objxy);
free(weights);
free(likelihood);
free(xj);
free(yj);
free(arrayX);
free(arrayY);
free(CDF);
free(u);
free(ind);
}
int main(int argc, char * argv[]){
char* usage = "openmp.out -x <dimX> -y <dimY> -z <Nfr> -np <Nparticles>";
//check number of arguments
if(argc != 9)
{
printf("%s\n", usage);
return 0;
}
//check args deliminators
if( strcmp( argv[1], "-x" ) || strcmp( argv[3], "-y" ) || strcmp( argv[5], "-z" ) || strcmp( argv[7], "-np" ) ) {
printf( "%s\n",usage );
return 0;
}
int IszX, IszY, Nfr, Nparticles;
//converting a string to a integer
if( sscanf( argv[2], "%d", &IszX ) == EOF ) {
printf("ERROR: dimX input is incorrect");
return 0;
}
if( IszX <= 0 ) {
printf("dimX must be > 0\n");
return 0;
}
//converting a string to a integer
if( sscanf( argv[4], "%d", &IszY ) == EOF ) {
printf("ERROR: dimY input is incorrect");
return 0;
}
if( IszY <= 0 ) {
printf("dimY must be > 0\n");
return 0;
}
//converting a string to a integer
if( sscanf( argv[6], "%d", &Nfr ) == EOF ) {
printf("ERROR: Number of frames input is incorrect");
return 0;
}
if( Nfr <= 0 ) {
printf("number of frames must be > 0\n");
return 0;
}
//converting a string to a integer
if( sscanf( argv[8], "%d", &Nparticles ) == EOF ) {
printf("ERROR: Number of particles input is incorrect");
return 0;
}
if( Nparticles <= 0 ) {
printf("Number of particles must be > 0\n");
return 0;
}
//establish seed
int * seed = (int *)malloc(sizeof(int)*Nparticles);
int i;
for(i = 0; i < Nparticles; i++)
seed[i] = time(0)*i;
//malloc matrix
int * I = (int *)malloc(sizeof(int)*IszX*IszY*Nfr);
long long start = get_time();
//call video sequence
videoSequence(I, IszX, IszY, Nfr, seed);
long long endVideoSequence = get_time();
printf("VIDEO SEQUENCE TOOK %f\n", elapsed_time(start, endVideoSequence));
//call particle filter
const unsigned long long full_program_start = current_time_ns();
particleFilter(I, IszX, IszY, Nfr, seed, Nparticles) ;
const unsigned long long full_program_end = current_time_ns();
printf("full_program %llu ns\n", full_program_end - full_program_start);
;
long long endParticleFilter = get_time();
printf("PARTICLE FILTER TOOK %f\n", elapsed_time(endVideoSequence, endParticleFilter));
printf("ENTIRE PROGRAM TOOK %f\n", elapsed_time(start, endParticleFilter));
free(seed);
free(I);
return 0;
}
unsigned long long hclib_current_time_ms() {
return current_time_ns() / 1000000;
}
