void quicksort(uint length, uint *feld,bool median){
quicksort(0,length-1,feld,median);
}
float KDTree::chooseCut(Object3d** p_objList, int p_objCount, BoundingBox* p_bbox, int p_axis) {
float* list = (float *)malloc(sizeof(float) * p_objCount);
float min, max;
if(p_axis == KD_AXIS_X) {
for(int i = 0; i < p_objCount; i++) {
list[i] = p_objList[i]->getBoundingBox()->x0;
}
min = p_bbox->x0;
max = p_bbox->x1;
}
if(p_axis == KD_AXIS_Y) {
for(int i = 0; i < p_objCount; i++) {
list[i] = p_objList[i]->getBoundingBox()->y0;
}
min = p_bbox->y0;
max = p_bbox->y1;
}
if(p_axis == KD_AXIS_Z) {
for(int i = 0; i < p_objCount; i++) {
list[i] = p_objList[i]->getBoundingBox()->z0;
}
min = p_bbox->z0;
max = p_bbox->z1;
}
quicksort(list, 0, p_objCount - 1);
float median = list[p_objCount/2];
delete list;
return median;
/*
BoundingBox* bbox = p_bbox;
float cut = 0;
if(p_axis == KD_AXIS_X) {
float xd = bbox->x1 - bbox->x0;
cut = bbox->x0 + xd * 0.5f;
}
if(p_axis == KD_AXIS_Y) {
float yd = bbox->y1 - bbox->y0;
cut = bbox->y0 + yd * 0.5f;
}
if(p_axis == KD_AXIS_Z) {
float zd = bbox->z1 - bbox->z0;
cut = bbox->z0 + zd * 0.5f;
}
return cut;
*/
/*
int minIndex;
Object3d* minObj;
float minValue;
Object3d* obj;
float value;
Object3d* objTmp;
int halfArrayIndex = p_objCount / 2;
for(int i = 0; i < halfArrayIndex; i++) {
minIndex = i;
minObj = p_objList[i];
if(p_axis == KD_AXIS_X) {
minValue = minObj->getBoundingBox()->x0;
} else if(p_axis == KD_AXIS_Y) {
minValue = minObj->getBoundingBox()->y0;
} else if(p_axis == KD_AXIS_Z) {
minValue = minObj->getBoundingBox()->z0;
}
for(int j = i + 1; j < p_objCount; j++) {
obj = p_objList[j];
if(p_axis == KD_AXIS_X) {
value = obj->getBoundingBox()->x0;
} else if(p_axis == KD_AXIS_Y) {
value = obj->getBoundingBox()->y0;
} else if(p_axis == KD_AXIS_Z) {
value = obj->getBoundingBox()->z0;
}
if(value < minValue) {
minIndex = j;
minValue = value;
}
}
if(minIndex != i) {
objTmp = p_objList[minIndex];
p_objList[minIndex] = p_objList[i];
p_objList[i] = objTmp;
}
}
return minValue;
*/
}
int main(int argc, char* argv[])
{
/* argument variables */
int f1=0,f2=0,fi=1;
char *defdir = ".";
char *inbase = NULL, *postname = NULL;
char *indir = defdir;
/* fild variables */
FILE *fid;
char fname[100];
/* data variables */
int i,*cpuid,*property,ntype;
long p,n,*pid,*order;
float header[20],**data,time[2],*typeinfo=NULL;
/* Read Arguments */
for (i=1; i<argc; i++) {
/* If argv[i] is a 2 character string of the form "-?" then: */
if(*argv[i] == '-' && *(argv[i]+1) != '\0' && *(argv[i]+2) == '\0') {
switch(*(argv[i]+1)) {
case 'd': /* -d <indir> */
indir = argv[++i];
break;
case 'i': /* -i <basename> */
inbase = argv[++i];
break;
case 's': /* -s <post-name> */
postname = argv[++i];
break;
case 'f': /* -f <# range(f1:f2:fi)>*/
sscanf(argv[++i],"%d:%d:%d",&f1,&f2,&fi);
if (f2 == 0) f2 = f1;
break;
case 'h': /* -h */
usage(argv[0]);
break;
default:
usage(argv[0]);
break;
}
}
}
/* Checkpoints */
if (inbase == NULL)
sort_error("Please specify input file basename using -i option!\n");
if (postname == NULL)
sort_error("Please specify posterior file name using -s option!\n");
if ((f1>f2) || (f2<0) || (fi<=0))
sort_error("Wrong number sequence in the -f option!\n");
/* ====================================================================== */
for (i=f1; i<=f2; i+=fi)
{
fprintf(stderr,"Processing file number %d...\n",i);
/* Step 1: Read the file */
sprintf(fname,"%s/%s.%04d.%s.lis",indir,inbase,i,postname);
fid = fopen(fname,"rb");
if (fid == NULL)
sort_error("Fail to open output file %s!\n",fname);
/* read header */
fread(header,sizeof(float),12,fid);
fread(&ntype,sizeof(int),1,fid);
if (i == f1)
typeinfo = (float*)calloc_1d_array(ntype,sizeof(float));
fread(typeinfo,sizeof(float),ntype,fid);
fread(&time,sizeof(float),2,fid);
fread(&n,sizeof(long),1,fid);
data = (float**)calloc_2d_array(n,7,sizeof(float));
property = (int*)calloc_1d_array(n,sizeof(int));
pid = (long*)calloc_1d_array(n,sizeof(long));
cpuid = (int*)calloc_1d_array(n,sizeof(int));
order = (long*)calloc_1d_array(n,sizeof(long));
/* read data */
for (p=0; p<n; p++)
{
fread(data[p],sizeof(float),7,fid);
fread(&(property[p]),sizeof(int),1,fid);
fread(&(pid[p]),sizeof(long),1,fid);
fread(&(cpuid[p]),sizeof(int),1,fid);
}
fclose(fid);
/* Step 2: sort the particles */
for (p=0; p<n; p++)
order[p] = p;
quicksort(cpuid, pid, order, 0, n-1);
/* Step 3: write back the ordered particle list */
fid = fopen(fname,"wb");
/* write header */
fwrite(header,sizeof(float),12,fid);
fwrite(&ntype,sizeof(int),1,fid);
fwrite(typeinfo,sizeof(float),ntype,fid);
fwrite(time,sizeof(float),2,fid);
fwrite(&n,sizeof(long),1,fid);
/* write data */
for (p=0; p<n; p++)
{
fwrite(data[order[p]],sizeof(float),7,fid);
fwrite(&property[p],sizeof(int),1,fid);
fwrite(&pid[order[p]],sizeof(long),1,fid);
fwrite(&cpuid[order[p]],sizeof(int),1,fid);
}
free_2d_array(data);
free(property);
free(pid);
free(cpuid);
free(order);
fclose(fid);
}
if (typeinfo != NULL)
free(typeinfo);
return 0;
}
/**
* quick sort algorithm.
*
* @see cel_sort#cel_shell_sort
*/
CEL_API void cel_quick_sort(
void *a, uint_t length, uint_t size, cel_compare_fn_t comp)
{
uchar_t * arr = (uchar_t *) a;
quicksort(arr, size, comp, 0, length - 1);
}
int main(int argc, char *argv[])
{
FILE *fp = stdin;
double *x, *mean, *med = NULL, **mtmp = NULL, **cov = NULL, **invcov =
NULL, *var = NULL, conf = CONFLEV, *upper = NULL, *lower = NULL, t, err;
int leng = LENG, nv = -1, i, j, k = 0, lp = 0, m, outtype = 0, count = 0;
Boolean outmean = OUTMEAN, outcov = OUTCOV, outconf = OUTCONF,
outmed = OUTMED, diagc = DIAGC, inv = INV, corr = CORR;
if ((cmnd = strrchr(argv[0], '/')) == NULL)
cmnd = argv[0];
else
cmnd++;
while (--argc)
if (**++argv == '-') {
switch (*(*argv + 1)) {
case 'l':
leng = atoi(*++argv);
--argc;
break;
case 'n':
leng = atoi(*++argv) + 1;
--argc;
break;
case 't':
nv = atoi(*++argv);
--argc;
break;
case 'o':
outtype = atoi(*++argv);
--argc;
break;
case 'c':
conf = atof(*++argv);
--argc;
break;
case 'd':
diagc = 1 - diagc;
break;
case 'i':
inv = 1 - inv;
break;
case 'r':
corr = 1 - corr;
break;
case 'h':
usage(0);
default:
fprintf(stderr, "%s : Invalid option '%c'!\n", cmnd, *(*argv + 1));
usage(1);
}
} else
fp = getfp(*argv, "rb");
if (conf < 0 || conf > 100) {
fprintf(stderr,
"%s : Confidence level must be greater than 0 and less than 1.0!\n",
cmnd);
}
switch (outtype) {
case 1:
outcov = FA;
break;
case 2:
outmean = FA;
break;
case 3:
outcov = FA;
outconf = TR;
break;
case 4:
outcov = FA;
outmean = FA;
outmed = TR;
break;
}
if (diagc && corr)
diagc = FA;
if (diagc && inv)
diagc = FA;
if (corr && inv)
corr = FA;
mean = dgetmem(leng + leng);
x = mean + leng;
if (outmed) {
if (nv == -1) {
typedef struct _float_list {
float *f;
struct _float_list *next;
} float_list;
float_list *top = NULL, *prev = NULL, *cur = NULL;
top = prev = (float_list *) getmem(1, sizeof(float_list));
prev->next = NULL;
while (freadf(x, sizeof(*x), leng, fp) == leng) {
cur = (float_list *) getmem(1, sizeof(float_list));
cur->f = fgetmem(leng);
for (i = 0; i < leng; i++) {
cur->f[i] = (float) x[i];
}
count++;
prev->next = cur;
cur->next = NULL;
prev = cur;
}
k = count;
mtmp = (double **) getmem(leng, sizeof(*mtmp));
mtmp[0] = dgetmem(leng * k);
for (i = 1; i < leng; i++)
mtmp[i] = mtmp[i - 1] + k;
med = dgetmem(leng);
for (j = 0, cur = top->next; j < k; j++, cur = cur->next) {
for (i = 0; i < leng; i++) {
mtmp[i][j] = (double) cur->f[i];
}
}
} else {
k = nv;
mtmp = (double **) getmem(leng, sizeof(*mtmp));
mtmp[0] = dgetmem(leng * k);
for (i = 1; i < leng; i++)
mtmp[i] = mtmp[i - 1] + k;
med = dgetmem(leng);
for (j = 0; j < k; j++) {
for (i = 0; i < leng; i++) {
freadf(&mtmp[i][j], sizeof(**mtmp), 1, fp);
}
}
}
if (k % 2 == 0) {
fprintf(stderr, "%s : warning: the number of vectors is even!\n",
cmnd);
}
for (i = 0; i < leng; i++) {
quicksort(mtmp[i], 0, k - 1);
if (k % 2 == 1) {
med[i] = mtmp[i][k / 2];
} else {
med[i] = ((mtmp[i][k / 2] + mtmp[i][k / 2 - 1]) / 2);
}
}
fwritef(med, sizeof(*med), leng, stdout);
return (0);
}
if (outcov) {
if (!diagc) {
cov = (double **) getmem(leng, sizeof(*cov));
cov[0] = dgetmem(leng * leng);
for (i = 1; i < leng; i++)
cov[i] = cov[i - 1] + leng;
if (inv) {
invcov = (double **) getmem(leng, sizeof(*invcov));
invcov[0] = dgetmem(leng * leng);
for (i = 1; i < leng; i++)
invcov[i] = invcov[i - 1] + leng;
}
} else
var = dgetmem(leng);
}
if (outconf) {
var = dgetmem(leng);
upper = dgetmem(leng);
lower = dgetmem(leng);
}
while (!feof(fp)) {
for (i = 0; i < leng; i++) {
mean[i] = 0.0;
if (outcov) {
if (!diagc)
for (j = 0; j < leng; j++)
cov[i][j] = 0.0;
else
var[i] = 0.0;
}
if (outconf) {
var[i] = 0.0;
}
}
for (lp = nv; lp;) {
if (freadf(x, sizeof(*x), leng, fp) != leng)
break;
for (i = 0; i < leng; i++) {
mean[i] += x[i];
if (outcov) {
if (!diagc)
for (j = i; j < leng; j++)
cov[i][j] += x[i] * x[j];
else
var[i] += x[i] * x[i];
}
if (outconf) {
var[i] += x[i] * x[i];
}
}
--lp;
}
if (lp == 0 || nv == -1) {
if (nv > 0)
k = nv;
else
k = -lp - 1;
for (i = 0; i < leng; i++)
mean[i] /= k;
if (outcov) {
if (!diagc)
for (i = 0; i < leng; i++)
for (j = i; j < leng; j++)
cov[j][i] = cov[i][j] = cov[i][j] / k - mean[i] * mean[j];
else
for (i = 0; i < leng; i++)
var[i] = var[i] / k - mean[i] * mean[i];
}
if (outconf) {
for (i = 0; i < leng; i++) {
var[i] = (var[i] - k * mean[i] * mean[i]) / (k - 1);
}
t = t_percent(conf / 100, k - 1);
for (i = 0; i < leng; i++) {
err = t * sqrt(var[i] / k);
upper[i] = mean[i] + err;
lower[i] = mean[i] - err;
}
}
if (corr) {
for (i = 0; i < leng; i++)
for (j = i + 1; j < leng; j++)
cov[j][i] = cov[i][j] =
cov[i][j] / sqrt(cov[i][i] * cov[j][j]);
for (i = 0; i < leng; i++)
cov[i][i] = 1.0;
}
if (outmean)
fwritef(mean, sizeof(*mean), leng, stdout);
if (outcov) {
if (!diagc) {
if (inv) {
for (i = 0; i < leng; i++) {
for (j = i + 1; j < leng; j++) {
cov[j][i] /= cov[i][i];
for (m = i + 1; m < leng; m++)
cov[j][m] -= cov[i][m] * cov[j][i];
}
}
for (m = 0; m < leng; m++) {
for (i = 0; i < leng; i++) {
if (i == m)
invcov[i][m] = 1.0;
else
invcov[i][m] = 0.0;
}
for (i = 0; i < leng; i++) {
for (j = i + 1; j < leng; j++)
invcov[j][m] -= invcov[i][m] * cov[j][i];
}
for (i = leng - 1; i >= 0; i--) {
for (j = i + 1; j < leng; j++)
invcov[i][m] -= cov[i][j] * invcov[j][m];
invcov[i][m] /= cov[i][i];
}
}
fwritef(invcov[0], sizeof(*invcov[0]), leng * leng, stdout);
} else
fwritef(cov[0], sizeof(*cov[0]), leng * leng, stdout);
} else
fwritef(var, sizeof(*var), leng, stdout);
}
if (outconf) {
fwritef(upper, sizeof(*upper), leng, stdout);
fwritef(lower, sizeof(*lower), leng, stdout);
}
}
}
return (0);
}
void escreverEmArquivo(char *nomeArquivo)
{
FILE *file;
long long int i;
int numero;
int *vet, *vet2,*vet3;
vet = (int*)malloc(sizeof(int)*tam);
vet2 = (int*)malloc(sizeof(int)*tam);
vet3 = (int*)malloc(sizeof(int)*tam);
double inicio, fim, tempo, inicio2, fim2, tempo2, inicio3, fim3, tempo3;
file = fopen(nomeArquivo, "w+b");
if( file == NULL)
{
printf("erro ao abrir arquivo\n");
exit(0);
}
for(i = 1; i < tam ; ++i) //escreve caracteres
{
numero = tam - i;
//printf(" [%d] ",numero);
fwrite(&numero, sizeof(int), 1, file);
}
rewind(file);
i = 1;
while(!feof(file))
{
fread(&vet[i],sizeof(int),1,file);
//fread(&vet2[i],sizeof(int),1,file);
//fread(&vet3[i],sizeof(int),1,file);
i++;
}
//imprimindo os valores do primeiro vetor;
inicio = clock();
heapsort(vet,tam);
fim = clock();
tempo = (fim-inicio)/CLOCKS_PER_SEC; // recebendo o tempo ;
i = 1;
while(!feof(file))
{
fread(&vet[i],sizeof(int),1,file);
//fread(&vet2[i],sizeof(int),1,file);
//fread(&vet3[i],sizeof(int),1,file);
i++;
}
i = 1;
while(!feof(file))
{
fread(&vet[i],sizeof(int),1,file);
//fread(&vet2[i],sizeof(int),1,file);
//fread(&vet3[i],sizeof(int),1,file);
i++;
}
//imprimindo os valores do segundo vetor;
printf("\n\n");
inicio2 = clock();i = 1;
quicksort(vet,1,tam);
fim2 = clock();
tempo2 = (fim2-inicio2)/CLOCKS_PER_SEC; //recebendo o tempo;
i = 1;
while(!feof(file))
{
fread(&vet[i],sizeof(int),1,file);
//fread(&vet2[i],sizeof(int),1,file);
//fread(&vet3[i],sizeof(int),1,file);
i++;
}
printf("\n\n");
inicio3 = clock();
mergesort(vet,1,tam);
fim3 = clock();
tempo3 = (fim3-inicio3)/CLOCKS_PER_SEC;
printf("\n\n");
printf(" TEMPO HEAP: %4.4lf\n",tempo);
printf(" TEMPO QUICK: %4.4lf\n",tempo2);
printf(" TEMPO MERGE: %4.4lf\n",tempo3);
fclose(file);
}
void quicksort(int* arr,int length){
quicksort(arr,0,length-1);
}
static PyObject* PositionWeightMatrix_nullScoreDistribution( PositionWeightMatrix* self, PyObject* args )
{
PyObject* seq;
int reps;
if( !PyArg_ParseTuple( args, "Oi", &seq, &reps ) )
return NULL;
seq = PySequence_Fast( seq, "argument must be iterable" );
if( !seq ) return NULL;
PyObject** elems = PySequence_Fast_ITEMS( seq );
size_t len = PySequence_Fast_GET_SIZE( seq );
//allocate array to put elems in
long* array = malloc( len * sizeof(long) );
size_t i;
for( i = 0 ; i < len ; ++i )
{
array[i] = PyInt_AsLong( elems[i] );
}
//check for error
if( PyErr_Occurred() )
{
free( array );
return NULL;
}
double** matrix = self->matrix;
int n = self->n;
int m = self->m;
//allocate scores array
size_t nmers = ( len - n + 1 );
size_t lenScores = nmers*reps;
double* scores = malloc( lenScores * sizeof( double ) );
//for each replicate, shuffle and score the sequence
int rep = 0;
while( rep < reps )
{
shuffle( array, len, sizeof( long ) );
scoreAllLong( matrix, n, m, array, len, scores, nmers * rep );
++rep;
}
//sort the scores
quicksort( scores, lenScores, sizeof( double ), &compareDoubleDescending );
//free the long array
free( array );
//build a python list to return
PyObject* list = PyList_New( lenScores );
for( i = 0 ; i < lenScores ; ++i )
{
PyList_SetItem( list, i, PyFloat_FromDouble( scores[i] ) );
}
//free the scores array
free( scores );
return list;
}
void s_samplesort(int fd, off_t size,char *base_name, int t){
struct queue_buf *buff;
struct queue_buf **file_buff;
int *files, *sizes, *keys;
int i,j,cur,r,ret,ret1;
char name_file[500];
int k;
k = t * min(ceildiv(size,M),ceildiv(M,B));
r = 0;
buff = qb_new(M);
files = (int *)malloc(sizeof(int)*k);
sizes = (int *)malloc(sizeof(int)*k);
keys = (int *)malloc(sizeof(int)*(k+1));
file_buff = (struct queue_buf **)malloc(sizeof(struct queue_buf *)*k);
select_keys(keys,fd,size,k);
for(i = 0; i < k; i++){
file_buff[i] = qb_new(B);
sizes[i] = 0;
sprintf(name_file, "%s_%d\0", base_name,i);
if((files[i] = open(name_file, O_RDWR | O_CREAT, S_IRWXU|S_IRWXG|S_IRWXO))== -1){
printf("%s_%d fail1\n", base_name,i);
perror("aca");
exit(1);
}
results.io_rand++;
}
j=1;
while(1){
if((ret1 = qb_refill(buff,fd))== -1){
printf("%s_%d fail2\n", base_name,i);
perror("aca");
exit(1);
}
results.io_rand++;
results.io_acc+=ceildiv(buff->n_elems,B);
if (qb_empty(buff)) break;
while(!qb_empty(buff)){
cur = qb_dequeue(buff);
i = bucket(cur,keys,k+1);
qb_enqueue(file_buff[i], cur);
if(qb_full(file_buff[i])){
results.io_rand++;
results.io_acc+=ceildiv(file_buff[i]->n_elems,B);
sizes[i] += file_buff[i]->n_elems;
if((ret = qb_flush(file_buff[i],files[i]))== -1){
printf("%s_%d fail3\n", base_name,i);
perror("aca");
exit(1);
}
}
}
}
close(fd);
remove(curr_name);
/* Pueden quedar restos en los bufs. Hay que vaciar todo */
for (i = 0; i < k; i++) {
if (!qb_empty(file_buff[i])) {
results.io_rand++;
results.io_acc+=ceildiv(file_buff[i]->n_elems,B);
sizes[i] += file_buff[i]->n_elems;
qb_flush(file_buff[i], files[i]);
}
close(files[i]);
qb_free(file_buff[i]);
}
free(file_buff);
free(keys);
for(i = 0; i < k; i++){
sprintf(name_file, "%s_%d\0", base_name,i);
if(sizes[i] == 0){
remove(name_file);
}
else if(sizes[i] <= M){
if((files[i] = open(name_file, O_RDWR | O_CREAT, S_IRWXU|S_IRWXG|S_IRWXO))== -1){
printf("%s_%d fail1\n", base_name,i);
perror("aca");
exit(1);
}
if((ret1 = qb_refill(buff,files[i]))== -1){
printf("%s_%d fail2\n", base_name,i);
perror("aca");
exit(1);
}
results.io_rand++;
results.io_acc+=ceildiv(buff->n_elems,B);
quicksort(buff);
lseek64(files[i],0,SEEK_SET);
results.io_rand++;
qb_flush(buff,files[i]);
results.io_rand++;
results.io_acc+=ceildiv(buff->n_elems,B);
close(files[i]);
remove(name_file);
}
}
qb_free(buff);
for (i = 0; i < k; i++) {
sprintf(name_file, "%s_%d\0", base_name,i);
if (sizes[i] > M) {
if((files[i] = open(name_file, O_RDWR | O_CREAT, S_IRWXU|S_IRWXG|S_IRWXO))== -1){
printf("%s_%d fail1\n",base_name, i);
perror("aca");
exit(1);
}
results.io_rand++;
curr_name= name_file;
s_samplesort(files[i],sizes[i],name_file,t);
}
}
free(sizes);
free(files);
}
int main(void)
{
const int MAX_STALLS = 100001;
int t;
long long int *stall_positions;
stall_positions = malloc(MAX_STALLS * sizeof(*stall_positions));
scanf("%d\n", &t);
while (t-- > 0)
{
long int no_stalls, no_cows, i;
scanf("%ld %ld\n", &no_stalls, &no_cows);
for (i=0; i<no_stalls; i++) scanf("%lld\n", stall_positions+i);
// heapsort(stall_positions, no_stalls);
quicksort(stall_positions, 0, no_stalls-1);
// printf("sorted stall_positions:"); for (i=0; i<no_stalls; i++) printf(" %lld", stall_positions[i]); printf("\n");
long long int max_distance, start, distance, end;
start = 0;
end = stall_positions[no_stalls-1] - stall_positions[0] + 1;
max_distance = 0;
while (start < end)
{
distance = (end+start)/2;
// count how many elements exist in stall_positions such that the difference between two such elements is greater than distance
long int count;
long long int prev;
count = 1;
prev = stall_positions[0];
for (i=1; i<no_stalls; i++)
if (stall_positions[i]-prev >= distance)
{
prev = stall_positions[i];
count += 1;
}
if (count >= no_cows)
{
// it is possible to place all cows in stalls such that the distance between two stalls with cows is gte distance
if (distance > max_distance) max_distance = distance;
start = distance+1;
}
else
{
end = distance;
}
}
printf("%lld\n", max_distance);
}
free(stall_positions);
return 0;
}
int main(int argc, char* argv[]) {
int* bodies_off;
int* n_bodies_split;
int n_local_bodies;
const MPI_Comm comm = MPI_COMM_WORLD;
FILE *inputf;
FILE *outputf;
double clockStart, clockEnd;
int rc, n_proc, rank;
rc = MPI_Init(&argc, &argv);
if (rc != MPI_SUCCESS) {
puts("MPI_Init failed");
exit(-1);
}
MPI_Comm_size(comm, &n_proc);
MPI_Comm_rank(comm, &rank);
//creazione datatype per mpi!
MPI_Datatype bodytype;
MPI_Datatype type[6] = { MPI_LB, MPI_DOUBLE, MPI_DOUBLE, MPI_DOUBLE, MPI_DOUBLE, MPI_UB };
int block_len[6] = {1, 1, 3, 3, 3, 1};
MPI_Aint disp[6];
leaf_t example[2];
MPI_Get_address(&example[0], &disp[0]);
MPI_Get_address(&(example[0].mass), &disp[1]);
MPI_Get_address(&(example[0].pos), &disp[2]);
MPI_Get_address(&(example[0].vel), &disp[3]);
MPI_Get_address(&(example[0].acc), &disp[4]);
MPI_Get_address(&(example[1].acc), &disp[5]);
// int i;
// for(i = 6; i >= 0; --i)
// disp[i] -= disp[0];
disp[1] = disp[1] - disp[0];
disp[2] = disp[2] - disp[0];
disp[3] = disp[3] - disp[0];
disp[4] = disp[4] - disp[0];
disp[5] = disp[5] - disp[0];
MPI_Type_create_struct(6, block_len, disp, type, &bodytype);
MPI_Type_commit(&bodytype);
bodies_off = malloc((n_proc + 1) * sizeof(int));
n_bodies_split = malloc((n_proc) * sizeof(int));
bodies = malloc(nbodies * sizeof(node_t*));
leafs = malloc(nbodies * sizeof(leaf_t));
char* inputfile = argv[1];
inputf = fopen(inputfile, "r");
if (inputf == NULL) {
printf("impossibile leggere da file");
exit(1);
}
fscanf(inputf, "%d", &nbodies);
fscanf(inputf, "%d", &steps);
fscanf(inputf, "%lf", &dt);
fscanf(inputf, "%lf", &eps);
fscanf(inputf, "%lf", &tol);
fclose(inputf);
if (rank == 0) {
int i;
create_bodies();
quicksort(0, nbodies - 1);
// bublesort();
// int i = 0;
// for (i = 0; i < nbodies; i++) {
// printf("%lf, %lf, %lf \n", bodies[i]->pos[0], bodies[i]->pos[1],
// bodies[i]->pos[2]);
// }
n_local_bodies = nbodies / n_proc;
//split delle particelle secondo shark & fish
// split_bodies(n_proc, bodies_off, n_bodies_split);
// n_local_bodies = n_bodies_split[rank];
//
// MPI_Bcast(n_bodies_split, n_proc, MPI_INT, 0, comm);
MPI_Bcast(leafs, nbodies, bodytype, 0, comm);
dthf = 0.5 * dt;
epssq = eps * eps;
itolsq = 1.0 / (tol * tol);
clockStart = MPI_Wtime();
int step = 0;
root = NULL;
for (step = 0; step < steps; step++) {
compute_center_and_diameter();
root = malloc(sizeof(struct node_t)); // "new" is like "malloc"
double mass_root = 0.0;
root->type = 1;
root->mass = &mass_root;
root->pos = center;
root->cell.childs[0] = NULL;
root->cell.childs[1] = NULL;
root->cell.childs[2] = NULL;
root->cell.childs[3] = NULL;
root->cell.childs[4] = NULL;
root->cell.childs[5] = NULL;
root->cell.childs[6] = NULL;
root->cell.childs[7] = NULL;
double radius = diameter * 0.5;
int i = 0;
for (i = 0; i < nbodies; i++) {
insert(root, bodies[i], radius); // questo è il modo per passare i dati per riferimento... cioè mandare l'indirizzo della struttura puntata dal puntatore
}
curr = 0;
compute_center_of_mass(&(*root));
for (i = 0; i < n_local_bodies; i++) {
compute_force(&(*root), &(*bodies[i]), diameter, step);
}
// for (i = 0; i < nbodies; i++) {
// }
deallocate_tree(root);
//inserire all gather
MPI_Allgather(leafs, n_local_bodies, bodytype, leafs,
n_local_bodies, bodytype, comm);
for (i = 0; i < nbodies; i++) {
advance(&(*bodies[i]));
}
// int p = 0;
// for (p = 0; p < nbodies; p++)
// printf("%lf, %lf, %lf \n", bodies[p]->pos[0], bodies[p]->pos[1],
// bodies[p]->pos[2]);
// printf("*************************************** \n");
}
// int i = 0;
// dopo l'esecuzione!!
// int proc_rec = 1;
// while (proc_rec < n_proc) {
// MPI_Status status;
// int proc_rank;
// int cap = nbodies / n_proc;
// node_t temp[cap];
// MPI_Recv(temp, cap, bodytype, MPI_ANY_SOURCE, MPI_ANY_TAG, comm,
// &status);
// proc_rank = status.MPI_SOURCE;
//
// int idx = 0;
// for (idx = proc_rec * (cap); idx < cap; idx++)
// *bodies[idx] = temp[idx];
// proc_rec++;
// }
clockEnd = MPI_Wtime();
if (nbodies == 16384) {
system("echo 'Host:' `hostname` >> output16384 ");
outputf = fopen("output16384", "a");
fprintf(outputf, "Tempo di esecuzione: %lf \n", clockEnd
- clockStart);
for (i = 0; i < nbodies; i++) {
fprintf(outputf, "%lf, %lf, %lf \n", bodies[i]->pos[0],
bodies[i]->pos[1], bodies[i]->pos[2]);
}
} else if (nbodies == 32768) {
system("echo 'Host:' `hostname` >> output32768 ");
outputf = fopen("output32768", "a");
fprintf(outputf, "Tempo di esecuzione: %lf \n", clockEnd
- clockStart);
for (i = 0; i < nbodies; i++) {
fprintf(outputf, "%lf, %lf, %lf \n", bodies[i]->pos[0],
bodies[i]->pos[1], bodies[i]->pos[2]);
}
} else if (nbodies == 65536) {
system("echo 'Host:' `hostname` >> output65536 ");
outputf = fopen("output65536", "a");
fprintf(outputf, "Tempo di esecuzione: %lf \n", clockEnd
- clockStart);
for (i = 0; i < nbodies; i++) {
fprintf(outputf, "%lf, %lf, %lf \n", bodies[i]->pos[0],
bodies[i]->pos[1], bodies[i]->pos[2]);
}
} else {
system("echo 'Host:' `hostname` >> output ");
outputf = fopen("output", "a");
fprintf(outputf, "Tempo di esecuzione: %lf \n", clockEnd
- clockStart);
for (i = 0; i < nbodies; i++) {
fprintf(outputf, "%lf, %lf, %lf \n", bodies[i]->pos[0],
bodies[i]->pos[1], bodies[i]->pos[2]);
}
}
fflush(outputf);
fclose(outputf);
printf("Esecuzione completata\n");
} else {
int low = 1, up = 0;
int i;
dthf = 0.5 * dt;
epssq = eps * eps;
itolsq = 1.0 / (tol * tol);
// if (PAPI_library_init(PAPI_VER_CURRENT) != PAPI_VER_CURRENT) {
// printf("Inizializzazione della libreria di papi fallita \n");
// exit(1);
// }
//
// if (PAPI_create_eventset(&event_set) != PAPI_OK) {
// printf("E' andata a male la creazione dell'eventSet \n");
// exit(1);
// }
//
// if (PAPI_add_events(event_set, events, 2) != PAPI_OK) {
// printf("E' andata a male l'aggiunta degli eventi\n");
// exit(1);
// }
n_local_bodies = nbodies / n_proc;
MPI_Bcast(leafs, nbodies, bodytype, 0, comm);
int step = 0;
root = NULL;
low += (rank * n_local_bodies);
up = low + n_local_bodies;
// PAPI_start(event_set);
// clockStart = PAPI_get_real_usec();
for (step = 0; step < steps; step++) {
compute_center_and_diameter();
root = malloc(sizeof(struct node_t)); // "new" is like "malloc"
root->type = 1;
*(root->mass) = 0.0;
root->pos = center;
root->cell.childs[0] = NULL;
root->cell.childs[1] = NULL;
root->cell.childs[2] = NULL;
root->cell.childs[3] = NULL;
root->cell.childs[4] = NULL;
root->cell.childs[5] = NULL;
root->cell.childs[6] = NULL;
root->cell.childs[7] = NULL;
double radius = diameter * 0.5;
for (i = 0; i < nbodies; i++) {
bodies[i] = malloc(sizeof(node_t));
bodies[i]->cell.leaf = &leafs[i];
bodies[i]->mass = &leafs[i].mass;
bodies[i]->pos = leafs[i].pos;
insert(&(*root), &(*bodies[i]), radius); // questo è il modo per passare i dati per riferimento... cioè mandare l'indirizzo della struttura puntata dal puntatore
}
curr = 0;
compute_center_of_mass(&(*root));
for (i = low; i < up; i++) {
compute_force(&(*root), &(*bodies[i]), diameter, step);
}
// for (i = 0; i < nbodies; i++) {
// }
deallocate_tree(root);
local_leafs = &leafs[low];
//inserire all_gather
MPI_Allgather(local_leafs, up - low, bodytype, leafs, up - low,
bodytype, comm);
for (i = 0; i < nbodies; i++) {
advance(&(*bodies[i]));
}
// int p = 0;
// for (p = 0; p < nbodies; p++)
// printf("%lf, %lf, %lf \n", bodies[p]->pos[0], bodies[p]->pos[1],
// bodies[p]->pos[2]);
// printf("*************************************** \n");
}
// clockEnd = PAPI_get_real_usec();
// PAPI_stop(event_set, values);
// int i = 0;
// MPI_Send(bodies[low], up - low + 1, bodytype, 0, MPI_ANY_TAG, comm);
}
MPI_Finalize();
return 0;
}
// -----------------------------------------------------------------------------
// Hauptprogramm
int main(int argc, char *argv[])
{
float *v; // Feld
int iter; // Wiederholungen
if (argc != 2) { // Benutzungshinweis
printf ("Vector sorting\nUsage: %s <NumIter>\n", argv[0]);
return 0;
}
iter = atoi(argv[1]);
v = (float *) calloc(NUM, sizeof(float)); // Speicher reservieren
// seriell
printf("Perform vector sorting %d times...\n", iter);
double start = omp_get_wtime();
for (int i = 0; i < iter; i++) { // Wiederhole das Sortieren
for (int j = 0; j < NUM; j++) // Mit Zufallszahlen initialisieren
v[j] = (float)rand();
quicksort(v, 0, NUM-1); // Sortierung
}
double s_t = omp_get_wtime() - start;
printf ("Done. This task took %f seconds.\n", s_t);
if (is_sorted(v))
printf ("The vector is sorted correctly.\n");
else
printf ("The vector is NOT sorted correctly.\n");
printf ("------------------------------------------\n");
// parallel V1
start = omp_get_wtime();
for (int i = 0; i < iter; i++) { // Wiederhole das Sortieren
for (int j = 0; j < NUM; j++) // Mit Zufallszahlen initialisieren
v[j] = (float)rand();
quicksortP1(v, 0, NUM-1); // Sortierung
}
double p1_t = omp_get_wtime() - start;
printf ("Parallel with sections done. This task took %f seconds.\n", p1_t);
if (is_sorted(v)){
printf ("The vector is sorted correctly.\n");
printf ("Speedup: %f\n", s_t/p1_t); // AUFGABE 3: Berechne Speedup
}
else {
printf ("The vector is NOT sorted correctly.\n");
}
printf ("------------------------------------------\n");
// parallel V2
start = omp_get_wtime();
for (int i = 0; i < iter; i++) { // Wiederhole das Sortieren
for (int j = 0; j < NUM; j++) // Mit Zufallszahlen initialisieren
v[j] = (float)rand();
#pragma omp parallel
{
#pragma omp single nowait
quicksortP2(v, 0, NUM-1);
} // Sortierung
}
double p2_t = omp_get_wtime() - start;
printf ("Parallel with tasks done. This task took %f seconds.\n", p2_t);
if (is_sorted(v)) {
printf ("The vector is sorted correctly.\n");
printf ("Speedup: %f\n", s_t/p2_t); // AUFGABE 3: Berechne Speedup
}
else {
printf ("The vector is NOT sorted correctly.\n");
}
printf ("------------------------------------------\n");
// parallel V3
start = omp_get_wtime();
for (int i = 0; i < iter; i++) { // Wiederhole das Sortieren
for (int j = 0; j < NUM; j++) // Mit Zufallszahlen initialisieren
v[j] = (float)rand();
quicksortP3(v, 0, NUM-1);
}
double p3_t = omp_get_wtime() - start;
printf ("Parallel with tasks done. This task took %f seconds.\n", p3_t);
if (is_sorted(v)) {
printf ("The vector is sorted correctly.\n");
printf ("Speedup: %f\n", s_t/p3_t); // AUFGABE 3: Berechne Speedup
}
else {
printf ("The vector is NOT sorted correctly.\n");
}
printf ("------------------------------------------\n");
return 0;
}
double SearchLambda(int *nn,double *eta,int len) {
int i,j,k,ind[2];
int *preindex=malloc(sizeof(int)*len),*index=malloc(sizeof(int)*len);
double step,*val=malloc(sizeof(double)*len),pre1,pre2,*val_c=malloc(sizeof(double)*len),del=0;
double *preval=malloc(sizeof(double)*(1+len));
memcpy(preval,eta,sizeof(double)*(len+1));
for(i=0; i<len; i++) {
val[i]=psi(nn[i]+eta[i+1])-psi(eta[i+1]);
val_c[i]=val[i];
index[i]=i;
del-=val[i];
}
quicksort(val_c,index,0,len-1);
//coarse to fine
for(k=0;k<8;k++){
step=S_STEP*(int)pow(10,7-k);
//for(k=0;k<7;k++){
//step=S_STEP*(int)pow(10,6-k);
memset(preindex,0,len*sizeof(int));
while(samevector(preindex,index,len)==0) {
memcpy(preindex,index,sizeof(int)*len);
for(i=0; i<(int)len/2; i++) {
ind[0]=index[i];
ind[1]=index[len-1-i];
index[i]=i;
index[len-1-i]=len-1-i;
/*for all pairs */
//printf("aaa %d,%d,%f,%f,%f,%f\n ",ind[0],ind[1],eta[1+ind[0]],eta[1+ind[1]],val[ind[0]],val[ind[1]]);
//
//if((eta[1+ind[0]]<0 || eta[1+ind[1]]<0)){
//printf("input err: %d,%f,%f,%f\n", k,eta[1+ind[0]], eta[1+ind[1]],step);
//}
if(val[ind[0]]>val[ind[1]]){
printf("sort err: %d,%d,%f,%f,%f,%f,%f\n", k,i,val_c[i],val_c[len-i-1],val[ind[0]], val[ind[1]],step);
printv1(len,index);
//printv1(len,nn);
printvector(len,val_c);
printvector(len,val);
}
while(val[ind[0]]<val[ind[1]] && eta[1+ind[0]]-step>0) {
pre1=val[ind[0]];
pre2=val[ind[1]];
eta[1+ind[0]]-= step;
eta[1+ind[1]]+= step;
val[ind[0]]=psi(eta[1+ind[0]]+nn[ind[0]])-psi(eta[1+ind[0]]);
val[ind[1]]=psi(eta[1+ind[1]]+nn[ind[1]])-psi(eta[1+ind[1]]);
//if(eta[1+ind[0]]>0){
//printf("gg %d,%d,%f,%f,%f,%f\n ",ind[0],ind[1],eta[1+ind[0]],eta[1+ind[1]],val[ind[0]],val[ind[1]]);
//}
// if( eta[1+ind[0]]<0 || eta[1+ind[1]]<0){
// printf("iteration err: %d,%f,%f,%f,%f,%f\n", k,val[ind[0]],val[ind[1]],eta[1+ind[0]], eta[1+ind[1]],step);
// }
}
//be careful about the equal case...
if(val[ind[0]]>val[ind[1]]) {
/*one step back*/
eta[1+ind[0]]+= step;
eta[1+ind[1]]-= step;
//if( eta[1+ind[0]]<0 || eta[1+ind[1]]<0){
//printf("add back err: %d,%f,%f,%f,%f,%f,%f,%f\n", k,val[ind[0]],val[ind[1]],pre1,pre2,eta[1+ind[0]], eta[1+ind[1]],step);
//}
val[ind[0]]=pre1;
val[ind[1]]=pre2;
//printf("fff %d,%d,%f,%f,%f,%f\n ",ind[0],ind[1],eta[1+ind[0]],eta[1+ind[1]],val[ind[0]],val[ind[1]]);
//if(val[ind[0]]>val[ind[1]]) {
// printf("pre_wrong...%d,%f,%f\n",k,val[ind[0]],val[ind[1]]);
//}
}
}
if (len%2==1){
index[(len-1)/2]=(len-1)/2;
}
memcpy(val_c,val,sizeof(double)*len);
quicksort(val_c,index,0,len-1);
//printv1(len,index);
//printv1(len,nn);
//printvector(len+1,eta);
}
}
//printvector(len,val);
//printvector(len+1,eta);
for(i=0; i<len; i++) {
del+=val[i];
}
free(preindex);
free(index);
free(val);
free(val_c);
return del;
}
//Synarthsh quick sort gia na mporesei na thn xrhsimopoihsei h quick.c anadromika
void quicksort(Record * x,int first,int last,int atr_num)
{
int pivot,j,i;
Record temp;
if(first<last)
{
pivot=first;
i=first;
j=last;
while(i<j)
{ if (atr_num==0)
{
while(x[i].ssn<=x[pivot].ssn&&i<last)
{
i++;
}
while(x[j].ssn>x[pivot].ssn)
{
j--;
}
if(i<j)
{
anathesh(&temp,x[i]);
anathesh(&(x[i]),x[j]);
anathesh(&(x[j]),temp);
}
}
else if(atr_num==1)
{
while((strcmp(x[i].FirstName,x[pivot].FirstName)<=0)&&i<last)
{
i++;
}
while(strcmp(x[j].FirstName,x[pivot].FirstName)>0)
{
j--;
}
if(i<j)
{
anathesh(&temp,x[i]);
anathesh(&(x[i]),x[j]);
anathesh(&(x[j]),temp);
}
}
else if(atr_num==2)
{
while((strcmp(x[i].LastName,x[pivot].LastName)<=0)&&i<last)
{
i++;
}
while(strcmp(x[j].LastName,x[pivot].LastName)>0)
{
j--;
}
if(i<j)
{
anathesh(&temp,x[i]);
anathesh(&(x[i]),x[j]);
anathesh(&(x[j]),temp);
}
}
else if(atr_num==3)
{
while(x[i].income<=x[pivot].income&&i<last)
{
i++;
}
while(x[j].income>x[pivot].income)
{
j--;
}
if(i<j)
{
anathesh(&temp,x[i]);
anathesh(&(x[i]),x[j]);
anathesh(&(x[j]),temp);
}
}
}
anathesh(&temp,x[pivot]);
anathesh(&(x[pivot]),x[j]);
anathesh(&(x[j]),temp);
quicksort(x,first,j-1,atr_num);
quicksort(x,j+1,last,atr_num);
}
}
void QuickSort::sort(int array[], int size)
{
quicksort(array, 0, size - 1);
}
int main() {
// Initialize RNG
dsfmt_gv_init_gen_rand(0);
double t, tmin;
// fib(20)
assert(fib(20) == 6765);
int f = 0;
tmin = INFINITY;
volatile int fibarg = 20; // prevent constant propagation
for (int i=0; i<NITER; ++i) {
t = clock_now();
for (int j = 0; j < 1000; j++)
f += fib(fibarg);
t = clock_now()-t;
if (t < tmin) tmin = t;
}
print_perf("fib", tmin / 1000);
// parse_bin
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
t = clock_now();
char s[11];
for (int k=0; k<1000 * 100; ++k) {
uint32_t n = dsfmt_gv_genrand_uint32();
sprintf(s, "%x", n);
uint32_t m = (uint32_t)parse_int(s, 16);
assert(m == n);
}
t = clock_now()-t;
if (t < tmin) tmin = t;
}
print_perf("parse_int", tmin / 100);
// // array constructor
// tmin = INFINITY;
// for (int i=0; i<NITER; ++i) {
// t = clock_now();
// double *a = ones(200,200);
// free(a);
// t = clock_now()-t;
// if (t < tmin) tmin = t;
// }
// print_perf("ones", tmin);
//
// // A*A'
// //SUBROUTINE DGEMM(TRANSA,TRANSB,M,N,K,ALPHA,A,LDA,B,LDB,BETA,C,LDC)
// double *b = ones(200, 200);
// tmin = INFINITY;
// for (int i=0; i<NITER; ++i) {
// t = clock_now();
// double *c = matmul_aat(200, b);
// free(c);
// t = clock_now()-t;
// if (t < tmin) tmin = t;
// }
// free(b);
// print_perf("AtA", tmin);
// mandel
/* The initialization on the next line is deliberately volatile to
* prevent gcc from optimizing away the entire loop.
* (First observed in gcc 4.9.2)
*/
static volatile int mandel_sum_init = 0;
int mandel_sum2 = mandel_sum_init;
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
int *M;
t = clock_now();
for (int j = 0; j < 100; j++) {
M = mandelperf();
if (j == 0) {
int mandel_sum = 0;
// for (int ii = 0; ii < 21; ii++) {
// for (int jj = 0; jj < 26; jj++) {
// printf("%4d", M[26*ii + jj]);
// }
// printf("\n");
// }
for (int k = 0; k < 21*26; k++) {
mandel_sum += M[k];
}
assert(mandel_sum == 14791);
mandel_sum2 += mandel_sum;
}
free(M);
}
t = clock_now()-t;
if (t < tmin) tmin = t;
}
assert(mandel_sum2 == 14791 * NITER);
print_perf("mandel", tmin / 100);
// sort
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
t = clock_now();
double *d = myrand(5000);
quicksort(d, 0, 5000-1);
free(d);
t = clock_now()-t;
if (t < tmin) tmin = t;
}
print_perf("quicksort", tmin);
// pi sum
double pi;
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
t = clock_now();
pi = pisum();
t = clock_now()-t;
if (t < tmin) tmin = t;
}
assert(fabs(pi-1.644834071848065) < 1e-12);
print_perf("pi_sum", tmin);
// rand mat stat
struct double_pair r;
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
t = clock_now();
r = randmatstat(1000);
t = clock_now()-t;
if (t < tmin) tmin = t;
}
// assert(0.5 < r.s1 && r.s1 < 1.0 && 0.5 < r.s2 && r.s2 < 1.0);
print_perf("rand_mat_stat", tmin);
// rand mat mul
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
t = clock_now();
double *C = randmatmul(1000);
assert(0 <= C[0]);
free(C);
t = clock_now()-t;
if (t < tmin) tmin = t;
}
print_perf("rand_mat_mul", tmin);
// printfd
tmin = INFINITY;
for (int i=0; i<NITER; ++i) {
t = clock_now();
printfd(100000);
t = clock_now()-t;
if (t < tmin) tmin = t;
}
print_perf("printfd", tmin);
return 0;
}
void *quicksort(void *arg) {
int pivot;
int i,j,tmp;
pthread_t lthread, rthread;
// used only for keeping initial field values
struct qs qs_info;
// the address of the struct is passed as a function parameter
// and for this reason we dont want the same address in both qs calls
struct qs *qs_left;
struct qs *qs_right;
struct qs *qs_temp;
// we use heap memory instead of static to avoid loss of local data
// before being read from their callees
qs_left = (struct qs *)malloc(sizeof(struct qs));
qs_right = (struct qs *)malloc(sizeof(struct qs));
// intializing struct fields
qs_temp = (struct qs *)arg;
qs_info.left = qs_temp->left;
qs_info.right = qs_temp->right;
qs_info.curr_layer = qs_temp->curr_layer;
qs_left->curr_layer = qs_temp->curr_layer;
qs_right->curr_layer = qs_temp->curr_layer;
qs_info.max_layer = qs_temp->max_layer;
qs_left->max_layer = qs_temp->max_layer;
qs_right->max_layer = qs_temp->max_layer;
free((struct qs *)arg);
trace(&qs_info);
// pivot calculation
if (qs_info.left < qs_info.right) {
pivot = qs_info.left;
i = qs_info.left;
j = qs_info.right;
while(i<j) {
while((a[i]<=a[pivot]) && (i < qs_info.right))
i++;
while(a[j] > a[pivot])
j--;
if (i<j){
tmp = a[i];
a[i] = a[j];
a[j] = tmp;
}
}
tmp = a[pivot];
a[pivot] = a[j];
a[j] = tmp;
//decide wether to continue creating threads or not
if (qs_info.curr_layer == qs_info.max_layer) {
qs_left->left = qs_info.left;
qs_left->right = j-1;
quicksort(qs_left);
qs_right->left = j+1;
qs_right->right = qs_info.right;
quicksort(qs_right);
}
else {
qs_left->curr_layer++;
qs_left->left = qs_info.left;
qs_left->right = j-1;
trace(qs_left);
if(pthread_create(<hread, NULL, quicksort, qs_left)) {
perror("pthread_create");
exit(1);
}
qs_right->curr_layer++;
qs_right->left = j+1;
qs_right->right = qs_info.right;
trace(qs_right);
if(pthread_create(&rthread, NULL, quicksort, qs_right)) {
perror("pthread_create");
exit(1);
}
}
}
return(NULL);
}
void qsort_wrapper(int a[], size_t n)
{
quicksort(a, 0, n-1);
}
/* sorts items alphabetically and indexes them */
void menu_sort_menu(menu_st *menulist)
{
quicksort(menulist->items, 0, menulist->num_items-1);
}
void sort(Item a[], int l, int r)
{ quicksort(a, l, r);
}
mxArray* ompcore(double D[], double x[], double DtX[], double XtX[], double G[], mwSize n, mwSize m, mwSize L,
int T, double eps, int gamma_mode, int profile, double msg_delta, int erroromp)
{
profdata pd;
mxArray *Gamma;
mwIndex i, j, signum, pos, *ind, *gammaIr, *gammaJc, gamma_count;
mwSize allocated_coefs, allocated_cols;
int DtX_specified, XtX_specified, batchomp, standardomp, *selected_atoms;
double *alpha, *r, *Lchol, *c, *Gsub, *Dsub, sum, *gammaPr, *tempvec1, *tempvec2;
double eps2, resnorm, delta, deltaprev, secs_remain;
int mins_remain, hrs_remain;
clock_t lastprint_time, starttime;
/*** status flags ***/
DtX_specified = (DtX!=0); /* indicates whether D'*x was provided */
XtX_specified = (XtX!=0); /* indicates whether sum(x.*x) was provided */
standardomp = (G==0); /* batch-omp or standard omp are selected depending on availability of G */
batchomp = !standardomp;
/*** allocate output matrix ***/
if (gamma_mode == FULL_GAMMA) {
/* allocate full matrix of size m X L */
Gamma = mxCreateDoubleMatrix(m, L, mxREAL);
gammaPr = mxGetPr(Gamma);
gammaIr = 0;
gammaJc = 0;
}
else {
/* allocate sparse matrix with room for allocated_coefs nonzeros */
/* for error-omp, begin with L*sqrt(n)/2 allocated nonzeros, otherwise allocate L*T nonzeros */
allocated_coefs = erroromp ? (mwSize)(ceil(L*sqrt((double)n)/2.0) + 1.01) : L*T;
Gamma = mxCreateSparse(m, L, allocated_coefs, mxREAL);
gammaPr = mxGetPr(Gamma);
gammaIr = mxGetIr(Gamma);
gammaJc = mxGetJc(Gamma);
gamma_count = 0;
gammaJc[0] = 0;
}
/*** helper arrays ***/
alpha = (double*)mxMalloc(m*sizeof(double)); /* contains D'*residual */
ind = (mwIndex*)mxMalloc(n*sizeof(mwIndex)); /* indices of selected atoms */
selected_atoms = (int*)mxMalloc(m*sizeof(int)); /* binary array with 1's for selected atoms */
c = (double*)mxMalloc(n*sizeof(double)); /* orthogonal projection result */
/* current number of columns in Dsub / Gsub / Lchol */
allocated_cols = erroromp ? (mwSize)(ceil(sqrt((double)n)/2.0) + 1.01) : T;
/* Cholesky decomposition of D_I'*D_I */
Lchol = (double*)mxMalloc(n*allocated_cols*sizeof(double));
/* temporary vectors for various computations */
tempvec1 = (double*)mxMalloc(m*sizeof(double));
tempvec2 = (double*)mxMalloc(m*sizeof(double));
if (batchomp) {
/* matrix containing G(:,ind) - the columns of G corresponding to the selected atoms, in order of selection */
Gsub = (double*)mxMalloc(m*allocated_cols*sizeof(double));
}
else {
/* matrix containing D(:,ind) - the selected atoms from D, in order of selection */
Dsub = (double*)mxMalloc(n*allocated_cols*sizeof(double));
/* stores the residual */
r = (double*)mxMalloc(n*sizeof(double));
}
if (!DtX_specified) {
/* contains D'*x for the current signal */
DtX = (double*)mxMalloc(m*sizeof(double));
}
/*** initializations for error omp ***/
if (erroromp) {
eps2 = eps*eps; /* compute eps^2 */
if (T<0 || T>n) { /* unspecified max atom num - set max atoms to n */
T = n;
}
}
/*** initialize timers ***/
initprofdata(&pd); /* initialize profiling counters */
starttime = clock(); /* record starting time for eta computations */
lastprint_time = starttime; /* time of last status display */
/********************** perform omp for each signal **********************/
for (signum=0; signum<L; ++signum) {
/* compute DtX */
if (!DtX_specified) {
matT_vec(1, D, x+n*signum, DtX, n, m);
addproftime(&pd, DtX_TIME);
}
/* initialize alpha := DtX */
memcpy(alpha, DtX + m*signum*DtX_specified, m*sizeof(double));
/* mark all atoms as unselected */
for (i=0; i<m; ++i) {
selected_atoms[i] = 0;
}
/* initialize residual norm and deltaprev for error-omp */
if (erroromp) {
if (XtX_specified) {
resnorm = XtX[signum];
}
else {
resnorm = dotprod(x+n*signum, x+n*signum, n);
addproftime(&pd, XtX_TIME);
}
deltaprev = 0; /* delta tracks the value of gamma'*G*gamma */
}
else {
/* ignore residual norm stopping criterion */
eps2 = 0;
resnorm = 1;
}
/* main loop */
i=0;
while (resnorm>eps2 && i<T) {
/* index of next atom */
pos = maxabs(alpha, m);
addproftime(&pd, MAXABS_TIME);
/* stop criterion: selected same atom twice, or inner product too small */
if (selected_atoms[pos] || alpha[pos]*alpha[pos]<1e-14) {
break;
}
/* mark selected atom */
ind[i] = pos;
selected_atoms[pos] = 1;
/* matrix reallocation */
if (erroromp && i>=allocated_cols) {
allocated_cols = (mwSize)(ceil(allocated_cols*MAT_INC_FACTOR) + 1.01);
Lchol = (double*)mxRealloc(Lchol,n*allocated_cols*sizeof(double));
batchomp ? (Gsub = (double*)mxRealloc(Gsub,m*allocated_cols*sizeof(double))) :
(Dsub = (double*)mxRealloc(Dsub,n*allocated_cols*sizeof(double))) ;
}
/* append column to Gsub or Dsub */
if (batchomp) {
memcpy(Gsub+i*m, G+pos*m, m*sizeof(double));
}
else {
memcpy(Dsub+i*n, D+pos*n, n*sizeof(double));
}
/*** Cholesky update ***/
if (i==0) {
*Lchol = 1;
}
else {
/* incremental Cholesky decomposition: compute next row of Lchol */
if (standardomp) {
matT_vec(1, Dsub, D+n*pos, tempvec1, n, i); /* compute tempvec1 := Dsub'*d where d is new atom */
addproftime(&pd, DtD_TIME);
}
else {
vec_assign(tempvec1, Gsub+i*m, ind, i); /* extract tempvec1 := Gsub(ind,i) */
}
backsubst('L', Lchol, tempvec1, tempvec2, n, i); /* compute tempvec2 = Lchol \ tempvec1 */
for (j=0; j<i; ++j) { /* write tempvec2 to end of Lchol */
Lchol[j*n+i] = tempvec2[j];
}
/* compute Lchol(i,i) */
sum = 0;
for (j=0; j<i; ++j) { /* compute sum of squares of last row without Lchol(i,i) */
sum += SQR(Lchol[j*n+i]);
}
if ( (1-sum) <= 1e-14 ) { /* Lchol(i,i) is zero => selected atoms are dependent */
break;
}
Lchol[i*n+i] = sqrt(1-sum);
}
addproftime(&pd, LCHOL_TIME);
i++;
/* perform orthogonal projection and compute sparse coefficients */
vec_assign(tempvec1, DtX + m*signum*DtX_specified, ind, i); /* extract tempvec1 = DtX(ind) */
cholsolve('L', Lchol, tempvec1, c, n, i); /* solve LL'c = tempvec1 for c */
addproftime(&pd, COMPCOEF_TIME);
/* update alpha = D'*residual */
if (standardomp) {
mat_vec(-1, Dsub, c, r, n, i); /* compute r := -Dsub*c */
vec_sum(1, x+n*signum, r, n); /* compute r := x+r */
/*memcpy(r, x+n*signum, n*sizeof(double)); /* assign r := x */
/*mat_vec1(-1, Dsub, c, 1, r, n, i); /* compute r := r-Dsub*c */
addproftime(&pd, COMPRES_TIME);
matT_vec(1, D, r, alpha, n, m); /* compute alpha := D'*r */
addproftime(&pd, DtR_TIME);
/* update residual norm */
if (erroromp) {
resnorm = dotprod(r, r, n);
addproftime(&pd, UPDATE_RESNORM_TIME);
}
}
else {
mat_vec(1, Gsub, c, tempvec1, m, i); /* compute tempvec1 := Gsub*c */
memcpy(alpha, DtX + m*signum*DtX_specified, m*sizeof(double)); /* set alpha = D'*x */
vec_sum(-1, tempvec1, alpha, m); /* compute alpha := alpha - tempvec1 */
addproftime(&pd, UPDATE_DtR_TIME);
/* update residual norm */
if (erroromp) {
vec_assign(tempvec2, tempvec1, ind, i); /* assign tempvec2 := tempvec1(ind) */
delta = dotprod(c,tempvec2,i); /* compute c'*tempvec2 */
resnorm = resnorm - delta + deltaprev; /* residual norm update */
deltaprev = delta;
addproftime(&pd, UPDATE_RESNORM_TIME);
}
}
}
/*** generate output vector gamma ***/
if (gamma_mode == FULL_GAMMA) { /* write the coefs in c to their correct positions in gamma */
for (j=0; j<i; ++j) {
gammaPr[m*signum + ind[j]] = c[j];
}
}
else {
/* sort the coefs by index before writing them to gamma */
quicksort(ind,c,i);
addproftime(&pd, INDEXSORT_TIME);
/* gamma is full - reallocate */
if (gamma_count+i >= allocated_coefs) {
while(gamma_count+i >= allocated_coefs) {
allocated_coefs = (mwSize)(ceil(GAMMA_INC_FACTOR*allocated_coefs) + 1.01);
}
mxSetNzmax(Gamma, allocated_coefs);
mxSetPr(Gamma, mxRealloc(gammaPr, allocated_coefs*sizeof(double)));
mxSetIr(Gamma, mxRealloc(gammaIr, allocated_coefs*sizeof(mwIndex)));
gammaPr = mxGetPr(Gamma);
gammaIr = mxGetIr(Gamma);
}
/* append coefs to gamma and update the indices */
for (j=0; j<i; ++j) {
gammaPr[gamma_count] = c[j];
gammaIr[gamma_count] = ind[j];
gamma_count++;
}
gammaJc[signum+1] = gammaJc[signum] + i;
}
/*** display status messages ***/
if (msg_delta>0 && (clock()-lastprint_time)/(double)CLOCKS_PER_SEC >= msg_delta)
{
lastprint_time = clock();
/* estimated remainig time */
secs2hms( ((L-signum-1)/(double)(signum+1)) * ((lastprint_time-starttime)/(double)CLOCKS_PER_SEC) ,
&hrs_remain, &mins_remain, &secs_remain);
mexPrintf("omp: signal %d / %d, estimated remaining time: %02d:%02d:%05.2f\n",
signum+1, L, hrs_remain, mins_remain, secs_remain);
mexEvalString("drawnow;");
}
}
/* end omp */
/*** print final messages ***/
if (msg_delta>0) {
mexPrintf("omp: signal %d / %d\n", signum, L);
}
if (profile) {
printprofinfo(&pd, erroromp, batchomp, L);
}
/* free memory */
if (!DtX_specified) {
mxFree(DtX);
}
if (standardomp) {
mxFree(r);
mxFree(Dsub);
}
else {
mxFree(Gsub);
}
mxFree(tempvec2);
mxFree(tempvec1);
mxFree(Lchol);
mxFree(c);
mxFree(selected_atoms);
mxFree(ind);
mxFree(alpha);
return Gamma;
}
void quicksort(int a[], int l, int r){
int i;
i = partition(a, l, r);
quicksort(a, l, i-1);
quicksort(a, i+1, r);
}
void main()
{
int i,j,k,num,a[max];
fp1=fopen("BIGFILE","w");
for(i=0;i<64*max;i++)
{
num=rand() % 1000;
fprintf(fp1,"%d\n",num);
}
fclose(fp1);
fp1=fopen("BIGFILE","r");
rewind(fp1);
for(i=1;i<=64;i++)
{
name[4]=i+48;
fp2=fopen(name,"w");
for(j=0;j<max;j++)
{
fscanf(fp1,"%d",&num);
fprintf(fp2,"%d\n",num);
}
fclose(fp2);
}
fclose(fp1);
for(i=1;i<=64;i++)
{
name[4]=i+48;
fp2=fopen(name,"r");
for(j=0;j<max;j++)
{
fscanf(fp2,"%d",&a[j]);
}
fclose(fp2);
quicksort(a,0,max-1);
fp2=fopen(name,"w");
for(j=0;j<max;j++)
{
fprintf(fp2,"%d\n",a[j]);
}
fclose(fp2);
}
i=1;
for(k=0;k<8;k++) //8 way merge
{
name2[5]=k+48+1;
FILE *f1,*f2,*f3,*f4,*f5,*f6,*f7,*f8,*f88;
int a1,a2,a3,a4,a5,a6,a7,a8,arr[8],pos,min,z;
name[4]=i+48;
f1=fopen(name,"r");
rewind(f1);
fscanf(f1,"%d",&a1);
arr[0]=a1;
i++;
name[4]=i+48;
f2=fopen(name,"r");
rewind(f2);
fscanf(f2,"%d",&a2);
arr[1]=a2;
i++;
name[4]=i+48;
f3=fopen(name,"r");
rewind(f3);
fscanf(f3,"%d",&a3);
arr[2]=a3;
i++;
name[4]=i+48;
f4=fopen(name,"r");
rewind(f4);
fscanf(f4,"%d",&a4);
arr[3]=a4;
i++;
name[4]=i+48;
f5=fopen(name,"r");
rewind(f5);
fscanf(f5,"%d",&a5);
arr[4]=a5;
i++;
name[4]=i+48;
f6=fopen(name,"r");
rewind(f6);
fscanf(f6,"%d",&a6);
arr[5]=a6;
i++;
name[4]=i+48;
f7=fopen(name,"r");
rewind(f7);
fscanf(f7,"%d",&a7);
arr[6]=a7;
i++;
name[4]=i+48;
f8=fopen(name,"r");
rewind(f8);
fscanf(f8,"%d",&a8);
arr[7]=a8;
i++;
f88=fopen(name2,"w");
for(z=0;z<8*max;z++)
{
pos=minpos(arr,8);
min=minele(arr,8);
printf("%d--%d ",pos,min);
fprintf(f88,"%d\n",min);
if(pos==0)
fscanf(f1,"%d",&arr[0]);
else if(pos==1)
fscanf(f2,"%d",&arr[1]);
else if(pos==2)
fscanf(f3,"%d",&arr[2]);
else if(pos==3)
fscanf(f4,"%d",&arr[3]);
else if(pos==4)
fscanf(f5,"%d",&arr[4]);
else if(pos==5)
fscanf(f6,"%d",&arr[5]);
else if(pos==6)
fscanf(f7,"%d",&arr[6]);
else if(pos==7)
fscanf(f8,"%d",&arr[7]);
}
fclose(f88);
}
i=1;
for(k=0;k<4;k++) //4 way merge
{
name3[5]=k+48+1;
FILE *f1,*f2,*f3,*f4,*f44;
int a1,a2,a3,a4,arr[4],pos,min,z;
name2[5]=i+48;
f1=fopen(name2,"r");
rewind(f1);
fscanf(f1,"%d",&a1);
arr[0]=a1;
i++;
name2[5]=i+48;
f2=fopen(name2,"r");
rewind(f2);
fscanf(f2,"%d",&a2);
arr[1]=a2;
i++;
name2[5]=i+48;
f3=fopen(name2,"r");
fscanf(f3,"%d",&a3);
rewind(f3);
arr[2]=a3;
i++;
name2[5]=i+48;
f4=fopen(name2,"r");
rewind(f4);
fscanf(f4,"%d",&a4);
arr[3]=a4;
i++;
f44=fopen(name3,"w");
for(z=0;z<4*max;z++)
{
pos=minpos(arr,4);
min=minele(arr,4);
printf("%d--%d ",pos,min);
fprintf(f44,"%d\n",min);
if(pos==0)
fscanf(f1,"%d",&arr[0]);
else if(pos==1)
fscanf(f2,"%d",&arr[1]);
else if(pos==2)
fscanf(f3,"%d",&arr[2]);
else if(pos==3)
fscanf(f4,"%d",&arr[3]);
}
fclose(f44);
i=1;
}
i=1;
for(k=0;k<1;k++) //2 way merge
{
name4[5]=k+48+1;
FILE *f1,*f2,*f22;
int a1,a2,arr[2],pos,min,z;
name3[5]=i+48;
f1=fopen(name3,"r");
rewind(f1);
fscanf(f1,"%d",&a1);
arr[0]=a1;
i++;
name3[5]=i+48;
f2=fopen(name3,"r");
rewind(f2);
fscanf(f2,"%d",&a2);
arr[1]=a2;
i++;
f22=fopen(name4,"w");
for(z=0;z<2*max;z++)
{
pos=minpos(arr,2);
min=minele(arr,2);
printf("%d--%d ",pos,min);
fprintf(f22,"%d\n",min);
if(pos==0)
fscanf(f1,"%d",&arr[0]);
else if(pos==1)
fscanf(f2,"%d",&arr[1]);
}
fclose(f22);
}
}
void sort (array_t &v) {
quicksort(v, 0, v.size());
}
void quicksort(int array[], size_t size) {
quicksort(array, 0, size);
}
/*
Main method:
-generate random list
-time sequential quicksort
-time parallel quicksort
-time standard qsort
*/
int main(int argc, char *argv[])
{
struct timeval start, end;
double diff;
srand(time(NULL)); //seed random
int NUM = DNUM;
int THREAD_LEVEL = T_LEVEL;
if (argc == 2) //user specified list size.
{
NUM = atoi(argv[1]);
}
else if (argc == 3) {
/* code */
NUM = atoi(argv[1]);
THREAD_LEVEL = atoi(argv[2]);
}
//Want to compare sorting on the same list,
//so backup.
double *lystbck = (double *) malloc(NUM * sizeof(double));
double *lyst = (double *) malloc(NUM * sizeof(double));
//Populate random original/backup list.
for (int i = 0; i < NUM; i++) {
lystbck[i] = 1.0 * rand() / RAND_MAX;
}
//copy list.
memcpy(lyst, lystbck, NUM * sizeof(double));
//Sequential mergesort, and timing.
gettimeofday(&start, NULL);
quicksort(lyst, NUM);
gettimeofday(&end, NULL);
if (!isSorted(lyst, NUM)) {
printf("Oops, lyst did not get sorted by quicksort.\n");
}
//Compute time difference.
diff = ((end.tv_sec * 1000000 + end.tv_usec)
- (start.tv_sec * 1000000 + start.tv_usec)) / 1000000.0;
printf("Sequential quicksort took: %lf sec.\n", diff);
//Now, parallel quicksort.
//copy list.
memcpy(lyst, lystbck, NUM * sizeof(double));
gettimeofday(&start, NULL);
parallelQuicksort(lyst, NUM, THREAD_LEVEL);
gettimeofday(&end, NULL);
if (!isSorted(lyst, NUM)) {
printf("Oops, lyst did not get sorted by parallelQuicksort.\n");
}
//Compute time difference.
diff = ((end.tv_sec * 1000000 + end.tv_usec)
- (start.tv_sec * 1000000 + start.tv_usec)) / 1000000.0;
printf("Parallel quicksort took: %lf sec.\n", diff);
//Finally, built-in for reference:
memcpy(lyst, lystbck, NUM * sizeof(double));
gettimeofday(&start, NULL);
qsort(lyst, NUM, sizeof(double), compare_doubles);
gettimeofday(&end, NULL);
if (!isSorted(lyst, NUM)) {
printf("Oops, lyst did not get sorted by qsort.\n");
}
//Compute time difference.
diff = ((end.tv_sec * 1000000 + end.tv_usec)
- (start.tv_sec * 1000000 + start.tv_usec)) / 1000000.0;
printf("Built-in quicksort took: %lf sec.\n", diff);
free(lyst);
free(lystbck);
pthread_exit(NULL);
}
int gmres_dr(spinor * const P,spinor * const Q,
const int m, const int nr_ev, const int max_restarts,
const double eps_sq, const int rel_prec,
const int N, matrix_mult f){
int restart=0, i, j, k, l;
double beta, eps, norm, beta2=0.;
complex *lswork = NULL;
int lwork;
complex tmp1, tmp2;
int info=0;
int _m = m, mp1 = m+1, np1 = nr_ev+1, ne = nr_ev, V2 = 12*(VOLUMEPLUSRAND)/2, _N = 12*N;
spinor ** solver_field = NULL;
const int nr_sf = 3;
if(N == VOLUME) {
init_solver_field(&solver_field, VOLUMEPLUSRAND, nr_sf);
}
else {
init_solver_field(&solver_field, VOLUMEPLUSRAND/2, nr_sf);
}
double err=0.;
spinor * r0, * x0;
cmone.re = -1.; cmone.im=0.;
cpone.re = 1.; cpone.im=0.;
czero.re = 0.; czero.im = 0.;
r0 = solver_field[0];
x0 = solver_field[2];
eps=sqrt(eps_sq);
init_gmres_dr(m, (VOLUMEPLUSRAND));
norm = sqrt(square_norm(Q, N, 1));
assign(x0, P, N);
/* first normal GMRES cycle */
/* r_0=Q-AP (b=Q, x+0=P) */
f(r0, x0);
diff(r0, Q, r0, N);
/* v_0=r_0/||r_0|| */
alpha[0].re=sqrt(square_norm(r0, N, 1));
err = alpha[0].re;
if(g_proc_id == g_stdio_proc && g_debug_level > 0){
printf("%d\t%e true residue\n", restart*m, alpha[0].re*alpha[0].re);
fflush(stdout);
}
if(alpha[0].re==0.){
assign(P, x0, N);
finalize_solver(solver_field, nr_sf);
return(restart*m);
}
mul_r(V[0], 1./alpha[0].re, r0, N);
for(j = 0; j < m; j++){
/* solver_field[0]=A*v_j */
/* Set h_ij and omega_j */
/* solver_field[1] <- omega_j */
f(solver_field[1], V[j]);
/* assign(solver_field[1], solver_field[0], N); */
for(i = 0; i <= j; i++){
H[i][j] = scalar_prod(V[i], solver_field[1], N, 1);
/* G, work and work2 are in Fortran storage: columns first */
G[j][i] = H[i][j];
work2[j][i] = H[i][j];
work[i][j].re = H[i][j].re;
work[i][j].im = -H[i][j].im;
assign_diff_mul(solver_field[1], V[i], H[i][j], N);
}
_complex_set(H[j+1][j], sqrt(square_norm(solver_field[1], N, 1)), 0.);
G[j][j+1] = H[j+1][j];
work2[j][j+1] = H[j+1][j];
work[j+1][j].re = H[j+1][j].re;
work[j+1][j].im = -H[j+1][j].im;
beta2 = H[j+1][j].re*H[j+1][j].re;
for(i = 0; i < j; i++){
tmp1 = H[i][j];
tmp2 = H[i+1][j];
_mult_real(H[i][j], tmp2, s[i]);
_add_assign_complex_conj(H[i][j], c[i], tmp1);
_mult_real(H[i+1][j], tmp1, s[i]);
_diff_assign_complex(H[i+1][j], c[i], tmp2);
}
/* Set beta, s, c, alpha[j],[j+1] */
beta = sqrt(_complex_square_norm(H[j][j]) + _complex_square_norm(H[j+1][j]));
s[j] = H[j+1][j].re / beta;
_mult_real(c[j], H[j][j], 1./beta);
_complex_set(H[j][j], beta, 0.);
_mult_real(alpha[j+1], alpha[j], s[j]);
tmp1 = alpha[j];
_mult_assign_complex_conj(alpha[j], c[j], tmp1);
/* precision reached? */
if(g_proc_id == g_stdio_proc && g_debug_level > 0){
printf("%d\t%e residue\n", restart*m+j, alpha[j+1].re*alpha[j+1].re);
fflush(stdout);
}
if(((alpha[j+1].re <= eps) && (rel_prec == 0)) || ((alpha[j+1].re <= eps*norm) && (rel_prec == 1))){
_mult_real(alpha[j], alpha[j], 1./H[j][j].re);
assign_add_mul(x0, V[j], alpha[j], N);
for(i = j-1; i >= 0; i--){
for(k = i+1; k <= j; k++){
_mult_assign_complex(tmp1, H[i][k], alpha[k]);
/* alpha[i] -= tmp1 */
_diff_complex(alpha[i], tmp1);
}
_mult_real(alpha[i], alpha[i], 1./H[i][i].re);
assign_add_mul(x0, V[i], alpha[i], N);
}
for(i = 0; i < m; i++){
alpha[i].im = 0.;
}
assign(P, x0, N);
finalize_solver(solver_field, nr_sf);
return(restart*m+j);
}
/* if not */
else {
mul_r(V[(j+1)], 1./H[j+1][j].re, solver_field[1], N);
}
}
j=m-1;
/* prepare for restart */
_mult_real(alpha[j], alpha[j], 1./H[j][j].re);
assign_add_mul(x0, V[j], alpha[j], N);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("alpha: %e %e\n", alpha[j].re, alpha[j].im);
}
for(i = j-1; i >= 0; i--){
for(k = i+1; k <= j; k++){
_mult_assign_complex(tmp1, H[i][k], alpha[k]);
_diff_complex(alpha[i], tmp1);
}
_mult_real(alpha[i], alpha[i], 1./H[i][i].re);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("alpha: %e %e\n", alpha[i].re, alpha[i].im);
}
assign_add_mul(x0, V[i], alpha[i], N);
}
/* This produces c=V_m+1*r0 */
for(i = 0; i < mp1; i++) {
c[i] = scalar_prod(V[i], r0, N, 1);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("c: %e %e err = %e\n", c[i].re, c[i].im, err);
}
}
for(restart = 1; restart < max_restarts; restart++) {
/* compute c-\bar H \alpha */
_FT(zgemv) ("N", &mp1, &_m, &cmone, G[0], &mp1, alpha, &one, &cpone, c, &one, 1);
err = sqrt(short_scalar_prod(c, c, mp1).re);
if(g_proc_id == 0 && g_debug_level > 0) {
printf("%d\t %e short residue\n", m*restart, err*err);
}
/* Compute new residual r0 */
/* r_0=Q-AP (b=Q, x+0=P) */
if(g_debug_level > 0) {
f(r0, x0);
diff(r0, Q, r0, N);
tmp1.im=sqrt(square_norm(r0, N, 1));
if(g_proc_id == g_stdio_proc){
printf("%d\t%e true residue\n", m*restart, tmp1.im*tmp1.im);
fflush(stdout);
}
}
mul(r0, c[0], V[0], N);
for(i = 1; i < mp1; i++) {
assign_add_mul(r0, V[i], c[i], N);
}
if(g_debug_level > 3) {
tmp1.im=sqrt(square_norm(r0, N, 1));
if(g_proc_id == g_stdio_proc){
printf("%d\t%e residue\n", m*restart, tmp1.im*tmp1.im);
fflush(stdout);
}
}
/* Stop if satisfied */
if(err < eps){
assign(P, x0, N);
finalize_solver(solver_field, nr_sf);
return(restart*m);
}
/* Prepare to compute harmonic Ritz pairs */
for(i = 0; i < m-1; i++){
alpha[i].re = 0.;
alpha[i].im = 0.;
}
alpha[m-1].re = 1.;
alpha[m-1].im = 0.;
_FT(zgesv) (&_m, &one, work[0], &mp1, idx, alpha, &_m, &info);
for(i = 0; i < m; i++) {
G[m-1][i].re += (beta2*alpha[idx[i]-1].re);
G[m-1][i].im += (beta2*alpha[idx[i]-1].im);
}
if(g_proc_id == 0 && g_debug_level > 3){
printf("zgesv returned info = %d, c[m-1]= %e, %e , idx[m-1]=%d\n",
info, alpha[idx[m-1]-1].re, alpha[idx[m-1]-1].im, idx[m-1]);
}
/* c - \bar H * d -> c */
/* G contains H + \beta^2 H^-He_n e_n^H */
/* Compute harmonic Ritz pairs */
diagonalise_general_matrix(m, G[0], mp1, alpha, evalues);
for(i = 0; i < m; i++) {
sortarray[i] = _complex_square_norm(evalues[i]);
idx[i] = i;
}
quicksort(m, sortarray, idx);
if(g_proc_id == g_stdio_proc && g_debug_level > 1) {
for(i = 0; i < m; i++) {
printf("# Evalues %d %e %e \n", i, evalues[idx[i]].re, evalues[idx[i]].im);
}
fflush(stdout);
}
/* Copy the first nr_ev eigenvectors to work */
for(i = 0; i < ne; i++) {
for(l = 0; l < m; l++) {
work[i][l] = G[idx[i]][l];
}
}
/* Orthonormalize them */
for(i = 0; i < ne; i++) {
work[i][m].re = 0.;
work[i][m].im = 0.;
short_ModifiedGS(work[i], m, i, work[0], mp1);
}
/* Orthonormalize c - \bar H d to work */
short_ModifiedGS(c, m+1, ne, work[0], mp1);
for(i = 0; i < mp1; i++) {
work[nr_ev][i] = c[i];
}
/* Now compute \bar H = P^T_k+1 \bar H_m P_k */
for(i = 0; i < mp1; i++) {
for(l = 0; l < mp1; l++) {
H[i][l].re = 0.;
H[i][l].im = 0.;
}
}
_FT(zgemm) ("N", "N", &mp1, &ne, &_m, &cpone, work2[0], &mp1, work[0], &mp1, &czero, G[0], &mp1, 1, 1);
_FT(zgemm) ("C", "N", &np1, &ne , &mp1, &cpone, work[0], &mp1, G[0], &mp1, &czero, H[0], &mp1, 1, 1);
if(g_debug_level > 3) {
for(i = 0; i < ne+1; i++) {
for(l = 0; l < ne+1; l++) {
if(g_proc_id == 0) {
printf("(g[%d], g[%d]) = %e, %e\n", i, l, short_scalar_prod(work[i], work[l], m+1).re,
short_scalar_prod(work[i], work[l], m+1).im);
printf("(g[%d], g[%d]) = %e, %e\n", l, i, short_scalar_prod(work[l], work[i], m+1).re,
short_scalar_prod(work[l], work[i], m+1).im);
}
}
}
}
/* V_k+1 = V_m+1 P_k+1 */
/* _FT(zgemm) ("N", "N", &_N, &np1, &mp1, &cpone, (complex*)V[0], &V2, work[0], &mp1, &czero, (complex*)Z[0], &V2, 1, 1); */
for(l = 0; l < np1; l++) {
mul(Z[l], work[l][0], V[0], N);
for(i = 1; i < mp1; i++) {
assign_add_mul(Z[l], V[i], work[l][i], N);
}
}
/* copy back to V */
for(i = 0; i < np1; i++) {
assign(V[i], Z[i], N);
}
/* Reorthogonalise v_nr_ev */
ModifiedGS((complex*)V[nr_ev], _N, nr_ev, (complex*)V[0], V2);
if(g_debug_level > 3) {
for(i = 0; i < np1; i++) {
for(l = 0; l < np1; l++) {
tmp1 = scalar_prod(V[l], V[i], N, 1);
if(g_proc_id == 0) {
printf("(V[%d], V[%d]) = %e %e %d %d %d %d %d %d %e %e\n", l, i, tmp1.re, tmp1.im, np1, mp1, ne, _m, _N, V2, H[l][i].re, H[l][i].im);
}
}
}
}
/* Copy the content of H to work, work2 and G */
for(i=0; i < mp1; i++) {
for(l = 0; l < mp1; l++) {
G[i][l] = H[i][l];
work2[i][l] = H[i][l];
work[l][i].re = H[i][l].re;
work[l][i].im = -H[i][l].im;
}
}
for(j = ne; j < m; j++) {
/* solver_field[0]=A*v_j */
f(solver_field[1], V[j]);
/* Set h_ij and omega_j */
/* solver_field[1] <- omega_j */
/* assign(solver_field[1], solver_field[0], N); */
for(i = 0; i <= j; i++){
H[j][i] = scalar_prod(V[i], solver_field[1], N, 1);
/* H, G, work and work2 are now all in Fortran storage: columns first */
G[j][i] = H[j][i];
work2[j][i] = H[j][i];
work[i][j].re = H[j][i].re;
work[i][j].im = -H[j][i].im;
assign_diff_mul(solver_field[1], V[i], H[j][i], N);
}
beta2 = square_norm(solver_field[1], N, 1);
_complex_set(H[j][j+1], sqrt(beta2), 0.);
G[j][j+1] = H[j][j+1];
work2[j][j+1] = H[j][j+1];
work[j+1][j].re = H[j][j+1].re;
work[j+1][j].im = -H[j][j+1].im;
mul_r(V[(j+1)], 1./H[j][j+1].re, solver_field[1], N);
}
/* Solve the least square problem for alpha*/
/* This produces c=V_m+1*r0 */
for(i = 0; i < mp1; i++) {
c[i] = scalar_prod(V[i], r0, N, 1);
alpha[i] = c[i];
if(g_proc_id == 0 && g_debug_level > 3) {
printf("c: %e %e err = %e\n", c[i].re, c[i].im, err);
}
}
if(lswork == NULL) {
lwork = -1;
_FT(zgels) ("N", &mp1, &_m, &one, H[0], &mp1, alpha, &mp1, &tmp1, &lwork, &info, 1);
lwork = (int)tmp1.re;
lswork = (complex*)malloc(lwork*sizeof(complex));
}
_FT(zgels) ("N", &mp1, &_m, &one, H[0], &mp1, alpha, &mp1, lswork, &lwork, &info, 1);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("zgels returned info = %d\n", info);
fflush(stdout);
}
/* Compute the new solution vector */
for(i = 0; i < m; i++){
if(g_proc_id == 0 && g_debug_level > 3) {
printf("alpha: %e %e\n", alpha[i].re, alpha[i].im);
}
assign_add_mul(x0, V[i], alpha[i], N);
}
}
/* If maximal number of restart is reached */
assign(P, x0, N);
finalize_solver(solver_field, nr_sf);
return(-1);
}
void QuickSort::sort(int* aArray, int aLength)
{
quicksort(aArray, 0, aLength-1);
}
void sorter(int fd)
{
int i, tempfile;
size_t readsize;
int buffer[(int) allowedBuffer/sizeof(int)];
int flno;
int file1, file2, target;
size_t size1, size2; /* Amount of bytes read from file1 and file 2 */
int a, b; /* Elements from file 1 and file 2 */
for (i=0; i<fileAmount; i++)
{
char newfile[10];
sprintf(newfile, "file%d", i);
tempfile=creat(newfile, 0666);
readsize=read(fd, &buffer, allowedBuffer);
quicksort(buffer, 0, (int) allowedBuffer/sizeof(int));
/*qsort(buffer, (size_t) readsize/sizeof(int), sizeof(int), comparator); */ /* In case things go south with wiki-quicksort */
write(tempfile, buffer, readsize);
close(tempfile);
}
/* printf("Finished splitting and sorting temporary files\n"); */
lseek(fd, 0, SEEK_SET);
/* printf("Amount of temp files is (%d)\n", fileAmount); */
for (flno=0; flno<fileAmount; flno+=2) /* This is a \"tail\" file. So we are taking 1st and 2nd file, then 2nd and 3rd... We don't want when we take n-1 and n fileto go out of reach */
{
/* printf("Entered %d iteration\n", flno); */
int buffer[(int) allowedBuffer/sizeof(int)];
char newfile[10];
sprintf(newfile, "file%d", flno);
/* printf("Old tempfile to work with is %s\n", newfile); */
file1=open(newfile, O_RDWR, 0666);
sprintf(newfile, "file%d", flno+1);
/* printf("Old tempfile to work with is %s\n", newfile); */
file2=open(newfile, O_RDWR, 0666);
fileAmount++;
sprintf(newfile, "file%d", fileAmount);
/* printf("Created new tempfile to work with - %s\n", newfile); */
target=creat(newfile, 0666);
size1=read(file1, &a, sizeof(int));
size2=read(file2, &b, sizeof(int));
/* printf("a=[%d], b=[%d]\n", a, b); */
while (((int) size1>0) && ((int) size2>0))
{
if (a<b)
{
write(target, &a, sizeof(int));
size1=read(file1, &a, sizeof(int));
}
else
{
if (a>b)
{
write(target, &b, sizeof(int));
size2=read(file2, &b, sizeof(int));
}
else
{
write(target, &a, sizeof(int));
write(target, &b, sizeof(int));
size1=read(file1, &a, sizeof(int));
size2=read(file2, &b, sizeof(int));
/* printf("a=[%d], b=[%d]\n", a, b); */
}
}
}
if ((int) size1>0) write(target, &a, sizeof(int));
if ((int) size2>0) write(target, &b, sizeof(int));
/* printf("Finished writing, size1 and size 2 are %d and %d\n", size1, size2); */
while ((int) size1>0)
{
size1=read(file1, buffer, allowedBuffer);
write(target, buffer, size1);
/* printf("%d\n", size1); */
/* printf("Writing 1 %d\n", flno); */
}
while ((int) size2>0)
{
size2=read(file2, buffer, allowedBuffer);
write(target, buffer, size2);
/* printf("Writing 2 %d\n", flno); */
}
/*
while ((size2=read(file2, buffer, allowedBuffer))!=0) write(target, buffer, size2); */
close(file1);
sprintf(newfile, "file%d", flno);
remove(newfile);
/* printf("Removed [%s]\n", newfile); */
close(file2);
sprintf(newfile, "file%d", flno+1);
remove(newfile);
/* printf("Removed [%s]\n", newfile); */
close(target);
}
printf("Finished merging temporary files\n");
}
int quicksort(int N,int *v){
int i,j=0,pivot,*less,*great,*pivots;
int m=0,n=0,o=0;
//pick middle point of array
pivot=N/2;
less=(int *) malloc(N *sizeof(int));
great=(int *) malloc(N *sizeof(int));
pivots=(int *) malloc(N *sizeof(int));
if (N<=1) {
return *v;
}else{
//printf("\n");
//FOR LOOP判斷v[i]與pivot的大小
for (i=0; i<N; ++i) {
if (v[i]< v[pivot]) {
less[m]=v[i];
//printf("pivot:(%d,%d)\n",v[i],v[pivot]);
m++;
}
else if (v[i] > v[pivot]){
great[n]=v[i];
//printf("pivot:(%d,%d)\n",v[i],v[pivot]);
n++;
}
else if (v[i] == v[pivot]){
pivots[o]=v[i];
//printf("pivot:(%d,%d,%d)\n",v[i],v[pivot],pivots[o]);
o++;
}
/*
for(k=0;k<o;++k){
printf("%d,%d,%d\n",i,k,pivots[k]);
}*/
}
}
//對less部分做qsort
quicksort(m, less);
//對great部分做qsort
quicksort(n,great);
//將less、pivot、great組合在一起
//printf("\nless:");
for(i=0;i<m;++i){
//printf("\n%d,%d,",i,m);
//printf("%d,",less[i]);
v[j]=less[i];
j++;
}
//printf("\npivots:");
for(i=0;i<o;++i){
//printf("\n%d,%d,",i,o);
//printf("%d,",pivots[i]);
//printf("%d",j);
v[j]=pivots[i];
j++;
}
//printf("\ngreat:");
for(i=0;i<n;++i){
//printf("\n%d,%d,",i,n);
//printf("%d,",great[i]);
//printf("%d",j);
v[j]=great[i];
j++;
}
//printf("\n");
return *v;
}