int main(int argc, char *argv[])
{
unsigned char str[154];
unsigned int arr[] = {9,2,5,8,4,2,4,1,6,9,1,8,9,9,6,1,5,7,0,7,7,4,3,7,6,3,9,5,4,2,3,0,4,4,1,5,3,3,7,2,3,3,7,0,9,4,5,2,8,4,6,\
2,1,3,4,1,4,2,6,0,8,5,1,7,3,1,4,4,7,0,5,3,4,4,8,9,1,1,9,8,3,5,1,8,3,4,4,8,3,2,8,1,2,8,7,4,1,8,1,8,0,4,\
8,4,2,4,4,5,4,9,1,8,3,4,9,5,6,3,3,1,4,6,4,1,0,2,0,2,5,1,4,8,5,9,9,6,9,4,0,3,6,5,5,9,5,4,2,2,3,7,8,5,9,7};
long i;
double t1, t2, Itime;
int provided;
/* Allocation */
v1 = (vector *) malloc (VLEN * sizeof (vector));
v2 = (vector *) malloc (VLEN * sizeof (vector));
v3 = (vector *) malloc (VLEN * sizeof (vector));
fin_sum = (mp_limb_t *) malloc ((2*LIMBS+1) * sizeof (mp_limb_t));
result = (mp_limb_t *) malloc ((2*LIMBS+1) * sizeof (mp_limb_t));
q = (mp_limb_t *) malloc (LIMBS * sizeof (mp_limb_t));
MPI_Init_thread (&argc, &argv, MPI_THREAD_MULTIPLE, &provided);
MPI_Comm_rank (MPI_COMM_WORLD, &id);
MPI_Comm_size (MPI_COMM_WORLD, &p);
MPI_Type_contiguous (2*LIMBS+1, MPI_UNSIGNED_LONG_LONG, &mpntype0);
MPI_Type_commit (&mpntype0);
MPI_Type_contiguous (LIMBS, MPI_UNSIGNED_LONG_LONG, &mpntype1);
MPI_Type_commit (&mpntype1);
MPI_Op_create ((MPI_User_function *)addmpn, 1, &mpn_sum);
for (i=0; i<154; ++i) str[i] = (unsigned char)arr[i];
mpn_set_str (q, str, 154, 10);
//if (!id) gmp_printf ("Modulus: %Nd\n", q, LIMBS);
MPI_Barrier (MPI_COMM_WORLD);
/* Setting limits for 2 MPI nodes */
VOffset = BLOCK_LOW(id,p,VLEN);
VChunk = BLOCK_SIZE(id,p,VLEN);
/* Setting limits for NCORES-1 threads */
for (i=0; i<NCORES-1; ++i)
{
VStart[i] = VOffset + BLOCK_LOW(i,NCORES-1,VChunk);
VEnd[i] = VOffset + BLOCK_HIGH(i,NCORES-1,VChunk);
}
for (i=0; i<VLEN; ++i) mpn_random (v1[i], LIMBS);
for (i=0; i<VLEN; ++i) mpn_random (v2[i], LIMBS);
for (i=BLOCK_LOW(id,p,VLEN); i<=BLOCK_HIGH(id,p,VLEN); ++i) mpn_random (v3[i], LIMBS);
MPI_Barrier (MPI_COMM_WORLD);
t1 = MPI_Wtime ();
for (i=0; i<NCORES; ++i)
pthread_create(&threads[i], &attr, VectMul, (void *) i);
for (i=0; i<NCORES; ++i)
pthread_join (threads[i], NULL);
t2 = MPI_Wtime ();
Itime = t2 - t1;
if (!id) printf ("Total time taken: %lf\n",Itime);
if (!id) gmp_printf ("Result: %Nd\n", cnum, LIMBS);
MPI_Op_free(&mpn_sum);
MPI_Request_free (&Rrqst);
MPI_Request_free (&Srqst);
MPI_Finalize ();
return 0;
}
int main (int argc, char *argv[])
{
//define variables
int n;
double elapsed_time;
int p;
int id;
int low_value;
int high_value;
int size;
int proc0_size;
char* marked;
int index;
int prime;
int first;
int count;
int global_count;
int i;
//Initialize MPI
MPI_Init (&argc, &argv);
MPI_Barrier(MPI_COMM_WORLD);
elapsed_time = -MPI_Wtime();
MPI_Comm_rank (MPI_COMM_WORLD, &id);
MPI_Comm_size (MPI_COMM_WORLD, &p);
//Check for proper command line parameters, must include N
if (argc != 2) {
if (!id) printf ("Command line: %s <m>\n", argv[0]);
MPI_Finalize();
exit(1);
}
//Convert parameter string to integer
//N represents the number up to which we need to calculate primes
n = atoi(argv[1]);
//Low and high values for each processor
low_value = 2 + BLOCK_LOW(id,p,n-1);
high_value = 2 + BLOCK_HIGH(id,p,n-1);
size = BLOCK_SIZE(id,p,n-1);
//largest prime is sqrt(n), so first processor has all primes if
//p is less than sqrt(n). We need to check we don't have more processors
//than we need.
proc0_size = (n-1)/p;
if ((2 + proc0_size) < (int) sqrt((double) n)) {
if (!id) printf ("Too many processes\n");
MPI_Finalize();
exit (1);
}
//allocate memory for block, error if unable to
marked = (char *) malloc (size);
if (marked == NULL) {
printf ("Cannot allocate enough memory\n");
MPI_Finalize();
exit (1);
}
/* Begin Sieve of Eratosthenes Algorithm */
//First fill marked[] with zero/false for all items in block
for (i = 0; i < size; i++) marked[i] = 0;
if (!id) index = 0;
//first prime is 2
prime = 2;
do {
if (prime * prime > low_value)
first = prime * prime - low_value;
else {
if (!(low_value % prime)) first = 0;
else first = prime - (low_value % prime);
}
//increment by prime, marking the non-primes with 1, or true
for (i = first; i < size; i += prime) marked[i] = 1;
if (!id) {
while (marked[++index]);
prime = index + 2;
}
MPI_Bcast (&prime, 1, MPI_INT, 0, MPI_COMM_WORLD);
} while (prime * prime <= n);
/* End Sieve of Eratosthenes Algorithm */
/*Begin count of primes */
count = 0;
//for all elements in block, if prime is 1/true, increment count
for (i = 0; i < size; i++)
if (!marked[i]) count++;
//Sum count of primes from each process
MPI_Reduce (&count, &global_count, 1, MPI_INT, MPI_SUM,
0, MPI_COMM_WORLD);
elapsed_time += MPI_Wtime();
//print results on main processor
if (!id) {
printf ("%d primes are less than or equal to %d\n",
global_count, n);
printf ("Total elapsed time: %10.6f\n", elapsed_time);
}
MPI_Finalize();
return 0;
}
int BLOCK_SIZE(int id,int p,int n) {return BLOCK_HIGH(id,p,n)-BLOCK_LOW(id,p,n)+1;}
/* the optimization loop */
int main(int argc, char **argv) {
cmaes_t evo; /* an CMA-ES type struct or "object" */
double *arFunvals, *xfinal, *const*pop;
int i,j;
int numberDipoles;
int id; //Rank
int p; //Number processors
double elapsed_time;//Time from beginning.
double bestValue;
int lambda;
int maxLambda;
int * sendCnts; //For MPI_Alltoallv for arFunVals
int * sdispls; //For MPI_Alltoallv for arFunVals
int * recvCnts; //For MPI_Alltoallv for arFunVals
int * rdispls; //For MPI_Alltoallv for arFunVals
int * sendCntsPop; //For MPI_Alltoallv for pop
int * sdisplsPop; //For MPI_Alltoallv for pop
int * recvCntsPop; //For MPI_Alltoallv for pop
int * rdisplsPop; //For MPI_Alltoallv for pop
int canTerminate;
int canTerminateBuffer;
//Start MPI
MPI_Init(&argc, &argv);
MPI_Barrier(MPI_COMM_WORLD);
elapsed_time = -MPI_Wtime(); //Set initial time.
MPI_Comm_rank(MPI_COMM_WORLD, &id); //Set id
MPI_Comm_size(MPI_COMM_WORLD, &p); //set p
for (i=0;i<32;i++)
{
observations[i]/=1000.0;
}
//Set number of dipoles, either first argument or default value of 2.
numberDipoles=2;
if (argc>=2)
{
numberDipoles=atoi(argv[1]);
}
//Set lambda based on entry, default of 40
maxLambda=40;
if (argc>=3)
{
maxLambda=atoi(argv[2]);
}
if (id==0)
{
printf("Dipoles:%d MaxLambda:%d\n",numberDipoles,maxLambda);
}
//Allocate lambda pieces to each processor, based on the size of maxLambda and the number of processors.
lambda = BLOCK_SIZE(id,p,maxLambda);
printf("Id:%d Lambda:%d\n",id,lambda);
//Setup send and receive buffers for function evaluations and populations that resulted in those evaluation.
sendCnts = malloc(p*sizeof(int));
sdispls = malloc(p*sizeof(int));
recvCnts = malloc(p*sizeof(int));
rdispls = malloc(p*sizeof(int));
sendCntsPop = malloc(p*sizeof(int));
sdisplsPop = malloc(p*sizeof(int));
recvCntsPop = malloc(p*sizeof(int));
rdisplsPop = malloc(p*sizeof(int));
for (i=0;i<p;i++)
{
sendCnts[i]=lambda;//Same for all others
sdispls[i] = BLOCK_LOW(id,p,maxLambda);//Same for all others
recvCnts[i] = BLOCK_SIZE(i,p,maxLambda);//Depends on which we receive from.
rdispls[i] = BLOCK_LOW(i,p,maxLambda);
sendCntsPop[i]=lambda*((numberDipoles*6+2));//Same for all others
sdisplsPop[i] = BLOCK_LOW(id,p,maxLambda)*(numberDipoles*6+2);//Same for all others
recvCntsPop[i] = BLOCK_SIZE(i,p,maxLambda)*(numberDipoles*6+2);//Depends on which we receive from.
rdisplsPop[i] = BLOCK_LOW(i,p,maxLambda)*(numberDipoles*6+2);
}
for (i=0;i<p;i++)
{
printf("Id: %d recvCnts[%d]=%d\n",id,i,recvCnts[i]);
printf("Id: %d rdispls[%d]=%d\n",id,i,rdispls[i]);
printf("Id: %d recvCntsPop[%d]=%d\n",id,i,recvCntsPop[i]);
printf("Id: %d rdisplsPop[%d]=%d\n",id,i,rdisplsPop[i]);
}
/* Initialize everything into the struct evo, 0 means default */
//arFunvals = cmaes_init(&evo, 0, NULL, NULL, 0, 0, "initials.par");
// printf("0\n");
//The maxLambda parameter has been added so all of them will have enough space to store the results
arFunvals = reinit(&evo, maxLambda, numberDipoles);
//outputCMAES_t(evo,1);
// printf("1\n");
resetSignals(&evo, numberDipoles); /* write header and initial values */
//Reset the seed value based on processor (so they don't all come out the same!
evo.sp.seed=evo.sp.seed*(id+1)/p;
printf("proc: %d seed: %d\n",id,evo.sp.seed);
//outputCMAES_t(evo,0);
// printf("2\n");
// printf("%s\n", cmaes_SayHello(&evo));
// i=40;
// for (i=32;i<40;i*=2)
// {
// arFunvals = reinit(&evo, i);
//outputCMAES_t(evo);
evo.sp.lambda=lambda;
canTerminate = (0==1);
/* Iterate until stop criterion holds */
while(!canTerminate)
{
/* generate lambda new search points, sample population */
pop = cmaes_SamplePopulation(&evo); /* do not change content of pop */
/* Here you may resample each solution point pop[i] until it
becomes feasible, e.g. for box constraints (variable
boundaries). function is_feasible(...) needs to be
user-defined.
Assumptions: the feasible domain is convex, the optimum is
not on (or very close to) the domain boundary, initialX is
feasible and initialStandardDeviations are sufficiently small
to prevent quasi-infinite looping.
*/
/*for (i = 0; i < lambda; ++i)
{
cmaes_ReSampleSingle(&evo, i);
}*/
for (i = 0; i < lambda; ++i)
{
while (!is_feasible(evo.rgrgx[i],(int) cmaes_Get(&evo, "dim")))
{
cmaes_ReSampleSingle(&evo, i);
}
}
for (i=0;i<lambda;i++)
{
for(j=0;j<(6*numberDipoles)+2;j++)
{
evo.rgrgx[BLOCK_LOW(id,p,maxLambda)+i][j]=evo.rgrgx[i][j];
}
}
/* evaluate the new search points using fitfun from above */
for (i = BLOCK_LOW(id,p,maxLambda); i <= BLOCK_HIGH(id,p,maxLambda); ++i) {
arFunvals[i] = fitfun(evo.rgrgx[i], (int) cmaes_Get(&evo, "dim"));
//printf("ID:%d, arFunvals[%d]=%lf\n",id,i,arFunvals[i]);
}
//Now communicate the arFunvals around
MPI_Alltoallv(arFunvals,sendCnts,sdispls,MPI_DOUBLE,arFunvals,recvCnts,rdispls,MPI_DOUBLE,MPI_COMM_WORLD);
//Now communicate the populations being looked at around
MPI_Alltoallv(&evo.rgrgx[0][0],sendCntsPop,sdisplsPop,MPI_DOUBLE,&evo.rgrgx[0][0],recvCntsPop,rdisplsPop,MPI_DOUBLE,MPI_COMM_WORLD);
/* update the search distribution used for cmaes_SampleDistribution() */
cmaes_UpdateDistribution(&evo, arFunvals);
//Test for any that can terminate.
canTerminate = cmaes_TestForTermination(&evo);
if (canTerminate)
{
printf("id:%d can terminate for reason:%s\n",id,cmaes_TestForTermination(&evo));
}
MPI_Allreduce(&canTerminate,&canTerminateBuffer,1,MPI_INT,MPI_MAX,MPI_COMM_WORLD);//Get the max, if any are >0, then someone has terminated.
canTerminate = canTerminateBuffer;//Reset so the loop will exit.
/* read instructions for printing output or changing termination conditions */
// cmaes_ReadSignals(&evo, "signals.par");
// fflush(stdout); /* useful in MinGW */
}
// printf("Stop:\n%s\n", cmaes_TestForTermination(&evo)); /* print termination reason */
// cmaes_WriteToFile(&evo, "all", "allcmaes.dat"); /* write final results */
elapsed_time += MPI_Wtime();
/* get best estimator for the optimum, xmean */
xfinal = cmaes_GetNew(&evo, "xmean"); /* "xbestever" might be used as well */
bestValue = fitfun(xfinal, (int) cmaes_Get(&evo, "dim"));
printf("Proccesor:%d has last mean of:%lf elapsedTime:%lf\n",id,bestValue,elapsed_time);
for (i=0;i<6*numberDipoles;i++)
{
printf("(%d:%d:%lf)\n",id,i,xfinal[i]);
}
// cmaes_exit(&evo); /* release memory */
/* do something with final solution and finally release memory */
free(xfinal);
free(sendCnts);
free(sdispls);
free(recvCnts);
free(rdispls);
free(sendCntsPop);
free(sdisplsPop);
free(recvCntsPop);
free(rdisplsPop);
MPI_Finalize();
//}
return 0;
}
//wh
int main (int argc, char *argv[])
{
int count; ///局部素数个数
double elapsed_time; ///运行时间
int first; ///每组中第一个是prime倍数的数
int global_count; ///全局素数个数
int high_value; ///每组的最后一个元素值
int i; ///用于循环
int id; ///程序ID
int index; ///子程序0数组标号
int low_value; ///每组的第一个元素值
char * marked; ///指向局部数组
int n; ///求素数的范围
int p; ///子程序数目
int proc0_size; ///子程序0的数组大小
int prime; ///下一个要删除其倍数的素数
int size; ///局部素数个数
MPI_Init (&argc, &argv); ///MPI初始化
MPI_Barrier(MPI_COMM_WORLD);///所有程序同时开始
elapsed_time = -MPI_Wtime();///获得当前时间的负值
MPI_Comm_rank (MPI_COMM_WORLD, &id);///得到该通信子程序的ID号
MPI_Comm_size (MPI_COMM_WORLD, &p);///得到通信关联组大小
///判断传入参数个数是否正确，不正确则终止程序
if (argc != 2)
{
if (!id) printf ("Command line: %s <m>\n", argv[0]);
MPI_Finalize();
exit (1);
}
n = atoi(argv[1]);///从传入参数得到的大小
low_value = 3 + 2*(BLOCK_LOW(id,p,(n-1)/2));///算出该子程序持有的数组的最小值
high_value = 3 + 2*(BLOCK_HIGH(id,p,(n-1)/2));///算出该子程序持有的数组的最大值
size = BLOCK_SIZE(id,p,(n-1)/2);///算出该子程序持有的数组元素个数
proc0_size = ((n-1)/2)/p;///算出进程0的数组元素个数
///如果进程0所控制的最大数小于n的平方根，异常退出
if ((3 + 2*proc0_size) < (int) sqrt((double) n))
{
if (!id) printf ("Too many processes\n");
MPI_Finalize();
exit (1);
}
marked = (char *) malloc (size);///分配数组空间
///如果分配空间失败，异常退出
if (marked == NULL)
{
printf ("Cannot allocate enough memory\n");
MPI_Finalize();
exit (1);
}
for (i = 0; i < size; i++) marked[i] = 0;///初始化数组，0表示未标记
if (!id) index = 0;///初始化index，只有子程序0使用该变量
prime = 3;///第一个素数为3(偶数已经去掉)
/*************************核心算法*****************************/
do
{
if (prime * prime > low_value)
first = (prime * prime - low_value)/2;
else
{
if (!(low_value % prime)) first = 0;
else
first = (prime - low_value % prime+1)/2+((prime-1)/2)*((prime - low_value % prime)%2);
}
for (i = first; i < size; i += prime)
{
marked[i] = 1;
}
if (!id)
{
while (marked[++index]);
prime = 2*index + 3;
}
MPI_Bcast (&prime, 1, MPI_INT, 0, MPI_COMM_WORLD);
}
while (prime * prime <= n);
/*************************************************************/
///计算局部数组个数
count = 0;
for (i = 0; i < size; i++)
if (!marked[i])
{
count++;
}
///将所有局部素数数目求和发送给子程序0
MPI_Reduce (&count, &global_count, 1, MPI_INT, MPI_SUM,
0, MPI_COMM_WORLD);
elapsed_time += MPI_Wtime();///计算运行时间
///子程序0负责输出信息
if (!id)
{
printf ("%d primes are less than or equal to %d\n",
global_count+1, n);
printf ("Total elapsed time: %10.6f\n", elapsed_time);
}
MPI_Finalize ();///MPI终止
return 0;
}
grid *grid_create(double startx, double endx, int nx,
double starty, double endy, int ny,
double startz, double endz, int nz)
{
MPI_Comm cart_comm;
int periodics[3];
int i, j, k;
int ind;
int np, rank;
grid *grd = (grid*) malloc(sizeof(grid));
MPI_Comm_size(MPI_COMM_WORLD, &np);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
grd->id = 0;
if (nz > 0)
grd->nd = 3;
else
grd->nd = 2;
grd->num_global[0] = nx;
grd->num_global[1] = ny;
grd->num_global[2] = nz;
grid_decomp(grd, np);
periodics[0] = periodics[1] = periodics[2] = 0;
MPI_Cart_create(MPI_COMM_WORLD, grd->nd, grd->num_procs, periodics, 1, &cart_comm);
MPI_Comm_rank(cart_comm, &rank);
MPI_Cart_coords(cart_comm, rank, grd->nd, grd->cart_coord);
grd->nx = BLOCK_SIZE(grd->cart_coord[0], grd->num_procs[0], nx);
grd->ny = BLOCK_SIZE(grd->cart_coord[1], grd->num_procs[1], ny);
grd->num_local[0] = grd->nx;
grd->num_local[1] = grd->ny;
grd->num_pts = grd->nx * grd->ny;
grd->is[0] = BLOCK_LOW(grd->cart_coord[0], grd->num_procs[0], nx) + 1;
grd->ie[0] = BLOCK_HIGH(grd->cart_coord[0], grd->num_procs[0], nx) + 1;
grd->is[1] = BLOCK_LOW(grd->cart_coord[1], grd->num_procs[1], ny) + 1;
grd->ie[1] = BLOCK_HIGH(grd->cart_coord[1], grd->num_procs[1], ny) + 1;
if (grd->nd == 3) {
grd->nz = BLOCK_SIZE(grd->cart_coord[2], grd->num_procs[2], nz);
grd->num_local[2] = grd->nz;
grd->num_pts = grd->num_pts * grd->nz;
grd->is[2] = BLOCK_LOW(grd->cart_coord[2], grd->num_procs[2], nz) + 1;
grd->ie[2] = BLOCK_HIGH(grd->cart_coord[2], grd->num_procs[2], nz) + 1;
}
grd->comm = MPI_COMM_WORLD;
grd->xyz = (double*) malloc(grd->num_pts*3*sizeof(double));
grd->x = &grd->xyz[0];
grd->y = &grd->xyz[grd->num_pts];
grd->z = &grd->xyz[grd->num_pts*2];
grd->hx = (endx - startx) / (grd->num_global[0] - 1);
grd->hy = (endy - starty) / (grd->num_global[1] - 1);
grd->hz = (endz - startz) / (grd->num_global[2] - 1);
startx = startx + (grd->is[0]-1)*grd->hx;
starty = starty + (grd->is[1]-1)*grd->hy;
startz = startz + (grd->is[2]-1)*grd->hz;
if (grd->nd == 3) {
for (k = 0; k < grd->nz; k++) {
for (j = 0; j < grd->ny; j++) {
for (i = 0; i < grd->nx; i++) {
ind = k*grd->nx*grd->ny + j*grd->nx + i;
grd->x[ind] = startx + i*grd->hx;
grd->y[ind] = starty + j*grd->hy;
grd->z[ind] = startz + k*grd->hz;
}
}
}
} else {
for (j = 0; j < grd->ny; j++) {
for (i = 0; i < grd->nx; i++) {
ind = j*grd->nx + i;
grd->x[ind] = startx + i*grd->hx;
grd->y[ind] = starty + j*grd->hy;
}
}
}
for (i = 0; i < grd->nd; i++) {
grd->periodic[i] = 0;
}
return grd;
}
/* Function to be executed by the workers */
void slave(Params * p)
{
srand(2);
/* Timer */
double start, end;
int i;
int * array;
array = (int *) malloc(sizeof(int) * p->array_size);
MPI_Status status;
int count = 0;
int done = 0;
int temp = 0;
int low = BLOCK_LOW(p->rank-1,p->size,p->max_num+1);
int high = BLOCK_HIGH(p->rank-1,p->size,p->max_num+1);
start = MPI_Wtime();
// Repeat until done
while(!done)
{
// Master work
if(p->rank == 0)
{
// Increase count until array_size
if(count++ < p->array_size)
{
// Add random number to list, send it up the pipeline to rank 1
temp = rand() % p->max_num;
MPI_Send(&temp, 1, MPI_INT, 1, NUM_TAG, MPI_COMM_WORLD);
}
else
{
// Signal end of the list
MPI_Send(&temp, 1, MPI_INT, 1, TERM_TAG, MPI_COMM_WORLD);
done = 1;
}
}
// Slave work
else
{
// receive a number from the previous stage
MPI_Recv(&temp, 1, MPI_INT, p->rank-1, MPI_ANY_TAG, MPI_COMM_WORLD, &status);
// check for termination
if(status.MPI_TAG == TERM_TAG)
{
// make sure I'm sending to a valid rank and foward termination
done = 1;
if(p->rank != p->size)
{
MPI_Send(&temp, 1, MPI_INT, p->rank+1, TERM_TAG, MPI_COMM_WORLD);
}
}
// got number
else if(status.MPI_TAG == NUM_TAG)
{
// check to see if I keep it
if(temp >= low && temp <= high)
array[count++] = temp;
else
// pass it on
MPI_Send(&temp, 1, MPI_INT, p->rank+1, NUM_TAG,MPI_COMM_WORLD);
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
int * sizes = (int *) calloc(p->size+1,sizeof(int));
MPI_Gather(&count, 1, MPI_INT, sizes, 1, MPI_INT, 0 , MPI_COMM_WORLD);
int * disp = (int *) calloc(p->size+1,sizeof(int));
if(p->rank == 0)
{
sizes[0] = 0;
disp[0] = 0;
for(i = 1; i < p->size+1; ++i)
{
disp[i] = disp[i-1] + sizes[i-1];
}
}
// sort the array, except for the master
if(p->rank != 0)
qsort(array,count,sizeof(int),compare);
if(p->rank == 0) count = 0;
// gather results
MPI_Gatherv(array, count, MPI_INT, &array[0], sizes, disp, MPI_INT,0,MPI_COMM_WORLD);
// stop the clock, print results
end = MPI_Wtime();
if(p->rank ==0) fprintf(stderr,"[%d] Elapsed time: %f\n",p->rank,end-start);
return;
}
bool processOptions (int argc, char *argv[], INFO *info) {
int c = 0;
char *base_fn = NULL;
char *co_fn = NULL;
unsigned int num_clusters = 0;
unsigned int seed = UINT_MAX;
unsigned int maxiter = 0;
unsigned int snapshot = UINT_MAX;
bool verbose = false;
bool debug = false;
bool textio = false;
bool rounding = false;
bool no_output = false;
/* Usage information if no arguments */
if (argc == 1) {
usage (argv[0]);
}
while (1) {
int option_index = 0;
static struct option long_options[] = {
{"base", 1, 0, 0},
{"cooccur", 1, 0, 0},
{"clusters", 1, 0, 0},
{"seed", 1, 0, 0},
{"maxiter", 1, 0, 0},
{"snapshot", 1, 0, 0},
{"openmp", 1, 0, 0},
{"verbose", 0, 0, 0},
{"debug", 0, 0, 0},
{"text", 0, 0, 0},
{"rounding", 0, 0, 0},
{"nooutput", 0, 0, 0},
{0, 0, 0, 0}
};
c = getopt_long (argc, argv, "", long_options, &option_index);
if (c == -1) {
break;
}
switch (c) {
case 0:
if (strcmp (long_options[option_index].name, "cooccur") == 0) {
co_fn = wmalloc (strlen (optarg) + 1);
co_fn = strcpy (co_fn, optarg);
}
else if (strcmp (long_options[option_index].name, "clusters") == 0) {
num_clusters = atoi (optarg);
}
else if (strcmp (long_options[option_index].name, "seed") == 0) {
seed = atoi (optarg);
}
else if (strcmp (long_options[option_index].name, "base") == 0) {
base_fn = wmalloc (strlen (optarg) + 1);
base_fn = strcpy (base_fn, optarg);
}
else if (strcmp (long_options[option_index].name, "maxiter") == 0) {
maxiter = atoi (optarg);
}
else if (strcmp (long_options[option_index].name, "snapshot") == 0) {
snapshot = atoi (optarg);
}
else if (strcmp (long_options[option_index].name, "openmp") == 0) {
#if HAVE_OPENMP
/* Check the previously set value, which is the maximum for the system */
if (atoi (optarg) > info -> threads) {
fprintf (stderr, "==\tError: The number of threads requested exceeds the number available in the system (%u).\n", info -> threads);
exit (-1);
}
info -> threads = atoi (optarg);
omp_set_num_threads (info -> threads);
#else
fprintf (stderr, "==\tError: OpenMP is not enabled; --openmp meaningless.\n");
exit (-1);
#endif
}
else if (strcmp (long_options[option_index].name, "verbose") == 0) {
verbose = true;
}
else if (strcmp (long_options[option_index].name, "debug") == 0) {
debug = true;
}
else if (strcmp (long_options[option_index].name, "text") == 0) {
textio = true;
}
else if (strcmp (long_options[option_index].name, "rounding") == 0) {
rounding = true;
}
else if (strcmp (long_options[option_index].name, "nooutput") == 0) {
no_output = true;
}
break;
default:
printf ("?? getopt returned character code 0%o ??\n", c);
exit (EXIT_FAILURE);
}
}
info -> base_fn = base_fn;
info -> co_fn = co_fn;
info -> num_clusters = num_clusters;
info -> seed = seed;
info -> maxiter = maxiter;
info -> snapshot = snapshot;
info -> verbose = verbose;
info -> debug = debug;
info -> textio = textio;
info -> rounding = rounding;
info -> no_output = no_output;
/* Set the range of clusters this process will handle */
info -> block_start = BLOCK_LOW (info -> world_id, info -> world_size, info -> num_clusters);
info -> block_end = BLOCK_HIGH (info -> world_id, info -> world_size, info -> num_clusters);
info -> block_size = BLOCK_SIZE (info -> world_id, info -> world_size, info -> num_clusters);
return true;
}
int main(int argc, char * argv[])
{
/* Constant Declarations */
//long const SET_SIZE = 7920;
/* Variable Declarations */
int count = 0; // local count
double elapsed_time = 0.00; // time elapsed
int first; // index of first multiple
int global_count = 1; // global count
int high_value; // highest value on processor
char hostname[MPI_MAX_PROCESSOR_NAME]; // host process is running on
int i; // counter variable
int id; // process id number
int index;
int init_status; // initialization error status flag
bool initialized = false; // mpi initialized flag
int len; // hostname length
int low_value; // lowest value on the processor
char* marked; // portion of 2 to n that is marked
int n; // number of elements to sieve
int n_sqrt; // square root of n
int p; // number of processes
int prime;
int proc0_size; // size of process 0's subarray
int size; // elements in marked
int* sqrt_primes; // primes up to the square root
int sqrt_primes_index; // index in the square root primes array
char* sqrt_primes_marked; // numbers up to sqrt marked prime or not
int sqrt_primes_size; // size of square root primes array
/* Function Declarations */
//int is_prime( int );
/* Initialization */
MPI_Initialized( &initialized ); // set initialized flag
if( !initialized ) // if MPI is not initialized
init_status = MPI_Init( &argc, &argv ); // Initialize MPI
else
init_status = MPI_SUCCESS; // otherwise set init_status to success
if( init_status != MPI_SUCCESS ) { // if not successfully initialized
printf ("Error starting MPI program. Terminating.\n"); // print error message
fflush(stdout);
MPI_Abort(MPI_COMM_WORLD, init_status); // abort
}
MPI_Get_processor_name( hostname, &len ); // set hostname
MPI_Comm_rank( MPI_COMM_WORLD, &id ); // set process rank
MPI_Comm_size( MPI_COMM_WORLD, &p ); // set size of comm group
//printf("Process rank %d started on %s.\n", id, hostname); // print start message
//fflush(stdout);
//MPI_Barrier(MPI_COMM_WORLD );
/* Start Timer */
MPI_Barrier( MPI_COMM_WORLD ); // synchronize
elapsed_time = - MPI_Wtime(); // start time
/* Check that a set size was passed into the program */
if(argc != 2) {
if(id==0) {
printf("Command line: %s <m>\n", argv[0]);
fflush(stdout);
}
MPI_Finalize();
exit(1);
}
n = atoi(argv[1]);
n_sqrt = ceil(sqrt((double)n));
//if(id==0)
// printf("square root: %i\n", n_sqrt);
// debug
//if(id==0) {
//printf("n sqrt: %i\n", n_sqrt);
//fflush(stdout);
//}
sqrt_primes_marked = (char *) malloc(n_sqrt + 1);
sqrt_primes_marked[0] = 1;
sqrt_primes_marked[1] =1;
for(i = 2; i <= n_sqrt; ++i) {
sqrt_primes_marked[i] = 0;
}
prime = 2;
sqrt_primes_size = n_sqrt;
//printf("sqrt primes size: %i\n", sqrt_primes_size);
do {
for(i = prime * prime; i < n_sqrt; i+=prime) {
sqrt_primes_marked[i] = 1;
//sqrt_primes_size--;
}
while(sqrt_primes_marked[++prime]);
} while (prime * prime <= n_sqrt);
//printf("sqrt primes size: %i\n", sqrt_primes_size);
sqrt_primes = (int *) malloc(sqrt_primes_size);
sqrt_primes_index = 0;
//sqrt_primes_size = 0;
for(i = 3; i <= n_sqrt; ++i) {
if(!sqrt_primes_marked[i]) {
sqrt_primes[sqrt_primes_index] = i;
// printf("%i, ", sqrt_primes[sqrt_primes_index]);
sqrt_primes_index++;
}
}
sqrt_primes_size = sqrt_primes_index;
//printf("sqrt primes size: %i\n", sqrt_primes_size);
//fflush(stdout);
/* Set process's array share and first and last elements */
low_value = 2 + BLOCK_LOW(id,p,n-1);
high_value = 2 + BLOCK_HIGH(id,p,n-1);
size = BLOCK_SIZE(id,p,n-1);
//printf("Process %i block low: %i\n", id, low_value);
//fflush(stdout);
//printf("Process %i block high: %i\n", id, high_value);
//fflush(stdout);
//printf("Block size: %i\n", size);
//fflush(stdout);
if(low_value % 2 == 0) {
if(high_value % 2 == 0) {
size = (int)floor((double)size / 2.0);
high_value--;
}
else {
size = size / 2;
}
low_value++;
}
else {
if(high_value % 2 == 0) {
size = size / 2;
high_value--;
}
else {
size = (int)ceil((double)size / 2.0);
}
}
//printf("Process %i block low: %i\n", id, low_value);
//fflush(stdout);
//printf("Process %i block high: %i\n", id, high_value);
//fflush(stdout);
//printf("Block size: %i\n", size);
//fflush(stdout);
//proc0_size = (n-1)/p;
/* if process 0 doesn't have all the primes for sieving, then bail*/
/*if((2+proc0_size) < (int)sqrt((double)n)) {
if(id==0) {
printf("Too many processes\n");
fflush(stdout);
}
MPI_Finalize();
exit(1);
}
*/
/* Allocate share of array */
marked = (char *) malloc(size);
if(marked == NULL) {
printf("Cannot allocate enough memory\n");
fflush(stdout);
MPI_Finalize();
exit(1);
}
/* Run Sieve */
//printf("made it to sieve\n");
//fflush(stdout);
for(i = 0; i < size; i++)
marked[i] = 0;
if(id==0)
first = 0;
sqrt_primes_index = 0;
prime = sqrt_primes[sqrt_primes_index];
//printf("first prime: %i\n", prime);
//fflush(stdout);
//for(i = 0; i < sqrt_primes_size; i++) {
// printf("%i,", sqrt_primes[i]);
// fflush(stdout);
//}
do {
if(prime >= low_value)
first = ((prime - low_value) / 2) + prime;
else if(prime * prime > low_value) {
first = (prime * prime - low_value) / 2;
}
else {
if(low_value % prime == 0)
first = 0;
else {
first = 1;
while ((low_value + (2 * first)) % prime != 0)
++first;
}
}
//printf("first: %i\n", first);
//fflush(stdout);
for(i = first; i < size; i += (prime))
marked[i] = 1;
//printf("made it to prime assignment\n");
prime = sqrt_primes[++sqrt_primes_index];
//printf("prime: %i\n", prime);
//fflush(stdout);
} while(prime * prime <= n && sqrt_primes_index < sqrt_primes_size);
count = 0;
for(i = 0; i < size; i++) {
if(!marked[i])
count++;
}
//printf("size: %i\ncount: %i\n", size, count);
// for( i=id; i<SET_SIZE; i+=p ) // interleaved allocation
// count += is_prime( i ); // check if prime w/ sieve of eratosthenes
/* Reduce Sum */
MPI_Reduce( &count, &global_count, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD ); // reduce the primes count, root: proces 0
/* Stop Timer */
elapsed_time += MPI_Wtime(); // end time
//printf("Process %i found %i primes.\n", id, count);
//fflush(stdout);
//printf("Process %d is done in %d, running on %s.\n", id, elapsed_time, hostname); // print process done message
if( id == 0 ) { // rank 0 prints global count
printf("There are %d primes in the first %i integers.\nExecution took %10.6f.\n",
global_count, n, elapsed_time);
fflush(stdout);
// printf("Debug:\n");
// fflush(stdout);
// printf("sqrt primes size: %i\n", sqrt_primes_size);
// fflush(stdout);
for(i = 0; i < sqrt_primes_size; i++) {
if(!sqrt_primes[i]){
printf("%i,", sqrt_primes[i]);
fflush(stdout);
}
}
}
MPI_Barrier(MPI_COMM_WORLD);
// printf("rank: %i\nlow value: %i\nhigh value: %i\ncount: %i\n", id, low_value, high_value, count);
//fflush(stdout);
MPI_Finalize(); // finalize
return 0;
}
int main (int argc, char *argv[])
{
int count; /* local prime count */
double elapsed_time; /* execution time */
int first; /* index of the first sieve */
int global_count; /* global count of prime numbers */
int high_value; /* highest value assigned to this process */
int i; /* loop counter */
int id; /* this process id */
int index; /* index of the current sieve */
int low_value; /* lowest value assigned to this process */
int *marked; /* array elements to be marked */
int n; /* value of the largest number */
int p; /* number of processes */
int proc0_size; /* number of elements assigned to process zero */
/* this is to find if process zero has all primes */
int prime; /* current prime or sieve */
int size; /* elements in marked array */
int seed_size;
char cpu_name[MPI_MAX_PROCESSOR_NAME];
int namelen;
MPI_Init (&argc, &argv);
/* start timer */
MPI_Barrier(MPI_COMM_WORLD);
elapsed_time = -MPI_Wtime();
MPI_Comm_rank (MPI_COMM_WORLD, &id);
MPI_Comm_size (MPI_COMM_WORLD, &p);
MPI_Get_processor_name(cpu_name, &namelen);
if (argc != 2)
{
if (!id) printf ("Command line: %s <m>\n", argv[0]);
MPI_Finalize();
exit (1);
}
n = atoi(argv[1]);
/* find how many elements are assigned to this process */
low_value = BLOCK_LOW(id,p,n);
high_value = BLOCK_HIGH(id,p,n);
size = BLOCK_SIZE(id,p,n);
seed_size = SEED_SIZE(n);
proc0_size = (n-1)/(2*p);
// - main loop works only for prime * prime <= n
// - this means it only runs as long as
// prime <= sqrt(n)
// - In this setup, the program will exit if proc0 doesn't hold all starting primes (i.e.
// it will exit if a starting prime will need to be chosen from another process - which this
// program is not prepared to do.
/*
if ((OFFSET + proc0_size) < (int) sqrt((double) n))
{
if (!id) printf ("Too many processes\n");
MPI_Finalize();
exit (1);
}*/
// There is too many processes when we cannot split
// up what is left of the numbers after taking out
// the SEED section (which encloses sqrt(n))
if (BLOCK_SXN_SIZE(n) < p)
{
if (!id) printf ("Too many processes\n");
MPI_Finalize();
exit (1);
}
marked = (int *) malloc ((seed_size + size) * sizeof(int));
if (marked == NULL)
{
printf ("Cannot allocate enough memory\n");
MPI_Finalize();
exit (1);
}
for (i = 0; i < (seed_size + size); i++) marked[i] = 0;
index = 0;
prime = OFFSET;
if(!id) printf("[%d-%s] SEED: low[ 0]= 3, high[%6d]=%6d (%d)\n", id, cpu_name, (SEED_SIZE(n)-1), SEED_HIGH(n), seed_size);
MPI_Barrier(MPI_COMM_WORLD);
printf("[%d-%s] ARRAY: low[%6d]=%6d, high[%6d]=%6d (%d)\n", id, cpu_name, seed_size, low_value, seed_size + size-1, high_value, size);
do {
//printf("[%d] *prime: %d\n", id, prime);
// SEED marking - mark the multiples of the seed - within
// the seed block. Start marking from the value of
// the prime * prime, which is at the position
// index + prime
for (i = index + prime; i < seed_size; i += prime)
{
//printf("[%d] marked: %d\n", id, (OFFSET + (i * 2)));
marked[i] = 1;
}
// Now, we need to continue to the block section, and keep
// marking. But we first need to find out where to start.
// For each process, we need to find the first element to mark.
// The first number that we would have to mark is prime * prime
// If the first number to mark (prime * prime) is at least
// above the low bound of this process ...
if (prime * prime > low_value)
{
// ... Then the first index is that number (prime * prime)
// - the low_value. E.g:
// prime = 7
// low_value = 41
// first = [(7*7) - 41]/2 = (49 - 41)/2 = 8/2 = 4
// marked[4] will be marked first
first = (prime * prime - low_value)/2;
//printf("[%d] first(a): %d(%d)\n", id, first, low_value + (2 * first));
}
else
{
// This section is for "run-on" arrays, e.g.
// prime=3
// p0 [ 3| 5] 3*3 > 3
// p1 [ 7| 9] 3*3 !> 7 "run-on", need mark 9
// p2 [11|13] 3*3 !> 11 "run-on"
if (!(low_value % prime))
{
// If it is, then the first element of the
// array will be the starting place
first = 0;
//printf("[%d] first(b): %d(%d)\n", id, first, low_value + (2 * first));
}
else
{
// (3 - (11 % 3))/2
// (3 - (2))/2s
int tmp = prime - (low_value % prime);
first = tmp % 2 == 0 ? tmp/2 : (tmp + prime)/2;
//printf("[%d] first(c): %d(%d)\n", id, first, low_value + (2 * first));
}
}
// 0 sqrt(n)
// [SEED ][ BLOCK_SXN ]
// SEED_SIZE -------> ---> first
// ----------------------> first + SEED_SIZE
// Now, mark all multiples of the prime.
// 'first' is a multiple of the prime, so += prime
// is also a multiple.
for (i = first; i < size; i += prime)
{
//printf("[%d] marked: %d\n", id, (low_value + (i * 2)));
marked[i + seed_size] = 1;
}
// Increase 'index' which pointed to the last
// smallest prime, until it reaches the next
// smallest prime.
while (marked[++index]);
// Remember, each 'index' in process 0
// represents the number = index + 2
prime = (2 * index) + OFFSET;
} while (prime * prime <= n);
count = 0;
// Total up the amount of items that are not marked
// i.e. the number of local primes
// REMOVE LATER!!!!
//sleep(1);
MPI_Barrier(MPI_COMM_WORLD);
//
//printf("[%d] ", id);
if (!id)
{
for (i = 0; i < seed_size; i++)
{
if (!marked[i])
{
//printf("%d,", ((2*i)+low_value));
count++;
}
}
}
for (i = 0; i < size; i++)
{
if (!marked[i + seed_size])
{
//printf("%d,", ((2*i)+low_value));
count++;
}
}
//printf("\n");
// process 0 will receieve the sum of the number
// of local primes
MPI_Reduce (&count, &global_count, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_WORLD);
elapsed_time += MPI_Wtime();
if (!id) {
global_count++; // To account for 2
printf ("%d primes are less than or equal to %d\n",
global_count, n);
printf ("Total elapsed time: %10.6f\n", elapsed_time);
}
MPI_Finalize ();
return 0;
}