0 
#include <math.h> 
#include <stdio.h>
#include <stdlib.h>
#include <mpi.h>
#include <time.h>

#define h 1
#define XY0 0
#define MAX_XY 5
#define N 2 //particles per subdomain
#define BLOCKS 4
#define a  1
#define b  1

float velocityX(float x, float y);
float velocityY(float x, float y);  
int malloc2dfloat(float ***array, int length);

int main (int argc, char **argv)
{
typedef struct {
    float xcoord;
    float ycoord;
    float velx;
    float vely;
} particle;

int points= (int) floor((MAX_XY - XY0)/h) + 1;
int procsize = 2;
int myid, nproc;

MPI_Datatype particletype, oldtypes[1];
MPI_Aint offset[1], extent;

MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &nproc);
MPI_Comm_rank(MPI_COMM_WORLD, &myid);
int startElementX, startElementY, endElementX, endElementY;
particle* sub_pars = (particle*)malloc(sizeof(particle)*N);

offset[0] = 0;
int blockcounts[1];
blockcounts[0] = 4;
oldtypes[0] = MPI_FLOAT;
MPI_Type_struct(1, blockcounts, offset, oldtypes, &particletype);
MPI_Type_commit(&particletype);
particle* particles = (particle*)malloc(sizeof(particle) * N * procsize*procsize);
if (nproc != procsize*procsize){
    printf("Must use np=4 -- split into 4 blocks");
    MPI_Abort(MPI_COMM_WORLD,1);
}

srand(time(NULL)+myid); 
if (myid == 0)
{

    float mins[4]; 
    startElementX = 0;
    startElementY = 0;
    endElementX = (points/procsize)-1;
    endElementY = (points/procsize) -1;
}
else if (myid == 1)
{
    startElementX = 0;
    startElementY = (points/procsize);
    endElementX = (points/procsize) -1;
    endElementY = points - 1;
}
else if (myid == 2)
{
    startElementX = (points/procsize);
    startElementY = 0;
    endElementX = points - 1;
    endElementY = (points/procsize) -1;
}
else
{
    startElementX = (points/procsize);
    startElementY = (points/procsize);
    endElementX = points-1;
    endElementY = points-1;
}


int i;
float localmin;
float mag;
for (i=0; i<N; i++)
{
    sub_pars[i].xcoord = ((startElementX + rand()/(RAND_MAX / (endElementX-startElementX+1)+1)))*h + XY0;
    printf("%f\n", sub_pars[i].xcoord);
    sub_pars[i].ycoord = ((startElementY + rand()/(RAND_MAX / (endElementY-startElementY+1)+1)))*h + XY0;
    sub_pars[i].velx = velocityX(sub_pars[i].xcoord, sub_pars[i].ycoord);
    sub_pars[i].vely = velocityY(sub_pars[i].xcoord, sub_pars[i].ycoord);
    mag = sqrt(sub_pars[i].velx*sub_pars[i].velx + sub_pars[i].vely*sub_pars[i].vely);
    if (i==0 || localmin > mag) localmin = mag; 
}
printf("localmin of %d is %.2f \n", myid, localmin);

MPI_Allgather(&sub_pars, 1, particletype, particles ,1, particletype, MPI_COMM_WORLD);

MPI_Finalize();
if(myid == 0)
{
int k;
for (k=0; k<N*4; k++)
{
    printf("test %.2f \n", particles[i].xcoord); 
}
}

return 0;

}
float velocityX(float x, float y)
{
 float temp = (a+(b*(y*y-x*x))/((x*x+y*y)*(x*x+y*y)));
 return temp;
}

float velocityY(float x, float y)
{
 float temp = (-1*(2*b*x*y)/((x*x+y*y)*(x*x+y*y)));
return temp;
}

すべてのパーティクルに対して同じ値を返すだけですが、各スレッド内で正しく計算されていることはわかっているため、MPI_Allgather に問題があります。どのように見えるべきか説明してもらえますか?

4

1 に答える 1

0

非常によくある間違いを犯しました:&渡す最初の引数の (address-of) 演算子MPI_Allgatherは不要です。は既にポインターであり、 withをsub_pars呼び出すと、実際のデータへのポインターではなく、ポインター(ルーチンのスタック フレーム内のどこかの場所) へのポインターが渡されます。MPI_Allgather&sub_parsmain()

于 2013-10-12T16:29:22.993 に答える