I have a pointer of type double which points to 12 elements of type double and i can print them like this.
void printArray(double *array, int size) {
int i;
for (i = 0; i < size; i++) {
printf("%f\n", array[i]);
}
}
and i print the values in my code by printArray(X,12) which displays :
-1.000000
-0.909091
1.000000
-1.000000
-0.909091
1.000000
-1.000000
-0.909091
1.000000
-1.000000
-0.909091
1.000000
I am trying to scatter these elements into 4 elements per processes for a total of 3 processes so for example process 0 as values :
-1.000000
-0.909091
1.000000
-1.000000
process 2 has :
-0.909091
1.000000
-1.000000
-0.909091
etc and then do some processing on these submatrices i-e multiply them by 2 and then return the result back to root process. To this effect i tried to study the following Scatter a Matrix - MPI and How are MPI_Scatter and MPI_Gather used from C? . I tried to implement it as follows :
int count = 4; //number of rows for submatrix
int blocklength = 1; //number of columns for submatrix
int stride = 4; //width of the submatrix
double *Adata = X;
double *localdata;
int size, rank;
int strip_size, A_row, A_col;
double *strip_A = NULL; //used to store all the
//results from gather operation size 12
double *stripdata; // used to store submatrices
// or vectors of size 4 x 1
MPI_Datatype strip;
int i, j;
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
A_row = 4;
A_col = 1;
strip_size = 4; // total no. of elements should be divided based on no of processes.
//For this example it is 4 (number of elements(12) / processes(3).
if (rank == 0) {
printf("Processor %d has data: \n", rank);
printArray(Adata, 12);
}
/* Broadcasting the strip size*/
MPI_Bcast(&strip_size, 1, MPI_INT, 0, MPI_COMM_WORLD);
/* defining a datatype for sub-matrix */
/*MPI_Type_vector(int count = 3,
int blocklength = 1,
int stride = 3,
MPI_Datatype oldtype = MPI_DOUBLE ,
MPI_Datatype *newtype)*/
MPI_Type_vector(count, blocklength, stride, MPI_DOUBLE, &strip);
MPI_Type_commit(&strip);
//4 rows 1 column .
//Assign sufficient memory for vectors.
stripdata = (double *) malloc(sizeof(double) * A_row * A_col);
// assign suffiecient memory to hold end data
strip_A = (double *) malloc(size * sizeof(double) * 12);
//localdata at each node 4 rows 1 column
localdata = (double *) malloc(sizeof(double) * 4 *1);
/*int MPI_Scatter(const void *sendbuf = Adata, int sendcount = 1, MPI_Datatype sendtype = strip,
void *recvbuf = strip_A, int recvcount = 1, MPI_Datatype recvtype = strip, int root = 0,
MPI_Comm comm = MPI_COMM_WORLD)
* */
MPI_Scatter(Adata, 1, strip, &stripdata, 1, strip, 0, MPI_COMM_WORLD);
MPI_Type_free(&strip);
free(strip_A);
free(stripdata);
if (rank == 0) {
free(Adata);
}
However i am getting segmentation fault if i try to run this by using mpirun -np 3 ./test for example. I believe it is because of some pointers or maybe something else is incorrect. I am very novice at mpi. Can someone kindly help in scattering the values pointed by the (double *) pointer, do some simple compuatation and then return the result back to root with an easy to understand example?
