topic I had done that already and, in Intel® oneAPI Math Kernel Library

scalapack psgemm fails

Arrigoni__Viviana — Thu, 20 Jun 2019 09:11:16 GMT

Hi.

I want to use psgemm to multiply a matrix A by its transpose on its left-hand-side, namely: C = At * A

I am trying to run a simple program with 9 processors placed in a 3x3 grid. Each processor generates a random square matrix, Ablock, of size block_dim x block_dim (choosing here block_dim = 10), that is a submatrix of A. Hence the global matrix A is a 30x30 matrix.

myrow and mycol are the row and column grid indexes of processors.

idesca and idescc are the descriptors of matrices A and C.

C is initialized as a block_dim x block_dim array.

I call psgemm as follows, but I get a segmentation fault:

psgemm_('T', 'N', block_dim, block_dim, block_dim, &one, Ablock, myrow * block_dim, mycol * block_dim, idesca, Ablock, myrow * block_dim, mycol * block_dim, idescal, &zero, C, myrow * block_dim, mycol * block_dim, idescc);

what am I doing wrong?

Viviana, could you give us

Gennady_F_Intel — Mon, 24 Jun 2019 05:15:55 GMT

Viviana, could you give us the test which we may compile and run on our side? and also show how did you link this case and version of mkl did you try.

Here is what I do before

Arrigoni__Viviana — Mon, 24 Jun 2019 08:20:56 GMT

Here is what I do before calling psgemm:

int n_procs, id;
int nprow, npcol, myrow, mycol, np, info;
int idesca[9], idescc[9];
int block_dim, n;
Cblacs_pinfo(&id, &n_procs);

np = (ceil)(sqrtf((double)n_procs));
n = 27;
block_dim = n / np;

Cblacs_get( -1, 0, &icon );

Cblacs_gridinit( &icon,"r", np, np ); 

Cblacs_gridinfo( icon, &nprow, &npcol, &myrow, &mycol);

descinit_(idesca, &n, &n  , &block_dim, &block_dim , &izero, &izero, &icon, &block_dim, &info);

descinit_(idescc,  &n, &n, &block_dim, &block_dim, &izero, &izero, &icon, &block_dim, &info);

float *Ablock = (float*)calloc(block_dim * block_dim, sizeof(float));

float *C = (float*)calloc(block_dim * block_dim, sizeof(float));

for (int k = 0; k < n_procs; ++k){
                generate_rand_mtx(Ablock, block_dim, block_dim);
}

Where generate_rand_mtx is the following function:

void generate_rand_mtx(float *A, int m , int n){
    int seed = time(NULL);
    srand(seed);
    for (int i = 0; i < m; ++i){
        for (int j = 0; j < n; ++j)
                A[i * n + j] = (float)rand()/(float)(RAND_MAX);
    }
    return;
}

I compile the code in this way:
mpiicc -std=c99 -DMKL_LP64 -I${MKLROOT}/include -o atasc AtA_scalapack.c -L${MKLROOT}/lib/intel64 -lmkl_scalapack_lp64 -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lmkl_blacs_intelmpi_lp64 -lpthread -lm -ldl

and I run it with 9 processors:
mpirun -np 9 ./atasc

-DMKL_LP64 - what is that?

Gennady_F_Intel — Mon, 24 Jun 2019 08:58:19 GMT

-DMKL_LP64 - what is that? Do you want to link with ILP64 API, then please set -DMKL_ILP64 and link with ILP64 libs

I had done that already and,

Arrigoni__Viviana — Mon, 24 Jun 2019 09:10:42 GMT

I had done that already and, together with the segmentation fault, I also got the error that parameter number 9 had an illegal value in descinit. In the past I had solved this problem for another program by linking with LP6, as somebody suggested me in this forum.