topic Pawan, in Intel® oneAPI Math Kernel Library

Example of submatrix multiplication

pawan_k_ — Thu, 01 May 2014 17:08:03 GMT

Hi,

Are there any example of using dgemm for multiplying submatrices of two matrices?

Thanks,

Pawan

Pawan,

Sridevi_A_Intel — Fri, 02 May 2014 20:43:23 GMT

Pawan,

Can you look into the examples directory inside MKL Installation folder?

-Sridevi

Hi,

pawan_k_ — Wed, 07 May 2014 06:38:45 GMT

Hi,

I looked into the example on dgemm. It is not quiet clear how to set lda etc for subblocks.

Here is an example I create.

A = [ 0 1 2 3 4 5 6 7 8 9

10 11 12 13 14 15 16 17 18 19

20 21 22 23 24 25 26 27 28 29

30 31 32 33 34 35 36 37 38 39

40 41 42 43 44 45 46 47 48 49

50 51 52 53 54 55 56 57 58 59

]

B = [

0 1 2 3 4 5

6 7 8 9 10 11

12 13 14 15 16 17

18 19 20 21 22 23

24 25 26 27 28 29

]

C = [ 0 1 2 3 4 5 6 7

8 9 10 11 12 13 14 15

16 17 18 19 20 21 22 23

24 25 26 27 28 29 30 31

32 33 34 35 36 37 38 39

40 41 42 43 44 45 46 47

]

All right! Now I want to operate on submatrices. Let us say I want to multiply a 4x3 submatrix of A as

follows

sub_matrix_A = [

14 15 16

24 25 26

34 35 36

44 45 46

]

with a 3x2 sub-matrix of B as follows

sub_matrix_B = [

7 8

13 14

19 20

]

to put in the 4x2 sub-matrix of C as follows

sub_matrix_C = [

12 13

20 21

28 29

36 37

]

Could someone work out this example for me and future MKL users ? Below are

the parameters to be filled.

cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, 
                m, n, k, alpha, A, k, B, n, beta, C, n);

I want to know the right values of m, n, k to perform

sub_matrix_C = sub_matrix_A * sub_matrix_B

for the example above.

Regards,

Pawan

Could anyone workout this

pawan_k_ — Mon, 19 May 2014 09:13:59 GMT

Could anyone workout this example?

Regards,

Pawan

I suggest that you perform

mecej4 — Tue, 20 May 2014 20:00:00 GMT

I suggest that you perform this operation in two steps. Keeping in mind that CBLAS functions are wrappers around the underlying Fortran BLAS routines, and that terms such as "leading dimension" are confusing and possibly incorrect if the C convention of storing matrices as a 1-D array obtained from concatenation of the rows, examine the attached Fortran example and see if it does what you want.

After noting that the results are as expected, we can move on to replicating the same using C and CBLAS, as in the second attached file.

Hi Pawan,

Ying_H_Intel — Wed, 21 May 2014 07:02:57 GMT

Hi Pawan,

You can calculate the sub_matrix multiply by the below parameters

I attached the cpp file for you reference.

Best Regards,

Ying

#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <mkl.h>

void init_arr(int N, double* a);

int main(int argc, char* argv[])
{
   double A[6][10];
   double B[5][6];
   double C[6][8];

   init_arr(60,&A[0][0]);
   init_arr(30,&B[0][0]);
   init_arr(48,&C[0][0]);

double alpha = 1.0, beta = 0.;
   int m=4; int k=3; int n=2;
   int lda=10; // the stride of row
   int ldb=6;
   int ldc=8;
   double * subA=&A[0][0]+14;
   double * subB=&B[0][0]+7;
   double * subC=&C[0][0]+12;

   cblas_dgemm(CblasRowMajor,CblasNoTrans,CblasNoTrans,m,n,k,alpha,subA,lda,subB,ldb,beta,subC,ldc);

printf("Sub Matrix C(mxn): %d x %d\n", m, n);

   for (int i=0;i<m;i++){
   for (int j=0;j<2;j++) {
           printf("%g\t",subC[ldc*i+j]);
       }
   printf("\n");}
   return 1;
}

void init_arr(int N, double* a)
{
   int i=0;
   for(i=0; i<N;i++)
           a = i;
}

build it with mkl library and get

The result is as below

Sub Matrix C(mxn): 4 x 2
597 642
987 1062
1377 1482
1767 1902
Press any key to continue . . .