Thread Subject:

about dgemm in lapack

I am using lapack and blas libraries in my C code. I have to multiply A(3 x 3) and B(3 x 1000) to get C(3 x 1000). Could anyone tell me what should be the leading dimension of A , B and C?? Since I am working in C the index starts at 0 rather than at 1 in MATLAB. Please also tell me how do you decide the leading dimension while working in C??

Thanks

Dear Kumar,

> I am using lapack and blas libraries in my C code. I have to multiply A(3 x 3) and B(3 x 1000) to get C(3 x 1000). Could anyone tell me what should be the leading dimension of A , B and C??

Beside the method of reading the BLAS documentation, you could easily try to insert some 3s, 2s, 1000s, 999s in the call and check the results. I know aksing the newsgroup is more sophisticated, but I'm an experimental physicist.

Good luck, Jan

PS. Don't forget to save all open files in Matlab at first! The wrong numbers might cause a funny crash.

"kumar vishwajeet" wrote in message <ies94g$621$1@fred.mathworks.com>...
> I am using lapack and blas libraries in my C code. I have to multiply A(3 x 3) and B(3 x 1000) to get C(3 x 1000). Could anyone tell me what should be the leading dimension of A , B and C?? Since I am working in C the index starts at 0 rather than at 1 in MATLAB. Please also tell me how do you decide the leading dimension while working in C??

For the dimensions in the BLAS libraries, think MATLAB (or Fortran) dimensions, not C dimensions. The memory layout expected by the BLAS routines is the same as MATLAB and Fortran. So the leading dimensions of A, B, and C are all 3. Remember, though, that you are passing everything by reference, not by value. So *don't* just put a 3 in the argument list of the call. You will need to create a variable to hold the 3 value and then pass the address of that to dgemm. e.g., here is the interface

http://www.netlib.org/blas/dgemm.f

      SUBROUTINE DGEMM(TRANSA,TRANSB,M,N,K,ALPHA,A,LDA,B,LDB,BETA,C,LDC)
* .. Scalar Arguments ..
      DOUBLE PRECISION ALPHA,BETA
      INTEGER K,LDA,LDB,LDC,M,N
      CHARACTER TRANSA,TRANSB
* ..
* .. Array Arguments ..
      DOUBLE PRECISION A(LDA,*),B(LDB,*),C(LDC,*)

And according to the MATLAB doc you should be using mwSignedIndex for the integer types being passed. So that means you should be using the following:

char TRANSA, TRANSB;
mwSignedIndex M, N, K, LDA, LDB, LDC;
double ALPHA, BETA;
double *A, *B, *C;

TRANSA = 'N';
TRANAB = 'N';
M = 3;
N = 1000;
K = 3;
LDA = 3;
LDB = 3;
LDC = 3;
ALPHA = 1.0;
BETA = 1.0;
A = whatever; // your 1st input
B = whatever; // your 2nd input
C = whatever; // your pre-allocated output
dgemm( &TRANSA, &TRANSB, &M, &N, &K, &ALPHA, A, &LDA, B, &LDB, &BETA, C, &LDC);


James Tursa

"James Tursa" wrote in message <ievv76$43k$1@fred.mathworks.com>...

> BETA = 1.0;

Correction. For a first call where you are not adding to C you would do this:

BETA = 0.0;

> dgemm( &TRANSA, &TRANSB, &M, &N, &K, &ALPHA, A, &LDA, B, &LDB, &BETA, C, &LDC);

Also, on Unix systems you may need to call dgemm_ (i.e., with an underscore at the end).


James Tursa

"James Tursa" wrote in message <ievv76$43k$1@fred.mathworks.com>...

> TRANAB = 'N';

Typo. Should be:

 TRANSB = 'N';


James Tursa