cblas_dgemm - correct parameters : incorrect error message

Question

I am trying to compute: C = 1*(A*B') + 0*C using cblas_dgemm(). As far as I can tell, the parameters are correct. The error message itself does not make sense:

"ldb must be >= MAX(K,1): ldb=3 K=3Parameter 11 to routine cblas_dgemm was incorrect" 

But, ldb = k = 3! Here is the detailed output of all three matrices and the parameters.

 A:  (m x n: 4x3), lda = 4
 B': (n x k: 3x9), lda = 3
 C:  (m x k, 4x9), ldc = 4


A: 
Dense matrix: 0xfe5cf0, nrows = 4, ncols = 3, ColumnMajor = 1
0.1246  0.5407  0.1822 
0.1020  0.4639  0.3164 
0.3058  0.9872  0.3348 
0.8375  0.9343  0.5893 

B: 
Dense matrix: 0xfe5cd0, nrows = 9, ncols = 3, ColumnMajor = 1
0.4387  0.3447  0.2999 
0.1582  0.6505  0.5076 
0.1497  0.6515  0.7486 
0.3936  0.1065  0.7850 
0.7182  0.5477  0.0220 
0.3291  0.0453  0.6630 
0.4005  0.3075  0.1818 
0.4071  0.0083  0.0840 
0.1868  0.0998  0.3279 

C: 
Dense matrix: 0xfe7180, nrows = 4, ncols = 9, ColumnMajor = 1
0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 
0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 
0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 
0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000  0.0000 


Transpose A ta = 111 (CblasNoTrans)
Transpose B tb = 112 (CblasTrans)
m = 4, k = 3, n = 9 
lda = 4, ldb = 3, ldc = 4
alpha = 1, beta = 0

CALLING CBLAS_DGEMM:
cblas_dgemm( CblasColMajor, ta, tb, m, n, k, alpha, A->d, lda, B->d, ldb, beta, C->d, ldc );

I am sure I am making a silly mistake, I just can't spot it. Any help will be appreciated.

many thanks,

Russ

Answer 1

My understanding of ldb was incorrect. ldb refers to the rows(B) and NOT rows(B')! Sorry to have wasted your time. Somehow just posting seems to help.

Answer 2

Thanks for posting the solution, I was having that exact problem. I think the documentation is a bit tricky as i understood it the same way you did. Thanks again!