Question in using Sgemm

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi all,

I have a question on using Sgemm this call.

Currently I have two matrices: (d_A)13 and (d_B)31 as

{1.0,2.0,3.0}

and

{3.0}

{2.0}

{1.0}

And I store them up in a array like this

{1.0,2.0,3.0}

{3.0,2.0,1.0}

The result of this two matrices multiplication should be 10.0

However I got a 0.0 result by a Sgemm call like this:

cublasSgemm('n', 't', 1, 1, 3, 1, d_A, 3, d_B, 3, 0, d_C, 1);

Is there any place I have set the flag wrongly in the call?

2

CUBLAS is column-major and from your setting, lda = 1.

cublasSgemm(‘n’, ‘n’, 1, 1, 3, 1, d_A, 1, d_B, 3, 0, d_C, 1);

3

Thanks for your help LSChien