Matrix Multiplication SDK question Awkward matrix dimensions

Hi, I’ll apologize in advance for the awkward question but, AFAIK, if we multiply A matrix with B matrix and get C, then the following dimensions hold:

A      X B      = C

 [NxM]    [MxL]    [NxL]

So, how come when I run the matrixMul on a Tesla C2050 I get the following output on the console:

Using Matrix Sizes: A(640 x 960), B(640 x 640), C(640 x 960)

, which translates to

A     X      B       =    C

 [640x960]    [640x640]    [640x960]

Even if we accessed A transposed the output matrix C would be [960x640], unless the result is transposed.

“Equivalently” if its B x A, then the dimensions are correct. Which one holds?


Yeah, probably they are doing C = B x A, since (A x B would not be proper).

You can do really easy array-based multiplications with ArrayFire, which might be useful to your efforts. Cheers!

Right. Thank you for your link. Tell me, can I overload the matrix multiplication so that it supports finit field arithmetics? Matrix-matrix memory instructions remains the same, only the arithmetic instructions change. Is it possible?


I’m not knowledgeable enough to answer that, but you could schedule a call with one of our GPU matrix multiply developers to discuss.