Problem in fast matrix multiplication

[thecoder3]

Hey guys, i am trying to implement the “Fast” matrix multiplication as stated in this power point presenation:

http://www.google.pl/url?sa=t&rct=j&q=&esrc=s&frm=1&source=web&cd=5&ved=0CFkQFjAE&url=http%3A%2F%2Fcoachk.cs.ucf.edu%2Fcourses%2FCDA6938%2Fs08%2Fmatrix_multiplication_hongliang.ppt&ei=MueOT5cjw-O1BoLQsKQJ&usg=AFQjCNFnfnU3X2SCsEXNcaVFxtYau2NWKg

It does have a chunk of code and i am trying to implement it but at one point, on the slides there are using:

comp16(b, &ashare[k][0],c)

and i have no idea what it does. I tried googling for it but with no luck. Can anyone shed any light on it and if anyone has implemented this method for matrix multiplication, is it actually faster?

I am struggling big time, so i hope you can help me out a bit.

Thanks again.

[vvolkov]

It does have a chunk of code and i am trying to implement it but at one point, on the slides there are using:

comp16(b, &ashare[k][0],c)

and i have no idea what it does. I tried googling for it but with no luck. Can anyone shed any light on it and if anyone has implemented this method for matrix multiplication, is it actually faster?

This possibly originates from my matrix multiply code published here: The Official NVIDIA Forums | NVIDIA Check calls to saxpy() there.

That code was optimized for G80 and was faster than CUBLAS at the time. I don’t think it is fast/faster on Fermi and Kepler.