Using flops_div() from the lightspeed toolbox by tom minka..

I'm using the Lightspeed toolbox created by Tom Minka to find the FLOP count for a project I'm using. For multiplication I used: addflops(flops_mul(matrix1, matrix2)); But for division (flops_div) it doesn't allow any input arguments, has anyone else used this toolbox and know how to properly use this function?

Answer 1

[Walter Roberson]

Multiply the results of flops_div by the number of scalars you are dividing.

flops_div is for scalar division, not for matrix algebra division.

[Gina DiCarlo]

So if I'm simply dividing 20 by 3 or something, then the answer would just be 8 (1*8)?

[Gina DiCarlo]

Or by the number of scalars do you mean the length of the number, so in this case it would be 2 (2 and 0) and the answer would be 16?

[Walter Roberson]

For the case of A ./ B

div_count = max(numel(A),numel(B)) * flops_div

This covers the case where one of the two might be a scalar, and also covers the case where the two are the same size. The case where the two are non-scalars of different size is an error for the ./ operator.

For the case of A / B then if B is a scalar then the operation is the same as A ./ B . For the case of A / B where B is not a scalar then that is matrix algebra division and flops_div does not cover that.