Is it possible to improve performance of the Kron operator by making it a built in function rather than using m-file implementation?

I use the kron operator as well as row-wise and column-wise Kronecker operators extensively in my work and in packages I write in Matlab. Given the importance of these operators it would be nice to have intrinsic functions rather than m-files implementations. For years I have relied on MEX file versions that I wrote which can be hundreds of times faster than the m-file versions, especially for sparse matrices. I would like you to consider providing intrincis versions of these operators. I've attached my MEX file versions which you are welcome to use.

Answer 1

[MathWorks Support Team]

The functionality requested is currently unavailable in MATLAB. This has been forwarded to our Developers as an enhancement request.

[Matt J]

Fabricius Somogyi's comment moved here:

Unfortunately I cannot find the MEX files that you attached in this thread. Could you kindly re-attach them? I am having similar issues with kron() being too slow for my purposes. Thanks!

[Matt J]

amir sh's comment moved here:

is it possible for you to re-attach the MEX file?

Answer 2

[Matt J]

I wouldn't expect an intrinsic implementation to improve much. Most of the computational intensity of KRON is in all the multiplication operations required and those are well-accelerated through calls to BSXFUN.

Also, there are many operations involving Kronecker product matrices that are more efficiently done without computing the product explicitly with KRON. See here for a tool that facilitates this,

https://www.mathworks.com/matlabcentral/fileexchange/25969-efficient-object-oriented-kronecker-product-manipulation