Ged Ridgway

Extremely useful -- thank you!

I'm glad someone has implemented this. Two comments:

Apparently an O(N*log(N)) algorithm is available:
http://en.wikipedia.org/wiki/Matrix_chain_multiplication#An_Even_More_Efficient_Algorithm
though I don't know how this compares to the current O(N^3) one for typical values of N (e.g. the constants could make N*log(N) worse until N gets above a certain level).

More importantly (for me, anyway), I think the computation of the ordering should be a separate function, and a set ordering should then be usable in a multiplication function. E.g. if I have some code which repeatedly computes a chain multiplication in a loop, where the matrices' sizes are always the same and are known in advance, but their values vary, then much better than using mmtimes(A,B,C,...) would be to find the correct ordering first, and then hard-code that ordering in the loop. Similarly, if their sizes are constant but not known in advance, then a two-function division of labour would allow the ordering to be computed on the first iteration of the loop, and then reused without recomputation in successive iterations.

c'est un travail interessant!