At first search for the most time consuming part of the function using the PROFILEr. It would be useless to optimize a part, which takes 0.1% of the processing time.
Then examine the found "bottleneck" and check for a proper pre-allocation, the possibility to use integer types instead of doubles, simplify the contents of loops and avoid all repeated calculations. Finally check if a vectorization could be possible.
Does the algorithm exhaust your RAM? Then installing enough memory can give a speed-up of factor 1000.
Can the problem be parallelized? Then PARFOR might be helpful.
After or before the above steps have been performed, buy a computer with the double speed. Then the computation will take 50% of the time.
