Posted on October 03, 2011 at 19:14I sometimes do code-profiling by setting up a regular interrupt which peeks at the return address and then adds one to the appropriate ''bin'' (array). This way I can see where the processor spends most of its time and thereby tell which functions might benefit by hand-optimised assembly-language.
But there are often greater savings that can be made by doing the processing in a more-efficient mannar e.g. heapsort/quicksort rather than bubble-sort, use of fixed-point-word-length variables rather than float/double/long long for calculations.
And sometimes the right approach is to make your code ''feel'' more responsive by arranging it in threads of different priority.
Hope this helps,
Danish
