Performance Bottlenecks

I've been looking at the Borland groups for some time, but what drawed
me to this specfic forum was all the Fastcode stuff going on, which I
find interesting for several reasons based on my own project I have
going on now.

So in terms of looking at some of the code, I'm wondering something.
What is generally best to try for something generic that you want to
work in every environment you work on (i.e. portability concerns)? I
guess I'm asking "How does one select which routine(s) you want to try
out?"

Another side question: If I wanted to try to contribute some input
(there are a couple of the categories I saw on the list that I've had
some extensive testing myself for the project above), how would I do
that? Just posting here? I'm not interested in a "contest" or
anything like that, just looking for intellectual enlightenment and
contribution, more or less. Hopefully I can be of some help
somewhere.
.