Finding the best optimization sequence order that can improve the performance for even
a simple program is not an easy task. However, the modern compilers provide dozens of
optimizations, making it not a practical solution to try all the optimization sequences
manually to find the optimal one. In this paper, genetic algorithm is proposed to select
the best optimization sequence for a cluster of similar programs. However, wide set of
programs are elected to cover as much as possible all the features. The set of the
programs are classified into three clusters depending on them features. Thus, the genetic
algorithm in this work is learning method. This means any new program, unseen
program, can take the optimization sequence of the cluster that has similar features to it.
Moreover, two scenarios are proposed using genetic algorithm to find the best
optimization sequence for each cluster. In the first scenario, programs are classified into
three clusters according to program dynamic features. The genetic algorithm with
Tournament selection method is applied on each cluster independently to obtain a good
optimization sequence for a cluster. Moreover, the proposed method improved the
execution time on average by 77% compared with the O2. The second scenario was
exactly similar to the first one. While, different selection methods are used for each
cluster