Report 9 – Melih Mert
In this week, we have worked on improving our algorithm most of the time. We had implemented the last parallel version of our code by using CUDA architecture. This time, we focused on optimization features that this architecture provided us.
First of all, we built our special clusters for blocks. Then, we controlled that the distribution of gates into these clusters were balanced or not for some test net-lists. Then, we left a number of levels at the top part to be executed by a single block. After arranging the clusters and blocks, we made a number of tests with both parallel and sequential version of code.
At the beginning, we got very bad results for parallel version when it was compared with sequential one. Then, we had to make certain optimizations. As memory optimizations increased in our code, we saw that time of parallel simulation decreased very much. In order to make optimizations in the code, we were quite careful in analyzing our code.
During next week, we will continue with optimization work and try new techniques.
Melih Mert
