| The main contribution of this dissertation is exploring the impact of, and improving memory instruction processing to improve the efficiency of the processor. In particular, we propose and evaluate five techniques that tackle certain inefficiencies that exist in memory instruction processing units of conventional processors. Our proposals leverage certain commonly-occurring behavioral patterns in memory instructions which allow for relatively simple and scalable designs to improve processor efficiency.;One of the main aspects of this thesis is the study of memory dependence prediction. We explore the fundamental concepts in memory dependence prediction and propose, to the best of our knowledge, one of the most high-performing and power-efficient memory dependence predictors. Using the fundamental principles and benefits of memory dependence prediction, we then propose an application of it in simultaneous multi-threaded processors that reduces the resource contention in, and improves the performance of these processors.;In our third proposal, we focus on a useful corollary of the predictability in memory dependences, which is the predictability in memory data forwarding, to simplify, reduce and even eliminate some of the power-hungry and area-inefficient resources required by memory instructions in conventional processors. The designs proposed have low implementation complexity and high scalability. We then focus our attention on another memory instruction behavioral pattern, that is predictability in instruction criticality and leverage this behavior to improve the performance of the processor as well as reduce its power consumption and area requirements.;Finally we explore out-of-order allocation and deallocation of certain microarchitectural resources used by memory instructions in the processor which allows us to reduce the energy consumption and the area overhead of the processor. Memory instructions constitute a significant portion of the instructions in traditional programs and as such their processing is very important to the performance, power, area, design complexity and scalability of modern processors.;This research presented in this dissertation improves the processing of these instructions so that the above-stated parameters can be optimized and improved, which can lead to the design of highly efficient processors. |
