Anderson Faustino da Silva

Compiling for performance and power efficiency

Performance and power efficiency are issues that can be addressed by the compiler, due to the fact that the compiler will generate code that will exercise several parts of the processor. The mainstream compilers apply several optimizations in order to improve the quality of the final code, but not all optimizations will result in a performance gain and/or power efficiency. In fact, in some cases, some optimizations can cause performance loss and increase the power consumption, due to the program characteristics do not fit the characteristics of optimizations. Therefore, it is a challenge, even for the most expert programmer, to know which optimizations and in which order will generate the best target code for a program in terms of multiple goals. The goal of this paper is to describe COSPpp, a case-based reasoning approach that automatically selects a compiler optimization set for a program that outperforms a well-engineered compiler optimization level, in terms of multiple goals. The results obtained by the proposed approach indicates that it achieves improvement in most cases. In fact, COSPpp achieves a balanced performance and power efficiency ratio close to 7%.

Register Allocation with Graph Coloring by Ant Colony Optimization

The goal of register allocation is to allocate an unbounded number of program values to a finite number of machine registers. In this paper, we describe a new algorithm for intraprocedural register allocation called CA-RT-RA, an algorithm that extends a classic graph coloring register allocator to use our graph coloring algorithm Color Ant-RT. The experiments demonstrated that our algorithm is able to minimize the amount of spills, thereby improving the quality of the generated code. CA-RT-RA is interesting in applications where compile time is not a concern, but the code quality.

Toward Better Performance of ColorAnt ACO Algorithm

This paper presents the Color Ant-RT algorithm version 3, an algorithm for graph coloring problems which is based on the Ant Colony Optimization metaheuristic and uses Tabu Search as local search. The experiments demonstrated that ColorAnt3-RT is a promising option in finding good approximations to the best known results for geometric random, geometric standard and le450 graphs of DIMACS benchmark in an acceptable runtime, also, it is good in minimizing the amount of conflicts, the main problem of graph coloring with a fixed number of colors.

Register Allocation by Evolutionary Algorithm

Graph coloring is a highly effective approach to intraprocedural register allocation. In this paper, we describe a new algorithm for intraprocedural register allocation called HECRA, an algorithm that extends a classic graph coloring register allocator to use a hybrid evolutionary coloring algorithm. The experiments demonstrated that our algorithm is able to minimize the amount of spills, thereby improving the quality of the generated code. Besides, HECRA is interesting in contexts where compile time is a concern, and not only the quality of the generated code.

Dynamic Binary Translation -- A Model-Driven Approach

Dynamic Binary Translation (DBT) is a technique used to allow execution of non-native machine code. It is an extensively explored research topic and lately it is being given much attention due to the increased demand for system virtual machines. This paper describes a DBT process model, which has been succesfully implemented on a 6502 emulator for x86_64 machines. Among other objectives, the model was conceived to allow the use of an existing interpreter core to implement a DBT-enabled one. Our preliminary results indicates that even a DBT scheme with no runtime information optimizations can lead to performance gain, at the same time saving the effort of implementing instruction code by reusing existing interpreter code.

EProf: An Accurate Energy Consumption Estimation Tool

Energy efficiency is a primary concern in Wireless Sensor Networks because in general, wireless sensor nodes run on non-rechargeable batteries. Therefore, the estimating a expected node lifetime is indispensable. Thus, energy is an important issue that should be optimized by the application developer. However, in order for the developer to realize that the program may not be developed in accordance with this power restriction, it is important to estimate the energy that will be consumed by the application. In this paper we present Eprof, a cycle-accurate instruction level simulator developed to estimate the energy consumption of applications running on UNB-RISC16 processor. Eprof is able to measure detailed time-critical phenomena, and can be used to shed new light on design issues for large-scale sensor network applications.

Yet Another Intelligent Code-Generating System: A Flexible and Low-Cost Solution

Modern compilers apply various code transformation algorithms to improve the quality of the target code. However, a complex problem is to determine which transformation algorithms must be utilized. This is difficult because of three reasons: a number of transformation algorithms, various combination possibilities, and several configuration possibilities. Over the last few years, various intelligent systems were presented in the literature. The goal of these systems is to search for transformation algorithms and thus apply them to a certain program. This paper proposes a flexible, low-cost and intelligent system capable of identifying transformation algorithms for an input program, considering the program’s specific features. This system is flexible for parameterization selection and has a low-computational cost. In addition, it has the capability to maximize the exploration of available computational resources. The system was implemented under the Low Level Virtual Machine infrastructure and the results indicate that it is capable of exceeding, up to 21.36%, performance reached by other systems. In addition, it achieved an average improvement of up to 17.72% over the most aggressive compiler optimization level of the Low Level Virtual Machine infrastructure.

myBee: An Information System for Precision Beekeeping.

Over the last few years, beekeepers have become aware of the necessity of integrating Information Technology into Apiculture. As a result, Precision Beekeeping emerged, which is an area that applies the said technology to monitor bees and, consequently, the state of the colony. However, the development of these platforms is complex due to the requirement of them adapting to various heterogeneous environments and constantly-updated technologies. Thus, the objective of this paper is to propose and develop a Precision Beekeeping Information System, called myBee, whose infrastructure is flexible, secure, fault tolerant, and efficient in decision-making. A real case study shows that myBee is an efficient information system that supports beekeepers in maintaining their apiaries.

Evaluating knowledge representations for program characterization

Knowledge representation attempts to organize the knowledge of a context in order for automated systems to utilize it to solve complex problems. Among several difficult problems, one worth mentioning is called code-generation , which is undecidable due to its complexity. A technique to mitigate this problem is to represent the knowledge and use an automatic reasoning system to infer an acceptable solution. This article evaluates knowledge representations for program characterization for the context of code-generation systems. The experimental results prove that program Numerical Features as knowledge representation can achieve 85% near to the best possible results. Furthermore, such results demonstrate that an automatic code-generating system, which uses this knowledge representation is capable to obtain performance better than others codegenerating systems.

Combining Machine Learning with a Genetic Algorithm to Find Good Complier Optimizations Sequences.

Artificial Intelligence is a strategy applied in several problems in computer science. One of them is to find good compilers optimizations sequences for programs. Currently, strategies such as Genetic Algorithms and Machine Learning have been used to solve it. This article propose an approach that combines both, Machine Learning and Genetic Algorithms, to solve this problem. The obtained results indicate that the proposed approach achieves performance up to 3.472% over Genetic Algorithms and 4.94% over Machine Learning.