Search

With the start of the parallel computing era, due to power and thermal considerations, there is a growing need to bridge the gap between parallel hardware and software. The unintuitive nature of parallel programming and the high learning curve often prove a bottleneck in the development of quality parallel software. We propose HAMP - A Highly Abstracted and Modular Programming paradigm for expressing parallel programs. We provide the developer with high level modular constructs that can use to generate hardware specific optimized code. HAMP abstracts programs into important kernels and provides scheduling support to manage parallelism. By abstracting the scheduling and hardware features from the developer, we cannot only, considerably reduce the learning curve, but also increase software lifetime [ABSTRACT FROM AUTHOR]

Module management support is very rough in the C and C++ programming languages. Modules must be separated in interface and implementation files, which will store declarations and definitions, respectively. Ultimately, only text substitution tools are available, by means of the C/C++ preprocessor, which is able to insert an interface file in a given point of a translation unit. This way of managing modules does not take into account aspects like duplicated inclusions, or proper separation of declarations and definitions, just to name a few. While the seasoned programmer will find this characteristic of the language annoying and error-prone, students will find it not less than challenging. In this document, a tool specially designed for improving the support of modules in C++ is presented. Its main advantage is that it makes it easier to manage large, module-based projects, while still allowing to use classic translation units. This tool is designed for students who have to learn modular programming; not only those in the computer science discipline, but also those in other engineerings in which programming is part of the curriculum. [ABSTRACT FROM AUTHOR]

In all but the simplest cases, performing data analysis based on Bayesian reasoning requires the use of advanced algorithms. The Bayesian Analysis Toolkit (BAT) provides a collection of algorithms and methods that facilitate the application of Bayesian statistics to user-defined problems of arbitrary complexity. With BAT.jl, we present a modern rewrite of BAT in the Julia programming language. Through the use of a modular software design that is capable of running parallel and distributed, and by extending the tool with new sampling and integration algorithms, BAT.jl is a high-performance framework for Bayesian inference, meeting the requirements of modern data analysis. [ABSTRACT FROM AUTHOR]

Code modularity is one of the criteria for testing software quality. Nevertheless, novice programmers often tend to avoid writing modular code. In this study, we aim to explore the circumstances in which novice programmers choose to write modular code. To address this aim, two student groups, twenty in each, were given a programming task, each in different configuration. Group-1 was given the task in several phases, each add complexity to the previous one, while Group-2 was given the entire task at once. The entire task was rather complex and its modular solution requires the definition of few classes and methods, each responsible to one of the problem's aspects. The students' solutions were analyzed using the dual-process theory, cognitive dissonance theory and content analysis methods to explore the extent of code modularity. The analysis revealed the following: (1) as regards to Group-1, a minor increase in the number of modular solutions was found while they progressed along the phases; (2) The number of modular solutions of Group-2 was higher than of Group-1. Analysis of students' justifications for lack of code modularity in Group-1 revealed the following. The first phase of the problem was perceived as rather simple hence many students did not find any reason to invest in designing a modular solution. When the task got complex in the following phases, some students realized that a modular solution would fit better, hence a cognitive dissonance was raised. Nevertheless, many of them preferred to continue their course of non-modular solution instead of re-designing a modular new one. In order to decrease the dissonance they claimed that this decision saved time. Students of both groups also attributed their non-modular code to lack of explicit criteria for the evaluation of the code quality that lead them to focus on functionality alone. [ABSTRACT FROM AUTHOR]

For purposes of test beam off-line data analysis a modular software framework has been developed and described in theses proceedings. The TbGaudi package allows to handle the test beam data for a set of runs, as well as the set of different devices under test in one-go, and finally obtain an integrated workflow to present the results. All code is written in C++ programming language. A class based design makes it flexible to add any new features of the device under investigation following a plug-in scheme. Currently, the toolkit handles different types of analysis such as charge collection efficiency, track resolution and eta correction, implemented for non-uniform irradiated sensors. The framework is applied for the LHCb upgrade studies for the VELO and UT groups. [ABSTRACT FROM AUTHOR]

McGNASH is a modern statitistical/preequilibrium nuclear reaction code, being developed at Los Alamos, which can simulate neutron-, proton- and photon-induced reactions in the energy range from a few-keV to about 150 MeV. It is written in modern Fortran 95 scientific language, offering new capabilities both for the developer and the user. McGNASH is still in a development stage, and a first public release is planned for later in 2005. © 2005 American Institute of Physics [ABSTRACT FROM AUTHOR]