Section: Overall Objectives

Overall Objectives

Postale is an Inria Saclay Île-de-France team in the area of high-performance computing (HPC), parallel architectures and compilation. The Postale acronym stands for "Performance Optimization by Software Transformation and Algorithms & Libraries Enhancement". Postale focuses on providing software and hardware means to help programmers to deal with the ever growing complexity of programming state-of-the-art parallel and distributed architectures and to develop optimized HPC applications. The Postale team involves researchers from Laboratoire de Recherche en Informatique (LRI) - University Paris-Sud - and have expertise in various domains including algorithms for HPC, programming languages, compilers, and architectures. The project is structured around two main research issues:

• Develop methods and software for program transformations/optimizations for a given algorithm/application and take advantage of programmer knowledge to develop efficient codes through programmer/compiler interface and domain specific languages (DSL),

• Provide innovative algorithms and efficient implementations in high-performance computing libraries for current highly parallel and heterogeneous or embedded architectures, and explore current barriers to performance.

Following the Inria terminology, the Postale team belongs to the field “Algorithmics, Programming, Software and Architecture’’ in the category `Architecture and Compiling’’. The specificity of this project among other Inria teams addressing similar topics is that it does not focus only on architecture caracteristics and low level aspects of program execution but it takes into account the dimension of the user or program developer and their specific domain of application. In particular it aims at developing paths between programmers at the application level and computing resources. The targeted applications are high-performance scientific or image processing applications that require efficient use of ever developing highly parallel and heterogeneous systems. Since the applications are at the heart of our research, the members of the Postale team share the common goal of providing users with the most adequate compiler/user interface and software for their scientific application. In this project, we address issues which are transverse to most research objectives but with a different point of view, depending on if we work at the compiler or at the algorithm level. Namely, these issues are related to minimizing energy consumption and the amount of communication or synchronizations, optimizing performance and data locality, proposing user interfaces as close as possible to application domains.