Extensions and Improvements for the Parallel Particle Mesh Environment

Datum

22.09.2016

Zeit

15:00 - 16:00

Sprecher

Tobias Nett

Zugehörigkeit

Institut für Technische Informatik, Professur Compilerbau

Sprache

en

Hauptthema

Informatik

Andere Themen

Informatik

Beschreibung

Domain-specific languages (DSLs) are of utmost importance in scientific high-performance computing to reduce development costs, raise the level of abstraction and, thus, make scientific programmer’s life easier. The parallel particle-mesh environment (PPME) is a DSL and projectional editor for numerical simulations based on the particle method. PPME implements a generative approach: it generates parallel Fortran code that links with the parallel particle-mesh library (PPM), which is also implemented in Fortran. PPM provides efficient implementations of the particle and mesh abstractions, discrete numerics, as well as an abstraction layer on the underlying HPC hardware. In its current state, PPME supports built-in abstractions such as particles, properties, fields, loops and computation phases. Moreover, systems of partial differential equations, differential operators such as the laplacian and fractals can be written using conventional mathematical notations. These concepts have been developed and tested using the example of a Gray-Scott reaction diffusion system, which is discretized and simulated using particles. However, while this example greatly shows the potentials of PPME w.r.t. particle-based simulations, it only includes a small set of equations and only two simulation phases—initialization and solve. Moreover, the editor lacks proper analysis features such as, for instance, type analysis and dead-code analysis. Also, potential optimizations to improve the efficiency of generated code or user experience were not considered yet. This thesis therefore addresses these problems by extending and improving the PPME. In detail, it has the following goals:

• investigation and integration of additional simulation example(s) with multiple move and solve phases (e.g., Lennard-Jones),
• extension of PPME as needed to support the example(s),
• design and implementation of a type system (with support for physical units) and pending analyses,
• statically improve floating-point accuracy using abstract program equivalence graphs and
• provide an evaluation of the implemented optimization(s).

Diese Veranstaltung wird unterstützt von Professur Compilerbau.

Letztmalig verändert: 22.09.2016, 09:44:46