A Domain-Specific Compiler for a Parallel Multiresolution Adaptive Numerical Simulation Environment
SessionCompilation for Enhanced Parallelism
Session ChairHironori Kasahara
Event Type
Paper
Applications
Intermediate
Introductory
Programming Systems
TimeWednesday, November 16th2pm - 2:30pm
Location355-D
DescriptionThis paper describes the design and implementation of a layered domain-specific compiler to support MADNESS – Multiresolution Adaptive Numerical Environment for Scientific Simulation. MADNESS uses k-d trees to represent spatial functions and operators like addition, multiplication, differentiation, integration on the numerical representation of functions. The MADNESS runtime system provides global namespace support and a task-based execution model including futures. MADNESS is deployed on large scale supercomputers and is used for numerical solution of differential and integral equations arising in many scientific domains. The current distribution of MADNESS is deployed on massively parallel supercomputers and has enabled many science advances. However, since all optimization is based on runtime decisions, scalability is less than desirable. This paper describes a layered domain-specific compiler that enables significant performance improvement of the MADNESS framework.
Download PDF
Paper provided by the IEEE Computer SocietyPaper also available from the ACM Digital Library
Authors
Samyam Rajbhandari (presenting)
