Core Concepts
Regent enables efficient parallel meshfree solver development for heterogeneous HPC platforms.
Abstract
The paper introduces Regent, a parallel programming language for heterogeneous platforms.
A meshfree solver based on the least squares kinetic upwind method is developed in Regent.
Performance comparisons are made between Regent and CUDA-C implementations on GPUs.
Details on data communication, synchronization, and performance analysis are provided.
Results show the computational efficiency of the Regent solver compared to CUDA-C.
Stats
The meshfree solver is verified with standard test cases for inviscid flows.
Benchmark simulations assess solver performance on various point distributions.
Quotes
"Regent can infer data dependencies between tasks."
"Applications written in Regent can be executed on various system configurations."