/Barnes2019-Landscape

Landscape evolution models and graph processing on the GPU

Primary LanguageC++

DOI

Barnes2019-Landscape

Title of Manuscript: Accelerating a fluvial incision and landscape evolution model with parallelism

Authors: Richard Barnes

Corresponding Author: Richard Barnes (richard.barnes@berkeley.edu)

DOI Number of Manuscript: https://arxiv.org/abs/1803.02977

Code Repositories

This repository contains a reference implementation of the algorithms presented in the manuscript above, along with information on acquiring the various datasets used, and code to perform correctness tests.

Abstract

Solving inverse problems and achieving statistical rigour in landscape evolution models requires running many model realizations. Parallel computation is necessary to achieve this in a reasonable time. However, no previous algorithm is well-suited to leveraging modern parallelism. Here, I describe an algorithm that can utilize the parallel potential of GPUs, many-core processors, and SIMD instructions, in addition to working well in serial. The new algorithm runs 43 x faster (70 s vs. 3,000 s on a 10,000 x 10,000 input) than the previous state of the art and exhibits sublinear scaling with input size. I also identify methods for using multidirectional flow routing and quickly eliminating landscape depressions and local minima. Tips for parallelization and a step-by-step guide to achieving it are given to help others achieve good performance with their own code. Complete, well-commented, easily adaptable source code for all versions of the algorithm is available as a supplement and on Github.

Compilation

Run

make

to compile all non-GPU code using your default compiler.

Run

make -f Makefile.summitdev

to compile all GPU code using the PGI compiler.

Correctness

Run

./compare.sh

to ensure that all algorithms are producing the same output

Empirical Comparisons

Run

./tests/tests.sh

To run all tests.