HiPer

March 3, 2021
SMP Pencil CFDNet
Evolution of physical-based simulations and Computational Fluid Dynamics (CFD) in particular has fundamentally reshaped the design and engineering process in the last several decades. However, in spite of noteworthy success, today's CFD still remains limited to a small design space. A critical bottleneck is the inability of current numerical methods and algorithms to predict turbulent-separated flows.

At the same time, today’s supercomputing platforms are based on massive parallelism and heterogeneous processor designs to deliver exascale performance. This project aims at transforming CFD simulation capabilities through a combination of novel numerical schemes and deep learning techniques suited for hybrid architectures and efficient parallel algorithms and implementation strategies to scale on large-scale distributed-memory systems.

HiPer is a new CFD solver that incorporates the above research advances to enable high-performance fluid simulations. Check-out our new multi-block structured mesh partitioner (SMP), pipelined distributed stencil algorithm (Pencil), and DL-based accelerator (CFDNet) thats part of HiPer.

This project is currently supported by an NSF OAC CAREER award.

Publications

ADARNet: Deep Learning Predicts Adaptive Mesh Refinement

Octavi Obiols-Sales, Abhinav Vishnu, Nicholas Malaya, Aparna Chandramowlishwaran
August 2023 International Conference on Parallel Processing (ICPP)
PDF Project Project

Lessons Learned on MPI+Threads Communication

Rohit Zambre, Aparna Chandramowlishwaran
November 2022 Proc. ACM/IEEE International Conference for High Performance Computing, Networking, Storage, and Analysis (SC)
PDF Project

Mosaic flows: A transferable deep learning framework for solving PDEs on unseen domains

Hengjie Wang, Robert Planas, Aparna Chandramowlishwaran, Ramin Bostanabad
January 2022 Computer Methods in Applied Mechanics and Engineering
PDF Project Project

SURFNet: Super-resolution of Turbulent Flows with Transfer Learning using Small Datasets

Octavi Obiols-Sales, Abhinav Vishnu, Nicholas Malaya, Aparna Chandramowlishwaran
September 2021 Proc. 30th International Conference on Parallel Architectures and Compilation Techniques (PACT)
PDF Project Project

Train Once and Use Forever: Solving Boundary Value Problems in Unseen Domains with Pre-trained Deep Learning Models

Hengjie Wang, Robert Planas, Aparna Chandramowlishwaran, Ramin Bostanabad
April 2021 arXiv
PDF Project Project

Pencil: A pipelined algorithm for distributed stencils

Hengjie Wang, Aparna Chandramowlishwaran
November 2020 Proc. ACM/IEEE International Conference for High Performance Computing, Networking, Storage, and Analysis (SC)
PDF Project

CFDNet: A deep learning-based accelerator for fluid simulations

Octavi Obiols-Sales, Abhinav Vishnu, Nicholas Malaya, Aparna Chandramowlishwaran
June 2020 Proc. ACM International Conference on Supercomputing (ICS)
PDF Project Project Video

Multi-criteria partitioning of multi-block structured grids

Hengjie Wang, Aparna Chandramowlishwaran
June 2019 Proc. ACM International Conference on Supercomputing (ICS)
PDF Code Project Slides

Roofline Guided Design and Analysis of a Multi-stencil CFD Solver for Multicore Performance

Bahareh Mostafazadeh, Ferran Marti, Feng Liu, Aparna Chandramowlishwaran
May 2018 Proc. 32nd IEEE Int’l. Parallel and Distributed Processing Symp. (IPDPS)
PDF Project Slides

Unsteady Navier-Stokes Flow on GPU Architectures

Bahareh Mostafazadeh Davani, Ferran Marti, Behnam Pourghassemi, Feng Liu, Aparna Chandramowlishwaran
June 2017 Proc. 23rd AIAA Computational Fluid Dynamics Conference
PDF Project

Talks

Transferable Deep Learning Surrogates for Solving PDEs

May 6, 2021 — Worldwide online event

SPCL_Bcast online seminar at ETH Zurich

Aparna Chandramowlishwaran
Project Project Watch

CFDNet - A deep learning-based accelerator for fluid simulations

June 29, 2020 — July 2, 2020 Worldwide online event

ACM International Conference on Supercomputing (ICS)

Octavi Obiols Sales
Project Project Watch