Mixed-Precision Performance Portability of FFT-Based GPU-Accelerated Algorithms for Block-Triangular Toeplitz Matrices

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Venkat, Sreeram, Swirydowicz, Kasia, Wolfe, Noah, Ghattas, Omar
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912624989437952
author Venkat, Sreeram
Swirydowicz, Kasia
Wolfe, Noah
Ghattas, Omar
author_facet Venkat, Sreeram
Swirydowicz, Kasia
Wolfe, Noah
Ghattas, Omar
contents The hardware diversity in leadership-class computing facilities, alongside the immense performance boosts from today's GPUs when computing in lower precision, incentivizes scientific HPC workflows to adopt mixed-precision algorithms and performance portability models. We present an on-the-fly framework using hipify for performance portability and apply it to FFTMatvec - an HPC application that computes matrix-vector products with block-triangular Toeplitz matrices. Our approach enables FFTMatvec, initially a CUDA-only application, to run seamlessly on AMD GPUs with excellent performance. Performance optimizations for AMD GPUs are integrated into the open-source rocBLAS library, keeping the application code unchanged. We then present a dynamic mixed-precision framework for FFTMatvec; a Pareto front analysis determines the optimal mixed-precision configuration for a desired error tolerance. Results are shown for AMD Instinct MI250X, MI300X, and the newly launched MI355X GPUs. The performance-portable, mixed-precision FFTMatvec is scaled to 4,096 GPUs on the OLCF Frontier supercomputer.
format Preprint
id arxiv_https___arxiv_org_abs_2508_10202
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Mixed-Precision Performance Portability of FFT-Based GPU-Accelerated Algorithms for Block-Triangular Toeplitz Matrices
Venkat, Sreeram
Swirydowicz, Kasia
Wolfe, Noah
Ghattas, Omar
Distributed, Parallel, and Cluster Computing
Numerical Analysis
Performance
65Y20, 65Y05, 65Y10, 68Q25, 68W40, 65M32, 5B05
F.2; G.4; C.4
The hardware diversity in leadership-class computing facilities, alongside the immense performance boosts from today's GPUs when computing in lower precision, incentivizes scientific HPC workflows to adopt mixed-precision algorithms and performance portability models. We present an on-the-fly framework using hipify for performance portability and apply it to FFTMatvec - an HPC application that computes matrix-vector products with block-triangular Toeplitz matrices. Our approach enables FFTMatvec, initially a CUDA-only application, to run seamlessly on AMD GPUs with excellent performance. Performance optimizations for AMD GPUs are integrated into the open-source rocBLAS library, keeping the application code unchanged. We then present a dynamic mixed-precision framework for FFTMatvec; a Pareto front analysis determines the optimal mixed-precision configuration for a desired error tolerance. Results are shown for AMD Instinct MI250X, MI300X, and the newly launched MI355X GPUs. The performance-portable, mixed-precision FFTMatvec is scaled to 4,096 GPUs on the OLCF Frontier supercomputer.
title Mixed-Precision Performance Portability of FFT-Based GPU-Accelerated Algorithms for Block-Triangular Toeplitz Matrices
topic Distributed, Parallel, and Cluster Computing
Numerical Analysis
Performance
65Y20, 65Y05, 65Y10, 68Q25, 68W40, 65M32, 5B05
F.2; G.4; C.4
url https://arxiv.org/abs/2508.10202