arXiv subjects

Jack Dongarra

Publications and source records attributed to Jack Dongarra.

At least 19 recordsLinked to original sources

arXiv2026

Analysis of Floating-Point Matrix Multiplication Computed via Integer Arithmetic

Ootomo, Ozaki, and Yokota [Int. J. High Perform. Comput. Appl., 38 (2024), p. 297-313] have proposed a strategy to recast a floating-point matrix multiplication in terms of integer matrix products. The factors A and B are split into integer slices, the product of these slices is computed exactly, and AB is approximated by accumulating these integer products in floating-point arithmetic. This technique is particularly well suited to mixed-precision matrix multiply-accumulate units with integer support, such as the NVIDIA tensor cores or the AMD matrix cores. The number of slices allows for performance-accuracy tradeoffs: more slices yield better accuracy but require more multiplications, which in turn reduce performance. We propose an inexpensive way to estimate the minimum number of multiplications needed to achieve a prescribed level of accuracy. Our error analysis shows that the algorithm may become inaccurate (or inefficient) if rows of A or columns of B are badly scaled. We perform a range of numerical experiments, both in simulation and on the latest NVIDIA GPUs, that confirm the analysis and illustrate strengths and weaknesses of the algorithm.

math.NA↗

arXiv2025

HPL-MxP Benchmark: Mixed-Precision Algorithms, Iterative Refinement, and Scalable Data Generation

We present a mixed-precision benchmark called HPL-MxP that uses both a lower-precision LU factorization with a non-stationary iterative refinement based on GMRES. We evaluate the numerical stability of one of the methods of generating the input matrix in a scalable fashion and show how the diagonal scaling affects the solution quality in terms of the backward-error. Some of the performance results at large scale supercomputing installations produced Exascale-level compute throughput numbers thus proving the viability of the proposed benchmark for evaluating such machines. We also present the potential of the benchmark to continue increasing its use with proliferation of hardware accelerators for AI workloads whose reliable evaluation continues to pose a particular challenge for the users.

math.NA↗

arXiv2025

The Stability of Block Eliminations and Additive Modifications

The block elimination with additive modifications (BEAM) method was recently proposed as a alternative to LU with partial pivoting requiring less communication. Because of the novelty of BEAM, the existing theoretical analysis is lacking. To that end, we analyze both the numerical stability of the underlying block LU factorization and the effects of additive modifications. For the block LU factorization, we are able to improve the previous results of Demmel et al. from being cubic in the element growth to merely quadratic. Furthermore, we propose an alternative measure of element growth that is better aligned with block LU; this new measure of growth allows our analysis to apply to matrices that cannot be factored with pointwise LU. In the second part, we analyzed the modifications produced by BEAM and the effect they have on the condition number and growth factor. Finally, we show that BEAM will not apply any modifications in some cases that regular block LU can safely factor.

math.NA↗

arXiv2024

Hardware Trends Impacting Floating-Point Computations In Scientific Applications

The evolution of floating-point computation has been shaped by algorithmic advancements, architectural innovations, and the increasing computational demands of modern technologies, such as artificial intelligence (AI) and high-performance computing (HPC). This paper examines the historical progression of floating-point computation in scientific applications and contextualizes recent trends driven by AI, particularly the adoption of reduced-precision floating-point types. The challenges posed by these trends, including the trade-offs between performance, efficiency, and precision, are discussed, as are innovations in mixed-precision computing and emulation algorithms that offer solutions to these challenges. This paper also explores architectural shifts, including the role of specialized and general-purpose hardware, and how these trends will influence future advancements in scientific computing, energy efficiency, and system design.

math.NA↗

arXiv2024

XaaS: Acceleration as a Service to Enable Productive High-Performance Cloud Computing

HPC and Cloud have evolved independently, specializing their innovations into performance or productivity. Acceleration as a Service (XaaS) is a recipe to empower both fields with a shared execution platform that provides transparent access to computing resources, regardless of the underlying cloud or HPC service provider. Bridging HPC and cloud advancements, XaaS presents a unified architecture built on performance-portable containers. Our converged model concentrates on low-overhead, high-performance communication and computing, targeting resource-intensive workloads from climate simulations to machine learning. XaaS lifts the restricted allocation model of Function-as-a-Service (FaaS), allowing users to benefit from the flexibility and efficient resource utilization of serverless while supporting long-running and performance-sensitive workloads from HPC.

cs.DC↗

arXiv2023

Generalizing Random Butterfly Transforms to Arbitrary Matrix Sizes

Parker and Lê introduced random butterfly transforms (RBTs) as a preprocessing technique to replace pivoting in dense LU factorization. Unfortunately, their FFT-like recursive structure restricts the dimensions of the matrix. Furthermore, on multi-node systems, efficient management of the communication overheads restricts the matrix's distribution even more. To remove these limitations, we have generalized the RBT to arbitrary matrix sizes by truncating the dimensions of each layer in the transform. We expanded Parker's theoretical analysis to generalized RBT, specifically that in exact arithmetic, Gaussian elimination with no pivoting will succeed with probability 1 after transforming a matrix with full-depth RBTs. Furthermore, we experimentally show that these generalized transforms improve performance over Parker's formulation by up to 62\% while retaining the ability to replace pivoting. This generalized RBT is available in the SLATE numerical software library.

math.NA↗

arXiv2023

Randomized Numerical Linear Algebra : A Perspective on the Field With an Eye to Software

Randomized numerical linear algebra - RandNLA, for short - concerns the use of randomization as a resource to develop improved algorithms for large-scale linear algebra computations. The origins of contemporary RandNLA lay in theoretical computer science, where it blossomed from a simple idea: randomization provides an avenue for computing approximate solutions to linear algebra problems more efficiently than deterministic algorithms. This idea proved fruitful in the development of scalable algorithms for machine learning and statistical data analysis applications. However, RandNLA's true potential only came into focus upon integration with the fields of numerical analysis and "classical" numerical linear algebra. Through the efforts of many individuals, randomized algorithms have been developed that provide full control over the accuracy of their solutions and that can be every bit as reliable as algorithms that might be found in libraries such as LAPACK. Recent years have even seen the incorporation of certain RandNLA methods into MATLAB, the NAG Library, NVIDIA's cuSOLVER, and SciKit-Learn. For all its success, we believe that RandNLA has yet to realize its full potential. In particular, we believe the scientific community stands to benefit significantly from suitably defined "RandBLAS" and "RandLAPACK" libraries, to serve as standards conceptually analogous to BLAS and LAPACK. This 200-page monograph represents a step toward defining such standards. In it, we cover topics spanning basic sketching, least squares and optimization, low-rank approximation, full matrix decompositions, leverage score sampling, and sketching data with tensor product structures (among others). Much of the provided pseudo-code has been tested via publicly available MATLAB and Python implementations.

math.NA↗

arXiv2022

Proposed Consistent Exception Handling for the BLAS and LAPACK

Numerical exceptions, which may be caused by overflow, operations like division by 0 or sqrt(-1), or convergence failures, are unavoidable in many cases, in particular when software is used on unforeseen and difficult inputs. As more aspects of society become automated, e.g., self-driving cars, health monitors, and cyber-physical systems more generally, it is becoming increasingly important to design software that is resilient to exceptions, and that responds to them in a consistent way. Consistency is needed to allow users to build higher-level software that is also resilient and consistent (and so on recursively). In this paper we explore the design space of consistent exception handling for the widely used BLAS and LAPACK linear algebra libraries, pointing out a variety of instances of inconsistent exception handling in the current versions, and propose a new design that balances consistency, complexity, ease of use, and performance. Some compromises are needed, because there are preexisting inconsistencies that are outside our control, including in or between existing vendor BLAS implementations, different programming languages, and even compilers for the same programming language. And user requests from our surveys are quite diverse. We also propose our design as a possible model for other numerical software, and welcome comments on our design choices.

cs.MS↗

arXiv2022

Reinventing High Performance Computing: Challenges and Opportunities

The world of computing is in rapid transition, now dominated by a world of smartphones and cloud services, with profound implications for the future of advanced scientific computing. Simply put, high-performance computing (HPC) is at an important inflection point. For the last 60 years, the world's fastest supercomputers were almost exclusively produced in the United States on behalf of scientific research in the national laboratories. Change is now in the wind. While costs now stretch the limits of U.S. government funding for advanced computing, Japan and China are now leaders in the bespoke HPC systems funded by government mandates. Meanwhile, the global semiconductor shortage and political battles surrounding fabrication facilities affect everyone. However, another, perhaps even deeper, fundamental change has occurred. The major cloud vendors have invested in global networks of massive scale systems that dwarf today's HPC systems. Driven by the computing demands of AI, these cloud systems are increasingly built using custom semiconductors, reducing the financial leverage of traditional computing vendors. These cloud systems are now breaking barriers in game playing and computer vision, reshaping how we think about the nature of scientific computation. Building the next generation of leading edge HPC systems will require rethinking many fundamentals and historical approaches by embracing end-to-end co-design; custom hardware configurations and packaging; large-scale prototyping, as was common thirty years ago; and collaborative partnerships with the dominant computing ecosystem companies, smartphone, and cloud computing vendors.

cs.DC↗

arXiv2021

Efficient Exascale Discretizations: High-Order Finite Element Methods

Efficient exploitation of exascale architectures requires rethinking of the numerical algorithms used in many large-scale applications. These architectures favor algorithms that expose ultra fine-grain parallelism and maximize the ratio of floating point operations to energy intensive data movement. One of the few viable approaches to achieve high efficiency in the area of PDE discretizations on unstructured grids is to use matrix-free/partially-assembled high-order finite element methods, since these methods can increase the accuracy and/or lower the computational time due to reduced data motion. In this paper we provide an overview of the research and development activities in the Center for Efficient Exascale Discretizations (CEED), a co-design center in the Exascale Computing Project that is focused on the development of next-generation discretization software and algorithms to enable a wide range of finite element applications to run efficiently on future hardware. CEED is a research partnership involving more than 30 computational scientists from two US national labs and five universities, including members of the Nek5000, MFEM, MAGMA and PETSc projects. We discuss the CEED co-design activities based on targeted benchmarks, miniapps and discretization libraries and our work on performance optimizations for large-scale GPU architectures. We also provide a broad overview of research and development activities in areas such as unstructured adaptive mesh refinement algorithms, matrix-free linear solvers, high-order data visualization, and list examples of collaborations with several ECP and external applications.

cs.DC↗

arXiv2020

Integrating Deep Learning in Domain Sciences at Exascale

This paper presents some of the current challenges in designing deep learning artificial intelligence (AI) and integrating it with traditional high-performance computing (HPC) simulations. We evaluate existing packages for their ability to run deep learning models and applications on large-scale HPC systems efficiently, identify challenges, and propose new asynchronous parallelization and optimization techniques for current large-scale heterogeneous systems and upcoming exascale systems. These developments, along with existing HPC AI software capabilities, have been integrated into MagmaDNN, an open-source HPC deep learning framework. Many deep learning frameworks are targeted at data scientists and fall short in providing quality integration into existing HPC workflows. This paper discusses the necessities of an HPC deep learning framework and how those needs can be provided (e.g., as in MagmaDNN) through a deep integration with existing HPC libraries, such as MAGMA and its modular memory management, MPI, CuBLAS, CuDNN, MKL, and HIP. Advancements are also illustrated through the use of algorithmic enhancements in reduced- and mixed-precision, as well as asynchronous optimization methods. Finally, we present illustrations and potential solutions for enhancing traditional compute- and data-intensive applications at ORNL and UTK with AI. The approaches and future challenges are illustrated in materials science, imaging, and climate applications.

cs.LG↗

arXiv2020

Improving the Performance of the GMRES Method using Mixed-Precision Techniques

The GMRES method is used to solve sparse, non-symmetric systems of linear equations arising from many scientific applications. The solver performance within a single node is memory bound, due to the low arithmetic intensity of its computational kernels. To reduce the amount of data movement, and thus, to improve performance, we investigated the effect of using a mix of single and double precision while retaining double-precision accuracy. Previous efforts have explored reduced precision in the preconditioner, but the use of reduced precision in the solver itself has received limited attention. We found that GMRES only needs double precision in computing the residual and updating the approximate solution to achieve double-precision accuracy, although it must restart after each improvement of single-precision accuracy. This finding holds for the tested orthogonalization schemes: Modified Gram-Schmidt (MGS) and Classical Gram-Schmidt with Re-orthogonalization (CGSR). Furthermore, our mixed-precision GMRES, when restarted at least once, performed 19% and 24% faster on average than double-precision GMRES for MGS and CGSR, respectively. Our implementation uses generic programming techniques to ease the burden of coding implementations for different data types. Our use of the Kokkos library allowed us to exploit parallelism and optimize data management. Additionally, KokkosKernels was used when producing performance results. In conclusion, using a mix of single and double precision in GMRES can improve performance while retaining double-precision accuracy.

math.NA↗

arXiv2020

A Survey of Numerical Methods Utilizing Mixed Precision Arithmetic

Within the past years, hardware vendors have started designing low precision special function units in response to the demand of the Machine Learning community and their demand for high compute power in low precision formats. Also the server-line products are increasingly featuring low-precision special function units, such as the NVIDIA tensor cores in ORNL's Summit supercomputer providing more than an order of magnitude higher performance than what is available in IEEE double precision. At the same time, the gap between the compute power on the one hand and the memory bandwidth on the other hand keeps increasing, making data access and communication prohibitively expensive compared to arithmetic operations. To start the multiprecision focus effort, we survey the numerical linear algebra community and summarize all existing multiprecision knowledge, expertise, and software capabilities in this landscape analysis report. We also include current efforts and preliminary results that may not yet be considered "mature technology," but have the potential to grow into production quality within the multiprecision focus effort. As we expect the reader to be familiar with the basics of numerical linear algebra, we refrain from providing a detailed background on the algorithms themselves but focus on how mixed- and multiprecision technology can help improving the performance of these methods and present highlights of application significantly outperforming the traditional fixed precision methods.

cs.MS↗

arXiv2016

Bidiagonalization with Parallel Tiled Algorithms

We consider algorithms for going from a "full" matrix to a condensed "band bidiagonal" form using orthogonal transformations. We use the framework of "algorithms by tiles". Within this framework, we study: (i) the tiled bidiagonalization algorithm BiDiag, which is a tiled version of the standard scalar bidiagonalization algorithm; and (ii) the R-bidiagonalization algorithm R-BiDiag, which is a tiled version of the algorithm which consists in first performing the QR factorization of the initial matrix, then performing the band-bidiagonalization of the R-factor. For both bidiagonalization algorithms BiDiag and R-BiDiag, we use four main types of reduction trees, namely FlatTS, FlatTT, Greedy, and a newly introduced auto-adaptive tree, Auto. We provide a study of critical path lengths for these tiled algorithms, which shows that (i) R-BiDiag has a shorter critical path length than BiDiag for tall and skinny matrices, and (ii) Greedy based schemes are much better than earlier proposed variants with unbounded resources. We provide experiments on a single multicore node, and on a few multicore nodes of a parallel distributed shared-memory system, to show the superiority of the new algorithms on a variety of matrix sizes, matrix shapes and core counts.

cs.MS↗

arXiv2014

Standards for Graph Algorithm Primitives

It is our view that the state of the art in constructing a large collection of graph algorithms in terms of linear algebraic operations is mature enough to support the emergence of a standard set of primitive building blocks. This paper is a position paper defining the problem and announcing our intention to launch an open effort to define this standard.

cs.MS↗

arXiv2014

Designing LU-QR hybrid solvers for performance and stability

This paper introduces hybrid LU-QR al- gorithms for solving dense linear systems of the form Ax = b. Throughout a matrix factorization, these al- gorithms dynamically alternate LU with local pivoting and QR elimination steps, based upon some robustness criterion. LU elimination steps can be very efficiently parallelized, and are twice as cheap in terms of floating- point operations, as QR steps. However, LU steps are not necessarily stable, while QR steps are always stable. The hybrid algorithms execute a QR step when a robustness criterion detects some risk for instability, and they execute an LU step otherwise. Ideally, the choice between LU and QR steps must have a small computational overhead and must provide a satisfactory level of stability with as few QR steps as possible. In this paper, we introduce several robustness criteria and we establish upper bounds on the growth factor of the norm of the updated matrix incurred by each of these criteria. In addition, we describe the implementation of the hybrid algorithms through an exten- sion of the PaRSEC software to allow for dynamic choices during execution. Finally, we analyze both stability and performance results compared to state-of-the-art linear solvers on parallel distributed multicore platforms.

math.NA↗

arXiv2013

Optimal Checkpointing Period: Time vs. Energy

This short paper deals with parallel scientific applications using non-blocking and periodic coordinated checkpointing to enforce resilience. We provide a model and detailed formulas for total execution time and consumed energy. We characterize the optimal period for both objectives, and we assess the range of time/energy trade-offs to be made by instantiating the model with a set of realistic scenarios for Exascale systems. We give a particular emphasis to I/O transfers, because the relative cost of communication is expected to dramatically increase, both in terms of latency and consumed energy, for future Exascale platforms.

cs.DC↗

arXiv2012

A hybrid Hermitian general eigenvalue solver

The adoption of hybrid GPU-CPU nodes in traditional supercomputing platforms opens acceleration opportunities for electronic structure calculations in materials science and chemistry applications, where medium sized Hermitian generalized eigenvalue problems must be solved many times. The small size of the problems limits the scalability on a distributed memory system, hence they can benefit from the massive computational performance concentrated on a single node, hybrid GPU-CPU system. However, new algorithms that efficiently exploit heterogeneity and massive parallelism of not just GPUs, but of multi/many-core CPUs as well are required. Addressing these demands, we implemented a novel Hermitian general eigensolver algorithm. This algorithm is based on a standard eigenvalue solver, and existing algorithms can be used. The resulting eigensolvers are state-of-the-art in HPC, significantly outperforming existing libraries. We analyze their performance impact on applications of interest, when different fractions of eigenvectors are needed by the host electronic structure code.

math.NA↗