arXiv subjects

Olaf Schenk

Publications and source records attributed to Olaf Schenk.

25 records · Page 2Linked to original sources

arXiv2019

A Recursive Algebraic Coloring Technique for Hardware-Efficient Symmetric Sparse Matrix-Vector Multiplication

The symmetric sparse matrix-vector multiplication (SymmSpMV) is an important building block for many numerical linear algebra kernel operations or graph traversal applications. Parallelizing SymmSpMV on today's multicore platforms with up to 100 cores is difficult due to the need to manage conflicting updates on the result vector. Coloring approaches can be used to solve this problem without data duplication, but existing coloring algorithms do not take load balancing and deep memory hierarchies into account, hampering scalability and full-chip performance. In this work, we propose the recursive algebraic coloring engine (RACE), a novel coloring algorithm and open-source library implementation, which eliminates the shortcomings of previous coloring methods in terms of hardware efficiency and parallelization overhead. We describe the level construction, distance-k coloring, and load balancing steps in RACE, use it to parallelize SymmSpMV, and compare its performance on 31 sparse matrices with other state-of-the-art coloring techniques and Intel MKL on two modern multicore processors. RACE outperforms all other approaches substantially and behaves in accordance with the Roofline model. Outliers are discussed and analyzed in detail. While we focus on SymmSpMV in this paper, our algorithm and software is applicable to any sparse matrix operation with data dependencies that can be resolved by distance-k coloring.

cs.DC↗

arXiv2019

State-of-The-Art Sparse Direct Solvers

In this chapter we will give an insight into modern sparse elimination methods. These are driven by a preprocessing phase based on combinatorial algorithms which improve diagonal dominance, reduce fill-in, and improve concurrency to allow for parallel treatment. Moreover, these methods detect dense submatrices which can be handled by dense matrix kernels based on multithreaded level-3 BLAS. We will demonstrate for problems arising from circuit simulation, how the improvements in recent years have advanced direct solution methods significantly.

cs.DS↗

arXiv2019

Structure Exploiting Interior Point Methods

Interior point methods are among the most popular techniques for large scale nonlinear optimization, owing to their intrinsic ability of scaling to arbitrary large problem sizes. Their efficiency has attracted in recent years a lot of attention due to increasing demand for large scale optimization in industry and engineering. A parallel interior point method is discussed that exploits the intrinsic structure of large-scale nonlinear optimization problems so that the solution process can employ massively parallel high-performance computing infastructures. Since the overall performance of interior point methods relies heavily on scalable sparse linear algebra solvers, particular emphasis is given to the underlying algorithms for the distributed solution of the associated sparse linear systems obtained at each iteration from the linearization of the optimality conditions. The interior point algorithm is implemented in a object-oriented parallel IPM solver and applied for the solution of large scale optimal control problems solved in a daily basis for the secure transmission and distribution of electricity in modern power grids.

math.OC↗

arXiv2018

Optimizing gas networks using adjoint gradients

An increasing amount of gas-fired power plants are currently being installed in modern power grids worldwide. This is due to their low cost and the inherent flexibility offered to the electrical network, particularly in the face of increasing renewable generation. However, the integration and operation of gas generators poses additional challenges to gas network operators, mainly because they can induce rapid changes in the demand. This paper presents an efficient minimization scheme of gas compression costs under dynamic conditions where deliveries to customers are described by time-dependent mass flows. The optimization scheme is comprised of a set of transient nonlinear partial differential equations that model the isothermal gas flow in pipes, an adjoint problem for efficient calculation of the objective gradients and constraint Jacobians, and state-of-the-art optimal control methods for solving nonlinear programs. As the evaluation of constraint Jacobians can become computationally costly as the number of constraints increases, efficient constraint lumping schemes are proposed and investigated with respect to accuracy and performance. The resulting optimal control problems are solved using both interior-point and sequential quadratic programming methods. The proposed optimization framework is validated through several benchmark cases of increasing complexity.

cs.CE↗

arXiv2017

Selected inversion as key to a stable Langevin evolution across the QCD phase boundary

We present new results of full QCD at nonzero chemical potential. In PRD 92, 094516 (2015) the complex Langevin method was shown to break down when the inverse coupling decreases and enters the transition region from the deconfined to the confined phase. We found that the stochastic technique used to estimate the drift term can be very unstable for indefinite matrices. This may be avoided by using the full inverse of the Dirac operator, which is, however, too costly for four-dimensional lattices. The major breakthrough in this work was achieved by realizing that the inverse elements necessary for the drift term can be computed efficiently using the selected inversion technique provided by the parallel sparse direct solver package PARDISO. In our new study we show that no breakdown of the complex Langevin method is encountered and that simulations can be performed across the phase boundary.

hep-lat↗

arXiv2017

Compile-Time Symbolic Differentiation Using C++ Expression Templates

Template metaprogramming is a popular technique for implementing compile time mechanisms for numerical computing. We demonstrate how expression templates can be used for compile time symbolic differentiation of algebraic expressions in C++ computer programs. Given a positive integer $N$ and an algebraic function of multiple variables, the compiler generates executable code for the $N$th partial derivatives of the function. Compile-time simplification of the derivative expressions is achieved using recursive templates. A detailed analysis indicates that current C++ compiler technology is already sufficient for practical use of our results, and highlights a number of issues where further improvements may be desirable.

cs.SC↗

arXiv2005

On Large Scale Diagonalization Techniques for the Anderson Model of Localization

We propose efficient preconditioning algorithms for an eigenvalue problem arising in quantum physics, namely the computation of a few interior eigenvalues and their associated eigenvectors for the largest sparse real and symmetric indefinite matrices of the Anderson model of localization. We compare the Lanczos algorithm in the 1987 implementation by Cullum and Willoughby with the shift-and-invert techniques in the implicitly restarted Lanczos method and in the Jacobi-Davidson method. Our preconditioning approaches for the shift-and-invert symmetric indefinite linear system are based on maximum weighted matchings and algebraic multilevel incomplete $LDL^T$ factorizations. These techniques can be seen as a complement to the alternative idea of using more complete pivoting techniques for the highly ill-conditioned symmetric indefinite Anderson matrices. We demonstrate the effectiveness and the numerical accuracy of these algorithms. Our numerical examples reveal that recent algebraic multilevel preconditioning solvers can accelerative the computation of a large-scale eigenvalue problem corresponding to the Anderson model of localization by several orders of magnitude.

math.NA↗