Evaluating and improving upon the use of high-performance libraries in linear and multi-linear algebra
Psarras, Christos; Bientinesi, Paolo (Thesis advisor); van de Geijn, Robert (Thesis advisor); Morrison, Abigail Joanna Rhodes (Thesis advisor)
Aachen : RWTH Aachen University (2023)
Dissertation / PhD Thesis
Dissertation, RWTH Aachen University, 2023
This dissertation aims to improve the process of translating (mapping) expressions containing matrices and tensors to high-performance code and thus contributes to the linear and multi-linear algebra domains, respectively. The two domains are examined separately. On the one hand, in the linear algebra domain, standardized building-block libraries (BLAS/LAPACK) have long existed, offering a set of highly optimized functions called kernels. The existence of those libraries has made the task of evaluating mathematical expressions containing matrices equivalent to mapping those expressions to a sequence of kernel invocations. However, end-users are progressively less likely to go through the time-consuming process of directly using said kernels; instead, languages and libraries, which offer a higher level of abstraction, are becoming increasingly popular. These languages and libraries offer mechanisms that internally map an input program to lower level kernels. Unfortunately, our experience suggests that, in terms of performance, this mapping is typically suboptimal. In this dissertation, we formally define the problem of mapping a linear algebra expression to a set of available building blocks as the "Linear Algebra Mapping Problem" (LAMP). Then, we investigate how effectively a benchmark of test problems is solved by popular high-level programming languages and libraries. Specifically, we consider Matlab, Octave, Julia, R, Armadillo (C++), Eigen (C++), and NumPy (Python); the benchmark tests both compiler optimizations, as well as linear algebra specific optimizations, such as the optimal parenthesization of matrix products. Finally, we provide guidelines to language developers and end-users for how to prioritize different optimizations when solving a LAMP. On the other hand, in the multi-linear algebra domain, the absence of standardized, building-block libraries has led to significant replication of effort when it comes to software development. While we point out the benefit that such libraries would provide to a number of scientific fields, at the same time we identify and focus on cases where such a library-based approach would not suffice. Specifically, we investigate two such cases in one of the most prominent fields in multi-linear algebra, chemometrics: the Canonical Polyadic (CP) tensor decomposition and jackknife resampling, a statistical method used to determine uncertainties of CP decompositions.For the first case, the CP decomposition, a broadly used method for computing it relies on the Alternating Least Squares (ALS) algorithm. When the number of components is small, regardless of its implementation, ALS exhibits low arithmetic intensity, which severely hinders its performance and makes GPU offloading ineffective. We observe that, in practice, experts often have to compute multiple decompositions of the same tensor, each with a small number of components (typically fewer than 20). Leveraging this observation, we illustrate how multiple decompositions of the same tensor can be fused together at the algorithmic level to increase the arithmetic intensity. Therefore, both CPUs and GPUs can be utilized more efficiently for further speedups; at the same time the technique is compatible with many enhancements typically used in ALS, such as line search, and non-negativity constraints. We introduce the Concurrent ALS (CALS) algorithm and library, which offers an interface to Matlab, and a mechanism to effectively deal with the issue that decompositions complete at different times. Experimental results on artificial and real datasets demonstrate that the increased arithmetic intensity results in a shorter time to completion.For the second case, we examine jackknife resampling. Upon observation that the jackknife resampled tensors, though different, are nearly identical, we show that it is possible to extend the CALS technique to a jackknife resampling scenario. This extension gives access to the computational efficiency advantage of CALS for the price of a modest increase (typically a few percent) in the number of floating point operations. Numerical experiments on both synthetic and real-world datasets demonstrate that the proposed workflow can be several times faster than a straightforward workflow where the jackknife submodels are processed individually.
- Department of Computer Science 
- Neural Computation Teaching and Research Area