arXiv subjects

Shoji Itoh

Publications and source records attributed to Shoji Itoh.

3 recordsLinked to original sources

arXiv2024

Xabclib:A Fully Auto-tuned Sparse Iterative Solver

In this paper, we propose a general application programming interface named OpenATLib for auto-tuning (AT). OpenATLib is designed to establish the reusability of AT functions. By using OpenATLib, we develop a fully auto-tuned sparse iterative solver named Xabclib. Xabclib has several novel run-time AT functions. First, the following new implementations of sparse matrix-vector multiplication (SpMV) for thread processing are implemented:(1) non-zero elements; (2) omission of zero-elements computation for vector reduction; (3) branchless segmented scan (BSS). According to the performance evaluation and the comparison with conventional implementations, the following results are obtained: (1) 14x speedup for non-zero elements and zero-elements computation omission for symmetric SpMV; (2) 4.62x speedup by using BSS. We also develop a "numerical computation policy" that can optimize memory space and computational accuracy. Using the policy, we obtain the following: (1) an averaged 1/45 memory space reduction; (2) avoidance of the "fault convergence" situation, which is a problem of conventional solvers.

cs.MS↗

arXiv2020

Structure of the polynomials in preconditioned BiCG algorithms and the switching direction of preconditioned systems

We present a theorem that defines the direction of a preconditioned system for the bi-conjugate gradient (BiCG) method, and we extend it to preconditioned bi-Lanczos-type algorithms. We show that the direction of a preconditioned system is switched by construction and by the settings of the initial shadow residual vector. We analyze and compare the polynomial structures of four preconditioned BiCG algorithms.

math.NA↗

arXiv2019

Structure of the preconditioned system in various preconditioned conjugate gradient squared algorithms

An improved preconditioned conjugate gradient squared (PCGS) algorithm has recently been proposed, and it performs much better than the conventional PCGS algorithm. In this paper, the improved PCGS algorithm is verified as a coordinative to the left-preconditioned system, and it has the advantages of both the conventional and the left-PCGS; this is done by comparing, analyzing, and executing numerical examinations of various PCGS algorithms, including another improved one. We show that the direction of the preconditioned system for the CGS method is determined by the operations of $α_k$ and $β_k$ in the PCGS algorithm. By comparing the logical structures of these algorithms, we show that the direction of the preconditioned system can be switched by the construction and setting of the initial shadow residual vector.

math.NA↗