Development of Rational Plans for Parallel Execution of Programs on Processors With Very Long Machine Words
Year: 2024
Type: conference-paper
Abstract: The purpose of the work is to develop rational (aiming for optimal) methods for obtaining plans (schedules) for parallel execution of programs for processors with very long machine words, based on information analysis of algorithms. A natural method for this case is used to obtain the Stacked Parallel Form (SPF) for the Information Graph Algorithm (IGA) with its subsequent targeted equivalent (without changing information connections) transformations to the given parameters. Transformations are performed using heuristic methods described in the built-in scripting language (a Lua virtual machine is used). The main target criteria were to achieve maximum code density and minimize the computational complexity of obtaining a given parallel execution plan. The size and lifetime distribution of internal (local) data were also analyzed and candidate operators for placement in the processor cache were evaluated. In the course of research, a phenomenon was found that implies the profitability (in the sense of minimizing the computational complexity of transformations) of using SPF in the initial so-called “bottom” form (when all operators are on the SPF tiers closest to the end of the program execution).
Author: V.M. Bakanov
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Keywords: parallel program scheduling, code density
Open Access status: bronze
OpenAlex ID: W4405259274
