Original scientific paper

https://doi.org/10.31803/tg-20241206111011

Framework for Improving Parallelization of Loops through Loop-Fission Transformation for Many-Core Systems

Akash Yadav ; GLA University, Mathura, UP- 281406, India / Malaviya National Institue of Technology, Jaipur, Rajasthan-302017, India *
Mushtaq Ahmed ; Malaviya National Institue of Technology, Jaipur, Rajasthan-302017, India

* Corresponding author.

Abstract

Many-core systems rely on parallel processing to achieve high-performance computing, but writing efficient parallel code remains challenging, especially for researchers with limited programming expertise. Automatic parallelization tools such as autoPar, Cetus, and Par4All help transform sequential programs into parallel ones, yet they often fail to handle certain programming constructs efficiently. This paper proposes a loop-fission transformation-based framework to improve the parallelization of dependency-free loops with control flow statements. The framework restructures loops to maximize parallel execution, overcoming limitations in existing tools. Experiments include computational statements like array assignments, addition, and multiplication to evaluate performance. Results show that the proposed approach reduces execution time by 4% to 15% on many-core systems compared to existing tools. These findings highlight the inefficiencies of current automatic parallelization techniques and demonstrate the effectiveness of the proposed method in enhancing execution performance.

Keywords

Loop Parallelization; Loop-Fission; OpenMP; Parallelizing Compilers; Superword-Level Parallelization

Hrčak ID:

351724

URI

https://hrcak.srce.hr/351724

Publication date:

15.12.2026.

Visits: 0 *