Year: 2025
Type: article
Abstract: Conventional tools and techniques of bytecode profiling require a lot of resources to generate bytecode sequences. Embedded devices have limited resources, and using conventional profilers to analyze bytecode execution traces and generate basic blocks is time-consuming. However, it often negates the benefits of creating Super instructions in some scenarios. Furthermore, these techniques provide complete basic blocks to be used as Super instructions. This paper proposes a simple and high-performance technique based on partial control flow analysis to identify basic blocks by locating control transfer instructions and their targets. Subsequently, the proposed approach extracts repeating patterns within the basic blocks to determine the most frequently repeated sequences to maximize the benefit from the super instructions. The SPECjvm 2008 benchmark suite and a custom bubble sort example were used in this study for the prototype implementation of the proposed framework. The prototype implementation of the technique identified the sequences of length 2 to 50 from the basic blocks, and then their frequency, ranging from 2 to 50, was extracted. It was observed that most basic blocks consist of 2 to 10 instructions in the actual code. The repeating sequences within basic blocks give better results compared to full basic blocks used as a super instruction. This approach enhances the execution performance and reduces the storage requirements of Java bytecode in embedded devices.
Authors: Adeel Iqbal, Minhaj Ahmad Khan
Institution: Bahauddin Zakariya University
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7130368696