Cross-Architecture Execution Optimization for Heterogeneous Computing Platforms

Abstract

With the proliferation of heterogeneous computing platforms, enabling efficient execution of cross-architecture software has become increasingly important. Although dynamic binary translation and virtualization technologies provide functional compatibility, existing solutions often suffer from high translation overhead and limited adaptability to emerging instruction set architectures.

This paper presents a modular cross-architecture execution framework that integrates specification-driven translation generation, hybrid static–dynamic optimization, and runtime profiling mechanisms. The proposed system constructs an intermediate representation layer from formal ISA descriptions and applies workload-aware optimization strategies to hot execution paths. A lightweight virtualization environment is adopted to support aggressive just-in-time compilation without compromising system stability.

Comprehensive evaluations are conducted using SPEC CPU2017 and PARSEC benchmarks under full-system Linux environments. Experimental results demonstrate notable improvements in execution efficiency and translation scalability. The proposed framework significantly reduces engineering effort for new architecture support while maintaining high performance.