In this paper, we present a new approach for mapping LLVM ir to binary machine code for overcoming the current limitations of host-based simulations of performance-critical embedded software imposed by compiler optimi...
详细信息
ISBN:
(纸本)9783031045806;9783031045790
In this paper, we present a new approach for mapping LLVM ir to binary machine code for overcoming the current limitations of host-based simulations of performance-critical embedded software imposed by compiler optimizations. Our novel, fully automated mapping approach even copes with aggressive compiler optimizations without requiring any modification to the compiler or the need of expert supervision. Experimental results show that accurate mappings are produced even when compiling with the highest level of optimization (average error below 2%). The proposed simulation methodology provides a speedup of at least 26 compared to the widely used gem5 simulator.
暂无评论