Framework for Machine Instruction Usage Analysis

Abstract

When migrating software to new hardware architectures, including the development of optimizing compilers for new platforms, there is a need for statistical analysis of data on the use of different machine instructions or their groups in the machine code of programs. This paper describes a new framework useful for statistical research on machine opcodes that is designed to be extensible and a dataset that can be used by other researchers. We automatically collect data on different GNU/Linux distributions and architectures and provide facilities for its statistical analysis and visualization. Related technical issues are discussed, and solutions to some of them are proposed.