VESPA: static profiling for binary optimization

Over the past few years, there has been a surge in popularity of binary optimizers such as BOLT, Propeller, Janus and HALO. These tools use dynamic profiling information to make optimization decisions. Although effective, gathering runtime data presents developers with inconveniences unrepresentative inputs, need accommodate software modifications, longer build times. In this paper, we revisit static technique proposed by Calder et al. late 90’s, investigate its application drive optimizations, context BOLT optimizer, replacement for profiling. A core modifications al.’s original proposal, consisting new program features regression model, are sufficient enable some gains obtained through An evaluation powered our profiler on four large benchmarks (clang, GCC, MySQL PostgreSQL) yields binaries that 5.47 % faster than executables produced clang -O3.