Compiler-assisted out of order execution for memory operations

As out of order techniques in modern processors become harder to scale, one place to look for improved performance is the compiler and its large scope of analysis. This talk presents recent work leveraging LLVM's dependency analysis for use in conjunction with Gem5's memory dependence predictor, leading to performance gains. Future work in integrating this work with the PHAST predictor (Kim and Ros 2024) is also explored, as well as overviewing ongoing work in labelling stack operations for memory renaming.