Integrated Programmable Compiler Optimization For Scientific Applications


DOE/LLNL subcontract B607283

  • Total amount $53,793
  • Period : 2/1/14 - 12/31/15
  • Priciple investigators: Qing Yi
  • co-investigators: none
  • Students supported: TBA
  • Project Summary : This research aims to extend the ROSE C/C++/Fortran source-to-source compiler infrastructure led by Dr. Quinlan within DOE/LLNL with a scripting language interface to support programmable control, flexible parameterization, and easy customization of its optimization analysis and transformations. Specifically, the project will seek to unify the internal AST (Abstract Syntax Tree) representation of the ROSE compiler with that of POET, an interpreted program transformation language designed and implemented by Dr. Yi, so that POET can serve as an external programming interface for specifying and composing specialized program analysis and optimizations within ROSE, and any compiler/translator implemented within ROSE can invoke the large collection of loop optimizations within the POET library with automated tuning support and can utilize the special language support for flexible composition of transformations and domain-specific code generation within POET. This project essentially embeds the POET language within the ROSE compiler infrastructure, so that ROSE translators can be written in POET combined with C/C++, and a scripting interface can be used to directly invoke the large collection of internal program analyses and optimizations within ROSE.