Polyhedral Dataflow Graphs

This research proposes an intermediate compiler representation designed for optimization, with an emphasis on temporary storage reduction and execution schedule guided optimization decisions. The representation is expressed as a dataflow graph that includes computational statements and data mappings. The targeted applications include both regular and irregular scientific applications. This intermediate representation can be integrated into existing compiler infrastructures. A specification language describes the graph components, the transformations that can be applied, visual representation, and code generation into source, or another IR. The IR can be implemented in an existing compiler toolchain to optimize code for stencil applications, PDE solvers, or sparse matrix and tensor operations