IntelLabs / ParallelAccelerator.jl

The ParallelAccelerator package, part of the High Performance Scripting project at Intel Labs
BSD 2-Clause "Simplified" License
294 stars 32 forks source link

Refactor Parallel IR to have optimization as its own pass #93

Open ninegua opened 8 years ago

ninegua commented 8 years ago

Right now the lowering from Domain IR to Parallel IR and optimizations at Parallel IR level are in the same pass, which has to deal with nested processing in a complicated manner. A refactoring to separate the lowering and optimization as their own passes would make the optimization pass easier to manage since all variables would already be in the same scope.