Open Quuxplusone opened 4 years ago
Implement the lowering of parallel loops to gpu.launch operations. The mapping of loop dimensions to hardware indices should be based on attributes on the parallel loop to decouple mapping decisions from actual code generation.
I started this in https://reviews.llvm.org/D73893.
Is this still relevant or is it done?
Implement the lowering of parallel loops to gpu.launch operations. The mapping of loop dimensions to hardware indices should be based on attributes on the parallel loop to decouple mapping decisions from actual code generation.
I started this in https://reviews.llvm.org/D73893.