Code Generation and Scheduling for Parallelization of Multi-Dimensional Perfectly Nested Loops

Tiling transformation is widely used to efficiently execute loops on parallel architectures. Automatic code generation for tiled loops can be a very complicated compiler work due to non-rectangular tile shapes and arbitrary iteration space bounds. A wave-front approach can simplify code generation for tiled loops. In this paper wave-fronts are applied to generate code for multi-dimensional irregular tiled loops. Also, an improved block scheduling strategy is presented to exploit wave-front parallelism for multi-dimensional tiled spaces. Experimental results show that the proposed scheduling strategy outperforms likely most currently implemented strategies.

Keywords:

Loop Parallelization , Wave-Front , Code Generation , Scheduling