Automatic code generation for executing tiled nested loops onto parallel architectures

This paper presents a novel approach for the problem of generating tiled code for nested for-loops using a tiling transformation. Tiling or supernode transformation has been widely used to improve locality in multi-level memory hierarchies as well as to efficiently execute loops onto non-uniform memory access architectures. However, automatic code generation for tiled loops can be a very complex compiler work due to non-rectangular tile shapes and iteration space bounds. Our method considerably enhances previous work on rewriting tiled loops by considering parallelepiped tiles and arbitrary iteration space shapes. The complexity of code generation for tiling transformation is now reduced to the complexity of code generation for any linear transformation. Experimental results which compare all so far presented approaches, show that the proposed approach for generating tiled code is significantly accelerated.