Abstract: A system and method for multi-level iterative filtering of a data structure, e.g., an image, wherein elements of the data structure form the zero layer in the zero level and the data layer in each subsequent level is given by the results of one iteration. First, the method of the present system includes subdividing each level into a plurality of regions, there being data dependency between the data in one data layer in one level and the data layers in any other level of a region. Second, the method includes filtering each level by lapped-region processing. Lastly., the method includes scheduling the data processing of each level to provide substantially regional synchronization of the filtering at each level. In one embodiment, the sequence for traversing the regions is selected so that outputs from processing the regions are scheduled to occur at substantially equal time intervals.

Abstract: Selected code is modeled in a polyhedral dependency graph (PDG). A placement optimizer maps each element of the PDG to an optimally placed PDG. An ordering optimizer maps the placed PDG to an optimally ordered PDG. The PDG, place PDG, and ordered PDG are combined to produce a transformation script. The transformation script is applied to the selected specification description to produce optimized selected code. Optimized selected code is combined with original code to generate a control-flow optimized code. In addition, memory directives are derived from the ordered PDG model. The memory directives and optimized code are used to generate target code for simulation or software compilation.