Think Unlimited and Compress Data Automatically

Developing an application which, when unoptimized, consumes more memory resources than physically or financially available demands a lot of expertise. In this work, we show that with the right tools and language abstractions, writing such programs for a given class of applications can stay within reach of non-expert developers. We explore the potential of a compiler-based data layout transformation from dense array to a compressed tree data structure. This transformation allows easy application prototyping, provides compression and carries information that can be used with more advanced optimization, e.g., adaptive and approximate computing techniques. We are primarily targeting partial differential equation solvers and signal processing applications. We evaluate the compression ratio and error originating from this compressed representation. We suggest multiple exploration paths to produce an automatic adaptive code transformation with compressing capabilities from the multiresolution information produced during the transformation.