Locality optimization in a compiler for wireless applications

Abstract

A strong need exists now for compilers of embedded systems to find effective ways of optimizing series of loop-nests. This is especially so for applications wherein the majority of the memory references occurs in the form of multi-dimensional arrays, indexed primarily with linear functions of iterators and parameterized constants. One major reason is the emergence of the new wireless standards, e.g. 802.11n, WiMAX, Bluetooth, HIPERMAN, 3GPP-LTE and WiBro, where the codes are predominantly of the type described above. These standards provide high bitrate and mobility but are also extremely power and performance hungry. For an even wider commercial applicability of these standards it is important to heavily optimize their energy consumption, so as to increase the mobile battery life time. We propose a novel solution to the multiple loop-nest optimization problem by using the concept of propagating constraints, and by splitting the problem into an access and layout locality-optimization phase, instead of adhering to the traditional split of temporal and spatial locality optimization. Experiments show that our technique leads to 47.5% reduction in external memory accesses over state-of-the-art.