Approximate Buffers for Reducing Memory Requirements: Case Study on SKA

Abstract

The memory requirements of digital signal processing and multimedia applications have grown steadily over the last several decades. From embedded systems to supercomputers, the design of computing platforms involves a balance between processing elements and memory sizes to avoid the memory wall. This paper presents an algorithm based on both dataflow and approximate computing approaches in order to find a good balance between the memory requirements of an application and the quality of the result. The designer of the computing system can use these evaluations early in the design process to make hardware and software design decisions. The proposed method does not require any modification in the algorithm's computations, but optimizes how data are fetched from and written to memory. We show in this paper how the proposed algorithm saves 23.4% of memory for the full SKA SDP signal processing computing pipeline, and up to 68.75% for a wavelet transform in embedded systems.

Design Space Exploration for Memory-Oriented Approximate Computing Techniques

“…The DSE method presented in this paper automates and extends the AxB technique presented in [12], adding the support of representations in addition to the FxP format.…”

“…Our generic DSE algorithm is designed to minimize the memory footprint of an application, while maintaining the result above a specified threshold of a desired quality metric. Our DSE relies on the AxB concept detailed in [12], allowing data to be represented and stored in memory on an arbitrary number of bits, without the usual 2 n constraints. Alternative data representations from Section II-A are added within AxBs, requiring no additional modification of the application original source code, aside from the initial modification needed for AxBs usage.…”

“…In this paper, we propose DSE method to ease the implementation of a precision optimization technique aiming to reduce the memory footprint of an application while respecting a user-defined quality constraint. Our method, relying on [12], is fast enough to be integrated into the design stream of an application.…”

“…The Approximate Buffer (AxB) technique presented in [12] consists in concatenating data into buffers to reduce the memory footprint. Data are represented using the Fixed-Point (FxP) format with an arbitrary bit-width, with every data from an AxB using the same format.…”