Publications

This document describes a series of kernel benchmarks for the PCA program. Each kernel benchmark is an operation of importance to DoD sensor applications making use of a PCA architecture. Many of these operations are a part of the composite example applications described elsewhere. The kernel-level benchmarks have been chosen to stress both computation and communication aspects of the architecture. "Computation" aspects include floating-point and integer performance, as well as the memory hierarchy, while the "communication" aspects include the network, the memory hierarchy, and the I/O capabilities. The particular benchmarks chosen are based on the frequency of their use in current and future applications. They are drawn from the areas of signal processing, communication, and information and knowledge processing. The specification of the benchmarks in this document is meant to be high-level and largely independent of the implementation.

Real-time signal processing consumes the majority of the world's computing power. Increasingly, programmable parallel processors are used to address a wide variety of signal processing applications (e.g., scientific, video, wireless, medical, communication, encoding, radar, sonar, and imaging). In programmable systems, the major challenge is no longer hardware but software. Specifically, the key technical hurdle lies in allowing the user to write programs at high level, while still achieving performance and preserving the portability of the code across parallel computing hardware platforms. The Parallel Vector, Signal, and Image Processing Library (Parallel VSIPL++) addresses this hurdle by providing high-level C++ array constructs, a simple mechanism for mapping data and functions onto parallel hardware, and a community-defined portable interface. This paper presents an overview of the Parallel VSIPL++ standard as well as a deeper description of the technical foundations and expected performance of the library. Parallel VSIPL++ supports adaptive optimization at many levels. The C++ arrays are designed to support automatic hardware specialization by the compiler. The computation objects (e.g., fast Fourier transforms) are built with explicit setup and run stages to allow for runtime optimization. Parallel arrays and functions in Parallel VSIPL++ also support explicit setup and run stages, which are used to accelerate communication operations. The parallel mapping mechanism provides an external interface that allows optimal mappings to be generated offline and read into the system at runtime. Finally, the standard has been developed in collaboration with high performance embedded computing vendors and is compatible with their proprietary approaches to achieving performance.

The Vector/Signal/Image Processing Library (VSIPL) is a DARPA initiated effort made up of industry, government and academic representatives who have defined an industry standard API for vector, signal, and image processing primitives for real-time signal processing on high performance systems. VSIPL supports a wide range of data types (int, float, complex, ...) and layouts (vectors, matrices and tensors) and is ideal for astronomical data processing. The VSIPL API is intended to serve as an open, vendor-neutral, industry standard interface. The object-based VSIPL API abstracts the memory architecture of the underlying machine by using the concept of memory blocks and views. Early experiments with VSIPL code conversions have been carried out by the High Performance Computing Program team at the UCSD. Commercially, several major vendors of signal processors are actively developing implementations. VSIPL has also been explicitly required as part of a recent Rome Labs teraflop procurement. This poster presents the VSIPL API, its functionality and the status of various implementations.