A polymorphous computing fabric
Christophe Wolinski, Maya B. Gokhale, K. McCave · IEEE Micro · 2002
We propose a polymorphous computing fabric-based system (FBS) well suited to digital signal processing (DSP) and image processing applications. We have implemented our design on a system on a programmable chip (SoPC). The fabric's highly parameterized cellular architecture enables customized synthesis of fabric instances to achieve high performance for different classes of applications. The system's innovative global memory provides a host control processor with random access to all the variables and instructions on the fabric. The fabric supports several computing models including multiple instruction, multiple data (MIMD); single program, multiple data (SPMD); and systolic flow and permits dynamic reconfiguration of communication patterns. To illustrate the capabilities of our approach, we present two fabric instances with implementations of representative applications including a k-means clustering algorithm, a bank of finite impulse response (FIR) filters, an N-tap FIR filter (N is the number of taps of the filter), and a vector-by-matrix multiplication. Each fabric instance holds 52 cells on the Altera Excalibur ARM embedded processor system.