Automatic synthesis of systolic arrays from uniform recurrent equations

Abstract

We describe a systematic method for the design of systolic arrays. This method may be used for algorithms that can be expressed as a set of uniform recurrent equations over a convex set D of Cartesian coordinates. Most of the algorithms already considered for systolic implementation may be represented in this way. The methods consists of two steps: finding a timing-function for the computations that is compatible with the dependences introduced by the equations, then mapping the domain D onto another finite set of coordinates, each representing a processor of the systolic array, in such a way that concurrent computations are mapped onto different processors. The scheduling and mapping functions meet conditions that allow the full automation of the method. The method is exemplified on the convolution product and the matrix product.

Parallelizing HMMER for Hardware Acceleration on FPGAs

“…Proving that a schedule is conflict-free when each PE executes computations in a one dimensional domain is relatively straightforward (see [14]), but this is more involved for higher dimensional domains. To solve this problem, we use Darte et al [2] results on juggling schedules, which can be summarized as follows.…”

Acceleration of a content-based image-retrieval application on the RDISK cluster

“…Our approach is inspired from systolic-array design methodologies [16] and especially partitioning techniques. The parallelization methodology in itself is out of the scope…”

“…On the other hand, most recent processors now integrate SIMD instruction sets targeted to multimedia applications. These instruction sets operate on short fixed-point data types (8,16 or 32-bit). As an example, the MMX instruction set provides instructions that can perform up to 8×8-bit operations in parallel.…”

A unifying framework for systolic designs

“…expression (1) into a ne~form which consists of possibly many recurrences each characterized by constant data dependencies; non-constant data dependencies may only occur at the bOWldaries between the recurrences. For each such recurrence it will be possible to determine a linear time-space rransforrnation into a systolic array, by applying the transformational method described in [11,13]. We briefly review the tranformational method for Wliform recurrenceS.…”