Lower and upper bounds on the switching activity in scheduled data flow graphs

Abstract

Kruse@OFFIS.Uni-Oldenburg.DE 1. ABSTRACT In this paper we present an approach to calculate lower and upper bounds for the switching activity in scheduled data flow graphs. The technique can be used to prune the design space in high level synthesis for low power before allocation and binding of functional units and registers. The low power allocation and binding problem is formulated. It is shown that this problem can be relaxed to the bipartite weighted matching problem which is solvable in O(n3) where n is the number of functional units or registers, respectively. The application of the technique on benchmarks shows the tightness of the bounds. Most of the investigated bounds were less than 1% off the minimum respectively maximum solutions.

Lower bound estimation for low power high-level synthesis

“…Lower and upper bound estimation procedures are given for scheduled DFGs without resource constraints. This paper extends the work of [10] by improving the bounds if the number of resources is constrained. Some researchers have addressed HLS for low power problems that are closely related to our work.…”

“…In [9] a technique is described that estimates a lower bound on the performance of schedules from a DFG with resource constraints. To the best of our knowledge, LBE techniques for low power at the higher levels of abstraction are first addressed in [10]. Lower and upper bound estimation procedures are given for scheduled DFGs without resource constraints.…”

“…Its calculation follows the approach described in [10] that originates from the work presented in [2,3,6]. A square power cost matrix (PCM) for the variables and for each operation type present in the DFG (e.g.…”

“…We first repeat Theorem 4.1 without proof from [10] which defines a lower bound on the power consumption of the low power allocation and binding problem, i.e. the binding problem without resource constraints: is associated with arc .…”

“…As stated in [10] the ILP of Theorem 4.1 can be efficiently solved by the Hungarian Method in because it describes the Assignment Problem [11].…”