A hardware/software codesign approach for programmable IO devices

Abstract

This work presents a new HW/SW codesign approach for programmable controllers (ICs or IPs) managing IO devices. A novel cost function is defined to evaluate a device driver's performance and formulate its exact minimization as a zero-one integer linear programming problem. The cost function is used to explore the design trade-off between the code performance of a device driver and the number of programmable registers in an IO controller. This work proposes a heuristic algorithm based on the cost function, which iteratively minimizes the software cost until the hardware constraint is violated. The proposed algorithm was implemented in C language and evaluated with a set of real devices. Compared with original designs, the proposed approach can obtain design alternatives that reduce both software and hardware costs.

Optimal Allocation of I/O Device Parameters in Hardware and Software Codesign Methodology

“…To the best of our knowledge, the paper [5] should be the first work to investigate how the parameter allocation can affect the performance of drivers and to try finding an optimal allocation. However, it does not address hardware optimization.…”

“…The strategy is applied to both software and hardware optimization. Each refinement process relies on proven lemmas, which have been presented in [5]. To show these lemmas, a new notation is defined as follows: The lemmas exploited in our approach are described in the followings, whose proofs can be directly derived by subtracting C s (Φ) from C s (Φ').…”