Approximate Computing with Runtime Code Generation on Resource-Constrained Embedded Devices

Damien Couroussé (1) , Caroline Quéva (1) , Henri-Pierre Charles (1)

1 CEA-TECH-Reg - CEA Tech en régions

—Approximate computing systems aim at slightly reducing the output quality of service, or precision, of a program in order to save computing operations, reduce the execution time and the energy consumption of the system. However, to the best of our knowledge, in all the approximate computing systems presented in the research literature, the implementation of the components that support the approximation is left to the developer. In this paper, we describe the implementation of a precision-aware computing library that saves the developer from the implementation of approximated functions. Efficient implementations of the approximated functions are achieved with runtime code generation. Our implementation of runtime code generation is fast and memory-lightweight, and its overhead can is amortised in a few executions of the generated code. We illustrate the performance and the lightness of our implementation on the WisMote, a MSP430-based platform with only 16 kB of RAM and 256 kB of flash memory. When the generated code is specialised on one of the input arguments of the approximated function, we achieve a speedup above 7×.

Keywords: Wireless Sensor Nework; Index Terms—approximate; runtime code generation; com-pilation; code specialization; precision; floating-point

Domains: Computer Science [cs]; Performance [cs.PF]; Software Engineering [cs.SE]

Damien Couroussé, Caroline Quéva, Henri-Pierre Charles. Approximate Computing with Runtime Code Generation on Resource-Constrained Embedded Devices. 2nd Workshop On Approximate Computing (WAPCO 2016), Jan 2016, Vienna, Austria. ⟨cea-01296569⟩