Design, Implementation and Evaluation of Built-in Functions on Parallel Programming Model in SMYLE OpenCL

In this paper, we propose built-in functions on parallel programming model in SMYLE OpenCL to extend the original OpenCL semantics giving our system's original limitation and interpretation for embedded many-core architecture. On a platform using FPGA to evaluate embedded many-core architecture SMYLEref, data parallel and task parallel programming models supported by the OpenCL execution model are implemented. Then, high-level API based on an OpenCL framework, named SMYLE OpenCL, has been developed. Math functions, integer functions, common functions, geometric functions, relational functions, and barrier function of synchronization functions in built-in functions are implemented for OpenCL version 1.2. And the routines for floating point emulation are developed in order to compute the ranges of float and double on the device core which OS is not installed in. This paper describes the design and its implementation of built-in functions on parallel programming model in SMYLE OpenCL, and shows how to solve the implementation issues. The high-speed low-power performance using parallel benchmark applications compared with a former technology is demonstrated.