A framework for high level synthesis of transcendental functions

High level synthesis using C/C++ code of applications is rapidly gaining ground. However, support for calculations is restricted to elementary algebraic operations of addition, multiplication, subtraction and division. Support for transcendental functions is generally unavailable and is inefficient where available. Transcendental functions are an important part of high performance computing. A framework to support their high level synthesis is presented here. The framework encompasses parallelism extraction, bounds, bit-width and precision analysis, and use of efficient low level implementations for high level synthesis. A method to account for latency and data sampling time due to drop-in low level modules is also demonstrated. High level synthesis of a computational kernel from GPS application involving transcendental functions is demonstrated.