Inlining of mathematical functions in HP-UX for Itanium® 2

HP-UX compilers inline mathematical functions for Itanium processor family (IPF) systems to improve throughput 4X-8X versus external library calls, achieving speeds comparable to highly tuned vector functions, without requiring the user to code for a vector interface and without sacrificing accuracy or edge-case behaviors. This paper highlights IPF architectural features that support implementation of high-performance, high-quality mathematics functions for inlining. It discusses strategies for utilizing the features and developing inlineable sequences on a large scale, and it presents requisite compiler features and language extensions. Also, this paper describes compiler mechanisms that produce inlineable code and inline it.