Quantitative selection of media benchmarks

Over the last decade, significant advances have been made in compilation technology and microprocessor architectures for capitalizing on instruction-level parallelism (ILP). While most of the processors containing superscalar, VLIW and SIMD structures that are well matched to the needs and capabilities of the ILP compilers are targeted at embedded applications, the majority of all ILP compilation work has been conducted in the context of general-purpose computing. As a consequence, there currently exists a gap between the compiler community and embedded application developers. We present a quantitative benchmark selection and validation technique to close the gap. It is based on more quantitatively rigorous metrics to accurately measure usefulness and effectiveness of benchmarks. We assemble a collection of DSP and communication applications written in a high-level language. An experimental selection of benchmarks from the collected applications is conducted and its applicability and usefulness are verified through an application specific system synthesis.