High-level floating-point synthesis

Abstract

MOODS (Multiple Objective Optimisation in Data and control path Synthesis) is a high-level synthesis system which provides the ability to synthesise a system level behavioural description into a structural representation. The thesis represents an enhancement to the original MOODS system to allow the designer to manipulate floating-point and complex variables on an equal footing with all other data types; the additional complexities arising from floating-point manipulation are completely hidden from the user.

Originally, the data processed by MOODS was fixed (occasionally variable) width integers, and the functional units available were relatively unsophisticated (adders, subtractors, multipliers, multiplexers and so on). The floating-point synthesis system described here provides a library of high-level floating-point functions (trigonometric, transcendental, and complex) to support the synthesis of behavioural designs incorporating floating-point operations.

The floating-point library components themselves are implemented using a number of base techniques, namely table lookup, the CORDIC algorithm, and iterative series. Decisions about the mapping of base techniques onto functional units are left to a floating-point optimiser, which makes individual binding choices based on global knowledge of the overall design, allowing the internal sub-structures of these units to be shared which results in a dramatic decrease in the overall hardware resources required to implement the design.

Finally, an exemplar is designed and analysed in detail: a cubic equation solver synthesised using the floating-point capability integrated within the MOODS environment.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata