Patent · US Expired

Method and apparatus for providing high numerical accuracy with packed multiply-add or multiply-subtract operations

US5862067A · kind A · utility

61Cited by
16References
24Claims
0Family size

Assignee

Inventors

Key dates

Filing dateDec 29, 1995
Grant dateJan 19, 1999
Priority date
Expiry dateDec 29, 2015

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG06F2207/3828
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

A method and apparatus for combining at least two packed multiply-accumulate instructions (or equivalent operations) to compute a filter result from coefficients having more bits than can be processed by a single multiply-accumulate instruction (or equivalent operation). This achieves greater accuracy in computing transforms and digital filters without requiring more expensive hardware to implement multiply-accumulate instructions for larger operands. Typical applications are compression/decompression algorithms, modem, audio and video. The invention is scalable, permitting additional multiply-accumulate instructions to be added as incrementally larger coefficients are required. Additionally, the invention permits different numbers of multiply-accumulate instructions for each coefficient depending on the sensitivity to that particular coefficient.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.