Universal Numbers Library
  • What is Universal?
    • Why custom number systems?
    • Number System Hierarchy
    • Performance
    • Next steps
  • Guides
    • Universal Docker Image
    • Local install
    • Development and testing
  • Fundamentals
    • Number Systems
      • integer<nbits>
      • fixpnt<nbits, rbits>
      • cfloat<nbits, es>
      • posit<nbits, es>
      • unum<nbits,es>
      • lns<nbits, base>
      • decimal
      • rational
    • Precision
    • Accuracy
    • Reproducibility
  • Use Cases
    • For Application Developers
      • High-Performance Computing
      • Embedded Computing
    • For Computational Scientists
      • Numerical performance
      • Tailoring Precision and Dynamic Range to the application
    • For Computational Mathematics
      • Numerical Encodings
  • Extras
    • Keyboard Shortcuts
Powered by GitBook
On this page
  • Guides: Getting started
  • Guides: Fundamentals
  1. What is Universal?

Next steps

PreviousPerformanceNextUniversal Docker Image

Last updated 3 years ago

If you are an application or algorithm designer and you are working on software that is exhibiting a performance or energy-consumption bottleneck, you are primed to take advantage of Universal. It provides tools to identify these bottlenecks and gather information about possible solutions. By following one simple rule of software engineering you will enable your application to take advantage of custom hardware solutions.

The first rule of mixed-precision algorithm optimization:

Parameterize computational kernels by arithmetic type

This rule will decouple your bottleneck operator(s) to ad hoc arithmetic type decisions that will limit your software to take advantage of custom hardware accelerators that would offer your applications orders of magnitude performance benefits.

The rest of the documentation will guide you through leveraging Universal to amplify your application or service.

Guides: Getting started

Follow these guides to get started on the basics as quickly as possible:

Guides: Fundamentals

Gain a deeper understanding of Universal and mixed-precision algorithm optimization:

Universal Docker Image
Local install
Development and testing
Number Systems
Precision
Reproducibility