E-Ink News Daily

Back to list

Double-double: 31 digits of precision without leaving the FPU

The article explains double-double arithmetic, a technique that combines two doubles to achieve ~31 decimal digits of precision without leaving the FPU, using error-free transformations. It benchmarks this approach against float128 and MPFR, showing it sits performance-wise between standard doubles and arbitrary-precision libraries, with a practical demonstration using deep Mandelbrot set rendering.

Background

Double-double arithmetic is a well-known technique in high-performance numerical computing, popularized byoshida and others, that provides extended precision using standard double operations. It is particularly useful in scientific computing applications requiring more accuracy than IEEE 754 double offers without the overhead of arbitrary-precision libraries.

Source
Lobsters
Published
Aug 20, 2026 at 11:31 PM
Score
6.0 / 10