Double-double: 31 digits of precision without leaving the FPU
Summary
Technical blog post by Marek Fiser explains double-double arithmetic, which combines two doubles to achieve about 31 decimal digits of precision without heap allocation. It covers error-free transformations (twoSum, quickTwoSum, twoProduct), benchmarks against MPFR, and the practical limits of the approach in fractal rendering. The piece also discusses when this technique is advantageous and when a real arbitrary-precision library is preferable.