DigiNews

Summary

Bruce Dawson's article explores how intermediate floating-point precision is determined by compiler, architecture, and runtime settings. It shows that the same code can run with float, double, or extended precision depending on flags like /arch:SSE2, x87 control words, and the platform (x86 vs x64), with implications for performance and results. The piece offers practical guidance for developers who need deterministic, portable results or optimized performance.