20 February 2025: Training Gray Scott Thursdays : Computing precision (NCC France), online

We compute, ideally fast with simple objects (we call them ‘primitive types’) : floating point numbers. Float or double, some languages call them Real. This webinar will highlight some surprising aspects of their behaviour. These numbers which can be represented as floating point and induced errors or the other. Moreover, defferent use cases will be presented to spot traps of floating point computing, and find the perfect balance between performance, precision and correction. An important point is the difference of performance which appears on several hardwares when exotic numbers appears in the computation such as denormals, NaN, inf. By Vincent Lafage & Hadrien Grasland, IJCLab – Online information & registration