How accurate we are? A refresher on the floating-point computations and the standard library.

Boguslaw Cyganek

60 minute presentation

In my talk I’ll present the basics of floating-point computations and how to avoid some pitfalls such as under- and overflow, the catastrophic cancellations, and how the NaN can pop up. I’ll also talk if the std::accumulate is good for all accumulations and what can we gain launching std::transform_reduce with the std::execution::par? Then, what are the most accurate and the fastest summation methods and how easy it is to use them in our code. I’ll also touch on such topics as accurate computation of the inner product of vectors and the fast matrix multiplication. Everybody is welcome.

std::cout << "See my talk " << ( 0.1+0.2==0.3 ? "absolutely" : "& all are welcome" );

numerical computations
compensated summation
parallel systems

Boguslaw Cyganek

I'm a researcher and lecturer at the Department of Electronics, AGH University of Science and Technology in Cracow, Poland. I worked as a software engineer for a number of companies in USA, UK, Germany and Poland. For the recent years I have been conducting research in AI and teaching C++ at the AGH University. Currently I'm finishing the book entitled “Introduction to Programming with C++ for Engineers”.