Mixed-precision computing has the potential to significantly reduce the cost of exascale computations, but determining when and how to effectively implement it in our programs can be challenging. This talk will present a methodology for enabling mixed-precision with the help of computer arithmetic tools. By evaluating metrics across three dimensions: accuracy, time-to-solution, and energy-to-solution, the talk will illustrate how this methodology can be applied to real-world applications.
Yanxiang Chen (Tue,) studied this question.