Date: 2025-05-15
Time: 15:00–15:50
Room: Breakout Room 2
Level: Advanced
In the past 20 years, year-on-year CPU clock speed increases have slowed. Performance increases now mainly come from doing more work per clock cycle. How much work a given CPU can do per clock cycle does not only depend on the CPU's design, it depends on the code that it's executing.
This talk will take the audience through the world of modern superscalar CPUs and discuss some of the tricks they utilize to execute code as fast as possible, touching on subjects like branch prediction and cache prefetching. With each concept, we'll talk about possible adjustments that can be made to the code to make the CPU's job easier and allow it to execute the code faster.
Finally, the talk will touch on some tools that can be used to find performance bottlenecks so that people know where to focus their efforts in making improvements.