Technical presentation - 30 minutes (including q&a)
Feedback Driven Optimizations are a set of toolchain techniques that use traces from program execution to improve the performance of output binaries. In this talk, we will discuss the support for AutoFDO available for native code (C/C++) in Android userspace and kernel, along with the performance impact we see on Google Pixel devices. Additionally, we'll outline our roadmap to make native profiling more broadly available - to use cases such as Android Apps, using post-link optimization techniques, and fleet-wide sampling. We will highlight some milestones required from the ecosystem that are necessary towards achieving this goal.
Pirama Arumuga Nainar leads the Android LLVM Team at Google with interests spanning compiler optimizations, code health, and developer tools. During his time at Google, Pirama has been part of the Android LLVM team with the mission of bringing the performant and modern tools from the Clang/LLVM ecosystem to Android. Pirama has a PhD from the University of Wisconsin-Madison where he worked on software fault isolation using statistical analysis of field traces. Prior to Google, Pirama worked as a Performance Engineer at Oracle.