Google is introducing a new optimization technology to the Android kernel to make the system run faster by analyzing how users actually use their phones. This technology, called Automatic Feedback-Directed Optimization (AutoFDO), will bring faster application startup, smoother performance experience, and longer battery life to Android devices.
What is AutoFDO?
During the standard software compilation process, the compiler makes thousands of tiny decisions based on static code hints, such as whether to embed a certain function and which branch of a conditional judgment is likely to be executed more often. However, these heuristics do not always accurately predict code execution behavior during actual phone use.
AutoFDO changes this by guiding the compiler through real-world execution models. These patterns represent the most common instruction execution paths in actual use of the program code, and are collected by recording the branch history of the CPU. Google tested Pixel phones in a controlled lab environment, simulating the use of the top 100 most popular Android apps, while using profiling tools to track the most commonly used parts of the core. The system identifies which code is “hot” (frequently used) and which is “cold” (rarely used).
Why optimize the core?
The Android core is one of the most important parts of the operating system. It sits at the heart of Android and manages communication between the CPU, applications, and hardware. According to Google, cores take up roughly 40% of an Android device’s CPU time, so even small improvements can have a noticeable impact on real-world performance.
How much is the actual performance improvement?
Google says it has seen “impressive improvements” in key Android core metrics. According to Google’s official blog, after the Android system introduces AutoFDO (Automatic Feedback-Directed Optimization, automatic feedback-oriented optimization) technology, the “performance improvement percentage” obtained by various system indicators is as follows:
-
Binder-rpc (21.7%)、Hwbinder (20%) and Binder Transactions (addints) (12.3%): Binder is the core “inter-process communication” (IPC) mechanism in the Android system. These data show that the speed of transferring data and instructions between different services, hardware layers and applications within the system has been significantly improved by 12% to nearly 22%.
-
System Call Efficiency (bionic mmap) (9.3%): Improved system call efficiency. It is particularly pointed out here that Android’s basic C library (Bionic) has improved efficiency by 9.3% when processing memory mapping (mmap) operations, which is very helpful for the overall memory management of the system.
-
Cold App Launch Time (4.3%): Cold start application time (that is, the time when the application is loaded from scratch without being cached at all in the background) is accelerated by 4.3%. This means that when users click on the App icon, the waiting time for loading is shortened.
-
Boot Time (2.1%): Overall device boot time is reduced by 2.1%.
These figures are more than just theoretical improvements. They actually translate into a more responsive interface, faster application switching, extended battery life, and an overall more responsive device for the end user.
Google has applied AutoFDO in userspace to optimize native executables and libraries, achieving about a 4.3% improvement in cold application startup and a 2.1% reduction in boot time. Now, this technology extends further to the core level, bringing even more significant performance improvements.
How it works: The complete process
Google’s deployment strategy involves a complex process to ensure that profiles remain relevant and performance remains stable.
Step 1: Configure file collection
While Google relies on internal test equipment to analyze user-space binaries, they have moved to a controlled lab environment for the Generic Kernel Image (GKI). Decoupling analytics from device release cycles allows for flexible, on-the-fly updates independent of deployed core versions. Crucially, testing confirms that this lab-based data delivers comparable performance gains to real-world devices.
Google uses the top 100 most popular apps from the Android Application Compatibility Test Suite (C-Suite) to build representative workloads. To capture the most accurate data, they focus on:
- Application launch: Optimize for the most visible user latency
- Whole system monitoring: Captures not only front-end application activity, but also critical background workloads and cross-program communications
This synthetic workload shows 85% similarity to execution patterns collected from on-premises devices.
Step 2: Configure file processing
Google post-processes the raw trace data to ensure it is clean, valid, and ready to be fed to the compiler. They consolidate data from multiple test executions and devices into a single system view.
Step Three: Core Construction and Deployment
When rebuilding the core using these profiles, the compiler can make smarter optimization decisions based on actual Android workloads. This allows frequently used parts of the system to perform more efficiently.
When will it be updated?
Google is currently deploying core AutoFDO in the android16-6.12 and android15-6.6 branches: this is the version used by recent Android versions, with plans to extend it to future core versions.
Over time, Google hopes to apply these optimizations to more parts of the core, including additional system components and hardware drivers added by phone manufacturers, such as camera and modem drivers. This means that future Android devices may also have improvements in camera performance, network connectivity, and other hardware-related features.
What does it mean for users?
For the average Android user, the impact of this technology will likely be gradual but noticeable. While Google makes many performance improvements every year, AutoFDO represents a more fundamental approach that isn’t about patching specific problems, but rather about comprehensively re-optimizing the entire system based on how people actually use their phones. Over time, you may notice:
- Apps start faster
- Multitasking is smoother
- The overall system is more responsive
- Battery life could improve slightly
Another important aspect of this technology is that it does not require any user action or settings. All optimizations are performed automatically at the system level, and users will automatically benefit from the next update. It’s the latest in Google’s ongoing efforts to make Android faster and more efficient.
Source: KOCPC Chinese