Benchmarking Jetpack Compose

Introduction to Compose Benchmarking

Benchmarking Jetpack Compose involves specific considerations compared to the View system. In microbenchmarks we generally track the three phases of rendering: Recomposition, Layout, and Draw, but in macrobenchmarks we look more at metrics like frame duration and jank.

It is important to understand the difference between the two types of benchmarks we use:

• Macrobenchmarks: Measure high-level app interactions (e.g., app startup or scrolling a list). They test the full out-of-process app performance.
• Microbenchmarks: Measure specific function calls in a tight loop (e.g., recomposition of a specific UI component).

For a core explanation of the differences between Microbenchmarks and Macrobenchmarks, see the AndroidX Benchmarking Overview.

Standard Reference Device: Oriole (Pixel 6)

When running benchmarks for Compose to establish baselines or test regressions, Oriole (Pixel 6) is the standard reference device.

• It provides a consistent baseline for our measurements and is representative of modern device topologies.
• For hardware setup instructions to reduce variance (e.g., locking clocks, disabling JIT), please refer to the AndroidX Microbenchmarking Guide.

Interpreting Results and Metrics

When reviewing benchmark output, you will typically focus on metrics such as Time to Initial Display (TTID), Time to Full Display (TTFD), Jank Rate and frame duration.

For a breakdown of Compose performance metrics, how they are calculated, and what they signify, please read our dedicated Compose Metrics Overview.

Skia Perf Dashboards

You can view CI results for Compose benchmarks on the AndroidX Skia Perf Dashboard. See the Performance Monitoring Guide for how to filter and view results.