Runping Lai

This commit introduces **new capabilities** to the **Wattson power estimation system** by adding support for **TPU (Tensor Processing Unit) v1 and v2 power metrics**. It extends the existing **Android power metrics** within the **Trace Processor** to include detailed TPU subsystem power consumption. This involves updating **protocol buffer definitions** in `protos/perfetto/metrics/android/wattson_in_time_period.proto` to accommodate new TPU estimate fields and modifying several **SQL metric scripts** to calculate and integrate `tpu_mw` into various power rail analyses. Specifically, `src/trace_processor/metrics/sql/android/wattson_rail_relations.sql` and `src/trace_processor/perfetto_sql/stdlib/wattson/aggregation.sql` are updated for TPU power calculations. The change provides a more comprehensive view of device power usage, allowing users to analyze **TPU power consumption** alongside CPU and GPU metrics.

This commit introduces a **new capability** to the **Wattson UI** by adding **TPU power estimate visualization**, mirroring the existing CPU and GPU power tracks. It extends the `Wattson` plugin's functionality to detect TPU support and dynamically render **TPU subsystem estimate tracks** within the UI. Specifically, the `onAreaChange` logic in `estimate_aggregator.ts` is updated to process these new tracks, while `index.ts` gains new functions like `hasWattsonTpuSupport` and `addWattsonTpuElements` to manage their activation and display. This enhancement provides users with a more comprehensive view of power consumption across various hardware components.

This commit **adds a new capability** to the **Wattson power estimation framework** within the `trace_processor` `stdlib`, introducing a **multi-dimensional TPU power model**. This model estimates **TPU power consumption** by integrating frequency, cluster, and parallel request concurrency, utilizing a non-linear power curve lookup table. The **feature addition** involves extracting TPU frequency and request states from trace events, which are then used to provide continuous TPU power estimates for the UI and enhance the overall `_windowed_system_state_mw` calculations. This significantly expands Wattson's ability to analyze power usage for TPU-accelerated workloads.

This commit introduces the **Wattson Metric v2 implementation**, a **new capability** that significantly refactors and enhances power attribution analysis within the **Perfetto trace processor's SQL stdlib**. Essential metric calculation logic, including detailed thread, rail, and base component power aggregations, has been moved into new `stdlib` SQL files like `aggregation.sql` and `windows.sql`, which define standardized analysis windows. This update flattens the previously nested metric proto structure, providing a more efficient and standardized way to consume power data and improving the overall **Wattson power analysis subsystem**.

This commit **adds PowerVR GPU power estimation support** to the **`wattson` module** within **`perfetto_sql`**. It introduces new power curve data for the **'Tensor G5' GPU** and generalizes the GPU power state inference logic, including the `_gpu_power_state` view, to accommodate multiple GPU architectures. This **new capability** also involves a **refactoring** of the UI's platform validation logic, specifically the `hasWattsonGpuSupport` function, to correctly identify supported GPU types. The changes enable more comprehensive power analysis for a broader range of devices by extending `wattson`'s capabilities.

This commit **integrates a new `power/gpu_power_state` ftrace tracepoint** into the **Perfetto tracing system**, enabling the collection of PowerVR GPU power states (ON/OFF/PG). It introduces the `GpuPowerStateFtraceEvent` proto definition and updates the `traced` component to recognize this event. Crucially, the **`trace_processor`** now includes dedicated parsing logic, specifically the `ParseGpuPowerState` function, to correctly interpret these new events. This **new capability** significantly enhances debugging and analysis of GPU power consumption and state transitions within Perfetto traces.

This commit **refactors** the **Wattson plugin**'s warning logic to improve user experience and address issues from a previous revert. It now ensures that **Wattson warnings** and blank tracks are only displayed if the user has explicitly enabled Wattson Markers, preventing unsolicited UI clutter for non-Wattson users. The work involves moving warning-related functions into a new `ui/src/plugins/org.kernel.Wattson/warning.ts` file, centralizing checks for sufficient CPU trace configurations for `power/cpu_frequency` and `devfreq/devfreq_frequency` data. This **maintenance** and **bug fix** significantly improves the **UI experience** by making Wattson features opt-in and more targeted.

This commit **refactors** the **Wattson performance analysis module** within the `stdlib` by **removing hardcoded GPU IDs** from its GPU frequency calculations. It introduces a new Perfetto table, `_gpuid_map`, in `src/trace_processor/perfetto_sql/stdlib/wattson/device_infos.sql` to store device-specific mappings from device names to GPU IDs. The `src/trace_processor/perfetto_sql/stdlib/wattson/gpu/freq_idle.sql` script is updated to dynamically determine the `gpu_id` using this new map, rather than a static value. This **maintenance improvement** enhances the **accuracy and portability of GPU frequency analysis** across different devices.

This commit **updates the Perfetto SQL documentation** to include a newly identified best practice for using `MACRO`s. It adds a **new recommended use case** for defining simple, frequently referenced constants with `MACRO`s within `docs/analysis/perfetto-sql-syntax.md`. This **maintenance** change clarifies the performance implications of choosing between `MACRO`s and `FUNCTION`s for scalar SQL values, emphasizing that `MACRO`s prevent significant overhead in **performance-sensitive queries** due to their inlining capabilities. The update helps users optimize their SQL by avoiding the performance penalties associated with non-inlinable `FUNCTION`s for constants.

This commit **refactors** the `trace_processor`'s SQL standard library by **replacing the magic number `255`** with a formal, named function `_dsu_dep()` for representing DSU CPU dependency. This **refactoring** effort defines `_dsu_dep()` within `wattson/utils.sql` and integrates it across several **`wattson` curve definitions**, including `device_cpu_2d`, `device_l3`, `tg5_cpu_2d`, and `tg5_l3`. The change improves code clarity and maintainability by centralizing the definition of this dependency. Future code should now explicitly reference `_dsu_dep()` instead of using the opaque numerical value, enhancing the **readability and robustness** of the `wattson` module.

This commit **refactors** the **Wattson plugin** within the Perfetto UI, separating the logic for generating and displaying warnings. The configuration checking logic has been extracted from `index.ts` into a new, dedicated file, `warning.ts`, improving code organization and extensibility. This **enhancement** introduces new checks for `atrace_categories` and `predefined cpu_idle/freq_ms`, leading to more comprehensive and accurate Wattson warnings regarding CPU configuration. Additionally, the **CPU track** is now consistently displayed, ensuring users always see relevant warning messages and improving the diagnostic experience.

This commit introduces **warning messages** within the **Perfetto UI** to inform users when their trace configuration is missing essential `ftrace_events` for the **Wattson plugin** to display CPU track data. This **new capability** significantly improves the **user experience** by preventing confusion when Wattson tracks appear empty due to incomplete trace setups. Specifically, the `ui/src/plugins/org.kernel.Wattson` module now checks for sufficient CPU trace configurations using `hasWattsonSufficientCPUConfigs` and modifies the display of individual **CPU tracks** via `addWattsonCpuElements` to show a list of missing events. Users will see a red warning message, generated by comparing configured events against a minimum set, when hovering over the "help" icon on affected tracks, guiding them to correct their trace configurations.

This commit **uniforms the displayed columns** for **Wattson Power Estimates** at the **package-level** within the **Perfetto UI**. It **adds missing columns** such as `Idle transitions overhead`, `Total energy`, and `% of total energy` to align with thread and process-level views, while also removing an extraneous `Total Duration` column. Furthermore, this **bug fix** addresses a discrepancy in `Total Energy` values caused by a `NULL` UID for the "swapper" thread, ensuring **accurate and consistent power estimation data** across all aggregation levels. The changes primarily affect the **Wattson package-level aggregation logic** within `ui/src/plugins/org.kernel.Wattson/package_aggregator.ts`. This work significantly improves the **consistency and reliability of power consumption metrics** presented to users.