Low-Latency Stylus for Chrome OS

Alpha 1.0.3

Updated 5 August 2021

A low-latency stylus library for Android apps on Chrome OS. This library provides mechanisms to reduce the touch-to-draw latency on Chrome OS devices by using direct rendering and prediction.

A simple demo app is provided showing both the CPU and GPU-driven low-latency implementations as well as a side-by-side comparison for each implementation with a regular (not low-latency) canvas.

Please file bugs and features requests to help us improve the library as we move toward launch.

How it works

The low-latency stylus library achieves fast stylus-to-screen response by leveraging two mechanisms.

1. Direct-rendering - by rendering pen strokes directly through the hardware compositor, delays due to OS compositing can be avoided
2. Prediction - there will always be some latency in drawing response to screen response time, hardware limitations, and required OS functionality. The library attempts to compensate for this remaining latency using prediction - guessing where the next part of the stroke will be drawn based on the current speed and direction of the stroke - and can achieve what feels like “zero-latency”.

Getting the library

1. In your top-level build.gradle file, be sure you have google() as one of your maven repositories:

   repositories { 
       // low-latency libraries are stored in google's maven repository
       google() 
       mavenCentral() 
   }
   

2. Put the following in the dependencies section of the app-level build.gradle file:

   implementation 'com.google.chromeos:chromeos-lowlatencystylus:1.0.1
   

Using the library

First, decide if you would like to use the CPU-based version of the library or the GPU-based version.

• CPU: The CPU version is easier to implement but may not be suitable for apps with complex brushes and graphics requirements.
• GPU: The GPU version requires some OpenGL knowledge to use and offers some performance improvements over the CPU version and allows for more advanced brushes.

Demo applications and instructions are provided for both versions of the library.

NOTE: To use the library, you will need Chrome OS version M91 or greater and an Android runtime (ARC) version greater than 7316937. To check this, go to chrome://version in the browser and read the ARC line. You may need to change your device's update channel to the beta channel or dev channel to get this version.

CPU Library

The best way to get started using the CPU library is to examine the LowLatencyStylusDemo app. Import it into Android Studio and use the cpu package and the LowLatencyStylusActivityCpu activity as your reference. You may also want to reference the API documentation provided here.

Create your main canvas

Create a drawing canvas for your main drawing surface that will receive MotionEvents and show the completed drawing gestures. In the demo app, a reference class is provided called SampleInkView.

Create an InkOverlayView

Create an InkOverlayView that will sit on top of your main canvas. This is the view that will show the current, incomplete, drawing gesture as well as the predicted path of the current gesture. Correctly tuned prediction can give the appearance of zero-latency drawing.

Extend InkDelegate

Drawing gestures passed to an InkOverlayView will be processed via an InkDelegate. Extending the InkDelegate class allows you to control how the current drawing gesture and predicted gestures will be drawn. An example of how this is shown in the SampleInkDelegate class of the demo app.

Pass current gestures to the InkOverlayView

Your main canvas should receive MotionEvents, keep track of them, and show any completed drawing gestures. The MotionEvents for the current, incomplete, drawing gesture should be passed to the InkOverlayView using InkOverlayView.onTouch.

Commit to main canvas

When a stroke is finished or cancelled (like due to palm rejection), and lines representing the real completed strokes have been drawn to the main canvas, the InkOverlayView should be cleared. In the demo app, this is done in the onTouchEvent method of the SampleInkView class.

Using/disabling prediction

Set latency compensation level for the prediction algorithm in ms using setPredictionTarget on your InkOverlayView. You can disable prediction by setting this to 0. Use the Prediction slider in the demo app to tune the prediction value you want. Note: setting this value too high will result in noticeable over-prediction.

Testing the library

You should notice a decrease in latency on Chrome OS devices when using the library. Test it out for a given device using the demo application. You can test the difference with and without low-latency by using the side-by-side comparison Activity provided in the demo application.

Note: pressing [SPACE] will clear the canvas in the demo application.

GPU Library

The best way to get started using the GPU library is to examine the LowLatencyStylusDemo app. Import it into Android Studio and use the gpu package and the LowLatencyStylusActivityGpu activity as your reference. You may also want to reference the API documentation provided here.

Create your main canvas

The main low-latency canvas will be subclassed from GLInkSurfaceView. To set up this view:

1. Set the prediction target in milliseconds (or disable prediction by setting target to 0)
2. Set the GLInkRenderer to handle the rendering
3. Receive input events and queue them correctly to the GLInkRenderer

Input events are generally handled by looking for ACTION_DOWN, ACTION_MOVE, and ACTION_UP events in the onTouchEvent callback. Note that ACTION_MOVE can contain a list of intermediate points that have been batched by the OS’s input logic. See Batching in the Android documentation for more details.

To draw on the low-latency surface, MotionEvents must be passed down to the GLInkSurfaceView.onTouch method. This method will automatically request unbuffered dispatch of input events for you to ensure that data is received as fast as possible.

Note: If you are not directly passing down MotionEvents received from the system - for example if you are interpreting the events and constructing new ones before sending them to GLInkSurfaceView.onTouch - you should call View.requestUnbufferedDispatch manually every time you receive a system ACTION_DOWN event, in order to get the lowest input latency.

Set your GLInkRenderer

Your GLInkRenderer should handle your brushes/shaders as well as manage incremental damage regions to provide the best drawing performance. Reference the included SampleInkRenderer as a starting point.

To configure the renderer:

1. Manage OpenGL shaders for drawing brush strokes
2. Keep track of damaged regions
3. Handle surface related callbacks (clear, onSurfaceCreated, onSurfaceChanged)
4. Override beforeDraw to handle input events received from the GLInkSurfaceView to be correctly passed to shaders and to return the damage area
5. Override onDraw to execute the draw

Note: the damage Rect returned from beforeDraw will be passed to an internal glScissor call before the render call is made.

Brushes/Shaders

You will need to create OpenGL shaders for your different brushes to handle the actual rendering of brush strokes received from the GLInkRenderer. The provided BrushShader class gives an example of a simple line brush (GL_LINES) as well as a bitmap-based brush shader.

To try your own bitmap files with the demo, create a 100x100px bitmap file with 0% alpha for blank areas and non-0% alpha for areas to paint.

Note: currently, changing between the line shader and the brush shader will re-draw all previous strokes with the newly selected shader.

Using/disabling prediction

As with the CPU library, the setPredictionTargetMs method in GLInkSurfaceView allows you to set your desired level of prediction. You can disable prediction by setting this to 0.

View and projection matrices

In the demo app, the view and projection matrices are set to the identity matrix. The API allows you to change this. If you are using different matrices in your application, please let us know. We would like feedback on this part of the API.

Testing the library

You should notice a decrease in latency on Chrome OS devices when using the library. Test it out for a given device using the demo application. You can test the difference with and without low-latency by using the side-by-side comparison Activity provided in the demo application.

Note: pressing [SPACE] will clear the canvas in the demo application.

Known Issues

• Prior to Chrome OS version M93 and Android runtime version 7434780, the GPU prediction target was fixed at 25ms. Adjusting the slider in the demo app or using setPredictionTargetMs will have no effect on these versions. To check your OS and Android runtime (ARC) versions, go to chrome://version in the browser. GPU prediction should work correctly on Chrome OS M93+ and ARC versions >= 7434780.

• Display scaling: in order to leverage direct compositing, the user needs to have their Chrome OS display resolution set to the “default” value. With other settings, additional GPU calculations may be needed to scale the output which makes direct rendering not possible.

Feedback and Bugs

We appreciate you exercising this library and strongly value your feedback. Please file bugs and feature requests here on the issue tracker.

This is not an officially supported Google product

LICENSE

---

Copyright 2021 Google LLC

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.