Enabling the Next 100,000,000 Android Devices

1
Enabling the Next
100,000,000 Android Devices
The Android Software Ecosystem
ARM Android Ecosystem Seminar
Seth Bernsen
May 18
th
, 2011
Taiwan
2
Application
Developers
OEM Partners
Software
Partners
Semiconductor
Partners
Open Source
Community
Engagement Model for Differentiation
Common Foundation
Differentiated Products

Enable apps developers

Connecting OEM/ODM to
software solutions
providers

Enable software and
services vendors

Ensure robust support of
key applications and
capabilities

Organize and rallying the
Open Source community

Open Source engineering
Enabling the Software Ecosystem
3

Over 400,000+ new
application client devices
per day
400,000+ Android Devices Shipping
Per Day
Devices from phones and tablets to
netbook’s
and TV’s running Android on ARM Cortex
-
A Processors
Android represents software trends

Open Source

Application ecosystems

Stack built for services
Range of Premium Content & Services Promotes Platform
4
Anatomy of Google Android

Heavily leverages open source software on ARM

Built around browser and application framework
Linux Kernel
Android
Middleware
SKIA 2D
Graphics
Mobile
Browser
Webkit
V8 JavaScript
Engine
Mobile
AIR
Flash
LLVM
VP8
Dalvik
LLVM
Linux Kernel
Linux
Middleware
SKIA 2D
Graphics
Chrome
Browser
Webkit
V8 JavaScript
Engine
Chrome OS
X Windows
Flash
LLVM
VP8
Linux Kernel
Android
Middleware
SKIA 2D
Graphics
Chrome
Browser
Webkit
V8 JavaScript
Engine
Tablet
Flash
LLVM
LLVM
Dalvik
VP8
Linux Kernel
Android
Middleware
SKIA 2D
Graphics
Chrome
Browser
Webkit
V8 JavaScript
Engine
Google TV
Flash
LLVM
LLVM
Dalvik
VP8
Google

Application
framework

Web browser

JavaScript
engine

Flash

Graphics

System libraries

Linux kernel
5
Dalvik
Java Optimizations for ARM

Dalvik
Java VM for ARMv7

Interpreter support

Instructions/co
-
procs
for
interpreter

UXFB, TBB… VFP, NEON

Tracing JIT support

Focus on hot code paths

Increases performance by 2
-
5x

Minimal memory bloat

Thumb or Thumb
-
2 code
generation

Support for VFP or Soft FP
Dalvik
JIT Benchmark Results
Speedup relative to
Dalvik
Interpreter
JIT Total Memory Usage (in
kBytes
)
6
Native Development Kit (NDK) for ARM

NDK is a comprehensive tool kit to enable application
developers to write directly for the ARM processor

Enhancements in latest NDK r5(b)
Android applications can be
written in Java, native ARM
code, or a combination of
the two
ARM
Improvements

Improved
performance
and
code density
with GCC 4.4.3

Optimizations for Cortex
-
A9

Support for VFPv3
General
highlights

New
NativeActivity
feature
eliminates need to write Java

Addition of default C++ STL

New API’s

Input subsystem, sensor data

Windows, surface subsystem

OpenSL
ES Audio API

Access to APK graphics assets

EGL library to create and manage
OpenGL ES textures and services
7
NativeActivity
for Pure Native Apps

New class called
NativeActivity
enables creation of strictly C
and ASM level applications

Moves “main()” from Java to C

Apps still run in a
Dalvik
sandbox, but
no Java code required

Two methods

Modify the
NativeActivity
class

Full flexibility of main app
thread control

Use native app glue helper
library built on top of
NativeActivity
class

Provides ready made
infrastructure for threading
and blocking behavior
Example main function
written for native code
using NDKr5
8
0
1
1
2
2
3
3
4
4
5
Improvement x times
Benefit of Native Code
-
Scimark 2
Nexus One
Droid
0.0
20.0
40.0
60.0
80.0
100.0
120.0
Tegra A9
Nexus One
Motorola
Droid
MFLOPS
Linpack (scaled to 1GHz)
Dalvik VM
Nativec
2.9 x
2.4 X faster
ARM Native Code vs.
Dalvik

Applications that use ARM
code can result in up to 4x
faster performance than
Dalvik

Access to processor features,
such as ARMv7 NEON
Advanced SIMD instructions

Fast reuse of existing and
cross platform software IP,
including C/C++/
Asm
code
9
GAME
(sample)
ARM
Native Lib
Neon
Optimized
ARM Native Code Examples

Significant impact on a
range of applications

Games

Media Players

Browsers

VoIP clients

All 3D OpenGL ES
content leverages ARM
code
Some of the many applications that
benefit from ARM native code
10
Android Browsing Optimized for ARM

From 2.3/Gingerbread Google
integrates the Chrome browser

3
rd
party browsers like Firefox,
UCWeb
, Opera and many others

ARM working continuously to enable
the three main components of HTML5
JavaScript
CSS
HTML
High performance JavaScript
drives HTML and CSS
Cascading Style Sheets V3.0
change visualization on the fly
HTML5 new markup for
video, audio, graphics, location,
local storage and more
11
ARM JavaScript JIT Support

JavaScript acceleration, all projects accelerate JavaScript by
compiling to native ARM code

ARM optimizations yielded up to
5x+ improvement

ARM contributing optimizations to All projects

Continuing to invest and improve
Java
Optimal JIT
Code
Generation
ActionScript
Virtual Machine
Adobe Flash
Adobe AIR
JavaScript
Virtual Machine
Web Browser
Runtime
Firefox Browser
Fennec Mobile
Browser
Internet
Application
Runtime
Web
Applications
Runtime
Yahoo Widgets
ARM
Investment
JavaScript Project
Foundation
/Package
TraceMonkey
Mozilla
Squirrelfish Extreme
WebKit
V8
Google
Tamarin
nanoJIT
Adobe
Example:
downstream
impact of
nanoJIT
optimization
12

First released as part of Android 2.2 for ARM

VERY fast pure JIT generating native ARM code

Cortex
-
A9 @1GHz completes V8 test suite in <2
secs

Continuous improvement for ARM by both ARM and Google

V8 in Android 3.0 is
>
2x faster than in 2.3!
Google V8 JavaScript Improvements
0
200
400
600
800
1000
1200
1400
1600
1800
2000
Score
Richards
DeltaBlue
Crypto
RayTrace
EarlyBoyer
RegExp
Splay
Cortex
-
A9
-
2.3
Cortex
-
A9
-
3.0
Google V8 Benchmark Suite
-
Run on the same Cortex
-
A9 based
SoC
Google V8 Performance
Improvements for ARM
in Android 3.0
(Higher is Better)
13
Investments in HTML5 Markup

HTML5 introduces new DOM constructs

Audio, video, 2D and 3D graphics, etc…

Optimizing new HTML DOM constructs
for ARMv7 NEON
Construct
Enables
NEON Optimized
Packages
<canvas>
2D bitmap graphics
Cairo,
Skia
<
svg
>
2D vector graphics
Cairo,
Skia
<audio>
Ogg
/
Theora
ffmpeg
<video>
H.264,
TheorARM
,
VP8
WebM
, X.264,
ffmpeg
14
WebM
Open Media Project

Launched May of 2010 to address lack of open video
standard for the web,
WebM
includes:

VP8 video codec licensed under a BSD
-
style, royalty
-
free license

The
Vorbis
open source audio codec

A container format based on a subset of the
Matroska
media
container

Extensively ARM optimized

Over
10,000
lines of ARMv7
NEON code at launch

VP8 on Cortex
-
A9 achieves
better clock for clock
performance than both
Atom and MIPS
Source: ARM benchmarking of
opensource
libvpx
codec running on 400MHz Cortex
-
A9
testchip
with 200MHz DDR
15
WebM
Proven on ARM SMP

Recent
WebM
VP8 source updates
improve threading capability

~45% gain
on dual thread version
running on production ARM Cortex
-
A9
silicon

Decoder results on quad core
Cortex
-
A9 @500MHz

1 thread: 14,8 fps

2 threads: 21,5 fps

3 threads: 24 fps

4 threads: 26 fps
0
10
20
30
1
2
3
4
fps for
720p
1Kbps
0
5
10
15
20
25
One
thread
Two
threads
fps for
1Mbps
720p video
Shields 720p 1024kbps
stream with 2 partitions
Source: ARM benchmarking of
opensource
libvpx
codec running on 500MHz Cortex
-
A9
testchip
with 200MHz DDR
16
Roadmap for CSS and SMP

CSS acceleration with NEON and Mali

Gradients, shadows, opacity, reflection

Hue/saturation/luminance

Transitions and transforms

Animation

Further SMP optimization work on browsers

Collaborating with Mozilla and Szeged

Mozilla Electrolysis

Moving key elements into separate processes
and threads

Video, HTML5 parsing, I/O, JavaScript garbage
collection, graphics compositing
HTML5 JavaScript
Web Workers
(Threads)
enhanced by ARM SMP
TODAY
17
Bringing Flash to Android
Flash
10
Software Rendering
Open GLES 2.0 Open Max IL
Tamarin Central JIT
Codec Optimization
>100 Man years effort
porting and optimizing
Focus on:

Performance Maximization

Power Optimization

Memory Conservation
“It’s great to see that our extensive
technical collaboration on this with
our partners is now becoming a reality.”
-
Paul
Betlem
, Sr. Director
Flash Player Engineering
http://bit.ly/AdobeARMOptimization
18
Adobe Flash & AIR Support for ARM

Adobe Flash 10.3 for Android 2.2 and above

Available from Android Marketplace today!

Adobe AIR for Android 2.2 and above

Browser
-
less Rich Internet Applications

Package as tradition .
apk
Android Apps

Available from the Android Marketplace today!

Broad Adobe Flash10.3 ARM device support

Flash10.3 for ARM through the Open Screen Project

ARM continuing work with Adobe

ARMv7 NEON for graphics and Pixel Bender
operations

Optimizations for SMP architecture
Adobe Flash 10 on a
Nexus One
Adobe AIR in the Marketplace
19
Skia Graphics Engine Optimizations

2D Graphics package still used extensively within Android

Used by Chrome browser and UI framework

Skia
optimized using ARMv7
NEON
intrinsics
and assembly
code

30%
performance uplift on UI use
cases

10%
-
30%
performance uplift on
browser use cases

S
ome
Skia
routines see as much as
500% lift
from NEON
Skia
2D
Graphics
Android
Chrome
Chrome
OS
20
Renderscript
Introduced in 3.0

Enables GPU agnostic
shader
and compute code for high
performance 3D graphics with native like performance

Augments traditional framework API’s and native OpenGL ES

Heavily leverages LLVM

Used by
Xcode
,
OpenCL
,
ActionScript
and
pNaCl

ARM contributes optimizations to
LLVM project

Renderscript
not for all

Increased application complexity

Limited debug visibility

Reduced platform portability
–
exclusive to Android
To ARM Native
21
Android Platform Highly Optimized for ARM
Dalvik
JIT Support
Renderscript
V8
Javascript
Ogg
Theora
VP8
WebKit
Browser
ARM Native Applications
NDK Tools
Google and ARM are investing significant effort to optimize both Android and
Chrome platforms for the ARM architecture
1,200 files of ARM Assembler
LLVM for
Renderscript
22
Anatomy of an Android Application
AirPlaySDK
System
Libraries
Native
Activity App
Custom
Native
Library
Dalvik
EE
Android System Libraries
Dalvik
Native Libraries
Flash+AIR
EE
Gaya3D
EE
Mono
EE
C/C++/
Asm
Java
ActionScript
, C#,
GayaScript
Dalvik
Native Libraries
Languages
Execution Environments
Native code
Intermediary Byte Code
Android Application
–
Run as a
Dalvik
process
Android Platform
ARM
23
DS
-
5 for Android App Development

Roadmap from ARM for Android
app development

Integrates with Eclipse framework
alongside Android SDK

Professional native debugging

DS
-
5 to include NDKr5(b) compatible
gdbserver
for native device debug

Streamline
™
performance analyzer*

Timelines with performance counters

Call path and time
-
based profiling
Compiler
Eclipse
Debugger
Device Configuration Database
Hardware Debug
Streamline
IDE
Simulation
DS
-
5
* Integrated kernel module required
24

ARM ecosystem for Android is incredibly rich

Flash, HTML5, 2D/3D graphics, applications, tools, etc...

The ecosystem grows with the popularity
of Android

Enabling this ecosystem takes
sustained investment

ARM has a history of ecosystem
enablement and continues to invest
in Android

More to come from our software engagements in
2011!
Summary
25
Thank You!