Android Optimization: Myth and Reality

Android Optimization:
Myth and Reality

Jim Huang ( 黃敬群 ) <jserv@0xlab.org>
Developer, 0xlab
http://0xlab.org/

July 8, 2011

Rights to copy
© Copyright 2011 0xlab
http://0xlab.org/
contact@0xlab.org
Attribution – ShareAlike 3.0
You are free Corrections, suggestions, contributions and
translations are welcome!
to copy, distribute, display, and perform the work
to make derivative works Latest update:July 10, 2011
to make commercial use of the work
Under the following conditions
Attribution. You must give the original author credit.
Share Alike. If you alter, transform, or build upon this work, you may distribute the
resulting work only under a license identical to this one.
For any reuse or distribution, you must make clear to others the license terms of this
work.
Any of these conditions can be waived if you get permission from the copyright holder.
Your fair use and other rights are in no way affected by the above.
License text: http://creativecommons.org/licenses/by-sa/3.0/legalcode

Agenda (1) Android Meets
Hardware Revolution
(2) Unproven "Optimization"
(3) Android Evolution
(4) Mythbusters

Revolution
(1) an attempt, by a large number of people, to change the government of a country,
especially by violent action ( 革命 )
(2) a great change in conditions, ways of working, beliefs, etc. that affects large numbers
of people ( 巨變 )

Evolution
(1) the gradual development of plants, animals, etc. over many years, from simple to
more complicated forms ( 進化 )
(2) the gradual development of something ( 演變 ; 發展 ; 漸進 )

Quote from Google Translation

My interpret of Android:
Hardware is Revolution;
Sotware is basically Evolution;
Android is Hardware-driven Software Revolution

Android Meets Hardware Revolution

Take ARM for Example

• 32-bit instructions, with extension to support 16-bit
Thumb & Thumb-2 instructions.
• Single unified memory address space (i.e. all
peripherals and I/O are accessed like normal
memory, at certain specific memory locations).
• Relatively low power consumption.
• Run at wide range of clock cycle
– (June 1, 2011) Qualcomm announced qual-
core MSM8964/APQ8064
– 2.5 GHz / 2MB L2 cache Do you rember the specifications
of your old laptop computers?
– sample in early 2012

Classic ARM Processors (1)

• ARM7TDMI family
– Based on ARMv4T architecture with 3-stage pipeline
– supports the 16-bit Thumb instruction set
– supports the JTAG Debugger
– includes a fast Multiplier to support DSP algorithm
– supports the In-Circuit Emulation interface
• ARM9TDMI family
– Based on ARMv4T with Harvard cache architecture
– 5-stage pipeline
– ARM920T is based on ARM9TDMI with a memory
management unit (MMU)


• ARM9E family / Intel’s Xscale
– Based on ARMv5E architecture
– Enhanced with DSP instructions
– Hardware support of Java bytecodes execution
• ARM10 family
– Based on ARMv5E with MMU
• ARM11 family
– Based on ARMv6 architecture The first shipped Android
phone belongs to ARM11 family.
– Supports SIMD instructions
– Reduce context switching cost

Android 1.5 (2009/04/27)
ARMv5te optimizations: libc (thumb, memcpy, memset, clz),
dalvik (fast interpreter), skia, surfaceflinger, pixelflinger, audioflinger

Android 2.0 (2009q4)
ARMv7-a optimizations: libc (memcpy/neon), dalvik (JIT), skia
• Cortex families
– Based on ARMv7 architecture
– Supports the new Thumb-2 instruction set
The era of Android:
The era of Android:
– NEON SIMD engine
High quality
High quality
– Cortex-A: For complex OS based applications mobile computing
mobile computing
• Production: Cortex-A8 / Cortex-A9
– Cortex-R: For real-time embedded applications
– Cortex-M: For deeply embedded, microcontroller type cost
sensitive applications
• Cortex-A15
– Triple dispatch superscalar, out-of-order
– High performance prefetcher
– Up to 4M integrated L2 cache
– 1TB of physical address space
– Enterprise class performance, mobile power

ARM Optimizations in Android

• Thumb2 optimizations in Dalvik VM
– On real memory systems with latency the
additional instructions in I-cache latency, it
makes Thumb2 higher performance than ARM
(logic: 1.23x)
• NEON/VFP optimization in skia & dalvik
– 1.45x in dalvik; 2-5x microbench in skia
• Thumb2 + VFP in V8 JavaScript engine
• ARMv6 atomic/rev in libc & dalvik
• NEON + VFP in StageFright
• SMP in libc, libcutils, dalvik
Again, we were informed of the improvements from
Again, we were informed of the improvements from
hardware and dedicated software.
hardware and dedicated software.
But, is it enough?
But, is it enough?

“Optimization" → 最佳化 (?)

「最佳化的迷思」
操作績效不理想或想再創佳績策略核心邏輯不是藤原拓海等級
於是就開始進行程式最佳化參數再怎麼調
在操盤軟體上東調西調
也調不出來一個藤原拓海
嚐試在變化萬千的參數搭配中找尋聖杯
果然出現創紀錄的績效時
就以為得到聖杯回到以策略為中心點
拿到市場實地操作檢視策略在買賣點、停損停利點與資金
卻發現落差很大控管上
可以改善的空間
最佳化的對象是誰才能突破你自己
程式還是策略
拿頭文字 D 來比喻吧
程式好比藤原拓海的車子
車子當然可以越改越好
但是藤原拓海才是致勝靈魂
對比程式策略才是致勝關鍵

http://tw.myblog.yahoo.com/trend-trading/article?mid=18&prev=22&next=16

Key Concept: Strategy

Even runtime is (claimed as) fully
optimized, we still have to look into
details carefully.

Android Services in Action
Content Telephony Bluetooth Connectivity Location
Manager Service Service Service Manager

Window Activity Package Power
…
Manager Manager Manager Manager
Add Service
Service Audio Flinger
Add Service
Manager
daemons Surface
Surface
Service Flinger
Flinger
usbd Manager

adbd System
runtime Server
debuggerd

rild Init Zygote
Zygote Dalvik VM
Dalvik VM

Kernel Android Service
Linux Kernel
Dalvik Framework

 Properties
Android SurfaceFlinger
 Can combine 2D/3D surfaces and surfaces from multiple applications
 Surfaces passed as buffers via Binder IPC calls
 Can use OpenGL ES and 2D hardware accelerator for its compositions
 Double-buffering using page-flip

Optimizing SurfaceFlinger Techniques
• Measure/Profile the execution flow
• Scenario-driven system analysis
– 2D/3D animation
– Video Playback
– Camera preview/recording
• Check list:
– Color space converting
– Eliminate the invalid cases (quality of HAL)
– Allow zero copy among framework/HAL/Driver
• Shared memory / IPC cost
– Any misconfiguration?
– Cache (write-through vs. write-back)
– The last: essential routines

Write bandwidth greatly
affected if caches are
configured as write-through

Configure caches as write-
back, with allocate-on -write

Memory bandwidth

Single Instance 2 Instances

Possibly Premature optimizations
• “Premature optimization is the root of all evil”
– Donald Knuth
• Bionic
– Glibc-incompatible, No SysV IPC, partial Pthread,
incomplete prelink
– inactive/incorrect kernel header inclusion
– May not re-use existing system utilities
• Assumed UI behaviors
– Input event dispatching and handler
– Strict / non-maintainable state machine (policy)
– Depending on a certain set of peripherals
• Unclear HAL design and interface
– Wifi, Bluetooth, GPS, ...

Android is Hardware-driven Software Revolution

Android Software Evolution
• Excluding the trickery, Android is a huge software project
and publishes large parts of source repositories to
AOSP (Android Open Source Project).
• “Hardware-driven software revolution” is the trickery. To
fairly develop, it has to turn into "evolution".
– SoC vendors built up the community(-like) hosted
Android variants. → Qualcomm, TI, Samsung, ST-
Ericsson, ...
• Quality-control is another key factor
– 0xlab delivered 0xbench and related validation
infrastructure.
• Solution to Android ecosystem is to continue to grow the
AOSP community
– need to unify to move forward
– Reasons why Linaro exists

Functional View (1.5)
Applications
Applications
Gallery Phone Web Browser Google Maps ・・・・・

Android Framework
Android Framework
Activity Window Content Notification
View System

Package Telephony Resource Location

System Library
System Library Android Runtime
Android Runtime
SurfaceFlinger
SurfaceFlinger OpenCORE
OpenCORE SQLite Class Library

Dalvik Virtual Machine
OpenGL|ES
OpenGL|ES AudioFlinger
AudioFlinger WebKit

SGL
SGL OpenSSL bionic libc
bionic libc Freetype

Linux Kernel
Linux Kernel

Functional View (2.3)
Applications
Applications
Gallery Phone Web Browser Google Maps ・・・・・

Android Framework
Android Framework
Activity Window Content Notification
View System

Package Telephony Resource Location
Manager Manager Manager Manager RenderScript

OpenGL|ES 2.x accelerated. System Library
System Library Android Runtime
Android Runtime
Drop 2D accel
SurfaceFlinger
SurfaceFlinger StageFright SQLite Class Library
V8 bridge Dalvik Virtual Machine
GLES 2.0 OpenGL|ES
OpenGL|ES AudioFlinger
AudioFlinger WebKit
SMP improvements
Skia
Skia OpenSSL bionic libc
bionic libc Freetype
JIT compiler
Skia supports GPU backend SMP fixes
In Android 3.x Linux Kernel
Linux Kernel

Frequently Asked Questions

• "What is the minimal hardware requirement for
running Android?"
→ UX is the key.
• "May I reduce BOM cost in my Android product?"
→ cost down is always possible, but it depends
on the overall performance.
• "How can I upgrade Android without pain?"
• "I need XXX feature from newer version of
Android. How can I get it into the current and old
codebase?"
→ Partially upgrading is feasible for system
libraries, but always difficult for Android
framework

Think Difficult

• The Effects of Convergence
– Mobile computing platform
– Ecosystem
• Content providers lead the development of
containers (devices)
• Innovation is integrated by open source software
model. The cost of proprietary software gets much
higher.
• To optimize or not to optimize, that is the question.
– Merge Local optimizations != Optimized globally
– Hardware revolution sharpens optimization
strategies and product directions.
– The role of your team

Android Optimization: Myth and Reality

More Related Content

What's hot

Similar to Android Optimization: Myth and Reality

More from National Cheng Kung University

Recently uploaded

Android Optimization: Myth and Reality