intel® vtune™ amplifier 2016 for systems

Intel® VTune™ Amplifier 2016 for SystemsPerformance and Power Profiling on Android* Devices

Copyright © 2015, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others.Optimization Notice

Agenda

2

Overview

Intel® System Studio 2016

Intel® VTune™ Amplifier 2016 for Systems

Intel® Energy Profiler

What’s New

Other Tool Options for Android*

Basic Performance Analysis

Steps to do “Basic Hotspots”

Additional Features

Performance Optimization Basics

Other Collection Options

Viewpoints/Grouping/Filtering

How to Collect

Advanced Topics

Power Analysis

Power Optimization Basics

Power Views in the VTune Amplifier GUI

How to Collect:

SoCWatch

Advanced Topics

Installation Overview

Device Requirements

Host Installation

Target Installation


Overview

3


Intel® System Studio for Android*Deep System Insights for Mobile System Developers

4

IA Coverage

OS Support

From single to multicore

High performance libraries

SoC, CPU, and GPU analysis

System Debug & Trace

In-depth Analysis & Debug

Smartphone and Tablet

Support for Latest Intel Processor and SoC

Advanced system debug & trace for greater system stability

SoC-wide analysis for enhanced power efficiency and performance

Graphics Performance Analysis and optimization tools for graphics-intensive applications

Industry-leading performance from exceptional C++ Compiler and libraries

Boost Performance

Windows* and Linux* host

Android* Target


Intel® System Studio for Android* Overview

Debug

• Intel® JTAG Debugger

System Application

Analyze• Intel® VTuneTM Amplifier• Intel® Graphics Performance Analyzers

(System Analyzer)• Intel® Energy Profiler

Power & Performance

Write and Test Code

JTAGInterface

Intel® Processor-based Mobile Systems

Integrated software tool suite that provides deep system-wide insights to help:

Accelerate Time To Market

Strengthen System Reliability

Boost Power Efficiency and Performance

• Intel® C/C++ Compiler• Intel® Integrated Performance Primitives

Preview • Intel® Hardware Acclerated Execution

Manager

System and Application Code

5


Intel® System Studio 2015 Components

6 Target OS Support

Linux* 1, 5 Android* 5 Windows* VxWorks*

Category Component

Co

mp

ose

r E

dit

ion

Pro

fess

ion

al

Ed

itio

n

Ult

ima

te E

dit

ion

Co

mp

ose

r E

dit

ion

Pro

fess

ion

al

Ed

itio

n

Ult

ima

te E

dit

ion

Co

mp

ose

r E

dit

ion

Pro

fess

ion

al

Ed

itio

n

Co

mp

ose

r E

dit

ion

Host Operating Systems Linux*, Windows* Linux*, Windows* Windows*Linux*,

Windows*

Integrated Development EnvironmentEclipse*, Wind River*

Workbench*Eclipse* Visual Studio*

Wind River* Workbench*

Compiler & Libraries

Intel® C++ Compiler √ √ √ √ √ √ √ √ √ 2

Intel® Integrated Performance Primitives √ √ √ √ √ √ √ √ √ 2

Intel® Math Kernel Library √ √ √ √ √

Intel® Threading Building Blocks √ √ √ √ √ √ √ √

Application Debugger

Intel-enhanced GDB* Application Debugger √ √ √ √ √ √

Analyzers

Intel® VTune™ Amplifier for Systems √ √ √ √ √

Intel® Energy Profiler √ √ √

System Analyzer √ √ √

Frame Analyzer 4 √ √ √

Platform Analyzer 4 √ √ √

Intel® Inspector for Systems √ √ √

System Debugger

Intel® System Debugger (JTAG) 3 √ √

1 Linux*, Embedded Linux, Wind River* Linux*, Yocto Project*, Tizen*2 Delivered with Wind River* VxWorks* platform*3 Via Intel® ITP-XDP3 probe, OpenOCD*, Macraigor* usb2demon* and EDKII* for UEFI*4 Available on Windows* host only5 Linux* and Android* target support available in a single product


Intel® VTune™ Amplifier for SystemsPerformance Profiler

Get the Tuning Data You Need

− Low overhead “hotspot” analysis with call stacks

− Advanced analysis for cache, branching, …

Find Answers Fast

− Powerful analysis & data mining

− Results mapped to C/C++ or Java source

Easy to Use

− Remote analysis from the User Interface

− Windows or Linux Host analyzes Linux or Android target

Available now as part of Intel® System Studio

7

Optimize Your Software Performance


Intel® Energy ProfilerEnergy and Power Profiler for System Software Developers

Optimize Software for Extended Battery Life

Find the Cause of Wake Ups That Waste Energy

− Interrupts mapped to the IRQ/device

− Timers mapped to the scheduling process

− Data correlated with Android Wake Locks

Available now for Linux and Android

Part of Intel® System Studio

8

Get Actionable Data to Extend Battery Life

Requires specific SOCs. On Android, a rootable OS is required with version compatible device drivers. See release notes for details.


Android Support including…

Basic hotspots, Locks & Waits and EBS with stacks for RT kernel and RT applications for Linux Targets

EBS based stack sampling for kernel mode threads

Support for Intel® Atom™ x7 Z8700 & x5 Z8500/X8400 processor series (Cherry Trail) including GPU analysis

Automated remote EBS analysis on SoFIA (by leveraging existing sampling driver on target)

Super Tiny display mode added for the Timeline pane to easily identify problem areas for results with multiple processes/threads

Platform window replacing Tasks and Frames window and providing CPU, GPU, and Bandwidth metrics data distributed over time

General Exploration analysis views extended to display confidence indication (greyed out font) for non-reliable metrics data resulted, for example, from the low number of collected samples

GPU usage analysis for OpenCL™ applications extended to display compute-originated batch buffers on the GPU software queue in the Timeline pane (Linux* target only)

New filtering mode for command line reports to display data for the specified column names only

Continually expanding Mobile Development Kit Program - http://software.intel.com/mdk

Many other features for embedded OS’s, improvements to the GUI, and various bug fixes…

See Release Notes9

What’s New In Intel® VTune™ Amplifier 2016 for Systems

9

http://software.intel.com/mdk


10

Other Intel® Software Developer Tools for Android*

10


Basic Performance AnalysisUsing Intel® VTune™ Amplifier 2016 for Systemson Android Systems

11


Basic HotspotsStart Here - Makes It Easy

12

#1 used feature

Easiest feature to use

Enables the most important feature of identifying the hotspot

Works on non-rooted (and rooted) Intel® architecture devices

Collects samples using OS-timer event for a specific application/process

Associate samples to:Module/thread/functionC/C++ source or assemblyJITted Java/Dalvik functions/ART functions/assembly/dex/source

Collects User Mode Stacks (default)


Basic HotspotsStep 1) Attach device to host System

13

Host SystemWindows* or Linux*

Android* DeviceIntel® architecture based

adb

• ADB connectivity installed on host system• “Enable USB Debugging” (Under Developer Options on Target Device)


Basic HotspotsStep 2) Create Project in Intel® VTune™ Amplifier

14

Create Project in VTune Amplifierset target type: “Android Device (ADB)”set target type: Launch Android Packageset target system: Your deviceset Package or Process NameOptionally set other options


Click “Launch New Analysis” via play buttonThen select “Basic Hotspots” under “Analysis Type”Then click “Start”

15

Basic HotspotsStep 3) Start Hotspot Analysis


Now interact with your app… make it run whatever it is you want to tune.

Tell VTune Amplifier to Stop

VTune Amplifier will then copy the data files back to the host system for post analysis

16

Basic HotspotsStep 4) Run benchmark & stop VTune Amplifier collection


17

Basic HotspotsStep 5) Identify hottest functions


18

Basic HotspotsOptional Step 5) Enable C/C++ Source View

amplxe-cl [options] –search-dirs=Path-to-Symbols-on-host

Binaries with symbols traditionally are located in:[AndroidAppBuildDir]/out/target/product/[your target]/obj

[AndroidOSBuildDir]/out/target/product/[your target]/symbols, or[AndroidOSBuildDir]/out/target/product/[your target]/obj[AndroidOSBuildDir]/out/target/product/[your target]/linux/kernel


Basic Hotspots is

• Easy

• Works on all Intel architecture-based Android* devices

• Extremely useful for diagnosing performance issues

Simple set of steps presented

19

Basic HotspotsSummary


Additional Features of Intel® VTune™ AmplifierUsing Intel® VTune™ Amplifier 2016 for Systemson Android Systems

20


1. Determine desired performance

2. Build Application/System for optimization not debug (w symbols avail)

1. System = Userdebug Build (rootable)

2. Application = Compile C/C++ binaries w optimizations (debuggable attribute disables this)

3. Find hotspots (Where the application/system is spending time)

4. [Optional] Determine the efficiency of the hotspots

Determine architectural issues of the hotspots

5. [Optional] Find changes to hotspots between builds

6. Make 1 code change to 1 hotspot to improve performance

Remove hotspot

Improve algorithm

Change code to better utilize hardware

Get new hardware

7. Validate that change improved performance

8. Go to step 3 – until desired performance achieved

21

Performance Analysis Steps (condensed)


Advanced HotspotsTo get more information

22

Identifies the hotspot using hardware counters (PMU) of Intel® processors

Allows system-wide collection Allowing you to see all processes running on system

For single application can collect:User-Stacks + Kernel-Stacks Context SwitchesCall Counts

Associate samples to:Process/Module/functionCore/threadC/C++ source or assemblyJITted Java/Dalvikfunctions/assembly/dex/source

System Wide:

amplxe-cl --collect advanced-hotspots --duration=<N>--target-system=android

Stacks, Context & Counts:

amplxe-cl –collect advanced-hotspots -knob collection-detail=stack-and-callcount–-target-process=<appName> --target-system=android


Uses hardware counters (PMU) to identify microarchitectural issues in your system/application

Makes it easy to select appropriate counters for each Intel® microarchitecture to find issues in…

Memory (Cache, TLB, Reissues, Bus-Locks)

Branch mis-prediction

Machine Clears, Floating Point Stalls

Efficiency (CPI, uOp(s)-Retired)

Applies Formulas/Heuristics developed by Intel engineers to highlights issues

Easy to Interpret – If it is Pink – Examine in more detail

amplxe-cl --collect advanced-hotspots --duration=<N>

--target-system=android

23

General ExplorationTo Diagnose Microarchitecture Bottlenecks

amplxe-cl --collect [atom-general-exploration | snb-general-exploration] --target-system=android

Pink


ViewpointsEnable looking at the data the way you need it

24

It is a pre-defined view that determines what needs to be displayed in the grid and timeline for a given analysis type

An analysis type may support more than one view points


Groupings

Each analysis type has many viewpoints

Each viewpoint has pre-defined groupings

Allows you to analyze the data in different hierarchies and granularities

25

Intel® VTune™ Amplifier Key Concepts

Click


For example, pre-defined groupings can be used to determine load imbalance

26

Groupings

Change to Function/Thread


Filtering (Time or Group) - Lets you focus on what’s important

27

FilteringOnly view data related to your issue

Copyright © 2015, Intel Corporation. All rights reserved. *Other names and brands may be claimed as the property of others.Optimization Notice28

Caller/Callee ViewGUI Layout Select a function in the

Bottom-Up and find the caller/callee

List of functions sorted by CPU Time

List of callers and their stacks

List of calleesand their stacks


29

Adding User Marks to the TimelineGUI Controls

Start application without data collection

Resume data collection when needed

Observe paused region on the Time Line

Click “Mark Timeline” during collection

Observe the mark on the Time Line


Repeat: Overview of Remote/Attached Collection Procedure/Architecture for Android*

Uses “adb” protocol/binary for collection & data transfer (must be in path)

Flexible collection configuration + control (pause/resume/stop)30

Target device

amplxe-runss

Host

VTuneGUI

VTune result

VTune collector binary runs on target and stores result on target

Data is opened in GUI and symbols are

resolved using modules stored in result dir

User can specify search dir with separate debug

files if needed

amplxe-cl

control collection

transfer data/modules

VTune result

driver

adb

adb

Transfers data collected remotely back to host automatically together with stripped application modules for symbol resolution

GUI Collector Control

Some collection types require signed drivers accessed from rooted device


How to Collect

31

Via the command line

amplxe-cl --collect hotspots --target-process=<appName> [Other Options] --target-system=android

amplxe-cl --collect advanced-hotspots[Other Options] --target-system=android

amplxe-cl --collect [atom-general-exploration | snb-general-exploration] [Other Options] --target-system=android

Via the GUI1)Attach device to host – via adb2)Create Project in VTune

and set target system3)Click “Launch New Analysis”

then Select Analysis Typethen click “Start”

4)Wait till collection finishes or click Pause, Resume, or Stop Collection.

Click “Command Line….” – dialog will displaythe Command Line for that analysis type


Advanced TopicsIntel® VTune™ Amplifier for Systems

32


Choose Specific Processor Events

Change Sampling Rates

Collect Samples for Entire System, Single Application, Kernel Space Only, User Space Only

Many, Many More Options….

Note: Very flexible option set – (with defaults presented earlier) lets user collect as much or as little potential information in a single run, with the caveat that the more you collect the higher the overhead – which will impact the performance of what you are measuring

33

Custom Analysis Typesallows many advanced options for collection

More details in Intel® VTune™ Amplifier Help: User's Guide


Quickly identify cause of regressions.

Run a command line analysis daily

Identify the function responsible so you know who to alert

Compare 2 optimizations – What improved?

Compare 2 systems – What didn’t speed up as much?

34

Compare Results Quickly - Sort By Difference


Power AnalysisUsing Intel® Energy Profiler in Intel® VTune™ Amplifier for Systems

35


Traditional optimization Race to Idle

Perform Operation Faster – Then Sleep

Achieved by

Use of new instructions

Increase core parallelism

Use Standard Performance Optimization Tools – like Intel® VTune™ Amplifier for performance Analysis

New optimization Increase uninterrupted idle time

Use SoC components as needed

Achieve by

Reduce the frequency of activity

Consolidate activities

Run code on appropriate SoC block

Turn off components (or system)when not in use

Increase Power Efficiency

Minimize Wake-ups from Timers and Interrupts


Intel® Energy ProfilerEnergy and Power Profiler for System Software Developers

Optimize Software for Extended Battery Life

Find the Cause of Wake Ups That Waste Energy

− Interrupts mapped to the IRQ/device

− Timers mapped to the scheduling process

− Data correlated with Android Wakelocks

Android Tools

• SoCWatch -collector (command line only)

• VTune Amplifier 2016 for Systems UI – for displaying data

37

Get Actionable Data to Extend Battery Life

Requires specific SOCs. A rootable version of the OS is required, with version compatible device drivers (or kernel sources + signing key)


CPU Sleep States

38

• Flexible C-States to Select Idle Power Level vs. Responsiveness

Core voltage*

Core clock

PLL

L1 caches

L2 cache

Wakeup time*

Idle power*

off

Active state

off

flushed

off

off

flushed

off

partial flush

active

* Rough approximation

C6

off

off

off

off

C0 C1 C2 C3


Processor Power and Processor Frequency

Copyright © 2004 Intel Corporation. All rights reserved.

39

9

59

109

159

209

259

309

359

0 0.5 1 1.5 2 2.5 3 3.5

Po

we

r (w

)

Frequency (GHz)

Power vs. Frequency Curve for Single Architecture

Small Increases in Processor Speed

Results in Large Increases in Power

39


40

CPU C-States / P-States

C1

C2

C3

C4

C6

Pn

P1

P0 CPU

Active

CPU

Sleep

P0 - CPU active at highest frequency (HFM)

Pn - CPU active at lowest frequency (LFM)

C0 - CPU active (In any P-state)C0

C1 - Core clock is Off

C3/C4 - Reduced Voltage, Partial L2 cache flush

C6 - Core Off, L2 cache flush, state saved to SRAM

The deeper the sleep state

more power saving

but longer to wake up

Po

we

r H

igh

er

La

ten

cy G

rea

ter

40


Find Process/Thread Waking System up

41

C-State Wakeup

Identify the object which woke the CPU up the most often

Reduce the # of wakeups

Identify if the Processor was asleep (C1-C6) mostly or awake (C0)


Small Increases in Processor Speed Results in Large Increases in Power

Determine when the CPU Frequency went up

Determine what frequency the CPU was running at and for how long…

42

Determine the CPU Frequency

Hovering the mouse cursor over a point in the timeline will bring up a pop-up box showing more detailed information such as specific measured frequency at that measurement time.


Component Device StatesFind Components Wasting Power

43

Intel Device States:

DOi0 = On

DOi1-DOi2 = Intermediate

D0i3 = Off

Find:

• Which Devices are on/off?

• For this example no media use and only periodic rendering?

• When they got turned on/off?

• When a device is not in use… the software needs to turn it off


C-States (CPU wakeups)

P-States (CPU frequency)

SOi States (System State)

Android Wakelocks

Temperature (Core, Skin, SoC, PMIC)

N(orth) C(luster) Device States

S(outh) C(luster) Device States

45

Potential Collections


Correlate Metricsto find patterns

46

See all the metrics at once

Find patterns

Find the cause


47

Correlate CPU Frequency, Sleep State, Wake-up Objects, etc...


Command Line Tool focused on Power Analysis

Correlates key hardware and OS data providing complete system view

Selects the best collection method based on user input

Tracing: 100% accurate

Collects every state change

Snapshot: minimum overhead

Read at start and end of collection, provide difference

Polling: reads values 10 times/sec (configurable)

48

SoCWatch for Android


49

SoCWatch Process


./socwatch –f sys –f wakelocks –t 10

-f sys // collects all metrics

-t 10 // defines duration of collection

Snapshots , Traces PStates for 10 seconds, create default SocWatchOutputfiles

Import into VTune Amplifier on host via:

adb pull <path-on-target>/SocWatchtOutput.sw1

amplxe-cl -import ./SocWatchOutput.sw1 –r <project name>

Open Results in VTune Amplifier GUI

50

Example Command Line Usage

More details in SoCWatchForAndroid_v1_3_0.pdf


51

CPU Collectable Metrics

parameter Metric Description Collection Method(s)

-f cpu-cstate App Processor Utilization

% of time spent executing instructions

Snapshot

-f cpu-pstate App Processor Frequency –

PStates

% of time spent in each PState or timeline of PState

transitions

Trace

-m –f cpu-cstate

App Processor Wakeups

Reasons why the App Processor exits a CState

Trace

-f cpu cpu-cstate & cpu-pstate


52

SoC Collectable MetricsMetric Description Collection

Method(s)

-f nc-dstate SoC South Cluster

Component Power States –SC D0ix States

% of time each SC SoC component spent in

each D0ix state

snapshot

-f sc-dstate SoC North Cluster

Component Power States –NC D0ix States

Statistical % of time each NC SoC

component spent in each D0ix state or

statistical timeline of D0ix behavior

Poll

-f device nc-dstate + sc-dstate


53

Temperature Collectable MetricsMetric Description Collection

Method(s)

-f core-temp App Processor Temperature

Statistical % of time spent in each temperature or statistical timeline of

temperature behavior

poll

-f soc-temp SoC Temperature

poll

-f pmic-temp PMICTemperature

poll

-f skin-temp Skin Temperature

Poll

-f temp core-temp + soc-temp +pmic-temp + skin-temp


54

System Collectible Metrics

Metric Description Collection Method(s)

-f acpi-state ACPI Suspend-To-RAM State – S3

% of time spent in S3 state or timeline of S3

transitions

Trace

-f s0i-state SoC Power States -S0ix States

% of time spent in each S0ix state

snapshot

-f wakelock Traces both user and kernel wakelocks used during the collection

trace

-f sys cpu + device + temp +acpi-state + s0i-state


Advanced reporting options

Configure polling intervals

Command line text report output format

VTune Amplifier GUI offers same Viewpoints/Grouping/Filtering and regression analysis capabilities as presented earlier.

55

More Options Available for SoCWatch

More details in SoCWatchForAndroid_v1_5_4.pdf


Installation Overview

56


Installing Intel® System Studio to the Host System

57

Linux* Windows*

1) ./online_install.sh 1) w_cembd_2016.0.xxx_online.exe

…or… …or…

2) tar l_cembd_p_2016.0.xxx.tgzl_cembd_p_2016.0.xxx.tg/install.sh

2) w_cembd_2016.0.xxx.exe

Default Install Location

/opt/intel/system_studio_2016.0.xxx/vtune_amplifier_for_systems

C:\Program Files (x86)\Intel\System Studio 2016.0.xxx\vtune_amplifier_for_systems

Location of Target Files

<path-to-SystemStudio>/target/system_studio_target.tgz

<path-to-SystemStudio>\target\system_studio_target.tgz


Performance Collection Requirements for VTune Amplifier 2016 for Systems:

Basic Hotspots including user mode stacks to C/C++ Source

− All available Android devices based on Intel® Architecture such as those at http://software.intel.com/en-us/android/get-device.

− Note: Install to target device is not needed

Java Functions with Java Source, JIT assembly or DEX Drill Down

− Rootable device

− Instrumented Java/Dalvik/ART JVM (Available in most Intel® Atom processor builds)

Hardware-Event Based Sampling (Advanced Hotspots, General Exploration, System [Wide] Profiling, Kernel Stacks, and Custom Analysis Types)

− Rootable device

− Version compatible signed performance analysis drivers

58

http://software.intel.com/en-us/android/get-device


Requirements for Intel® Energy Profiler

−Rootable device

−Version compatible signed Intel® Energy Profiler drivers

59


Best option is to get the drivers integrated into the build

Reference builds from Intel® Corporation already have them integrated –sometimes the latest version of driver needs to be updated…

Some OEM builds have the driver

Note: This is “much, much” easier….

Else you need to build the drivers

The install package contains the sources and scripts needed to build the drivers If you need to build the drivers against the existing kernel

You need to have:– The public/private key for that kernel

– The exact source for the kernel used to build that kernel

If you do not have the above then you will need to build a new kernel on your host system, install it on the device, and use the newly generated key to build and sign the drivers.

Instructions in Help/User’s Guides

60

Building Device Drivers


Install Intel® Energy Profiler to Remote Target

61

Linux* Windows*

$tar –zxvf <path-to-SystemStudio>/target/system_studio_target.tgz

“unzip” <path-to-SystemStudio>\target\system_studio_target.tgz

$adb root $adb root

$system_studio_target/socwatch_android_vXX.XX/socwatch_android_install.sh

$system_studio_target/socwatch_android_vXX.XX/socwatch_android_install.bat

See SoCWatchForAndroid_vXX_XX_XX.pdf for full details


62

Steps to start SoCWatch

SoCWatch

$adb shell

#cd /data/socwatch

#. ./setup_socwatch_env.sh

#insmod /lib/modules/socperf1_2.ko

#insmod /lib/modules/SOCWATCH1_5.ko

#./socwatch --help


Capabilities of VTune Amplifier 2016 for Systems on Android*

To Identify Performance Issues

Hotspot, Advanced-Hotspots, General Exploration

Other Advanced Options(Custom Collections, Regressions, Frames)

To Identify Power Issues

CPU Wake-ups, Frequency, Device States, Wakelocks

To Zoom In on your Issue via the GUI

Grouping, Filtering, Sorting, Comparing

Workflow for

Performance and Power Analysis Steps

Installation

Including System Requirements

Collecting

Viewing Data

63

Summary


Intel® System Studio 2016 provides deep system-level insights into power, reliability and performance to help accelerate time to market of Intel Architecture-based embedded and mobile systems

For other versions contact:

Your Intel representative or …

[email protected]

Note: Most features presented here require access to a rootable Android* device, and version compatible device drivers.

64

Call to Action

For more information, to evaluate, or purchase:http://intel.ly/system-studio

mailto:[email protected]


http://intel.ly/system-studiohttp://software.intel.com/en-us/intel-vtune-amplifier-for-systems

http://software.intel.com/en-us/intel-energy-profiler

Premier Support: https://premier.intel.com

Forums: http://software.intel.com/en-us/forum/intel-system-studio/

Email: [email protected]

Release Notes:

http://software.intel.com/sites/default/files/release_notes_amplifier_for_android_linux.pdf

VTune Amplifier Help Documentation:

http://software.intel.com/en-us/vtuneampxe_2013_ug_linSubTopic-> Intel VTune Amplifier User’s Guide : Running Analysis Remotely

http://software.intel.com/sites/default/files/managed/c8/f9/SoCWatchForAndroid_v1_3_0.pdf

http://software.intel.com/sites/default/files/managed/9d/59/WakeUpWatch_v3_1_6.pdf

KB Articles: http://software.intel.com/en-us/articles/intel-system-studio-articles

http://software.intel.com/en-us/articles/android-features-in-intel-vtune-amplifier-2014-for-systems-requirements

http://software.intel.com/en-us/articles/using-intel-vtune-amplifier-on-non-rooted-android-devices

http://software.intel.com/en-us/articles/how-to-use-the-intel-energy-profiler-in-intel-system-studio-2014

65

Additional Resources

65



https://premier.intel.com/

http://software.intel.com/en-us/forum/intel-system-studio/

mailto:[email protected]

http://software.intel.com/sites/default/files/release_notes_amplifier_for_android_linux.pdf

http://software.intel.com/en-us/vtuneampxe_2013_ug_lin

http://software.intel.com/sites/default/files/managed/9d/59/WakeUpWatch_v3_1_6.pdf

http://software.intel.com/en-us/articles/intel-system-studio-articles

http://software.intel.com/en-us/articles/android-features-in-intel-vtune-amplifier-2014-for-systems-requirements

http://software.intel.com/en-us/articles/using-intel-vtune-amplifier-on-non-rooted-android-devices

http://software.intel.com/en-us/articles/how-to-use-the-intel-energy-profiler-in-intel-system-studio-2014


INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS”. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO THIS INFORMATION INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT.

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products.

Copyright © 2014, Intel Corporation. All rights reserved. Intel, Pentium, Xeon, Xeon Phi, Core, VTune, Cilk, and the Intel logo are trademarks of Intel Corporation in the U.S. and other countries.

66

Legal Disclaimer & Optimization Notice

Optimization Notice

Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Notice revision #20110804


Backup SlidesUsing Intel® VTune™ Amplifier 2016 for Systemson Android Systems

User APIs

WuWatch

68


69

Intel® VTune™ Amplifier for SystemsUser APIs

Enable you to

• control collection

• set marks during the execution of the specific code

• specify custom synchronization primitives implemented without standard system APIs

To use the user APIs, do the following:

• Include ittnotify.h, located at <install_dir>/include

• Insert __itt_* notifications in your code

• Link to the libittnotify.lib file located at <install_dir>/lib

• New feature allows creating a csv file instead of using user

API’s… see Help : Creating a CSV File with External Data


70

Intel® VTune™ Amplifier User APIs

Collection Control APIs

Thread naming APIs

void __itt_pause (void) Run the application without collecting data. VTune™

Amplifier reduces the overhead of collection, by collecting

only critical information, such as thread and process

creation.

void __itt_resume (void) Resume data collection. VTune™ Amplifier resumes

collecting all data.

void __itt_thread_set_name (const

__itt_char *name)

Set thread name using char or Unicode string, where

name is the thread name.

void __itt_thread_ignore (void) Indicate that this thread should be ignored from

analysis. It will not affect the concurrency of the

application. It will not be visible in the Timeline pane.


71

Intel® VTune™ Amplifier Collection Control APIs

int main(int argc, char* argv[]){

doSomeInitializationWork();

__itt_resume();while(gRunning) {

doSomeDataParallelWork();}__itt_pause();

doSomeFinalizationWork();return 0;

}


• Useful to observe when certain events occur in your application or identify how long certain regions of code take to execute

• Event APIs enables you to annotate an application when certain events occur

__itt_event __itt_event_create(char *, int);

__itt_event_start(__itt_event);

__itt_event_end(__itt_event);

72

Intel® VTune™ Amplifier User Event APIs


73

Intel® VTune™ Amplifier User Event APIs

__itt_event __itt_event_create(const

__itt_char *name, int namelen ); Create a user event type with the specified name. This API

returns a handle to the user event type that should be

passed into the following APIs as a parameter. The

namelen parameter refers to the number of characters,

not the number of bytes.

int __itt_event_start( __itt_event event ); Call this API with an already created user event handle to

register an instance of that event. This event appears in

the Timeline pane display as a tick mark.

int __itt_event_end( __itt_event event ); Call this API following a call to __itt_event_start() to show

the user event as a tick mark with a a duration line from

start to end. If this API is not called, the user event

appears in the Timeline pane as a single tick mark.


74

Intel® VTune™ AmplifierUser Events - using APIs

DWORD WINAPI aiWork(LPVOID lpArg){

int tid = *((int*)lpArg);__itt_event aiEvent;aiEvent = __itt_event_create("AI Thread Work",14);

while(gRunning) {WaitForSingleObject(bSignal[tid], INFINITE);__itt_event_start(aiEvent);doSomeDataParallelWork();__itt_event_end(aiEvent);SetEvent(eSignal[tid]);

}return 0;

}


75

Intel® VTune™ AmplifierVisualizing Events in the Timeline View

User defined task


Frame Analysis –Analyze Long Latency Activity

Frame: a region executed repeatedly (non-overlapping).

• API marks start and finish

Examples:

• Game – Compute next graphics frame

• Simulator – Time step loop

• Computation – Convergence loop

76

Intel® VTune™ Amplifier Frame Analysis Application

voidalgorithm_1();voidalgorithm_2(int myid);doubleGetSeconds();DWORD WINAPI do_xform (void * lpmyid);bool checkResults();__itt_domain* pD = __itt_domain_create (“myDomain”);

while( gRunning ) {__itt_frame_begin_v3(pD, NULL);. . .

//Do Work. . .__itt_frame_end_v3(pD, NULL);

}

for (int k = 0; k < N; ++k) {int ik = i*N + k;int kj = k*N + j;c2[ij] += a[ik]*b[kj];}

Region (Frame)


__itt_domain* __itt_domain_create( const __itt_char *name );

Create a domain with a domain name. Since the domain is expected to be static over the application's execution time, there is no mechanism to destroy a domain. Any domain can be accessed by any thread in the process, regardless of which thread created the domain. This call is thread-safe.

void __itt_frame_begin_v3(const__itt_domain *domain, __itt_id *id);

Define the beginning of the frame instance. A __itt_frame_begin_v3 call must be paired with a __itt_frame_end_v3 call. Successive calls to __itt_frame_begin_v3 with the same ID are ignored until a call to __itt_frame_end_v3 with the same ID. •domain is the domain for this frame instance. •id is the instance ID for this frame instance, or NULL.

void __itt_frame_end_v3(const__itt_domain *domain, __itt_id *id);

Define the end of the frame instance. A __itt_frame_end_v3 call must be paired with a __itt_frame_begin_v3 call. The first call to __itt_frame_end_v3 with a given ID ends the frame. Successive calls with the same ID are ignored, as are calls that do not have a matching __itt_frame_begin_v3 call. •domain - The domain for this frame instance •id - The instance ID for this frame instance, or NULL for the current instance.

77

Intel® VTune™ Amplifier Frame APIs


78

Frame Analysis Using APIs

__itt_domain* pD = __itt_domain_create ("SimDomain");

while(gRunning) {__itt_frame_begin_v3(pD, NULL);

start = clock();//Wait all threads before moving into the next frameWaitForMultipleObjects(FUNCTIONAL_DOMAINS, eSignal, TRUE,

INFINITE);stop = clock();//Give all threads the "go" signalfor (int i = 0; i < FUNCTIONAL_DOMAINS; i++)

SetEvent(bSignal[i]);if (frame % NETWORKCONNETION_FREQ == 0) {

//Start network threadSetEvent(bNetSignal);

}__itt_frame_end_v3(pD, NULL);

}


Summary View / Frame Rate Chart

Adjust the frame rate then Apply changes


80

Frame Analysis Find Slow Frames With One Click

(1) Regroup Data

… (Partial list shown)

Before: List of Functions Taking Time

After: List of Slow Frames


81

Just 2 more clicks shows where to focus tuning…

Slow functions in slow framesResult: Functions taking a lot of time in slow frames

(1) Only show slow frames

(2)Regroup: Show functions


• A task is a logical unit of work performed by a particular thread

• Tasks can be nested

• You can use task APIs to assign tasks to threads

• One thread executes one task at a given time

• Tasks may correspond to functions, scopes, or a case block in a switch statement

82

Intel® VTune™ AmplifierTask APIs


83

Task API primitivesUse This Primitive To Do This

void ITTAPI__itt_task_begin ( const__itt_domain *domain, __itt_id taskid, __itt_idparentid, __itt_string_handle *name)

Create a task instance on a thread. This becomes the current task instance for that thread. A call to __itt_task_end() on the same thread ends the current task instance.

void ITTAPI__itt_task_begin_fn ( const__itt_domain *domain, __itt_id taskid, __itt_idparentid, void *fn)

Begin a task instance on a thread.

void ITTAPI__itt_task_end ( const__itt_domain *domain)

End a task instance on a thread.

Parameter Description

__itt_domain The domain of the task.

__itt_id taskid This is a reserved parameter.

__itt_id parentid This is a reserved parameter.

__itt_string_handle The task string handle.

*fn This is a reserved parameter.


84

Task API usage__itt_domain* domain = __itt_domain_create(L"Task Domain");__itt_string_handle* UserTask = __itt_string_handle_create(L"UserTask");__itt_string_handle* UserSubTask = __itt_string_handle_create(L“UserSubTask");

int main(int argc, char* argv[]){

...__itt_task_begin (domain, __itt_null, __itt_null, UserTask);//create many threads to call work()__itt_task_end (domain);...

}

work(){

__itt_task_begin (domain, __itt_null, __itt_null, UserSubTask);do_foo();__itt_task_end (domain);return 0;

}


85

Using Task APIHotspots analysis – Bottom-up pane


86

Using Task APIHotspots analysis – Tasks pane


Collection and reporting

WuWatch

– High-K next generation Intel® Atom™ processor (codenamed Medfield, Lexington, or Clovertrail Plus)

SoCWatch

– Intel 22 nm ultra-mobile processor (code name: Silvermont)

– 4th Generation Intel Core Processors (code name:Haswell)

Collection runs on target system (through adb)

Reporting options

Text/CSV style reports from command line

– Some advanced reports only available from command line

Python scripts will generate summary reports

Intel VTune Amplifier GUI (Linux or Windows)

87

Collection Tools for Power State Analysis

WuWatch no longer used


88

Wakeup Watch Process

wuwatch

(collect)

Raw data

(WW1 file)

wuwatch

(process)

Raw text trace

(TXT file)

Summary_data_vX.py

Summary results

(CSV or TXT file)

VTune Amplifier

(correlate / visualize)



In adb root shell on target device:

Get Help>./wuwatch --help

Execute wuwatch command on Android target>./wuwatch -cs –ps –kb –ss –ds -dn -wl -t 60 –o /data/results/test

Execute Benchmark (usually via Android UI)

Import into VTune Amplifier GUI

adb pull /data/wuwatch/results/test.ww1

amplxe-runss --import-socwatch-data ./test.ww1 [-r testresults]

89

WuWatch 3.1 Collection Usage

More details in WakeUpWatchForAndroid.pdf



Potential Collections:

Monitors and traces sleep state transitions (c-states) causing wakeups to srccode that scheduled the timer. switch = -cs

Monitors processor frequency changes (p-states). switch = -ps

Platform sleep states (S–states). switch = -ss

Device sleep states (D-states). switch = -ds

Traces Android* wakelocks to the process. switch = -wl

90

WakeUp Watch (wuwatch)



Drivers needed:

SOCWATCH1_3.ko or apwr3_1.ko

vtsspp.ko (only needed for call stack info)

pax.ko

sep3_10.ko*

For Intel® reference builds, drivers are located in either:

/lib/modules or /system/lib/modules

Verify if the drivers are on your system via:

adb shell ls /lib/modules /system/lib/modules

91

Drivers Needed for VTune Amplifier 2016 for Systems


(Re)Installing VTune™ Amplifier Remote Collector to Target

92

Linux* Windows*

“unzip” <path-to-SystemStudio>\Target\system_studio_target.tgz

$adb root $adb root

$vtune_amplifier_2016_for_systems/vtune_amplifier_2016_for_systems/bin[64|32]/amplxe-androidreg.sh --package-command=install --jitvtuneinfo=[src|none|jit|dex]

$system_studio_target\vtune_amplifier_2016_for_systems/bin32/amplxe-androidreg.bat --package-command=install --jitvtuneinfo=[src|none|jit|dex]

See Help -> Intel® VTune™ Amplifier User's Guide -> Running Analysis Remotely -> Preparing a Target Android* System for Remote Analysis

This step is only needed to support Java.

intel® vtune™ amplifier 2016 for systems

Documents