28
© 2018 Arm Limited Arm Technical Symposia 2018 Beyond Hardware IP An overview of Arm development solutions

Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

© 2018 Arm Limited

• Arm Technical Symposia• 2018

BeyondHardware IP

An overview of Arm development solutions

Page 2: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

2 © 2018 Arm Limited

IC design complexity and cost aren’t slowing down

51.3

106.3

174.4

297.8

542.2

28nm 16nm 10nm 7nm 5nm

Ad

van

ced

fir

st d

esig

n c

ost

(U

S$ m

illio

n)

Process nodeSource: IBS

Software represents

35%-40%of cost at 5nm

Page 3: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

3 © 2018 Arm Limited

Explosion in code size across the board

Windows OS Linux kernel Luxury car

0

50

100

150

200

250

300

350

Luxury car(2010)

Luxury car(L3 ADAS, 2020)

0

5

10

15

20

25

30

Linux Kernel 2.6 (2003) Linux Kernel 4.14(2017)

0

10

20

30

40

50

60

Windows 2000(2000)

Windows Vista(2007)

Mill

ion

s o

f lin

es o

f co

de +200M4x1.8x

Sources: https://informationisbeautiful.net/visualizations/million-lines-of-code/ and Arm

Page 4: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

4 © 2018 Arm Limited

Page 5: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

5 © 2018 Arm Limited

Arm and partners enable an awesome ecosystem

Applications

Middleware

Driver & OS

Security

Tools

Models

100,000s

1,000s

100s

10s

10s AR

M e

nab

lem

en

t

Collective effort across the industry

Arm focuses on key building blocks, close to the architecture

Partners add value with application-specific contributions

Part

ne

rs E

nab

lem

en

t

Page 6: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

6 © 2018 Arm Limited

RTOS

Middleware codecs

10+

IDE, compiler

13+

Debug & Trace

21+

RTOS

40+

World #1 embedded ecosystem

Page 7: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

7 © 2018 Arm Limited

Open source compilers: It takes an ecosystem

GNU toolchain LLVM

Arm

Ecosystem

• Top 3 contributor mid/backend1

• Focus on architecture support• Optimization for enterprise workloads• ARM GNU Embedded/Linux

• 2,100+ commits2

• Focus on architecture support• Across embedded, mobile and HPC• Commercial ARM Compiler

• 3rd party micro-architecture• Tools and libraries maintenance and

optimization

• 3rd party micro-architecture• Collaboration on particular

architecture features

1 Between GCC 4.9 and 7.1. 2 From 2013 to Jul-2017 in LLVM and Clang

developer.arm.com/open-source/gnu-toolchain

Page 8: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

8 © 2018 Arm Limited

Continuous investment in engineering efficiencyArm’s development solutions history pre-dates Arm foundation

1987Launches

Arm Compiler

2008Launches

CMSIS

2015Acquires

Carbon Design Systems

2016Acquires

Allinea

2001Acquires

Noral

2005Acquires

Keil

1990Arm is born from Acorn

Page 9: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

9Copyright © 2018 Arm TechCon, All rights reserved.

#ArmTechCon

Arm Development Solutions: focus areas

Servers

SoC Design Embedded & Safety Ready

Mobile & Graphics Servers & HPC

Page 10: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

10Copyright © 2018 Arm TechCon, All rights reserved.

#ArmTechCon

Arm Development Solutions: SoC Design

Servers

SoC Design Embedded & Safety Ready

Mobile & Graphics Servers & HPC

Page 11: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

11 © 2018 Arm Limited

• Run your code on a cycle accurate virtual prototype of your SoC, and ensure you get it right first time

• Extensive library of cycle-accurate models

• Deep visibility into your system

• Easy to learn, fast to modify

• Integration in SystemC/EDA environments

• Auto-generate cycle accurate models from custom RTL

SoC Design

IP evaluation

Architecture exploration

Design & Prototyping

Production

Page 12: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

12 © 2018 Arm Limited

• Avoid mistakes in the configuration of Arm processor, system and debug IP, and shorten your SoC schedule

• Easy to use, menu driven setup of complex interconnect IP

• Automatic generation of CoreLink interconnect and CoreSight debug subsystems

• Fast and convenient graphical stitching of AMBA interfaces

SoC Design

IP evaluation

Architecture exploration

Design & Prototyping

Production

Page 13: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

13 © 2018 Arm Limited

• Accelerate go-to-market by developing software on a virtual prototype

• Extensive library of functionally-accurate models

• Deep visibility into your system

• Functional accuracy: validated against RTL

• 10-100x faster than emulators

• Integration in SystemC/EDA environments

SoC Design

IP evaluation

Architecture exploration

Design & Prototyping

Production

Page 14: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

14 © 2018 Arm Limited

SoC Design

IP evaluation

Architecture exploration

Design & Prototyping

Production

Arm trainingLearn your way, anytime

Face to face, Live virtual and Online training enables a blended approach to learning, allowing you to learn in the way that suits you best.

Arm ApprovedFind the right partner

The Arm Approved program enables a rich ecosystem of design partners and training partners, providing our licensees with confidence in selecting partners to trust with their project success.

Design ReviewsGet your design right first time

Arm experts can review customer’s designs at key points in their projects. Catching potential errors at the right time avoids costly mistakes and long project delays.

75%Projects had serious issues identified ahead of tapeout

20+Approved design or training

partners worldwide

Page 15: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

15 © 2018 Arm Limited

DesignStart: Fast access to Arm IP and ecosystem

• Cortex-M0 and Cortex-M3 CPUs

• 1,000s of physical IP libraries

• License for $0 upfront, only success-based royalties

DesignStart for custom SoCs/ASICs

Cortex-M1 and Cortex-M3 CPUs

Soft IP in Xilinx FPGA

✓No license fee, no royalties

DesignStart FPGA

Get started today: DesignStart.arm.com

Page 16: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

16Copyright © 2018 Arm TechCon, All rights reserved.

#ArmTechCon

Arm Development Solutions: Embedded & Safety Ready

Servers

SoC Design Embedded & Safety Ready

Mobile & Graphics Servers & HPC

Page 17: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

17 © 2018 Arm Limited

One tool suite. All compute requirements. All design stages.

featuring Keil MDK

Software Architecture DevelopmentIntegration and testing

Deployment and

maintenance

SoC Design

IP evaluation

Architecture exploration

Design & Prototyping

Production

Page 18: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

18 © 2018 Arm Limited

Page 19: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

19 © 2018 Arm Limited

Most complete architectural

support

5000+ CMSIS devices

supported

Up to 30% faster code*

Streamlined workflow

* Based on CoreMark score comparing Arm Compiler 6.11 to 6.6 on Cortex-M33

Page 20: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

20 © 2018 Arm Limited

Safety Ready software stack and tools

Arm Cortex-M processor

User Application code

FuSa RTX RTOS

FuSa C library

Software test library (STL)

Self-test code for run-time verification

FuSaEvent

Recorder

FuSa CMSIS-Core(Arm-Core specific)

CMSIS-Core(device-specific)

Ready-to-use software framework for embedded applications

RTOS SchedulerTime

Thread

Events Mutex

Semaphore

Memory

Safety Read

y Arm

Co

mp

iler

Developer focus on added-value

Page 21: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

21Copyright © 2018 Arm TechCon, All rights reserved.

#ArmTechCon

Arm Development Solutions: Mobile & Graphics

Servers

SoC Design Embedded & Safety Ready

Mobile & Graphics Servers & HPC

Page 22: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

22 © 2018 Arm Limited

Graphics performance analysis and debug

AnalyzeDS-5 Streamline

• Profile CPUs and Mali GPUs

• Timeline

• HW Counters

• OpenCL visualizer

DebugMali Graphics Debugger

• API Trace & Debug

• Vulkan, OpenGL ES, OpenCL

• Debug and improve performance at frame level

OptimizeMali Offline Compiler

• Analyze shader performance

• Command line tool

• Number of cycles

• Registers utilization

Page 23: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

23 © 2018 Arm Limited

Example Streamline comparisonsML workloads on CPU vs GPU

CPU GPU

Page 24: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

24 © 2018 Arm Limited

Arm Development Solutions: Servers & HPC

Servers

SoC Design Embedded & Safety Ready

Mobile & Graphics Servers & HPC

Page 25: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

25 © 2018 Arm Limited

Debug and performance analysis on very large systems

• The Arm DDT debugger and MAP performance analyser are designed specifically for HPC systems with tensof thousands of nodes. They focus on common issues in HPC

• Variable comparisons across threads

• Identification of over-subscribed cores and sleeping threads

• Synchronization, communication and workload imbalance for MPI and I/O

Page 26: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

26 © 2018 Arm Limited

Arm Performance Libraries for HPC applications

• Optimized BLAS, LAPACK and FFT routines to boost the performance of typical HPC scientific applications

• Validated with NAG’s test suite, a de-facto standard

• Maintained and supported by Arm for a wide range of 64-bit Armv8-A based platforms

• Tuned for Armv8-A partner cores like Cavium in collaboration with Silicon Partners

DGEMM performance compared to OpenBLAS

Page 27: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

27 © 2018 Arm Limited

© 2

01

8A

rm L

imit

ed

Three things to remember from this talk

Software development complexity and cost continues to grow Arm provides wide

range of products and services to boost

engineering efficiency

It takes an ecosystem

Page 28: Beyond Hardware IP - armkk-event.com · 2 © 2018 Arm Limited IC design complexity and cost aren’t slowing down 51.3 106.3 174.4 297.8 542.2 28nm 16nm 10nm 7nm 5nm n) Process node

2828

The Arm trademarks featured in this presentation are registered trademarks or trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere. All rights reserved. All other marks featured may be trademarks of their respective owners.

www.arm.com/company/policies/trademarks

© 2018 Arm Limited