Increasing  programmability with  domain-­‐specific  languages  

 Kris6an  Sandahl,  Peter  Fritzson,  Christoph  Kessler  

PELAB  Department  of  Computer  and  Informa6on  Science  

Linköping  University  

PELAB  at  a  Glance

•  Programming  environments  laboratory,  founded  in  1981  •  Staff:  5  professors,  4  assistant  professors,  11  phd  students,  4  engineers  •  Current  topics:  

•  Compiler  technology    •  SoQware  composi6on    •  Parallel  compu6ng    •  Design  and  implementa6on  of  programming  and  modeling  languages  and  systems    •  Equa6on-­‐based  object-­‐oriented  modeling  and  simula6on  languages    •  Large-­‐scale  soQware  engineering    •  Requirements  engineering    •  SoQware  tes6ng  and  debugging  

 

Problem

• Produc6vity  and  Quality  of  human-­‐made  soQware  

Means Raise  the  abstrac6on  level  

Model-­‐based  SoQware  Engineering  

Domain-­‐specific  Languages   Skeletons  and  paWerns  

SoQware  models   System  models  

MODELICA  SySML  

Simula6on   Visualiza6on   Valida6on  

Parallel  programming  and  mul6core  processing  

SkePU  skeleton  programming  Sequen6al  interface  

Heterogeneous  systems   Pla[orm  agnos6c   High-­‐level  constructs  

Use  

September 25, 2014 STEW Workshop Peter Fritzson Professor at Linköping University, Sweden Vice Chairman of Modelica Association Director of Open Source Modelica Consortium Senior Member, IEEE peter.fritzson@liu.se

Modelica – A Cyber-Physical Modeling Language for Systems Engineering and the OpenModelica

Environment

6

What is Modelica?

•  Robotics •  Control •  Automotive •  Aircraft •  Satellites •  Power plants •  Systems biology

A language for modeling of complex cyber-physical systems

7

What is Special about Modelica?

Multi-Domain Modeling

Cyber-Physical Modeling

Physical

Cyber

3 domains -  electric -  mechanics -  control

8

inertialxy

axis1

axis2

axis3

axis4

axis5

axis6r3Drive1

1r3Motor

r3ControlqdRef1

S

qRef1

S

k2

i

k1

i

qddRef cut joint

q: angleqd: angular velocity

qdd: angular acceleration

qd

tn

Jmotor=J

gear=i

spring=c

fric=Rv0

Srel

joint=0

S

Vs

-

+diff

-

+pow er

emf

La=(250/(2*D*wm))

Ra=250

Rd2=100

C=0.004*D/w m

-

+OpI

Rd1=100

Ri=10

Rp1=200

Rp2=

50

Rd4=100

hall2

Rd3=

100

g1

g2

g3

hall1

g4

g5

rw

cut in

iRef

qd q

rate2

b(s)

a(s)

rate3

340.8

S

rate1

b(s)

a(s)

tacho1

PT1

Kd

0.03

wSum

-

sum

+1

+1

pSum

-

Kv

0.3

tacho2

b(s)

a(s)

q qd

iRefqRef

qdRef

Courtesy of Martin Otter

What is Special about Modelica?

Visual Acausal Hierarchical Component

Modeling

Multi-Domain Modeling

Hierarchical system modeling

Srel = n*transpose(n)+(identity(3)- n*transpose(n))*cos(q)- skew(n)*sin(q); wrela = n*qd; zrela = n*qdd; Sb = Sa*transpose(Srel); r0b = r0a; vb = Srel*va; wb = Srel*(wa + wrela); ab = Srel*aa; zb = Srel*(za + zrela + cross(wa, wrela));

9

Graphical and Textual View

•  A DC motor can be thought of as an electrical circuit which also contains an electromechanical component model DCMotor Resistor R(R=100); Inductor L(L=100); VsourceDC DC(f=10); Ground G; ElectroMechanicalElement EM(k=10,J=10, b=2); Inertia load; equation connect(DC.p,R.n); connect(R.p,L.n); connect(L.p, EM.n); connect(EM.p, DC.n); connect(DC.n,G.p); connect(EM.flange,load.flange); end DCMotor

load

EM DC

G

R L

10

Application of Modelica in Robotics Models Real-time Training Simulator for Flight, Driving

•  Using Modelica models generating real-time code

•  Different simulation environments (e.g. Flight, Car Driving, Helicopter)

•  Developed at DLR Munich, Germany

•  Dymola Modelica tool

Courtesy of Martin Otter, DLR, Oberphaffenhofen, Germany

11

Modelica Standard Library Open Source, Developed by Modelica Association

•  Blocks Library for basic input/output control blocks •  Constants Mathematical constants and constants of nature •  Electrical Library for electrical models •  Icons Icon definitions •  Fluid 1-dim Flow in networks of vessels, pipes, fluid machines, valves, etc. •  Math Mathematical functions •  Magnetic Magnetic.Fluxtubes – for magnetic applications •  Mechanics Library for mechanical systems •  Media Media models for liquids and gases •  SIunits Type definitions based on SI units according to ISO 31-1992 •  Stategraph Hierarchical state machines (analogous to Statecharts) •  Thermal Components for thermal systems •  Utilities Utility functions especially for scripting

The Modelica Standard Library contains components from various application areas, including the following sublibraries:

12

•  Advanced Interactive Modelica compiler (OMC) •  Supports most of the Modelica Language •  Modelica and Python scripting

•  Basic environment for creating models •  OMShell – an interactive command handler •  OMNotebook – a literate programming notebook •  MDT – an advanced textual environment in Eclipse

12

•  OMEdit graphic Editor •  OMDebugger for equations •  OMOptim optimization tool •  OM Dynamic optimizer collocation •  ModelicaML UML Profile •  MetaModelica extension •  ParModelica extension

The OpenModelica Open Source Environment www.openmodelica.org

13

Industrial members •  Bosch Rexroth AG, Germany • Siemens PLM, California, USA • Siemens Turbo, Sweden •  CDAC Centre, Kerala, India •  Creative Connections, Prague •  DHI, Aarhus, Denmark •  EDF, Paris, France •  Equa Simulation AB, Sweden •  Fraunhofer IWES, Bremerhaven •  Frontway AB, Sweden •  IFP, Paris, France

Open-source community services •  Website and Support Forum •  Version-controlled source base •  Bug database •  Development courses •  www.openmodelica.org

Code Statistics •  Austrian Inst. of Tech, Austria •  Linköping University, Sweden •  UC Berkeley, USA • TU Berlin, Insti UEBB, Germany •  FH Bielefeld, Bielefeld, Germany • TU Braunschweig, Germany •  Univ Calabria, Italy •  Danish Technical Univ, Denmark • TU Dortmund, Germany • TU Dresden, Germany •  Université Laval, Canada •  Georgia Inst. Technology, USA

University members

OSMC – International Consortium for Open Source Model-based Development Tools, 46 members Mar 2014

Founded Dec 4, 2007 •  GTI, USA •  ISID Dentsu, Tokyo, Japan •  ITI, Dresden, Germany •  Maplesoft, Canada •  Ricardo Inc., USA •  STEAG, Dehli, India • TLK Thermo, Germany • Sozhou Tongyuan, China •  VTI, Linköping, Sweden •  VTT, Finland •  Wolfram MathCore, Sweden

•  Ghent University, Belgium •  Halmstad University, Sweden •  Heidelberg University, Germany •  TU Hamburg/Harburg Germany •  KTH, Stockholm, Sweden •  Univ of Maryland, Syst Eng USA •  Univ of Maryland, CEEE, USA •  Politecnico di Milano, Italy •  Ecoles des Mines, CEP, France •  Mälardalen University, Sweden •  Univ Pisa, Italy • Telemark Univ College, Norway

14

OpenModelica MDT Eclipse Plug-in with OpenModelica Algorithmic Code Debugger

15

General Tool Interoperability & Model Exchange Functional Mock-up Interface (FMI)

•  FMI development initiated by Daimler •  Improved Software/Model/Hardware-in-the-Loop Simulation, of physical

models and of AUTOSAR controller models from different vendors for automotive applications with different levels of detail.

•  Open Standard, standardized by Modelica Association •  14 automotive use cases for evaluation •  > 40 tool vendors are supporting it

The FMI development a result of the MODELISAR 29-partner project

Engine with ECU

Gearbox with ECU

Thermal systems

Automated cargo door

Chassis components, roadway, ECU (e.g. ESP)

etc.

functional mockup interface for model exchange and tool coupling courtesy Daimler

16

Simulation and Requirements Checking Tank system example

Req. 001 is instantiated 2 times (there are 2 tanks in the system)

tank-height is 0.6m

Req. 001 for the tank2 is violated

Req. 001 for the tank1 is not violated

17

ParModelica – Modelica Parallel Algorithmic Programming Execution on GPUs and CPUs via portable OPENCL code generation

Gained speedup •  Intel Xeon E5520 CPU (16 cores) 26 •  NVIDIA Fermi-Tesla M2050 GPU (448 cores) 115

Speedup comparison to sequential algorithm on Intel Xeon E5520 CPU

32 64 128 256 512 CPU E5520 (Serial) 0.093 0.741 5.875 58.426 465.234

CPU E5520 (Parallel) 0.137 0.17 0.438 2.36 17.66

GPU M2050 (Parallel) 1.215 1.217 1.274 1.625 4.057

0.0625 0.125

0.25 0.5

1 2 4 8

16 32 64

128 256 512

Sim

ulat

ion

Tim

e (s

econ

d)

4.36 13.41 24.76 26.34 0.61 4.61

35.95

114.67

64 128 256 512 Parameter M (Matrix sizes MxM)

Speedup CPU E5520 GPU M2050

Matrix Multiplication application

18

Get More Information, Download Software

Peter Fritzson Principles of Object Oriented Modeling and Simulation with Modelica 3.3 A Cyber-Physical Approach Wiley-IEEE Press, 2014 1250 pages

•  OpenModelica •  www.openmodelica.org

•  Modelica Association •  www.modelica.org

The image cannot be displayed. Your computer may not have enough memory to open the image, or the image may have been corrupted. Restart your computer, and then open the file again. If the red x still appears, you may have to delete the image and then insert it again.High-Level Programmingfor Heterogeneous Systems

using Skeletons*"Christoph Kessler

Linköping University

Sweden

*This research is funded by EU FP7 projects PEPPHER and EXCESS, and by SeRC"

EXCESS

20" "

The Computer Engineer’s Solution Today: Heterogeneous Multicore"

Programmable Hardware Accelerator

"

Offload certain

computation#

Data transfer#

###

Device#memory#

Core#

Core#

Core#

Core#

Main Memory#

J More energy-efficient than general-purpose sequential resp. general-purpose multicore CPU#J Hardware-tailored parallel programming models#J Detailed resource control#

General-purpose" CPU"

21" "

The Application Programmer’s Nightmare: Heterogeneous Multicore"

Programmable Hardware Accelerator

"

Offload certain

computation#

Data transfer#

###

Device#memory#

Core#

Core#

Core#

Core#

Main Memory#

L Distributed Memory or Non-coherent Shared Memory#L  Different Programming Models#L  Low Abstraction Level#L  Require Architecture- specific Optimization#

General-purpose" CPU"

22" "

Common: GPU-Based Systems"

CPU"

GPU "

Offload heavy

computation#

Data transfer#

###

Device#memory#

Core#

Core#

Core#

Core#

Main Memory#

L Distributed Memory#L  Different Programming Models#L  Low Abstraction Level#L  Require Architecture- specific Optimization#

23" "

Common: GPU-Based Systems"

CPU"

GPU "

Offload heavy

computation#

Data transfer#

###

Device#memory#

Core#

Core#

Core#

Core#

Main Memory#

###

Device#memory#

…#GPU "

L Distributed Memory#L  Different Programming Models#L  Low Abstraction Level#L  Require Architecture- specific Optimization#

24" "

Programming of GPU-based Systems … with OpenCL™ (?)- Code portability J - Programmability L (low level) - Performance portability L (requires reoptimization)#

CPU"

GPU "

Offload heavy

computation#

Data transfer#

###

Device#memory#

Core#

Core#

Core#

Core#

Main Memory#

25" "

Skeleton Programming"

  A skeleton is a pre-defined generic software component that # Models a common computation / dependence pattern#

à well-defined semantics; metadata implicit# Can be customized with sequential user code" Provides a sequential interface# Encapsulates all platform-specific implementation details

(parallelism, heterogeneity, memory management, …)#

26" "

SkePU www.ida.liu.se/~chrke/skepu#

  C++ template library targeting GPU-based systems#  6 dataparallel skeletons#

  Map, Reduce, Scan, MapReduce, MapArray, MapOverlap (stencil)#

  1 task-parallel skeleton: farm#  STL-based containers wrapping operand data#

  Smart containers Vector<…>, Matrix<…>optimizing data transfers and memory management at runtime#

  Generation of platform-specific variants for user functions#  Multiple back-ends: C, OpenMP, OpenCL, CUDA, StarPU, (MPI)#

  Hybrid CPU-GPU execution (with StarPU backend)#  Multi-GPU support #

  Low overhead#  Auto-tunable [Dastgeer et al. IWMSE’11, MULTIPROG’12, APPT’13; PhD’14]#  Open-source#

27" "

Summary"  Domain-specific languages#

 High level of abstraction# Automatic code generation#

  Model-based development of Cyber-Physical Systems# OpenModelica open-source environment#

  Skeleton programming # Parallelism, heterogeneity, communication managed internally# Performance, portability and automatic optimizations for free!#