Sim, Render, Repeat

1
Sim, Render, Repeat

• An Analysis of Game Loop Architectures
• Daniel Martin and Michael Balfour

March 23, 2006
2
Who?

• Introduction

3
Introduction The Speakers

• Technical directors at EA-Tiburon
• MS degrees
• 7 years in the game industry
• Exposed to many games and architecture
• Also programmed on Apple s and mainframes ?

4
Introduction Motivation

• Hidden complexities
• Little coverage
• Highest-level game architecture

5
Introduction Agenda

• Definitions
• Timeline
• Architecture decisions
• Case study
• Questions
• Slides will be made available on the GDC website

6
What?

• Game loop architectures defined

7
Definition Background

• Source Code
• Digger
• Duke3D
• Freespace2
• Harry Potter
• Madden
• Books
• 3D Game Engine Architecture, First Edition
• Core Techniques and Algorithms in Game
  Programming
• Game Coding Complete, Second Edition

• NASCAR
• Popcap Engine
• Quake2
• Unreal Tournament v432
• Wumpus

8
Definition Pseudocode pattern
GameLoop() Startup() while (!done)
GetInput() Sim()
Render() Shutdown()
9
Definition Formal

• Game Loop
• A thread of execution consisting of
• a startup phase,
• a looping phase that processes inputs/outputs,
• a shutdown phase.
• We use Game Loop and Loop interchangeably

10
Definition Formal

• Video Game
• A collection of one or more game loops processing
  inputs and outputs for entertainment purposes.

11
Definition Formal

• Game Loop Architecture
• An architecture consisting of one or more game
  loops and the infrastructure to manage their
  execution and dependencies.

12
When?

• Game Loop Timeline

13
1947 Cathode Ray Tube Amusement Device
1950
1960
1940
1970
1980
1990
2000

• First patented video game
• Game Fire a missile at a target
• Hardware loop run by master clock

14
1952 Noughts and Crosses
1950
1960
1970
1980
1990
2000
1940

• First software-based game
• Game Tic Tac Toe
• Software loop

15
1940-1970 Analysis

• Game loops have existed for 50 years
• We still use the same game loop patterns today
• Seems to be the most intuitive architecture for
  games
• Hardware and software loops
• Independently developed
• Similar results

16
1975 Gun Fight
1950
1960
1940
1970
1980
1990
2000

• First CPU-based arcade video game
• Game Two players shoot at each other
• Software Sim loop
• Hardware Render and Audio loops

17
1978 Gypsy Juggler
1950
1960
1940
1970
1980
1990
2000

• First multi-CPU game
• Game Move a gypsy and juggle eggs
• Concurrent Software Sim/Audio loops
• Two symmetric S2650 CPUs

18
1970-1980 Analysis

• Multi-CPU arcade games have existed for 28 years
• Multi-CPU common in arcade games
• Typically 2-3 CPUs
• Common use of specialized processors
• Moved hardware Render/Audio Loops to software

19
1988 Genesis
1950
1960
1940
1970
1980
2000
1990

• First multi-CPU console
• 68000 CPU for Sim/Render
• Z80 CPU for Audio

20
1994 Saturn
1950
1960
1940
1970
1980
1990
2000

• First symmetric Multi-CPU console
• Two Hitachi RISC processors
• 6 processors for video/sound/control

21
1980-2000 Analysis

• Multi-CPU Consoles have existed for 18 years
• Many consoles with asymmetric CPUs
• Few consoles with symmetric CPUs
• One very fast central processor would be
  preferable I think only one out of 100
  programmers is good enough to get that kind of
  speed out of the Saturn. Yu Suzuki

22
2003 MHz Barrier
1950
1960
1940
1970
1980
2000
1990
CPU frequency 1995-2006
MHz
Year
23
2000-2006 Analysis

• Paradigm Shift
• The Free Lunch is Over
• Parallel by necessity
• Legacy code wont run magically faster
• More performance requires mastering concurrency
• Single CPU no longer the norm

24
How?

• Game Loop Architecture Decisions

25
Architecture Game loop complexity
Concurrency
Coupling
Time
Complexity
26
Architecture Game loop complexity
Concurrency
Coupling
Time
Complexity
27
Architecture Time

• Problem
• Running smoothly in real-time
• Solution
• Frequency-Driven Game Loop
• Divide time into discrete iterations
• Attempt to run fast enough and smooth enough
• Consequence
• Each Loop iteration is a time slice

28
Architecture Time

• Challenge
• How to maintain a loop frequency with variable
  execution time
• Factors
• Performance
• Tolerance
• Architecture Decisions
• Scheduling
• Time Step

• Determinism
• Simplicity

29
Architecture Time

• Scheduling
• Controls when a loop iteration starts

30
Architecture Legend

• Time Axis
• Real time
• Ticks at desired frequency
• Box
• Loop iteration
• Length is processing time
• Arrows
• Time step
• Length is simulated time
• Data
• Data used by a loop

31
Architecture Time
Scheduling

• Description
• As fast as it can
• Factors
• Performance
• - Tolerance
• Determinism
• Simplicity

Immediate
Exact
Faster
Slower
Varied
32
Architecture Example
Scheduling Immediate
while (1) GetInput() Sim()
Render()

• Early adventure games

33
Architecture Time
Scheduling

• Description
• Averaging
• Tries to
• Maintain frequency
• Start at exact times
• Catch up
• Factors
• Performance
• Tolerance
• Determinism
• - Simplicity

Best-Fit
Exact
Faster
Slower
Varied
34
Architecture Time
Scheduling

• Description
• VSynced
• Factors
• - Performance
• - Tolerance
• Determinism
• Simplicity

Aligned
Exact
Faster
Slower
Varied
35
Architecture Time

• Time Step
• Controls interpretation of time
• Simulated time

36
Architecture Time
Time Step

• Description
• Ignores time
• Factors
• Performance
• Tolerance
• Determinism
• Simplicity

None
Exact
Faster
Slower
Varied
37
Architecture Time
Time Step

• Description
• Time step always constant
• Factors
• Performance
• - Tolerance
• Determinism
• - Simplicity

Fixed-Value
Exact
Faster
Slower
Varied
38
Architecture Example
Scheduling Time Step
Best-Fit Fixed-Value
const float TIMESTEP 1 / 60.0f while (1)
curTime GetTime() if ((curTime lastTime)
gt TIMESTEP)
Sim(TIMESTEP) lastTime curTime

• Deterministic real-time games

39
Architecture Example
Scheduling Time Step
Immediate Fixed-Value
const float TIMESTEP 1 / 60.0f while (1)
Sim(TIMESTEP)

• Early PC games, assumed CPU bottleneck

40
Architecture Time
Time Step

• Description
• Sim Time Real Time
• Factors
• Performance
• Tolerance
• - Determinism
• - Simplicity

Real-Time
Exact
Faster
Slower
Varied
41
Architecture Example
Scheduling Time Step
Aligned Real-Time
while (1) WaitForVSync() curTime
GetTime() step curTime lastTime
Clamp(step, 0, MAX_STEP) Render(step)
lastTime curTime

• Interpolating Render loop, aligned to VSync

42
Architecture Game Loop Complexity
Concurrency
Coupling
Time
Complexity
43
Architecture Coupling

• Problem
• Supporting systems at different frequencies
• Solution
• Multiple game loops
• Consequence
• Loop Coupling
• Dependencies between every pair of loops

44
Architecture Coupling

• Challenge
• How to split code and data into multiple loops
• Factors
• Performance
• Tolerance
• Simplicity
• Architecture Decisions
• Frequency Coupling
• Data Coupling

• Video Modes
• Memory
• Scalability

45
Architecture Coupling

• Frequency
• How much a loop relies on anothers frequency

46
Architecture Coupling
Frequency

• Description
• 11
• Lockstep
• Factors
• - Video modes
• Memory
• Performance
• - Tolerance
• - Scalability
• Simplicity

Equal
Time (Loop 1)
Time (Loop 2)
47
Architecture Coupling
Frequency

• Description
• N1
• Multiple of frequency
• Sim _at_ 60 Hz
• Render _at_ 30 Hz
• Factors
• - Video modes
• Memory
• Performance
• - Tolerance
• - Scalability
• Simplicity

Multiple
Time (Loop 1)
Time (Loop 2)
48
Architecture Coupling
Frequency

• Description
• NM
• Two independent rates
• Factors
• Video modes
• - Memory
• Performance
• Tolerance
• Scalability
• - Simplicity

Decoupled
Time (Loop 1)
Time (Loop 2)
49
Architecture Example
Scheduling Time Step Frequency
Aligned Fixed-Value Decoupled
Best-Fit Fixed-Value Decoupled
Scheduling Time Step Frequency
while (1) time GetTime() if ((time
lastTime) gt SIM_STEP)
Sim(SIM_STEP) lastTime time
while (1) Render(RENDER_STEP)
WaitForVSync()

• Decoupled Sim and Render loops

50
Architecture Coupling

• Data Coupling
• The amount and method of sharing data

51
Architecture Coupling
Data Coupling

• Description
• Data structure reliance
• Factors
• Video modes
• Memory
• Performance
• Tolerance
• - Scalability
• - Simplicity

Tight
52
Architecture Coupling
Data Coupling

• Description
• Formalized data passing
• Factors
• Video modes
• - Memory
• - Performance
• Tolerance
• Scalability
• Simplicity

Loose
53
Architecture Example
Scheduling Time Step
Frequency Data Coupling
Aligned N/A
Decoupled Loose
Scheduling Time Step
Best-Fit N/A

• Sim at 60 Hz, Render at 50 Hz

54
Architecture Coupling
Data Coupling

• Description
• Fully independent
• No data passing
• Factors
• Video modes
• Memory
• Performance
• Tolerance
• Scalability
• Simplicity

None
55
Architecture Game loop complexity
Concurrency
Coupling
Time
Complexity
56
Architecture Concurrency

• Problem
• Hardware makers need cutting edge performance
• Solution
• Hardware contains multiple CPUs
• Consequence
• Concurrency
• Mapping between game loops and CPUs

57
Architecture Concurrency

• Challenge
• How to manage simultaneous execution
• Factors
• Performance
• Simplicity
• Scalability
• Architecture Decisions
• Low-Level Concurrency
• High-Level Concurrency

58
Architecture Concurrency

• Low-Level
• Parallelism within a game loop

Low-Level
None
Instruction
Functions
59
Architecture Concurrency

• Low-Level
• Covered extensively
• Easiest transition to NextGen
• Start small and grow
• OpenMP
• Bottom up approach
• Can be postponed

60
Architecture Concurrency

• High-Level
• Parallelism across a pair of game loops

High-Level
Sequential
Interleaved
Parallel
61
Architecture Concurrency
High-Level

• Description
• No overlap of loop execution
• Factors
• - Performance
• - Scalability
• Simplicity

Sequential
Exact
Varied
62
Architecture Concurrency
High-Level

• Description
• Loops iterate in parallel
• Sequential at instruction level
• Concurrency on 1 CPU
• Factors
• - Performance
• Scalability
• - Simplicity

Interleaved
Exact
Varied
63
Architecture Concurrency
High-Level

• Description
• Instructions execute at same time
• Factors
• Performance
• Scalability
• - Simplicity

Parallel
Exact
Varied
64
Why?

• Case Study

65
Case Study Madden Xbox vs Madden X360
VS
66
Case Study Madden Xbox vs Madden X360
Sim Render Audio
Scheduling
Immediate
Best-Fit
Aligned
Sim
Render Audio
67
Case Study Madden Xbox vs Madden X360
Audio
Sim Render
Sim
Render Audio
68
Case Study Madden Xbox vs Madden X360
Sim-Audio Render-Audio
Sim-Render
Frequency Coupling
Equal
Multiple
Decoupled
Sim-Render Sim-Audio Render-Audio
69
Case Study Madden Xbox vs Madden X360
Sim-Render
Render-Audio
Sim-Audio
Render-Audio
Sim-Render Sim-Audio
70
Case Study Madden Xbox vs Madden X360
Sim Render Audio
Low-Level Concurrency
None
Instruction
Functions
Sim Render Audio
71
Case Study Madden Xbox vs Madden X360
Sim Render
Audio
High-Level Concurrency
Sequential
Interleaved
Parallel
Sim Render Audio
72
!

• Conclusion

73
Conclusion

• Hidden complexities
• Make key architecture decisions early
• Our decisions
• Use multiple game loops
• Decouple data

74
???

• Questions?

Slides, speaker notes and references
available http//www.gdconf.com
75
Where?

• References

76
References Literature

• Game Loops in Literature
• 3D Game Engine Architecture, First Edition,
  David Eberly
• http//www.elsevier.com/wps/find/bookdescription.c
  ws_home/704123/descriptiondescription
• Core Techniques and Algorithms in Game
  Programming, Daniel Sanchez-Crespo
• http//www.novarama.com/users/dani/docs/book/index
  -en.shtml
• Game Coding Complete, Second Edition, Mike
  McShaffry
• http//www.mcshaffry.com/GameCode/
• Task Manager Papers
• Game Engineering for a Multiprocessor
  Architecture, Abdennour El Rhalibi, Dave
  England, and Steven Costa
• http//www.gamesconference.org/digra2005/papers/a4
  458b374e5b16f16e725fdbeed4.doc
• Compositional Constraints Generation for
  Concurrent Real-Time Loops with Interdependent
  Iterations, I. Assayad and S. Yovine
• http//www-verimag.imag.fr/yovine/articles/i2cs05
  .pdf
• Hierarchical Finite State Machines with Multiple
  Concurrency Models, Alain Girault, Bilung Lee,
  and Edward A. Lee
• http//ptolemy.eecs.berkeley.edu/papers/99/starcha
  rts/starcharts.pdf

77
References Video Game History

• Patent 2455992 Cathode Ray Tube Amusement
  Device, Thomas Goldsmith Jr and Estle Ray Mann
• http//www.pong-story.com/2455992.pdf
• PONG-Story, David Winter
• http//www.pong-story.com/intro.htm
• The Edsac Simulator, Martin Campbell-Kelly
• http//www.dcs.warwick.ac.uk/edsac/
• The First Video Game, Brookhaven National
  Laboratory
• http//www.bnl.gov/bnlweb/history/higinbotham.asp
• Gun Fight, Midway Mfg. Co.
• http//www.arcadedocs.com/vidmanuals/G/gunfight_2.
  pdf
• Atari Jaguar, Wikipedia
• http//en.wikipedia.org/wiki/Atari_JaguarTechnica
  l_specifications
• SEGA SATURN F.A.Q., John Hokanson Jr.
• http//www.classicgaming.com/saturn/content/vault/
  faqs/saturn/system/saturnfaqb.txt

78
References Concurrency

• The Free Lunch is Over A Fundamental Turn
  Toward Concurrency in Software, Herb Sutter
• http//www.gotw.ca/publications/concurrency-ddj.ht
  m
• Software and the Concurrency Revolution, Herb
  Sutter and James Larus
• http//www.acmqueue.org/modules.php?nameContentp
  ashowpagepid332page1
• Debugging Concurrency, Philippe Paquet
• http//www.gamasutra.com/features/20050606/paquet_
  01.shtml
• Real-World Case Studies Threading Games for
  High Performance on Intel Processors, Sara
  Sarmiento
• http//cache-www.intel.com/cd/00/00/20/40/204080_2
  04080.pdf
• Trials and Tribulations of Debugging
  Concurrency, Kang Su Gatlin
• http//acmqueue.com/modules.php?nameContentpash
  owpagepid226
• Software Challenges in Nanoscale Technologies,
  James Larus
• http//research.microsoft.com/larus/Talks/CRA20A
  rchitecture20Challenge.ppt
• Threading 3D Game Engine Basics, Henry Gabb and
  Adam Lake
• http//www.gamasutra.com/features/20051117/gabb_01
  .shtml
• Compiler Technology for Scalable Architectures,
  Alexandre E. Eichenberger, Kevin OBrien, and
  Kathryn OBrien
• http//domino.research.ibm.com/comm/research_proje
  cts.nsf/pages/cellcompiler.index.html
• The Next Mainstream Programming Language A Game
  Developers Perspective, Tim Sweeney
• http//www.cs.princeton.edu/dpw/popl/06/Tim-POPL.
  ppt

79
References Public Source Code

• Digger, Windmill Software
• http//www.digger.org/download.html
• Duke Nukem 3D Source Code, 3D Realms
• http//www.3drealms.com/downloads.html
• Freespace2 Source Code, Volition Software
• http//www.gamespot.com/pc/sim/freespace2/download
  .html?sid2862847
• PopCap Engine, PopCap Games
• http//developer.popcap.com/
• Quake II Source Code, Id Software
• http//www.idsoftware.com/business/techdownloads/
  
• Unreal Tournament v432 Source Code, Epic Games
• http//www.fileshack.com/file.x?fid308
• Wumpus 1, Gregory Yob
• http//www.atariarchives.org/morebasicgames/showpa
  ge.php?page178