Stream Processing for Computer Generated Forces

1

• Stream Processing for Computer Generated Forces
• Kickoff Meeting
• Maria Bauer
• RDECOM-STTC

2
Agenda

• Project Summary
• Technical Information
• Team Members and Roles
• Schedule
• Deliverables
• Information needed
• Communication
• Upcoming Events

3
Project Summary GPU Research

• Customer
• DARPA
• User
• PM OneSAF
• Contract Type
• Cost
• Period of Performance
• 3/18/2004 12/18/2005
• Subcontractor
• SAIC

4
Technical Information - SOW Description

• Statement of Work (SOW)
• UNC shall develop new algorithms for real-time
  path and route planning in complex terrain and
  urban environments.
• UNC shall develop GPU-based visibility and
  line-of-sight (LOS) computations for real-time
  interaction and communication between very large
  numbers of entities (thousands).
• UNC shall develop GPU-based real-time collision
  detection for terrain simulation.
• UNC shall leverage early PCA program
  developments, including streaming architecture
  and streaming languages (e.g. Brook developed at
  Stanford)
• Developed algorithms shall be integrated with the
  OneSAF Objective System (OOS).
• UNC shall develop software libraries that can be
  distributed freely to DOD labs and also used for
  other applications.
• UNC shall deliver quarterly management reports.
  Each report shall cover progress attained during
  the reporting period, contract funds expended and
  contract funds remaining.

5
Team Members

• Dinesh Manocha (UNC Chapel Hill)
• Ming C. Lin (UNC Chapel Hill)
• Michael Macedonia (PEO STRI)
• Maria Bauer (RDECOM)
• Audrey Rubel (RDECOM)
• Latika Eifert (RDECOM)
• Angel Rodriguez (RDECOM)
• Brett Butler (SAIC)
• Pat Hanrahan (Stanford)

6
Program Players and Roles

• Task I Use of GPUs to speed-up OneSAF
  computations
• Algorithm Development UNC
• Experiment Design and Data Collection Plan
  UNC/SAIC
• Integrate in OneSAF UNC/SAIC
• Test Performance AnalysisUNC/SAIC
• Final Report UNC

7
Program Players and Roles

• Task II Use of Streaming Languages
  Architectures for OneSAF Computation
• Development Delivery of Brook Stanford
• Redesign visibility, collision avoidance and
  route planning algorithms using Brook
  UNC/Stanford
• OneSAF performance Evaluation using Brook
  UNC/Stanford/SAIC
• Evaluating Brook and Streaming Architectures for
  Computer-Generated Forces UNC/Stanford/SAIC
• Final Report UNC/Stanford

8
Deliverables

• Initial Report
• Report on experiment design and data collection
  plan
• Due 6 months after contract award
• Mid Report
• Detailed architecture of integrating routines
  into OOS. Performance of line-of-sight and
  collision computations using GPUs
• Information regarding progress of language
  development for GPUs
• Due 12 months after contract award
• Final Report
• Will document the comparison of performance
  between the traditional (OOS) CPU centric
  solution versus a GPU centric solution for the
  terrain placement, collision detection and LOS
  services
• The final report will contain all information
  regarding use of Brook for CGF application
• Due 30 days before the end of the contract

9
Schedule
10
Schedule
11
Upcoming Events

• GP2 Workshop on General Purpose computing on
  Graphics Processors
• August 7-8, 2004, Los Angeles
• Participants from computer graphics, computer
  architecture, simulation, high performance
  computing

12

• Questions?