ICCAD-2002 Open Source Panel Andrew B. Kahng UC San Diego CSE

1
ICCAD-2002 Open Source PanelAndrew B. KahngUC
San Diego CSE ECE Depts.Igor L. MarkovUniv.
of Michigan EECS Dept.abk_at_ucsd.eduhttp//vlsica
d.ucsd.edu/abk/TALKS/
2
Experience With Open Source Project

• Project UCLA Physical Design Tools (MLPart,
  Capo, )
• Open-source partitioning and placement tools
  MIT license
• LEF/DEF interface, runs on Linux and Solaris (
  Windows)
• Students Andrew Caldwell (Simplex/Cadence),
  Igor Markov (Univ. of Michigan) Saurabh Adya
  (Univ. of Michigan)
• Much effort in installation scripts, revision
  control, code clarity
• Ongoing maintenance, improvements Capo8.5 (2002)
  gtgt Capo8.0 (2000)
• Academia CMU, GIT, UCLA, UCSD, UCSB, UCSC,
  UMinn, UMich,
• Industry IBM, Intel, Cadence, Philips,
  Synplicity, Ammocore,
• Available in MARCO GSRC Bookshelf, openeda.org,
  etc.
• GSRC Bookshelf http//gigascale.org/bookshelf/
• Repository for CAD-IP reuse
• One of three initiatives to close Design
  Technology Productivity Gap

3
Experience With Open Source Project

• Uses in Academia
• Work on competing placers (20th C TimberWolf
  21st C Capo)
• UCLA, ICCAD-2000 Binghamton, DAC-2001, etc.
• Improvement by pre- or post-processing (UCSB,
  ICCAD-2002)
• Embed into larger design flows (Michigan,
  ISPD-2002)
• Evaluate home-grown sub-solvers (plug in place of
  Capo components)
• Uses in Industry
• Use source code for prototyping commercial tools
  (e.g., Cadence)
• Benchmark internal tools (e.g., Intel, IBM)
• Educational uses (IBM, Intel, Synplicity)
• Plan B uses (IBM, Synplicity)
• Component-based plug-and-play (Intel)
• Try novel PD flows, e.g., use Capo before
  synthesis

4
Problems and Frustrations

• Nothing unexpected
• Very non-trivial software expertise required
• OS, compilers, C, software engineering
• Serious level of commitment required
• Students graduate !
• Non-trivial effort to support platforms,
  configurations
• E.g., Magma was using an older C compiler we
  could not support
• Compilers are changing all the time
• Unfair comparisons and claims
• Make simple preprocessor ? claim new tool with
  better results
• Tune competing software to specially-made
  benchmarks
• Tune benchmarks to competing software and avoid
  standard test cases
• Very few others are open-sourcing slow culture
  change

5
How We Overcame Problems/Frustrations

• Perseverance
• Leveraging the open-source nature of the project
• We know it is a win
• Better software is developed
• Easier to get funding
• Useful in teaching
• Motivates graduate students
• Keeps our research moving (no artificial barrier
  to competition)
• DAC-2000 Capo paper is the most-cited VLSI CAD
  paper from 2000, according to Citeseer
• Problems and Frustrations are NOT an issue
  Open Source Is The Right Thing To Do