Search: 
 
View
  By topic
  As outline
  Fully expanded

Forum topics
  compiler-infrastructure
  arch-summaries
  compiler-summaries
  moc's
  network-drivers
  Presentations
  Princeton Presentation
  Simulator
  Reading List
  Design Examples
  Architecture Discussion
  Related Conferences
  Downloads
  DEPRECATED
  DEPRECATED
  Taxonomies

Articles
  Intro
  SUIF
  IMPACT
  Trimaran
  EDG
  Apogee
  SUIF vs. Trimaran
Home
People
Publications
Multiprocessor Architecture Watch
Concurrency Abstraction Watch
Laboratory
 
Modern Embedded Systems: Compilers, Architectures, and Languages
Forum
Compiler infrastructure evaluation
Previous topic  |  This topic  |  Next topic
Previous article  |  This article  |  Next article

Trimaran 2.0 Evaluation
Ashok Sudarsanam, 13 Sep 1999

The Trimaran 2.0 compiler infrastructure, which was recently made publicly available,  is a joint effort between HP Labs, the University of Illinois, and New York University.  The front-end of Trimaran is IMPACT 2.33a, which was developed by the University of Illinois.  The back-end of Trimaran is called Elcor, and was developed by HP Labs.  NYU has developed a graphical user interface for the Trimaran package.  Trimaran 2.0 possesses the following advantages:
 

  • Elcor back-end very usable, written in C++ (very SUIF-like)
  • powerful suite of machine-independent code optimizations (from IMPACT)
  • powerful suite of VLIW data structures and code optimization algorithms
  • GUI extremely useful for experimentation, can be easily extended
  • documentation fairly reasonable (much better than IMPACT)
  • good support team
  • we can leverage of Bob Rau's CAR Research Group at HP Labs


Trimaran possesses just a couple of disadvantages:
 

  • Elcor currently supports PlayDoh, which is a parameterized non-clustered VLIW architecture.  Thus, Elcor may need to be modified substantially if we wish to target architectures that differ significantly from PlayDoh (e.g. clustered VLIW architectures)
  • Only Linux and HP-UX ports are currently available -- this is not really  a disadvantage

 
 
Previous topic  |  This topic  |  Next topic
Previous article  |  This article  |  Next article
Send feedback to mescal@gigascale.org
You are not logged in
©1998-2009 GSRC