Differences between revisions 1 and 36 (spanning 35 versions)
Revision 1 as of 2008-09-10 20:35:22
Size: 1688
Editor: 163
Comment:
Revision 36 as of 2008-11-13 20:12:22
Size: 4761
Editor: versailles
Comment:
Deletions are marked like this. Additions are marked like this.
Line 1: Line 1:
= Graphite Workshop (around Super Computing'08) = = Graphite Workshop =
A Graphite Workshop is organized on AMD's campus in Austin on November 16 and 17, 2008.
Line 3: Line 4:
A Graphite Workshop is organized on AMD's campus in Austin around the Super Computing conference on November 16 and 17, 2008. === Day 1: Nov 16 ===

 * Tentative schedule:
  * 9am breakfast
   * Introduction (1h)
   * Polyhedral Compilation Package (PCP) (1h)
  * 12pm lunch
   * Front-end (2h)
  * 3pm break
   * Front-end (continued) (3h)
  * 6pm end

=== Day 2: Nov 17 ===

 * Tentative schedule:
  * 9am breakfast
   * Middle-end (3h)
  * 12pm lunch
   * Middle-end (continued) (2h)
  * 3pm break
   * Testing and Performance Evaluation (1h)
   * Community Interaction (1h)
  * 5pm end
Line 6: Line 29:
 * Short term
  * PPL
   * Migration of dependence analysis to PPL.
   * Better integration of PIPLib and Omega Test with PPL.
   * Improve PPL's support for code generation and analysis algorithms (sorting, simplification in context, integers, complexity, improvements, caching).
  * CLooG
   * Backend improvements: unrolling, modulo removal.
   * Algorithmic improvements: parametric tiling.
  * Integration with the array middle end.
  * Minimal but correct handling of function calls in SCoPs.
  * Fuzzy (yet conservative) dependence and array data-flow analysis.
  * Integrate with and extend vectorizer cost model.
  * Cache and locality models.
  * More tolerant SCoP formation (boolean expressions, if-conversion, modulo and division).
  * Support for reductions in the dependence analysis (and parallel code generation).
  * Test suite.
  * Regression testing.
  * Robustness (including algorithmic complexity issues).
Line 25: Line 30:
 * Long term
  * Interprocedural array region analysis (w/ or w/o LTO).
  * Irregular control-flow.
  * Shared-memory multi-processing performance models (NUMA, heterogeneous).
  * Array expansion (privatization, renaming).
  * Array contraction (a.k.a. storage mapping optimization).
 * Introduction: Led by Sebastian Pop, Albert Cohen. (1h)
  * History and current status of Graphite (tools, representations,
    SCoP formation algorithm, gloog pretty-printer, analyses,
    transformations, limitations).
Line 32: Line 35:
 * Besides technical topics, there are a bunch of things to talk about:
  * organization,
  * split of the work and responsibilities,
  * publication policy,
  * integration of external polyhedral tools (at the polyhedral representation level).
 * Polyhedral Compilation Package (PCP): (1h) (Jan Sjodin)

 * Front-end, static analysis and conversion to polyhedral representation. (5h)
  * SCoP formation vs. code generation. (Tobias Grosser) (20min)
  * Data dependence and reaching definition analysis (Sebastian Pop)
   * Instance-wise dependences and reaching definitions. (Albert Cohen) (10min)
   * Fuzzy analysis. (Albert Cohen) (20min)
   * Interprocedural array region analysis (w/ or w/o LTO). (Honza Hubicka) (20min)
  * Polyhedral representations: Led by Albert Cohen, Roberto Bagnara, Sven Verdoolaege
   * Integration with PIPlib and Omega Test. (Cedric Bastoul, Sebastian Pop, Sven Verdoolaege) (20min)
   * Parma Polyhedral Library. (Roberto Bagnara) (40mn)
   * An Integer Set Library for CLooG. (Sven Verdoolaege) (20min)
   * Improvements: sorting, simplification in context, caching, relaxation. (Albert Cohen) (20min)
  * Interaction with the array middle-end. (Richard Guenther) (20min)
  * Support for reductions. (Albert Cohen) (20min)
  * Direct support for irregular control flow. (Cedric Bastoul) (20min)

 * Middle-end: how to add time, space and machine descriptions (5h)
  * Performance Modeling and Heuristics: Harsha Jagasia, Dwarak Rajagopal.
   * Machine descriptions, cache and locality models. (Harsha Jagasia) (20 min)
   * Computing a schedule. (Harsha Jagasia, Sebastian Pop) (20 min)
   * Parametric tiling support. (Sanjay Radjopadhye, Cedric Bastoul, Albert Cohen) (20 min)
   * Direct search in Farkas coefficient spaces. (Albert Cohen, Cedric Bastoul) (20 min)
  * Array expansion and contraction. (Richard Guenther) (20 min)
  * Unrolling support. (Cedric Bastoul) (20 min)
  * Modulo guard elimination support. (Tobias Grosser) (20 min)
  * Automatic parallelization: Albert Cohen, Jan Hubicka, Sebastian Pop.
   * Loop parallelization with interprocedural analysis. (Honza Hubicka, Richard Guenther) (20 min)
   * Loop selection (with interchange and distribution). (Sebastian Pop) (20 min)
   * Streamization (post/wait synchronizations). (Sebastian Pop) (20 min)
   * Automatic extraction of functional parallelism. (Albert Cohen, Sebastian Pop) (20 min)
  * Vectorization and parallelization (more discussions):
   * Interaction with vectorization. (group discussion) (20 min)
   * Interaction with parallelization. (group discussion) (20 min)

 * Testing and Performance Evaluation: (1h) (Sebastian Pop, Dwarak Rajagopal)
   Reproducibility challenges, regression testing, algorithmic complexity:
   * GCC testsuite
   * PCP testsuite
   * Cloog testsuite
   * Polyhedral operations testsuite

 * Community Interaction (1h)
  * Other compilers
  * Work sharing/splitting, responsibilities.
  * Publication policy.
  * Integration of external polyhedral tools (at the level of the polyhedral representation).

== Attendees ==
If you think you can attend the workshop, please add your name to the list.

 * Roberto Bagnara, University of Parma
 * Cedric Bastoul, INRIA
 * Andrey Belevantsev, Institute for System Programming of RAS
 * Albert Cohen, INRIA
 * Tobias Grosser, Student Uni Passau
 * Richard Guenther, SuSE
 * Gautam Gupta, COREquations
 * Christophe Harle, AMD
 * Jan Hubicka (Honza), SuSE
 * Harsha Jagasia, AMD
 * Michael Lai, AMD
 * Sebastian Pop, AMD
 * Dwarakanath Rajagopal, AMD
 * J. (Ram) Ramanujam, Louisiana State University
 * Sanjay Rajopadhye, Colorado State University
 * Jan Sjodin, AMD
 * Sven Verdoolaege, Katholieke Universiteit Leuven

Graphite Workshop

A Graphite Workshop is organized on AMD's campus in Austin on November 16 and 17, 2008.

Day 1: Nov 16

  • Tentative schedule:
    • 9am breakfast
      • Introduction (1h)
      • Polyhedral Compilation Package (PCP) (1h)
    • 12pm lunch
      • Front-end (2h)
    • 3pm break
      • Front-end (continued) (3h)
    • 6pm end

Day 2: Nov 17

  • Tentative schedule:
    • 9am breakfast
      • Middle-end (3h)
    • 12pm lunch
      • Middle-end (continued) (2h)
    • 3pm break
      • Testing and Performance Evaluation (1h)
      • Community Interaction (1h)
    • 5pm end

Agenda

  • Introduction: Led by Sebastian Pop, Albert Cohen. (1h)
    • History and current status of Graphite (tools, representations,
      • SCoP formation algorithm, gloog pretty-printer, analyses, transformations, limitations).
  • Polyhedral Compilation Package (PCP): (1h) (Jan Sjodin)
  • Front-end, static analysis and conversion to polyhedral representation. (5h)
    • SCoP formation vs. code generation. (Tobias Grosser) (20min)
    • Data dependence and reaching definition analysis (Sebastian Pop)
      • Instance-wise dependences and reaching definitions. (Albert Cohen) (10min)
      • Fuzzy analysis. (Albert Cohen) (20min)
      • Interprocedural array region analysis (w/ or w/o LTO). (Honza Hubicka) (20min)
    • Polyhedral representations: Led by Albert Cohen, Roberto Bagnara, Sven Verdoolaege
      • Integration with PIPlib and Omega Test. (Cedric Bastoul, Sebastian Pop, Sven Verdoolaege) (20min)
      • Parma Polyhedral Library. (Roberto Bagnara) (40mn)
      • An Integer Set Library for CLooG. (Sven Verdoolaege) (20min)
      • Improvements: sorting, simplification in context, caching, relaxation. (Albert Cohen) (20min)
    • Interaction with the array middle-end. (Richard Guenther) (20min)
    • Support for reductions. (Albert Cohen) (20min)
    • Direct support for irregular control flow. (Cedric Bastoul) (20min)
  • Middle-end: how to add time, space and machine descriptions (5h)
    • Performance Modeling and Heuristics: Harsha Jagasia, Dwarak Rajagopal.
      • Machine descriptions, cache and locality models. (Harsha Jagasia) (20 min)
      • Computing a schedule. (Harsha Jagasia, Sebastian Pop) (20 min)
      • Parametric tiling support. (Sanjay Radjopadhye, Cedric Bastoul, Albert Cohen) (20 min)
      • Direct search in Farkas coefficient spaces. (Albert Cohen, Cedric Bastoul) (20 min)
    • Array expansion and contraction. (Richard Guenther) (20 min)
    • Unrolling support. (Cedric Bastoul) (20 min)
    • Modulo guard elimination support. (Tobias Grosser) (20 min)
    • Automatic parallelization: Albert Cohen, Jan Hubicka, Sebastian Pop.
      • Loop parallelization with interprocedural analysis. (Honza Hubicka, Richard Guenther) (20 min)
      • Loop selection (with interchange and distribution). (Sebastian Pop) (20 min)
      • Streamization (post/wait synchronizations). (Sebastian Pop) (20 min)
      • Automatic extraction of functional parallelism. (Albert Cohen, Sebastian Pop) (20 min)
    • Vectorization and parallelization (more discussions):
      • Interaction with vectorization. (group discussion) (20 min)
      • Interaction with parallelization. (group discussion) (20 min)
  • Testing and Performance Evaluation: (1h) (Sebastian Pop, Dwarak Rajagopal)
    • Reproducibility challenges, regression testing, algorithmic complexity:
    • GCC testsuite
    • PCP testsuite
    • Cloog testsuite
    • Polyhedral operations testsuite
  • Community Interaction (1h)
    • Other compilers
    • Work sharing/splitting, responsibilities.
    • Publication policy.
    • Integration of external polyhedral tools (at the level of the polyhedral representation).

Attendees

If you think you can attend the workshop, please add your name to the list.

  • Roberto Bagnara, University of Parma
  • Cedric Bastoul, INRIA
  • Andrey Belevantsev, Institute for System Programming of RAS
  • Albert Cohen, INRIA
  • Tobias Grosser, Student Uni Passau
  • Richard Guenther, SuSE
  • Gautam Gupta, COREquations
  • Christophe Harle, AMD
  • Jan Hubicka (Honza), SuSE
  • Harsha Jagasia, AMD
  • Michael Lai, AMD
  • Sebastian Pop, AMD
  • Dwarakanath Rajagopal, AMD
  • J. (Ram) Ramanujam, Louisiana State University
  • Sanjay Rajopadhye, Colorado State University
  • Jan Sjodin, AMD
  • Sven Verdoolaege, Katholieke Universiteit Leuven

None: Graphite_Workshop_Nov08 (last edited 2009-04-20 09:27:43 by kotnet-147)