Skip to end of metadata
Go to start of metadata

Meeting Details

  • Moderator: cds
  • Protocol: cds
  • Attendees:
    • als
    • cds
    • dalu
    • nbw
    • nir
    • psi
    • rvh
    • sdo
    • ssm
    • tga
  • Start: 9:50
  • End: 11:10

Agenda

Visualization of Things (psi)

  • Tool Gropius: A static analysis tool that analyzes multithreaded programs for data races.
    • The tool uses LLVM as well as initial task / thread entry specifications to base its analysis on.
    • New application: Find code areas that can be parallelized.
    • To do that, the tool builds a dependency graph...
  • Dependency graph
    • The dependency graph visualizes access to data. Each node in the graph represents loading or storing a variable and includes the following information:
      • The variable
      • Where in the code the operation occurs.
      • The call stack that is active when the operation occurs.
    • The question is how best to draw the dependency graph.
  • More complex graph
    • The dependency graph is generated from a graph that contains more and more complex information:
      • Function calls.
      • Function contexts (basically, identifiers for functions).
    • The graph can be read as "If this happens, the things its descendants describe must have happened as well."
  • Ultimate goal
    • The ultimate goal is to produce dependency graph layouts that assist finding sets of components that may be parallelized.
  • The data the visualization would be based on is kept in an SQL database.

Example Data

The dependency-graph-stuff.zip file contains examples.

  • No labels