CS/EE6760 Fall 1998 Project 2: Branch Prediction

• dates: out 20-Oct-98, due 29-Oct-98.
• handout: project2.ps, project2.pdf
• simulator and benchmarks: /users/projects/class/6760/fall98/project2/

• 30-Oct-98: OPTIONAL: it turns out that CC1 doesn't run long enough with the given dataset to let the fancy 21264 predictor really get going (it takes a long time to load up all those local state bits). If you want to see results from a more realistic example, try running CC1 on amptjp.i through your analyzer. It takes about 15 minutes or so.
• 27-Oct-98: One bug, several additions:
  • Bug: Problem 2.A.ii. should say an 8K (12,2) predictor.
  • Updated br.c to include more comments.
  • Updated Makefile so that it makes links to cc1.solaris, sor.i and also a central version of cc1.dis (which is 15MB).
• 23-Oct-98: The PATH in a default CoC account doesn't include the right C compiler (it also doesn't include the current directory). The Sun C compiler is /opt/SUNWspro/bin/cc. You can update your PATH or change the CC= line in the Makefile, as you prefer.

  ---------------------------------------------------------------------
  |  guam            | Ultra 1/140, 320MB    | (dedicated to 6760)    |
  |  midway          | Ultra 1/140, 320MB    | Ken's desktop          |
  ---------------------------------------------------------------------
  |  akron           |                       |                        |
  |  boise           |                       |                        |
  |  cheyenne        |                       |                        |
  |  dallas          |                       |                        |
  |  eugene          |                       | first-floor cluster    |
  |  fargo           | Ultra 1/140, 64MB     | (always defer to       |
  |  gary            |                       | anyone at the console) |
  |  winnipeg        |                       |                        |
  |  xenia           |                       |                        |
  |  yorktown        |                       |                        |
  |  zebulon         |                       |                        |
  ---------------------------------------------------------------------

• Defer to the console: Don't use a public-cluster machine remotely unless there's no one at the console.
• Use nice +19: On any machine, preface any job you expect to run for more than about 10 seconds with nice +19 to voluntarily lower its priority.
• Limit scripts: Avoid launching unattended jobs (e.g., scripts of multiple simulation) that you expected to run for more than about a half an hour.
• No parallel processing: limit yourself to at most one 6760-related simulation per person active at a time in the CoC.