1 / 12

An Integer Programming Approach to Instruction Scheduling

An Integer Programming Approach to Instruction Scheduling. Matt Streeter Carsten Schwicking. Objective. To use integer programming to obtain optimal instruction schedules Assume partioning has already been done (copies inserted, etc.). Optimization & Constraints. Optimize schedule length

vina
Download Presentation

An Integer Programming Approach to Instruction Scheduling

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. An Integer Programming Approach to Instruction Scheduling Matt Streeter Carsten Schwicking

  2. Objective • To use integer programming to obtain optimal instruction schedules • Assume partioning has already been done (copies inserted, etc.)

  3. Optimization & Constraints • Optimize schedule length • Constraints • Single execution (of each instruction) • Resource • Data dependence • Not yet dealing with crossbar constraint

  4. Formulation Source • Daniel Kaestner and Sebastian Winkel. “ILP-based Instruction Scheduling for IA-64,” in LCTES 2001. • Used a real multiple issue architecture • IA-64 has VLIW characteristics

  5. Integer Programming Formulation • Define a variable for each possible instruction, start time, and function unit tuple • Xkm t  {0,1} • m = instruction id • t = start time (cycle #) • k = execution unit #

  6. Constraints:Instruction Executed Once

  7. Constraints:Resource • Rk = # of execution units of type k • R(n) = all possible execution units for n

  8. Constraints:Data Dependence • where n must not execute until m finishes

  9. Preliminary results • Generated random data dependency trees • Each node has degree chosen uniformly at random from {0, 1, 2, 3}; durations chosen from {1, 2, 3, 4}; function unit assigned at random • 4 function units, one of each type • Maximum height H (=6 here)

  10. Preliminary results • Optimal scheduling only feasible for basic blocks of ~10 instructions • An alternative is to search for locally optimal schedules

  11. Preliminary results • Use simple heuristic (list scheduling) to generate a “center” schedule • Until 10 seconds have elapsed: • For r from 1 to rmax do: • Generate optimal schedule subject to the constraint that the cycle assigned to each instruction is within a distance r of the cycle assigned by the center schedule

  12. Preliminary results

More Related