Path Profile Estimation and Superblock Formation

1
Path Profile Estimation and Superblock Formation

• Jeff Pang
• Jimeng Sun

2
Motivation
Optimize
Compile
Run
Profile

• Continuous Optimization
• Dynamic Optimization
• Realistic Profiles

3
Challenges
Optimize
Compile
Run
Sample
Estimate
Profile
Path Profile

• Automate optimization
• Low overhead profiling
• Accurate profiling

4
Project Goals
Superblock Formation
Run with Simulated PMU
Path Profile
Sample
Data Mining based Path Estimation

• Simulate performance monitoring unit
• Like in Pentium 4, Itanium, PPC 970, etc.
• Allows sampling of last couple branches
• Estimate full path profile using samples
• Leverage data mining techniques similar to
  PageRank
• Validate by doing Superblock formation
• Powerful optimization to improve scheduling
  (especially on VLIW processors)

5
Superblock Formation
6
Project Outline
instrument
instrumented program
source
label
c2dil
superblock
optimized program
Offline estimator
path profile

• Implement PMU simulator and Superblock
  optimization as SUIF passes
• Implement Estimator offline using sampled branch
  profiles and SUIF CFG

7
Completed
instrument
instrumented program
source
label
c2dil
superblock
optimized program
Offline estimator
path profile

• PMU Simulation modified HALT profiling lib
• Initial offline estimator (?)

8
Initial Results
9
Todo
instrument
instrumented program
source
label
c2dil
superblock
optimized program
Offline estimator
path profile

• Superblock SUIF pass
• Have c2dil build hyperblocks using superblocks
• Refine path estimation heuristics

10
Questions or Comments?

• Anyone have a good scheduler from Assignment 2?
• Better scheduler better comparison of
  superblock scheduling performance
• The HALT lib maybe useful
• If you need to uniquely label branches, basic
  blocks, loads, stores, etc.
• Insert instrumentation at those points

11
Extra Slides
12
Related Work

• Shye, et al. 2005
• Used heuristics to rebuild path profiles using
  PMU partial path samples
• Achieved 80-99 path accuracy with lt5 overhead
• Did not evaluate impact on optimizations
• Chen, et al. 2003
• Used PMU partial path sampling to dynamically
  form hot traces
• Used to adapt to dynamic phase transitions