Course information

• Title: EE 4702 -- Software Optimizations
• Professor: J. (Ram) Ramanujam, 324 Johnston Hall (578-8979) & 345 EE Building, 578-5628
• Email: jxr [at] ece [period] lsu [period] edu)
• Office Hours: To Be Decided
• Grader/TA: To be announced
• Time and Place: MW 3:40pm - 5:00pm in 307 Frey 145 EE Building
• Text: No required text; the class will be based on handouts.
  However, you are encouraged to use the reference books listed below.

  References:

  1. High Performance Compilers for Parallel Computing, Michael Wolfe, Addison-Wesley, 1996.

  2. Optimizing Compilers for Modern Architectures, R. Allen & K. Kennedy, Morgan Kaufman, 2001.
  

  Attention: Graduate students WILL get graduate credit for this course.
  

  Prerequisites: graduate standing OR ((data structures: CSC 3102) AND (computer organization EE 3755))
  

Description

The course deals with principles and practice of software optimizations for modern architectures including CPUs and accelerators such as GPUs. Issues in compilation of programs for high performance computers and accelerators will be discussed in detail. Data dependence testing and program transformations constitute the core of this course.

Course Topics

This set of material is in the class notes. You will find relevant material in Chapters 4-7, parts of 8-10 and 12 from the book by Michael Wolfe. Additional material will be covered if we have time.

• Introduction
• Independence and parallelism detection
• Data dependence testing for array-based codes
• Data dependence testing for other data strauctures
• Standard scalar transformations (constant propagation, ...)
• Loop parallelization and scheduling
• Loop transformations (interchange, skewing, ...)
• Data locality optimizations (i.e., for caches)
• Optimizations for accelerators (GPUs)
• Optimizations for (i) embedded processors, (ii) power/energy reduction

Tentative Grading (will be finalized by last day to add)

2 tests (22.5% each), homework/projects (25%), final (30%).