EE 7700-1              Fall 2003              Call No.: 1842
Compiler Optimizations

Professor:	J. (Ram) Ramanujam, 345 EE Bldg., 578-5628    Email:
Office Hours:	MWF 8:30 - 9:40; MWF 11:30 - 12:00
Time:	10:40 - 11:30 MWF, 149 EE
Reference:	Class notes and recent papers from the literature
Prerequisites:	Graduate standing

Description

The course deals with principles and practice of compiler optimizations. The interaction between architectures and the choice of compiler optimizations will be discussed. The issues of data dependence and flow analysis, and program transformations for a variety of objectives for a wide range of processor architectures (single processor systems, shared- and distributed-memory parallel processors, special-purpose processors, etc.) will be covered. Transformations that enhance data locality and optimize memory usage will get a lot of coverage. In addition, compilation starting from high-level specifications will be a key focus.

Course Outline

1. Review:     processor architectures; latency, locality, synchronization
   
   
2. Techniques for Dependence Detection:
   
   
   • dependence analysis: integer-programming based tests, value-based tests
   • control flow graph, dominators, interval analysis, dataflow analysis
   • program dependence graph (PDG) and static single assignment (SSA) form
   • interprocedural analysis
   
   
3. Program Restructuring:
   
   
   • scalar transformations; vectorization; parallelization
   • unimodular transformations: loop interchange, skewing, wavefront
   • loop fusion and distribution
   • non-unimodular transformations
   • iteration space tiling
   • communication optimizations
   • transformations for data locality: registers, caches, I/O
   • exploiting instruction-level parallelism
   • storage related optimizations; memory usage minimization
   • arithmetic optimizations; high-level specification
   • transformations for special-purpose processors