15-745:  Optimizing Compilers

Andersen's Pointer Analysis for CASH compiler framework

Deepak Garg (dg AT cs.cmu.edu)
Himanshu Jain (hjain AT cs.cmu.edu)

We present the design and results of our implementation of Andersen's
pointer analysis on the Pegasus intermediate representation. We use a
constraint based formulation of this analysis and use the Banshee
constraint solving framework to solve the generated constraints. The
results of the pointer analysis are then used for analyzing the
token dependencies between the various memory related operations in
the Pegasus IR. Our experimental results indicate that significant
percentage of token dependencies in Pegasus IR are redundant. These
dependencies can be removed to increase the parallelism in the Pegasus IR.

Project proposal: (pdf) (ps)
Checkpoint presentation: (ppt)
Poster presentation slides: (ppt)
Final report: (pdf) (ps)