-
Notifications
You must be signed in to change notification settings - Fork 223
Solvers
WALA includes several types of solvers applicable for various styles of dataflow analysis. You may find one of these helpful for setting up and solving a particular problem.
The simplest solver is the fixed point solver,
DefaultFixedPointSolver
. This solver simply manages a worklist of
"statements", and "evaluates" statements until there are no further
changes to the solution.
The main implementation benefits of using the WALA solver are a space-efficient representation of dependencies between statements, and the ability to evaluate statements based on various approximations to topological order. The latter feature is crucial for rapid convergence in some problems.
WALA's default IR construction and flow-insensitive pointer analysis, along with several other analyses, use the common fixed-point solver at the lowest level.
The com.ibm.wala.dataflow
package provides a layer on top of the
fixed-point solver to facilitate implementation of Killdall-style
dataflow frameworks. The abstractions here help set up a dataflow system
induced over nodes and/or edges of a graph.
To solve backwards dataflow problems, simply reverse graph edges using
GraphInverter.invert()
, and set up flow functions appropriately.
The TabulationSolver
provides an implementation of the RHS algorithm
for Sharir-Pneuli functional context-sensitive dataflow analysis,
including IFDS. The TabulationSolver
implementation has been tuned for
space efficiency, and includes various extensions to help with features
such as exceptional control flow.
To solve backwards dataflow problems, start with the
BackwardsSupergraph
.