This experimental patch implements an algorithm to detect control structures in the CFG, such as If-Then, If-Then-Else, Self-Loop etc. This analysis is kind of an extension of interval/region analysis to detect more than just loops. The algorithm and data structures used in this implementation are based on the description given in "Advanced Compiler Design and Implementation" -- Steven. S. Muchnik (Section 7.7 "Structural Analysis"). The result of the analysis is a tree called the control-structure-tree of the control flow graph of the function begin analysis.
Typical uses for this analysis are:
- Perform control flow transformations based on the control structure tree. Some control transformations are easier to reason about using this tree.
- IR to source transformations. Detecting source level constructs in the CFG is usually useful in generating high level code from an already reduced IR.
- Similar to the uses of interval/region analysis, control structure analysis also can be used to speed up data flow analyses.
The code, although tested and used internally, needs further work:
- Make it an actual LLVM analysis.
- Provide a richer set of APIs for updates.
- Improve analysis time.
I'm not sure how useful such an analysis would be to the community. If people find it to be useful, I'm open to working on it further and submitting it. Any feedback will be appreciated.