Points-To Set Backends

include/Alias/Infrastructure/PtsSet/ contains alternative data structures for representing points-to sets inside alias analyses.

Backends in this tree:

DenseHashPtsSet for hash-based storage.
ChunkedSparseBitsetPtsSet and BloomBitsetPtsSet for compact bitset variants.
BDDAndersPtsSet for a BDD-backed representation.

Some of these implementations are experimental or selectively wired into the current analyses, but the directory is the natural home for set-representation experiments.

Common set interface

The backends are designed around the operations used in propagation loops: membership tests, insertion, union, subset checks, intersection tests, iteration, clearing, and size queries. In particular, insert and unionWith report whether the set changed, allowing a solver to enqueue dependent constraints only when new points-to facts are discovered.

Choosing a representation

DenseHashPtsSet is a simple baseline built on llvm::DenseSet and is a good fit for irregular, sparse identifiers. ChunkedSparseBitsetPtsSet stores only populated fixed-size chunks, which favors sparse sets with large identifier ranges. BloomBitsetPtsSet combines an exact dynamic bitset with a Bloom filter that can accelerate negative membership and intersection checks; the bitset remains the authority, so Bloom-filter false positives do not change results. BDDAndersPtsSet uses CUDD and supports BDD variable reordering, trading a heavier dependency for compact symbolic representation when it suits a workload.

The header comments identify which alternatives are currently standalone. Before replacing a production backend, benchmark representative inputs and verify that the surrounding solver relies only on operations implemented by the candidate set.